Split (1)

train · ~5.33M rows (showing the first 168k)

id string	text string	source string	added string	created string	metadata dict
PMC1087873	Editorial ========= \"In France, one can be an anarchist, a communist or a nihilist, but not an anti-Pastorian. A simple question of science has been made into a question of patriotism.\" August Lutand \"Pasteur et la Rage\" 1887 \" \[Pasteur\] was the most perfect man who has ever entered the kingdom of science.\" Stephen Paget \"The Spectator\" 1910 To continue the series of *\"The Classics of Immunology\"we turn next to Louis Pasteur\'s development of the live attenuated anthrax vaccine, first published in 1881 \[[@B1]\]. In this publication, Pasteur claimed that he could attenuate the anthrax organism by simply culturing it in air at elevated temperatures. Together with his announcement a year previously of a vaccine for chicken cholera \[[@B2]\], he thus introduced the concept of attenuating the virulence of organisms so that they could be used as live vaccines. Audaciously, he speculated that all microbial diseases could be prevented this way: one only had to learn how to culture the offending agent. Of course we now know that it is not quite as simple as Pasteur originally claimed, particularly for bacterial infections. A brief review of this interesting period in early immunology is warranted before one can fully appreciate our present day knowledge. When the anthrax spores were sent through the mail soon after the terrorist attacks on 9/11/2001, I was perplexed. I really had no idea what disease, or diseases, that were caused by the anthrax microbe. I vaguely recalled from my medical school days that the anthrax microbe is a Gram-positive rod. I also recalled that Louis Pasteur had developed a vaccine against anthrax, but that was about all that I could remember. Accordingly, I consulted my microbiology texts and went on-line, to see what I could learn. From my texts I learned that the anthrax organism has the unusual property of undergoing a metamorphosis from a multiplying vegetative bacteria into a spore(from the Greek, sporos= seed), which is defined as an inactive resting or resistant form produced within the body of a bacterium. I also learned that anthrax was the first disease actually proven to be caused by a microbe, by Dr. Robert Koch, who reported on his experimental findings in 1876 \[[@B3]\]. This piece of information really piqued my interest, because I had no idea that anthrax was so central to the history of medicine. Like most other physicians of my generation, I had never seen nor heard of an actual case of anthrax. However, I did know that anthrax was often mentioned as a possible bio-weapon that could be used by terrorists, but I did not know why it was so dangerous, or exactly what symptoms were produced by the infection. I remembered visiting one of my friends at the Frederick Cancer Research Facility (FCRF) in Frederick Maryland in the 1970\'s. The FCRF was originally Fort Dietrich, which from 1943 to 1969 was the U.S. Army base devoted to chemical and biological warfare research. My friend pointed out a large 5--6-story building that had all of the windows and doorways covered over with concrete blocks. This building had housed all the anthrax research, and it was still contaminated by anthrax spores, which were extremely deadly, so that the building had been hermetically sealed, rather than razed, when President Nixon discontinued chemical and biological warfare research. As far as I know, that building is still there in Frederick. To learn more about Robert Koch and anthrax, I turned to a book by Eli Metchnikoff, published originally in 1905, entitled \"The Founders of Modern Medicine: Pasteur, Koch, and Lister\"\[[@B4]\]. Metchnikoff was a Russian zoologist, who observed the phenomenon of phagocytosis for the first time, while studying starfish larvae off the coast of Naples in the 1870\'s \[[@B5]\]. Subsequently, in 1885, he was recruited by Pasteur to become the first Chef de Serviceat the newly formed Institut Pasteur in Paris, where he championed the idea that cells actually are responsible for immunity. Beginning with Ignaz Semmelweis\' observations on the possible cause of puerperal fever (childbed fever) in 1850 \[[@B6]\], the notion that small microscopic living things might cause disease and death began. Then, Louis Pasteur\'s demonstration in 1857 that fermentation of lactic acid into alcohol and carbon dioxide was actually caused by living organisms (animal infusoria), and not by the then popular theory of \"spontaneous generation\", set the stage for the importance of microbes in everyday processes \[[@B7]\]. Subsequently, Joseph Lister\'s 1867 descriptions of the use of antiseptics in the practice of treatment of compound fractures and at surgery \[[@B8],[@B9]\], promoted a growing notion that removal of microbes isolated from wounds and other degenerative tissues could improve the outcome of the patient. However, the most common belief still held was that any microbes found in suppurating tissues were the result and not the cause of the fetid, morbid state. The morbidity was thought to arise spontaneously via chemical reactions. Any association with living microbes was considered fortuitist. In the words of Metchnikoff, \"A powerful impulse was necessary to change this inchoate idea of organized (chemical) ferments into a rigorously proven scientific truth that microbes were responsible (for putrifaction and disease). Robert Koch started such an impetus in his 1876 paper on anthrax. This young health officer in the little city of Wolstein, a god-forsaken hole in Posen (Prussia), suddenly came into the limelight of science. His work was indeed a model of true scientific creativeness. Living in a region in which anthrax was endemic, he set about to study it, without the help of laboratory or library, and was always thrown back on his own resources. He worked in his own rooms where for lack of gas illumination he was obliged to use a petroleum lamp. By means of plates covered with moist sand he constructed a semblance of an apparatus for growing cultures of bacteria. Nevertheless, he achieved results superior to anything yet accomplished. He was the first to succeed in changing the thread-like microscopical corpuscles identified by others (in France) into identifiable filaments (chains of rods) and then into beads consisting of minute grains, the spores. This great discovery of the spore of anthrax removed all doubts regarding the role of bacteria in the causation of anthrax, for it illuminated all points hitherto left unexplained.\" Throughout medieval times, anthrax was a disease primarily of livestock, and it still is considered so, which explains why I was unfamiliar with it. In humans, the most common affliction is a skin inflammation that matures into a very characteristic ugly black eschar, from whence the disease was named from the Greek: anthrax = coal, carbuncle. In the 19^th^and early 20^th^centuries, cutaneous anthrax was also known as \"wool sorters disease\", because farmers and woolen workers would contract it from handling animals and wool contaminated with anthrax spores. For the livestock industry, anthrax was a serious problem, in that many animals would succumb to a more severe disease manifested by both gastrointestinal and pulmonary symptoms. Once animals died and their corpses were allowed to disintegrate in a pasture, it was well known that the particular pasture was thereafter suspect, in that the reintroduction of fresh animals in the spring often resulted in the reappearance of the disease. Thus, as a result of Koch\'s experiments, it was realized that the ability of the microbe to sporulate enabled it to withstand the harsh temperatures and conditions that often occur during the winter months. Nowadays the livestock industry is protected from anthrax by vaccination. This protection of farm animals extends to farmers and other humans, such as textile workers and vets, so that anthrax infection of humans has become exceedingly rare, especially since the time of Koch. However, what is the situation with an anthrax vaccine for humans? In our country, the only vaccine available is only being administered to soldiers. Since the postal anthrax scare of 2001, the Administration, via the Defense Department ordered all new recruits to receive the vaccine. However, the vaccine is reportedly not 100% effective, requires 6 injections over a period of a year and a half, and is associated with side effects/toxicities that have led some army personnel to refuse it. Louis Pasteur introduced a live attenuated anthrax vaccine more than 100 years ago. So why is the currently used vaccine so cumbersome and toxic? Also, is the current vaccine similar to the one introduced by Pasteur? Is the vaccine that is used for animals the same as the one used for humans? When contemplating these questions, I remembered that in 1998, while in France, I happened to read an article in Le Figaro, which announced that the anthrax vaccine introduced by Pasteur in 1881 was in fact not the live attenuated vaccine that Pasteur had suggested he used at the time. Instead, the vaccine was a chemically killed vaccine that had been developed and introduced by one of Pasteur\'s rivals, a Dr. Toussaint, who was a veterinarian from Toulouse, France. To understand the implications of the announcement by a leading French newspaper that the icon of the French scientific accomplishment and integrity had committed what amounts to scientific fraud, it is necessary to research the source documents of Pasteur\'s experiments and publications. After Pasteur, the chemist, had dispensed with the theories of \"spontaneous generation\" as responsible for the chemical changes responsible for fermentation of sugar into alcohol in 1857, he went on over the next 20-years to perform a series of careful microbiological experiments in applied science in studies of bacterial contamination confronting the silk worm industry, as well as the wine, vinegar and beer industries, thereby establishing the importance of microbes for everyday endeavors. In the process of doing so, he became almost deified in France, if not the rest of the world as the icon of a scientist. Thus, in April of 1878, just two years after Koch\'s revolutionary publication proving the microbiological cause of anthrax, Pasteur presented a \"Summary\" to the Academy of Sciences, essentially claiming priority of the germ theory of disease \[[@B10]\]. According to Pasteur: \"The only way currently available to science to experimentally prove that a microscopic organism is the cause of both the illness itself and its transmission, is to subject the microbe to serial cultures.\" Pasteur then goes on to describe his experiments with the anthrax bacillus, never mentioning that Koch had already demonstrated the culture of the anthrax microbe two years earlier. In concluding, he states that: \"I ask the Academy not to dismiss these curious results before I demonstrate one important theoretical conclusion. We insist on demonstrating at the start of these studies(that are opening a whole new world of knowledge) a proof that the cause of transmissible, contagious and infectious diseases resides essentially and uniquely in the presence of microorganisms.\"* Not yet two years later, in February of 1880, Pasteur again presented to the members of the Academy a treatise entitled \"Of Infectious Diseases, Especially the Diseases of Chicken Cholera\"\[[@B2]\]. In this presentation, Pasteur first reminded the members that the theory of spontaneous generation was false, as demonstrated by his very own experiments performed more than 20 years previously. He then set the stage by stating: \"Infectious diseases consist of most of the major disasters, such as smallpox, scarlet fever, rubella, syphilis, glanders, anthrax, yellow fever, typhus, and bovine plague.\" Pasteur then discussed the phenomenon of vaccination as introduced by Sir Edward Jenner almost 100 years before as something already known by the common man, and essentially claims immunity for all other microbes for himself: \"The practices of vaccination and variolization have been known in India for the longest time. Even before Jenner demonstrated the efficacy of vaccinia, people of the countryside where he practiced already knew that cowpox protected against smallpox. The facts about vaccinia are unique, but the facts about nonrecurrance of virulent diseases are more general. The organism never expresses twice the effect of chicken pox, scarlet fever, typhus, plague, smallpox, syphilis and others, as the immunity lasts for a long time at least.\" Pasteur then introduced the problem of chicken cholera, and mentioned that M. Toussaint, a professor at the veterinary school of Toulouse had been the first to culture and isolate the microbe that he thought to be responsible for the cause of the disease of chickens. Pasteur went on to say that he had discovered an improved culture medium for the microbe, and\.... \"We can diminish the microbe\'s virulence by changing the mode of culturing. This is the crucial point of my subject. I ask the Academy not to criticize for the time being, the confidence of my proceedings that permit me to determine the microbe\'s attenuation, in order to save the independence of my studies and to better assure their progress.\" With this presentation to the Academy, Pasteur merged the science of microbiology with that of what subsequently became known as immunology for the first time. As well, this presentation to the public revealed a crucial aspect of Pasteur\'s experiments and thinking as to his perception of the importance of his findings. In France at the time it was common practice to submit a sealed note (called a pli cachete) on an important scientific discovery to the Academy of Sciences to secure or protect one\'s priority. By comparison, an official patent application (brevet d\'invention) was necessary to establish one\'s right to the commercial exploitation of that discovery. Pasteur thus kept it a secret as to exactly how he had attenuated the virulence of the chicken cholera microbe for more than 9 months, until October of 1880. Eventually, Pasteur disclosed that his methods simply involved culturing the microbe exposed to atmospheric oxygen for prolonged intervals, i.e. longer than 2--3 months. However, he never explained why oxygen should weaken the microbe\'s virulence, especially as the chicken cholera microbe is an aerobic organism. It is likely that he did not want to risk others trying to repeat his methods, both from the standpoint of the fear of their success as well as their failure. Pasteur then described using the \"live attenuated\" cholera vaccine to immunize animals against lethal challenges of the microbe, and stated that \" It seems as if the initial microbe inoculations(of the live attenuated vaccine) have*depleted*a certain element that healing does not reconstitute and that the absence of which hinders the development of this small organism (when re-inoculated a second, third, and fourth time). *This explanation will without doubt, become general and applied to all infectious diseases.* I would like to point out to the Academy two main consequences to the facts presented:*the hope to culture all microbes and to find a vaccine for all infectious diseases that have repeatedly afflicted humanity, and are a major burden on agriculture and breeding of domestic animals.\"* The importance of Pasteur\'s theory, i.e. that it was possible to attenuate the virulence of allmicrobes, simply by passing them in special culture conditions can only be appreciated by understanding the competition that developed between Pasteur and Toussaint in the summer of 1880 involving different approaches to the creation of a vaccine for anthrax. Pasteur had begun working on a vaccine for anthrax 3 years previously, soon after Koch\'s announcement on the isolation of the causative anthrax bacillus \[[@B11]\]. On July 12, 1880, Henri Bouley (a fellow veterinarian and friend of Toussaint) read before the Academy of Sciences a report from Toussaint (who was not a member of the Academy), which described the initial results of his experimental vaccine trials. In contrast to Pasteur\'s \"live attenuated\" vaccine, Toussaint generated his vaccine simply by killing the bacilli by heating for 10 minutes at 55°C. Using this vaccine, Toussaint had conducted trials on 8 dogs and 11 sheep. Of the 8 dogs, 4 had been injected with the vaccine and had survived a series of 4 successive injections of virulent live anthrax. By comparison, all 4 unvaccinated dogs succumbed to the first injection. A similar result was obtained with the sheep. In August, while vacationing, Pasteur heard the news of Toussaint\'s vaccine experiments from Bouley. He responded as follows \[[@B11]\]: \"My very good colleague, Since yesterday morning, when I received your letter, the extracts of the journals, and the Summary of the Academy of Sciences-all at the same time -I have been in astonishment and admiration over the discovery of M. Toussaint-in admiration that it exists, in astonishment that it can be. It overturns all the ideas I had on viruses, vaccines etc. I no longer understand anything. Ten times yesterday, I had the idea of taking the train to Paris. I really cannot believe this surprising fact until I\'ve seen it, seen it with my own eyes, though the observation that establishes the fact makes me want to confirm it to my own satisfaction. The Academy of medicine has thus received a severe lesson. It will surely have grasped that one does not deal lightly with facts of this order in public, that contemplation is appropriate in the face of such solutions to such problems. I am too moved to write more fully. I have dreamed about it, both asleep and awake, all through the night. Best to you and thanks. L. Pasteur Pasteur\'s expression of surprise and agitation makes sense only in the context of his general theoretical views on diseases and immunity. Because of his successes in his studies of the metabolism of living microbes, Pasteur naturally extended his concepts to immunity. Linking immunity with the biology of microbes, especially the nutritional requirements of the virulent microbe, he suggested that the tissues of the invaded host might contain only trace amounts of some nutrients required for the growth and survival of the microbe, just as some culture media contained only trace amounts of vital nutrients. If so, the invading microbe might soon exhaust the supply of these trace substances, rendering the host an unsuitable medium for the microbe\'s subsequent cultivation. Thus, the host would not support the growth of a subsequent infection by the virulent microbe, and would be \"immune\" (Latin, immunis; free, exempt). Also, an attenuated microbe would be one that had been stressed by cultivation either in vitroor in vivoin an environment that was limiting in essential nutrients, thereby somehow loosing its virulence. Thus, central to Pasteur\'s conception of immunity, was the biological activity of a living, if attenuated, microbe that depletedthe host of essential nutrients. It was Toussaint\'s claim that he had in fact produced a \"dead\" vaccine against anthrax that moved Pasteur to state that \"it overturns all the ideas I had on viruses, vaccines, etc.\" As one might imagine, given Pasteur\'s theory, and his statements already made to the Academy, his lance had been planted. He could not, and would not, graciously admit that he was wrong. The story only goes downhill from this point. In the public critique that Pasteur was soon to issue against Toussaint\'s work, his central theoretical concern was precisely the question of \"live vs. dead\" vaccines. In August, 1880, soon after announcing his heat killing method of vaccine production, Toussaint switched his procedures and had begun to subject the bacilli to the action of carbolic acid, which had long been used as a disinfectant and had more recently become famous as Joseph Lister\'s \"antiseptic\" of choice for the treatment of surgical wounds. Pasteur did not announce the discovery of his own \"live attenuated\" anthrax vaccine until February 28, 1881. Of significance, Pasteur linked his new vaccine with his earlier chicken cholera vaccine by ascribing attenuation in both cases to the action of atmospheric oxygen. However, there was an important difference between the production methods of the two vaccines. Unlike the chicken cholera microbe, the anthrax bacillus formed spores that \"resisted the attenuating effects of atmospheric oxygen\". It had taken much time and effort to ascertain that a spore-free culture of anthrax could be produced at a temperature of 42°--43°C. Subsequently, on March 21^st^, Pasteur reported further successful results testing his vaccine in sheep, which stimulated a challenge by a veterinarian, Hippolyte Rossignol from Pouilly-Le-Fort, to test the new vaccine at his farm in Melun, 40 kilometers from Paris. Examination of Pasteur\'s lab notebooks \[[@B11]\] reveals that he had been conducting small trials, testing his vaccines in animals during this time, with less than conclusive results as to the protective efficacy of the live atmospheric oxygen attenuated vaccine. However, at the same time, Pasteur\'s lab was testing a vaccine prepared by M. Chamberland, who was experimenting with a \"dead\" vaccine prepared by chemical treatment with potassium-bichromate. In small-scale tests this vaccine was working. If Pasteur had failed to accept Rossignol\'s challenge, he would certainly have damaged his priority competition with Toussaint. Moreover, there were already rumors that Pasteur was really seeking to profit financially from his \"secret remedies\" against livestock diseases. Therefore, Pasteur \"impulsively\" accepted the challenge and on April 28, 1881, and he signed a detailed and demanding protocol, which was performed in May. There is a wonderfully detailed accounting of the drama of the public trial that Pasteur publicly presided over on June 2, 1881 \[[@B12]\]. There were more than 200 observers, including government officials, local politicians, veterinarians, farmers, agriculturists, cavalry officers and newspaper reporters. Of 50 sheep in the trial, half were vaccinated on May 5^th^and May 17^th^, while the other half served as unvaccinated controls. All of the sheep were then challenged with a virulent culture of anthrax bacilli on May 31st. Just 3 days later, all of the vaccinated sheep were alive, while most of the unvaccinated sheep were already dead, with the remaining obviously very ill. Only Pasteur and his collaborators knew of the real nature of the vaccine used for this famous trial. Pasteur had not used the live attenuated vaccine that he had emphasized was so important for his success with chicken cholera. Instead, the \"dead\" vaccine of Toussaint prepared by Chamberland by treatment with potassium-bichromate was used \[[@B11]\]. The up-shot of this public demonstration of Pasteur\'s vaccine was that he received credit for developing the first successful vaccine against anthrax. Toussaint subsequently published only 2 more scientific papers before he died in 1890 at the age of 43, after suffering a mental breakdown \[[@B11]\]. It was not until 1998, that the French government officially recognized Toussaint\'s vaccine as the first effective vaccine against anthrax. It is noteworthy that Robert Koch, who became one of Pasteur\'s chief competitors, hailed Toussaint as the worthy inventor of vaccination against anthrax, and persistently denigrated Pasteur\'s contributions to microbiology \[[@B4]\]. There are many other questions that remain unanswered, such as the nature of the vaccine that Pasteur\'s laboratory supplied to the many people who requested doses for their animals. Parenthetically, it is noteworthy that the vaccine was manufactured commercially by Pasteur\'s team, and yielded a substantial income for the new Pasteur Institute, which was initiated four years later, in 1885 \[[@B11]\]. As well, what of the others in the Pasteur group, all of who knew of the real nature of the vaccine that was used at Pouilly-Le-Fort? Probably of most importance, what became of the concept of attenuating microbes by exposing them to atmospheric oxygen? Surely, all competent bacteriologists who worked in the early and mid 20^th^century had to know that Pasteur had been wrong, and that it was impossible to attenuate aerobic microbes by simply culturing them in the open air. Why was this not aired? Fast-forwarding to the present, I have asked Julia Wang and Michael Roehrl to bring us up-to-date on the present state of the art of anthrax vaccine research. As detailed in their excellent review, the nature of the anthrax vaccine that is in use presently to immunize at-risk wool mill workers, veterinarians, laboratory workers, livestock handlers, and members of the Armed Service is a cell-free filtrate. The vaccine was developed in the 1950s and 1960s for use in humans and was licensed by the FDA in 1970. It has undergone extensive testing in monkeys and has been found to be effective in protecting against pulmonary anthrax after an experimental aerosol challenge. The remarkable virulence of anthrax, which makes it such an attractive microbe for bio-warfare, resides in several unique features, including its capability to sporulate, thereby surviving extremes of the environment, its capsule, which impairs phagocytosis, and its toxins, as well as how the toxins interact with and eventually incapacitate the immune system. A lot has changed since the days of Pasteur and Toussaint. We are fortunate that now we have a better understanding of bacteria in general and the anthrax bacillus in particular. Now it is possible to make a safe and effective vaccine for such a virulent organism, based not on a live attenuated vaccine as proposed by Pasteur, but a vaccine more like the inactivated preparation originally developed by Toussaint. Access to the papers referred to in this editorial can be obtained at \[[@B13]\]	PubMed Central	2024-06-05T03:55:56.190446	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087873/", "journal": "Med Immunol. 2005 Apr 18; 4:5", "authors": [ { "first": "Kendall A", "last": "Smith" } ] }
PMC1087874	Review ====== Introduction ------------ In the last decade, protein aggregation has moved beyond being a mostly ignored area of protein chemistry to become a key topic both in medical and biotechnological sciences \[[@B1]\]. The biological significance of protein deposition has been shown to be much higher than formerly thought. First, because the presence of insoluble protein deposits in human tissues correlates with the development of some debilitating human disorders of growing incidence such as Alzheimer\'s disease, Parkinson\'s disease, type II diabetes and the transmissible spongiform encephalopathies \[[@B2]-[@B4]\]. Second, because it has been shown than under cellular stress conditions, such us severe heat, massive protein misfolding exceeds the buffering capacity of the folding quality machinery and results in the aggregation of proteins, which usually results in cell death \[[@B5],[@B6]\]. Finally, the use of bacteria as factories for recombinant expression is limited by their intrinsic tendency to accumulate the target protein into inactive insoluble aggregates, called inclusion bodies (IBs). IBs are dense, amorphous protein deposits that can be found in both the cytoplasmic and periplasmic space of bacteria \[[@B7]-[@B11]\]. In fact, the formation of IBs is the main bottleneck in protein production, narrowing the spectrum of relevant polypeptides obtained by recombinant techniques and hampering the development of top priority research areas such as the de novodesign of novel proteins, the rational modification of natural proteins or structural and functional genomics. The rising recognition of the crucial significance of protein aggregation has resulted in a number of recent reviews \[[@B12]-[@B19]\]. This review focuses mainly on the role played by intrinsic polypeptide properties in protein aggregation. One should distinguish between precipitates, in which proteins maintain the native folded conformation and aggregates, in which proteins adopt new non-native structures. The first type of self-assembly is generated during random precipitation of already native protein due to an environment promoted reduction of solubility in the polypeptide chain. Examples of these processes are salting out by ammonium sulfate or isoelectric precipitation. Reducing ionic force or shifting solution\'s pH results in immediate dissolution of these precipitates. The second type of macromolecular structures exhibits, without exception, an increase in β-sheet secondary structure content relative to the native conformation and very high concentrations of denaturants or detergents are needed to dissolve them into mainly unfolded polypeptide chains. We will focus our attention on these aggregates, which include amyloid fibrils, thermal aggregates and bacterial IBs. The progress made on the control of their aggregation propensities by means of primary sequence modulation is discussed. Protein aggregation is usually a specific process ------------------------------------------------- Protein aggregation has long been considered to be an unspecific process driven by the establishment of non-native contacts between proteins in totally or partially unfolded conformations to form a disordered precipitate. This idea was sustained in part by the diversity of morphologies of aggregates that were observed by techniques such as electron microscopy and atomic force microscopy \[[@B20]\]. This way, the typical amyloid aggregate is a long, straight and unbranched fibril with a diameter between 40 Å and 120 Å \[[@B21]\], whereas inclusion bodies appear as bigger globular electro-dense structures seen as refractile bodies under phase contrast microscopy usually with near 1 micron in diameter \[[@B22]\] and thermal aggregates are usually amorphous \[[@B23]\]. Recent work shows however that often aggregation is a much more specific event than previously expected at least in amyloid fibrils and bacterial IBs \[[@B24]-[@B27]\]. In fact, for many biotechnologically relevant proteins, isolation of the IBs is an efficient initial step in the purification process, since they contain usually more than 90% of recombinant protein \[[@B28]\], other proteins trapped in the aggregates are proteolytic fragments of the aggregating protein \[[@B29]\], other aggregation-prone polypeptides deposited by titration of chaperones during recombinant expression \[[@B30],[@B31]\] or even contaminants from the purification process \[[@B11]\]. Similarly in Alzheimer and related neurodegenerative diseases in vivoamyloid plaques are composed primarily of the pathogenic aggregating protein rather than resulting from a widespread recruitment of other amyloidogenic proteins, although proteins such as proteases or chaperones have been also found to co-localize in the amyloid deposits \[[@B32]\]. Amyloid fibrils are thought to form trough self-assembly of protein monomers via a nucleation dependent pathway similar to the highly ordered process of protein crystallization \[[@B33]\]. This mechanism is also behind physiological ordered protein aggregation processes as viral coat assembly, microtubule formation or flagellum formation \[[@B33]\]. All these processes are characterized by an initial slow nucleation phase, in which the protein associates to form ordered oligomeric nucleus followed by a growth phase, in which the nucleus rapidly growths to form larger insoluble polymers. Addition of preformed protein nucleus during the lag time results in immediate polymerization. All these aggregation processes and in particular amyloid fibril formation are highly specific. This way, in the aggregation of β-amyloid protein, islet amyloid peptide, transthyrretin, and prion protein the formation of amyloid fibrils is not seeded by preformed fibrils of similar amyloidogenic proteins \[[@B34]-[@B36]\]. Although it has been shown that some amyloid fibrils can accommodate up to 1% of a foreign peptide, indicating than some co-aggregation can occur \[[@B37]\], the efficiency of this event decreases rapidly as differences in protein sequence of co-aggregating proteins increases showing that specific protein-protein interactions are needed for amyloid fibril formation to occur \[[@B38]\]. Aggregation into IBs during recombinant protein expression has been usually though to occur via non-specific association of hydrophobic patches on the surface of folding intermediates. However, the reduced number of IBs (usually one) formed during recombinant protein expression in bacteria suggested that the may be formed by the growth of a reduced number of \"founder\" aggregates in a nucleation-like mechanism. In this respect, the aggregation of the P22 coat protein has been extensively characterized \[[@B39]-[@B42]\] and it was demonstrated that when partially folded species of this protein where mixed in vitrowith those of tailspike protein, no co-aggregation occurs, despite the fact that both form IBs when expressed individually in bacteria \[[@B43]\]. The folding intermediates for each protein preferred to self-associate indicating specificity in the in vitroaggregation process and suggesting that specific interactions may underlie IBs formation in the cell. Very recently, we have confirmed this extent by showing that the preformed IBs of an aggregation-prone β-galactosidase variant are able to act as effective aggregating cores for the aggregation of its soluble, partially folded counterpart in a dose-dependent manner \[[@B27]\]. Moreover, the aggregation process is highly specific as shown by the fact that preformed IBs promote deposition of homologous but not heterologous polypeptides. Both protein sequence and conformation appear to play a role in the establishment of specific intermolecular contacts between aggregating polypeptide chains to form IBs, since aggregated β-galactosidase moiety in IBs do not recognize the properly folded tetrameric enzyme \[[@B27]\]. Inclusion bodies in mammalian cells, the so-called aggresomes, are far more complex structures that those in bacteria containing many proteins, including molecular chaperones, components of the ubiquitin-proteasome system, centrosomal material, and cytoskeletal proteins \[[@B44]\]. This suggested that co-aggregation of misfolded, damaged, or mutant proteins with normal cellular proteins could explain both the presence of multiple proteins in IBs and the toxicity associated with protein aggregation in many neurodegenerative diseases \[[@B45]\]. However, also in this complex system, protein aggregation into IBs exhibits exquisite specificity even among extremely hydrophobic substrates expressed at very high levels \[[@B46]\]. Thus, independent of the source, both amyloid fibril formation and IBs aggregation depend, at least partially, on the formation of specific protein-protein interactions between non-native species. Different polypeptides aggregate into similar structures -------------------------------------------------------- The formation of amyloid fibrils was initially associated to a reduced number of proteins related to recognized pathological situations. Nevertheless, a growing number of globular proteins not related to disease can be induced to generate similar fibrils in vitro, albeit in some cases only in non-native conditions, leading to the suggestion than the ability to form amyloids is intrinsic to many or all polypeptides when their normal folding pathways are compromised \[[@B47]-[@B50]\]. This appears to be true for IBs as well since deposition in such structures has been reported in the recombinant expression of many, but not all, heterologous genes and in the high level expression of several endogenous genes \[[@B7],[@B51],[@B52]\]. No sequence or structural similarities are apparent between any of the proteins that display the ability to form amyloids. Prior to fibrillation, amyloidogenic polypeptides may be rich in β-sheet, α-helix, β-helix, or combine α-helices and β-sheets. They may be globular proteins with a stable unique conformation in the native state or belong to the class of natively unfolded proteins. Despite these differences, the fibrils formed by different polypeptides display many common properties including high content of β-sheet secondary structure forming a core cross-β architecture in which continuous β-sheets are formed with β-strands running perpendicular to the long axis of the fibrils \[[@B53]\]. As in the case of amyloids, proteins incorporated in IBs are not related either structurally or sequentially and deposition during heterologous expression in bacteria has been reported for small, large, monomeric or multimeric proteins. The internal architecture of IBs has long thought by molecular biologists to be amorphous, despite the fact that several observations in the early 90\'s pointed to the presence of ordered structure in IBs \[[@B54]-[@B56]\]. The use of attenuated total reflectance FTIR in IBs formed by all-α, all-β or α +β showed that in all cases, even for all-β proteins, significant new β structure, compared to that in the native conformation, was observed. Interestingly, the amount of secondary structure in the inclusion body varies from one protein to another, as does the amount of disordered structure. More recently, others and we have recapitulated these studies in previously unexplored protein systems, showing clearly that the intermolecular interactions leading to aggregation in IBs in the cell involve β-sheet like interactions \[[@B27],[@B57]\]. Although the exact nature of the intermolecular interactions is unknown, and could be different in different IBs, the overall FTIR data suggest that the newly formed β-sheets in IBs are tightly packed with short hydrogen bonds providing them high stability. These features are reminiscent of those stabilizing the structure of amyloid fibrils \[[@B53]\]. In addition, Thioflavin-T and Congo red, two dyes used for the diagnostic of amyloid structures also bind to IBs, confirming thus certain resemblance in the internal organization of both kinds of aggregates \[[@B27]\]. Also, even if we still lack structural information on thermal aggregates purified directly from bacteria under stress conditions, it has been shown that in vitroheat denaturation leads to the formation of thermal aggregates that display the β-sheet signature as analyzed by FTIR \[[@B58]\] and are also able to bind amyloid dyes \[[@B59]\]. Despite the fact that the different types of aggregates share similar characteristics, they are obviously not identical and exhibit a series of distinctive features. First, most amyloid fibrils are SDS-insoluble, whereas SDS can usually dissolve IBs. This observation is in agreement with the higher extent of β-sheet content of amyloids relative to that in IBs, in which the presence of some native or disordered structure can be still detected \[[@B27],[@B60]\]. As a result amyloids would display more and stronger intermolecular non-covalent interactions that would provide them with higher order and stability in front of denaturation, while sharing similar overall connectivity between polypeptide chains than this present in IBs. Also, the regulation of amyloid and bacterial aggregates formation in vivoappears to be somehow different. In this sense, it has been demonstrated that in yeast the formation of amyloids by the Sup 35 prion is highly dependent on the presence of the Hsp 104 chaperone \[[@B61]\]. In contrast, the role of the bacterial Hsp 104 homologue, ClpB, in the regulation of inclusion body formation in E. coli is more controversial, some studies indicating that, as in the case of Hsp 104, it binds preferentially to the hydrophobic surface of aggregated protein \[[@B62]\], while others suggesting only a moderate role in the process of aggregation, which is mainly controlled by the chaperones DnaK and GroEL \[[@B63]\]. Interestingly, the bacterial chaperone GroEL is able to modulate both in vitro\[[@B64]\] and in vivoin mammalian cells \[[@B65]\] the aggregation of proteins involved in amyloid pathologies, suggesting that in spite of the constrains imposed by the different cellular contexts some similitude may still exist between the mechanisms of bacterial and eukaryotic protein aggregation. Regardless of the existence of some structural or functional differences between the aggregates formed in bacteria and those in eukaryotic cells, in both cases there is an inherent tendency to kidnap misfolded protein in the interior of such supra-molecular structures. It is suggested that this is a mechanism evolved to reduce the potential toxicity of partially folded monomers or small oligomers, which by exposing large hydrophobic surfaces could interact inappropriately with a wide range of cellular components, hampering this way cell function \[[@B66]\]. In these sense, specific aggregation could be a conserved strategy playing a cellular protective role. Sequence modulates protein aggregation -------------------------------------- One of the major unanswered questions of protein aggregation is the specificity with which primary sequence determines both the aggregation propensity and the specific details of the aggregated structure. The hypothesis that the ability of proteins to form ordered aggregates is a general property of the polypeptide chain rather to be limited to a restricted set of proteins \[[@B2]\] seems reasonable, especially if the main driving force for aggregation is the formation of an inter-backbone hydrogen-bonded network leading to the above described β-sheets structures, since all polypeptides regardless of sequence share the polypeptide backbone. In this regard, IBs and amyloid formation abilities has not been associated a priori to particular protein sequences, being this fact, an additional obstacle to predict the yield of a given protein in a new production process or its cellular toxicity. However, in recent times it is coming clear that sequence modulates aggregation, giving a chance to control the unwanted protein deposition phenomena. A first indication that sequence controls deposition comes from the observation that not all regions of a polypeptide are equally important for determining the aggregation propensities. This way, we have proved recently that very short specific amino acid stretches can act as facilitators or inhibitors in the incorporation of globular proteins into amyloid fibrils \[[@B67]\]. These relevant regions are usually known as aggregation \"hot spots\". Aggregation-prone regions are blocked in the native state of globular proteins because their side chains are usually hidden in the interior of the protein hydrophobic core or already involved in the establishment of the network of native contacts that stabilizes a protein. This is the reason why globular proteins rarely aggregate from their native states. Destabilization usually results in an increased population of partially folded molecules and is well established as a trigging factor in disorders associated with the deposition of proteins that are globular in their normal functional states \[[@B68]\]. Accordingly, peptides and proteins involved in the most prevalent human neurodegenerative diseases are mostly unstructured within the cell \[[@B3]\]. In these disorders, protein deposition does not require the unfolding of a globular native conformation and occurs by direct self-assembly of the unstructured polypeptide chains, in which aggregation-prone, usually hydrophobic, regions are already exposed to solvent. The presence of aggregation \"hot spots\" have been already described in the peptides and proteins underlying Alzheimer\'s, Creutzfeldt-Jakob disease, or some systemic amyloidogenic disorders \[[@B69]-[@B71]\]. Independent of the native conformation and stability of the protein, the high level of expression during recombinant production results in a large number of polypeptides emerging from the ribosome in at least partially unfolded conformations which usually associate to form IBs. Even if not yet proved, it is thinkable that the presence of aggregation prone sequences in these conformers will influence at least partially the equilibrium between aggregated and folded protein during recombinant expression. Interestingly, it is observed that proteins assembled into amyloid in vitrousually render insoluble during recombinant protein expression. For example, this happens for proteins involved in disease such us Aβ42 amyloid peptide, β-2-microglobulin, mammalian prions and human islet amyloid polypeptide \[[@B72]-[@B75]\]. The study of the effects of mutations on the formation of amyloid fibrils and IBs also point to the role of sequence as an aggregation controller. Two types of mutations should be distinguished according to their ability to destabilize or not significantly the native state of the protein. As stated before, destabilizing mutations favour aggregation by originating an ensemble of partially unfolded conformations allowing this way the establishment of inter-molecular interactions. In addition, it has been shown that punctual mutations can also facilitate aggregation without affecting the native state stability when they promote the conversion of already unfolded or partially folded polypeptides into oligomeric forms that further aggregate to form insoluble species. In these cases, protein aggregation has been found to be tuned by mutations that change the polarity, the secondary structure propensity or the net charge of the polypeptide. In general, increases in hydrophobicity and β-sheet propensity result in increased aggregation whereas an increase in the overall net charge decreases this tendency \[[@B24],[@B76],[@B77]\]. There are a good number of protein systems in which it has been shown that point mutations may dramatically affect the amount of aggregate formation; these include the P22 tailspike protein, single-chain antibodies, interferon-γ, colicin A, Che Y, immunoglobulin domains, and interleukin-1β for IBs formation \[[@B43],[@B78]-[@B83]\] and SH3-domains, acylphosphatase, amylin, prion peptides, α-synuclein, amyloid-β-peptide and tau for amyloid formation \[[@B25],[@B67],[@B84]-[@B88]\]. Notably, mutant proteins with reduced in vitroamyloid propensity are expressed usually in E. colias more soluble proteins than the natural occurring ones \[[@B89]\], whereas providing a previously in vitrosoluble protein increased amyloid propensity results in accumulation as IBs during recombinant expression \[[@B90],[@B91]\]. Moreover, when amyloid proteins have been designed de novo, all proteins displaying amyloid properties in vitroaccumulated in vivoas bacterial IBs \[[@B92]\], but the rational introduction of point mutations that convert these aggregation-prone proteins into monomeric β-sheet forms allowed their expression in bacteria in soluble forms \[[@B93]\]. These observations strongly suggest that both aggregation phenomena are related and depend in last term on tendency to self-aggregate associated to individual protein sequences. This way, it appears that the study of bacterial models may contribute significantly in the future to the understanding of protein misfolding and aggregation, since they are fast, simple and biologically relevant experimental systems. Conversely, it is thinkable that the application of successful anti-depositional strategies derived from the numerous studies dealing with amyloid fibril formation to the less explored area of protein aggregation within the cell may provide clues to optimize biotechnological protein production. In this regard, simple sequence-based computational approaches have been developed very recently which permit to predict with reasonable accuracy the aggregation propensity of polypeptides \[[@B94]-[@B97]\]. In particular, TANGO a statistical mechanics algorithm based on the physico-chemical principles of β-sheet formation, extended by the assumption that the core regions of an aggregate are fully buried, accurately predicts the aggregation propensity of a data set of more than 200 different peptides \[[@B95],[@B96]\]. Without doubt, these new algorithms born in the sinus of the amyloid area are going to be very useful tools for the rational modification of polypeptides for biotechnological applications, opening the door to a fully automated, sequence-based design strategies to improve the solubility of proteins of industrial interest. Perspectives: Towards rational design of protein solubility ----------------------------------------------------------- There is an increasing need for the efficient production of genetically engineered proteins as a result of the success of the genome sequencing projects. From the different host that may be used to produce this large set of proteins, bacteria, mainly E. coli, still appears as the default option, particularly when the biological activity of the protein does not depends on post-translational modifications. E. coliis fast and inexpensive to culture, easy to handle and manipulate genetically and usually renders high levels of recombinant products. However, expression of recombinant proteins in E. colioften results in the accumulation of the protein product in inactive IBs in the cell. The recovery of bioactive proteins from IBs is a complex process. Still, IBs formation is such a frequent phenomena in protein production that a large number of in house and commercial protocols and solutions have been developed in order to obtain pure, active and soluble protein from IBs \[[@B17],[@B98]\]. Nevertheless, the purification of protein from IBs usually requires the optimization of refolding conditions for each individual target, the recovery yields are usually poor and one should be sure that the refolding procedure does not affect the integrity and activity of the recovered protein. In addition, purification of over-expressed soluble proteins is faster and cheaper than obtaining it in a pure form from IBs, especially at large scale. Overall, optimizing the levels of soluble protein is nowadays a more attractive strategy to increase pure and active protein yield than recovering highly expressed protein in aggregated form \[[@B99]\]. The observation that natural proteins are usually soluble in their biological environments may help to maximize soluble expression levels in recombinant approaches. This way, nature has provided proteins with a reasonable conformational stability in the native state, in which most of the hydrophobic residues, amide and carboxyl groups and aggregation-prone sequence stretches are buried or involved in intra-molecular interactions. This appears as a very successful strategy used to avoid aggregation, since few proteins are able to aggregate from its stable native conformation. Along with this, over-stabilized proteins of thermophilic organisms are usually expressed in soluble forms during recombinant protein production \[[@B100]-[@B102]\] and a positive correlation between thermostability and solubility has been recently reported \[[@B103]\]. In addition, the analysis of protein databases has shown that highly aggregating sequences are less frequent in proteins than innocuous amino acid combinations and that, if present; they are surrounded by amino acids that disrupt their aggregating capability \[[@B94]\]. These evidences support the suggestion that natural protein sequences have evolved in part to code for structural characteristics other than those included in the native fold, such as avoidance of aggregation. According to this, protein aggregation results from a failure of the natural protective strategies under special circumstances, such as recombinant protein expression. Using rational design to engineer target proteins in order to emulate and reinforce natural anti-aggregation mechanisms, taking advantage of the above mentioned computational methods to predict aggregation, appears thus as a reasonable approach to overwhelm protein deposition and optimizing the levels of soluble protein in biotechnological processes. Few, but successful experimental steps have been taken already in this direction. First, improving thermodynamic stability by rational mutation has been shown to render more soluble heterologous protein versions \[[@B104]\]. Second, it has been proven that decreasing the intrinsic propensity to aggregate of the partially unfolded state of an aggregation-prone protein by modulating the net polypeptide charge and introduction of electrostatic repulsions also results in increased solubility \[[@B105]\]. Finally, the analysis, identification and disruption by mutation of sequential \"hot spots\" of aggregation has allowed the recovering from the E. colisupernatant of previously aggregated polypeptides \[[@B67],[@B93],[@B106]\]. Conclusion ========== The raising interest to understand the mechanisms underlying protein aggregation in the cell has crystallized in a good number of recent relevant studies in an area whose biological significance is coming of central importance in biotechnology. The scenario emerging from these efforts is especially encouraging because one can foresee a future in which rational design of protein solubility based on natural laws will allow to tune aggregation, permitting to over-pass the main bottleneck in high throughput expression projects. Acknowledgements ================ The author is recipient of a \"Ramón y Cajal\" contract awarded by the MCYT-Spain and co-financed by the Universitat Autonoma de Barcelona (UAB) and founded by PNL2004-40 from UAB.	PubMed Central	2024-06-05T03:55:56.193659	2005-4-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087874/", "journal": "Microb Cell Fact. 2005 Apr 22; 4:11", "authors": [ { "first": "Salvador", "last": "Ventura" } ] }
PMC1087875	Review ====== Introduction ------------ Expression of recombinant proteins from a variety of host organisms is now a common practice. However, production of properly folded proteins with high yield and purity may not always be achieved. Issues such as folding, solubility, protein stability, transcription and translation efficiency, posttranslational processing, secretion, metabolic burden and other stress responses resulting from recombinant protein production, as well as protein purification, need to be addressed in order to obtain biologically active recombinant proteins with high purity and yield \[[@B1]\]. In this regard, genetically encoded fluorescent reporters provide ample new opportunities to better tackle these issues. Since the demonstration of the Aequorea victoriagreen fluorescent protein (GFP) as a versatile reporter \[[@B2]\], several additional GFP-like fluorescent proteins with various colors have been discovered and their genes cloned \[[@B3]\]. Synthetic fluorescent protein variants have also been developed, exhibiting traits distinct from their wild-type counterparts. The properties of selected fluorescent protein variants derived from the A. victoriaGFP and the Discosomared fluorescent protein (DsRed) \[[@B4]\] are summarized in Table [1](#T1){ref-type="table"}. Fluorescence spectra of enhanced GFP variants along with DsRed are shown in Figure [1](#F1){ref-type="fig"}\[[@B5]\], and fluorescence of purified protein variants derived from DsRed \[[@B6]\] are shown in Figure [2](#F2){ref-type="fig"}. Readers are referred to the work of Labas et al. \[[@B7]\] for information of additional fluorescent proteins. These GFP-like proteins each has its own unique properties, while sharing common structural, biochemical and photophysical characteristics \[[@B3]\]. GFP-like proteins are relatively small (25--30 kDa) and their fluorescence mechanism is self-contained, requiring no cofactors. These unique properties make GFP-like proteins very attractive tools in non-invasive biological monitoring applications. As a tool to improve recombinant protein production, fluorescent proteins can be used to monitor the protein product or the cellular processes relevant to recombinant protein production. Monitoring protein production, secretion, and culture growth ------------------------------------------------------------ Fluorescent proteins are commonly used as a reporter for a protein of interest, normally by tagging the fluorescent protein reporter to the protein of interest viagenetic fusion. Functional fusion of AequoreaGFP to a broad range of protein partners at either N- or C- terminus has been reported, and a direct quantitative correlation between the GFP fluorescence intensity and the titer or even the functional activity of the fusion partner can often be established \[[@B8],[@B9]\]. To minimize potential interference by the GFP tag on its fusion partner, it is desirable and sometimes necessary to incorporate a peptide linker to allow sufficient spatial separation of the two protein moieties to assure fusion protein stability and functionality. Flexible linkers lacking large bulky hydrophobic residues (e.g.GSAGSAAGSGEF \[[@B10]\]) are commonly used, while hydrophilic helix-forming linker peptides have been reported to be superior to flexible linkers in some cases \[[@B11]\]. To allow removal of the GFP tag, an enzymatic cleavage site (e.g.enterokinase or Factor Xa cleavage sites) can be engineered into the linker. It is preferred to splice the GFP/linker to the N-terminus of the target protein, provided such fusion does not impair the target protein function and stability. With the majority of the enzymes commonly used for tag removal, this fusion orientation enables elimination of the tag without leaving extraneous amino acid residues on the target protein after cleavage. Alternatively, chemical cleavage based on cyanogen bromide, formic acid, or hydroxylamine may be considered, provided the target proteins are not susceptible to cutting by these chemical agents. Further information of tag removal can be found in a comprehensive review by Hearn and Acosta \[[@B12]\]. In addition to tandem fusion, insertional fusion (i.e.by inserting the protein of interest into GFP or vise versa) may also be feasible \[[@B13]\]. Recombinant protein production can be monitored non-invasively, in situ, and almost in real time, by monitoring culture fluorescence using on-line optical sensors \[[@B14],[@B15]\]. This information is useful in determining the optimal product harvest time to avoid product degradation \[[@B16]\] and to devise process control strategies to optimize culture/operating conditions to improve recombinant protein production \[[@B8],[@B17]\]. GFP has also been used to monitor recombinant virus titers in cell cultures \[[@B18]\], and cell density in microbial \[[@B15]\], animal \[[@B19]\], and plant cell cultures \[[@B8]\]; the cell growth information can be used, in turn, to optimize the culture process for improved recombinant protein production (e.g.by optimizing the feeding profiles of the limiting nutrient or the promoter inducer, or by determining the optimal product harvest time \[[@B8],[@B17]\]). Additionally, GFP-fusion coupled with flow cytometric analysis is useful for profiling recombinant protein expression among different cell subpopulations, and selection of high-producing cells \[[@B9]\]. GFP-fusion can also be used for monitoring protein secretion and other subcellular protein localization and trafficking events. In the event direct GFP fusion hampers protein secretion, alternative protein fusion strategies may be sought. One plausible approach is to express GFP and a protein of interest as a cleavable chimeric polyprotein. By targeting the polyprotein to the secretory pathway, the target protein and the GFP tag may become separated and secreted as individual proteins. Feasibility of such approach has been demonstrated in fungi and plant cells for the successful in-vivocleavage of an glucoamylase-interleukin-6 fusion protein \[[@B20]\] and a fusion antimicrobial polyprotein \[[@B21]\], respectively. Cellular processing in plant cells of a polyprotein that consists of DsRed-GFP fusion linked by a Kex2 cleavage sequence \[[@B22]\] is currently being investigated in the author\'s laboratory. In another possible approach, one may express the target gene and GFP in a dicistronic vector by incorporating an internal ribosome entry site (IRES) \[[@B23]\]. One additional possibility would be to express target protein and GFP as separate genes, but from the identical promoter. If a constant ratio between GFP fluorescence and target protein concentration could be established, the independent GFP could be used for fluorescent monitoring. Fluorescent protein reporters can also be used to probe protein-protein interactions that regulate the protein secretion process \[[@B24]\], potentially leading to development of molecular strategies that improve recombinant protein secretion. Monitoring protein purification ------------------------------- The fact that GFP fluorescence is readily detectable makes it a very attractive tool for optimizing purification of recombinant proteins. Poppenborg et al. \[[@B25]\] optimized immobilized metal affinity separation of a histidine-rich protein tagged with GFP by tracking the fluorescence of the fusion protein. Since GFP is a highly hydrophobic protein, recovery of GFP-fusion proteins can be facilitated by using hydrophobic interaction chromatography (HIC) \[[@B26]\]. GFP has also been engineered to allow affinity purification. Paramban et al \[[@B27]\] developed a chimeric GFP tag having an internal hexa-histidine sequence. Such a GFP tag allows efficient purification of GFP-fusion proteins based on immobilized metal affinity separation, as well as maximum flexibility for protein or peptide fusions since both termini of the GFP are available. Monitoring protein folding -------------------------- Correct folding is one of the key challenges in recombinant protein production using simple hosts like Escherichia coli. GFP has been proposed as a folding reporter. By fusing GFP to a panel of proteins, Waldo et al\[[@B10]\] demonstrated that display of GFP fluorescence in the E. colicolonies expressing the fusion proteins indicated proper folding of GFP\'s fusion partner. De Marco \[[@B28]\] however cautioned that observation of GFP fluorescence from the fusion protein may not guarantee the fusion partner have reached its native structure. Recently, Waldo and coworkers \[[@B29]\] reported a novel split-GFP system that consists of a small (GFP β strand 11; GFP 11) and a complementary large fragment (GFP β strand 1--10; GFP 1--10). Neither fragment by itself displays fluorescence, but GFP fluorescence emerges upon self-association of the two complementary fragments. These researchers demonstrated that by tagging target proteins with the GFP-11 tag, the solubility of the target proteins can be checked by mixing with the GFP 1--10 fragment in vitroor by co-expressing the GFP 1--10 fragment in vivo. Appearance of GFP fluorescence suggests the target protein is soluble. This split-GFP may also be useful for detecting protein-protein interaction in vivo, similar to the luciferase fragment complementation systems \[[@B30],[@B31]\]. Monitoring cellular responses ----------------------------- In addition to direct monitoring of protein expression, processing, and secretion, fluorescent proteins can also be used to monitor cellular events that are directly or indirectly related to recombinant protein production. For instance, FRET (fluorescence resonance energy transfer)-based GFP sensors have been developed to detect proteolysis in vivo. FRET-based GFP sensors have also been developed for measuring intracellular concentration of nitric oxide, calcium, cAMP, zinc, activation of G protein-coupled receptor (GPCR), and PKA-mediated phosphorylation \[[@B9]\]. Redox sensitive GFP variants have been developed for monitoring the cellular oxidative states, which strongly affect protein folding. In addition, intracellular pH can be measured using GFP. Metabolic stresses induced by recombinant protein over-expression may also be monitored in vivousing GFP linked to stress-induced promoters as a reporter. Optical sensing of culture GFP fluorescence ------------------------------------------- The presence of cell aggregates, debris, and other light absorbing/scattering compounds in the culture medium contributes to the \"inner filter effect\" (IFE) that could distort the optical measurement of culture GFP fluorescence. Real-time compensation of IFE in monitoring cultures expressing GFP-fusion proteins typically involves establishing a mathematical model to link the IFE to cell density, and to use an on-line laser turbidity sensor to report the biomass density needed in the calculation of the IFE \[[@B32]\]. An obvious drawback of such an approach is the requirement of a turbidity sensor in addition to the optical sensor for monitoring culture fluorescence. We recently developed a technique that allows real-time compensation of IFE during on-line monitoring of culture GFP fluorescence, without the need for an additional biomass sensor \[[@B33]\]. This was achieved by developing a model-based state observer, using the extended Kalman filter (EKF) and on-line measurement of GFP culture fluorescence using an optical light-rod sensor. Applications involving multiple fluorescent protein variants ------------------------------------------------------------ Given the many facets of its applications, multiple fluorescent protein reporters could potentially be used in parallel for multi-color in vivosensing. For instance, GFP may be used to tag the recombinant protein product, while a red fluorescent protein could be linked to a stress-responsive promoter such as the heat-shock promoter groELto monitor stress induced by recombinant protein over-expression in E. coli\[[@B34]\]. Having a large repertoire of fluorescent proteins with diverse spectral properties is also necessary for multiplex FRET-based sensing applications. Through both structure-based modification and evolutionary methods for protein engineering, several robust variants of the AequoreaGFP have been created with blue, cyan, and yellow colors \[[@B35]\]. In a recent paper by Shaner et al \[[@B6]\], development of a panel of novel monomeric red, orange and yellow fluorescent proteins derived from the DiscosomaDsRed was reported. These new variants also show additional traits that are useful for monitoring recombinant protein production. Monomeric fluorescent protein variants -------------------------------------- Wild type DsRed is an obligate tetramer. When fused to a protein of interest, the fusion protein often forms aggregates, hampering normal localization, trafficking and protein-protein interactions of the protein of interest. A monomeric DsRed variant, called mRFP1, was developed by disrupting each subunit interface viainsertion of arginines, and then using directed evolution to accelerate chromophore maturation and to restore fluorescence, which takes 33 substitutions \[[@B36]\]. mRFP1 was further improved by subjecting to additional rounds of directed evolution \[[@B6]\]. The resulting eight variants display corresponding emission peaks ranging from 537 to 610 nm. Some of these new variants also show better tolerance to N- and C- terminal fusions, higher extinction coefficients, quantum yields, and photostability, though no single variant has acquired all the desirable traits \[[@B6]\]. Since GFP is known to have high tolerance to either N- or C- terminal fusions, and DsRed and GFP share similar structures, Shaner et al \[[@B6]\] engineered GFP-type termini into mRFP1, rendering improved tolerance to protein fusion in the new variant. Among the new monomeric variants, mCherry (excitation at 587 nm, emission at 610 nm) is the most red-shifted, and has the best photostability, fastest maturation (15 min) and excellent pH resistance and tolerance to N-terminal fusions. The mOrange variant (excitation at 548 nm, emission at 562 nm) has high extinction coefficient and quantum yield, and is shown to be a superior FRET acceptor for GFP variants. Conclusion ========== The repertoire of fluorescent protein variants has continued to expand, and is now covering almost the entire color spectrum. With the advances in directed evolution techniques \[[@B37]\], each of these new proteins is likely to be further improved. The availability of improved multicolor fluorescent protein reporters will undoubtedly lead to development of innovative techniques that enable more effective multiplex cellular sensing, and allow more efficient on-line optimization of recombinant protein production. Acknowledgements ================ The author is grateful to the funding supports from the United States National Science Foundation (BES97-12916 and BES01-26191) and the United States Department of Agriculture (01-34135-11295 and 58-3148-9-080). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Fluorescence spectra of EGFP variants and DsRed (reproduced with the permission of \[5\]). ::: ![](1475-2859-4-12-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Fluorescence of DsRed variants developed by Shaner et al \[6\] (from left to right, mHoneydew, mBanana, mOrange, tdTomato, mTangerine, mStrawberry, mCherry; reproduced with the permission of \[6\]; refer to the same reference for details). ::: ![](1475-2859-4-12-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Properties of selected fluorescent proteins ::: Fluorescent Protein Excitation Peak (nm) Emission Peak (nm) Extinction Coefficient (M^-1^cm^-1^) Fluorescence Quantum Yield Reference ------------------------- -------------------------- ------------------------ ------------------------------------------ -------------------------------- --------------- EBFP 383 445 31,000 0.25 \[5\] ECFP 434 477 26,000 0.40 \[5\] Cerulean CFP 433 475 43,000 0.62 \[38\] EGFP 489 508 55,000 0.60 \[5\] EYFP 514 527 84,000 0.61 \[5\] Venus YFP 515 528 92,200 0.57 \[39\] Citrine YFP 516 529 77,000 0.76 \[40\] DsRed 558 583 75,000 0.79 \[6\] mRFP1 584 607 50,000 0.25 \[6\] mHoneydew 487/504 537/562 17,000 0.12 \[6\] mBanana 540 553 6,000 0.70 \[6\] mOrange 548 562 71,000 0.69 \[6\] mTangerine 568 585 38,000 0.30 \[6\] mStrawberry 574 596 90,000 0.29 \[6\] mCherry 587 610 72,000 0.22 \[6\] :::	PubMed Central	2024-06-05T03:55:56.197147	2005-4-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087875/", "journal": "Microb Cell Fact. 2005 Apr 25; 4:12", "authors": [ { "first": "Wei Wen", "last": "Su" } ] }
PMC1087876	Introduction ============ Pancreatic cancer is one of the most aggressive human malignancies with an overall five-year survival rate of less then 5% \[[@B1]\]. Although the reasons for the aggressive growth behavior of pancreatic cancer are not completely understood, recent molecular biological studies have revealed several factors that are involved in the pathogenesis of pancreatic cancer. These include genetic changes, such as k-ras, p53, p16, and Smad4 mutations \[[@B2]\], as well as epigenetic alterations, such as overexpression of a number of growth factors and their receptors \[[@B3],[@B4]\]. Membrane-associated heparin-sulfate proteoglycans (HSPGs) are thought to play an important role in many aspects of cellular physiology including growth factor signaling. HSPGs are required for the optimal activity of heparin-binding growth factors, such as for example fibroblast growth factors (FGFs) \[[@B5],[@B6]\]. One member of the HSPG family, glypican-1 is over-expressed in pancreatic cancer and influences heparin binding growth factor signaling in this disease \[[@B7],[@B8]\]. The heparan-sulfate (HS) chains of HSPGs seem to interact with the ligands (e.g. FGF-2) and high-affinity FGF-receptors, to increase ligand-receptor binding and signaling \[[@B9]\]. The enzyme Hsulf-1 is a recently identified human sulfatase, which exhibits arylsulfatase activity \[[@B10]\]. Hsulf-1 expression is down-regulated in ovarian cancers, and lost in a proportion of liver cancers \[[@B11],[@B12]\]. Absence or low levels of Hsulf-1 in hepatocellular, ovarian, and head and neck squamous cell carcinoma cell lines were associated with up-regulation of heparin-binding growth factor signaling \[[@B11]-[@B13]\]. Since HSPGs such as glypican-1 play an important role in pancreatic cancer and since Hsulf-1 can influence the sulfation state and the biological function of HSPGs, the expression and functional role of Hsulf-1 was analyzed in pancreatic cancer. Results ======= Hsulf-1 mRNA expression in pancreatic tissues --------------------------------------------- Utilizing DNA arrays the expression of nine sulfatase family members in pancreatic cancer, pancreatic cancer metastasis, chronic pancreatitis and the normal pancreas was screened. This analysis revealed that Hsulf-1 was significantly over-expressed in pancreatic cancer and chronic pancreatitis compared to normal pancreatic tissues. Thus, Hsulf-1 mRNA expression levels were increased 9.1-fold in primary pancreatic cancer, 4.5-fold in pancreatic cancer metastasis, and 3.4-fold in CP tissues compared to normal pancreatic tissues. In contrast, there were only minor or no changes in the mRNA levels of the other members of the sulfatase family (Table [1](#T1){ref-type="table"}). In order to better quantify Hsulf-1 expression quantitative RT-PCR was carried out in normal pancreatic tissue samples (n = 19), chronic pancreatitis (n = 22) and pancreatic cancer tissue samples (n = 31). The samples from normal tissues had a mean (+/- SEM) number of Hsulf-1 transcripts/μl of 114 ± 23, while Hsulf-1 mRNA levels increased in both chronic pancreatitis and pancreatic cancer, with mean (+/- SEM) transcripts levels of 2054 ± 911 in chronic pancreatitis and 2566 ± 420 in pancreatic cancer. 10 of 22 (45%) CP and 22 of 31 (71%) pancreatic cancer tissue samples displayed higher copy numbers of Hsulf-1 mRNA than the highest Hsulf-1 mRNA level observed in normal pancreatic tissue samples (Figure [1](#F1){ref-type="fig"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Expression of different sulfatases in normal pancreas (No), chronic pancreatitis (CP), primary pancreatic cancer (CA) and pancreatic cancer metastasis (Mx) tissues as determined by DNA array analysis. ::: Fold change -------------------------------------------------- ----------------- -------------- -------------- Ca vs No Mx vs No CP vs No hSulf-1 9.1 4.5 3.4 Arylsulfatase A 0.6 0.8 0.9 Arylsulfatase B 1.8 1.6 1.8 Arylsulfatase C 1.6 1.8 1.6 Arylsulfatase D 0.5 0.2 0.8 Arylsulfatase E 0.5 0.6 0.7 galactosamine (N-acetyl)-6-sulfate sulfatase 0.8 0.8 0.9 glucosamine (N-acetyl)-6-sulfatase 2.5 2.2 1.6 iduronate 2-sulfatase 1.0 1.2 1.0 ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Expression of Hsulf-1 mRNA in pancreatic tissues and cell linesQuantification of Hsuf-1 mRNA levels in the normal pancreas, chronic pancreatitis (CP), pancreatic cancer tissues (PC), by real time QRT-PCR as described in the Material and Methodssection. Values are normalized to housekeeping genes (cyclophilin B and HRPT), and presented as single values and mean (horizontal line). ::: ![](1476-4598-4-14-1) ::: Localization of Hsulf-1 in pancreatic tissues --------------------------------------------- To identify the local expression pattern of Hsulf-1 in the normal pancreas, chronic pancreatitis and pancreatic cancer tissues, in situhybridization analysis was carried out. Weak Hsulf-1 mRNA expression was observed in the acini of normal and chronic pancreatitis tissues. Hsulf-1 mRNA was localized in the smooth muscle cells and the endothelium of blood vessels, as well as in fibroblasts of the connective tissue (Figure [2 A--C](#F2){ref-type="fig"}). In addition, Hsulf-1 mRNA expression was present in tubular complexes of chronic pancreatitis tissues (Figure [2 D](#F2){ref-type="fig"}). In pancreatic cancer tissues, Hsulf-1 mRNA was mainly expressed in tubular complexes (Figure [2 E](#F2){ref-type="fig"}), in the cancer cells themselves (Figure [2 F, G](#F2){ref-type="fig"}) as well as in fibroblasts of the connective tissue (Figure [2 F](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Expression and localization of Hsulf-1 mRNA in pancreatic tissuesIn situ hybridization was performed as described in Material and Methodssection. Hsulf-1 localization in: normal pancreatic acini (A); blood vessels (B), nerves (C), and tubular complexes (D) of CP; tubular complexes (E), cancer cells (F) of pancreatic cancer. Note the staining in the control section probed with the antisense riboprobe (G), compared with absent staining in the sense-probed section (H). ::: ![](1476-4598-4-14-2) ::: Hsulf-1 expression in pancreatic cancer cell lines -------------------------------------------------- QRT-PCR analysis was carried out in 8 cultured pancreatic cancer cell lines. This analysis revealed relatively high expression of Hsulf-1 mRNA in Su-8686 and moderate expression in T3M4, Colo-357, and BxPc-3 pancreatic cancer cell lines. In the other cell lines, Hsulf-1 expression was below the detection level (Figure [3 A](#F3){ref-type="fig"}). Panc-1 pancreatic cancer cells were selected for Hsulf-1 transfection, since Hsulf-1 expression was below the level of detection in this cell line. To confirm successful transfection of Panc-1 cells with the full-length Hsulf-1 construct, Northern blot analysis was carried out using a Hsulf-1 antisense riboprobe. A total of number 36 clones were screened, of which 10 clones clearly expressed Hsulf-1 mRNA. Two Hsulf-1 positive clones (sulf-26 and sulf-38) were selected for use in further experiments and compared to empty vector-transfected (EV) and non transfected wild type (WT) Panc-1 cells (Figure [3 B](#F3){ref-type="fig"}). To confirm the expression of Hsulf-1 in the positive clones on the protein level, immunoblotting was performed. Since the Hsulf-1 expression plasmid contained a c-myc tag \[[@B10]\], it was possible to detect the expression of the Hsulf-1-myc fusion protein in the selected clones by immunoblot analysis with an anti-c-myc antibody (Figure [3 C](#F3){ref-type="fig"}). To determine the activity of the expressed sulfatase, cellular extracts prepared from both control (wild type and empty vector) and transfected clones (sulf-26, sulf-38) were analyzed. 4-Methylumbelliferyl-sulfate, which represents a substrate for a variety of sulfatases, including cellular steroid sulfatases, was used as the substrate for sulfatase activity. Upon transfection, sulfatase activity was most prominently increased in clone sulf-38 (Figure [3 D](#F3){ref-type="fig"}). However, since this assay could not differentiate between different sulfatases, high sulfatase activity was also observed in the control cells. Therefore, to further confirm successful transfection and increased Hsulf-1 activity, immunofluorescence with the 10E4 anti-HSPG monoclonal antibody, which recognizes N-sulfated glucosamine-containing HSPGs, was carried out. Prominent staining of the cell membrane was observed in both wild type (Figure [4A](#F4){ref-type="fig"}) and empty vector Panc-1 cells (Figure 4 C). In contrast, markedly diminished staining of the cell membrane was observed in the two Hsulf-1 expressing clones (Figure 4 B, D), indicating that Hsulf-1 desulfates HSPGs at the cell surface. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Expression of Hsulf-1 in pancreatic cancer cell lines(A)Quantification of Hsulf-1 mRNA levels in pancreatic cancer cell lines by real time QRT-PCR as described in the Materials and Methodssection. Values are normalized to housekeeping genes (cyclopilin B and HRPT), and presented as mean ± SD. (B/C)Panc-1 cells were stable transfected with a Hsulf-1 sense expression plasmid as described in the Material and Methodssection. (B)Hsulf-1 sense RNA expression in Panc-1 cells was verified by Northern blot analysis using a radiolabeled Hsulf-1 antisense riboprobe. A sample Northern blot of 2 controls and 2 transfected clones is shown. (C)Expression of c-myc tagged Hsulf-1 (arrow) by immunoblot analysis as described in the Materials and Methodssection. Equal loading of the protein samples was confirmed using anti-γ-tubulin antibodies. (D)Sulfatase activity was measured as described in Material and Methodssection in control and positive transfected clones. Data are expressed as relative fluorescence and presented as mean ± SD. ::: ![](1476-4598-4-14-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Hsulf-1 decreases the sulfation of cell surface HSPGsImmunofluorescence was performed as described in Material and Methodssection with a specific 10E4 anti-HSPG monoclonal antibody, which recognizes native heparan-sulfate containing the N-sulfated glucosamine moiety. WT (A), EV (C), Sulf-26 (B), Sulf-38 (D). ::: ![](1476-4598-4-14-4) ::: Functional consequences of Hsulf-1 expression in Panc-1 pancreatic cancer cells ------------------------------------------------------------------------------- Next, the effects of Hsulf-1 expression on the growth of Panc-1 cells were assessed. Analysis of basal growth revealed that the 2 control clones displayed an average exponential doubling time of 50.3 ± 3.2 hours, which was significantly shorter compared to the 2 Hsulf-1 transfected clones (68.2 ± 4.3 hours) (Figure [5 A](#F5){ref-type="fig"}). To determine anchorage-independent growth rates, soft agar assays were carried out. The 2 control clones showed an average colony number of 223 as compared to 31 colonies for the Hsulf-1 transfected cells (Figure [5 B](#F5){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Hsulf-1 expression decreases Panc-1 pancreatic cancer cell growth(A) Basal cell growth as determined by the MTT assay. Data are expressed as mean ± SD of three independent experiments. Data are presented for control cells (WT, EV), as well as for Hsulf-1 transfected clones. (Sulf-26, Sulf-38). (B) Anchorage-independent cell growth for individual clones was measured by the soft agar assay as described in the Material and Methodssection. Data are presented for controls and positive clones as indicated. Data are presented as mean ± SEM obtained from three independent experiments. ::: ![](1476-4598-4-14-5) ::: Hsulf-1 decreases FGF-2 mediated cell proliferation and signaling in Panc-1 pancreatic cancer cells --------------------------------------------------------------------------------------------------- It has been shown previously that a variety of growth factors such as FGF-2, EGF, HB-EGF, and IGF-1 are over expressed in pancreatic cancer and that they have the potential to act as mitogens for pancreatic cancer cell lines \[[@B3],[@B4]\]. Therefore, we further investigated whether over-expression of Hsulf-1 could modulate the function of these growth factors in pancreatic cancer cells. Two Hsulf-1 transfected clones and two controls (wild type and empty vector) were selected to perform growth assays in the presence or absence of different doses of the indicated growth factors. Hsulf-1 expression significantly attenuated FGF-2 (50 ng/ml) induced cell growth by around 50%, from +28.0 ± 3.8% in controls to +14.4 ± 1.0 % in Hsulf-1 clones (Figure [6](#F6){ref-type="fig"}). In contrast, there was no difference in the response towards IGF-1, EGF or HB-EGF in the control versus Hsulf-1 expressing cells. Since Hsulf-1 expression reduced FGF-2 but not EGF or HB-EGF induced cell proliferation, next we sought to investigate whether Hsulf-1 expression would influence FGF-2 and EGF/HB-EGF downstream signaling. EGFR phosphorylation was not changed in response to EGF or HB-EGF in Hsulf-1 expressing clones compared to controls (Figure [7 A](#F7){ref-type="fig"}). In addition, there was also no difference of EGF and HB-EGF induced MAPK phosphorylation between control cells and positive clones (Figure [7 A](#F7){ref-type="fig"}). In contrast, control cells showed increased MAPK phosphorylation after FGF-2 stimulation, while this FGF-2 induced phosphorylation was markedly attenuated in Hsulf-1 transfected cells (Figure [7 B](#F7){ref-type="fig"}). Next, the basal and FGF-2 induced invasion capacity of tumor cells was analyzed. This analysis revealed a significant reduction in the invasiveness of FGF-2 exposed Hsulf-1 expressing cells compared to Hsulf-1 negative clones. As demonstrated in Figure [7C](#F7){ref-type="fig"}, FGF-2 (10 μg/ml) significantly stimulated the invasion of the control cells by +83.4 ± 24.2% after 24 h of incubation. In contrast, the invasion ability of Hsulf-1 positive cells was significantly less stimulated (+27.4 ± 35.5%) by exposure to FGF-2 (Figure [7 C](#F7){ref-type="fig"}). ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Effects of Hsulf-1 expression on growth factor induced proliferation. Clones were cultured in 1% FBS medium and incubated with increasing concentrations of the indicated growth factors for 72 h. Cell growth was measured by the MTT assay. Percent growth stimulation was determined by comparison with control cell growth. Values shown are the mean ± SEM obtained from three independent experiments. ::: ![](1476-4598-4-14-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### (A-B). Effects of Hsulf-1 on EGF/HB-EGF and FGF-2 induced receptor phosphorylation and MAPK phosphorylation.Panc-1 pancreatic cancer cells were cultured in 1% FBS medium overnight and then incubated with 10 μg/ml of the indicated growth factors for 10 min. Phosphorylation of MAP kinase and receptor phosphorylation (EGFR) was determined by immunoblotting with antibodies specific for phospho-p44/42 MAPK and phospho-EGFR. Equal loading was determined by re-blotting the membranes with an γ-tubulin antibody. The figure is representative of three independent experiments. (C) Hsulf-1 attenuates FGF-2 stimulated invasion.An in vitrocell invasion assay was performed using 8 μM filters coated with Matrigel as described in the Material and Methodssection. Panc-1 cells (1.25 × 10^5^) were seeded onto the filters in 1% serum overnight, and then treated as indicated for 24 h. The values shown are the mean ± SEM obtained from three independent experiments. ::: ![](1476-4598-4-14-7) ::: Effects of Hsulf-1 expression on chemosensitivity ------------------------------------------------- Pancreatic cancers exhibit variable degrees of chemotherapy resistance. To determine whether Hsulf-1 expression might influence the sensitivity of Panc-1 cells to chemotherapeutic agents, cells were treated with gemcitabine, 5-FU, or actinomycin-D and the GI50 concentration was calculated. The GI50 concentration of gemcitabine in the Panc-1 WT control cells was 3.9 nM, and in EV Panc-1 cells approximately 50 nM. Interestingly Hsulf-1 expressing cells exhibited GI50 values of more than 100 nM (Figure [8](#F8){ref-type="fig"}). In contrast no significant changes were observed in the sensitivity of Hsulf-1 positive clones and control cells towards 5-FU and actinomycin-D. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Effects of Hsulf-1 on the sensitivity towards chemotherapeutic agents. Panc-1 cells were cultured in complete medium and incubated in the absence (control) or presence of increasing concentrations of the indicated drugs for 48 h. Cell growth was measured by the MTT assay. Percent growth inhibition was determined by comparison with control cell growth. Values shown are the mean ± SEM obtained from three independent experiments. ::: ![](1476-4598-4-14-8) ::: Discussion ========== Sulfatases are a family of enzymes that catalyse the hydrolysis of sulfate ester bonds from a wide variety of compounds. They are classified into arylsulfatases and nonarysulfatases according to their ability to hydrolyse the sulfate ester bonds of aromatic compounds such as p-nitrocatechol sulfate and 4-methylumbelliferyl sulfate \[[@B14]\]. Hsulf-1 is a newly identified member of the sulfatase family, which exhibits arysulfatase activity and removes sulfate from the C-6 position of glucosamine within the specific sub regions of intact heparin \[[@B10]\]. In the present study a significant up-regulation of Hsulf-1 in primary pancreatic cancer, pancreatic metastasis and CP compared to normal pancreatic tissues was demonstrated, whereas there was no significant difference in the expression of other members of the sulfatase family in these tissues. This indicates that Hsulf-1 might play a specific role in the pathogenesis and evolution of CP and pancreatic cancer. Normal pancreatic tissues are composed mainly of a homogenous population of acinar cells (and a low percentage of ductal and islet cells), whereas both CP and pancreatic cancer tissues contain a variable amount of desmoplastic areas, inflammatory cells, degenerating acini, tubular complexes (and cancer cells). Thus, the observed wide range of expression of Hsulf-1 mRNA in both CP and pancreatic cancer tissues is most likely due to the different individual composition of these tissues. To confirm this hypothesis, in situhybridization was utilized to localize Hsulf-1 mRNA expression in normal, CP and pancreatic cancer tissues. This analysis demonstrated that Hsulf-1 mRNA expression was weakly present in normal acinar cells, and at high levels in the endothelium and smooth muscle cells of blood vessels, as well as in fibroblasts and tubular complexes in CP tissues and additionally in the malignant cells in pancreatic cancer tissues. The observed increased levels of Hsulf-1 in pancreatic cancer tissues seem to be in contrast to the down-regulation of Hsulf-1 in HCC and ovarian tumors \[[@B11],[@B12]\]. However, while in ovarian cancer markedly diminished levels were observed in approximately 75% of the cases \[[@B11]\], the percentage was much smaller in HCCs (30%) \[[@B12]\], suggesting that reduced Hsulf-1 expression is not universally observed in all tumor types. It has been hypothesized that enhanced expression of Hsulf-1 is related to c-myc amplification in HCCs \[[@B12]\]. It could be speculated that also in pancreatic cancer high Hsulf-1 levels are related to c-myc amplification \[[@B15]\], at least in a subset of tumors. Another interesting aspect is the generally low Hsulf-1 expression level in cultured cancer cell lines. Thus, Hsulf-1 expression is absent in 71% of ovarian cancer cell lines \[[@B11]\], in 82% of HCC cell lines \[[@B12]\], and in 50% of pancreatic cancer cell lines (present study). To evaluate the functional importance of Hsulf-1 in pancreatic cancer cells, Panc-1 cells, which do not express Hsulf-1 at detectable levels, were stably transfected with a Hsulf-1 expression vector. Over-expression of Hsulf-1 in Panc-1 cells resulted in reduced anchorage-dependent and -independent cell growth, suggesting an important growth regulatory role of this gene in pancreatic cancer. These tumors are characterized by enhanced expression of a variety of growth factors and their receptors, which have the capacity to influence different cellular functions, such as cell proliferation, migration and angiogenesis \[[@B3],[@B4]\]. Some of these growth factors are heparin-binding growth factors, such as FGFs, VEGF and HB-EGF. We hypothesized that Hsulf-1 expression would attenuate the effects of these growth factors by desulfation of HSPGs resulting in a growth disadvantage as suggested for other tumors \[[@B11]-[@B13]\]. FGF-2 stimulated cell proliferation was attenuated by the expression of Hsulf-1. Nonetheless, FGF-2 still induced growth in Hsulf-1 expressing cells, but to a lesser extent compared with control cells. It is conceivable that the HSPG/FGF receptor complex can facilitate FGF-2 signaling, but may not be strictly required for binding of FGF-2 to its receptor; it only increases the affinity of the FGF-2/FGF receptor interaction to a certain degree. Hsulf-1 expression in Panc-1 cells also partially blocked FGF-2 induced MAPK phosphorylation and invasion, further supporting the hypothesis that Hsulf-1 interferes with FGF-2 signaling in pancreatic cancer cells. In contrast, no difference between Hsulf-1 expressing and control cells was observed upon stimulation with HB-EGF -- another heparin-binding growth factor- as well as EGF and IGF-1 suggesting that these growth factors and their receptors do not require sulfated HSPGs for effective signaling. The observation that Hsulf-1 expression does not interfere with HB-EGF signaling in pancreatic cancer cells is in contrast to recent studies in ovarian cancer cells \[[@B11]\], suggesting cell type specific differences. Previously, it has been shown that Hsulf-1 expression enhances cisplatin-induced apoptosis in HCC cell lines \[[@B12]\]. In the present study, we did not observe increased sensitivity towards chemotherapeutic agents in Hsulf-1 expressing versus control cells. In contrast, Hsulf-1 expressing Panc-1 cells were more resistant to gemcitabine than the control cells, thereby suggesting that Hsulf-1 over-expression might confer increased chemoresistance to pancreatic cancer cells and thus provide them with a growth advantage. However, the reason behind this effect is currently not known and requires further analysis. In conclusion, Hsulf-1 is up-regulated in pancreatic cancer and chronic pancreatitis compared to normal pancreatic tissues, mainly due to over-expression in the desmoplastic and cancerous tissue elements. Expression of Hsulf-1 in Panc-1 cells negatively influences growth and invasion by attenuating FGF-2 signaling, suggesting that Hsulf-1 plays a specific role in the pathogenesis of pancreatic cancer. Further experimental approaches, especially in vivostudies, will help to assess in more detail the role of this enzyme in human pancreatic cancer. Materials and methods ===================== Materials --------- DMEM, trypsin-EDTA, and penicillin-streptomycin were purchased from Invitrogen (Mannheim, Germany); FBS from PAN Biotech (Aidenbach, Germany); Gene screen hybridization transfer membranes from PerkinElmer Life Science (Boston, MA, USA); ^32^P CTP from Amersham Pharmacia Biotech (Freiburg, Germany); Lipofectamine reagent™ and TRIzol Reagent from Invitrogen (Karlsruhe, Germany); RiboMAX™ Large Scale RNA production system, antibiotic G-418 sulfate from Promega (Mannheim, Gemany); Gemcitabine-Hydrochloride from Lilly (Eli Lilly and Company Limited, Hampshire, UK); Recombinant human FGF-2, Recombinant human IGF-1, Recombinant human HB-EGF from R&D systems (Wiesbaden-Nordenstadt, Germany); p-EGFR (Tyr 1173) antibody from Santa Cruz Biotechnology (Santa Cruz, CA, USA), p-P44/42 MAPK (Thr202/Tyr204) antibodies from Cell Signaling Technology (Frankfurt, Germany); EGF from Upstate Biotechnology (Hamburg, Germany); monoclonal anti-Heparan sulfate (10-E4 epitope) from Seikagaju Corporation (Tokyo, Japan); Labeled goat anti-mouse IgM antibodies from Molecular Probes (Leiden, Netherlands); Anti-c-Myc antibody from Invitrogen (Karlsruhe, Germany); Anti-goat IgG HRP- linked antibodies, anti-mouse IgG HRP- linked antibodies, anti-rabbit IgG HRP- linked antibodies and ECL immunoblotting detection reagents from Amersham Biosciences (Freiburg, Germany); complete mini-EDTA-free protease inhibitor cocktail tablets from Roche (Mannheim, Germany); DIFCO Noble agar from DIFCO Laboratories (Detroit, MI,) All other reagents were from Sigma (Munich, Germany). DNA array --------- The GeneChip^®^HG-U95Av2 array used in this study was fabricated by Affymetrix Inc. (Santa Clara, CA). Poly (A)^+^RNA isolation, cDNA synthesis, cRNA in vitro transcription, product purification and fragmentation was performed as described \[[@B16],[@B17]\]. Hybridization of the fragmented in vitro transcription product to oligonucleotide arrays was performed according to the manufacturer instructions (Affymetrix Inc.). Patients and tissues collection ------------------------------- 31 pancreatic cancer samples were obtained from patients (median age 62.5 years; range, 41--78 years), who underwent pancreatic resections for pancreatic cancer at the University Hospitals of Berne (Switzerland) and Heidelberg (Germany). 22 chronic pancreatitis samples were obtained from patients who underwent resection for chronic pancreatitis (median age 44 years range 22--66 years). 19 normal human pancreatic tissue samples were obtained from previously healthy individuals through an organ donor program (median age 45 years range 20--74 years). Immediately upon surgical removal, tissue samples were either snap-frozen in liquid nitrogen and then maintained at -80°C until use (for RNA extraction) or fixed in 5% formalin and embedded in paraffin after 24 h. All studies were approved by the Ethics Committees of the University of Heidelberg, and the University of Bern. Written informed consent was obtained from all patients. Cell culture ------------ Pancreatic cancer cell lines were routinely grown in DMEM medium (Panc-1 and Mia-PaCa-2) or RPMI medium (Aspc-1, BxPc-3, Capan-1, Colo-357, SU-8686 and T3M4) supplemented with 10% fetal calf serum (FCS), 100 U/ml penicillin, and100 μg/ml streptomycin (complete medium). Cells were maintained at 37°C in a humid chamber with 5% CO~2~and 95% air atmosphere. Real-time quantitative polymerase chain reaction (QRT-PCR) ---------------------------------------------------------- All reagents and equipment for mRNA and cDNA preparation were purchased from Roche (Roche Applied Science, Mannheim, Germany). mRNA was prepared by automated isolation using the MagNA Pure LC instrument and isolation Kit I (for cells) and Kit II (for tissues). RNA was reverse transcribed into cDNA using the 1^st^Strand cDNA Synthesis Kit for RT-PCR (AMV) according to the manufacturer\'s instructions. QRT-PCR was performed with the Light Cycler Fast Start DNA SYBR Green kit as described previously \[[@B18]\]. The number of specific transcripts was normalized to housekeeping genes (cyclophilin B and hypoxanthine guanine phosphoribosyltransferase, HPRT), and presented as adjusted transcripts / μl cDNA. All primers were obtained from Search-LC (Heidelberg, Germany). In situ hybridization --------------------- Specific human Hsulf-1 riboprobes were generated by reverse-transcription polymerase chain reaction using the following primer pairs: Hsulf-1: sense, 5\'-ACT GTA CCA ATC GGC CAG AG-3\'; antisense, 5\'-CCT CCT TGA ATG GGT GAA GA -3\'. The resulting polymerase chain reaction products were subcloned into the pGEM-T easy vector (Promega GmbH, Mannheim, Germany) containing promoters for DNA-dependent SP6 and T7 RNA polymerases. The authenticity of the subcloned Hsulf-1 fragment was confirmed by sequencing (Qiagen GmbH, Hilden, Germany). Plasmids were linearized using SpeI and NcoI restriction enzymes. T7 and SP6 RNA polymerases were used to construct sense and antisense complementary RNA riboprobes. Biotin complementary RNA labeling was performed using the biotin RNA labeling kit according to the manufacturer\'s instructions (Roche Diagnostics, Mannheim, Germany). Tissue sections (3 μm) were deparaffinized, rehydrated with 1× phosphate-buffered saline, and incubated in 0.2 M/L HCl for 20 minutes at room temperature. After rinsing the slides in 2× standard saline citrate, sections were treated with proteinase K (Roche Diagnostics) at a concentration of 25 μg/ml for 15 minutes at 37°C. After postfixation with 4% paraformaldehyde in phosphate-buffered saline for 5 minutes and washing in 2× standard saline citrate, samples were acetylated in 2.5% acetic anhydride and 1.5% triethanolamine for 10 minutes. Subsequently, sections were prehybridized at 78°C for 2 hours in 50% formamide, 4× standard saline citrate, 2× Denhardt\'s reagent, and 250 μg RNA/ml. Hybridization was performed overnight at 78°C in 50% formamide, 4× standard saline citrate, 2× Denhardt\'s reagent, 500 μg RNA/ml, and 10% dextran sulfate. The final concentration of the biotin-labeled probes was 0.8 ng/μl. After hybridization, excess probe was removed by washing the slides 3 times in Dako stringent wash solution (Dako) at 78°C for 15 minutes. The samples were then incubated with streptavidin alkaline phosphatase conjugate (Dako) for 30 minutes at room temperature. For the color reaction, 5-bromo-4-chloro-3-indolyl phosphate/nitroblue tetrazolium substrate (Dako) was used. Stable transfection ------------------- The Hsulf-1 expression plasmid (pcDNA 3.1/myc-His) \[[@B10]\] was a kindly provided by S.D. Rosen (University of California, San Francisco). Panc-1 pancreatic cancer cells were stably transfected with the Hsulf-1 plasmid and with the empty control vector using the lipofectamine reagent \[[@B7],[@B8]\]. Briefly, after reaching confluence, cells were split 1:10 into selection medium (complete medium supplemented with 1200 μg/ml G418) and single clones were isolated after 2--4 weeks. After expansion, cells fromeach individual clone were screened for the expression of Hsulf-1 by Northern blot analysis. Parental Panc-1 pancreatic cancer cells were also transfected with an empty expression vector carrying the neomycin-resistance gene as a control. Positive clones were routinely grown in selection medium. RNA extraction and Northern Blot analysis ----------------------------------------- The pGEMT-easy vector containing the Hsulf-1 fragment was linearized with SpeI, and a ^32^P -labeled riboprobe was synthesized with the RiboMAX™ large scale RNA production system kit using T7 polymerase and ^32^P CTP. Total RNA was extracted by the single step-acid guanidinium thiocyanate phenol chloroform method \[[@B7],[@B8]\]. RNA (15 μg/lane) was size fractionated on 1.2 % agarose/1.8 M formaldehyde gels. Gels were stained with ethidium bromide for verification of RNA integrity and loading equivalency. Fractionated RNA was transferred onto Genescreen membranes and cross linked by UV irradiation. Blots were then prehybridized for 12 h at 65°C in 50% formamide, 1% SDS, 5 × Denhardt\'s, 100 μg/ml salmon sperm DNA, 50 mM Na~2~PO4, pH 7.4, 10% dextran, 75 mM NaCl and 5 mM EDTA. Blots were then hybridized for 24 h at 65°C in the presence of ^32^P CTP labeled riboprobe, rinsed twice with 2 × SSC and washed twice with 0.2 × SSC/2%SDS at 65°C for 20 min, respectively. All blots were exposed at -80°C to Kodak BiomaxMS films with Kodak-intensifying screens. Immunoblot analysis ------------------- Cell culture monolayers were washed twice with ice-cold PBS and lysed with lysis buffer (50 mM Tris-HCl, 100 mM NaCl, 2 mM EDTA,1% SDS) containing one tablet of complete mini-EDTA-free protease inhibitor cocktail (in 10 ml buffer). Protein concentration was determined by the BCA protein assay (Pierce Chemical Co, Rockford, IL,). Cell lysates (30 μg/lane) were separated on SDS-polyacrylamide gels and electroblotted onto nitrocellulose membranes. Membranes were then incubated in blocking solution (5% nonfat- milk in 20 mM Tris-HCl, 150 mM NaCl, 0.1% Tween-20), followedby incubation with the indicated antibodies at 4°C overnight. The membranes were then washed in blocking solution and incubatedwith HRP-conjugated secondary antibodiesfor 1 hour at room temperature. Antibody detection was performedby an enhanced chemiluminescence reaction. Sulfatase assay --------------- To assay sulfatase activity in whole cell extracts, cells were lysed in lysis buffer (10 mM HEPES, 150 mM NaCl, 1% NP-40, 10% glycerol, 1.0 mM PMSF, and 1 mM EGTA), and the lysates were incubated on ice for 10 minutes. 4-Methylumbelliferyl-sulfate was used as the substrate. Cell lysates with 100 μg protein were diluted with SIE (250 mmol/l sucrose, 3 mmol/L imidazole, 0.1% absolute ethanol, pH 7.4) to a total volume of 100 μl. One hundred μl of 1 μmol/l 4-methylumbelliferylsulfate was added to each tube, mixed, and incubated at 37°C for 12 hours. 2 ml of stop solution (50 mM glycine, 5 mM EDTA, pH 10.4) was then added and mixed, and released 4-methylumbelliferone was measured using a fluorometer (excitation wavelength 360 nm, emission wavelength 460 nm). Immunofluorescence assay ------------------------ Cells were grown in complete medium overnight in 10-well chambers, washed with PBS, fixed with 4% formaldehyde/PBS for 20 min at room temperature (RT), incubated in ice-cold methanol for 5 min (RT) and subsequently in acetone for 2 min (RT). Cells were incubated overnight with an anti-mouse antibody that recognizes negative heparan sulfate that contains the N-sulfated glusamine residue (10E-4 mAb; 1:30 dilution) (Seikagaku Corporation Tokyo, Japan). Cells were then washed 3 times with PBS, incubated with Alexa Fluor labeled goat anti-mouse IgM antibodies (Molecular Probes, Leiden, Netherlands) for 1 hour at room temperature, washed with PBS and mounted with DAPI and anti-fading medium (Gel/mountTM, Abcam, Cambridgeshire, UK). Confocal microscopic analysis was performed using the Spectral Confocal Microscope Leica TCS SL (Leica Microsystems GmbH, Heidelberg, Germany). Proliferation assay ------------------- Anchorage-dependent cell growth was determined by the 3-(4,5-methylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) colorimeric growth assay \[[@B7]\]. Briefly, 2,000 cells/well were plated in 96-well plates and cultured for up to 7 days. Each day cell growth was determined by adding MTT solution (50 μg /well) for 4 hours. Cellular MTT was solubilized withacidic isopropanol and optical density was measured at 570 nm. The doubling time was calculated for the exponential growth phase. To measure and calculate the GI~50~of the chemotherapeutic agent (the concentration that causes 50% cell growth inhibition), graded concentrations of drugs were added to triplicate wells and GI~50~was calculated using the formula 100 x (T-T~0~)/(C-T~0~) = 50, where T is the optical density of the test well after 48 hours period of exposure to drugs, T~0~is the optical density at time zero, and C is the control optical density after 48 hours \[[@B19]\]. All experiments were performed in triplicate. Soft agar assay --------------- Cells (1 × 10^3^/well) were suspended in 3 ml of 0.3% Difco Noble agar supplemented with complete culture medium. This suspension was layered over 1.5 ml of 0.5% agar-medium base layer in 12-well plates. After 14 days, cells were stained with MTT (400 μg/well) for 24 h and colonies larger than 0.05 mm were counted. Invasion assay -------------- 8 μm filters were coated with Matrigel and placed in chambers. Cells (1.25 × 10^4^) cells were cultured in DMEM medium containing 1% FBS overnight. FGF-2 was added to the top chambers. After a subsequent 24 h incubation at 37°C, non-invaded cells were scraped off, and the cells that migrated to the lower surface of the filter inserts were fixed with 25% acetic acid and 75% methanol for 10 min and stained with 1% Toluidine blue in 1% sodium borate solution. The invasion index was expressed as the ratio of the percent invasion of the treated cells over the percent invasion of the control cells. Statistical analysis -------------------- Results were expressed as mean ± SEM, unless indicated otherwise. For statistical analysis, the Student\'s t test was used. Significance was defined as p\< 0.05. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= J.L, I.A, H.K carried out the in situ hybridization, immunoblotting, cell proliferation, invasion assays, and soft agar experiments. N.A.G. and T.G carried out the QRT-PCR analysis. J.K, M.W.B. and H.F. conceived the study and participated in its design and coordination. J.K. and K.F. drafted the manuscript. All authors read and approved the final version.	PubMed Central	2024-06-05T03:55:56.198991	2005-4-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087876/", "journal": "Mol Cancer. 2005 Apr 7; 4:14", "authors": [ { "first": "Junsheng", "last": "Li" }, { "first": "Jörg", "last": "Kleeff" }, { "first": "Ivane", "last": "Abiatari" }, { "first": "Hany", "last": "Kayed" }, { "first": "Nathalia A", "last": "Giese" }, { "first": "Klaus", "last": "Felix" }, { "first": "Thomas", "last": "Giese" }, { "first": "Markus W", "last": "Büchler" }, { "first": "Helmut", "last": "Friess" } ] }
PMC1087877	Introduction ============ Our perception of temperature is a finely tuned element of our somatosensory system, fundamentally allowing us to avoid thermal conditions that may be potentially harmful in nature. The preponderance of studies into thermosensation have focused on noxious heat, with the best characterized populations of thermosensitive afferents being those that have \'moderate\' and \'high\' heat thresholds of \~43 and 52°C, respectively \[[@B1]\]. The first insights into the molecules mediating thermosensation came from the cloning of the capsaicin receptor, or TRPV1, a non-selective cation channel activated by temperatures in the \"moderate\" heat range \[[@B2]\], and a TRPV1 homologue, TRPV2, that responds to temperatures near the \"high\" heat thresholds \[[@B3]\]. Taken together, these channels are considered critical for our perception of noxious, painful heat and provided the first clues in the molecular logic for thermosensation \[[@B4]\]. When do we feel cold? ===================== In contrast to the definitive thermal thresholds of noxious heat-sensitive nerves, similar distinctions for cool- and noxious cold-sensitive fibers have been problematic. In general, the perception of non-painful, cool temperatures is reported to occur when the skin is cooled as little as 1°C from normal body temperature \[[@B5]\]. In fiber recordings, temperatures in the range of 30--15°C will activate both Aδ- and C-fibers \[[@B5]-[@B7]\]. These cold-sensitive afferents will fire continuously at body temperature, with cold stimuli inducing an increase in their rate of firing, while warm temperatures reduce this activity \[[@B5]\]. However, peak responsiveness can vary between studies, falling between the temperatures of 25 and 15°C \[[@B5],[@B8]\]. Once temperatures approach 15°C, the perception of cold pain is evoked, with qualities described as burning, aching, and pricking \[[@B9]\]. However, the exact proportion of nociceptors that respond to noxious cold temperatures is not clear, with reported percentages ranging from 10 to 100% of Aδ- and C-fibers \[[@B6],[@B10]-[@B12]\]. Thus, it has been difficult to group cold-sensitive afferents in a manner similar to the distinct categorizations made of heat-sensitive fibers. More recently, a number of laboratories have studied cold using cultured dorsal root (DRG) or trigeminal (TG) ganglion neurons as in vitromodels of afferent nerves. In contrast to fiber recordings, it is consistently reported that approximately 10--20% of these cells will respond to cold temperatures, with thresholds for activation ranging from \~30 to near 15°C \[[@B13]-[@B16]\]. Moreover, two groups of neurons with distinct cold responses have been described, the predominant response characterized by a low-activation threshold temperature near 30°C and a second class of neurons with a high-activation threshold below 20°C \[[@B15],[@B17],[@B18]\]. Thus, the different activation thresholds suggest the former cells to be an in vitromodel for innocuously cool signaling afferents, while the latter may be analogous to those mediating noxious cold. Moreover, each neuronal population has a distinct response profile that suggests cold elicits these effects through different mechanisms (see below). Thus, these cultured sensory afferents have been a useful experimental model and provided insights into the mechanisms of cold signaling. Minty cool ========== The two populations of cold-sensitive cultured neurons described above can be furthered divided by their sensitivity to menthol, with the low-threshold cells being predominantly menthol-sensitive and the high-threshold population largely insensitive \[[@B15],[@B17]\]. As capsaicin elicits a sensation of burning and tingling heat, cooling compounds, such as menthol, elicit the psychophysical sensation of cold \[[@B19]\]. Menthol, a cyclic terpene alcohol found in leaves of the genus Mentha, is used in a wide range of products, such as confectionary, candy, toothpastes, vapo-rubs, and aromatherapy inhalations. When applied at low concentrations to the skin or mouth, menthol elicits a pleasant cool sensation, while higher doses can cause burning, irritation, and pain \[[@B19]-[@B21]\]. In fact, menthol was recently shown to evoke pain in humans through activation and sensitization of C-fibers \[[@B22]\]. Conversely, prolonged exposure to large doses of menthol will adapt or desensitize cold-sensitive neurons, a process analogous to that of capsaicin and heat-sensitive fibers \[[@B20],[@B23],[@B24]\]. Menthol application in the mouth can transiently prevent the irritancy of concomitant or subsequent capsaicin exposure, but very few studies have directly assessed the analgesic properties of menthol and it remains to be seen if menthol or other cooling compounds can be used as effective analgesics \[[@B25]\]. While it had been known for centuries that menthol produces a sensation of cold, the molecular site of action was not known, nor was it known if menthol and cold activate sensory afferents through similar mechanisms. Over fifty years ago, Hensel and Zotterman, using lingual nerve recordings, proposed that menthol exerts its actions on cold-sensitive fibers by raising their temperature threshold and suggested that menthol specifically acts upon a cold receptor \[[@B26]\]. Recent support for this hypothesis came from a number of studies in cultured sensory neurons where \~10--20% of excitable cells were menthol-sensitive \[[@B13],[@B14],[@B27]\], similar to the numbers reported to be cold-sensitive. Additionally, menthol and cold were both shown to elicit non-selective cation currents in these cells (threshold temperature near 28°C) and menthol-induced responses were temperature dependent \[[@B13],[@B14]\]. Therefore, it seemed likely that cold and menthol had a common molecular site of action, activating a Ca^2+^permeable channel. TRPM8, the minty-cool ion channel ================================= The above hypothesis was confirmed by the concurrent cloning of a cold and menthol receptor by two independent groups; one using menthol to expression-clone a cDNA, from rat TG neurons, that could confer menthol-sensitivity in heterologous expression systems \[[@B14]\], and another that used a genomics approach to identify TRP channels expressed in mouse DRG neurons \[[@B28]\]. This [c]{.underline}old and [m]{.underline}enthol [r]{.underline}eceptor, termed CMR1 or TRPM8, was activated at a temperature threshold of \~28°C, with currents increasing in magnitude down to 8°C \[[@B14],[@B28]\], thus spanning both innocuous cool to noxious cold temperatures. Additionally, the biophysical properties of cold- and menthol-induced currents (ion selectivity, rectification, menthol EC~50~, and temperature activation threshold) in heterologous cells expressing TRPM8 were reminiscent of those recorded in sensory neurons \[[@B13],[@B14]\]. TRPM8 transcripts are expressed in \<15% of small diameter sensory neurons, consistent with the proportion of excitable cells shown to be cold and menthol-sensitive \[[@B14],[@B28]\]. While the size of TRPM8 expressing neurons suggests them to be C-fibers, these cells do not express other markers such as TRPV1, neurofilament, or calcitonin gene-related peptide (CGRP) \[[@B28]\]. Thus, TRPM8 is not expressed in a class of afferents historically considered to be nociceptors \[[@B29]\]. In addition to menthol, a number of cooling agents, including icilin, eucalyptol, and WS-3, activate TRPM8 in vitro\[[@B14],[@B30]\]. The former of these compounds, icilin, is considered a super-cooling agent since it has higher potency and efficacy than menthol in cellular and behavioral studies \[[@B14],[@B31]\]. However, icilin appears to activate the channel in a manner that is divergent from other agonists, including cold. It was first reported that icilin can only activate TRPM8 in the presence of extracellular calcium \[[@B14]\], and Chuang et al. have further described the dependence of icilin activation of TRPM8 on calcium \[[@B32]\]. Indeed, TRPM8 acted as a coincidence detector of icilin and calcium in that a rise in intracellular calcium, either through influx via TRPM8 or release from intracellular stores, was required for icilin-induced TRPM8 currents. This study also mapped a critical amino acid residue required for icilin activity to a single glutamine in the third transmembrane domain of the channel. Interestingly, work from the same laboratory had previously determined that the capsaicin-binding site in TRPV1 maps to the same region \[[@B33]\], suggesting a conserved mechanism for ligand activation of these thermosensitive channels \[[@B32]\]. However, it should be noted that this single residue in TRPM8 does not appear to be involved in the menthol- and cold-sensitivity of the channel, thus suggesting that the TRPM8 can be gated by distinct mechanisms. Along with the number of cooling agents that activate TRPM8, several antagonists have been identified, including BCTC, thio-BCTC, capsazepine, and protons \[[@B30],[@B34]\]. The latter of these findings has further supported the notion of differential modulation of TRPM8 by various mediators. Specifically, lowering of the intracellular pH to below 7 was able to completely block TRPM8 currents elicited by either cold or icilin, but not menthol \[[@B34]\]. However, Behrendt et al. reported that both menthol- and icilin-responsiveness were reduced by lowering external pH (cold was not tested) \[[@B30]\]. Interestingly, in the former study, changes in intracellular pH dramatically altered the thermal threshold for activation of the channel, suggesting that intracellular acidity has some regulatory role in this regard. However, it still remains to be seen if the pH-dependence of TRPM8 can be placed in a physiological context, such as inflammatory injury. Cold receptors will adapt in vivowith prolonged cold stimulation \[[@B8],[@B35]\], a phenomena also observed in recordings of cultured sensory neurons \[[@B13],[@B36]\]. Menthol and cold-evoked currents in cells heterologously expressing TRPM8 will also adapt to prolonged stimuli in a manner that is dependent upon calcium \[[@B14]\], similar to capsaicin-induced desensitization of TRPV1 \[[@B37]\]. Interestingly, both menthol- and cold-induced adaptation were absent in recordings in excised patches from sensory neurons, suggesting this process is not an intrinsic property of the channel, but requires a cytoplasmic or membrane component lost upon membrane excision \[[@B24],[@B36]\]. While the mechanism of adaptation is not well understood, recent findings by Liu and Qin \[[@B38]\] have suggested that TRPM8, like many TRPM channels \[[@B39],[@B40]\], requires the presence of phosphatidylinositol 4,5-bisphosphate (PIP2) for activity. They demonstrated that menthol- and cold-evoked currents decreased or ran-down upon patch excision, a process that was inhibited under conditions of decreased phosphatase activity. Moreover, addition of exogenous PIP2 to the cytoplasmic face of the membrane recovered most of the menthol and cold-evoked currents. While the relationship between the effects of increased intracellular calcium in adaptation and PIP2-mediated channel rundown has not been established, these observations suggest that these two phenomena may be linked. Moreover, whether either is related to a mechanism of analgesia remains to be seen. Nonetheless, the cloning of TRPM8 established the first molecular detector of cold stimuli and its in vitroproperties are consistent with this role in vivo. Furthermore, TRPM8 confirmed Hensel and Zotterman\'s half-century old hypothesis \[[@B26]\] and established that TRP channels can confer thermal stimuli over broad ranges of temperature. TRPA1, a noxious cold sensor? ============================= While TRPV1 and TRPV2 established TRP channels as neuronal thermosensors, the cloning of TRPM8 suggested that detection of temperatures beyond the ranges of these channels may be conferred by other TRPs. Indeed, two members of the TRPV subfamily, TRPV3 and TRPV4, are involved in thermosensation of warmth \[[@B4]\]. In regard to cold sensation, as described above, a cold-sensitive, menthol-insensitive population of sensory neurons has been observed in culture, suggesting that these cells possess a cold thermosensor other than TRPM8 \[[@B15],[@B17]\]. Story et al. first suggested that the TRP-like channel TRPA1 (or ANKTM1) mediates cold-responsiveness in these cells when they reported that noxious cold temperatures activated the mouse orthologue of this ion channel \[[@B41]\]. This channel was first identified as a transformation-sensitive RNA transcript in human fibroblasts \[[@B42]\]. However, TRPA1 transcripts were later found in a population of sensory neurons distinct from those expressing TRPM8, but almost exclusively in nociceptive afferents that also express TRPV1, Substance P, and CGRP \[[@B41]\]. Calcium microfluorimetry and voltage-clamp recordings performed using mTRPA1-expressing Chinese Hamster Ovary (CHO) cells demonstrated that temperatures, with an aggregate threshold of \~17°C (range between 8--28°C), elicited non-selective cation currents that were blocked by ruthenium red, a blocker of several Ca^2+^-permeable channels. Moreover, the cooling compound icilin, a known agonist for TRPM8 \[[@B14]\], also activated TRPA1 currents, although with reduced potency compared to TRPM8 \[[@B41]\]. Thus, due to its expression pattern and temperature threshold, TRPA1 was proposed to be a detector of noxious cold in nociceptive afferents \[[@B41]\]. However, the above findings were questioned when Jordt et al. reported the rat and human orthologues of TRPA1 to be receptors for isothiocyanates, the pungent ingredients in wasabi and yellow mustard, and proposed the channel mediates the inflammatory and vasodilatory effects of these agents \[[@B43]\]. The controversy arose when this study did not observe cold-activation of TRPA1 currents when the channel was heterologously expressed in either a human embryonic kidney (HEK293) cell-line or Xenopusoocytes. Similar results were recently reported by Nagata et al. (see below) \[[@B44]\]. Furthermore, currents elicited by allyl isothiocyanate, or mustard oil (MO), were reduced upon a reduction in temperature to beyond the thermal thresholds reported by Story et al. \[[@B41],[@B43]\]. In addition to isothiocyanates, other pungent compounds were subsequently reported to activate TRPA1 in vitro, including the ingredients found in cinnamon (cinnamaldehyde), wintergreen, and clove oil, as well as ginger and methyl salicylate \[[@B45]\]. It should be noted that this latter report, from the same laboratory that originally reported cold-activation of TRPA1 \[[@B41]\], reproduced the earlier findings that the channel was sensitive to cold in both mammalian cells and Xenopusoocytes. In addition to these pungent compounds, the inflammatory peptide bradykinin also activated TRPA1 currents in a G-protein-coupled receptor-dependent manner, presumably via phospholipase C (PLC) \[[@B45]\]. Similarly, when TRPA1 was co-expressed with the PLC-coupled M1 muscarinic acetylcholine receptor (mACHR), application of acetylcholine elicited inward currents \[[@B43]\]. Thus, these data, and its expression in nociceptors, suggests that TRPA1 is involved in nociceptive signaling, and appears to mediate the distinct pungent sensations provided by a number of compounds. Moreover, it has also been postulated that TRPA1 plays an important role in inflammatory hypersensitivity in that the channel may be activated in a receptor-operated mechanism, perhaps through activation of PLC, by pro-algesic or pro-inflammatory mediators \[[@B43],[@B45]\]. While the ability of TRPA1 to respond to temperature in vitrois still unresolved, a number of studies using cultured sensory neurons have further confused the issue. In the initial description of TRPA1, Story et al. reported that cold (average threshold temperature of \~15°C) and capsaicin activate a menthol-insensitive population of mouse DRG neurons in culture \[[@B41]\]. Thus, the pharmacology of these responses, as well as the high-threshold temperature for activation, suggested that TRPA1 accounts for the cold-sensitivity of these cells. This same group further supported these original findings, reporting that \~70% and 90% of cold-sensitive, menthol-insensitive mouse DRG neurons responded to MO and cinnamaldehyde, respectively \[[@B45]\]. However, a number of alternative studies have failed to reproduce these findings. First, Jordt et al. did not find evidence for a population of cultured rat TG neurons that were sensitive to both cold and MO, but not menthol \[[@B43]\]. Greater than 90% of the cold-sensitive neurons were menthol-sensitive, while those few cells that were cold- and MO-sensitive (\~5%) also responded to menthol. Thus, the cold responses observed in this latter neuronal population were likely mediated by TRPM8. Secondly, Babes et al. recently suggested that TRPA1 does not mediate cold-responsiveness in cold-sensitive, menthol-insensitive neurons after they observed no correlation between cold sensitivity and MO responses in rat DRG cultures \[[@B17]\]. Lastly, two studies have provided indirect evidence supporting the notion that cold-sensitivity in high-threshold, menthol-insensitive cultured rat neurons is not mediated by TRPA1 \[[@B15],[@B18]\]. These reports showed that the majority of these cells were labeled with the isolectin B4 (IB4), a marker for non-peptidergic sensory afferents \[[@B46]\]. Thus, since TRPA1 expression was shown to be exclusively in CGRP-positive mouse DRG cell bodies \[[@B41]\], this would preclude TRPA1 expression in IB4-positive nerves. However, there may be some significant differences in the phenotype of afferents in culture versus in vivo, due to growth factor-dependent \[[@B41]\], or independent mechanisms \[[@B17]\]. It should also be noted that species variations (mouse versus rat) may be attributing to these discrepancies. The enigma of TRPA1 has been furthered by two additional findings. First, the Drosophila melanogasterorthologue of the channel was cloned and when expressed in heterologous expression systems, cold temperatures did not elicit membrane currents \[[@B47]\]. However, warm temperatures did activate dTRPA1 currents, within the range of 24--29°C, and flies with reduced expression of dTRPA1 exhibited deficits in normal thermotaxis to heat \[[@B48]\]. At the amino acid level, dTRPA1 is 32% identical and 54% similar to mTRPA1 \[[@B41]\]. In contrast, dTRPA1 is 22% identical and 39% similar to Painless, another drosophila TRP-like channel known to be involved in noxious thermal and mechanical signaling \[[@B49]\]. Thus, while dTRPA1 may be a relative sequence orthologue of the mammalian channels, it is not a functional one, and may be more related to Painless. Secondly, an alternative role for TRPA1 in sensory transduction has been proposed by Corey et al. and Nagata et al., reporting it as a candidate for the elusive mechanosensitive transduction channel in vertebrate hair cells \[[@B44],[@B50]\]. These findings were based upon localization of mTRPA1 transcripts and protein in hair cells, deficits in hair cell function after inhibition of TRPA1 protein expression, and similar biophysical and pharmacological properties of heterologously expressed TRPA1 and the hair cell transduction channel. Moreover, in the latter study cold could not elicit currents in cells heterologously expressing mTRPA1 \[[@B44]\], similar to the findings of Jordt et al. for the rat and human orthologues of the channel \[[@B43]\]. Thus, while cold activation of TRPA1 remains puzzling, the channel may have a diverse range of biological roles that depends upon the species and the cellular context in which the channel is expressed. Conclusion ========== The past few years have firmly established mammalian TRP channels as the primary detectors of thermal stimuli in the peripheral nervous system. These channels can account for temperature perception over the entire perceived temperature spectrum and also play fundamentally important roles in nociceptive signaling \[[@B4]\]. In comparison to the wealth of published data on heat- and capsaicin-sensitive nerves, and TRPV1, our knowledge of cold sensation and the involvement of TRPM8 and TRPA1 are still in their infancy. Thus, several key and fundamental issues regarding cold sensation and these channels remains (Figure [1](#F1){ref-type="fig"}). First, while the sensitivity range of TRPM8 encompasses both innocuous and noxious temperatures, the role of the channel in nociception is still unknown. The fact that high-doses of menthol can produce pain would suggest that TRPM8 is nociceptively-relevant. However, it is uncertain if TRPM8 is the only menthol receptor in sensory afferents, or if menthol affects other biological processes in a nociceptively-relevant manner \[[@B19]\]. Second, it is not known if TRPM8 or TRPA1 will provide good targets for as yet unidentified analgesics that may alleviate chronically painful conditions such as cold allodynia. Naturally occurring products, such as menthol, mustard oil, and cinnamon, have been used for centuries in nociceptively-relevant manners. Thus, now that molecular targets for these and related compounds have been identified, critical approaches can be developed to determine the role of these channels in nociception, and if compounds that modulate them can be used therapeutically. Lastly, whether or not TRPA1 is involved in cold-sensation needs to be reconciled by genetic approaches, such as TRPA1-null mice, or neuronal membrane current recordings combined with antibody labeling, studies undoubtedly currently underway in a number of laboratories. Until these and other, more directed approaches are performed, the evidence for or against TRPA1 functioning as a cold-sensor in vivois conflicting. Therefore, both TRPM8 and TRPA1 are and will remain fascinating molecules to study and, even though there is considerable debate over these thermosensors, as written by the 19^th^century essayist Lyman Beecher \"No great advance has ever been made in science, politics, or religion, without controversy.\"\[[@B51]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Molecular identity of cold-sensitive afferents based upon the expression of TRPM8 and TRPA1. TRPM8 and TRPA1 are found in distinct and non-overlapping populations of sensory afferents, with TRPA1 expressed exclusively in some, but not all, neurons that express the heat-gated channel TRPV1. These thermosensitive TRP channels respond to a number of naturally occurring pungent compounds, such as menthol (mint), allyl isothiocyanate (mustard oil), cinnamaldehyde (cinnamon), and capsaicin (\'hot\' chili peppers), thus providing a molecular explanation for how these compounds provide distinct sensations of cold, heat, or spiciness. Based upon in vitrocharacterizations of these channels, along with their distinct expression patterns, thermal stimuli activating TRPM8-expressing afferents elicit the sensation of cool to potentially noxious cold, while TRPA1 afferents will merge both noxious cold and noxious heat, due to the expression of TRPV1. ::: ![](1744-8069-1-16-1) ::: Acknowledgements ================ I would like to thank Emily Liman for her comments on the manuscript and this work was supported by a grant from the NIH to D.D.M.	PubMed Central	2024-06-05T03:55:56.202205	2005-4-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087877/", "journal": "Mol Pain. 2005 Apr 22; 1:16", "authors": [ { "first": "David D", "last": "McKemy" } ] }
PMC1087878	Background ========== Chronic renal failure (CRF) is associated with disturbances in the internal milieu, with repercussions on the immune, hematopoietic, gastrointestinal and endocrine systems and organs \[[@B1]-[@B4]\]. Patients with advanced CRF display a variety of hormonal abnormalities including perturbations of the hypothalamic-pituitary-thyroid hormone endocrine axis. The peripheral thyroid hormone metabolism is also altered in patients with CRF \[[@B5]-[@B9]\]. In uremia various thyroid hormone physiological characteristics are altered. Total and free tyroxine (T~4~) and 3,5,3\'-triiodothyronine (T~3~) levels in the serum are frequently reduced in patients with CRF \[[@B10],[@B11]\]. Reduced T~3~levels might be explained by decreased peripheral tissue conversion of T~4~to T~3~\[[@B12]\]. Most CRF patients, however, are considered to be euthyroid as evidenced by normal thyroid-stimulating hormone (TSH) levels \[[@B11],[@B13]\]. In addition, the prevalence of thyroid diseases, including goiter and hypothyroidism, are also higher in CRF patients than in the general population \[[@B10]\]. Thyroid hormones control numerous aspects of mammalian development and metabolism, of which most of these actions are mediated by specific thyroid hormone receptors (TRs). An important metabolic activity of thyroid hormones is to increase oxygen consumption of target tissues \[[@B14],[@B15]\]. In fact, in experimental renal failure and in uremic patients the expected increase in basal oxygen consumption following the administration of T~3~is not observed, suggesting that CRF is associated with resistance to thyroid hormone action \[[@B16]-[@B18]\]. However, it is currently not known whether thyroid hormone receptors play a role in the thyroid dysfunction observed in uremic patients. TRs are ligand-regulated transcription factors of the nuclear receptor superfamily which includes steroid hormones and vitamin D receptors and also PPARγ \[[@B15],[@B19],[@B20]\]. TRs modulate gene expression by binding specific DNA sequences, known as thyroid response elements (TREs), found in the promoters of TR-regulated genes. TREs are composed of repeats of the consensus half-site AGGTCA in a variety of different orientations, including direct repeats spaced by four nucleotides (DR-4), inverted palindromes (F2) and palindromes \[[@B21],[@B22]\]. In the presence of T~3~, TRs preferentially form heterodimers with the retinoid X receptors (RXRs) although unliganded TR binds to DNA as either homodimers or monomers \[[@B20],[@B23]\]. In recent years, it has become apparent that uremic toxins can impair the function of some nuclear receptors, such as the vitamin D receptor (VDR). Previous studies suggest that uremic toxins inhibit the binding of VDR to DNA and can contribute to the vitamin D resistance observed in CRF patients \[[@B24]-[@B28]\]. It is, therefore, conceivable that uremia also induces modifications in thyroid hormone receptors and consequently plays a role in the thyroid hormone dysfunction observed in uremic patients. We investigated the effects of uremia on TRβ1 function by studying the ability of TRs to bind to DNA sequences in the presence or absence of uremic plasma collected from CRF patients. Our results showed that uremic plasma significantly reduced the binding of TR heterodimers (TRβ1-RXRα), but not of homodimers (TRβ1-TRβ1) to DR-4. Furthermore, uremic plasma also inhibited the binding of a VDR heterodimer (VDR-RXR) to DR-3, while the binding of PPARγ to DR-1 remained unaltered. Moreover, hemodialysis (HD) diminished the inhibitory effect of CRF patients\' plasma on the binding of both TR and VDR heterodimers to DNA. When human promonocyte cells were incubated with ultrafiltrate collected Pre-HD the transcriptional activation induced by T~3~was inhibited. This inhibition was lost when the cells were treated with ultrafiltrate collected after-HD. Thus, we suggest that dialyzable uremic toxins selectively block the binding of TRβ1-RXRα and VDR-RXR heterodimers to DNA and reduce the transcriptional activities regulated by these receptors. These results indicate that the thyroid hormone dysfunction observed in uremia may be partially explained by T~3~resistance induced by impaired TRβ1 function. Results ======= Uremic plasma inhibits the binding of hTRβ1-hRXRα on DR-4 --------------------------------------------------------- To study the effects of uremic plasma on the ability of TRβ1 to bind to a specific TRE we analyzed the binding of hRXRα-hTRβ1 heterodimers to DR-4. In this assay, the protein-DNA complex was visualized by labeling the TRβ1 with ^35^S. In the presence of T~3~, the addition of increasing amounts (Figure [1](#F1){ref-type="fig"}; lanes 2--4) of plasma from normal individuals improved the binding of hRXRα-hTRβ1 to DNA. Conversely, TR incubation with uremic plasma (lanes 5--7) collected prior to hemodialysis significantly reduced the binding of heterodimers (RXR-TR) to DR-4. Band densitometry analysis of 5 independent experiments showed that uremic plasma reduced hRXRα-hTRβ1-DR-4 complex formation by 77 ± 15%, compared to plasma from normal subject (not shown). Similar results were observed for the thyroid response element F2 (inverted palindrome) in which uremic plasma also inhibited the RXR-TR binding to DNA (not shown). Pre-treatment of hTRβ1 with T~3~failed to improve the ability of the dimer hRXRα-hTRβ1 to bind to DNA. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Uremic plasma reduces hTRβ1-hRXRα complex formation on DR-4.Gel Shift experiments were performed using in vitro translated \[^35^S\] hTRβ1, cold hRXRα and DR-4. \[^35^S\] hTRβ1 was treated with T~3~10^-7^M for 30 min at 4°C and then incubated without (Control -- lane 1) or with increasing volumes (0.5, 1.0, 2.0 μL) of normal (lanes 2--4) or uremic (lanes 5--7) plasma for 30 min at 4°C. Cold DR-4 type TRE (5\'- AGCT TC AGGTCACAGG AGGTCAGAG - 3\'), cold hRXRα, and nonspecific DNA poly (dIdC) were subsequently added and incubated for 20 min. ::: ![](1478-1336-3-1-1) ::: We used uremic plasma from four different patients to determine whether the observed inhibition of RXRα-TRβ1 binding to DNA (DR-4) was patient specific. Uremic plasma samples from all four patients inhibited RXRα-TRβ1 binding to DR-4 to various degrees (not shown). However, we could not detect any correlation between TR binding impairment and abnormalities in plasma levels of urea, creatinine or thyroid hormone. To exclude the possibility that the inhibition of hRXRα-hTRβ1 binding to DNA induced by uremic plasma could be due to proteolytic degradation of TRβ1, we incubated ^35^S-labeled TRβ1, at the same conditions as in the gel-shift experiments, with normal or uremic plasma at 4°C, for thirty minutes. ^35^S-labeled TRβ1 samples were then analyzed by SDS-PAGE. As expected, the major translation product of TRβ1 was 53 kD and incubation of theses products with normal or uremic plasma did not modify the translated \[^35^S\]TRβ1 (Figure [2](#F2){ref-type="fig"}). These results indicated an absence of uremic proteolytic activity that might be involved in the decrease of TR-RXR complex formation on DNA. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Lack of proteolytic activity of uremic plasma on TRβ1. Recombinant \[^35^S\]TRβ1 was produced by translation in reticulocyte lysate and then incubated for 30 min at 4°C without (Control -- lane 1) or with increasing volumes (0.5, 1.0, 2.0 μL) of normal (lanes 2--4) or uremic (lanes 5--7) plasma. Products were analyzed by SDS-PAGE and autoradiography. ::: ![](1478-1336-3-1-2) ::: Hemodialysis improves hTRβ1-hRXRα binding to DR-4 ------------------------------------------------- It is yet unknown why uremic plasma diminished TRβ1-RXRα binding to DNA, when compared to non-uremic plasma. The observed effect could be ascribed either to a lack of some factor(s) typically present in normal plasma or to the presence of some inhibitory products present in uremic plasma. To test the second hypothesis, we analyzed the influence of uremic plasma, collected before and after hemodialysis, on TRβ1-RXRα-DR-4 complex formation using gel-shift assays. In these experiments, \[^35^S\] TRβ1 was incubated with normal or uremic plasma, collected before (pre-HD) or 4 h after hemodialysis (post-HD). As shown in Figure [3](#F3){ref-type="fig"}, uremic plasma collected before HD (lanes 5--7) decreased the binding of hTRβ1-hRXRα to DNA (DR-4), relative to normal plasma (lanes 2--4). However, when these receptors were pre-incubated with uremic plasma collected from the same patient after hemodialysis (lane 8--10), an important improvement of hTRβ1-hRXRα binding to DR-4 was observed. Although hemodialysis improved complex formation, it did not completely recover the inhibition caused by uremic plasma. Densitometry analysis demonstrated that hemodialysis increased by 50% the hTRβ1-hRXRα heterodimer binding to DNA when compared to pre-HD uremic plasma (not shown). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Hemodialysis reduces the inhibitory effect of uremic plasma on hTRβ1-hRXRα binding to DR-4.\[^35^S\] hTRβ1 was pre-incubated without (Control -- lane 1) or with increasing volumes (0.5, 1.0, 2.0 μL) of normal (lane 2--4) or uremic plasma (patient 1) collected before (pre-HD -- lanes 5--7) or after hemodialysis (pos-HD -- lanes 8--10) for 30 min at 4°C. Cold DR-4 type TRE, cold hRXRα, and nonspecific DNA poly (dIdC) were subsequently added and incubated for 20 min. ::: ![](1478-1336-3-1-3) ::: Hemodialysis reduces the inhibitory effect of uremic plasma upon hVDR-RXRα binding to DNA ----------------------------------------------------------------------------------------- In view that HD improved the binding of hTRβ1-hRXRα to DR-4 response element, and that previous studies showed that uremic solutions inhibited hVDR-RXRα binding to DR-3-response element (VDRE) \[[@B26],[@B27],[@B29]\] we decided to determine whether uremic plasma collected before and after hemodialysis has the same effect on hVDR-RXRα binding to DR-3. Accordingly, a similar experiment was carried out with hVDR treated with 1,25 (OH)~2~D~3~vitamin (VD~3~). \[^35^S\]hVDR was incubated with non-uremic plasma, uremic plasma collected before and after hemodialysis. As shown in Figure [4](#F4){ref-type="fig"}, in comparison to control (lane1), the incubation of VDR with plasma from normal individuals increased \[^35^S\]hVDR-RXRα binding to DR-3 (lanes 2--4). In contrast, when \[^35^S\]VDR was incubated with uremic plasma collected before HD, we observed a significant reduction of hVDR-hRXRα binding to DR-3 (lane 5--7). On the other hand, when compared to pre-HD uremic plasma pre-incubation with uremic plasma collected after hemodialysis (lanes 8--10) improved hVDR-hRXRα binding to DR-3. Taken together, these results suggest that dialyzable toxins were responsible for the reduced binding of hTRβ1-hRXRα and VDR-RXRα to DNA. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Uremic plasma inhibits VDR-RXRα-DR-3 complex formation and hemodialysis reduces this effect.Gel Shift experiments were performed using in vitro translated \[^35^S\] hVDR and cold hRXRα. \[^35^S\] hVDR was incubated with VD~3~vitamin for 30 min at 4°C and then without (Control -- lane 1) or with increasing volumes (0.5, 1.0, 2.0 μL) of normal (lanes 2--4) or uremic plasma, collected before (pre-HD -- lanes 5--7) and after hemodialysis (pos-HD -- lanes 8--10) from patient 1, for 30 min. Cold DR-3 type VDRE (5\'- AGCT TC AGGTCAAGG AGGTCAGAG - 3\'), cold hRXRα, and nonspecific DNA poly (dIdC) were subsequently added and incubated for 20 min. ::: ![](1478-1336-3-1-4) ::: Uremic plasma does not inhibit the binding of the hPPARγ-hRXR heterodimer to DR-1 --------------------------------------------------------------------------------- Next we decided to evaluate whether uremic plasma impairs DNA binding of other members of the nuclear receptor superfamily. We performed gel shift assays using the nuclear fatty acid receptor PPARγ and its response element, DR-1. Members of the PPAR family have been shown to play an important role in obesity and the plurimetabolic syndrome \[[@B30]\] and insulin resistance has also been described in uremic patients \[[@B31]\]. In contrast to what was observed with TRβ1 and VDR, Figure [5](#F5){ref-type="fig"} shows that the incubation of \[S^35^\] hPPARγ with uremic plasma failed to reduce the binding of PPARγ-hRXRα to DR-1 (lanes 2--4 compared to lanes 5--7). Such finding suggests that the inhibitory effect of uremic plasma does not extend to all members of nuclear receptor superfamily. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Uremic plasma does not decrease PPARγ-RXRα-DR-1 complex formation.Gel Shift experiments were performed using in vitro translated \[^35^S\] PPARγ and cold hRXRα. \[^35^S\] PPARγ was incubated without (Control -- lane 1) or with increasing volumes (0.5, 1.0, 2.0 μL) of normal (lanes 2--4) or uremic (lanes 5--7) plasma for 30 minutes. Cold DR-1 type TRE (5\'- AGCT TC AGGTCAG AGGTCAGAG - 3\'), cold hRXRα, and nonspecific DNA poly (dIdC) were subsequently added and the reaction was incubated for 20 min. ::: ![](1478-1336-3-1-5) ::: The inhibitory activity of uremic plasma is not thermo-labile ------------------------------------------------------------- Our findings demonstrated that hemodialysis partially corrected the inhibition of hTRβ1-hRXRα binding to DR-4 caused by uremic plasma. We therefore hypothesized that dialyzable toxins may cause the inhibitory effect of uremic plasma on protein-DNA complex formation. To gather more information on the nature of these toxins we heated normal and uremic plasma for 5 minutes at 100°C to test if the inhibitory factor was thermo-labile. We subsequently performed gel shift assays. To control for our experimental setup we used in these experiments \[^32^P\] labeled DR-4 with unlabelled TRβ1-RXRα (Figure [6](#F6){ref-type="fig"}). Furthermore, in order to avoid any effect from plasmatic phosphatases on \[^32^P\] DNA, we also incubated the normal and uremic plasmas with a phosphatase inhibitor cocktail (lane 6). ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Heating the uremic plasma does not diminish the inhibition of hTRβ1-hRXRα binding to DR-4.Gel Shift experiments were performed using in vitrotranslated cold hTRβ1, cold hRXRα and \[^32^P\]DR-4. Cold hTRβ1 was pre-incubated with normal or uremic plasma, heated or not, and in the presence or absence of phosphatase inhibitor for 30 min at 4°C. \[^32^P\] DR-4 type TRE, cold hRXRα, and nonspecific DNA poly (dIdC) were subsequently added and the reaction was incubated for 20 min. ::: ![](1478-1336-3-1-6) ::: Heating the plasma from normal individual did not affect the binding of the hTRβ1-hRXRα heterodimer to \[^32^P\] DR-4 (lane 1 compared to lane 2). As shown in prior experiments, uremic plasma significantly diminished the hTRβ1-hRXRα -- \[^32^P\] DR-4 complex formation (lane 4). However, heating the uremic plasma did improve the binding of the hTRβ1-hRXRα heterodimer to \[^32^P\] DR-4 (lane 5). In addition, the effects of uremic plasma persisted in the presence of phosphatase inhibitors (lane 6). Uremic Ultrafiltrate Impairs T~3~Transcriptional Activation ----------------------------------------------------------- Lastly we examined if the in vitroinhibitory effect of uremic plasma on TR binding to DNA could affect the functions of TR as a transcription regulator in a cell based assay (Figure [7](#F7){ref-type="fig"}). We transfected U937 cells with cDNA encoding the human TRβ1 and a reporter gene with a DR-4 response element upstream of the firefly luciferase coding sequence. We used RPMI-1640 medium treated with ultrafiltrate (UF) solution 10-fold concentrated from normal or uremic patients collected before or after hemodialysis. As shown in Figure [7](#F7){ref-type="fig"}, in the absence of any UF (Control), T~3~activated transcription by 5.7 ± 1.2 fold. When the cells were treated with UF from normal individuals, we did not observe significant changes in the transcription activation of the luciferase reporter (4.7 ± 0.74 fold; ns). Interestingly, when compared to the UF from normal individuals, UF collected Pre-HD from uremic patients reduced T~3~-dependent transcriptional activation by 66% (1.6 ± 0.35, p \< 0.05). On the other hand, U937 cells treated with UF collected Post-HD had no significant effect on T~3~-dependent transcription activation (4.8 ± 1.47 fold; ns). These results suggest that uremic toxins impair T~3~induced transcriptional activation and that hemodialysis reduces their inhibitory effect. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Uremic toxin (s) impair (s) T~3~-dependent transcriptional activation in U937 cells. A reporter construct consisting of two copies of a direct repeat thyroid response element (DR-4) 5\'AGGTCAcaggAGGTCA 3\' cloned upstream from the minimum thymidine kinase (TK) promoter, linked to the luciferase gene was examined in U937 cells. After electroporation, cells were transferred to fresh RPMI-1640 medium without (Control) or with normal or uremic ultrafiltrate solution, collected before (Pre-HD) or after hemodialysis (Post-HD). The cells were then plated in 12-well dish and treated with T~3~10^-7^M. After 24 h, cells were assayed for luciferase and β-galactosidase activities. \* p \< 0.05 versus control and normal ultrafiltrate. ::: ![](1478-1336-3-1-7) ::: Discussion ========== Uremia is a systemic chemical toxemia with repercussions on different organs and systems. Chronic renal failure patients demonstrate several endocrine dysfunctions, such as disturbance of thyroid hormone metabolism and are different from patients with the euthyroid sick syndrome. In the later, the conversion of T~4~to T~3~is reduced, but the generation of reverse T~3~(rT~3~) from T~4~is increased. In uremic patients, rT~3~is typically normal \[[@B32]-[@B34]\]. In addition, Lim et al. showed thyroid hormone resistance in hemodialysis patients with significantly reduced peripheral tissue sensitivity to thyroid hormone \[[@B17]\]. Recent data from our laboratory indicate that in order to maintain the euthyroid state showed that uremia increases T~3~influx across erythrocyte\'s membrane \[[@B35]\]. Taken together, these findings suggest that CRF affects thyroid function in multiple ways. The molecular mechanisms involved and the role of thyroid hormone receptor in this dysfunction, however, are not fully understood. In the present study we observed that uremic plasma impaired the ability of TRβ1 and VDR heterodimers (TRβ1-RXRα and VDR-RXRα) to bind to DNA (DR-4 and DR-3 respectively), whereas that the ability of PPARγ-RXRα to bind to DNA (DR-1) was not altered. Interestingly, there was no correlation between the inhibitory activity and the plasma levels of urea, creatinine, parathyroid and thyroid hormone of the patients enrolled in this study. However, the small number of patients precludes any definitive conclusions. To investigate whether these findings were secondary to the presence of uremic dialyzable toxins, we compared the effect of uremic plasma collected before and after hemodialysis, on TRβ1-RXRα or VDR-RXRα binding to DNA. Our results showed that the inhibitory effect of uremic plasma was significantly reduced by hemodialysis, suggesting that dialyzable toxins were in fact involved. We did not identify which toxin is responsible for this effect, but our results suggest the presence of thermo-resistant molecule(s). Further analyses of these dialyzable toxins are currently being conducted to identify and characterize the molecules responsible for this inhibitory effect. The mechanisms responsible for our findings are not clear. In uremic syndrome the reduced clearance of many toxins plays a key role in this pathogenesis. Although VDR degradation has been suggested in renal failure \[[@B36]\], the uremic inhibition of TRβ1-RXRα binding to DNA could not be explained by proteolytic activity of uremic plasma since our SDS-PAGE did not show any uremic plasma-dependent degradation of TRβ1. Our results are in agreement with other studies, which have shown that uremic toxins are involved in VD~3~resistance observed in patients with chronic renal failure \[[@B37]\]. Uremic ultrafiltrates derived from hemo or peritoneal dialyzed patients have been shown to inhibit the interaction of VDR with DNA \[[@B27],[@B28]\]. Further studies showed that the VDR complex formation on different types of VDREs can be reduced by uremic solutions collected from patients after hemo or peritoneal dialysis \[[@B28]\]. Our results allow us to speculate on the possibility of a common inhibitory mechanism involving the same uremic toxin(s) inhibiting both TR-RXR-DR-4 and VDR-RXR-DR-3 complex formation. To evaluate whether uremic toxins also affect other members of the nuclear receptor family, we studied the effect of uremic plasma on PPARγ-RXRα binding to DR-1. Contrary to what we observed with TRβ1 and VDR, pre-incubation of PPARγ with uremic plasma did not influence PPARγ-RXRα binding to DNA. This result suggests that the inhibition of protein-DNA complex caused by uremic plasma occurs only with some nuclear receptors. Taken together, our results indicate that uremic toxins exert their inhibitory effect by acting specifically on TRβ1 and VDR heterodimers. The molecular mechanism involved in this phenomenon is not clear. The fact that PPARγ-RXRα heterodimer was not affected by uremic plasma suggests that these toxins do not interact directly with RXRα. Another possible model to explain the effects of the toxins from uremic plasma on the binding of TR to DNA would be a direct action on DNA that would block its interaction with TRβ1 and VDR heterodimers. However, even though we used the DR-1 in PPARγ assay, in contrast to DR-4 (TRE) and DR-3 (VDRE), this hypothesis is not strongly supported by the results from this study, as PPARγ-RXRα heterodimers bind normally to DNA in the presence of uremic plasma. Another alternative that can not be excluded is an inhibitory effect of the uremic toxin on the surface of TR and VDR DNA binding domain (DBD), disrupting its ability to binding to DNA. Patel et al.attributed to the formation of Schiff bases between \"reactive aldehydes\" and lysine residues of the DBD of the VDR to explain the inhibitory effect of the uremic ultrafiltrate on the binding of VDR to DNA \[[@B26]\]. Nevertheless, in another study, point mutagenesis of different lysine residues in the DBD could not confirm this idea \[[@B28]\]. In addition, we should consider that the uremic toxins can interact with TR and VDR, causing structural conformational changes on these receptors, consequently, impairing heterodimers formation. We attempted to demonstrate the physiological relevance of these results by examining the effect of uremic toxins on T~3~transcriptional activation. Our results showed that uremic ultrafiltrate collected before hemodialysis inhibited T~3~-induced transcriptional activation, confirming the in vitrofindings. Conversely, in the presence of ultrafiltrate collected after hemodialysis, the transcriptional activation induced by T~3~was similar to the control group treated with ultrafiltrate collected from normal individuals. Therefore, we hypothesize that dialyzable toxins are responsible for the resistance to T~3~action documented in CRF patients. In summary, uremic toxins circulating in the plasma of CRF patients selectively reduced the binding of TRβ1-RXRα to DNA and impaired the TRβ1 transcriptional activation mediated by T~3~. Moreover, hemodialysis partially corrected this inhibitory effect, suggesting the presence of a dialyzable toxin. Since TRβ1 functions as a heterodimer with RXRα, these findings might explain some features of hypothyroidism and thyroid hormone resistance commonly found in CRF patients. Future studies are necessary to identify the toxins and further characterize the mechanisms involved in resistance to T~3~action in CRF patients. Materials and methods ===================== Patients and Clinical Procedures -------------------------------- Four patients from the chronic dialysis program of Soclimed Dialysis Clinic were enrolled in our study. All patients were men whose age ranged from 19 to 43 years with the mean age being 34 years. They appeared well nourished and clinically and laboratorial euthyroid; none had a history of thyroid disease, thyroid hormone therapy, treatment with amiodarone or clinically detectable goiter. Etiology of their chronic renal failure was as follows: chronic glomerulonephritis (2); hypertension (1); reflux nephropathy (1). Mean plasma urea level was 178 ± 44.8 mg/dL (120 to 233 mg/dL), while that of creatinine was 12.6 ± 2.7 mg/dL (9.9 to 16.3 mg/dL). Patients were on hemodialysis 3 times a week, during 4 hours using a 1.8 m^2^Fresenius^®^Polysulfone filter. Normal control subjects consisted of three healthy men, age ranging from 23 to 41 years, with the mean age of 32 years. The experimental protocol was approved by the Human Rights in Research Committee of the University of Brasilia and all patients and normal individuals gave their informed consent. For the in vitroDNA binding assay, uremic plasma was collected immediately before and after 4 h of hemodialysis, aliquoted into 20 μL samples and stocked at -20°C. Uremic ultrafiltrate (UF) was also collected pre and post 4 h hemodialysis. Lyophilisation was used to concentrate ultrafiltrate samples. The lyophilisates were re-suspended in bidistillated water to a 10-fold concentrated solution, as effects of the UF were not detectable at lower concentrations (1 fold, 2.5 fold, 5 fold concentrated; not shown). Samples were subsequently desalted by filtration using Centricon 3 filters. Following centrifugation, the pellet was re-suspended in RPMI-1640 medium, (10% newborn bovine serum; 2 mM glutamine; 50 units/mL penicillin; 50 μg/mL streptomycin) and pH corrected to 7. Normal UF was collected from control plasma of normal individuals. The treatment solution was prepared in the same manner as the uremic solution. All experiments were performed with the uremic sample from the same patient that showed the strongest inhibitory effect. Gel shift binding assay ----------------------- Gel shift assays were used to evaluate the binding of ^35^S-labeled TR synthesized in reticulocyte lysate on 600 fmoles of unlabeled DR-4 (5\'-AGCT TC [AGGTCA]{.underline} CAGG [AGGTCA]{.underline} GAG -3\') and inverted palindrome -- F2 (5\'-TTC [TGACCC]{.underline} CATTGG [AGGTCA]{.underline} GAG -3\'); ^35^S-labeled VDRs to unlabeled DR-3 5\'- AGCT TC AGGTCA AGG AGGTCA GAG - 3\') and ^35^S-labeled PPARγ to unlabeled DR-1 (5\'- AGCT TC AGGTCA G AGGTCA GAG - 3\'). Sensitivity and specificity of this assay have been previously characterized \[[@B23]\]. Briefly, the labeled protein will migrate in the nondenaturating polyacrylamide gel only when bound to DNA. Additionally, gel shift were also performed using unlabeled synthesized TRs and ^32^P-labeled DR-4. In vitroreceptor synthesis was performed using plasmids encoding hTRβ1 hRXRα, and hPPARγ \[[@B38]\] and hVDR \[[@B39]\] with the TNT-coupled Reticulocyte Lysate System (Promega, Madison, WI) containing a methionine-free aminoacid mixture, and either 20 μM cold methionine or ^35^S-labeled methionine. DNA plasmid (0.2--2 μg) was added to TNT Quick Master Mix and incubated in 50 μL for 90 min at 30°C. To confirm efficiency of the translation reaction ^35^S-labeled translated proteins were analyzed by sodium dodecyl sulfate gel electrophoresis (SDS-PAGE). The TR-DNA complex was visualized by labeling reticulocyte lysate-translated receptors with ^35^S or by using labeled DNA with ^32^P using polynucleotide kinase. Prior to incubation with DNA, the reticulocyte lysate-translated receptors were treated with TRβ1, VDR and PPARγ receptors ligands (T~3~, 1,25-dihydroxy-vitamin D~3~; and 9-cis-retinoic acid respectively) for 30 min at 4°C. Labeled nuclear receptor plus ligands were then incubated in different volumes (0.5, 1 and 2 μL) of uremic or normal plasma for 30 min at 4°C. Following plasma exposure, the nuclear receptors were incubated for another 20 min at room temperature (20--30°C) in a solution containing 2 μg non-specific DNA poly (dIdC) (Pharmacia LKB, Piscataway, NJ), cold specific response element (10 ng/reaction), nonradioactive RXRα and a binding buffer in a 20 μL reaction as previously described \[[@B23]\]. When using non-labeled nuclear receptors the gel shift experiment was performed using ^32^P-DR-4 (2000--5000 cpm). A phosphatase inhibitor cocktail 2 (Sigma, P 5726) was added when the DNA was labeled (^32^P-DR-4). The binding buffer contained 0.2 mM Na~2~HPO~4~, 0.2 mM NaH~2~PO~4~, 1 mM MgCl~2~, 0.5 mM EDTA, and 5% glycerol. Final samples were loaded on 5% nondenaturing polyacrylamide gel, previously run for 30 min at 200 V. To separate the protein-DNA complexes, the gel was run at 4°C for 90--180 min at 240 V, using a running buffer (pH 7.5 for 10X stock at room temperature) containing 6.7 mM Tris-base, 1 mM EDTA, and 3.3 mM Sodium Acetate. The polyacrylamide gel was dried at the end of electrophoresis and autoradiographed. Cell Culture, Transfections and Reporter Gene Assays ---------------------------------------------------- Human promonocyte U937 cells were maintained in culture as previously described \[[@B38]\]. For transfection assays, cells were collected by centrifugation and resuspended in transfection solution (0.5 mL/ 1.5 × 10^7^cells) containing PBS, 100 mM calcium and 0.1% dextrose and mixed with 2 μg of human TRβ1 expression vector, 4 μg of the luciferase (Luc) reporter and 500 ng control β-galactosidase vector. The reporter plasmid contained a synthetic TR response element containing two copies of DR-4 cloned immediately upstream of a minimal thymidine kinase (tk) promoter (-32/+45) linked to luciferase coding sequences \[[@B38]\]. The cells were transferred to a cuvette and electroporated using a Bio-Rad gene pulser at 300 V and 960 μF. Immediately after electroporation, the cells were transferred to fresh RPMI-1640 medium treated without or with normal or uremic ultrafiltrate solution (10 fold concentrated) collected before or after hemodialysis. Cells were then plated in 12-well dish and treated in triplicates with T~3~10^-7^M or ethanol (vehicle). After 24 h, cells were collected by centrifugation, lysed by the addition of 150 μL 1X lysis buffer (Promega) and assayed for luciferase (kit from Promega Corp.) and β-galactosidase (kit from Tropix, Inc., Bedford, MA) activities. All transfection experiments were performed at least three times. Statistics ---------- Data were analyzed by Kruskal-Wallis test followed by Dunn\'s Multiple Comparison Test when applicable. P \< 0.05 was considered statistically significant. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= G.M.S. carried out all the experiments and prepared the manuscript. C.J.A.B.P. helped with cells culture and manuscript preparation. A.C.S. and M.C.S. selected and monitored the patients and collected all uremic plasma samples enrolled in this study. L.A.S. and R.C.J.R. participated in design the experiment. N.L. helped with analysis of the data and in drafting the manuscript. F.A.R.N. conceived the study and participated in its design and coordination. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by Brazilian Research Council (Conselho Nacional de Desenvolvimento Científico e Tecnológico -- CNPq 520654/02-1; CNPq/PADCT SBIO 620003/02-2) and CAPES-COFECUB Program, grant 434/03. We thank all the staff from Soclimed Dialysis Clinic; Rilva Soares for all the technical assistance; Cristina Luisa Simeoni for help in preparation of this manuscript; Marília Barros for reviewing the manuscript and Prof John D Baxter for the helpful discussion.	PubMed Central	2024-06-05T03:55:56.204717	2005-4-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087878/", "journal": "Nucl Recept. 2005 Apr 4; 3:1", "authors": [ { "first": "Guilherme M", "last": "Santos" }, { "first": "Carlos J", "last": "Pantoja" }, { "first": "Aluízio", "last": "Costa e Silva" }, { "first": "Maria C", "last": "Rodrigues" }, { "first": "Ralff C", "last": "Ribeiro" }, { "first": "Luiz A", "last": "Simeoni" }, { "first": "Noureddine", "last": "Lomri" }, { "first": "Francisco AR", "last": "Neves" } ] }
PMC1087879	Background ========== It has been estimated that one-third of all cancer mortality in the United States (US) is related to diet\[[@B1]\]. Reviews of dietary studies have led groups, such as the American Institute for Cancer Research, to recommend that diet should largely be based on plant products with 400 grams of vegetables and fruits to provide more than 10 percent of energy consumed daily\[[@B2],[@B3]\]. The American Cancer Society added that intake of high-fat foods and alcohol should be limited\[[@B4]\]. The national objectives in both the US and Canada have been set at five or more servings per day of fruits and vegetables\[[@B5]\]. Average intake falls considerably short of this. In the US, intake is estimated to be 3.4 total servings of fruits and vegetables per day on average, but differs by age, ethnicity, and socioeconomic status\[[@B6]\]. Considerable recent research has focused on dietary change to increase fruit and vegetable consumption and to reduce fat consumption. The effectiveness of these interventions has been the subject of several systematic reviews\[[@B7]\]. There is some evidence that physician education in dietary counseling is an effective dietary intervention. However, there is no consistent evidence of effectiveness of other healthcare provider directed interventions. Interventions directed at individuals that were shown to have some effect in producing dietary change include: tailored interventions; multiple interventions; and provision of multiple contacts and environmental interventions. Media campaigns may result in increased knowledge and awareness of behaviors to reduce risks\[[@B7]\]. As the evidence grows for the effectiveness of dietary interventions, it is expected that more attention will be given to the dissemination and diffusion of these interventions to promote dietary change. The theoretical background for research dissemination and diffusion is complex and often contradictory. There are theoretical bases and models for dissemination and diffusion of research generally, and for behavior change of healthcare practitioners and the general public. These major fields of dissemination/diffusion and practitioner/client behavior change are inconsistently integrated into the development of interventions, and the field of cancer control is no exception. Closing the gap from knowledge generation to use in decision-making for practice or policy is conceptually and theoretically hampered by diverse terms and inconsistent definitions of terms, including diffusion, dissemination, knowledge transfer or translation or uptake or utilization, adoption, and implementation. There is a lack of distinction in the research between interventions to change behavior and strategies to disseminate that information. Furthermore, many studies have combined evaluation of both interventions and strategies within one study. Some activities (e.g., media campaigns, opinion leaders, and peer educators) can be characterized as both cancer control interventions and strategies to disseminate cancer control interventions to target audiences. This can lead to confusion about what is considered a cancer control intervention and what is considered dissemination of cancer control interventions. For the purpose of this evidence report, if an activity was used to provide educational information about the benefits of a desired cancer control behavior, it was classified as a cancer control intervention. If the activity was used to provide information about the availability or benefits of a cancer control intervention, it was classified as a strategy to disseminate a cancer control intervention. In keeping with Lomas\' views, this evidence report uses the term \"dissemination \" to refer to the active process of transferring cancer control interventions to target audiences and \"diffusion\" is used to refer to the passive spread of cancer control interventions\[[@B8]\]. Methods ======= The following question was addressed by this review: What strategies have been evaluated to disseminate cancer control interventions that promote the uptake of adult healthy diet?Primary studies of dissemination and diffusion strategies of dietary interventions were systematically reviewed. This review does not include studies of effectiveness of direct interventions to change dietary intake; rather, it includes those studies focused on dissemination of interventions, to adults and healthcare professionals. Primary studies were considered for inclusion if they were English language, published ≥ 1980 and evaluated dissemination of a cancer control intervention in one of the five topic areas. All primary studies regardless of study designs were eligible for inclusion. Reports exclusively focused on children or adolescents were excluded. Search strategies were developed as an iterative process in consultation with the McMaster Evidence based Practice Centre (EPC) librarian. The search strategy can be located at <http://www.ahrq.gov/clinic/evrptfiles.htm#cancontrl>, report name Cancer Control Interventions, Diffusion and Dissemination, file name 27appc.doc. Similar databases were searched for both objectives: \- MEDLINE, the U.S. National Library of Medicine (NLM) database \- PreMedline \- CancerLIT \- EMBASE the Excerpta Medica Database \- PsychINFO \- The Cumulative Index to Nursing and Allied Health Literature (CINAHL) \- Sociological Abstracts \- HealthSTAR \- Cochrane Database of Systematic Reviews (CDSR) \- Reference lists of pertinent articles and reviews; and \- The use of technical experts All data extraction forms were developed, pilot-tested, and revised by members of the local research team. Two reviewers completed data extraction independently for all reports. Any disagreements were resolved by consensus. The research team discussed differences that could not be resolved by these reviewers. Quality assessment was undertaken using standardized quality assessment tools developed by the Effective Public Health Practice Project. Tables were constructed to describe the most salient characteristics of the eligible studies. Meta-analysis was not undertaken because there were substantial differences across the studies, in terms of study design, intervention assessed, outcome measurements, methodological quality, and completeness of data reporting. Therefore, the report represents a systematic narrative review of the existing evidence. Results ======= Included Studies ---------------- The electronic database search identified 2,872 articles; 101 were retrieved for full text screening (Figure [1](#F1){ref-type="fig"}). Of these, nine reports of seven distinct studies are included: three reports about one study \[[@B9]-[@B11]\] and six other studies \[[@B12]-[@B17]\] are presented in Evidence Table 1(see [additional file 1](#S1){ref-type="supplementary-material"}). Ninety-two papers were excluded for lack of relevance; they did not address dissemination and diffusion strategies for dietary interventions. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Adult Healthy Diet: Search yield for studies evaluating dissemination strategies ::: ![](1475-2891-4-13-1) ::: Although the search inclusion criteria were broad, all of the eligible studies were conducted in the US. Six reports were published since 1998; the other four were published between 1989 and 1993\[[@B12],[@B13],[@B15],[@B16]\]. All seven projects were funded: five by the National Cancer Institute (NCI),\[[@B9],[@B14]-[@B17]\] one by the National Institute of Health (NIH),\[[@B12]\] and one by a private foundation\[[@B13]\]. One study achieved a rating of \"moderate\",\[[@B14]\] and all others were \"weak\" as defined by the standardized assessment tool\[[@B18]\]. The tool was adapted from those developed by Clarke et al.,\[[@B19]\]and Jadad et al\[[@B20]\]. As community interventions are often not evaluated by randomized trials, the tool reflects other possible study designs, and rates the following criteria: selection bias, study design, confounders, blinding, data collection methods (reliability and validity), withdrawals and dropouts, intervention integrity, and analyses. Based on a dictionary and standardized guide to assessing component ratings, each component was rated \"strong,\" \"moderate,\" or \"weak.\" Content and construct validity have been established\[[@B21]\]. A comparison of the tool used in this review was made with the tool used in the Guide to Community Preventive Health Services\[[@B22]\]. Four of the studies were randomized trials\[[@B9],[@B14],[@B16],[@B17],[@B23]\]. None of the other studies included a comparison group; three articles were descriptive,\[[@B11],[@B13],[@B15]\] one article was a cohort study\[[@B12]\] (Table 1)(see [additional file 1](#S1){ref-type="supplementary-material"}). Included studies were very diverse in the intervention that was disseminated and in strategies used for dissemination and diffusion. Only two studies compared two strategies\[[@B16],[@B17]\]. Of these, one study compared the effectiveness of a training workshop to postal delivery\[[@B17]\]. The second study evaluated whether the use of educational facilitators (academic detailing) plus a workshop was more effective than educational facilitators (academic detailing) only\[[@B16]\]. Each of the other studies evaluated the effectiveness of a single dissemination strategy. One strategy assessed was \"train-the-trainer\" to disseminate preventive medicine education to physicians;\[[@B12]\] two studies evaluated media campaigns for promoting access to a phone information services;\[[@B13],[@B15]\] one study assessed the effect of peer educators for improving fruit and vegetable consumption; \[[@B9]-[@B11]\] and one looked at the dissemination of intervention materials to control sites following the completion of a worksite nutrition intervention\[[@B14]\]. Outcomes were very diverse across studies and were not usually behavioral outcomes but rather process indicators, such as numbers of training sessions conducted,\[[@B12]\] numbers of physicians trained,\[[@B12]\] numbers of consumer telephone calls\[[@B13],[@B15]\], counts of peer-education strategies according to gender and ethnicity,\[[@B11]\] and uptake of materials by control sites following an intervention\[[@B14]\]. Client-based outcomes included knowledge\[[@B12]\] and intake of fruits and vegetables\[[@B9],[@B10]\]. Dissemination Studies That Targeted Healthcare Providers -------------------------------------------------------- ### Train-the-trainer One \"train-the-trainer\" study aimed at disseminating preventive medicine education to physicians\[[@B12]\]. Faculty from general internal divisions across the US were invited to apply for a month-long Stanford Faculty Development Program; 10 were chosen and trained to be Clinical Preventive Medicine facilitators. They then went to their home institutions and trained other faculty at their home site. Fidelity checks concluded that facilitators adhered closely to the curriculum they had been taught. Those medical faculty educated by the facilitators had an increase in knowledge and self-efficacy to use behavior changes to promote healthy diets. Subsequently, house staff physicians interacting with faculty who had attended the facilitator-run sessions reported an increase in the degree of preventive medicine content in teaching interactions and an increase in their ratings of self-efficacy to implement preventive medicine strategies\[[@B12]\]. While the train-the-trainer model shows some promise, it needs to be evaluated with a more rigorous design; furthermore, many biases are likely to be inherent in the selection of internists who were able to leave their work situation for a month of training. ### Academic detailing (educational facilitators) One Randomized Control Trial (RCT)\[[@B16]\] targeted dissemination to healthcare providers using academic detailing. In this trial by Dietrich et al., primary care medical practices were randomized to one of four groups: facilitator only, facilitator-plus-workshop, workshop only, or a control group. Practices in the facilitator-only group (n = 24) received three to four visits from a facilitator who provided detailed instruction and assistance in selecting and implementing non-computer-based office-system interventions. Practices in the facilitator-plus-workshop group (n = 26), in addition to receiving visits from an educational facilitator, had a physician from the practice attend a one-day workshop. The workshop session reviewed NCI\'s prevention and screening recommendations, but did not provide information on the use of office-system interventions. Practices in the workshop-only group (n = 24) attended the workshop. Practices in the control group (n = 24) received no information. Cross-sectional patient surveys were conducted before randomization and again at 12-month follow-up. The study reported on two diet-related outcomes: (1) the number of patients reporting that their physician had advised them to reduce their fat intake and (2) the number of patients reporting their physician had advised them to increase their fiber consumption. At 12-month follow-up, significantly more eligible patients in the facilitator-only group reported their physician had advised them to reduce their fat intake compared with patients in the control group (0.56 vs. 0.47, p \< 0.05). There was no significant difference in the number of patients reporting advice to decrease fat intake between the facilitator-plus-workshop group and the control group at 12-month follow-up (0.51 vs. 0.47). There was no significant increase in the number of eligible patients in the facilitator-only or facilitator-plus-workshop groups reporting advice to increase fiber consumption compared with patients in the control group at 12-month follow-up (facilitator vs. control 0.48 vs. 0.38; facilitator-plus-workshop vs. control 0.41 vs. 0.38). The overall conclusion from this RCT was that the use of educational facilitators to disseminate and implement office-system interventions could improve the provision of prevention and early detection services in community practices. The use of educational facilitators (academic detailers) to disseminate office-system interventions appears to be a promising strategy. Further research in this area is needed. ### Workshops The RCT Tziraki et al.\[[@B17]\] assessed the effectiveness of two strategies for promoting the use of an NCI nutrition manual by primary care physicians and their office staff. The nutrition manual was modeled after the NCI publication \"How to help your patients stop smoking\". Medical practices randomized to the workshop group (n = 244) were invited to send one staff member to a three-hour training workshop on how to use the nutrition manual. Training was provided in four major components of the manual: (1) how to organize the office environment, (2) how to screen for patient adherence, (3) how to provide dietary advice, and (4) how to implement a patient follow-up system. Medical practices assigned to the postal-delivery group (n = 256) received the nutrition manual in the mail with no further information. Medical practices in the control group (n = 255) did not receive the nutrition manual. Follow-up interviews with medical staff and observational assessments were conducted at four to six months after dissemination of the manual. Adherence scores were calculated for four areas: office organization, nutrition screening, nutrition advice or referral, and patient follow-up. There was low attendance at the workshop session; less than 50 percent of assigned practices sent representatives (120 of 244). The authors of the trial used an \"intent to treat\" approach for the primary statistical analyses and included all practices in the workshop group regardless of attendance. The workshop group was significantly more adherent to the manual\'s recommendations for office organization at follow-up than either the postal-delivery group (28.5 vs. 24.7 percent, p \< 0.005) or the control group (28.5 vs. 23.0 percent, p \< 0.001). Of those practices who sent a representative to the workshop, 30.6 percent were adherent to the recommendations for office organization. There was no significant difference between the postal-delivery group and the control group for office organization (24.7 vs. 23.0 percent). The workshop group was also significantly more adherent to the manual\'s recommendation for nutrition screening than either the postal-delivery group (23.5 vs. 21 percent, p \< 0.05) or the control group (23.5 vs. 20.5 percent, p \< 0.05). Of those practices that sent a representative to the workshop, 25 percent were adherent to the nutrition screening recommendations. There was no significant difference between the postal-delivery group and the control group for nutrition screening (21 vs. 20.5 percent). There was no statistically significant difference between the three groups for providing nutrition advice (workshop 54.9 percent, postal delivery 53 percent, control 52.3 percent), nor for patient follow-up (workshop 14.6 percent, postal delivery 13.6 percent, control 13.6 percent). A secondary analysis showed that those practices who attended the workshop were significantly more likely than either the postal-delivery group (57 vs. 53 percent, p \< 0.05) or the control group (57 vs. 52.3 percent, p \< 0.05) to provide nutrition screening. There was no significant difference observed for patient follow-up on secondary analysis. Training workshops appear to hold some promise as a dissemination strategy; however, motivating medical professionals to attend these sessions may be a difficult barrier to overcome. Further research in this area is needed. ### Postal delivery One RCT\[[@B17]\] evaluated the effectiveness of postal delivery as a dissemination strategy. This trial compared the effectiveness of postal delivery with a training workshop to disseminate an NCI nutrition manual to primary care practices. Postal delivery was not found to be an effective method to disseminate the nutrition manual. Please refer to the section above on Workshops for the detailed results of this study. Dissemination Studies That Targeted Worksites --------------------------------------------- ### Passive dissemination The Working Well Trial\[[@B14],[@B24]\] randomized 114 worksites of over 28,000 workers to test the effectiveness of health promotion activities that were planned and delivered with a high level of employee participation. The intervention phase lasted for two years, and then nutrition materials were disseminated to the control sites, followed by a further two-year assessment. The investigators were particularly interested to see if the control sites would utilize the materials. No information was given about the actual strategies used to get the nutrition intervention materials to the control group, nor was any report of measure of uptake given. No changes occurred in the level of nutrition activities in the control sites. An opinion leader strategy was tested using peer educators in the worksite intervention called \"5-A-Day: Healthier Eating for the Overlooked Worker\". While rated methodologically weak, it holds promise as an area for further research. It was an RCTof 5-A-Day intervention to increase fruit and vegetable consumption in an ethnically mixed population of 2,091 lower socioeconomic and trade employees\[[@B9],[@B10]\]. Both the intervention group and the control worksites received an 18-month intervention program of education materials through workplace mail, cafeteria promotions, and speakers. In the intervention group, naturally occurring work \"cliques\" were identified, and within those, ratings were given to each individual regarding their degree of \"centrality\" to communication ties and flow. Those rated highest in \"centrality\" became the peer educator for that clique, mimicking the \"opinion leader\" strategy. Peer educators attended a 16-hour training program where they were given information about health benefits of eating fruits and vegetables, cultural trends in dietary practices, peer educator\'s roles and responsibilities, and five persuasive communication strategies (foot-in-the-door, fear appeal, benefits, peer pressure, and questioning) and ways to initiate informal conversations about fruits and vegetables. They were instructed to engage in nutrition education of the co-workers for about two hours per week, on work time. They also distributed 5-A-Day materials produced specifically for this population: a nine-booklet resource guide, four issues of a newsletter, enabling gifts such as a recipe book, and vegetable seeds. The peer educator intervention lasted nine months, with consumption measured at the end of the intervention and six-month follow-up. The result was an increase in fruit and vegetable consumption of 0.77 total servings per day more in the intervention group compared with the controls (measured by recall, p \< 0.001) and an increase of 0.46 total daily servings (measured by food frequency, p \< 0.002)\[[@B9]\]. The effect was maintained at six-month follow-up for intake recall (increase of 0.41 daily servings, p = 0.034) but not for food frequency\[[@B9]\]. In analysis of the frequency and duration of peer-education contact with co-workers, greater contact with the peer educators was related to larger immediate increases in fruit and vegetable intake, particularly vegetable intake, but was not related to total intake at six-month follow-up\[[@B10]\]. A qualitative design, used to study the educational strategies used by the peer educators in the intervention group,\[[@B11]\] found that these studies differed by gender and ethnicity\[[@B11]\]. Hispanic educators were more likely to use individual, rather than group, change strategies than non-Hispanic educators; men more frequently used strategies such as \"mock competition\", \"giving materials\" and \"encouragement\", while female peer educators more often used \"creating context\", and \"keeping 5-A-Day visible\"\[[@B11]\]. Few worksite dissemination strategies have been evaluated. In one, the dissemination strategy was not evaluated\[[@B14]\]. The other study using an opinion leader strategy had at least a short-term impact on consumption. Dissemination Studies That Targeted Individuals ----------------------------------------------- ### Media strategies Two studies evaluated multiple media channels (print, television, and radio) to assess the impact of the media campaigns on telephone calls to an information telephone line\[[@B13],[@B15]\]. \"Project Lean\" (Low-Fat Eating for America Now) was a three-year initiative, begun in 1989, to reduce dietary fat consumption. The media campaign led to hotline access of 300,000 consumer calls in 18 months (25,000 to 28,000 calls/month), but the calls declined as publicity declined, and the line was terminated due to expense, estimated to be US \$300,000 per year\[[@B13]\]. While these outcomes were not assessed in a direct comparison, some important lessons were learned in this study:\[[@B13]\] that well-placed advertising may be the most appropriate and effective communications strategy for a national nutrition social marketing campaign as it can, more easily than Public Service Announcements (PSA\'s), be tailored to the particular audience; can communicate information more directly and can reduce the need for an information hotline or follow-up materials. Furthermore, building a network of state and local programs and partnerships with the food service industry allowed the campaign to reach a broader audience\[[@B13]\]. A second primary study was identified which was an analysis of calls to the Cancer Information Service (CIS) hotline. Callers were asked, \"How did you first find out about the CIS?\" Records of a subsample of people (214,472) who inquired about smoking, nutrition, Pap smears, and breast self-evaluation were reviewed. Television was the most frequently reported source of learning about the information line, regardless of age, gender, or ethnic group (except callers of Asian or Pacific heritage, who reported publications as the more common source of information about the hotline)\[[@B15]\]. The media dissemination strategies, particularly television messages, can make people aware of information lines and prompt them to call. However, from these two studies, it appears that the lines are expensive to advertise and maintain. Discussion and implications =========================== There has been increased recognition of the need for processes to transfer new knowledge into routine practice. Traditional methods of knowledge transfer such as journals and conferences have not proven effective in changing behavior\[[@B25]\]. Emphasis has been placed on the importance of research examining the dissemination of evidence-based knowledge and its uptake by the targeted recipients. Target audiences include providers, policymakers and the general public. There are several limitations of this review. It does not include the effectiveness of the dietary interventions themselves, but of the dissemination interventions to get others to know about the interventions. The results and conclusions are based on information available in published English-language reports. Contact with authors could have compensated for any reporting difficulties that resulted in a lower quality rating of the studies. Meta-analysis was deemed inappropriate due to the diversity in the target groups, interventions and outcome measures. There are few studies of dissemination of dietary interventions for cancer prevention. Overall, the quality of the evidence is not strong and is primarily descriptive rather than evaluative. Either process measures (numbers of calls, numbers of physicians educated, or number of educations sessions held) are reported or outcomes are often non-validated self-report measures. Controlled studies need to be done for any dissemination strategies, and dissemination and diffusion strategies with different messages and different target audiences need to be compared. More studies of healthcare providers with strategies such as opinion leaders or academic detailing should be done. The idea of a peer educator who is identified more as an opinion leader warrants further exploration. Cost-effectiveness needs to be established for any interventions. Most of the focus of research on healthy diet and cancer has been on evaluating interventions to promote behavior change. There is a lack of information on how to disseminate these findings to the community. Questions to address in future research include: What is the effectiveness of reminder strategies for health professionals to give interventions in-patient encounters? What innovative technologies can be brought to the dissemination strategies? Once media strategies have alerted the public to services, can effective interventions then be disseminated to individuals in such a way that they will utilize them to change dietary habits? Or is there an effective combination or sequencing of strategies that will result in dietary change? What policy level strategies are effective in at promoting dissemination of healthy diet interventions? What maintenance strategies can be incorporated to maintain the uptake and utilization of the evidence? Competing interests =================== The author(s) declare that they have no competing interests. Authors\' Contributions ======================= D. Ciliska: review of the literature, conceptualization, writing and editing P. Robinson: review of the literature and comments on the draft T. Armour: review of the literature and comments on the draft P. Ellis: review of the literature and comments on the draft M. Browers: review of the literature and comments on the draft M. Gauld: review of the literature and comments on the draft F. Baldassarre: review of the literature and comments on the draft P. Raina: review of the literature and comments on the draft Supplementary Material ====================== ::: {.caption} ###### Additional File 1 This table details the included studies in this review: citation, design, methodologic quality rating, strategy evaluated and findings. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This research was performed under contract to the Agency for Healthcare Research and Quality (Contract No. 290-97-0017), Rockville, MD. The National Cancer Institute, NIH, provided funding for the project and contributed to the design and production of the research. Dr. Raina holds a Canadian Institutes of Health Research (CIHR) Investigator Award. We would also like to thank Roxanne Cheeseman for help in the preparation of this article.	PubMed Central	2024-06-05T03:55:56.207595	2005-4-8	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087879/", "journal": "Nutr J. 2005 Apr 8; 4:13", "authors": [ { "first": "Donna", "last": "Ciliska" }, { "first": "Paula", "last": "Robinson" }, { "first": "Tanya", "last": "Armour" }, { "first": "Peter", "last": "Ellis" }, { "first": "Melissa", "last": "Brouwers" }, { "first": "Mary", "last": "Gauld" }, { "first": "Fulvia", "last": "Baldassarre" }, { "first": "Parminder", "last": "Raina" } ] }
PMC1087880	Background ========== With the rapid emergence of the HIV-1 and AIDS pandemic, tremendous effort has been directed towards development of effective treatments and vaccines. Currently, HAART is the only therapeutic option available for seropositive and symptomatic individuals, and is comprised of targeted inhibitors of HIV-1 reverse transcriptase (NNRTIs and NRTIs) and/or protease (PI) and the newly FDA approved gp41-inhibitor Fuzeon/T20 \[[@B1]\]. Though HAART is effective in prolonging life, its use, coupled with other factors, engenders rapid development of multiple drug-resistant strains. Therefore, the comprehensive elucidation of HIV-1-mediated effects on host cellular networks is urgently needed for rational therapeutic targets. HIV-1 infection, pathogenesis, and AIDS development are largely due to the various retroviral structural, regulatory, and accessory proteins, but more importantly due to efficient \'hijacking\' of cell regulatory machineries, including the differential expression of receptors, transcription, mRNA processing, and translation factors. While there has been much research on the effects of viral proteins on host cellular pathways, HIV-1 Tat appears to be the most critical for viral transcription and replication. HIV-1 Tat is absolutely required for productive, high titer viral replication. Though its sequence and a number of its functions have been uncovered, there is still much to learn about its replication-driven and pathogenic mechanisms, including the identification and characterization of Tat-regulated cellulargenes. With the advent of microarray technologies, it is now possible to assay the entire human genome for the effects of a single gene product, viral infection, or drug treatment. Many laboratories have previously demonstrated the effects of Tat on cell cycle-regulated transcription \[[@B2]-[@B4]\]. The finding that Tat activates gene expression at both the G~1~(TAR-dependent) and G~2~(TAR-independent) phases of the cell cycle demonstrates a concerted effort by Tat to take full advantage of cell cycle regulatory checkpoints. These findings prompted us to explore the effects of constitutive Tat expression on the expression profile of 1,200 host cellular genes in HIV-1 infected unsynchronized cells \[[@B5]\]. We observed that while the majority of cellular genes were down-regulated, especially those with intrinsic receptor tyrosine kinase activity, numerous S phase and translation-associated genes were up-regulated. These findings and the fact that inducing a G~1~/S block on infected cells dramatically reduces viral transcription and progeny formation \[[@B6]-[@B8]\], prompted us to follow and elucidate the effects of Tat on the host transcriptional profile throughout the entire cell cycle. Here, we report the HIV-1 Tat-mediated effects on the host expression profile relative to the cell cycle. We first performed microarray experiments in unsynchronized Tat-expressing cells compared to empty vector-transfected cells. We subsequently performed similar experiments in synchronized cells at the G~1~/S and G~2~/M phase boundaries. Cells were then collected at 0 h, 3 h, 6 h, and 9 h post-release per treatment corresponding to a specific cell cycle stage, and cytoplasmic RNA was isolated for microarray analysis. After microarray analysis using the Affymetrix U95Av2 gene chip, we found a wide variety of gene ontologies that were affected by Tat through cell cycle progression. We confirmed that Tat differentially regulates the expression of a variety of genes at different phases of the cell cycle, with an overall inhibition of the cellular transcription profile. Using siRNA technology to \'knock-down\' protein expression, we screened several of these genes as possible therapeutic targets for inhibition of HIV-1 replication. We generated a comprehensive list of Tat-induced genes at each cell cycle phase, particularly the G~1~/S phase transition, and expanded the list of Tat-regulated cellular proteins and potential therapeutic targets. Results and Discussion ====================== Microarray design and analysis ------------------------------ To understand which cellular genes were affected by Tat, we analyzed the transcription profile of \~12,000 gene transcripts using the Affymetrix U95Av2 gene chip. Cells were either transfected with the eTat plasmid or a pCep4 control vector. We chose to perform experimental and control conditions in duplicate to account for inter-chip variability. Figure [1A](#F1){ref-type="fig"} illustrates the cross-validity of the duplicate synchronized cell cycle experiments run for the eTat samples. The scatter plot graph logarithmically plots the probe set signal intensity values from the first experiment against those from the second experiment (average R^2^value = 0.912). Yellow spots represent gene probes with absent or marginal calls and the blue spots correspond to probes with present and marginal calls. Blue spots show less correlation and the yellow spots indicate the lowest level of correlation. Red spots represent those probes that displayed present calls in both experiments and thus demonstrate the highest level of correlation. The fold change lines indicate two-fold, three-fold, and ten-fold changes. Figure [1A](#F1){ref-type="fig"} shows the correlation of signal and detection values between the two experiments for each probe set, as well as the reliability of one dataset compared to its replicate. Similar results were observed for this analysis between the duplicate control pCep4 samples (data not shown). Though previous microarray experiments performed by us and others have used total nuclear and cytoplasmic RNA, we chose to isolate only cytoplasmic RNA because nuclear RNA would include RNAs that have been improperly spliced, or uncapped, and may have contain inappropriate poly-A tails, while cytoplasmic RNAs would yield almost a complete RNA population that has been properly processed prior to nuclear export and translation. As seen in Figure [1B](#F1){ref-type="fig"}, the RNA samples for both experiments show good RNA integrity with defined 18S and 28S bands. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Cross-validity of Tat samples and RNA isolation.(A) Cross-validity of the duplicate Tat samples analyzed. With a total of 32 gene chips, we analyzed the reliability of the gene chip samples relative to their respective replicate. The scatter graph logarithmically plots the signal intensity values of probe sets for one sample against those for a sample replicate. Each graph point indicates a common probe set between the two data sets and the value is determined by the intersection of the x and y values for that probe set. 2-fold, 3-fold, and 10-fold change lines are defined by the following equations: y = 2x and y = 1/2x, y = 3x and y = 1/3x, y = 10x and y = 1/10x, y = 30x and y = 1/30x. Yellow spots represent probes with absent-absent, absent-marginal, marginal-absent, and marginal-marginal detection calls on sample replicates. Blue spots represent those with absent-present, present-absent, marginal-present, and present-marginal calls, while red spots represent probe sets with present-present detection calls. (B) Cytoplasmic RNA was isolated from all experimental and corresponding control samples, and quantitated by UV spectrophotometric analysis; 3 μg was run on a 1% agarose gel for visual inspection. (C) IP/Westerns for Tat protein. Lanes 1--3 are from eTat extracts and Lanes 4--6 are from control pCep4 cells; unsynchronized cells are shown in Lanes 1 and 4. ::: ![](1742-4690-2-20-1) ::: We first studied the effects of constitutive Tat expression on the host cell transcription profile in unsynchronized cells and then relative to the cell cycle phases. Initially, a heterogenous cell population of Tat-expressing cells was compared to one expressing the pCep4 vector to create a global Tat-induced transcription profile. In the latter experiment, samples were treated with either hydroxyurea (Hu) or nocodazole (Noco) for 18 h to obtain either a G~1~/S or G~2~/M block, respectively. Cells blocked with Hu were 60% at G~1~, 35% at S, and 5% at the G~2~/M phase, while cells blocked with Noco were 6% at G~1~, 24% at S, and 70% at the G~2~/M phase (data not shown). Following cell cycle arrest, cells were washed and released in complete media. The 0 h time point following Hu treatment is representative of the G~1~/S phase of the cell cycle, while the 3 h, 6 h, and 9 h time points correspond to the early S, late S, and G~2~phases, respectively. Noco, a G~2~/M phase blocker, was added to the cell populations and the cells were likewise released. Samples were taken at the 0 h, 3 h, 6 h, and 9 h time points to obtain cells in the M and early, middle, and late G~1~phases, respectively. Immunoprecipitation and western blot analysis of tat protein were also carried out to verify the presence of tat in the unsynchronized and synchronized Tat-expressing cells and those expressing the pCep4 vector (Figure [1C](#F1){ref-type="fig"}). Thus, we obtained and analyzed the HIV-1 Tat-induced transcription profile at every cell cycle stage. All cell cycle phase populations were confirmed using FACS analysis as previously shown \[[@B2]\]. Gene expression analysis in unsynchronized Tat-expressing cells --------------------------------------------------------------- We analyzed the differential gene expression of a Tat-expressing cell population relative to that of a control population. This microarray analysis consisted of looking at \~12,000 genes in unsynchronized cells to ascertain the global effect of HIV-1 Tat-mediated transcriptional regulation on the host cell genome. Overall, we observed Tat-induced/-repressed differential expression of 649 genes (\~5% of genes screened) belonging to a wide variety of gene ontologies (Figure [2A](#F2){ref-type="fig"}). Figure [2B](#F2){ref-type="fig"} depicts gene ontologies for genes showing increased/decreased expression between the eTat and pCep4 samples. A few genes were represented as belonging to a variety of classifications and were placed into multiple categories. We observed the greatest effect (\~3%) of Tat on genes encoding for cellular enzymes; secretory, metabolic, and apoptotic pathways; and RNA binding, DNA binding, cytoskeletal, protein synthesis, and receptor proteins, while the other gene ontologies were less affected by Tat expression. We also observed that \~60% of the Tat affected genes were down-regulated. These findings are consistent with the previously published results by us and other laboratories \[[@B5],[@B9],[@B10]\]. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Gene ontologies present on the human U95Av2 chip and those specifically induced by Tat.(A) The U95Av2 gene chip was surveyed to determine the ontology of genes represented on the chip, as well as the corresponding number of genes belonging to each category. The percentages next to each classification correspond to the percentage of genes affected by Tat. (B) HIV-1 Tat-induced/repressed genes in an unsynchronized HeLa-eTat cell population. The number of genes induced/repressed by Tat, as well as the various classifications, is shown. ::: ![](1742-4690-2-20-2) ::: HIV-1 Tat-induced transcription profile --------------------------------------- Using a two-fold threshold to constrain our gene lists to those genes only significantly induced by Tat, we observed many genes that were expressed during all cell cycle phases, with fewer genes that were exclusive to only one cell cycle phase. This can be seen in both the self-organizing maps (SOMs) and k-means analysis graphs \[Figures [4](#F4){ref-type="fig"} and [3](#F3){ref-type="fig"}, respectively & Additional Files [5](#S5){ref-type="supplementary-material"}, [6](#S6){ref-type="supplementary-material"}, and [7](#S7){ref-type="supplementary-material"}\]. In the 3 sets of SOMs generated using three separate filtering rules, we observed many genes that were relatively consistent in their expression patterns through most cell cycle phases. This was also evident in the k-means graphs that contain gene clusters whose expression was relatively linear \[see [Additional File 7](#S7){ref-type="supplementary-material"}: sets 1, 10, 11, and 14\]. In the k-means analysis, the y-axis represents the normalized intensity values for the genes analyzed and the x-axis contains two sets of eight time points for each condition. K-means clustering allows for the elucidation of those genes with similar temporal expression profiles. As shown in \[[Additional File 7](#S7){ref-type="supplementary-material"}\], the various graphs correspond to separate clusters of genes whose expression is similar in Tat-expressing cells relative to cell cycle progression. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### K-Means clustering analysis of Tat-induced genes.The temporal differential gene expression in Tat cells was compared to respective control samples and analyzed using the k-means clustering algorithm. The coordinated expression profiles are representative of the 32 chips analyzed (16 eTat and 16 pCep4). The y-axis represents the log scale of the normalized intensity of the genes shown (data was normalized against the corresponding control samples). The x-axis corresponds to the various cell cycle phases: 1) M phase, 2) early G~1~, 3) middle G~1~, 4) late G~1~, 5) G~1~/S, 6) early S, 7) late S, and 8) G~2~. Fifteen clusters were found based on the parameters used \[see [Additional File 7](#S7){ref-type="supplementary-material"}\] and three are shown in 3A-C. Figure 3A shows altered genes at the G1/S for cathepsins, and various cellular receptors, while Figure 3B shows a close-up of apoptotic regulated genes, signal transduction and transcription factors. Figure 3C shows genes that dramatically oscillate at every stages of cell cycle in Tat expressing cells, including ribosome and actin/cytoskeleton genes. ::: ![](1742-4690-2-20-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Temporal SOM analysis of HIV-1 Tat-induced cellular genes in synchronized Tat cells.3 separate filters were applied to remove genes that did not display at least a 1.5, 2, or 3-fold change at each time point analyzed in the 16 eTat chips (see Methods); each filter produced a discrete dataset that was applied to SOM analysis. The third and most restrictive dataset is shown here. Genes that were significantly up (red) and down-regulated (blue) are shown. The U-matrix identifies which genes are similar to each other in terms of expression profile (blue) separated by a \"boundary\" (red). This SOM graph contains 17 rows and 6 columns of neurons, represented as coordinates. The arrows adjacent to the G~1~/S SOM indicate those genes significantly up-regulated during this transition and S phase, and those that show decreased expression in the G~1~phase. ::: ![](1742-4690-2-20-4) ::: Based on the k-means clustering methods, we observed a coordinated up-regulation of 228 genes during the G~1~/S phase transition in set 14 (Figure [3B](#F3){ref-type="fig"}) and 54 genes in set 12 (Figure [3A](#F3){ref-type="fig"}). On the other hand, set 5 (Figure [3C](#F3){ref-type="fig"}) displays genes whose expression peaks at different time points in the cell cycle, but are specifically down-regulated at the G~1~/S boundary. Set 12 (Figure [3A](#F3){ref-type="fig"}) was very similar to the results seen with the G~1~/S SOM (Figure [4](#F4){ref-type="fig"}), in which genes were up-regulated at the G~1~/S phase and continued to be highly expressed until the G~2~phase. Set 12 illustrates the increased expression of various cathepsins (L, L2, Z, PPGB), receptors (EGFR, lamin B, poliovirus), solute/ion carrier transporters, and MHC molecules (HLA-C, HLA-A, GRP58). In set 14 (Figure [3B](#F3){ref-type="fig"}), genes whose expression peaked at the G~1~/S phase transition were observed, though a greater number of genes relative to set 12 with similar expression patterns and functions were found. For example, we observed up-regulation of apoptosis regulators (UDP-galactose ceramide glucosyltransferase, BAX, BAX inhibitor 1, TRAIL receptor 2, thioredoxin peroxidase, CD47, API5-like 1), receptors/adhesion proteins (CCRL2, LIFR, EGFR, FGFR1, syndecan 4, syndecan 1, IL-4R, IL-13R, lymphotoxin B receptor), signaling mediators (Grb2, AKAP1, IRAK1, CaM-kinase II, calcineurin), and proteins involved in transcriptional regulation (BAF60C, NFI/C, ATF6). Interestingly, 26 genes in this cluster were related to the ER-Golgi protein transport pathway, suggesting a dependence on efficient protein processing and intracellular transport. These findings suggest an increase in Tat-induced receptor-mediated signaling and transcription, and most importantly, the increased expression of membrane proteins and antigens involved in promoting HIV-1 replication and immune evasion. On the other hand, set 5 (Figure [3C](#F3){ref-type="fig"}) shows 20 genes whose expressions peaked at late G~1~, early S, and then again at G~2~, while their expressions were lowest at early G~1~. This set contains primarily ribosomal subunit genes. We previously observed very similar results in our microarray experiment using Tat-expressing H9 cells \[[@B5]\], where we saw a significant up-regulation of numerous ribosomal subunit genes and translation initiation factors. The dramatic temporal expression of the ribosomal subunits for the 40S and 60S components in early S, as seen in set 5, may be indicative of a critical coupling of transcription and translation for efficient viral RNA production. Tat-mediated gene expression during G~1~/S phase ------------------------------------------------ Using a complementary technique for unsupervised clustering, we looked at those genes that were induced by HIV-1 Tat during the late G~1~phase and the G~1~/S phase transition since our previous findings indicated that these cell cycle phases were starting points for transcription of the HIV-1 long terminal repeat (LTR) and activated viral transcription \[[@B2]\]. The SOM analysis makes it easier to visualize the dramatic cell cycle effects of Tat on the total gene dataset. In this analysis, red areas indicate up-regulated genes, while blue indicates down-regulated genes, and yellow represents minor effects on gene expression. The U-matrix allows visualization of those clusters in the SOM that show significant expression changes. Each hexagon or neuron corresponds to a group of genes with similar expression patterns. We performed 3 filters to generate SOMs, with the last filter being the most restrictive (Figure [4](#F4){ref-type="fig"}). The most restrictive list includes genes that show a 3-fold increase or decrease in expression between the experimental and control samples at each time point. For this particular SOM, genes were removed if their average signal ratio fell between 0.333 and 3.0 across all time points tested and displayed absent calls at any time point. Using the SOM analysis from the third filter (Figure [4](#F4){ref-type="fig"}), we observed a similar transcription profile throughout the G~1~phase, with a marked difference at the G~1~/S transition. This is seen with the dramatic induction of those genes represented in the red and dark red neurons at the bottom right portion of the G~1~/S SOM. Repression of genes on the left side of the G~1~component plane, when cells enter the G~1~/S transition, was also observed. Interestingly, the G~1~/S profile remained relatively constant through the S phase, while upon entering G~2~, there was an overall reduction in Tat-mediated gene activation. This can be seen with the greater percentage of blue neurons at the G~2~phase concomitant with a reduction of dark red neurons. We generated a list of genes up-regulated at the G~1~/S transition that were seen in both k-means and SOM clustering analyses (Table [1](#T1){ref-type="table"}). Bolded genes are those that have already been shown to be involved in HIV-1 infection. It is important to note that there were a significant number of genes that were identified as similarly dysregulated by using both the k-means and SOM analyses across all time points. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### SOM and K-means Analysis of Tat-upregulated genes at the G~1~/S phase.^a^ ::: Gene Ontology Accession \# Gene Title Gene Symbol Unigene ID ----------------------------- ------------------ ----------------------------------------------------------------------- ----------------- ---------------- Transcription/ D83782 SREBP cleavage-activating protein SCAP Hs.437096 DNA binding AC004770 fatty acid desaturase 3 FADS3 Hs.21765 Enzymes Y08685 serine palmitoyltransferase, long chain base subunit 1 SPTLC1 Hs.90458 D50840 UDP-glucose ceramide glucosyltransferase UGCG Hs.432605 AF038961 mannose-P-dolichol utilization defect 1 MPDU1 Hs.95582 U67368 exostoses (multiple) 2 EXT2 Hs.75334 M22488 bone morphogenetic protein 1 BMP1 Hs.1274 AF002668 degenerative spermatocyte homolog, lipid desaturase (Drosophila) DEGS Hs.299878 AB016247 sterol-C5-desaturase-like SC5DL Hs.287749 X15525 acid phosphatase 2, lysosomal ACP2 Hs.75589 D13643 24-dehydrocholesterol reductase DHCR24 Hs.75616 AF020543 palmitoyl-protein thioesterase 2 PPT2 Hs.332138 AL050118 fatty acid desaturase 2 FADS2 Hs.388164 M16424 beta-hexosaminidase A (alpha polypeptide) HEXA Hs.411157 L13972 sialyltransferase 4A (beta-galactoside alpha-2,3-sialyltransferase) SIAT4A Hs.356036 Membrane/ D79206 syndecan 4 (amphiglycan, ryudocan) SDC4 Hs.252189 Antigens M90683 HLA-G histocompatibility antigen, class I, G HLA-G Hs.512152 X58536 major histocompatibility complex, class I, C & B HLA-C, B Hs.77961 AF068227 ceroid-lipofuscinosis, neuronal 5 CLN5 Hs.30213 U72515 putative protein similar to nessy (Drosophila) C3F Hs.530552 X85116 stomatin STOM Hs.439776 Z26317 desmoglein 2 DSG2 Hs.412597 S90469 P450 (cytochrome) oxidoreductase POR Hs.354056 Receptors/Ligands U97519 podocalyxin-like PODXL Hs.16426 AI263885 interleukin 27 receptor, alpha IL27RA Hs.132781 U60805 oncostatin M receptor OSMR Hs.238648 M63959 low density lipoprotein receptor-related protein associated protein 1 LRPAP1 Hs.75140 L25931 lamin B receptor LBR Hs.435166 X00588 epidermal growth factor receptor EGFR Hs.77432 M25915 clusterin CLU Hs.436657 X87949 heat shock 70 kDa protein 5 (glucose-regulated protein, 78 kDa) HSPA5 Hs.310769 Proteases AF032906 cathepsin Z CTSZ Hs.252549 AB001928 cathepsin L2 CTSL2 Hs.87417 Y00264 Amyloid beta (A4) precursor protein APP Hs.177486 Protein transport/Chaperone D83174 serine (or cysteine) proteinase inhibitor, clade H, member 1 SERPINH1 Hs.241579 Z49835 glucose regulated protein, 58 kDa GRP58 Hs.110029 X97335 A kinase (PRKA) anchor protein 1 AKAP1 Hs.78921 X90872 gp25L2 protein HSGP25L2G Hs.279929 D49489 thioredoxin domain containing 7 (protein disulfide isomerase) TXNDC7 Hs.212102 AF013759 calumenin CALU Hs.7753 AL008726 protective protein for beta-galactosidase (galactosialidosis) PPGB Hs.118126 Z50022 pituitary tumor-transforming 1 interacting protein PTTG1IP Hs.369026 AA487755 FK506 binding protein 9, 63 kDa FKBP9 Hs.497972 Ion channel/transporter U81800 solute carrier family 16, member 3 SLC16A3 Hs.386678 M23114 ATPase, Ca++ transporting, cardiac muscle, slow twitch 2 ATP2A2 Hs.374535 J04027 ATPase, Ca++ transporting, plasma membrane 1 ATP2B1 Hs.20952 AL049929 ATPase, H+ transporting, lysosomal accessory protein 2 ATP6AP2 Hs.183434 AL096737 solute carrier family 5, member 6 SLC5A6 Hs.435735 Unknown/Other AF052159 protein tyrosine phosphatase-like, member b PTPLB Hs.5957 D14658 KIAA0102 gene product KIAA0102 Hs.87095 AI867349 nicastrin-like protein NICALIN Hs.24983 AL031228 solute carrier family 39 (zinc transporter), member 7 SLC39A7 Hs.66776 X57398 nodal modulator 1, 2, 3 NOMO1, 2, 3 Hs.429975 ^a^Bolded genes indicate those genes upregulated at the G1/S transition (found using both SOM and k-means analyses) ::: Numerous signaling receptors were shown to be up-regulated upon Tat expression. The oncostatin M receptor is normally bound by the IL-6 cytokine family member and is increased in HIV-1 infection \[[@B11]\]. Interestingly, oncostatin M has been shown to stimulate the production of immature and mature T cells in the lymph nodes of transgenic mice \[[@B12]\]. It has also been shown that cdk9, a component of pTEFb, can also bind gp130, which is a common subunit recognized by the IL-6 cytokine family \[[@B13]\]. Expression of the 4-1BBL cytokine, a T-cell co-stimulatory molecule (i.e. induces IL-2 production and T-cell proliferation) that is involved in the antigen presentation process and generation of a CTL response was also increased \[[@B14],[@B15]\]. Similarly, we observed the up-regulation of LFA-3, ICAM-1, and other membrane proteins and receptors. These membrane proteins serve as additional activation signals and molecules involved in the transmission of free virus to bystander, uninfected cells \[[@B16]-[@B18]\]. Interestingly, a recent report illustrates the ability of soluble ICAM (sICAM) to promote infection of resting cells and cell cycle progression after initiating B and T cell interactions \[[@B19]\]. Syndecan 4 was also up-regulated by Tat at the G~1~/S phase. Syndecans are a type of heparan sulfate proteoglycan (HSPG) that is able to efficiently attach to HIV-1 virions, protect them from the extracellular environment, and efficiently transmit the captured virions to permissive cells \[[@B20]\]. We also observed the up-regulation of the CXCR4 co-receptor that is critical for infection by X4 HIV-1 strains. Likewise, the SDF receptor 1 had increased expression. SDF-1 is the ligand for the CXCR4 co-receptor and can block HIV-1 infection via co-receptor binding. Therefore, the expression of the SDF receptor 1 could serve as an alternate binding site for SDF-1, allowing CXCR4 to be available for HIV-1 gp120/gp41-binding. Fractalkine, the ligand for the CX3CR1 receptor, has been shown to be important in the adhesion, chemoattraction, and activation of leukocytes \[[@B21]\], was also up-regulated by Tat expression. Overall, these proteins serve to increase the efficiency of HIV-1 infection, transmission to other cells, activation of T cells, and the recruitment of circulating leukocytes to infection sites. A critical feature of HIV-1 infection is its ability to evade host immune responses and subsequently create a state of immunodeficiency. Previous studies have shown the ability of HIV-1 Nef to decrease the expression of CD4, HLA-A, and HLA-B, while having no effect on HLA-C or HLA-D, which allows for host cell survival and permits productive viral progeny formation prior to immune recognition and eventual apoptosis \[[@B22],[@B23]\]. HLA-A and HLA-B allow for efficient CD8^+^cytotoxic T lymphocyte (CTL) detection. Since it has been demonstrated that HLA-C and HLA-E are needed for protection from natural killer (NK) cell-mediated death \[[@B23]\], the up-regulation of HLA-C by Tat suggests similar host cell survival-directed functions for both Tat and Nef. Interestingly, HLA-G has been shown to be up-regulated in both monocytes and T lymphocytes of seropositive individuals, though its relation to infection and pathogenesis remains to be determined \[[@B24]\]. Collectively, SOM and k-means analyses catalog a set of genes representative of a close interplay between promoting and inhibiting factors induced by Tat. These findings, coupled with the up-regulation of signaling receptors involved in cell growth and survival, illustrate an intrinsic ability of HIV-1 Tat in regulating immune evasion, viral transmission, cell cycle progression and subsequent apoptosis. Importantly, these results delineate a variety of cellular gene products, both previously characterized with respect to HIV-1 and those uncharacterized, to be directly or indirectly induced by Tat expression. A plausible notion is that during activated transcription, HIV-1 hijacks the host cell machineries to promote its own replication, while concurrently directing a certain minimal level of cell survival until the virus reaches its critical point of progeny formation and subsequent virus-induced cell cycle block and apoptosis at the G~2~phase. siRNA-mediated validation of cellular HIV-1 therapeutic targets --------------------------------------------------------------- Using siRNAs targeted at several Tat-induced host cellular gene products, we examined the significance of our synchronized microarray data on a few genes we thought were critical for productive viral progeny formation. Based on the 32 arrays (16 eTat and 16 pCep4) in this study, we generated a list of Tat-induced genes that included those genes displaying two or more present calls on the eTat chips (present on at least 2 of 16 chips) while having 16 absent calls in the control pCep4 chips. We hypothesized that genes which were consistently (at various cell cycle phases) induced/repressed by Tat and were absent from the control pCep4 chips, would be the most important and specific for the Tat-mediated effects on the viral life cycle or host cell cycle progression. We also identified genes that displayed at least four and at least eight present calls across all 16 eTat chips and displayed all absent calls across all 16 pCep4 chips \[see [Additional File 4](#S4){ref-type="supplementary-material"} and Methods\]. Finally, the two present call gene list was screened against the Hu95 microarray data indexed at the Children\'s National Medical Center (CNMC) in Washington, D.C. This analysis was executed to identify those genes only induced by Tat, while never induced in a myriad of other human genetic diseases and tissues whose data is hosted at CNMC. Those genes that were 100% absent or 50.1% to 99.9% absent across all the Hu95 data in the database were compiled and listed (Table [2](#T2){ref-type="table"}). This list of genes has potential to be very specific cellular therapeutic targets. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Tat-upregulated genes not induced in other genetic diseases profiled. ::: Accession \# Fold Change Gene Name ------------------ ----------------- ------------------------------------- D13243 1.9 Pyruvate kinase L Z49194 4.1 Pou2AF1 (OBF-1) AF072099 3.1 LILRB4 U61836 0.2 SMOX J00117 10.8 CGB X02612 2.2 Cytochrome P(1)-450 (CYP1A1) Y12851 0.8 P2X7 receptor AI349593 0.6 Similar to hemoglobin epsilon chain AF055007 3.9 MARCH-III AB002449 3.9 Hypothetical gene AA203545 1.9 Unknown ::: Based on a literature search of our initial list of dysregulated genes (from the K-means, SOMs, and present call gene list analyses) and from the CNMC screen, we have a comprehensive list of potential targets. Through the exhaustive literature search, we looked for genes that were previously characterized as necessary for HIV-1 replication and/or progeny formation and identified HIV-1 Rev binding protein 2, Pou2AF1 (OBF-1), cyclin A1, PPGB, EXT2, and HEXA for further analysis. The HIV-1 Rev binding protein 2 has been characterized as having high homology to the S. cerevisiaeKrr1p protein, which is a nucleolar protein, and has been shown to be critical for 18S rRNA synthesis and subsequent 40S ribosome synthesis and cell viability \[[@B25]-[@B27]\]. Therefore, ablation of the HIV-1 Rev binding protein 2 should theoretically inhibit virus replication and possibly direct infected cells towards apoptosis. The HIV-1 LTR contains four potential binding sites for the Oct-1 transcription factor and Oct-1 has been shown to interact with Tat \[[@B28]\]. OBF-1 interacts with Oct-1 and Oct-2, acting as a B lymphocyte-specific transcriptional coactivator of B cell activation and maturation, as well as induction of immunoglobulins. It is also activated in T cells upon TCR signaling \[[@B29]\]. Recently, OBF-1 was found to up-regulate CCR5 co-receptor surface expression and fusion to the Env protein of R5 strains, the predominant strain found during initial infection \[[@B29]\]. Therefore, we predict that this factor is repressed upon the onset of AIDS, which is usually correlated with a R5 to X4 HIV-1 strain switch. Cyclin A1, which binds and regulates cdk2 and cdk1, was also chosen for targeted inhibition since it is important during the S and G~2~phases of the cell cycle, both of which are important for the viral life cycle \[[@B5],[@B30]\]. Cyclin A1 has also been shown to bind Rb family members, the p21/waf1 family of endogenous cdk inhibitors, as well as the E2F-1 transcription factor, all of which are important in the regulation of cell cycle progression and HIV-1 progeny formation \[[@B4],[@B6],[@B31]-[@B34]\]. Based on the importance of viral attachment, entry, and membrane fusion in the course of infection, we also chose to inhibit expression of the PPGB protein, which forms a heterotrimeric complex with the lysosomal enzymes β-galactosidase and neuraminidase (NA). Though there have been no reports on the contribution of PPGB in HIV-1 infection, a number of reports have illustrated the importance of NA in increasing the efficiency of viral binding and entry \[[@B35],[@B36]\]. NA is a sialidase that exposes sites on the HIV-1 gp120 surface protein, enabling greater interaction between gp120 and the CD4/co-receptor complex, which consequently increases syncytium formation and single-round infection by both X4 and R5 HIV-1 isolates. These findings coupled with the importance of HSPGs, illustrate the importance of membrane proteins and their modifications on both viral attachment and entry processes. Cellular proteins involved in the fusion and entry processes of infection may play a greater role in extracellular Tat-mediated effects, such as bystander cell infection. The EXT2 and HEXA gene products were also targeted since they displayed present calls in at least half of the eTat chips and showed no induction in the pCep4 chips \[see [Additional File 4](#S4){ref-type="supplementary-material"}\]. EXT2 is a putative tumor suppressor with glycosyltransferase activity that is involved in the chain elongation step of heparan sulfate biosynthesis \[[@B37]\]. HEXA is involved in ganglioside GM2 degradation and is a member of a subfamily of glycosyl hydrolases \[[@B38]\]. It has been established that GM2 levels are significantly increased in HIV-1 infection, as is seen both in vitroand in vivofrom seropositive individuals \[[@B39],[@B40]\]. Surprisingly, both groups showed that anti-GM2 IgM antibodies caused complement-mediated cytolysis of infected cells. We propose that inhibiting HEXA would increase the levels of circulating GM2 in vivo, thereby creating a more pronounced level of infected cell cytolysis. Using HIV-1 latently infected OM 10.1 T cells, which contain a single copy of silent full length wild type infectious provirus, we transfected 10 μg of each siRNA (2 for each representative gene) into cells. After 48 hrs, TNF-α was added for 2 hours to induce the latent virus and normal cell cycle progression. Samples were collected at 72 hrs post-TNF-α treatment and subjected to p24 Gag ELISA and western blot analysis. Cells that were not transfected with any siRNA were used as the negative control sample, while cdk2 and cdk9-targeted siRNAs served as positive controls. As seen in Figure [5A](#F5){ref-type="fig"}, the majority of siRNAs demonstrated some efficacy in inhibiting p24 expression. Ablation of EXT2 had a moderate effect (\~2 fold reduction), while the HEXA siRNA had a negligible effect (\<1 fold reduction). While the cdk2- and cdk9-mediated inhibition of HIV-1 replication was expected \[[@B41],[@B42]\], the potency of the other siRNAs were very dramatic. Interestingly, the most effective siRNAs were involved in cell cycle progression and/or transcription (cdk2, cdk9, cyclin A1, and OBF-1), RNA pathways (HIV-1 Rev binding protein 2), or membrane protein modification (PPGB). While EXT2 has been shown to be important in heparan sulfate synthesis, HSPGs are most important for cells that do not express large amounts of CD4, such as macrophages \[[@B20]\]. Thus, EXT2 degradation should only affect infection and replication in cells devoid of CD4. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Representative siRNA-directed inhibition of HIV-1 replication.(A) Using two candidate siRNAs per gene shown, each siRNA was transfected into HIV-1 latently infected OM-10.1 cells at mid log phase of growth. Following transfection, viral activation, and treatment, supernatants were collected and analyzed for p24 Gag expression by ELISA. The white crossed bars represent the first set of experiments, while the black bars represent the second run performed in an identical manner. (B) For Western blots, protein samples (one hundred micrograms of each extract) were separated on SDS-PAGE and then transferred to an Immobilon-P (polyvinylidene difluoride; Millipore) membrane and blocked with 5% fat-free milk (in TNE50/0.1% Nonidet P-40). Membranes were incubated overnight with various primary antibodies, and reactive complexes were developed with protein G-labeled ^125^I and visualized with a PhosphorImager scanner (Amersham Biosciences). ::: ![](1742-4690-2-20-5) ::: We also performed series of western blots to measure the efficiency of inhibition from each of siRNAs tested. As shown in Figure [5B](#F5){ref-type="fig"} most siRNA treatments dropped the protein level by more than 90%, except for the HEXA gene. None of siRNAs inhibited actin gene expression or PARP degradation (an indicator of active apoptosis), implying that the siRNA targets were not toxic in these transient experiments. We finally performed simple FACS analysis using PI staining and saw no apparent cell cycle or apoptotic effects (Figure [6](#F6){ref-type="fig"}). Although, we have never been able to inhibit HEXA translation completely in OM10.1 cells (or three other infected cell lines), data on HEXA indicates that even a 50% drop in protein levels maybe sufficient to increase GM2 levels, thereby increasing a more pronounced rate of viral production. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### FACS analysis of PI stained OM10.1 cells. The stained cells were analyzed for red fluorescence (FL2) on a FACScan (Becton Dickinson, San Jose, CA), and cell distribution in the G~1~, S, and G~2~/M phases of the cell cycle was calculated from the resulting DNA histogram with Cell FIT software, based on a rectangular S-phase model. A sub-G1 population was considered as an apoptotic population. ::: ![](1742-4690-2-20-6) ::: Next, we performed a similar set of experiments in PBMCs infected with a HIV-1 field isolate and treatment with various siRNAs. Activated PBMCs were first treated with 10 μg of each siRNA for 48 hours and subsequently infected with a field HIV-1 isolate (UG/92/029 Uganda strain, sub-type A envelope). Supernatants were collected every six days for Gag p24 assay. Results in Figure [7A](#F7){ref-type="fig"} indicate that siRNA\'s against cdk9, cdk2, HEXA, and Rev-BP2 were the most potent inhibitors, followed by siRNAs against cyclin A, OBF-1 and PPGB, and the least amount of inhibition with EXT-2 siRNA. Control experiments using antibody staining against CD4 on activated PBMCs treated with each siRNA for 48 hours prior to HIV-1 infection showed no appreciable differences, except a minor drop with cdk2 siRNA (\~5%) in CD4 levels (Figure [7B](#F7){ref-type="fig"}), and a PI staining of the same cells also showed no significant apoptosis except for a minor drop with cyclin A siRNA (\~5%, Figure [7C](#F7){ref-type="fig"}), implying that the siRNA treatment in general did not significantly alter the expression of CD4 levels prior to viral infection. Collectively, these results are somewhat similar to the latent OM10.1 treatments and imply that these genes could be a potential target in both cell lines and primary infections. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Effect of representative siRNA treatment in PBMC field isolate HIV-1 infection.Approximately 5 × 10^6^Phytohemagglutinin-activated PBMC were kept in culture for two days prior to infection. PBMC were first treated for 48 hrs with 10 μg of the various siRNAs and then infected with SI (UG/92/029 Uganda strain, subtype A envelope, 5 ng of p24 gag antigen) strain of HIV-1 obtained from the National Institutes of Health (NIH) AIDS Research and Reference Reagent Program. After 8 h of infection, cells were washed and fresh media was added. Samples were collected every sixth day and stored at -20°C for p24 gag enzyme-linked immunosorbent assay (ELISA). Media from infected cell lines was centrifuged to pellet the cells and supernatants were collected and diluted to 1:100 to 1:1,000 in RPMI 1640 prior to analysis. Supernatants from the infected PBMC were collected and used directly for the p24 antigen assay. The p24 gag antigen level was analyzed using the HIVAG-1 Monoclonal Antibody Kit (Abbott Laboratories, Diagnostics Division). (B) PBMCs stimulated with PHA were treated with appropriate siRNA prior to HIV infection and stained for presence of surface CD4 on activated cells. Prior to infection, 1/5 of the samples were processed for CD4 and PI staining. Cells were then collected and washed twice with PBS containing FCS and NaN~3~. Cells were stained on ice for with human tri-color-labeled anti-CD4 (Catalog Laboratories) at a 1:10 dilution. Stained cells were next washed two times in PBS containing FCS and NaN~3~and fixed in paraformaldehyde followed by analysis by FACS. (C) FACS analysis of PI stained cells from panel B. Sub-G1 population was scored as apoptotic population in each siRNA treated cell. ::: ![](1742-4690-2-20-7) ::: Finally, we asked whether the identified gene lists from our siRNA experiments were specific to HIV-1 transcription or could they also inhibit other viral activated transcriptions. We therefore performed CAT assays with either HIV-LTR-CAT and its activator Tat (as positive controls, Figure [8](#F8){ref-type="fig"}, Lanes 1--3) or HTLV-LTR-CAT and its positive activator Tax (Figure [8](#F8){ref-type="fig"}, lanes 4--14). Results in Figure [8](#F8){ref-type="fig"} show that HIV-1 activated Tat can be suppressed with cdk2, however none of the siRNA treatments inhibited HTLV-1 Tax activated transcription except cdk9 siRNA. This result is somewhat expected since cdk9 is known to be involved in general transcription elongation, and is consistent with a recent report indicating that Tax might have a role in transcription elongation \[[@B43],[@B44]\]. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### CAT assays with HIV-LTR-CAT and its activator Tat, and HTLV-LTR-CAT and its positive activator Tax.Lymphocyte (CEM, 12D7) cells were grown to mid log phase and were processed for electroporation according to a procedure published previously \[52\]. The cells were washed with phosphate-buffered saline and resuspended in RPMI 1640. They were next transfected with reporter constructs (HIV-LTR-CAT or HTLV-LTR-CAT; 3 ug of each), their respective activators (Tat or Tax; 4 ug each) or with various siRNAs (10 ug each). Lanes 1--3 serve as positive controls for basal, activated transcription and effect of cdk2 siRNA on inhibition of HIV-1 LTR. Lanes 4--14 are basal, activated transcription and effect of various siRNAs on HTLV- LTR-CAT. Only cdk9 siRNA showed an appreciable amount of suppression on Tax activated HTLV-LTR (lane 8). CAT % conversations are listed below the diagram. ::: ![](1742-4690-2-20-8) ::: Conclusion ========== Potential therapeutic targets of HIV-1 Tat-induced cellular genes ----------------------------------------------------------------- We believe that our current results are by no means the ultimate list of genes altered by HIV-1 Tat. Some of the limitations of our experiments include: constant presence of Tat in cells as compared to possible transient expression of Tat in HIV-1 infected cells, possible indirect effect of Tat on gene expression, and lack of using various Tat clades (i.e., from clades B, E, and C), which may have a different rate and set of activated genes in vivo. However, we believe the current study is an ongoing attempt to narrow down which cellular genes are critical in Tat regulation and therefore define a minimal set of potential targets for therapy. Based on exhaustive and stringent data analysis, we have compiled a list of gene products that may serve as potential therapeutic targets for the inhibition of HIV-1 replication (Table [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). Table [1](#T1){ref-type="table"} specifies Tat-induced cellular genes at the G~1~/S transition, while Table [2](#T2){ref-type="table"} lists those genes that were observed to be up-regulated by Tat while displaying no induction in the myriad of genetic diseases and diverse tissues and cell types screened at CNMC. As observed in both tables and the initial screening of genes displaying at least two present calls, several genes have been established as important for HIV-1 infection and replication, including OBF-1 \[[@B29],[@B45]\], complement factor H related 3 \[[@B46]\], CD4 receptor, ICAM-1 \[[@B18]\], NA \[[@B35],[@B36]\], and cyclin A1 \[[@B8],[@B47]\]. There were also several genes that have not been published in relation to HIV-1 infection and may also be novel and efficacious therapeutics. These include FGFR and EGFR, the latter of which has been targeted against various cancers and inhibits cancer-associated angiogenesis and subsequent metastasis \[[@B48]\]. Concerning HIV-1 infection and replication, some potentially important proteins that have not been previously characterized with respect to HIV-1 and thus necessitate further study, seem to be the CAP-binding protein complex interacting protein, tropomyosin 2 beta, BTG3, the IL-10R, PPGB, and cathepsins Z and L2 \[see [Additional File 4](#S4){ref-type="supplementary-material"} and Tables [1](#T1){ref-type="table"} &[2](#T2){ref-type="table"}\]. Though not established, the CAP-binding protein complex is most likely involved in translation processes. Tropomyosin 2 beta was found to interact with FRP1, which is important in the regulation of HIV-1 virus-mediated cell fusion and possibly syncytium formation \[[@B49]\]. Also, therapeutics against individual gene products or a cocktail containing inhibitors for ICAM-1, LFA-3, DC-SIGN, all syndecan isoforms, PPGB, clusterin and other adhesion/membrane proteins important in viral transmission may, alone or in combination with Fuzeon/T20, significantly abrogate the infection of circulating lymphocytes and other cells that are able to support viral infection and replication. Recently a report by Krishnan and Zeichner described experiments associated with changes in cellular gene expression that accompany the reactivation of the lytic viral cycle in cell lines chronically infected with HIV-1. They found that several genes exhibited altered expression in the chronically infected cells compared to the uninfected parental cells prior to induction into lytic replication including genes encoding proteasomes, histone deacetylases, and many transcription factors \[[@B50]\]. Although it is difficult for us to compare our results with Krishnan and Zeichner due to difference in cell types, presence of all HIV-1 ORFs as compared to our study where there was only Tat present, and the difference in cell cycle stages, however, we did a general comparison and found some overlap between our list of dysregulated genes and theirs -- this overlap includes genes coding for splicing factors, proteasomes, and heat shock proteins. We compared our SOM and k-means analyses (Table [1](#T1){ref-type="table"}) from which we found genes that displayed differential expression at the G1/S phase and found three intersecting genes as well as some genes that are very closely related to genes listed in the Krishnan table (e.g. genes coding for a different subunit of a protein); these genes are listed in Table [3](#T3){ref-type="table"}. The first part of Table [3](#T3){ref-type="table"} contains three genes that fell in both our SOM and k-means analyses and the Krishnan table (bold genes) and the genes from our SOM and k-means analyses that are closely related to genes in the Krishnan table. Collectively, the list of common genes indicates the involvement of HIV-1 Tat in splicing, transport of RNA, an acceleration of cell cycle stages. All of these genes fall into pathways that have previously been reported to be regulated by Tat, including stabilization of critical transcription units (i.e., Hsp70 stabilization of Cdk9/cyclin T1 complex), splicing and nuclear transport (i.e., the SR protein ASF/SF2; Tat-SF1), translation (5\'-terminal TAR recognition by eukaryotic translation initiation factor 2), and degradation of critical factors needed for cell cycle progression using the proteosome pathway (i.e., analogous to HPV E6 binding to p53 and its degradation resulting in loss of check point, ubiquitin/proteasome degradation of IkappaB(alpha) and release of active NFkB, or CD4 glycoprotein degradation through the ubiquitin/proteasome pathway). Therefore these results imply that Tat regulates these apparently discrete pathways, at least in case of pre-mRNA processing, where transcription initiation/early elongation complex directly controls every aspect of subsequent pre-mRNA processing including capping at the 5\' end, intron recognition and removal by splicing, the 3\' end cleavage and polyadenylation, and release of the mature mRNA from the site of transcription and export to the cytoplasm for translation \[[@B51]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### A set of common genes regulated by Tat in both Tat expressing cells and HIV-1 infected cells. ::: Probe Set ID Accession \# Gene Description ------------------ ------------------ ------------------------------------------------------------------------- 34083\_at AA311181 splicing factor, arginine/serine-rich 9 35323\_at U78525 eukaryotic translation initiation factor 3, subunit 9 (eta, 116 kD) 31858\_at X07315 nuclear transport factor 2 32165\_at L41887 splicing factor, arginine/serine-rich 7 (35 kD) 32556\_at X64044 U2 (RNU2) small nuclear RNA auxiliary factor 2 33372\_at AI189226 RAB31, member RAS oncogene family 39628\_at AI671547 RAB9A, member RAS oncogene family 2029\_at N36267 Rho GTPase activating protein 5 35255\_at AF098799 RAN binding protein 7 1191\_s\_at AB003102 proteasome (prosome, macropain) 26S subunit, non-ATPase, 11 1192\_at AB003103 proteasome (prosome, macropain) 26S subunit, non-ATPase, 12 37350\_at AL031177 proteasome (prosome, macropain) 26S subunit, non-ATPase, 10 1104\_s\_at M11717 heat shock 70 kD protein 1A 36614\_at X87949 heat shock 70 kD protein 5 (glucose-regulated protein, 78 kD) 35467\_g\_at W73046 DnaJ (Hsp40) homolog, subfamily B, member 12 ::: While some of these proteins have available inhibitors, the majority of the potential cellular targets for HIV-1 therapeutics do not have known specific inhibitors. Thus, much effort must be allocated for the elucidation and design of specific inhibitors, concurrent with the growing plausibility of siRNA-based therapeutics. Another important factor in designing inhibitors for cellular targets, as shown with potential cancer therapeutics, is the necessity to target cellular gene products with redundant functions. If a certain cellular protein kinase, receptor, membrane protein, or cytokine/chemokine is inhibited, it may have adverse effects that make the drug impractical for clinical trials and use. However, the presence of two or more proteins with similar functions, with only one being critical for HIV-1 and thus targeted, may allow for the decreased possibility of side effects. This is especially true for targeting redundant molecules (i.e., cdk2), where they are nonessential during mammalian development and are likely replaced by other kinases. Similarly, once specific inhibitors are elucidated, a major resulting challenge is generating a combinatorial therapeutic regimen that is effective in sub-lethal doses (submicromolar or nanomolar range). Methods ======= Cell culture ------------ HeLa CD4^+^cells containing either an epitope-tagged (the influenza epitope at the C terminus of Tat 1--86) eTat plasmid or the parental control vector pCep4 were used \[[@B2]\]. All cells were cultured in RPMI 1640 containing 10% fetal bovine serum, 1% streptomycin/penicillin, and 1% L-glutamine (Quality Biological) at 37°C in 5% CO~2~. Cytoplasmic RNA isolation ------------------------- Cells were centrifuged at 4°C, 3000 rpm for 10 min., quickly washed with D-PBS without Ca^2+^/Mg^2+^, and centrifuged twice. Pelleted cells were immediately frozen at -80°C until all time points were collected. Cytoplasmic RNA was isolated utilizing the RNeasy Mini Kit (Qiagen, Valencia, CA) according to manufacturer\'s directions with the addition of 1 mM dithiothreitol in Buffer RLN. Isolated RNA was quantitated by UV spectrophotometric analysis and 3 μg of RNA was visualized on a non-denaturing 1% agarose TAE gel for quality and quantity control. Lymphocyte Transfection ----------------------- Lymphocyte (CEM, 12D7) cells were grown to mid log phase and were processed for electroporation according to a procedure published previously \[[@B52]\]. The cells were centrifuged and then washed with phosphate-buffered saline without Mg2+ or Ca2+ twice and resuspended in RPMI 1640 at 4 × 10^5^cell/0.25 ml. The CEM cells (0.25 ml) were transfected with the plasmid DNAs of HIV-LTR-CAT or HTLV-LTR-CAT (3 ug of each) either alone or in combination with Tat or Tax (4ug each). 10 μg of the various siRNAs were also mixed in with reporter and/or appropriate transactivator prior to electroporation. The mixture of cells, plasmid DNAs, and siRNAs were then transferred to a cuvette and electroporated with fast charge rate, at 230 V, and capacitance of 800 microfarads. Cells were then plated in 10 ml of complete RPMI 1640 medium for 18 h prior to harvest and CAT assay. For CAT assays, standard reaction was performed by adding the cofactor coenzyme A to a microcentrifuge tube containing cell extract and radiolabeled chloramphenicol, in a final volume of 50 μl and incubated at 37°C for 1 h. The reaction mixture was then extracted with ethyl acetate. It was then separated by TLC on silica gel plates (Baker-flex silica gel TLC plates) using the chloroform:methanol (19:1) solvent system. The resolved reaction products were then detected by exposing the plate to a PhosphorImager cassette. Immunoprecipitation/Western Blot Analysis ----------------------------------------- Immunoprecipitations of tat protein were performed as described previously \[[@B2]\]. Cellular protein (100 μg) was mixed with monoclonal 12CA5 antibody (2.5 μg) for 2 h at 4°C. Protein A + G agarose beads (5 μl; Calbiochem, Inc.) were added and incubated at 4°C for another 2 h. The immunoprecipitated complex was then spun down and washed with buffer D containing 500 mM KCl (three times; 1 ml each). Samples were eluted with HA- peptide for 4 hrs at 37 C on a rotator, and eluted complexes were separated on a 4--20% SDS-polyacrylamide gel electrophoresis gel, and Western blot analysis was performed with anti-Tat monoclonal antibody. Antigen/antibody complexes were detected with ^125^I Protein G. CD4 staining of human cells --------------------------- Human PBMCs stimulated with PHA were treated with appropriate siRNA prior to HIV infection. Activated PBMCs were first treated with 10 μg of each siRNA for 48 hours and subsequently infected with a field HIV-1 isolate (UG/92/029 Uganda strain, subtype A envelope, 5 ng of p24 gag antigen) \[[@B53]\]. Prior to infection, 1/5 of the samples were processed for CD4 and PI staining. Cells were then collected and washed twice with PBS containing 5% FCS and 0.05% NaN~3~. Cells were stained on ice for 30 minutes with human tri-color-labeled anti-CD4 (Catalog Laboratories) at a 1:10 dilution. Stained cells were next washed two times in PBS containing 5% FCS and 0.05% NaN~3~and fixed in 1% paraformaldehyde followed by analysis by FACS. Cell cycle analysis ------------------- The eTat and pCep4 cells were either blocked with hydroxyurea (G~1~/S blocker, 2 mM) or nocodazole (G~2~/M blocker, 50 ng/ml). Cells were washed with PBS and released with complete medium. Samples were collected every 3 hrs and cytoplasmic RNA was isolated. Single-color flow cytometric analysis of DNA content (PI staining) was performed on both cell types \[[@B2]\]. Stained cells (including OM10.1) were analyzed for red fluorescence (FL2) on a FACScan (Becton Dickinson, San Jose, CA), and cell distribution in the G~1~, S, and G~2~/M phases of the cell cycle was calculated from the resulting DNA histogram with Cell FIT software, based on a rectangular S-phase model. PBMC infection -------------- Phytohemagglutinin-activated PBMC were kept in culture for two days prior to each infection. Isolation and treatment of PBMC were performed by following the guidelines of the Centers for Disease Control. Approximately 5 × 10^6^PBMC were first treated for 48 hrs with 10 μg of the various siRNAs and then infected with SI (UG/92/029 Uganda strain, subtype A envelope, 5 ng of p24 gag antigen) strain of HIV-1 obtained from the National Institutes of Health (NIH) AIDS Research and Reference Reagent Program. After 8 h of infection, cells were washed and fresh media was added. Samples were collected every sixth day and stored at -20°C for p24 gag enzyme-linked immunosorbent assay (ELISA). For HIV-1 p24 ELISA, media from infected cell lines was centrifuged to pellet the cells and supernatants were collected and diluted to 1:100 to 1:1,000 in RPMI 1640 prior to analysis. Supernatants from the infected PBMC were collected and used directly for the p24 antigen assay. The p24 gag antigen level was analyzed using the HIVAG-1 Monoclonal Antibody Kit (Abbott Laboratories, Diagnostics Division). siRNA analysis -------------- siRNA sequences were designed using the Oligoengine Workstation <http://www.oligoengine.com> and were purchased from Qiagen-Xeragon. Candidate sequences were chosen based on general siRNA design criteria, including a %GC content between 45--55 % and avoiding more than three consecutive guanosines. Selected target sequences were also BLASTed <http://www.ncbi.nlm.nih.gov/BLAST/> with a standard nucleotide-nucleotide BLAST to ensure they were not homologous to other genes. Each candidate siRNA was generated from the 5\' end and consisted of 19 nucleotides with a d(TT) overhang. The following genes were chosen for siRNA analysis with the GenBank accession numbers in brackets: HIV-1 Rev-binding protein 2 \[U00943\], Pou2AF1 (OBF1) \[Z49194\], cyclin A1 \[U66838\], PPGB \[NM\_000308\], cdk2 \[AF512553\], cdk9 \[AF517840\], EXT2 \[U67368\], and HEXA \[M16424\]. 2 candidate siRNAs were chosen for each of the 8 genes to ensure protein expression silencing. For each duplex siRNA, the first sequence represents the sense sequence (\"s\"), and the second, the antisense sequence (\"as\"): ### HIV-1 Rev-binding protein 2 1\. s: GGUCCAAUGGCUGAAACUG, as: CAGUUUCAGCCAUUGGACC 2\. s: ACAGUCAUGCUGCCUUCGA, as: UCGAAGGCAGCAUGACUGU ### Pou2AF1 (OBF-1) 1\. s: GAGGAUAGCGACGCCUAUG, as: CAUAGGCGUCGCUAUCCUC 2\. s: UGUCACGACAAGAAGCUCC, as: GGAGCUUCUUGUCGUGACA ### Cyclin A1 1\. s: ACUGCAGCUCGUAGGAACA, as: UGUUCCUACGAGCUGCAGU 2\. s: GUAGACACCGGCACACUCA, as: UGAGUGUGCCGGUGUCUAC ### PPGB 1\. s: CUAAUGACACUGAGGUCGC, as: GCGACCUCAGUGUCAUUAG 2\. s: UGCGUGACCAAUCUUCAGG, as: CCUGAAGAUUGGUCACGCA ### Cdk2 1\. s: AUCCGCCUGGACACUGAGA, as: UCUCAGUGUCCAGGCGGAU 2\. s: UCCUCCUGGGCUGCAAAUA, as: UAUUUGCAGCCCAGGAGGA ### Cdk9 1\. s: CCACGACUUCUUCUGGUCC, as: GGACCAGAAGAAGUCGUGG 2\. s: CCGCUGCAAGGGUAGUAUA, as: UAUACUACCCUUGCAGCGG ### EXT2 1\. s: GCACCUCGAGCUAUGCAAC, as: GUUGCAUAGCUCGAGGUGC 2\. s: CUCCGUCUUUGGCCUGACA, as: UGUCAGGCCAAAGACGGAG ### HEXA 1\. s: CCUGGUCACAAAAGAGCCU, as: AGGCUCUUUUGUGACCAGG 2\. s: GUGUGAAUGGCGUUAGGGU, as: ACCCUAACGCCAUUCACAC HIV-1 latently infected OM-10.1 T lymphocytes were treated with 10 μg of the various siRNAs listed above for 48 hrs prior to TNF-α treatment. siRNAs were electroporated into OM-10.1 cells at 5 × 10^6^(mid log phase of growth) cells/ml. 48 hrs later cells were treated with TNF-α (5 μg/ml for 2 hrs) to induce viral transcription and progeny formation, washed, and complete media was added to cells. Samples were collected at 72 hrs post-TNF-α treatment for presence of HIV-1 p24 Gag by ELISA. Presence of p24 Gag in the supernatant is indicative of mature infectious virion particles released from HIV-1 infected cells. Expression profiling -------------------- Six μg of cytoplasmic RNA from each sample were converted to double-stranded cDNA using the Superscript Choice System kit and T7-(dT)24 primer (100 pmol/μL) (Invitrogen). The cDNA was cleaned and purified using phenol/chloroform extraction and ethanol precipitation. The cDNA was then used to perform in vitrotranscription using the BioArray HighYield RNA Transcript Labeling Kit (T7) (Enzo, Farmingdale, NY). The biotin-labeled cRNA was cleaned using the RNeasy Mini Kit (Qiagen) and was quantified by spectrophotometric analysis and analyzed on a 1% agarose TAE gel. The biotin-labeled cRNA was then randomly fragmented to \~35--200 base pairs by metal-induced hydrolysis using a fragmentation buffer according to the Affymetrix Eukaryotic Target Hybridization protocol. The Human U95Av2 microarrays (Affymetrix) were washed, primed, and stained on the Affymetrix Fluidics Station 400 following the Affymetrix protocol. cRNA was first detected through a primary scan with phycoerythrin-streptavidin staining and then amplified with a second stain using biotin-labeled anti-streptavidin antibody and a subsequent phycoerythrin-streptavidin stain. The emitted fluorescence was scanned using the Hewlett-Packard G2500A Gene Array Scanner, and the intensities were extracted from the chips using Microarray Suite 4.0 (MAS4.0) software. All raw chip data was scaled in MAS4.0 to 800 to normalize signal intensities for inter-array comparisons. A statistical algorithm in MAS4.0 assigns present, marginal, and absent calls based on probe pair intensities where one probe is a perfect match of a reference sequence and the other is a mismatch probe that has a single base change at the 13th position within the 25-base oligonucleotide reference sequence. Quality Control --------------- Report files generated by MAS4.0 were reviewed to ensure all quality control standards were met -- these include percentage of present calls, presence of spike controls, signal scaling factors per chip, and the GAPDH 3\'/5\' ratios. All raw data files containing the signal and detection values for each probe set and supplemental data files are posted on a Translational Genomics (TGen) data site, <http://www.tgen.org/research/index.cfm?pageid=142>, as well as on the Gene Expression Omnibus (GEO) online repository <http://www.ncbi.nlm.nih.gov/geo> as identified by GEO accession number \[see [Additional File 1](#S1){ref-type="supplementary-material"}\]. Data analysis ------------- Comparative analyses were performed in MAS4.0 between replicate samples to determine gene expression behavior changes between every sample set; calls assigned by MAS4.0 can be either increase, marginally increase, decrease, marginally decrease, or no change. Comprehensive microarray data analysis was performed using GeneSpring software (v4.2; Silicon Genetics, Redwood City, CA). Using the synchronized cell cycle data, a gene list was generated by filtering for genes that had (1) a minimum of 2 present calls (detection as determined by MAS4.0) out of a total of 32 calls (1 call per chip), (2) a maximum p-value of 0.05 where, in this case, the p-value represents the probability that the signal intensity for a gene is due to chance alone, and (3) a greater than 2-fold expression change between control pCep4 samples and respective eTat samples. To divide the genes in this list into groups based on similar expression patterns through the cell cycle, k-means clustering (of 15 clusters as selected based on Genespring\'s expressed validity value) was applied and gene lists for each cluster were consolidated \[see Additional Files [3](#S3){ref-type="supplementary-material"} and [7](#S7){ref-type="supplementary-material"}\]. A complementary analysis was also performed using SOMs \[[@B54]\]. The input gene list for this analysis was generated using several filters against the entire list of probe sets, which represent the gene transcripts on the U95Av2 array: (1) filter for at least 2 present calls, (2) any probe sets that generated an absent call across all cell cycle time points were eliminated, (3) any probe sets that did not have three out of four marginal increase or increase calls, or marginal decrease or decrease calls in at least one of the eight cell cycle time points, were removed (based on comparative analyses generated by MAS4.0) to control for replicate consistency. The signal log ratio of each gene in the resulting list was calculated (using the two replicate eTat samples and 2 replicate pCep4 samples per time point for each gene): ![](1742-4690-2-20-i1.gif) Three sets of gene lists were created based on 3 separate filtering rules: \(1) 0.666 \< ratio \< 1.500 \(2) 0.500 \< ratio \< 2.000 \(3) 0.333 \< ratio \< 3.000 For a single rule, if a gene had average signal ratios at every time point that fell within the specified boundary, the gene was removed from the list. Separate gene lists were generated for each rule. For the first rule, 464 genes were removed and 2330 genes were used for clustering; the second rule, 1644 genes were removed and 1150 were used for analysis; and for the third rule, 2415 genes were eliminated and 379 were used for clustering. The gene ratios in each of the three lists were log transformed (natural base), median centered, applied to separate SOMs, and visualized using the U-matrix and component planes representation \[for each SOM see Additional Files [5](#S5){ref-type="supplementary-material"} and [6](#S6){ref-type="supplementary-material"}, and Figure [4](#F4){ref-type="fig"}, respectively\] \[[@B54],[@B55]\]. The algorithm incorporates a batch learning algorithm with Euclidean distance, and all computations were performed using MATLAB (The MathWorks) with the SOM-toolbox with parameters set to defaults as described \[[@B56]\]. Defined groups of neurons that displayed expression differences from one time point to the next in the component planes representation, as well as clusters appearing in the U-matrix were noted. Neurons in the same position across the component planes contain the same genes; thus, coloring of the neurons allows for direct interpretation of the differences in expression levels between time points. Gene lists corresponding to the first and third filters were consolidated \[see [Additional File 2](#S1){ref-type="supplementary-material"}\]. The original gene list of synchronized sample data was also filtered for those genes that had all absent calls in the control cells and at least 2 present calls in the experimental cells. The resulting gene list was surveyed against 540 Affymetrix Hu95 chips whose data is hosted at the Children\'s National Medical Center (CNMC) in Washington, D.C. <http://microarray.cnmcresearch.org>. These human data include all control and experimental data produced from the study of different genetic diseases in a variety of human tissues and cultured cells. Those genes from our gene lists that were 100% absent or 50.1% to 99.9% absent across all Hu95 data in the database were compiled and noted to provide an estimate of the drug target specificity. Gene classification/ontologies ------------------------------ Genes were classified as functionally relevant to HIV-1 after exhaustive literature review of publications indexed on the Entrez PubMed website. Affymetrix probe set identifiers from the increasing and decreasing expression lists were queried on the Affymetrix website <http://www.affymetrix.com> using the NetAffx analysis tool to determine gene names and functions. The genes in the resulting lists were classified into ontologies to show the genes having increased or decreased expression (organized based on their respective functions). For the gene ontology for the entire human U95Av2 genechip, ontology lists specific to the classifications available on Genespring v5.0.3 were first obtained. The remaining classifications were queried on the Affymetrix website with the NetAffx tool <http://www.affymetrix.com/analysis/index.affx>. Abbreviations ============= HIV-human immunodeficiency virus PBMC-peripheral blood mononuclear cells HAART-highly active retroviral therapy NNRTI-non-nucleoside reverse transcriptase inhibitor NRTI-nucleoside reverse transcriptase inhibitor TAR-transactivation response Hu-hydroxyurea Noco-nocodazole CNMC-Children\'s National Medical Center NA-neuraminidase FGFR-fibroblast growth factor receptor EGFR-epidermal growth factor receptor ELISA-enzyme-linked immunosorbent assay Competing Interests =================== The author(s) declare that they have no competing interests. Authors\' Contributions ======================= WSL performed the data analyses and helped to draft the manuscript. AM, and EA performed the siRNA experiments, coordinated data analysis, and helped to draft the manuscript. TT performed the expression profiling protocol on all samples. CdlF isolated RNA and contributed to the expression profiling experiment. KK, CdlF, and SD helped with the gene expression profiling, westerns, and FACS. SH ran the self-organizing map analyses. AP provided some of the supervision for the manuscript and support for the Kashanchi lab members. DAS coordinated the expression profiling and analytical methodology. FK participated in the design, coordination, and validation of the study. DAS and FK funded the studies. All authors have read and approved the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 5 Self-organizing map (SOM) for filter 1 (refer to Methods) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 Self-organizing map (SOM) for filter 2 (refer to Methods) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 K-means clustering (15 graphs and corresponding close-ups shown) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 Gene lists filtered for all absent in pCep4 samples and at least 2 present calls in eTat samples (Excel worksheet, \"2P\"), 4 present calls in eTat samples (Excel worksheet, \"4P\"), and 8 present calls in eTat samples (Excel spreadsheet, \"8P\") ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 1 GEO accession numbers for each sample ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 K-means clustering gene lists (three Excel worksheets, \"set1--5,\" \"set6--10,\" \"set11--15\") ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Self-organizing map (SOM) gene lists for the first and third filters (two Excel worksheets, \"HIV\_SOM\_Filt\_1a\" & \"HIV\_SOM\_Filt\_3a\") ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was supported by grants from the George Washington University REF funds to A. Vertes and F. Kashanchi, NIH grants AI44357, AI43894 and 13969 to F.K, and grant 1U24NS043571-01 for the NINDS/NIMH Microarray Consortium. A.M. and W.S.L. contributed equally to this work. F.K. and D.A.S. share senior authorship on this work.	PubMed Central	2024-06-05T03:55:56.210386	2005-3-21	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087880/", "journal": "Retrovirology. 2005 Mar 21; 2:20", "authors": [ { "first": "Winnie S", "last": "Liang" }, { "first": "Anil", "last": "Maddukuri" }, { "first": "Tanya M", "last": "Teslovich" }, { "first": "Cynthia", "last": "de la Fuente" }, { "first": "Emmanuel", "last": "Agbottah" }, { "first": "Shabnam", "last": "Dadgar" }, { "first": "Kylene", "last": "Kehn" }, { "first": "Sampsa", "last": "Hautaniemi" }, { "first": "Anne", "last": "Pumfery" }, { "first": "Dietrich A", "last": "Stephan" }, { "first": "Fatah", "last": "Kashanchi" } ] }
PMC1087881	Background ========== The Dominique Dormont Conferences provide an international forum for the promotion of exchanges between clinicians and fundamental scientists, including team leaders and young researchers, involved in interdisciplinary research on chronic infections. They provide an occasion for researchers with common interests to get together for two or three days of synthesis and intense discussion on the most recent advances in their field, to crystallize new research directions and collaborations. Contacts with young scientists are strongly encouraged during the conference and such exchanges are facilitated by limiting attendance at each conference to 150 participants, on a first-come, first-served basis. The participants include prestigious invited speakers for state-of-the-art introductions to each scientific session, followed by abstract-driven talks, most presented by young investigators. Abstract-driven poster sessions also provide a space for scientific exchanges between team leaders and young researchers. An International Scientific Program Committee establishes the final program and abstracts are selected according to the highest scientific standards. The Dominique Dormont International Conferences will be held every year on different specific thematic topics related to \"Host-Pathogen Interactions in Chronic Infections\". In December 2004, the topic of the first conference was \"Viral and Host Determinants of HCV, HCMV and HIV Infections\", considering the pathogenesis of these chronic viral infections as a priority topic for the development of interdisciplinary research and collaborations between clinicians and fundamental scientists worldwide. The International Scientific Program Committee selected original presentations on six topics of particular interest in the field. Over the course of two days, stimulating exchanges and discussions between team leaders and young researchers, highlighting key issues in this research field, were achieved, raising hopes of opening the way to further international investigations on interactions between host and viral determinants. Summary of the scientific sessions ================================== Session 1: Receptor and viral entry ----------------------------------- *Chairs:F. Arenzana and U. Koszinowski; Keynote Lecture:\"New aspects on hepatitis C virus replication, assembly and entry into host cells\" by R. Bartenschlager. ### Factors enhancing viral entry Viral entry depends in part on expression of the corresponding cellular receptors. The CD4 molecule has been identified as the receptor for HIV, whereas the identity of the receptors for HCV and HCMV is far from clear. Viruses may also use alternative receptors (e.g. CXCR4 for some HIV strains) or different receptors in different cell types (e.g. epithelial GalCer for HIV). Furthermore, additional molecules (coreceptors) may be required for entry (such as chemokine receptors for HIV, CD13 for HCMV and the scavenger receptor B1 for HCV). Finally, as recently shown, the efficiency of viral entry is conditioned by multiple cellular factors that may enhance infection in cisor in trans. Bartenschlager et al. presented a review of recent findings on HCV replication. Bartosch et al. then discussed their observation that human serum components increase the infectivity of HCV pseudoparticles. These components are primate-specific as sera from chimpanzees and rhesus monkeys increase HCV infectivity whereas sera from rabbits and cows do not. The enhancement of infectivity by serum components has been described for other viruses (Ebola virus). However, for HCV, this effect is not mediated by immunoglobulins or complement, but is instead due to interplay between the HVR1 region of the HCV E2 glycoprotein, SR-B1 and serum HDL. Coating HCV with such serum lipoproteins may also help the virus to evade neutralizing antibodies. Bomsel et al. discussed their observation that HIV-1 transcytosis is more efficient with infected cells than with cell-free virus. They showed that the transcytosis induced by HIV-1-infected cells involves the adhesion-mediating RGD-containing protein. They also showed that the scaffolding protein HSPG Agrin is expressed at the apical epithelial surface and acts as an attachment factor for HIV-1, by interacting with the gp41 P1 lectin site, synergizing binding to the epithelial receptor galactosyl ceramide. GalCer is also expressed in immature dendritic cells (iDCs) and may mediate the internalization of HIV and its transfer to CD4+ cells in a lipid raft-dependent manner, independently of DC-SIGN. The role of DC-SIGN in viral retention and enhancement is not fully understood and depends on the cell line studied. The data presented by Nobile et al. suggest that HIV-1 X4 viruses can replicate in iDC and Raji-DC-SIGN cells, but only covertly and slowly. They show that transfer from iDC or DC-SIGN+ to CD4+ T cells occurs during the first few hours after exposure to virus, whereas only replicating viruses (i.e. not single-cycle virions) are transmitted several days after exposure, suggesting that long-term transmission is associated with replication in DC rather than with the retention of infectious particles through DC-SIGN. ### The fusion step of entry After interaction of the HIV-1 envelope surface glycoprotein with host cell receptors and coreceptors, the fusogenic transmembrane glycoprotein undergoes a series of conformational changes that lead to the insertion of the fusion peptide into the membrane of the target cells, bring the viral and cellular membranes in close proximity, and finally triggering virus-cell membrane fusion. Each of these steps provides an opportunity to prevent viral entry. Several agents targeting viral entry are currently being developed, and one -- T-20 or Efarvirtide -- has been approved by the US Food and Drug Administration for use as an antiretroviral drug. Tokunaga et al. have studied a highly fusogenic proviral HIV1-pL2 clone, with a view to identifying the amino acids responsible for this enhancement of fusion activity. By making recombinant envelopes combining parts of pL2 with parts of the less fusogenic pNL4.3, they identified glycine 36 of gp41 as essential for fusion enhancement. This glycine is conserved in almost all HIV-1 isolates, but not in the prototypic pNL4.3. In pNL4.3, an aspartic acid is found in position 36, preventing the correct formation of the helix hairpin, thereby decreasing fusion activity and infectivity. Interestingly, HIV-1 mutants escaping the effects of the fusion inhibitor T-20 frequently present variations in amino acid 36 of gp41. Münch et al. constructed peptide libraries from the hemofiltrates of patients with chronic renal failure and searched for antiviral peptide agents involved in the innate antiviral response. They identified a 20-residue peptide -- VIRIP (virus inhibitory peptide) -- that specifically inhibited infection with various HIV-1 isolates. This peptide, corresponding to a C-terminal fragment of 1-antitrypsin (the most abundant circulating serine protease inhibitor), seems to exert its antiviral effects by direct interaction with the gp41 fusion peptide. Hovanessian et al. identified a caveolin-1-binding motif within the ectodomain of gp41. This motif is conserved in all HIV-1 isolates and seems to be functional, as gp41 is found complexed with caveolin in infected cells. These researchers designed peptides (CBD1) corresponding to the consensus domain of gp41 and showed that these peptides bound caveolin-1 specifically. The peptides also elicited the production of specific anti-CBD1 antibodies, which inhibited the infection of primary CD4 cells by laboratory and primary HIV-1 isolates. The antibodies act at two different steps: they prevent the infection of cells by HIV particles and aggregate gp41 at the plasma membrane of HIV-1-infected cells, resulting in the production of defective particles. Unlike other gp41 epitopes, the CBD1 epitope is not a transient conformational epitope. ### Receptors and signaling Does the interaction between HIV-1 envelope glycoprotein and cell receptors, including the chemokine receptors CCR5 and CXCR4, for viral entry induce signals relevant to viral replication and cell function? This is a key issue in this field. Chakrabarti et al. presented convincing data showing that X4-tropic Env gp120, whether monomeric or in its natural conformation on inactivated HIV-1 virions, triggers a signaling array similar to that induced by SDF-1, the natural ligand for CXCR4, in unstimulated primary CD4 T cells. At concentrations (200 nM) close to the Kd for CXCR4, HIV-1 gp 120 efficiently activates Ga proteins and induces calcium mobilization, and activation of the MAP and PI3 kinase pathways. Inactivated virus and gp120 trigger CXCR4-dependent actin cytoskeleton rearrangements and cell chemotaxis. Thus, gp120 may function as a chemokine, inducing structural and functional changes favoring viral entry and replication, and may affect the trafficking and functional responses of unstimulated CD4+ T cells. The interactions between CD4 and CCR5 at the plasma membrane and their role in HIV-1 entry were explored by F. Bachelerie and coworkers, using FRET on living cells transduced with both receptors. The results of this group suggest that the two molecules are colocalized on the cell membrane, where they interact in a stable fashion. The disruption of this interaction inhibits R5 HIV-1 infection. The same team also presented data on the relationships between CCR5 activation, signaling and β-arrestin-mediated endocytosis and chemotaxis, suggesting that different CCR5 structural determinants may be involved in these responses. Session 2: Viral sanctuary -------------------------- Chair:R. Pomerantz; Keynote Lecture:\"HIV residual disease: The main barrier to viral eradication in the era of HAART\" by R. Pomerantz. Roger Pommeranz introduced the session with his keynote speech on \"Viral reservoirs as major obstacles for viral eradication despite effective highly active antiretroviral therapy (HAART)\". The combination of at least 3 different antiretroviral drugs in the clinical management of HIV-1 infection has improved the prognosis of HIV-1 infected patients. However, despite this therapy, HIV-1 has not been eradicated, at least partly due to latent HIV-1 replication occurring in the resting TCD4+. ### Combination therapy to induce out of latency The HIV-1 replication cycle includes a large number of possible stages for latency and persistence. Two mechanisms have been described: pre-and post-integration into the human genome. A large body of data has accumulated to indicate that the cells of HIV-1- infected patients may contain proviral DNA but produce only a small amount of viral RNA. In pre-integration HIV-1 latency, differences in latently infected cells may be observed, depending on the severity of disease. The virus may maintain cellular latency via various molecular mechanisms, which may depend on cell type. Understanding the basis of viral latency would make it easier to design new strategies for viral eradication. Because resting CD4+ T-cells are a major component of the reservoir of circulating cells in vivo, Pomerantz suggested that persistently infected cells should be activated in order to purge the viral reservoirs, making it then possible to control the production of new viruses by HAART. For example, IL-2 treatment could be combined with d4T/3TC/Efavirenz; or treatment with OKT3 anti-T cell receptor monoclonal antibody could be combined with ddI. However, experimental data have shown viral rebound after cell stimulation with IL-2, for example. The new viruses produced upon activation came from follicular dendritic cells, lymph nodes, cells in sanctuary sites and other tissues. Most of the HIV produced from reservoirs and during rebound are defective in the V3 sequence of the HIV-1 envelope gene. Resting cells reflect the state of the immune system. In contrast to the results obtained with IL-2, therapeutic strategies based on the stimulation of cells with IL-7 associated with HAART have yielded promising results. IL7 alone is indeed a more potent activator of latent infected cells than IL-2. Most of the newly produced viruses had a CXCR4 and CCR5 phenotype. Although these approaches show promise for circulating resting CD4+ T-cells, new strategies must be defined to target novel pharmacological drugs to viral sanctuaries such as the brain or testes. A combination of both approaches may facilitate eradication of the viral reservoir in HIV-1-infected individuals. Massips et al. have followed three HIV-1-infected patients with viral loads below the detection threshold and who have been on HAART for 7 years. Viral DNA was detected in the memory and naive CD4+ T-cell subsets and in CD14+ monocytes, but not in CD56+CD3-NK cells. Phylogenetic analysis demonstrated that the various types of blood cell in two of the three patients harbored genetically different quasispecies. This suggests that the virus populations within each type of blood cell evolved independently and may originate from difference sources (differences in CXCR4 and/or CCR5). Real cellspecific compartmentalization of residual virus populations is thus observed in patients on HAART. For instance, in one of the three patients investigated, CCR5 variants were found in naive CD4+ T-cells and in memory CD4+ T cells. However, both CXCR4 and CCR5 variants were present in CD14+ monocytes. In another HIV-1-infected patient, CXCR4 variants were found in CD14+ monocytes and in naive CD4+ T cells, whereas both CXCR4 and CCR5 variants were found in resting memory CD4+ T cells. Ivan Hirsh et al. reported that CCR5 HIV-1 variants predominantly infect CD62Lnegative memory T cells, which selectively express the CCR5 receptor. The predominance of CXCR4 HIV-1 variants in less differentiated memory CD4+ T cells may be related to their activation state, as suggested by the expression of both CD45RA and CD45RO molecules on their membrane. In addition, most viruses isolated from peripheral blood resting cells of HIV-1-infected patients with levels of viral RNA in plasma below the detection threshold have few mutations conferring drug resistance. The CCR5 HIV-1 variants, which predominantly infected memory T cells, were found to be resistant to nucleoside reverse transcriptase inhibitors (NRTIs) such as zidovudine and lamivudine. As pointed out by J. Ghosn and colleagues, resistance mutations acquired by HIV-1 during primary infection may correspond to the dominant viral population, and are archived in cellular reservoirs at an early time point, despite treatment. In summary, virological failure in the resting memory CD4+ T cells, the emergence of a dominant pool of HIV-1-resistant virus very early in primary infection and the difficulties involved in getting drugs into viral sanctuary sites, once again raise questions as to the best combination of approaches for eradicating HIV-1 from infected individuals. ### HCV and IFN-alpha Feray et al. reported the effect of interferon-alpha in patients infected with hepatitis C virus (HCV). Differences in the composition of HCV quasispecies between plasma and peripheral blood mononuclear cells (PBMCs) suggest that PBMCs support viral replication. The frequency of compartmentalization in 119 naive patients chronically infected with HCV was determined and found to be correlated to virological response to inteferon-alpha. A significant proportion of HCV patients responding well to IFN-alpha treatment were found to be coinfected with variants not found in plasma. This relationship was independent of route of infection, plasma genotype and duration of infection. Session 3: Restriction of viral replication ------------------------------------------- Chairs:B. Cullen and D. Moradpour; Keynote Lecture:\"Defensive arts: innate intracellular immunity against retroelements\" by D. Trono. ### Innate and adaptive immunities to HCV in the host Most viral infections are successfully controlled by conventional innate and adaptive immune responses developed by the host. Viruses such as HIV, HCMV and HCV are able to persist in their host in the long term thanks to multiple strategies aimed at shutting down antiviral defenses. However, one major difference between HIV and HCMV on the one hand, and HCV on the other, is that HCV infections may, in some cases, resolve spontaneously or under treatment. Critical immunological events may thus take place early in viral infection that lead to viral clearance. Our understanding of these early events in the antiviral immune response has led to great efforts in recent years to diagnose HCV infection during the acute phase. This trend was illustrated by the two presentations on HCV in this session. F. L. Cosset focused on the analysis of neutralizing antibodies in a cohort of 17 individuals acutely infected with a single nosocomial outbreak strain of genotype 1b HCV. Neutralizing activity, evaluated by assessing the ability of the patients\' sera to inhibit the infection of HuH7 cells by HCV pseudotyped particles, was monitored, together with viral load and phylogenetic analysis of the predominant viral strains was also carried out. The patients studied could be divided into two subgroups on the basis of the infecting genotype 1b strains. Group 1, infected with strain A, showed a very large decrease in viral load within nine weeks of infection whereas group 2, infected with strain B, maintained high viremia. One major finding of this study was that group 1 patients display potent neutralizing activity that is inversely correlated with viremia. The sera from group 2 patients were found to facilitate infection with HCVpp rather than neutralizing such infections. This study strongly suggests that neutralizing antibodies are involved in the control of HCV infection -- an observation in apparent contradiction with recent reports \[[@B1]-[@B3]\] and with data reported by C. Bain in this same session. Bain\'s study was performed on a cohort of seven intravenous drug users acutely infected with genotype 3 HCV treated with pegylated IFN-alpha. In this study, the neutralizing activity of the patients\' sera neatly paralleled titers of antibodies specific for the envelope E2 glycoprotein. However, these neutralizing antibodies were found both in patients who responded to antiviral therapy and in those who did not, calling into question the role of neutralizing antibodies in therapeutic resolution of acute HCV infection. Longitudinal analysis of T-cell immune responses did not result in the identification of immune correlates of the therapeutic resolution of acute HCV infection. T-cell responses were surprisingly weak throughout follow-up, as shown by comparison with recently published data \[[@B4],[@B5]\] but were improved by treatment with immunomodulators. The initiation of IL-2 treatment strongly increased not only the vigor, but also the breadth of HCV-specific immune responses, revealing significant reactivity to the newly described alternative reading frame protein (ARFP) of HCV, in particular. However, therapeutic recovery from HCV infection could be achieved in the presence of T-cell suppressive mechanisms, suggesting that the presence of immunosuppressive T cells is not in itself responsible for therapy failure and subsequent chronic infection and that these cells probably modulate detrimental immune responses to maintain persistently low levels of liver inflammation in chronic HCV infection. The existence of immune responses to ARFP provides further evidence that this protein is synthesized in natural HCV infection and, like conventional HCV antigens, is expressed during the early steps of HCV infection. Together with published studies, these two presentations highlighted the difficulties involved in identifying immune correlates of viral clearance in a viral infection that may resolve spontaneously. ### Intrinsic host immunity to HIV In contrast to what has been reported for HCV, some HIV-infected individuals may control disease progression, but they never eliminate viral infection altogether, suggesting that the conventional immune system is unable to control viral replication. In addition to conventional innate and acquired immune responses, complex organisms have developed so-called \"intrinsic\" immunity, mediated by constitutively expressed restriction factors that efficiently prevent or limit viral infections \[[@B6]\]. Two major classes of factor have been shown to restrict retroviral infections by blocking incoming retroviral particles (Fv1 and TRIM5a) or by the specific deamination of dC residues to generate dU, leading to the hypermutation of viral DNA and blocking viral replication (APOBEC3; class: cytidine deaminases). Obviously, viruses have evolved strategies to overcome these restriction mechanisms. D. Trono, in the opening lecture of the session, summarized recent data on the cytidine deaminase superfamily, mostly focusing on APOBEC3G \[[@B7]\]. This restriction factor, primarily found in T lymphocytes and macrophages, is packaged into HIV virions in the absence of Vif protein, via specific interaction with the NC/p6 domain of the Gag polyprotein precursor (B. Cullen) or non-specific RNA binding \[[@B8]\]. Upon the infection of new target cells, APOBEC3G deaminates the nascent minus strand DNA, resulting in a less stable uracyl-containing minus-strand DNA, which is degraded or yields hypermutated plus-strand DNA liable to encode defective viral proteins. In the presence of Vif, APOBEC3G is targeted for proteasome degradation. Bet protein, derived from primate foamy virus, can partially rescue Vif-deleted virions (B. Cullen). APOBEC3G has been shown to block a wide range of retroviruses and unrelated viruses such as hepatitis B virus (HBV) \[[@B9]\]. If hepatoma HuH7 cells are cotransfected with a plasmid containing the HBV genome and a plasmid encoding APOBEC3G, intracellular levels of core-associated HBV DNA are significantly lower than those in cells transformed with the viral genome alone. Although this effect is inhibited by HIV-1 Vif, catalytically inactive APOBEC3G continues to have an inhibitory effect on HBV DNA, suggesting that APOBEC3G may act on HBV and retroviruses via different mechanisms. However, one unresolved question concerns the potential relevance of such an interaction as APOBEC3G is expressed in lymphoid cells and HBV mostly infects hepatocytes. Anti-HIV-1 treatments targeting Vif protein may eventually come out of this work. ### RNAi targeted to HIV R. Benarous presented work on another antiretroviral strategy, the use of RNA interference to block the interaction between HIV integrase and a cellular protein, the lens epithelium-derived growth factor/transcription coactivator p75 (LEDGF/p75) protein. HIV-1 replication is strongly inhibited by the presence of siRNA targeting the 3\' end of the LEDGF coding region, suggesting that this protein is required for HIV infection. Further experiments with HIV integrase (Gln168) mutants displaying defective HIV-1 DNA integration, demonstrated the involvement of LEDGF in the targeting of integrase to chromosomes. HIV-1 can infect the central nervous system (CNS), where it causes progressive cognitive and motor dysfunctions. Astrocytes have been shown to be target cells for HIV-1 in the CNS but these cells allow only limited replication of HIV-1. They can also be infected with HIV-1 in vitrobut such infections are generally of very low and transient productivity, suggesting that astrocytes may contain a factor that restricts HIV-1 replication. ### Rev and RNA transport In this session, S. Kramer-Hämmerle reported an abnormal distribution of HIV-1 Rev in astrocytes, with a blockade of its nucleocytoplasmic shuttling function leading to the inhibition of nuclear export of HIV-1 mRNAs. Using a cDNA library from astrocytes, a double-hybrid strategy in yeast and then in mammalian cells, Kramer-Hämmerle identified a cellular factor -- 16.4.1 -- that colocalized with Rev in transfected cells. This factor also interacted with an exportin, CRM1, a member of the karyopherin family of nucleocytoplasmic transport factors and a cellular cofactor for the Rev-dependent export of HIV-1 RNAs. 16.4.1, which is probably part of a larger protein, reduces Rev activity. These data illustrate the huge diversity and complexity of mechanisms developed by these two viruses for the establishment of chronic infection. However, these two viral infections differ primarily in that HCV-infected patients, unlike HIV-infected patients, may recover spontaneously from infection. This may explain why the study of conventional immune responses has always been a major research field for HCV whereas HIV research is gradually turning to the investigation of more intrinsic interactions between host and viral proteins. Session 4: Viral infection and innate immunity ---------------------------------------------- Chairs:C. Soderberg-Naucler & L. Zitvogel; Keynote Lecture:\"Immunopathology of prion infection\" by A. Aguzzi* ### Prions This session began with a keynote lecture by A. Aguzzi, presenting data on two aspects of prion infection. He first presented an immunointervention strategy for modulating the course of scrapie in mice, based on a chimeric PrP molecule consisting of two PrP fused to the constant fragment of an IgG (PrP-Fc2). The aim was to interfere with the PrPsc -- PrPc interaction, which results in there being two PrPsc conformers and spreads \"infection\", as a means of limiting disease. Aguzzi\'s team hypothesized that an Fc-linked dimer of PrPc would interact with PrPsc without transconformation, thereby blocking prion progression. Such an interaction was demonstrated to exist as PrP-Fc2 precipitated PrPsc from diseased brains. Crossing WT mice and transgenic mice expressing PrP-Fc2 delayed the onset of scrapie (by up to 150 days) and decreased PrPsc accumulation. Moreover, PrP-Fc2, which normally settles in the bottom layer of membrane fraction gradients, was redistributed to the raft layer, which is the site of PrPsc is in infected brain preparations. Nevertheless, the delay in scrapie onset may not be entirely due to higher levels of PrPsc clearance through the reticulo-endothelial system, as the Fc fragment was deleted from its FcgR interaction site. These encouraging data led to the transfer of PrP-Fc2 into WT mice brain via lentiviral vector, which conferred clinical resistance to scrapie for up to 265 days. The PrP-Fc2 transferred by the lentivirus decreased astrogliosis in the injected hemisphere whereas the contralateral hemisphere continued to displaye strong GFAP reactivity. The PrPsc signal was cleared only near the injection site. The protection conferred by PrP-Fc2 requires central expression, as peripheral injection is not protective, although PrP-Fc2 expression can be targeted to oligodendrocytes, a cell type not infected by prions, in the periphery. Aguzzi then rapidly presented data for transgenic mice displaying targeted tissue-specific expression of lymphotoxin antibody. In these mice, which displayed tertiary lymphoid tissue development in the liver, the kidney or the pancreas, the replication responsible for infectivity occurs in these organs. This raises questions of food safety, if animals with inflammation sites are used for meat, but may also open up new possibilities for the use of peripheral preventive strategies such as PrP-Fc2 injection during the invasion phase of spongiform encephalopathies. ### NK cells and HCV, HIV, and HCMV The session then moved on to more conventional viruses and dealt with the effects of HCV, HIV and HCMV actions on natural killer cells and monocytes/macrophages. U. C. Meier presented comparative data on NK cell modulation in response to HCV and HIV infection. The major subpopulation of NK cells in uninfected humans is CD3-/CD56 dim NK cells. These cells are highly cytolytic and display strong NK receptor expression, and low levels of trafficking and cytokine production. The minor CD3-/CD56 bright subpopulation displays the opposite phenotype with respect to these characteristics. In response to HCV and HIV infections, the number of NK cells in the blood decreases and there is a shift toward the CD56 bright subpopulation, with no change in CD57 expression on NK cells. This results in a decrease in the percentage of perforin-bright NK cells in favor of perforin-dim cells. In HCV patients, this decrease was shown not to correspond to NK cell accumulation in the liver. The response of NK cells to HCV and HIV infections differed in that interferon production under IL12 + IL18 stimulation decreased in the NK cells of HIV patients but increased in those of HCV patients. The decrease in frequency of NK cells may be the consequence of a loss of IL15 expression, as the serum concentration of this cytokine is low in both infections, or of an impaired response to the IL15 survival signal. Such an impaired response to IL15 was demonstrated only in HIV infection, in terms of survival and cytolysis. ### HCV, HCMV and monocyte activation Assessment of the effect of HCV and HCMV on monocyte activation and differentiation as a means of estimating viral persistence was the subject of two talks, by P. Balard and S. Gredmark. HCV persistence is thought to be associated with a Th2 bias, which is demonstrated by a decrease in IL-12 production and an increase in CD36 membrane expression on monocytes. Chêne et al. showed that HCV core protein induces the overproduction of PGJ2 by the PLA2 -- Cox2 cascade, with Cox2 overproduced. PGJ2 is a ligand for PPARl, which is activated in HCV-core-treated monocytes, and involved in CD36 and IL-12 modulation. These results are consistent with the notion that the HCV present in the patient\'s serum may establish a chronic infection by inducing an M2 orientation of monocyte activation, leading to a biased T-cell response. Monocytes-macrophages are also critical to HCMV infection, as this virus can be reactivated in vitrofrom macrophages. HCMV strategies for escaping immune surveillance include decreases in the expression of MHC class I and class II molecules, the impairment of T-cell activation, and a decrease in NK cell-mediated lysis. Gredmark et al. found that a suspension of HCMV inhibited the differentiation of monocytes into mature macrophages, resulting in the production of monocytoid cells with impaired migration and phagocytosis and low levels of β-chemokine production. This inhibition was achieved with inactivated HCMV, but not with HCMV suspension supernatant; nor was it reproduced with HIV or measles virus. The viral effector was identified as the gpB protein of HCMV, which binds to CD13 and signals by means of Ca^2+^flux, through this receptor. CD13 is an N-aminopeptidase involved in monocyte-macrophage adhesion and migration. Using monoclonal CD13 antibodies, Gredmark were able to mimic or to anatagonize the effect of HCMV on macrophage differentiation, depending on the clone used. These two talks strongly suggested that monocytes-macrophages are, together with NK cells, a major target for the prevention of viral persistence and infection chronicity. However, viruses may use several different strategies, involving numerous mechanisms to establish chronic infections. Session 5: Chemokines and inflammatory cytokines ------------------------------------------------ *Chairs:K. Klenerman and G. Poli; Keynote Lecture:\"CD4 T-cell homeostasis in HIV infection: role of the thymus\" by R. Sekaly. ### CD4 T-cell homeostasis in HIV infection: role of the thymus In chronic viral infections, CD4+ T-cell responses are associated with disease control. R. Sekaly reported stronger proliferative HIV-specific CD4+ T-cell responses in aviremic than in viremic patients. Long-term CD4+ T-cell memory depended on IL-2-producing CD4+ T cells whereas cells producing only IFN-γ were short-lived. Sekalt characterized the ex-vivophenotype of CD4+ T cells in more detail by genomic and proteomic analysis, and identified genes differentially expressed along the CD4+ T-cell differentiation pathway: 1) TOSO, which inhibits Fas- and TNF-mediated apoptosis, and PIM2 and DAD1 were more strongly expressed in naive and central memory CD4+ T cells than in effector/memory and effector CD4+ T cells. These genes were also expressed more strongly in samples from healthy donors than in samples from viremic patients; 2) Conversely, Rab27a, which indicates the activation state of T-cell maturation, was expressed more strongly in effector and effector/memory CD4+ T cells than in naive and central memory CD4+ T cells. These data provide new insights into CD4+ T-cell homeostasis during HIV infection. ### Cytokine production in the livers of HCV+HIV- and HCV+HIV+ individuals As cytokines play a crucial role in controlling the immune responses against viral persistence, G. Paranhos-Baccala et al. measured intrahepatic levels of IFN-gamma, TNF-alpha, TGF-β, IL-2, IL-4, IL-8, IL-10 and IL-12p40 by real-time PCR in 12 HCV+HIV- and 14 HCV+HIV+ individuals. They showed that the detection rates for individual cytokines were higher for the HCV+HIV- group than for the HCV+HIV+ grou. However, only the detection rates for TNF-alpha, IL-8 and IL-10 differed significantly between the two groups. Moreover, median levels of IFN-gamma, IL-8 and IL-10 were significantly higher in the HCV+HIV+ group. This study demonstrated the existence of a global defect in cytokine signaling in HCV+HIV+ individuals, which may contribute to HCV persistence. ### HIV interactions with other pathogens in coinfected human lymphoid tissues Recent epidemiological studies have reported examples of of the inhibition of HIV replication by microbial interactions. In a study of ex vivo-infected human lymphoid tissue, L. Marogolis et al. showed that two microbes (measles virus (MV) and Toxoplasma gondii(TG)) inhibited the replication of both CXCR4-tropic (X4) and CCR5-tropic (R5) HIV-1. This inhibitory effect was particularly marked for R5 virus and was mediated by a parasite-encoded cyclophilin, C18, in TG-infected tissues, and by a CC chemokine, RANTES, in MV-infected tissues. These microbes were also found to display a moderate cytopathic effect on lymphocytes, decreasing the number of R5 and X4 HIV-1 targets in co-infected tissue. This study highlighted the crucial role of the cytokine/chemokine network in interactions between microbes in the human host. ### Early induction of an anti-inflammatory environment may temper T-cell activation during SIVagm infection During primary SIVagm infection, African green monkeys (AGM) can display a transient decline in CD4+ T-cell counts together with transient T-cell activation until the end of primary infection. Cytokine gene expression was assessed in a longitudinal studycarried out by Ploquin et al., before infection and at intervals of two to three days during primary infection (PI), and then regularly until day 430 postinfection. The following observations were made in SIVagm-infected AGMs: 1)A significant increase in TGF-b1 and Foxp3 gene expression beginning in the first week after infection, coinciding with expansions of the populations of CD4+CD25+ and CD8+CD25+ T cells; 2) An increase in IL-10 gene expression during the 2nd and 3rd week p.i, with no change in TNF-alpha gene expression at any point in the study; 3) Changes in the plasma concentration of cytokines correlated with gene expression changes. In conclusion, the harmful generalized immune activation levels observed during the post-acute phase of SIVagm infections may be controlled by the early induction of anti-inflammatory cytokines, as observed in this study. ### HIV infection: role of IL-7 in immune reconstitution after HAART or HAART plus IL-2 and preclinical assessment of its therapeutic potential As plasma IL-7 levels are negatively correlated with CD4 counts during HIV disease progression and antiretroviral therapy, J. Theze suggested that IL-7 is part of a feedback loop regulating the size of the CD4 pool. In this study, plasma IL-7 levels at the start of HAART were found to be positively correlated with an increase in CD4 counts during the first two years of HAART. Plasma IL-7 concentrations increased in HIV-infected patients receiving HAART plus IL-2. Theze assessed the therapeutic potential of IL-7 by studying IL-7R expression in CD4 and CD8 T lymphocytes from three groups of patients (group 1: naive for antiretroviral therapy (plasma viral load \> 10,000 copies /ml and CD4 count \> 350 cells /mm^3^); group 2 : HAART-treated patients with CD4 \> 400 cells/mm^3^and plasma viral load \< 50 copies /ml; group 3: HAART-treated patients with CD4 counts remaining low (CD4 \< 250 cells /mm^3^) despite good control of plasma viral load (\< 50 copies /ml)). The major findings of this study were: 1) CD127 was less strongly expressed on CD4 lymphocytes from group 1 and group 3 patients than on those from group 2 patients; 2) CD8+ lymphocytes from HIV-infected patients were mostly CD27-CD45RO+ and CD27-CD45RO-; 3) High viremia was correlated with IL-7R dysfunction, whereas HAART-treated patients recovered a functional IL-7R. These concluded that the IL-7/IL-7R system plays a role in HIV disease and that IL-7 could be used in immune interventions to treat HIV infection. ### The HIV-1 mediated induction of ET-1 in the CNS increases the secretion of markers of blood-brain barrier failure, which are altered by HIV-1 protease inhibitors, nelfinavir N. Didier et al. suggested that endothelin-1 (ET-1) is involved in the neuropathogenesis of HIV-1 infection because ET-1 levels have recently been shown to be correlated with the degree of encephalopathy in HIV-1-infected individuals. Using a model of the blood-brain barrier (BBB), N. Didier et al. showed that the production of ET-1 by brain endothelial cells in response to HIV-1 leads to disruption of the BBB by the pro-inflammatory cytokines (IL-1, IL-6 and IL-8) produced by astrocytes. As proteases play an important role in inflammatory processes, nelfinavir decreases the level of cytokine secretion, and may therefore be useful in HAD. Session 6: Dendritic cells and activation of T-cell antiviral responses ----------------------------------------------------------------------- Chairs:B. Autran & A. Hosmalin; Keynote Lecture:\"Combat between cytomegalovirus and dendritic cells in T-cell response\" by C. Davrinche; ### Combat between cytomegalovirus and dendritic cells in the T-cell response During HCMV infection, innate (apoptosis, IFNα/β, complement, NK cells and dendritic cells) and adaptive (CD4+, CD8+ and antibodies) immune responses are generated. The main target proteins for CD4 and CD8 T cells are IE1 and pp65 (early proteins). In a model consisting of dendritic cells (DC) cocultured with HCMV-infected fibroblasts, C. Davrinche showed that the fibroblasts rapidly became apoptotic. The DC acquired pp65 from infected fibroblasts via a mechanism requiring cell-to-cell contact and, after 6 hours, DC produced TNFα and IL6. In the presence of PBMC, a large number of pp65-specific CD8 T cells were generated and a peak of IFNγ production was observed 24 h after incubation. DC maturation (upregulation of CD83) was induced by incubation with HCMV-infected fibroblasts, and a peak in CD83 expression was observed after 6 h, with levels decreasing after 48 h and 72 h. This maturation seems to be a prerequisite for efficient T-cell stimulation. C. Davrinche has identified a soluble factor (TGF-β) secreted at a late stage of HCMV infection in fibroblasts that downregulates CD83. He has also shown that the IL10 homolog carried by HCMV interferes with DC maturation and cross-presentation. Overall, the results presented suggested that cross-presentation must occur soon after infection by HCMV to prevent the soluble factor-mediated viral escape mechanism. This may explain why the main target proteins for T-cell responses are IE1 and pp65, which are available early in infection. ### HIV-1-induced dysfunction of naive CD8+ T cells D. Favre showed that in the SCID-hu thymus/liver mouse model, HIV infection of the thymus resulted in a CD8 functional defect due to impaired signaling via the TCR complex, with effects on calcium flux and IL-2 responses (cytokine production and proliferation). After the transplantation of a human thymus/liver graft in SCID mice, thymocytes from SCID-hu mice were infected in week 18 with HIV-1 NL4-3, BaL, or primary stocks and the infected animals were compared with mock-infected animals. HIV infection of the thymus induced the upregulation of MHC-I in thymocytes, correlated with increases in HIV RNA levels and the development of single-positive CD8^low^(SP8) thymocytes. Following polyclonal stimulation (anti-CD3/CD8) viathe TCR, a significantly weaker calcium flux response and lower proliferative capacity, as measured by CFSE, were observed in SCID-hu thymus/liver mice than in control mice. Thus, in the SCID-hu thymus/liver mouse model, HIV infection results in the selection of CD8^low^T cells with defective calcium flux signaling. Favre also presented data concerning the activation status of circulating CD8^+^T cells from 40 HIV-1-infected patients at various stages of the disease. In patients with progressive disease, a decrease in CD8^+^naive (CD45RA^+^CD27^+^) T-cell counts was observed, with low levels of CD8 expression, associated with chronic immune activation, as assessed with the CD38 marker. A dysfunction in calcium flux and IL-2 responses is also observed in patients with progressive HIV disease. In conclusion, the CD8^low^T cells observed after experimental HIV infection of the thymus and in the peripheral blood of patients with progressive HIV disease seem to display MHC-I upregulation and defect in signaling across the TCR, associated with chronic immune activation (CD38). Fabre suggested that the higher density of MHC-I on cells in the thymus might lead to high-avidity interactions with TCRs on developing thymocytes and hence to supranormal levels of negative selection, but it remains unclear how these CD8^low^T cells are generated. Such dysfunctional CD8^low^T cells would contribute to the profound immunodeficiency associated with HIV disease progression. ### Role of HIV-1 Nef in viral replication in lymphocytes The results presented by Nathalie Sol-Foulon demonstrated a requirement for ZAP70 for efficient HIV replication in Jurkat cells and the severe impairment of replication in Nef-deleted virus in Zap-deleted Jurkat cells. In these experiments, Jurkat cells or PBLs were infected with a wild-type HIV or Nef-deleted HIV and stimulated by PMA iono or superantigen. IL-2 production was then evaluated. Sol-Foulon showed that HIV infection increased activation (as assessed by determining IL-2 production) in response to T-cell stimulation viathe TCR or the MAP kinase signaling pathways. Infection with wild-type HIV or Nef-deleted HIV had no significant effect on IL-2 production (53% and 43%, respectively) so Nef does not significantly affect this process. The absence of ZAP70 is known to cause a major defect in the TCR. HIV replication is strongly affected in Zap-deleted Jurkat cells but it is unclear which step of the viral cycle is affected and the effects of Zap on viral replication in primary T cells and the links between transduction pathways and HIV replication are unknown: Sol-Foulon is currently investigating these aspects. ### The extent of CD4+ T cell apoptosis during primary SIV infection is predictive of the rate of progression to AIDS J. Estaquier showed that the rate of CD4+ T-cell apoptosis was correlated with subsequent viremia levels whereas levels of CD8+ T cells were not. In rhesus macaques experimentally infected with the pathogenic SIVmac251 isolate, peak numbers of apoptotic cells in the lymph node T-cell areas were significantly higher in future rapid progressors than in the slow progressors during the first two weeks of infection. No correlation was found between the rate of viral replication within lymph nodes and the extent of FasL-mediated apoptosis in CD4+ T cells. The mechanism of apoptosis seems to be independent of the caspase and AIF pathways. The role played by mitochondria was also evaluated in SIVmac251-infected macaques and the results presented indicated that the Bak gene is involved in SIV-mediated CD4+ T cell apoptosis. Estaquier concluded that memory T cells are lost early in infection and that levels of apoptotic CD4+ T cells are predictive of disease progression. ### A T-cell based HCV vaccine capable of blunting acute viremia and protecting against acute and chronic disease induced by heterologous viral challenge in chimpanzees Alfredo Nicosia presented his results for HCV-vaccination with an MRK adenovirus at weeks 0 and 25 and a DNA EP boost in week 35. Chimpanzees were challenged with a heterologous virus in week 49, and the vaccination was shown to have elicited potent, broad-range and durable effector T-cell responses. The immunogen used was from a non structural region of HCV corresponding to genotype 1b, the most frequent strain in USA and Europe. The challenge involved H77, corresponding to a genotype 1a. In this study, five animals were vaccinated and five others received the control vector. Specific IFNγ-CD8+ responses were maximal in week 37, after the booster. Polyspecific HCV- CD8+ responses were detected in peripheral blood and in the liver. These specific immune responses, induced by vaccination, were also elicited by the with challenge strain, demonstrating cross-reaction. Nicosia showed that eight weeks after challenge, viral load in vaccinated animals was less than one hundredth that in control animals (P= 0.009). He also demonstrated an absence of liver damage in vaccinated animals, whereas ALT and GGT levels were high in control animals. He concluded that this vaccine can prevent hepatitis and protect animals against chronic infections caused by heterologous viruses. ### Cross-presentation by dendritic cells of HIV antigens from live infected CD4+ T lymphocytes e Hosmalin showed that dendritic cells (DC) can capture, and cross-present to specific-CD8+ T cell lines, HIV antigens from live, infected cells as efficiently as antigens from apoptotic infected CD4+ T cells. When MDDC + LPS were cultured with various sources of HIV antigens (peptides from Gag, RT, free virus, CD4+ T cell lines infected with HIV) and presented to CD8+ T cell lines specific for Pol 476--484, the cross-presentation of HIV antigens from apoptotic infected CD4+ T cells was more efficient than direct DC infection or other sources of HIV antigens. Hosmalin also presented other data, showing that similar levels of cross-presentation are also observed in live infected CD4+ T cells. She performed similar experiments with live infected CD4+ T cells and ex vivoPBMC from HIV-infected patients. In HIV-infected patients, circulating CD8+ T cells recognized cross-presented HIV antigens from live infected T cells. Thus, anti-HIV immunity begins before the induction of apoptosis. Moreover, the proportion of CD83+ mature DC increased when DC were incubated with primary CD4+ T cell blasts, whether apoptotic or not, and independent of HIV infection. Hosmalin concluded that, during HIV infection, live or apoptotic HIV-infected T lymphocytes can supply antigens and costimulation signals for MHC class-I-restricted presentation by DC or induce tolerance in patients with low CD4 counts and impaired CD4 T-cell functions. Acknowledgements ================ Conference Organizing Committee:Conference chair: Françoise Barré-Sinoussi; Conference cochairs: Patrick Gourmelon & Roger Le Grand; Secretary: Daniel Béquet; Vice-Secretary: Hervé Fleury; Treasurer: Pascal Clayette; Scientific Advisors: Henry Agut, Paul Brown, Jean-François Delfraissy, Jacques Grassi, Geneviève Inchauspé, Olivier Schwartz. Sponsors:*Agence Nationale de Recherche sur le SIDA (ANRS, Paris, France), Aventis-Pasteur (Marcy-l\'Etoile, France), BD Biosciences (Le Pont de Claix, France), BioMérieux (Lyon, France), Biorad (Marnes la Coquette, France), Commissariat à l\'Energie Atomique (CEA, Paris, France), Direction Générale pour l\'Armement (DGA, Paris, France), Institut de l\'Ecole Normale Supérieure (ENS, Paris, France), Gilead Sciences (Paris, France), Novartis (Bale, Suise), Spi-Bio (Montigny le Bretonneux, France).	PubMed Central	2024-06-05T03:55:56.218439	2005-4-6	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087881/", "journal": "Retrovirology. 2005 Apr 6; 2:24", "authors": [ { "first": "Elisabeth", "last": "Menu" }, { "first": "Mickaela C", "last": "Müller-Trutwin" }, { "first": "Gianfranco", "last": "Pancino" }, { "first": "Asier", "last": "Saez-Cirion" }, { "first": "Christine", "last": "Bain" }, { "first": "Geneviève", "last": "Inchauspé" }, { "first": "Gabriel S", "last": "Gras" }, { "first": "Aloïse M", "last": "Mabondzo" }, { "first": "Assia", "last": "Samri" }, { "first": "Françoise", "last": "Boutboul" }, { "first": "Roger Le", "last": "Grand" } ] }
PMC1087882	The silent majority =================== In his US presidential campaigns, Richard Nixon famously popularized the concept of a \"silent majority\". Fundamentally what Nixon meant was that the stalwart contributors in any given population tend to be a faceless and non-vocal \"majority\". Applying Nixon\'s term to science, where is our \"silent majority\"? I recently returned from a large international meeting where numerous awards were presented to wunderkinds (i.e. Young Investigator Awards) and senior statesmen/stateswomen of science (i.e. Lifetime Achievement Awards). Curiously, I found subjectively that the majority of the best work was being presented by a glum-looking bunch of scientists traversing \"midlife crises\" (i.e. individuals who uncannily resemble me!). It struck me that arguably the most productive denizens of science are those in the 45 to 60 age group (who are no longer young investigators and not yet competitive for lifetime achievements); and they are our equivalent of Nixon\'s \"silent majority\". The RetrovirologyPrize ======================== To my knowledge, there is no significant scientific prize that is age-restricted to mid-career scientists (i.e. between the ages of 45 to 60). Although mid-life scientists do and have competed ably for numerous non-age restricted awards, a unique niche is fulfilled by a prize that specifically recognizes this increasingly large and productive age group. Moreover, I am not aware of a prize limited to retrovirologists. Beginning this year, Retrovirologywill recognize annually through the RetrovirologyPrize one 45- to 60- year-old retrovirologist. The selection process ===================== This editorial kicks-off the 2005 call for nominations for the RetrovirologyPrize. We envisage an approximate alternation of the Prize yearly between a non-HIV retrovirologist (2005 and odd years) and an HIV retrovirologist (2006 and even years). There can be some discretion on this criterion exercised from time-to-time by the selection committee. A nomination includes a statement (1000 words or less) of significant contributions to retrovirology research and a curriculum vitae of the nominee. The selection committee is composed of the Editors (currently, M. Benkirane, B. Berkhout, M. Fujii, K.T. Jeang, M. Lairmore, A. Lever, and M. Wainberg) of Retrovirology. All nominations to the selection committee must come through an Editorial Board member of Retrovirology. Candidates may self-nominate, but they must ask a RetrovirologyEditorial board member to communicate their nominations to the selection committee. A list of Editorial Board members can be found at the Retrovirologywebsite www.retrovirology.com. All RetrovirologyEditors and Editorial Board members are eligible for nomination with the exception of the Editor-in-Chief who will administer the final selection decision. For 2005, all nominations must be received by June 1. The RetrovirologyPrize consists of an attractive trophy, a cash award, and a biographical article published in Retrovirologyabout the winner and his or her scientific contributions to retrovirology. Acknowledgements ================ I thank M. Benkirane, B. Berkhout, M. Fujii, M. Lairmore, A. Lever, M. Wainberg, R. King, T. Bloom, and G. Baynes for critical readings of this editorial.	PubMed Central	2024-06-05T03:55:56.223266	2005-4-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087882/", "journal": "Retrovirology. 2005 Apr 15; 2:26", "authors": [ { "first": "Kuan-Teh", "last": "Jeang" } ] }
PMC1087883	Background ========== Excessive mucus secretion is one of the major clinical manifestations of chronic airway diseases such as asthma, chronic bronchitis, and cystic fibrosis \[[@B1]\]. The excessive mucus is attributed to goblet cell hyperplasia (GCH) and submucosal gland hypertrophy, which are hallmarks of airway remodeling in chronic airway diseases \[[@B2],[@B3]\]. Air pollution aggravates respiratory symptoms in patients with chronic airway diseases. Chronic obstructive pulmonary disease (COPD) patients living in communities exposed to high levels of air pollution have faster rates of decline in lung function than patients living in areas with low pollution \[[@B4]\]. The level of environmental particles is also positively correlated with exacerbation of asthma \[[@B5]\]. Airborne particulate matter less than 10 μm in aerodynamic diameter (PM10) is a complex mixture of organic and inorganic compounds containing sulfates and various metals such as aluminum, calcium, copper, iron, lead, magnesium, titanium, and zinc \[[@B6]\]. Clinically, PM10 particles are thought to provoke airway inflammation with the release of mediators that are capable of exacerbating lung disease in susceptible individuals \[[@B5],[@B7]\]. This assumption is based on experimental evidence of airway inflammation following direct instillation or inhalation of PM10 particles in animal models \[[@B8]\]. Furthermore, inhaled particles directly stimulate macrophages and epithelial cells to produce inflammatory cytokines such as TNF-α, GM-CSF and IL-8 \[[@B9],[@B10]\], which induce neutrophil- and eosinophil-mediated airway inflammation, and eventually lead to GCH. Recently, particle exposure favors the antigen -- sensitized lung toward Th2 environment with over secretion of IL-13, IL-4 \[[@B11]\] and IL-5 \[[@B12]\]. Beside the inflammatory cell mediated -- GCH, IL-13 directly induces GCH and Muc5AC gene expression through the signaling of IL-4Rα and IL-13Rα \[[@B13],[@B14]\]. Therefore, we hypothesized that particles induce GCH via over-production of IL-13 by recruited inflammatory cells. Titanium dioxide (TiO~2~) particles, one component of PM10, are found in dusty workplaces such as industries involved in the crushing and grinding of the mineral ore rutile \[[@B15]\]. It was reported that 50% of TiO~2~-exposed workers had respiratory symptoms accompanied by reduction in pulmonary function \[[@B16]\]. Because acute and chronic exposure to TiO~2~particles induces inflammatory responses in the airways and alveolar spaces of rats \[[@B17],[@B18]\], TiO~2~-- instilled rat may be a good model to study the particle induced -- airway injury. In this study, we evaluated the role of neutrophilic and eosinophilic inflammation by pretreatment with cyclophosphamide inducing neutropenia \[[@B19]\] and the association of IL-13 by pretreatment with dexamethasone suppressing IL-13 gene expression \[[@B20]\]. Methods ======= Treatment protocols ------------------- Particles of TiO~2~(mean diameter = 0.29 μm, DuPont, Wilmington, DE) were suspended in endotoxin-free saline. The endotoxin concentration of the TiO~2~suspension was less than \<0.32 EU/ml as measured with a limulus amebocyte lysate kit (QCL-1000; BioWhittaker, Inc., Walkersville, MD). Seven-week-old male Sprague-Dawley rats (Charles River Technology Inc.) received a single intratracheal instillation of homogeneous suspension of TiO~2~particles (4 mg/kg in 200 μl of endotoxin free water). In a pretreatment group, cyclophosphamide (CPA) (100 mg/kg, i.p.) was given 5 days before instillation of TiO~2~and a second injection of CPA (50 mg/kg, i.p.) 1 day before TiO~2~instillation. In the second pretreatment group, dexamethasone (DEX) (0.25 mg/kg, i.p.; Sigma, St. Louis, MO) was administered 24 h before TiO~2~instillation. The Institutional Animal Care and Use Committee of Soonchunhyang University approved the study protocols. Preparation of lung tissues and morphological analysis ------------------------------------------------------ Rats were sacrificed at 4, 24, 48 and 72 hr after TiO~2~instillation by being anesthetized with pentobarbital sodium (65 mg/kg, i.p.) and bronchoalveolar lavage (BAL) was performed by 5 times instillation of 1 ml normal saline and gentle retrieval. Cell numbers were measured using a hemacytometer and differential cell counts were performed on slides prepared by cyto-centrifugation and Diff-Quik staining (Scientific Products, Gibbstowne, NJ). Immediately following BAL, the trachea was snap-frozen for RNA extraction or fixed with 4% paraformaldehyde in PBS and embedded in paraffin. The tissues were subjected to periodic acid-Schiff (PAS) and toluidine blue staining to permit measurement of goblet cells and mast cells, respectively. Morphometric analysis was performed under light microscopy at ×400 magnification. PAS positive epithelial cells and total epithelial cells were counted on the length of 250 μm basement membrane at each of four predetermined sites (12, 3, 6, 9 o\'clock; 12 o\'clock was the membranous portion) using a soft program (Nikon DXM 1200, Nikon Inc. N.Y. USA & Image Pro Plus 4.01 software, Media Cybernetics, Maryland, USA). Results are expressed as the percentage of goblet cells among the epithelial cells. Mast cells in the airway wall were counted on the membranous portion. The results are expressed as the number of cells staining positive for toluidine blue per area of 0.01 mm^2^. Reverse transcription-polymerase chain reaction (RT-PCR) -------------------------------------------------------- Total RNA was isolated using the modified guanidium thiocyanate-phenol-chloroform extraction method \[[@B21]\]. DNase I (10,000 U/ml; Stratagene, La Jolla, CA)-treated RNA was reverse-transcribed by incubating with 0.5 mM dNTP, 2.5 mM MgCl~2~, 5 mM DTT, 1 μl of random hexamer (50 ng/μl) and SuperScript II RT (200 unit/μl; Life Technologies, Grand Island, NY) at 42°C for 50 min, and heat inactivated at 70°C for 15 min. cDNA was aliquoted into tubes containing specific primer pairs for rat GAPDH, Muc1, Muc2 and Muc5ac genes for amplification (300, 403, 421, and 382-bp fragments, respectively). Nucleotide sequences of the primers were as follows. GAPDH-forward ; 5\'GGCATTGCTCTCAATGACAA3\', GAPDH-reverse; 5\'AGGGCCTCTCTCTTGCTCTC3\', Muc1-forward; 5\' AGAGCTATGGGCAGCTGG 3\', Muc1-reverse; 5\' ACTACCCCAGTGTCCCTC 3\', Muc2-forward; 5\' TACTGCTGATGACTGTAT 3\', Muc2-reverse; 5\'GGCCACAGGCCTGATACT3\', Muc5ac-forward; 5\' TACAAGCCTGGTGAGTTC 3\', Muc5ac-reverse; 5\' TCACAGTGCAGCGTCACA 3\'. Amplification was performed for 40 cycles (one cycle: 1 min at 94°C, 1 min at 52°C, and 1 min at 72°C) with initial denaturation at 94°C for 5 min and a final extension at 72°C for 10 min. Immunohistochemical identification of Muc5ac-expressing epithelial cells and IL-13-expressing cells --------------------------------------------------------------------------------------------------- Muc5ac-positive (+) epithelial cells and IL-13-positive (+) cells were identified by immunohistochemical staining. Three-micron tissue sections of the trachea were treated with 0.3% H~2~O~2~-methanol for 20 min to block endogenous peroxidase, and then incubated at 4°C overnight with anti-rat Muc5ac mouse monoclonal antibody (1:200 dilution; Neomarkers, Fremont, CA) or biotinylated anti-rat IL-13 antibody (1:5 dilution; Biosource, Camarillo, CA). After the slides had been incubated with avidin-biotin peroxidase complex (ABC kit, Vector Laboratories, Burlingame, CA), color was developed with 3,3\'-diaminobenzidine tetrachloride (DAB, Zymed Laboratories, South San Francisco, CA). The Muc5ac expressing epithelial cells and total epithelial cells were counted on the length of 250 μm epithelial basement membrane at each of four predetermined sites (12, 3, 6, 9 o\'clock; 12 o\'clock was the membranous portion). Results are expressed as the percentage of Muc5ac (+) cells among the epithelial cells. IL-13 (+) cells was counted on the membranous portion in the same way as mast cells were counted. The results are expressed as the positive rate of mast cells for IL-13 stain per area of 0.01 m^2^. Measurement of IL-13 concentration in BAL fluids ------------------------------------------------ The levels of IL-13 in the BAL fluids were measured with a quantitative sandwich enzyme-linked immunoassay kit (Biosource, Camarillo, CA). The lower limit of detection was approximately 1.5 pg/ml. Values below this limit were considered as zero for statistical analysis. Inter- and intra-assay coefficients of variance were less than 10%. Statistical analysis -------------------- Differences between independent samples were compared using the Spearman test for continuous data. If differences were found significant, the Mann-Whitney U test was applied to compare differences between two samples. Differences were considered significant when the p value was less than 0.05. Results are expressed as means ± standard error of the mean (SEM) unless otherwise stated. The correlations were analyzed between the ratio of Muc5ac (+) expressing epithelial cell and the concentration of IL-13 in BAL fluid and the number of mast cell and the IL-13 positive rate of mast cells by Spearman\'s non-parametric correlation using SPSS (version 10.0, Chicago, USA) Results and Discussion ====================== Expression of Muc gene transcripts in the trachea of TiO~2~or saline -- instilled rats -------------------------------------------------------------------------------------- Total RNA was extracted from the trachea 24 h following treatment with saline or TiO~2~and analyzed for Muc1, Muc2, and Muc5ac transcripts by RT-PCR. As shown in Figure [1](#F1){ref-type="fig"}, Muc1, Muc2 and Muc5ac mRNAs were practically undetectable in sham-treated rats. In contrast, TiO~2~treatment markedly increased Muc5ac mRNA, but only modestly increased Muc2 mRNA. Muc1 mRNA was not seen in TiO~2~-treated rats. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The expression of Muc1, Muc2 and Muc5ac mRNA in TiO~2~treated rats. Rats were treated with TiO~2~, as described in Methods. Twenty-four hours after treatment, the levels of the Muc gene transcripts in the trachea were quantified using RT-PCR. GAPDH was used to ensure an equal loading of RNA samples. This figure is representative of 4 experiments. ::: ![](1465-9921-6-34-1) ::: The effect of TiO~2~instillation on Muc5ac-positive and PAS-positive epithelial cells in trachea ------------------------------------------------------------------------------------------------ Rats were given a single intratracheal instillation of saline or TiO~2~and the percentage of Muc5ac-positive (Muc5ac (+)) and PAS-positive (PAS (+)) epithelial cells were measured. At 24 h after saline instillation, almost no PAS (+) or Muc5ac (+) epithelial cells were found in the trachea (Fig. [2Aa, b](#F2){ref-type="fig"}). TiO~2~instillation, however, induced PAS (+) or Muc5ac (+) cells in the trachea at 24 h (Fig. [2Ac, d](#F2){ref-type="fig"}). The percentage of Muc5ac (+) cells was significantly higher at 24 hr (p \< 0.05) and further increased (Fig. [2B](#F2){ref-type="fig"}) in TiO~2~-- instilled rats and maintained until 72 h when compared with those of sham rats (p \< 0.01). The percentage of PAS (+) cells was very similar to that of Muc5ac (+) cells at 48 h after TiO~2~instillation (Figure [2B](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Light microscopic analysis of the trachea and the percentage of Muc5ac, PAS (+) epithelial cells. Rats were treated intratracheally with saline or TiO~2~, and the tracheas were prepared for morphometric analysis of PAS (+) and Muc5ac (+) cells as described in Methods. A. Histology of trachea 24 hr after saline or TiO~2~treatment. PAS (+) cells were stained red whereas Muc5ac (+) cells dark brown. Note that the trachea obtained from the saline-treated group contained little or no PAS (+) (Aa) or Muc5ac (+) cells (Ab) while the trachea from TiO~2~-treated group contains significant number of PAS (+) (Ac) and Muc5ac (+) cells (Ad). B. Time (4, 24, 48,72 h) dependent change in the percentage of Muc5ac (+) cells following saline (open bar, n = 8) or TiO~2~treatment (closed bar, n = 8). Note that the percentage of PAS (+) cells was similar to that of Muc5ac (+) cells at 48 hr after TiO~2~instillation. \* p \< 0.05, \\ p \< 0.01 as compared with the saline treated group. ::: ![](1465-9921-6-34-2) ::: Effects of cyclophosphamide and dexamethasone on the number of inflammatory cells and IL-13 levels in BAL fluid of TiO~2~-treated rats -------------------------------------------------------------------------------------------------------------------------------------- The numbers of eosinophils and neutrophils are markedly increased in the BAL fluids at 48 h after TiO~2~instillation when compared with those in saline-treated rats (p \< 0.05, respectively) (Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). Also, the levels of IL-13 in BAL fluids were significantly higher in TiO~2~-- treated rats than those of sham rats at 48 h after treatment (p \< 0.05) (Fig. [3D](#F3){ref-type="fig"}). Pretreatment with CPA prior to TiO~2~instillation significantly decreased the numbers of neutrophils and eosinophils in BAL fluids when compared with those in rats at 48 h after treatment with TiO~2~alone (p \< 0.05, Fig. [3A](#F3){ref-type="fig"} &[3B](#F3){ref-type="fig"}). Pretreatment with CPA, however, did not affect both the ratio of PAS (+) cells in the trachea and the IL-13 levels in BAL fluids of TiO~2~-treated rats (p \> 0.05, Fig. [3C](#F3){ref-type="fig"} &[3D](#F3){ref-type="fig"}). Pretreatment with DEX prior to TiO~2~instillation significantly decreased the number of eosinophils in BAL fluid (p \< 0.05, Fig. [3A](#F3){ref-type="fig"}), the ratio of PAS (+) cells in the trachea (p \< 0.05, Fig [3C](#F3){ref-type="fig"}) and the levels of IL-13 in BAL fluid (p \< 0.05, Fig. [4D](#F4){ref-type="fig"}) compared with those of rats instilled by TiO~2~alone. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The cell distribution in BAL fluid of TiO~2~instilled rats with or without pretreatment. Rats were pretreated with CPA (n = 6) or DEX (n = 6) and then treated intratracheally with TiO~2~. Saline (n = 8) or TiO~2~(n = 8) was treated without pretreatment. At 48 h post-treatment, BAL fluids were collected and analyzed for the numbers of eosinophils (A), neutrophils(B), and the levels of IL-13(D). PAS (+) cells (C) were measured in the trachea as described in Methods. \* p \< 0.05 as compared with saline -- treated group, † p \< 0.05 as compared with TiO~2~-- treated group. ::: ![](1465-9921-6-34-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The effects of cyclophosphamide (CPA) or dexamethasone (DEX) on the IL-13 (+) expressing cells. Rats were pretreated intratracheally with saline (Fig. A, E ; n = 8), CPA (Fig. C, E ; n = 6) or DEX (Fig. D, E ; n = 6) prior to treatment with TiO~2~Eight rats were treated with TiO~2~alone (Fig. B, E ; n = 8) as described in Methods. At 48 h post-treatment, IL-13 (+) cells are stained brown whereas toluidine blue (+) mast cells are stained dark purple. Note that saline -- treated group contained little or no IL-13 (+) cells (Aa) in spite of the presence of mast cells (Ab). TiO~2~-treated group showed significantly increasing numbers of mast cells when compared with sham group (E) and the mast cells (Ba) showed strong positivity for IL-13 protein (Bb). CPA pretreatment did not affect the TiO~2~induced-increase in the number of IL-13 (+) cells (Ca) or mast cells (Cb & E). On the other hand, DEX pretreatment significantly decreased the number of mast cells (Db & E) and reduced the IL-13 (+) cells (Da). \* p \< 0.05 as compared with saline treated group, † p \< 0.05 as compared with TiO~2~treated group. ::: ![](1465-9921-6-34-4) ::: Effects of cyclophosphamide and dexamethasone on the number and IL-13 expression of mast cells in TiO~2~-treated rats --------------------------------------------------------------------------------------------------------------------- Toluidine blue -- stained mast cells were observed in and around the muscle layer of the trachea in saline-treated rats. The shape of the cells was relatively round with a single nucleus and a large cytoplasm containing granules (Fig. [4Ab](#F4){ref-type="fig"}). In TiO~2~-instilled rats, some mast cells showed an elongated and branching shape of the cytoplasm (Fig. [4Bb](#F4){ref-type="fig"}). The trachea of the saline-treated group contained no IL-13 (+) cells (Fig. [4Aa](#F4){ref-type="fig"}) in spite of the presence of mast cells (Fig. [4Ab](#F4){ref-type="fig"}). TiO~2~-instilled rats increased the number of mast cells when compared with the saline control group (p \< 0.05, Figs. [4Bb](#F4){ref-type="fig"} and [4E](#F4){ref-type="fig"}). Serial section slides of the trachea showed that IL-13 protein was expressed exclusively on the mast cells in TiO~2~-- treated rats (Fig. [4Ba](#F4){ref-type="fig"}). CPA pretreatment did not affect the TiO~2~-induced increase in the number of toluidine blue (+) mast cells positive for IL-13 (p \> 0.05, Fig. [4Ca, 4Cb](#F4){ref-type="fig"} &[4E](#F4){ref-type="fig"}). However, DEX pretreatment significantly decreased the number of toluidine blue (+) mast cells expressing IL-13 compared to those of TiO~2~-- treated rats (p \< 0.05, Fig. [4Da, 4Db](#F4){ref-type="fig"} &[4E](#F4){ref-type="fig"}). The correlation between the number of IL-13 expressing mast cells, the concentration of IL-13 in BAL and Muc 5ac positive epithelial cells in the airway -------------------------------------------------------------------------------------------------------------------------------------------------------- The number of mast cells in the trachea was significantly correlated with percentage of Muc5ac (+) epithelial cells and concentration of IL-13 in BAL fluid of TiO~2~-- treated (n = 7) and sham (n = 6) rats (p \< 0.001 and p \< 0.0001, respectively, Table [1](#T1){ref-type="table"}). However, the number of eosinophil and neutrophils in BAL fluids were correlated with neither the percentage of Muc5ac (+) epithelial cells nor the concentration of IL-13 in BAL fluid (p \> 0.05, Table [1](#T1){ref-type="table"}). In addition, there were significant correlations of IL-13 (+) rate of mast cells in the trachea with IL-13 concentration in BAL fluid (r = 0.782, p \< 0.01, Fig. [5A](#F5){ref-type="fig"}) and with the percentage of Muc5ac (+) cells in the sham and TiO~2~treated rats (r = 0.604, p \< 0.05, Fig [5B](#F5){ref-type="fig"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The correlation of Muc5ac(+) cells or the IL-13 concentration with the number of inflammatory cells. The correlation between percentage of Muc5ac (+) epithelial cells or concentration of IL-13 in BAL fluid and number of eosinophil, neutrophil and mast cell in sham (n = 6) and TiO~2~-- instilled rats (n = 7). ::: Correlation (ρ) Eosinophils No. in BAL fluid Neutrophils No. in BAL fluid Mast cells No. in trachea ------------------------------------- ------------------------------ ------------------------------ --------------------------- \% of Muc5ac (+) 0.156 (p = 0.549) -0.195 (p = 0.438) 0.813 (p = 0.001\) Concentration of IL-13 in BAL fluid 0.447 (p = 0.138) 0.193 (p = 0.57) 0.903 (p = 0.0001\\) \ p \< 0.05, \\ p \< 0.01 ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### The correlation of the IL-13(+) mast cells with Muc5ac(+) epithelial cells and the IL-13 concentration. The percentage of IL-13 (+) mast cells was correlated with concentration of IL-13 in BAL fluid (r = 0.782, p \< 0.01) and the percentage of Muc5ac (+) cells (r = 0.604, p \< 0.05) (open circle; sham, open square; TiO~2~-- instilled rats). ::: ![](1465-9921-6-34-5) ::: Discussion ========== Although air pollution contains heavy metallic environmental particles that increases morbidity and mortality of the patients with chronic airway diseases \[[@B4],[@B5]\], the underlying mechanisms of mucus hyperproduction causing airway obstruction has not been revealed in detail. In this study, we demonstrated that a single instillation of TiO~2~is able to induce GCH within 24 h. The TiO~2~-induced GCH is associated with a dramatic increase in Muc5ac gene and protein expression in the present study (Figure [1](#F1){ref-type="fig"} &[2](#F2){ref-type="fig"}). Up regulation of Muc5ac gene in TiO~2~-- induced GCH is thought to be a common pathway in the process of GCH because MUC5AC has been demonstrated to be a major MUC gene during the process of GCH observed in the other non-particulates experimental model of airway diseases \[[@B22]-[@B25]\] and the asthmatics \[[@B26]\]. GCH is known as associated with airway inflammation and can be experimentally induced by various inflammatory agents such as LPS \[[@B22]\], neutrophil elastase \[[@B27]\], cathepsin B \[[@B23]\], IL-4 \[[@B25]\], IL-9 \[[@B28]\], and IL-13 \[[@B29],[@B30]\]. The exact mechanism of GCH, however, may differ in the experimental models. Neutrophils or eosinophils have been implicated in the induction of GCH in some animal models \[[@B30],[@B31]\]. Neutrophils and eosinophils depleted rats using CPA or specific antibodies inhibit granulocyte in agarose plug-induced and IL-13-induced GCH model \[[@B29],[@B31]\]. The epidermal growth factor receptor cascades are showed to be involved in underlying mechanism of the neutrophils -- induced GCH \[[@B29],[@B31]\]. However, in the present study we showed that depletion of these inflammatory cells by pretreatment with CPA similar dose used in the previously study \[[@B29],[@B31]\] did not prevent TiO~2~-induced GCH (Figure [4](#F4){ref-type="fig"}). Because cyclophosphamide effectively suppressed the number of neutrophils and eosinophils in peripheral blood (data not shown) and airways in the present study although not complete (Figure [4](#F4){ref-type="fig"}), our data indicates that these inflammatory cells may be not responsible for the TiO~2~-induced GCH. The dissociation of GCH from airway eosinophilia has been well documented in murine asthma models, in which anti-IL-5 (TRFK-5) \[[@B32]\], or IL-5 deficiency \[[@B33]\] reduced airway eosinophilia without affecting the induction of GCH. Therefore, depending on the experimental models investigated, the induction of GCH may not require neutrophils and eosinophils. Furthermore, IL-13 is known to induce GCH without any help of other inflammatory cells \[[@B24]\] and has been clearly shown to play a single, common pathway by which GCH is induced by CD4+ cells and IL-9 \[[@B34]\]. This process needs IL-4 receptor alpha, but not IL-4 or IL-5 \[[@B33],[@B34]\]. These data suggested a possibility that IL-13 is also involved in the particle -- induced GCH. In the present study, the levels of IL-13 in BAL fluids increased after TiO~2~instillation concomitantly with the development of GCH and the increase of IL-13 was completely abolished by pretreatment with DEX (0.25 mg/Kg), but not by that with CPA (Figure [4](#F4){ref-type="fig"}). These results suggest that the elevation of IL-13 may be associated with particles such as TiO~2~-induced GCH without any assistance of neutrophils or eosinophils. The in vivo effect of dexamethasone has been also demonstrated in allergic asthma model \[[@B35]\]. Dexamethasone (4 mg / kg) effectively abolishes allergic airway inflammation in mice by suppression of IL-13 m-RNA and protein expression \[[@B35]\]. The exact biochemical mechanism of GCH induction by IL-13 is not fully understood. One possible explanation is that IL-13 converts the bronchial epithelium from an absorptive to a secretory phenotype through loss of an amiloride-sensitive current and an increase in calcium-sensitive apical anion conductance \[[@B36]\]. The increase in apical anion conductance in the airway epithelium is most likely due to the ability of IL-13 to induce expression of hCLCA1/mCLCA3, which encodes a calcium-activated chloride channel. This channel is necessary and sufficient for the development of GCH and mucus hypersecretion in some experiments \[[@B37]\]. Besides Th2 cells, IL-13 is produced by mast cells, eosinophils \[[@B38],[@B39]\], and macrophages \[[@B40]\]. Since IL-13 was not decreased in rats of which eosinophils depleted by pretreatment of CPA (Figure [4](#F4){ref-type="fig"}), we can exclude eosinophils as the source of IL-13. Interestingly, serial thin section slides revealed that the IL-13 positive cells are mast cells, as shown by staining with toluidine blue. Also, we found the significant correlation between the IL-13 (+) rate of mast in tissue, concentration of IL-13 in BAL fluid and Muc5ac positive cells (Figure [5](#F5){ref-type="fig"} and table [1](#T1){ref-type="table"}). Based on these data, mast cells may be the cellular source for IL-13 present in the airways of TiO~2~-treated rats. It is well known that mast cells produce IL-13 when stimulated with antigen \[[@B39]\] and that the synthesis can be suppressed by dexamethsone \[[@B20]\]. Our finding showed that TiO~2~instillation increased the numbers of IL-13 expressing mast cells and Muc5ac (+) goblet cells, both of which were decreased by dexamethsone pretreatment is a novel finding to our knowledge. It is not known whether TiO~2~-- induced IL-13 overproduction is specific to TiO~2~or generally related to other particulates. However, base on the findings of particles such as diesel exhaust particles or carbon black particle -- induced the deviation to Th2 environment in antigen sensitized lung \[[@B11],[@B12]\], TiO~2~-- induced GCH via over production of IL-13 may be a general finding attributed to the particulate matters, but it remains unproven. Conclusion ========== We demonstrated that a single intratracheal instillation of TiO~2~particles induces GCH and Muc5ac gene expression within 24 h in rats, and that this process may be associated with elevated amount of IL-13 derived from mast cells. The present study may provide experimental evidences to support that patients with chronic airway disease may aggravate their symptoms and airway functions in the heavily polluted environment of particulate matters. Acknowledgements ================ The authors are indebted to Hwan-man Shin, Myong-ran Lee, and Eun-young Kim for their excellent animal care and technical support throughout the study. The authors express thanks to at least two professional editors, both native speakers of English for their kind editing for grammar and topographic error <http://www.textcheck.com>. This work was supported by a grant from the Korea Health 21 R&D Project, Ministry of Health and Welfare, Republic of Korea (01-PJ3-PG6-01GN04-003).	PubMed Central	2024-06-05T03:55:56.223781	2005-4-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087883/", "journal": "Respir Res. 2005 Apr 13; 6(1):34", "authors": [ { "first": "Mi-Hyun", "last": "Ahn" }, { "first": "Chun-Mi", "last": "Kang" }, { "first": "Choon-Sik", "last": "Park" }, { "first": "Sang-Jun", "last": "Park" }, { "first": "Taiyoun", "last": "Rhim" }, { "first": "Pyeong-Oh", "last": "Yoon" }, { "first": "Hun Soo", "last": "Chang" }, { "first": "Soo-Ho", "last": "Kim" }, { "first": "Hiroko", "last": "Kyono" }, { "first": "Kwang Chul", "last": "Kim" } ] }
PMC1087884	Background ========== In recent years, several population-based studies -- e.g. in Europe or the USA \[[@B1]-[@B5]\] -- have, by applying both classical epidemiological and molecular strain-typing techniques, revealed a high frequency of transmission of tuberculosis (TB), even in countries with a low TB incidence. Mycobacterium tuberculosisstrains with DNA fingerprint patterns that are identical in respect of the insertion sequence IS6110indicate possible transmission chains, and on the basis of this it has been concluded \[[@B6]\] that recently transmitted infections with rapid progress to active TB seem generally to play an important part in the spread of TB. However, there is a lack of information on the contribution of recent transmission to occupationally acquired TB. Up to now, only a few studies have been performed that have applied modern molecular DNA-fingerprint techniques capable of tracing directly routes of transmission attributable to occupational exposure, e.g. among health-care workers \[[@B5],[@B7]-[@B9]\]. For Germany, no prospectively acquired data on this topic have been available up to now. The incidence of TB in Germany is relatively low and steadily decreasing, with a total of 7866 new cases reported to the Robert Koch Institute in 2001 (9.6 per 100,000 inhabitants \[[@B10]\]). However, in the city of Hamburg (with its 1.7 million residents the second largest city in Germany) the overall incidence was 16.3 cases per 100,000 in 2001 and, against the national downward trend, is currently rising \[[@B11]\]. In 2001, as in previous years, Hamburg had a TB incidence rate higher than in any of the other fifteen German federal states. In order to identify the pathways of TB transmission and to determine the predictors of clustering of identical isolates in this metropolis, a long-term, prospective, population-based molecular-epidemiological study has been in progress in Hamburg since 1 January 1997 and is ongoing. As a preliminary result of the first three study years (1997--1999) referring to a sample of only 398 culturally proven TB patients it could be shown that conventional contact tracing prior to IS6110RFLP analysis by far underestimates the amount of recent TB transmission in a metropolis like Hamburg. In Germany -- as well as in most other European countries -- profession is not declared when TB disease is reported; thus the excess risk for health-care workers over members of the general population is unknown. Because precise data on the occupational risk of TB infection in low incidence settings are urgently needed for the development of better-aimed TB control, this study -- now comprising more than twice as many patients within an observation period of six years -- should help to evaluate the risk of recently transmitted disease in health-care settings. Methods ======= Study population ---------------- The study includes all patients in Hamburg in whom M. tuberculosiscomplex is confirmed by culture (\"definite cases\" \[[@B12]\]) as reported to each of the seven district public health departments from 1 January 1997 with a cut-off for the present analysis of 31 December 2002. Case data were collected prospectively by trained public-health staff using a standardised questionnaire. By interviewing each patient, information was obtained on: the patient\'s sex, country of birth, nationality, immigration status (if applicable), current address (or whether the patient was homeless), the nature of the patient\'s current employment (if any), details of any previous known exposure to other persons with tuberculosis. The subgroup of health-care workers refers to all the paid and unpaid persons working in health-care settings and potentially exposed to M. tuberculosis\[[@B13]\]. To acquire clinical data, the following were also included in the questionnaire: date of first onset of illness, date and reason for the diagnostic investigation, latency due to the patient\'s delay in seeking medical help, associated medical problems (especially HIV infection), chest radiographic findings, results of microbiological analyses and presence of alcoholism (according to the WHO-ICD 10-classification \[[@B14]\]). IS6110-RFLP analysis ---------------------- Extraction of DNA from mycobacterial strains and DNA fingerprinting using IS6110as a probe were performed according to standard protocols \[[@B15],[@B16]\]. The IS6110fingerprint patterns of mycobacterial strains were analysed by using the Bionumerics software (Applied Maths, Kortrijk, Belgium) as described previously \[[@B17]-[@B19]\]. Clusters were defined as groups of patients with M. tuberculosisstrains showing identical RFLP patterns. Patients with isolates with fewer than five bands were not included in this study. Statistical analysis -------------------- All patients included were classified into two groups, \"clustered\" and \"non-clustered\". Categorical data were compared by the chi-square test (or Fisher\'s exact test). Wilcoxon\'s rank sum test was performed to determine whether the distribution of treatment duration as a continuous variable differed between two groups. All tests were performed as two-sided tests. Pvalues below 0.05 were considered significant. Potential confounders and statistics ------------------------------------ Odds ratios (OR) and 95% confidence intervals (CI) were calculated by using logistic regression analysis. As tuberculosis patients aged less than 16 years (n = 22) could not contribute to the analysis of occupationally transmitted tuberculosis, they were excluded from the analysis of the relationship between specific outcome variables and clustering. The following potential confounders were judged separately with respect to their influence on the OR for health-care work: age; sex; AIDS; foreign birth; unemployment; homelessness; alcohol abuse; drug addiction; previous history of tuberculosis; sputum smear positivity; latency period between first symptoms and TB diagnosis; history of contact tracing. Variables were included in the final regression model for health-care work if they changed the odds ratio for health-care work by 10% or more. Therefore, we kept the following confounders in the final regression model for health-care work: foreign birth; unemployment; alcohol abuse; drug addiction; the latency period between first symptoms and TB diagnosis. The OR for variables other than health-care work were adjusted for the same variables that had been included in the final regression model for health-care work. For variables other than health-care work, the selection of confounders started with the final regression model for health-care work. Afterwards, for each variable other than health-care work, the following potential confounders were judged separately with respect to their influence on the odds ratio: age; sex; AIDS; homelessness; previous history of tuberculosis; sputum smear positivity; history of contact tracing. Non-health-care variables were included in the regression model if they changed the odds ratio for the specific outcome variable by 10% or more. According to this selection strategy, the ORs for homelessness and previous history of TB were additionally adjusted for history of contact tracing; the OR for drug addiction was additionally adjusted for AIDS. Results ======= Up to 31 December 2002, 895 cases of pulmonary and 66 cases of non-pulmonary tuberculosis reported to the public health offices were identified as culture-positive for M. tuberculosiscomplex. Culture isolates from 848 patients\' isolates were available for RFLP fingerprinting (88.2%). Their relevant characteristics are given in Table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Univariate analysis of risk factors for patients belonging to IS6110RFLP clusters ::: Characteristic Non-cluster group Cluster group All patients *p* ------------------------------------------ ----------------------- ------------------- ------------------ --------- Age (yr) mean ± SD 44.9 ± 19.7 43.2 ± 17.2 44.4 ± 18.9 n.s. AIDS (n, %) 27 (4.8) 21 (7.3) 48(5.7) n.s. Resistance to any drug (n, %) 52 (9.3) 11 (3.8) 63 (7.4) 0.005 MDR (n, %) 6 (1.1) 4 (1.4) 10 (1.2) n.s. Foreign-born (n, %) 270 (48.0) 96 (33.6) 366 (43.2) \<0.001 Male (n, %) 335 (59.6) 190 (66.4) 525 (61.9) n.s. Female (n, %) 227(40.4) 96 (33.6) 323 (38.1) n.s. Drug abuse (n, %) 30 (5.3) 40 (14.0) 70 (8.3) \<0.001 Alcohol abuse (n, %) 80 (14.2) 113 (39.5) 193 (22.8) \<0.001 Homelessness (n, %) 41 (7.3) 29 (10.1) 70 (8.3) n.s. History of contact tracing (n, %) 42 (7.5) 70 (24.5) 112 (13.2) \<0.001 Site of tuberculosis - Pulmonary (n, %) 513 (91.3) 268 (93.7) 781 (92.1) n.s. - Extrapulmonary (n, %) 49 (8.7) 18 (6.3) 67 (7.9) n.s. - Cavitary disease (n, %) 109 (19.4) 93 (32.5) 202 (23.8) \<0.001 Sputum smear positivity (n, %) 170 (30.2) 118 (41.3) 288 (34.0) 0.001 Previous history of Tb (n, %) 62 (11.0) 43 (15.0) 105 (12.4) n.s. Diagnosis due to symptoms (n, %) 492 (87.5) 186 (65.0) 678 (80.0) \<0.001 Unemployment (n, %) 108 (19.2) 163 (57.0) 271 (32.0) \<0.001 Contact persons with disease (n, %) 5 (8.9) 60 (21.0) 65 (7.7) \<0.001 Diagnosis due to contact tracing (n, %) 8 (1.4) 30 (10.5) 38 (4.5) \<0.001 Health-care worker (n, %) 2 (0.4) 8 (2.8) 10 (1.2) \<0.004 Latency from first symptoms to diagnosis \< 6 months (n, %) 444 (79.0) 164 (57.3) 608 (71.7) \<0.001 \>= 6 months; \< 12 months (n, %) 98 (17.4) 79 (27.6) 177 (20.9) \<0.001 \>= 12 months (n, %) 20 (3.6) 43 (15.0) 63 (7.4) \<0.001 n.s., not significant ::: The average age was 44.4 ± 18.9 years (mean ± SD; Table [1](#T1){ref-type="table"}). The majority of patients were male (525/848; 61.9%). In the study population, 366 patients (43.2%) were born outside Germany; 48/848 (5.7%) showed clinical symptoms of HIV infection (AIDS); 70 (8.3%) were homeless at the start of the study; 193 (22.8%) appeared to be alcohol-dependent; 70 (8.3%) regularly injected drugs of abuse; 105 (12.4%) had a past history of TB; 112 (13.2%) had been previously identified as contact persons of infectious TB sources; 63 (7.4%) showed resistance to antituberculotic drugs, but of these only 10 (1.2%) cases were reported with resistance to at least INH and RIF, and were thus multidrug-resistant. Two hundred and eighty-six patients (33.7%) shared an identical RFLP pattern with one or more other patients and were thus classified, as clustered cases, into 76 clusters ranging in size from 2 to 39 persons. Among the clusters, 39 (51.3%) comprised only two patients each, thus representing 78/146 (53.4%) of the cluster patients. Characterisation of cluster patients and identification of risk factors associated with clustering -------------------------------------------------------------------------------------------------- In order to identify significant differences between the 286 patients in clusters and the 562 patients not in clusters, univariate analyses were performed. The results for the different variables analysed are summarised in Table [1](#T1){ref-type="table"}. There was no significant difference in age distribution between the two groups, as determined by Wilcoxon rank sum test (p= 0.63). Chi-square tests showed that the patients in clusters were more likely than the non-clustered patients to be drug abusers (p\< 0.001), to be alcoholics (p\< 0.001), to be health-care workers (p= 0.004), to be unemployed (p\< 0.001), to be sputum-smear positive (p= 0.001) and to have a cavitary disease (p\< 0.001), and they had more known contacts to patients with active tuberculosis (p\< 0.001). A history of involvement in the contact tracing of tuberculosis patients was also significantly associated with clustering (p\< 0.001). In the \"cluster\" group fewer patients were diagnosed because of symptomatic disease (p\< 0.001), but more were found through contact tracing, irrespective of symptoms (p\< 0.001). Drug resistance was more common in patients not found in a cluster (p= 0.005), and the proportion of foreign-born patients was smaller among the clustered patients (p\< 0.001). The latency period between the first onset of symptoms and confirmed diagnosis was far longer in cluster patients (p\< 0.001). In multivariate analyses (Table [2](#T2){ref-type="table"}), the OR for employment as a health-care worker was found to be the strongest predictor (OR = 17.9; CI 3.6--89.3), followed by a latency period between the first onset of symptoms and confirmed diagnosis of at least 12 months (OR = 7.5: CI 3.7--13.9) or of at least 6 months (OR = 6.0; CI 3.4--10.4), unemployment (OR = 5.2; CI 3.6--7.5) and a history of contact tracing (OR = 2.8; CI 1.7--4.6). Alcoholism -- the leading predictor in our first evaluation three years ago \[[@B6]\] -- represented a further considerable risk of being in a cluster (OR = 2.6; CI 1.7--3.8), whereas drug addiction (OR = 1.7; CI 1.0--3.1), sputum smear positivity (OR = 1.2; CI 0.9--1.8), AIDS (OR = 0.8; 0.4--1.8), male sex (OR = 1.1; CI 0.6--1.8), foreign origin (OR = 0.5; CI 0.3--0.7) and age (see the results of several groups; Table [2](#T2){ref-type="table"}) were not significant independent risk factors. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Health-care work and clustering (multivariate analysis). ::: Variable Non-clustered TB Clustered TB Crude OR 95% CI Adj. OR^b^ 95% CI ---------------------------------------------------- ---------------------- ------------------ -------------- ------------ ---------------- ------------ ------ ----------- Health-care worker No 545 99.6 269 97.1 1.0 \- 1.0 \- Yes 2 0.4 8 2.9 8.1 1.7--38.4 17.9 3.6--89.3 AIDS No 520 95.1 256 92.4 1.0 \- 1.0 \- Yes 27 4.9 21 7.6 1.6 0.9--2.8 0.8 0.4--1.8 Foreign-born No 284 51.9 185 66.8 1.0 \- 1.0 \- Yes 263 48.1 92 33.2 0.5 0.4--0.7 0.5 0.3--0.7 Sex Female 221 40.4 90 32.5 1.0 \- 1.0 \- Male 326 59.6 187 67.5 1.4 1.0--1.9 1.1 0.8--1.6 Drug abuse No 517 94.5 237 85.6 1.0 \- 1.0 \- Yes 30 5.5 40 14.4 2.9 1.8--4.8 1.7 1.0--3.1 Alcohol abuse No 468 85.6 164 59.2 1.0 \- 1.0 \- Yes 79 14.4 113 40.8 4.1 2.9--5.7 2.6 1.7--3.8 Homelessness No 510 93.2 249 89.9 1.0 \- 1.0 \- Yes 37 6.8 28 10.1 1.6 0.9--2.6 0.9 0.5--1.7 History of contact tracing No 505 92.3 211 76.2 1.0 \- 1.0 \- Yes 42 7.7 66 23.8 3.8 2.5--5.7 2.8 1.7--4.6 Sputum smear positivity No 377 68.9 161 58.1 1.0 \- 1.0 \- Yes 170 31.1 116 41.9 1.6 1.2--2.2 1.2 0.9--1.8 Previous history of TB No 485 88.7 234 84.5 1.0 \- 1.0 \- Yes 62 11.3 43 15.5 1.4 0.9--2.2 1.7 1.0--2.7 Unemployment No 441 80.6 114 41.2 1.0 \- 1.0 \- Yes 106 19.4 163 58.8 5.9 4.3--8.2 5.2 3.6--7.5 Latency period from first symptoms to TB diagnosis \< 6 months 432 79.0 156 56.3 1.0 \- 1.0 \- ≥ 6 months; \<12 months 95 17.4 78 28.2 2.3 1.6--3.2 6.0 3.4--10.4 ≥ 12 months 20 3.7 43 15.5 2.6 1.7--3.8 7.5 3.7--13.9 ^a^TB patients \< 16 years (n = 22) excluded from the analysis ^b^Adjusted for foreign birth; unemployment; alcohol abuse; drug addiction; and the latency period between first symptoms and TB diagnosis; the odds ratios for homelessness and previous history of TB are additionally adjusted for history of contact tracing; the odds ratio for drug addiction is additionally adjusted for AIDS ::: Epidemiological analysis of clustered cases and efficiency of classical contact tracing --------------------------------------------------------------------------------------- As in our preliminary study \[[@B6]\], in addition to acquiring data from standardised patient questionnaires, we also performed intensive epidemiological investigations, including additional interviews for patients within clusters. Recent transmission, verified by epidemiological relationships, could be confirmed for 146 of the 286 clustered patients (51.0%) in 35 of the 76 clusters (46.1%). Of these 146 patients, 43 were source cases suffering from tuberculosis, labelled as index persons for the purpose of the investigation and identified by earlier onset of disease, and 103 were infected by these persons, with onset of disease during the study period (Figure [1](#F1){ref-type="fig"}); 50 of the 103 (49%) had already been reported as contact persons to the public health bureaus. Only 20 of these 50 (40%) had been identified by contact tracing procedures; a further 15 (30%) had independently sought medical attention because of symptoms, and the remaining 15 (30%) were identified by other means (screening as part of the asylum procedures, or diagnostic measures performed because of other diseases). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Distribution of cases discovered by contact tracing and alternative sources ::: ![](1465-9921-6-35-1) ::: Fifty-three patients could only be identified retrospectively, by RFLP fingerprinting, as contact persons of their respective index cases. Of these, six were discovered to be tuberculosis patients through contact tracing (owing to index persons other than their respective index cases), 36 by their symptomatic disease and 11 as a consequence of other medical examinations. In total, conventional investigation of the patients\' contacts conducted before RFLP typing identified only 26 (25.2%) of the 103 clustered patients who had became ill between exposure and contact tracing as contact persons with confirmed epidemiological links. Conversely, contact tracing led to the discovery among the cluster patients of four TB patients with no known epidemiological connection to the other members of their cluster, and also to the fortuitous discovery of eight infected persons who had been named as contact persons but were not members of any cluster. This means, remarkably, that contact tracing based on retrospectively incorrect information in terms of molecular-epidemiological links led to the detection of almost one-half (12/26) as many cases of disease as the contact tracing that was (correctly) rooted in an epidemiological context. Eight (5.5%) of the 146 cluster patients with an epidemiological connection to others in the study population were employees in the health-case sector, as will be described below. Apart from these cases there were two other patients with sporadic TB, i.e. due to reactivation rather than to recent transmission, not cluster members; both were nurses (a 28-year-old Rumanian and a 52-year-old Russian) who might have been infected as children in their country of origin. Description of the cluster relationships ---------------------------------------- Out of 10 health-care workers in this study 8 were in the cluster group. A short description of these cases is given below: ### Cluster A In August 1996, a 32-year-old homeless alcoholic in the intensive-care ward of hospital 1 was diagnosed as having sputum-positive, progressive tubercular pneumonia. He died of multiple causes seven weeks after admission. He had named four close contact persons from the time before his admission; none of these developed disease. Eight months later, in March 1997, a 27-year-old nurse working in the same intensive-care ward was diagnosed as having tubercular pleuritis. In a routine examination of staff, which took place shortly before the admission of the patient who later died, the result of a Mendel-Mantoux intracutaneous test (10TU) had been negative. The cluster analysis revealed the causal relationship: the cluster contained only these two patients. No connection between the deceased patient and the infection of the nurse -- a classic instance of fresh transmission -- had been made, either by the physician responsible for the staff or by the public health authority in the search for the source of infection. Only afterwards was the occupational health investigation required by German law (BK3101) carried out. ### Cluster B On 9 September 1997, a 34-year-old unemployed alcoholic, admitted to hospital 2 for investigation of pneumonia of unknown origin, was found to have sputum-positive pulmonary tuberculosis. He had already been examined in three separate contact tracings between 14 August 1995 and 21 February 1996, as a close contact person of another unemployed alcoholic with known TB, but the result had each time been negative. He was nursed between 9 September and 16 September 1997 by a 33-year-old male nurse, who seven months later became ill with tubercular pleuritis. The required occupational health investigation was initiated, and the alcoholic patient could be assumed to be the most likely source of disease before the result of the RFLP analysis became known. However, it was found that the cluster was not restricted to these two cases; on the contrary, by September 2002 its known membership had increased to nine. Of the remaining seven members, all with pulmonary TB, six were alcoholics and frequented various bars, although no epidemiological connection between these could be established. The seventh was a Rumanian prostitute had lived in Hamburg illegally for only a few months and who had presumably been infected by a client. ### Cluster C A 67-year-old ear, nose and throat (ENT) specialist became ill in May 2002 with culturally confirmed pulmonary TB. In his view no connection with any former patient could explain this. RFLP analysis revealed a strain identical to that in a Nepali waiter who three years earlier, in May 1999 -- then 21 years of age -- had received out-patient treatment for earache from the same doctor. As he had only visited this doctor before the TB diagnosis and had not mentioned him as a contact person, the connection could only be established retrospectively. ### Cluster D An 86-year-old woman was admitted to hospital 3 with fever of unknown cause. During the routine diagnostic procedures, in early September 2002, a medical technician in the hospital laboratory pricked herself in the right index finger with a sharp metal object after withdrawing incubated material from a liquid-culture bottle. The infection was followed up in an occupational health investigation and, because of the clear causality, was registered as being work-related infection before the RFLP analysis was performed. On 12 September the patient was diagnosed as having urogenital TB. Shortly after this the technician developed a protracted skin granuloma that was found to be tubercular on 6 November. ### Cluster E In April 2000 a 73-year-old patient in the ENT department of hospital 4 received a tympanoplasty because of chronic otitis media. A smear test for TB bacteria was not performed at the time, and an infiltrate in the upper left pulmonary lobe, detected by X-ray, was not investigated. In early July 2000 the patient was admitted to the abdominal surgery ward of the same hospital with an abdominal aortic aneurism. During the pre-operative tests an enlargement of the infiltrate was detected, and two days later sputum-positive pulmonary TB was diagnosed. An ear smear was performed and also found to be TB-positive. Contact tracing was carried out retrospectively on the medical and nursing personnel who had regular contact with the patient. On 3 January 2001 a 21-year-old assistant nurse, who had given a negative tuberculin skin test in the autumn of 1999, now gave a positive tuberculin test and also tested culturally positive for pulmonary tuberculosis. ### Cluster F In late February 1998 a 26-year old homeless man with an i.v. drug addiction was admitted to hospital 5 with interstitial pneumonia and general respiratory failure. He was given mechanical respiration in the intensive-care unit. In November 1996 he was diagnosed as having an HIV infection (Category B according to the CDC classification). On 4 March 1998 sputum-positive pulmonary TB was diagnosed. In November 2000 the 55-year-old domestic responsible for disinfection of equipment in this ward began to experience increasing weight loss and general weakness. On 7 March 2001, sputum-positive pulmonary TB was found as well. An occupational cause was not suspected at first; only the retrospective investigation on the basis of an RFLP analysis showed that the domestic had shaken out the tubes of the respiration apparatus, which explained the fresh airborne infection. ### Cluster G On 30 March 2001 a 43-year-old male geriatric nurse fell ill with culturally confirmed pulmonary TB; initially, no occupational cause was suspected. He belonged to a cluster that comprised five members with no established epidemiological interrelationships; of the other four, three were Africans and one was the German-born daughter of a Nigerian, 30 months old and with pulmonary TB that had been culture-confirmed by a gastric-fluid sample on 15 July 2000. The source of infection of the girl was her father, who was suffering from AIDS-induced encephalopathy; his consequent mental confusion prevented a diagnosis of sputum-positive TB until long after diagnosis on 9 June 2000. The culture from the father could not be fingerprinted, so this patient was at first not considered in the cluster analysis, and only the correspondence between the strain of the daughter with that of the geriatric nurse prompted a revision of the medical history of the father. It was found that the nurse had, during practical training in the ward for general internal diseases at the hospital 6, looked after the father when the latter had been admitted with pneumonia of unknown origin. As the training period had ended before the father\'s TB was diagnosed, the geriatric nurse had not been included in the contact tracing. ### Cluster H In early January 1997, a 58-year-old homeless alcoholic was diagnosed as having sputum-positive pulmonary TB. Treatment was initiated but was repeatedly interrupted; the patient continued to drink and intermittently gave positive sputum cultures. By the end of 2002, eight additional cluster members had been identified, of whom two were known as direct contacts from the initial tracing, while the other seven belonged to the bar milieu of Hamburg\'s red-light district. On 3 September 1998, a hospital physician at hospital 7 fell ill with tubercular pleuritis. Contact between the index patient and the physician could only be established retrospectively: the two persons had been in contact for a brief period in March 1998 in the admissions ward, where the patient had been admitted for alcohol withdrawal. Discussion ========== The purpose of conducting this population-based epidemiological study was to determine which factors might influence TB transmission in the city of Hamburg, and how these contribute to the community epidemiology of tuberculosis. Available information about the risk of transmission to employees in the medical sector in other cities is at present sparse, and partly contradictory. Therefore, we especially evaluated the occupational risk of TB infection performed an in depth investigation of the transmissions identified. In the classical study by Small et al. \[[@B5]\], previous care in a tuberculosis clinic was included as a risk variable and was identified as the only significant, independent risk factor in patients aged 60 years or above (OR = 5.7). However, transmission appeared to occur only between patients; there was no evidence for recent transmission between hospital staff and patients. The frequently cited study by Sepkowitz et al. \[[@B9]\] distinguished between medical and non-medical staff, and concluded that the risk of a recent TB transmission for health-care workers was nearly three times as great as for others. However, this study was not based on the general population, as only patients from six hospitals -- classified as \"hospital staff\" or \"outside patients\" are compared with one another. The professions of almost 30% of the 201 patients were unknown and, among the 20 infected health-care workers, 8 of the 13 cluster members turned out to be HIV-positive. The conclusion of this study, which took place at the height of the New York HIV epidemic, \"that many of the apparently sporadic cases of tuberculosis among health-care workers may be due to unrecognised occupational transmission\", does not strictly follow from the study\'s data, and can in any case hardly be applied to European conditions. Van Deutekom et al. \[[@B7]\], in their population-based \"Amsterdam\" study, were able to assign 47% of 459 patients registered between 1 July 1992 and 1 January 1995 to clusters. They reached a conclusion opposite to that of Sepkowitz et al.: only 6 out of a total of 17 patients (8 Dutch, 9 immigrants) working in the health-care sector were cluster members. \"Health-care working\" thus emerged as a negative predictor of clustering. Lemaitre et al. \[[@B8]\] looked for nosocomial transmission in a Parisian teaching hospital. A total of 161 RFLP isolates were studied. Interestingly, only 5 (13%) of the 40 cluster patients but 34 (28%) of the 121 non-clustered patients had been admitted to hospital before the diagnosis of TB. None of the 40 members of, in all, 12 clusters showed evidence of transmission within the hospital. (However, the study period was relatively short: 1 March 1993 to 30 April 1995, with only sputum-positive patients being investigated up to 28 February 1994.) The results of the present study indeed confirm strikingly the (relative) impact of health-care employment on recent transmission. Of course, in view of the small number of cases of active disease we cannot conclude that there is a very high risk in acquiring TB in health-care settings. However, a highly significant association between health-care work and clustering was found (Table [2](#T2){ref-type="table"}). Most cases (except the employees in clusters B and D) remained unrecognised until the result of RFLP fingerprinting became available. A conspicuous observation in our study was that none of the infections took place in a chest clinic, but rather in general-internal or ENT wards, (in one case, involving an out-patient); thus, in none of these cases had precautions against airborne infection been taken -- e.g., an isolation room with negative pressure relative to the surroundings, or use of disposable splash-proofed masks (N95 or HEPA) by persons entering the room. Furthermore, transmission was always between an infected patient and a health-care worker, and never between patients; thus, there were no nosocomial outbreaks. The risk of TB infection for nursing-home employees is believed to be considerably higher in general -- a 1990 CDC study reported that it was as much as three times higher \[[@B20]\] -- than the rate found in other adult employees of similar age, race, and sex. For this reason, the new German law on infectious diseases (IfSG paragraph 36, section 4) that came into force on 1 January 2001 requires that new residents provide a medical certificate that they are free of TB. Because the law\'s protective effect on TB infection had been limited only to the last 12 months of our study, it was a considerable surprise that not one case of recent transmission within this sector was recorded in the present study, in spite of its long duration. It is well known that investigation of hospital contacts is often difficult to conduct because the movement of patients, and the changing work assignments of personnel, make it difficult to assess the nature and extent of contact between an index patient and staff members with actual contact \[[@B21]\]. As our broad, comparative cluster analysis shows, the principal problem is that at the beginning of a patient contact the possibility of a TB infection is not considered. This means that adequate account is not taken of the infectiousness of the index case and -- because the diagnosis comes too late -- a previous exposure will be forgotten, or only the people with the most frequent exposure will ultimately be identified. It is clearly insufficient to record retrospectively only the co-patients of an index person who shared the patient\'s room and the nurses or doctors who were directly assigned to care for him/her on the ward. Even employees with brief contact or -- as described above -- members of the cleaning staff responsible for decontamination may be exposed to infection spread by droplets. According to the classical finding of Wells and Riley \[[@B22]\] that a single viable TB bacillus, once inhaled, is, sufficient to produce infection, the risk of infection is not necessarily a function of duration of contact. Thus, the risk arising from a prolonged period of proximity need not be greater than that arising from a short phase of heightened contagiousness. In our study the risk of clustering is at least about six-fold higher in patients whose diagnosis of TB disease is confirmed more than six months after the beginning of symptoms. Thus, spreading of TB by infectious persons who remain unidentified for an unexpectedly long time may play an important role in TB transmission. The most crucial prerequisite for effective contact tracing is the verification that (a) a case of infectious TB is reported to the hospital concerned as promptly as possible, and (b) ideally, it should be a matter of routine to note, beforethe disease is diagnosed, as completely as possible who and in which period has looked after the index patient. As in the management of an outbreak, an organised and continuously supervised procedure is necessary, in order to establish the personal, spatial and temporal framework of contact. Above all, the need to obtain a detailed medical history from each patient has to be stressed, especially if the patient has received previous treatment for tuberculosis or belongs to defined high risk groups. Another reason why it is important to find every contact person is the converse one: not every RFLP cluster will demonstrate a recent epidemiological link \[[@B6],[@B23],[@B24]\]. Owing to the possibility that a portion of clustered patients are infected with circulating strains that are prevalent in a given community, spreading over a long period, or that were imported from endemic areas of the world developed an active TB by chance within the study period, mixed clusters are not uncommon. In these, cases of recent transmission are associated with cluster members between whom there is no epidemiological connection. Examples of this in the present study are clusters B and G, in which the respective transmission rates n-1 (i.e., cluster size minus 1) are not 8 and 4, but rather 1 and 0. However, one should take into account the fact that TB infection may occur among highly infectious source cases and their recipients through casual contacts. Thus, a conventional epidemiological investigation -- even if it is conducted as meticulously as possible -- is unlikely to be able to pick up all of the infected persons, possibly leading to an overestimate of the total proportion of cluster members between whom no epidemiological linkage is assumed. The predominance of health-care employment as an independent risk factor in this study might have important implications for future tuberculosis control in inner cities. The effectiveness of traditional contact tracing according to the \"stone-in-the-pond\" principle \[[@B25],[@B26]\] in social risk groups such as alcoholics, drugs addicts or homeless people is obviously largely dependent on their individual willingness to co-operate \[[@B27]\]. Thus, it is not surprising that in Hamburg only approximately 25% (26/103) of clustered patients with confirmed transmission links have been identified by conventional contact tracing up to now in our study. In health-care settings, however, the success of determining transmission links is -- as the examples of our clusters show -- largely associated with the quality of each case management as defined in current guidelines \[[@B28]\]. This implies that IS6110RFLP typing can help to define more precisely where the deficiencies are, and whose standards have to be adapted in awareness of possible exposure to transmission. In conclusion, the results of our study demonstrate that health-care workers -- even those outside high-risk-settings -- are at particular risk of recent TB transmission. Since most cases remained unrecognised, there is a need for improved strategies for contact tracing that avoid ineffective procedures and allow a better-targeted identification of cases. Continual prospective RFLP fingerprinting is essential to assess the efficiency of such new standards, and should therefore play an integral part in communal TB control. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Roland Diel: Conception and design of the study, acquisition, analysis and interpretation of data, drafting and revising of the article, giving final approval to the version to be published Andreas Seidler: Statistical analysis and interpretation of data, drafting and revising of the article, giving final approval to the version to be published Albert Nienhaus: Statistical analysis and interpretation of data, drafting and revising of the article, giving final approval to the version to be published Sabine Rüsch-Gerdes: Interpretation of data, drafting and revising of the article, giving final approval to the version to be published Stefan Niemann: Conception and design of the study, acquisition, analysis and interpretation of data, drafting and revising of the article, giving final approval to the version to be published Acknowledgements ================ The authors would like to thank K. Ott, B. Schlüter, I. Radzio, T. Ubben and P. Vock (Borstel, Germany) for excellent technical assistance. We thank the staff of the departments of Tuberculosis Control at the Hamburg Bureaus of Public Health, without whom this study would not have been possible. Parts of this work were supported by the Robert Koch Institute Berlin, Germany, and the EU Concerted Action project \"New generation genetic markers and techniques for the epidemiology and control of tuberculosis\" (QLK2-CT-2000-00630).	PubMed Central	2024-06-05T03:55:56.225926	2005-4-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087884/", "journal": "Respir Res. 2005 Apr 14; 6(1):35", "authors": [ { "first": "Roland", "last": "Diel" }, { "first": "Andreas", "last": "Seidler" }, { "first": "Albert", "last": "Nienhaus" }, { "first": "Sabine", "last": "Rüsch-Gerdes" }, { "first": "Stefan", "last": "Niemann" } ] }
PMC1087885	Background ========== Lung fibrosis is the final common and often irreversible pathway of different lung diseases, such as idiopathic interstitial pneumonitis (idiopathic pulmonary fibrosis) and granulomatous diseases (sarcoidosis) \[[@B1]-[@B3]\]. Though these diseases are different in their etiology, all are characterized by zones of lung injury where varying numbers of fibroblasts proliferate and contribute to the accumulation of extracellular matrix (ECM). Interstitial and intralumal deposition of connective tissue then disrupts the lung architecture and impairs respiratory function. Recent studies have shown that the development of lung fibrosis is accompanied by the differentiation of normal lung fibroblasts into myofibroblasts. These myofibroblasts express α-smooth muscle actin, and they are thought to be the major source of collagen and profibrogenic growth factors in the fibrosing lung \[[@B4]\]. Additionally, decreased apoptosis of these cells may contribute to the remodeling of lung tissue during chronic inflammation. Apoptosis is a physiological process that is highly selective in eliminating aged and injured cells. In addition to internal pathways that mainly trigger apoptosis in response to cytotoxic stress, apoptosis can also be induced by cell-membrane-anchored signaling pathways of the TNF-superfamily: the CD95-receptor/CD95-ligand-system (Fas/FasL or APO-1) and the tumor necrosis factor (TNF)-related apoptosis inducing ligand (TRAIL or APO-2L) with the TRAIL receptors 1 and 2 (TRAIL-R1 and R2) and the decoy receptors DcR1 (TRAIL-R3) and DcR2 (TRAIL-R4). TRAIL induces programmed cell death in many tumor cells, but not in normal, non-neoplastic cells \[[@B5]\]. The mechanisms through which stimulation of Fas by FasL initiate apoptosis have been extensively investigated. It is also known that mesenchymal, fibroblast-like cells express Fas. Alterations in the susceptibility of these cells to Fas-induced cell death contribute to the pathogenesis of lung fibrosis, \[[@B6],[@B7]\] and myofibroblasts are susceptible to the suppression of apoptosis by transforming growth factor-β1 (TGF-β1) \[[@B6]\] and resistant to interleukin (IL)-6-induced apoptosis \[[@B8]\]. However, the molecular mechanisms which regulate these alterations in resistance to proapoptotic signals, and thus contribute to decreased apoptosis of fibroblasts during chronic inflammation, are not known in detail. Apoptosis is regulated by a complex system consisting of numerous proteins and cascading proteolytic and phosphorylation steps. The contribution of isolated elements of the system to the regulation of apoptosis resistance is less well characterized. The binding of soluble or cell surface bound FasL with surface Fas may initiate apoptosis. Consequently, the intensity and the stochiometry of the Fas-FasL interaction could play a crucial role in the regulation of apoptosis. In this study we systematically investigated the expression and interplay of the Fas/FasL system in fibroblasts obtained from patients with and without lung fibrosis. We aimed to clarify the possible involvement of the Fas/FasL system in the survival of lung myofibroblasts and the development of lung fibrosis. Methods ======= Human tissues ------------- Tissue samples from patients with (n= 5) and without lung fibrosis (n= 6) were obtained from diagnostic open lung biopsies (fibrotic samples) and from healthy tissue areas during pneumonectomy for tumor resection (non-fibrotic samples). The fibrotic samples comprised the following diseases: usual interstitial pneumonia (UIP, two patients), non-specific interstitial pneumonia (NSIP, one patient), bronchiolitis obliterans-organizing pneumonia (BOOP, two patients) \[[@B3]\]. Although these diseases are different in clinical behavior and outcome, they are characterized by increased amounts of activated fibroblasts and increased matrix deposition. The histology of these tissue samples was recently partially described \[[@B9]\]. The lung fibroblasts were compared to synovial fibroblasts (n = 5), which were obtained from patients with osteoarthritis during joint replacement surgery. An experienced surgical pathologist (C.R.) examined the tissue specimens. All tissue samples were obtained immediately after surgery and used for the isolation of fibroblasts. The tissue sampling was approved by the local ethical committee. Characterization of fibroblasts by flow cytometry ------------------------------------------------- Fibroblasts were trypsinized. For extracellular staining, fibroblast-specific antibodies \[clones AS02 (anti-Thy-1) and D7-Fib; Dianova, Hamburg, Germany\], a macrophage-specific anti-CD68 (clone KP1; Signet Laboratories, Inc., Dedham, MA) and a pan-leukocyte anti-CD45 antibody were used. Cells were incubated with the primary antibodies for 30 min and with a fluorescein isothiocyanate (FITC)-labeled goat-anti-mouse IgG for 20 min. In addition, intracellular fluorescence staining was performed with anti-prolyl-4-hydroxylase antibodies (clone 5B5; DPC Biermann, Bad Nauheim, Germany) using the Fix and Perm reagent (Dianova) according to the instructions of the manufacturer. For the analysis, a FACSCalibur (Becton Dickinson, Heidelberg, Germany) flow cytometer was used. Cell culture ------------ Fibroblasts were obtained by mincing freshly excised lung parenchyma into \~1 mm^3^pieces, followed by digestion with collagenase IV (1 mg/ml, Sigma, Deissenhofen, Germany) for 30 min at 37 °C. Fibroblasts were cultured in a 75-ml tissue culture flask containing Iscove\'s modified Dulbecco\'s medium with 10% (w/v) fetal calf serum (FCS), 10^-3^M glutamine and antibiotics, at 37°C and 5% (v/v) CO~2~until they reached confluence. Only fibroblasts between passages 3 and 8 were used for the experiments. Determination of collagen and ECM deposition -------------------------------------------- Collagen secretion and deposition into the ECM was assessed by proline incorporation assays originally developed by Peterovsky and Diegelmann \[[@B10]\] and described in detail earlier \[[@B11],[@B12]\]. All assays were performed in triplicate. Briefly, 5 × 10^4^fibroblasts were seeded into 24-well plates (Falcon, Heidelberg, Germany) in culture medium containing 10 % FCS. After 16 h, the medium was changed to low serum medium (Dulbecco\'s modified Eagle\'s medium supplemented with 0.1% FCS, 100 μg/ml L-ascorbic acid) containing \[2,3,4,5-^3^H\]-lL-proline (2 μCi/ml, NEN, Boston, MA). When indicated, E64d was added (10 μM). After 72 h, the culture medium was removed and the remaining fibroblasts were lysed with distilled water (10 min, room temperature). The ECM was ethanol fixed (70% ethanol, 15 min, RT). Half of the wells were incubated with 30 U/ml collagenase (Clostridium histolyticum, Sigma, Deissenhofen, Germany) in collagenase assay buffer (50 mM Tris-HCL, pH 7.5, 5 mM CaCl~2~, 2.5 mM N-ethylmaleimide) for 4 h at 37°C. The remaining wells were incubated with assay buffer. The supernatants were removed and residual ECM was solubilized by overnight incubation in 0.3 M NaOH-1% SDS. Equal numbers of aliquots of supernatants obtained after collagenase digestion and supernatants containing the residual ECM were subjected to liquid scintillation counting. The counts measured in supernatants after collagenase treatment represent the collagen content. The amount of \[^3^H\]proline measured after solubilization of the remaining ECM represents non-collagenous ECM. The total of both counts was equal to the counts from solubilized ECM without collagenase treatment and represents the total proline incorporation. Relative ECM synthesis can be calculated by the established formula \[[@B12]\]: ECM = CPM in collagen + (5.4 × CPM in non-collagen ECM). The formula contains the factor 5.4 to correct for the 5.4-fold higher proline or hydroxyproline content of collagens compared with that of other proteins. Induction and detection of apoptosis ------------------------------------ Fibroblasts were stimulated with 100 ng/ml recombinant human FasL for 16 h as described \[[@B13]\]. When indicated, cells were preincubated with TNF-α or cycloheximide (100 μg/ml) for 24 h. Subsequently, apoptosis was determined using a histone fragmentation assay (Cell Death Detection ELISAPlus, Roche Diagnostics, Mannheim, Germany) according to the manufacturer\'s instructions. This assay is based on a quantitative sandwich-enzyme-immunoassay using mouse monoclonal antibodies against DNA and histones that allow for the specific, quantitative determination of cytoplasmatic histone-associated-DNA-fragments (mono- and oligonucleosomes) in the cell lysates. The ELISA plates were read at 405 nm (490 nm reference). We have shown before that the results obtained using this assay correlate to the amount of apoptotic cells found after TUNEL staining \[[@B14]\]. Additionally, apoptosis was measured in lung fibroblasts using TUNEL staining (ApoBrdU kit, Pharmingen, Heidelberg, Germany). Briefly, following induction of apoptosis, cells were fixed in 1% paraformaldehyde and incubated with Br-dUTP in the presence of TdT enzyme, which results in the incorporation of Br-dUTP into exposed 3-OH DNA ends. Br-dUTP sites were then labeled with FITC-conjugated anti-Br-dUTP antibodies. The number of apoptotic cells was measured using flow cytometry (FACS Calibur, Becton Dickinson), and labeling with Br-dUTP was compared with that of unstimulated controls. Measurement of Fas/CD95 mRNA ---------------------------- Expression levels of Fas/CD95 mRNA were analyzed by quantitative real time PCR using a fluorogenic 5\'-nuclease assay (TaqMan©, Applied Biosystems, Weiterstadt, Germany) on a ABI Prism 7900 HT Sequence Detection system. For each experiment, total RNA was extracted from 10^5^cells using the RNeasy system (Qiagen, Hilden, Germany). Total RNA was reverse transcribed using random hexamer primers. For quantitative PCR, the appropriate primers and FAM-TAMRA labeled probes were purchased as 20-fold concentrated predeveloped assays from Applied Biosystems and used according to the instructions of the manufacturer. 18S rRNA gene was co-amplified as an internal standard. Data were calculated with the ΔΔCt method as described \[[@B15]\]. Measurement of soluble Fas (sFas) in cell culture supernatant and cell surface bound Fas ---------------------------------------------------------------------------------------- For the detection of sFas in the cell culture supernatants of the fibroblasts, a commercially available ELISA (Quantikine Assays, R&D Systems, Wiesbaden, Germany) was used according to the manufacturer\'s instructions. The amount of cell-surface-bound Fas was measured by flow cytometry after staining the fibroblasts with FITC-labelled anti-Fas/CD95 antibodies (Becton Dickinson). Statistical analysis -------------------- All statistical analyses were performed with SPSS 10.0 for Windows (SPSS, Chicago, IL). Results were presented as mean values ± SE. Mean values were compared by Student\'s t-Test. In addition the data were analysed using the non-parametric Mann-Whitney-U-Test. Differences were considered to be significant if the p-values were below 0.05 in both tests. Results ======= Characterization of lung fibroblasts ------------------------------------ The fibroblasts isolated from lung tissue specimens were characterized with respect to the expression of lineage-specific marker proteins (Fig. [1A](#F1){ref-type="fig"}). The majority of the cells stained with antibodies directed against Thy-1, an antigen that is specific for fibroblasts. These cells also expressed the fibroblast specific antigen D7-Fib and the enzyme prolyl-4-hydroxylase, which is involved in collagen synthesis. Neither CD68, a marker of monocytes/macrophages, nor CD45, a leukocyte membrane protein, were detected. We found no significant differences in the phenotypic characteristics of fibroblasts which were derived from different patient groups. The matrix production of fibroblasts isolated from fibrotic (fibrotic fibroblasts, f-fibs) and non-fibrotic lung tissues (normal fibroblasts, n-fibs) was also analyzed: F-fibs produced significantly more ECM proteins, including collagen, than n-fibs (Fig. [1B](#F1){ref-type="fig"}) independent from the underlying disease. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *Lung fibroblasts from patients with fibrotic lung diseases are resistant to apoptosis.A:*Characterization of isolated fibroblasts using flow cytometry with lineage-specific monoclonal antibodies. Immunoreactivity was detected after staining with the fibroblasts specific anti-Thy-1, D7-fib and anti-P-4-H antibodies, but not with anti-CD45 and anti-CD68 antibodies. The specificity of the immunostaining (solid line) was tested using irrelevant isotype controls (dotted line). All fibroblasts samples were analyzed and representative figures were presented. B:Matrix deposition by isolated fibroblasts as determined by a \[^3^H\]proline incorporation. Fibroblasts derived from patients with lung fibrosis deposited more extracellular matrix (solid circles) and collagen (open circles) than non-fibrotic fibroblasts (solid and open squares). C:Increased apoptosis in lung fibroblasts (upper panel) compared to synovial fibroblasts (lower panel). Apoptosis was induced by incubation with Fas ligand (rhFasL) and measured after TUNEL staining. Lung fibroblasts showed significantly more apoptosis resistance. D:Increased resistance to pro-apoptotic signals in fibrotic fibroblasts (square) compared to control fibroblasts (circle). Apoptosis was induced by rhFasL after pre-incubation with TNF-alpha. Apoptosis was measured by quantification of histone-bound DNA fragments. ::: ![](1465-9921-6-37-1) ::: Resistance to Fas induced apoptosis in lung fibroblasts ------------------------------------------------------- Recently we showed that fibroblasts derived from patients with different inflammatory joint diseases display different susceptibilities to FasL-induced apoptosis \[[@B14]\]. Comparing fibroblasts derived from human lung tissues, we applied the same conditions for induction of apoptosis and found that lung fibroblasts are generally more resistant to FasL-induced apoptosis than synovial fibroblasts (Fig. [1C](#F1){ref-type="fig"}). Similar results were found after staurosporin treatment of fibroblasts (Fig. [1C](#F1){ref-type="fig"}) and after induction of apoptosis by anti-Fas antibodies (not shown). The percentage of apoptotic cells, as determined by TUNEL staining, was at the detection limit. Therefore this method could not be used for the comparison of apoptosis in lung fibroblasts. On the other hand, we found low but measurable apoptosis after quantification of the amount of histone-associated DNA fragments in the cellular supernatants. Comparison of f-fibs and n-fibs confirmed that f-fibs were more resistant to FasL-induced apoptosis. Incubation of the cells with TNF-α slightly increased the susceptibility of n-fibs to apoptosis, but it had no affect on apoptosis in f-fibs (Fig. [1D](#F1){ref-type="fig"}). Recently Tanaka et al. have shown that the resistance to anti-Fas-induced apoptosis in lung fibroblasts is mediated by the overexpression of the specific inhibitors of apoptosis X-chromosome-linked inhibitor of apoptosis (ILP) and FLICE-like inhibitor protein (FLIP) \[[@B16]\]. Those authors found that suppression of protein synthesis using cycloheximide decreased the concentration of these short-lived inhibitory proteins and led to increased susceptibility to Fas-mediated apoptosis. We used cycloheximide to analyse whether the differences in sensitivity to FasL-induced apoptosis depended on different expression levels of short-lived inhibitory proteins. As expected, preincubation of fibroblasts with cycloheximide increased apoptosis as determined by TUNEL-staining. However, the percentage of apoptotic cells among f-fibs was still lower than among n-fibs (Fig. [2](#F2){ref-type="fig"}), which suggests that short-lived inhibitory proteins do not contribute to the difference in apoptosis resistance between n-fibs and f-fibs. Therefore, other mechanisms are involved in the regulation of the resistance to FasL-induced apoptosis in f-fibs. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Resistance to pro-apoptotic signals in fibrotic fibroblasts after incubation with cycloheximide and Fas ligand.A:Representative histograms of non-fibrotic (upper panel) and fibrotic fibroblasts (lower panel). The cells were incubated with medium, FasL, cycloheximide or FasL+cycloheximide. Only the incubation with FasL and cycloheximide resulted in significant amounts of apoptotic cells. B:*Fibrotic fibroblasts (circles) showed increased resistance to the induction of apoptosis by FasL and cycloheximide in comparison to non-fibrotic fibroblasts (squares). Apoptotic cells were detected by flow cytometry after TUNEL staining. The cumulative data of all samples are represented as mean ± SEM, \\p \< 0.01. ::: ![](1465-9921-6-37-2) ::: Expression of Fas ----------------- Expression of Fas plays a crucial role in FasL-induced apoptosis. Therefore, we investigated Fas-mRNA expression using quantitative RT-PCR analysis. Fas-mRNA levels were normalized to 18S-ribosomal RNA. Surprisingly, we found increased expression of Fas-mRNA in f-fibs (Fig. [3A](#F3){ref-type="fig"}). To determine whether increased Fas mRNA in f-fibs translates into increased levels of cell surface Fas, we used flow cytometry to analyze the expression of Fas on the cell surface of lung fibroblasts. We found that 65 ± 3% (mean fluorescence intensity 41 ± 5%) of n-fibs and 41 ± 5% of f-fibs (mean fluorescence intensity 24 ± 2%) expressed Fas at the cell surface (Fig. [3B, C](#F3){ref-type="fig"}). Based on these data, we then determined the concentration of soluble Fas in the culture supernatant and found an increased concentration of soluble Fas in the supernatant of f-fibs (Fig. [3D](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Expression of soluble and surface-bound Fas in normal and fibrotic lung fibroblasts.A:*Increased levels of Fas-mRNA were found in fibrotic fibroblasts (circles) in comparison to non-fibrotic fibroblasts (squares). Fas-mRNA levels were measured using quantitative RT-PCR. B:Fas surface expression on isolated fibroblasts. Representative histograms of Fas-immunostaining on isolated fibroblasts (solid line). The specificity of the immunostaining was shown using irrelevant isotype-matched control antibodies (dotted line). Non-fibrotic fibroblasts (upper panel) expressed more surface bound Fas than fibrotic fibroblasts (lower panel). C:Cumulative data of all analysed samples showed that the percentages of Fas-positive cells (left panel) as well as the mean fluorescence intensities (right panel) were lower in fibrotic fibroblasts (circles) compared to non-fibrotic fibroblasts (squares). The results are represented as mean ± SEM. D:*Increased concentration of soluble Fas in the supernatant of fibrotic fibroblasts (circles) in comparison to non-fibrotic fibroblasts (squares). The concentration of soluble Fas was measured by ELISA. The results are represented as mean ± SEM, \p \< 0.05, \\p \< 0.01. ::: ![](1465-9921-6-37-3) ::: Discussion ========== Lung fibrosis remains a devastating clinical condition with very limited therapeutic options. A number of experimental approaches have been investigated in clinical trials, including the modulation of key cytokines and growth factors, and treatment with corticosteroids or immunosuppressants. In a number of patients, especially those with UIP, these treatments have little effect on patient outcome \[[@B17]-[@B19]\]. The persistence of fibrotic lesions, which characterize lung fibrosis and lead to organ dysfunction, suggests that decreased apoptosis of myofibroblasts may play a major role in the pathology of lung fibrosis. The present study provides evidence that fibroblasts derived from lung tissues of patients with lung fibrosis are characterized by a relative resistance to Fas-mediated apoptosis. In this context we have shown that the resistance to apoptosis depends not only on the expression of short-lived intracellular anti-apoptotic proteins, but that increased production of soluble Fas adds to this process. The resulting long-lived cells may contribute to increased matrix-deposition, and thus to altered tissue remodeling in the diseased lung. In our study we used fibroblasts from patients with different fibrotic lung diseases, which were characterized by a 1.6-fold increase in production of extracellular matrix proteins, particularly collagen. The findings suggest that these cells retained their fibrotic differentiation state in short-term culture. Previously, it was shown that lung fibrosis is characterized by predominant differentiation of fibroblasts into myofibroblasts \[[@B20]\]. These cells are characterized by increased ECM production. The data concerning apoptosis in these cells are conflicting. Whereas Ramos and coworkers reported increased spontaneous apoptosis in fibroblasts obtained from patients with UIP, \[[@B20]\] TANAKA et al. found a high resistance to Fas-mediated apoptosis in lung fibroblasts \[[@B16]\]. In addition, it was reported that the apoptosis of myofibroblasts is suppressed by TGF-β1 \[[@B6]\] and that increasing amounts of TGF-β1 are produced by f-fibs \[[@B20]\]. In our experiments we found very low levels of apoptosis in all the lung fibroblasts investigated. After TUNEL-staining, the spontaneous amount of apoptotic cells was generally below 2%. This was similar to the data derived from synovial fibroblasts \[[@B14]\]. However, in contrast to synovial fibroblasts, apoptosis remained low after incubation of lung fibroblasts with FasL, anti-Fas antibodies or staurosporin. The data are consistent with the findings of Tanaka et al. who investigated normal lung fibroblasts and the lung fibroblast cell line WI-38 \[[@B16]\]. Only the application of very sensitive detection systems allowed us to quantify apoptosis in these cells. Using these assays we were able to demonstrate that f-fibs are more resistant to Fas-mediated apoptosis than n-fibs. However, we found no direct correlation between the matrix production of isolated fibroblasts and the amount of apoptosis. Apart from interfering with receptor activation at the cell-surface, apoptosis can also be blocked by intracellular anti-apoptotic proteins. For example FLICE-like inhibitory proteins (FLIPs) can prevent the recruitment and activation of caspase 8 (FLICE) to the Fas-associated protein with death domain (FADD), and thus inhibit the formation of the death inducing signaling complex (DISC). In addition, anti-apoptotic members of the bcl-family, e.g., Bcl-2, MCL-1 and A1, prevent the mitochondrial cytochrome c release. Recently, the inhibitor of apoptosis (IAP) family of genes was identified \[[@B21]\]. The X-linked IAP (ILP) suppresses apoptosis by direct inhibition of caspase 3. In a variety of experimental systems it has been shown that the overexpression of these anti-apoptotic proteins results in resistance to pro-apoptotic signals \[[@B21]-[@B23]\]. Therefore, it was tempting to speculate that a differential expression of these anti-apoptotic proteins in n-fibs and f-fibs may cause the resistance to Fas-mediated apoptosis in lung fibroblasts. Anti-apoptotic bcl-2 proteins and IAPs are characterized by a very short half-life. \[[@B22],[@B24]\]. Cycloheximide, which blocks protein synthesis, was shown to decrease the concentration of ILP and FLICE in human lung fibroblasts on the one hand, and to increase the sensitivity of these cells to Fas-mediated apoptosis on the other hand \[[@B16]\]. Our experiments showed that short-lived anti-apoptotic proteins are generally involved in the apoptosis resistance of lung fibroblasts. However, they did not contribute to the different susceptibilities of n-fibs and f-fibs. Finally, we found a difference in the Fas-mRNA levels. F-fibs had higher Fas-mRNA levels with lower levels of surface-bound Fas-receptor than n-fibs. At the same time, f-fibs exhibited higher expression of soluble Fas, which exerts an anti-apoptotic function \[[@B25]\]. It has been shown that soluble Fas is produced as an alternatively spliced variant of Fas. On the other hand increased soluble Fas concentrations were found in patients with rheumatoid arthritis and the release was correlated to increased activities of matrix metallo proteases \[[@B26]\]. We have recently shown that fibrotic fibroblasts expressed increased amounts of the potent protease cathepsin K \[[@B9]\]. An important role of matrix metalloproteases was shown by other groups \[[@B27],[@B28]\]. In summary the release of soluble Fas can be regulated by different mechanisms. Part of them is activated in fibrotic fibroblasts. From these data, we conclude that the increased resistance to pro-apoptotic signals in lung fibroblasts obtained from patients with fibrosis is mediated at least in part by increased amounts of soluble Fas. Authors\' contributions ======================= FB cultured the fibroblasts, drafted the manuscript and participated in the design of the study. AW measured apoptosis by flow cytometry. AB performed quantitative RT-PCR analyses. IM carried out sFas analyses and cells death assays. OW did the bronchoscopy and tissue biopsies. CR carried out the histo-morphological classification of tissue samples. TW conceived the study, participated in the design of the study and coordinated the tissue sampling. TP conceived the study, established quantitative RT-PCR and helped to draft the manuscript. All authors read and approved the manuscript. Acknowledgements ================ The authors wish to thank Yvonne Peter, Gabriele Weitz, Sybille Pietzke, DesireWeber and Susann Weinholz for their technical assistance. The work was supported by the Deutsche Forschungsgemeinschaft (DFG We2292/2-1, DFG Pa 698/2-1)	PubMed Central	2024-06-05T03:55:56.230295	2005-4-17	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087885/", "journal": "Respir Res. 2005 Apr 17; 6(1):37", "authors": [ { "first": "Frank", "last": "Bühling" }, { "first": "Aline", "last": "Wille" }, { "first": "Christoph", "last": "Röcken" }, { "first": "Olaf", "last": "Wiesner" }, { "first": "Anja", "last": "Baier" }, { "first": "Ingmar", "last": "Meinecke" }, { "first": "Tobias", "last": "Welte" }, { "first": "Thomas", "last": "Pap" } ] }
PMC1087886	Heparin-induced thrombocytopenia (HIT) is recognized as a rare, but potentially devastating complication of heparin therapy because of its association with arterial and venous thrombosis \[[@B1]\]. HIT is mediated by antibodies which recognize an antigen formed by the binding of platelet factor 4 to heparin \[[@B2]\]. The now widespread use of low-molecular-weight heparins for a variety of indications previously reserved exclusively for unfractionated heparin has generated interest in comparing the relative rates of HIT between the two classes of heparin. In 1995 Warkentin examined rates of HIT in patients undergoing elective hip arthroplasty who had been randomized to receive either unfractionated heparin (UFH) or low molecular weight heparin (LMWH) for thromboprophylaxis \[[@B3]\]. Warkentin reported that HIT occurred in 9 of 332 patients who received UFH and in none of 333 patients who received LMWH (2.7 percent vs. 0 percent). In addition, development of heparin-dependent IgG antibodies and thrombotic events associated with thrombocytopenia were more common in patients treated with UFH than in those treated with LMWH. Recently, using different criteria for HIT (an absolute drop in platelet count of 50% or greater vs. platelet count less than 150,000 cells /ml), Warkentin reanalyzed these same data and found the difference in the observed rate of HIT was even more pronounced, 8 times greater (4.8% vs. 0.6%) in UFH compared to LMWH for prophylaxis of venous thromboembolism (VTE) in patients undergoing elective hip arthroplasty \[[@B4]\]. Warkentin\'s results are supported by a recent study by Walenga et al which carefully evaluated sera from three clinical studies \[[@B5]\]. Walenga found that LMWH was less likely to generate H-PF4 antibodies than UFH and less likely to result in clinical HIT. Walenga also noted that LMWH were more likely to generate IgA and IgM antibodies rather than IgG antibodies, which are associated with clinical HIT. However, the sera reviewed by Walenga all were from orthopedic surgical patients. Authors in the medical literature often generalize Warkentin\'s results, applying them to medical as well as surgical patients \[[@B6]-[@B8]\]. Further, it has been suggested that differences in rates of HIT represent an advantage of LMWH over UFH in VTE prophylaxis in non-surgical patients \[[@B7],[@B9]\]. However, we do not think the Warkentin data can be applied with confidence to non-surgical patient populations. In non-surgical (medical) patients, the rate of HIT with UFH is reported to be much lower than in Warkentin\'s analysis of surgical patients. For example, an earlier study cited in Warkentin\'s 1995 paper reported an incidence of HIT of only 0.3% for non-surgical patients who received therapeutic intravenous UFH \[[@B10]\]. To evaluate the relative rates of HIT in non-surgical patients we reviewed recent studies that compared UFH to LMWH in either the treatment or prevention of VTE in medical (non-surgical) patients. We chose studies available to us through PubMed. In our review, we found 11 trials that reported either HIT (which, as Warkentin points out, does not have a uniform definition), thrombocytopenia, \"severe thrombocytopenia\" or some combination of the above. Our findings are listed in the table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Treatment duration and reported adverse event rates in studies comparing UFH vs. LMWH heparin in the treatment and prophylaxis of VTE ::: Treatment of VTE ------------------------ ------------------------------- ---------------------------- ------------ ---------- --------- ------- --------- Study Treatment duration (days) Reported Adverse Event UFH LMWH UFH LMWH % n/N % n/N Merli \[11\] ≥ 5^‡^ ≥ 5^‡^ T\* 1.4 4/290 2.0 12/610 Koopman \[12\] 6.1 6.5 T(u) 2.5 5/198 1.5 3/202 Levine \[13\] 5.5 5.8 T 1.9 3/253 2.0 5/247 T w/o exp. 0.4 1/253 0.4 1/247 Hull \[14\] not stated T(u) 1.0 1/103 3.1 3/97 Columbus Invest \[15\] 5.8 6.3 HIT 0.3 1/308 0.0 0/304 Simmoneau \[16\] 7.0 7.3 HIT 0.3 1/308 0.0 0/304 Prophylaxis of VTE Study Treatment duration (days) Reported Adverse Event UFH LMWH UFH LMWH % n/N % n/N Harenberg, 1990 \[17\] 10 10 \\ -/82 -/84 Harenberg, 1996 \[18\] 10 10 T(d) 0.5 4/780 0.0 0/810 PRIME \[19\] 7 7 T (n) 0.0 0/482 0.0 0/477 Bergmann \[20\] 10 10 HIT† 0.4 1/223 0.0 0/216 PRINCE \[21\] 10 10 T NR NR Key: T:Thrombocytopenia, defined as platelet count less than 100,000 cells/ml, \One case of \"immune thrombocytopenia reported in this study\". Case was in the LMWH group but had received UFH prior to randomization. T(u) -- Thrombocytopenia-undefined in study T w/o exp. -- Thrombocytopenia with \"no apparent explanation\" NR: not reported, \\*: \"Thrombocyte count did not change in either group\". T(d): A decrease in platelet count (values ranging between 40,000 and 80,000/microliter) was observed in four patients with UF and in none with LMW heparin. No severe thrombocytopenia was observed. T(n): \"There was no decrease in platelet count due to enoxaparin or Ca-heparin.\" † One patient with drop in platelet count from 149 K cells/ml to 87 K during treatment. Platelet count rose to 280 K post study. No sequallae from thrombocytopenia. ^‡^Average length of treatment not stated. Treatment length \"at least five days\". ::: The studies cited in the tables are heterogeneous in the endpoints used. Given the variability in definition of thrombocytopenia among the trials and the lack of standardized and routine evaluation for HIT in any of the above studies we feel it is currently impossible to estimate the relative risk of HIT in UFH vs. LMWH in either the treatment of VTE or prevention of VTE in non-surgical patients. However, the data do suggest that thrombocytopenia is rather uncommon with either heparin therapy. Insofar as the rate of HIT must be less than that of thrombocytopenia, HIT is likely to be an infrequent event as well. A recently published study provided a rigorous analysis of H-PF4 antibodies in patients treated for deep vein thrombosis with LMWH vs. UFH \[[@B22]\]. In this study, H-PF4 antibodies (measured by a commercial ELISA method) developed in 9.1% of patents in the UFH group vs. 2.8% of patients in the LMWH group (both treated for 5--7 days). However, there was only one occurrence of HIT with thrombosis among 356 patients in the UFH group vs. no occurrences of HIT among 374 patients in the LMWH group. This study, we feel, is consistent with both Warkentin\'s data regarding orthopedic surgery patients and the data presented in our table above; namely, LMWH induces H-PF4 antibodies at a lower rate than UFH but that clinical incidence of HIT in non-surgical patients is too small to statistically differentiate. Unfortunately, the generalization to medical patients of Warkentin\'s data regarding HIT rates for orthopedic surgical patients persists in the literature. As recently as 2004, a meta-analysis comparing heparins for the treatment of pulmonary embolism cited \"the lower risk for\...heparin-induced thrombocytopenia\" as an advantage of LMWH over UFH \[[@B7]\]. As justification, this paper references a review article \[[@B8]\] which in turn, references the 1995 Warkentin paper \[[@B3]\] which, as discussed above, involved exclusively orthopedic surgical patients. One possible explanation as to why the studies of treatment and prevention of VTE in medical patients have not demonstrated a difference in thrombocytopenia or HIT rates is because the length of treatment in these studies may be too short for patients to develop HIT. In a study which clearly demonstrated a difference in HIT rates for LMWH vs. UFH following cardiopulmonary bypass surgery, 4 patients developed significant levels of Heparin-PF4 antibodies in days 3--5 postoperatively vs. 75 patients in days 7--10 post operatively \[[@B23]\]. Similarly, in Warkentin\'s data, thrombocytopenia typically developed 6--13 days after surgery (and of heparin therapy) and thrombotic events occurred 7--17 days after surgery \[[@B3]\]. In a recent study, Girolami et al reviewed 598 consecutive patients admitted to a medical ward with either a therapeutic or prophylactic indication for UFH \[[@B24]\]. HIT was not observed in any of the 238 patients who received UFH for a therapeutic indication. The authors speculate that HIT was not observed in these patients because duration of heparin was less than one week. There were 5 cases of HIT observed in the 598 patients (0.8%), all in those patients who received heparin for prophylactic indication. These cases occurred from day 8 to day 22 of therapy and the three observed associated thrombotic events occurred on days 10--21 of therapy. Such data are consistent with the College of American Pathologists (CAP) 2002 recommendations that platelet monitoring for HIT should focus on days 4--10 after starting heparin \[[@B25]\]. In addition, the CAP position on monitoring for HIT differentiates medical vs. surgical patients, with more frequent platelet count monitoring recommended for surgical patients. The CAP calls \"postoperative\" patients receiving UFH at \"highest\" risk for HIT, while \"medical\" patients receiving UFH are considered at \"intermediate\" risk \[[@B25]\]. The Girolami study further support the position that, in most cases, the use of heparin for the treatment of VTE is limited to the first 5--7 days of treatment and that heparin (either LMWH or UFH) is discontinued before clinical HIT, as evidenced by either thrombocytopenia and/or thrombosis, generally occurs. Similarly, the strong correspondence of length-of-treatment with the likelihood of development of H-PF4 antibodies and HIT is likely important in prophylaxis of non-surgical patients as well. In the studies we reviewed, the length of heparin pharmacoprophylaxis was generally 7--10 days. Despite this length of treatment being ostensibly long enough for patients to develop laboratory-evident HIT we suggest that 7--10 days of therapy is too short of a duration for many cases of potential clinically-evident HIT to manifest. This could limit the potential clinical import of differences in rates of HIT for UFH and LMWH. Indeed, our experience is that length-of-stay in our institution for most of our medical patients eligible for pharmacoprophylaxis is 10 days or less. Additionally, in the studies of \"medical patients\" we reviewed, the risk low and similar risk of either thrombocytopenia, HIT or thrombotic complications of HIT in the UFH and LMWH groups may also be, in part, due to the absence of surgical activation of PF-4 in these patients. Accurate assessment of the risks and benefits of competing therapies is paramount to sound cost-effective decision-making. At our institution, acquisition costs for branded LMWH are approximately 15 times that of generic UFH, a factor that would certainly favor the latter if efficacy and safety are similar. In Europe and Canada, where cost differences between LMWH and UFH are less pronounced, clinical decision-making and cost-modeling may be different than in the United States. Indeed, the latest Amercian College of Chest Physicians\' (ACCP) guidelines on antithrombotic therapy recognize that \"the cost for low-molecular-weight heparin (LMWH) is high in the United States, but low in most European countries. Thus, in instances in which small benefits accrue to patients from the use of LMWH in comparison to the use of unfractionated heparin, the choice in favor of LMWH may be clear in Europe, but much less clear in North America\" \[[@B26]\]. However, given the serious (and expensive) nature of complications from HIT, true differences in clinical HIT with thrombosis between UFH and LMWH would affect significantly safety considerations as well as total health care cost-modeling between the two therapies. Unfortunately, we feel sufficient information in this area are currently lacking. We encourage investigators to make a rigorous evaluation of HIT using new definitions proposed by Dr. Warkentin in his 2003 paper \[[@B4]\] as part of any future studies comparing LMWH and UFH for either the treatment or prevention of VTE in non-surgical patients to better define the risk of this important clinical problem.	PubMed Central	2024-06-05T03:55:56.232677	2005-4-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087886/", "journal": "Thromb J. 2005 Apr 4; 3:4", "authors": [ { "first": "Charles FS", "last": "Locke" }, { "first": "John", "last": "Dooley" }, { "first": "Jonathan", "last": "Gerber" } ] }
PMC1087887	Background ========== Internationally there are several models available to assess the long-term risk of cardiovascular disease and they are broadly used in clinical practice and research \[[@B1]-[@B3]\]. A 10-year risk prediction model based on clinical and laboratory data plays an integral part in planning cardiovascular prevention \[[@B1]\]. However, this applies only for the arterial side of the vascular system. We are not aware of any model to predict the long-term risk for venous thromboembolism (VTE) in a similar way. A prediction of the absolute risk of venous thromboembolism can only be developed on the basis of a specifically designed long-term cohort study. The study reported here is the first long-term, community-based cohorts study that observed genetic and clinical thromboembolic risk factors and their multivariate impact on VTE incidence to address this issue in young women. A population-based thromboembolic risk factor study started in the mid-1990s in Bavaria, the *BAvarian ThromboEmbolic Risk study (BATER), focused on women in the reproductive age \[[@B4]-[@B6]\]. Clinical and laboratory risk factors for VTE, the lifetime history of relevant conditions or medications, and the family history of cardiovascular diseases were documented from 1993 throughout the follow-up period until 2003, i.e., carefully reviewing complaints or findings possibly related to the occurrence of venous clots. This study provides a probability scheme that enables identification of women at high risk for VTE compared to women with virtually no risk for a theoretical period of maximal 10 years (defined by the model used). These findings could contribute to weigh the prognostic importance of clinical and laboratory data for medical decisions and better counseling of patients. Methods ======= Material and methods of this long-term cohort study has been described in detail in earlier publications \[[@B4]-[@B6]\]. In brief, we used a cohort of 4337 young women (18--55 years) in Bavaria (Germany) who had at least one follow-up. We collected data from demographics, reproductive life, lifestyle pattern, conditions/diseases, and particularly potential risk factors for VTE through a questionnaire for self-administration. Whenever possible, time-related information was documented. Using this method we were able to set up the common starting point for the cohort as 1993. Telephone enquiries were made to supplement, clarify and verify the data in the questionnaires. The primary source for the data on VTE was the follow-up questionnaire (self-reported VTE or symptoms potentially compatible with VTE). This information was completed by telephone interviews with the woman and with the treating physician. All available information about diagnostic and therapeutic measures taken was recorded. Clinical data and/or invasive or non-invasive diagnostic procedures were assigned to one of the following categories of likelihood of a VTE: Definite VTE ------------ Unequivocal positive finding in at least one imaging test, e.g., phlebography or duplex sonography for deep venous thrombosis (DVT); pulmonary angiogram, VQ scan, spiral computed tomography (spiral CT) for pulmonary embolism (PE). Probable VTE ------------ Typical clinical symptoms for VTE without unequivocal imaging test but positive findings in other diagnostic tests (e.g. Doppler US or plethysmography) [and]{.underline} subsequent specific therapy over a longer period (low-dose heparin or other anticoagulants). Possible VTE ------------ Typical clinical symptoms for VTE and unknown or equivocal result on imaging, only suspicion of VTE suggested by a non-imaging diagnostic tests (such as Doppler US, plethysmography, ECG, blood gas analysis or others for PE) [and]{.underline} no subsequent specific therapy (e.g., only short-term low-dose heparin and bandage). Potential VTE ------------- Typical clinical symptoms for VTE without further diagnostic tests or negative results or diagnostics unknown. Unspecific therapy -- but nonetheless the treating physician maintained the diagnosis VTE based on clinical findings. All possible and potential \"VTE cases\" were excluded from the analyses in this paper because of diagnostic uncertainty, i.e. lacking information whether they should be better classified as VTE cases or non-cases. Women with a history of cancer or with known antiphospolipid syndrome were not in our follow-up study. ### Laboratory methods After having given informed consent the women included in this study gave a blood sample at one time point during the observational period (1996/97). An independent ethics committee approved all study related activities. Whole blood samples were obtained from resting subjects. Blood was put into tubes with trisodium citrate. Plasma was prepared soon after venipuncture by centrifugation for 15 minutes with 3000 to 4000 / min at room temperature and stored at -20°C. Protein C and antithrombin activities in plasma were measured by chromogenic substrate assays (Dade Behring, Marburg, Germany). For \"antithrombin\" the activity against factor IIa was determined, Protein C activity was measured after activation of the proenzyme by snake venom. Plasma activities are given as percentage (units/dl) of pooled human normal plasma. Genomic DNA was isolated by mean of QIAmp^®^DNA Blood Kit (Qiagen) according to the manufacturer\'s instructions. The genetic polymorphisms Factor V R506Q (G1691A), the prothrombin promoter G2010A and the 5-, 10-methylenetetrahydrofolate reductase (MTHFR) A223V (C677T) were determined using a multiplex PCR with allele-specific primers slightly modifying a previously described method \[[@B7]\]. All blood tests were performed blinded, i.e. the investigators had no clinical information, and had no access to the clinical database. ### Method of data analysis Due to the importance of the temporal relationship the database was structured to accommodate both concurrent as well as time-dependent variables. Concurrent variables are variables, which describe the woman\'s status at the time of questionnaire response, whereas the outcome variable (VTE) is time-dependent. While concurrent variables were held in a fixed data set, a periodic data set containing information on lifetime exposures and the occurrence of VTE events along a time axis was created for the time-dependent variables of each participant, using months as a unit of measurement. The exposures of interest in this publication, such as VTE risk factors including genetic markers, refer to the baseline time point. Some of the variables in the database (age, BMI, Protein C, AT) were continuous. These variables were dichotomized in order to define a categorical exposure status (exposed -- non-exposed) for the analyses based on incidence or logistic regression. We arbitrarily separated the continuum in two roughly equal intervals such as age under/over 30 or BMI under/over 25 in order to have sufficient case numbers for analyses with further stratification. For protein C and AT we used the 5th percentile (lower 5% of the distribution in non-cases) as cut off point. This limit was considered as usual definition for deficiency and clinically relevant \[[@B8]\]. All analyses concerning the occurrence of VTE events over time were performed by adding up individual observation time (1993 until the last contact) for different exposure-cohorts and in total. The incidence rate of VTE was calculated per 10,000 women years of observation (WY). The predictive model was developed using the discriminant function analysis technique \[[@B9]\]. This technique permits to determine which of the clinical and laboratory data discriminate best between two groups: future VTE cases vs. non-cases. In other words, this multivariate analysis determines which variables at baseline are the best predictors of a future event. We multivariately ranked the predictive power of the variables with suspected effect on occurrence of VTE (i.e., possible or established risk factors). Other available information in the database or information not known at baseline (e.g. later occurring conditions like surgery or longhaul flights; see discussion) were not included into the model since they cannot contribute to a predictive model (i.e. the setting for the application of the results of this study is women consulted by a physician independent of an acute VTE event). Technically, we used stepwise discriminant analysis with forward inclusion or backward elimination of variables. The p-value for entry into the model was 0.49 and for removal 0.50. The p-value of the parameter provided by the discriminant analysis at the last step determines its rank. Variables with the smallest p-value get the highest rank. This permits the comparison of predictive power of potential risk factors -- documented at baseline (1993) -- for later occurring VTEs. This technique permits to classify persons by the discriminant function value (separated by the case status). The true incidence of new VTE cases was determined in strata of persons with different pattern of risk profile to characterize groups with lower or higher absolute VTE risk according their risk profile at entry. We used for this analysis those variables that depicted the highest 5 ranks in the stepwise discriminant analysis. All analyses were performed with the statistical packages SPSS 10.2, SAS 8.2 or STATA 8.2. Results ======= Description of the cohort ------------------------- The overall cohort encompasses 4337 women with sufficient information in 1993 and one follow-up at minimum. The observational period for our current analysis was 32,656 WYs since 1993. The follow-up was continued until 2003 at most, or was otherwise terminated at the time when the last contact was possible to get information about new conditions that may have had occurred. 2076 women could be followed up until 2002/3 (47.9 %), 595 (13.7%) women dropped out between 1999 and 2001, and the largest proportion of women dropped out before 1999 (38.4%). Thus, the follow-up period was censored some time before 2002/3 for approximately half of the cohort members. Thirty-four new cases of VTE occurred in the observational period. These cases were finally confirmed and categorized according to diagnostic certainty by an independent medical reviewer as definite (n = 31) or probable (n = 3). Cases with possible/potential VTE (n = 17) were excluded from further analyses because of low diagnostic certainty, i.e. it was not clear whether to classify them in the group \"cases\" or \"non-cases\". Out of the 34 definite or probable VTE cases 18 cases (= 52.9%) were associated with \"clinical causes for VTE\" and 16 (= 47.1%) were so-called \"idiopathic\" VTEs. The following previously described \"acute clinical causes\" for VTE the following were observed: 4 with previous VTE, 3 with pregnancy/delivery, 4 after accident, 2 after surgery, 3 with immobilization, and 2 after long travel. Table [1](#T1){ref-type="table"} depicts the profile of relevant data available at baseline (1993) to get an impression of the group under follow-up. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Distribution of 10 clinically and 5 genetically relevant variables in a cohort of young women at baseline of the observational period 1993 -- 2003. The total number of women in this analysis is 4320, i.e. excluding 17 women with a final diagnosis of a possible/potential VTE. Deviations from this number are due to missing data ::: Variables ---------------------------------------------- ------ ------ -------------- Continuous variables n Mean (SD) Age (years) 4320 26.0 (8.6) Life births, number 1910 1.7 (0.8) BMI^§^ 4309 23.3 (4.1) Protein C (unit/dl) 4315 102.4 (15.8) AT III (unit/dl) 4316 98.4 (11.3) Categorical parameters Percent (%) Own history of VTE No 4279 99.0 Yes 41 1.0 Family history No 3840 88.9 Yes 480 11.1 Age, alternative \<30 2843 65.8 ≥ 30 1477 34.2 Family history of varicous veins No 2395 55.4 Yes 1925 44.6 Family history of MI No 3830 88.7 Yes 490 11.3 BMI, alternative \<25 3218 74.7 ≥ 25 1091 25.3 Ever use of hormone replacement No 4031 93.7 Yes 270 6.3 Family history of stroke No 4013 92.9 Yes 307 7.1 Ever use of oral contraceptives No 346 8.0 Yes 3973 92.0 Education level: University entrance diploma No 3119 73.2 Yes 1139 26.8 Ever smoker No 2022 46.8 Yes 2296 53.2 Laboratory parameters Factor V Leiden mutation^1^ No 4035 93.7 Yes 271 6.3 Prothrombin mutation^1^ No 4088 96.6 Yes 142 3.4 MTHFR^1^ No 1798 42.5 Yes 2432 57.5 Protein C: ≥ 77(unit/dl) No 4117 95.4 \<77 (5th percentile; unit/dl) Yes 198 4.6 AT III: ≥ 81(unit/dl) No 4106 95.1 \<81 (5th percentile; unit/dl) Yes 210 4.9 ^1^Homozygote & heterozygote together ^§^Body mass index (kg/m^2^) ::: The mean age was 26 ± 8.6 years, however, for the dichotomized age variable we used as cut-off point 30 years resulting in strata that contained VTE cases in both age groups. The frequency of other conditions, family history of potentially relevant diseases, lifestyle pattern and genetic tests is provided in the table [1](#T1){ref-type="table"}. Homo-and heterozygote carriers of mutations were analyzed together because of small numbers of homozygote carriers. Ranking risk factors according predictive power for VTE ------------------------------------------------------- The stepwise discriminant analysis was used to rank relevant variables at baseline according to their power to predict the case and non-case status many years later. We initially used 7 clinically available, potential risk factors for VTE and 5 laboratory parameters. Later, we combined genetic markers (Factor V Leiden and Prothrombin mutation G20210A) together in one composite variable due to low prevalence in the cases: FVL (n = 4) and PTM (n = 2). We used only categorical variables in the model, i.e., dichotomized continuous variables. We got the following ranking of the predictive power to explain the occurrence of new VTE cases during a theoretical 10-year period (defined by our model) in declining order: Medical history of VTE, family history of VTE (1^st^degree relatives), age at baseline 1993, body mass index (BMI), factor V Leiden (FVL) or prothrombin mutation (PTM), family history of varicose veins, protein C, AT level, hormone ever use, MTHFR carrier status, and OC ever use. We considered only the 5 highest-ranking variables for the development of our predictive model for practical use (Table [2](#T2){ref-type="table"}). Seven other variables with lower predictive importance were left out because their multivariate impact was too low and the practical application of a model with more than 5 variables is not easy to handle in practice. In addition, models that included other or more than the selected 5 variables brought no further improvement of the prediction (data not shown). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Ranking order of clinical and laboratory data according, possibly relevant for VTE. Analysis with stepwise discriminant analysis ::: Rank order* P value Independent variables ---------------- ------------- ---------------------------------------------------- 1 0.000 Medical history of VTE (yes, no) 2 0.005 Family history of VTE (yes, no) 3 0.012 Age at baseline 1993: \<30 vs. ≥ 30 years 4 0.082 BMI: \<25 vs. ≥ 25 5 0.112 FVL or PTM carrier: any positive vs. all negativ 6 0.153 Family history of varicose veins (yes, no) 7 0.536 Protein C: \<77 vs. ≥ 77(unit/dl) 8 0.767 AT III: \<81 vs. ≥ 81(unit/dl) 9 0.773 HRT ever use (yes, no) 10 0.776 MTHFR carrier (yes, no) 11 0.798 OC ever use (yes, no) ::: Taking the VTE incidence in all combinations of the five variables into account, we formed four levels of future VTE risk (Table [3](#T3){ref-type="table"}). Some of the combinations of the five risk markers had not sufficient data, i.e. the observation period (WY) was too short to observe new VTE cases. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Expected risk level for VTE within next 10 years in four categories in the BATER study population: No/low risk, moderate risk, high risk, very high risk. ::: Future risk level Study population WY^1^ VTE cases VTE incidence ----------------------- ---------------------- ----------- --------------- ------------------- ------- No/low 2634 61.3 19282 8 4.1 Moderate 907 21.1 7314 9 12.3 High 257 6.0 2117 10 47.2 Very high 40 0.9 352 6 170.5 No data^2^ 457 10.6 3300 0 0.0 ^1^WY = representing the women-years of observation in the respective category of the BATER cohort) ^2^No cases observed in these sub-groups ::: The majority of the study population (61.3%) had a small VTE risk (no/low in Table [3](#T3){ref-type="table"}), about 20% depicted a \"moderate risk\", 6% a \"high risk\", and 0.9% a \"very high risk\". About 10% could not be classified due to lacking VTE cases in the observation period (\"no data\"). The cut points for the four risk levels were arbitrarily defined: The average VTE incidence per 10^4^WYs steeply increases across the five groups: 4.1 (no/low), 12.3 (moderate), 47.2 (high), and 170.5 (very high). Figure [1](#F1){ref-type="fig"} provides a scheme to support the individual decision about a future VTE risk based on information on five or less variables. The small number of new VTE cases in our model, however, prevented us from drawing a complete decision tree, i.e. some branches of the tree cases were not observed yet or only one case (considered as not sufficient to be included in the scheme). For example, the small group of VTE cases with a previous history of a VTE in our cohort (n = 4) did not allow for further specification by age, family history of VTE, BMI, and genetic marker: this group is associated with a very high risk altogether, but it is not clear if subgroups may have a lower or higher risk. Other examples were women without previous VTE history but positive family history of VTE, higher age, and higher BMI: there were no cases to distinguish between carriers of genetic markers (FVL or PTM positive) and those without any positive genetic markers. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Predicted risk for VTE within next 10 years in four categories: No/low risk (blank circles), moderate risk (dotted), high risk (hatched), very high risk (black circles). BMI = body mass index; FVL + PTM = Factor V Leiden mutation (hetero-&homozygote) and /or prothrombin G20210A mutation (hetero-&homozygote) ; n.d. = no data ::: ![](1477-9560-3-5-1) ::: Women without VTE history, no family history, young and slim are privileged with low future VTE risk. With increasing age and BMI the VTE risk increases, particularly if associated with positive family history and markers for inherited VTE risk. Using this scheme, it is also possible to predict future VTE risk without knowledge about genetic risk factors and to give appropriate advice. Discussion ========== To our knowledge, long-term, community-based cohort studies with the aim to evaluate or compare the predictive power of clinical as well as genetic risk markers for VTE are lacking. Most studies related to VTE risk factors were restricted to clinically available markers such as age, BMI, previous VTE, family history, or acute factors (immobilization, surgery, accidents, pregnancy, and also hormone use) and usually based on clinical observations or case-control studies or studies in administrative databases, and also a few cohort studies (overview about incidence and risk factor studies in \[[@B10],[@B11]\]). One recent publication of a historic cohort \[[@B12]\] assessed carefully the impact of clinical risk factors for the prevalence of VTE and determined relative VTE risk estimates. Cohort studies in the population rarely included/reported genetic markers for thrombophilia and acquired risk factors, except the Physicians Health Study for example -- however only for males over 40 years of age \[[@B13]\]. An important recently published cohort study in Denmark analyzed specifically the incidence of VTE and compared carriers of FVL compared with non-carriers \[[@B14]\]. Other studies with focus on markers for hereditary thrombophilia were performed in patients (e.g. in anticoagulation clinics), in relatives of carriers of genetic mutations but not in the \"normal female population\" \[[@B15]-[@B18]\]. In addition, the evaluation of the importance of genetic markers for VTE risk does rarely consider the impact of clinically available risk factors and the design is often restricted to case-control studies. Overall different study designs and restricted views may come to similar conclusions in an ideal world, but not necessarily. Our BATER cohort study covers more than 4,000 cohort members with a fairly long observation period (1993 -- 2003), translating into over 32,000 WYs. Thirty-four VTE cases, classified as definite or probable, occurred within this period. This is equivalent to an incidence of about 10 per 10,000 WYs. It is important to realize that we put great effort on the detection of potential cases and -- even more important -- we included all definite and probable cases, whereas most reported incidence rates refer only to definite and so-called \"idiopathic VTE\", i.e. most reported rates excluded all cases that occurred in temporal relationship to possible \"acute\" causes such as pregnancy/delivery, surgery, and immobilization. Idiopathic VTEs, however, reflect only a part of all confirmed VTE cases \[[@B17]\]. We found in our cohort study roughly 50% so-called \"idiopathic\" VTE cases, and the other 50% of cases had a previous VTE in the past, pregnancy, delivery, surgery, accident, or immobilization/long bed-rest shortly prior to the VTE event. Thus, the incidence of \"idiopathic VTE\" observed in this study would be 5 per 10,000 WYs and thereby likely to be in the same range as other reported incidence rates in the general population. The incidence estimates for definite VTE ranges between 1 to 6 per 10^4^WYs in OC non-users and 2 to 10 per 10^4^WYs in OC users \[[@B10]\]. Older studies depicted almost always-higher incidence rates than more recently performed studies (see overview in \[[@B10]\]). A recent systematic review \[[@B11]\] came to a pooled incidence of definite VTE for the general population of 5 per 10,000 person years, similar in males and females, and found that around 40% of VTE cases were \"idiopathic\". A large cohort study found a similar incidence rate in males aged 40--49 years: 4.7/10^4^person-years \[[@B13]\]. The objective of this study was to provide a simple algorithm for medical practice to predict the future VTE risk with a simple scheme based on usually available information, i.e. to discriminate women with virtually no VTE risk in the foreseeable future from those at a high absolute risk to suffer from VTE. Incidence rates associated with different clinical and genetic factors will be published separately \[[@B6]\]. It is a limitation of this long-term cohort study, however, that the number of confirmed (definitive and probable), incident VTE cases was still too small in absolute numbers (n = 34). In other words, some sub-cohorts with certain combinations of risk factors did not contain one single new VTE case. The consequence was that the number of subgroups at risk was minimized to the extent possible to make it a feasible tool for the practice. In so far, the results and conclusions should be considered as rough but the best we can possibly do at this stage, i.e. future analyses will benefit from an improved point of departure (more cases, longer observation). Another limitation is that we did not have the chance yet to test the validity of the model in another, independent cohort. This is a task for the future. Therefore we focused this paper on a simple scheme with a rough classification of the future VTE risk. The interested (or worried) women and her treating physician might like to know (or to get confirmation) whether the future VTE risk is higher than \"normal\" (no/low risk). This information could have an impact on further medical surveillance, especial counseling, proposals as how to reduce of changeable risk factors or on suggestions for preventive measures under certain circumstances and -- of course- with respect to the compliance regarding preventive measures. Using stepwise discriminant analysis the rank order of 12 (11) clinical or laboratory data at baseline (1993) was multivariately determined concerning the power to predict future VTEs. This was the information needed to select a minimal set of parameter combinations to build a \"VTE prediction model\". Finally we ended up with a model covering the five variables with highest ranking (impact) regarding predictive importance for future VTE only: history of previous VTE, family history of VTE, higher age, higher body mass index, and carrier of FVL or PTM. The decision to form the composite genetic marker \"FVL or PTM\" was guided by the small numbers of cases who were carrier of this two mutations and the low predictive power of all other lab parameters we had in the data set (see table [2](#T2){ref-type="table"}). Four levels of future VTE risk were arbitrarily defined-based on a steeply increasing absolute VTE risk: No/low risk(4 per 10^4^WY), moderate(12/10^4^WY), highrisk (47/10^4^WY), or very high(171/10^4^WY). The low-risk group was chosen to reflect an assumed VTE risk of the normal population (see above), the group with \"moderate risk\" because VTE risk over 10/10^4^is indicative for an increased risk, and the \"high and very high risk\" groups are clearly out of the normal range. One should also consider in this context, that these cut-off points reflect an average risk with an assumed variation within these groups -- as can be seen in the scheme of a decision tree (figure [1](#F1){ref-type="fig"}). In accordance with clinical experience the overwhelming majority (61%) depicts a low risk of a future VTE. Only a minority of 6% and 0.9% is facing a high or very high VTE risk. The women who fall into the two highest risk categories have a previous own history of VTE or a positive VTE family history, have a higher BMI or a genetic mutation (FVL or PTM). Even though, the contribution of genetic appears to be limited. Using an analysis based only on clinically available data, i.e., without use of the information about lab parameters, we arrived at very similar three risk categories with almost identical absolute VTE risk (data not shown separately but are part of Figure [1](#F1){ref-type="fig"}). Another point for discussion is the impression suggested by figure [1](#F1){ref-type="fig"} that persons with previous VTE do not require genetic testing because they are in the \"high risk\" category without any further considerations. The risk might well be different for persons with/without inherited risk (family history), younger/higher age, or overweight. This however we cannot further disentangle because we are lacking new VTE events particularly in this high-risk group of our study. Thus, the conclusions are rather crude as discussed before and require clinical experience for the interpretation of individual cases. The need for genetic testing depends on the judgment in a specific clinical situation and the usefulness of this additional information for the physician and/or the patient (family). Decisions based on clinical variables about preventive measures will be made in any case -- even if no genetic information is available. The possible approaches are a matter of a current controversy in the literature \[[@B15],[@B20],[@B21]\]. Clinical reports point often towards a high VTE recurrence rate in patients with previous VTE \[[@B22]\], but despite being the \"best\" single predictor we found this phenomenon only in 4 of our 34 incident VTE cases. The predictor variables used in our model seem to be plausible and consistent with the clinical experience: History of previous VTE, age and obesity are indeed important clinical information for the VTE risk assessment, and also genetic marker were discussed as predictors of a future VTE. These are also risk parameter that are commonly used when recommending preventive measures in situations like long-haul flights, immobilization (such as accidents, surgery) and are also labelled as risk factor in drugs containing sexual hormones (e.g., oral contraceptives or hormone therapy). We assume that physicians will appreciate these results as a possibility to double-check if their own decision are supported by evidence coming from this large cohort study or may even alter their decision. At least the proposed model contains a reassuring element. It should be underlined that these algorithms do not obviate the need of weighing individual risks and benefits. We conclude from our observation that the prediction of future VTEs can well be done on clinical data alone -- at least until better genetic markers are established. In other words, well-established genetic parameters alone are relatively weak long-term risk factors, the occurrence of VTE requires interaction of both inherited and acquired risk factors \[[@B23]\]. Results of several recent studies support arguments against the possibility that testing for thrombophilia could help to better predict future VTEs \[[@B15],[@B20],[@B21]\]. Nonetheless one can argue that genetic testing in families with significant family history of VTE or previous experience of a VTE might well give additional information for clinical decisions and may increase efforts to comply with preventive measures. The limitation of our study is that we cannot further divide the risk spectrum in these sub-groups due to small numbers of new events or too short total WY of observation. In any case, when a genetic test is recommended, the physician should know how a positive test would influence his/her clinical judgment \[[@B15]\]. These early results of our cohort study contribute to this interpretation or decision-making, respectively. Forecast of VTE risk cannot be based on genetic characteristics alone but only in combination with important clinical data (acquired risk information). Genetic markers play obviously a limited role in the long-term prediction of VTE -- at least in the age group under 50. Genetic markers together with these \"personal characteristics\" constitute the disposition. Family history of cardiovascular events, specifically venous events have to be taken into account. As described before and confirmed by our data the probability whether the disposition translates into an event is obviously more influenced by \"personal characteristics\" such as higher age, or higher BMI. However, there are obviously other important, more acutely affecting environmental factors such as immobilization, surgery, accidents, and treatment with drugs that influence coagulation. The latter factors can be used to reduce the risk as estimated by the model (or own clinical experience). Another issue for discussion is the validity of our calculated incidence rates: The lowest VTE incidence level observed in this model was 4 per 10,000 WYs. Due to an active search for findings compatible with the diagnosis of VTE, the inclusion of definite and probable diagnosis as well as of so-called \"non-idiopathic VTEs\", the incidence figures were expected to be higher than in normal \"medical statistics\" or administrative databases as discussed above. The equivalent incidence rate for only definite and idiopathic VTE could be expected to be approximately 2 per 10^4^per year and therefore very low for women who were using oral contraceptives as the average population. We like to stress that our study was rigorous in documenting the VTE diagnosis. We conclude that the data can be generalized for the female population in the fertile age range. In analogy, the incidence might be compared with results of a prospective, community-based cohort study \[[@B13]\] that found in males aged 40--49 years a VTE incidence rate of 2.7 primary VTEs cases per 10^4^person-years (equal to idiopathic: no previous VTE history, no cancer, no surgery or trauma). The influence of other, more acutely acting risk modifiers -- such as immobilization, surgery, long-haul flights, and use of drugs (e.g. OCs and other hormones) was intentionally excluded from this model. Only parameters that were available at baseline and likely to affect the long-term development were eligible for this prognostic model. We saw no possibility to introduce parameters in the model that may or may not operate later, shorter or longer during the observational period. In other words, only long-term characteristics at baseline (both clinical and genetic variables) were initially included into the modeling. Other influential risk factors or preventive measures have to be considered when discussing activities to reduce a predicted increased risk in the medical practice. It was not the aim of the study and data are neither available to test the effect of preventive measures nor the effect of additional risk factors in the immediate period before the event occurred. This would require another study design and a separate study with sufficient power for such questions. The variables selected for the model fit the expectations of the skilled clinician. The model was developed to assist physicians- we hope for feedback from medical practice. This \"prognostic model\" seems to be worthwhile to be tested in clinical practice. There is a minority of women that would need additional genetic testing, intense counseling, suggestions for risk reduction (if possible), and efforts to prevent avoidable risk situations (e.g. treatment with OCs or hormones) or to take other appropriate preventive measures in situation of an acute risk (immobilization, surgery, long-haul flights and others), e.g. compression stockings/ heparin. Several other options to reduce the VTE risk profile are principally available, but sometimes not easy to achieve (e.g. reduction of BMI). Prevention of VTE in medical practice can be improved if the main risk factors are known (importance to document medical history and established risk factors), but also knowledge of their relative importance in the risk-network as well as of interactions with environmental factors. The predictive models discussed in this paper may assist doctors to pay particular attention to labeling prior to prescription (e.g. oral contraceptives or hormones) and to use convincing evidence-based data when counseling women of a predicted higher risk. We abstained -- at the current stage -- from the temptation to use a complex equation to calculate an apparently exact risk for the individual person (e.g., using a \"risk calculator\") because it suggests inadequate accuracy and we rather prefer to provide a very simple scheme that can be handled during routine work. Moreover, we are planning a validation of this model in an independent cohort study, as the first step the part related to clinical risk factors. Conclusion ========== Our prognostic tool -- containing clinical information (and if available also genetic data) -- seems to be worthwhile testing in medical practice in order to confirm or refute the positive findings of this study. Our cohort study will be continued to include more VTE cases and to increase predictive value of the model. Competing interests =================== The five authors from research institutes (LAJH, TDM, AA, WS, MS) and the two authors from industry (RS, JH) see no conflict of interest. Authors\' contributions ======================= LAJH: designed together with WS the cohort study and both are the principal investigators, LAJH planned all analyses, wrote a first draft of the manuscript. DMT: developed and maintained the database, performed the majority of analysis, and contributed to the manuscript. AA: responsible for running all field work, performing quality control and designing the validation of diagnoses, contributed to the manuscript. WS: one of the PIs, contributed to the manuscript. RS: responsible together with JH for interpretation of the findings, major contributions to the manuscript. JH see RS. MS: responsible for the haemostasiological lab work during the entire study period, contributed to the manuscript. Acknowledgements ================ We thank Professor Dr. L. Will-Shahab for the external medical review of all suspected VTE cases. We also thank Sabine Möhner for executing the follow-up over the years, and Andrea Dick for their work with the blood samples and determination of lab data.	PubMed Central	2024-06-05T03:55:56.234042	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087887/", "journal": "Thromb J. 2005 Apr 18; 3:5", "authors": [ { "first": "Lothar AJ", "last": "Heinemann" }, { "first": "Thai", "last": "DoMinh" }, { "first": "Anita", "last": "Assmann" }, { "first": "Wolfgang", "last": "Schramm" }, { "first": "Rolf", "last": "Schürmann" }, { "first": "Jan", "last": "Hilpert" }, { "first": "Michael", "last": "Spannagl" } ] }
PMC1087888	Background ========== The antiphospholipid syndrome is characterized by clinical evidence of arterial or venous thrombosis and repeated presence of antiphospholipid autoantibodies. The disease, first described by Hughes in 1983 \[[@B1]\], may occur in a variety of conditions including neoplasms, infections, other autoimmune disease such as lupus erythematodes, and after administration of certain drugs. If found without any other demonstrable disease, it is termed primary antiphospholipid syndrome. The autoantibodies, of which the most important are lupus anticoagulant and cardiolipin-antibodies, comprise a heterogeneous group which are mainly directed against complexes of anionic phospholipid with some phospholipid-binding proteins such as beta-2-glycoprotein I and human prothrombin. In a proportion of individuals the circulation of the antibodies may induce thrombosis of virtually any vein or artery. The most common are deep vein thrombosis, pulmonary, or cerebrovascular embolisms. In addition, other complications such as cardiomyopathy, hepatitis, hemolytic anemia, bleeding, vasculitis and renal failure have been reported \[[@B2]-[@B5]\]. The complexity of manifestations and the risk of severe complications, led us to conclude that this syndrome is an important differential diagnosis in patients with the described symptoms. Case Report =========== In September 2001, a previously healthy, Caucasian, 14-year-old boy was admitted to our hospital. Two weeks prior to his arrival, he had developed abdominal pain, a recurring fever of up to 39°C, and an intermittent cough. These symptoms were worsening and his general condition was deteriorating. On examination he had diffuse abdominal pain located primarily in the upper abdomen, and his temperature was 38.5°C. His skin and the further physical examination was normal. Pertinent laboratory investigations are listed in Table [1](#T1){ref-type="table"}. Of note, the activated partial thromboplastin time and prothrombin time were both prolonged, mild thrombopenia and leukocytosis of 14/nl were present, and C-reactive protein was elevated (6.1 mg/dl). The activity of factors II, V, VII and VIII was normal. There was no history of autoimmune diseases or coagulation disorders in his family. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Expression of important laboratory findings ::: Variable Day 1 Day 5 Day 10 Day 15 Day 25 Day 50 12 month Normal range ------------------------------------------ ------- ---------- ---------- ---------- ---------- ---------- ---------- -------------- IgG anti-cardiolipin antibody (GPL-U/ml) 44.7 21 60.2 30.7 37.7 \<12 Lupus anticoagulant positive positive positive positive positive absent Anti-β-2-Glycoprotein 1 (U/ml) 2 6 3 negative \<5 Anti-Phosphatidylserin (U/ml) 72.8 10.8 31.1 \<15 Anti-Phosphatidylethanolamin (U/ml) 45.2 23.9 18.5 \<15 Antinuclear antibodies negative negative absent ENA negative absent Double strand-DNA negative absent anti-adrenal-antibodies negative absent Cortisol 0800 h 26 1.1 22 28 6--26 Renin activity 100 \<3β ACTH (pg/ml) 468 \<50 Aldosteron (pg/l) \<10 12--125 DHEA-sulfate (g/dl) \<10 \<280 Bleeding time (min) 10.55 5.30 \<7 Partial thromboplastin time (s) 39 37 48 57 58 36 41 \<35 Prothrombin time (%) 64 55 55 73 78 45 36 70--100 Sodium (mmol/l) 134 140 111 139 142 143 136 132--145 Potassium (mmol/l) 3.9 4.8 3.6 4.2 4.0 3.5 3.9 3.1--5.1 C-reactive Protein (mg/dl) 6.1 6.1 12.1 3.7 1.0 0.4 0.3 \<0.5 Aspartate aminotransferase (U/l) 15 17 81 37 21 16 18 \<15 Alanine aminotransferase (U/l) 20 18 67 69 40 24 22 \<14 LDH (U/l) 201 442 309 256 181 \<240 Creatinine (mg/dl) 0.7 0.7 1.0 0.7 0.8 0.93 0.3 0.6--1.3 β2-microglobulin (mg/dl) 0.18 0.27 0.19 \<0.08 Hemoglobin (g/l) 13.6 11.1 7.6 10.8 11.8 12.1 12.4 11.8--16.8 Platelet count /nl 129 177 56 154 97 153 174 150--350 White cell count /nl 14.5 10.3 10.2 6.2 5.4 5.6 7.8 4.3--10.0 ::: The initial abdominal ultrasound revealed a tumor in the right pararenal space. However, a clear anatomical relation to the adrenal gland could not be established (Figures 1 and 2- see Additional file [1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}). The left adrenal gland was slightly enlarged. Doppler ultrasonography showed no thromboses of abdominal vessels and serum cortisol was normal at this stage. The CT scan identified the tumor as hematoma. Additional tests were positive for lupus anticoagulant and anti-cardiolipin antibodies. Bleeding time was prolonged to 11 minutes but all other coagulation tests eliminated the possibility of acute bleeding. In fact a repeated test of bleeding time a few days later was normal. Under intravenous treatment with cefuroxim the patient\'s temperature normalized within two days. But by day five, C-reactive protein increased to 12 mg/dl, and the patient\'s temperature rose again so antibiotic therapy was extended by gentamycin. Ten day after being admitted, the boy suddenly suffered generalized tonic-clonic seizures due to severe hyponatremia of 111 mmol/l. Serum transaminases raised to levels five times above normal range, and after three days hemoglobin dropped from 11.8 g/l to 7.6 g/l. Since LDH was elevated and haptoglobin was reduced, we diagnosed hemolytic anemia and transfused him with erythrocyte concentrates. Two days before the acute onset of adrenal insufficiency, microhematuria began. Proteinuria increased to 1 g/l and β-2-microglobulin raised to 0.16 mg/dl but there were no clinical signs of urinary infection. As the C-reactive protein rose to 16.1 mg/dl, intravenous antibiotic treatment was changed to erythromycin, ceftriaxon, tobramycin and flucloxacillin. However, two blood cultures, urine culture, stool culture and liquor culture remained sterile. Furthermore, serum serological tests were negative for the following infections: borrelia-burgdorferi, HIV, hepatitis B virus, hepatitis C virus, listeria, leptospira, epstein-barr-virus, treponema pallidum, parvovirus B 19, cytomegalie virus, mycoplasma pneumonia, coxsackivirus type B1-B6, chlamydia pneumonia and toxoplasmosis. Tuberculosis was ruled out by mendel-mantoux skin test, microscopy of Ziehl-Nelson stained urine samples, PCR of a urine and liquor sample, and chest x-ray. The boy was transferred to the intensive care unit and hyponatremia was balanced within the next 12 h with intravenous application of NaCl 5.85%. Over the course of the next 5 h until his serum sodium was above 125 mmol/l, he had repeated convulsions for up to 30 minutes. Due to status epilepticus and marked agitation, intubation and artificial ventilation was required. A second CT scan on the next day showed a novel hemorrhage of the left adrenal as well and bleeding in the right pararenal region had enlarged (Figure 3- see [Additional file 3](#S3){ref-type="supplementary-material"}). A CT scan of the head was normal, cortisol was low and the adrenals were unresponsive to the Synacten test. We started treatment with hydrocortisone, fludrocortisone and 10 days later with phenprocoumon. Under this treatment the boy improved clinically within two days and was extubated. As well, laboratory signs of hepatitis normalized within the next three weeks and hemoglobin remained in normal range after blood transfusion. The C-reactive protein normalized within 10 days and antibiotic treatment was stopped as no other clinical or laboratory signs of infection were present. We continued to monitor the patients condition. As shown in table [1](#T1){ref-type="table"}, lupus anticoagulant and cardiolipin antibodies remained positive. He did not develop any other autoimmune diseases (Table [1](#T1){ref-type="table"}). The boy\'s parents and his brother were negative for lupus anticoagulant and cardiolipin antibodies. Discussion ========== The antiphospholipid syndrome is a rare disease in childhood, significantly related to thrombotic events. However, in our patient, initial acute intraabdominal hemorrhage was the major finding. Due to the large hematoma, no clear anatomical association to retroperitoneal organs could be established. Although acute bleeding may sometimes be associated with this syndrome, in this case, it was obvious that microthrombotic events in the adrenal veins led to secondary hemorrhage, because coagulation tests were never critically changed (Tab. [1](#T1){ref-type="table"}). Additionally, in the ultrasounds during the first week, the left adrenal gland was slightly enlarged. This was an indirect sign of thrombosis, which proceeded the hemorrhaging 10 days later. In most reported cases, adrenal failure has been linked to thrombotic complications of veins leading to bilateral hemorrhage \[[@B4],[@B6],[@B7]\]. There are five other reported cases of antiphospholipid syndrome and adrenal failure in childhood \[[@B6],[@B7],[@B9]-[@B11]\]. However, the clinical presentation is very variable (see Tab. 2- see [Additional file 4](#S4){ref-type="supplementary-material"}) and gastrointestinal symptoms are not always present. The acute adrenal failure 10 days after admission, documented by the novel hemorrhage of the left adrenal gland in the second CT scan, was followed by acute hyponatremia caused by bilateral adrenal insufficiency. Infarction or hemorrhage of the adrenal is a rare but known complication of the antiphospholipid syndrome. Signs of adrenal insufficiency appear when more than 90% of the adrenal cortex is destroyed. The possibility of infectious or autoimmune adrenalitis, by far the most common cause of adrenal insufficiency, was excluded from consideration. In our patient, the onset of organ insufficiency occured very rapidly as cortisol and serum sodium were still normal six and two days before, respectively. Thus, the failure of the complete adrenal function has to be attributed to the second hemorrhage. In addition, acute hyponatremia was accompanied by hemolytic anemia and elevated transaminases. Recent studies suggest that these symptoms are autoimmune mediated, however, the pathogenic mechanism is not well understood \[[@B2]-[@B4]\]. One different hypothesis is due to the unique nature of the vascular anatomy of the adrenal glands since they are rich with arteial supply (3 arteries with up to 60 branches) but limited with venous drainage by a single vein. This however may predispose thrombotic events. Another complication in our patient was the development of renal symptoms combined with hematuria and proteinuria. Although renal function and blood pressure was appropriate, β-2-microglobulin, a marker for tubular injury, increased during the phase of hematuria (Table [1](#T1){ref-type="table"}). In a recent study Nochy et al. reported 16 patients with APS. All of them had biopsy proven vascular nephropathy caused by small vessel vaso-occlusive lesions \[[@B8]\]. Therefore, it is possible that involvement of the small renal vessels also played a role in our patient. This case demonstrates that APS in childhood may present with a diverse and severe set of symptoms. Adrenal hemorrhage due to thrombosis should be considered as a manifestation of the antiphospholipid syndrome. These patients must be closely monitored because consecutive thrombosis with secondary hemorrhage may rapidly cause adrenal insufficiency. Competing Interests =================== The author(s) declare that they have no competing interests. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table 2 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Figure 1: scanned ultrasound photography ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Figure 2: scanned ultrasound photography ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 Figure 3: scanned photography of computer tomography ::: ::: {.caption} ###### Click here for file :::	PubMed Central	2024-06-05T03:55:56.238098	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087888/", "journal": "Thromb J. 2005 Apr 18; 3:6", "authors": [ { "first": "Patrick", "last": "Gerner" }, { "first": "Michael", "last": "Heldmann" }, { "first": "Peter", "last": "Borusiak" }, { "first": "Vladimir", "last": "Bures" }, { "first": "Stefan", "last": "Wirth" } ] }
PMC1087889	Introduction ============ NK cells are key mediators of innate immunity contributing to immunosurveillance by recognizing and killing tumor and virus-infected cells. They are cytolytic and produce inflammatory cytokines \[[@B1],[@B2]\]. Mature NK cells are CD3-CD56+ and variably CD16+. The molecule CD56 is a 120--180 KD N-linked glycosylated isoform of the neural cell adhesion molecule (NCAM) \[[@B3]\]. It is expressed on NK cells, NK-T cells, and a subset of dendritic cells. NK cells originate in the bone marrow from a CD34+Lin- common lymphoid progenitor cells \[[@B4]\]. In the absence of bone marrow stroma, NK cell generation requires a combination of IL-2 or IL-15 \[[@B5],[@B6]\] and stem cell factor (SCF) \[[@B7]\]. However, the early stages of CD56+ cell generation and the origin of diversification into mature CD56+ cell types are not well characterized. We previously found that in culture with IL-2 and SCF CD34+ cells differentiate into several CD56+ subpopulations -- a minor myeloid subset consisting of large CD56^dim^CD33+ macrophage-like cells and a major lymphoid subset of CD56^bright^cells. Both cell types had low or absent perforin and no granzyme B \[[@B8]\]. In studying the function of immature CD56+ cells, we observed that they had negligible cytotoxicity. Here we describe a novel cell-contact dependent proliferation inhibition of cell lines by cultured CD56+ cells which suggests that immature CD56+ cells may have novel growth regulatory properties. Materials and methods ===================== Antibodies and reagents ----------------------- Fluoroscein isothiocyanate (FITC)-conjugated anti-CD56, anti-CD16, anti-CD33, anti-CD3, anti-CD2, anti-CD11a, anti-CD94, anti-CD80, anti-CD44, anti-granzyme A, Allophycocyanin (APC)-conjugated anti-CD56, anti-CD11c, anti-CD38, anti-CD69, anti-CD117, (Pe)-conjugated anti-CD117, NKB1 (KIR3DL1), anti-CD3, anti-CD16, anti-CD56, anti-perforin, PerCP-conjugated anti-CD3, anti-CD69, anti-CD8 and matching isotype mouse mAbs were purchased from Becton and Dickinson (S Jose, CA). Pe-conjugated anti-CD34, P58.1 (NK2DL1), P58.2(NK2DL2), NKG2A were purchased from Immunotec (Marseille, France). Magnetic beads-conjugated anti-CD56 and mini Macs magnet were purchased from Miltenyi Biotec (Auburn, CA). (APC)-conjugated anti-CD95 and (Pe)-conjugated anti-CD95L were purchased from Caltag (Burlingame, CA). Hyaluronic acid was purchased from Sigma (S Louis, Mo) Cell isolation, activation and expansion ---------------------------------------- CD34+ cells were positively selected from normal donor G-CSF-mobilized peripheral blood stem cells, (PBSC) counted, and frozen in liquid nitrogen until use. All donors gave written informed consent to donate stem cells in NIH protocols 99-H-0046 and 95-H-0049. Peripheral blood mononuclear cells (PBMC) were separated by Ficoll-hypaque density separation. Cells were cultured in RPMI 1640 supplemented with 10% AB or 10% FCS serum, glutamine (2 mM) gentamicin, hereafter referred to as complete medium. CD34+ cells were cultured in complete medium, in 24 or 12 or 96 U well plates (Costar), for a minimum of 10 to a maximum of 70 days. Cells were stimulated every 5--7 days with SCF (20--50 ng/ml) (Peprotech, Rocky Hill, NJ) with or without IL-2 (200 U/ml) or IL-15 (1--100 ng/ml). To obtain pure NK populations, CD34+ cells stimulated for 15--21 day with IL-2 were stained with a Pe-conjugated anti-CD56 Moab and CD56+ cells were isolated by electronic sorting using an EPICS ALTRA Flow cytometer (Beckman Coulter, Miami, FL). In some experiments, immature CD56+ cells were selected with an anti-CD56 conjugated magnetic bead column (Miltenyi Inc.). Vigorous mechanical pressure eluted the CD56+cells retained in the column. Peripheral blood NK cells were negatively selected by magnetic sorting using a Miltenyi isolation kit. Positively and negatively selected peripheral blood NK cells were further expanded in vitro as follows: 100 μl of 1 × 10^5^/ml CD56+ were mixed with 100 μl of 75 Gy irradiated LCL cells in complete medium supplemented with IL-2 (10 U/ml) and 15 % conditioning medium were plated in Costar 96/w round bottom plates. Cells were stimulated every 3 days with CM supplemented with 15% CM for the required time. Flow cytometric analysis ------------------------ In some experiments, cells were stained with Pe-Conjugated anti-CD56 or anti-c-Kit (one color). In other experiments, cells were incubated with FITC-anti-CD56 and Pe-anti-c-Kit (two colors) or a combination of Pe, FITC, PerCP, and APC-conjugated antibodies specific for the desired molecules (four colors). In all cases the cells were stained on ice, for 30 minutes, washed twice, fixed in 1% paraformaldehyde (PFA). For intracellular staining experiments (IC), 1 million cells were first stained with a Pe-conjugated anti-CD56 for 15 minutes at room temperature (RT) in the dark than 2 ml of FACS lysing solution was added to the cell mixture. After 10 minutes incubation at RT, cells were washed and permeabilized with 0.5 ml FACS perm mix for 10 minutes at RT. Cells were than stained with an anti-Pe-conjugated anti-Perforin Moab and a FITC-conjugated anti-granzyme A moAb or Pe and FITC-conjugated mouse control moAb isotypes for 30 minutes at RT, washed and fixed in 200 ul of 1% PFA. Cells were analyzed by a 4 filters BD FACScan Excalibur. For trans-well experiments, after an initial equilibration (1 hour at 37 C) with complete medium, 200 μl of 5 × 10^4^- 1 × 10^5^CD34 cell- derived NK cells or IL-2 activated NK cells cultures were incubated in complete medium in the upper trans-well compartment of a 12-transwell plate (Costar, Cambridge, MA) in the presence or absence of 5 × 10^3^- 5 × 10^4^K562, while 1 ml culture of 5 × 10^3^K562 were cultured in the lower compartment. A 0.4 μm polycarbonate membrane separated upper and lower compartments. After a 2-day incubation, aliquots of 200 μl of K562 cell culture were removed from the lower compartment and labeled with 3H-TdR. Breaking and washing the transmembrane then permitted harvesting the upper compartment. Cells were counted or labeled with 3H-TdR. Microcytotoxicity assay ----------------------- For reverse antibody-directed cell cytotoxicity (ADCC) 6 replicates of 20 μl of effector cells were incubated in a 60-well (40 μl depth) Terasaki plate for 30 m at room temperature in the presence or absence of 5 μl (10μg/ml) 3g8 (anti-CD16). At the same time, FcR+P815 (2 × 10^6^) were incubated in 1 ml of complete medium supplemented with 10 μl of Calcein-AM (Molecular Probes, Junction City, OR) for 30\' at 37°C, washed four times and diluted to 1 × 10^5^/ml. After diluting the effector cells, 10 μl of target cells were added, plates were centrifuged and incubated at 37°C for 4 hours. In some experiments, effectors with no mAbs were challenged with K562 cells. A few minutes before scanning the plates using a fluorescent detector 5 μl fluoro-quench was added to each well. The percent of lysis was calculated as follows: 1-(mean test-mean blank)/(mean max-mean blank) ×100. Reverse transcriptase polymerase chain reaction ----------------------------------------------- Total RNA was extracted from 1 × 10^6^positively selected G-CSF-mobilized peripheral blood CD34+ cells, PBMC, resting or IL-2 activated total peripheral blood CD56+ cells and NK cells using Triazol reagent; complementary DNA (cDNA) first strand was produced using Maloney murine leukemia reverse transcriptase (Roche) with an oligo(dt)^12--18^anti-sense primer (Roche). The cDNA of interest was amplified as shown in table [I](#T1){ref-type="table"}. Amplified fragments were analyzed in 1.5 % agarose gel electrophoresis in the presence of ethidium bromide (Sigma, S Louis, MO). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### List of primers used in this study ::: ------------ ------------------------------------------ -------------------------- ------------ Sequences Sense Anti-Sense Transcript Perforin cggctcacactcacagg ctgccgtggatgcctatg 369 Granzyme B ggggaagctccataaatgtcacct tacacacaagagggcctccagagt 431 NKp30 cagggcatctcgagtttccgacatggcctggatgctgttg gatttattggggtcttttgaag 606 NKp44 tacttcaaagtgtggcag tcacaaagtgtgttcatcatc 751 NKp46 aaaagcaagtgaccatct aagaacatgcttgttgcagt 337 NKp80 caagatgaagaaagataca gagaaccatccacccaagt 568 NKG2D gaaggcttttatccacaa ttacacagtcctttgcat 761 ------------ ------------------------------------------ -------------------------- ------------ NKp30 NKG2D sense12, perforin and granzyme B11have been already described ::: Proliferation assay ------------------- The proliferation of K562 and P815 was measured using the tritiated thymidine incorporation (3H-TdR) assay. The first three U-wells of each horizontal row of 96 well plates were filled with 200 μl of negatively selected NK cells or positively selected peripheral blood CD56+ cells, then 100 μl of cultured cells were serially diluted in the remaining wells previously filled with 100 μl of CM. Later, 1 × 10^4^K562 or P815 cells were added to the cell cultures. After 2 days incubation, cells were pulsed with 1 μCi of 3H-TdR per well (Amersham Biosciences, Piscataway, NJ) Eighteen hours later, 3H-TdR was measured using a beta scintillation counter. Results ======= Kinetics of CD56+ cell generation --------------------------------- After three weeks stimulation with IL-2 and SCF, G-CSF mobilized peripheral blood CD34+ cells generated CD2-, CD56^high^(10.6 ± 11.4) (range 3--39 %) and CD56^low^(3.4 ± 2.0) (range 0.7--5 %) in a sample size of 7 normal donors (REF). Some CD56^high^cells expressed CD94 and NKG2A but lacked expression of KIR with immunoglobulin like domains KIR2DL2, while CD56^low^were KIR negative (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Phenotypic analysis of G-CSF mobilized CD34+ cells cultured in SCF and IL-2 After 3 week stimulation with SCF (100 ng/ml) and IL-2 (200 U/ml), G-CSF mobilized CD34+ cells were stained with for the indicated NK cell surface antigens using specific moAb. ::: ![](1479-5876-3-15-1) ::: Immature CD56+ cells lack granzyme B and perforin ------------------------------------------------- We next evaluated the stage of differentiation of molecules associated with cytotoxicity in immature CD56+ cells. Perforin and granzyme B mRNA expression was measured by RT-PCR in CD34+ cells stimulated with IL-2. Perforin was observed on day 0--1 of culture, but disappeared by day 7. Intracellular staining of immature CD56+ cells revealed a granzyme A content comparable to that found in IL-2 activated PBL (Fig. [2](#F2){ref-type="fig"}). Notably, the reduction of GAPDH band intensity may reflect loss of viability of the cells upon medium term culture. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Cytotoxic granule content of immature CD56+ cells: A. Absence of perforin and granzyme B in CD34+ cells incubated with IL-2 and SCF for up to 15 days compared with control (representative of three experiments). BCD56+ cells from three week cultures showing presence of granzyme A but not perforin compared with control IL-2 stimulated PBL. ::: ![](1479-5876-3-15-2) ::: Induction of NKp46 in immature CD56+ cells ------------------------------------------ We then sought to determine whether the stimulation of CD34+ cells with IL-2 and SCF induced NK natural cytotoxic receptor (NCR) expression. Within two week of culture, immature CD56+ cells expressed some NK activation molecules including CD44, CD69, and CD38 (data not shown) but did not express NCR genes until at least 6 weeks. At than time CD56+ cells expressed NKp46 but not NKp30. Traces of NKG2D and NKP80 RNA expression were also detected by PCR (Fig. [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Gene expression of NK activating molecules on CD34-derived CD56+ cells upon stimulation with SCF+IL-2. CD34+ cells were stimulated with SCF and IL-2. At the indicated time, RNA was isolated. NK activating molecules mRNA gene expression was analyzed. By day 15 incubation 39% of CD56+ cells were detected by flow cytometry in the cell culture. ::: ![](1479-5876-3-15-3) ::: Functional assays of CD56+ cells -------------------------------- IL-2 activated immature CD56+ cells showed negligible cytotoxicity against K562 compared with the potent lysis exhibited by peripheral blood NK cells. Similarly in a reverse ADCC assay, immature CD56+ cells coated with an anti-CD16 (3g8) showed only low cytotoxicity against the FcR+ cell line, P815, while control peripheral blood NK cells were strongly cytotoxic. We then investigated the effect of CD56+ cells on proliferation of cell lines. Resting CD34+ cells did not inhibit K562 proliferation, while flow sorted, three week cultured CD56+ cells cultures strongly inhibited K562 cell proliferation in a dose dependent manner, reaching 90% inhibition at an E:T ratio of 10:1. The proliferation inhibition induced by immature CD56+ cells ranged between 29 and 96.5 % (Table [2](#T2){ref-type="table"}). Immature CD56+ cells also comparably inhibited the proliferation of the NK resistant cell line P815, indicating that the proliferation inhibition was independent of the resistance of this line to NK-mediated cytotoxicity. Unlike immature CD56+ cells, peripheral blood-derived CD56+ cells proliferated in the presence of IL-2. Nevertheless the proliferation attributable to K562 cells was also abolished when cultured with CD56+ cells (Fig. [4](#F4){ref-type="fig"}). Trans-well experiments were used to determine whether the inhibition of K562 cell proliferation was mediated by cell-interaction or by soluble factors. In three experiments, the average 3H-TdR incorporation of K562 cells when separated by a membrane from immature CD56+ cells was 5361 ± 1967 versus 5110 ± 1539 cpm for K562 growth in the absence of CD56+ soluble factors. This suggested that proliferation was primarily blocked by cell-cell contact (Table [3](#T3){ref-type="table"}). To further examine the mechanism of K562 proliferation inhibition, we cultured K562 with either magnetically sorted immature CD56+ cells in a transwell upper chamber and K562 cells in the lower chamber for two days. Viable K562 cells (\> 95% trypan blue negative cells) were recovered from immature CD56+ cells but not from mature CD56+ cultures. After depleting the K562/immature CD56+ cultures of CD56+ cells using antibody-coated magnetic beads, residual K562 cells again proliferated, indicating that in the absence of cytotoxicity, proliferation inhibition was reversible (data not shown). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Functional features of immature CD56+ cell upon SCF+IL-2 stimulation. A, cytotoxicity of immature CD56+ cells (open squares) compared with IL-2 activated peripheral blood NK cells (closed circles). B. Cytotoxicity against P815 cells by immature CD56+ cells incubated with (open circles) or without 3G8 (open squares), compared with peripheral blood NK (closed symbols in the presence (closed triangles) or absence 3G8 (closed circles). This was a representative experiment of series of three. CResting CD34+ cells (open squares) or IL-2 activated peripheral blood NK cells (open circles) were incubated at different ratio with K562. D. After 21 days stimulation with IL-2 and SCF, immature CD56+ cells were isolated from whole cells by electronic sorting and incubated in the absence (open squares) or presence (close squares) of K562. IL-2 activated peripheral blood NK cells were cultured in the absence (close circles) or presence (open circles) of K562 (representative of ten experiments). E. Inhibition of proliferation of the NK resistant cell line P815 by IL-2 activated peripheral blood NK cells (open circles) or with IL-2 activated peripheral blood NK cells (closed squares). Cells were then labeled with 3H-TdR and radioactivity (CPM) was measured by a β-counter (representative of three experiments). ::: ![](1479-5876-3-15-4) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### CD34-derived CD56+ cells inhibited K562 proliferation ::: ----------- ---------------------------- ----------------- ----------------------------- 3H-TdR (CPM) ImmatureCD56+ cells + K562 K562 only \% proliferation inhibition 1 214 6183.5 96.5 2 267 6183.5 95.7 3 8183 11172 27 4 1805 5283 66 5 1944.5 4720 59 6 743 11232 93.4 7 2708 10323 74 8 355 2524 86 9 138 2524 94.6 10 829 7336 89 Mean ± SD 1718.7 ± 2433.5 6748.1 ± 3255.5 78 ± 22.2 ----------- ---------------------------- ----------------- ----------------------------- Electronically or magnetically sorted CD34-derived CD56+ cells and incubated 10:1 ratio with K562 and cell proliferation was evaluated after 2-day culture. The difference in the mean CPM incorporation between immature CD56+cells+K562 and K562 alone was highly significant with a P value of 0.0011. The range of CPM incorporation in the presence of immature CD56+ cells and K562 was 138--8103 while in the presence of K562 alone was 2524--11232. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Transwell Experiments ::: 3H-TDR (CPM) ------- ----------------- ----------------- ------------ \Immature CD56 \\Mature CD56 \\\K562 Exp 1 6781 ± 51 5941 ± 560 4429 ± 197 Exp 2 3116 ± 65 2534 ± 196 4030 ± 239 Exp 3 6187 ± 512 2932 ± 181 6873 ± 34 Sorted Immature \* or mature CD56+ cells\\ or K562 were incubated 10:1 ratio in the upper chamber at of Transwell plates while K562 were incubated in the lower chamber. After 2 days, lower K562 were collected labeled overnight with 3H-TDR and analyzed in a β-counter 0.4 μm polycarbonate transwell membrane were removed and the upper compartment cells were harvested and evaluated for viability by trypan blue exclusion assay. ::: Discussion ========== Prolonged culture of human CD34+ cells with IL-2 or IL-15, with or without bone marrow stroma cells can generate cytotoxic CD56+ cells \[[@B9],[@B5]\]. These cells are CD94+ and NKG2A+ but it is not clear whether they express other KIRs \[[@B9]-[@B11]\]. Here we studied the early stages of NK cell generation from G-CSF mobilized CD34+ cells. After three weeks stimulation with SCF+IL-2, we identified an immature NK population of CD94+ NKG2A+ KIR- CD56+ cells. We previously found that these immature CD56+ populations are heterogeneous. Notably there is a minor subset of CD56^dim^CD33+ cells that may be precursors to a novel population of CD56+ monocytes \[[@B12]\]. The major CD56+ population with bright CD56+ expression and lymphoid features include a c-Kit+ CD11a- cell and a more mature c-Kit- CD11a+subset. The c-kit+ cell may be the precursor to the c-kit- cell which has lost responsiveness to SCF and has acquired integrins necessary for formation of effector-target conjugates and killing \[[@B13]\]. However, all these CD56+ subsets were immature with respect to functioning cytotoxic apparatus. Mature cytotoxic NK cells can be generated from CD34+ cells when cultured with an IL-15-producing human spleen fibroblast cell line \[[@B14]\]. However, the generation of fully functional NK cells takes many weeks of cell culture. Using a stroma cell line stimulated with IL-15 Sivori et al did not observe NK mediated cytotoxicity against K562 within the first of month culture. After one month, cells exhibited cytotoxicity but remained KIR negative \[[@B15]\]. Similarly, in our experiments, long term culture induced some mature NK markers -- NKp46 and NKp80 gene expression. However within the first month of culture CD34-derived CD56+ cells exerted negligible cytotoxicity. In the absence of any T, B or myeloid cell markers and some NK marker (including NK activation markers CD38, CD44, and CD69), we consider the major population of CD56+ cells to be immature NK cells \[[@B16]-[@B19]\]. Their failure to exert cytotoxicity is consistent with perforin and granzyme B knock-out mouse models which lack cell-mediated cytotoxicity, while granzyme A knock-out mice retain cytotoxicity \[[@B20]-[@B22]\]. As a consequence, while mature NK cells undergo functional anergy and apoptosis on contact with K562 cells \[[@B23],[@B24]\] mixed cultures of immature CD56+ and K562 maintained cell numbers. Because immature CD56+ cells produce a variety of cytokines \[[@B6]\], we explored the possibility that they might have functional properties other than cytotoxicity. Remarkably, purified immature CD56+ cells strongly inhibited proliferation of both K562 cells and the NK resistant cell line P815 at low E:T ratios. The absence of detectable perforin excluded the possibility that the effect was due to contamination by a small population of surviving NK cells. Maximum inhibition of proliferation was only seen when effector-target contact occurred. This suggests that immature CD56+ cells, while lacking cytotoxicity, had a novel cell-contact dependent cytostatic effect. Our findings support earlier observations suggesting that bone marrow NK cells regulate hematopoiesis through a non-cytotoxic pathway \[[@B25]-[@B28]\]. It is not clear what is the mechanism utilized by immature CD56+ cells for K562 and P815 proliferation inhibition. Cell surface TGF-beta expression on immature CD56+ cells may be responsible for such inhibition since CD34+ cells upon SCF and IL-2 stimulation acquire TGF-beta gene expression (data not shown). In conclusion, we describe here early stages of NK cell generation from G-CSF mobilized CD34+ cells in the presence of SCF and IL-2. Although immature NK cells lack the cytotoxic apparatus required for classical NK-like cytotoxicity they had the unusual property of inhibiting proliferation of K562 and P815 cell lines. In our future studies we are planning to assess whether the antiproliferative effects of immature CD56+ cells will also involve cells of myeloid lineages and non hematopoietic cell lines. It is possible that these cells normally reside in the bone marrow and have a regulatory effect on hematopoiesis Acknowledgements ================ We thank Philip McCoy and Keyvan Keyvanfar (NHLBI, HB, NIH, Bethesda, USA) that kindly provided us technical assistance for electronic cell sorting.	PubMed Central	2024-06-05T03:55:56.240478	2005-4-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087889/", "journal": "J Transl Med. 2005 Apr 14; 3:15", "authors": [ { "first": "Giuseppe", "last": "Sconocchia" }, { "first": "Maurizio", "last": "Provenzano" }, { "first": "Katayoun", "last": "Rezvani" }, { "first": "Jongming", "last": "Li" }, { "first": "Jos", "last": "Melenhorst" }, { "first": "Nancy", "last": "Hensel" }, { "first": "A John", "last": "Barrett" } ] }
PMC1087890	Background ========== Avian metapneumovirus (aMPV) causes turkey rhinotracheitis (TRT) and is associated with swollen head syndrome (SHS) of chickens that is usually accompanied by secondary bacterial infections which can increase morbidity and induce mortality. Avian metpnuemovirus was first reported in South Africa during the early 1970s and was subsequently isolated in Europe, Israel and Asia \[[@B1],[@B2]\]. During 1997, mortality due to aMPV infections among commercial turkeys in the U.S. ranged from zero, to 30% when accompanied by bacterial infections, with condemnations due to air sacculitis. This was the first reported outbreak of aMPV infections in the U.S. which was previously considered exotic to North America. The virus causing disease was designated a new aMPV type C genetically different from European counterparts \[[@B3]-[@B5]\] and was subsequently demonstrated to be most closely related to human metapneumovirus (hMPV) from diverse geographic locations \[[@B6],[@B7]\]. Infections among commercial turkeys with aMPV/C continue in the north-central U.S. resulting in substantial economic loss to the poultry industry \[[@B6],[@B8],[@B9]\]. Pneumoviruses are members of the family Paramyxoviridaethat contain a nonsegmented, negative-sense RNA genome of approximately 15 kb in length. Viruses related to aMPV include human, bovine, ovine and caprine respiratory syncytial viruses and pneumonia virus of mice \[[@B10]\], as well as the recently identified hMPV \[[@B11]\]. Although genome length is similar, pneumoviruses generally encode ten genes, compared to six or seven in other paramyxoviruses. These include the nonstructural proteins (NS1 and NS2), nucleoprotein (N), phosphoprotein (P), matrix protein (M), small hydrophobic protein (SH), surface glycoprotein (G), fusion protein (F), second matrix protein (M2) and a viral RNA-dependent RNA polymerase (L). The pneumoviruses have an F protein that promotes cell fusion, but these viruses do not hemagglutinate, nor do they have neuraminidase activity in their G attachment protein. This is an important distinguishing characteristic from the other paramyxoviruses \[[@B10]\]. Because of a limited genome size, many non-segmented RNA viruses, including the pneumoviruses, have devised mechanism to increase protein coding capacities. This may occur at two levels: 1) transcriptional mRNA processing or modification \[[@B12]-[@B14]\] or 2) translational, in which proteins may be produced from alternative open reading frames (ORFs) or from translational initiation at non-AUG or downstream AUG codons \[[@B15]-[@B17]\]. Among the pneumoviruses, secondary coding usage has only been documented for the M2 gene, which encodes two proteins. The M2-1, a transcription antitermination factor, is required for processive RNA synthesis and transcription read-through at gene junctions. The M2-2 is involved with the shift between viral RNA transcription and replication \[[@B18]\]. In this report, we present evidence for utilization of a secondary open reading frame, within the N gene encoding a truncated nucleoprotein (N2) among aMPV/C/Co and aMPV/A/UK/3b infected cells. Results ======= Avian metapneumovirus N gene possess several putative AUG start sites --------------------------------------------------------------------- The aMPV/C/US/Co nucleoprotein is encoded by the N gene with a predicted molecular weight of 42--45 kD \[[@B7],[@B19]\]. The N gene ranges from 1191 to 1206 nucleotides in length \[[@B6],[@B19]\], with the first AUG at nucleotide position 14 (Fig. [1](#F1){ref-type="fig"}) in all three subtypes (A, B, and C). The aMPV/C/US/Co N gene has additional putative start sites at nucleotide positions 212, 350, 416, 758, 785, 827, 896, and 1022 with \"true\" Kozak sequences \[[@B20]\] at nucleotide positions 413 (ACC[AUG]{.underline}G) and 893 (GAG[AUG]{.underline}G), with predicted translation products of 28.5 kD and 10.78 kD, respectively. The aMPV/A/UK/3b N gene has additional putative start sites at nucleotide positions 161, 212, 293, 410, 413, 605, 722, 749, 749, 776, 818, 887, and 1013 with \"true\" Kozak sequences \[[@B20]\] at nucleotide positions 602 (AGG[AUG]{.underline}G), 719 (AGG[AUG]{.underline}G), and 884 (AAA[AUG]{.underline}G), with predicted translation products of 21.26 kD, 16.73 kD, and 10.54 kD, respectively. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Alignment of avian metapneumovirus type A and C nucleoprotein genes demonstrating presence of multiple start sites. Underlined sequences denote hypothesized alternative in-frame start sites and the stop codon. Primer sequences utilized for cDNA synthesis of nucleoprotein genes are also illustrated. ::: ![](1743-422X-2-31-1) ::: Avian metapneumovirus-infected cells produce two proteins (N1 and N2) encoded by two open-reading frames within the N gene -------------------------------------------------------------------------------------------------------------------------- Five peptides within the aMPV N gene (Fig. [2](#F2){ref-type="fig"}) were utilized to generate affinity-purified rabbit peptide antibodies. This approach was exploited to determine if any of the alternative start sites of the aMPV N gene were utilized during an active cell infection. aMPV/N-peptide antibody directed against aMPV/C/US/Co N protein amino acids 10--29 (DLSYKHAILKESQYTIKRDV) with only 3 changes in both aMPV types A and B at amino acid positions 12 (S to E), 19 (K to D) and 26 (K to R) reacted with all three full length nucleoproteins by western blot (Fig. [3A](#F3){ref-type="fig"}, Lanes 3, 4, and 5), but did not react with any proteins in uninfected Vero cells (Fig. [3A](#F3){ref-type="fig"}, Lane 2). All three virus nucleoproteins were between 42--45 kD based on SDS-PAGE/western blot analysis (Fig. [3A](#F3){ref-type="fig"}). We then tested the aMPV/C-N2 peptide antibody directed against amino acids 128--148 in the mid-portion of the of the aMPV/C/US/Co isolate (Fig. [2](#F2){ref-type="fig"}) by western blot which would recognize any downstream translation products encoded by the N gene and utilization of any secondary start sites. Western blot analysis revealed two putative N gene products in aMPV/C/US/Co-infected Vero cells, the first, the full-length nucleoprotein with a molecular weight of approximately 43 kD (Fig. [3B](#F3){ref-type="fig"}, Lane 3) and the second, a smaller protein of approximately 35--36 kD (Fig. [3B](#F3){ref-type="fig"}, Lane 3). The peptide antibody to amino acids 303 to 393 (aMPV/C-N4) synthesized to be reactive to the C-terminal N protein from aMPV/C also recognized two proteins as in Fig. [3B](#F3){ref-type="fig"}, Lane 3 (data not shown). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Relative position of peptides within the avian metapneumovirus nucleoproteins utilized for generation of affinity purified polyclonal antibodies. ::: ![](1743-422X-2-31-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Detection of avian metapneumovirus (aMPV) nucleoprotein gene products among infected cells utilizing affinity purified peptide antibodies. A. Antibody reacted against an N-terminal portion of the nucleoprotein designed to detect all aMPV serotypes N1. Lane 1: molecular size markers; Lane 2: uninfected cell proteins; Lane 3: aMPV/A infected cell proteins; Lane 4: aMPV/B infected cell proteins; Lane 5: aMPV/C infected cell proteins. B. Antibody detection of a C-terminal portion of the aMPV/C nucleoprotein. Lane 1: uninfected cell proteins; Lane 2: aMPV/C infected cell proteins reacted with N1 peptide antibodies; Lane 3: aMPV/C infected cells reacted with aMPV/C-specific N2 peptide antibodies. C. Antibody detection of a C-terminal portion of the aMPV/A nucleoprotein. Lane 1: uninfected cell proteins; Lane 2: aMPV/A infected cell proteins reacted with N1 peptide antibodies; Lane 3: aMPV/A infected cell proteins reacted with N3 peptide antibodies; Lane 4: aMPV/A infected cells reacted with N5 peptide antibodies. ::: ![](1743-422X-2-31-3) ::: To evaluate whether the utilization of alternative start sites was unique to members of the aMPV type C group, or whether this also occurred in other aMPV types, we utilized aMPV/A-N3 and aMPV/A-N5 peptide antibodies (anti-aMPV/Type A, N protein, amino acids 126--145 and 380--390, respectively). Unlike aMPV/C-N2 peptide antibody, aMPV/A-N3-peptide antibody (amino acids 126--145) reacted to only a full length nucleoprotein (Fig. [3C](#F3){ref-type="fig"}, lane 3) similar to the aMPV/N-peptide antibody (Fig. [3C](#F3){ref-type="fig"}, lane 2), while aMPV/A-N5-peptide antibody (amino acids 380--390) reacted with both the full length nucleoprotein of approximately 41--43 kD (Fig. [3C](#F3){ref-type="fig"}, lane 4) and a smaller protein of approximately 28--30 kD (Fig. [3C](#F3){ref-type="fig"}, lane 4). Finally, all aMPV type-specific antibodies were not cross active with other metapneumoviruses (data not shown). Expression of the N1 and N2 ORF of avian metapneumovirus type C/Colorado in eukaryotic cells -------------------------------------------------------------------------------------------- Sequence analysis of the aMPV/C/US/Co and aMPV/A/UK/3b N gene nucleotide sequences revealed that downstream of the first AUG (position 14) were multiple putative start sites as described above (Fig. [1](#F1){ref-type="fig"}). We therefore utilized sequence analysis software to analyze the N gene putative open reading frames and the predicted translation products from each putative start site for products that would result in proteins of approximate size as the smaller reactive band that was detected by western blot (Fig. [3B](#F3){ref-type="fig"}, lane 3 and Fig. [3C](#F3){ref-type="fig"}, lane 4). Two predicted proteins in the aMPV/C/US/Co sequences corresponding to a predicted molecular weight of approximately 31.12 kD (third AUG) and another at 28.5 kD (fourth AUG) were detected in the N gene sequence. Since SDS-PAGE analysis is not necessarily an accurate measurement of molecular size, both starts sites could result in a protein observed at approximately 35--36 kD by SDS-PAGE, and therefore either site could result in the second ORF product. We therefore used two primer sets N1/N1189C and N212/N1189C which spans either the full length of ORF1 or the ORF2 and any down stream putative ORFs of aMPV/C/US/Co, respectively (Fig. [2](#F2){ref-type="fig"}) to amplify both ORFs by RT-PCR. Both ORFs were amplified and cloned into a eukaryotic expression vector. Western blot analysis of the Vero cell expressed N1 and N2 ORFs revealed one reactive band in the pCR3.1-N1ORF transfected Vero cells with the aMPV/N antibody (Fig. [4](#F4){ref-type="fig"}, lane 4) corresponding to the full length nucleoprotein of aMPV, similar to that observed in aMPV-infected Vero cells (Fig. [4](#F4){ref-type="fig"}, lane 3). This protein was not visualized in the pCR3.1-N2ORF transfected Vero cells (Fig. [4](#F4){ref-type="fig"}, lane 5), as was expected since the N212 primer is downstream of the peptide (aMPV/N, amino acids, 10--29) utilized to synthesize aMPV/N peptide antibody. However, when the aMPV/C-N2 (peptide antibody directed to amino acids 383--393 of aMPV/C N protein) was used for western blot analysis, two proteins were reactive in the pCR3.1-N1ORF Vero cells, the first at approximately 43 kD (Fig. [4](#F4){ref-type="fig"}, lane 8), similar to that observed in aMPV-infected Vero cells (Fig. [4](#F4){ref-type="fig"}, Lane 7) and the second, a protein of approximately 35 kD (Fig. [4](#F4){ref-type="fig"}, Lane 8), slightly smaller than the N2 ORF protein in aMPV-infected Vero cells (Fig. [4](#F4){ref-type="fig"}, Lane 7). Western blot analysis of the pCR3.1-N2ORF induced Vero cells demonstrated one reactive band of approximately 35 kD (Fig. [4](#F4){ref-type="fig"}, Lane 9), similar to the smaller reactive band in the pCR3.1-N1ORF transfected Vero cells. The full-length nucleoprotein, as expected was not present in the pCR3.1-N2ORF transfected Vero cells, since the N212 primer is downstream of the first AUG start site (position 14). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Expression of N1 and N2 open reading frames of avian metapneumovirus type C in transfected eukaryotic cells by an expression vector. Lane 1: molecular size markers; Lane 2: uninfected control cells; Lane 3. aMPV/C infected cells reacted with antibodies to peptide N1. Lane 4: Cells transformed with aMPV/C-N gene complete ORF reacted with antibodies to peptide N1. Lane 5: Cells transformed with expression plasmid with truncated N2ORF reacted to antibodies to peptide N1; Lane 6: uninfected control cells; Lane 7: aMPV/C infected cells reacted to antibodies to peptide N4. Lane 8: Cells transformed with aMPV/C-N gene complete ORF reacted with antibodies to peptide N2. Lane 9: Cells transformed with expression plasmid with truncated N2ORF reacted to antibodies to peptide N2. ::: ![](1743-422X-2-31-4) ::: Discussion ========== The utilization of alternative open reading frames for the expansion of genetic information in negative-stranded RNA viruses has been well documented \[[@B10],[@B16],[@B17],[@B21],[@B22]\]. There are, however, various mechanisms for accessing this genetic information. The phosphoprotein of measles virus encodes a single mRNA, which is read in two independently initiated overlapping reading frames \[[@B17]\], while transcripts of influenza virus gene segments 7 and 8 are spliced within the nucleus for production of two different sizes of mRNAs sharing the same 5\'-proximal AUG initial codon \[[@B16]\]. The P gene of Sendai virus is reported to be transcribed into two polycistronic mRNAs, P/C and V/C, which are translated to synthesis the P, C, C\', Y1, and Y2 proteins from independent start sites in two overlapping reading frames \[[@B23]-[@B25]\]. Within the Paramyxoviridae, Newcastle disease virus possesses a polycistronic phosphoprotein (P) gene. Transcriptional modification of the NDV P gene mRNA allows for potential expression of two smaller putative proteins, designated V and W \[[@B12]\], that appears to be a result of polymerase stuttering at the editing site sequences \[[@B13],[@B14]\], leading to the insertion of non-template G nucleotides within the P gene \[[@B12]\]. Consequently, during translation there is a frame shift resulting in production of the V or W protein, dependent on the number of G nucleotides inserted \[[@B12]\]. It was previously suggested that NDV \[[@B26]\] potentially utilized an alternative in-frame AUG start site for expression of an accessory protein similar to the Sendai virus X protein \[[@B21]\] that was recently demonstrated to not be utilized during infection of cells in culture \[[@B27]\]. Pneumonia virus of mice, human and bovine respiratory syncytial viruses, and avian metapneumovirus also possess polycistronic gene(s) \[[@B28]-[@B30]\]. The M2 gene of all the pneumoviruses contains two partially overlapping open reading frames, with the 5\'-proximal open reading frame favored for utilization by the criteria of location and sequence of its start site \[[@B28],[@B29]\]. The P gene of the pneumonia virus of mice is the only known polycistronic phophoprotein gene in the pneumoviruses, and utilizes internal initiation of in-frame AUG initiation codons to generate up to four additional carboxy co-terminal products \[[@B30]\]. In this present study, we demonstrated that the nucleoprotein gene of the avian metapneumovirus subtypes A and C are putatively polycistronic. This may occur by utilization of a second in-frame initiation site (AUG) for the generation of a truncated nucleoprotein present among infected Vero cells. Sequence analysis demonstrated the presence of multiple putative initiation (AUG) start sites along the N gene, however only one alternative start site at nucleotide positions 212 and 410 for APV/C and APV/A, respectively appear to be utilized to transcribe the N2 protein seen in infected cells. The N protein of Pneumoviruses ranges in size from 42--45 kD, based on SDS-PAGE relative mobility, and is highly conserved among metapneumoviruses \[[@B7]\]. The N protein, which protects the RNA genome from ribonucleases, is associated with other viral proteins (P, M2, and L), which together form the transcription complex. The nucleocaspid is the template for transcription and replication; the RNA genome by itself cannot fulfill the role of template. Pneumovirus infection in cells results in the accumulation of the N protein in cytoplasmic inclusion bodies that can be visualized by immunofluorescence \[[@B31]\] or immunohistochemistry \[[@B7]\] as relatively large dots that are usually close to the nucleus of infected cells. Mapping of several paramyxovirus N proteins, including Sendai and measles virus, indicated that the N protein has two major domains; the amino terminal domain appears to be required for nucleoprotein formation, containing the domains necessary for RNA binding and N-N interactions; while the carboxy-domain interacts with the phosphoprotein (P), particularly when it is part of the polymerase complex \[[@B32],[@B33]\]. In bovine respiratory syncytial virus (bRSV), removal of the C-terminal 32 amino acids of the N protein inhibits the interactions with the P protein, whereas the removal of 32 amino acids from the N-terminus has a minimal effect \[[@B32]\]. However, almost all of the N from amino acids 2--391 is required to support bRSV minigenome RNA synthesis \[[@B34]\]. The truncated N2 protein encompasses 328 amino acids (250 for aMPV/Type A) of the carboxy terminus of the full-length N protein, suggesting that N2 may not be involved in the polymerase complex. However the domains responsible for RNA binding of N-N and N-P binding remain intact, suggesting that N2 may play an alternative role in cells during viral infection. Methods ======= Cells and viruses ----------------- Vero cells were maintained as monolayer cultures in minimal essential media (MEM) supplemented to contain 8 % fetal bovine serum with 100 units/ml penicillin G, 0.025 μg/ml amphotericin B, and 100 units/ml streptomycin. The aMPV/C/US/Co and aMPV/A/UK/3b isolates were obtained from the National Veterinary Services Laboratory (NVSL, APHIS, USDA, Ames, Iowa). Viruses were propagated on 95% confluent Vero cell monolayers in MEM supplemented to contain 2% FBS and antibiotics as described previously \[[@B3]\]. Cells were infected at multiplicity of infection of 10 (moi = 10), and virus was adsorbed for 1 hour at 37°C. Media was added and cells were incubated at 37°C, 5% CO~2~for 72 hours or until 90% cytopathic effect was observed by light microscopy. Cells were scraped and harvested by centrifugation at 8000 × g. Computer analyses, peptide synthesis and antibody production ------------------------------------------------------------ The nucleoprotein (N) gene sequences of aMPV serotypes A, B, and C (Genbank accession numbers: AAC55065, AAG42499, and AAF05909) were analyzed in the GeneWorks (Intelligentics, Mountain View, CA) and Mac Vector (Accelrys, San Diego, CA) computer analysis programs to determine hydrophilicity, antigenicity, and identity of the deduced amino acid sequences. The sequences were aligned for maximum similarity, and a consensus sequence was determined using the most prevalent amino acid for each residue. Five peptides with sequences: 1) aMPV/N: DLSYKHAILKESQYTIKRDV; 2) aMPV/C-N2: DKEARKTMASATKDNSGPIPQ; 3) aMPV/A-N3: ERTTREAMGAMVREKVQLTK; 4) aMPV/C-N4: LNINEEGQNDY; and 5) aMPV/A-N5: LGGDDERSSKF were chosen based on antigenicity and hydrophilicity to be utilized for generation of aMPV peptide-based antibodies. Peptides were synthesized by Research Genetics (Huntsville, AL) according to the manufacturer\'s protocol. Briefly, rabbit aMPV/N peptide antibodies were produced by Research Genetics (Huntsville, AL) according to manufacturer\'s protocol. Two rabbits were injected with 0.1 mg of KLH-conjugated peptide emulsified with Freud\'s complete adjuvant and injected into four subcutaneous (SQ) sites on day 1. On days 14, 42, and 56 rabbits were injected again (boosters) with 0.1 mg of KLH-conjugated peptide emulsified with Freud\'s complete adjuvant \[[@B35]\]. Sera were collected at days 0, 28, 56 and 70. Rabbit pre-immune sera were used as negative controls for rabbit assays. SDS-PAGE and Western blot assay ------------------------------- Protein concentration of the supernatant fraction from infected cells was measured for protein concentration by Bradford\'s reagent (Bioworld, Dublin, OH) at 595 nm. Infected supernatants were denatured in Laemmli\'s sample buffer (BioRad, Hercules, CA) and boiled for 5 min. Denatured polypeptides (6 μg protein/lane) were separated in a sodium dodecyl sulfate 4--20% polyacrylamide Criterion (Biorad, Hercules, CA) gel gradient by electrophoresis (SDS-PAGE) at 120 V for 2 hours \[[@B36]\]. Polypeptides were transferred to nitrocellulose by applying a constant voltage of 10 V for 1 hour on a Biorad (Hercules, CA) Trans-Blot SD Semi-Dry Transfer cell \[[@B37]\]. Blots were blocked with BLOTTO (20% dry milk in PBS) overnight at 4°C or for 1 hour at 37° and washed 3 X in phosphate buffered saline (PBS). Affinity purified rabbit anti-peptide antibody (diluted 1:100) was used as the source of the primary antibodies and incubated for 1 hour at 37°C followed by 3 washes in PBS. Secondary antibody (α-rabbit IgG-alkaline phosphatase, Sigma, The Woodlands, TX) was added (1:500), incubated 1 hour at 37°C, washed 3 X in PBS and developed using a alkaline phosphatase substrate kit (Vector, Burlingame, CA). Viral RNA Isolation accompanied by RT-PCR Amplification of aMPV/C/US/Co N1 and N2 ORF nucleotide sequences ---------------------------------------------------------------------------------------------------------- Total RNA was isolated \[[@B38]\] from aMPV/C/US/Co-infected Vero cell lysates using Qiagen\'s \"RNeasy\" kit (Qiagne, Valencia, CA) according to the manufacturer\'s protocol. RNA was analyzed for purity by agarose gel electrophoresis in a 1.5% agarose gel, at 125 volts, and stained with 10 μg/ml of ethidium bromide (Sigma, The Woodlands, TX). The aMPV N1 and N2 ORFs were reverse transcribed using either the N1 (5\'-GAAATGTCTCTTCAGGGGATTCAG-3\') and N1185C (5\'-AATCATTCTGGCCTTCCTCAT-3\') primer pair or the N212 (5\'-ATGCAGTACGTGAGCACC-3\') and N1185C (5\'-AATCATTCTGGCCTTCCTCAT-3\') primer pair, followed by 30 cycles of PCR \[[@B39]\]. RT-PCR amplification products were analyzed by agarose gel electrophoresis and the full length N1 ORF product and the N2 ORF product were excised and purified before cloning into the expression vector pCR3.1-Topo (Invitrogen, Carlsbad, CA). Molecular cloning, nucleotide sequencing, and eukaryotic expression of pCR3.1-N1ORF and pCR3.1-N2ORF ---------------------------------------------------------------------------------------------------- The N1 ORF and N2 ORF fragments of aMPV/C/US/Co were cloned into the eukaryotic expression vector pCR3.1-Topo (Invitrogen, Carlsbad, CA) according to the manufacturer\'s protocol. Plasmid DNA was isolated using Qiagen\'s miniprep kit (Qiagen, Valencia, CA). Double stranded sequencing with Taqpolymerase (Applied Biosystems Inc.) and fluorescent labeled dideoxynucleotides was performed with an automated sequencer \[[@B40]\] on both amplification products to verify identity and insure that no changes in the ORFs had been made relative to the original N gene. The pCR3.1-N1ORF and pCR3.1-N2ORF vectors were transfected into Vero cells using lipofectamine (Invitrogen, Carlsbad, CA). Protein was induced with IPTG (Sigma, The Woodlands, TX) at 24 hours post-transfected and total proteins were harvested by scraping. An aliquot of uninduced and induced cells were lysed in 2 X Laemmli\'s buffer, boiled for 5 minutes and separated by SDS-PAGE on a 4--20% Criterion (Biorad, Hercules, CA) gradient gel, followed by electroblotting onto nitrocellulose as previously described. Competing Interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Dr. Alvarez was a post-doctoral associate and conducted the primary experimentation following design of peptides and production of anti-sera under the direction of Dr. Seal. Dr. Alvarez initiated writing of the draft manuscript with subsequent editing and revisions by both authors. Acknowledgements ================ This research was supported by ARS, USDA CRIS project No. 6612-32000-015-00D-085 and U.S. Poultry & Egg Association grant no. 404 to BSS which supported synthesis of peptides and immunization for antibodies commercially.	PubMed Central	2024-06-05T03:55:56.243015	2005-4-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087890/", "journal": "Virol J. 2005 Apr 12; 2:31", "authors": [ { "first": "Rene", "last": "Alvarez" }, { "first": "Bruce S", "last": "Seal" } ] }
PMC1087891	Background ========== Dengue infection (DI) is amongst the most important emerging viral diseases transmitted by mosquitoes to humans, in terms of both illness and death \[[@B1]\]. The worldwide large-scale reappearance of dengue for the past few decades has turned this disease into a serious public health problem, especially in the tropical and subtropical countries \[[@B2]-[@B4]\]. It is estimated that 52% of the global population are at the risk of contracting Dengue fever (DF) or dengue hemorrhagic fever (DHF) lives in the South East Asian Region. Although all the four serotypes have been circulating in this region, ecological and climatic factors are reported to influence the seasonal prevalence of the dengue vector, Aedes aegypti, on the basis of which countries in this region are divided in to four zones with different DF/DHF transmission potential \[[@B5]\]. In most of the countries, dengue epidemics are reported to occur, during the warm, humid and rainy seasons, which favor abundant mosquito growth and shorten the extrinsic incubation period as well \[[@B6]-[@B9]\]. DF has been known to be endemic in India for over two centuries as a benign and self-limited disease. In recent years, the disease has changed its course manifesting in the severe form as DHF, with increasing frequencies \[[@B10]\] Delhi City (India) is home to more than 13 million people and is endemic for DI \[[@B11]\]. Overpopulation has consequently led to poor sanitary conditions and water logging at various places. A major epidemic of DHF from Delhi was last reported in the year 1996 after which DI became a notifiable disease and a number of policies were formulated to bring the DI as well as its vector under control. The retrospective studies, one conducted by us during the period, 1997--2001 and another by National Institute of Communicable Diseases (NICD), New Delhi during the year 1997, have observed a decline in the number of cases having either DF or DHF in the following years \[[@B12],[@B13]\]. Although, the vector mainly responsible for the spread of DI is present all the year around in Delhi, studies on the relative prevalence and distribution have shown the highest A. aegyptilarval indices during the monsoon and post monsoon period \[[@B13]-[@B15]\]. In the year 2003, India had experienced one of the wettest monsoons in 25 years, which led to a spate of mosquito growth creating an alarming situation of mosquito borne diseases in many states. Delhi experienced an outbreak of DF this year, after 6 years of silence. Studies conducted in the countries like Brazil, Indonesia and Venezuela, where DI is present either in epidemic or endemic form have suggested a correlation between weather and pattern of DI. Rain, temperature and relative humidity are suggested as important factors attributing towards the growth and dispersion of this vector and potential of dengue outbreaks \[[@B2]-[@B4]\]. Since limited data is available on the association of climatic conditions and the pattern of DI from this geographical region, this study was conducted to find out the relationship of dengue infection with climatic factors such as the rainfall, temperature and relative humidity during the dengue outbreak in the year 2003. Results ======= Seropositivity -------------- All blood smears microscopically screened for malarial parasite were found to be negative. Analytical interpretations presented in this study were based upon instructions mentioned in the Pan Bio Rapid Strip Test procedure manual. During the outbreak period, blood samples were collected from 1550 patients experiencing a febrile illness clinically consistent with DI over the period of one year from January to December 2003. Eight hundred ninety three cases (57.36%) were confirmed as serologically positive, out of which 199 (22.28%) cases were positive for dengue-specific IgM antibodies indicating primary infection and 381 (42. 67%) cases were positive for both dengue-specific IgM and IgG antibodies indicating secondary infection (Figure [1](#F1){ref-type="fig"}). IgG antibodies alone were also detected in 313 (35.05%) cases and these cases were presumed to be either suspected secondary dengue infection as IgG positivity alone could also be due to cross reactivity with other flaviviruses. The difference between numbers of serologically positive cases reported during different months was significant (p \< 0.05). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Month wise distribution of primary and secondary serologically positive cases during the outbreak period in the year 2003. ::: ![](1743-422X-2-32-1) ::: DI is observed to be a seasonal disease in Delhi. According to intensity of rainfall, weather data was divided in three periods, namely; pre monsoon period: from February- May, monsoon period: from June -- September and post monsoon period: from October -- January. Few cases clinically suspected of dengue infection in the pre monsoon period were later found to be serologically negative for dengue-specific antibodies. During the monsoon period, only 3 cases (0.34%) were confirmed serologically positive in the month of August, and 68 cases (7.6%) in the September. Dengue-specific antibody positive cases were mainly reported during the post monsoon period with maximum number of cases 583 (65.3%) cases reported during the month of October followed by 230 (25.76%) cases in the November (Table [1](#T1){ref-type="table"}). The difference between numbers of serologically positive cases as compared to serologically negative ones in post monsoon period was significantly higher (p \< 0.001), than during the remaining period with 92% of total annual cases reported during this period. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Month wise distribution of clinically diagnosed and serologically positive cases amongst primary and secondary cases during the DF outbreak, 2003 ::: Month Total Suspected cases Serologically Positive cases (%) Primary infection (IgM Positivity) Secondary infection (IgM+ IgG Positivity) Suspected secondary infection (IgG Positivity) ----------- ----------------------- ---------------------------------- ------------------------------------ ------------------------------------------- ------------------------------------------------ August 12 3 (0.34%) 1 (0.5%) 1 (0.26%) 1 (0.32%) September 157 68 (7.6%) 17 (8.6%) 24 (6.3%) 27 (8.6%) October 982 583 (65.3%) 126 (63.3%) 246 (64.57%) 211 (67.4%) November 362 230 (25.76%) 49 (24.6%) 110 (28.87%) 71 (22.68%) December 37 9 (1%) 6 (3%) 0 (0%) 3 (1%) Total 1550 893 (57.36%) 199 (22.28%) 381 (42.67%) 313 (35.05%) ::: Distribution by age ------------------- Out of 893 serologically positive cases, 687 cases belonged to the adult\'s age group (\> 12 years) and 206 cases to pediatric age group (≤ 12 years) in this study. Larger proportions of serologically positive cases were observed among adults, with a positive prevalence of 56.4% among children and 58% among adults, distribution was however, not significantly different when compared with pediatric age group (p \> 0.05). The difference between numbers of serologically positive cases among adult and pediatric group in post monsoon period as compared to the rest of the season was also not significant (p \> 0.05) (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Month wise distribution of serologically positive cases amongst children and adults during the DF outbreak, 2003 ::: Month Dengue-specific Antibody Positive cases ----------- ----------------------------------------- ------------------------- ----------------------- Total Children (Positivity %) Adults (Positivity %) August 3 0 3 (25%) September 68 18 (48.6%) 50 (41.7%) October 583 133 (69.6%) 450 (57%) November 230 54 (44.3%) 176 (83.8%) December 9 1 (11.1%) 8 (28.6%) Total 893 206 (56.4%) 687 (58%) ::: Climatic influence ------------------ Fig. [2a](#F2){ref-type="fig"} indicates that outbreak coincided mainly with the post monsoon period of subnormal rainfall (Cumulative rainfall = 30.3 mm) from October to December 2003 and was followed by relatively heavy rainfall during the monsoon period; from June to September 2003. The difference in the rainfall and temperature between three seasonal periods was found to be significant (p \< 0.05) (Fig. [2a](#F2){ref-type="fig"} &[2b](#F2){ref-type="fig"}). Mean ambient temperature was 25.4°C during the pre monsoon period, which increased to 30.9°C during the monsoon period; the period preceding the outbreak and decreased to 20.3°C (Mean temperature from October to December) in the actual outbreak months during the post monsoon period. The difference between relative humidity during the three periods was not significant. The mean relative humidity was 71.2% during the pre monsoon period. It increased during the monsoon period to 85% and increased further during the post monsoon period to 90% (Fig. [2c](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### a: Month wise distribution of serologically positive cases of dengue fever /dengue hemorrhagic fever and rainfall in Delhi for the year 2003 b:Month wise distribution of serologically positive cases of dengue fever /dengue hemorrhagic fever and temperature in Delhi for the year 2003 c: Month wise distribution of serologically positive cases of dengue fever /dengue hemorrhagic fever and relative humidity in Delhi for the year 2003 ::: ![](1743-422X-2-32-2) ::: Discussion ========== In the year 2003, India had experienced one of the wettest monsoons in 25 years, which led to a spate of mosquito growth creating an alarming situation of mosquito borne diseases in Delhi and many other states \[[@B16]\]. As a consequence to this unusually heavy rain, an outbreak of dengue fever was once again reported from Delhi after a silence of six long years. Most of vector borne diseases exhibit a distinctive seasonal pattern and climatic factors such as rainfall, temperature and other weather variables affect in many ways both the vector and the pathogen they transmit \[[@B17]\]. Worldwide studies have proposed that ecological and climatic factors influence the seasonal prevalence of both the A. aegyptiand dengue virus \[[@B2]-[@B4]\]. The vector mainly responsible for the spread of DI is present at the basal level all the year around in Delhi, however, studies on the relative prevalence and distribution have shown the highest A. aegyptilarval indices during the monsoon and post monsoon period \[[@B13]-[@B15]\]. Since limited data is available on the affect of climatic factors on the pattern of DI, this study was planned to carry out the month wise detailed analysis of three important climatic factors such as rainfall, temperature and relative humidity on the pattern of DI. Observations on the seasonality were based on a single year\'s data as the intensity of sampling was at its maximum during this outbreak period. The outbreak coincided mainly with the post monsoon period of subnormal rainfall, which was followed, by relatively heavy rainfall during the monsoon period; from June to September 2003. The difference in the total rainfall and temperature during three seasonal periods was found to be statistically significant (p \< 0.05). Monthly weather data showed that temperature variations were more amongst different months during the pre monsoon and post monsoon period as compared to the monsoon period. Even though, the monsoon season began in mid- June, there was no respite from the heat as there was not much difference in the temperature during the last month of pre monsoon; May and beginning of monsoon in the June. Unusual heavy rainfall subsequently led to decrease in temperature during the later part of monsoon period. The temperature showed a decline and remained almost constant during the months of July and August (30.2°C), continuous heavy rainfall subsequently led to further decrease in the temperature during the month of September to 29°C. Relative humidity increased during the rainy season and remained high for several weeks. An in-depth analysis of these three factors thus led to a proposal that optimum temperature with high relative humidity and abundant stocks of fresh water reservoirs generated due to rain, developed optimum conditions conducive for mass breeding and propagation of vector and transmission of the virus. Our study was in tune with a previous study by NICD of seasonal variations and breeding pattern of A. aegyptiin Delhi, which showed that there are two types of breeding foci, namely; primary and secondary breeding foci. Primary breeding foci served as mother foci during the pre monsoon period. A. aegyptilarvae spread to secondary foci like discarded tyres, desert coolers etc., which collect fresh water during the monsoon period \[[@B14]\]. This study supported the proposal that all the three climatic factors studied could be playing an important role in creating the conducive condition required for breeding and propagation of this vector, the basal level of which is present all round the year. This prospective study therefore highlighted the major important factors, which could alone or collectively be responsible for an outbreak. In our study, the largest proportion of serologically positive cases was recorded in the post monsoon period, which is in agreement with our previous study \[[@B12]\]. Our findings were in coordination with study by other groups from this geographical region \[[@B13]-[@B15]\]. The seasonal occurrence of positive cases has shown that post monsoon period is the most affected period in Bangladesh as well \[[@B18]\]. However, a retrospective study from Myanmar during 1996--2001 reported the maximum cases of dengue during the monsoon period \[[@B19]\]. Study by group of Rebelo from Brazil has also emphasized the importance of season. They have observed that dengue cases were higher during rainy season showing the importance of rain in forming prime breeding sites for A. aegyptithus spread of DI \[[@B20]\]. Study of eco-epidemiological factors by Barrera et al \[[@B21]\] showed that DF has a positive correlation with the relative humidity and negative relation with evaporation rate. Peaks of dengue cases were observed to be near concurrent with rain peaks in this study from Venezuela showing a significant correlation of intensity of DI with the amount of rain \[[@B21]\]. In this study we have observed that temperature tends to decrease towards the end of monsoon period, specially remains moreover constant during the later months of rainy season. India and Bangladesh fall in the deciduous, dry and wet climatic zone. The temperature remains high during the pre monsoon period. It is continuous rain pour for a couple of days that brings down the temperature during the monsoon period, which may also be responsible for an increase in the relative humidity and decrease in the evaporation rate thus maintaining secondary reservoirs containing rain water. More studies are needed to establish the relationship between the climatic changes and dengue outbreaks, which would help in formulating the strategies and plans to forecast any outbreak in future, well in advance. Very little dengue is found in adults in Thailand, presumably because people acquire complete protective immunity after multiple DI as children \[[@B1]\], as DI is highly endemic in Thailand \[[@B22]\]. On the other hand, DI especially DHF is an emerging disease in India; probably this may be the reason that people of all the age are found to be sensitive to infection in our study. Even though more adults were reported of having anti dengue antibodies, the difference in the number of positive cases was not significant as compared to pediatric age group. The severity of this outbreak was lesser as compared to the DHF epidemic that occurred in year 1996 caused by the serotype Den-2 \[[@B23]\]. Serotype Den-2 is reported to be the one mainly associated with DHF, the more severe form of the disease \[[@B24],[@B25]\]. More studies in this regard can further elucidate correlation of serotypes with severity of disease from this geographical region. Conclusion ========== This prospective study highlighted rain, temperature and relative humidity as the major and important climatic factors, which could alone or collectively be responsible for an outbreak. More studies in this regard could further reveal the correlation between the climatic changes and dengue outbreaks, which would help in making the strategies and plans to forecast any outbreak in future well in advance. Materials and methods ===================== Study design, population and sample size ---------------------------------------- The present study was conducted retrospectively for a period of one year during the recent outbreak of dengue fever in Delhi in the year 2003. The study population comprised individuals of all age groups, attending the outpatient and inpatient departments of Lok Nayak Hospital, a tertiary care hospital in Delhi. Blood samples were collected from 1550 patients experiencing a febrile illness clinically consistent with dengue infection, selected according to the following inclusion and exclusion criteria. Case-inclusion criteria ----------------------- A case was included if there was high fever with clinical symptoms suggestive of dengue infection as per WHO criteria \[[@B26]\]. Case-exclusion criteria ----------------------- A case was excluded, if routine laboratory testing suggested bacterial or any viral infection other than dengue infection or any other disease \[[@B26]\]. Microscopy for malaria identification ------------------------------------- Venous blood was used for blood slide preparation for malaria parasite examination. Thick and thin blood films were prepared on the same slide, stained with Giemsa and examined for the presence of malaria parasite. Laboratory confirmation of dengue infection by serology ------------------------------------------------------- Dengue Duo IgM and IgG Rapid Strip test (Pan Bio, Australia) was used for the detection of dengue-specific antibodies. 1 μl of serum was mixed with 75 μl of buffer (supplied in the kit) and test strip was dipped in to the diluted serum. Results of the test were read after 30 minutes. Serum antibodies of the IgM or IgG class, when present bind to anti-human IgM or IgG immobilized in two lines across the test strip. Colloidal gold-labeled anti-dengue monoclonal antibodies form complexes with the dengue antigen that is captured by dengue specific IgM or IgG in the patient\'s serum. These complexes were visualized as pink/purple line(s). The presence of anti-dengue IgM antibodies alone indicated primary infection. In contrast, presence of anti-dengue IgG antibodies with or without IgM indicated secondary infection. (IgG antibodies alone was considered as suspected secondary infection as it could also be due to cross reactivity with other flaviviruses). Analysis of metrological data ----------------------------- Monthly details of total rainfall, temperature and relative humidity for all the months of the year, 2003 were obtained from Meteorological Department of Delhi, Mausum Bhawan, New Delhi and retrospectively analyzed in relation to total number of dengue cases. According to the intensity of the rainfall, weather data was divided in three periods namely; pre-monsoon period: from February- May, monsoon period: from June -- September and post monsoon period: from October -- January. Competing Interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= It is stated that both the authors 1) have made substantial contributions to conception and design, or acquisition of data, or analysis and interpretation of data; 2) have been involved in drafting the article or revising it critically for important intellectual content; and 3) have given final approval of the version to be published. Acknowledgements ================ We thank the Metrological Department of Delhi, Mausum Bhawan, India for providing the monthly weather details of rainfall, temperature and relative humidity for the year 2003.	PubMed Central	2024-06-05T03:55:56.245002	2005-4-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087891/", "journal": "Virol J. 2005 Apr 14; 2:32", "authors": [ { "first": "Anita", "last": "Chakravarti" }, { "first": "Rajni", "last": "Kumaria" } ] }
PMC1087892	Background ========== Plant pararetroviruses are classified as members of the family Caulimoviridaewhich comprises 6 genera \[[@B1]\]. Like retroviruses, members of this group of viruses use reverse transcriptase for replication of the genome \[[@B2],[@B3]\]. However, they differ in two major points: retroviruses have an RNA genome whereas pararetroviruses have a DNA genome; and, a proviral form of retroviruses is integrated into the host chromosome whereas the DNA of pararetroviruses accumulates within the nucleus as multiple copies of a circular chromosome \[[@B4]\]. Retroviruses and pararetroviruses show similarities in their molecular organization and replication process and are phylogenetically related. These groups of viruses direct the production of a terminally redundant RNA which is used as a replicative intermediate and as mRNA. Many of the genes encoded by pararetroviruses are homologous in sequence and/or analogous in function to those of retroviruses. The genome of all replication-competent retroviruses consists of three major genetic elements that are arranged in the order gag-pol-env (structural-replication-envelope proteins). Each protein is produced as a result of frameshifting during translation, or suppression of stop codons in the polyproteins. Products of the gag ORF represent the structural components of the viral matrix, i.e. capsid and nucleocapsidproteins; the pol domains generally comprise the protease, reverse transcriptase, ribonuclease and an endonuclease/integrase. Pararetroviruses encode the gag-pol core, but lack an integrase, as viral DNA integration into the host chromosome is not required \[[@B5]\]. Retrovirus and pararetrovirus polyproteins are believed to be processed by their own aspartate proteases \[[@B3],[@B6],[@B7]\]. These proteases contain several conserved regions when compared with each other and share consensus sequences in the active site; however, they show no homology with other viral proteases. Protein cleavages by these and other proteases are dependant on amino acid sequence and conformation near the cleavage site \[[@B6]\]. Rice tungro is a major rice disease in southeast asia and is caused by a double infection by two viruses \[[@B8],[@B9]\]: Rice tungro bacilliform virus (RTBV) \[[@B10],[@B11]\], a member of the genus Badnavirusin the family Caulimoviridae\[[@B1]\], and Rice tungro spherical virus (RTSV), a single-stranded RNA virus and member of the genus Waikivirusin the family Sequiviridae\[[@B12],[@B13]\]. RTBV is responsible for symptoms of the disease \[[@B8]\] and RTSV is required for the transmission of the two viruses by the leafhopper vector Nephotettix virescens\[[@B14],[@B15]\]. RTBV genome is a double stranded DNA of 8.0 kbp with two site-specific discontinuities resulting from replication by reverse transcriptase \[[@B3],[@B16]\]. The RTBV genome has four open reading frames (ORF) \[[@B17]\]. ORF3 has similarities with the gag-pol core of retroviruses. The polyprotein (P3) contains a putative movement protein (MP), the capsid protein (CP), the aspartate protease (PR), and the reverse transcriptase (RT) with a ribonuclease H activity. The N- and C- terminal amino acid (aa) sequences of the CP were deduced from MALDI-TOF mass spectral analysis \[[@B18]\]. The location of the CP domain, which is encoded by aa 477--791 in P3, was confirmed by immunodetection reactions using antibodies raised against different segments of ORF3 \[[@B18]\]. P3 sequence aa 985--995 shows homologies to the active site of aspartate proteases encoded by retroviruses and pararetroviruses, with the sequence DSGS believed to be the RTBV protease active site. Changing of D to A (aspartic acid to an alanine) in this sequence affects the proteolytic processing of the RT \[[@B19]\]. Detection of RTBV gene products in vivorevealed a major protein of 13.5 kDa in virus preparations, by using an antibody raised against the domain aa 881--1098 \[[@B10],[@B20]\]. The antibody was used in immunogold labelling reactions and showed binding on the surface of the virus particle, suggesting that the PR is located at or near the surface of the virus particle \[[@B20]\]. In this paper, we used mass spectrometry analysis to characterize the RTBV protease domain in a protein associated with purified virions, and confirm the hypothesis by in silicomodelling and in vitroexpression studies. Analysis of products of in vitroprocessing of P3 suggests that RTBV aspartate protease is involved in the release of the CP from the polyprotein P3. Results ======= Determination of molecular mass ------------------------------- In-line HPLC electrospray ionization was used to determine the molecular mass of proteins that were isolated in preparations of RTBV. Several peaks were observed using this method, none of which correspond to the known molecular mass of the CP subunits \[[@B18]\]. Many of the primary peaks were accompanied by low amounts of proteins of similar but not identical charge: we presumed that these peaks represented various charge statesof the peptide/protein. Reconstruction of the data, after algorithms were applied to convert the family of ion peaks, gave a protein with molecular mass of 13,794 ± 4 Da (Figure [1](#F1){ref-type="fig"}). The lack of recovery or detection of proteins of the size of CP subunits is likely the result of retention on the HPLC column due to the very basic charge of the protein; estimated isoelectric point (pI) of CP is 9.43. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Mass spectrometry analysis performed on RTBV virions. In-line HPLC electrospray ionization mass spectrometry analysis performed on RTBV virions. Virus sample was denatured with guanidium hydrochloride 4.8 M prior, to injection onto the column. Analysis of peptide which coeluted from the in-line HPLC column in various charge states gave a molecular mass of 13,794 ± 4 Da. ::: ![](1743-422X-2-33-1) ::: Localization of the protease domain ----------------------------------- By comparing the known sequence of P3 protein in RTBV with aspartate proteases encoded by other retroviruses we predicted that five possible proteins could be derived with a molecular mass of 13,794 ± 4 Da that could contain the putative active site domain DSGS (at aa 987--990) and the IIG sequence. The IIG sequence motif is conserved amongst retroviral proteases and is located at aa 1063--1065 in P3. The five predicted proteins (Figure [2](#F2){ref-type="fig"}) were covering amino acids 965--1085, 967--1086, 971--1091, 982--1102, and 984--1104, with predicted molecular masses of 13,790.71 Da, 13,790.71 Da, 13,796.7 Da, 13,791.7 Da and 13,795.7 Da, respectively. These predictions of the five domains are based on the combination of molecular mass and the presence of conserved motifs, and not on putative protease cleavage sites that flank the five predicted proteins. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Putative protease domains for RTBV. Position of five peptides that include the active protease domains with molecular mass that correspond to the mass spectrometry analysis. Peptide A has a predicted molecular mass of 13,790.71 kDa; peptide B of 13,790.71 kDa; peptide C of 13,796.70 kDa; peptide D of 13,793.60 kDa; peptide E of 13,795.70 kDa. Underlined sequences represent the active site of the protease. The grey box indicates a conserved motif among retroviral proteases. Numbers above arrows indicate position of amino acids in P3. ::: ![](1743-422X-2-33-2) ::: Structural model of the protease -------------------------------- The overall sequence similarity between aspartate proteases is quite low (below 25%) based on pairwise scoring criteria; therefore we applied fold recognition methods to detect remote homologues. The sequence alignment for the proteins selected for analysis showed strong conservation in the active site of the proteins (Figure [3](#F3){ref-type="fig"}). The pairwise alignment between the Rous sarcoma virus (RSV) template and the RTBV sequence was submitted to comparative modelling. As represented in the model (Figure [4](#F4){ref-type="fig"}), there is a strong resemblance between the structures predicted for RTBV and RSV proteases. The lack of identity in the predicted structures may be due to inherent inaccuracy of the modelling programs, due to inherent differences in the protein and substrate, and other characteristics. Taking these differences into account, as deduced by a folding assignment system \[[@B21]\] (where reliable scores were obtained) the protease sequence had a predicted folding structure that was highly similar to several other aspartate proteases. The overall fold surrounding the active site was well conserved between the putative RTBV protease and RSV protease used as template (Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Structural sequence alignment of the RTBV protease with other retroviral proteases. Sequence alignment of the RTBV protease amino acid sequence with proteases of Rous sarcoma virus (RSV), Equine infectious anemia virus (EIAV) and Human immunodeficiency virus (HIV). The color scheme corresponds to percentage of similarity (based on physico-chemical properties). Black background and white foreground indicate 100%, grey background and white foreground indicate 80%, grey background and black foreground indicate 60%. Lower similarity values are not shown. Numbers over the alignment indicate the alignment length. Secondary structure elements from the RSV sequence are represented over the alignment. The numbering of the elements follows the RSV numbering based on structure. Boxes indicate beta strand elements assigned as β. The helix is represented as a cylinder and indicated as α. Thick lines connecting the elements are loops and dashed lines indicate a break in the sequence. The black triangle indicates the location of the active site. ::: ![](1743-422X-2-33-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Structural modelling of the RTBV protease. Structural modelling of the RTBV protease (A), and Rous sarcoma virus (RSV) protease used as template (B). The sphere indicates the N-terminal end, aspartic acid of active site is shown in the stick model. In red is the RSV protease inhibitor 39 coupled to the active site. The first residues of RTBV PR could not be modelled. Conservation of the active site and overall fold recognition analyses with modelling building show that the RTBV sequence resembles greatly a protease fold. ::: ![](1743-422X-2-33-4) ::: Induction of putative protease domain ------------------------------------- Specific primers (Ab-PR-F and Ab-PR-R) were designed to amplify the putative protease domain deduced from the mass spectrometry analysis (Figure [2-A](#F2){ref-type="fig"}). DNA used for PCR amplification included the full-length RTBV clone pBSR63A, and plasmid pBS-mp/PR; the reactions led to isolation of peptide PR and mPR, respectively (Table [1](#T1){ref-type="table"} and Figure [5](#F5){ref-type="fig"}). PR and mPR differ from each other only at amino acid 987, in which the residue was changed from Aspartic acid (D) to Alanine (A). Peptides were expressed in E. coliand, following induction of gene expression the extracts were subjected to SDS-PAGE, and gels were stained with coomassie blue (Figure [6-A](#F6){ref-type="fig"}). The expected size of the peptides was about 14 kDa. We did not detect a protein in cultures that contain pTr-PR. However, in extracts of cultures containing pTr-mPR, the protein mPR was easily detected at the expected position on the gel. Western immunoblot assays using antibodies raised against mPR (Ab-PR) confirmed that the 14 kDa peptide was mPR (Figure [6-C](#F6){ref-type="fig"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Methodology for creating the constructs used in the analysis ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Cloning into pBlueScript KS Clone pBS-CP/PRwas obtained by using primers CP-PR-F and CP-PR-R on full-length RTBV clone pBSR63A and cloning the PCR fragment into plasmid pBlueScript KS. Clone pBS-PR1was obtained by digesting pBS-CP/PR with XbaI and HindIII and cloning the 0.8 kbp fragment into plasmid pKS. Clone pBS-mPR1was obtained by digesting pBS-mpr/RT 19 with PstI and EcoRV and cloning the 0.7 kbp fragment into pBS-PR1 digested with PstI and EcoRV. Clone pBS-CP/mPRwas obtained by digesting pBS-mPR1 with PstI and EcoRV and cloning the 0.7 kbp fragment into pBS-CP/PR digested with PstI and EcoRV. Cloning into pTrHis. Clone pTr-CP/PRwas obtained by digesting pBS-CP/PR with BamHI and HindIII and cloning the 2.5 kbp fragment into pTrHisA digested with BamHI and HindIII. Clone pTr-PRwas obtained was obtained by using primers Ab-PR-F and Ab-PR-Ron full-length RTBV clone pBSR63A and cloning the PCR fragment into pTrHis A digested with NcoI and HindIII. Clone pTr-mPRwas obtained by using primers Ab-PR-F and Ab-PR-R on plasmid pBS-mpr/RT and cloning the PCR fragment into pTrHis A digested with NcoI and HindIII. Cloning into pET. Clone pET-MPwas obtained by using primers Et-MP-F and Et-MP-R on full-length RTBV clone pBSR63A and cloning the PCR fragment into pET-28a digested with NdeI and BamHI. Clone pET-MP/PRwas obtained by digesting pBS-CP/PR with ScaI and HindIII and cloning the 2.3 kb fragment into pET-MP digested with ScaI and HindIII. Clone pET-MP/mPRwas obtained by digesting pBS-CP/mPR with SacI and HindIII and cloning the 2.3 fragment into pET-MP. Clone pET-P3was obtained by different steps. First, PstI digested product (2.1 kbp) from pBS-PR/RT \[19\] containing PR and RT was cloned at the PstI site of pTr-CP/PR. The resulting plasmid was digested by BamHI, and the 3.9 kb fragment was clone into pTrHis A, to obtain pTr-CP-RT. The later was digested with ScaI and BamHI and the 3.8 fragment was cloned into pET-MP digested with ScaI and BamHI to obtain pET-P3. Clone pET-mP3was obtained using the same strategy than above, by using pBS-mpr/RT instead of pBS-PR/RT. Lists of primers CP-PR-F : 5\'-GAAAGAGGGATCCAAAATGGCAATAGTAGAAG-3\' CP-PR-R : 5\'-GTTTTTCAAAAGCTTCTTAATCTGCTGGCGTG-3\' Ab-PR-F: 5\'-CATGCCATGGCACATCATCATCATCATCATCATGCAGGATGTTATGTA-3\' Ab-PR-R : 5\'-TATTCCCGAAGCTTTTTATATAGTTATATAATC-3\' Et-MP-F : 5\'-GTAAGTGCCCATATGAGCCTTAGACCATTTACTGG-3\' Et-MP-R : 5\'-AGGGCTGTGGGATCCTCATTCAGGTCTATCACCTC-3\' ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Polyprotein P3 peptide domains cloned in different constructs. Visualization of P3 peptide domains cloned in different constructs. Parts in grey are sequence derived from the full-length RTBV clone pBSR63A 11. Parts in black are sequence imported from plasmid pBS-mp/RT 19, containing the protease mutated active site. Parts in white are sequences from vectors. Underlined restriction enzymes are sites that are present in ORF3. ::: ![](1743-422X-2-33-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Induction of the putative protease domain. Expression of peptides in E.coli. Numbers on the left are estimated sizes in kDa of the molecular weight marker. (A) Coomassie blue-stained gel of induced peptides in E.coli. Lane 1: pTr-PR; Lane 2: pTr-mPR. (B) Western blot performed on induced peptides using antibodies raised against RTBV (Ab-RTBV). Lane 3: pTr-PR; Lane 4: pTr-mPR. (C) Western blot performed on induced peptides using antibodies raised against PR domain (Ab-PR). Lane 5: pTr-mPR. Peptide PR could not be induced from pTr-PR. pTr-mPR induced a specific peptide of about 14 kDa, corresponding to the protease domain (with mutation), and recognized by Ab-PR. ::: ![](1743-422X-2-33-6) ::: Analysis of the processing of polyprotein P3 -------------------------------------------- Pulse-chase labelling techniques were used to reveal possible processing of P3 by the protease in E. coli. After 1 hour induction cultures were labelled with ^35^S-methionine for five min after which protein extracts were subjected to SDS-PAGE, followed by autoradiography. Multiple bands were observed in cultures that contain each plasmid (Figure [7](#F7){ref-type="fig"}). Protein patterns represent peptides that are expressed and processed between 60 and 65 min after induction. Patterns of peptides induced from construct pET-MP-PR and pET-MP-mPR were very similar with one visible difference, namely the presence of a band at 37 kDa in cells that contain pET-MP-PR. This band is absent in cells containing pET-MP-mPR, the plasmid with mutant PR. This difference was also observed between cultures that contain constructs pET-P3 and pET-mP3. The 37 kDa band was not present for the clone pET-MP and pET-28 (no insert) but was in cells that contain constructs that code a peptide that contains the CP and the protease, i.e. pET-MP-PRand pET-P3. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **In vitroreleasing of the coat protein from the polyprotein P3*. Autoradiography of an SDS-PAGE of induced peptides from different pET-vectors induced in E. coli. ^35^S radiolabelled methionine was added for 5 minutes after 60 minutes of induction with IPTG. Numbers (in kDa) on the left indicate mobility of the molecular weight markers. Lane 1: pET(no insert); Lane 2: pET-MP; Lane 3: pET-MP-PR; Lane 4: pET-MP-mPR; Lane 5: pET-P3; Lane 6: pET-mP3. Arrow shows the presence of a peptide (estimated molecular mass of 37 kDa) that is present only for constructs that code a peptide that contains the coat protein and the protease (pET-MP-PR; pET-P3). ::: ![](1743-422X-2-33-7) ::: Discussion ========== Retroviral proteases have been studied for many years as they are essential in the control of the replication of retroviruses. Retroviral proteases contain about 100 amino acids residues in length, and contain one copy of the active site Asp-Thr-Gly or Asp-Ser-Gly \[[@B22]\]. It was proposed that they are active as homo-dimers \[[@B23]\]. Crystallographic structures of Rous sarcoma virus (RSV) and Human immunodeficiency virus (HIV) showed that the PR of the viruses are dimeric, with two copies of the active site being brought into close proximity at the junction between the dimer partners \[[@B24],[@B25]\]. To date, numerous retroviral proteases have been investigated, and their structures determined \[[@B26]\]. The retroviral protease is formed by duplication of four structural elements: a hairpin, a wide loop, an α-helix and a second hairpin. Active site sequences are placed in the extended loop of the structural model, implying that the active site is a number of amino acids away from the N-terminal of retroviral proteins. The CP domain was previously characterized by MALDI-TOF (matrix-assisted laser desorption/ionization-time of flight) mass spectrometry analysis \[[@B18]\]. A single CP domain was identified, with positions of the amino- and carboxy- termini of the CP at aa 477 and 791, respectively. The molecular mass of the CP was determined to be 37,303 Da with an estimated pI of 9.43. A basic pI can explain the absence of peaks related to the CP in our mass spectral analysis. In this paper, virions of RTBV were subjected to in-line HPLC electrospray ionization mass spectrometry analysis. The HPLC column is expected to have retained the CP prior to mass spectrometry analysis. The molecular mass of the protein found by these analyses was determined to be 13,794 ± 4 Da, with a putative isoelectric point of 6.3. We suspected that this protein was the viral protease that was encoded in P3 protein. Five peptide domains, with the predicted mass of the protein identified in virions, were found to contain the active site DSGS, the active site of an aspartate protease, and the conserved region IIG (Figure [2](#F2){ref-type="fig"}). We compared the derived sequences of the five putative proteins with sequences of retroviral proteases, a process that led us to predict that the peptide comprising aa 965--1085 represents the protease, based on the position of the active site within the predicted sequence. In the predicted protein the Aspartic acid residue of active site would be 23 amino acid away from N-terminus; the active site is 25 and 37 aa from the N-terminus for HIV and RSV, respectively. To confirm if the target sequence is or is not a protease, we used a combination of sequence and structural predictions procedures to build a structural model for the RTBV protease (Figure [4-A](#F4){ref-type="fig"}) with RSV protease as a template (Figure [4-B](#F4){ref-type="fig"}). The consistency in fold recognition was given by reliable scores against different templates for the same SCOP classification (SCOP classification b.50.1). The proteins of this fold show a closed beta barrel. In the model, some of the beta strands elements are missing, however the remaining beta sheets can be arranged in a predicted conformation as they are product of sequence duplications. Keeping in mind that the protease domain is part of a multi-domain protein, additional interactions among the domains may influence the folding of individual subunits. The conservation in the predicted active site, and the predicted overall folding of sequences led us to predict that the domain posseses a protease activity. Subsequent experiment evidence that confirmed that the predicted active site can be inactive by mutation of D to A, the predicted secondary structure and the fold recognition analyses with model building led us to conclude that the protein comprises an aspartate protease. Plasmids that encode peptides PR and mPR were introduced for expression in E. coli: however, PR did not accumulate in E.coliwhile mPR was expressed normally and reacted with Ab-RTBV and with Ab-PR antibodies. In other studies, the mutation of D to A in the active site of the RTBV protease was shown to affect its activity \[[@B19]\]. We suggest that PR did not accumulate in E. colibecause the peptide was an active protease that was not tolerated in the host. Previous attempts to express proteases in E. colihave had similar outcomes \[[@B27],[@B28]\]. An antiserum against the region aa 881--1098 of the P3 was produced in previous studies \[[@B20]\]; this peptide includes the protease domain. Using this antibody a protein of approximately 13.5 kDa was detected by western blots assays in virus preparations and in infected tissues, suggesting that the protein that was detected represented the viral protease \[[@B20]\]. Furthermore, the antiserum was used to label virus particles, and revealed that the label was attached to virions. The characterization of PR domain, and the immunodetection reactions performed in the present studies are in agreement with the previous results, and also with studies performed by Marmey et al., \[[@B18]\], which investigated a peptide comprising aa 806--961 of P3, that was referred to as IR. IR did not react with Ab-RTBV serum, suggesting that the IR region did not contain sequences related to the CP. Our results support the hypothesis that peptide IR corresponds to the intervening region between CP and PR, and that it may be involved in the processing of P3 \[[@B18]\]. With the present study the N- and C- terminal amino acid sequences are now characterized for CP, PR and RT-Rnase H \[[@B18],[@B19]\]. However, we did not identify apparent sequence similarities between the cleavage sites that would be used by the PR (Figure [8](#F8){ref-type="fig"}). Such a lack of sequence similarity is usual for viral aspartate proteases \[[@B29],[@B30]\]. Other details that remain to beclarified in the organization of P3 include: characterization of the movement domain, and the order and rates in which the various sites on P3 are cleaved. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Protease cleavage site sequences in the RTBV polyprotein P3. Protease cleavage site sequences in the RTBV polyprotein P3. The designation of amino acid residues spanning the cleavage site is according to \[40\]. MP: movement protein; IR: intervening region; CP: capsid protein; PR: Protease; RT: Reverse transcriptase ; Rnase H: Ribonuclease H. Cleavage site sequences MP/IR has not been determined yet. A lack of significant sequence similarities is observed, a characteristic of aspartate proteases. ::: ![](1743-422X-2-33-8) ::: A previous work conducted in insect cells using baculovirus based constructs, including constructs in which the active site of the protease was mutated revealed that RT was processed by PR \[[@B19]\]. In the present work, in vitroprocessing of CP by PR was demonstrated in E.coli. If immunoprecipitations with antibodies were not achieved for technical reasons, presence of the 37 kDa peptide was associated with co-existence of CP and active PR in the construct. It is the first time that such a processing is demonstrated for pararetroviruses (e.g. Commelina yellow mottle virus, Banana streak virus, Cacao swollen shoot virus), where CP and PR are components of the same polyprotein. Our results clarify the organization of P3, and its processing by its own protease and lead to a more complete understanding of the replication process and possible points of control of pararetroviruses. Methods ======= RTBV strain used for the analysis --------------------------------- The RTBV strain used for the analysis was from the International Rice Research Institute (IRRI, Los Banos, Philippines). Sequence of the genome was published \[[@B11]\], with accession number \[GenBank:M65026\]. Mutation of the active site of protease --------------------------------------- Plasmid pBS-mp/RT \[[@B19]\] contain the putative mutated protease and reverse transcriptase of RTBV. The aspartic acid in the sequence [D]{.underline}SGS (RTBV P3, amino acid 987) was changed to alanine, resulting in the sequence [A]{.underline}SGS. Plasmid pBS-mp/RT was used for further sub-cloning. Analysis by mass spectrometry ----------------------------- In-line HPLC electrospray ionization mass spectrometry \[[@B31]\] was employed. The experimental protocol was similar to that described \[[@B19]\]. MacBioSpec algorithms (Sciex) were used to convert the family of ion peaks, which result from the protein being in various charge states, to an accurate molecular mass. Virus sample was denatured with guanidium hydrochloride at 4.8 M prior to injection onto the column. Sequence and structural prediction analyses ------------------------------------------- The RTBV sequence was used as a query for the BLAST program against non redundant databases and PDB databases. No significant hits were identified with suitable e-values by these queries. The sequence was then submitted to fold recognition methods at the metaserver <http://genesilico.pl>\[[@B32]\]. Reliable templates were found with high scores, all of which were found in retroviral proteases (all beta proteins: SCOP classification b.50.1.1). Several pairwise alignments between RTBV and templates were checked using SQUARE \[[@B33]\] and further submitted to homology modelling using the Swissmodel program \[[@B34]\]. Models were evaluated using PSQS <http://www1.jcsg.org/psqs/>\[[@B35]\] and Whatif <http://www.cmbi.kun.nl/gv/servers/>\[[@B36]\] tools. The structure of the Rous sarcoma virus (RSV) protease (pdb code 1bai\_A) provided the best model as a viral aspartate protease and was chosen for this purpose. Illustrations for the model were generated using MolMol \[[@B37]\]. Constructions of plasmids ------------------------- The full-length RTBV clone pBSR63A \[[@B11]\] was used as DNA matrix for PCR reactions to amplify specific regions of ORF3 using specific primers that were designed to amplify specific sequences from the RTBV genome. Constructs were obtained by cloning the PCR fragments into vectors and/or by cloning fragments obtained after digestion of constructs with restriction enzymes and pBS-mp/PR (Table [1](#T1){ref-type="table"}). Cloning was conducted in pBluescript KS vector (Stratagene/USA), in pTrHis vector (Invitrogen/USA) and in pET vector (Novagen/USA). Restriction enzymes were used according to manufacturer (Gibco-BRL, USA). Expression of proteins ---------------------- All pTrHis based vectors were transformed into E.coliDH5-α. The resulting plasmids were designated pTr-PR, pTr-mPR and resulted in synthesis of peptides, named PR, mPR, corresponding to regions between residues 965--1085 (Figure [5](#F5){ref-type="fig"}), with plasmid pTr-mPR encoding peptide with amino acid 987 mutated from D to A. Analysis of in vitroprocessing -------------------------------- All pET vectors were transformed into E.coliBL21/DE3 (pLys S). The resulting plasmids were designated pET-MP, pET-MP/PR, pET-P3, pET-MP/mPR, pET-mP3. These plasmids encode (in order) peptides, named MP, MP-PR, P3, MP-mPR, mP3 corresponding to regions between residues 1--606, 1--1195, 1--1675, 1--1195, 1--1675, respectively (Figure [5](#F5){ref-type="fig"}). Plasmids pET-MP/mPR and pET mORF3 encode peptides with amino acid 987 mutated from D to A. Bacteria were induced with IPTG in M9 salts medium, using rifampicin and an amino acid mixture that lacks methionine. At 60 minutes, ^35^S radiolabelled methionine was added for five minutes. Bacteria were centrifuged for three minutes and resuspended in Laemli sample buffer \[[@B38]\]. Samples were subjected to electrophoresis, the gel was dried and exposed to x-ray film overnight. Antibodies and Western blot analysis ------------------------------------ Antibodies were obtained as previously described \[[@B18]\]. Peptide mPR, corresponding to region between residues 965--1085 and expressed with construct pTr-mPR, was used to produce Ab-PR. An antiserum (Ab-RTBV) was also raised against purified virions. Proteins were subjected to electrophoresis in SDS/PAGE, and transferred to a nitrocellulose membrane. The blot was incubated with antiserum at a 1:1000 dilution. Immunogenic proteins were detected using an alkaline phosphatase goat anti-rabbit IgG at 1:10000 dilution. Proteins were visualized in the presence of nitroblue tetrazolium and 5-bromo-4-chloro-3-indoyl phosphate, or using the Biomax chemiluminescent detection system (Kodak/USA). Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= PM carried out the designing of primers, the construction of various plasmids, the in vitro*expression analysis and pulse chase labelling experiments; ARM performed the structural modelling analysis; AdK analyzed various sequences by computer; RNB and CMF were principal investigators. All authors read and approved the final manuscript. Acknowledgements ================ We thank Dr. S.B.H. Kent for performing mass spectral analysis, and Dr. F. Mathieu-Daudé for critical reading of the manuscript. Work was performed at The Scripps Research Institute in ILTAB (International Laboratory for Tropical Agricultural Biotechnology).	PubMed Central	2024-06-05T03:55:56.246739	2005-4-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087892/", "journal": "Virol J. 2005 Apr 14; 2:33", "authors": [ { "first": "Philippe", "last": "Marmey" }, { "first": "Ana", "last": "Rojas-Mendoza" }, { "first": "Alexandre", "last": "de Kochko" }, { "first": "Roger N", "last": "Beachy" }, { "first": "Claude M", "last": "Fauquet" } ] }
PMC1087893	Background ========== Cholesterol-rich microdomains like rafts and caveolae are specialized regions of the plasma membrane and play an important role for several cellular processes e.g., signal transduction, and for the life cycle of certain viruses (e.g., the entry and exit steps). These domains are enriched in cholesterol, sphingomyelin, ganglioside GM1 and caveolin proteins \[[@B1]\]. The cholesterol molecules are intercalated between the lipid acyl chains and cause a decrease of the fluidity of these membrane regions leading to their resistance against treatment with non-ionic detergents like Triton X-100 at 4°C \[[@B1]\]; therefore, these regions are also referred to as detergent resistant microdomains (DRMs). The specific lipid composition of DRMs leads to the selective incorporation and concentration of specific cellular proteins (reviewed in \[[@B1]\]). Recently, the envelope protein (Env) of the ecotropic murine leukaemia virus (E-MLV) as well as of human immunodeficiency virus type 1 (HIV-1) were shown to associate with DRMs after transport to the plasma membrane \[[@B2],[@B3]\]. Similarly, Gag proteins of HIV-1 prefer DRMs as cellular destinations after synthesis in the cytoplasm \[[@B4]-[@B6]\]. As HIV-1 and E-MLV bud from plasma membrane regions where the viral capsid and envelope proteins are enriched \[[@B7],[@B8]\] the DRM-association of the viral proteins led directly to the idea that DRMs are platforms for assembly and budding (reviewed in \[[@B9]\]). Glycosyl phosphatidylinositol (gpi) anchoring and fatty acylation have been shown to direct proteins to lipid rafts (reviewed in \[[@B10],[@B11]\]). Mutation of HIV-1 Env or E-MLV Env palmitoylation sites \[[@B2],[@B3]\] or the HIV-1 Gag myristoylation site \[[@B4]\] impaired the association of these proteins with DRMs. Furthermore, knock out of Env palmitoylation sites led to a decreased viral titer due to a reduced Env incorporation into the viral particles \[[@B3]\]. Viral budding from DRMs should lead to a viral membrane composition, which resembles the lipid composition of DRMs and differs from the average distribution of lipids in the plasma membrane \[[@B9]\]. For example, the enrichment of the membrane of HIV-1 with sphingomyelin and cholesterol \[[@B12],[@B13]\] strongly supports a role for DRMs in HIV-1 budding (reviewed in \[[@B9]\]). In a recent report, we showed that a 1.4 fold increase of the cholesterol content of the plasma membrane of NIH3T3 cells resulted in a more than 3-fold increase of viral membrane cholesterol of amphotropic MLV (A-MLV) released from these cells \[[@B14]\]. We suggested that this phenomenon could be due to the involvement of DRMs in assembly and budding of A-MLV. To address this issue, we have here performed density gradient centrifugation, immunocytochemical staining and co-localization experiments using A-MLV Env expressing NIH3T3 and 293 cells. Results ======= Triton X-100 insolubility of A-MLV Env -------------------------------------- To investigate the association of A-MLV Env with DRMs via density gradient centrifugation, 293T cells were transiently transfected with a pHIT-derived plasmid encoding the A-MLV envelope protein \[[@B15]\]. Moreover, expression plasmids encoding enhanced green fluorescent protein (eGFP) (pEGFP-N1, Clontech) were transiently transfected in 293T cells and used as non-DRM marker. Forty-eight hours after transfection, the cells were treated for 10 minutes with 1% TX-100 at 4°C and the resulting cell lysates were loaded on discontinuous density gradients. Due to the insolubility of DRMs, these membrane regions as well as their associated proteins float to the top of the gradient \[[@B16]\]. Confirming the routine of the fractionation experiments, unmodified eGFP, which is localized in the cytoplasm, was exclusively found in the soluble fractions 5 and 6 (Fig [1A](#F1){ref-type="fig"}). Therefore, these fractions were considered as detergent soluble fractions. A-MLV Env floated predominantly to the DRM fractions 2, 3, and 4. Fractions 5 and 6 contained high background signals, in which no A-MLV Env specific band could be detected (Fig [1A](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A-MLV envelope protein associates with detergent resistant microdomains (DRMs).A)293T cells producing A-MLV were treated with TX-100 at 4°C and loaded on a discontinuous sucrose gradient. Western blot analyses were performed. Fraction 1 corresponds to the top and fraction 6 corresponds to the bottom of the tube. Fractions 1 to 4 contain the DRMs, fractions 5 to 6 the non-DRM membrane fractions. A-MLV Env is found predominantly in the DRM fractions 2, 3 and 4. EGFP, which is localized in the cytoplasm, remains in the soluble fractions 5 and 6. B)NIH3T3 cells releasing A-MLV were treated with TX-100 at 4°C and loaded on a discontinuous sucrose gradient. Dot blot analyses were performed. Fraction 1 corresponds to the top and fraction 6 corresponds to the bottom of the tube, respectively. B is the background of the dot blot. Fractions were processed in parallel for immunological detection of cav-1 and A-MLV Env. C)Quantification of the dot blot shown in B) using image analysing software. The amounts shown are determined as percentages of the total of all dots; DRM (fractions 1 to 3), non-DRM (fractions 4 to 6). D)Detergent soluble supernatant (non-DRM) and insoluble pellet (DRM) of A-MLV producing NIH3T3 cells treated with TX-100 at 4°C or 37°C were investigated for the amount of envelope protein using dot blot analysis. The results of two independent experiments are shown. The amounts shown are determined as percentages of the total of all dots. ::: ![](1743-422X-2-36-1) ::: Additional fractionation experiments were performed using A-MLV producing NIH3T3 cells. Analysis of the resulting Dot Blots revealed that at least 60% of the viral Env protein was localized within the detergent insoluble fractions when the cells were treated with TX-100 at 4°C (Fig [1B](#F1){ref-type="fig"} and [1C](#F1){ref-type="fig"}). As unspecific background was found only in detergent soluble fractions (Fig [1A](#F1){ref-type="fig"}), an overestimation of the amount of DRM associated A-MLV Env is unlikely. In addition, TX-100 treatment of the cells at 37°C dissolved rafts and drastically reduced the percentage of Env associated with the detergent insoluble fractions (Fig [1D](#F1){ref-type="fig"}). In summary, these data imply that a large fraction of A-MLV Env is localized in DRMs. A-MLV Env exhibits properties of DRM-associated proteins -------------------------------------------------------- To verify A-MLV Env association with DRMs at the cell level, a set of immunocytochemical experiments were performed employing DRM (caveolin-1 (cav-1)) and non-DRM (CD71) markers. Moreover, the cell surface receptor for cholera toxin, the glycolipid GM1, was detected with fluorescent labelled subunits of the cholera toxin, which represents a standard method for DRM identification \[[@B17]\]. Cav-1 is a major component of caveolae, which are flask-shaped invaginations of the plasma membrane involved in endocytic processes. Cav-1 is also present in lipid rafts, which are thought to be precursors of caveolae (\"pre-caveolae\") \[[@B18]\]. The transferrin receptor (CD71) is localized in clathrin coated pits or in other plasma membrane regions, but is absent from DRMs \[[@B17],[@B19]\]. Wild-type A-MLV releasing NIH3T3 cells grown on chamber slides were washed with PBS or 0.5% TX-100 at 4°C and subsequently fixed to the glass surface with paraformaldehyde. The cells were treated with filipin, a cholesterol-binding fluorescent dye \[[@B20]\], and stained for the DRM markers GM1 and cav-1 using FITC labelled cholera toxin or anti cav-1 antibody, respectively, and for CD71 using an anti CD71 antibody. A-MLV Env was detected using an anti-Env antibody (83A25 \[[@B21]\]). The relatively mild TX-100 treatment was sufficient to disperse CD71, which is not associated with DRMs, over the plasma membrane while the DRM markers GM1 and cav-1 as well as A-MLV Env remained as discrete spots (Fig [2A](#F2){ref-type="fig"}, compare left and middle columns). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Immunocytochemical investigations of the association of proteins with DRMs.A)NIH3T3 cells producing A-MLV were treated with PBS, TX-100 or MBCD as indicated and subsequently subjected to TX-100 extraction and stained for cav-1, GM1, CD71 and A-MLV Env as indicated. B)Background of the secondary antibody used for cav-1 staining. C)Background of the secondary antibody used for A-MLV Env staining. D)NIH3T3 cells (Env negative) stained for A-MLV Env, negative control (see text for details). Photographs were taken using an oil immersion objective, original magnification 1000×. ::: ![](1743-422X-2-36-2) ::: In another set-up, the cells were first treated for 30 min with 5 mM methyl-beta-cyclodextrin (MBCD) at 37°C and subsequently with 0.5% TX-100 at 4°C prior to paraformaldehyde fixation and immunocytochemical staining (Fig. [2A](#F2){ref-type="fig"}, right column). MDCB is known to extract cholesterol from plasma membranes and is widely used to disrupt DRMs \[[@B22]\]. Enzymatic cholesterol determination revealed that approximately 60% of the cholesterol was removed from the plasma membrane upon MBCD treatment (data not shown). Due to disruption of the DRM structure, a combined MBCD/TX-100 treatment should result in dispersal of DRMs and proteins concentrated therein. Indeed, the combined MBCD/TX-100 treatment resulted in even distribution of GM1 as well as A-MLV Env fluorescence in the plasma membrane while cav-1 still was detectable in discrete spots in MBCD/TX-100 treated cells (Fig. [2A](#F2){ref-type="fig"}, right column). With respect to the distribution in the plasma membrane, TX-100 resistance, and MBCD extraction, A-MLV Env exhibits similar properties as the DRM marker GM1 and distinct properties compared to CD71. These findings are in agreement with the results obtained from density gradient centrifugations showing that A-MLV Env to a high degree is associated with DRMs. A-MLV Env co-localizes with DRM markers --------------------------------------- Finally, we performed co-localization studies of Env proteins with DRM markers in immunocytochemical experiments. Again, we used a combination of TX-100 treatment and immunocytochemical stainings. Wild-type A-MLV producing NIH3T3 cells grown on chamber slides were washed with PBS or 0.5% TX-100 at 4°C and subsequently fixed to the glass surface by paraformaldehyde treatment. The cells were incubated with filipin, a cholesterol-binding fluorescent dye \[[@B20]\], and DRM markers GM1 and cav-1 were detected using FITC labelled cholera toxin or anti cav-1 antibody, respectively. A-MLV Env was detected using an anti-Env antibody (83A25 \[[@B21]\]). As expected for a DRM-associated protein and from the results of the Dot Blot analysis (Fig. [1B](#F1){ref-type="fig"} and [1C](#F1){ref-type="fig"}), approximately 50% of A-MLV Env co-localized with cholesterol-rich spots (Fig. [3A](#F3){ref-type="fig"}). In accordance with the experiment shown in figure [2A](#F2){ref-type="fig"}, A-MLV Env did not disperse in the plasma membrane after TX-100 treatment (Fig. [3A](#F3){ref-type="fig"}). In addition, A-MLV Env also co-localized with cav-1 and GM1 resulting in yellow spots in merged photographs (Fig. [3B](#F3){ref-type="fig"} and [3C](#F3){ref-type="fig"}). No co-localization was observed when cells were stained for A-MLV Env and the non-DRM marker CD71 (data not shown). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A-MLV Env co-localization with cholesterol, GM1 and cav-1.A)A-MLV Env co-localization with cholesterol. NIH3T3 cells producing wild-type A-MLV were treated with filipin for cholesterol detection (left column) and with an A-MLV Env specific antibody (second column) after fixation and treatment with PBS (top) or TX-100 at 4°C (bottom). Co-localization result in pink spots (merged images, third column). The column on the right shows the result of the co-localization finder plugin of the ImageJ program \[[@B30]\] merged with the original A-MLV Env staining. Turquoise colour indicates co-localization of A-MLV Env with cholesterol. B)A-MLV Env and cav-1 co-localization monitored by fluorescence microscopy. Immunofluorescent detection of cav-1 (left) and the A-MLV Env (middle) after treatment with TX-100 at 4°C in NIH3T3 cells producing A-MLV. Co-localization result in yellow spots (right). C)A-MLV Env (left) and GM1 (middle) were detected by immunofluorescence in A-MLV producing NIH3T3 cells after PBS (top) or TX-100 treatment at 4°C (bottom). Co-localization result in yellow spots (right). All photographs were taken using a fluorescence microscope and oil immersion objective, original magnification 1000×. ::: ![](1743-422X-2-36-3) ::: Taken together, the immunocytochemical data confirm that of A-MLV Env to a large extent is associated with DRMs. Discussion ========== A number of previous investigations have shown that the plasma membrane of animal cells is a heterogeneous lipid bilayer that contains distinct cholesterol-rich micro-domains like DRMs, which are responsible for a number of biological functions e.g., concentrating and sorting of proteins \[[@B1]\]. A variety of viruses like HIV-1 and measles virus exploit DRMs for their assembly and budding \[[@B6],[@B23]\] after association of certain structural proteins with DRMs. Here we show that the major portion of plasma membrane A-MLV Env is associated with DRMs. Using biochemical and immunocytochemical methods we found that approximately 60--80% of A-MLV Env is localized in these microdomains. Similarly, Li et al. have reported that the closely related envelope protein of the Moloney murine leukaemia virus (MoMLV), which shows 62% identity to A-MLV Env on the protein level \[[@B2],[@B24]\], is associated with rafts. Similar to MoMLV Env, A-MLV Env is not completely localized within DRMs. This is not uncommon for DRM-associated proteins as it has been shown for, e.g., HIV-1 p17 and gp41 \[[@B6]\]. The immunocytochemical method used here for investigation of the DRM association of A-MLV Env was shown to be suitable. The markers for DRM (cav-1, GM1) and non-DRM regions (CD71) of the plasma membrane exhibited the properties expected when the cells were treated with the non-ionic detergent TX-100. These experiments showed that A-MLV Env resembles GM1 or cav-1 upon treatment with TX-100. MBCD is known to dissolve DRMs by extracting cholesterol from the plasma membrane. As expected for a DRM associated protein, cholesterol extraction and subsequent treatment of the cells with TX-100 dispersed GM1 and A-MLV Env spots at the plasma membrane. In contrast, cav-1-positive spots were still detectable even when these were depleted of cholesterol (data not shown). This is in accordance with a previous investigation demonstrating that only a negligible amount of cav-1 could be released through MBCD treatment \[[@B22]\]. Probably, MBCD resistance of caveolin-spots is due to the fact that the caveolin proteins build up a close network on the luminal side of the plasma membrane \[[@B25]\]. Furthermore, A-MLV Env co-localizes with the DRM markers cholesterol, cav-1 and GM1 confirming that A-MLV Env to a high degree is associated with DRMs. Retrovirus assembly and release is solely driven by the viral Gag polyprotein \[[@B28]\], thus virus-like particles are formed in the absence of any other viral proteins or genome. Since, the spatial neighbourhood of Env and Gag proteins is a prerequisite for release of functional viral particles, the localisation of A-MLV Env within DRMs may be indicative of viral budding from these regions. This model is supported by the fact that a 1.4 fold increase of the cholesterol content of the plasma membrane of NIH3T3 cells resulted in a more than 3-fold increase of viral membrane cholesterol of amphotropic MLV (A-MLV) released from these cells \[[@B14]\]. Our finding may have consequences for the understanding of A-MLV assembly and budding, which is known to be a specific and coordinated process. In the case of A-MLV, previous data indicated that the viral components assemble and bud at the cellular plasma membrane (reviewed in \[[@B8]\]). Recent investigations of Sandrin et al., however, demonstrate intracellular co-localization of A-MLV Env and MoMLV core proteins in the endocytic pathway in late endosomes including multivesicular bodies (MVBs). They suggest that the interaction of MLV Env and core proteins in these compartments could influence virus particle formation \[[@B27]\]. According to general belief DRM like microdomains are already formed in the Golgi, and it is thus possible that A-MLV Env and core proteins are already sorted intracellularly in the same compartment and transported together to the plasma membrane. Co-localization has been suggested to be sufficient for incorporation of cellular proteins into virions \[[@B26]\]. Since cav-1 and A-MLV Env co-localize in mouse NIH3T3 cells the putative presence of cav-1 in A-MLV virions would indicate that A-MLV buds from cav-1 containing DRMs. Interestingly, we have found that cav-1 is incorporated into A-MLV virions, whereas no CD71 could be detected (Beer and Wirth, unpublished data). However, whether cav-1 plays a specific role in viral protein sorting to the plasma membrane and viral assembly is presently not known, but this issue is subject of current investigations. Nevertheless, based on the specific properties of individual DRMs, like rafts or caveolae, rafts seem to be most suitable for virus assembly and budding. The invagination of caveolae within the plasma membrane of the cells, their involvement in endocytic processes and, moreover, their compact coat of caveolin-oligomeres \[[@B25]\] presumably exclude caveolae as suitable regions for viral budding and suggest rafts as budding platforms for A-MLV. Conclusions =========== Taken together, our findings provide evidence that A-MLV Env is localized in DRMs, similar to the Env of the closely related E-MLV \[[@B2],[@B26]\] and lentiviral HIV-1 Env \[[@B6]\] These results suggest that rafts are budding platforms for A-MLV in NIH3T3 and 293T cells. Methods ======= Cells ----- NIH3T3 (ATCC CRL-1658) and 293T (ATCC CRL-11268) cells were propagated in DMEM supplemented with glutamine and 10% FCS. Antibody producing hybridoma cells were grown in RPMI 1640 medium supplemented with glutamine and 1% ultra low IgG FCS (Gibco). All cells were grown at 37°C, 5% CO2 and 95% humidity. Plasmids, transfection and helper virus approach ------------------------------------------------ pMLVampho contains the complete genome of A-MLV cloned into pBluescript (Genethon, France received via J.-C. Pages). A-MLV producing NIH3T3 cells resulted from transfection of pMLVampho \[[@B29]\] and a subsequent infection of NIH3T3 cells with replication-competent MLV-A. Antibodies and antibody production ---------------------------------- Hybridoma cell lines were used for the production of rat monoclonal immunoglobulin G (IgG) antibodies against MLV SU (83A25, kindly provided by L.H.Evans \[[@B21]\]). To concentrate the antibodies, the cell suspension was centrifuged at 2000 × g for 10 minutes. 29.1 g ammoniumsulfat per 100 ml were added and the supernatant stirred for 1 hour at 4°C. After centrifugation (27000 × g, 4°C, 1 h), the pellet was resuspended in PBS and the antibody solution dialyzed against PBS. For Western and dot blot analysis rabbit anti rat IgG coupled to horseradish peroxidase (HRP) (Sigma) was used. Antibody to mouse CD71 was purchased from ebioscience and to caveolin from BD Bioscience. Fluorescein isothiocyanate (FITC)-conjugated goat anti rat IgG was obtained from Sigma and Texas Red labelled goat anti rabbit IgG was purchased from Calbiochem. Texas Red conjugated goat anti rat IgG and FITC-conjugated goat anti rabbit IgG were obtained from Jackson Immunoresearch. Triton X-100 extraction and sucrose gradient -------------------------------------------- To investigate the association of the A-MLV envelope protein with cholesterol-rich microdomains, 293T cells were transfected with either pEGFP-N1 (Clontech) or A-MLV Env encoding plasmids \[[@B15]\] using the calcium phosphate precipitation method. 48 hours after transfection, the cells were washed with 1×PBS, overlaid with 1×PBS and washed from the cell culture flask surface. The cells were pelleted with 300×g at 4°C and resuspended in icecold 1×PBS containing 1% TritonX-100 and 1 mM Pefabloc (Sigma). The cells were incubated 30 min on ice and adjusted to 40% sucrose or OptiPrep and loaded into SW60Ti-tubes. The samples were overlaid with a discontinuous sucrose or OptiPrep gradient (35% -- 5%). The gradient was centrifuged at 4°C with 40000 rpm for 20 h in a SW60Ti rotor. Six fractions were collected from the top of the tube. An equal volume of acetone was added to the fraction and incubated at -20°C. The precipitated proteins were pelleted by centrifugation and dried at room temperature. The pellet was resuspended in 1×SDS gel loading buffer. The fractions were analysed for their egfp-N1 or A-MLV Env protein content using a 12% SDS gel and Western Blot. Anti-gfp antibody was obtained from Abcam (AB290). A-MLV Env was detected using antibodies produced by the hybridoma cell line 83A25. Dot immunoassay --------------- To investigate the association of proteins with cholesterol-rich microdomains via Western or Dot Blot the extraction of TX-100 soluble proteins was performed as described previously \[[@B6]\] with the following modifications. NIH3T3 cells were washed with PBS, overlaid with 4°C cold 0.5% Triton X-100 in the presence of a protease inhibitor cocktail (Pefabloc, Sigma) and gently shaked at 4°C for 1 min. The supernatant was removed and stored on ice. The remaining cells were suspended in PBS and homogenized in a RiboLayser tube at 6000 rpm. The stored soluble protein fraction was adjusted to 40% sucrose in TKM buffer (50 mM Tris-HCl, pH 7.4; 25 mM KCl; 5 mM MgCl~2~; 1 mM EDTA) and loaded into SW40Ti-tubes. The sample was overlaid with 35% to 5% sucrose (5% steps). The gradient was overlaid with the homogenized cell pellet. The gradient was centrifuged at 4°C with 38000 rpm for 20 h. Six fractions were collected from the top of the tube. 100 μl portions of each fraction were diluted with 400 μl PBS, filled into the wells of a Bio Dot apparatus (BioRad) and gently suctioned onto nitrocellulose membranes (MilliPore). The membrane strips were blocked for 1 h with Tris-buffered saline containing 10% horse serum and 3% bovine serum albumin. To detect A-MLV envelope and cav-1 proteins the membrane was incubated over night with antibodies against the proteins in blocking buffer at a 1:200 (Env) and 1:5000 (cav-1) dilution. The secondary antibodies, rabbit anti rat and goat anti rabbit coupled to HRP, were used at a 1:1000 dilution. The dot blots were developed with TMB stabilized substrate for HRP (Promega). The spot intensities were quantified using Easy Win 32 (Herolab). MBCD treatment -------------- To extract cholesterol out of the cellular plasma membrane NIH3T3 cells were overlaid with 5 mM Methyl-β-cyclodextrin (MBCD, Sigma). After slightly shaking at 37°C for 30 min, the cells were used for further treatment with Triton X-100. Immunofluorescent staining -------------------------- NIH3T3 cells were seeded onto chamber slides (Nunc) and grown to 80% confluency. After washing once with PBS, the cells were overlaid with 200 μl PBS or 0.5% Triton X-100 (4°C) and incubated for 1 minute at 4°C (gently shaking at 8 rpm). Afterwards the cells were immediately overlaid with 4% paraformaldehyde and incubated for 15 min at RT. After washing with PBS and blocking with Tris-buffered saline containing 10% horse serum and 3% bovine serum albumin antibodies against A-MLV Env, and cav-1 were added. The cells were overlaid with secondary antibodies after washing with PBS. After a final washing step with PBS the slides were mounted with immunofluorescence mounting medium (Dako). For co-localization studies the cells were blocked a second time after incubation with the secondary antibody and stained for GM1 with FITC-conjugated cholera toxin (Calbiochem, 8 μg/ml), for cholesterol with filipin (Sigma, 50 μg/ml) or cav-1 as described above. A fluorescence microscope (Axiovert TV135, Zeiss; filter sets: filipin -- XF113, 387/450 nm (Em/Ex), FITC -- 495/520 nm, Texas Red -- 595/615 nm; Omega filters) at 1000× magnification was used for the detection of the stained proteins. Images were taken using a cooled CCD camera (PXL 1400, Photometrics), digitalized, pseudo-coloured and merged (IPLab Spectrum). Brightness and contrast were adjusted. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= CB conceived of the study, carried out the experimental work and helped to draft the manuscript. MW participated in the design of the study, supervision of conduction of the experiments and drafted the manuscript. LP helped with coordination and design of the density gradients. All authors read and approved the final manuscript. Acknowledgements ================ Part of the work presented in this article was funded from the German Academy of Natural Scientists Leopoldina (BMBF-LPD 9901/8-81) (C.B.) and the Lundbeck Foundation, the Novo Nordisk Foundation, the Danish Medical Research Council (Grant 22-03-0254) (L.P.).	PubMed Central	2024-06-05T03:55:56.249668	2005-4-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087893/", "journal": "Virol J. 2005 Apr 19; 2:36", "authors": [ { "first": "Christiane", "last": "Beer" }, { "first": "Lene", "last": "Pedersen" }, { "first": "Manfred", "last": "Wirth" } ] }
PMC1087894	Background ========== Viruses commonly enter cells by receptor-mediated endocytosis; an entry pathway involving clathrin-coated pits and vesicles derived from them. These vesicles generally transport the virus to the endosomal/lysosomal compartment, where acidic conditions trigger virus disassembly and genome release \[[@B1]\]. Earlier experimental findings demonstrated that the entry pathway for simian virus 40 (SV40) might differ in important ways from the more common virus entry pathway. First, SV40 infection was found to be independent of the low pH of the endosomal/lysosomal compartment \[[@B2]\]. Second, electron microscopy studies showed that entering SV40 traffics to the endoplasmic reticulum (ER), rather than to endosomes \[[@B3]\]. More recently, SV40 was shown to enter cells via caveolae, rather than clathrin-coated pits \[[@B4],[@B5]\]. Indeed, these were the first reports of a virus entering cells by means of caveolae. In addition, SV40 entry is signal-dependent, in contrast to the constitutive endocytosis of viruses that enter via clathrin-coated pits \[[@B6]-[@B8]\]. Finally, SV40 particles disassemble in the ER, rather than in the endosomal/lysosomal compartment \[[@B9]\]. Caveolae are small (70 to 100 nm) invaginations of the plasma membrane that are distinguished from clathrin-coated pits by their size, distinctive flask-like shape, and lack of a visible coat in thin sections. Expression of the protein caveolin-1 triggers the formation of caveolae in microdomains of the plasma membrane that are enriched in sphingolipids and cholesterol, and which are known as lipid rafts \[[@B10],[@B11]\]. The cellular functions of caveolae are not yet entirely clear, but they have been implicated in organizing signal transduction pathways, and in sorting and trafficking through the endocytic and secretory pathways \[[@B12]-[@B14]\]. The endocytic trafficking of any ligand from the plasma membrane to the ER is most exceptional. Thus, it was important to ascertain the pathway taken by SV40. Cholera toxin (CT) provided a precedent for a possible SV40 entry pathway since it earlier had been shown to enter cells via caveolae, and also to traffic to the ER \[[@B15]\]. Interestingly, CT also can enter cells via clathrin-coated pits. Yet pharmacologically impairing the clathrin-mediated entry pathway had little effect on the toxic effects of CT. In contrast, selective inhibition of caveolae-mediated uptake prevented CT toxicity \[[@B15]\]. Importantly, productive SV40 infection likewise was prevented by pharmacologically impairing caveolae-mediated endocytosis, but not by blocking clathrin-mediated entry \[[@B4]\]. These experimental findings demonstrated that a caveolae-mediated entry pathway, rather than a clathrin-mediated pathway, is the physiologically relevant one for both the toxin and SV40. Moreover, these results also imply that the clathrin-coated pit-mediated pathway and the caveolae-mediated pathway do not mix or intersect at any point. Importantly, CT traffics through the Golgi en routeto the ER \[[@B16]\]. Transport of the toxin from the Golgi to the ER is mediated by the Golgi-to-ER retrieval pathway, which normally retrieves resident ER proteins that have escaped to the Golgi with the anterograde flux. In contrast, rather than trafficking to the Golgi, SV40 was seen to traffic to a new organelle, the caveosome. SV40 is then transported directly from the caveosome to the ER \[[@B17]\]. Little is known about caveosomes. They contain caveolin-1, but are non-acidic, and they do not contain markers for endosomes, lysosomes, or the ER. Importantly, caveosomes also do not contain the Golgi markers TGN46 or mannosidase II, implying that they are distinct from the Golgi \[[@B17]\]. Prior to the report that SV40 traffics to the ER via caveosomes \[[@B17]\], we set out to ask whether SV40, like CT, might traffic to the ER via the Golgi. We selected β-COP as our Golgi marker since this protein is best known as a component of the COPI coatamer complexes that mediate the retrograde retrieval pathway from the Golgi to the ER \[[@B18]-[@B21]\]. We found that SV40 indeed co localizes with β-COP before it enters the ER. However, in light of the report that SV40 traffics through caveosomes, rather than through the Golgi \[[@B17]\], we interpreted our findings and redirected our subsequent experiments as follows. First, note that the recycling pathway from the Golgi to the ER is mediated by COPI coatamers that assemble on Golgi membranes \[[@B22]-[@B24]\]. Thus, since SV40, like CT, is transported from an intermediate compartment to the ER (from caveosomes in the case of SV40, and from the Golgi in the case of CT), we hypothesized that the SV40 pathway from caveosomes to the ER likewise might be mediated by COPI coatamers. This premise accounts both for the co localization of SV40 with β-COP \[[@B9]\], and the targeting of the virus to the ER. Considering the above, we asked whether β-COP indeed might be present on caveosomes, and whether it might mediate trafficking from caveosomes to the ER. First, we demonstrated that β-COP in fact does co localize with caveolin-1 on an organelle that contains input SV40. Second, the simultaneous co localization of SV40 with both β-COP and caveolin-1 is seen before the virus appears in the ER \[[@B9]\]. Thus, the β-COP-containing intermediate organelle appears to be the caveosome \[[@B17]\]. Third, we and others reported that transport of SV40 from the intermediate organelle to the ER is blocked by the drug brefeldin A, which specifically inhibits the Ras-like GTPase ARF-1 that regulates assembly of COPI coat complexes \[[@B9],[@B25]\]. Note that CT transport to the ER likewise is blocked by brefeldin A, as well as by microinjected antibodies against β-COP \[[@B23],[@B26],[@B27]\]. These experimental results confirm that SV40 indeed traffics through a caveolin-1-containing compartment, which most likely is the caveosome, en route to the ER. Moreover, they demonstrate that the caveosome, like the Golgi, is marked by β-COP. Finally, the pathway from the caveosome to the ER, like the retrograde pathway from the Golgi to the ER, is dependent on assembly of COPI coat complexes \[[@B9]\]. Now, for several reasons, we believe that it is important to revisit the question of whether SV40 is transported to the ER via the Golgi. Most importantly, there is the precedent provided by CT for a caveolae-mediated endocytic pathway that accesses the ER via the Golgi. This precedent becomes more compelling in view of the more recent discovery that CT traffics through a caveolin-1-containing \"endosomal\" compartment on its path to the Golgi \[[@B16]\]. Moreover, that compartment may well be identical to caveosomes, as demonstrated by the finding that when cells are allowed to simultaneously endocytose CT and SV40, these ligands are observed to co localize in caveolin-1-positive endosomes \[[@B16]\]. Yet CT then traffics to the Golgi en route to the ER, whereas SV40 is said to traffic directly from caveosomes to the ER. Finally, in most cultured cells caveolin-1 is seen in the Golgi, as well as at the cell surface. In the current study, we sought to confirm our earlier finding that SV40 co localizes with β-COP \[[@B9]\]. In addition, we ask whether SV40 co localizes with two standard Golgi markers: golgin 97 and BODIPY-ceramide. Results and Discussion ====================== In agreement with our earlier report \[[@B9]\], SV40 indeed co localized with β-COP at all three time points examined (3, 5, and 10 hours), although co localization was diminished by 10 hours (Figure [1A, B](#F1){ref-type="fig"}, and [1C](#F1){ref-type="fig"}, 3-hour sample shown). Regarding the latter observation, we demonstrated earlier that SV40 appears in the ER between five and ten hours, and most of the virus is in the ER by 10 hours \[[@B9]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### SV40 co localizes with β-COP (A, B, C, 3-hour sample), but not with the more stringent Golgi markers, golgin 97 (D, E, F, 5-hour sample) and BODIPY-ceramide (G, H, I, 3-hour sample). ::: ![](1743-422X-2-38-1) ::: In contrast to the early co localization of SV40 with β-COP, at no time was SV40 seen to co localize with Golgi markers golgin 97 (Figure [1D, E](#F1){ref-type="fig"}, and [1F](#F1){ref-type="fig"}, 5-hour sample shown) and BODIPY-ceramide (Figure [1G, H](#F1){ref-type="fig"}, and [1I](#F1){ref-type="fig"}, 3 hour sample shown). The golgins in general are Golgi-localized proteins, characterized by an extensive coiled-coil structure throughout the entire molecule. The anti-human golgin 97 monoclonal antibodies used here recognize a 97 kD protein called golgin 97, a peripheral membrane protein that appears to be localized exclusively on the cytoplasmic face of the Golgi \[[@B28]\]. The exact function of golgin 97 is not known, although there is evidence that it may act to regulate transport between endosomes and the Golgi \[[@B29]\]. Fluorescent ceramide analogs, such as BODIPY-ceramide, are used extensively as selective stains for the Golgi \[e.g., \[[@B30]\]\]. They accumulate in this organelle, presumably because of its role in lipid biosynthesis and trafficking. Indeed β-COP also is used as a Golgi marker. It preferentially associates with the lateral rims of the cisand medial Golgi cisternae, and the buds and vesicles derived from them. However, β-COP is not entirely specific to the Golgi since it also is associated with endosomal vesicles scattered throughout the cytoplasm \[[@B31],[@B32]\]. Our own recent experimental results strongly implied that β-COP also is associated with caveolin-1-containing caveosomes \[[@B9]\]. Based upon the differential co localization of SV40 with β-COP, but not with the less promiscuous Golgi markers, golgin 97 and BODIPY-ceramide, we conclude that SV40 does not traffic through the Golgi en routeto the ER. Both SV40 and CT traffic from the plasma membrane to a common intermediate compartment, which appears to be the caveosome \[[@B16]\]. Since SV40 then traffics directly to the ER, whereas CT is first transported to the Golgi, the caveosome compartment would seem to have the ability to sort its different cargos for transport to different destinations. How this sorting might be achieved remains to be discovered. Caveolin-1 is not likely to play a role in this differential sorting of SV40 and CT, since that protein does not traffic with either ligand to its next destination \[[@B16],[@B17]\]. Moreover, in cells lacking caveolin-1 expression and, therefore, caveolae, SV40 enters via lipid rafts, and still traffics to the ER via neutral organelles that resemble caveosomes, except that they do not contain caveolin-1 \[[@B33]\]. Understanding the determinants of these atypical trafficking pathways is fundamentally important since virtually all host ligands that are not internalized via clathrin-coated pits (e.g., sphingolipids, GPI-linked proteins) enter via sphingolipid and cholesterol-enriched membrane domains, and at least some traffic to a caveosome-like compartment, from which they then are sorted. This clathrin-independent mode of endocytosis likely enables these ligands to access sites that can not be accessed from the clathrin-dependent endocytic pathway. SV40 binds to major histocompatibility complex (MHC) class I molecules at the cell surface \[[@B34],[@B35]\]. However, MHC class I molecules do not internalize with the virus \[[@B36]\]. An interaction of SV40 with the ganglioside GM1 at the plasma membrane recently was shown to greatly enhance infectivity of the virus \[[@B37]\]. Perhaps SV40 uses GM1 as a co-receptor to deliver the virus into the cell. Interestingly, several bacterial toxins likewise use gangliosides for their cell surface receptors. In particular, CT binds to GM1 via its B subunit, and that interaction is necessary for the retrograde transport of CT via the Golgi to the ER \[[@B38]\]. It will be interesting to identify the factors which determine that SV40 takes a direct route from the caveosome to the ER. Conclusion ========== SV40 does not traffic through the Golgi en routefrom the cell surface to the ER. Based upon the current report and the earlier work of others regarding SV40 and cholera toxin \[[@B16]\], the caveosome appears to be an organelle able to sort its different cargos for transport to different destinations within the cell. Moreover, there are at least two distinct but related routes by which a ligand might traffic from the plasma membrane to the endoplasmic reticulum; one involving transport through the Golgi, and the other not involving the Golgi. Methods ======= Cell cultures and infections ---------------------------- CV-1 cells (from the American Type Culture Collection) were seeded on 8 well Lab-Tek chamber slides (Nalge Nunc). SV40 was adsorbed to cells for 1 h at 4°C, at a multiplicity of infection of 50 to 100 plaque-forming units per cell. Cultures then were incubated at 37°C in Dulbecco modified Eagle medium plus 10% newborn calf serum (Atlanta Biologicals). At 3, 5, and 10 hours post infection, samples were washed five times in phosphate-buffered saline and fixed with 70% methanol at -20°C for 10 min. Confocal immunofluorescence microscopy -------------------------------------- Confocal immunofluorescence microscopy was carried out using an epifluorescence Nikon E600light microscope. An ORCA-ER-cooled CCD camera (Hamamatsu) and OPENLAB software (Improvision) were used for all image acquisition and processing. Primary antibodies were monoclonal anti-β-COP antisera (Sigma), monoclonal anti-golgin 97 antisera (Molecular Probes), and our rabbit anti-SV40 antisera \[[@B9]\]. BODIPY (TR)-ceramide was from Molecular Probes. Secondary antibodies were fluorescein-conjugated goat anti-rabbit immunoglobulin G (IgG), Texas Red (TR)-conjugated goat anti-rabbit IgG, fluorescein-conjugated donkey anti-mouse IgG, and TR-conjugated donkey anti-mouse IgG (Jackson Laboratories, West Grove, PA). All antisera were diluted 1:100. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= LN conceived and supervised the study and drafted the manuscript. DK carried out all of the experimental work and data acquisition, taking important initiatives toward those ends. Both authors approved the final manuscript. Acknowledgements ================ This work was supported by Public Health Service Grant CA100479 from the National Cancer Institute to LN.	PubMed Central	2024-06-05T03:55:56.251886	2005-4-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087894/", "journal": "Virol J. 2005 Apr 19; 2:38", "authors": [ { "first": "Leonard C", "last": "Norkin" }, { "first": "Dmitry", "last": "Kuksin" } ] }
PMC1087895	Background ========== Totally implantable venous access devices are widely used for infusion of chemotherapy or parenteral nutrition \[[@B1]-[@B4]\]. Implantation and use of these systems offer a high level of safety and convenience for patients and physicians. Device associated complications include technical operative problems, infections, paravasal infusions and catheter or punction chamber dislocation \[[@B1],[@B2]\]. We present the case of a patient with the rare complication of an intrapulmonal catheter dislocation in a totally implantable venous access system. Case presentation ================= A 49-year-old female with metastatic breast cancer (supraclavicular lymph node metastases) presented with dyspnoea, intermittent coughing and general weakness. The patient had undergone chemotherapy with treosulfane via a totally implantable venous access port catheter the day before. The venous silicone catheter system (Fresenius Intraport, Fresenius Kabi, Bad Homburg, Germany) had been implanted 2 years before via the left cephalic vein by tangential incision of the vein after distal ligation without intra- or postoperative complications in our department. Correct placement had been documented by x-ray of the chest immediately after the implantation (figure [1](#F1){ref-type="fig"}). Following implantation, the catheter had been used for three months without any problems. Thereafter, the system was regularly flushed with heparin-saline solution. On presentation, injection into the port catheter was freely possible but caused instant reflectory coughing. Aspiration of blood via the port was not possible. Laboratory findings showed a mild leukocytosis of 11.0/nl and hypokalemia of 2.6 mmol/l. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Postoperative x-ray after catheter implantation. The catheter tip is placed correctly in the superior caval vein. ::: ![](1477-7819-3-19-1) ::: Further radiological diagnostics including injection of contrast medium into the catheter documented a dislocation of the catheter tip into the upper lobe of the right lung with paravasation into the bronchial system (figure [2](#F2){ref-type="fig"}). In addition, thoracic computerised tomographic (CT) scan showed a large pleural effusion in the right pleural cavity (figure [3](#F3){ref-type="fig"}). There was no evidence for mediastinal or intrapulmonal tumour growth or lymph node metastases at the site of perforation. The patient was referred to the department of thoracic surgery for further therapy. The intrapulmonal tip of the port catheter was cut off and extracted thoracoscopically. The remaining catheter retracted into the superior caval vein lumen. As the patient had undergone mammarial gland ablation on both sides with consecutive radiation there was a severe dermatitis with ulcerations at the implantation site of the port catheter. Due to the high-risk of infection and wound healing complications and the limited life-expectancy of the patient, it was decided to leave the injection chamber and catheter remnant in situwithout any further use for injections or infusions. The patient recovered from the intervention without complications. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Injection of contrast medium via the port catheter. Paravasal and intrabronchial drainage of the applied contrast. ::: ![](1477-7819-3-19-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### CT scan showing the catheter tip in the right upper lobe of the lung and a large dorsobasal pleural effusion. ::: ![](1477-7819-3-19-3) ::: Discussion ========== Totally implantable venous access devices are broadly used for application of chemotherapy or intravenous nutrition, especially in patients with poor peripheral vein conditions \[[@B1]-[@B4]\]. Early, mainly surgical complications can occur, such as bleeding, pneumothorax, nerve lesions or catheter misplacement. Wound and catheter infection (4--5%), thrombosis (3--3.5%), catheter fracture or disconnection (0.5%) and secondary dislocation (1.5--2%) of the catheter are the most important long-term complications \[[@B1],[@B2]\]. In the presented case, the catheter tip of the venous access device perforated the superior caval vein and dislocated into the upper lobe of the right lung. To our knowledge, this pulmonary complication has not been reported in the literature before. Common events of secondary dislocation include migration of the catheter tip into the internal jugular vein or the contralateral subclavian vein. A perforation of the caval vein and migration of the catheter into mediastinal structures or the pericardium has been reported \[[@B5]-[@B7]\]. The venous access device was implanted 2-years before and the correct placement of the catheter at the atrial-caval junction was documented by intra- and postoperative x-ray (figure [1](#F1){ref-type="fig"}). Moreover, the system worked properly since then, excluding a surgical problem in the presented case. The reason for the perforation remains obscure. CT scan showed no local inflammation, pathologic lymph nodes or tumour growth at the site of perforation. A spontaneous catheter perforation may be explained by a cranial dislocation of the catheter tip. A significant increase in catheter malfunctions has been reported by Petersen et al, \[[@B8]\] when catheters are placed primarily in a too high position in the caval vein. In the presented case this position might have resulted from a secondary dislocation. As the catheter was introduced via the left subclavian vein, its tip could get into a right angled position to the right lateral vessel wall of the caval vein. In this position a mechanical irritation leading to a chronic decubitus of the vessel wall with consequent perforation into the lung is possible. A similar event of an intrabronchial migration has been reported with a broken intraatrial pacing device \[[@B9]\]. However, these pacing catheter tips contain metallic material and therefore are more rigid than silicone port catheters, favouring a spontaneous perforation of these devices. Another explanation for a chronic damage of the vessel wall might be endothelial cytotoxicity of the applied chemotherapy itself. In recent studies such an effect has been observed, especially when vinorelbin or 5-fluorouracil had been administered via a central venous catheter \[[@B10]-[@B12]\] leading to injury of the right phrenic nerve by direct cytotoxic effects. The authors of these studies postulated a damage of the endothelial barrier by the chemotherapeutic agent. In the presented case, first-line chemotherapy had included paclitaxel and epirubicin. Additionally, treosulfane had been administered immediately before the perforation became evident. None of these agents has yet been reported to cause endothelial damage as mentioned above. Therefore, a direct cytotoxic effect to the vessel wall seems rather unlikely. Possible consecutive complications of the perforation itself include the risk of bleeding and air embolism as well as the paravasal application of fluids and especially aggressive chemotherapeutic agents via the dislocated catheter. There are case reports of accidental intrapericardial and intramediastinal applications of chemotherapeutic drugs as well as subcutaneous applications of chemotherapy due to wrong positioning of the puncture needle \[[@B5]-[@B7]\]. In most of the cases, patients did not suffer from adverse effects of these paravasats. In the presented case, treosulfane was administered intrabronchially, leading to coughing and dyspnoea, as well as a large pleural effusion which was absorbed without consecutive problems. Especially no interstitial pneumonia or evident tissue necrosis occurred. Guidelines on how to avoid accidental paravasal infusion during the long-term use of port catheters include puncture and spilling of the catheter with saline solution, which is broadly accepted as a safety test before using the catheter. Blood aspiration before injection can be performed in addition. However, many catheters shows a \"ventil\" mechanism following long-term use. Therefore, blood aspiration may not be possible, although the catheter can still be used for infusion. A standard chest x-ray may be used to discover catheter dislocation, but does not reveal functional problems. The gold standard for diagnosis of catheter dislocation and function is the radiographic visualization with contrast medium application via the catheter. This is certainly no routine procedure prior to each application of chemotherapeutic drugs as it is associated with x-ray exposure of the patient, the risk of contrast-medium related complications, requires radiological facilities and high costs. Therefore, the only recommendation is to puncture and use port catheters with the highest accuracy. Each new symptom, even when not obviously related to the catheter, e.g. coughing following infusion, should be carefully documented and evaluated by expert physicians to avoid severe catheter-associated complications. Conclusion ========== Intrapulmonal dislocation of the catheter tip is a rare complication of a totally implantable venous access device. However, it can cause severe complications and may be difficult to recognize due to unspecific symptoms. Authors\' contributions ======================= TH, CT review of literature and manuscript preparation AK, HD surgical management BS, MWB review of manuscript All authors have read and approved the manuscript in the presented form. Competing Interests =================== The author(s) declare that they have no competing interests. Acknowledgements ================ Patient\'s written consent was obtained for scientific use of anonymized personal data and publication of case report and images.	PubMed Central	2024-06-05T03:55:56.253267	2005-4-11	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087895/", "journal": "World J Surg Oncol. 2005 Apr 11; 3:19", "authors": [ { "first": "Thilo", "last": "Hackert" }, { "first": "Christin", "last": "Tjaden" }, { "first": "Angelika", "last": "Kraft" }, { "first": "Bernd", "last": "Sido" }, { "first": "Hendrik", "last": "Dienemann" }, { "first": "Markus W", "last": "Buchler" } ] }
PMC1087896	Background ========== Soft tissue sarcomas account for less than 1% of malignant neoplasms. Approximately 140 different histological types of sarcomas have been described \[[@B1]\]. Sarcomas arise in tissues of mesenchymal or ectodermal origin and may thus occur anywhere in the body The surgical treatment of choice is the wide surgical excision of the tumor in combination with radiotherapy \[[@B2],[@B3]\]. Currently, about 30--50% of all patients die within 5 years of primary diagnosis. This overall poor prognosis is mainly due to metastatic disease at the time of diagnosis. Approximately 40% of patients with high-grade sarcomas develop pulmonary metastases in spite of local tumor control \[[@B4]\]. The median duration of survival with apparent metastases is presently 8--12 months \[[@B2],[@B4]\]. Soft tissue sarcomas are notoriously resistant against most chemotherapeutic agents \[[@B5]\]. As first line treatment, only doxorubicin (adriamycin), epirubicin and ifosfamide have achieved a single-agent activity of more than 15% resulting in response rates of 18%-29% \[[@B6],[@B7]\]. The combination of anthracyclines and ifosfamide resulted in response rates of 40%-50%, with 10% of patients in complete remission \[[@B8]\]. ite the reported association between remission and prolonged survival, an overall superiority of combination chemotherapy over the administration of single-agent doxorubicin has not been established yet. Ifosfamide can play a major role as a second-line treatment with response rates of up to 30%-50% following the failure of anthracylin \[[@B2],[@B7]\]. Despite the disparate appearance and histology of sarcomas, only few drugs and combinations have been used as treatment regimes \[[@B5],[@B9]\]. Recently, new strategies have been pursued to improve treatment options for well defined subgroups of sarcomas. One example is the successful use of imatinib hydrolasis, a specific tyrosine kinase inhibitor, in gastrointestinal stromal tumors \[[@B10],[@B11]\]. Several new drugs such as exatecan, TZT 1027, trofosfamid and topotecan are under investigation \[[@B12]-[@B15]\]. Another promising agent for the treatment of soft tissue sarcomas is ET-743. Several clinical trials have shown response rates of up to 20% in untreated and 10% in pretreated sarcoma patients. Approximately 50% of patients in these series have shown long-lasting stabilization of disease \[[@B4],[@B9],[@B16],[@B17]\]. In the present study we used a testing system, the ATP-based tumor sensitivity assay (ATP-TCA) (Fig. [1](#F1){ref-type="fig"}) The successful use of ATP-TCA has been demonstrated for pretesting of chemosensitivity in melanoma, breast and ovarian cancer \[[@B18]-[@B20]\] as well as other tumors \[[@B21]-[@B26]\]. Based on these results, we performed in vitro chemosensitivity testing with the ATP-TCA in patients with soft tissue sarcoma. The aim of this study was to evaluate the feasibility of ATP-TCA in soft tissue sarcomas and to gain information about their chemosensitivity profiles with regard to subtype, grade of differentiation and tumor recurrence. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Principle of chemosensitivity testing using the ATP luminescence method (ATP-TCA). TDC, test drug concentration ::: ![](1477-7819-3-20-1) ::: Patients and Methods -------------------- After informed consent was obtained, the tumors were resected in 50 patients with diagnosed sarcomas of the extremities. The group of tumors included liposarcomas (n = 17), malignant fibrous histiocytomas (MFH/NOS; n = 16), extraskeletal chondrosarcomas (n = 8), rhabdomyosarcomas (n = 5) and malignant peripheral nerve sheet tumors (MPNST;n = 4). Clinical staging of the patients was performed in accordance with the criteria of the American Joint Committee on Cancer. The mean age of the patients was 63 years, ranging from 32--76 years. The sex distribution was 20 females and 30 males. The sarcomas were primary tumors in 21 patients and recurrencies in 29 patients. None of the patients had received adjuvant chemotherapy, neither before nor after surgery. All patients with recurrent tumors had radiotherapy following the excision of the tumor. All patients underwent wide, complete resection of the tumor. Histological grading showed GII in 17 tumors and GIII in 33 tumors (Tab. [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Basic data of patients with soft tissue sarcomas (n = 50) ::: n age (mean-range) primary tumors recurrent tumors grade G2 grade G3 ------------------------------ ---- ------------------ ---------------- ------------------ ---------- ---------- Liposarcoma 17 61 (32--73) 9 8 6 11 NOS-Sarcoma/MFH 16 64 (42--69) 7 9 7 9 Extraskeletal Chondrosarcoma 8 65 (62--72) 1 7 2 6 Rhabdomyosarcoma 5 68 (64--76) 4 1 1 4 MPNST 4 59 (55--67) 0 4 1 3 ::: Tumor tissue was sampled under sterile conditions by a histopathologist and stored in cell culture medium (DMEM, Sigma™, Germany) with antibiotics (100 U/ml penicillin and 100 μg/ml streptomycin) at 4°C. In vitro chemosensitivity testing was performed with a ATP-based assay (ATP-TCA, DCS Innovative Diagnostic Systems^®^, Hamburg, Germany) \[[@B27]\]. At least 1 cm^3^of tumor tissue was minced and dissociated enzymatically. The single-cell suspension was depleted of red blood cells and cellular debris through Ficoll-Hypaque density gradient centrifugation. After assessing the viability of the cells, a cell count was conducted. The cells were then incubated in polypropylene round-bottom 96-well plates containing 2 × 10^4^cells per well. Different cytotoxic agents in different concentrations were added to the wells, whereas some wells received no supplements and served as negative controls. Applied chemotherapeutics were doxorubicin, epirubicin, ifosfamide, dacarbazine (DTIC), actinomycin D, cisplatin and vincristine at six different dilutions (6.25--200%) of the test drug concentrations (TDC) shown in Tab. [2](#T2){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Cytotoxic drugs used in the ATP-TCA ::: drug 100% TDC\* (μg/ml) -------------------------- -------------------- doxorubicin (adriamycin) 0,5 epirubicin 0,5 ifosfamide (mafosfamide) 3,0 dacarbazine (DTIC) 10,0 actinomycin D 0,1 cisplatin 3 vincristine 0,4 \TDC, test drug concentration ::: While the concentration of a chemotherapeutic agent is typically measured in plasma or serum with defined precision, sensitivity, and specificity, it is possible to clinically and analytically validate the data to be used as test drug concentration in other biological samples or in cell cultures <http://www.aacc.org/divisions/tdm/default.stm>. The individual TDC for each cytotoxic drug was determined previously (Andreotti et al. 1995) by reference to known pharmacokinetic and response data. After 7 days of incubation at 37°C, 5% CO~2~and 100% humidity, the cells were lysed and the ATP content of each well was measured by a luciferin-luciferase-based luminescence assay with a Dynatech ML1000 luminometer (Fig. [1](#F1){ref-type="fig"}). Individual sensitivity indices for each test drug or drug combinations were calculated from the obtained data curves by summing up the percentages of tumor inhibition for each drug concentration tested (6.25%, 12.5%, 25%, 50%, 100%, 200%) followed by the subtraction of 600. A sensitivity index of 600 indicates unrestrained tumor cell growth and minimal drug sensitivity, whereas a sensitivity index of 0 reflects a complete tumor inhibition and maximal drug sensitivity. Moreover, the chemosensitivity of the test samples was classified according to the tumor inhibition at different TDC. The chemotherapeutic single agents and combinations were arranged in groups starting from high sensitivity, over moderate sensitivity and low sensitivity, to resistant. The grouping was performed according to the threshold values shown in Tab. [3](#T3){ref-type="table"}. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Classification of ATP-TCA chemosensitivity testing results ::: Tumor growth inhibition at 200% TDC\ Tumor growth inhibition at 25% TDC\* ---------------------- --------------------------------------- -------------------------------------- High sensitivity \>95% \>70% Moderate sensitivity \>95% 50--70% Low sensitivity \>95% \<50% \<95% \>50% Resistant \<95% \<50% \TDC, test drug concentration ::: Statistical analysis -------------------- The statistical significance of differences between sub entities of soft tissue sarcomas, tumor grading and the comparison between primary and recurrent tumors was calculated by the Mann-Whitney-U-test and the Chi squared test. P\< 0.05 was considered statistically significant. Results ======= ATP-TCA testing was done on 53 tumor samples from 53 patients. 50 (94%) of these tumor samples were included in the study. 3 samples had to be excluded due to bacterial infection (n = 2) or mycosis (n = 1). The sensitivity profiles of the sarcomas exhibited a significant heterogeneity in response to the array of different cytotoxic agents (Tab. [4](#T4){ref-type="table"}) ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Chemosensitivity testing in 50 soft tissue sarcoma patients ::: Cytotoxics Resistant Low Sensitivity Moderate Sensitivity High Sensitivity Mean Sensitivity Index\ ---------------------------- ------------- ----------------- ---------------------- ------------------ -------------------------- Actinomycin D 00/50 (00%) 10/50 (20%) 03/50 (06%) 37/50 (74%) 137 Doxorubicin (Adriamycin) 03/50 (06%) 02/50 (04%) 10/50 (20%) 35/50 (70%) 140 Ifosfamide 10/50 (32%) 00/50 (00%) 02/50 (04%) 32/50 (64%) 234 Epirubicin 08/50 (16%) 00/50 (00%) 12/50 (24%) 30/50 (60%) 245 Cisplatin 39/50 (82%) 11/50 (22%) 00/50 (00%) 00/50 (00%) 444 Dacarbazine 45/50 (95%) 05/50 (5%) 00/50 (00%) 00/50 (00%) 532 Vincristine 42/50 (84%) 08/50 (16%) 00/50 (00%) 00/50 (00%) 513 Doxorucicin + Ifosfamide 03/50 (06%) 00/50 (00%) 11/50 (22%) 36/50 (72%) 139 Actinomycin D + Ifosfamide 01/50 (2%) 08/50 (16%) 03/50 (06%) 38/50 (76%) 163 VAC\\ 03/50 (06%) 05/50 (10%) 07/50 (14%) 35/50 (70%) 218 CYVADIC\\\* 02/50 (04%) 00/50 (00%) 15/50 (30%) 33/50 (66%) 255 \Data are means from individual chemosensitivity indices for each test drug or drug combination, which were calculated by summing up the tumor growth inhibition for each drug concentration tested followed by the subtraction of 600. A sensitivity index of 600 indicates unrestrained tumor cell growth and minimal drug sensitivity, whereas a sensitivity index of 0 reflects complete tumor growth inhibition and maximal drug sensitivity. \\VAC: Vincristine, Actinomycin D, Cyclophosphamide \\\CYVADIC: Cyclophosphamide, Vincristine, Doxorubicin, Dacarbazine ::: The strongest chemoresistance was demonstrated for the single use of dacarbazine with 95% of the samples being resistant. The application of vincristine and cisplatin resulted in rates of resistance of 84% and 82%, respectively. The highest response rate was found for actinomycin D, with 74% of the samples showing a high sensitivity without any resistances. These positive results were followed by doxorubicin and ifosfamide with a highly sensitive response in 70% and 64% and rates of resistance in 6% and 12%, respectively. In contrast, epirubicin had a markedly lower effect on inhibition, with high sensitivity in 60% and resistance in 16% of cases (Tab. [4](#T4){ref-type="table"}). The tumor inhibition of the corresponding concentration of the 7 single agent chemotherapeutics is shown in Fig. [2](#F2){ref-type="fig"}. It reveals the superior efficiency of actinomycin D, doxorubicin, dacarbazin and ifosfamide, compared to cisplatin, vincristine and dacarbazin (Fig. [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Original data curves obtained with ATP-TCA in 50 soft tissue sarcomas. Each plot shows the corresponding test results with seven different cytotoxics as single agents. ::: ![](1477-7819-3-20-2) ::: In addition to single drug testing, four different combinations of agents were submitted to ATP-TCA testing. Chemosensitivity was less pronounced for CYVADIC (cyclophosphamide, vincristine, doxorubicin and dacabazine) with 66% of the samples showing high sensitivity versus 4% of resistant samples. The VAC combination (vincristine, actinomycin D and cyclophosphamide) resulted in 70% of high sensitivity and 6% resistance. Doxorubicin and ifosfamide in combination yielded highly sensitive responses in 72% and 6% of resistant cases. The combined use of actinomycin D and ifosfamide resulted in the overall strongest chemosensitivity of all combinations used with 76% of high sensitivity and 2% of resistance (Tab. [4](#T4){ref-type="table"}). The tumor inhibition of the tested chemotherapeutic agent combinations is demonstrated in correlation to the agent\'s concentrations in Fig. [4](#F4){ref-type="fig"}. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Effect of Actinomycin D as single agent measured by ATP-TCA in different types of soft tissue sarcomas (Liposarcomas:n = 17, NOS-Sarcomas/MFH:n = 16, Chondrosarcomas:n = 8, Rhabdomyosarcomas:n = 5, MPNST:n = 4) (p \< 0.05). ::: ![](1477-7819-3-20-4) ::: The cytotoxic treatment with actinomycin D resulted in significantly divergent profiles of chemosensitivity depending on the histological type of soft tissue sarcoma (Fig. [4](#F4){ref-type="fig"}). Rhabdomyosarcomas were especially chemoresistant against actinomycin D (p \< 0.05). This is in contrast to the response pattern of other types of sarcomas to this particular chemotherapy. There was no difference in chemosensitivity between rhabdomyosarcomas and other tumor types for the remaining cytotoxic drugs used. Complementary, doxorubicin revealed significantly lower cytotoxic effect on extraskeletal chondrosarcomas compared to all the other tumors (p \< 0.05) (Fig. [5](#F5){ref-type="fig"}). The remaining chemotherapeutic agents showed no difference in chemosensitivity between extraskeletal chondrosarcomas and the other tumor sub entities tested. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Effect of Doxorubicin (Adriamycin) as single agent measured by ATP-TCA in different types of soft tissue sarcomas (Liposarcomas:n = 17, NOS-Sarcomas/MFH:n = 16, Chondrosarcomas:n = 8, Rhabdomyosarcomas:n = 5, MPNST:n = 4) (p \< 0.05). ::: ![](1477-7819-3-20-5) ::: There also was a strong statistical correlation between the histological grading and the chemosensitivity. GIII tumors (n = 33) demonstrated a significantly higher chemosensitivity for actinomycin D, doxorubicin and ifosfamide than GII tumors did (p \< 0.05) (n = 17) (Fig. [6](#F6){ref-type="fig"}). ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Chemosensitivity profiles of 50 soft tissue sarcomas divided by tumor grade: G2 (n = 17), G3 (n = 33). Results shown for Actinomycin D, Doxorubicin and Ifosfamide (p \< 0.05). ::: ![](1477-7819-3-20-6) ::: Similar results were seen when chemosensitivity of primary tumors was compared to sensitivity of recurrent tumors. We found the recurrent tumors to be more chemosensitive to actinomycin D, doxorubicin and ifosfamide than the primary ones (p \< 0.05) (Fig. [7](#F7){ref-type="fig"}). ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Chemosensitivity profiles of 50 soft tissue sarcomas divided by primary- (n = 21) and recurrent- (n = 29) tumors. Results shown for Actinomycin D, Doxorubicin and Ifosfamide (p \< 0.05). ::: ![](1477-7819-3-20-7) ::: Discussion ========== In view of the limited number of effective chemotherapeutic drugs with response rates of less than 20%, the need for more detailed information about the tumor specific chemoresistance seems obvious \[[@B4],[@B5],[@B28]\]. A testing system would have to produce repeatable and comparable results of the chemosensitivity of various tumor types. Chemosensitivity testing with ATP-TCA proved a success in acquiring resistance profiles for different chemotherapeutic agents in metastasizing carcinoma of the breast \[[@B21]\] and recurrent tumors of the ovaries \[[@B24]\]. In these tumors, the intra- and inter-assay variability was less than 15%, which demonstrated a high sensitivity and linearity \[[@B29]\]. In the present study, the ATP-TCA test was practicable in 94% of all cases and thus represented a feasible method in soft tissue sarcomas. Only a limited number of cytotoxic drugs, such as doxorubicin, epirubicin and ifosfamide, have been reported to produce response rates of more than 15% in the treatment of sarcomas, either as single agents or as part of drug combinations. There are no clinical trials providing evidence of prolonged survival rates after either adjuvant or neoadjuvant chemotherapy \[[@B30]-[@B32]\]. Currently, the EORTC is not conducting any clinical trials on the efficacy of chemotherapy in soft tissue sarcomas due to the lack of a prospected benefit. Chemotherapy is presently used in metastasized disease only which may account for the low acceptance of this therapeutic option in adult soft tissue sarcoma \[[@B30],[@B33]-[@B36]\]. Further limiting factors regarding the feasibility of large-scale trials are the low incidence of sarcomas and the enormous heterogeneity of the types of tumors. At the moment, pathologists know more than 140 entities that may explain some problems of classification. The application of chemotherapy is thus not differentiated according to the histological type of tumor, yet. Doxorubicin has been reported to achieve response rates of 20% to 26% and represents the most effective chemotherapeutic agent among the available first line single drug therapies \[[@B37]\]. In our study, the ATP-TCA testing of doxorubicin showed that 70% of tumors reacted highly sensitive. However, among all cytotoxic agents tested, doxorubicin proved to be only a second rate treatment regimen. The overall highest sensitivity of all single therapeutic agents was demonstrated for actinomycin D. This is in contrast to the poor clinical response rates of approximately 17% rendered by this drug \[[@B15],[@B38]\]. Especially actinomycin D is in common use as single agent and combination chemotherapy for treatment of rhabdomyosarcomas in childhood and hyperthermia/isolated limb perfusion \[[@B15],[@B39]-[@B41]\]. Ifosfamide serves as a prime drug in first line treatments under clinical conditions and as a high-dose cytotoxic agent in second line applications with response rates of more than 25% \[[@B7],[@B42]-[@B45]\]. Yet, the ATP-TCA chemosensitivity testing found it to be significantly inferior to actinomycin D and doxorubicin. The ATP-TCA confirmed strong chemoresistance of soft tissue sarcomas, known from several clinical trials for vincristine, cisplatin and dacarbazine as single agents \[[@B2],[@B32],[@B42]\]. Similar results were found for combined chemotherapies. The CYVADIC combination of Gottlieb has been in use for over a decade \[[@B46]\], however, the ATP-TCA testing showed significantly higher chemosensitivities with the combination of actinomycin D and ifosfamide (Fig. [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Original data curves obtained with ATP-TCA in 50 soft tissue sarcomas. Each plot shows the corresponding test results with four different drug combinations. ::: ![](1477-7819-3-20-3) ::: The analysis of respective histological entities showed, that specific types of tumors respond in different ways. Rhabdomyosarcomas were significantly more resistant against actinomycin D than any of the other tumors tested (Fig. [4](#F4){ref-type="fig"}). Another example of sub-entity-dependant resistance was found in the chemosensitivity of extraskeletal chondrosarcomas on doxorubicin. The ATP-TCA testing showed a considerable chemoresistance of these tumors against doxorubicin, when compared to other types of soft tissue sarcomas (Fig. [5](#F5){ref-type="fig"}). Different test results were also found depending on grading and whether the sarcoma presented as a primary or a recurrent tumor. The results outlined in Fig. [7](#F7){ref-type="fig"} show that the chemoresistance of primary tumors was considerably higher than the resistance of recurrent tumors. These results correspond to the regularly higher differentiation and the lower turn over time of primary tumors compared to recurrent tumors. The superior chemosensitivity of recurrent tumors is in sharp contrast to the worse prognosis of recurrencies \[[@B47]-[@B51]\]. GII tumors (n = 17) exhibited a significantly higher chemoresistance when compared to GIII tumors (n = 33) (Fig. [6](#F6){ref-type="fig"}). This could be due to the higher mitotic activity of tumor cells in the GIII group. The test results stress the importance of the state of cellular differentiation as a crucial parameter for the responsiveness of soft tissue sarcomas to cytotoxic therapies \[[@B32],[@B36]\]. The results indicate that protocols for chemotherapy should be designed based on tumor specificity, sub entity specificity and according to the tumors grading to improve the response rates of soft tissue sarcomas. Conclusion ========== The pre-therapeutic chemosensitivity testing of cytotoxics with ATP-TCA offers the opportunity of improving the hitherto moderate success rates of conventional chemotherapy in soft tissue sarcomas. Sarcomas usually are of sufficient size to gain enough tissue for the testing procedure, a problem often encountered with small carcinomas of the breast. The test poses no risk or disadvantage for the patient and should be performed in cooperation with the oncologist. To create an individualized chemotherapy protocol, evaluation of ATP-TCA chemosensitivity testing correlated with clinical data and in-vitro results are required. Due to the limited course of time of this study, the results were not yet associated with data of clinical responses in long-term follow-up observations. Future trials on any positive correlations of this kind should prove the validity of the testing method. Authors\' contributions ======================= ML: coordinated the work, developed the study design and prepared the manuscript TM: prepared the manuscript and figures carried out statistical analyses, CK: carried out histopathology of the tumors DB: carried out ATP-TCA-Assays HUS: developed the idea, have given final approval of the version to be published HJ: carried out ATP-TCA-Assays LS: carried out ATP-TCA-Assays OM: conceived the study, have given substantial contribution to conception and design HHH: conceived the work and participated in study design, gave final approval Competing interests =================== The author(s) declare that they have no competing interests. Acknowledgements ================ This work was supported by FoRUM-grant, Ruhr-University-Bochum, Germany	PubMed Central	2024-06-05T03:55:56.254367	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087896/", "journal": "World J Surg Oncol. 2005 Apr 18; 3:20", "authors": [ { "first": "Marcus", "last": "Lehnhardt" }, { "first": "Thomas", "last": "Muehlberger" }, { "first": "Cornelius", "last": "Kuhnen" }, { "first": "Daniel", "last": "Brett" }, { "first": "Hans U", "last": "Steinau" }, { "first": "Hamid Joneidi", "last": "Jafari" }, { "first": "Lars", "last": "Steinstraesser" }, { "first": "Oliver", "last": "Müller" }, { "first": "Heinz H", "last": "Homann" } ] }
PMC1087897	The most important prognostic factor in the management of head and neck cancer is the presence of cervical nodal metastasis. Once the tumor involves neck nodes, survival drops by almost 50%. Management of cervical metastasis has gone through an evolution since the beginning of the last century. The classic radical neck dissection, where all the neck nodes are removed along with 3 important structures -- sternocleidomastoid muscle, internal jugular vein and accessory nerve -- was popularized in the landmark article by George Crile. It subsequently became the standard of care in the management of neck nodes for almost 75 years. The major complication from radical neck dissection was very apparent to most clinicians; shoulder dysfunction, which led to modifications in neck dissection techniques. Oswaldo Suarez gets the credit for popularizing functional neck dissection wherein the accessory nerve is carefully preserved to the extent tumor involvement allows. He also popularized the facial envelope and oncologic safety of modified neck dissection, which gained acceptance with the teachings of Itore Bocca, Javier Gavilan, and Richard Jesse. In the 1980\'s there was a tremendous switch from radical neck dissection to modified neck dissection to maintain patient quality of life and preserve shoulder function. At the same time, the patterns of nodal metastasis were studied in detail. Publications from Lindberg and Shah described the location of metastatic disease in the neck depending upon the primary site, and justified modifications in neck dissection \[[@B1]\]. By the mid 1980\'s there was confusion and disagreement about the nomenclature of various modifications, and every institution had their own modification. In an effort to standardize the nomenclature, the American Academy of Otolaryngology-Head and Neck Surgery developed a systematic approach to neck dissection, dividing it into comprehensive and selective neck dissections \[[@B2],[@B3]\]. In the recent modification of the standardization of neck dissection, the Committee on Head and Neck Oncology divided Levels I, II and V into A&B groups \[[@B3]\]. This division appears to be anatomically sound in relation to metastatic disease to the neck. For example, Level IA is rarely involved in metastatic disease. At the same time, Level IIB nodes are rarely involved in metastatic tumor unless there is a bulky metastatic disease at Level II. One may consider super-selective neck dissections. There appears to be a trend in deleting specific names for the modification of neck dissection and use selective neck dissections with the structures and group of lymph nodes removed as specific types. The comprehensive neck dissection removes all lymph nodes in the neck and its modification includes preservation of the sternocleidomastoid muscle, accessory nerve or jugular vein. The selective neck dissection addresses a select group of lymph nodes based on the location of the highest incidence of metastatic disease, thus supraomohyoid neck dissection became very popular as a staging procedure for cancer of the oral cavity \[[@B4]\]. Interest developed in understanding the prognostic factors of metastatic disease in the neck, such as tumor size, location, and extranodal spread. Postoperative radiation therapy in patients with cervical neck node metastasis became the standard practice in the early 1990\'s. The high incidence of metastatic disease to neck nodes from cancer of the oral tongue was noted at the same time. Supraomohyoid neck dissection became standard practice in patients with cancer of the oral cavity and N0 neck. Clinical evaluation was supplemented with imaging studies, such as computerized tomography (CT) and magnetic resonance imaging (MRI) scanning. Van den Brekel from Amsterdam popularized the role of ultrasound and ultrasound-guided needle biopsy \[[@B5]\]. Even though the researchers from Amsterdam reported excellent correlation, ultrasound did not become very popular in day-to-day clinical surgical practice. It does play an important role in the initial evaluation of cervical metastasis and in patient follow-up, particularly in those who have received chemo-radiation therapy. Ultrasound is an easy outpatient test which can be performed at frequent intervals during follow-up. In the mid 1990\'s an interest developed in using PET scanning to diagnose neck metastasis. Even though PET scanning can be a useful tool, particularly FDG uptake evaluation, it has not helped to make clinically definitive decisions as to the presence or absence of metastatic disease. Small volume disease, especially in the N0 neck, is difficult to image with a PET scan. PET scanning appears to be an important investigative tool in patients who are being followed after nonsurgical treatment, such as chemo-radiation therapy. Further studies are necessary, however, to standardize the SUV (Standard Uptake Value) in the PET scan. Postoperative radiation therapy was routinely recommended in patients with large nodal metastasis and extranodal spread. Extranodal spread was considered to indicate grave prognosis in patients with cervical metastasis. These patients had a high incidence of local recurrence and distant metastasis. With this in mind, the EORTC and RTOG conducted randomized prospective trials of the use of postoperative chemo-radiation therapy in patients with cervical nodal metastasis. It is interesting that the two groups reported their results in the same issue of the New England Journal of Medicine. Cooper, et al, \[[@B6]\] from the Radiation Therapy Oncology Group reported results from a randomized prospective trial in the New England Journal of Medicine. There were 231 patients randomly assigned to receive postoperative radiation therapy alone, and 228 patients to receive identical treatment plus concurrent chemotherapy with cisplatinum 100 mg per m^2^on days 1, 22 and 43. They reported the estimated 2 year rate of local and regional control as 82% in the combined therapy group, as compared with 72% in the radiotherapy alone group. Disease-free survival was significantly longer in the combined therapy group than in the radiotherapy group; however, interestingly the overall survival was not altered by the addition of chemotherapy. The incidence of acute adverse effects of Grade III or greater was reported in 34% of the radiotherapy group and 77% in the combined therapy group. Four patients who received combined therapy died as a direct result of treatment. The authors concluded that among high risk patients in the postoperative setting, concurrent chemo-radiation therapy significantly improves the rates of local and regional control and disease-free survival. However, the combined treatment is associated with a substantial increase in adverse effects. In the same issue of the New England Journal of Medicine, Bernier et al, \[[@B7]\] reported on the European Organization for Research and Treatment Cancer Trial 22931. They randomly assigned 167 patients in each group to receive postoperative radiation therapy or radiation and chemotherapy. They also used 100 mg cisplatinum per m^2^on days 1, 22 and 43 of the radiotherapy regimen. They reported a 5 year progression-free survival of 47% compared to 36% with radiation therapy alone. They reported an overall survival rate of 53% in patients who received chemo-radiation therapy compared to 40% with radiation therapy alone. The estimated 5-year cumulative incidence of local or regional relapses was 31% after radiotherapy and 18% after combined therapy. This group also reported severe adverse effects from combined therapy (41%) and radiotherapy alone (21%). Even though these two studies are prospective randomized trials, further confirmation needs to be obtained through continued interest in such trials including reduction of adverse effects. The most important complication of combined chemo-radiation therapy that has been noted in organ preservation protocols is severe mucositis and pharyngeal stricture. Pharyngeal stricture is a disastrous complication from this treatment and has a major impact on the quality of life of the patients. It appears that in patients with poor prognostic factors with cervical metastasis there is an increasing interest in treatment with chemo-radiation therapy, rather than radiation alone. Obviously one needs to keep in mind the complications related to chemo-radiation therapy. Such complications are well recognized in patients who undergo an organ preservation protocol for laryngopharyngeal tumors or for oropharyngeal cancers. There is a high incidence of neutropenia, Grade 4 mucositis, and pharyngeal stricture. Development of pharyngeal stricture continues to be a difficult problem to manage in clinical practice and leads to discussion regarding the quality of life. Long term dependency on gastrostomy is extremely frustrating to patients who may have been cured of their neck disease. Since organ preservation and chemo-radiation therapy has become the most prevalent treatment for patients with oropharyngeal and laryngopharyngeal cancers, it has generated controversy about the management of patients with nodal metastasis and their follow-up. The general consensus of opinion is that N1 disease can be easily controlled with chemo-radiation therapy. The problem comes with patients who present with N2 and N3 neck disease. Even though approximately 50% of the patients can be cured with chemo-radiation therapy, the remaining patients may persist with microscopic nodal metastasis. Even though there is no unanimous consensus today about how to manage N2--N3 neck after chemo-radiation therapy, there appears to be a general trend to consider close follow-up of these patients with clinical exam, CT, MRI and PET scan. If there is a residual thickening or presence of nodal disease, neck dissection is routinely recommended. Approximately 40--50% of patients may have viable tumor, which also depends upon the location of the primary tumor. The patients who recur in their neck nodes after previous surgery and radiation therapy are clearly a major challenge to the head and neck surgeon. Every attempt is made to resect the tumor if that is possible, along with (in select patients) additional local radiation therapy or brachytherapy. If brachy catheters are to be used, the carotid artery needs additional protection, preferably with myocutaneous flaps. Carotid resection is considered in very select circumstances where the patient has satisfactory carotid blood flow from the opposite side, and appropriate reconstruction is considered with carotid replacement by either a gortex graft or saphenous vein. There is a high incidence of neurologic complications when the carotid artery is resected. One needs to keep in mind that the carotid artery may not be the only limiting factor in such patients with metastatic disease to the neck. The surrounding structures, such as the vagus nerve, sympathetic trunk, and scalene muscles are also directly involved by the tumor. An appropriate and satisfactory surgical resection should be undertaken only if possible. With increasing interest in sentinel node biopsy in melanoma and breast cancer, some investigators have extended this technology to the squamous carcinoma of the upper aerodigestive tract, especially cancer of the oral cavity \[[@B8]\]. The American College of Surgeons Oncology Group has launched a prospective study of sentinel node biopsy in tumors of the oral cavity. At this stage, the sentinel node biopsy should be used only as an investigational tool. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= ARS: Conceptualized the editorial prepared the draft and edited it.	PubMed Central	2024-06-05T03:55:56.256335	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1087897/", "journal": "World J Surg Oncol. 2005 Apr 18; 3:22", "authors": [ { "first": "Ashok R", "last": "Shaha" } ] }
PMC1088007	Background ========== The QTc interval is a heart rate corrected value that measures the time between the onset and the end of electrical ventricular activity. Prolongation of this interval is considered a marker of the arrhythmogenic potential of a drug specifically linked to an increased risk of torsade de pointes ventricular tachycardia \[[@B1]\]. According to a document presented by the Committee for Proprietary Medicinal Products (CPMP) in 1997, normal subjects can be divided into three groups based on QTc interval length. For males, QTc values less than 430 ms are normal, between 431 and 450 ms are borderline and over 450 ms are prolonged. Whereas for females QTc values less than 450 ms are normal, between 451 and 470 ms are borderline and over 470 ms are prolonged \[[@B2]\]. This sex difference appears to be androgen driven and not determined by female hormones: at birth, QTc interval measurements are the same for male and female infants. At puberty, the male QTc interval shortens and remains shorter than its female counterpart by about 20 ms until ages 50 to 55 years, coincident with a decline in testosterone levels moreover, baseline QTc interval duration doesn\'t show significant fluctuations during the menstrual cycle and Hormone Replacement Therapy in postmenopausal age doesn\'t affect QTc interval \[[@B3]\]. In the above mentioned CPMP document it was also suggested that individual changes of QTc length of between 30 and 60 ms from baseline raises concern for the potential risk of drug induced arrhythmias \[[@B2]\]. Antipsychotics such as thioridazine, ziprasidone, quetiapine, risperidone, olanzapine or haloperidol have been suggested to prolong QTc interval \[[@B4]-[@B7]\]. Some authors reported that antidepressant drugs, including Selective Serotonine Reuptake Inhibitors (SSRI) (fluvoxamine, paroxetine and sertraline), Tricyclic Antidepressants (TCA) (amytriptiline, clomipramine, imipramine), and lithium can also prolong QTc interval \[[@B8]-[@B10]\]. Almost all drugs causing significant QT prolongation are known to interact with repolarizing potassium channels, particularly with the rapid component of delayed rectifier potassium currents (I~kr~), encoded by the human Ether-a-go-go relatedgene (HERG) \[[@B11]\]. However, TCA agents may affect the QTc interval primarily by their effect on sodium channels during depolarization \[[@B12]\]. Nonetheless TCA can also affect HERG potassium channels \[[@B13]\]. Drug trapping and structure-function studies suggest that the inner cavity of HERG channels is larger than other voltage-gated potassium channels and is therefore able to accommodate diverse chemical structures \[[@B14]\]. Among those drugs there are also SSRI like fluvoxamine \[[@B15]\], citalopram \[[@B16]\] and fluoxetine \[[@B17]\]. The combination of antipsychotic and antidepressant agents seems to have addictive effects on QTc interval \[[@B18]\]. We investigated the effects of polypharmacy on QTc in two groups of psychiatric female inpatients. Our null hypothesis was that patients treated with antipsychotics plus antidepressants or lithium would not have a greater QTc prolongation, if any, than patients treated with antipsychotics alone. Methods ======= A prospective naturalistic observational study was conducted in the Department of Psychiatry of San Matteo Hospital in Pavia. The study was approved by local Ethics Review Committee. We chosed to recruit only women because of their higher risk of developing drug related arrythmias. Consecutive female inpatients admitted from August 2003 to April 2004, with schizophrenia, bipolar disorder or schizoaffective disorder, were considered eligible for the study. Diagnoses were made by two staff psychiatrists (one attending and one resident psychiatrist), after reaching a clinical consensus in accordance to the DSM IV. Pharmacological and medical history were obtained. Included patients had to be free from psychiatric medications for at least 48 hours. Patients who were taking fluoxetine untill three days before recruitment were excluded, because of the long half life of this drug (3--5 days). Also patients treated with depot preparations were excluded. Non psychoactive drugs, like cardiovascular drugs, were allowed only if they were not reported to alter QT interval. Patients with disturbances of cardiac rate and rhythm, history of prolonged QTc, family history of sudden death, QTc interval greater than 470 ms in the ECG performed at admission, alterations of hepatic or renal function and substance abusers patients were also excluded. For each subject, therapy was started according to the clinical evaluation of psychiatrist in charge of the patient. The first group (Group 1) included women who were treated with only an antipsychotic (concomitant benzotropine treatment was permitted, as also zolpidem for insomnia was), the second group (Group 2) was composed of female patients who started treatment with antipsychotics in association with either an antidepressant or lithium. The dosage equivalent of haloperidol was calculated \[[@B19]\]. Two ECGs were obtained for each patient, the first before the beginning of treatment and the second after four days of treatment with the patients on the full therapeutic daily dose of antipsychotic prescribed by the clinician, generally after one week from the beginning of the treatment, except for the patients treated with clozapine, who had the second ECG four days after the end of titration (generally after two weeks of therapy). On the same day of the first ECG, blood samples were drawn for the evaluation of potassium serum levels from all the participants. When the second ECG was administered, blood was drawn to obtain potassium and additionally magnesium serum levels as well as serum levels of the antipsychotic agent. Samples for plasmatic levels determination were drawn before the first drug dose in the morning. Serum antipsychotic levels were analyzed by high-performance liquid chromatography with ultraviolet detection. The ECGs were obtained by standard 3-leads resting ECG procedure in the supine position and analyzed by a resident cardiologist (A. V.) who was blind to the patient\'s condition, study hypothesis, treatment status, serum levels of antipsychotics and was not involved in patient care. The QTc interval was calculated with the Bazzett formula. The QT interval was assessed in both DII and V2 leads. It was decided to focus on DII leads measurements due to the higher variability of measurements in precordial leads. Statistical Analysis -------------------- A repeated measures analysis of variance was used to test the effect of treatments on QTc (within subject factor: time of ECG examination, between subject factor: therapy group). Fisher Exact Test was used to compared the number of patients who exceed the threshold of borderline QTc values in Group 1 and Group 2. Indipendent T-Sample Tests were used to test the differences between baseline QTc values, ages and duration of illness using the statistical package Stata 7.0 (Stata Coorporation, 2001). Results ======= Patients\' characteristics -------------------------- Our sample consisted of thirty eight women. Ninenteen were included in the Group 1 and ninenteen in the Group 2. Age and duration of illness were comparable between the two groups (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Diagnosis, duration of illness, psychoactive treatment before recruitment and age of patients ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Diagnosis Comorbid Disorders (N) Duration of illness (yr) Psychoactive treatment 48 hours before recruitment (N) Age (yr) Mean ± SD Age (yr) Range --------------------------------- -------------- ------------------------ -------------------------- -------------------------------------------------------- -------------------- ---------------- Group 1 Patients in monotherapy Schiz 17\ 0 5,8 ± 5 Haloperidol 3\ 45,7 ± 15 22--77 Schizoaf 2 Risperidone 2 Group 2 Patients in politherapy Schizoaf 10\ An Nervosa 2\ 4,2 ± 3 Venlafaxine 2\ 45,79 ± 12,8 26--74 Bip Dis 3\ Alc Abuse 2 Haloperidol 1\ Schiz 6 Risperidone 2\ Atenolole 2\ Lacipidine 1\ Amlodipine 1\ Ranitidine 1 --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Schizoaf: Schizoaffective Disorder; Schiz: Schizophrenia; Bip Dis: Bipolar Disorder; An Nervosa: norexia Nervosa; Alc Abuse: Alcohol Abuse yr: years ::: Data on diagnosis and previous pharmacological treatment of patients are reported in Table [1](#T1){ref-type="table"}. Among the nineteen patients in Group 1, five were treated with haloperidol, five with olanzapine, five with risperidone and four with clozapine; among the nineteen patients of group 2, five started haloperidol, eight olanzapine, four risperidone and two quetiapine. Antidepressant used were escitalopram (two patients), citalopram (three patients), mirtazapine (four patients), paroxetine (one patients), sertraline (two), fluvoxamine (one patient), venlafaxine (three patients), clomipramine (two patients). Three patients started also lithium treatment. The mean antipsychotic doses, equivalent doses and mean plasmatic levels are reported in Table 2 (See [Additional file 1](#S1){ref-type="supplementary-material"} ). Potassium and magnesium serum levels were always within the normal range for all subjects. QTc interval ------------ Mean baseline QTc intervals were similar in the two groups: 422 ± 26 ms in group 1 and 414 ± 22 ms in group 2 (Indipendent T-test p \> 0,5). One patient in the monotherapy group, who started clozapine treatment, was excluded from this analysis and from pre-post treatment comparisons because QTc in DII was not measurable with sufficient accuracy. Three patients in the first group had QTc values that exceeded 450 ms at baseline while no patients in Group 2 exceeded this threshold before starting treatment. The average QTc interval after treatment was 421 ± 20 ms in the monotherapy group (range 391--452) and 438 ± 30 and in the polytherapy group (range 379--488) (Indipendent T test, p \< 0,05) (see figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Mean QTc (bars indicate standard deviations) values at baseline (T0) and after four days at full dosage (T1) of antipsychotic therapy, in the monotherapy (1) and politherapy (2) groups. ::: ![](1744-859X-4-1-1) ::: Compared with the baseline, mean QTc change after treatment was -- 1 ± 30 ms in Group 1 and 24 ± 21 ms in Group 2 (repeated measures ANOVA p \< 0,05). After treatment only one patient in Group 1 reached the threshold for borderline values of QTc interval in comparison to seven patients in Group 2 (Fisher Exact Text p \< 0,05). Moreover, in Group 2 two patients had a QTc exceeding 480 ms. The highest prolongations of QTc intervals (66 ms and 55 ms) were found in two patients taking risperidone, the first in association to clomipramine and the second in association to escitalopram. However in these two cases plasmatic dosages of antipsychotics were not higher than in other patients who reported a shorter QTc prolongation. Discussion ========== We found that the psychiatric population treated with antipsychotic monotherapy had much less risk of developing an increase in QTc interval compared to those treated with antipsychotics plus an antidepressant or lithium. Two main mechanisms seem to operate in determining the prolongation of QTc interval during treatment with different combinations of psychoactive drugs. The first is the synergic blockade of the HERG potassium channels, the second is the increase in drug levels (with subsequent augmented risk of cardiotoxicity) due to metabolic interactions between drugs that share the same metabolic pathway \[[@B20]\]. This mechanism may be particularly relevant in subject with genetic-determined impairment of CYP2D6 and CYP3A4 drug-metabolizing enzymes (poor metabolizer subjects) \[[@B21]\]. In our study, metabolic interactions leading to abnormal elevation of serum levels of antipsychotics did not seem to be the principal determinant of the greater QTc prolongation in the group with combined therapy. Indeed serum levels of antipsychotics were all within or under the expected range after therapeutic dosing in both groups, they were not higher in Group 2 compared to Group 1 and the highest prolongations observed were not associated with the highest antipsychotic serum levels. Recently, Harringan et al \[[@B22]\] analyzed, in a prospective randomized study, the effects of six antipsychotics on the QTc interval; they found that each of the antipsychotics were associated with measurable QTc prolongation which was not augmented by concomitant use of metabolic inhibitors, even if in their study plasmatic levels of antipsychotics raised after the addition of the specific metabolic inhibitor. In our study, the combination of different drugs doesn\'t seem to cause strong interactions on drug metabolism. However, in our sample, the combination of drugs that specifically interfere in their own methabolism, like fluvoxamine and olanzapine, paroxetine and risperidone, were avoided by clinicians. Antidepressant used in our study have a mild inhibitory action on antipsychotic methabolism and this can explain why antipsychotic serum levels didn\'t raise in Group 2 compared to Group 1. It is reassuring to find that significant pharmacokinetic interactions do not occur when the antipsychotics studied were coadministered with antidepressant commonly used in clinical practice. If the metabolic interactions do not seem to be the most important explanation for our results, an alternative explanation might be the synergic actions of different drugs on ion channels. The two patients with the highest prolongations were both taking risperidone, which is noted to block the I~kr~current \[[@B23],[@B24]\]. Actually, many psychotropic drugs share this capacity to inhibit I~kr~current, including not only antipsychotic agents but also antidepressant agents like citalopram, fluoxetine, paroxetine \[[@B16],[@B23]\]. Those agents may have synergic effect when used in combination. This study has several limitations, most of them related to the naturalistic setting of this study: we chosed to administer the second ECG after four days of therapy at full dosage (generally after one week from recruitment) because all antipsychotic used reached the steady-state in 3--5 days. Actually we couldn\'t chose a longer interval between the first and the second ECG because the average duration of recovery in our ward is 8,5 days. Dosing was clinically determined for symptom response by the treating psychiatrist and hence, doses varied within and between groups. Moreover statistical comparison of QTc interval changes among agents was not possible because of the small number of the samples. Finally we didn\'t measure antidepressant serum levels. Consistently with data reported in literature, we thought that antipsychotic would have been the principal drugs involved in QTc prolongation, while antidepressant would have only a role of potentiating agents. Actually, serum levels of antidepressants would have helped to explain the greater prolongation observed in Group 2. Conclusions =========== Prolongation of the QT interval by a non cardiovascular drug including notably an antipsychotic agent is considered a good marker of the arrythmogenic potential of that agent \[[@B1]\]. Psychiatric patients had been identified as a population at risk for cardiovascular problems \[[@B12],[@B25]\]. Mortality rates are higher in psychiatric patients than in the general population \[[@B26]\] and the pharmacological treatment itself might produce side effects that affect mortality from causes other than suicide \[[@B27]\]. There is a vast amount of evidence available showing the effect of a single antipsychotic on QTc interval however there is not much evidence obtained for clinical populations treated with a different combinations of drugs. Interestingly, the response to repolarization prolonging stimuli is \"patient-specific\" \[[@B9]\]. Thus, detecting the patients with a reduced repolarization reserve \[[@B28],[@B29]\] will lead to personalized psychotropic therapy according to the predisposition of that patient to develop cardiac side effects with a certain drug. In view of the fact that psychiatric patients are considered high risk subjects and since they frequently show electrolytes unbalances \[[@B30]\], an accurate monitoring of the QTc interval before and after the beginning of treatment appears warranted, particularly for patients taking multiple psychoactive drugs, sharing QTc prolonging properties. Further observational studies on larger samples of patients, comparing QTc intervals, plasmatic levels of antipsychotics and daily doses of psychotropic drugs are necessary to perform statistical comparisons for each kind of antipsychotic and for each kind of antipsychotic-antidepressant association commonly used in clinical practice. Competing interest ================== The author(s) declare that they have no competing interests. Author\'s contributions ======================= MS recruited and assessed participants and conceived of the study, and participated in its design and coordination. AV and GMD read ECGs and measured the QTc and analysed the results. CM ran the statistical analysis. AB and EC participated in the assessment of participants. MP, PB, JRSJ and FB participated in its design and coordination and the interpretation of results. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Additional file 1: Table 2.doc ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ Dr. Michela Sala was supported by a fellowship from Fondazione Polizzotto- Milano (Italy).	PubMed Central	2024-06-05T03:55:56.257574	2005-1-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088007/", "journal": "Ann Gen Psychiatry. 2005 Jan 25; 4:1", "authors": [ { "first": "Michela", "last": "Sala" }, { "first": "Alessandro", "last": "Vicentini" }, { "first": "Paolo", "last": "Brambilla" }, { "first": "Cristina", "last": "Montomoli" }, { "first": "Jigar RS", "last": "Jogia" }, { "first": "Eduardo", "last": "Caverzasi" }, { "first": "Alberto", "last": "Bonzano" }, { "first": "Marco", "last": "Piccinelli" }, { "first": "Francesco", "last": "Barale" }, { "first": "Gaetano M", "last": "De Ferrari" } ] }
PMC1088008	Introduction ============ Adolescence is a critical developmental period that may also be characterized as a period of great risk to healthy development \[[@B1]\]. Adolescents are often subjected to a multitude of traumatic events in their daily lives. Those who are victimised and/or traumatised often lag behind those who are not, in terms of their behavioural and physical growth \[[@B2]\]. PTSD is one syndrome that may result from exposure to extreme trauma and is characterized by persistent reexperiencing, avoidance/numbing, and hyperarousal symptoms, present for more than one month and associated with significant distress and/or functional impairment \[[@B3]\]. Although community violence is highly prevalent in South Africa, a lack of awareness that children and adolescents may be adversely affected both in the short- and long- term \[[@B4]\], has contributed to a dearth of systematic data on youth exposed to violence and PTSD. Much of the work done has focused on the impact of political violence in the 1980\'s \[[@B5],[@B6]\]. Although politically inspired violence has been in decline, criminal and domestic violence continues to prevail in local communities \[[@B7],[@B8]\]. This has seen large numbers of children and adolescents being exposed to, and directly involved in, acts of violence \[[@B4]\]. For example, Peltzer\'s study on rural children in South Africa found that 67% had directly or vicariously experienced a traumatic event (elicited from direct interviewing of the child or from parent report) while 8% fulfilled criteria for PTSD \[[@B9]\]. Studies in the Western Cape have also noted high rates of traumatisation and PTSD among youth. In Cape Town, a retrospective chart review found PTSD to be one of the most common disorders at the Child and Adolescent Psychiatry Unit at Tygerberg Hospital \[[@B10]\]. In a community study in Khayelitsha, Ensink, Robertson, Zissis and Leger \[[@B11]\] used self-report measures to determine exposure to violence, as well as structured questionnaires and non-standardized clinical assessments to elicit symptoms and make psychiatric diagnoses in children aged 6 to 16 years. All children reported exposure to indirect violence. Ninety-five percent had witnessed violent events, 56% had experienced violence themselves, and 40% met the criteria for one or more DSM-III-R diagnoses. 22% met criteria for PTSD. The most commonly reported PTSD symptoms were: avoidance of thoughts and activities associated with the trauma, difficulties in sleeping, and hypervigilance. A recent school survey of 307 Grade Ten pupils in the Western Cape, found that adolescents reported an average of 3.5 childhood traumatic experiences, and 12.1% met DSM-IV criteria for PTSD on self-report measures \[[@B12]\]. The most commonly reported symptoms were: avoiding thoughts about the event (34.4%), irritability (28.2%), difficulty showing emotion (26.5%), emotional upset at being reminded of the trauma (24.9%), and intrusive recollections of the event (19.4%). A significant positive correlation between multiple trauma exposure and PTSD symptoms was also found. These aforementioned studies suggest that South African children, as a whole, are exposed to high levels of trauma and that a significant proportion develop PTSD. In order to develop preventative and ameliorative strategies for dealing with trauma, reliable and valid measurements of posttraumatic stress responses are needed. Although several instruments for assessing childhood disorders and symptoms have been developed over the past two decades \[[@B13]\], most have originated in the United States \[[@B14]\]. PTSD assessment instruments need preferably to be standardized in local samples to improve detection of the disorder. In South Africa, increasingly limited resources such as few school psychologists and large classrooms make it difficult to accurately identify traumatized children. Nevertheless, identification of children at risk for PTSD post-trauma may lead to the more efficient use of resources that are currently available. The present study compared the psychometric properties of two instruments designed to assess trauma exposure and PTSD symptomatology and asked the question: \" Do the K-SADS (a diagnostic clinical interview) and the CATS (a self-report scale) elicit similar information with regards to rates of trauma exposure and PTSD symptoms in a sample of South African adolescents?\" Methodology =========== Sample ------ A random sample of Grade 11 adolescents (n = 67) was selected from two Cape Town schools (36 from school A and 31 from school B). Of the 67 who were selected, 58 (17 males and 41 females) agreed to participate. Their mean age was 16 years, 8 months (SD: 0.59; range: 16--18 years). All spoke English as a first language. The majority of participants were non-White (n = 39 Coloured, n = 1 Asian, n = 18 White). The schools selected had participated in an aforementioned school survey of three schools that were conveniently sampled. Anonymous self-report questionnaires of trauma exposure and PTSD symptoms were utilized \[[@B12]\]. Lack of resources (time and money) did not allow for all participants in that study to be included in the present one. Instrumentation --------------- ### 1. Demographic Questionnaire This was clinician-administered and devised for the present study. It included information on age, sex, residential address, parental marital status, and occupation. ### 2. Kiddie Schedule for Affective Disorders and Schizophrenia for School-Age Children -- Present and Lifetime Version (K-SADS-PL) \[[@B15]\] The K-SADS-PL is a standardised, DSM-IV based, clinician administered diagnostic interview, designed to provide an overview of current and lifetime psychopathology \[[@B16]\]. The K-SADS-PL has demonstrated good reliability and validity \[[@B17]\]. Abrosini \[[@B18]\] reported inter-rater reliability of 0.67 and 0.60 for current and lifetime PTSD, respectively. Construct validity \[[@B18]\] and criterion validity \[[@B19]\] have also been established. Based on DSM-III-R and DSM-IV criteria, the K-SADS-PL has an initial 82 item screen interview that surveys key symptoms for current and past episodes of twenty different diagnostic areas, some of which screen for multiple disorders. Symptoms that have been present in the previous two months are recorded as current. For the purpose of this study, in order to make the K-SADS comparable to the CATS, PTSD symptoms judged to have been present in the past month were recorded as \'current\'. Furthermore, only the PTSD and depression sections of the K-SADS-PL were administered. The screen interview and diagnostic supplement format is unique to the K-SADS-PL and greatly facilitates administration of the instrument with normal controls and patient populations \[[@B16]\]. Most items on the K-SADS are rated on a zero to three point scale with a score of zero indicating no information is available; \'1\' suggesting the symptom is not present; \'2\' indicating sub-threshold levels of symptomatology; and \'3\' representing threshold criteria \[[@B15]\]. The Clinician-Administered PTSD Scale (CAPS-CA), arguably the current \"gold standard\" clinical interview for childhood PTSD, was not chosen for this study as it does not make use of screening questions for PTSD and is too lengthy to administer. Although the K-SADS is designed to be administered to both parent and participant, it was administered to participants only. The reasons for this are twofold. First, our sample comprised older adolescents (16 to 18 years of age) and it was felt that the information gathered would be reasonably reliable. Second, as the primary objective of our study was to directly compare a clinician-assessment with a self-report, we did not consider it imperative that parental collateral be obtained. Previous studies have noted that parents may not always be aware of what their children are experiencing and may, therefore, not always be accurate historians \[[@B20]\]. ### 3. Child and Adolescent Trauma Survey (CATS) \[[@B21]\]) The CATS is a self-report index of PTSD qualifying stressors, non-PTSD life events, and PTSD symptoms. It is also a self-report measure of PTSD modelled on the Multidimensional Anxiety Scale for Children (MASC) \[[@B22]\] and the DSM-IV criteria for PTSD. The CATS is, however, not a DSM score scale but is derived using Item Response Theory (IRT). It includes stable indices of non-PTSD life events and provides a reliable and valid survey of secondary adversities, PTSD qualifying stressors, as well as a psychometrically sound symptom scale \[[@B21]\]. Unlike other self-rating scales, the CATS includes both a trauma exposure list and a PTSD inventory. Most self-rating scales focus on one or the other. The trauma list includes both direct (happened to me) and vicarious (happened to someone I know well) lifetime exposure. For example, the participant is required to indicate if s/he or someone s/he knows well has been badly beaten, or has been kidnapped during the participant\'s lifetime. Participants were also asked to note which was the worst event experienced and to report PTSD symptoms, experienced in the last month, in relation to this event. In the PTSD section, participants are asked to rate how often in the past month they have had experiences that inventory the major symptom domains of PTSD -- reexperiencing, avoidance and hyperarousal -- on a four-point Likert scale. For example, participants are asked to indicate how often they are jumpy and nervous, or how often they sleep poorly -- never, rarely, sometimes or often. Each of the DSM-IV PTSD criterion variables is represented by at least two questions \[[@B22]\]. According to March \[[@B23]\] a score of 27 and above on the PTSD scale should be taken to indicate that a child is at risk of clinically significant levels of PTSD. The CATS shows excellent internal \[[@B22]\] and test-retest reliability (March and Amaya-Jackson, unpublished data, 1997, \[in \[[@B24]\]\]). Neither the K-SADS nor the CATS has been cross-culturally validated. However, the K-SADS is widely used internationally as a diagnostic measure in children and adolescents. Procedures ---------- Permission to carry out the study was obtained from the Western Cape Department of Education and the Ethics Committees of the Universities of Stellenbosch and Cape Town. Consent from the heads of schools, parent bodies, and parents and learners was also obtained. Learners and their parents/guardians were informed that participation was entirely voluntary. Consent forms were handed to parents/guardians for signing prior to the interviews. Learners who opposed participation, or whose parents/guardians opposed participation, were excluded. All evaluations were conducted in private in rooms allocated by school staff. The order of administration of the CATS and K-SADS was counterbalanced to control for practice effects. The evaluation per participant took approximately 45 minutes. Data Analysis ------------- Microsoft Excel and Statistica software were used for data analysis. Student t-tests were used to determine significances for numeric data. The difference of two proportions test and the McNemar chi-square test were used to determine significances for categorical data. Cohen\'s kappa coefficients (K) were used to measure the level of agreement between the measures. As initial analysis revealed a statistically significant difference and low level of agreement between the measures, a more sensitive cut-off point was established for the CATS using a Receiver Operator Characteristic (ROC) Analysis. Results ======= Exposure to Traumatic Events ---------------------------- On interview (Table [1](#T1){ref-type="table"}), 86% of participants reported lifetime exposure to at least 1 traumatic event, (mean = 2.3; SD = 1.7; range = 0--10), while on self-report (Table [2](#T2){ref-type="table"}), 91% of participants reported direct or indirect lifetime exposure to at least 1 traumatic event (mean = 3.7; SD = 3.2; range = 0--14). The difference of two proportions test revealed that the number of participants who reported experience of a traumatic event on each measure was not significantly different (p = 0.36). The level of agreement between the measures was 0.74 (SE = 0.15; CI = 0.46--1.0). These events were random, rather than politically-motivated experiences of trauma. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Frequencies of reported traumas on the K-SADS ::: Event Number % -------------------------------- ------------ ------- Car accident 4 6.9 Other accident 9 15.5 Fire 2 3.4 Witness of a disaster 4 6.9 Witness of a violent crime 14 24.1 Victim of a violent crime 6 10.3 Confronted with traumatic news 33 56.9 Witness to domestic violence 18 31 Physical abuse 2 3.4 Sexual abuse 5 8.6 Other 11 19 (n = 58) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Frequencies of reported traumas on the CATS ::: Event Happened to Me Happened to Someone I Know Well ---------------------------------------------------- -------------------- ------------------------------------- ---- ------ Badly bitten by a dog or another animal 8 13.8 15 25.9 Badly scared or hurt by a gang or criminal 4 6.9 17 29.3 Badly beaten 1 1.7 14 24.1 Shot or stabbed 0 0 16 27.6 Terrible fire or explosion 0 0 7 12.1 Chemical or other deadly poisoning 1 1.7 4 6.9 Bad storm, flood, tornado, hurricane or earthquake 2 3.4 6 10.3 Bad car, boat, bike, train, or plane accident 3 5.2 18 31 Other very bad accident 5 8.6 9 15.5 Got sick and almost died or died 5 8.6 28 48.3 Kidnapped or held captive 0 0 5 8.6 Suicide attempt or died from suicide 4 6.9 19 32.8 I was taken away from my family 1 1.7 I saw something terrible happen to a stranger 16 27.6 Other shocking or terrifying event 5 8.6 2 3.4 (n = 58) ::: Differences in Reporting of Trauma Exposure Between Measures ------------------------------------------------------------ When both direct (\"happened to me\") and vicarious (\"happened to someone I know well\") trauma exposure on the CATS was considered, significantly more traumas were endorsed on the CATS (mean = 3.7) than on the K-SADS (mean = 2.3) (t = -3.94; p = \< 0.01). However, when vicarious exposure was excluded on the CATS, the number of traumas reported on the K-SADS was significantly higher (t = 5.68; p = \< 0.01). PTSD Diagnoses -------------- 11 participants (19%) received a diagnosis of PTSD on the K-SADS, while only 1 participant (1.7%) received a diagnosis of PTSD on the CATS using a cut-off of 27, as suggested by March \[[@B23]\]. This difference was significant (χ^2^= 50.3; p \< 0.01) with the level of agreement between the measures (K) being 0.14 (SE = 0.25; CI = -0.35--0.62). Three participants diagnosed with PTSD(27.3%) on the K-SADS appeared to have developed it in response to sexual assault trauma, as did the one participant screened with PTSD on the CATS. ROC Analysis ------------ Given the low level of agreement using a CATS cut-off of 27, an ROC analysis (Table [3](#T3){ref-type="table"}) was done in order to establish a CATS cut-off score that would be more appropriate for the sample. First, using the K-SADS as the \"gold\" standard for a diagnosis of PTSD (a measure that identifies those individuals who have or do not have a disorder), the sensitivity and specificity for various CATS cut-off scores were established. In addition to sensitivity(the proportion of true positives that are test positives \[true positive probability\]) and specificity(the proportion of true negatives that are test negatives \[true negative probability\]); \'1 -- specificity\' (false positive probability), the gradientsbetween each point, and the positive and negative predictor valueswere calculated (the predictive value of a positive test is the proportion of those with a positive test result who actually have the disorder, while the predictive value of a negative test is the proportion of those with a negative test result who do not have the disorder). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Receiver Operator Characteristic (ROC) and Predictive Values ::: Cut-off Values Sensitivity Specificity 1-specificity Gradient Predictive Values -------------------- ----------------- ----------------- ------------------- -------------- ----------------------- ----- 0 100 0 100 \-- \-- \-- 1 100 2 98 0 22 100 3 100 17 83 0 22 100 4 91 21 79 2.25 21 91 5 91 23 77 0 22 92 7 91 28 72 0 23 93 8 91 32 68 0 24 94 9 91 36 64 0 25 94 10 82 36 64 ∞ 23 90 11 82 40 60 0 24 91 12 82 47 53 0 27 92 13 82 53 47 0 29 93 14 82 62 38 0 33 94 15 73 70 30 1.12 36 92 16 64 81 19 0.82 44 91 17 64 87 13 0 44 91 18 55 87 13 ∞ 50 89 19 55 92 8 0 60 90 20 55 94 6 0 67 90 21 55 98 2 0 86 90 22 55 100 0 0 100 90 23 36 100 0 ∞ 100 87 25 18 100 0 ∞ 100 84 27 9 100 0 ∞ 100 83 ::: An ROC curve graph (sensitivity and 1 -- specificity) was also plotted (Figure [1](#F1){ref-type="fig"}). The area under the curve (sensitivity of the scale) was found to be 0.805. A cut-off that gives a gradient closest to 1 is usually chosen as appropriate because it maximises both sensitivity and specificity. With a cut-off of 15, 22 participants had scores indicative of PTSD on the CATS. However, the difference between the number of participants diagnosed with PTSD on the K-SADS and the CATS remained significant (χ^2^= 19.9; p \< 0.01). While significance was in the expected direction (i.e. the prevalence on self-report was higher than on interview), but the level of agreement was doubled (K = 0.31; SE = 0.14; CI = 0--0.59). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### ROC curve ::: ![](1744-859X-4-2-1) ::: A t-test comparing the scalar scores of participants with a PTSD diagnosis on the K-SADS (mean = 18.5, SD = 7.8) and participants without a PTSD diagnosis (mean = 10.4, SD = 6.4) showed the difference between the two measures to be significant (t = 3.64; p \< 0.01). PTSD Symptom Clusters --------------------- On the K-SADS, 18 participants met DSM-IV criteria for re-experiencing symptoms (Criterion B), 15 participants met criteria for avoidance symptoms (Criterion C), and 18 participants met criteria for hyperarousal symptoms (Criterion D). Since the CATS is not a DSM-IV PTSD score scale, the number of participants meeting individual DSM-IV criteria could not be established. However, the CATS does include six items for Criterion B, two for Criterion C, and four for Criterion D, so a scalar score for each of these factors could be derived. Student t-tests comparing mean Criterion B, C, and D CATS scalar scores for participants fulfilling criteria B, C, and D, respectively, on the K-SADS, with those not meeting criteria, revealed significant differences for all three symptom clusters at the 0.05 level. The Criterion B mean scalar score for participants meeting Criterion B on the K-SADS was 7.3 (SD = 4.3) compared to 4.8 (SD = 3.1) for those not meeting criteria (t = -2.45; p = 0.02). Participants with Criterion C on the K-SADS had a mean scalar score of 3.9 (SD = 1.8), while those not meeting Criterion C had a mean score of 2.2 (SD = 2.0) (t = -3.12; p = 0.03). The mean scalar score for participants meeting Criterion D was 5.1 (SD = 3.3) compared to a mean score of 3.1 for those not meeting this criterion (SD = 2.5) (t = -2.50; p = 0.02). PTSD Symptoms ------------- Table [4](#T4){ref-type="table"} compares the percent endorsement of PTSD symptoms on the K-SADS and the CATS. Student t-tests were used to compare number of symptoms reported on the K-SADS (mean = 3.3, SD = 5.0) with number of symptoms reported on the CATS (mean = 3.7, SD = 2.8). No significant differences were noted (t = -0.83; p = 0.41). Kappa\'s were then done to measure the level of agreement between the measures for symptoms that could be directly compared for the sample as a whole, and for participants with and without PTSD on the K-SADS (Table [5](#T5){ref-type="table"}). Items assessing sleep problems, distress at reminders of event, and exaggerated startle responses evidenced the best agreement across instruments. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### PTSD symptoms ::: Rate of PTSD symptoms on the K-SADS Rate of PTSD symptoms on the CATS ----------------------------------------------------------------------------------- --------------------------------------- ------------------------------------------------------------------ ---- symptom \% symptom \% Comparable Symptoms Recurrent thoughts or images of events 28 I go over and over what happened in my mind 40 Efforts to avoid thoughts or images associated with the trauma 28 I try not to think about what happened 47 Insomnia 22 I sleep poorly 26 Irritability or outbursts of anger 24 I am grouch and irritable 36 Distress at reminders of event 16 When something reminds me of what happened I get tense and upset 21 Exaggerated startle response 17 I am jumpy and nervous 29 Nightmares 16 I have bad dreams about what happened 9 Difficulty concentrating 19 I have trouble keeping my mind on things 28 Efforts to avoid activities or situations that arouse recollections of the trauma 21 I try to stay away from things that remind me of what happened 21 Non-comparable Symptoms Sense of foreshortened future 3 I worry that what happened will happen again 57 Feelings of detachment or estrangement 21 I get scared when I think about what happened 38 Inability to recall important aspects of the trauma 10 I have unwanted thoughts about what happened 21 Restricted affect 28 Hypervigilance 17 Physiological reactivity upon exposure to reminders 9 Dissociative episodes, illusions or hallucinations 21 Diminished interest in activities 22 Repetitive play related to event / reenactment 2 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Levels of agreement for comparable PTSD symptoms ::: PTSD Symptom 95% Confidence Interval ------------------------------------------- ------- ------- ------ ----------------------------- ------- Recurrent thoughts or images of event (i) 0.02 0.14 -0.31 0.26 (ii) -0.57 0.22 -1.01 -0.13 (iii) 0.01 0.18 -0.33 0.35 Trying not to think about the event (i) 0.25 0.13 -0.003 0.51 (ii) -0.14 0.56 -1.24 0.96 (iii) 0.07 0.17 -0.26 0.41 Sleep problems (i) 0.44 0.15 0.15 0.72 (ii) 0.61 0.25 0.11 1 (iii) 0.16 0.23 -0.3 0.62 Anger and irritability (i) 0.24 0.14 -0.05 0.52 (ii) 0.24 0.3 -0.35 0.83 (iii) 0.13 0.19 -0.24 0.49 Distress at reminders of event (i) 0.48 0.16 0.17 0.79 (ii) 0.44 0.28 -0.1 0.98 (iii) 0.17 0.29 -0.4 0.74 Exaggerated startle response (i) 0.39 0.15 0.09 0.68 (ii) 0.3 0.35 -0.38 0.98 (iii) 0 0.3 -0.59 0.59 Nightmares (i) 0.2 0.23 -0.26 0.65 (ii) 0.23 0.26 -0.28 0.73 (iii) -0.05 0.45 -0.93 0.82 Difficulty concentrating (i) 0.19 0.17 -0.04 0.51 (ii) 0.35 0.26 -0.15 0.86 (iii) 0.03 0.23 -0.41 0.48 Efforts to avoid reminders of event (i) 0.27 0.15 -0.03 0.56 (ii) -0.31 0.3 -0.89 0.28 (iii) 0.34 0.18 -0.02 0.7 \(i) Total sample (N = 58); (ii) Participants with PTSD on the K-SADS (N = 11); \(iii) Participants without PTSD on the K-SADS (N = 47). ::: Participants with and without a diagnosis of PTSD based on the K-SADS were compared on percentage endorsement of each CATS symptom. The difference of two proportions test showed a significant difference in only five of the twelve symptoms (recurrent thoughts about the event, exaggerated startle response, difficulty concentrating, avoidance of physical reminders of the event, and nightmares). The other symptoms did not discriminate well between participants with and without PTSD. Internal Consistency -------------------- Alphas of 0.96, 0.97 and 0.93 were obtained for the K-SADS PTSD Criterion B, C, and D respectively. These were not improved by the removal of any items within a symptom category (Criterion B, C, and D). Alphas of 0.79 and 0.67 were obtained for Criteria B and D in the CATS, which were not improved by the removal of any items. An alpha was not calculated for Criterion C as there are only two items comprising that category. A total internal consistency of 0.99 was obtained for the PTSD section of the K-SADS and a total internal consistency of 0.86 was obtained for the CATS. Discussion ========== Compared with other international community-based studies \[e.g. \[[@B25],[@B26]\]\], our study found high rates of trauma exposure on both clinician-administered and self-report measures in adolescents, with the majority (86% on the KSADS and 91% on the CATS) reporting exposure to at least one traumatic event in their lifetime. These rates are consistent with previous South African studies \[e.g. \[[@B12]\]\]. Consistency in reporting of traumatic events was low between the measures and participants were more likely to endorse a trauma on the CATS than on the K-SADS. This may be attributable to the fact that more vicarious traumatisation as compared to directly experienced or witnessed traumas is asked about in the CATS, or to the relative privacy of the self-report format- participants may have felt more comfortable in admitting to traumatic experiences on a self-report scale which may be perceived as less intrusive \[[@B27]\]. 19% of adolescents in the sample were diagnosed with PTSD on the K-SADS. This rate is comparable with the PTSD rate found in a larger sample of adolescents who were sampled in the same geographical region \[[@B28]\]. The rate of 19% is, however, higher than that documented in a previous survey of which this sample constituted a sub-sample \[[@B12]\] The passage of time (i.e. more than a year between assessments) may be one reason for the higher rate of PTSD in the sub-sample. Most other South African community-based studies in adolescents (with the exception of a study by Ensink et al. \[[@B11]\], that have used self-report measures of assessment, have documented lower rates of PTSD than was found in this study. The differing rates of PTSD between the K-SADS and the CATS (using a cut-off 27 on the CATS), suggests that this cut-off may be too high in our setting. The ROC analysis yielded a cut-off of 15 on the CATS. This cut-off maximizes both the number of true positives and true negatives and may be more appropriate. Using a cut-off of 15, 22 participants (38%) were diagnosed with PTSD. While there still remained significant differences in the rates of PTSD using this cut-off, the level of diagnostic agreement was higher than with a cut-off of 27. Our findings are consistent with studies that have demonstrated that self-report measures \[e.g. \[[@B29],[@B30]\]\] yield higher rates of psychiatric diagnoses than clinician-based interviews \[e.g. \[[@B25],[@B27]\]\]. Moreover, significant differences in CATS severity scores between participants with and without PTSD, suggests that the CATS discriminates well between those with and without the disorder. Further, significant differences were found between mean CATS scores for Criterion B (intrusive), C (avoidance), and D (hyperarousal) symptoms between participants meeting DSM-IV criteria for these clusters on the K-SADS, and those not meeting criteria. The two symptoms that were most frequently endorsed on both the K-SADS and the CATS (recurrent thoughts/ images of event and efforts to avoid thoughts of the event) are also among the symptoms most frequently reported in other studies \[[@B11],[@B12]\], suggesting that careful inquiry of these symptoms is important. However, the level of agreement for specific symptoms appeared to be suboptimal: overall, participants who reported symptoms on the K-SADS did not necessarily report the same symptoms on the CATS. That said, participants with PTSD were more consistent in their reporting than those without PTSD. Nevertheless, the lack of significant differences in the numbers of symptoms reported between the measures suggests that these measures may be comparable in eliciting the average number of symptoms experienced post-trauma. The CATS appeared to discriminate well between those with and without PTSD on five of twelve items (recurrent thoughts about the event, exaggerated startle response, difficulty concentrating, avoidance of physical reminders of the event, and nightmares), suggesting that these symptoms may be more sensitive indicators of PTSD. General Implications of Findings -------------------------------- The K-SADS and CATS yield different information about the level and type of trauma exposure, therefore researchers and clinicians should be cautious when substituting one for the other. The K-SADS is likely to yield more detailed information on witnessing traumatic events, while the CATS is likely to yield more information on vicarious trauma exposure. Adolescents are also more likely to endorse a trauma on the CATS than they are on the K-SADS. The significantly larger proportion of adolescents with scores indicative of PTSD on the CATS, compared to the K-SADS, indicates that the CATS may be better utilized as a PTSD screening device (as suggested by its author), with a cut-off threshold of 15 instead of the original threshold of 27, in the South African context. This will identify over one third of all participants with PTSD while making few false positive identifications. This will, however, require replication in a larger South African cohort. For an actual PTSD diagnosis, a clinician-based diagnostic interview may be more appropriate even though it is likely to be more time consuming. Several limitations are worth mentioning. First, the K-SADS was not administered to both parents and learners as it is intended to be, thus participants\' responses were not verified by collateral information from parents and legal guardians. Second, the sample comprised predominantly female adolescents of mixed race. Even though this constitutes the majority ethnic group in the province, the small sample and truncated age limits the generalizability to the larger population. Further, socio-demographic variables (e.g. social class, family income and race) were not accounted for in the analysis. Third, cultural influences may favour certain symptoms of trauma over others \[[@B31]\] and it has been noted that there is a need to identify other post-traumatic expressions of distress, such as somatization \[[@B32],[@B33]\]. Both the K-SADS and the CATS do not attempt to capture these experiences. However, PTSD has been widely documented in traumatized cohorts from different ethnocultural backgrounds and those from non-Western cultures who meet PTSD diagnostic criteria often show a similar clinical course and response to treatment \[[@B33]\]. Fourth, we used the DSM-IV concept of trauma to compare these instruments and some authors, for example Summerfield \[[@B34]\], have highlighted some of the difficulties with the concept of trauma as defined in the DSM. It may be that events counted and endorsed as traumas were too broad to ascertain their level of agreement on the K-SADS and the CATS. Fifth, while we attempted to compare traumatic events and symptoms across instruments, it must be noted that these instruments are not necessarily suited to direct comparison. For example, the two instruments measure different traumatic events, automatically placing a cap on the level of agreement. In view of the high levels of violence in South African youth, identification of those children and adolescents with PTSD is important and necessary to allow for appropriate interventions. Owing to limited resources, administration of diagnostic clinical interviews to all youth is not feasible. Self-report scales, even though they do not replace clinical interviews, may be useful in identifying those youth in the community who are most at risk. This may help to facilitate more targeted and efficient treatments. While this study has limitations, some tentative conclusions can nevertheless be drawn. High rates of trauma exposure and PTSD characterize South African children and adolescents. Self-report scales may be better utilized as screening instruments rather than as diagnostic tools. To establish more efficient ways of diagnosing PTSD and other post-traumatic sequelae in the South African setting, future studies (using self-rating scales and brief PTSD diagnostic measures) should be conducted in larger samples, more representative of the South African population. In particular, we need to establish and verify more suitable cut-off values on these instruments to enable the identification of those children and adolescents who are at higher risk for PTSD and other disorders. Competing Interests =================== The author(s) declare that they have no competing interests. Acknowledgements ================ This work is supported by the Medical Research Council (MRC) Unit on Anxiety and Stress Disorders, Department of Psychiatry, University of Stellenbosch.	PubMed Central	2024-06-05T03:55:56.259656	2005-1-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088008/", "journal": "Ann Gen Psychiatry. 2005 Jan 31; 4:2", "authors": [ { "first": "S", "last": "Suliman" }, { "first": "D", "last": "Kaminer" }, { "first": "S", "last": "Seedat" }, { "first": "DJ", "last": "Stein" } ] }
PMC1088009	Since January 1^st^, 2005 the title of our journal has changed from Annals of General Hospital Psychiatry (AGHP) to Annals of General Psychiatry (AGP). The AGHP was launched in September 2002 as an Open Access, peer-reviewed, online journal covering the wider field of Psychiatry, Neurosciences and Psychological Medicine. It was aimed at publishing articles on all aspects of psychiatry. However, it seems that the title caused a certain degree of confusion. As a reflection of this confusion, we received a lot of questions from colleagues asking whether their paper would fit the journal or not. These colleagues had authored general psychiatry papers, definitely inside the scopes of the journal, but their impression was that the journal, because of its title, had to do only with \'psychological medicine\', \'psychosomatic medicine\', \'consultation-liaison psychiatry\' and similar areas, and not with general psychiatry as part of the medical model and tradition. Repeated efforts from our side to change this impression were largely unsuccessful. So, changing the title seemed to be the only solution left in order to cure this misunderstanding. So, now the Annals of General Psychiatry (AGP) is born! The web site has changed to <http://www.annals-general-psychiatry.com>. Our editorial board has been enlarged to include eminent figures from the field of Psychiatry. It is still an Open Access, peer-reviewed, online journal covering the wider field of Psychiatry, Neurosciences and Psychological Medicine, and aims at publishing articles on all aspects of psychiatry. Primary research articles are the journal\'s priority, and both basic and clinical neuroscience contributions are encouraged. The AGP strongly supports and follows the principles of evidence-based medicine. The AGP is not a new journal. Instead, it is the continuation of the AGHP. It emphasizes a biopsychosocial approach to illness and health, and where possible a wider dialogue will be opened on specific articles thereby placing the research into a wider framework. Online publication with Open Access, are powerful tools to achieve these goals. AGP continues to be published by BioMed Central, an independent publisher committed to ensuring peer-reviewed biomedical research is Open Access. That means it is freely and universally accessible online, it is archived in at least one internationally recognized free access repository, and its authors retain copyright, allowing anyone to reproduce or disseminate articles, according to the BioMed Central copyright and license agreement. AGP takes this a step further by ensuring that all published articles are Open Access. AGP\'s articles are archived in PubMed Central, the US National Library of Medicine\'s full-text repository of life science literature, and also in repositories at the University of Potsdam in Germany, at INIST in France and in e-Depot, the National Library of the Netherlands\' digital archive of all electronic publications. BioMed Central is working closely with the Institute for Scientific Information to ensure that citation analysis of articles published in AGP will be soon available. Also, by being part of BioMed Central, AGP is a member of the Committee on Publication Ethics (COPE). Recently both the BioMed Central and the AGP joined the World Health Organization Department of Mental Health and Substance Abuse initiative and signed the January 2004 joint Statement which was published in the AGHP \[[@B1]\]. This gives the AGP a new additional role, that is to contribute to the efforts of the WHO to galvanize mental health research and publishing in middle and low-income countries. We hope that the change in the journal\'s name will cure the confusion caused by its previous title and help to achieve the journal\'s aims and scope, that is to help the world-wide promotion of research and publishing in the mental health area. Competing interests =================== Dr KN Fountoulakis has received support for research and educational reasons as well as honoraria and financial assistance with registration, accommodation, and/or travel expenses for conferences and meetings from Eli Lilly, Janssen-Cilag, AstraZeneca, Pfizer, Sanofi-Synthelabo, Novartis, Wyeth, Organon and Bristol-Myers Squibb.	PubMed Central	2024-06-05T03:55:56.265461	2005-2-1	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088009/", "journal": "Ann Gen Psychiatry. 2005 Feb 1; 4:3", "authors": [ { "first": "Konstantinos", "last": "Fountoulakis" } ] }
PMC1088010	Background ========== Major cities in countries with religious, economical, or tourist attractions have experienced an increase in the influx of visitors; some of whom are mentally unwell, or subsequently become ill whilst in a foreign country. Ødegaard (1932) described the tendency to travel in people with schizophrenia; more recent literature describe \"crisis-flight\" as a way of finding a geographical solution to internal problems \[[@B1]\], and airports as concrete representation of subjective conflicts related to separation and reunion at times of crisis \[[@B2]\]. Mental health care models of delivery, such as de-institutionalisation, the legal framework for admissions, the social acceptance of mental illness (including stigma and alienation) vary across countries world-wide \[[@B3]\]. Moreover, the perception and experience by vulnerable individuals of these issues in their own country might also be factors that contribute to individuals with mental illness travelling abroad. Psychological \[[@B4]\], artistic -\"Stendhal syndrome\" \[[@B5]\], religious -\"Jerusalem syndrome\" \[[@B6]\], and time zone changes \[[@B7]\], among others, are described as factors related to psychiatric decompensation in travellers. There is little data to show the number of these patients admitted to National Health Service (NHS) hospitals in the United Kingdom (UK), therefore it is difficult to know the real impact of overseas patients\' admissions on the NHS. This study describes the different aspects concerning the referral, admission, treatment, and outcome of overseas visitors (persons who are not ordinarily resident in the UK) admitted under a Mental Health Team (MHT) at a NHS psychiatric hospital in inner London. Ethical, legal, and economic implications are discussed. The involvement of consulates in the admission process of overseas patients is suggested and the benefits of their involvement discussed. Setting ------- The multidisciplinary MHT for this study serves a population of around 29, 000 local residents, in addition to the homeless and transient people in the area of Westminster in central London. The area is close to major international rail and bus terminals, has direct connection with large international airports, and has a number of business and tourist attractions. There are mixed affluent and under-privileged sectors in the area with an average of 41 psychiatric beds for 100, 000 habitants, and a Jarman index (an index of social deprivation, ranging from -32.79 \[less deprivation\] to 54.89 \[more deprivation\]) of 22.7. The MHT had a number of beds allocated for admission at the 75-bed psychiatric hospital (the Hospital), which is also used by other mental health teams operating in annexed geographical areas. Methods ======= Review of all overseas nationals between 18--65 years of age, admitted to Hospital between 1 January 1999 and 31 December 1999. The sample included patients admitted before 1 January 1999 who were still inpatients by 31 December 1999. Foreign residents in the UK, transient foreign nationals attending outpatient clinics, foreign nationals pursuing immigration into the UK, or patients seeking, or under refugee status were not included in this study. The medical team assigned diagnoses using the ICD-10 (Classification of mental and behavioural disorders: clinical description and diagnostic guidelines. WHO, Geneva, 1992), and additionally using the ICD-10: DCR-10 (Classification of mental and behavioural disorders: diagnostic criteria for research. WHO, Geneva, 1993). Both authors, FJC and AMP, were directly involved in the management of the patients in this study. Data was obtained from: • Medical notes. • Discharge summaries from previous admission in the UK (if applicable). • Medical and psychiatric reports from patients\' country of origin (if applicable). • Database archives of the Mental Health Team. • The Hospital\'s Human Resources department. • Social Services reports. • Police reports. • Assessment and interview of patients and relatives (when available), by AMP and FJC. Fisher\'s exact test was used in the statistical analysis to examine the relationship between two categorical variables. The relationship between cost and other continuous variables was measured using Spearman\'s rank correlation test. The relationship between cost and categorical variables was assessed using the Mann-Whitney U test. Results ======= Demographic characteristics (Table [1](#T1){ref-type="table"}) -------------------------------------------------------------- ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics ::: No of patients (n = 31) % ---------------------------------------------------------- -------------------- ----------------------------- ------- Gender Female 13 42 Male 18 58 Age (years) Range 23--52 Mean 38 Mean male age 35 Mean female age 41 Marital status Single (14 male-7 female) 21 68 Married (all male) 3 10 Divorced (all female) 6 19 Widowed (male) 1 3 Nationality EU nationals (includes five with adopted EU nationality) 22 71 Other nationalities 9 29 Ethnicity White 26 84 Black 3 10 Other 2 6 Language Did not speak English 2 6 Spoke basic English 15 48 Spoke fluent English 14 45 Required interpreter 17 55 Did not require interpreter 14 45 Mobility before arrival in England Travelled directly to England 21 68 Travelled to other countries before arriving in England 10 32 Mental health on arrival in England Unwell on arrival 23 74 Became unwell in England 4 13 Not ascertained 4 13 Purpose of travel to England Tourism 20 65 To \"escape persecution\" in their country 5 16 To visit friends-relatives 5 16 Business 1 3 Support in England (other than statutory services) None 26 84 From friends or relatives 5 16 ::: Of 163 (100%) admissions under the care of the MHT between 1 January 1999 and 31 December 1999, 31 (19%) were overseas patients. 58% were male; age range (years) was 23--52. 90% were unattached. 71% came from Europe; most were white (84%). 45% spoke fluent English, 48% spoke basic English; 55% required an interpreter for assessments. 68% travelled directly from their home country to England; 32% had been to other countries before arriving in England. 74% were mentally unwell on arrival in England. 65% travelled as tourists; 16% gave \"escaping persecution\" as a reason for travelling. Only 16% had support from friends or relatives in England. The socio-economic status of overseas patients in their home country showed one (3%) homeless and 97% housed. Of these, 13 patients lived independently, 13 lived with relatives, and 4 were housed by social services. 52% had been employed and 48% unemployed, with 4 of them receiving social benefits. In England, 61% overseas patients were homeless, 13% were housed by local services, and 26% lived in rented accommodation, with relatives, or with friends. 10% had financial income from employment, 3% received benefits, 26% received financial help from family or other sources, and 61% patients had no financial income. Admission, assessment and treatment (Table [2](#T2){ref-type="table"}) ---------------------------------------------------------------------- ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Admission, assessment and treatment ::: No of patients (n = 31) % -------------------------------------------------------------------- ---- ----------------------------- ------- Mode of contact with the Mental Health Team Police referral to mental health team for assessment 13 42 Police referral to hospital (section 136 of the Mental Health Act) 7 23 Assessment by mental health team (community-hospital) 11 35 Appeals against section of the Mental Health Act Appealed 9 29 Tribunals (5 patients) 7 Not discharged 6 Deferred discharge 1 Discharged from section by MHT before hearing 3 Symptoms on admission Delusions-hallucinations-thought disorder 25 81 Mania-hypomania-elated mood 4 13 Depression-delusions 2 6 Impaired insight 25 81 Length of illness before admission \>1 year 10 32 6--12 months 7 23 1--5 months 8 26 Not ascertained 6 19 Psychiatric history Had contact with psychiatrist 21 68 \>1 year before admission (range 1--10 years) 14 3 months before admission 7 None 4 13 Known to social-primary care, but not to psychiatric team 2 6 Not ascertained 4 13 Dual diagnosis Diagnosed 0 00 History of drug use 3 10 Used drugs regularly 1 3 Forensic history Had history 6 19 No history 17 55 Not ascertained 8 26 Medication Refused, or given \"if required\" 5 16 Given regularly 26 84 Atypical neuroleptics 15 Typical neuroleptics 10 Antidepressants 1 Took medication in the past 15 48 ::: Forty two per cent of patients were referred to the MHT for assessment at a police station. The police brought 23% of patients to Hospital, for assessment under section of the Mental Health Act 1983 (MHA) -- see Table [4](#T4){ref-type="table"} for further explanation of relevant sections of the MHA. 35% were assessed in the community or self-presented to hospital. No immediate discharges were granted on seven appeal hearings to review formal admissions; one (3%) patient received a deferred discharge. 81% presented with delusions, hallucinations, or thought disorder, alone or in combination. 81% had impaired insight. 32% had been ill for at least one year before the current admission. 68% had a psychiatric history, 13% had no psychiatric history; 6% were known to social and primary care services, but had not been assessed by a psychiatric team. Table [3](#T3){ref-type="table"} shows the diagnoses according to the International Classification of Diseases-10^th^edition (WHO, Geneva 1992). 74% had psychotic disorders, all of them with positive symptoms of the illness. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Diagnosis ::: Diagnosis ICD-10 (WHO) classification Number of patients Total (%) ---------------------------- --------------------------------- ------------------------ --------------- Schizophrenia F 20.0 16 F 20.00 1 F 20.02 1 18 (58) Acute psychotic disorder F 23.2 2 F 23.9 2 4 (13) Schizoaffective disorder F 25.2 1 1 (3) Bipolar affective disorder F 30.1 1 F 31.2 5 6 (19) Drug induced psychosis F 14.55 1 1 (3) Not determined \- 1 1 (3) ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Discharge and outcome ::: No of patients (n = 31) % --------------------------------------------------------------- -------------- ----------------------------- ------- Mental state on discharge No or little improvement 6 19 Moderate improvement 11 35 Major improvement 14 45 Contact with relatives-care team in country of origin Before admission 1 3 At some point after admission 23 74 With care team 18 patients With family 14 patients With care team and family 9 patients No contact made 7 23 Contact with consulates-embassies Contacted 16 52 Gave information 9 Provided travel documents 4 Could not help 3 Not contacted 15 48 Care Programme Approach Enhanced CPA 3 10 Follow up by mental health team 2 Initiated but discontinued 1 Standard CPA 28 90 Patients-relatives agreement with discharge plan Agreed 27 87 Disagreed 4 13 Absent without leave 2 Deferred discharge by MHRT 1 Ongoing review under s.86 MHA\* 1 Outcome on discharge Repatriated by the hospital 16 52 Discharged to return to country of origin 6 19 Taken home by relatives 2 6 Absent without leave 2 6 Discharged with follow up by the Mental Health Team 4 13 Application made for section 86 MHA\* 1 3 Medication on discharge Supplied to take home 25 81 Not supplied 6 19 Unreliable 4 absent without leave 2 Average length of treatment (days) 43.4 Range 1--365 Length of treatment according to Mental Health Act status Voluntary (mean 22.3 days) 7 23 Section 4\\ (mean 4 days) 1 3 Section 2\\\* (mean 21 days) 13 42 Section 3\\\\ (mean 91.5 days) 10 32 \Section 86 of the Mental Health Act 1983: Allows the Home Secretary to authorise the removal to another country of patients, who are neither British nor Commonwealth citizens having the right of abode in the UK, who are receiving treatment for mental illness in hospital under section of the MHA. \\Section 4: compulsory admission and detention for up to 72 hours for assessment. \\\Section 2: compulsory admission and detention for up to 28 days for assessment or assessment followed by treatment for mental disorder. \\\\Section 3: compulsory detention for up to six months for treatment. ::: One (3%) patient used drugs regularly, 10% had a history of drug use. There was no dual diagnosis. 55% had no forensic history; one patient was referred to the MHT by the local forensic team. On admission, 84% took medication regularly; 16% refused or had medication \"If required\", usually for agitation. 48% had taken medication for mental health problems in the past. Two patients had been admitted under the MHT on a previous visit to London; at that time they had been repatriated and subsequently admitted to hospital in their country, returning back to London after discharge from hospital. One patient had been admitted to two other psychiatric hospitals in London before admission to the MHT. One patient had been assessed by the MHT on a previous visit to London. Figure [1](#F1){ref-type="fig"} shows the MHA status on admission and discharge, and the sections of the MHA used. 77% of patients, including two patients admitted informally and placed under section of the MHA shortly after admission, were admitted and 48% were discharged under section of the MHA. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Mental Health Act 1983 status on admission and discharge ::: ![](1744-859X-4-4-1) ::: Discharge and outcome (Table [4](#T4){ref-type="table"}) -------------------------------------------------------- Nineteen per cent of patients showed no-little improvement in mental state; 35% showed moderate improvement, 45% showed a major improvement. The mental state was assessed regularly at weekly review meetings. No outcome scales were used. The presence of insight was taken as indicator of major improvement. One (3%) patient\'s relatives were contacted before admission; relatives of 45% of patients were contacted at some point after admission. Consulates of 52% of patients were contacted, most of them provided information, and in some cases supplied emergency travel documents. 10% were on Enhanced and 90% on Standard Care Programme Approach (CPA), the statutory planning and provision of mental health and social after-care. The MHT followed up 10% of patients after discharge (two on Enhanced and one on Standard CPA). Agreement with patients and/or relatives to a discharge plan was achieved in 87% of cases. 52% of patients were repatriated by the Hospital. These took place by air, accompanied by two members of staff, following the Hospital policy. 19% made their own arrangements to return home after discharge; relatives took 6% home. The MHT organised follow up for four patients, of these one decided to return home after the persecutory delusions had subsided. 81% of patients were supplied with medication (usually a two weeks supply) to take home. The average length of treatment in Hospital was 43.4 days per individual (range 1--365 days). One patient had been admitted before 1.1.1999 and was still admitted by the 31.12.1999. Patients under section 3 of the MHA spent the longest in Hospital (mean 91.5 days). Voluntary patients and those under section 2 spent similar numbers of days in Hospital (mean 22.3 and 21 days respectively). The total cost of the 31 admissions of overseas patients was GBP350, 600 (\$577, 490). The average individual cost of admission was GBP11, 116 (\$18, 230); the range was GBP200 -- GBP81, 000. The costs were for nursing care and repatriation. Other costs, such as translators, special nursing observations, or legal costs, were not included. Spearman\'s rank correlation test showed a highly significant positive correlation between length of admission and cost (P \< 0.01). Mann-Whitney U tests showed a significant difference in cost between patients with and without housing in England (P = 0.02), and between patients with and without financial help in England (P = 0.01). Patients with housing had a median cost of GBP4, 500 compared to GBP11, 000 for those without housing; patients with financial help had a median cost of GBP4, 500 compared to GBP12, 000 for those without help. Discussion ========== Overseas patients form a significant proportion (19%) of the admissions under the care of the MHT. It is estimated that overseas patients account for 10% of admissions in central London \[[@B8]\], whilst research in the same geographical area as this study report rates of 16% \[[@B9]\]. Studies in Jerusalem, where all psychiatric admissions of tourists are channelled into one central hospital, report an average of 40--50 admissions a year \[[@B4]-[@B10]\], whilst in Florence 107 tourists were admitted to a central hospital between 1978 and 1986 \[[@B5]\]. Homelessness in England among overseas patients in this study (61%) differs significantly from rates of homelessness among local (3%) (Parshall & Carranza, European Congress of Psychiatry, Madrid, 2001), and other patients admitted in Westminster -25% \[[@B11]\]. Geographical mobility has been linked to disruption in the continuity of care of patients, lack of accountability in census figures \[[@B12]-[@B14]\], and for service planning and provision \[[@B15]\]. These problems also apply to overseas patients, whose mobility is likely to have influenced the length of untreated illness and the level of contact with health services before admission. This may be illustrated by four overseas patients\' previous contacts with mental health services in London, which resembles the \"revolving door\" phenomenon, widespread in psychiatric services in England. A comparison of UK and European studies on attitudes towards the mentally ill describes British respondents as one of the most tolerant with little fear of the mentally ill, who consider mental illness as a universal condition, and favour community-based interventions as opposed to institutionalised care \[[@B16]\]. The perception of British attitudes towards mental illness, coupled with some familiarity with the English language, may have encouraged an \"international drift\" to the United Kingdom in individuals already unwell. In this study no specific factors could be identified as causes for overseas patients\' mental breakdown. Police involvement in the referral process is a significant predictor of admission to psychiatric hospitals \[[@B17]\]. Overseas patients assessments under section 136 of the MHA (Table [2](#T2){ref-type="table"}) are likely to contribute significantly to the rate of these referrals to the psychiatric services in Westminster reported as one of the highest in the United Kingdom \[[@B18]\]. Fisher\'s exact test showed a significant association between mode of contact with the MHT and MHA status on admission (P \< 0.001), with only 5% voluntary hospitalisations via the police, compared to 73% voluntary admissions via the MHT. The proportion of overseas patients\' admissions via the police (65%) (Table [2](#T2){ref-type="table"}) is similar to reports from London \[[@B19]\], and Jerusalem \[[@B4]\], and differs from rates reported among UK (24%) and local patients (6%) admitted under the MHT (Parshall & Carranza, European Congress of Psychiatry, Madrid, 2001). Offences by overseas patients leading to contact with the police were mainly behavioural and non-violent (e.g. bizarre conduct in public places, or not paying fees for services). One overseas patient was admitted via the Court Liaison Service, compared to the reported 15% of other admissions to the MHT from that service \[[@B20]\]. Overseas patients\' admissions under section of the MHA (77%) correspond with reports of admissions from Heathrow airport -81% \[[@B7]\], 69% \[[@B19]\], and the local Hospital -- 76% (Hospital MHA Officer\'s data), and differ from rates for England, where less than one third of admissions are under the MHA \[[@B21]\]. Rates of overseas patients with schizophrenia or related disorders (74%) (Table [3](#T3){ref-type="table"}) are comparable to figures from studies of travellers in New York -74% \[[@B2]\], London -50% \[[@B7]\], 46% \[[@B19]\], Jerusalem -63% \[[@B10]\], 85% \[[@B4]\], and Florence -68% \[[@B5]\]. These rates differ from figures of admissions with schizophrenic psychosis in inner London -30% \[[@B22]\], and Westminster -38% \[[@B23]\]. All overseas patients with schizophrenia presented with \"positive symptoms\" (delusions, hallucinations, and thought disorder). These are prevalent in urban populations with schizophrenia \[[@B12]\], and have been associated with high mobility \[[@B24]\] and homelessness \[[@B25]\]. \"Negative symptoms\" such as marked apathy, paucity of speech, blunting or incongruity of emotional responses (ICD-10: DCR-10) are associated with prefrontal dysfunction \[[@B26]\], and impairment of brain executive functions \[[@B27]\]. Patients with negative symptoms may find the planning and execution of foreign trips too challenging, and might also explain their absence in this study. Monopolar depression, personality disorder, neurotic or stress related disorders, or disorders other than the ones shown in Table [3](#T3){ref-type="table"}, were not found in this study. The absence of patients with dual diagnosis (substance misuse problems and mental illness in the same patient at the same time) contrasts with reports of 50% substance use among the mentally ill in the UK and substance misuse problems in 36% of patients with psychosis in London \[[@B28]\]. The low number of patients on the Enhanced component of the CPA, reflects the difficulties found on establishing responsibilities for the provision of services and care planning in overseas patients, and misrepresents the severity of the problems with which these patients present. A limiting factor is the difficulty in setting up care plans for patients whose aftercare is to be implemented by agencies abroad. Mental Health Review Tribunals and Managers\' Hearings discharged no overseas patients. Figures for England and Wales show discharge rates between 14.4% and 15.6% \[[@B21]\] and 7.0% in high security hospitals \[[@B29]\]. Discharge from hospital on grounds other than medical (e.g. request for repatriation by relatives) may explain the percentage of overseas patients discharged with little or moderate improvement in mental state (55%), and discharges from hospital under section of the MHA (48%) (Figure [1](#F1){ref-type="fig"}). Overseas patients\' refusal to return to their country, where a health and care system may or may not be in place, poses an ethical and legal challenge to services. Section 86 of the MHA (see Table [4](#T4){ref-type="table"}) is rarely used, perhaps due to the lengthy process and the varied factors to consider for its application. The Department of Health\'s recommendation to treat patients as close to home as possible \[[@B30]\], and the need for a \"substrate for health\" -looking not only at psychiatric interventions, but also at the individual\'s basic needs, housing, and a social network \[[@B31]\], need careful consideration when making decisions on repatriation. Since October 2000 contravention against the European Convention on Human Rights (ECHR) \[[@B32]\] can be challenged in UK courts. Problems with language translation and interpretation, usually evident on admission coincidental with an acute stage of patients\' mental state, are common when treating overseas patients. These can give rise to ethical and legal issues for example, when assessing capacity and consent to treatment. Current legislation states that all patients should be given information both orally and in writing on their legal position and rights (MHA)\[[@B33]\], of the reasons for their detention (ECHR \[[@B32]\], Mental Health Act Code of Practice \[[@B34]\]) but section 132 of the MHA is silent on this point, in a language that the person understands (ECHR)\[[@B32]\]. Failure to do so may be challenged under article 5(2) of the ECHR. Particularly relevant to overseas patients is the issue of deportation under section 86 of the MHA, which may be challenged under article 3 of the ECHR. Delays on discharging a patient because of failure to set-up aftercare services may breach article 5(4). Difficulty of access to information on the grounds for detention to apply for a hearing may breach article 6. Discrimination in the provision of services, such as individual therapies, multidisciplinary team involvement, or treatment in locked units may breach article 14 of the Act. The Eighth Principle of the Data Protection Act 1998 -personal data should not be transferred outside the European Economic Area unless that country ensures its adequate protection \[[@B35]\], is difficult to guarantee when dealing with foreign agencies on behalf of patients, and may give rise to breach of article 8(2) of the ECHR. Conversely, the lack of consultation and provision of information to a nearest relative on patients\' admissions may be challenged under the same article 8(2). The Council of Europe determines that family and other people close to a patient should be consulted on involuntary placement and treatment \[[@B36]\]. The MHA provides legislation on ascertaining the nearest relative of patients from England and Wales, but gives no indication on how to proceed in the case of foreign nationals. The lack of nearest relative in overseas patients has ethical and legal implications, particularly on issues of risk assessment, information about their power to discharge a patient, to delegate their role, advanced directives, and repatriation. At present, consular representations play, to a major or lesser degree and at an informal level, a role in some ways similar to that of a nearest relative, which is not recognised by mental health law. Contact with embassies is described as ranging from lack of involvement, particularly when patients are in need of repatriation \[[@B8]\], to full cooperation with contact and liaison with services abroad, particularly from European embassies \[[@B37]\]. A way forward for future legislation could be for the consular representations to take formally the role of nearest relative, which could revert back to the patient\'s relatives when practicable. The Expert Committee Review of the MHA recommends that the powers of the nearest relative should be reduced and for the provision of advocates independent from the service provider \[[@B38]\]. Proposals in the Government\'s Draft Mental Health Bill include the patient\'s choice of a \"nominated person\" to replace the figure of nearest relative, and a duty to provide sufficient advocates \[[@B39]\]. A feasible option would be for consulates to fulfil the role of nearest relative, which would automatically encompass the role of advocate; the advantages include: • The prompt nomination of a nearest relative when it is not possible to identify one, or when they have been displaced of their role by the Court. • To prevent problems with confidentiality e.g. when trying to contact relatives, who may not speak English, and services abroad. • Provisions under the MHA do not apply to voluntary patients; thus they may receive less information on issues related to their admission. In these patients, as in detained patients, consulates could be useful in establishing links locally, with services abroad, and as a reference point e.g. in overseas patients missing in their country who present to health services abroad. • Admissions under the MHA require the involvement of social services. There may be a negative perception or reluctance to accept the input from social services by patients when the Court appoints a social worker as the nearest relative, e.g. when a relative cannot be identified. • As advocates, consulates are better prepared to assist patients with lessening the impact of transcultural barriers, relaying information, which could assist patients on making decisions e.g. on medico-legal matters. • From the patient\'s perspective, familiarity with the person representing the nearest relative may reassure them on issues of the service\'s independence and lack of bias, leading to better co-operation with their treatment and care plans. The pressure on mental health services in inner London may be a consequence of changes in patients\' characteristics- younger, increasingly mobile, more likely to be unattached and unemployed \[[@B40]\], features that also correspond with the average patient\'s profile in this study (Table [1](#T1){ref-type="table"}). Furthermore, patients with these characteristics who are less able to live independently increase the costs of care \[[@B41]\]. Likewise, overseas patients have a high degree of dependence on care services, and their high mobility is likely to have an influence on levels of provision and possibly on the reported underestimate of needs in inner London by measures of service requirement, such as the Mental Illness Needs Index (MINI) \[[@B42]\]. Mobility is also likely to be an obstacle for overseas patients\' inclusion in audit, service planning, and mental health strategies aimed at improving standards of care. Conclusions =========== The sample size in this study is small, which makes our findings difficult to generalise. The figures in this paper represent the results of one mental health team, among the more than 50 mental health teams in central London, which suggests a higher scale to this problem. Research is much needed in this area. Our findings replicate at international level the \"social drift\" seen in people affected by psychiatric morbidity into deprived inner city areas \[[@B43]\]. A high proportion of patients in this study, particularly patients with schizophrenia, fall into what has been described as \"double drift\" \[[@B44]\], by virtue of moving from one country to another, and then into a socially isolated urban area where they become part of a low socio-economic group. High mobility among overseas patients had a marked impact on homelessness, contact with services, care and service planning and delivery, Mental Health Act reviews\' outcomes and status on admission and discharge. Psychotic disorders with positive symptoms were prevalent. Police involvement in the referral process was high, correlated positively with the high rate of involuntary admissions, and negatively with the type of offences attributed to these patients. A highly significant correlation was observed between length of admission and cost, with a significant cost difference between overseas patients with and without social and financial support. An enhanced role for consulates as representative bodies for overseas patients receiving psychiatric treatment needs to be explored and formalised. Service providers need mechanisms better able to identify and to evaluate overseas patients\' needs. This would allow patients\' data to count in audit, research, and financial planning; thus facilitating their inclusion in user and information groups, and strategies aimed at improving standards of care. Recent changes to the Charging Regulations for treatment under the NHS of non-resident patients \[[@B45]\] need to take into account the characteristics and problems common to overseas patients with psychiatric illnesses and to adapt legislation accordingly. As the boundaries between domestic and international health matters become blurred, countries need to pursue a global integration of policies aimed at helping people with mental illness in general, and patients with high mobility in particular. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= FJC conceived the study, collected data and drafted the manuscript. AMP participated in the design of the study and reviewed the manuscript. Both authors read and approved the final manuscript. Acknowledgements ================ Helen Goodman, Librarian, for help with literature research. Statistical analysis supported by the West London Mental Health Research and Development Consortium.	PubMed Central	2024-06-05T03:55:56.266025	2005-2-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088010/", "journal": "Ann Gen Psychiatry. 2005 Feb 14; 4:4", "authors": [ { "first": "Fredy J", "last": "Carranza" }, { "first": "Alice M", "last": "Parshall" } ] }
PMC1088011	Background ========== The use of mood stabilizing antiepileptic drugs has increasingly been explored for the treatment of different psychiatric conditions. Topiramate is a novel neurotherapeutic agent approved in more than 75 countries for adjunctive treatment for refractory partial-onset seizures or primary generalised tonic-clonic seizure in adults and children over 2 years of age and migraine prophylaxis in USA. Several mechanisms of action of topiramate support the hypothesis for its putative actions in bipolar affective disorders, unipolar depression, schizophrenia, posttraumatic stress disorder, disordered eating behaviour. This article reviews the pharmacology of topiramate and describes adverse events and the outcomes observed in published and unpublished studies. Particular interest is focused on topiramate related weight loss and its clinical implications. Pharmacokinetic and pharmacodynamic profile ------------------------------------------- Topiramate is a sulfamate substituted, derivative of the monosaccharide D-fructose \[[@B1]\]. It is absorbed in 1--4 hours, its oral bioavailability is about 80% and its plasma protein binding is 15%. It pharmacokinetic profile is linear in relation to dose \[[@B2]\]. It does not affect liver enzymes, it is excreted unchanged in the urine, and has a high therapeutic index \[[@B3]\]. In renal impairment, the clearance of topiramate is decreased and elimination half-life is prolonged, usually between 19 and 23 hours \[[@B4]\]. Moderate, not clinically significant, increases in plasma concentrations have been observed in the presence of hepatic disease \[[@B2],[@B4]\]. It is not extensively metabolised, and six inactive metabolites have been identified \[[@B4]\]. Topiramate half-life (18--23 hrs) is decreased by carbamazepine \[[@B5]\]. It may compromise the efficacy of oral contraceptive agents by reducing mean total exposure to the estrogen component \[[@B6]\]. Similarly to carbamazapine and valproate, topiramate reduces the seizure threshold and the after-charge duration in the amygdale-kindled rat \[[@B7]\]. It may increase cerebral GABA concentrations in humans \[[@B8]\], enhancing the inhibitory GABAergic transmission by binding to allosteric GABA-A receptors, probably through a non-benzodiazepine mechanism and second-messenger systems \[[@B9],[@B10]\]. Also, topiramate may inhibit brain glutamate release, by antagonising α-amino-3-hydroxy-5-methyl-4-isoxazolapropionate (AMPA) kainate type of glutamate receptors, and may inhibits NA (+) and L-type Ca (2+) channel neuronal activities \[[@B11],[@B12]\]. Topiramate is also suggested to be an inhibitor of specific carbonic anhydrase isoenzymes \[[@B13]\]. Rationale for evaluating topiramate in psychiatric disorders ------------------------------------------------------------ The use of topiramate in bipolar spectrum disorders is based on the putative shared biological mechanism between epilepsy and bipolar disorders suggested by the amygdala-kindled seizures in animal models \[[@B14]-[@B16]\] and the high rate of co-morbid psychiatric conditions in epilepsy \[[@B17]\]. However, there is inadequacy of current treatment strategies \[[@B18]\]. The efficacy of lithium, valproate, and carbamazapine in prophylaxis of bipolar spectrum disorders is rather modest \[[@B19]-[@B22]\]. Mixed or rapid cycling disorders are particularly characterised by a poor response to lithium treatment, which reaches 72--82% \[[@B23],[@B24]\]. Twenty-five to fifty percent of patients need reduction or discontinuation of lithium therapy due to adverse effects \[[@B25]\] and up to 55 % of patients develop resistance to lithium after 3 years of treatment \[[@B26]\]. Pharmacological interventions are also limited in bipolar depression extensively treated with antidepressants \[[@B27]\] in the absence of replicated controlled studies \[[@B21]\], and with the recognised risk of induced hypomanic switching or cycles acceleration \[[@B28]-[@B30]\]. Lamotrigine has demonstrated stabilising properties in bipolar I depression and rapid cycling bipolar II disorder \[[@B31],[@B32]\] highlighting the role of newer mood stabilisers in the treatment of this condition. In unipolar major depression the role of double-blind placebo controlled trials confirm that lithium is effective in about 40--50% of patients and there is scope for the use of mood stabilising agents such as carbamazapine and sodium valproate \[[@B33]\]. In schizophrenia, the postulated action of anticonvulsants is based on some evidence supporting a reciprocal interaction between glutamatergic and dopaminergic systems. It is postulated that the striatum, which has rich D1 and D2 dopamine innervation, receives cortical, limbic and thalamic excitatory glutamatergic afferents. Striatal activation by glutamate leads to inhibition of the thalamic sensory outflow to the cortex. This effect seems to be mediated by inhibitory gabaergic neurons acting via thalamic circuits \[[@B34]\]. Phencyclidine binds non-competitively to a site adjacent to N-methyl-D-aspartate (NMDA) receptor of glutamate exercising an inhibitory effect that can mimicry schizophrenia. This model constitutes the theory for the \'hypothesis of glutamatergic hypofunction\' based on receptor hypofunction or \'glutamatergic deficiency\' in the pathophysiology of schizophrenia \[[@B35]-[@B37]\]. In humans with schizophrenia, elevated levels of N-acetyl-aspartyl-glutamate, a naturally occurring acidic dipeptide, could dampen or antagonize NMDA receptor mediated neurotransmission. Elevated levels of N-acetyl-aspartyl-glutamate could rise from diminished activity of glutamate carboxypeptidase II, a hydrolytic enzyme enriched at glutamatergic nerve terminals and located on the membrane of astrocytes \[[@B35],[@B38]\]. Topiramate mechanisms of action could optimise the imbalanced availability of glutamate and/or GABA in the subcortical circuitation. Posttraumatic stress disorder can be a difficult condition to treat especially if the course is chronic \[[@B39]\]. Current pharmacological interventions are limited to serotoninergic reuptake inhibitors (SSRIs). Hypothesis on the aetiology of posttraumatic stress disorder, have suggested that after exposure to traumatic events, limbic nuclei may become kindled and sensitized. Consequently, drugs known to have anti-kindling or anticonvulsant effects might have potential in the treatment of posttraumatic stress disorder \[[@B40]\]. Carbamazapine and valproate may be effective \[[@B41],[@B42]\]. Particularly carbamazapine has shown efficacy in reducing re-experiencing and arousal symptoms whilst valproate decreased avoidance/numbing and arousal symptoms \[[@B43]\]. Most recently lamotrigine has also shown some efficacy \[[@B44]\]. The use of topiramate in eating disorders derives from the observation of appetite suppression and weight loss in controlled trials in patients with epilepsy \[[@B45]\]. Animal models suggest that stimulation of the lateral hypothalamus, by glutamate agonists (like kainate/AMPA agonists), causes an intense rapid dose-dependent increase in food intake \[[@B46],[@B47]\]. Antagonists of kainate/AMPA glutamate receptors like topiramate might contribute to suppress appetite and to regain control over eating, a typical feature observed in eating disorders \[[@B48]\]. Clinically, this would be in agreement with EEG abnormalities found in bulimia nervosa. Another postulated mechanism might be linked to a recent observation that topiramate down-regulates neuropeptide Y1 and 5 receptor subtypes in rats \[[@B49]\]. Current pharmacological approaches to treatment of binge eating disorders are limited to SSRIs \[[@B50]\] and imipramine \[[@B51]\] whereas desipramine \[[@B52]\] and d-fenfluramine \[[@B53]\] have not been associated with weight loss. Similarly, in bulimia nervosa, SSRIs constitute the main pharmacological resource \[[@B54]\], with some possible effectiveness for carbamazapine \[[@B55]\] and phenytoin \[[@B50],[@B56],[@B57]\]. In alcohol dependence, antiepileptic medications share neurochemical effects with alcohol by inhibiting neuronal excitation. Carbamazapine, gabapentin, and valproic acid have been reported to reduce alcohol consumption \[[@B58]\]. Chronic alcohol intake is linked to decreased GABA receptor activity in the ventral tegmental area with disinhibition of dopaminergic neurons \[[@B59]\]. Similarly, hippocampal and cortical GABA neurons projecting to the midbrain might facilitate dopaminergic neurotransmission in the midbrain at glutamate binding sites \[[@B60]\] such as kainate/AMPA receptors \[[@B61]\]. The putative efficacy of topiramate in the treatment of alcohol dependence is based on reversing chronic changes induced by alcohol resulting in dopamine-facilitated neurotransmission in the midbrain. In psychiatry, drug induced severe obesity plays an important role \[[@B62]\] and substantive weight gain has been described with several psychotropic medications \[[@B63]-[@B65]\]. Obesity is associated with an increase risk of co-morbid medical conditions such as hypertension, diabetes and cardiovascular disease \[[@B66]\]. Diabetes mellitus reaches nearly 10% prevalence among hospitalized subjects with bipolar disorder in USA \[[@B67]\]. Topiramate induced weigh loss in the 5--10% range is associated with significant reduction in blood pressure and changes in total cholesterol, low-density lipoproteins and triglycerides \[[@B68]\]. There are no clear mechanisms underlying weight changes but it may be dependent of glycemic control as suggested by Chengappa et al. \[[@B69]\]. Methods ======= A comprehensive search from a range of electronic databases, including BNI, CancerLit, Cochrane Library, EMBASE, Medline, Psychinfo, and Pub MED was conducted for the period from the introduction of topiramate to December 2003. Key words used to identify the studies were: TOPIRAMATE or ANTICONVULSANTS and PSYCHIATRIC DISORDERS, PSYCHIATRY, PSYCHOSIS, AFFECTIVE DISORDERS, EATING DISORDERS, SCHIZOPHRENIA, SCHIZOAFFECTIVE DISORDERS. The search was also complemented by manual search of bibliographic cross-referencing. Researchers who had expressed an interest in the subject were contacted for any non-published information. Janssen-Cilag Ltd medical information was also contacted. There was no restriction on the identification of studies in terms of publication status, language and design type. Papers were identified if presented original data and addressed the question, \'use of topiramate in treating psychiatric conditions\'. Studies were screened for design type, diagnosis according to diagnostic criteria, topiramate dose, titration regime, response onset, response rate, duration of treatment, outcome measures, and adverse events. Presence of weight loss (preferably expressed as ≥5% reduction in baseline weight) was also considered. Response was preferably indicated by significant score reduction in rating scales or objective measures. Randomised controlled studies if available were considered primary source of evidence, followed by naturalistic studies, case series and case reports. Reports or posters presented to meetings and subsequently re-considered in larger numbers or published were excluded. Results ======= Thirty-two published reports met criteria for inclusion, 4 controlled and 28 uncontrolled studies (see [Additional file-table 3](#S1){ref-type="supplementary-material"}). Five unpublished controlled studies were identified in the treatment of acute mania ([table 2](#T2){ref-type="table"}). Details are given below. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Characteristics of the studies included in the review ::: Condition Number of studies Design Outcome ------------------------------------------------------------------------- ----------------------- ---------------------------- ------------- Bipolar disorders Bipolar mania 8 5 controlled (\) Negative 3 open label (70--72) Positive Rapid-cycling bipolar disorders* 1 Open label, add-on (75) Positive Adjunctive therapy (refractory bipolar disorders) 12 Open label (76--87) Positive Bipolar depression 2 1 Controlled, add-on (88) Positive 1 Open label, add-on (89) Positive Bipolar disorders in children and adolescents as adjunctive treatment 1 Open label, add-on (90) Positive Unipolar depression 2 1 Case report (91) Negative 1 Chart review (92) Positive Schizophrenia, schizoaffective disorders and psychosis unspecified 3 2 Case series (93, 94) Negative 1 Case report (95) Positive Eating disorders and disordered eating 4 2 Controlled (96, 97) Positive 1 Open label, add on (98) Positive 1 Case series, add on (99) Positive Posttraumatic stress disorder 1 Open label, add on (100) Positive Alcohol dependence 1 Controlled (101) Positive Gilles de la Tourette\'s syndrome 1 Case series (102) Positive Emotional unstable personality disorder 1 Case reports (103) Positive (\) Unpublished ::: Bipolar disorders ----------------- ### Bipolar mania Following encouraging results from preliminary reports in acute mania \[[@B70]-[@B72]\], topiramate was compared with placebo in one double-blind randomised trial \[[@B73]\]. Two different dosages of topiramate (250 and 500 mg/day) were studied in a 3-week trial among hospitalised patients. The final analysis found no significant differences in efficacy in the three groups. Four subsequent large unpublished placebo controlled studies, unavailable for review, failed to demonstrate efficacy of topiramate in mania compared to placebo, leading to the discontinuation of development programs \[\[[@B74]\]; Calabrese, personal communication\]. ### Rapid-cycling bipolar disorders Kusumakar et al.\[[@B75]\] studied 27 women with ultra rapid, ultradian, and chaotic biphasic bipolar disorder type I/II refractory to treatment for 16 weeks and more than 29% weight gain over the previous 24 months. The study had a prospective open label, add-on design. Topiramate was introduced at a dose of 25 mg/day, and increased by 25 mg/day every 5--7 days until clinical response or tolerability was reached. The dose range was 100--150 mg/day. Rating scales used in this study were the Hamilton depression rating scale, 21 items (HAM-D-21), the Young mania rating scale (YMRS), and daily assessments of mood, sleep pattern, and weight loss. Among the 23 patients who completed the study, clinical response was noted within 12 weeks for 15 patients who remained euthymic for at least 4 weeks. Weight loss \>5% was recorded in 9 patients and of 1--4% in 5 patients. The rest of the subjects experienced no weight change and in 1 case weight gain was recorded. Only 4 patients discontinued the study because of adverse events (drowsiness and dizziness, ataxia, confusion, inability to concentrate). Adjunctive therapy in treatment-refractory bipolar disorders ------------------------------------------------------------ Marcotte et al.\[[@B76]\] in an open-label study examined retrospectively 58 in-out patients with different psychiatric disorders, refractory to conventional mood stabilisers, and with psychiatric and medical co-morbid conditions. Forty-four patients had rapid cycling bipolar disorder (manic, hypomanic and mixed), 9 had schizoaffective disorder, 3 had dementia, and 2 had psychotic illness. The range of duration of psychiatric illness was from 7 months to 40 years. The mean duration of topiramate treatment was 16.0 weeks with a mean dosage of 200 mg/day (range 25--400 mg/day). The initial dose was 25 mg twice daily, slowly increased by 50 mg every 7 days. Response was regarded as \'marked\' or \'moderate\' improvement based on a Likert global assessment scale including quality of sleep, appetite, mood, and concentration during therapy. Twenty-three (52%) of the 44 rapid cycling bipolar disorder patients and 36 (62%) of the whole sample showed \'marked\' or \'moderate\'. Six (46%) of the 13 patients with rapid cycling bipolar disorder and substance misuse showed marked or moderate improvement when topiramate was added. Adverse effects were minor and 6 (10%) patients discontinued due to adverse events (delirium, grand mal seizures, increased panic attacks, confusion, frequent bowel movements, nausea, somnolence, fatigue, impaired concentration and memory, paraesthesias). In a larger cohort continuation of open treatment with topiramate showed additional clinical improvement with longer drug exposure \[[@B77]\]. Chengappa et al.\[[@B78]\] examined prospectively in a 5-week naturalistic study 18 patients with a diagnosis of bipolar disorder type I (manic, hypomanic, mixed phase and rapid cycling) and 2 patients with schizoaffective disorder (bipolar type), all refractory to previous mood stabilizing therapies. Topiramate was added on to existing pharmacotherapy and it was initiated at a dosage of 25 mg/day, increased by 25--50 mg/day every 3--7 days. The target dose was in the 100--300-mg/day range. The YMRS, HAM-D-21, and the clinical global impression scale for bipolar disorder (CGI-BD) were used in the evaluation. Response was defined as 50% or greater reduction in the total Y-MRS scores and CGI-BD score of \'much or very much improved\'. Twelve of the patients (60%) responded to topiramate, within 2--4 weeks after treatment initiation. Progressive decline in weight and body mass index (BMI) occurred during the course of therapy. Topiramate was well tolerated and adverse events were minor. The average weight loss was 1.5--2 lb/week. Subjects with BMI of 30 or more (i.e. obese) lost more weight. McElroy et al.\[[@B79]\] studied 56 outpatients participating in the Stanley Foundation Bipolar Outcome Network in a prospective study with an open label add-on design. Patients had bipolar disorder type I/II, psychotic disorder not otherwise specified and schizoaffective disorder bipolar type, inadequately responsive or poorly tolerant to one or more standard mood stabilizers. The YMRS, CGI-BD and the Inventory of Depressive Symptoms (IDS) were used in the assessments. The baseline YMRS reflected only mild mania. The initial dose was 25--50 mg/day, given either at night or in divided doses, subsequently increased every 3--14 days by 25--50 mg/day, according to patients\' response and side effects. The maximum dose utilised was 1200 mg/day. The mean dose at 10 weeks was 193.2 mg/day (SD = 122.0) and 244.7 mg/day (SD = 241.7) at last evaluation. Thirty manic and 11 depressed patients completed the 10 weeks acute phase, of which 19 manic (63.3%) and 3 depressed (27.3%) were \'much or very much improved\' so regarded as responders, according to YMRS, CGI-BP-Mania and IDS but not CGI-BP-Depression. Thirty-seven patients continued open maintenance treatment with topiramate for a mean ± SD of 294.6 ± 145.3 days (i.e., more then 7 months): 22 manic, 5 depressed and 10 euthymic patients. At last evaluation, 12 manic patients (55%) were rated as much or very much improved and 10 minimally or no changed, 1 depressed patient was rated very much improved and 4 displayed no or minimal change, 9 euthymic displayed minimal or no change and 1 had worsened with mixed symptoms. In total 29 (52%) discontinued topiramate during the acute and maintenance phase (up to a year). The main reasons for discontinuation were increased depressed (N = 7) or hypomanic/manic (N = 4) symptoms, discontinuation of medication (N = 1) and side effects (N = 6). Ten patients (18%) discontinued topiramate because of side effects. Topiramate was associated with reduction in BMI and body weight. Patients who began topiramate for depressive symptoms or relative euthymia did not display notable changes in ratings at most time points. Sacks et al.\[[@B80]\] treated 14 patients with treatment resistant bipolar disorder and a variety of co-morbid conditions for a mean of 22.4 +/- 22.0 weeks with adjunctive topiramate in a retrospective trial. The mean dose of topiramate was 50 mg/day (SD = 27.4). Among the 11 patients who remained on treatment for longer than 2 weeks, 4 experienced decreased severity of bipolar illness by more than 1 CGI score and 8 experienced significant improvement in their primary co-morbid condition. Four patients with BMI of 28 or more experienced a mean weight loss of 13.5 +/- 7.4 kg whilst on topiramate. Discontinuation occurred in 5 patients due to adverse effects (paraesthesias, rash, cognitive impairment, sedation) and in 2 due to lack of efficacy. Eads et al.\[[@B81]\] studied 17 treatment resistant patients with bipolar disorder type I (N = 11) and II (N = 3). The study was retrospective in design and with a mean duration of 22.4 (SD = 22.0) weeks. Patients were evaluated with the Global Assessment of Functioning (GAF) scores. Topiramate was added to other medications and titrated to a mean dose of 826 mg/day in divided doses. Nine patients completed the study and 8 patients discontinued due to adverse effects (cognitive impairment, sedation, paraesthesias). All nine patients responded to topiramate with 8--20 improvement on the GAF scale. Eight experienced clinically significant improvement in their primary co-morbid condition as measured by the Clinical global impression scale for improvement (CGI-I) (anorexia nervosa N = 1, bulimia N = 3, obesity N = 1, obsessive compulsive disorder N = 1, Tourette\'s N = 1). Patients with BMI of 28 or more (N = 4) experienced a weight loss of 29.75 lb (SD = 16.29). Ghaemi et al.\[[@B82]\] in a retrospective open label study reviewed 76 charts of outpatients with refractory bipolar disorder type I/ II or psychotic disorder non otherwise specified (depressive phase N = 33, rapid cycling N = 24, mixed episodes N = 8 and prophylaxis N = 8, hypomania N = 3). In all the patients topiramate had been added on or used in monotherapy. The main dose of topiramate used was 96.1 mg/day (SD = 94.19) (range 12.5--400 mg/day) for a mean duration of 17.5 (SD = 16.7) weeks (range 0.5--65 weeks). Response was measured with the CGI-I rating scale as \'moderate\' to \'marked\' improvement. The overall response rate to topiramate was 13.2% (10/76). Response rates remained similar when assessed on indication of treatment. Responders received a higher dose of topiramate (180 mg/day, SD = 120.1) than non-responders (83.2 mg/day, SD = 83.7, p = 0.002) and higher in the high rather than the low dose group (p = 0.04, Fischer\'s exact test). Topiramate was not higher in patients receiving monotherapy (N = 6). Response rate between subjects receiving mood stabilisers (p = 0.27 Fischer\'s exact test) or antidepressant (p = 0.48 Fischer\'s exact test) and those who did not wasn\'t significant. Weight estimates were based on patient self-report. Weight loss was experienced by 51.6% of the sample with 14.2 lb (SD = 6.2) (range 5--25 lbs). Topiramate dose was also higher in those subjects who lost weight (138.3 mg/day) than in those who did not (70 mg/day, p = 0.007) but not the amount of weight (p = 0.49). There was no difference if concomitant medication were used (p = 0.43). Side effects were reported by 81.6% of the sample. Topiramate was discontinued in 51.3% (N = 39) of the sample with 27 (69.2%) for side effects (paraesthesias, nausea, fatigue, insomnia, slowed thinking, sedation, ataxia, headache, agitation, frequent peristalsis) and 7 for lack of efficacy. Vieta et al.\[[@B83]\] designed a prospective, 6-week open label study with an add-on design. The authors studied 21 patients with poor response or intolerance to mood stabilisers and with a diagnosis of bipolar disorder type I/II in a manic (N = 9), mixed (N = 2), hypomanic (N = 3) and depressed (N = 6) phase or schizoaffective manic (N = 1). The YMRS, HAM-D-17 and CGI rating scales were used. At study entry, patients had a minimum score of 12 on YMRS and HAM-D and a minimum score of 4 on CGI. Topiramate was introduced at the dose of 25 mg/day and increased by 25--50 mg every 3--7 days to a mean dose of 158 mg/day. At end point, among the 15 patients who completed the study, 6 (28.5% by intention to treat) were responders with 50% or greater decrease in YMRS or in HDRS-D-17 scores and 2 or more in the CGI-BP. Patients in the depressed phase only obtained a reduction equal to 50% in HDRS-17. Six patients discontinued for lack of efficacy and side effects (paraesthesias, impaired concentration, anxiety) (N = 1), poor compliance (N = 1) and loss of follow up (N = 3). Ten patients experienced moderate weight loss. Saxena et al.\[[@B84]\] assessed the efficacy of topiramate as adjunctive treatment in 9 bipolar disorder patients resistant to conventional mood stabilisers, in a prospective 10--24 week open label trial. Significant decrease in YMRS and HAM-D were observed in four patients. Decreases in CGI-I in the Global assessment scale (GCI-S) scores of at least one point from baseline to endpoint were noted in all patients and no relapses were observed. Topiramate was titrated according to efficacy with a mean dose at endpoint of 488 mg/day. It was well tolerated at doses of up to 600 mg/day. The mean weight loss during the follow up period was 5.39 kg. Only one patient discontinued due to side effect (anxiety, sleep disturbance, lack of libido). Vieta, Torrent et al.\[[@B85]\] completed a 6-month open trial with 34 treatment resistant bipolar patients (type I = 28, type II = 3, not otherwise specified = 2 and schizoaffective = 1) in different phases (manic = 17, depressive = 11, hypomanic = 3, mixed = 3). Topiramate therapy was added on current medication and the dose titrated slowly. The dose at end point was 202 mg/day (SD = 65). Outcome measures included the YMRS, HAM-D, and CGI for severity. Twenty-five patients (74%) completed the study, 9 subjects discontinued due to lost of follow up (N = 4), worsening of symptoms (N = 2), side effects (N = 1), hospitalization (N = 1) and non-compliance (N = 1). Response occurred within 2--6 weeks. Fifty-nine percent of manic patients and 55% of depressed patients responded to the drug by intention to treat analysis expressed as significant reduction in rating scales. Only one patient discontinued due to side effects (paraesthesias) and topiramate was generally well tolerated. Vieta, Ros, Valle et al.\[[@B86]\] evaluated 61 refractory bipolar patients, in a 12-week preliminary multicentre study. Outcome measures included the YMRS, HDRS and CGI-BP. The mean YMRS at baseline was 27.8. Among the 55 patients who completed the study, 43 patients (70%) were considered responders with 50% or more reduction in YMRS score. Also 25 patients (41%) met criteria for remission with YMRS score of 8 or less. Weight loss was recorded in 24 (39%) patients. Those with the highest BMI at baseline (\>40) experienced the greatest weight loss (mean 3.3 kg) during the follow up. Highly significant reduction in HDRS (p = 0.004) and CGI-BP (p \< 0.0001) from baseline to endpoint were also noted. Only 6 patients discontinued the study due to loss of follow up (N = 2), non-compliance (N = 2), lack of efficacy (N = 1), and side effects (paraesthesias) (N = 1). The mean topiramate dose at endpoint was 214 mg/day. McIntyre et al.\[[@B87]\] enrolled 109 subjects with bipolar disorder type I/II in manic (N = 3), hypomanic (N = 18), mixed (N = 33), depressed (N = 40), rapid cycling (N = 15) phases, resistant to conventional antipsychotics in a 16-week, add-on, naturalistic trial. Different co-morbid disorders were present in 24 subjects. The baseline YMRS score was 13 or greater, the Montgomery and Asberg depression rating scale (MADRAS) was 12 or greater, and the CGI-S was \'moderate\', \'marked\' or \'extremely severe\'. Topiramate mean dose was 140.8 mg/day (range 25--400 mg/day). Seventy patients completed the study but 99 were evaluable at end point. Seventy percent of subjects (N = 69) responded to topiramate treatment with a reduction of 50% or more on YMRS score. Twenty-five subjects obtained remission at endpoint expressed as YMRS of 8 or less. The MADRAS score decreased in the all population studied throughout the study period, with a more pronounced decrease in subjects not on antidepressants (N = 57). Sixty percent (N = 59) of patients responded to topiramate according to MADRAS expressed as 50% or more reduction in score and 37 obtained remission defined as a score of 12 or less. Thirty-nine subjects discontinued because of adverse events (paraesthesias, nausea, fatigue, somnolence, frequent peristalsis, blurred vision, headache, dizziness) (N = 12), lack of efficacy (N = 6), missed doses (N = 3), protocol violation (N = 5), withdrew consent (N = 9), lost at follow up (N = 3), other reasons (N = 1). Adverse events occurred in 131 patients. Tremor, scored with the VAS severity scale (1--10 range), showed a reduction in severity from 3.84 at baseline to 2.06 at week 16 (p \< 0.001). Subjects\' satisfaction with treatment was also considered with only 10% of patients rated \'completely dissatisfied\', \'somewhat dissatisfied\', \'neither satisfied nor dissatisfied\'. Weight change was noted in 107 subjects: 65 lost weight, 24 gained weight and 18 maintained their weight. It was not evaluable in 2 patients. The mean weight change at endpoint was -- 1.8 Kg (p \< 0.001). Bipolar depression ------------------ McIntyre et al.\[[@B88]\] conducted a study where topiramate was added to current medication and randomly compared to bupropion in the treatment of 36 subjects for bipolar disorder type I/II in depressive phase. This was an 8 weeks single blind (rater blinded) study developed in outpatients setting, with intent to treat analysis. Topiramate was introduced at the dose of 50 mg/day and titrated every two weeks until clinical response was obtained to a maximum of 300 mg/day. The mean dose of topiramate was 176 mg/day (SD = 102 mg/day). Fifty-six percent of patients on topiramate and 59% for bupropion obtained 50% or more decrease from baseline in HDRS-17 scores. Response to treatment ranged from two to four weeks. Significant reduction in YMRS and CGI-I scores were also observed at week-8 similarly in both the topiramate and the bupropion SR groups with no significant difference between the two. Weigh loss was recorded in both treatment groups; the mean weight loss was of 1.2 Kg in the bupropion SR group and 5.8 Kg in the topiramate group. Adverse events were reported in eleven (61%) patients receiving topiramate and nine (50%) receiving bupropion SR. In total 8 of patients receiving topiramate and 5 of patients in the bupropion SR group discontinued prematurely. Six patients in the topiramate group and 4 patients in the bupropion SR group discontinued for adverse events (topiramate group: paraesthesias, nausea, sweating, decreased/increased appetite, anxiety, slow memory, word finding difficulty, tremor, blurred vision and headache). The two further discontinuations in the topiramate group were attributable to lack of efficacy (N = 1) and withdraw of consent (N = 1). Hussein et al.\[[@B89]\] studied the efficacy of topiramate as adjunctive treatment with a 3-year, naturalistic study in patients with bipolar disorder type I (N = 65) and II (N = 18) in a moderately severe depressive phase, refractory to mood stabilisers. Depressive symptomatology was assessed with the HAM-D-17 scale. Topiramate was commenced at a dose of 50 mg/day and titrated every 2 days to a mean dose of 275 mg/day (range 100--400 mg/day). Forty-one patients completed the study but 65 were evaluable with 35 (54%) who showed great improvement (HAM-D score at endpoint 0--5) and 6 (9%) partially responded (HAM-D score 6--10). The response occurred within the first 4 weeks of treatment. Nineteen patients (29%) abandoned the study because adverse events (paraesthesias, nausea, dizziness). The average weight loss in 36 months was 38 pounds. Bipolar disorders in children and adolescents as adjunctive treatment --------------------------------------------------------------------- DelBello and associates\[[@B90]\] evaluated topiramate as open label, adjunctive treatment for children and adolescents with bipolar disorder type I/II for 4.1 months (SD = 6.1). The charts of 26 subjects were retrospectively reviewed using the CGI and CGA scales separately for mania and overall bipolar illness. The dose at end point was 104 mg/day (SD = 77). Response rate defined as improvement of 2 or more points on the rating scales was 73% for mania and 62% for overall bipolar disorder. No serious adverse events were reported. Unipolar depression ------------------- Gordon and Price\[[@B91]\] reported topiramate lack of efficacy in a case report of recurrent major depression. Topiramate was used as adjunctive treatment for 8 weeks at a dose of 300 mg/day. Anxiety and depressive features supervened leading to discontinuation. A significant weight loss of 15 lb occurred. Carpenter and associates\[[@B92]\] reviewed the charts of 16 females patients with treatment resistant unipolar depression and obesity (mild to moderate) treated with open label adjunctive topiramate. Self reported symptoms and clinician ratings were assessed regularly. Only 36% of patients were considered responders at 5.5 weeks (SD = 1.2) and 44% at end point 17.7 weeks (SD = 13.4). The initial dose of topiramate was 25--100 mg daily, increased variably according to the individual\'s symptomatology and side effects; the final dose was 277 ± 101 mg/day (range 100--400 mg/day). Four subjects discontinued due to adverse events (paraesthesias, memory concerns, lack of concentration, dysgeusia). Body mass index decreased significantly with a mean weight loss of 6.1 % (SD = 8.2). Schizophrenia, schizoaffective disorders and psychosis unspecified ------------------------------------------------------------------ Millson et al.\[[@B93]\] in a case series treated 3 men and 2 women with chronic schizophrenia adding topiramate to current medication. The initial dose was 50 mg/day and titrated at 50 mg/week to a mean dose of 250 mg/day (range 200--300 mg/day). Current medication dose was held constant. Positive and negative symptoms were monitored with the Positive and Negative Syndrome Scale for schizophrenia before commencing topiramate and a month after the maximum dose was administered. A deterioration of both positive and negative symptoms was noted in all the subjects. Dursun and Deakin\[[@B94]\] augmented antipsychotic medication with either topiramate or lamotrigine in 26 outpatients with treatment resistant schizophrenia. The case series had an open label, add-on design with 24-week duration. Psychopathology was assessed periodically with the Brief Psychiatric Rating Scale (BPRS) and the baseline score was of at least 30. Nine patients received topiramate in addition to their current treatment and did not show significant reduction at end point compared to the baseline score. Topiramate was initiated at a dose of 25 mg/day and increased to a maximum of 300 mg/day with a range of 225--300 mg/day at end point. Tolerability and side effects were not assessed systematically but no clinically significant or serious side effects were reported. Weight change was not assessed. Drapalski et al.\[[@B95]\] suggested an improvement in negative symptoms in a patient with schizophrenia when added to a stable regimen of antipsychotic medication. The patient described was a participant in a 17 weeks duration open label study with an on-off design. An initial 4-week titration phase was followed by 8-week maintenance phase, 1-week tapering phase and 4-week follow-up. Negative symptoms were assessed with the Negative Scale of the Positive and Negative Syndrome Scale (PANSS) at baseline (Negative Scale score = 24), 4-week, 8-week and follow-up after discontinuation of topiramate. There was a significant 7 points improvement at the end of medication phase (from 24 to 17). When topiramate was discontinued there was an increase in the Negative Scale score (follow up score = 24). The dosage of topiramate was tailored cautiously by 25--50 mg every 4--7 days and the maximum dosage was 175 mg/day in two divided doses. No side effects were reported. Eating disorders and disordered eating -------------------------------------- McElroy et al.\[[@B96]\] designed a randomized, placebo-controlled trial, investigating the therapeutic benefit of topiramate in treating binge eating disorder associated with obesity. For this 14-week, flexible dose (25--600 mg/day) trial, 61 outpatients (53 women and 8 men) with a body mass index of 30 or more, and a diagnosis of binge eating disorder according to the Structured Clinical Interview for DSM-IV were randomly assigned to receive topiramate (N = 30) or placebo (N = 31). The number of binges and binge days during the previous week were assessed at the initial screening visit together with psychiatric and medical history, physical examination, vital sign monitoring, routine blood chemical and haematological tests including fasting glucose, insulin and lipids, electrocardiogram and urinalysis. Monitoring of medication dose and compliance (review of patients\' take-home diaries and tablet count), adverse events, use of non-study medications, weight and vital signs, efficacy measures, was achieved with regular visits. Topiramate was introduced at a dose of 25 mg/day and the dose titrated by 25 mg to 50 mg on day 4. It was then increased by 25--50 mg to 75--100 mg/day on day 7; the dose was subsequently increased by 50 mg/week for 4 weeks to maximum dose of 300 mg/day at 6 weeks and by 75 mg/week for 4 weeks to a maximum of 600 mg/day at 10 weeks. The dose was not changed from treatment period weeks 10 through 14 unless a medical reason supervened. If a patient did not tolerate any dose increase, the dose could be decreased to a tolerable one. The primary efficacy measure was binge frequency but the CGI severity scale, the Yale-Brown Obsessive Compulsive Scale (YBOCS) modified for binge eating, the Hamilton Depression Rating Scale, body mass index, weight were also used. Waist-to-hip ratio, percent and total body fat (measured by bioelectrical impedance), blood pressure, fasting blood glucose, insulin and lipids were also considered as secondary measures of efficacy at the last visit. Safety measures such as adverse events, clinical laboratory data, physical examination findings and vital signs were assessed. The baseline score on the YBOCS was 15 or more, suggestive of marked distress regarding binge-eating behaviour. Twenty-six subjects (42.6%) discontinued the study (Topiramate N = 14) but analysis included all patients with at least one post-randomization efficacy measure (intent to treat analysis) with a repeated-measures random regression with treatment-by-time as the effect measure. Topiramate was associated with a statistically significant reduction in binge eating frequency (topiramate 94% vs. placebo 46%) and binge day frequency (topiramate 93% vs. placebo 46%). The CGI severity scale and the Yale-Brown Obsessive Compulsive Scale showed improvement scores at the last visit and were greater in the treatment arm. The rate of decrease in Hamilton Depression Rating Scale scores did not differ between treatment groups. The mean weight loss for topiramate treated subjects was 5.9 kg compared to 1.2 kg in the placebo group. Median topiramate dose was 212 mg/day (range 50--600). Twenty-six patients discontinued. Nine patients (topiramate = 6) because adverse events with paraesthesias and headache as the most common side effects. Topiramate was associated with a significant change in diastolic blood pressure at the last visit compared with placebo among the intent to treat group. There was no significant difference between groups in mean change for the fasting metabolic measurements of insulin, glucose, LDL cholesterol, triglycerides and total cholesterol. Hoopes et al., \[[@B97]\] enrolled 69 patients with DSM-IV bulimia nervosa in a randomised, double blind, placebo controlled trial. Sixty-four patients (33 in the placebo group vs. 31 in the topiramate group) were included in the intent to treat analysis. The primary efficacy measure, mean weekly number of binge and/or purge days, decreased 44.8% from baseline in the topiramate group versus 10.7% in the placebo group (p = 0.004). This was confirmed by significant reduction in scores on the Bulimic Intensity Scale, 37% for topiramate vs. 14% for placebo. The trial lasted for 10 weeks and the median dose was 100 mg/day (range 25--400). Topiramate, administered in monotherapy, was commenced at 25 mg/day for the first week. The dose was titrated by 25--50 mg increments per week to a maximum of 400 mg/day. Response supervened within 10 weeks. Only 3 patients discontinued from the trial (2 placebo, 1 topiramate) due to adverse events (topiramate: nausea). Shapira et al.\[[@B98]\] studied 13 female patients with binge eating disorder in a naturalistic, open label, add-on study. All the patients had co-morbid diagnoses. Treatment was begun at 25 mg/day and subsequently increased by 25--50 mg/week according to response and side effects to 1400 mg/day, given in divided doses. Response and side effects were valuated retrospectively as recalled by patients at monthly appointments. Outcome was measured as decrease in binge-eating episodes: none (0% to \<25% reduction), mild (25 to \<50% reduction), moderate (50 to \<70% reduction), marked (75 to \<100% reduction) or remission (complete cessation of binge eating episodes). Patient weight and BMI at beginning of treatment and at end point were recorded and statistically correlated. Nine patients displayed a moderate or marked response of binge eating disorder that was maintained for 18.7 +/- 8.0 months (range: 3 to 30 months), 7 continued to display the improvement at 21.1 +/- 6.0 (range 13--30 months), whilst 1 patient continued treatment because stabilised her bipolar disorder. Two patients displayed moderate or marked response that subsequently declined. The remaining two patients had a mild or no response. The mean topiramate dose was 492.3 +/- 467.8 mg/day for all 13 patients. The main weight at beginning of treatment was 99.3 +/- 26.4 kg and 87.5 +/- 20.4 kg at the end (z = -2.4, df = 1, p = .02) but only 7 patients lost 5 or more kg of weight. The mean dose of topiramate was higher in those who lost 5 kg or more (725.0 +/- 529.3 mg/day) compared to those who lost \<5 kg (220.8 +/- 156.9 mg/day). Topiramate was well tolerated. However, 2 patients reported side effects (cognitive impairment and dyspepsia) which subsided with discontinuation and slower reintroduction of the dose. Two patients reported worsening of co-morbid bipolar (manic) symptoms. A mixed response of co-morbid condition was also noted (obsessive compulsive disorder, compulsive buying, major depressive disorder). Barbee\[[@B99]\] treated a series of five patients with adjunctive topiramate. All the patients had a long history of severe bulimia nervosa combined with significant different co-morbid conditions (major depression, bipolar disorder II, substance misuse, post traumatic stress disorder, dysthymia, social phobia, border line personality disorders and general anxiety disorder). The dose was titrated slowly to 95--400 mg/day according to clinical response. During a follow up period of 7--18 months, 3 patients responded to topiramate, 1 did not respond and 1 subject discontinued treatment because of gastro-intestinal related side effects. Only 1 case reported simultaneous improvement in the co-morbid affective disorder. Adverse events occurred in 2 patients (paraesthesias and constipation). Posttraumatic stress disorder ----------------------------- Berlant and van Kammen \[[@B100]\] retrospectively reviewed 35 patients with chronic posttraumatic stress disorder treated with topiramate as add-on treatment (N = 28) or monotherapy (N = 7). Dosage titration was slow with an initial dose of 12.5--25 mg/day, increased by 25--50 mg every 3--4 days until therapeutic response was achieved. The main duration of treatment was 33 weeks (range 1--119 weeks). Topiramate decreased nightmares in 79% (19/24) and flashbacks in 86% (30/35) of patients, with full suppression of nightmares in 50% and of intrusions in 54% of patients with these symptoms. Nightmares and intrusions partially improved in a median of 4 days (mean 11+/-13 days) and were fully absent in a median of 8 days (mean 35 +/- 49 days). Response was seen in 95% of partial responders at a dosage of 75 mg/day or less and in 91% of full responders at a dosage of 100 mg/day or less. The last 17 patients completed the PTSD Checklist-Civilian Version (PCL-C) before treatment and at week-4. Mean reduction in PCL-C score from baseline to week-4 was highly significant (baseline score = 60 vs. week-4 score = 39, p \< .001), with similar reductions in re-experiencing, avoidance, and hyper-arousal criteria symptoms. Thirteen patients discontinued for various reasons during the study period. There were no serious side effects reported a part from a case of acute secondary narrow-angle glaucoma. Response assessment used the last observation carried forward. Alcohol dependence ------------------ Johnson and associated \[[@B101]\] conducted a double blind randomised controlled 12-week clinical trial comparing topiramate to placebo for treatment of 150 individuals with alcohol dependence. Of these 150 individuals, 75 were assigned to receive topiramate (escalating dose of 25--300 mg per day) and 75 had placebo as an adjunct to weekly-standardised medication compliance management. Primary variables were: self reported drinking (drinks per day, drinks per drinking day, percentage of heavy drinking day, percentage of day abstinent) and plasma gamma-glutamyl transferase as an objective index of alcohol consumption. The secondary efficacy variable was self-reported craving measured on the 14-item obsessive compulsive drinking scale. In the topiramate group 55 subjects completed the study versus 47 in the placebo group. The authors adopted intention to treat analysis. Response supervened between 6 and 8 weeks. At study end, participants on topiramate, compared with those on placebo, had 2.88 (95% CI -4.50 to -1.27) fewer drinks per day (p = 0.0006), 3.10 (-4.88 to -1.31) fewer drinks per drinking day (p = 0.0009), 27.6% fewer heavy drinking days (p = 0.0003), 26.2% more days abstinent (p = 0.0003), and a log plasma gamma-glutamyl transferase ratio of 0.07 (-0.11 to -0.02) less (p = 0.0046). Topiramate induced differences in craving were also significantly greater than those of placebo, of similar magnitude to the self-reported drinking changes, and highly correlated with them. There were no discontinuations due to side effects and topiramate was generally well tolerated. Gilles de la Tourette\'s syndrome --------------------------------- Abuzzahab et al. \[[@B102]\] described 2 cases of Tourette\'s syndrome successfully treated with topiramate respectively at 50--200 mg for 8 months and 100 mg nocte for a month. In both cases, previous medication were tapered down and discontinued during the first two weeks of treatment. Significant weight loss was noted: weight dropped from 183 to 145 lb for case 1 and 12.5 lb weight loss for case 2. Lacks of concentration, loss of appetite, thirst and lethargy sensitive to dose reduction were reported. Emotional unstable personality disorder --------------------------------------- Cassano et al.\[[@B103]\] described a case of bipolar mood disorder and border line personality disorder complicated by self mutilating behaviour, which responded to topiramate administration with an on-off-on design. Although depressive symptoms persisted, topiramate controlled self-injurious acts within 2 weeks at a dose of 200 mg/day. No side effects were reported. Teter et al. \[[@B104]\] published a case of an inpatient with psychotic disorder not otherwise specified and border line personality disorder treated with topiramate at the dose of 200 mg/day. Borderline symptoms improved in 6 weeks. Considerable weight loss was also reported. Adverse events, safety and tolerability --------------------------------------- Topiramate was generally well tolerated. General observations suggested that side effects occurred with high dose titration and frequently resolved or lessened with time and/or dosage reduction. Conversely, slow dose titration was associated with a lower rate of side effects \[e.g. \[[@B73],[@B75],[@B78],[@B84]\]. This is in agreement with data from epilepsy clinical trials, which suggest possible appearance of adverse reactions and treatment discontinuation following rapid dose titration and a target dose greater than 400 mg/day \[[@B105]-[@B109]\]. This indicates that individuals on complex pharmacological treatments are more vulnerable to side effects, particularly with sodium valproate and lithium \[[@B80]\]. The commonest adverse events (table [1](#T1){ref-type="table"}) across the studies analysed in this review were paraesthesias/numbness (N = 116, 12.9%), nausea and vomiting (N = 56, 6.2%), cognitive impairment (N = 48, 5.4%), headache (N = 46, 5.1%), dizziness (N = 45, 5.0%), sedation/drowsiness (N = 44, 4.9%), fatigue (N = 38, (4.2%), decreased appetite (N = 24, 2.7%), frequent peristalsis (N = 20, 2.2%), somnolence (N = 19, 2.1%), blurred vision (N = 18, 2.0%), slow memory (N = 16, 1.8%). There was one reported case of psychotic features \[[@B71]\], a case of delirium in a patient who overmedicated with 800 mg of topiramate and tranylcypromine sulphate (170 mg) combined with alcohol \[[@B76]\], a case of acute narrow angle glaucoma \[[@B100]\] and 2 cases of hematuria \[[@B92]\]. Occurrence of hematuria is consistent with the known 2 to 4 increased risk of nephrolithiasis during topiramate treatment \[[@B45]\]. Rare but serious adverse events have been described with topiramate (e.g. metabolic acidosis, acute myopia, acute glaucoma, oligohidrosis, hyperthermia) leading topiramate to be under review by regulatory authorities in several jurisdictions. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Adverse events in order of frequency ::: Adverse events \ Topiramate (N = 896) N (%)** ------------------------------- ---------------------------------- Paresthesia/numbness 116 (12.9) Nausea/vomiting 56 (6.2) Cognitive impairment 48 (5.4) Headache 46 (5.1) Dizziness 45(5.0) Sedation/drowsiness 44 (4.9) Fatigue 38 (4.2) Decreased appetite 24 (2.7) Frequent peristalsis 20 (2.2) Somnolence 19 (2.1) Blurred vision 18 (2.0) Slow memory 16 (1.8) Lack of concentration 11 (1.2) Influenza-like-symptoms 10 (1.1) Panic/anxiety 9 (1.0) Dysgeusia 8 (0.9) Dry mouth 8 Nervousness 7 (0.8) Rash 7 Ataxia 7 Insomnia 7 Constipation 7 Reduced libido 5 (0.6) Memory concerns 5 Dyspepsia 5 Unwanted weight loss 4 (0.4) Increased thirst 4 Word-finding difficulty 4 Impaired concentration 4 Tremor 4 Itching 4 Sweating 3 (0.3) Confusion 3 Slowed thinking 3 Psychosis 3 Slurred speech 3 Increased salivation 3 Sleep disturbance 3 Backache 3 Increased appetite 2 (0.2) Gastrointestinal disturbances 2 Agitation 2 Cold sensitivity 2 Worsening of symptoms 2 Increased libido 2 Amenorrhea 1 (0.1) Hematuria 1 Dysuria 1 Urticaria 1 Increased suicidality 1 Glaucoma 1 Water retention 1 Delirium 1 Grand mal seizures 1 (\*) From the studies reviewed only ::: Weight loss ----------- Topiramate weight loss was reported in 21 of the 32 studies analysed \[[@B70],[@B72],[@B75],[@B78]-[@B84],[@B86]-[@B89],[@B91],[@B92],[@B96],[@B98],[@B101],[@B102],[@B104]\] and reached 5% reduction of the baseline weight prior to treatment initiation in 5 studies \[[@B75],[@B79]-[@B81],[@B92]\]. Weight change was not systematically evaluated in 11 trials \[[@B71],[@B76],[@B85],[@B90],[@B93]-[@B95],[@B99],[@B97],[@B100],[@B103]\]. Among the studies, a frequent finding was that the initial BMI affected topiramate-induced weight loss and that greater weight loss was associated with higher BMI at baseline \[e.g. \[[@B73],[@B78],[@B80],[@B81],[@B86]\]. In diabetic patients, topiramate induced weight loss was also associated with glycemic control and normalization of blood pressure in hypertensive subjects \[[@B78],[@B96]\]. Conclusion ========== Preliminary reports \[[@B70]-[@B72]\], available for review, suggested a trend towards improvement in acute mania but more recent unpublished controlled studies, not available for review, showed lack of efficacy \[\[[@B74]\]; Calabrese, personal communication\]. It emerges that in the light of current evidence, there is limited scope for the use of topiramate in acute mania. The only randomised single blind study by McIntyre et al. \[[@B88]\], in the treatment of refractory bipolar disorders in depressive phase, showed a significant improvement in 56% of the subject treated with topiramate versus 59% in the bupropion group. This study, according to the authors, was not powered to detect a difference in efficacy between the two treatment groups and, given the small sample size, it only aimed to corroborate the antidepressant property of topiramate already shown in naturalistic studies \[[@B89]\]. If demonstrated efficacious in further adequately powered controlled studies, topiramate could fill the therapeutic vacuum in the treatment of bipolar depression as alternative or adjuvant to mood stabilisers. The role of topiramate in the treatment of rapid cycling bipolar disorders \[[@B75]\], and as adjunctive treatment in refractory bipolar disorder in adults \[[@B76]-[@B87]\] and children \[[@B90]\], is limited by the open label nature of the published studies: lack of randomisation and blindness, heterogeneous patient, population resistant to conventional treatment regimes, incomplete information on current or past treatment for illness, concomitant medications with possibly inflating side effects profile and therapeutic effect, self-reported weight and side effects, qualitative assessment of response to treatment, various settings and variegated level of symptoms, co-morbid psychiatric and medical conditions. Although there is no sufficient evidence for its use in these conditions, its trend towards improvement warrants controlled studies. However, it may not be sustained in randomised studies as observed in acute mania. The effectiveness of topiramate in schizophrenia is similarly based on uncontrolled studies. Only Drapalski et al. \[[@B95]\] reported a positive outcome in treatment of negative symptoms with adjunctive topiramate. Millson et al. \[[@B93]\] observed a post-treatment deterioration in PANSS scores in 5 patients treated with topiramate. Dursum and Deakin \[[@B94]\] also noted no reduction in BRPS scores when topiramate was augmented with antipsychotic medications. These controversial results, conveys doubts about the efficacy of topiramate in schizophrenia and uncertain the postulated stabilizing properties of topiramate in the interaction between the glutamatergic and dopaminergic systems. Alternatively, these observations may reflect that patients analysed were a heterogeneous group in many aspects of their illness and future studies would probably require more strict research criteria. Evidence for the use of topiramate in binge eating disorders and bulimia nervosa is encouraging and suggest a complementary role of topiramate in the treatment of these conditions together with established treatment strategies (e.g. SSRIs). McElroy et al. \[[@B96]\], reported efficacy of topiramate in reducing binge eating episodes by 93% in the treatment arm compared to 46% in the placebo group. Hoopes at al. \[[@B97]\] reported a decrease in the mean weekly number of binge and/or purge days by 44.8% from baseline in the topiramate group versus 10.7% in the placebo group (p = 0.004) and a significant reduction in scores on the BIS by 37% in the topiramate group vs. 14% in the placebo group. The only study in the treatment of PTSD by Berlant and van Kammen \[[@B100]\] was suggestive of efficacy in treating nightmares (79%) with full suppression in 50% of cases, flashbacks (86%) and intrusions (54%). The relative short duration of the trial did not allow exploration of a possible prophylactic role of topiramate. However, similarly to bipolar spectrum disorder, the naturalistic nature of this report constitutes a limitation for its validity. The only study by Johnson et al. \[[@B101]\] in the treatment of alcohol dependence was controlled. It indicated that topiramate, as an adjunct to standardised medication, is more efficacious in reducing alcohol consumption than placebo. This study warrants further investigation and indicates that topiramate could be included in the rather limited pharmacological armamentarium to defeat alcohol dependence. The effectiveness of topiramate in unipolar depression \[[@B91],[@B92]\], emotionally unstable personality disorder \[[@B103],[@B104]\] and Gilles de la Tourette\'s syndrome \[[@B102]\] is entirely based on case reports and case series. The evidence is sometimes controversial and at the time of writing there is no clear indication for treatment with topiramate. Weight change was not always systematically reported across the studies. However, findings are encouraging considering the rather disappointing success rates of efficacious prevention programs \[[@B110]\] and have potential implications in reversing increased body weight and obesity induced by psychotropic medication \[[@B111],[@B112]\]. Weight loss was also proportional to the initial BMI and it was associated with glycemic control and normalization of blood pressure in hypertensive subjects. Topiramate was generally well tolerated and serious adverse events were rare. Polypharmacy often contributed to an increased rate of side effects. Competing interest ================== The author(s) declare that they have no competing interests. Drug names ========== Topiramate (Topamax) Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Details of published studies included in the review. ::: ::: {.caption} ###### Click here for file ::: Acknowledgement =============== I would like to acknowledge Dr LJ Simmons\'s contribution to the completion of the manuscript.	PubMed Central	2024-06-05T03:55:56.270399	2005-2-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088011/", "journal": "Ann Gen Psychiatry. 2005 Feb 16; 4:5", "authors": [ { "first": "Danilo", "last": "Arnone" } ] }
PMC1088012	Background ========== Religious faith, spirituality and spiritual beliefs were rarely discussed in the psychological or medical literature of the \'80s \[[@B1],[@B2]\]. However, over the past decade an increasing interest on the part of health care professionals in spirituality and spiritual care emerged, and related articles were frequently published in medical and nursing journals \[[@B3]-[@B8]\]. Religious faith and spirituality are now widely recognized as important components of subjective human wellness \[[@B9]\], of health care outcomes \[[@B10]-[@B14]\], of holistic nursing care \[[@B15],[@B16]\] and of the quality of hospital care \[[@B17]-[@B19]\]. Nevertheless, the terms spirituality and religiousness have been used in different ways by different authors, sometimes interchangeably because of the elusiveness of both concepts. Narayanasamy argues that the lack of any authoritative definition of these two terms has resulted in a multiple definition and hence in the confusion surrounding these concepts \[[@B20]\]. Furthermore, Mokuau et al. aptly state: \"\...the difficulties in developing standardized definitions and measures relate to varying interpretations of religiousness and spirituality\...\"and they stress that the potential for providing quality care that integrates religiousness and spirituality to a large extent depends upon the development of measures that are at least psychometrically sound \[[@B21]\]. Studies on religiousness, spirituality and spiritual care are rare in Greece \[[@B17],[@B22]\] and this may be the reason for the present lack of a valid instrument. Sapountzi et al. mention that \"\...the lack of valid instruments is a widespread phenomenon, concerning not only Greece but rather all non-English speaking countries, owing to the fact that in this day and age, English is the language of science, just as Greek was in biblical times\...\"\[[@B23]\]. In 1995 King et al. developed a scale, \"The Royal Free Interview for Religious and Spiritual Beliefs\", in order to evaluate religious and spiritual beliefs in a variety of populations. Some years later, this scale was modified in order to make it more functional \[[@B24],[@B25]\]. The Royal Free Interview for Religious and Spiritual Beliefs is a valid and reliable scale; in the English version \[[@B24]\], it is short and simple and can be easily completed by most people. Moreover, it focuses on the strength and the consequences of faith, rather than on the specific nature of each belief and as King et al. stress, the \"\...interview was designed specifically to avoid a focus on any one religious system or type of spiritual belief and thus comparisons were impossible to make\...\"\[[@B24]\]. At the time that The Royal Free Interview for Religious and Spiritual Beliefs was published, the researchers of the present paper were looking for an instrument capable of measuring spirituality and religiousness in the Greek population \[[@B24]\]. Following careful consideration, they decided to translate an already valid instrument instead of developing a new one, choosing a European rather than an American one, because they believe that although cultural differences exist among European populations, they also share many common cultural issues. In addition, taking into account that the vast majority of the population living in Greece belongs to the Greek Orthodox Church, it was desirable to find a scale designed specifically to avoid a focus on any one religious system or type of spiritual belief. The Royal Free Interview for Religious and Spiritual Beliefs fulfilled most of the criteria set, and therefore the researchers approached Professor King, asking for his permission to translate the scale into the Greek language and validate it. After obtaining written permission, the Royal Free Interview for Religious and Spiritual Beliefs was translated into Greek in 2003 \[[@B23]\]. In a second stage, the research team continued evaluating the psychometric properties of the scale. Let us point out that as a result of extensive discussions among the research team and with other researchers, we decided to revise the question \"How would you describe yourself? (tick one or more)\". More precisely, the modified question in the Greek version asked participants to tick a specific part of Greece as their place of origin. Given that over the past twenty years, Greece has been faced with immigration from the Balkan countries and the former USSR, the question was adapted to include these newcomers\' origin as well. The question was thus revised, and instead of asking respondents to indicate specific parts of Greece as their place of origin, they are now requested to indicate their origin from an international perspective. The modification was made in order to make the question suitable for people from a variety of ethnic backgrounds living in contemporary Greece. This paper reports on test -- retest reliability and the psychometric properties of the Greek version of the Royal Free Interview for Spiritual and Religious Beliefs. Methods ======= Cross-cultural validation of an existing scale, such as the Royal Free Interview for Spiritual and Religious Beliefs, has the great advantage of avoiding the initial stages of development of a new questionnaire, which is a lengthy process \[[@B26]\]. Furthermore, translation and adaptation of a scale into different languages makes it possible to use the questionnaires in comparative international multi-center studies. This is why we decided first to translate, re-translate and then proceed to check the validity, reliability and psychometric properties of the scale for a Greek population \[[@B27]-[@B29]\]. The translation and the cultural adaptation of the Royal Free Interview for Spiritual and Religious Beliefs was carried out at an early stage \[[@B23]\]. A panel of experts who were also bilinguals was requested to translate the English version of the Royal Free questionnaire. Furthermore, in order to assess the linguistic accuracy of the translated instrument, a pilot study using bilingual persons was carried out. Details related to the translation and adaptation of the Royal Free questionnaire for spiritual and religious beliefs are mentioned elsewhere \[[@B23]\]. Test -- Retest -------------- To evaluate the stability of a translated instrument, it is recommended that it be tested in the target culture based on a test-retest design \[[@B30]\]. It is difficult to establish standards for retest reliability since many factors need to be considered, such as the time between pre-test and post-test, learning obtained from the pre-test or between tests, and the type of test (trait or state). Some researchers believe that it is sufficient to know that the retest coefficient is statistically significant from zero, although Huck & Cormier have warned against such use \[[@B31]\]. The Spearman Rank Correlation Coefficient (rho) between scores produced at the first and second testing was calculated to assess the test-retest reliability. However, the calculation of correlation coefficients is not a sufficient method to test reliability and reproducibility of a method and its results, because it is an index of correlation and not an index of agreement \[[@B32]-[@B34]\]. There is less agreement about intra-class correlation coefficients. For questions with categorical responses, such as questions 1 and 12 to 14, kappa statistics were used. Face validity and content validity were established during the stage of translation and modification of the scale into Greek \[[@B23]\]. Data analysis ------------- All items were coded and scored, and the completed questionnaires were included in the data analysis set. Individual unanswered items were excluded from the analysis. Statistical Package for Social Sciences 10.0 computer software was used for the statistical analysis of the data obtained \[[@B35]\]. The Pearson correlation coefficient was used to calculate the linear correlation of two continuous variables. The Chi-squared test was used between two nominal variables. The t-test assessed whether the means of two groups were statistically different from each other. Values less than 0.05 were considered statistically significant, unless otherwise stated. Sample ------ Potential subjects meeting the following inclusion criteria were selected to participate in the study: (1) willing to participate, (2) over 18 years of age (3) capable of speaking and reading Greek and (4) no cognitive impairment, according to the research team\'s assessment. Potential subjects were recruited from the community on the basis of their availability. They received a brief explanation of the purpose and the aim of the study, and those who agreed to participate were asked to sign an informed consent form. We finally approached 209 people (77 men and 132 women) informally through hospitals and institutions of learning. Participants were divided in two major groups: Group A:209 participants who answered to the Royal Free questionnaire for the first time, thus called the test group and Group B:139 participants who were given the Royal Free questionnaire a second time two weeks later (the retest group). Group A was the test group and the group B was the retest group. The mean age of group A was 28.33 ± 9.44 and that of group B was 28.93 ± 9.60 years. In the test group, there was a statistically significant difference between the two genders and their mean age, with women being younger than men \[26.73 (sd 9.02) cf 31.08 (sd 9.56), mean diff 4.34, CI 1.69 to 7.00, df (unequal variance) 151.707, p = .001\]. A similar difference was observed in the retest group \[26.97 (sd 9.25) cf 30.82 (sd 9.41), mean diff 3.86, CI 0.46 to 7.25, df (unequal variance) 86.778, p = .027\]. Participants in both groups were mainly white, single students. Table [1](#T1){ref-type="table"} shows the distribution of the sample according to demographic characteristics. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of the sample ::: Group A Group B --------------------------------------------------------- ------------- ------------- ----- ------ *Gender* Male 77 36.8 46 33.3 Female 132 63.2 92 66.7 *Marital status* Married 43 20.8 31 22.5 Cohabiting 35 16.9 22 15.9 Divorced 6 2.9 4 2.9 Separated 6 2.9 4 2.9 Single 117 56.5 77 55.8 *Ethnic group* White Europeans from EU 198 96.1 131 95.6 Russian 4 1.9 2 1.5 White Europeans from Eastern Europe countries out of EU 3 1.5 3 2.2 Albanian 1 0.5 1 0.7 *Occupation* Employed 51 28.2 33 28.5 Unemployed seeking work 17 9.4 12 10.3 Student 101 55.8 63 54.3 Retired 1 0.6 1 0.9 Home manager 11 6.0 7 6.0 ::: Results ======= The majority of participants in both groups were Orthodox Christians (for group A: N = 106, 62.4% for group B: N = 68, 58.6%) as compared to smaller percentages of those with no religious faith (for group A: N = 5, 2.9% for group B: N = 4, 3.4%) Roman Catholics (for group A: N = 19, 11.2% for group B: N = 12, 10.3%) Protestants (for group A: N = 18, 10.6% for group B: N = 14, 12.1%) Sunni Muslims (for group A: N = 1, 0.6% for group B: N = 1, 0.9%) and those belonging to other religions (for group A: N = 21, 12.4% for group B: N = 17, 14.7%). A total of 18 out of 207 participants of the test group (8.7%) stated that they had no religious or spiritual understanding of their life; 44 (21.1%) reported a religious belief; 18 (8.7%) told of a spiritual belief and the vast majority of participants (61.4%) reported both a religious and a spiritual belief. In group B, responses were similar. Those with a spiritual or religious understanding of their life explained that they believed in God, the Saints, the Holy Trinity, a superior creature, or the Bible. Spiritual belief differed in a statistically significant manner between the two genders of the test population (x^2^= 25.808, df = 3, p \< .001) as shown in table [2](#T2){ref-type="table"}. The vast majority of women (66.8%), rather than men (41.4%), reported religious and spirituals beliefs and only 2.1% of the women and 12.6% of the men had neither religious nor spiritual beliefs. Logistic regression analysis indicated that women (OR 2.3, 95%CI 1.1--4.6) were significantly more likely to hold a religious and spiritual view of life. Table [2](#T2){ref-type="table"} presents the gender-specific difference of mean scores of spiritual scale. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Difference of mean scores of spiritual scale between two genders ::: Question Men Women --------------------------------- --------- ------------- ----- ------------- -------- Q3: strength of belief 83 6.95 (2.21) 183 7.10 (2.12) .593 Q7: practice of faith 83 4.94 (2.87) 180 5.91 (2.64) .008 Q8: influence of power or force 85 5.36 (3.14) 182 6.66 (2.38) \<.001 Q9: enable you to cope 85 5.44 (3.10) 180 6.72 (2.29) \<.001 Q10: influence on world affairs 85 3.48 (2.90) 181 4.69 (2.77) .001 Q11: natural disasters 85 4.26 (3.27) 182 5.45 (2.93) .003 ::: Four persons (1.5%) answered that they underwent an intense experience at a time when they almost died, but were eventually revived. Four other persons were uncertain as to whether or not they had had such an experience. For these four persons, the mean effect of this experience on their lives was moderate (4.69 ± 2.9). The majority of persons participating in our study answered that they prayed by themselves (N = 215, 77.3%) or with others (N = 24, 8.6%). Only 57 persons (20.5%) stated that religious ceremonies play a central role in their faith; 43.2% said the same of meditation, 60.8% of reading and study, 36.3% of contact to religious leaders and 7.2% none of the above. After we tallied the scores of participants for questions 3, 7 to 11, which make up the spiritual scale, we found that during the test phase the t-test was significant (t = -3.562, df = 256, p \< .001) and showed that the mean scores of the participants on the spiritual scale differed significantly between women (mean = 36.10 ± 10.54) and men (mean = 31.80 ± 11.25), as shown in table [2](#T2){ref-type="table"}. The retest did not lead to the same result, as the t-test showed no statistical difference between the two genders (p = .563). There was a gender-specific difference (Fisher\'s exact test p = .001), though, between those who believed that prayer plays an important role in their faith: the majority of women (75.6%) like to pray alone, instead of men (60.9%) who prefer to pray with others. One hundred and sixty-five participants (62%) answered that they do not communicate with a spiritual power at all, i.e. by way of prayer or contact via a medium. Eleven persons (4.1%) were uncertain. Women communicate more often by praying than do men (Chi-Square test 7,595 df = 2 p = .022). Twenty-five percent of the men, as against 45% of the women communicate with a spiritual power in some other way. The percentage of those who believe that we continue to exist in some form after death (37.5%) was similar to those who were uncertain (39.3%). Thirty-two participants (23.9%) had had an intense experience through which they sensed some deep new meaning in life that lasted for a few moments, hours or even days. Eleven of them felt it once (44%) and ten of them had had eleven to fourteen such intense experiences (40%). This experience lasted for 7.96 ± 8.53 days. For some, it lasted 4.88 ± 0.61 hours, for others 5.02 ± 0.65 minutes and for a few individuals it lasted for 5 seconds. When asked to describe this experience, they answered that the incident was a prophetic dream, the recovery from an accident or an illness, the presence of an invisible force, the birth of their child, a detachment of the material body, a strong feeling of happiness, of feeling protected from harm. Further analysis of this open question (describe the intense experience) put to those participants who had had an intense experience rendered a description of a dreamy journey to an unknown place, filling them with a strong euphoric feeling of happiness, wellness and inner peace. The main characteristic of this experience was the lack of any sense of control over the circumstances. They felt no fear, although it was an unusual experience -- perhaps precisely because it was a perceived spiritual experience, providing an opportunity to become closer to God or clarifying the grace of God. It was an isolated moment of physical and bodily sensation, or a sharing emotion. For others it was a unique and private emotional experience, or simply a prophetic reassurance of future recovery from a difficult situation. Some of the respondents were not able to describe it, even though their experience had been very intense. Between those 90 participants (67.2%) who had never felt an intense experience, two persons mentioned that they had undergone an intense experience at a time when they almost died but were eventually revived. It was an incident that changed their lives. Fourteen (88.9%) of the 18 participants with a spiritual belief communicate in some way with a spiritual power, compared to 26 (59.1%) of the 44 persons with a religious faith and 67 (52.8%) of the 127 persons with both a religious and a spiritual belief (Chi Squared 18.20, df = 6, p = .006). Respondents who reported having intensely experienced a power were more likely to believe in a spiritual power that influences the universe (5.52 ± 3.30) than those who had never had such an experience (4.25 ± 2.54). Belief in the existence of some form of life after death was more common among Muslims and those who do not belong to a specific religion. The majority of Orthodox Christians did not believe in life after death (73.6%). Forty-nine (38.9%) of the 126 persons who spoke of both a religious and spiritual faith believed in life after death, as against forty-five (35.7%) persons who were uncertain. There was a major difference in mean scores on the spiritual scale between persons who expressed a religious/spiritual view of life on the Royal Free Questionnaire and those who did not \[37.2 (sd 10.2) cf 23.1 (sd 18.3), mean diff 14.0, CI 6.6 to 21.5, df (equal variance) 131, p \< 0.0001\]. Those who communicate with a spiritual power obtained higher mean scores on the spiritual scale than those who do not \[42.4 (sd 8.9) cf 31.1 (sd 9.7), mean diff 11.3, CI 8.5 to 14.1, df (unequal variance) 156.55, p \< 0.0001\]. Women expressed a higher spirituality than did men \[37 (sd 12) cf 31 (sd 10.7), mean diff -5.97, CI -9.4 to -2.5, df (unequal variance) 125,64, p \< 0.0001\]. Internal consistency of the spiritual scale and test -- retest reliability -------------------------------------------------------------------------- The internal consistency of the spiritual scale for the test group proved to be good, as standardised inter-item reliability / Cronbach\'s alpha was 0.83. This is above the accepted limit of 0.70 \[[@B35]\]. Item-total correlations ranged from 0.51 to 0.73. They indicated very good levels of differentiation, thus showing that questions were appropriate. Internal consistency of the spiritual scale for the retest group proved as good as for the test group. Standardized inter-item reliability / Cronbach\'s alpha was 0.84. Item-total correlations ranged from 0.52 to 0.75. Kappa statistics for categorical items are summarized in table [3](#T3){ref-type="table"}. Kappa is a chance-corrected measure of agreement, as it represents the proportion of agreement obtained after removing the proportion of agreement that could be expected to occur by chance. Kappa is always less than or equal to 1. A value of 1 implies perfect agreement and values of less than 1 imply less than perfect agreement. Kappa coefficients ranged from 0.70 to 0.74, indicating good agreement \[[@B32]\]. Table [3](#T3){ref-type="table"} presents the results of the test -- retest reliability of the Greek version of the Royal Free Interview for Spiritual and Religious Beliefs, while in table [4](#T4){ref-type="table"} the mean test-retest score, intra-class correlation coefficient (ICC), test-retest correlation (rho) and p value are presented. As shown in table [4](#T4){ref-type="table"}, the Pearson correlation coefficient for the total test-retest score of the spiritual scale was 0.754 (p \< 0.001). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Test retest reliability of spiritual scale ::: Questions with categorical responses Kappa ------------------------------------------------------ ------------------- ---------------------- Present study King et al. 2001 Q1. Belief system 0.73 0.79 Q12. Do you communicate with this power? 0.74 0.76 Q13. Do you think we exist in some form after death? 0.70 0.79 Q14. Have you ever had an intense experience? 0.72 0.78 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### The mean test-retest score, intraclass correlation coefficient (ICC), test retest correlation (rho) and p value ::: Scale item ICC\* Mean (sd) score^+^ Rho p ---------------- ----------- ------------------------ --------------- ------------- ------- -------- 3 0.73 7.20 (2.07) 7.08 (2.24) 0.12 (0.17) 0.581 \<.001 7 0.80 5.76 (2.87) 5.76 (2.77) 0 (0.10) 0.670 \<.001 8 0.80 6.37 (2.66) 6.34 (2.61) 0.03 (0.05) 0.675 \<.001 9 0.81 6.38 (2.61) 6.24 (2.58) 0.13 (0.03) 0.681 \<.001 10 0.87 4.47 (2.92) 4.53 (2.96) 0.06 (0.04) 0.770 \<.001 11 0.86 5.37 (3.10) 5.33 (3.09) 0.04 (0.01) 0.755 \<.001 Total Score 0.86 35.55 (11.71) 35.28 (11.04) 0.27 (0.67) 0.754 \<.001 \p \< 0.001 for all ICCs, ^+^t-test for paired comparisons not significant. ::: The bivariate scatterplot between the total test and retest scores of the spiritual scale are shown in figure [1](#F1){ref-type="fig"}. It gives a good visual picture of the relationship between the two variables and facilitates the interpretation of the regression model. As we see in figure [1](#F1){ref-type="fig"}, the points of test-retest plot are very close to or follow the regression line. This finding reasserts the test-retest reliability of the spiritual scale. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Bivariate scatterplot between the test and retest total scores of the spiritual scale ::: ![](1744-859X-4-6-1) ::: Discussion ========== Judging from the results obtained the Greek version of the Royal Free Interview for Religious and Spiritual Beliefs proved to have satisfactory psychometric properties for a Greek population. The spiritual scale displayed good reliability, with sound internal consistency as assessed by coefficient α, and a degree of test-retest reliability similar to that reported by King et al. \[[@B24]\]. The excellent Pearson correlation coefficient for the test-retest of spiritual scale suggests that any repetition of the test would be likely to render the same results. The tool therefore proved to be reliable. The intra-class correlation coefficient for continuous variables ranged from 0.73 to 0.86, and for total spiritual scale score it was 0.86. Coefficient kappa ranged from 0.70 to 0.74. In King et al., intra-class research correlation ranged from 0.72 to 0.89 and 0.94 for the total spiritual scale score. Coefficient kappa ranged from 0.76 to 0.79. These findings are evidence of cross-cultural test-retest agreement of scale items. The majority of respondents replied that they believe in God and go to Church. Furthermore, the vast majority of participants (58.9%) reported both a religious and a spiritual belief. This is a constant finding of other research studies in the U.S.A. \[[@B37]-[@B39]\], with a percentage varying from 59% to 74%. It would seem that Greek respondents dissociate themselves from both spirituality and religion, and that they have a more traditional attitude toward religion. A similar attitude is mentioned by Streib in a research study on spirituality and religious orientation in adolescents in Germany \[[@B40]\]. However, we might interpret this evidence in the context of Christianity and with a view to the age structure of the respondents, who were very young. The low mean age of the sample may explain the distinction between spirituality and religiousness, as young people perceive the spiritual meaning of religiousness and distinguish it from church-related spirituality. In our sample, women were significantly more likely to hold a religious and/or spiritual view of life \[[@B24]\] and expressed a greater spirituality than men did. The term \"spirituality\" is multidimensional, allowing for various interpretations with its many connotations and vague structure. Parker Palmer mentions that spiritual questions always revolve around angels or ethers or include the word God: Spiritual questions are the kind that we and our patients ask every day of our lives, as we yearn to connect with the largeness of life: \"Does my life have a meaning and a purpose?\" \"How do I deal with suffering?\" \"What is the real meaning of my life?\"*\[[@B41]\] The last question accounts for the core meaning of the term spirituality for Greeks, as it includes all features that give meaning and purpose to life. All these approaches challenge the perceptual and conceptual framework of any instrument assessing spirituality and religiousness. The Greek version of the Royal Free Interview for Religious and Spiritual Beliefs captures all dimensions of this construct. Further evaluation of the scale might include a wider population, encompassing ethnic minorities and persons representing a greater variety of religions. There are limits to the study to be discussed here. During its initial stage, we did not select a random sample. However, we decided to proceed with a convenient sample because it is very difficult to locate and approach certain groups of persons holding strong religious beliefs. Even if one succeeds in approaching them, it is difficult to persuade them to participate in such a study, since they tend to feel embarrassed and suspicious. Nevertheless, some attempts were made to approach persons belonging to Christian Orthodox groups, as it is generally believed that they have a stronger spiritual belief and commitment to Orthodox doctrine. Despite lengthy efforts, we did not achieve our objective. As a result, we approached not only Orthodox Christians -- the vast majority of the Greek population -- but also Roman Catholics and Protestants. A further concern was that due to language barriers, our sample consisted chiefly of persons fluent in spoken and written Greek. Despite these restrictions, we decided to proceed with checking the psychometric properties of our instrument, as we believed that this study could serve as a precursor for future research by contributing important insights to the psychometric properties of this instrument, i.e. pertaining to populations from minority groups and persons with a wider range of spiritual beliefs. Conclusion ========== In summary, the self-report version of the Royal Free Interview for Religious and Spiritual Beliefs appears to be a valid and a reliable test-retest measure of spirituality in a general Greek population. Future validation studies with multiple populations and a longitudinal design will be required in order to refine the instrument as an additional scale of spirituality.	PubMed Central	2024-06-05T03:55:56.275390	2005-3-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088012/", "journal": "Ann Gen Psychiatry. 2005 Mar 4; 4:6", "authors": [ { "first": "Despina", "last": "Sapountzi-Krepia" }, { "first": "Vasilios", "last": "Raftopoulos" }, { "first": "Marcos", "last": "Sgantzos" }, { "first": "Evangelia", "last": "Kotrotsiou" }, { "first": "Zoe", "last": "Roupa-Darivaki" }, { "first": "Kalliope", "last": "Sotiropoulou" }, { "first": "Ioanna", "last": "Ntourou" }, { "first": "Alexandra", "last": "Dimitriadou" } ] }
PMC1088013	Background ========== Depression, according to recent epidemiological surveys might affect almost 25% of the general population at some point of their lives. The definition of \'depression\' according to both classification systems \[[@B1]-[@B3]\], is based on the definition of the depressive episode. Modern classification systems recognise melancholic (\'somatic) and atypical features. In spite of early reports \[[@B4]-[@B7]\], today the only report which seems to survive is not the favourable response of atypical patients to MAOIs, but their resistance to TCAs. One of the theories concerning the etiopathogenesis of depression suggests that a disturbance of biological rhythms is the core feature \[[@B8]\]. This disturbance is better studied in Seasonal Affective Disorder (SAD), which is a form of depression which responds to light therapy. It is possible that similar disturbances might be also present in non-seasonal depression, since these patients respond to sleep deprivation, especially in combination to light therapy. Additionally, there is a direct connection of the hypothalamus with the retina (retinohypothalamic tract) and some authors believe that at least 40% of brain neurones carry or process visual information \[[@B9]\]. A neglected area concerns the contribution of the visual system to the genesis of the circadian rhythms of the organism. Especially the direct assessment of retinal function would be valuable \[[@B10]\]. The suprachiasmatic nucleus is believed to be the center of the production of these rhythms. It processes information originating from the retina. Our group has already published papers on the visual system of depressives \[[@B11],[@B12]\] and Alzheimer disease patients \[[@B13]\] using pupillometry. In a recent study of our group \[[@B14]\] the use of PR-VEPs revealed that there might be an underactivation of the anterior right hemisphere in melancholic depressives (anterior to the chiasm) and a hyperactivation of the same region in atypical depressives. The question which arises is whether there is a specific dysfunction at the level of the pigmentum epithelium or the retina responsible for these findings. The present study aimedto investigate the outer part of the visual system of depressed patients and to provide evidence for further localization of a suggested anterior right hemisphere dysfunction in depression. Also aimed to compare the results of normal controls with those of depressed patients and to compare depressed subtypes between each other. Materials and methods ===================== Study Participants ------------------ Fifty (50) patients (15 males and 35 females) aged 21--60 years (mean = 41.0, standard deviation = 11.4) and 15 controls (4 males and 11 females) aged 20--55 years (mean 35.2, standard deviation = 9.2) suffering from Major Depression according to DSM-IV \[[@B2]\], and depression according to ICD-10 \[[@B15]\] criteria, took part in the study. All provided written informed consent. Fourteen of them fulfilled criteria for atypical features, 16 for melancholic features and 32 for somatic syndrome (according to ICD-10). Also, 9 patients did not fulfilled criteria for any specific syndrome (undifferentiated patients). All were inpatients or outpatients of the 3^rd^Department of Psychiatry, Aristotle University of Thessaloniki, University Hospital AHEPA, Thessaloniki Greece. They constituted the total number of patients during a two-years period that fulfilled the criteria to enter in the study. These criteria demanded that patients: 1\. Be free of any medication for at least two weeks prior to the first assessment and diagnosis. In no case medication was interrupted in order to include the patient in the study. 2\. Be physically healthy with normal clinical and laboratory findings, including EEG, ECG and thyroid function. 3\. Opthalmological examination should be normal and patients should have normal or corrected visual acuity and went through a full ophthalmologic investigation. 4\. No patient should fulfill criteria for catatonic or psychotic features or for seasonal affective disorder. 5\. Also, no patient should fulfill criteria for another DSM-IV axis-I disorder, except from generalised anxiety disorder and panic disorder 6\. No past history of manic or hypomanic episode. 7\. Psychiatric history of no more than five distinct episodes including the present one (mean 1.16 ± 1.53). 8\. Patients should be right-handed and the right eye to be the dominant one. 9\. All should be born and lived in the area of Thessaloniki, Greece (Latitude 40--40.1° North). 10\. All should be depressed during testing. Finally, the study sample of the current paper is identical with that of our previous study on PR-VEPs in depression \[[@B14]\]. ### Clinical Diagnosis The Schedules for Clinical Assessment in Neuropsychiatry version 2.0 (SCAN v 2.0) \[[@B16]\] were used for the clinical diagnosis. Each one of the symptoms (according the lists of both classification systems) was recorded and correlated with the laboratory findings. ### Laboratory Testing It included ECG, EEG, blood and biochemical testing, test for pregnancy, T3, T4, TSH, B12 and folic acid. ### Psychometric Assessment Its aim was the quantification of depression and anxiety \[[@B17],[@B18]\]. This was achieved with the use of the Hamilton Depression Rating Scale (HDRS) \[[@B19],[@B20]\] and the Hamilton Anxiety Scale (HAS) \[[@B21]\] and their subscales. The assessment of the endogeneity of depression was achieved with the use of the Newcastle Scales (1965 Newcastle Depression Diagnostic Scale-1965-NDDS and 1971 Newcastle Depression Diagnostic Scale-1971-NDDS) and the Diagnostic Melancholia Scale (DMS). These three scales have a different rational in assessing the \'endogenous-melancholic\' and the \'neurotic\' syndromes of depression. The General Assessment of Functioning Scale (GAF) \[[@B22]\] was used to assess the severity of depression. The questionnaire of Holmes \[[@B23]\] was used to search for stressful life events during the last 6 months before the onset of the symptomatology. Psychophysiological Methods --------------------------- It included: ### 1. Electro-oculogram (EOG) which is a method with which one can study the electrical and metabolic activity of the outer layers of the retina. During the adaptation of the retina to dark, the amplitude of the EOG gradually decreases, reaching a nadir (dark trough). During the adaptation to light (ganzfeld, 1200 lux) it gradualy increases reaching a zenith (light peak). The systematic development of the method of electro-oculogram was made mainly by Arden \[[@B24],[@B25]\] and the conditions for EOG recording have been coded by the International Society for Clinical Electrophysiology of Vision (ISCEV) \[[@B26]\] and this was kept in the current study. However some deviations from these conditions were inevitable. These included the use of 3 instead of 4 electrodes, the recording every 2 min for 12 minutes duration instead of every minute for a 15 minutes duration and not dilatated pupils. A video camera was used to verify that the patients were following the instructions and moved eyes to catch the alternating lights. EOG was recorded by two electrodes attached in the outer canthous (Lc and Rc) and a third in the mideye (Mr). The movement of the eyes produces a change of potential, which is recorded by the electrodes. After the recording of several movements of the eyes, the averaging of potentials gives the mean potential for the given conditions (interaction of time with lighting conditions). The procedure includes recordings of eye movents every 2 minutes, for 12 minutes in dark and subsequently 12 minutes in light. The resulting recording is shown in figure [1(a)](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A. Electro-oculogram (EOG). Recording in a normal control (upper), an atypical (middle-continuous line) and a melancholic patient (lower-dotted line). The control subject has Arden ratio = 224, the melancholic Arden ratio = 295, and the atypical patient Arden ratio = 248. However, although all ratios are within normal limits, the curves of the depressed patients have lower amplitude. B. flash-ERG. Upper: normal latency of a and b waves (control subject) Lower: slightly increased than normal latency of a and b waves (melancholic patient) All recordings are within normal range. ::: ![](1744-859X-4-7-1) ::: There is no difference of the recorded EOG curves between the two eyes \[[@B27]\]. The most widely used indices for the interpretation of the EOG are the Arden ratio: ![](1744-859X-4-7-i1.gif) The normal values of this index lie between 162 and 228, but values under 180 should be considered as borderline. Another index, which also takes into consideration the baseline potential is the A criterion \[[@B28]\]: A Criterion = light peak-\[0, 61\baseline potential+0,91\dark trough\]. According to Pinckers over of 70% of healthy subjects have A-Criterion values over 80 and all over zero. ### 2. Flash-Electroretinogram This is a method of recording potentials of the retina after the fall of light stimuli. The Electroretinogram (ERG) can be recorded after flash (f-ERG) or Pattern-Reversal (PR-ERG) stimulation. In the current study, binocular f-ERG was used. ERG recording have been coded by the International Society for Clinical Electrophysiology of Vision (ISCEV) and this was kept in the current study. However some deviations from these conditions were inevitable. These included the use of skin electrodes, and lack of maximum dilatation of the pupil. The f-ERG curve includes mainly the waves a and b. Wave a is photochemical in origin and is produced in the photoreceptors as their respond to a light stimuli and under specific conditons (scotopic conditions) the a wave may be split to ap and as waves \[[@B29]\]. It is believed that the ap wave comes from the cones and as wave from the rods \[[@B30]\]. The b-wave is produced by the bioelectrical activity of the neurons of the inner grannule layer and the bipolar cells. It is neuronal in origin. It can also be split (under scotopic conditions) in two waves, named bp and bs. In the current study, f-ERG was recorded from two electrodes, attached below the eyes (Lr and Rr) and a reference electrode at the mid-eye (Mr), under photopic conditions from both eyes simultaneously (binocular). ### 3. Specific Issues All recordings were conducted around mid-day (12:00 h to 16:00 h) and there was no difference in the times of the day or the season of the year the groups were studied. Gold-plated silver electrodes were used and the impedance was \<4 kohms. All patients came from North Greece (Latitude 40--40.1° North). Statistical analysis -------------------- It included Analysis of Covariance (ANCOVA) with age as a covariate and Pearson\'s product moment correlation coefficient. Student\'s t-test was used for post-hoc comparisons. Since 8 ANCOVAs were performed, the Bonferonni method suggests that the appropriate p-level should be \<0.00625, and for practical reasons the level p \< 0.005 was chosen and used also in post-hoc comparisons. Results ======= Depressed patients and controls had similar gender composition and did not differ in age (p = 0.107, table [1](#T1){ref-type="table"}). Melancholic patients seemed to be marginally older (table [2](#T2){ref-type="table"}). This is why age was used as covariate. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Results of Electrooculographic and flash-Electroretinographic recordings of depressed patients and controls and p-values after ANCOVA with age as covariate. ::: depressed patients N = 50 Controls N = 15 ------------------------------- ------------------------------- --------------------- ---------- ---------- --------------------------- --------------------------- Mean S.D. Mean S.D. P (ANCOVA) P (post-hoc) Age 41.0 11.4 35.2 9.2 0.107 EOG results *[0.001]{.underline}* Left Dark Trough 178.54 55.93 284.08 109.46 *[0.000]{.underline}* Left Light Peak 455.22 127.11 659.17 195.72 *[0.000]{.underline}* Right Dark Trough 169.32 54.74 283.08 96.72 *[0.000]{.underline}* Right Light Peak 402.40 104.13 646.58 183.92 *[0.000]{.underline}* Left Arden Ratio 261.04 49.67 241.88 44.81 0.227 Right Arden Ratio 248.47 53.87 233.81 34.94 0.374 Left A-Criterio 153.21 71.40 142.11 65.66 0.287 Right A-Criterio 118.40 66.41 131.85 69.12 0.534 F-ERG Photopic Conditions 0.147 Lr a wave, ampl 4.54 1.96 4.63 1.57 Lr a wave, lat 13.99 1.15 13.17 1.47 Lr b wave, ampl 8.30 2.58 10.83 8.22 Lr b wave, lat 31.85 2.85 29.53 4.62 Rr a wave, ampl 4.51 1.75 4.30 1.71 Rr a wave, lat 13.88 1.50 13.37 1.71 Rr b wave, ampl 7.92 2.59 11.08 8.14 Rr b wave, lat 31.94 2.87 29.57 4.68 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison between melancholic and atypical patients and controls (ANCOVA with age as covariate; significant are p-values below 0.005). ::: mean s.d. mean s.d. p p p p p p ----------------------------------- ------------------------------ --------------------------------- ---------------- ------------------ --------------------------- ------------------ --------------------------- --------------------------- --------------------------- --------------------------- Atypical features N = 14 Melancholic features N = 16 A/M ANCOVA A/M Post-hoc A/C ANCOVA A/C Post-hoc M/C ANCOVA M/C Post-hoc Age 37.00 7.79 47.00 13.03 0.016 0.575 0.007 Age of Onset 27.21 8.74 34.68 12.83 0.070 \- \- Number of previous episodes 1.21 1.37 1.21 1.69 0.995 \- \- GAF 60.36 10.65 41.16 12.10 *[0.000]{.underline}* \- \- Number of PD diagnosed 0.50 0.65 0.16 0.37 0.066 \- \- Number of PD criteria fulfilled 3.29 4.14 2.26 3.96 0.477 \- \- Number of stressful life events 3.64 2.27 1.21 1.44 *[0.001]{.underline}* \- \- HDRS-17 23.14 3.82 28.53 6.54 0.010 \- \- EOG 0.510 *[0.004]{.underline}* *[0.001]{.underline}* Left dark trough 187.86 47.31 181.84 49.22 0.006 *[0.001]{.underline}* Left light peak 488.50 132.98 443.16 124.36 0.015 *[0.001]{.underline}* Right dark trough 175.57 48.39 191.74 54.57 *[0.001]{.underline}* *[0.002]{.underline}* Right light peak 421.14 133.78 422.21 85.18 *[0.001]{.underline}* *[0.000]{.underline}* Left Arden ratio 262.95 47.09 246.03 40.03 0.256 0.790 Right Arden ratio 242.73 44.30 230.94 56.65 0.579 0.876 Flash-ERG Photopic Conditions 0.175 0.714 0.142 Lr a wave, ampl 4.08 1.97 4.70 1.89 Lr a wave, lat 13.38 1.42 14.37 0.75 Lr b wave, ampl 7.44 2.19 8.80 1.50 Lr b wave, lat 30.00 3.42 33.00 1.86 Rr a wave, ampl 4.12 1.35 4.67 1.84 Rr a wave, lat 13.31 1.30 14.14 1.63 Rr b wave, ampl 6.73 2.64 8.51 1.95 Rr b wave, lat 30.12 3.42 33.19 2.23 ::: a. EOG ------ Depressed patients (as a whole), manifested a decrease of both dark trough and light peak values in comparison to controls. This did not hold true for Arden ratios or A-Criterion values which both were within normal range (table [1](#T1){ref-type="table"}). This was true both for melancholic and atypical patients. The comparison between melancholic and atypical patients provided no significant results (table [2](#T2){ref-type="table"}). However, both groups differed from controls. Correlation analysis included only depressed patients. Both Arden ratios related negatively with the score in NDDS 1965, but this was significant only from the left eye (R = -0.48, p \< 0.01). Left Arden ratio marginally correlated with the number of life events (R = 0.46), and the HDRS anxiety index (R = -0.47). Concerning the existence of individual symptoms, according to DSM-IV and ICD-10 lists, patients with \'distinct quality of depressed mood\' had lower right Arden ratio values (p \< 0.001); patients who were \'worse in the morning\' had lower right Arden ratio and right A-Criterion values (p \< 0.001) and higher right dark trough values (p \< 0.001). b. flash-ERG ------------ Flash-ERG results suggested no differences between depressed patients and controls (table [1](#T1){ref-type="table"}), nor between specific symptoms and controls exist (table [2](#T2){ref-type="table"}). There were correlations between b-wave latency and GAF (left eye, R = -0.55), number of atypical features (right eye, R = -0.50), number of life events (left eye, R = -0.49), non-specific HDRS index (bilaterally, R = 0.51). There was also a positive correlation between HDRS depressed index and b-wave amplitude bilaterally (R = 0.52). Concerning the existence of individual symptoms, according to DSM-IV and ICD-10 lists, patients with \'melancholic anhedonia\' had bilaterally larger b- wave latency and those with \'thoughts of death\' (present at the time of clinical interview) had prolonged b- wave latency (p \< 0.001) Discussion ========== The alteration between light and dark produces electrochemical changes in the retina. The electro-oculogram (EOG) is a technique suitable for the study of the electrical and metabolic activity of the outer layers of the retina. The fall of a light stimuli on the retina produces early and late potentials. The method of recording late potentials is the Electroretinogram (ERG). ERG provides information about the functioning of the photoreceptors and the neuronal elements of the retina. Both EOG and ERG in fact are useful indices reflecting dopamine activity. There are several studies in the international literature concerning the relationship of dopamine with specific depressive symptoms. There is no report in the international literature on a combined use of EOG and ERG in depression. There are only papers using either method. This is one of the reasons the results and interpretations are inconclusive and problematic. a. EOG ------ The current study reports that although Arden ratios and A-criteria were within normal limits, both dark trough and light peak were reduced in all subtypes of depression. However, different mechanisms are reported to underlie them \[[@B31]\]. The light peak is related mostly to the intensity of the stimuli, while the dark through does not. Also, the light peak is related to the pre-adaptation level of the retina, while the dark trough is stable after only 2 minutes in dark. Generally, the standing potential of the eye manifests a diurnal rhythm, similar to that of the body temperature. It seems that after 15 minutes of adaptation to darkness, the amplitude of the dark trough is related only to the diurnal rhythm (in normal subjects). The correlations between EOG variables and clinical picture and psychometric scales suggested that the core feature was the relationship of the dark trough with melancholic symptoms (NDDS 1965 score). Here again should be stressed that NDDS 1965 takes into consideration premorbid personality and personal history of affective illness, while the rest melancholic scales are largely cross-sectional and do not include personality assessment. Dark trough was of course lower than in controls, but this finding suggests that the more melancholic features the patient fulfilled, the closer its dark trough amplitude was to normal. It is believed that the biochemical alterations, which produce the EOG potentials take place in the pigmentum epithelium. The origin of the light peak and dark trough probably lies in the interaction between photoreceptors and pigmentum epithelium \[[@B32]\], and dopamine seems to hold a major role \[[@B24],[@B25],[@B32]-[@B35]\]. The role of melatonin which is also reported to dysfunction in depression \[[@B12]\] remains elusive \[[@B36]\]. Thus, the EOG findings of the current study could receive two different interpretations: either dopamine activity is decreased, or an advance of the circadian cycle might be present, as already some authors have proposed \[[@B37],[@B38]\]. Of course a combination of them could be present but this is not in accord with the results of the current study, since ERG findings were not significant. It is also possible that one of them could be the result of the other. Another important finding was the relationship of dark trough with melancholia. There are no reports in the international literature concerning the different subtypes of depression. There are only a few papers, and focus on seasonal depression. Reports are inconclusive \[[@B39]-[@B45]\]. Light therapy acts on the photoreceptors, at least in the initial phase \[[@B46]\]. Lam \[[@B47]\] studied the EOG in 19 seasonal patients and reported the presence of subtle disorders in the retina, at the photoreceptors level, resulting in a decreased light sensitivity, evident from lower Arden ratios in depressed patients in comparison to controls. Terman et al \[[@B48]\] concluded that it is possible that some environmentally induced, but genetically determined state disorders of the photoreceptors contribute to the development of seasonal depression. They also suggested that these patients had light hypersensitivity due to cone hypereactivity. Beersma \[[@B49]\], suggested that this light hypersensitivity disturbs the information arriving to the hypothalamus via the retinohypothalamic tract (single neurone) and subsequently the functioning of the suprachiasmatic nucleus which seems to posses properties of an endogenous pacemaker which regulates the rhythms of the organism \[[@B50]\]. On the contrary, Reme \[[@B51]\] argues in favor of a reduced sensitivity to light in seasonal patients. The disturbed functioning \[[@B52]\] does not affect vision, but only those functions which demand prolonged exposure to light (similar to light therapy). Leaving the area of seasonal depression, which is not the direct focus of the current study, two are the only papers investigating non-seasonal depression with EOG. Seggie et al \[[@B40]\] reported that there were no differences in the Arden ratios between 20 depressed patients and equal number of controls, however depressed patients had lower dark trough values. A careful study of the paper reveals that there was no similar finding concerning the light peak, probably due to small study sample, and if the study sample was larger, such a finding could be possible. The results of that study is to a large degree similar to ours. Seggie et al concluded that depressed patients were light supersensitive and located the disturbance at the receptor level, and specifically in the rods. The authors of the current study consider that these conclusions do not really fit the data of that study. Economou and Stefanis \[[@B39]\] studied unipolar and bipolar patients and reported lower Arden ratios in unipolar and higher in bipolars in comparison to controls. They concluded that the existence and the quality of psychomotor symptomatology and not the mood of the patients is of prime importance, and related their results to disorders of dopamine activity. So, conclusively, in spite of the differences in interpretation, which is a difficult issue when only EOG is applied, the results reported in the international literature are in accord with the results of the current study. b. flash-ERG ------------ The a- wave is produced in the photoreceptors as they respond to a light stimuli. The b-wave is neuronal in origin and largely reflects dopamine activity. There are only scattered and unpublished reports (e.g. Seggie et al: Electroretinographic Changes in Depression, Proceedings of the 2^nd^Canadian Workshop on Epiphysis, 1990), and all suggest that there is an increased amplitude and decreased latency of both the rods and the cones response to the flash-ERG. These findings support the existence of light hypersensitivity in depression. Similar observations were made in animals during the transition from light to dark conditions \[[@B53]\]. It has also been suggested that the retinal disorders might relate to a toxic effect of higher neurosteroid levels, which are produced on the basis of excitatory impulses from NMDA receptors through GABA~A~receptors. This arc is also influenced by the light of the environment \[[@B54]\]. The results of the current study do not confirm the finding of light hypersensitivity. Correlation results suggest that melancholic features related positively with the photoreceptor sensitivity in darkness, and this relationship seems to lie on a continuum. c. Synthesis of findings ------------------------ The major theories related to our findings are \[[@B55]\]: a\. The phase advance hypothesis (Wehr and Wirz-Justice), which postulated that depressed patients get asleep too late in comparison to the rest of their rhythms. b\. The S deficiency hypothesis (Borbely and Wirz-Justice), which postulates that there is a disturbance in the homeostatic S procedure of sleep (reflecting the need of the organisation for sleep) c\. The adrenergic-cholinergic imbalance hypothesis of depression of Janowsky \[[@B56]\]. d\. The proposal of von Zerssen et al \[[@B57]\] which suggests that rhythms are independent from depression and just intensify or attenuate the clinical picture in the same way they affect normal mood. e\. The internal coincidence theory, which basically focuses to the time of awakening. Wehr and Wirz-Justice again suggested that there is a \'depressiogenic switch\' which normally is triggered and simultaneously inhibited by other synchronous activities; however in depressive patients the triggering occurs too early. Wehr et al \[[@B58]\] tested the above theories by depriving 4 depressed patients (however only one unipolar) patients from the environment, and thus isolating the endogenous part of the rhythms. It is important to note that all patients were impressively eager to accept this deprivation and all were improved. They all expressed discomfort when the experiment ended. This last observation is of prime importance, since it can provide further data on the relationship between psychophysiological methods and abnormal but different response to light stimuli under different conditions, and stressful life events. Another key report is that sleep deprivation, according to the review of Wu et al \[[@B59]\] immediately improves 67% of melancholic and 48% of neurotic depressives. If we combine this observation with the correlation of melancholia with the dark trough, one could conclude that higher dark trough values could predict better response to sleep deprivation. On the other hand, melancholics are considered not to respond well to light therapy and atypical (neurotic) patients share common clinical manifestations with seasonal depression. Since all depressed patients (according to the results of the current study) had low dark trough and light peak values in comparison to controls, but normal ERG, it is most possible that the initial cause could lie in the pigmentum epithelium, which secondary could affect the functioning of the receptors. The change of rhythms could cause mild affective symptomatology in normal subjects \[[@B60]\], but in depression it is unlikely to be the prime disorder. Since lesions in the pigmentum epithelium have not been yet detected, this change in the functioning should be attributed to the change of the firing of the raphe nucleus, which is considered to be an endogenous pacemaker. There is no possibility of a spreading of the frontal lobe metabolism dysfunction seen in depression, to the retina, since, in the vast majority of cases, the ophthalmic artery stems from the internal carotid artery. However, since no differences were evident between melancholic and atypical patients, the source of the difference in PR-VEPs latency between these two depressive subtypes \[[@B14]\] should be traced posterior to the retina and anterior to the chiasm. The problem is that the neurons that constitute the optic nerve have their body located in the ganglionic layer of the optic nerve, which constitutes the outer layer of the cerebral stratum, while their axons terminate in the lateral geniculate body. It is obvious that the part of the optic nerve from the retina to the chiasm constitutes only a part of the optic nerve axon, and thus it is very difficult to explain any dysfunction, which is so narrowly localized. The only thing that differentiates this specific area is the fact that its blood supply come from small vessels originating mainly from the anterior cerebral artery \[61, 62 and 63\] There is another possibility. EOG, ERG and PR-VEPs are three different methods which can not be used simultaneously. Therefore, there might be some specific features (e.g. eye micromovements) which have different influence on each of these tests or are activated or deactivated during anyone of these tests, and thus contribute to the results reported. In this case, our effort to localize the dysfunction on the base of the results of our studies so far is in vain. The advantages of the current study include the precise diagnosis according to modern diagnostic criteria and the detailed psychometric assessment. The major disadvantage is the deviations from the International Society for Clinical Electrophysiology of Vision (ISCEV) standards for the recordings of EOG and ERG. Conclusion ========== The main finding of the current study concerns the lower dark trough and light peak values while ERG findings were normal in all depressive subtypes. The above provide the empirical foundation in order to incorporate the reports of the international literature in a comprehensive theory, which could explain the mechanism of action of therapies like sleep deprivation and light therapy. Competing interests =================== The author(s) declare that they have no competing interests.	PubMed Central	2024-06-05T03:55:56.278999	2005-3-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088013/", "journal": "Ann Gen Psychiatry. 2005 Mar 29; 4:7", "authors": [ { "first": "Konstantinos N", "last": "Fountoulakis" }, { "first": "Fotis", "last": "Fotiou" }, { "first": "Apostolos", "last": "Iacovides" }, { "first": "George", "last": "Kaprinis" } ] }
PMC1088276	Introduction {#s1} ============ The antigen-specific interaction between B cells and helper T cells in secondary lymphoid organs is an essential step in T-dependent humoral immune responses \[[@pbio-0030150-b01]\]. These early B cell--T cell (B-T) interactions occur in the border area between follicles and T zones after B and T cells are antigen primed in B cell follicles and in T cell zones, respectively. During cognate B-T interaction, B cells present a specific antigen to helper T cells and receive cytokine signals requisite for their survival, proliferation, and differentiation \[[@pbio-0030150-b02]\]. The relocation of antigen-engaged B and T cells to the boundary from two adjacent but distinct microenvironments is believed to contribute to the prompt encounter of B and T cells with cognate-antigen specificity \[[@pbio-0030150-b01],[@pbio-0030150-b03]\]. The localization of antigen-engaged B cells at the B-zone--T-zone (B/T) boundary, which was initially discovered by tracking hapten-specific memory B cells \[[@pbio-0030150-b04]\], has been well characterized for naive cells by examining the distribution of immunoglobulin-transgenic (Ig-tg) B cells at various times after antigen exposure \[[@pbio-0030150-b05]\]. These immunohistochemical studies established that the relocalization process is rapid, being complete within 6 h. A recent study showed this relocalization is driven by upregulation in B cells of CCR7 ---the receptor for T-zone chemokines CCL21 and CCL19 \[[@pbio-0030150-b06]\]. However, these experiments have not established whether movement to the boundary occurs by directed, chemotactic movement or by random migration followed by retention by adhesive interactions at the boundary. Antigen-specific B-T conjugates in the boundary area have been revealed by immunohistochemical analysis of Ig-tg B cells and T cell receptor (TCR)--transgenic CD4^+^ T cells in spleen and lymph node sections \[[@pbio-0030150-b07],[@pbio-0030150-b08]\]. When equal numbers of antigen-specific cells were given, cognate B-T contacts were more frequent than noncognate B-T contacts in lymph node snapshots \[[@pbio-0030150-b07]\]. The dynamics of B-T interactions and knowledge of the spatiotemporal organization of molecules at the B-T interface have also been studied by imaging living cells on cover slips, \[[@pbio-0030150-b02],[@pbio-0030150-b09]\], but it is unclear how these observations relate to events in vivo. Recently, approaches have been developed to image living cells within intact lymphoid tissues \[[@pbio-0030150-b10]--[@pbio-0030150-b14]\]. Two-photon microscopy provides advantages over other imaging approaches, including a greater imaging depth in tissues and less phototoxicity; this technique has been used to image the dynamic nature of naive lymphocytes and the interactions between T cells and dendritic cells (DCs) \[[@pbio-0030150-b13]--[@pbio-0030150-b18]\]. Imaging has been performed both in explanted lymph nodes incubated at 37 ^o^C with oxygen-perfused medium \[[@pbio-0030150-b11],[@pbio-0030150-b16]--[@pbio-0030150-b18]\] and by intravital microscopy of lymph nodes \[[@pbio-0030150-b13],[@pbio-0030150-b14]\]. Naive lymphocytes, antigen-primed T cells, and DCs showed very similar motility in these two experimental configurations. These studies have revealed that T lymphocytes move in lymph node T zones at 10--12 μm/min, and B cells move within follicles at about 6 μm/min, with both cell types exhibiting what appears as a random walk. Surprisingly, despite the many studies of lymphocyte chemotaxis in vitro and our understanding of chemokine requirements for lymphocyte homing and architecture in the lymphoid organs, evidence of directional migration of lymphocytes has not yet been obtained from ex vivo or in vivo microscopy of lymph nodes \[[@pbio-0030150-b19]\]. Furthermore, although much has recently been discovered about the properties of T cell--DC interactions \[[@pbio-0030150-b14],[@pbio-0030150-b16]--[@pbio-0030150-b18],[@pbio-0030150-b20],[@pbio-0030150-b21]\], no information is yet available on the dynamics of B-T interactions during T cell--dependent antibody responses. In this study we used two-photon microscopy of explanted mouse lymph nodes to study the behavior of antigen-primed B cells and to track their interactions with helper T cells during the first stages of a T cell--dependent antibody response. We show that relocation of antigen-engaged B cells to the B/T boundary is achieved by random migration far from the follicle boundary in combination with directional migration in the region extending up to approximately 140 μm from the boundary. Directional migration was dependent on CCR7 expression by the B cells and a gradient of CCL21 was detected extending into the follicle. Furthermore, we have tracked the dynamics of B-T encounters and demonstrate that cognate B-T interactions can last for at least 1 h, whereas noncognate interactions last only a few minutes. Antigen-specific B-T conjugates are usually monogamous and migrate extensively in the interfollicular region at approximately 9 μm/min. Results {#s2} ======= Directional Migration of Antigen-Engaged B Cells to the B/T Boundary {#s2a} -------------------------------------------------------------------- To investigate the migration dynamics of antigen-engaged B cells, we transferred fluorescently labeled hen egg lysozyme (HEL)-specific Ig-tg B cells (green) and nontransgenic (non-tg) B cells (red) into syngeneic mice, allowed 1 d for the cells to equilibrate within lymphoid tissues, and then tracked migration of the cells in inguinal lymph nodes excised from the recipient animals 1 h after intravenous HEL injection. By flow cytometric analysis of dissociated lymph node cells, the B cell receptors on the Ig-tg B cells were shown to be fully occupied by HEL antigen (data not shown). Extensive immunohistochemical analysis of the distribution of Ig-tg B cells in lymphoid tissues isolated at various times after HEL injection has established that the HEL-engaged B cells accumulate at the boundary of the follicle and T zone, including in interfollicular regions \[[@pbio-0030150-b05],[@pbio-0030150-b06],[@pbio-0030150-b22]\]. With our current microscope configurations, two-photon imaging through the capsule was restricted to depths of approximately 200 μm. This was too shallow to permit extensive imaging of the follicle boundary distal to the capsule, but it did allow analysis of boundary regions at the sides of follicles, which typically corresponded to interfollicular regions ([Figures 1](#pbio-0030150-g001){ref-type="fig"}A and [S1](#sv001){ref-type="supplementary-material"}). Images of intact lymph nodes at the time of isolation showed that HEL-specific B cells were distributed together with non-tg B cells throughout the follicles (0 min in [Figures 1](#pbio-0030150-g001){ref-type="fig"}A, [1](#pbio-0030150-g001){ref-type="fig"}B, and [S1](#sv001){ref-type="supplementary-material"}). Within 1 h of imaging, accumulation of HEL-specific B cells along the rim of the follicles became evident, whereas non-tg B cells remained uniformly distributed in the follicles ([Figure 1](#pbio-0030150-g001){ref-type="fig"}B and [1](#pbio-0030150-g001){ref-type="fig"}C; Videos [S1](#sv001){ref-type="supplementary-material"} and [S2](#sv002){ref-type="supplementary-material"}). Analysis of individual naive non-tg B cells revealed that these cells moved in an apparently random manner in the follicle with a velocity of approximately 7 μm/min ([Figure 1](#pbio-0030150-g001){ref-type="fig"}D and [1](#pbio-0030150-g001){ref-type="fig"}E), in agreement with previous studies \[[@pbio-0030150-b11]\]. By contrast, HEL-binding B cells tracked 1--3 h after antigen injection showed reduced velocities, moving at about 4 μm/min ([Figure 1](#pbio-0030150-g001){ref-type="fig"}E). Within 6 h after antigen challenge, the velocities of activated B cells were restored, and by 24 h, the cells became more motile than naive B cells ([Figure 1](#pbio-0030150-g001){ref-type="fig"}F; [Video S3](#sv003){ref-type="supplementary-material"}). ::: {#pbio-0030150-g001 .fig} Figure 1 ::: {.caption} ###### Antigen-Engaged B Cells Reduce Random Motility and Migrate Toward the Follicle--T Zone Boundary \(A) On the left is a diagram showing the region of the inguinal lymph node that was imaged. The right panel shows the xz projection view of an image stack collected immediately prior to time-lapse imaging, demonstrating the location of a B cell follicle containing transferred B cells (green and red). The collagen-rich lymph node capsule is visualized by second harmonic emission (blue). The dashed white rectangle shows the region used in the time-lapse image analysis. \(B) Time-lapse images of HEL-engaged Ig-tg B cells (green) clustering at the follicle--T zone boundary and naive non-tg B cells (red) in the follicle. The 0-min image is approximately 1 h after antigen exposure (see [Video S1](#sv001){ref-type="supplementary-material"}). The pathways of an Ig-tg B cell (light blue circle and dotted line) and a non-tg B cell (pink circle and dotted line) are traced as examples. The traced non-tg B cell moved out of the imaging stack at the 57.5-min timepoint. Scale as in (C). (C and D) Tracks of antigen-engaged Ig-tg (C) and naive non-tg (D) B cells in the xy-, xz-, and yz-planes. The boundary as defined by the area of Ig-tg B cell accumulation at 120 min is shown in lighter gray. Tracks cover 30 to 112 min (C) and 30 to 57 min (D). Circles indicate the end point of tracking. \(E) Velocity distribution for naive (upper histogram) and antigen-engaged (lower histogram) B cells. Data are shown for 44 Ig-tg cells and 31 non-tg cells. Medians are indicated by arrows. \(F) Velocity distributions for non-tg B cells at 1--20 h (red, n = 25) or Ig-tg B cells at 1--3 h (green, n = 29), 6--8 h (light blue, n = 21), or 18--20 h (dark blue, n = 39) after antigen injection. The data are for cells imaged in the boundary regions. Medians are indicated by arrows. \(G) Ratios of log (displacement) to log (path length) of naive B cells (20 cells), antigen-engaged B cells that moved to the boundary (22 cells), or antigen-engaged B cells that did not move to the boundary (22 cells). Points indicate data for individual cells collected at cumulative 30-sec time intervals. The formula for each regression line is indicated. The correlation coefficients, R ^2^, were 0.63, 0.80, and 0.67, respectively. Data are from three experiments and are representative of 12 to ∼18 cells from each recording. ::: ![](pbio.0030150.g001) ::: To determine whether antigen-engaged B cells accumulated at the B/T boundary because of directional movement or as a result of random migration followed by retention, we tracked the migration of HEL-specific and non-tg B cells. Of 44 HEL-specific cells that could be tracked for more than 30 min, 22 cells migrated to reach the B/T boundary ([Figure 1](#pbio-0030150-g001){ref-type="fig"}C, tracking data for 12 HEL-specific cells are shown, with seven of the cells reaching the boundary), whereas of 20 non-HEL-specific cells, only four were found at the boundary after 30 min ([Figure 1](#pbio-0030150-g001){ref-type="fig"}D, tracking data for 11 non-tg cells are shown, with two cells reaching the boundary). Of the Ig-tg B cells that failed to localize to the boundary, 11 cells were only marginally displaced from their starting positions (maximum displacement \<20 μm; two examples are shown in [Figure 1](#pbio-0030150-g001){ref-type="fig"}C as green and light-purple tracking lines). The remaining cells (11/44) did show substantial migration despite failing to move to the boundary (maximum displacement \>20 μm; examples shown in beige, dark-brown, and yellow tracking lines in [Figure 1](#pbio-0030150-g001){ref-type="fig"}C). To quantify the directionality of migration, we constructed double-logarithmic plots of the net displacement of cells from their starting point against their cumulative path length. In the case of linear motion, a slope of one is expected, whereas the net displacement increases as the square root of path length for random motion, resulting in a slope of 0.5 on log/log axes. The HEL-engaged cells that reached the boundary showed a slope closer to one than either HEL-engaged cells that failed to move to the boundary or naive cells ([Figure 1](#pbio-0030150-g001){ref-type="fig"}G). These data indicate that the migration path taken by antigen-engaged B cells that reached the boundary was closer to a straight line than that of cells not reaching the boundary or of naive B cells, suggesting that migration to the boundary was directional. A difficulty with measuring migration of activated B cells to the B/T boundary was that the cells being tracked would frequently leave the imaging volume before reaching the boundary. To further quantify directionality and also test its relationship to distance from the boundary, cells that migrated through randomly selected cubic volumes of follicle were tracked for short periods to determine whether they were migrating toward or away from the B/T boundary ([Figure 2](#pbio-0030150-g002){ref-type="fig"}). We defined 30 μm (x-axis) × 30 μm (y-axis) × 30 μm (z-axis) cubes at various positions within follicles that had been imaged from 1 to 3 h after antigen injection, and the migration of cells emerging from each cube was tracked. The xy positions of the follicle--T zone boundary were not greatly changed throughout the 50-μm z depth (not shown), and we therefore plotted the cell tracking lines in the xy coordinate of the z-projection view and counted the number of cells that crossed the sides of the square facing the follicle--T zone boundary ([Figure 2](#pbio-0030150-g002){ref-type="fig"}B and [2](#pbio-0030150-g002){ref-type="fig"}C, sides shown as solid lines) versus those that crossed the opposite sides ([Figure 2](#pbio-0030150-g002){ref-type="fig"}B and [2](#pbio-0030150-g002){ref-type="fig"}C, sides shown as dashed lines). For boxes located 20 to approximately 80 μm away from the B/T boundary (shown in blue in [Figure 2](#pbio-0030150-g002){ref-type="fig"}B), 50% of the cells emerged across the sides facing the boundary whereas 18% migrated across the opposite sides ([Figure 2](#pbio-0030150-g002){ref-type="fig"}C). Similarly, for the boxes located 80 to approximately 140 μm away from the follicular boundary (shown in purple in [Figure 2](#pbio-0030150-g002){ref-type="fig"}B), more cells migrated across the two sides facing the boundary than across the other sides ([Figure 2](#pbio-0030150-g002){ref-type="fig"}C). However, for boxes located 140 to approximately 200 μm away from the boundary (pink box in [Figure 2](#pbio-0030150-g002){ref-type="fig"}B), the number of cells that migrated across the sides facing the boundary was not significantly higher than the number of cells that migrated across the opposite sides ([Figure 2](#pbio-0030150-g002){ref-type="fig"}C). In contrast, the fraction of non-tg B cells that migrated across the sides of the box facing the B/T boundary versus the other sides was essentially even for every box examined, showing no directionality of migration toward the boundary ([Figure 2](#pbio-0030150-g002){ref-type="fig"}D). Although the B/T boundary was also located below the box, the number of cells leaving via the top and bottom of each box was not included in this analysis because the z distance to the boundary could not be accurately determined in two of the experiments. However, as with the above observations, we observed a tendency for more cells to leave via the bottom of the boxes than via the top (data not shown). Plots of displacement over path length for antigen-engaged B cells emerging from each of the cubes revealed that when cells were within approximately 140 μm of the boundary, they showed less turning when migrating toward the boundary than when migrating away from the boundary ([Figure 2](#pbio-0030150-g002){ref-type="fig"}E). By contrast, non-antigen-engaged B cells and antigen-engaged B cells that were more distant from the boundary showed low displacement to path-length ratios irrespective of their direction of migration ([Figure 2](#pbio-0030150-g002){ref-type="fig"}E). Taken together, these observations indicate that when antigen-engaged B cells are within approximately 140 μm of the B/T boundary, they are able to conduct directional migration toward the boundary. ::: {#pbio-0030150-g002 .fig} Figure 2 ::: {.caption} ###### Relationship between Directionality of Ag-Engaged B Cell Migration and Distance from the Follicular Boundary \(A) The xz projection view of an image stack collected immediately prior to time-lapse imaging, demonstrating the location of the B cell follicle containing transferred B cells (green and red). The collagen-rich lymph node capsule is visualized by second harmonic emission (blue). The dashed white rectangle shows the region used in the time-lapse image analysis. \(B) Time-lapse images of HEL-engaged Ig-tg B cells (green) clustering at the follicle--T zone boundary and naive non-tg B cells (red) in the follicle. The 0-min image is approximately 1 h after antigen exposure (see [Video S2](#sv002){ref-type="supplementary-material"}). Square boxes indicate regions used for directionality analysis, shown in (C). Scale bar is as shown in (A). \(C) Tracks of antigen-engaged B cells originating from 30-μm follicular cubes. The centers of the boxes are placed proportionally to the actual positions of the boxes shown in (A), with each box corresponding to a 30 μm × 30 μm × 30 μm cube. Tracks cover 8 to 95 min. The histograms show the percentage of cells that moved across the sides of the square (solid lines) that face the boundary (left histograms) and the percentage of cells that moved across the opposite sides (dashed lines) of the squares (right histograms). The bottom, middle, and top histograms show the results with cubes that are 20--80, 80--140, and 140--200 μm away, respectively, from the boundary. Data shown in the histograms are pooled from three experiments. \, p* \< 0.05; \\, p \< 0.01. \(D) Tracks of naive B cells originating from inside of the middle blue box shown in (A). Tracks cover 5 to 25 min. The histogram shows the percentage of cells that moved across the solid or dashed sides of cubes that are 20--80 μm away from the boundary. Data shown in the histogram are from three experiments. \(E) The dot plots show ratios of the displacement to the path length of 8-min tracks of antigen-engaged B cells (left three graphs) or naive B cells (far right graph) originating from the cubes described in (B--D). The distances from the cubes to the boundary are indicated above the graphs. The left and right plots in each graph are the data of tracks that cross the solid and dashed sides of the cubes, respectively. All (100%) of Ig-tg B cells and 94% of non-tg B cells that crossed the sides of the cubes could be tracked for 8 min. The means of each data group are shown as red bars. \, p* \< 0.05; \\, p \< 0.01. ::: ![](pbio.0030150.g002) ::: Detection of a CCL21 Gradient {#s2b} ----------------------------- As observed previously for cells in the spleen \[[@pbio-0030150-b06]\], CCR7 was required for antigen-engaged B cells in lymph nodes to localize at the B/T boundary, shown by comparing the distribution of Ig-tg B cells from CCR7^−/−^ and CCR7^+/+^ mice in HEL-injected recipients ([Figure 3](#pbio-0030150-g003){ref-type="fig"}A and [3](#pbio-0030150-g003){ref-type="fig"}B). We also observed that in the absence of antigen-receptor engagement, CCR7^−/−^ B cells tended to localize in regions of lymph node follicles that are distal to the T zone ([Figure 3](#pbio-0030150-g003){ref-type="fig"}C), suggesting that CCR7 is required for naive B cells to migrate efficiently through the region of the follicle near the T-zone border. The CCR7 ligand CCL21 is abundantly expressed within the T zone \[[@pbio-0030150-b23]\]. To test whether a CCL21 gradient could be detected extending from the T zone into the follicle, we performed immunohistochemical staining of adjacent lymph node sections for a B cell marker (B220), to locate the boundary between the follicle and the T zone ([Figure 3](#pbio-0030150-g003){ref-type="fig"}C), and for CCL21 ([Figure 3](#pbio-0030150-g003){ref-type="fig"}D). CCL21 was concentrated in the T zone as expected, but staining was also observed in regions of B cell follicles that are proximal to the T zone ([Figure 3](#pbio-0030150-g003){ref-type="fig"}C and [3](#pbio-0030150-g003){ref-type="fig"}D). To quantify the CCL21 staining in the follicle, we converted the images to gray scale and measured the intensity of gray pixels along lines drawn from the outer follicle to the T zone ([Figure 3](#pbio-0030150-g003){ref-type="fig"}E). This approach confirmed that CCL21 was present in an increasing gradient from regions of follicles that are distal to the T zone, to the B/T boundary. An increasing CCL21 gradient could also be observed extending 150--200 μm from inside the follicle to the interfollicular region, the boundary region typically imaged in this study by two-photon microscopy. The gradient was not observed in lymph nodes from plt/plt mice, which lack CCL21 expression in lymphoid tissues, although a weak signal was sometimes detected in interfollicular regions, perhaps due to small amounts of CCL21 entering the lymph node via afferent lymphatics ([Figure 3](#pbio-0030150-g003){ref-type="fig"}D and [3](#pbio-0030150-g003){ref-type="fig"}E, "plt/plt LN") \[[@pbio-0030150-b24]\]. Quantitation of the staining pattern generated with antibodies against CD3, another protein abundant in the T zone, failed to reveal any evidence of a gradient ([Figure 3](#pbio-0030150-g003){ref-type="fig"}C--[3](#pbio-0030150-g003){ref-type="fig"}E, bottom panels), showing that there was minimal diffusion of the fast-red reaction product generated during immunohistochemical staining and confirming that CCL21 was present as a gradient. ::: {#pbio-0030150-g003 .fig} Figure 3 ::: {.caption} ###### CCL21 Concentration Gradients in B Cell Follicles in Lymph Nodes (A and B) Lymph node sections of wild-type mice that received CCR7^+/+^ Ig-tg (A) or CCR7^−/−^ Ig-tg B cells (B) and 1 mg of HEL intravenously for 6 h, stained to detect all B cells (brown) and HEL-binding B cells (blue). \(C) B cell (upper three panels) or T cell (lower panel) staining of lymph node sections from wild-type and plt/plt mice as indicated. The wild-type lymph nodes shown in the upper panels were from Igh^a^ mice that had received CCR7^−/−^ Igh^b^ B cells 1 d before. Staining was to detect all B cells (B220, brown) and transferred CCR7^−/−^ B cells of the Igh^b^ allotype (dark blue) or T cells (CD3, red) as indicated. (D and E) Detection of CCL21 concentration gradients in follicle. In (D), adjacent sections were stained for CCL21. The dashed lines indicate the B/T boundary determined by the distribution of B220 staining. The pixel intensity at each position along the filled line extending from the follicle into the deep T zone (line 1) or the interfollicular T zone (line 2) was averaged across 50 μm perpendicular to the line (the width shown by the bar at the T zone end of each line) and plotted against distance from the follicular end of the line (E) in red for line 1 and blue for line 2. As a negative control, the pixel intensity was determined along a line located in the equivalent location in a CD3-stained section (bottom panel of \[C\] and shown as a green line in \[E\]). Pixel intensity was measured using Metamorph software after converting colored pixels to gray scale. A decrease in transmitted light intensity indicates an increase in CCL21. Data are representative of more than five lymph nodes from two mice of each type. ::: ![](pbio.0030150.g003) ::: CCR7 Is Required for Directional Migration of Antigen-Engaged B Cells {#s2c} --------------------------------------------------------------------- To further test whether directional migration of antigen-engaged B cells to the B/T boundary occurred in response to a CCR7 ligand gradient, we tracked the migration of carboxyfluorescein diacetate succinimidyl ester (CFSE)--labeled CCR7^−/−^ Ig-tg B cells in wild-type recipients 1--4 h after antigen exposure. 5-(and-6)-(((4-chloromethyl)benzoyl)amino)tetramethylrhodamine (CMTMR)--labeled CCR7^+/+^ Ig-tg B cells were co-transferred with the CCR7^−/−^ cells as an internal control. As expected, the CCR7^+/+^ cells were observed to accumulate at the edge of the follicle ([Figure 4](#pbio-0030150-g004){ref-type="fig"}; [Video S4](#sv004){ref-type="supplementary-material"}). By contrast, although the CCR7^−/−^ cells moved at a similar velocity as the wild-type cells ([Figure 4](#pbio-0030150-g004){ref-type="fig"}F), they failed to accumulate at the boundary ([Figure 4](#pbio-0030150-g004){ref-type="fig"}A; [Video S4](#sv004){ref-type="supplementary-material"}). Measurements of the number of cells that migrated from the boxed areas shown in [Figure 4](#pbio-0030150-g004){ref-type="fig"}A--[4](#pbio-0030150-g004){ref-type="fig"}C revealed that the CCR7^−/−^ cells did not display directional bias in their migration, in contrast to CCR7^+/+^ cells in the same region, which displayed preferential migration toward the boundary ([Figure 4](#pbio-0030150-g004){ref-type="fig"}B--[4](#pbio-0030150-g004){ref-type="fig"}D). As well as quantitating the direction in which cells exited from the boxes, we also measured the number of cells entering the boxes. More than twice as many wild-type Ig-tg B cells entered from the sides located away from the boundary as via the sides adjacent to the boundary, whereas approximately equal numbers of CCR7^−/−^ cells entered through the distal and adjacent sides (data not shown), confirming that CCR7^+/+^ but not CCR7^−/−^ cells showed directional migration toward the boundary. As observed above, wild-type B cells that migrated toward the boundary had an increased displacement to path-length ratio compared to cells migrating in the opposite direction ([Figure 4](#pbio-0030150-g004){ref-type="fig"}E). By contrast, CCR7^−/−^ cells that emerged from the boxes in the direction of the boundary had the same low displacement to path-length ratio as cells that were migrating in the opposite direction. These findings establish that directional migration of antigen-engaged B cells toward the B/T boundary is CCR7 dependent. ::: {#pbio-0030150-g004 .fig} Figure 4 ::: {.caption} ###### Antigen-Engaged CCR7^−/−^ B Cells Fail to Show Directional Migration toward the Follicle--T Zone Boundary \(A) Time-lapse images of HEL-engaged CCR7^−/−^ (green) and CCR7^+/+^ (red) Ig-tg B cells in an inguinal lymph node. The 0-min image is \~3.5 h after antigen exposure (see [Video S4](#sv004){ref-type="supplementary-material"}). Square boxes indicate regions used for directionality analysis, shown in (B). A CCR7^−/−^ Ig-tg B cell (light blue circle and line) and a CCR7^+/+^ Ig-tg B cell (pink circle and line) are traced as examples. (B and C) Tracks of antigen-engaged CCR7^+/+^ (B) and CCR7^−/−^ (C) Ig-tg B cells originating from 30-μm follicular cubes, analyzed as described in [Figure 2](#pbio-0030150-g002){ref-type="fig"}. Tracks cover 3--17 min for CCR7^+/+^ cells and 3--22 min for CCR7^−/−^ cells. \(D) Dot plot showing the percentage of cells that moved across the solid or dashed sides of the cubes. Filled symbols correspond to the data shown in (A--C) and [Video S4](#sv004){ref-type="supplementary-material"}, open symbols correspond to data obtained in a second time-lapse movie collected 2.5--3.5 h after antigen injection. The cubes were located approximately 20--80 μm from the site of accumulation of CCR7^+/+^ Ig-tg B cells. \(E) The dot plots show ratios of the displacement to the path length of 8-min tracks of antigen-engaged CCR7^+/+^ or CCR7^−/−^ Ig-tg B cells originating from the cubes described in (A--D). The left and right plots for each cell graph are the data of tracks that cross the solid and dashed sides of the cubes, respectively. A total of 74% of wild-type B cells and 81% of CCR7^−/−^ B cells that crossed the sides of the cubes could be tracked for 8 min. The means of each data group are shown as red bars. \, p* \< 0.01. \(F) Velocity distribution data for CCR7^−/−^ Ig-tg B cells (green, n = 27) and CCR7^+/+^ Ig-tg B cells (red, n = 40) tracked 2.5--4.2 h after antigen injection. The data are pooled from two time-lapse movies. Medians are indicated by arrows. ::: ![](pbio.0030150.g004) ::: Dynamics of Interactions between Antigen-Engaged B Cells and Helper T Cells {#s2d} --------------------------------------------------------------------------- After arrival at the B/T boundary, antigen-engaged B cells interact with activated helper T cells. In order to observe antigen-specific interactions between B cells and helper T cells, we utilized class II (I-A^b^)--restricted HEL-specific TCR-transgenic mice (TCR7; M. Neighbors, S. B. Hartley, and A. O\'Garra, unpublished data). The Ig-tg B cells were co-transferred with CD4^+^ TCR7 T cells into recipient mice that had been immunized subcutaneously with HEL plus adjuvant for 8 h ([Figure 5](#pbio-0030150-g005){ref-type="fig"}). The TCR7 T cells and Ig-tg B cells in draining lymph nodes started proliferating between 1 and 2 d after transfer ([Figure 5](#pbio-0030150-g005){ref-type="fig"}A and [5](#pbio-0030150-g005){ref-type="fig"}B). Consistent with a previous report \[[@pbio-0030150-b07]\], B cell proliferation was delayed compared to T cell proliferation and peaked 4 to 5 d after transfer. B cell proliferation was not observed if TCR7 T cells were not co-transferred ([Figure 5](#pbio-0030150-g005){ref-type="fig"}A). By contrast, T cell activation and proliferation occurred even in the absence of transferred B cells (data not shown), consistent with other studies indicating that after injection of antigen in adjuvant, T cells are rapidly activated within the T zone by antigen-bearing DCs \[[@pbio-0030150-b18],[@pbio-0030150-b25]\]. One day after transfer, Ig-tg B cells localized in compact clusters with TCR7 T cells in interfollicular regions ([Figure 5](#pbio-0030150-g005){ref-type="fig"}C). This pattern was distinct from the uniform distribution of Ig-tg B cells along the B/T boundary upon antigen challenge in the absence of transferred T cells (see [Figures 3](#pbio-0030150-g003){ref-type="fig"}A and [5](#pbio-0030150-g005){ref-type="fig"}G) \[[@pbio-0030150-b05]\], suggesting that B cell clustering is promoted by TCR7 helper T cells. Two days after transfer, when TCR7 T cells had proliferated robustly, Ig-tg B cells remained in interfollicular regions, although the clusters seemed to be less compact ([Figure 5](#pbio-0030150-g005){ref-type="fig"}D). After 5 d in the presence of antigen and helper T cells, Ig-tg B cells formed germinal centers and also differentiated into plasma cells based on syndecan-1 staining ([Figure 5](#pbio-0030150-g005){ref-type="fig"}E and [5](#pbio-0030150-g005){ref-type="fig"}F). ::: {#pbio-0030150-g005 .fig} Figure 5 ::: {.caption} ###### Kinetics of HEL-Specific B Cell Expansion and Differentiation in the Presence of HEL-Specific TCR-Transgenic Helper T Cells \(A) Numbers of Ig-tg B cells and TCR7 CD4^+^ T cells in draining (two inguinal) lymph nodes plotted against time after cell transfer. Shown are means ± standard errors of more than three experiments. Immunization of recipient mice with HEL in adjuvant was done 8 h before cell transfer. \(B) Time course of B and T cell division in draining lymph nodes, determined by CFSE dilution. (C--F) Immunohistochemical analysis of draining lymph nodes at days 1--5 following immunization, stained as indicated. Note redistribution of B cells into interfollicular clusters in the presence of helper T cells. An enlarged version of the boxed region in (C) is shown in [Figure S2](#sg002){ref-type="supplementary-material"}. \(G) Distribution of Ig-tg B cells at day 1 in the absence of helper T cells, stained as indicated. Objective magnification in (C), (D), and (G), 5×, and in (E) and (F), 10×. ::: ![](pbio.0030150.g005) ::: Having established that our adoptive transfer system supported T-dependent B cell antibody responses, we performed two-photon microscopy of fluorescently labeled Ig-tg B cells and TCR7 T cells to analyze the dynamics of B-T interactions in draining lymph nodes. One day after cell transfer, which corresponds to \~30 h after challenge with antigen in adjuvant, Ig-tg B cells and TCR7 T cells were observed in focal clusters and were highly motile. On the basis of immunohistochemical analysis ([Figures 5](#pbio-0030150-g005){ref-type="fig"}C, [5](#pbio-0030150-g005){ref-type="fig"}D, and [S2](#sv002){ref-type="supplementary-material"}) and the fact that the inguinal lymph nodes were imaged in the same orientation and at similar depths as in [Figures 1](#pbio-0030150-g001){ref-type="fig"} and [2](#pbio-0030150-g002){ref-type="fig"}, we presume that these clusters were in interfollicular regions. The Ig-tg B cells formed stable conjugates with TCR7 T cells, but typically the conjugate pair remained highly motile (see [Videos S5--S7](#sv005){ref-type="supplementary-material"}; Ig-tg B cells, red; TCR7 T cells, green). As conjugates moved, B cells always moved in front of helper T cells, and when making turns, B cells turned first and the T cell partner followed ([Figure 6](#pbio-0030150-g006){ref-type="fig"}A--[6](#pbio-0030150-g006){ref-type="fig"}C; see also Videos [S5](#sv005){ref-type="supplementary-material"} and [S6](#sv006){ref-type="supplementary-material"}), suggesting that the conjugate pair was led by the B cell partner. Moreover, the instantaneous three-dimensional velocity of B cells in monogamous conjugates (approximately 9 μm/min) was similar to the motility of B cells free of helper T cells ([Figure 6](#pbio-0030150-g006){ref-type="fig"}D and [6](#pbio-0030150-g006){ref-type="fig"}E). In contrast, the motility of T cells in monogamous conjugates (approximately 9 μm/min) was virtually identical to the motility of the activated B cells and lower than the motility of free helper T cells (approximately 14 μm/min; [Figure 6](#pbio-0030150-g006){ref-type="fig"}D and [6](#pbio-0030150-g006){ref-type="fig"}E). Further imaging made at various times between 40 and 64 h after immunization revealed similar dynamics of B-T interaction. [Video S7](#sv007){ref-type="supplementary-material"} provides four time-lapse image segments showing the spatiotemporal changes induced in Ig-tg B cell and TCR7 T cell distribution, motility, and interactions from the time of antigen challenge until 64 h after challenge, when extensive T cell division is occurring. ::: {#pbio-0030150-g006 .fig} Figure 6 ::: {.caption} ###### Dynamics of Antigen-Engaged B Cell--Helper T Cell Interactions \(A) Time-lapse images of Ig-tg B cells interacting with TCR7 CD4^+^ T cells \~30 h after immunization with HEL in adjuvant, showing T cells moving along behind B cells. The pathways of a B cell (pink dotted line) and a T cell (blue dotted line) remaining bound to each other for more than 1 h are shown (see also [Video S5](#sv005){ref-type="supplementary-material"}). \(B) Time-lapse images showing the dynamics of a B-T conjugate. \(C) The t-x, t-y, t-z plots of the interacting B (red line) and T (green line) cells traced in (A). Note the B cell makes turns before the T cell (arrows). (D and E) Velocity measurements of unpaired (D) and paired (E) B and T cells, showing that paired T cells slow to the velocity of the B cell. Velocity data are from 16 cells of each type. \(F) Time-lapse images of a B cell interacting with two T cells. \(G) Time-lapse images showing an encounter of a B cell and a T cell to form a conjugate. ::: ![](pbio.0030150.g006) ::: Although stable B-T conjugates were largely monogamous, polygamous conjugates consisting of one B cell sandwiched by multiple T cells were also observed. The polygamous conjugates were less motile than the monogamous conjugates ([Figure 6](#pbio-0030150-g006){ref-type="fig"}F; [Video S8](#sv008){ref-type="supplementary-material"}). On the other hand, we never observed a single T cell forming stable conjugates with multiple B cells for more than 5 min. In instances when one T cell contacted multiple B cells, the T cell typically followed one of the B cells within several minutes ([Video S9](#sv009){ref-type="supplementary-material"}). Some B cells were also observed to switch T cell partners instead of keeping contacts with both T cells ([Video S10](#sv010){ref-type="supplementary-material"}). Encounters of B and T cells took place in a variety of ways: T cells approached B cells, B cells approached T cells, and T cells and B cells collided while migrating. During the initial formation of a conjugate pair, T cells sometimes remained attached by a tether to the B cell before rounding up ([Figure 6](#pbio-0030150-g006){ref-type="fig"}G). As the conjugate pair moved, contact was sometimes maintained by a B cell tether extending from the trailing edge of the B cell to the T cell ([Figure 6](#pbio-0030150-g006){ref-type="fig"}G; [Video S11](#sv011){ref-type="supplementary-material"}). Such tethers are consistent with a motile B cell dragging a passive T cell behind it. To determine whether the formation of stable B-T conjugates was due to cognate-antigen specificity, we examined the interactions that occurred between B and T lymphocytes specific for two different antigens. OT-II CD4^+^ T cells, which have a class II--restricted ovalbumin (OVA)--specific TCR, were transferred together with Ig-tg B cells into recipient mice immunized as for the above studies except using a combination of HEL and OVA. By this approach, both B cells and T cells become activated, but the HEL-specific B cells are not expected to present OVA peptides to the T cells. In contrast to antigen-specific interactions, contacts between Ig-tg B cells and OT-II T cells were short lived ([Video S12](#sv012){ref-type="supplementary-material"}). Measurements of B-T contact times in time-lapse recordings made 30--50 h after immunization with antigen in adjuvant revealed that many (81/150, or 54%) of the antigen-specific contacts lasted longer than 8 min, whereas the great majority (90/93, or 97%) of the noncognate interactions lasted less than 8 min ([Figure 7](#pbio-0030150-g007){ref-type="fig"}). Of the 81 stable antigen-specific conjugates that were tracked, at least 15% (12/81) persisted longer than 40 min. It was difficult to determine entire contact periods for many (75/150, or 50%) of the long-lasting antigen-specific conjugate pairs because the cells entered or left the field of view as a conjugate. Of the 75 conjugates in which cell--cell association and dissociation were recorded, 23% (17/75) had a measured contact time of between 8 and 40 min ([Figure 7](#pbio-0030150-g007){ref-type="fig"}). Some of the conjugates that moved in or out of the field may also fall in this time window, so this measurement is most likely an underestimate. A small number of conjugates remained in the field of view and did not dissociate for the entire 60--90 min imaging period, establishing that B-T interactions can persist for at least 1.5 h ([Figure 7](#pbio-0030150-g007){ref-type="fig"}). ::: {#pbio-0030150-g007 .fig} Figure 7 ::: {.caption} ###### Contact Times of Antigen-Engaged B cell--Helper T Cell Conjugates \(A) The histogram shows contact time distribution for Ig-tg B cells and TCR7 CD4^+^ T cells 30 to 50 h after immunization with antigen with adjuvant. Open bars show conjugates that were tracked for the duration of contact and shaded bars show conjugates that could not be tracked for the entire period of contact because the cells entered the field as a conjugate, left the field as a conjugate, or both. \(B) Contact time distribution for Ig-tg B cells and OT-II CD4^+^ T cells 1 to 2 d after antigen priming. Open and shaded bars as in (A). ::: ![](pbio.0030150.g007) ::: Discussion {#s3} ========== Using time-lapse two-photon imaging, we have shown that after antigen engagement, B cells reduce their migration speed and display a clear bias in migration toward the B/T boundary in lymph nodes. By contrast, co-transferred CCR7-deficient Ig-tg B cells failed to show preferential migration toward the boundary. Moreover, the CCR7 ligand CCL21 was present as a gradient within the follicle, increasing toward the boundary over a region where B cell migration bias was observed by two-photon imaging. We propose that after antigen capture and CCR7 upregulation, B cells use this long-range chemokine gradient to navigate to the B/T boundary. Within 1 d of antigen encounter, the B cells had increased their motility, and they maintained this behavior while making stable conjugates with helper T cells. Many cognate interactions between antigen-engaged B cells and helper T cells lasted 10 to 40 min, and some interactions persisted for more than 1 h, whereas cells forming noncognate interactions dissociated in less than 10 min. Taken together, these experiments provide evidence of B cell chemotaxis in vivo; they also reveal that interactions between antigen-engaged B cells and cognate T cells are dynamic and may involve multiple, serial contacts. Chemotaxis has been implicated in many biological processes within multicellular organisms, including positioning of cells during development and homing of leukocytes during immune surveillance and in the immune response. However, despite extensive in vitro evidence in support of directed cell migration along chemoattractant gradients \[[@pbio-0030150-b26]\], only limited in vivo evidence of chemotaxis has so far been reported \[[@pbio-0030150-b27],[@pbio-0030150-b28]\]. Our findings indicate that in lymph nodes, antigen-stimulated B cells migrate up a chemokine gradient toward the B/T boundary, providing an in vivo example of chemotaxis in a multicellular organism. Our results suggest the following sequence of events during the spatiotemporal reorganization of antigen-engaged B cells within the follicle. During the first few hours after encountering antigen, B cells upregulate CCR7 \[[@pbio-0030150-b06]\]. Initially, the random motility of B cells promotes intermingling of individual cells within the follicle and allows them to get close to the follicle edge. Once B cells are within 80 to approximately 140 μm of the edge, antigen-engaged B cells begin to respond via CCR7 to the CCL21 gradient and move in a directed manner toward the B/T boundary. Deviations from a linear migration path occur due to collisions with randomly migrating naive B cells and small-scale irregularities in the CCR7 ligand gradient. At the boundary region, the cells continue to migrate extensively. When activated CD4 T cells are present, the B cells and T cells co-cluster in interfollicular regions, areas that also contain many DCs. Antigen-engaged B cells form conjugate pairs with T cells and continue to migrate actively in the interfollicular region, pulling the T cell partner behind. Activated B cells remain within these regions for 1--2 d, most likely undergoing many consecutive 10--40 min interactions with helper T cells before B cell proliferation begins. Despite the large displacement of many of the antigen-engaged B cells, all of the cells showed reduced migration speeds in the first hours after antigen-injection, a behavior that may be due to enhanced adhesion of B cells to follicular stromal cells or other neighboring cells. B cell--receptor (BCR) stimulation increases the avidity of integrins α~L~β~2~ and α~4~β~1~, promoting increased attachment of B cells on intercellular adhesion molecule-1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1) \[[@pbio-0030150-b29],[@pbio-0030150-b30]\], integrin ligands that are expressed by follicular dendritic cells and neighboring B cells \[[@pbio-0030150-b31],[@pbio-0030150-b32]\]. Early after antigen challenge, some HEL-engaged B cells in the part of the follicle distal to the T-zone were not significantly displaced during 30-min imaging experiments. In this regard, it is notable that antigen-receptor stimulation in T cells induces transient α~L~β~2~ activation that persists for about 30 min \[[@pbio-0030150-b33]\]. In vitro studies have indicated that as integrin-mediated adhesiveness increases above a critical threshold, cell motility decreases \[[@pbio-0030150-b34]\]. The BCR signaling may also affect intrinsic motility of B cells by rearranging their cytoskeleton through the Vav--Rho--GTPase pathway \[[@pbio-0030150-b35]\]. The finding that many antigen-engaged B cells showed CCR7-dependent directional migration toward the B/T boundary region is consistent with the conclusion that the cells are migrating in response to the gradient of CCL21 that extends from the T zone into the follicle. The factors determining the distribution of CCL21 are not clear, but it seems likely that the chemokine is distributed along stromal cells that are present throughout the follicle \[[@pbio-0030150-b36]\]. Previous studies have shown that T-zone CCL21 is concentrated on the stromal cell surface \[[@pbio-0030150-b23],[@pbio-0030150-b37]\]. We also consider it likely that the naive B cells migrating through the part of the follicle that is proximal to the T zone bind and transiently display CCL21 in a manner that permits engagement of receptors on other B cells. In addition to displaying directional migration in a CCR7-dependent manner, some antigen-engaged B cells appear to arrive at the B/T boundary by random migration from the follicle and, possibly, other locations because some CCR7-deficient cells were found at the boundary. However, CCR7-deficient cells fail to accumulate at the boundary over time \[[@pbio-0030150-b06]\], and it is likely that CCR7 functions both to promote directional migration of antigen-engaged cells and to help retain cells in the boundary region. A dense network of stromal cells is present on the T-zone side of the boundary, an area known as the outer cortex \[[@pbio-0030150-b38]\] or cortical ridge \[[@pbio-0030150-b37]\], and it will be important to determine whether these cells contribute to retaining antigen-engaged B cells in this region. The finding that antigen-engaged B cells at the B/T boundary are highly motile indicates that the mechanisms retaining cells in this region are unlikely to be solely adhesion based. At the start of our imaging experiments, the transferred B cells had equilibrated in the host for at least 1 d and the majority of Ig-tg B cells were located within lymphoid follicles. Because soluble HEL rapidly gains access to lymphoid follicles \[[@pbio-0030150-b39]\], we assume that most of the B cells in our studies are encountering the antigen while migrating within the follicle. Although we believe our findings should be generalizable for B cells encountering any type of follicular antigen, including antigens displayed by follicular dendritic cells, in some cases B cells may encounter antigens in the blood or at the point of entry into the lymph node, and in this situation, movement to the B/T boundary most likely occurs by a different path from the one examined in this study. The B/T boundary region imaged in our experiments corresponds to the boundary of the follicle and interfollicular region. Although our frozen-section analysis suggested that antigen-engaged B cells accumulated similarly along the length of the B/T boundary in lymph nodes and spleen, it remains possible that cell behavior will vary in different subregions of the boundary. Similar to B cells encountering a foreign antigen, autoantigen-binding B cells localize at the B/T boundary \[[@pbio-0030150-b05]\]. Because their chronic exposure to autoantigen excludes these cells from accessing follicles during their short life span, the term "follicular exclusion" has been used to describe this process for autoreactive B cells. In some cases, autoantigen-binding B cells appear to lodge at the B/T boundary because of reduced expression of CXCR5 rather than increased expression of CCR7 \[[@pbio-0030150-b40]\]. It will be interesting in future studies to examine the migratory behavior of autoantigen-binding B cells at the B/T boundary. Unexpectedly, conjugates of antigen-specific B and T cells were found to migrate extensively, being led by B cells. We propose that T cells in the conjugates are not themselves motile but are dragged by B cells through firm adhesion based on the following observations: (1) in B-T conjugate pairs, B cells turn first and T cells follow; (2) trailing T cells in a conjugate are often rounded and yet the conjugate pair is moving; and (3) conjugate pairs move with the velocity of activated B cells. Conjugate velocities (approximately 9 μm/min) were substantially greater than velocities of DC--T cell conjugates (\<4 μm/min), although here also the T cells followed along after their antigen-presenting partner \[[@pbio-0030150-b14],[@pbio-0030150-b18]\]. In vitro studies have established that T cells become polarized toward their interaction partner, and it seems likely that this reorganization of the cytoskeleton is not compatible with also maintaining motility \[[@pbio-0030150-b41],[@pbio-0030150-b42]\]. Our observations are also consistent with the "stop signal" concept that was developed from the finding that T cells stop crawling on artificial lipid bilayers containing MHC class II and ICAM-1 molecules when specific TCR ligands are given \[[@pbio-0030150-b43]\]. In vitro studies with T cells and a recent study that imaged cell movement in thymic slices by two-photon microscopy showed that motility arrest was caused by TCR-induced elevations in cytosolic Ca^2+^ concentration \[[@pbio-0030150-b44]--[@pbio-0030150-b46]\]. Contrary to our observations, a recent study using confocal laser scanning microscopy to image interactions between naive DO11.10 T cells and peptide-loaded B cells concluded that B cells were being pushed by actively migrating T cells \[[@pbio-0030150-b47]\]. However, most of those data were obtained by imaging cells in a collagen matrix, and although this system provides information on the spectrum of B-T interactions that are possible, these conditions differ substantially from the environment within a lymph node, in which much of the collagen is sheathed by stromal cells (reviewed in \[[@pbio-0030150-b36]\]). Moreover, no BCR stimulus was included in the experiments performed by Gunzer et al. \[[@pbio-0030150-b47]\], and this could also explain the differences in our observations. By migrating while conjugated with T cells, B cells may continue to survey for helper T cells, a process that could promote T cell exchange. During a response involving a diverse repertoire of T cells, continued surveillance might allow B cells to exchange weakly interacting, low-affinity T cells for cells with higher-affinity TCRs that are likely to provide more robust help \[[@pbio-0030150-b49]\]. Our results show that many cognate B-T interactions last for more than 8 min, with some interactions lasting more than 1 h, whereas non-antigen-specific interactions between B cells and helper T cells typically last less than 8 min. Interaction for more than 8 min is likely to provide sufficient time for the formation of immature immunological synapses between the interacting B-T interfaces \[[@pbio-0030150-b50],[@pbio-0030150-b51]\]. Although we were unable to obtain precise measurements of the persistence times of many of the stable conjugates because of their unexpectedly high motility, we can conclude that at least 21% of the conjugates that last longer than 8 min dissociate within 40 min. Therefore, the development of a stable antigen-specific conjugate does not automatically mean that the conjugate will persist for a period of hours. Reciprocally, at least some of the conjugates persisted for at least 40 min, and some for longer than 80 min. Given that the antigen-receptor specificities on all the B cells and T cells in our system are identical, the basis for the different contact duration is not clear at present. It is also notable that many of the noncognate B-T contacts lasted for several minutes, similar to the duration of noncognate interactions imaged in vitro and an amount of time that may be sufficient for initial clustering and scanning of MHC class II peptide complexes \[[@pbio-0030150-b45],[@pbio-0030150-b51]\]. A large fraction of cognate B-T contacts were as short as noncognate contacts, suggesting that many initial antigen-specific contacts fail to achieve adequate signaling and changes in adhesiveness to proceed to stable interactions. At present little is understood about the cumulative amount of T cell stimulation that is required for ensuring B cell commitment to proliferation and differentiation. Our studies reveal that interactions between B and T cells begin taking place within 1 d of immunization with antigen in adjuvant, whereas T cell--dependent B cell proliferation does not begin for another 12--24 h. Using a slightly different system, Garside et al. reported that cognate B-T contacts were not observed in lymph node sections until 2 d after antigen challenge, but again there was about a 1-d lag before B cell proliferation occurred \[[@pbio-0030150-b07]\]. Experiments with anti-CD40 antibodies have indicated that prolonged antibody exposure over a period of days is necessary to induce B cell proliferation, although these studies typically did not employ antigen-stimulated B cells \[[@pbio-0030150-b52]\]. In our studies so far, we have not observed differences in the duration of B-T contacts at day 1 versus day 2 of antigen priming (data not shown). Therefore, at present, we favor the view that most B cells interact consecutively with multiple T cell partners over a period of approximately a day before B cell proliferation is induced. Instead of being distributed along B/T boundaries, antigen-specific B cells were localized in interfollicular zones when TCR7 antigen--specific helper T cells were present. This clustering was not seen when recipients of Ig-tg B cells alone were immunized with antigen in adjuvant, indicating that clustering was not directly induced by the adjuvant and that the low, endogenous HEL-specific T cell response of the B6 mice was insufficient to cause the relocalization. The relocalization may be caused by the activated TCR7 T cells making chemokines or inducing chemoattractant production by other cells, such as antigen-bearing DCs. Ingulli et al. showed that 18 h after subcutaneous injection of fluorochrome-labeled OVA, a subpopulation of antigen-bearing CD11b^+^ DCs was present in paracortical regions adjacent to follicles \[[@pbio-0030150-b53]\]. Interfollicular regions are also rich in high endothelial venules, and other studies have indicated that DCs newly arriving in lymph nodes are often found near follicles or clustered around high endothelial venules \[[@pbio-0030150-b14],[@pbio-0030150-b17],[@pbio-0030150-b54]\], and small, soluble antigens can reach these regions via conduits that connect with the subcapsular sinus \[[@pbio-0030150-b25],[@pbio-0030150-b39]\]. The congregation of cells in interfollicular regions may also foster B cell--DC interactions, which themselves may be important during T cell--dependent antibody responses \[[@pbio-0030150-b02]\]. In regions that were not crowded with helper T cells, B cells mainly formed one-to-one contacts with helper T cells. However, antigen-engaged B cells occasionally formed stable contacts simultaneously with multiple helper T cells, especially in areas crowded with helper T cells. These observations suggest that in physiological conditions, antigen-specific B-T interactions are initially monogamous, but polygamous interactions may occur after antigen-specific T cells expand in number. The ability of B cells to form stable interactions with more than one T cell has also been observed in vitro and is consistent with evidence that B cells do not undergo cytoskeleton-dependent polarization toward the T cell \[[@pbio-0030150-b50],[@pbio-0030150-b51]\]. This may also be important in allowing the continued motility of the B cell in the B-T conjugate pairs. Future studies are needed to learn whether simultaneous interactions with multiple T cells induce different signaling outcomes in B cells compared to one-to-one interactions. In contrast, we did not find helper T cells forming stable conjugates with multiple B cells simultaneously. Instead, when contact was made with multiple B cells, helper T cells followed only one of the B cells. These findings are in good accord with in vitro studies showing that individual T cells use their cytoskeleton to actively polarize toward a single B cell partner \[[@pbio-0030150-b41],[@pbio-0030150-b45],[@pbio-0030150-b50],[@pbio-0030150-b51]\]. These observations suggest that each T cell forms one immunological synapse at a time in cognate B-T interactions in lymph nodes. Materials and Methods {#s4} ===================== {#s4a} ### Mice and cells {#s4a1} The HEL-specific Ig-tg mice and cOVA-specific TCR-transgenic mice were of the MD4 line and of the OT-II line, respectively \[[@pbio-0030150-b55],[@pbio-0030150-b56]\]. B6.Cg-Igh^a^ Thy1^a^ Gpi1^a^/J(B6-Igh^a^) and μMT mice were from the Jackson Laboratory (Bar Harbor, Maine, United States). The HEL-specific TCR-transgenic mice were of the TCR7 line that carries transgenic TCR-αβ-recognizing peptide HEL~74--88~ in I-A^b^ (M. Neighbors, S. B. Hartley, and A. O\'Garra, unpublished data). All transgenic lines were crossed with C57BL6 (B6)-CD45.1 mice (Jackson Laboratory). B cells were purified from spleen and lymph node cells of Ig-tg or B6 mice by immunomagnetic depletion of CD43-expressing cells using autoMACS (Miltenyi Biotec, Bergisch Gladbach, Germany). CD4^+^ T cells were enriched from spleen and lymph node cells of B6, TCR7, or OT-II mice by depleting CD8^+^ T cells, B cells, NK cells, DCs, macrophages, granulocytes, and erythrocytes (Miltenyi Biotec). Purity of B and T cells from B6 mice and transgenic BCR-positive B cells from Ig-tg mice was more than 90%. Purity of Vβ3^+^CD4^+^ T cells from TCR7 mice and Vα2^+^Vβ5^+^CD4^+^ T cells from OT-II mice was 70% to 80% because of the presence of CD4^−^CD8^−^ T cells in the periphery. After adoptive transfer, however, transgenic TCR-positive CD4^+^ T cells were more than 90% of transferred cells in recipient lymph nodes. CCR7^−/−^ mice \[[@pbio-0030150-b57]\] were tenth-generation backcrossed to B6 mice. CCR7^−/−^ Ig-tg B6 mice were frequently found to become stunted and sickly by 4--6 wk of age and could not be used as B cell donors. To generate sufficient B cells for transfer experiments, irradiated B6 or B cell--deficient (μMT) mice were reconstituted for 6--10 wk with 80% CCR7^−/−^ Ig-tg bone marrow and, to provide a source of wild-type T cells, 20% μMT bone marrow. Bone marrow chimeras were made as described by Reif et al. \[[@pbio-0030150-b06]\]. ### Flow cytometric analysis {#s4a2} Analyses of purified cells and transferred cells were performed on a FACSCalibur (Becton Dickinson, Palo Alto, California, United States). All antibodies were purchased from BD Pharmingen (San Diego, California, United States). For proliferation analysis of antigen-specific cells in lymph nodes, 6--10 million purified Ig-tg B cells and 5--8 million purified TCR7 CD4^+^ T cells were labeled with 7 μM CFSE (Molecular Probes, Eugene, Oregon, United States) for 25 min at 37 °C and transferred intravenously into B6 mice that had been immunized subcutaneously in the flank and the base of the tail with 200 μl of 1.3% alum emulsion containing 400 μg of HEL and 4 μg of recombinant murine TNFα (R&D Systems, Minneapolis, Minnesota, United States) 6--8 h before cell transfer. At 20--24 h after cell transfer, the animals were given a second intraperitoneal injection of 1 mg of HEL in saline. The superficial inguinal lymph nodes were removed after sacrificing the recipient mice and chopped in medium containing 1.6 mg/ml bovine collagenase type IV (Worthington Biochemical, Lakewood, New Jersey, United States) and 50 μg/ml bovine DNase I (Sigma, St. Louis, Missouri, United States). After 30 min incubation at 37 °C, the cell suspension was mashed through 70-μm filters and immunofluorescently stained to identify CD45.1^+^ populations of transferred cells. ### Immunohistochemistry {#s4a3} Cryostat sections (7--8 μm) of lymph nodes were fixed and stained as previously described \[[@pbio-0030150-b58]\]. CCL21 staining was performed with goat anti-mouse CCL21 antibody (R&D Systems) and biotinylated donkey anti-goat IgG antibody (Jackson ImmunoResearch, West Grove, Pennsylvania, United States). Biotin was detected with an ABC-AP kit (Vector Laboratories, Burlingame, California, United States). CCR7^−/−^ B cells transferred into B6-Igh^a^ mice were detected by biotinylated anti-IgM^b^ and anti-IgD^b^ antibodies (BD Pharmingen). PNA was from Sigma; anti-IgD was from The Binding Site (Birmingham, United Kingdom); and syndecan-1 was from BD Pharmingen. Ig-tg B cells were detected by HEL binding as previously described \[[@pbio-0030150-b06]\] or by staining for CD45.1 or IgM^a^. CFSE-labeled CD4^+^ T cells were stained with anti-fluorescein HRP antibody (PerkinElmer, Boston, Massachusetts, United States). All other antibodies were from BD Pharmingen. ### Two-photon imaging and analysis {#s4a4} For imaging antigen-engaged B cells relocalizing in the B/T boundary, 20--30 million wild-type Ig-tg B cells and 5--10 million B6 B cells were labeled with 10 μM CFSE and 10 μM CMTMR (Molecular Probes), respectively, and transferred to B6 mice. In two experiments, Ig-tg B cells were labeled with CMTMR, and this gave similar results (not shown). For experiments with CCR7^−/−^ Ig-tg B cells, 30--70 million cells were purified from approximately ten bone marrow chimeric donors, labeled with CFSE, and injected into single recipients that also received wild-type CMTMR-labeled Ig-tg B cells as above. Flow cytometric analysis of recipient lymph node cells established that more than 90% of the CFSE^+^ cells were HEL-binding CCR7^−/−^ Ig-tg B cells. One to 2 d after cell transfer, recipient mice were intravenously injected with 1.5 mg of HEL in saline. One, 5, or 18 h after HEL injection, superficial inguinal lymph nodes were isolated, maintained in 36 °C medium bubbled with 95% O~2~/5% CO~2~, and imaged through the capsule in a region distal to the efferent lymphatic by multi-dimensional (x, y, z, time, and emission wavelength) two-photon microscopy \[[@pbio-0030150-b11]\]. For time-lapse image acquisition, some experiments were performed exactly as described by Miller et al. \[[@pbio-0030150-b11]\]. In other experiments, each xy-plane spanned 240 μm by 288 μm at a resolution of 0.6 μm per pixel, and images of 18 xy-planes with 3-μm z spacing were formed by averaging ten video frames, using emission wavelengths of 500--540 nm (for CFSE-labeled cells), 567--640 nm (for CMTMR-labeled cells), and 360--440 nm (to detect second harmonic emission) every 30 s. For orientation purposes, z stacks of up to 200 μm were collected. For imaging B-T interactions, 6--10 million Ig-tg B cells and 5--10 million TCR7 or OT-II CD4^+^ T cells were labeled with 10 μM CSFE and 10 μM CMTMR, respectively, and transferred to B6 mice that had been immunized 8 h before transfer. In two experiments, the dyes were switched, and this gave similar results (not shown). Immunization was as described above for cell proliferation analysis. One day or in some cases 2 d after cell transfer, superficial inguinal lymph nodes were imaged as described above. Image acquisition was performed by Metamorph (Universal Imaging, Marlow, United Kingdom) or Video Savant software (IO Industries, London, Ontario, Canada). Three-dimensional cell tracking was performed using Metamorph software with manual tracking of individual cells between each 3-μm z-plane. In any instance where two cells in a given set of adjacent z-planes were indistinguishably overlapping, we excluded them from the analysis. Supporting Information {#s5} ====================== Figure S1 ::: {.caption} ###### Projection Views of Image Stacks Collected before and after Time-Lapse Image Analysis The xz, yz, and xy projection views of image stacks collected immediately prior to time-lapse imaging (A), or immediately after imaging (B), demonstrating the location of the B cell follicle containing transferred Ig-tg (green) and non-tg (red) B cells. The collagen-rich lymph node capsule is visualized by second harmonic emission (blue). The lymph node was isolated for imaging 1 h after HEL injection and imaging was performed for 3 h. The relocation of the antigen-engaged (green) B cells to the rim of the follicle can be seen most clearly in the xz projection view in (B). The time-lapse movie collected from this follicle corresponds to the third dataset included in [Figures 1](#pbio-0030150-g001){ref-type="fig"} and [2](#pbio-0030150-g002){ref-type="fig"}. Objective magnification, 20×. (6.4 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S2 ::: {.caption} ###### Enlarged View of an Interfollicular Cluster of Ig-tg B Cells and TCR7 T Cells in the Draining Lymph Node 1 d after Immunization Ig-tg B cells are shown in blue and TCR7 T cells are shown in brown. The area shown corresponds to the boxed region in [Figure 5](#pbio-0030150-g005){ref-type="fig"}C. Arrows indicate examples of interactions between antigen-specific B cells and T cells. (4.6 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S1 ::: {.caption} ###### Relocation of Antigen-Engaged B Cells from the Follicle to the B/T Boundary Time-lapse image sequence shows Ig-tg B cells (green) and non-tg B cells (red) in a lymph node approximately 1--3 h after HEL-antigen challenge. The white circle highlights an Ig-tg B cell that moves to the B/T boundary. Time indicated as h:min:s. Each image is 210 × 180 μm and projects 51-μm z stacks. Time compression is 300×. (3.1 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S2 ::: {.caption} ###### Second Example of Antigen-Engaged B Cell Relocalization from the Follicle to the B/T Boundary Time-lapse image sequence shows Ig-tg B cells (green) and non-tg B cells (red) in a lymph node approximately 1--4 h after HEL-antigen challenge. Time indicated as h:min:s. Each image is 288 × 240 μm and projects 51-μm z stacks. Time compression is 300×. (8 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S3 ::: {.caption} ###### Motility of Antigen-Engaged B Cells Localized on the B/T Boundary Time-lapse image sequence shows Ig-tg B cells (green) and non-tg B cells (red) 18--19 h after antigen challenge. Time indicated as h:min:s. Each image is 210 × 180 μm and projects 51-μm z stacks. Time compression is 300×. (1.6 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S4 ::: {.caption} ###### Migration of Antigen-Engaged Wild-Type but Not CCR7^−/−^ B Cells to the B/T Boundary Time-lapse image sequence is shown as a montage to facilitate improved tracking of the two B cell types, with CCR7^−/−^ Ig-tg B cells shown in the upper panel, wild-type Ig-tg B cells in the center panel, and the overlay of both cell types in the lower panel (CCR7^−/−^, green; wild-type, red). Image sequence corresponds to approximately 3.5--4.2 h after HEL-antigen challenge. Time indicated as h:min:s. Each image is 180 × 150 μm and projects 51-μm z stacks. Time compression is 300×. (2.5 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S5 ::: {.caption} ###### Interactions of HEL-Specific B Cells and HEL-Specific Helper T Cells within the Lymph Node Time-lapse image sequence shows Ig-tg B cells (red) and TCR7 transgenic T cells (green) \~30 h after challenge with antigen in adjuvant. Time indicated as h:min:s. Each image is 200 × 163 μm and projects 51-μm z stacks. Time compression is 270×. (3.9 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S6 ::: {.caption} ###### Migration Dynamics of B-T Conjugates within the Lymph Node Time-lapse image sequence shows Ig-tg B cells (red) and TCR7 transgenic T cells (green) \~30 h after challenge with antigen in adjuvant. Time indicated as h:min:s. Each image is 135 × 114 μm and projects 51-μm z stacks . Time compression is 300×. (1.9 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S7 ::: {.caption} ###### Spatiotemporal Changes Induced in Ig-tg B Cell and TCR7 Transgenic T cell Distribution, Motility, and Interactions after Antigen Challenge The video contains four time-lapse image segments representing initial random motility and segregation in the absence of cognate antigen (segment I), and three times after challenge with HEL in adjuvant: \~40 h (segment II), \~52 h (segment III), and \~64 h (segment IV). B cells shown in red; T cells shown in green. The yellow box in segment IV highlights a representative T cell blast that migrated into the imaging volume, paused, rounded up, and divided. As the daughter cells regain motility and move apart, long membrane tethers can be observed trailing behind. Scale bar = 50 μm; time compression is 450×. (4.5 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S8 ::: {.caption} ###### Time-Lapse Image Sequence Showing Migration of One HEL-Specific B Cell Simultaneously Interacting with Multiple HEL-Specific Helper T Cells B cells shown in red; T cells shown in green. Time indicated as h:min:s. Each image is 60 × 50 μm and projects 30-μm z stacks. Time compression is 300×. (2 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S9 ::: {.caption} ###### Time-Lapse Image Sequence Showing Brief, Simultaneous Interactions of One HEL-Specific Helper T Cell with Two HEL-Specific B Cells B cells shown in red; T cell shown in green. In this particular movie, it appears that the T cell switches partner before and after the multipartite interaction. Time indicated as h:min:s. Each image is 40 × 45 μm and projects 51-μm z stacks. Time compression is 270×. (547 KB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S10 ::: {.caption} ###### Time-Lapse Image Sequence Showing a HEL-Specific B Cell Switching HEL-Specific Helper T Cell Partners B cells shown in red; T cells shown in green. Time indicated as h:min:s. Each image is 40 × 35 μm and projects 51-μm z stacks. Time compression is 270×. (979 KB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S11 ::: {.caption} ###### Encounters between HEL-Specific B and T Cells after Antigen Challenge Time-lapse image sequence sequentially showing three examples of HEL-specific B cells (red) encountering HEL-specific helper T cells (green) to form conjugates. Arrows are included to highlight the initial tethering points between cells. Time indicated as h:min:s. Each image is 50 × 50 μm and projects 24-μm z stacks. Time compression is 270×. (2.4 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S12 ::: {.caption} ###### Migration Dynamics of HEL-Specific B Cells and OVA-Specific T Cells within the Lymph Node Time-lapse image sequence showing the brief interactions of HEL-specific B cells (red) and OVA-specific helper T cells (green) \~30 h after combined immunization with HEL and OVA in adjuvant. Time indicated as h:min:s. Each image is 200 × 163 μm and projects 51-μm z stacks. Time compression is 270×. (1.9 MB AVI). ::: ::: {.caption} ###### Click here for additional data file. ::: Accession Numbers {#s5a15} ----------------- The LocusLink (<http://www.ncbi.nlm.nih.gov/LocusLink/>) accession numbers for the gene products discussed in this article are CCL19 (LocusLink 24047), CCL21 (LocusLink 65956), and CCR7 (LocusLink 12775). We thank Lu Forrest, Caroline Low, and Olga Safrina for expert assistance in cell preparation and animal handling, and Hsiang Ho for help with cell tracking. TO was supported by the Japan Society for the Promotion of Science, and JGC is a Howard Hughes Medical Institute investigator. This work was supported by National Institutes of Health grants GM41514 (MDC) and AI45073 (JGC), and by a Sandler New Technology Award (MFK and JGC). Competing interests. The authors have declared that no competing interests exist. Author contributions. TO, MJM, MDC, and JGC conceived and designed the experiments. TO and MJM performed the experiments. TO, MJM, IP, MDC, and JGC analyzed the data. TO, MJM, IP, MFK, MN, SBH, AO, MDC, and JGC contributed reagents/materials/analysis tools. TO, MJM, IP, MDC, and JGC wrote the paper. ¤a Current address: Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, Missouri, United States of America ¤b Current address: Maxygen, Redwood City, California, United States of America ¤c Current address: Ainslie, ACT, Australia ¤d Current address: Laboratory of Immunoregulation, The National Institute for Medical Research, London, United Kingdom Citation: Okada T, Miller MJ, Parker I, Krummel MF, Neighbors M, et al. (2005) Antigen-engaged B cells undergo chemotaxis toward the T zone and form motile conjugates with helper T cells. PLoS Biol 3(6): e150. BCR : B cell receptor B-T : B cell--T cell B/T : B-zone--T-zone CFSE : carboxyfluorescein diacetate succinimidyl ester CMTMR : 5-(and-6)-(((4-chloromethyl)benzoyl)amino)tetramethylrhodamine DC : dendritic cell HEL : hen egg lysozyme ICAM-1 : intercellular adhesion molecule-1 Ig-tg : immunoglobulin-transgenic non-tg : nontransgenic OVA : ovalbumin TCR : T cell receptor	PubMed Central	2024-06-05T03:55:56.283916	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088276/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e150", "authors": [ { "first": "Takaharu", "last": "Okada" }, { "first": "Mark J", "last": "Miller" }, { "first": "Ian", "last": "Parker" }, { "first": "Matthew F", "last": "Krummel" }, { "first": "Margaret", "last": "Neighbors" }, { "first": "Suzanne B", "last": "Hartley" }, { "first": "Anne", "last": "O'Garra" }, { "first": "Michael D", "last": "Cahalan" }, { "first": "Jason G", "last": "Cyster" } ] }
PMC1088277	Introduction {#s1} ============ Although it is known that thymocytes relocalize from the cortex to the medulla after positive selection, the means by which this relocalization occurs is largely unknown \[[@pbio-0030160-b01]--[@pbio-0030160-b03]\]. For example, thymocytes may migrate by random walk throughout the cortex, with selected thymocytes being captured at the medulla by short-range guidance cues. Alternatively, thymocytes may migrate directionally across the cortex in response to long-range cues emanating from the medulla. Directional migration in response to such long-range cues may be induced by positive selection or may be a property of all cortical thymocytes. To address these and other questions, we undertook a real-time analysis of thymocyte migration in intact thymic lobes. Results/Discussion {#s2} ================== In order to track migrating thymocytes in situ, we generated chimeric mice in which a fraction of thymocytes express green fluorescent protein (GFP). We devised a protocol, based on a previously described method \[[@pbio-0030160-b04]\], in which adult bone marrow from GFP transgenic mice \[[@pbio-0030160-b05]\] is injected into newborn mice to generate partial GFP hematopoietic chimeras without the use of irradiation (see [Materials and Methods](#s3){ref-type="sec"}). The GFP^+^ thymocytes generated in this way comprised approximately 1% of total thymocytes and showed expected developmental profiles by flow cytometric analysis ([Figure S1](#sg001){ref-type="supplementary-material"}). At 4.5 to 5.5 wk of age, chimeric thymic lobes were harvested and imaged while being perfused with warmed oxygen-supplemented media. Imaging of intact lymph nodes under similar conditions revealed identical T cell and dendritic cell motility and behavior to that observed from intravital imaging of lymph nodes \[[@pbio-0030160-b06]--[@pbio-0030160-b09]\]. The objective was positioned above the center of the thymic lobe such that the movement of the stage (z direction) was perpendicular to the thymic capsule ([Figures S2](#sg002){ref-type="supplementary-material"} and [S3](#sg003){ref-type="supplementary-material"}). Regions of thymic cortex 104 × 104 × 40 μm were scanned at tissues depths between 80--200 μm below the surface of the capsule. The scan sequence was repeated every 37 s for up to 33 min to generate four-dimensional (4D) datasets (x, y, z, and time) ([Videos S1](#sv001){ref-type="supplementary-material"} and [S2](#sv002){ref-type="supplementary-material"}). The datasets were analyzed with a combination of 4D cell-tracking software ([Figure 1](#pbio-0030160-g001){ref-type="fig"}; [Video S3](#sv003){ref-type="supplementary-material"}) and mathematical treatment of the cell tracks. ::: {#pbio-0030160-g001 .fig} Figure 1 ::: {.caption} ###### Tracking Thymocyte Migration in 3D Tracking software identifies the positions of individual thymocytes over time. Trajectories of individual cells are shown as tracks, which are color coded to indicate increasing time from blue (start of imaging) to yellow (end of imaging) (see [Videos S1](#sv001){ref-type="supplementary-material"}--[S4](#sv004){ref-type="supplementary-material"}). Left panels show fluorescent signal from thymocytes at a single time point superimposed on cell tracks. Right panels show the positions of thymocytes (indicated by spheres) at a single time point. Top panels show a projection in which the z-axis is perpendicular to the viewer. In the bottom panels the image is rotated to display the z dimension. ::: ![](pbio.0030160.g001) ::: Analysis of the motility rates of individual GFP cortical thymocytes ([Figure 2](#pbio-0030160-g002){ref-type="fig"}A) showed that the vast majority of cells (88% of total cells) moved at low motility rates between 3--8 μm/min (MR^lo^) and exhibited peak instantaneous velocities of up to 9 μm/min ([Figure 2](#pbio-0030160-g002){ref-type="fig"}B). A small subset of imaged cells (approximately 7%) moved at motility rates that were 10 μm/min or greater (MR^hi^) and exhibited peak instantaneous velocities of up to 28 μm/min ([Figure 2](#pbio-0030160-g002){ref-type="fig"}A and [2](#pbio-0030160-g002){ref-type="fig"}B; [Video S4](#sv001){ref-type="supplementary-material"}). The distribution of motility rates among cortical thymocytes was highly reproducible between samples ([Figure S4](#sg004){ref-type="supplementary-material"}). The instantaneous velocities for each group showed no interconversion of motility rates throughout imaging ([Figure 2](#pbio-0030160-g002){ref-type="fig"}B), suggesting that these differing motility behaviors represented distinct cell populations. ::: {#pbio-0030160-g002 .fig} Figure 2 ::: {.caption} ###### Two Distinct Migratory Behaviors within Wild-Type Cortical Thymocytes \(A) Histogram showing the frequency distribution of average motility rates (MR) for cortical thymocytes compiled from over 1,250 tracked cells from four independently imaged thymic lobes. The vast majority of cells exhibited speeds ranging from 3 to 8 μm/min (MR^lo^). Approximately 7% exhibited speeds of 10 μm/min or greater (MR^hi^). Cells migrating between 10--13 μm/min represented approximately 5% of cortical thymocytes, and those with speeds of 14 μm/min or greater represented approximately 2% of cortical thymocytes. \(B) Instantaneous velocities versus time for representative MR^hi^ and MR^lo^ cells. Data are representative of 53 MR^hi^ cells and more than 200 MR^lo^ cells analyzed. No conversions between MR^hi^ or MR^lo^ behaviors were observed over a combined imaging time of more than 30 h. \(C) Five successive time frames showing the morphology associated with propulsion for an MR^hi^ and an MR^lo^ cell ([Video S5](#sv005){ref-type="supplementary-material"}). \(D) Graph of displacement versus time for four individual MR^hi^ and MR^lo^ cells. \(E) Graph of directional index (Traj/D) versus average motility rate. The bars indicate the average values for Traj/D computed from 50 MR^hi^ and 466 MR^lo^ cells. \(F) Graph of MR^lo^ cells (left), but not MR^hi^ cells (middle), shows linear relationship between the square of the displacement from origin versus time, indicative of random walk. Right graph shows a linear relationship between displacements from origin (as opposed to their square) with increasing time for MR^hi^ cells, indicative of ballistic motion (right). Analysis was done on 466 MR^lo^ cells and 50 MR^hi^ cells from three independently imaged thymic lobes. ::: ![](pbio.0030160.g002) ::: Further analysis showed striking differences between MR^lo^ and MR^hi^ cells with regard to their morphology and with other aspects of their migratory behavior. For example, MR^hi^ cells displayed a highly polarized morphology with a well-defined leading edge and uropod, and moved by a series of lurches followed by contraction ([Figure 2](#pbio-0030160-g002){ref-type="fig"}C; [Video S5](#sv005){ref-type="supplementary-material"}). In contrast, MR^lo^ cells exhibited a more spherical morphology that lacked obvious polarization. Propulsion by MR^lo^ cells involved only a modest protrusion of the cell\'s leading edge. These properties exclusively segregated with motility rates and remained constant over the entire imaging time. Also, whereas MR^lo^ cells showed frequent pausing during the course of their trajectories, MR^hi^ cells were never observed to pause (see [Video S3](#sv003){ref-type="supplementary-material"}; data not shown). Examination of a cell\'s displacement from origin relative to time can provide additional insight into the migratory behavior of cells. Individual MR^hi^ cells exhibit a linear relationship between displacement and time. In contrast, the displacement from origin for MR^lo^ cells revealed numerous turns back toward cell origin ([Figure 2](#pbio-0030160-g002){ref-type="fig"}D). The turning behavior of thymocytes is of interest in part because it may reflect the interaction of thymocytes with other cells or structures in the tissue environment. For example, thymocytes engaged in dynamic contacts with MHC-bearing stromal cells during positive selection and T cells contacting antigen-bearing dendritic cells in intact lymph nodes turn frequently and show little displacement over time \[[@pbio-0030160-b07]--[@pbio-0030160-b10]\]. We quantitated the degree of turning by computing the total length of a cell\'s trajectory divided by the absolute value of its displacement from origin. If a cell\'s migratory path shows little to no deviation from a straight line, this ratio will be close to one. An analysis of this directional index for each population showed a ratio of 5.9 for MR^lo^ cells as compared to 1.6 for MR^hi^ cells ([Figure 2](#pbio-0030160-g002){ref-type="fig"}E). Taken together with the frequent pausing observed for MR^lo^ cells, but not for MR^hi^ cells, these results suggest that the MR^lo^ thymocytes interact with their environment more extensively than do MR^hi^ thymocytes. A major aim of this study was to determine whether the localization of mature thymocytes to the medulla involves directed inward migration across the cortex, and if so, whether directed migration is a property of all cortical thymocytes or only thymocytes that have been selected to mature. To examine this question, we used graphical techniques borrowed from diffusion mechanics to distinguish movement by random walk versus directed migration \[[@pbio-0030160-b11]\] (see [Materials and Methods](#s3){ref-type="sec"}). Analysis of 1,250 MR^lo^ cells showed that the relationship between displacement from origin with respect to time was consistent with random walk statistics ([Figure 2](#pbio-0030160-g002){ref-type="fig"}F). In contrast, a graph of the mean displacement from origin (as opposed to the square of the displacement) versus time for MR^hi^ cells showed a linear relationship, indicative of directed migration. Thus, MR^lo^ and MR^hi^ thymocytes use distinct modes of migration as they move through the cortex. The observation that MR^hi^ cells moved by directed migration is consistent with the possibility that these cells are being directed to migrate toward the medulla. If this were the case, we would expect their trajectories to show a common orientation in the --z direction (away from the capsule). To examine this question, we used two methods of statistical analysis (see [Materials and Methods](#s3){ref-type="sec"}). First, vector analysis was performed in which the average displacement per cell in each direction of three-dimensional (3D) space for a fixed time interval of 3 min was calculated ([Figure 3](#pbio-0030160-g003){ref-type="fig"}A). If MR^hi^ cell tracks were randomly oriented in the cortex, then the average displacement in each of the six directions (+x, --x, +y, --y, +z, and --z) would be similar. On the other hand, if there were common directionality for MR^hi^ cell trajectories, we should see an increase in average displacement values for the preferred direction. As shown in [Figure 3](#pbio-0030160-g003){ref-type="fig"}A (and [Figure S5](#sg005){ref-type="supplementary-material"}), although the average displacements in the x and y directions were similar, there was greater displacement in both +z and -z directions with the greatest bias in the +z direction (toward the capsule). [Figure 3](#pbio-0030160-g003){ref-type="fig"}B provides a visual representation of track orientation shown along the x and z directions for both MR^lo^ (top panel) and MR^hi^ cells (lower panel). ::: {#pbio-0030160-g003 .fig} Figure 3 ::: {.caption} ###### MR^hi^ Thymocytes Show Preferential Movement Perpendicular to the Thymic Capsule \(A) Bar graph showing the average displacement in each direction by wild-type MR^hi^ cells in a 3-min interval. Data shown were computed from 53 MR^hi^ cells from four independently imaged thymic lobes. Data from individual runs are shown in [Figure S3](#sg003){ref-type="supplementary-material"}. \(B) The upper image is rotated to display the x and z dimensions showing tracks of MR^hi^ cells. Five of six MR^hi^ tracks show preferential orientation in the z direction. The lower image shows tracks of MR^lo^ cells from same dataset. \(C) The results of step analysis (see [Materials and Methods](#s3){ref-type="sec"}) on 172 MR^hi^ cells. Thymocytes are grouped according to their average motility rate (displayed on x-axis) and percentage of cells moving in either the positive or negative direction is displayed on the y-axis. Data are compiled from four independently imaged thymic lobes. ::: ![](pbio.0030160.g003) ::: To confirm and extend these results, we performed a step analysis on MR^hi^ cells (see [Materials and Methods](#s3){ref-type="sec"}) in which individual thymocytes were scored as showing net displacement in the positive or negative direction along each of three axes (x, y, and z). This analysis allowed us to correlate the tendency of thymocytes to migrate in a particular direction with their motility rates ([Figure 3](#pbio-0030160-g003){ref-type="fig"}C). As expected, thymocytes showed equal tendency to displace in the positive and negative directions along the x and y axes. In striking contrast, cells with motility rates greater than 13 μm/min were consistently scored as moving in the --z direction (away from the capsule, toward the medulla). Interestingly, the vast majority of cells with motility rates in the range of 10--12 μm/min were scored as moving in the +z direction (toward the capsule). We suspect that the population with intermediate motility rates is comprised of CD4^--^CD8^--^ double-negative (DN) thymocytes based on published evidence for outward migration of DN thymocytes in the adult thymus \[[@pbio-0030160-b12],[@pbio-0030160-b13]\]. CD4^+^CD8^+^ double-positive (DP) thymocytes express clonally variable versions of the T cell antigen receptor (TCR). Following somatic V(D)J rearrangement and cell surface expression of the αβTCR, cortical thymocytes test out their antigen receptors for their ability to bind self-peptide and MHC proteins expressed in the thymus. A small fraction of thymocytes expressing TCR with moderate avidity for self-peptide MHC receive signals that allow them to differentiate into more mature medullary CD4^+^CD8^--^ or CD4^--^CD8^+^ thymocytes, a process known as positive selection \[[@pbio-0030160-b02],[@pbio-0030160-b03],[@pbio-0030160-b14]\]. The low frequency of cortical thymocytes with motility rates greater than 13 μm/min, together with their biased movement away from the capsule, led us to hypothesize that these cells might represent thymocytes that had successfully undergone positive selection. To test this hypothesis, we generated chimeric mice in which a small fraction (approximately 1%) of thymocytes expressed both GFP and rearranged TCR transgenes that do or do not allow positive selection. As a positive-selecting TCR, we used the class I MHC-restricted P14 TCR transgene, which promotes the development of mature CD8 T cells in the H2^b^ (B6) background \[[@pbio-0030160-b15]\]. As a nonselecting TCR, we used the 5CC7 TCR transgene, which leads to neither positive nor negative selection in the B6 background \[[@pbio-0030160-b16]\]. Analysis of the motility rate distribution of cortical thymocytes expressing transgenic TCRs showed a striking correlation between positive selection and the frequency of thymocytes with high motility rates ([Figure 4](#pbio-0030160-g004){ref-type="fig"}A and [Figure S6](#sg006){ref-type="supplementary-material"}). A total of 34% of cortical thymocytes expressing the P14 TCR had motility rates greater than 13 μm/min compared to approximately 2% for wild-type thymocytes expressing diverse TCRs. In addition, thymocytes expressing the nonselecting 5CC7 TCR showed a nearly complete absence of cortical thymocytes moving at speeds greater than 13 μm/min. In fact, only three of the 1,275 5CC7 thymocytes that were analyzed exhibited motility rates greater than 13 μm/min. These values are significantly different from wild-type cortical thymocytes in which 50 of 1,670 thymocytes exhibited motility rates greater than 13 μm/min (p = 0.002). Moreover, although the majority of rapidly migrating wild-type thymocytes were migrating away from the capsule (44 of 50), all three of the rapidly migrating 5CC7 thymocytes were moving toward the capsule ([Figure 4](#pbio-0030160-g004){ref-type="fig"}). ::: {#pbio-0030160-g004 .fig} Figure 4 ::: {.caption} ###### Positive Selection Leads to an Increased Frequency of MR^hi^ Thymocytes Migrating away from the Thymic Capsule \(A) A histogram showing the frequency distribution of average motility rates for positively selecting (blue, P14) and nonselecting (black, 5CC7) transgenic thymocytes compiled from over 1,200 P14 and 875 5CC7 thymocytes from, respectively, four and three independently imaged thymic lobes. Data (from [Figure 2](#pbio-0030160-g002){ref-type="fig"}A) from wild-type (WT) thymocytes (red) were overlaid for comparison. P14 cells moving at motility rates greater than 13 μm/min were 34% of total imaged thymocytes ([Videos S6](#sv006){ref-type="supplementary-material"} and [S7](#sv007){ref-type="supplementary-material"}) compared to approximately 1% of wild-type cortical thymocytes. Analysis of 5CC7 thymocytes showed nearly complete absence of cells moving at motility rates greater than 13 μm/min, a value that differed significantly (p = 0.002) from wild-type cortical thymocytes. ([Video S8](#sv008){ref-type="supplementary-material"}) \(B) Image showing trajectories of representative P14 MR^hi^ cells. Note tracks for P14 thymocytes are relatively linear compared to the tracks of wild-type thymocytes (see [Figure 1](#pbio-0030160-g001){ref-type="fig"}). \(C) Bar graph showing the average displacement per cell moved in each direction over a 3-min time interval (left). Data was computed from more than 100 P14 MR^hi^ cortical thymocytes compiled from four independent experiments. Data from individual runs are shown in [Figure S5](#sg005){ref-type="supplementary-material"}. \(D) Results of step analysis on 412 P14 thymocytes as a function of motility rate are shown (left). Results of step analysis on 123 5CC7 thymocytes are shown for comparison (right). P14 cells moving at MR greater than 13 μm/min showed strong bias for movement in the --z direction (away from capsule) whereas 5CC7 thymocytes showed random use of both +z and --z directions. ::: ![](pbio.0030160.g004) ::: Our studies show that positive selection leads to a rapid directional migration pattern and are consistent with earlier studies showing that activated CD4^+^CD8^+^ cells migrate rapidly in vitro \[[@pbio-0030160-b17]\]. In an earlier study of thymocyte--stromal cell interactions in reaggregate thymic organ culture, we did not note a major difference in overall motility rates between positively selected and wild-type thymocytes \[[@pbio-0030160-b10]\]. This may be due to the fact that reaggregate thymic organ cultures lack the normal spatial distribution of chemokines and other guidance cues that are likely to control thymocyte migration patterns. As in the case of MR^hi^ cells of wild-type mice, P14 MR^hi^ cells displayed a highly polarized morphology, and their trajectories showed very little turning, with no incidence of pausing ([Figure 4](#pbio-0030160-g004){ref-type="fig"}B and data not shown). In addition, P14 MR^hi^ thymocytes displayed greater displacement in the positive and negative z directions (see [Figure 4](#pbio-0030160-g004){ref-type="fig"}C), and step analysis showed that the vast majority of cells moving at rates greater than 13 μm/min moved away from the capsule ([Figure 4](#pbio-0030160-g004){ref-type="fig"}D). The motility rates and directionality of P14 thymocytes were highly reproducible between samples ([Figures S6](#sg006){ref-type="supplementary-material"} and [S7](#sg007){ref-type="supplementary-material"}). When considering thymocytes with intermediate motility rates (10--12 μm/min), there were two notable differences between P14 and wild-type thymocytes ([Figure 4](#pbio-0030160-g004){ref-type="fig"}A and [4](#pbio-0030160-g004){ref-type="fig"}D). First, the frequency of these intermediate motility thymocytes was higher among P14 compared to wild-type cortical thymocytes ([Figure 4](#pbio-0030160-g004){ref-type="fig"}A). In addition, whereas the majority of wild-type thymocytes of intermediate motility migrated toward the capsule (see [Figure 3](#pbio-0030160-g003){ref-type="fig"}C, right panel), this trend was less clear among P14 thymocytes ([Figure 4](#pbio-0030160-g004){ref-type="fig"}D, left panel). These differences could be explained by the proposition that thymocytes with intermediate motility rates consist of a mixture of outwardly migrating CD4^--^CD8^--^ DN thymocytes and inwardly migrating positively selected CD4^+^CD8^+^ thymocytes, with the DN subset predominating in wild-type samples. In P14 samples, the increase in the number of intermediate motility, inwardly migrating thymocytes could be attributed to an increase in the fraction of thymocytes undergoing positive selection. Importantly, we observed directional migration of MR^hi^ thymocytes in each dataset corresponding to a region of the cortex that extends from, approximately, 80 μm to 200 μm below the thymic capsule. This suggests that thymocyte migration is directed by guidance cues that extend over a large area of the cortex. Although the nature of these guidance cues is currently unknown, there are a number of chemokines expressed in the medulla whose corresponding receptors are upregulated during positive selection \[[@pbio-0030160-b18]--[@pbio-0030160-b23]\]. These include CCL19/CCL21, whose receptor, CCR7, is upregulated on activated CD4^+^CD8^+^ thymocytes \[[@pbio-0030160-b19],[@pbio-0030160-b24]\]. Moreover, gain and loss of function mutations have implicated CCR7 in the positioning of mature SP thymocytes to the medulla \[[@pbio-0030160-b24],[@pbio-0030160-b25]\]. The contribution of CCR7 and other chemokine receptors in controlling the thymocyte migration patterns described here will be an important area for future investigation. A cortical thymocyte must travel a distance of hundreds of microns in order to reach the medulla. Based on the average distance from the capsule to the medulla in the adult mouse thymus, and the speed and directionality reported here, we estimate that a typical MR^hi^ thymocyte that we image in the cortex could arrive at the medulla in 1 to 2 h. This short time period for migration to the medulla is in contrast to the estimates of 2--3 d for a CD4^+^CD8^+^ thymocyte to complete positive selection \[[@pbio-0030160-b26],[@pbio-0030160-b27]\]. Moreover, we have previously shown that thymocytes frequently pause and turn during MHC-driven contacts with stromal cells \[[@pbio-0030160-b10]\], behaviors that differ strikingly from the behavior of MR^hi^ thymocytes described here. Based on these considerations, we suspect that thymocytes moving at rates greater than 13 μm/min represent cells at a relatively late stage in the positive-selection process and that migration from the cortex to the medulla may not require continuing MHC engagement. In contrast, the MR^lo^ cells are likely to include thymocytes that are actively engaging thymic stromal cells and receiving MHC-driven TCR signals. Future analysis of the signaling events and migratory patterns of these slowly migrating thymocytes may provide further insights into the process of positive selection in the thymus. Materials and Methods {#s3} ===================== {#s3a} ### Generation of GFP chimeric mice {#s3a1} Mice expressing a GFP transgene driven by the ubiquitin promoter \[[@pbio-0030160-b05]\] were used as bone marrow donors for the generation of GFP hematopoietic stem cell chimeras using a modification of a previously described procedure \[[@pbio-0030160-b04]\]. Whole bone marrow from a single adult GFP^+^ mouse was aseptically harvested and resuspended into a single cell suspension in sterile Hanks\' Balanced Salt Solution (Mediatech Cellgro). A total of 2--3 × 10^6^ bone marrow cells were injected into newborn mice (C57Bl/6) in a volume of 70 μl. The first injection was done at 12--24 h after birth and repeated every 2--3 d for a total of four injections. Resulting chimeric mice expressed GFP in 1--2% of their thymocytes. P14 TCR transgenic mice \[[@pbio-0030160-b15]\] on a Rag2^--/--^B6 background were obtained from Taconic. P14^+/+^Rag2^--/--^ mice were crossed with UBI-GFP transgenic mice to generate P14^+/-^Rag2^+/--^GFP^+/--^ mice, and these mice were then intercrossed to generate P14^+/+\ or\ +/--^Rag2^--/--^GFP^+/+\ or\ +/--^ mice. 5CC7 TCR transgenic mice \[[@pbio-0030160-b16]\] on a Rag2^--/--^B10 background were obtained from Taconic and were crossed once with UBI-GFP transgenic mice to generate 5CC7^+/--^GFP^+/--^Rag2^+/--^mice. Bone marrow from adult double transgenic mice was used to generate chimeric mice as described above. ### Two-photon imaging of intact thymic lobes {#s3a2} Thymi from 4.5--5.5 wk-old GFP chimeric mice were quickly harvested, lobes were separated, and the dorsal face of the lobe was adhered to 22 × 22 mm cover glass with single drop of Vetbond tissue adhesive (see [Figure S2](#sg002){ref-type="supplementary-material"}). Cover slip with lobe was immediately placed into a 60 × 15 mm polystyrene Petri dish containing Dulbecco\'s modified Eagle\'s medium (DMEM) without phenol red (Mediatech Cellgro). Petri dish was placed into a heated ring, and the sample was perfused with warmed media bubbled with a blend of 95% O~2~ and 5% CO~2~. Sample was maintained under perfusion and held at 36.5 °C to 37.5 °C throughout imaging. Thymic lobes maintained under these conditions for up to 6 h showed no changes in cell motility and no indications of tissue deterioration. Imaging was performed as previously described \[[@pbio-0030160-b08]\] using an upright Zeiss NLO 510 microscope equipped with a MaiTai Ti:Sapphire laser (Spectra-Physics). For each dataset, 20--33 min of imaging was performed with the objective oriented over the top center of the thymic lobe (corresponding to the ventral side of the organ; see [Figures S2](#sg002){ref-type="supplementary-material"} and [S3](#sg003){ref-type="supplementary-material"}). A total of 20 optical slices were acquired at 2-μm step intervals with a total acquisition time of 36.7 s/z stack. Using Imaris Bitplane software, z stacks (104 × 104 × 40 μm in dimension) were processed into 3D images and reiterated through time to generate a 3D movie of thymocyte migration. In most experiments, data were acquired as "blocks" of stacked movies; a first movie was made at a maximum depth below the capsule with a second movie acquired immediately above the bottom movie (--180 to --80 μm; see [Figure S3](#sg003){ref-type="supplementary-material"}B). Analysis of HNE-stained thymic tissue sections indicated that the area imaged invariably corresponded to cortex (data not shown). ### 4D data analysis {#s3a3} The 4D cell tracking was performed on 75--300 cells per movie using Imaris Bitplane software which identifies the x, y, and z coordinates for each cell at each given time point. These statistics were exported into an Excel spreadsheet for analysis. Average motility rates (MR) were computed as the total length of migratory path divided by the total time of tracking. The extent to which a cell\'s migratory path deviated from a straight line was quantitated as the total length of a cell\'s trajectory divided by the total displacement from origin. For the determination of movement by random walk for MR^lo^ cells, the mean-square displacement from origin was shown to be proportional to time. This relationship is given by 〈r^2^〉 = 6Dt, where r = the displacement from origin, 〈 〉 denotes the average r over numerous events at time t, and 6D is the motility coefficient \[[@pbio-0030160-b09]\], which characterizes the spread of cells in three dimensions. Movement by directed migration for MR^hi^ cells was indicated by a linear relationship between mean displacement from origin (as opposed to its square) and time when computed over many events. The correlation coefficient R^2^ for the best-fit line was computed in Excel using the least squares method. R^2^ at p = 0.001 is statistically significant at values of 0.801 or greater. For displacement analyses of MR^hi^ cells, the average displacement per cell in each direction was calculated from 3 min of tracking. For step analyses, thymocytes were grouped according to their average motility rates and scored as showing net displacement in the positive or negative directions along each of three axes (x, y, and z). The percentage of cells in each motility rate category which moved in the positive and negative directions was then graphed as a function of average motility rate. The computation of statistical significance between the frequencies of 5CC7 TCR thymocytes with high motility rates as compared to wild-type thymocytes was done in Excel using a paired t test. Supporting Information {#s4} ====================== Figure S1 ::: {.caption} ###### Developmental Profiles of GFP+ Thymocytes from Chimeric Mice Representative profiles obtained by flow cytometric analysis of P14 and 5CC7 chimeric thymii. As expected, GFP^+^-gated P14 thymocytes (top row) showed high levels of TCR within the CD4^+^CD8^+^ population and a high percentage of CD8^+^ SP thymocytes, indicating a high frequency of positive selection. As expected for expression of 5CC7 in a nonselecting host (bottom row), thymocytes remained arrested at the CD4^+^CD8^+^ stage of development and fail to upregulate TCR. (128 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S2 ::: {.caption} ###### Orientation of Imaging Relative to Thymic Lobes In Vivo Thymic lobes are depicted in their normal position relative to the heart. Thymic lobes were surgically removed and separated, and then the dorsal side (side facing the heart) of thymic lobe was adhered to glass cover slip. Imaging (see [Figure S3](#sg003){ref-type="supplementary-material"}A and [S3](#sg003){ref-type="supplementary-material"}B) was performed with the objective positioned over the center of the ventral side of lobe. (223 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S3 ::: {.caption} ###### Two-Photon Imaging of Thymocyte Migration in Intact Thymic Lobes \(A) Explanted GFP chimeric thymic lobe was placed in oxygen-perfused media and maintained at 37 °C throughout experiment. Objective was placed directly over the top of lobe and a total of 20 optical slices at 2-μm step intervals were acquired, which generated z stacks of 104 x 104 x 40 μm in the x, y, and z directions. The z stack acquisition was repeated every 37 s for 20--33 min. Stacks were rendered into 3D images and processed through time to yield 4D datasets (see [Videos S1](#sv001){ref-type="supplementary-material"}--[S8](#sv008){ref-type="supplementary-material"}). \(B) In most cases, a stack of movies was generated to increase the effective area of imaging. A bottom movie was generated by imaging starting at --160 to --200 μm below capsule and then a second movie was generated starting 2 μm above the bottom movie. (334 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S4 ::: {.caption} ###### Frequency Distribution of Average Motility Rates for Wild-Type Cortical Thymocytes Histograms showing the frequency distribution of average motility rates for wild-type cortical thymocytes were obtained from four individual runs. Compiled data are shown in [Figure 2](#pbio-0030160-g002){ref-type="fig"}A. (76 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S5 ::: {.caption} ###### Displacement Analyses of Wild-Type MR^hi^ Cells Results of displacement analyses of wild-type MR^hi^ cells from 4 individual experiments are shown. Bar graphs show the average displacement per MR^hi^ cell moved in each direction in a 3-min interval. Data shown were computed from 11--16 MR^hi^ cells from each dataset. The four runs made up two separate stacks of movies (see [Materials and Methods](#s3){ref-type="sec"} and [Figure S3](#sg003){ref-type="supplementary-material"}B). Compiled data are shown in [Figure 3](#pbio-0030160-g003){ref-type="fig"}A. (73 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S6 ::: {.caption} ###### Frequency Distribution of Average Motility Rates for P14 Cortical Thymocytes Histograms showing the frequency distribution of average motility rates for P14 cortical thymocytes were obtained from four individual runs. Compiled data are shown in [Figure 4](#pbio-0030160-g004){ref-type="fig"}A. (79 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S7 ::: {.caption} ###### Displacement Analyses of P14 MR^hi^ Cells Results of displacement analyses of P14 MR^hi^ cells from four individual experiments are shown. Bar graphs show the average displacement per MR^hi^ cell in each direction in a 3-min interval. Data shown were computed from 29--35 MR^hi^ cells from each dataset. Compiled data are shown in [Figure 4](#pbio-0030160-g004){ref-type="fig"}C. (76 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S1 ::: {.caption} ###### GFP Thymocytes within an Intact Thymic Lobe A representative 3D image of GFP thymocytes within an intact thymic lobe. Image is rendered from one z stack at a single time point and is shown in a 360° rotation. Image size is 164 × 164 × 40 μm. Image was recorded approximately 140 μm below the thymic capsule. Corresponds to [Figure 1](#pbio-0030160-g001){ref-type="fig"}. (3.9 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S2 ::: {.caption} ###### Wild-Type GFP Thymocytes Migrating through an Intact Thymic Lobe Time-lapse image of dataset used to generate [Video S1](#sv001){ref-type="supplementary-material"}. Image is shown as a maximum projection of all z stacks. Corresponds to [Figure 1](#pbio-0030160-g001){ref-type="fig"}. All movies were generated from 20 to 33 min of imaging and are played at six frames per second unless otherwise indicated. (2 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S3 ::: {.caption} ###### 4D Tracking of Wild-Type GFP Thymocytes Migrating through an Intact Thymic Lobe Same dataset as shown in [Video S2](#sv002){ref-type="supplementary-material"} with tracks highlighted. Tracks were generated using 4D cell-tracking software. The fluorescence signal from GFP thymocytes is shown in green, and the positions of individual cells as determined by tracking software are represented as grey spheres. Tracks are color coded for time from blue (start of imaging) to light yellow (end of imaging). (2.1 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S4 ::: {.caption} ###### 4D Tracking of Wild-Type Thymocytes Reveals Distinct Migratory Behaviors A time-lapse image of GFP thymocytes in an intact thymic lobe with selected tracks highlighted. Image size is 104 × 104 × 40 μm. Note that the majority of thymocytes migrate slowly and turn frequently, as exemplified by the three MR^lo^ tracks on the right side. A small percentage of thymocytes migrate more rapidly and follow straight trajectories as exemplified by the MR^hi^ track highlighted on the left side. (1.5 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S5 ::: {.caption} ###### MR^hi^ Cell Propulsion Is Associated with Polarized Morphology Time-lapse image of GFP thymocytes cropped to approximately 40 × 40 × 40 μm in the x, y, and z directions. Note the polarized morphology and dramatic shape changes of the MR^hi^ cell as it crawls from bottom to upper left corner. Video shown was generated from 5 min of imaging and is played at six frames per second. Corresponds to [Figure 2](#pbio-0030160-g002){ref-type="fig"}C. (418 KIB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S6 ::: {.caption} ###### P14 TCR Transgenic GFP Thymocytes in an Intact Thymic Lobe Time-lapse image of P14 TCR transgenic GFP thymocytes in an intact thymic lobe. The P14 TCR induces positive selection in this system. Note that a high proportion of thymocytes migrate rapidly and in straight trajectories compared to wild-type GFP thymocytes ([Videos S2](#sg002){ref-type="supplementary-material"} and [S3](#sg003){ref-type="supplementary-material"}). Corresponds to [Figure 4](#pbio-0030160-g004){ref-type="fig"}B. (1.4 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S7 ::: {.caption} ###### P14 GFP Thymocyte Migration is Biased in the z Direction Time-lapse image of P14 TCR transgenic GFP thymocytes in intact thymic lobe is shown rotated to display the x and z dimensions. The same dataset was used to generate [Video S6](#sv006){ref-type="supplementary-material"}. Tracks of MR^hi^ cells are highlighted. Note that the majority of MR^hi^ tracks are oriented in the z direction. (1.3 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Video S8 ::: {.caption} ###### 5CC7 TCR Transgenic GFP Thymocytes in an Intact Thymic Lobe Time-lapse image of 5CC7 TCR transgenic GFP thymocytes in an intact thymic lobe. The 5CC7 TCR is nonselecting in this system. Note the almost complete absence of rapidly migrating thymocytes. (1.1 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: We thank Philippe Bousso, BJ Fowlkes, and members of Robey lab for comments on the manuscript and Holly Aaron for assistance with microscopy. Competing interests. The authors have declared that no competing interests exist. Author contributions. CMW and ER conceived and designed the experiments and performed the experiments. CMW, SR, and AC analyzed the data. BS contributed reagents/materials/analysis tools. CMW and ER wrote the paper. ¤Current address: Department of Biomedical Engineering, University of California, Davis, California, United States of America Citation: Witt CM, Raychaudhuri S, Schaefer, Chakraborty AK, Robey EA (2005) Directed migration of positively selected thymocytes visualized in real time. PLoS Biol 3(6): e160. CCR7 : CC chemokine receptor 7 DN : double negative (CD4^--^CD8^--^) DP : double-positive (CD4^+^ CD8^+^) GFP : green fluorescent protein MR : motility rate MR^hi^ : high motility rate MR^lo^ : low motility rate SP : single-positive (CD4^+^CD8^--^ or CD4^--^CD8^+^) TCR : T cell receptor 3D : three-dimensional 4D : four-dimensional	PubMed Central	2024-06-05T03:55:56.289510	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088277/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e160", "authors": [ { "first": "Colleen M", "last": "Witt" }, { "first": "Subhadip", "last": "Raychaudhuri" }, { "first": "Brian", "last": "Schaefer" }, { "first": "Arup K", "last": "Chakraborty" }, { "first": "Ellen A", "last": "Robey" } ] }
PMC1088278	Introduction {#s1} ============ Genes, or regions of the genome, that have been affected by natural selection may show an excess of functionally important molecular changes, beyond what would be expected in the absence of selection. Genomic regions with such an excess of changes are said to have experienced positive selection, i.e., selection in favor of new genetic variants. The most common statistical technique for detecting positive selection takes advantage of the fact that mutations in coding regions of genes come in two classes: nonsynonymous mutations that change the resulting amino acid sequence of the protein and synonymous mutations, which do not change the encoded protein. An excess of nonsynonymous mutations over synonymous mutations, beyond what would be expected if the two types of mutations occur at the same rate, provides strong evidence for the past action of positive selection at the protein level. Using this logic, there have recently been numerous studies documenting positive selection in a variety of genes and organisms, including immune-response-related genes \[[@pbio-0030170-b01]--[@pbio-0030170-b03]\], viral genes \[[@pbio-0030170-b04]--[@pbio-0030170-b06]\], fertilization genes \[[@pbio-0030170-b07],[@pbio-0030170-b08]\], and genes involved in sensory perception and olfaction in humans \[[@pbio-0030170-b09]\]. Clark et al. \[[@pbio-0030170-b10]\] compared 7,645 genes from humans to their orthologs from the chimpanzee and the mouse. For each gene, they tested if there was an excess of nonsynonymous substitutions on the evolutionary lineage leading to humans. They showed that there was an excess of putatively positively selected genes in several functional classes, including genes involved in sensory perception, olfaction, and amino acid catabolism. They also showed that human genes that have been targeted by positive selection are significantly more likely to harbor variation associated with known genetic diseases. We here report the results of an analysis of 20,361 human and chimpanzee genes (of which 6,630 later were eliminated in a very conservative quality control), which includes the 7,645 genes analyzed by Clark et al. \[[@pbio-0030170-b10]\]. While the objective of the study by Clark et al. \[[@pbio-0030170-b10]\] was to find genes that have experienced accelerated evolution on the human lineage, using the mouse as an outgroup, the aim of the current study is to find genes that have been targeted by positive selection at any point in time during the evolution of humans and chimpanzees, based on a larger set of genes. We use a likelihood ratio test to identify positive selection and do extensive simulations to find the appropriate critical values of the test. Positive selection is inferred if the ratio of nonsynonymous substitutions per nonsynonymous site to synonymous substitutions per synonymous site (d~N~/d~S~) is statistically significantly greater than one in a test of the neutral null hypothesis d~N~/d~S~ = 1 \[[@pbio-0030170-b11],[@pbio-0030170-b12]\]. The method used for detecting positive selection takes transition/transversion rate biases and unequal codon and amino acid frequencies into account. The test for positive selection applied in this study is a traditional test of d~N~/d~S~ greater than one. It has more power than the test used in the Clark et al. study \[[@pbio-0030170-b10]\] if selection affects both the human and the chimpanzee lineages because it uses information from both lineages. Results {#s2} ======= Chimpanzee sequence was obtained by PCR using primers designed to flank exon sequence annotated in the human genome \[[@pbio-0030170-b10]\]. Our analysis begins with data from 20,361 coding regions, including 103,606 nucleotide differences and 403 indels among 17,687,331 aligned nucleotides. These numbers are significantly lower than the genome-wide averages \[[@pbio-0030170-b13],[@pbio-0030170-b14]\], presumably due to selective constraints in the coding regions. The distributions of nonsynonymous and synonymous nucleotide differences among genes are shown in [Figure 1](#pbio-0030170-g001){ref-type="fig"}. The average numbers of nonsynonymous and synonymous mutations per nucleotide site are 0.002578 and 0.003281, respectively. Eliminating reads without a hit to known genes in public databases (see [Materials and Methods](#s4){ref-type="sec"}), there are 71,896 nucleotide differences in 13,731 genes. The remaining analysis is restricted to this set of genes. Among them, 5,574 were eliminated from the positive selection analysis because they had fewer than three mutations, and 797 were eliminated because the sequence was less than 50 bp long. Additionally, 45 genes were eliminated because they contained internal stop codons, presumably due to erroneous annotations or sequencing errors. Among the remaining 8,079 genes, 3,913 were also analyzed by Clark et al. \[[@pbio-0030170-b10]\]. ::: {#pbio-0030170-g001 .fig} Figure 1 ::: {.caption} ###### Distribution of Mutations The figure shows the number of synonymous and nonsynonymous nucleotide differences in 13,731 human--chimpanzee orthologous gene pairs. ::: ![](pbio.0030170.g001) ::: The average level of sequence divergence was 0.60%, corresponding to a divergence level of 1.57% in silent sites. This figure matches well the level of divergence observed by Ebersberger et al. \[[@pbio-0030170-b14]\] for Chromosome 22 of 1.44% overall and 2.26% in CpG islands. Seven hundred thirty-three of the 8,079 genes evolved with d~N~/d~S~ greater than one, but only 35 had p-values less than 0.05, as determined by a likelihood ratio test of the null hypothesis of d~N~/d~S~ = 1 against the alternative hypothesis of d~N~/d~S~ greater than one. The number of significant genes at the 5% level, in this one-sided test, is lower than the nominal level because the vast majority of genes are conserved and evolve with d~N~/d~S~ less than one. Nonetheless, after using Simes\'s improved Bonferroni procedure \[[@pbio-0030170-b15]\] we can, at the 5% significance level, reject the hypothesis that none of the genes are evolving with d~N~/d~S~ greater than one. This also implies that a 5% false discovery rate set is nonempty. Even though the level of divergence between humans and chimpanzees is very low, there is statistically significant evidence for positive selection in the DNA sequences of these two species. Results for all genes are available in [Dataset S1](#sd001){ref-type="supplementary-material"}. Biological Processes Affected by Positive Selection {#s2a} --------------------------------------------------- To identify functional groups of genes with an overrepresentation of putatively positively selected genes, we used the PANTHER \[[@pbio-0030170-b16],[@pbio-0030170-b17]\] classification of biological processes and a Mann-Whitney U test (MWU) based on the p-values from the likelihood ratio test ([Table 1](#pbio-0030170-t001){ref-type="table"}). The classification based on the MWU identifies categories of genes with small p-values from the likelihood ratio test. It is important to notice that genes that evolve approximately neutrally will tend to have smaller p-values than genes evolving under strong functional constraints. The classification based on the MWUs, therefore, does not provide unambiguous evidence for positive selection, but it provides a key to which groups harbors the most candidates for positive selection. ::: {#pbio-0030170-t001 .table-wrap} Table 1 ::: {.caption} ###### Biological Process Categories with an Excess of Putatively Positively Selected Genes (Nominal p less than 0.05; MWU) among a Total of 133 Biological Process Categories ::: ![](pbio.0030170.t001) Note that the categories overlap; e.g., "T-cell-mediated immunity" is entirely nested within "Immunity and defense." ::: Immune-defense-related genes appear at the top of the list. It is not surprising that several of the genes experiencing most positive selection are involved in immune responses to viruses. Considering the speed at which many pathogens, such as viruses, evolve (e.g., \[[@pbio-0030170-b05]\]), a coevolutionary molecular arms race between pathogens and host cells might explain the presence of strong selection favoring new mutations in these genes. Other forces, including overdominant selection to diversify the spectrum of immune responses, may also cause positive selection in immune- and defense-related genes. Such explanations have previously been used to explain the presence of positive selection in the human major histocompatibility complex \[[@pbio-0030170-b18]\]. As in \[[@pbio-0030170-b10]\] we also identify genes involved in various forms of sensory perception, including olfaction and genes classified as "unknown biological function." Many of the genes with unknown biological function show sequence similarity with known transcription factors (data not shown). Much of the selection on sensory genes is driven by the selection on olfactory receptors previously found by Gilad et al. \[[@pbio-0030170-b09]\]. In contrast to Clark et al. \[[@pbio-0030170-b10]\], we also find that genes involved in spermatogenesis appear to have an excess of positively selected genes. The genes involved in spermatogenesis showing the strongest evidence for positive selection include several KRAB-containing zinc finger proteins that serve as repressors of transcription and are believed to be involved in determining the differentiation of pluripotent stem cells \[[@pbio-0030170-b19]\]. Expression Patterns and Positive Selection {#s2b} ------------------------------------------ We also categorized 3,464 of the 8,079 genes according to the tissue of expression in the Novartis Gene Expression Atlas \[[@pbio-0030170-b20]\]. Because of the relatively small number of tissue-selective genes in our dataset (204) and the large number of tissues analyzed (28), many tissues had fewer than 20 tissue-selective genes, providing little statistical power for further subdivision. Therefore, we examined instead whether the tissue of maximal expression for a gene was correlated with positive selection, since high expression levels and importance in tissue function are often, but not always, correlated. The set of genes that have their maximal expression in the testes is the only one showing an excess of positive selection, after a Bonferroni correction for multiple tests ([Table 2](#pbio-0030170-t002){ref-type="table"}). ::: {#pbio-0030170-t002 .table-wrap} Table 2 ::: {.caption} ###### Test for an Excess of Putatively Positively Selected Genes by Tissue Type ::: ![](pbio.0030170.t002) Small p-values (MWU; nominal p-values not corrected for multiple testing) indicate an excess of putatively positively selected genes in the tissue type. ::: Genes with their maximal expression in the brain do not have an excess tendency toward positive selection. In fact, genes expressed in the brain seem to be among the most conserved genes with the least evidence for positive selection. MWUs, comparing genes with their maximal expression in the brain (83 genes) to all other genes, show that these genes tend to have significantly higher p-values of the likelihood ratio test for positive selection (p = 0.035), indicating high levels of selective constraint. Genes that are expressed in the brain at a level of twice the expression level found in blood show an even stronger tendency toward avoidance of positive selection (p = 0.0002). Although studies of gene expression in the brain tissue are complicated by low-abundance transcripts and heterogeneous specialized brain regions \[[@pbio-0030170-b21]\], the overall evidence points toward a deficiency of positively, or fast evolving, genes among those expressed in the brain. The causes for the cognitive differences may instead be sought in adaptive changes in just a few genes, in changes in gene expression \[[@pbio-0030170-b22]\], or in changes in copy number and/or organization of genes relating to cognitive function \[[@pbio-0030170-b23]\]. Dorus et al. \[[@pbio-0030170-b24]\] found that genes expressed in the nervous system showed a relative increase in the rate in primates relative to rodents when compared to housekeeping genes, but provided no direct evidence for positive selection on these genes. Nervous-system-specific genes appear to be so conserved that it is unlikely that direct evidence for positive selection will be discovered in this group of genes. Positive Selection in the X Chromosome {#s2c} -------------------------------------- We also tested if any chromosomes show an excess of genes with evidence for positive selection. The only chromosome enriched in genes with small p-values from the likelihood ratio test for positive selection is the X chromosome (p = 0.0049; MWU). Several factors influence the contrast between the X and autosomes in tests of selection, including hemizygosity of the X in males, resulting in more effective selection against deleterious recessive and in favor of positive recessive mutations \[[@pbio-0030170-b25]\]. Male hemizygosity also results in mutations, with male-specific effects being more readily fixed by selection on the X \[[@pbio-0030170-b26]\]. This increased efficiency of selection for male-specific genes on the X may explain the excess of X-linked genes expressed in spermatogonia \[[@pbio-0030170-b27]\]. The observation that reproductive proteins generally evolve at a greater rate, coupled with the overrepresentation of male-specific genes on the X, could produce the excess positive selection seen on the X. However, after eliminating all genes with highest expression levels in the testis, or annotated as functioning in spermatogenesis, there is still an excess of putatively positively selected genes on the X chromosome (p = 0.0131; MWU). Thus, it appears that the elevated positive selection on the X is likely due to the general tendency of mutations to be recessive, regardless of their tendency to be male-limited in expression. Although other factors, such as an elevated male mutation rate \[[@pbio-0030170-b28]\], differences in the efficacy of genetic hitchhiking between autosomes and the X chromosome \[[@pbio-0030170-b29]\], and correlations between recombination rate and divergence \[[@pbio-0030170-b30]\], may cause differences in variability and substitution rate between autosomes and the X chromosome, none of these factors alone can explain the excess of positively selected genes on the X chromosome. Analysis of the 50 Genes Showing Strongest Evidence for Selection {#s2d} ----------------------------------------------------------------- We studied the 50 genes with the highest likelihood ratios in greater detail to further characterize the causes of positive selection and examine error rates ([Table 3](#pbio-0030170-t003){ref-type="table"}). To investigate the degree to which our results might be influenced by sequencing errors, we compared the data for these genes with the public data available for the same genes. In the regions with overlap between the public data and our data there were a total of 327 mutations in the public data and 306 mutations in our data. This demonstrates that there is not an excess of (potentially artifactual) mutations in our data in the genes that show evidence for positive selection. While most of the 50 genes also show strong evidence for positive selection in the public data, six of the genes do not. HC19953, HC2758, HC6579, HC7761, HC8067, and HC9844 do not have d~N~/d~S~ ratios larger than one in the public data. In most cases, the difference is caused by the fact that our database and the public database contain different regions of the genes. Not all regions of a gene are expected to be targeted by positive selection, but this does not challenge the evidence for positive selection in the regions of the genes included in this analysis. In any case, using the public data would not change the qualitative conclusions of the analysis of the genes presented here. ::: {#pbio-0030170-t003 .table-wrap} Table 3 ::: {.caption} ###### The Top 50 Genes Showing Evidence for Positive Selection ::: ![](pbio.0030170.t003) ^a^Reference number used in Dataset S1. ^b^Number of nonsynonymous differences between humans and chimps. ^c^Number of synonymous differences between humans and chimps. ^d^Number of nonsynonymous polymorphism in humans. ^e^Number of synonymous polymorphism in humans. ^f^Likelihood ratio from the likelihood ratio test of d~N~/d~S~ equals one versus d~N~/d~S~ is greater than one in the human--chimp alignment. ::: Immunity and Defense Genes Targeted by Positive Selection {#s2e} --------------------------------------------------------- The top 50 genes include many genes that we might a priori expect to be targets of positive selection, including four genes involved in olfaction (OR2W1, OR5I1, OR2B2, and C20orf185) and several genes involved in host--pathogen interactions, such as CMRF35H, CD72 antigen, pre-T-cell antigen receptor α (PTCRA), APOBEC3F, and granzyme H (GZMH). Only one of these genes was among the 50 most significant entries in the Clark et al. \[[@pbio-0030170-b10]\] model 2 analysis. APOBEC3F encodes an antiviral factor that has previously been demonstrated to be under positive selection by Sawyer et al. \[[@pbio-0030170-b03]\] who note that this gene has been associated with anti-HIV activity. Presumably, most of these genes have been targeted by positive selection throughout the primate and mammalian phylogeny. The widespread evidence for positive selection in immune-related genes confirms the hypothesis that much positive selection in the human and mammalian genomes may be driven by a coevolutionary arms race between host immune system and pathogens. Spermatogenesis- and Apoptosis-Related Genes {#s2f} -------------------------------------------- The list also contains many testis- or sperm-specific genes including Protamine-1 (PRM1), which previously has been shown to be under positive selection \[[@pbio-0030170-b31]\], possibly due to sperm competition (but see \[[@pbio-0030170-b32]\] for an alternative explanation). Other sperm-specific genes on the list include USP26, C15orf2, PEPP-2, TCP11, HYAL3, and TSARG1. The inclusion of these genes in the list of the genes showing the strongest evidence for positive selection is consistent with the results, based on the PANTHER annotation and the Novartis expression data, of excess positive selection in sperm/testis-specific genes. The possible causes include sperm competition (e.g., \[[@pbio-0030170-b31]\]), sexual conflict (e.g., \[[@pbio-0030170-b07],[@pbio-0030170-b08]\]), selection for reproductive isolation, pathogen-driven selection in the reproductive organs, and selection related to the occurrence of mutations causing segregation distortion. We notice that at least one of these genes (TSARG1) is involved in apoptosis during spermatogenesis. Apoptosis of germ cells is conspicuous during normal spermatogenesis, eliminating up to 75% of the potential spermazoa \[[@pbio-0030170-b33]--[@pbio-0030170-b35]\], affecting cells both before and after the meiotic division \[[@pbio-0030170-b36]\]. It has been hypothesized that the main cause for the high rate of apoptosis during spermatogenesis is to maintain a proper cell-number ratio between maturing germ cells and Sertoli cells \[[@pbio-0030170-b35]\]. The natural process of elimination of germ cells by apoptosis creates a genomic conflict in which each individual germ cell will benefit from avoiding apoptosis, but apoptosis of a certain fraction of germ cells may be beneficial to the mature organism. New mutations occurring in cells during spermatogenesis, which reduces the probability of apoptosis, will be positively selected. This effect will be particularly strong for mutations in genes expressed after the meiotic division, potentially resulting in segregation distortion. A mutant with an even very small increase in the probability of escaping postmeiotic apoptosis will have a strong selective advantage. Compensatory mutations, reducing or eliminating the effect of the apoptosis avoidance mutation, may then later occur. These dynamics may lead to recurrent events of positive selection in genes affecting spermatogenesis apoptosis. The 40 genes in this study involved in inhibition of apoptosis show an excess of evidence for positive selection compared to other categories (p = 0.0047; see [Table 2](#pbio-0030170-t002){ref-type="table"}). Many of the genes showing most evidence for positive selection are known to be involved in either spermatogenesis, apoptosis, or both. For example, the apoptosis-related gene showing the strongest evidence for positive selection (DFFA) is an inhibitor of Fas-mediated apoptosis, which has been shown to be involved in apoptosis during spermatogenesis \[[@pbio-0030170-b36]\]. This may suggest that genomic conflict due to spermatogenesis apoptosis may be driving positive selection in many of the included genes. Cancer-Related Genes {#s2g} -------------------- While we expected to find genes involved in olfaction, spermatogenesis, and immune defense among the 50 annotated genes showing the strongest evidence for positive selection, we were surprised to find a very large proportion of cancer-related genes, especially genes involved in tumor suppression, apoptosis, and cell cycle control. These genes include four putative tumor suppressors: HYAL3, DFFA, PEPP-2 (note that both HYAL3 and PEPP-2 also appear to be involved in spermatogenesis), and C16orf3, another gene associated with tumor progression (MMP26), and a gene with unknown function but high similarity to melanoma-associated antigens (FLJ32965). In addition, there are several genes involved in apoptosis (PPP1R15A, HSJ001348, TSARG1, and GZMH). Given that many of the genes have very little functional information, it is surprising to find such a large proportion of genes that may be related to tumor development and control. The factors causing positive selection on these genes are unknown, but genes important in tumor development and suppression may be positively selected due to other functional effects of the genes, particularly in immunity and defense or in spermatogenesis. Several of the genes involved in tumor suppression or progression show testis-specific expression, and models of genomic conflict may explain the presence of positive selection in these genes. It should be noted that there is no pattern of human-specific selection in these genes. The high number of nonsynonymous mutations in these genes is approximately evenly distributed between the human and the chimpanzee lineage (results not shown). PAML Analysis {#s2h} ------------- For each of the 50 genes, we searched public databases to find orthologous genes in other mammals. For 25 of the genes we were able to identify orthologs from mouse and rat, and for these 25 genes we estimated the d~N~/d~S~ ratio of each lineage of the underlying phylogeny using PAML \[[@pbio-0030170-b37]\]. The d~N~/d~S~ ratio was elevated (p \< 0.05) in 5/25 cases in just the human lineage, in 5/25 cases in just the chimp lineage, in 8/25 cases in both lineages, and in 7/25 cases significant in neither lineage. These results show that the elevated d~N~/d~S~ ratios are a consequence of positive selection in both the human and the chimpanzee lineage. Population Genetic Analysis {#s2i} --------------------------- To further investigate the effect of selection on the 50 genes showing the strongest evidence for positive selection, 20 European-American and 19 African-American individuals were sequenced for these genes. Forty-six of the genes contained intraspecific polymorphism, and there were a total of 55 synonymous polymorphisms and 116 nonsynonymous polymorphisms, showing that the d~N~/d~S~ ratio is also relatively high in the polymorphism data. The distribution of allele frequencies within these genes, as summarized by the allele frequency spectrum, provides additional support for positive selection. The frequency spectrum ([Figure 2](#pbio-0030170-g002){ref-type="fig"}) of synonymous polymorphisms does not deviate from the pattern expected under a standard neutral model \[[@pbio-0030170-b38]\]. However, this does not necessarily provide evidence for the adequacy of the standard neutral model, but may rather be caused by a cancellation of effects due to population growth, population subdivision, and linkage to selected mutations, low power due to the small sample size, or by an ascertainment bias described below. Other data in humans have shown an excess of rare derived alleles in synonymous sites, presumably caused by population growth \[[@pbio-0030170-b39],[@pbio-0030170-b40]\]. In contrast, we find that nonsynonymous single nucleotide polymorphisms (SNPs) show evidence for an excess of high-frequency-derived alleles in these genes ([Figure 2](#pbio-0030170-g002){ref-type="fig"}). The excess of high-frequency-derived nonsynonymous mutation supports the notion that these genes have been targeted by positive selection. An important caveat is that an ascertainment bias has been introduced because interspecific and intraspecific variability has been confounded when selecting genes with high d~N~/d~S~ ratios. To assess the impact of this ascertainment bias, we simulated 1,000 new neutral datasets, each dataset consisting of 13,731 genes with a similar distribution of d~N~/d~S~ ratios, mutation rates, and sequence lengths, as observed in the real data, and with both interspecific and intraspecific variation. From these datasets we selected the 50 genes with the largest d~N~/d~S~ ratios, as in the selection procedure applied to the real data. There is a clear effect of the ascertainment bias on synonymous sites, but there is essentially no effect on nonsynonymous sites ([Figure 2](#pbio-0030170-g002){ref-type="fig"}). The main effect of the ascertainment bias is to eliminate genes with many high-frequency-derived synonymous mutations. This shows that the excess of high-frequency-derived nonsynonymous mutations is not a result of the ascertainment bias. ::: {#pbio-0030170-g002 .fig} Figure 2 ::: {.caption} ###### Frequency Spectra The figure shows the frequency spectra of nonsynonymous (red) and synonymous (black) mutations among the 50 genes showing the strongest evidence for positive selection in the interspecific comparison. Also shown is the expectation from the standard neutral model, expectations from the neutral model taking the protocol used to select the 50 genes into account (see text), and from the prediction of the selection model. On the x-axis is the number of derived allele in a sample of size 30 chromosomes (Count), and on the y-axis is the proportion of sites expected in the sample with a particular frequency. ::: ![](pbio.0030170.g002) ::: In addition to selection, certain demographic factors, such as population bottlenecks and population subdivision \[[@pbio-0030170-b41],[@pbio-0030170-b42]\], and/or incorrectly inferred ancestral states may also enrich the sample with apparent high-frequency-derived mutations. Przeworski \[[@pbio-0030170-b42]\] has previously reported an excess of high-frequency-derived mutations in human data. To investigate this possibility we compared the frequency spectrum in our data to the frequency spectrum of the genes in the Seattle SNP database (SeattleSNPs; <http://pga.gs.washington.edu> \[01/10/03\]). These data also consist of a mixture of declared African Americans and European Americans and should, therefore, comprise a suitable sample for comparison. With 24 out of 116 and 37 out of 360 nonsynonymous mutations of frequency greater than 50% in our data and the Seattle data, respectively, there is a significant excess of high-frequency-derived mutations in our data compared to the Seattle data (p \< 0.01, chi-square test). The Seattle data shows a slight deficiency of nonsynonymous-derived mutations with frequency greater than 50%, primarily due to an excess of very low-frequency-derived mutations. These results strongly suggest that the pattern we observe is caused by ongoing positive selection and not by demographic effects. There are a total of 25/78 and 22/92 polymorphisms of frequency greater than 50% within the Caucasian and African-American groups, respectively. Analyzing each population separately gives an even more extreme excess of high-frequency-derived polymorphism, especially in the Caucasian population. There is a very high variance in the ratio of divergence to polymorphism in these genes (Hudson-Kreitman-Aguadé test; p less than 0.05). While the overall ratio of divergence to polymorphism is around two (2.06), a few genes stand out as having particularly high levels of polymorphism. For example, one of the olfactory receptors, OR5I1, has six substitutions and 11 polymorphisms. This raises the possibility that positive selection in the olfactory receptors may be a type of balancing selection. One possibility is heterozygote advantage driven by selection to increase the repertoire of olfactory receptors. Another gene with a low divergence to polymorphism ratio is RPP38 (four substitutions and seven polymorphisms), which is a subunit of RNase P. RPP38 is necessary for normal processing of stable RNA in human cells, but it is also a target for antisera from systemic sclerosis patients. It is likely that the positive selection in this gene is caused by selection to avoid an autoimmune response. Such a hypothesis is plausible if the sequence pattern of RPP38 influences the likelihood of developing systemic sclerosis. This hypothesis can be tested using linkage or linkage disequilibrium studies. Other genes show an apparent deficiency of polymorphisms. SCML1 has 16 substitutions (of which 15 are nonsynonymous) and zero polymorphisms. Such a pattern is consistent with repeated selective sweeps driving divergence between species, while eliminating variation within species. SCML1 is a repressor of expression of Hox genes and may play an important role in the control of embryonal development \[[@pbio-0030170-b43]\]. This gene may be a prime candidate for explaining developmental differences between humans and chimpanzees. Poisson Random Field (PRF) Analysis {#s2j} ----------------------------------- To further investigate the distribution of selection coefficients among mutations in these genes, we applied a PRF model \[[@pbio-0030170-b44]\]. In PRF models, the distribution of sample allele frequencies can be expressed as a function of the scaled selection coefficient, S, (S = 2Ns; N = population size, s = selection coefficient) acting on a mutation. We assumed that there were three types of mutations: negatively selected mutations (of frequency p ~--~), neutral mutations (of frequency p ~0~), and positively selected mutations (of frequency p~+~ = 1 -- p ~--~ -- p ~0~). We then estimated p ~--~, p ~0~, p ~+~, and the scaled selection coefficients of the mutations in the two selected categories (S ~--~ and S ~+~) using maximum likelihood. The maximum likelihood estimates of the parameters of the PRF model are p ~--~ = 0.748, p ~0~ = 0.172, p ~+~ = 0.080, S ~--~ = --34.96, and S ~+~ = 267.11; i.e., the estimated proportion of negatively selected mutations is approximately 75%, and the proportion of positively selected mutations is approximately 8%. The proportion of positively selected mutations is so high because we have analyzed the 50 genes showing the strongest evidence for positive selection among a very large pool of candidate genes. Likelihood ratio tests show that a model with three selected classes fits the nonsynonymous data significantly better than a model with fewer selective classes (see [Materials and Methods](#s4){ref-type="sec"}). We conclude that the allelic distribution in nonsynonymous sites is best described by a mixture of neutral, positively selected and negatively selected mutations. In this case, our best estimate of the proportion of mutations in these genes that are neutral is less than 18%. The predicted frequency spectrum under the estimated selection model is shown in [Figure 2](#pbio-0030170-g002){ref-type="fig"}. The results of this analysis should be interpreted with some caution because the effects of linkage have been ignored. The effects of linkage would be to underestimate the selection coefficient and, possibly, to overestimate the number of mutations that have been targeted by selection \[[@pbio-0030170-b45]\]. As previously discussed, these types of inferences are also sensitive to the demographic assumptions of a panmictic population of constant size \[[@pbio-0030170-b41],[@pbio-0030170-b42]\] and to the assumptions regarding unambiguous inference of the ancestral state from the chimpanzee. For these reasons, the exact values of the parameter estimates should not be overinterpreted, but may help suggest the magnitude of the selective forces necessary to explain the data in isolation. Discussion {#s3} ========== The statistical methods used for detecting selection have been the subject of debate over the past few years \[[@pbio-0030170-b46],[@pbio-0030170-b47]\]. This debate has mainly focused on the validity of methods that model variation in the d~N~/d~S~ ratio among sites. The current test does not model rate variation among sites and should, therefore, be uncontroversial. Unfortunately, this test may also have very low power. To determine the power of the test, we conducted power simulations under parameter values estimated from the data ([Figure 3](#pbio-0030170-g003){ref-type="fig"}). Notice first that the test does not result in excess significant results when d~N~/d~S~ = 1, and results in very few falsely significant results when d~N~/d~S~ is less than one. In fact, when d~N~/d~S~ = 1 the power is lower than the nominal significance level because of the possibility of ties. However, the power increases steadily when d~N~/d~S~ increases above one, and for a gene of length 500 codons, the test has more than 80% power when d~N~/d~S~ = 5. For a functional gene, in which most sites are expected to be under functional constraints and evolve with d~N~/d~S~ less than one, a significant value of the test is almost surely caused by positive selection. The fact that our data shows significant evidence for positive selection, when using this test with a correction for multiple tests, illustrates that positive selection can be detected from human--chimpanzee comparative data despite the very low levels of divergence. ::: {#pbio-0030170-g003 .fig} Figure 3 ::: {.caption} ###### Power of the Likelihood Ratio Test for Positive Selection The power is shown as a function of the proportion of the d~N~/d~S~ ratio, and for sequence lengths (n) of 150 and 500 codons. Power is defined as the proportion of tests that are significant at the 5% level. Simulation parameters, including codon frequencies, transition/transversion bias, and divergence times, are equal to the values estimated from the data. Notice the logarithmic x-axis. ::: ![](pbio.0030170.g003) ::: In the previous study by Clark et al. \[[@pbio-0030170-b10]\], an outgroup (mouse) was used to make inferences regarding human-specific processes. We have here analyzed a larger dataset but cannot, in general, distinguish between selection that is particular to the human evolutionary lineage and positive selection that tends to occur in both chimps and humans. While Clark et al. \[[@pbio-0030170-b10]\] found strongest evidence for positive selection in genes related to olfaction and sensory perception, we find the strongest evidence for positive selection in genes related to immunity and defense. The reason is probably that genes related to immunity and defense are targets for positive selection throughout the mammalian phylogeny, which the test used by Clark et al. \[[@pbio-0030170-b10]\] would not detect, whereas much of the selection on sensory perception and olfaction is likely to be specific to the distinct niches of humans and chimpanzees. Similar arguments may also explain why we obtain strong evidence for positive selection on genes related to spermatogenesis and inhibition of apoptosis, while Clark et al. \[[@pbio-0030170-b10]\] did not find any evidence for human-specific selection on genes related to spermatogenesis and apoptosis. In this paper we analyzed population genetic data from the 50 genes showing most evidence for positive selection. An excess of high-frequency-derived nonsynonymous mutations in these data supports the conclusions that these genes are targeted by positive selection. Although some demographic models also may cause an excess of high-frequency-derived mutations \[[@pbio-0030170-b41],[@pbio-0030170-b42]\], the excess observed in our data is beyond the level observed in other comparable human data. The use of the population genetic data may also help suggest the mode of positive selection acting on the gene. For example, a developmental gene (SCML1) had 16 fixed substitutions and zero polymorphisms, suggesting repeated selective fixations, whereas and olfactory receptor had six substitutions and 11 polymorphisms consistent with the action of balancing selection. The combined use of comparative and population genetic data may help, not only to identify positive selection, but also to help narrow down possible models of positive selection. With the increased availability of both comparative genomic data and SNP data, we expect to see many future studies that take advantage of the availability of both types of data. The discovery that many genes involved in spermatogenesis, apoptosis, and tumor suppression are positively selected may prompt further investigations into models of genomic conflict and other models predicting positive selection in these genes. In general, mutations that increase the expected number of functional sperm cells produced by a specific germ-line cell, such as mutations increasing the rate of cell division or decreasing the probability of apoptosis, will be favored. Such mutations will not necessarily increase the fitness of the mature organism, leading to a genomic conflict, in which selfish mutations causing avoidance of apoptosis are being counteracted by compensatory mutations in other loci. Many of the genes with evidence for positive selection encountered in this study play functional roles in cell cycle regulation, tumor suppression, apoptosis, or spermatogenesis. We suggest that a genomic conflict relating to the process of spermatogenesis may be responsible for much of the positive selection observed in this study. Because many of these genes are involved in inhibition of apoptosis, this may also explain the apparent excess of cancer-related genes targeted by positive selection. This raises the interesting prospect that the high prevalence of cancer in humans and other organisms may be related to selection for apoptosis avoidance in the germ line. Mutations that in general increase apoptosis avoidance will be selected in the germ line, but such mutations may at the same time increase the probability of cancer in somatic tissue. The relative high prevalence of cancer will, according to this hypothesis, be related to an evolutionary conflict between the selfish interests of a germ cell and selection at the level of mature organisms to decrease the cancer rate. We note that the fact that the same pathways (e.g., Fas-mediated apoptosis) are involved in the control of cancer and in apoptosis during spermatogenesis supports this hypothesis. At present we cannot exclude an alternative hypothesis, such as pathogen-driven positive selection or sperm competition. Future functional and evolutionary studies of the genes suggested to be under positive selection by this study may help determine which of these alternative evolutionary models are most plausible. Materials and Methods {#s4} ===================== {#s4a} ### DNA sequencing and alignment. {#s4a1} Sequences of chimpanzee genes were obtained by PCR amplification of individual exons from a single western chimpanzee male. PCR products were directly sequenced on automated sequencers at Celera Genomics. Details of primer construction, DNA sequencing, and alignment were described in Clark et al. \[[@pbio-0030170-b10]\] and references therein. Chimpanzee sequences were obtained for both strands from PCR products, filtered to remove base calls with Phred scores less than 30. Genes that did not have a hit in the curated accessions (NM\_ or NR\_ series) in the REFSEQ 3.0 database (<http://www.ncbi.nlm.nih.gov/RefSeq/>) and for which the best hit did not map to the same genomic location in NCBI 34 build of the human genome, were omitted from the functional analysis, to minimize the chance of including nonfunctional genes/regions. This was done using BLAT v. 27 (<http://www.genomeblat.com/genomeblat/index.asp>). Regions of the alignment that were not in the REFSEQ database were eliminated from the analysis. However, we provide the full dataset ([Dataset S2](#sd002){ref-type="supplementary-material"}) for future exploration. Indels were identified after alignment as all nonterminal gaps that could not be attributed to low base-calling scores. All pairwise alignments are available in Phylip and FASTA format in [Dataset S2](#sd002){ref-type="supplementary-material"}. Human polymorphisms were detected automatically from assembled sequencing traces using PolyPhred 4.0 \[[@pbio-0030170-b48]\] and RuleGen, a decision-tree-based method (S. Glanowski, unpublished data). Manual calls were employed if a potential SNP was not flagged by both programs. Validation of the automated pipeline using a set of several hundred manually called SNPs showed a sensitivity of 85% for all SNPs and up to 100% for SNPs with more than three minor alleles observed. Independent verification of several hundred SNPs using TaqMan assays indicated that validation rates of 95% for common SNPs and 90% for SNPs with only one minor allele were observed. ### Likelihood ratio tests {#s4a2} For each human--chimpanzee orthologous gene pair, a likelihood ratio test of the hypothesis of an equal d~N~/d~S~ ratio was performed using a codon-based likelihood model (see \[[@pbio-0030170-b12]\] for such tests). The test was performed as a one-sided test of the hypothesis H~0~: d~N~/d~S~ = 1 versus the alternative of H~A~: d~N~/d~S~ greater than one. To reduce the computational burden, the transition/transversion rate ratio was first estimated for genes in high-GC-content regions and low-GC-content regions separately. This parameter was then considered fixed for the remainder of the analysis. Because many of the sequence pairs showed very little divergence, the usual asymptotic assumptions of a chi-square distribution of the likelihood-ratio test statistic would not have been appropriate. Instead, simulations were performed to determine the appropriate distribution of the test statistic. The simulations were performed under the empirical distribution of the divergence time and other parameter estimates assuming d~N~/d~S~ = 1. The distribution of the test statistic, conditional on the observed number of nucleotide differences between the sequences (for each gene), was then determined. One of the advantages of using the conditional distribution is that the distribution becomes more robust to violations of the assumptions regarding the nuisance parameters, particularly the divergence times, and this will allow us to exclude genes with very little variability while maintaining the right size of the test. Genes with fewer than three nucleotide differences, or with fewer than 50 aligned codons, were excluded from the analysis. ### Functional analysis. {#s4a3} The functional annotation was performed as in \[[@pbio-0030170-b10]\], using the PANTHER database \[[@pbio-0030170-b16],[@pbio-0030170-b17]\]. Throughout, excesses of positively selected genes in a category were tested, using an MWU comparing the distribution of p-values obtained from the likelihood ratio tests in genes included in the category to the distribution of such values in genes not included in the category. Genes with fewer than three nucleotide differences between humans and chimpanzees were excluded from the identification of categories with an excess of putatively positively selected genes. The MWU does not in itself demonstrate that the evolution of a particular category of genes is affected by positive selection, but it shows that the category contains more evidence for positive selection than other genes in the study. Because genes of short sequence length are less likely to show strong evidence for positive selection, but are more likely to show spurious evidence for positive selection, the MWU (or any other categorization) may be affected by different sequence lengths in different categories. The reason for using an MWU, instead of reporting overall p-values for a category after correction for multiple testing, is that such an approach would be strongly influenced by just one or a few genes. However, correction for multiple testing reveals significant positive selection in several categories, including the immune and defense and the spermatogenesis category. ### Expression data. {#s4a4} Expression data from normal human tissues were obtained from the Novartis Gene Expression Atlas (\[[@pbio-0030170-b20]\]; <http://wombat.gnf.org/index.html>; 6,741 gene symbols could be matched unambiguously to the human--chimp alignments. All negative expression and values less than 20 were coded as 20. Tissue selectivity was determined by averaging probe expression values across samples and replicate tissues. In total, 61 samples were collapsed into 28 tissues. A probe was classified as tissue selective if it was expressed in only one tissue at a value of 200 or higher, and all other tissues were less than 100. Probes were then collapsed into genes. A gene was classified as tissue selective if at least one of its probes showed specificity. The tissue of maximal expression was determined by identifying the probe and sample (n = 61) with the highest expression value that was greater than 20 (85% of the genes had values greater than 200). Probes were then collapsed into genes. Tissue expression was determined by averaging the sample replicates (n = 28). A gene was considered expressed in a tissue if its expression value was greater than 200. ### PAML analysis. {#s4a5} To obtain orthologous sequences for the 50 annotated genes with the highest likelihood ratios, we downloaded "Unique Best Reciprocal Hits" between human and mouse and human and rat from the Ensembl Web site (<http://www.ensembl.org/>). Sets of human, mouse, rat, and chimpanzee sequences were translated and aligned using ClustalW \[[@pbio-0030170-b49]\]. Codon alignments were generated using the ClustalW alignments as a guide, then manually checked. Partial sequences covering less than 80% of the human sequence were eliminated, and ambiguously aligned regions were masked before analysis. The underlying phylogeny was assumed to be (\[chimpanzee, human\], \[mouse, rat\]) for all genes. The lineage-specific analysis was done in PAML \[[@pbio-0030170-b37]\] by allowing two values of the d~N~/d~S~ ratio along the lineages of the phylogeny, one for the human lineage and one for all other lineages. To test if the d~N~/d~S~ ratio was different on the human lineage, we then compared the maximum likelihood value in this model to the maximum likelihood value obtained, assuming the d~N~/d~S~ ratio was constant among lineages. If two times the log likelihood ratio was larger than 3.84, we rejected the model of constant d~N~/d~S~ ratio at the 5% significance level. This analysis was then repeated using the chimpanzee lineage as the focal lineage instead of the human lineage. ### Calculating the frequency spectrum. {#s4a6} Because of missing data for many polymorphisms, the frequency spectrum in a sample of size 30 is reported. The frequency spectrum was calculated in a sample of size 30 as where p~i,~ ~30~ is the frequency of SNPs with derived alleles that exist in i copies in a sample of size 30, n~j~ is the chromosomal sample size of the jth SNP, f~j~ is the frequency of the derived allele for the jth SNP, k is the number of SNPs, and (^i^ ~j~) = 0 = 0 \--\> if i is less than j. The polarity of the mutation was determined using the chimpanzee sequence as outgroup. ### Analysis of ascertainment bias. {#s4a7} To assess the impact of the ascertainment scheme in the tests that contrast human polymorphism data to the human--chimp divergence, new datasets were simulated, using standard neutral coalescence simulations (e.g., \[[@pbio-0030170-b38]\]). Each simulated dataset generated one chimp sequence and 78 human sequences for each of the 13,731 genes. For each simulated gene, one human sequence was randomly chosen and compared to the chimp sequence using a chi-square statistic for the goodness-of-fit test of d~N~/d~S~ = 1. The 50 genes with largest chi-square statistic among genes with d~N~/d~S~ greater than one were selected for population genetic analysis. This scheme was repeated 1,000 times to investigate the effect of the ascertainment protocol of the 50 genes. The parameters of the simulations were estimated from the data, using the observed distribution of sequence lengths, and synonymous-site mutation rate and humans--chimp divergence time estimated from the concatenated data. The distribution of d~N~/d~S~ ratios among genes was estimated assuming the d~N~/d~S~ ratios follow a γ distribution among genes, keeping the synonymous rate constant among them. ### Power analysis. {#s4a8} To analyze the power of the test for positive selection, we simulated pairs of sequences and performed likelihood ratio tests of H~0~: d~N~/d~S~ equals one versus d~N~/d~S~ is greater than one for each sequence pair. The simulations were done using the average value of synonymous sequence divergence observed in the data, while nonsynonymous divergence was varied. For more details regarding such simulations, see, e.g. \[50\]. ### PRF analysis. {#s4a9} Assume nonlethal mutations enter a population of constant size 2N according to a Poisson process and are assigned to one of three categories: neutral (S = 0), positively selected with selection coefficient S ~+~, and negatively selected with selection coefficient S ~--~, according to probabilities p ~0~, p ~+~, and p ~--~ (where p ~0~ + p ~+~ + p ~--~ = 1). Furthermore, assume mutations evolve independently. It follows from standard population genetic theory, the total law of probability, and the rules of conditional probability that the probability of an SNP being found at frequency i out of n chromosomes under this scheme \[[@pbio-0030170-b44]\] is where F(i,n,S) \--\> is given by The likelihood of observing counts x ~1~, x ~2~, . . ., x~S~ where S is the total number of segregating sites out of n ~1~, n ~2~, ..., n~S~ chromosomes is, thus, The maximum likelihood value and the maximum likelihood parameter estimates can then be obtained by numerically maximizing this function with respect to the parameters. Likelihood ratio tests can be constructed by constraining certain of the parameters to take on particular values. For example, setting p ~0~ = 1 defines a model with no selected mutations. Likewise, setting p ~0~ + p ~--~ = 1 defines a model that allows negative selection, but no positive selection. This analysis assumes that mutations are independent. Because of linkage and the possibility of epistasis, the independence assumption may not be met by the data. However, a full analysis taking the correlation among SNPs into account is not computationally feasible. Fortunately, the average correlation is low between SNPs because they have been sampled among 50 genes distributed throughout the genome. The effect of the correlation among SNPs on this analysis should, therefore, be minimal. The maximum log likelihood value for the full model is --234.19. However, the maximum log likelihood values for models assuming only neutral mutations, or a single class of selected mutations, are --243.82 and --240.88, respectively. Under the independence assumption, both of these simpler models can be rejected against the model with three classes of mutations, using a likelihood ratio test (p = 0.0006 and p = 0.004). Supporting Information {#s5} ====================== Dataset S1 ::: {.caption} ###### Results File (3.1 MB XLS). ::: ::: {.caption} ###### Click here for additional data file. ::: Dataset S2 ::: {.caption} ###### Alignment File (9.8 MB ZIP). ::: ::: {.caption} ###### Click here for additional data file. ::: Accession Numbers {#s5a3} ----------------- The sequence analyzed in this study has been submitted to GenBank (<http://www.ncbi.nlm.nih.gov/Genbank/>). The data from this paper were obtained from more than 18 million sequencing reads obtained from the Celera Genomics sequencing center in Rockville, Maryland. We especially acknowledge the technical contributions of J. Duff, C. Evans, S. Ferriera, C. Forbes, C. Gire, B. Murphy, M. A. Rydland, B. Small, and G. Wang. Competing interests. The authors have declared that no competing interests exist. Author contributions. SG, DMT, DC, TJW, JJS, MDA, and MC conceived and designed the experiments and performed the experiments. RN, CB, AGC, TBS, MJH, and AFA analyzed the data and contributed reagents/materials/analysis tools. RN, CB, AGC, MDA, and MC wrote the paper. ¤ Current address: Department of Genetics, Case Western Reserve University, Cleveland, Ohio, United States of America Citation: Nielsen R, Bustamante C, Clark AG, Glanowski S, Sackton TB, et al. (2005) A scan for positively selected genes in the genomes of humans and chimpanzees. PLoS Biol 3(6): e170. Note Added in Proof The version of this paper that was first made available on 3 May 2005 has been replaced by this, the definitive, version. (d~N~/d~S~) : ratio of nonsynonymous substitutions per nonsynonymous site to synonymous substitutions per synonymous site MWU : Mann-Whitney U test PRF : Poisson random field SNP : single nucleotide polymorphism	PubMed Central	2024-06-05T03:55:56.292728	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088278/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e170", "authors": [ { "first": "Rasmus", "last": "Nielsen" }, { "first": "Carlos", "last": "Bustamante" }, { "first": "Andrew G", "last": "Clark" }, { "first": "Stephen", "last": "Glanowski" }, { "first": "Timothy B", "last": "Sackton" }, { "first": "Melissa J", "last": "Hubisz" }, { "first": "Adi", "last": "Fledel-Alon" }, { "first": "David M", "last": "Tanenbaum" }, { "first": "Daniel", "last": "Civello" }, { "first": "Thomas J", "last": "White" }, { "first": "John", "last": "J. Sninsky" }, { "first": "Mark D", "last": "Adams" }, { "first": "Michele", "last": "Cargill" } ] }
PMC1088279	Introduction {#s1} ============ Many clinical conditions, including aspiration, trauma, and hemorrhagic shock, are frequently followed by pulmonary and systemic infectious and septic complications that lead to pulmonary dysfunction and subsequent lung failure. Acute lung injury or its more severe form, the acute respiratory distress syndrome (ARDS), occur with a frequency of approximately 130.000 cases and more than 50.000 deaths from ARDS per year in the United States alone \[[@pbio-0030174-b01]\]. Intubation with mechanical ventilation represent one of the most widely used prophylactic and therapeutic clinical interventions to counteract the insufficient pulmonary oxygen-delivering capacity in patients who suffer from severe lung inflammation. Although the majority of patients respond well to oxygen therapy, and oxygen toxicity is an uncommon occurrence in intensive care medicine \[[@pbio-0030174-b02]\], there remains the possibility that oxygen therapy may be suboptimal in ARDS patients, as it may promote deleterious pulmonary inflammation, which fuels this disease process. Since the magnitude and duration of lung inflammation has been shown to determine the final outcome of ARDS patients \[[@pbio-0030174-b03]\], it is important to carefully evaluate the possible adverse effects of oxygen on inflammatory processes. We assumed that lung tissues are protected from overactive immune cells by the same hypoxia-driven mechanism and immunosuppressive adenosine A~2A~ receptor (A~2A~R)-mediated mechanism that was recently shown to play a critical role in the down-regulation of inflammation and tissue damage in different models \[[@pbio-0030174-b04]--[@pbio-0030174-b07]\]. Accordingly, it is likely that bacterial toxin-activated immune cells (e.g., granulocytes) cause collateral lung tissue damage with impairment of the local microcirculation and blood supply, thereby contributing to the pathogenesis of acute lung injury. The ensuing tissue damage-associated hypoxia facilitates the accumulation of extracellular adenosine \[8--12, which then triggers the activation of immunosuppressive A~2A~Rs on activated immune cells and causes the accumulation of immunosuppressive intracellular cyclic adenosine 3′,5′-monophosphate (cAMP). This cAMP, in turn, inhibits signaling pathways that are required for synthesis and secretion of proinflammatory and cytotoxic mediators by immune cells, and thereby protects remaining healthy tissues from continuing immune damage. Since this physiological tissue-protecting mechanism depends on the hypoxia-produced extracellular adenosine \[8--12, and since the oxygenation of lungs in intubated patients is performed to increase oxygen tension---with the goal of abolishing the hypoxia but disrupting the adenosine accumulation---we reasoned that such interruption of the hypoxia → adenosine → A~2A~R pathway by oxygenation could lead to a disengagement of the critical tissue-protecting mechanism and to an unintended exaggeration of inflammatory lung damage (iatrogenic disease). Thus, oxygenation may eliminate this lung-protecting pathway and thereby contribute to pulmonary complications. We confirm our prediction in several in-vivo models of lung infection and inflammation and report that oxygenation does indeed strongly exacerbate inflammatory lung damage and accelerate death in mice by the interruption of the hypoxia → adenosine → A~2A~R pathway. We also suggest an effective and feasible therapeutic countermeasure to prevent deleterious effects of oxygenation: Exogenously added synthetic A~2A~R agonist compensated for the loss of endogenously formed adenosine in inflamed lungs of oxygenated mice, and thereby prevented inflammatory lung injury and prevented death. Results {#s2} ======= Exacerbation of Inflammatory Lung Injury and Death in Oxygenated Mice {#s2a} --------------------------------------------------------------------- To test our prediction, we subjected mice to inhalation of combined toxins from gram-positive and gram-negative bacteria. In this model of polymicrobial lung infection, intratracheal (IT) injection of both lipopolysaccharide (LPS) and staphylococcal enterotoxin B (SEB) strongly potentiates their toxicity \[[@pbio-0030174-b13]\]. The results of these assays confirmed the prediction of exaggerated lung injury in mice in conditions that mimic therapeutic oxygenation, and this is reflected in the dramatic increases in inflammatory lung damage in different in vivo and ex vivo assays ([Figures 1](#pbio-0030174-g001){ref-type="fig"} and [2](#pbio-0030174-g002){ref-type="fig"}). ::: {#pbio-0030174-g001 .fig} Figure 1 ::: {.caption} ###### Increased Death Rate upon Oxygenation of Mice with Acute Lung Injury Mice were IT injected with SEB and LPS to model polymicrobial infection and were exposed to 21% or 100% oxygen for 48--60 h. Determination of time-dependent survival curves was prohibited by considerations of unrelieved severe respiratory distress in NIH-approved animal care protocol, which required termination of experiments immediately after differences between groups became apparent. Major differences between groups occurred within 48--60 hours after IT injection of toxins, when the majority of oxygenated animals with inflamed lungs had died, while the nonoxygenated, obviously sick control mice with inflamed lungs were still alive. ::: ![](pbio.0030174.g001) ::: ::: {#pbio-0030174-g002 .fig} Figure 2 ::: {.caption} ###### Exacerbation of Inflammatory Lung Injury after Exposure of Mice to Different Concentrations of Oxygen \(A) Enhanced lung vascular permeability (left graph) and impairment of lung gas exchange (right graph) in mice breathing 100% O~2~ upon induction of acute lung injury. Following IT injection of mice with SEB and LPS, animals breathed 21% or 100% oxygen. After 48 h, lung vascular permeability and lung gas exchange were determined by the amount of protein recovered by BAL or by measuring pO~2~ values in arterial blood drawn, respectively, 15 min after return of mice to normal atmosphere. During this equilibration period, three out of seven mice previously exposed to 100% oxygen died, so that no arterial blood gas analyses could be performed, but BAL protein concentrations were determined immediately thereafter. \(B) Increased lung vascular permeability (left graph) and impairment of lung gas exchange (right graph) in mice with acute lung injury even upon exposure to lower levels of oxygen (60%), which are considered clinically safe in humans. Experimental conditions were the same as in (A), except oxygen concentration was 60% instead of 100%. ::: ![](pbio.0030174.g002) ::: Five times more mice with inflamed lungs died after exposure to 100% oxygen than those left at 21% ambient oxygen tension ([Figure 1](#pbio-0030174-g001){ref-type="fig"}). This was further confirmed by a much more pronounced increase in the alveolocapillary permeability and severe overall impairment of lung gas exchange, as evidenced by the increase in the amount of protein recovered from the alveolar space by bronchoalveolar lavage (BAL), as well as by the decrease in arterial oxygen partial pressure (pO~2~) values of previously oxygen-exposed mice when returned to normal atmosphere ([Figure 2](#pbio-0030174-g002){ref-type="fig"}A). Although exposure of mice to 100% oxygen alone (with no toxin inhalation) did induce a small accumulation of BAL fluid protein, the magnitude of that effect could not account for the dramatic increases in lung vascular permeability and impairment of lung gas exchange when both toxins and oxygenation were used ([Figure 2](#pbio-0030174-g002){ref-type="fig"}A). The exacerbation of inflammatory lung injury was also observed when mice were exposed to 60% oxygen, a concentration considered in human patients to be fairly safe. Compared to toxin-injected animals breathing 21% oxygen, those breathing 60% oxygen accumulated much more exudate proteins in the alveolar spaces and exhibited impaired arterial blood oxygen tensions ([Figure 2](#pbio-0030174-g002){ref-type="fig"}B). Exposure of toxin-injected mice to 60% oxygen, however, did not result in death in the short-term assays we used. These observations confirmed the prediction that inflamed lung injury is exacerbated by oxygenation; however, the development of therapeutic countermeasures requires testing of the validity of our underlying assumptions and conclusive identification of the molecular mechanisms of these proinflammatory effects of oxygen. Oxygenation Disrupts the Hypoxia → A~2A~R-Mediated Pathway {#s2b} ---------------------------------------------------------- The observed effects of oxygenation could not be accounted for by direct toxic effects of oxygen, since these effects of oxygen take much longer to manifest \[[@pbio-0030174-b02], [@pbio-0030174-b14], [@pbio-0030174-b15]\] and therefore are unlikely to fully account for the dramatic lung injury observed in our short-term experiments ([Figures 1](#pbio-0030174-g001){ref-type="fig"} and [2](#pbio-0030174-g002){ref-type="fig"}). To test whether oxygen enhances inflammatory tissue damage by eliminating the hypoxia → adenosine → A~2A~R pathway, which is hypothesized as being responsible for protecting inflamed lungs and down-regulating neutrophils, we (i) determined the role of neutrophils in inflammatory lung injury, (ii) tested for A~2A~R expression on these cells to demonstrate their susceptibility to modulation by endogenously produced adenosine, (iii) analyzed the degree of inflammatory lung tissue damage in A~2A~R gene-deficient mice or in wild-type mice treated with highly selective pharmacologic antagonists for A~2A~Rs, and (iv) tested whether the hypoxia has a lung-protective role and whether there is a direct linkage between hypoxia (upstream) and A~2A~Rs (downstream) in the proposed lung-protective hypoxia → A~2A~R pathway. To diminish the suffering of animals with severely inflamed and damaged lungs, in the experiments described in the next section we used the less-severe lung injury model established by the injection of LPS alone instead of the polymicrobial (SEB + LPS) model of lung damage. Neutrophils Are Involved in Inflammatory Lung Injury and Are Inhibited by A~2A~Rs {#s2c} --------------------------------------------------------------------------------- In agreement with an important pathogenic role for granulocytes in acute lung injury \[[@pbio-0030174-b16], [@pbio-0030174-b17]\], we confirmed that inflammatory lung injury elicited by IT injection of LPS in wild-type mice is in large part mediated by granulocytes, because the depletion of polymorphonuclear leukocytes (PMNs) with the anti-Gr-1 monoclonal antibody resulted not only in a substantial decrease in the number of cells in the BAL fluid but also in a much less pronounced endotoxin-induced lung injury ([Figure 3](#pbio-0030174-g003){ref-type="fig"}A). This finding was further supported by the correlation between the number of PMNs and the degree of lung vascular permeability as reflected by the amount of protein recovered from the alveolar space by BAL ([Figure 3](#pbio-0030174-g003){ref-type="fig"}B). ::: {#pbio-0030174-g003 .fig} Figure 3 ::: {.caption} ###### Role of PMNs in the Pathogenesis of LPS-Induced Lung Injury \(A) Depletion of granulocytes attenuates the endotoxin induced rise in alveolocapillary permeability. Pretreatment of mice with anti-Gr-1 was followed by a significant decrease in the number of granulocytes (left graph) and a significant reduction of the total amount of protein (right graph) recovered by BAL 48 h after IT LPS injection. \(B) The more granulocytes immigrated into the alveolar spaces, the higher the alveolocapillary permeability rose. Bivariate analysis according to Pearson revealed a statistically significant correlation (p \< 0,001) between the number of PMNs and the amount of protein in the BAL fluid 48 h after IT LPS injection, suggesting that inflammatory lung injury after IT injection of LPS is mostly mediated by granulocytes. ::: ![](pbio.0030174.g003) ::: To answer the question of whether murine neutrophils are susceptible to inhibitory effects of endogenously produced adenosine via A~2A~R signaling, we characterized the effects of increasing concentrations of the highly specific A~2A~R agonist CGS21680 on the oxygen radical production of granulocytes taken from the blood of healthy mice. Surprisingly, respiratory burst activity of such "naïve" granulocytes was only poorly inhibited by A~2A~R signaling ([Figure 4](#pbio-0030174-g004){ref-type="fig"}A, left graph). Therefore it was important to test whether "in vivo-activated" granulocytes isolated from inflamed lungs of IT LPS-injected mice might have up-regulated their A~2A~R expression. As expected, significant up-regulation of A~2A~R was observed in experiments in which CGS21680 greatly enhanced the chemotactic peptide N-formyl-methionyl-leucyl-phenylalanine (fMLP)-stimulated cAMP production of in vivo-activated neutrophils isolated from inflamed lungs from wild-type mice, but not of naïve granulocytes obtained from the bone marrow of healthy wild-type control mice ([Figure 4](#pbio-0030174-g004){ref-type="fig"}B, left graph). In a genetic control, the cAMP increase was absent in in vivo-activated granulocytes from A~2A~R gene-deficient mice ([Figure 4](#pbio-0030174-g004){ref-type="fig"}B, right graph). In addition, the adenosine-induced cAMP increase was inhibited in wild-type granulocytes by the selective antagonist of A~2A~Rs ZM241385 ([Figure 4](#pbio-0030174-g004){ref-type="fig"}B, right graph). Results of parallel assays confirmed increased expression of A~2A~R on activated granulocytes, since CGS21680 inhibited the functional response of in vivo-activated granulocytes (i.e., as evidenced by the production of tissue-damaging reactive oxygen species) much more strongly than in naïve cells ([Figure 4](#pbio-0030174-g004){ref-type="fig"}A, right graph). ::: {#pbio-0030174-g004 .fig} Figure 4 ::: {.caption} ###### Evidence for the Up-Regulation of Immunosuppressive A~2A~R Expression on In Vivo-Activated Granulocytes Isolated from Inflamed Lungs \(A) The selective A~2A~R agonist CGS21680 inhibited the fMLP-stimulated hydrogen peroxide production by granulocytes in blood of healthy mice to only a small degree, reflecting low levels of expression of A~2A~R on naïve blood granulocytes. In contrast, granulocytes recovered by BAL from inflamed lungs 48 h after IT LPS injection were much more inhibited by CGS21680, demonstrating functional up-regulation of A~2A~R on in vivo-activated cells. \(B) CGS21680 induces cAMP accumulation in in vivo-activated granulocytes isolated from lungs 48 h after IT LPS injection. No effects of the A~2A~R agonist were observed in naïve granulocytes obtained from bone marrow of healthy mice (left graph) or in in vivo-activated granulocytes recovered from inflamed lungs of A~2A~R gene-deficient mice (right graph). The CGS21680-stimulated cAMP production observed in lung granulocytes obtained from wild-type mice could also be antagonized by the selective A~2A~R antagonist ZM241385. Naïve bone marrow granulocytes were used for cAMP measurements, since it was impossible to isolate naïve cells from blood of healthy mice in sufficient numbers. \(C) Higher levels of A~2A~R-specific mRNA in in vivo-activated granulocytes. In parallel with the much stronger A~2A~R agonist-induced inhibition of hydrogen peroxide production and accumulation of cAMP in in vivo-activated granulocytes, the relative levels of A~2A~R-specific mRNA were much higher in in vivo-activated granulocytes obtained from inflamed lungs 48 h after IT LPS injection, as compared with naïve granulocytes isolated from the bone marrow of healthy mice (left graph). Up-regulation of A~2A~R mRNA in in vivo-activated granulocytes was confirmed in another set of experimental animals breathing 21% oxygen, but was increased to a much lesser extent in animals subjected to 100% oxygen (right graph). Levels of A~1~R mRNA did not change much in inflammatory lung granulocytes from animals breathing normal atmosphere, but were clearly increased in mice exposed to 100% O~2~. In the two sets of experiments (left and right graphs), granulocytes were pooled from five and six mice per treatment, respectively. Taken together, the results demonstrate that granulocytes recovered from alveolar spaces of inflamed lungs did, indeed, up-regulate their A~2A~R expression during these in vivo lung injury assays, thereby confirming and extending previous findings in other inflammation models \[[@pbio-0030174-b11], [@pbio-0030174-b53]\] ::: ![](pbio.0030174.g004) ::: The results of mRNA expression studies were in agreement with above functional studies of adenosine receptors, and a much stronger increase in A~2A~R mRNA levels was observed in in vivo-activated PMNs than in naïve cells, with only small changes in A~1~ receptor mRNA expression in the first experiment ([Figure 4](#pbio-0030174-g004){ref-type="fig"}C, left graph). These findings were further confirmed in a second experiment in which the effect of 100% oxygen was compared to that of normal atmosphere ([Figure 4](#pbio-0030174-g004){ref-type="fig"}C, right graph). When compared to the several-fold increase in A~2A~R mRNA expression levels in inflammatory granulocytes from mice exposed to 21% oxygen, these levels were up-regulated to a much lesser extent in mice breathing 100% oxygen. As in the first experiment, no changes were observed in A~3~ mRNA expression, while expression of A~2B~ receptors showed changes similar to those of A~2A~R mRNA (unpublished data). Interestingly, the expression of A~1~ mRNA levels was increased in inflammatory lung granulocytes obtained from hyperoxic animals. Thus, the decreased up-regulation of A~2A~R-specific mRNA that was observed in parallel with an increased A~1~ receptor-specific mRNA expression in inflammatory granulocytes from mice breathing pure oxygen is in support of hyperoxic exacerbation of lung injury. A~2A~Rs Protect Lung Tissue from Inflammatory Damage {#s2d} ---------------------------------------------------- The genetic evidence for the critical role of A~2A~R in lung protection was provided by the observation of many more PMNs in BAL from A~2A~R-deficient mice than in BAL from similarly treated wild-type mice ([Figure 5](#pbio-0030174-g005){ref-type="fig"}A). This was accompanied by an increase in lung vascular permeability as reflected by enhanced BAL protein levels in A~2A~R-deficient mice and a decrease in overall lung function, which was manifested by a decrease in arterial blood oxygen tension as compared to wild-type mice. ::: {#pbio-0030174-g005 .fig} Figure 5 ::: {.caption} ###### Evidence for the Critical Role of Immunosuppressive A~2A~R in Lung Protection from Inflammatory Damage \(A) In A~2A~R gene-deficient mice, number of PMNs (left graph) and amount of protein recovered (center graph) 48 h after IT LPS injection by BAL was significantly higher than in similarly treated wild-type control mice, reflecting increased lung damage in the absence of A~2A~R. The arterial oxygen tension (right graph) was lower in A~2A~R gene-deficient mice as compared with wild-type mice. \(B) Pharmacologic inactivation of A~2A~R leads to exacerbated inflammatory lung tissue damage and decreased lung funciton. After estimation of biologically relevant half-life of A~2A~R antagonist ZM241385 (ZM) in vivo (unpublished data), the IT LPS-injected mice were administered ZM241385 at a dose of 10 mg/kg body weight every 6 h subcutaneously to ensure sufficient levels of the antagonist. This dosing regimen of the A~2A~R antagonist caused significant more lung tissue damage, as reflected by increased number of PMNs (left graph) and protein levels (center graph) in the BAL fluid obtained after 48 h. In parallel experiments, the A~2A~R antagonist decreased lung function (right graph) as compared to untreated wild-type mice, in agreement with results of experiments with A~2A~R gene-deficient mice. ::: ![](pbio.0030174.g005) ::: In agreement with the genetic evidence in [Figure 5](#pbio-0030174-g005){ref-type="fig"}A, similar proinflammatory changes in numbers of granulocytes, levels of protein, and values of lung gas exchange were observed after pharmacological inactivation of A~2A~R with the antagonist ZM241385 ([Figure 5](#pbio-0030174-g005){ref-type="fig"}B). These observations establish that the A~2A~R is critical in limiting inflammatory lung injury; even higher lung injury would have resulted from the inflammatory stimuli we used, were it not for the lung protection in wild-type mice due to the inhibition of neutrophils by a functional hypoxia → A~2A~R pathway. Hypoxia Down-Regulates Neutrophils and Protects Lung Tissue from Inflammatory Damage {#s2e} ------------------------------------------------------------------------------------ To test whether anti-inflammatory lung-protective effects can be induced in mice by allowing them to breathe a hypoxic gas mixture, LPS-injected wild-type mice were exposed to 10% oxygen. Although this oxygen concentration was sublethal for LPS-injected mice, we chose to study the effects of this degree of hypoxia because 15% of ARDS patients die from therapy-refractory hypoxemia \[[@pbio-0030174-b18]\]. Accordingly, some deaths occurred in IT LPS-injected mice at hour 4--7 of hypoxic exposure. In the majority of surviving mice (over 90%), however, hypoxia for 48 h strongly inhibited acute neutrophilic inflammation and led to overall better lung protection compared to mice kept at 21% oxygen. This improvement was evidenced by a decreased pulmonary sequestration of PMNs ([Figure 6](#pbio-0030174-g006){ref-type="fig"}), inhibition of their capability to produce oxygen-reactive metabolites ([Figure 6](#pbio-0030174-g006){ref-type="fig"}A), decreased protein accumulation ([Figure 6](#pbio-0030174-g006){ref-type="fig"}B), and better lung function as reflected in higher arterial pO~2~ levels ([Figure 6](#pbio-0030174-g006){ref-type="fig"}A). This conclusion was further confirmed by histological studies ([Figure 6](#pbio-0030174-g006){ref-type="fig"}C) that revealed not only a decrease in PMN sequestration, but also a significant decrease in parameters of lung tissue damage as evidenced by the 7-fold reduction of the lung injury score (LI; [Figure 6](#pbio-0030174-g006){ref-type="fig"}C). ::: {#pbio-0030174-g006 .fig} Figure 6 ::: {.caption} ###### Hypoxia Down-Regulates Neutrophils and Protects Lung Tissue from Inflammatory Damage \(A) Exposure of IT LPS-injected mice to hypoxic (10%) oxygen levels for 48 h atmosphere leads to a significantly decreased accumulation of PMNs (left graph), production of LPS-triggered oxygen reactive metabolites in lungs (center graph), and improved lung gas exchange (right graph) compared to a control group of endotoxin-treated mice that were kept at ambient (21%) oxygen. To standardize conditions, the arterial blood samples were taken 15 min after return of the previously hypoxia-exposed animals to normal atmosphere. \(B) Treatment by a shorter period of hypoxia attenuates PMN sequestration (left graph) and lung vascular permeability (right graph). Hypoxic treatment of mice even for only 24 h was sufficient to delay PMN sequestration and to diminish the increase in lung vascular permeability. \(C) Histologic evidence for the hypoxic inhibition of pulmonary PMN sequestration. Quantitative analysis of lung slices by a pathologist blinded to the experimental design revealed inhibition of PMN sequestration in IT LPS-injected mice following 4-h exposure to hypoxia. Hypoxia not only attenuated PMN accumulation, but the lung tissue damage was also significantly decreased as assessed by the LIS (n = 9, mean ± standard deviation). The representative H&E-stained slices in the right two photomicrographs show less intravascular granulocyte sequestration, less thickening of the alveolocapillary membrane, and almost no granulocytes in the alveolar spaces as compared to IT endotoxin-injected animals breathing 21% O~2~. These observations demonstrate that hypoxia also inhibited the transmigration of granulocytes from capillaries into the alveolar spaces. ::: ![](pbio.0030174.g006) ::: Taken together, these data strongly suggest that tissue hypoxia is important in protecting pulmonary tissues from additional inflammatory damage. Hypoxia and A~2A~R-Triggered Signaling Function in the Same Lung Tissue-Protecting Pathway {#s2f} ------------------------------------------------------------------------------------------ It was important to establish whether inflamed and hypoxic lung tissues are protected by enhanced adenosine formation and subsequent A~2A~R engagement or by an anti-inflammatory role of a yet-to-be-uncovered hypoxia → unknown endogenous anti-inflammatory molecule "X"-mediated pathway. To distinguish between these two possibilities we tested the effects of genetic deficiency of the A~2A~R on inflammatory lung injury in hypoxic conditions. Exposure of healthy wild-type mice to 10% oxygen resulted in a drop of arterial blood pO~2~ values to levels observed during severe hypoxemia ([Figure 7](#pbio-0030174-g007){ref-type="fig"}A, left graph). As a result of insufficient systemic oxygen delivery and resulting hypoxemia, breathing of 10% oxygen caused an increase in production of endogenous adenosine, shown by the rise of extracellular plasma concentrations of the nucleoside ([Figure 7](#pbio-0030174-g007){ref-type="fig"}A, right graph). ::: {#pbio-0030174-g007 .fig} Figure 7 ::: {.caption} ###### Hypoxia and Extracellular Adenosine A~2A~R Function in the Same Anti-Inflammatory, Lung Tissue-Protecting Pathway \(A) Effects of breathing hypoxic (10%) oxygen on arterial blood oxygen tension (left graph) and plasma adenosine concentration (right graph) in healthy wild-type mice. As a control, data are also shown for healthy mice breathing 21% and 100% oxygen. \(B) No survival of A~2A~R gene-deficient mice was observed in acute hypoxic lung injury. Wild-type and A~2A~R gene-deficient mice were injected IT with LPS and exposed to hypoxia (10%). While the majority of wild-type mice survived, all of the A~2A~R gene-deficient mice died, indicating that expression of A~2A~R is required for survival of hypoxic lung inflammation; this experiment mimics the clinical situation in which lung inflammation increases to such severity that hypoxia occurs. \(C) Significantly higher levels of pulmonary and systemic inflammatory cytokine production in hypoxic A~2A~R-deficient mice. Observations of survival were supported by significantly higher BAL and serum (Se) levels of inflammatory cytokines in hypoxic A~2A~R-deficient mice compared to hypoxic wild-type mice. Cytokines were determined 2 h after IT LPS injection, because A~2A~R-deficient mice started to die soon after LPS administration and thus could not be used in comparative studies with wild-type control mice. The early mortality of A~2A~R-deficient mice also did not allow the comparative determination of effects of hypoxia on other late markers of inflammation such as PMN accumulation, lung vascular permeability, and pulmonary gas exchange, which in wild-type mice need about 48 h to develop after IT endotoxin injection. \(D) Degree of inflammation is independent from level of oxygen in A~2A~R-deficient mice but not in wild-type mice. While BAL fluid TNF-α concentration determined 2 h after IT LPS injection was significantly suppressed in hypoxic wild-type mice compared to animals breathing 100% oxygen, hypoxia had no effect on TNF-α BAL concentrations in A~2A~R gene-deficient mice, demonstrating that suppression of TNF-α formation by hypoxia is mediated through A~2A~R signaling. ::: ![](pbio.0030174.g007) ::: Hypoxic (10%) oxygen was also expected to provide an important readout of the degree of lung injury by the level of survival in the mice. In our experiments, the dose of LPS injected IT induced much higher levels of lung injury in A~2A~R-deficient mice than in control wild-type mice (see [Figure 5](#pbio-0030174-g005){ref-type="fig"}), but neither wild-type or A~2A~R deficient mice died if they were kept at 21% oxygen at these levels of lung injury. Different outcome with respect to death rates were expected, if A~2A~R-deficient and wild-type mice with LPS-inflamed lungs were kept at 10% oxygen. We reasoned that under these conditions the A~2A~R-deficient mice would have much more pronounced pulmonary inflammation and hence less healthy lung tissues left to adequately oxygenate vital organs. Thus, the interruption of the hypoxia → A~2A~R pathway and the resulting uninhibited inflammatory processes and increased lung damage would lead to accelerated death of A~2A~R-deficient mice in 10% oxygen. In contrast, these treatments would not result in pronounced lethality in lung-inflamed wild-type mice at 10% oxygen tension, because wild-type mice benefit from their lung-protecting hypoxia → A~2A~R pathway and would have much larger portions of still-healthy lungs left to ensure sufficient oxygen supply to tissues. Accordingly, we predicted that the majority of A~2A~R genetically deficient mice would die at 10% oxygen, while the majority of wild-type mice would survive. This prediction was based on the assumption that the hypoxia → A~2A~R pathway is nonredundant and would not be substituted by another hypoxia → "X" molecule receptor pathway. These expectations were confirmed by the data showing that exposure of the IT LPS-injected wild-type mice to hypoxia resulted in low mortality (only 20% \[4 out of 20\] of the mice died), whereas in the parallel group all (20 out of 20) similarly treated A~2A~R-deficient mice died ([Figure 7](#pbio-0030174-g007){ref-type="fig"}B). Control experiments with healthy, non-LPS treated, wild-type and A~2A~R-deficient mice subjected to hypoxia demonstrated that deficiency in the A~2A~R did not cause death per se, because all mice survived (unpublished data). These observations of survival were further supported by measurements of BAL and plasma levels of the cytokines tumor necrosis factor α (TNF-α) and interleukin 6 (IL-6), and the chemokine major intrinsic protein 2α (MIP-2α), all of which are considered to be early mediators of lethal endotoxic shock \[[@pbio-0030174-b19], [@pbio-0030174-b20]\]. Accordingly, the outcome of survival experiments ([Figure 7](#pbio-0030174-g007){ref-type="fig"}B) was in agreement with that from higher concentrations of TNF-α, IL-6, and MIP-2α in hypoxic (10% oxygen) A~2A~R-deficient mice as compared to hypoxic wild-type mice 2 h after IT LPS injection ([Figure 7](#pbio-0030174-g007){ref-type="fig"}C). No time points after 2 h could be compared in wild-type versus A~2A~R-deficient mice, because A~2A~R knockout mice started to die soon after IT LPS-injection ([Figure 7](#pbio-0030174-g007){ref-type="fig"}B). Interestingly, the increase in hypoxia A~2A~R-controlled TNF-α levels was accounted for by local accumulation in the lung, because BAL concentrations of this cytokine were on average ten times higher than in the systemic circulation, although the BAL volume was at least three times higher than that of a mouse\'s blood. By contrast, levels of IL-6 and MIP-2α were much higher in the systemic circulation, suggesting that TNF-α BAL levels may serve as a useful marker for lung inflammation ([Figure 7](#pbio-0030174-g007){ref-type="fig"}C). Moreover, the results shown in [Figure 7](#pbio-0030174-g007){ref-type="fig"}D support the view that pulmonary proinflammatory cytokine TNF-α levels are indeed under the negative control of the hypoxia → A~2A~R pathway. This is supported by data showing no differences in BAL TNF-α concentrations between A~2A~R gene-deficient mice breathing either 10% or 100% oxygen. In contrast, the BAL TNF-α concentrations were significantly lower in hypoxic wild-type mice (with functioning hypoxia → adenosine → A~2A~R pathway) than in mice breathing pure oxygen ([Figure 7](#pbio-0030174-g007){ref-type="fig"}D). Taken together, these experiments support the view that both hypoxia and A~2A~R are needed to down-regulate lung inflammation, and that oxygenation exacerbates lung injury due to interruption of the tissue-protecting hypoxia → A~2A~R anti-inflammatory pathway. Thus, these observations leave no room for yet another lung tissue-protecting hypoxia → X receptor signaling pathway in tested experimental conditions. This knowledge suggested a direct and effective therapeutic countermeasure to reap benefits of oxygenation without sacrificing still-healthy lung tissues to continuing inflammatory damage: use of the A~2A~R agonist CGS21680. An A~2A~R Agonist Compensates for the Loss of Lung-Protective Mechanisms and Prevents Death {#s2g} ------------------------------------------------------------------------------------------- [Figure 8](#pbio-0030174-g008){ref-type="fig"} shows that IT injections of the selective A~2A~R agonist CGS21680 significantly inhibited lung injury in LPS-treated mice. This treatment led to (i) significantly decreased accumulation of PMNs ([Figure 8](#pbio-0030174-g008){ref-type="fig"}A), (ii) reduced production of reactive oxygen metabolites ([Figure 8](#pbio-0030174-g008){ref-type="fig"}A), (iii) less-pronounced increases in lung vascular permeability ([Figure 8](#pbio-0030174-g008){ref-type="fig"}B), and (iv) improved lung gas exchange ([Figure 8](#pbio-0030174-g008){ref-type="fig"}B). Histological examination of CGS21680-treated mice revealed that therapeutic effects of the agonist were similar to those of exposure of mice to hypoxia. CGS21680 treatment resulted in inhibition of pulmonary PMN sequestration ([Figure 8](#pbio-0030174-g008){ref-type="fig"}C), and---as shown for hypoxia above---was followed by a significant reduction of lung tissue damage as assessed by the 4-fold decrease in the LIS ([Figure 8](#pbio-0030174-g008){ref-type="fig"}C). ::: {#pbio-0030174-g008 .fig} Figure 8 ::: {.caption} ###### Intratracheal Injection of A~2A~R Selective Agonist Mimics Protective Effects of Hypoxia \(A) IT injection of the A~2A~R agonist CGS21680 into endotoxin-inflamed lungs provides protection similar to that observed in hypoxia-treated mice. Number of PMNs recovered after 48 h by BAL from endotoxin-injected animals that were kept at normal 21% oxygen atmosphere was significantly diminished by IT injections of CGS21680 compared to placebo-treated mice. Lung PMNs (left graph) from A~2A~R agonist-treated animals also produced lower levels of reactive oxygen metabolites (H~2~O~2~; right graph). \(B) Significantly decreased lung vascular permeability (protein in BAL; left graph) and improved lung gas exchange (p~a~O~2~; right graph) in endotoxin-injected mice after treatment with the A~2A~R agonist CGS21680. \(C) Histologic evidence for the lung tissue-protecting effects of A~2A~R agonist during endotoxin- and oxygenation-induced lung damage. Quantitative analysis of lung histopathology by a pathologist blinded to the experimental design revealed inhibition of PMN sequestration in IT LPS-injected mice after treatment with the A~2A~R-selective agonist CGS21680 for 48 h. The lung tissue damage was also significantly decreased as assessed by the LIS (n = 9, mean ± standard deviation). Representative H&E-stained slices in the right two photomicrographs show less intracapillary PMN sequestration and almost no intraalveolar accumulation of PMNs in CGS21680-treated mice. These CGS21680-induced changes are similar to those observed for the effects of hypoxia on endotoxin- injected animals (compare with [Figure 6](#pbio-0030174-g006){ref-type="fig"}C). ::: ![](pbio.0030174.g008) ::: Treatment with CGS21680 was effective even when applied in the more severe polymicrobial toxin model of lung injury (i.e., LPS + SEB); IT injections of this agonist under hyperoxic conditions rescued the majority of mice from oxygenation-induced death. The death rate was 80% among oxygenated mice with inflammatory lung injury in the control group, but dramatically less among those treated with the A~2A~R agonist ([Figure 9](#pbio-0030174-g009){ref-type="fig"}). ::: {#pbio-0030174-g009 .fig} Figure 9 ::: {.caption} ###### IT Administration of A~2A~R Agonist Protects from Increased Death Rate upon Oxygenation of Mice with Acute Lung Injury Compensation for the oxygenation-associated loss of the hypoxia → adenosine → A~2A~R signaling pathway by IT injection of CGS21680 significantly decreased the oxygen-exacerbated death rate in mice with acute lung injury induced by IT injection of SEB and LPS. For further explanation, see legend for [Figure 1](#pbio-0030174-g001){ref-type="fig"}. ::: ![](pbio.0030174.g009) ::: Thus, the exogenously added, selective, synthetic A~2A~R agonist compensated for the loss of endogenously formed adenosine in oxygenated inflamed lungs, thereby decreasing lung injury and rescuing mice from death. In an important control, CGS21680 at the dosing regimen used to treat wild-type mice was proven to be selective, since it did not affect lung inflammation in A~2A~R gene-deficient mice (unpublished data). Discussion {#s3} ========== The oxygenation of hypoxic patients with impaired lung function is an important and life-saving therapeutic measure, but 15% of patients with ARDS still die from treatment-refractory hypoxia \[[@pbio-0030174-b18]\]. We here provide evidence supporting the hypothesis that, in a mouse model of lung inflammation, while oxygenation relieves the immediate life-threatening consequences of hypoxemia, it also further exacerbates acute lung injury and even may lead to death due to the interruption of the critically important, nonredundant hypoxia → adenosine → A~2A~R-mediated lung-protecting pathway. This conclusion is supported by (i) the strong increase in lung inflammation and mortality after short-term exposure of mice to high and even to moderately elevated concentrations of oxygen (see [Figures 1](#pbio-0030174-g001){ref-type="fig"} and [2](#pbio-0030174-g002){ref-type="fig"}); (ii) the causative pathogenic role of PMNs in inflammatory lung injury (see [Figure 3](#pbio-0030174-g003){ref-type="fig"}) and the up-regulation of A~2A~R expression on alveolar PMNs (see [Figures 4](#pbio-0030174-g004){ref-type="fig"}); (iii) the critical role of A~2A~R in lung protection from enhanced accumulation of PMNs as well as more pronounced vascular permeability and stronger impairment of lung gas exchange in A~2A~R genetically deficient mice (see [Figure 5](#pbio-0030174-g005){ref-type="fig"}A) and in A~2A~R antagonist-treated wild-type mice (see [Figure 5](#pbio-0030174-g005){ref-type="fig"}B); (iv) the strong lung-protective effects of hypoxia by suppression of PMN emigration, PMN activation, lung vascular permeability, and impairment of gas exchange in wild-type mice (see [Figure 6](#pbio-0030174-g006){ref-type="fig"}); (v) the evidence that hypoxia is upstream of A~2A~Rs in the same anti-inflammatory, lung tissue-protecting pathway, as shown by the failure of the adenosine-producing (see [Figure 7](#pbio-0030174-g007){ref-type="fig"}A) hypoxia to protect A~2A~R-deficient mice from death and excessive pulmonary TNF-α cytokine production (see [Figure 7](#pbio-0030174-g007){ref-type="fig"}B--[7](#pbio-0030174-g007){ref-type="fig"}D), and (vi) the confirmation of the hypoxia-elicited lung-protective effects in wild-type mice by the A~2A~R agonist CGS21680 (see [Figure 8](#pbio-0030174-g008){ref-type="fig"}), which (vii) could rescue mice from hyperoxia-accelerated death ([Figure 9](#pbio-0030174-g009){ref-type="fig"}). By confirming the molecular mechanism of the predicted exacerbation of lung injury by oxygenation, this mechanism also offers a direct and effective preventive measure by compensating for the oxygenation-related loss of endogenous adenosine. Accordingly, we suggest the reevaluation of relevant ventilation and oxygenation protocols and the consideration of compensatory therapeutic treatments of oxygenated-inflamed lungs with A~2A~R agonists (reviewed in \[[@pbio-0030174-b07]\]) to prevent uninterrupted inflammatory lung damage (see [Figure 8](#pbio-0030174-g008){ref-type="fig"}) and death ([Figure 9](#pbio-0030174-g009){ref-type="fig"}). Other anti-inflammatory drugs alone or in combination with an A~2A~R agonist should be also considered in future refinements of this approach, because an A~2A~R agonist offers the advantage of pharmacologically restoring the physiological tissue-protecting pathway \[[@pbio-0030174-b04], [@pbio-0030174-b11], [@pbio-0030174-b12]\], which is unintentionally weakened by therapeutic oxygenation. It is likely that these findings are most applicable to ARDS patients, although other clinical situations in which inflamed lungs are oxygenated should be also considered. The lungs of ARDS patients are known to be heterogenously ventilated, leaving substantial lung areas hypoxic with up to 33% of total lung volume being nonaerated \[[@pbio-0030174-b21]\]. These are the most damaged and therefore hypoxic lung regions, where---counterintuitively---hypoxia may protect still-healthy surrounding lung tissues from the additional inflammatory injury by promoting formation of extracellular adenosine \[[@pbio-0030174-b08]--[@pbio-0030174-b10]\] and strengthening the anti-inflammatory A~2A~R signaling pathway \[[@pbio-0030174-b04]--[@pbio-0030174-b07], [@pbio-0030174-b11], [@pbio-0030174-b12]\]. Supportive ventilation to increase lung tissue oxygen levels also eliminates the hypoxia-associated formation of anti-inflammatory endogenous adenosine, and this allows the unopposed continuation of the inflamed lung destruction. Thus, a necessary medical intervention (oxygenation of oxygen-deprived patients) may also cause an iatrogenic exacerbation of the very condition that led to the need for oxygenation in the first place. Such pathophysiological consequences of tissue oxygenation are expected to be most pronounced in collapsed hypoxic lung areas that would be recruited by any kind of ventilatory strategy \[[@pbio-0030174-b22]\]. In agreement with our findings are sporadic observations of effects of oxygenation in other clinical human protocols \[[@pbio-0030174-b23], [@pbio-0030174-b24]\] and experimental assays in animal models \[[@pbio-0030174-b25], [@pbio-0030174-b26]\]. For instance, intraoperative administration of 100% oxygen augmented proinflammatory cytokine production of alveolar macrophages within 2--8 h of the start of anesthesia and surgery in patients \[[@pbio-0030174-b23], [@pbio-0030174-b24]\]. Similarly, in animal models of lung injury, a synergistic action between infectious agents \[[@pbio-0030174-b25]\], bacterial toxins \[[@pbio-0030174-b26]\], or acid aspiration \[[@pbio-0030174-b27]\] and hyperoxia was demonstrated and difficult to explain by direct oxygen toxicity. Indeed, the time periods of these effects were much too short for the manifestation of oxygen toxicity, which usually takes more than 64--72 h to become clinically apparent in mice \[[@pbio-0030174-b14]\]. Of note, in the same mouse strain (C57BL/6) that we used in our studies, breathing of pure oxygen did not result in death before the fourth day \[[@pbio-0030174-b15]\]. Although it is likely that synergy between lung-damaging noxious agents and oxygen is mediated by oxygen radicals \[[@pbio-0030174-b27]\], the elimination by oxygenation of the natural anti-inflammatory hypoxia → A~2A~R signaling pathway was not appreciated before, which may account for a significant proportion of inflammatory complications in patients. Our observations of potent anti-inflammatory effects of hypoxia in lung injury (see [Figures 6](#pbio-0030174-g006){ref-type="fig"} and [7](#pbio-0030174-g007){ref-type="fig"}D) are in agreement with previously published effects of hypoxia, including significant attenuation of emigration of neutrophils to the site of inflammation in carrageenin-induced pleurisy \[[@pbio-0030174-b28]\], inhibition of granulocyte adhesion to endothelial cells \[[@pbio-0030174-b29]\], enhanced shedding of adhesion molecule CD11b from PMNs \[[@pbio-0030174-b30]\], and suppression of cytokine formation when PMNs are in contact with extracellular matrix proteins \[[@pbio-0030174-b31]\]. Use of cd39- and cd73-null animals revealed that extracellular adenosine produced through adenine nucleotide metabolism during hypoxia is a potent mechanism attenuating excessive tissue PMN accumulation \[[@pbio-0030174-b09]\]. Hypoxia was also shown to strongly inhibit production of MCP-1 \[[@pbio-0030174-b32]\], IL-1β \[[@pbio-0030174-b33]\], granulocyte-macrophage colony-stimulating factor \[[@pbio-0030174-b34]\], and induced down-regulation of cosignaling molecules (CD80) \[[@pbio-0030174-b35]\]. Hypoxia was further shown to cause decreased expression of Toll-like receptor 4 receptors by inhibiting translocation of activator protein 1 \[[@pbio-0030174-b36]\] and caused suppression of Escherichia coli-induced nuclear factor κB and activator protein 1 transactivation \[[@pbio-0030174-b37]\]. Finally, hypoxia was shown to stimulate phosphatidylinositol 3-kinase activity and thereby protect human lung microvascular endothelial cells and epithelial type II-like A549 cells from subsequent oxygen toxicity \[[@pbio-0030174-b38]\]. However, hypoxia was also shown to exert some proinflammatory effects in vitro \[[@pbio-0030174-b39]\] and in vivo \[[@pbio-0030174-b09], [@pbio-0030174-b40], [@pbio-0030174-b41]\] for time intervals ranging from 3 to 12 h, which may not reflect the long-term effects of hypoxia on inflammatory processes as assessed in our study. Importantly, the virtually complete lung protection from the proinflammatory effects of 100% oxygen by an immunosuppressive A~2A~R agonist ([Figure 9](#pbio-0030174-g009){ref-type="fig"}) supports the view that exacerbation of lung damage by 100% oxygen could be almost fully accounted for by immune mechanisms in addition to mechanical \[[@pbio-0030174-b42],[@pbio-0030174-b43]\] or radical-mediated mechanisms of lung injury \[[@pbio-0030174-b02]\]. It remains to be determined, however, whether the much longer time course of the direct toxicity of 100% oxygen in healthy, noninflamed lungs could be explained by the recruitment of immune cells \[[@pbio-0030174-b44]\], which then, in turn, would exacerbate the 100% oxygen-triggered lung tissue damage through prolongation of trauma-initiated inflammation. Finally, the observations and conclusions of this report may have implications for the widely used therapeutic oxygenation in patients with lung injury, and may answer the long-standing question as to why lungs are the most susceptible to inflammatory injury and why lung failure usually precedes multiple organ failure \[[@pbio-0030174-b45], [@pbio-0030174-b46]\]. We propose that the diseased lung might be uniquely susceptible to iatrogenic damage by oxygenation, because the lungs are exposed to the highest levels of oxygen tension of all organs. Therefore, this critical, A~2A~R signaling-mediated, hypoxia-driven, lung tissue-protecting anti-inflammatory mechanism is the most vulnerable to elimination by oxygenation in the hypoxic areas of damaged lung as compared to other vital organs. Materials and Methods {#s4} ===================== {#s4a} ### Mice {#s4a1} Female 6- to 8-wk-old C57BL/6 A~2A~R gene-deficient mice (N10) and age-matched wild-type mice were maintained in pathogen-free NIH animal facilities. The A~2A~R genotypes of mice were determined by Southern blot analysis as described previously \[[@pbio-0030174-b47]\]. All animal study protocols were approved by the NIH Animal Care and Use Committee. ### IT administration of LPS and SEB. {#s4a2} Female C57BL/6 mice were anesthetized by isoflurane anesthesia. Using a modified transtracheal illumination technique \[[@pbio-0030174-b48]\], mice were nontraumatically intubated by intratracheal insertion of a 24-gauge catheter (Abbocath-T, Abbott Ireland, Ireland) via direct laryngoscopy. SEB was injected IT at a dose of 1 mg/kg body weight in a total volume of 50 μl per mouse. After IT injection of unconscious but spontaneously breathing mice, animals were held in an upright position for 15 s and then briefly shaken in all directions to ensure homogenous fluid dispersion in the lung. After 1 h, mice were anesthetized again with isoflurane and injected with LPS at a dose of 4 mg/kg body weight in a total volume of 50 μl per mouse. The injected fluid was dispersed by the same positioning maneuver as applied for the intrapulmonary distribution of SEB. Control, sham-treated mice were intubated and injected with only the solvent that was used to inject toxins, i.e., phosphate-buffered saline. ### LPS administration {#s4a3} LPS was IT injected at a dose of 2 mg/kg body weight in a total volume of 50 μl per mouse under isoflurane anesthesia using the same technique as described above. ### Depletion of granulocytes in mice. {#s4a4} Mice were depleted in granulocytes by two IV injections of anti-GR-1 monoclonal antibody at a dose of 1 mg/kg body weight, administered 24 h before and briefly before IT injection of LPS. Control animals received isotype-matched antibody at the same dose and intervals. ### In vivo administration of A~2A~ receptor agonist and antagonist. {#s4a5} The A~2A~R agonist CGS21680 was dissolved in PBS and administered by IT injection at 0,1 mg/kg body weight in a total volume of 50 μl per mouse. IT injection of CGS21680 was repeated every 8 h until termination of the experiment. In control animals, the solvent of CGS21680, i.e., PBS only, was administered in the same way and on the same schedule. CGS21680 solution or solvent were injected within 15 min of administration of SEB and/or LPS in both models of lung injury. In studies using ZM241385, the A~2A~R antagonist was injected subcutaneously at a dose of 10 mg/kg body weight every 6 h until the end of the experiment. ZM241385 was dissolved in DMSO (10 mg/145 μl of DMSO) and then further diluted in 14,8 ml of PBS, yielding a 2 mM working concentration of which a volume of approximately 250--300 μl was injected per mouse. Control animals received the same volume of the solvents only. ZM214385 or control solution were injected subcutaneously 30 min before IT LPS administration. According to the pharmacokinetics of subcutaneously injected ZM214385 at 10 mg/kg body weight, plasma concentrations of the antagonist were higher than 50 nM, even 6 h after its administration. This concentration of ZM241385 is sufficient to exert maximum pharmacologic antagonism on the adenosine-induced cAMP response in murine thymocytes (for determination of pharmacokinetics of ZM214385 see also "Modulation and measurement of cAMP production of PMNs by A~2A~R stimulation and antagonism in vitro," below). ### Control of fraction of inspired oxygen. {#s4a6} In order to modify the fraction of inspired oxygen, mice were placed in airtight modular incubation chambers (Billups-Rothenberg, San Diego, California, United States), and the atmosphere was controlled by a constant gas flow (1,5 l/min) of desired composition (10% O~2~, 60% O~2~, or 100% O~2~). To prevent any CO~2~ retention in the chambers, the chambers\' bottom was covered with approximately 250 g of anesthetic CO~2~ absorber material (Sodasorb, Grace & Co, Chicago, Illinois, United States). Composition of gas atmosphere was tested intermittently by analyzing pCO~2~ and pO~2~ values in an equilibrated fluid sample drawn from a tube inside the chamber whenever the latter was opened for reinjection of mice. During the stay of mice in chambers, animals were fed and given fluids by offering them Transgel (Charles River Laboratories, Wilmington, Massachusetts, United States). ### Assessment of lung gas exchange and oxygen tension in peripheral blood. {#s4a7} At the end of the experiments, arterial blood specimen from the tail artery were sampled directly into heparinized glass capillaries that were closed on both ends with Parafilm and kept on ice until further analysis (Rapidlab 248 system; Chiron Diagnostics, Essex, United Kingdom). ### Histological analysis. {#s4a8} IT LPS-injected mice were sacrificed by isoflurane anesthesia, and the trachea was surgically exposed and cannulated with a 20-gauge needle. The thoracic cage was opened to allow lungs to expand during injection of 1 ml of 4% paraformaldehyde. After 15 min, the lungs were removed and fixed in 4% paraformaldehyde until processing and H&E staining (American Histolabs, Washington, Maryland). Quantitative analysis of lung neutrophil sequestration was performed as previously described \[[@pbio-0030174-b49]\], and lung tissue injury was semiquantitatively assessed in a blinded fashion by a professional pathologist (I. Bittmann, MD). Lung slices of IT LPS-injected groups of mice were first assessed for pathological changes in general, enabling identification of lung inflammation parameters to be further evaluated in a semiquantitative way by assignment of four degrees of increasing severity. Overall assessment of slides showed acute blood congestion, intraseptal and intraalveolar neutrophil sequestration, interstitial edema, intraalveolar macrophage accumulation, pneumocyte type II activation, endothelial cell activation, and endothelial adherence of leukocytes. No intraalveolar edema and no fibrin deposits were observed. Lung tissue injury was then evaluated by assignment of four degrees of severity (0--3) to the following parameters in H&E stained lung slides: interstitial edema formation, intraalveolar macrophage accumulation, pneumocyte activation, endothelial cell activation, and leukocyte adhesion to endothelial cells. The lung injury score of each lung slice was calculated as an average of each parameter\'s degree of severity, ranging from 0 for a healthy lung to 15 (= 5 × 3) for a maximally injured lung. Further details on definition of degrees of severity of pathological changes of each parameter are described in [Table 1](#pbio-0030174-t001){ref-type="table"}. ::: {#pbio-0030174-t001 .table-wrap} Table 1 ::: {.caption} ###### Determination of Lung Injury Score ::: ![](pbio.0030174.t001) MØ, macrophages. ::: ### BAL and cellular differentials. {#s4a9} Following sacrifice and surgical preparation of mice as outlined above for histological analysis, except the opening of the thoracic cavity, lungs were lavaged by injection of 0,6 ml of ice-cold HBSS, which after each collection was repeated four times. The recovered (over 90%) BAL fluid was processed for total cell counts and cellular differentials. Contaminating erythrocytes were lysed by 0,01% saponin, and total leukocytes counted with a Neubauer chamber. Differential counts were performed after cytospin onto glass slides and staining with Hema-3 stain. The remaining bronchoalveolar fluid was spun down and the supernatant collected for determination of protein concentration using the Bio-Rad Protein Assay (Bio-Rad, Hercules, California, United States) according to the manufacturer\'s manual. The cell pellet was used for flow cytometric detection of hydrogen peroxide production by neutrophils. ### Flow cytometric determination of hydrogen peroxide production by activated PMNs. {#s4a10} BAL cells or whole blood, the latter withdrawn from the tail vein and diluted in ice-cold heparinized (40 IE/ml) HBSS, were washed two times with HBSS (0 °C). Cell pellets were resuspended in HBSS containing dihydrorhodamine (2,5 μM). For measurement of spontaneous hydrogen peroxide production by oxidation of dihydrorhodamine to fluorescent rhodamine, cells were either left on ice or incubated at 37 °C for 30 min, then returned thereafter to ice. In parallel experiments designed to test the effects of CGS21680 on blood granulocytes, cells were preincubated with LPS (10 μg/ml) in the absence or presence of the A~2A~R agonist for 15 min, then stimulated by addition of the chemotactic tripeptide fMLP (10 μM). When pharmacologic effects of CGS21680 were tested on BAL granulocytes, no LPS was added. Cells were washed two times with ice-cold PBS that contained bovine serum albumin (0,5%). FcgII/III receptors were blocked by anti-mouse CD16/CD32 (10 min, 1 mg/ml) and cells were incubated with PE-anti-mouse Ly6G (Gr-1) (15 min, 0,5 mg/ml). After washing with PBS, cells were subjected to flow cytometry. Granulocytes were identified by staining for Gr-1 on Fl-2. Hydrogen peroxide production was determined by the fluorescence intensity of rhodamine on Fl-1. Appropriate compensation was set on fluorescence channels to avoid signal overlap. All data were collected in the log-amplified mode, and readings linearized using CellQuest software (Becton Dickinson, San Diego, California, United States). Spontaneous activity was calculated by subtraction of values of cells kept on ice from those incubated at body temperature. Chemotactically induced activity was determined by subtraction of values obtained at body temperature from those after addition of fMLP. ### Preparation of granulocytes from bone marrow. {#s4a11} Bone marrow was removed from femur and tibia by injection of HBSS medium, washed two times, and resuspended in HBSS medium. Remaining red blood cells were removed by dextran sedimentation, and neutrophils were recovered by discontinuous Ficoll density gradient centrifugation. Purity of cells (over 90% granulocytes) was checked and further used for cAMP and adenosine receptor mRNA determination as outlined below. ### Modulation and measurement of cAMP production of PMNs by A~2A~R stimulation and antagonism in vitro {#s4a12} In vivo-activated granulocytes recovered by BAL from inflamed lungs or naïve granulocytes obtained from bone marrow of healthy mice were incubated in the presence of rolipram (1 μM, 37 °C, 30 min) and activated with fMLP (10 μM) in the absence and presence of increasing concentrations of the A~2A~R agonist CGS21680. Specificity was tested by either pharmacological antagonism (A~2A~R antagonist ZM241385 1 μM) or using cells from A~2A~R^−/−^ mice. After 2 min of chemotactic activation, cAMP metabolism was stopped by addition of HCl and putting cells on dry ice. Cell suspensions were kept frozen (−80 °C) until analysis of cAMP levels as previously described \[[@pbio-0030174-b50]\]. Freshly prepared thymocytes from wild-type mice in which cAMP production was induced by CGS21690 (10 μM) in the absence or presence of ZM241385 were used as a positive control. Since ZM241385 concentration-dependently decreased the cAMP production of thymocyte elicited by A~2A~R activation with a fixed concentration of adenosine, the system was also used to determine the plasma concentrations of ZM241385 during the in vivo treatment of mice with the antagonist (unpublished data). ### Determination of A~1~R- and A~2A~R-specific mRNA levels in PMNs. {#s4a13} Quantification of A1- and A~2A~R-specific mRNA levels were determined in naïve or in vivo-activated granulocytes according to the method published previously \[[@pbio-0030174-b51]\]. ### Blood sampling and determination of plasma concentrations of adenosine. {#s4a14} Arterial blood samples were drawn from anesthetized mice by cardiac puncture using ice-cooled syringes pre-filled with a stop solution containing EHNA (20 μM), dipyridamole (200 μM), EDTA-Na~2~ (20 mM), EGTA (20 mM), and dl-α-glycerophosphate to prevent degradation or additional formation of plasma adenosine. Mice were breathing 21% oxygen or 10% oxygen atmosphere at least 5 min before cardiac blood sampling. By obtaining arterial blood samples from the left ventricle, the loss of the adenosine due to its rapid degradation by adenosine deaminases was minimized and plasma concentrations of adenosine better reflected those in lung tissue, although local levels of adenosine in hypoxic tissues are expected to be much higher. Drawn samples were processed and plasma concentrations of adenosine determined by HPLC as previously described \[[@pbio-0030174-b52]\]. ### Reagents {#s4a15} SEB was obtained from Toxin Technology (Sarasota, Florida, United States). LPS (E. coli E055:B5) and HBSS were purchased from Sigma Chemicals (St. Louis, Missouri, United States) and supplemented by 1 mM MgCl~2~ and 1 mM CaCl~2~ freshly before use. Dihydrorhodamine was purchased from Molecular Probes (Eugene, Oregon, United States). PE-labeled anti-mouse Ly6G (Gr-1) antibody and isotype control antibody were obtained from Caltag Laboratories (Burlingame, California, United States). Fc-receptor-blocking anti-CD16/CD32 antibodies were purchased from BD Biosciences Pharmingen (San Diego, California, United States). Hema 3 stain for white blood cell differential counts was purchased from Fisher Scientific (Swedesboro, New Jersey, United States). Dye reagent concentrate for protein determination was obtained from Bio-Rad Laboratories. A~2A~R-selective agonist CGS21680 and antagonist ZM241385 were purchased from Tocris (Ellisville, Missouri, United States). A kit for cAMP enzyme immunoassay was purchased from Amersham Pharmacia Biotech (Piscataway, New Jersey, United States). Isoflurane was obtained from Baxter (Deerfield, Illinois, United States). All other chemicals were purchased from Sigma Chemicals. ### Statistics. {#s4a16} Data are represented as individual values. In the graphs, the horizontal lines give the means of the individual values. Comparison between independent samples was performed by two-tailed nonparametric Mann-Whitney test. To test for the strength of the relationship between two variables, linear regression analysis was performed and Pearson correlation coefficients calculated. Differences between survival rates were tested by Chi^2^ test. The authors wish to thank Dr. William Paul for support, discussions, and help. MT was a visiting National Institutes of Health Fogarty Fellow on leave from the Clinic of Anesthesiology, Klinikum Grosshadern, University of Munich, and was partially supported by a grant from the German National Research Foundation to MT, TH733/2--1. Competing interests. The authors have declared that no competing interests exist. Author contributions. MT and MVS conceived and designed the experiments. MT, AC, AO, EJ, CC, and PS performed the experiments. DL and IB analyzed the data. MT and MVS wrote the paper. Citation: Thiel M, Chouker A, Ohta A, Jackson E, Caldwell C, et al. (2005) Oxygenation inhibits the physiological tissue-protecting mechanism and thereby exacerbates acute inflammatory lung injury. PLoS Biol 2(6): e174. A~2A~R : adenosine A~2A~ receptor ARDS : acute respiratory distress syndrome BAL : bronchoalveolar lavage cAMP : cyclic adenosine 3′,5′-monophosphate fMLP : N-formyl-methionyl-leucyl-phenylalanine IT : intratracheal(ly) LIS : lung injury score LPS : lipopolysaccharide MIP-2α : major intrinsic protein 2α PMN : polymorphonuclear leukocyte pO~2~ : oxygen partial pressure SEB : staphylococcal enterotoxin B TNF : tumor necrosis factor	PubMed Central	2024-06-05T03:55:56.297097	2005-5-3	{ "license": "Creative Commons Zero - Public Domain - https://creativecommons.org/publicdomain/zero/1.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088279/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e174", "authors": [ { "first": "Manfred", "last": "Thiel" }, { "first": "Alexander", "last": "Chouker" }, { "first": "Akio", "last": "Ohta" }, { "first": "Edward", "last": "Jackson" }, { "first": "Charles", "last": "Caldwell" }, { "first": "Patrick", "last": "Smith" }, { "first": "Dmitry", "last": "Lukashev" }, { "first": "Iris", "last": "Bittmann" }, { "first": "Michail V", "last": "Sitkovsky" } ] }
PMC1088280	Introduction {#s1} ============ In thalamus, high frequency bursting of action potentials has been suggested to maximize stimulus onset detection, while prolonged depolarizations switch the thalamic output mode to tonic firing (single spiking), yielding a more linear reflection of the excitatory synaptic drive \[[@pbio-0030175-b01]\]. Relatively long periods of quiescence followed by bursts are observed during wakefulness, when they may powerfully activate the cortex in response to salient sensory stimuli \[[@pbio-0030175-b01]--[@pbio-0030175-b06]\]. By contrast, continuous, oscillatory bursting of thalamic relay neurons is a hallmark of some stages of sleep and is thought to interfere with sensory stimulation of the cortex during sleep \[[@pbio-0030175-b07]\]. Output-mode switching is likely to be an important general mechanism through which neurons in other brain regions code for different stimulus features \[[@pbio-0030175-b08]\]. Understanding the mechanisms responsible for transitions between bursting and single-spiking output modes is an important goal, because it may provide insight not only into stimulus feature coding, but also how brain-state transitions influence attentional states. Lying at the interface between the hippocampus and the entorhinal cortex, the subiculum provides the major output from the hippocampus to a number of brain structures \[[@pbio-0030175-b09]\]. By analogy to the thalamus, the subiculum can be viewed as a relay structure that transmits information from the hippocampal formation to subcortical and cortical areas, such as the prefrontal and entorhinal cortex. In humans and rodents alike, the subiculum consists of intrinsically bursting neurons that exhibit transitions from bursting to single-spike firing \[[@pbio-0030175-b06],[@pbio-0030175-b10]--[@pbio-0030175-b12]\]. Unlike thalamic neurons, however, where bursting is voltage dependent and driven by inactivating, low-voltage--activated (T-type) Ca^2+^ channels \[[@pbio-0030175-b13],[@pbio-0030175-b14]\], bursting in the subiculum is voltage independent and driven primarily by non-inactivating, high-voltage--activated Ca^2+^ tail currents \[[@pbio-0030175-b15]\]. Consequently, depolarization alone is incapable of switching the output from bursting to single-spike firing in subicular neurons \[[@pbio-0030175-b12],[@pbio-0030175-b16]\]. Given this difference in the mechanism of bursting, we hypothesized that the mechanism for output-mode switching in the subiculum would be fundamentally different from that of the thalamus. We endeavored to determine the mechanism for subicular output-mode transitions, because it may provide general insight into how neurons conditionally regulate their output, and further our goal of understanding how the subiculum integrates hippocampal information during normal function and disease. Results {#s2} ======= Spontaneous extracellular single-unit activity and population field recordings from the dorsal subiculum of anesthetized adult rats (6--7 wk old) exhibited epochs of spontaneous firing with dominant delta (1--3 Hz) and theta (5--10 Hz) frequency bands ([Figures 1](#pbio-0030175-g001){ref-type="fig"} and S1). Each epoch of activity began with bursting and then shifted to predominantly single spiking in all subicular neurons recorded (n = 7; [Figure 1](#pbio-0030175-g001){ref-type="fig"}A). Interspike interval histograms of extracellular units showed a bimodal distribution, with a narrow interspike interval peak associated with the high-frequency bursts of two or three action potentials at intervals of 2--8 ms, and a broader second peak consisting of primarily single spiking at intervals of 40--200 ms ([Figure 1](#pbio-0030175-g001){ref-type="fig"}B and [1](#pbio-0030175-g001){ref-type="fig"}E). There was a positive correlation between the inter-event interval and the probability of bursting, such that cells with a higher rate of spontaneous action potential firing were least likely to burst ([Figure 1](#pbio-0030175-g001){ref-type="fig"}C). The average cumulative probability plot of pre-event silent periods indicates that 55% percent of all events were single spikes (range 14%--92%), while 45% were bursts (range 8%--86%, n = 7; [Figure 1](#pbio-0030175-g001){ref-type="fig"}D). The median interval before a single spike was 89 ms, while the median interval before a burst was 378 ms ([Figure 1](#pbio-0030175-g001){ref-type="fig"}D), indicating that single-spike and burst event probability is related to the pre-event silent period. To determine how event probability changed as a function of inter-event interval, a probability density plot was obtained by differentiating the cumulative probability curve fit. For intervals less than 229 ms (4.4 Hz) the probability of a single spike was greatest, while for longer intervals (over 229 ms) the probability of a burst was greatest ([Figure 1](#pbio-0030175-g001){ref-type="fig"}E). Consistent with this result, the average pre-event silent period was significantly longer for bursts (418 ms, 2.4 Hz) than for single spikes (153 ms, 6.5 Hz, n = 7; p \< 0.01). Thus, the transition from bursting to single spiking corresponds to the transition from the delta to the theta frequency band. Two possible mechanisms could account for this result: Either the subicular neurons in vivo burst in response to strong synaptic inputs following long silent intervals and switch to tonic firing as synaptic inputs depress, or intrinsic conductances of the subicular neurons may undergo changes in activation states that determine the output mode during constant synaptic drive. ::: {#pbio-0030175-g001 .fig} Figure 1 ::: {.caption} ###### In Vivo Action Potential Output Mode Transition in Bursting Dorsal Subicular Neurons \(A) Representative 30-s trace shows six epochs of bursting-to-single spiking output mode transition. Each epoch began with bursts that switched to single spikes. Bursts (BS, blue) and single spike (SS, red) events were extracted from the raw data (black trace) for each cell, and the percentages of spike events (burst or single spikes) were determined. For this cell, 22% of the total events were bursts, while 78% of events were single spikes. The lower inset depicts an expansion from one epoch showing the transition from bursting to single spiking. \(B) Interspike interval histogram for the neuron in (A) shows a bimodal distribution of short intervals of less than 8 ms (blue dashed lines show burst intervals), and of longer intervals between 40 and 200 ms. The arrow points to an expanded burst (blue) and single spike (red) extracellular waveform (scale bar, 2 ms). \(C) A strong positive correlation was observed between the percentage of burst events and the average event interval (n = 7). The filled black circle shows the relative position of the cell from (A) and (B). \(D) The cumulative probability plot shows the probabilities of total burst (45%) and single spike (55%) events across inter-event intervals for all cells (n = 7). A sigmoidal function was used to fit the data (half max SS \[red\] = 89 ms, BS \[blue\] = 378 ms). \(E) Probability density plot shows the fractional probability change across the inter-event interval. The peak probability density for bursting (blue) was longer than for single spiking (red) events; the arrow indicates the interval beyond which the probability of bursting exceeds that of single spiking. ::: ![](pbio.0030175.g001) ::: To address this issue, we studied the intrinsic properties of subicular neurons in the acute brain-slice preparation. Patch-clamp recordings were obtained from subicular pyramidal neurons in hippocampal slices prepared from mature rats ([Figure S2](#sg002){ref-type="supplementary-material"}) \[[@pbio-0030175-b12],[@pbio-0030175-b15],[@pbio-0030175-b16]\]. Neuronal responses to noisy current injections were found to undergo transitions from bursting to single spiking that mimicked the in vivo transitions ([Figure S3](#sg003){ref-type="supplementary-material"}). Similarly, in response to 5-Hz synaptic stimulation of CA1 afferents to subiculum, the majority of bursting neurons (seven out of eight) exhibited a transition from bursting to single spiking during the train. To determine whether the transition from bursting to single spiking was the result of a postsynaptic, intrinsic property of bursting neurons, we injected simulated excitatory synaptic currents (sEPSCs) that produced simulated excitatory postsynaptic potentials (EPSPs) with rise and decay kinetics similar to synaptically evoked EPSPs \[[@pbio-0030175-b16]\]. The sEPSCs were injected at frequencies of 1--10 Hz in the presence of blockers of glutamatergic, GABAergic, and muscarinic synaptic receptors (2.5 mM kynurenic acid, 2 μM SR 95531, and 1 μM atropine, respectively) ([Figure 2](#pbio-0030175-g002){ref-type="fig"}). At frequencies between 1 and 10 Hz, bursting neurons switched from bursting to single spiking ([Figure 2](#pbio-0030175-g002){ref-type="fig"}A and [2](#pbio-0030175-g002){ref-type="fig"}B). This transition was not associated with a progressive change in the peak of the single-spike fast afterdepolarization (ADP1 versus ADP2; [Figure 2](#pbio-0030175-g002){ref-type="fig"}F), suggesting that a change in the Ca^2+^ tail current driving the ADP was not responsible for the switch from bursting to single spiking. Furthermore, this transition persisted at a frequency (5 Hz) where the post-burst afterhyperpolarization (AHP) had decayed completely (n = 8; [Figure 2](#pbio-0030175-g002){ref-type="fig"}G). The mode switching was associated with a change in the initial slope of the ADP, however, suggesting that the mechanism of bursting is activated within 1 ms of the initial spike repolarization ([Figure 2](#pbio-0030175-g002){ref-type="fig"}D and [2](#pbio-0030175-g002){ref-type="fig"}E). Mode switching was also associated with a frequency-dependent and cumulative reduction in the rate of rise (dV/dt) of the initial action potential in a burst (a 17% ± 1% decrease in the peak rate at 10 Hz and a 96% ± 1% recovery at 1 Hz, n = 14). Analysis of this reduction in dV/dt indicated that the reduction in the process responsible for this change was frequency dependent, cumulative, and slow to recover ([Figure 3](#pbio-0030175-g003){ref-type="fig"}A). The dV/dt recovery to control level was fit by a double exponential (τ~fast~=10.3 ms, τ~slow~=545 ms). The relative contributions of the fast and slow components to the recovery of dV/dt were 80% and 20%, respectively. The action potential dV/dt is proportional to Na^+^ channel availability \[[@pbio-0030175-b17]\], so we focused our attention on voltage-gated Na^+^ channel inactivation as a possible mechanism for the transition from bursting to single-spiking mode. ::: {#pbio-0030175-g002 .fig} Figure 2 ::: {.caption} ###### In Vitro Transition from Bursting to Single Spiking Output Is Frequency-Dependent \(A) A representative cell shows the typical burst-to-single spike transition in response to current injections of five sEPSCs (τ~decay~ = 6 ms) delivered at frequencies between 2 and 10 Hz. Note: Only the 10 Hz sEPSC is shown. \(B) The average number of bursts in response to five sEPSC injections delivered at frequencies of 1--10 Hz (n = 17). \(C) The representative trace shows the lack of a change between the ADP peaks (ADP1 versus ADP2) that follow a burst (5 Hz) and the complete decay of the AHP before the transition from bursting to single spiking. (D and E) A representative trace (D) and averaged data (E) show that the initial slope of the ADP is decreased during the transition from bursting to single spiking at 5 Hz (n = 7). The arrow in (D) points to the beginning of the ADP slope in an overlay of a burst before and after a transition to a single spike at 5 Hz. \(F) The average peak of the ADP (in the absence of a second spike) does not change during a train (5 Hz; n = 4). \(G) The average decay of the AHP (time for V ~m~ to return to rest) is not different between two bursting events (B-B, n = 8) and transitions from bursting to single spike events (B-S, n = 8) at frequencies of 2--5 Hz. ::: ![](pbio.0030175.g002) ::: ::: {#pbio-0030175-g003 .fig} Figure 3 ::: {.caption} ###### Transitions from Bursting to Single Spiking Output Are Mediated by Slow Recovery from Na^+^ Current Inactivation \(A) Filled black circles show the dV/dt of the first action potential response to the fifth suprathreshold sEPSC input at frequencies between 1 and 10 Hz (n = 14; open circles, 5 mM BAPTA, n = 3; recorded at 34 °C). The inset shows a cumulative decrease in the action potential rise rate (dV/dt) (τ~1Hz~ = 17,980 ms, τ~2Hz~ = 925 ms, τ~10Hz~ = 250 ms; n = 14). \(B) Recovery from burst-induced (two or three step pulses \[2 ms\] from −100 mV to 0 mV) inactivation of Na^+^ current (black, n = 11 cell-attached patches) and Ca^2+^ current (blue; n = 9 \[pooled cell attached, n = 4, and nucleated patches, room temperature, n = 5\]). The inset shows examples of the burst-induced slow inactivation of Na^+^ current (black, vertical scale bar is 50 pA) and the lack of slow inactivation of the Ca^2+^ tail current (blue, vertical scale bar is 5 pA). I ~Na~ (black) is an overlay of multiple traces, each with an initial burst-like current followed by a single test pulse at different recovery intervals. \(C) The left tracing shows a nucleated outside-out patch recording (at room temperature) comparing Na^+^ current in response to a single step pulse (2 ms) in control (black) or 100 ms (10 Hz) after burst conditioning (red). Arrowhead shows the lack of reduction in the I ~Ca~ (tail). The right tracing shows a reduction of the Na^+^ current without a change in the I ~Ca~ (tail) after bath application of 1 nM TTX (black, control; red, TTX). Values are reported as mean ± standard error of the mean. ::: ![](pbio.0030175.g003) ::: Previous work has shown that recovery from prolonged inactivation of Na^+^ channels in CA1 pyramidal neurons \[[@pbio-0030175-b17]--[@pbio-0030175-b19]\] exhibits a time course similar to the transition from bursting to single spiking we observed in subiculum. To determine whether this property could be responsible for output mode transition, we measured currents in cell-attached and outside-out nucleated patches obtained from the soma of subicular neurons (see [Materials and Methods](#s4){ref-type="sec"}). In these patches, brief depolarizing voltage commands elicited rapidly inactivating Na^+^ currents (I ~Na~; τ~inact~ = 0.36 ± 0.024 ms; n = 9). We studied recovery from inactivation of I ~Na~ by using two or three brief (2 ms) depolarizations at 200 Hz to mimic Na^+^ channel inactivation during action potential bursting. Recovery of I ~Na~ was probed using a test pulse (2 ms) at variable times (10--2000 ms) after the initial burst-like depolarization ([Figure 3](#pbio-0030175-g003){ref-type="fig"}B). I ~Na~ recovery from prolonged inactivation was fit by a double exponential (τ~fast~ = 3.7 ms, τ~slow~ = 934 ms). The relative amplitudes of the fast and slow recovery from inactivation of I ~Na~ were 86% and 14%, respectively. The slow time constant is similar to the time course for mode switching and the slow and cumulative reduction in the dV/dt ([Figure 3](#pbio-0030175-g003){ref-type="fig"}A and [3](#pbio-0030175-g003){ref-type="fig"}B), suggesting that Na^+^ channel availability may regulate the transition from bursting to single-spiking (96% Na^+^ channel recovery at 1.2 s for nucleated patches compared to 96% recovery in the action potential dV/dt at 1 Hz stimulation). The Ca^2+^ tail current, shown to be important for the ADP that drives bursting \[[@pbio-0030175-b18]\], was measured in cell-attached or outside-out nucleated patches in the presence of blockers for voltage-gated Na^+^ and K^+^ channels (500 nM TTX, 5 mM 4-aminopyridine, 30 mM tetraethyl ammonium chloride, and 130 mM internal Cs^+^). The resulting Ca^2+^ current exhibited biexponential deactivation (τ~fast~ = 0.44 ± 0.024 ms, τ~slow~ = 1.65 ± 0.17 ms, voltage stepped from −70 mV to 0 mV, n = 7). There was no evidence for either fast or slow inactivation of the Ca^2+^ tail current when repeated at frequencies of 5--10 Hz ([Figure 3](#pbio-0030175-g003){ref-type="fig"}B). Similarly, outward K^+^ currents contributed the termination of the burst, but voltage-activated K^+^ currents evoked by brief depolarizations did not change when evoked repeatedly at 5--10 Hz ([Figure S4](#sg004){ref-type="supplementary-material"}). To more directly test the hypothesis that prolonged inactivation of I ~Na~ is responsible for the transition from bursting to single spiking, we used a low concentration of tetrodotoxin (TTX) to block a fraction of Na^+^ channels similar to that removed by prolonged inactivation. We reasoned that if small reductions in I ~Na~ are capable of inducing a switch from bursting to single spiking, then very low concentrations of TTX should influence the mode transition. Because most subicular neurons switch from bursting to single spiking in response to suprathreshold input spaced 100 ms apart (10 Hz; see [Figure 2](#pbio-0030175-g002){ref-type="fig"}B), we first determined that 11% ± 2% of I ~Na~ is unavailable 100 ms after two conditioning pulses mimicking a burst ([Figure 3](#pbio-0030175-g003){ref-type="fig"}B and [3](#pbio-0030175-g003){ref-type="fig"}C). A similar fraction of I ~Na~ was blocked by 1 nM TTX (16% ± 2%, n = 6; [Figure 3](#pbio-0030175-g003){ref-type="fig"}C). In current-clamp recordings, 1 nM TTX induced a transition from bursting to single spiking at low frequencies (2 Hz), where subicular neurons normally burst reliably (n = 5; [Figure 4](#pbio-0030175-g004){ref-type="fig"}A), and reduced the dV/dt of the initial action potential in a burst by 9% ± 2% (n = 5; [Figure 4](#pbio-0030175-g004){ref-type="fig"}B). The ability of 1 nM TTX to accelerate the output-mode transition was completely reversible in most cases and was not associated with a change in the subthreshold response to sEPSCs or a change in the resting potential during recording ([Figure 4](#pbio-0030175-g004){ref-type="fig"}A). Similar results were obtained at slightly higher concentrations of TTX (2--5 nM); however, at these concentrations, bursting was completely blocked and replaced by single spiking, even at very low frequencies of less than 0.1 Hz (unpublished data). A decrease in the initial slope of the ADP (−23% ± 6%, n = 9) was associated with a frequency-independent, TTX-induced mode transition ([Figure 4](#pbio-0030175-g004){ref-type="fig"}C) similar to the slope decrease (−33% ± 6%, n = 7) during the frequency-induced mode switch (see [Figure 2](#pbio-0030175-g002){ref-type="fig"}D). This finding indicates that Na^+^ channel inactivation regulates bursting by reducing the initial slope of the ADP, an effect that could be mediated by persistent Na^+^ current, rapid reactivation of Na^+^ current, or current returning to the soma as the action potential propagates into the dendrites \[[@pbio-0030175-b20]\]. ::: {#pbio-0030175-g004 .fig} Figure 4 ::: {.caption} ###### Modest Reduction of Na^+^ Current Induces a Frequency-Dependent Output-Mode Transition from Bursting to Single Spiking \(A) The number of bursts in response to a train of five suprathreshold sEPSC currents injections at 2 and 5 Hz delivered every 20 sec before, during (red), and after bath application of 1 nM TTX. The black (left), red, and black (right) insets show overlays of the action potential transition in response to five 2-Hz sEPSC inputs (I ~Command~) under baseline, TTX (1 nM), and wash conditions, respectively. Note the dashed line in the TTX (red) trace shows the lack of change in the ADP during the stimulus train. The lower panels show no change in a subthreshold sEPSP or resting potential (mV) in response to TTX. The insets show an expanded sEPSC (left; scale bar, 5 ms) and the corresponding sEPSP (right; scale bars, 1 mV vertical, and 20 ms horizontal) that was used to monitor changes in passive properties of the cell. \(B) A representative trace showing the reduction in the dV/dt of the action potentials before (black) and after 1 nM TTX (red). The average reduction in the first action potential dV/dt after 1 nM TTX was 9% ± 2% (n = 5). \(C) An overlay (left tracing) showing the reduction in the initial slope of the ADP before (black burst) and after (red single spike) 1 nM TTX (All traces in \[A\], \[B\], and \[C\] were taken from the same recording). The graph on the right shows that the low concentrations of TTX (1--5 nM) necessary to induce a switch from bursting to single spiking decrease the initial slope (as illustrated in \[C\]) of the ADP for a burst (black) and compared to the first transition to a single spike (red) at low frequencies (less than 0.1 Hz) that alone do not influence the burst-single spike transition (ADP reduction = 23% ± 6%, p \< 0.003, n = 9). ::: ![](pbio.0030175.g004) ::: Discussion {#s3} ========== Identifying the specific mechanisms that govern hippocampal/subicular output is important for a greater understanding of cognitive and mnemonic functions (e.g., working memory and sleep), as well as pathological conditions (e.g., schizophrenia, addiction, epilepsy, and Alzheimer\'s disease) that have been linked to the subiculum \[[@pbio-0030175-b06],[@pbio-0030175-b10],[@pbio-0030175-b11],[@pbio-0030175-b21]--[@pbio-0030175-b24]\]. Here we present evidence for a novel intrinsic mechanism that governs the output mode of bursting subicular neurons: a Na^+^ channel slow inactivation state. In vivo and in vitro, we observed that subicular neurons generate bursts near the beginning of epochs of activity (less than 10 s) and later switch to single-spike firing. Induction of prolonged Na^+^ channel inactivation produces a frequency-dependent output-mode transition from bursting to single spiking by reducing the initial slope of the ADP in a burst. Thus, while we have shown previously that the ADP is mediated largely by a Ca^2+^ tail current \[[@pbio-0030175-b15]\], a critical component also depends on voltage-gated Na^+^ channels, thus rendering bursting susceptible to Na^+^ channel inactivation. Slow (prolonged) inactivation of Na^+^ channels is distinct from conventional fast inactivation and has been studied in a number of preparations, including the squid giant axon; skeletal muscle; and neocortical, thalamic, and CA1 pyramidal neurons \[[@pbio-0030175-b17]--[@pbio-0030175-b19],[@pbio-0030175-b25]--[@pbio-0030175-b27]\]. Despite numerous biophysical studies of slow inactivation, very little is known about its functional consequences at the cellular or network levels. We and others have previously suggested a role for slow inactivation of Na^+^ channels in regulating the amplitude of back-propagating action potentials in the dendrites of CA1 pyramidal neurons \[[@pbio-0030175-b17]--[@pbio-0030175-b19]\]. In bursting neurons of subiculum, it appears that the slow inactivation state of Na^+^ channels has a fundamental role in regulating pyramidal neuron output mode. Transitions from bursting to single spiking have been observed in other cell types, such as thalamic relay cells and neocortical and hippocampal CA3 pyramidal cells. In the thalamus, sustained transitions from bursting to regular spiking are mediated by inactivation of T-type Ca^2+^ channels \[[@pbio-0030175-b13]\]. This is fundamentally different from the mechanism we describe here for pyramidal neurons of subiculum, both because the ion channels involved are different and because bursting is self-limiting in subiculum, but not in the thalamus (but see \[[@pbio-0030175-b28]\]). Self-limiting bursting has also been observed in pyramidal neurons of CA3 and neocortex \[[@pbio-0030175-b29],[@pbio-0030175-b30]\]. Although the mechanisms have not been experimentally determined, transitions from bursting to regular spiking in CA3 have been modeled by activation of slow, Ca^2+^-activated K^+^ channels \[[@pbio-0030175-b31]\]. This mechanism is not responsible for the output-mode shift in subiculum, because transitions from bursting to single spiking routinely occurred at intervals longer than the burst or single-spike AHP (see [Figure 2](#pbio-0030175-g002){ref-type="fig"}G). Furthermore, we found no evidence for activity-dependent increases in voltage-activated K^+^ currents (see [Figure S4](#sg004){ref-type="supplementary-material"}). As a determinant of bursting or single spiking output in the subiculum, prolonged inactivation of Na^+^ channels is likely to be an important mechanism for regulating network activity. Neuromodulators that decrease slow Na^+^ channel inactivation would enhance the predominance of bursting, while increasing slow inactivation would reduce bursting in subiculum, thus changing the way the hippocampus activates its target structures. Recent work indicates that such modulation may indeed occur; in layer 5 neurons of the prefrontal cortex, serotonin receptors (5-HT~2~) modulate the slow-inactivation state of Na^+^ channels and can influence spiking dynamics \[[@pbio-0030175-b32]\]. Furthermore, it is possible that slow inactivation of Na^+^ channels is a general mechanism that conditionally regulates the output mode of bursting prefrontal cortex projection neurons. Functionally, transient bursting followed by a shift to single spiking may serve to enhance the importance of new stimuli as they become represented within a network. For example, intrinsic bursting neurons are capable of converting brief, strong inputs into longer-lasting action potential outputs (bursts), which activate target structures more effectively than single spikes \[[@pbio-0030175-b04]\]. One major output pathway of subiculum is to the nucleus accumbens medium spiny neurons \[[@pbio-0030175-b33]\]. Hippocampal input to these neurons, via subiculum, quickly drives them from their very hyperpolarized (about −80 mV) "down" states to their more depolarized (about −55 mV) "up" states \[[@pbio-0030175-b34]\]. Subicular bursting may provide a "gating" signal at the beginning of the input, in order to drive target neurons into an activated "up" state. During sustained activity, bursting is gradually replaced by the weaker single-spike output mode, which may be sufficient to maintain the "up" state. By regulating spike-output mode in this way, prolonged inactivation of Na^+^ channels may be a key regulator of network function in the hippocampus and elsewhere in the brain. Materials and Methods {#s4} ===================== {#s4a} ### Extracellular single-unit recording. {#s4a1} Male rats (42--50 d old) were anesthetized with chloral hydrate (400 mg/kg intraperitoneally) and mounted in a stereotaxic apparatus. The tail vein was catheterized to administer supplemental anesthetic. Body temperature was maintained at 36.5--37.0 °C. Glass electrodes were filled with a solution of 2 M NaCl and 1% fast green dye. The tip of the electrode was less than 2 μm in diameter (impedance 1.8--2.2 MΩ in 0.9% saline). The electrode was advanced through a small burr hole in the skull with a hydraulic microdrive (David Kopf Instruments, Tujunga, California, United States) to the dorsal subiculum. Single-unit and bursting extracellular waveforms were identified using a high-impedance amplifier (Fintronic, Foster City, California, United States), bandpass filtered (low cutoff, 400 Hz; high cutoff, 500 Hz), and monitored with an oscilloscope, computer, and audio monitor. An analog-to-digital interface (Digidata 1200; Axon Instruments, Foster City, California, United States) was attached to a computer running AxoScope 7.0 software. Bursting units and single-spiking waveforms were analyzed using a template-matching algorithm in SPIKE 2 (Cambridge Electronic Design, Cambridge, United Kingdom) and custom analysis macro written in IGOR Pro 5.01 (Wavemetrics, Portland, Oregon, United States). Event detection threshold was set at six standard deviations above the mean of the entire recording. Multiunit recordings were not included. Bursting neurons were identified by their high frequency (130--300 Hz) clusters of two to four action potentials, decreasing spike amplitudes with each burst, and broadening spike widths within each burst (initial spike widths = 1.2--1.8 ms). ### Solutions and drugs {#s4a2} ACSF for slice experiments consisted of 125 mM NaCl, 25 mM glucose, 25 mM NaHCO~3~, 2.5 mM KCl, 1.25 mM NaH~2~PO~4~, 2 mM CaCl~2~, and 1 mM MgCl~2~ (pH 7.4) (bubbled with 5% CO~2~ and 95% O~2~). Kynurenic acid (2.5 mM), SR 95531 (2--4 μM), and atropine (1 μM) were added to the ACSF to block synaptic input. The whole-cell current-clamp recording solution contained 115 mM K-gluconate, 20 mM KCl, 10 mM Na~2~-phosphocreatine, 10 mM HEPES, 2 mM Mg-ATP, 0.3 mM Na-GTP, and 0.1% biocytin (pH 7.3). For nucleated voltage-clamp I ~Na~ recordings, intracellular solutions were either CsCl-based (130 mM CsCl, 10 mM Na~2~-phosphocreatine, 10 mM HEPES, 2 mM EGTA, 2 mM Mg-ATP, and 0.3 mM Na~2~-GTP \[pH 7.3\]) or Cs-gluconate-based (115 mM Cs-gluconate with 20 mM CsCl substituted for 130 mM CsCl \[pH 7.3\]). Membrane potentials were not corrected for a −8-mV liquid junction potential. For cell-attached patch recording of I ~Na~, the pipette solution contained 120 mM NaCl, 3 mM KCl, 10 mM HEPES, 2 mM CaCl~2~, 1 mM MgCl~2~, 30 mM tetraethyl-ammonium chloride (TEA), 5 mM 4-aminopyridine (4-AP) and NaOH (pH 7.4). For cell-attached patch recording of I ~Ca~, the pipette solution contained 120 mM NaCl, 3 mM KCl, 10 mM HEPES, 5 mM CaCl~2~, 1 mM MgCl~2~, 30 mM TEA, 5 mM 4-AP, 0.1 mM nicotine, and 500 nM TTX (pH adjusted to 7.4 with NaOH). All drugs were obtained from Sigma (St. Louis, Missouri, United States). ### Slice preparation {#s4a3} Hippocampal slices were prepared from mature male Wistar rats (36--50 d old). Slices were incubated for 20--40 min in a chamber containing warm (34--35 °C) ACSF and maintained at room temperature until they were moved to the recording chamber (32--35 °C). For recording, slices were transferred individually to a chamber on a fixed stage of a Zeiss (Oberkochen, Germany) Axioscop equipped with DIC optics. Recordings were obtained with visualization Dage-MTI (Michigan City, Indiana, United States) tube camera. ### Current-clamp recordings. {#s4a4} Whole-cell, current-clamp recordings were made from the soma using a BVC-700 amplifier (Dagan, Minneapolis, Minnesota, United States). Patch-clamp electrodes were fabricated from thick-walled borosilicate glass with resistances of 3--5 MΩ in ACSF. Synaptic stimulation of CA1 axons projecting to subiculum was performed using a bipolar stimulating electrode fabricated from theta glass and connected to a stimulus isolator (Axon Instruments, Union City, California, United States). Brief current injections were intended to mimic EPSCs by using a dual exponential function (τ~rise~ = 1 ms; τ~decay~ = 6 ms). Data were stored on a computer (Dell) via an ITC-16 analog-to-digital interface (Instrutech, Port Washington, New York, United States). Data acquisition and analysis was performed using Igor Pro. Group comparisons were made using paired or unpaired t-tests as appropriate. ### Voltage-clamp recordings. {#s4a5} Cell-attached and nucleated outside-out patch recordings were obtained from slices prepared from 14- to 16-day old male rats using an EPC-7 amplifier (Heka Elektronik, Lambrecht/Pfalz, Germany). Nucleated-patch experiments were performed at room temperature for improved stability of nucleated patches and improved clamp of Na^+^ currents. Glass electrodes (3--5 MΩ) were coated with Sylgard. Nucleated patches were obtained with capacitance and leak subtraction as previously described \[[@pbio-0030175-b18]\]. Analysis was performed using averages of five to 20 sweeps. ### Histological procedures {#s4a6} For in vivo experiments, the recording position was marked by local iontophoresis of fast green dye through the electrode. Electrode placement was verified using routine light microscopy from serial coronal brain sections (60 μm). For in vitro recording, the slices were placed in paraformaldehyde (4%) at 4 °C. We processed the biocytin-filled cells using an avidin, horseradish peroxidase, diaminobenzene reaction. Supporting Information {#s5} ====================== Figure S1 ::: {.caption} ###### Power Spectral Density Plot of Local Field Potentials Recorded in the Subiculum In Vivo Note the predominant delta-frequency band (1--4 Hz) and a weaker theta-frequency band (5--10 Hz) recorded in vivo. The inset shows a 10-s sweep illustrating the local field potential fluctuations. (8.5 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S2 ::: {.caption} ###### Transverse Section (300 μm Thick) of the Hippocampus Showing Paraformaldehyde-Fixed, Biocytin-Labeled Bursting Subicular Neuron (3.3 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S3 ::: {.caption} ###### In Vitro Current Clamp Recording of a Bursting Neuron in Response to a Noisy Current Stimulus Input Bursts (asterisks) transition to single spikes with sustained depolarization. The bottom trace shows the rate of rise (dV/dt) of the action potentials above. The dV/dt is high with each initial burst, decreases in size as the cell transitions to single spike mode, and recovers as the time between spike events increases. (8.4 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S4 ::: {.caption} ###### Activity-Dependent Changes in Voltage-Gated K^+^ Current Cannot Explain the Transition from Bursting to Single Spiking The inset depicts sample K^+^ currents evoked by 2 ms depolarizations from −70 to 0 mV in cell-attached patches. Three superimposed traces are shown, each beginning with the response to two-step depolarizations (to mimic a burst) and followed by a single test response. Note that the K^+^ current depresses during the burst, but recovers rapidly and does not facilitate, as would be required to explain the activity-dependent output-mode transition. On the main graph, the normalized current amplitude in the test response is plotted as a function of recovery time after the burst (n = 6). (10.4 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: We would like to thank Michela Marinelli for assistance with the extracellular recording and helpful comments on the manuscript. We would also like to thank members of the Spruston laboratory, especially Shannon Moore, Tim Jarsky, Juan Varela, and Bill Kath for helpful discussions and commentary during the project. This research was supported by grants from the National Science Foundation to NS (IBN-9876032) and the National Institutes of Health to NS (NS-35180) and DCC (NIH DA06089). Competing interests. The authors have declared that no competing interests exist. Author contributions. DCC and NS conceived and designed the experiments. DCC and SC performed the experiments. DCC analyzed the data and wrote the paper with NS. Citation: Cooper DC, Chung S, Spruston N (2005) Output-mode transitions are controlled by prolonged inactivation of sodium channels in pyramidal neurons of subiculum. PLoS Biol 3(6): e175. ADP : afterdepolarization AHP : afterhyperpolarization EPSP : excitatory postsynaptic potential I~Na~ : Na^+^ current TTX : tetrodotoxin sEPSC : simulated excitatory synaptic current	PubMed Central	2024-06-05T03:55:56.302311	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088280/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e175", "authors": [ { "first": "Donald C", "last": "Cooper" }, { "first": "Sungkwon", "last": "Chung" }, { "first": "Nelson", "last": "Spruston" } ] }
PMC1088281	Since humans and chimpanzees forged separate evolutionary paths some 5 million to 6 million years ago, we shed our hirsute coat and heavy brow, mastered bipedal locomotion, and acquired a knack for abstract thought while our next of kin learned to use tools, created a complex communication system, and developed the skills to construct tree-bound nests high above the forest floor. We differ by just a tad over 1% at the DNA sequence level, yet scientists predict that both species should harbor genetic footprints of our divergence. One way to find such genetic signatures is to search for genes that reveal signs of natural selection. The assumption is that genes or genomic elements touched by natural selection will show more functionally significant molecular changes than unaffected regions. A 2003 study by Andrew Clarke et al. used this approach to identify human genes affected by positive selection---that is, selection that preserves new genetic variants---by comparing 7,645 genes from humans to their chimp and mouse equivalents. Clarke et al. identified genes in several functional categories, including olfaction and hearing, and showed that positively selected genes are more likely to contain variations (called single nucleotide polymorphisms, or SNPs) associated with genetic diseases. In a new study, Rasmus Nielsen, Michele Cargill, and their colleagues (many of whom participated in the 2003 study) compared 13,731 known genes in humans to their equivalents in chimps to find positively selected genes in both species. Nielsen et al. also identified many genes involved in sensory perception, as well as spermatogenesis, but found the strongest evidence for positive selection in genes related to immune defense. Immunity genes, the authors explain, were likely targeted throughout mammalian evolution, while the perception and olfactory genes probably reflect primate-specific adaptations. They also found a "surprising number" of tumor-suppression and apoptosis genes.[](#pbio-0030202-g001){ref-type="fig"} ::: {#pbio-0030202-g001 .fig} ::: {.caption} ###### Young adult male chimpanzee (Photo: Frans de Waal, Emory University) ::: ![](pbio.0030202.g001) ::: Mutations in coding DNA can be broadly classified into two groups: a nucleotide change can cause an amino acid substitution that either alters the encoded protein (called nonsynonymous mutation) or has no effect (synonymous mutation). Nielsen et al. used a statistical method that denotes positive selection based on the ratio of nonsynonymous to synonymous mutations. Thousands of genes failed to unambiguously pass this test, leaving 8,079 for further study; those that passed were grouped into functional categories, revealing the genes related to immunity, sensory perception, and spermatogenesis. Of genes associated with specific tissues, only testis-specific genes showed evidence of positive selection. Genes expressed in the brain appear to be under selective constraints, indicating that the genetic roots of our cognitive distance from chimps lies elsewhere, perhaps in how genes are regulated or organized. The authors followed up the chimp--human comparison by analyzing the top 50 genes from their list in a group of African- and European-Americans. The data provide further support for the conclusion that these genes have undergone positive selection. Nielsen et al. were surprised to find so many genes involved in tumor development and control among the top 50 positively selected genes (in both primates). The factors behind this pattern are unclear, but the authors suggest that studying the genes\' other functions, in immunity or spermatogenesis, may offer clues to selective pressures---and it also raises some intriguing paradoxes. It could be, for example, that the overrepresentation of genes involved in tumor suppression, spermatogenesis, and apoptosis sets up competing interests on two fronts. Apoptosis normally eliminates up to 75% of sperm cells during spermatogenesis; mutations that protect sperm cells from apoptosis may be selected for, benefiting the cell but compromising the fitness of the organism. Positive selection for apoptosis avoidance in the germ line could subsequently increase the probability of cancer in body cells---apoptosis routinely eliminates unhealthy cells---pitting the "selfish interests of a germ cell" against the organism\'s interest in avoiding cancer. Future studies will have to determine whether these explanations---of an evolutionary arms race---prove plausible. We\'re a long way from understanding why we\'re so different from our closest primate cousins, but this study provides plenty of tools, and hypotheses, to mine the tiny differences in our DNA for more clues.	PubMed Central	2024-06-05T03:55:56.305054	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088281/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e202", "authors": [] }
PMC1088282	Ever since Robert Hooke startled the world with finely rendered illustrations of "minute bodies" in his 1665 book, Micrographia, our understanding of the microscopic world within and around us has mushroomed with each technological advance. Though imaging capability has developed light-years beyond the compound microscope that inspired Hooke\'s cork cell epiphany, biologists have only recently been able to observe living cells in chunks of tissue extracted from an organism---an approach that\'s critical for studying processes like cell differentiation and development. Ellen Robey and her group study the mechanisms of cell differentiation and cell fate by tracking T cell development in mouse models. In a new study, Colleen Witt, Ellen Robey, and their colleagues take advantage of a recent innovation called two-photon microscopy to visualize the migration of developing T cells, or thymocytes, in intact thymuses extracted from mice. They find that after cells undergo positive selection---which seals their fate as either helper T or killer T cells---they make a beeline for the thymus interior (called the medulla). Though it\'s been known that positively selected thymocytes migrate to the medulla, this study shows that migration follows a clear directional course, possibly guided by long-range signaling cues.[](#pbio-0030205-g001){ref-type="fig"} ::: {#pbio-0030205-g001 .fig} ::: {.caption} ###### Real-time visualization of thymocytes within intact thymic lobes using two-photon microscopy ::: ![](pbio.0030205.g001) ::: Because two-photon microscopy can penetrate tissue at high resolution without distorting or damaging the specimen, the authors could characterize thymocytes moving through their native tissue environment and interacting with the molecules and cells they would normally encounter. (For more on two-photon microscopy, see the Primer by David Piston \[[10.1371/journal.pbio.0030207](10.1371/journal.pbio.0030207)\] and the "Tracking the Details of an Immune Cell Rendezvous in 3-D" \[DOI: [10.1371/journal.pbio.0030206](10.1371/journal.pbio.0030206)\].) In the service of optimum immune defense, the mammalian adaptive immune system churns out billions of T cells a day. Precursor T cells originate in the bone marrow and migrate to the thymus, where their immune mettle is tested by a selection process that only about 1% will survive. Immature double-positive thymocytes---so-called because they have the protein markers associated with both helper (CD4) and killer (CD8) T cells---inhabit the outer thymic layer, called the cortex, while single-positive thymocytes---which have lost either the CD4 or CD8 marker following positive selection---are found in the central medulla. How a thymocyte reacts to other lymphocytes as it wends its way through the thymus determines whether it undergoes positive selection and matures into a helper or killer T cell or undergoes negative selection and programmed cell death. The signaling cues that guide this process remain obscure. Witt et al. engineered mice with thymocytes tagged with green fluorescent protein (GFP), removed their thymic lobes for microscopic analysis when they were 4.5 to 5.5 weeks old, then observed the behavior of the glowing cells. The GFP cells in the cortex showed distinct differences in motility, morphology, and migratory behavior: low-motility cells had a spherical, nonpolar shape, moved with a modestly protruding leading edge, and sometimes paused; high-motility cells had a clear leading edge that moved in fits and starts and never paused. Once on the move, high-motility cells mostly hewed to a single direction while low motility cells often retraced their steps. Unlike the low-motility cells, the high-motility cells traveled in a linear manner through the cortex, suggesting directed migration. Since there were so few of the fast-moving, inwardly migrating cells, the authors hypothesized that they had undergone positive selection---which they went on to confirm in transgenic mouse models. From these results, Witt et al. conclude that positive selection triggers a "rapid directional migration pattern." And because that migration corresponds to an area of the cortex that extends up to 200 microns below the outer layer of the thymus, it appears to be guided by long-range signaling cues. As often happens in biology, close observation of a process reveals more complexity and raises more questions about the mechanics underlying it. Homing in on the source of these long-range signaling cues and characterizing the migratory patterns of the large number of slow-moving cells will go a long way toward understanding how the major components of immunity acquire their defensive chops.	PubMed Central	2024-06-05T03:55:56.305751	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088282/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e205", "authors": [] }
PMC1088283	If you stopped to consider the potentially pathogenic encounters waiting outside your door on any given day, you might never leave home. Then you would just have to contend with the billions of microbes reproducing on your toothbrush and kitchen sponge. Of course, most of us safely navigate our microbe-filled world thanks to an immune system that manufactures billions of lymphocytes a day, in case any of those microbes proves malicious. Lymphocytes arise in the bone marrow, where B lymphocytes mature; T lymphocyte precursors migrate to the thymus to mature. Both T and B cells then travel to distinct regions in the spleen and lymph nodes---B cells amassing in follicles and T cells in T zones---in search of alien antigens. After encounters with antigen, B and T lymphocytes migrate to the edges of their respective zones and compare notes. These B cell--helper T cell interactions are essential for an effective B cell--mediated antibody response. One would not expect such crucial interactions to be left to chance alone, but evidence of directed migration has not been generated---until now. In a new study, Takaharu Okada and Jason Cyster at the University of California at San Francisco together with Mark Miller and Mike Cahalan at the University of California at Irvine and several colleagues used a groundbreaking technology called two-photon microscopy to visually inspect intact lymph nodes extracted from mice to investigate this lymphocyte rendezvous. They discovered a combination of random and directed behaviors: antigen-engaged B cells move randomly along the follicle outskirts, then undergo directed migration near the follicle/T-zone border as they home in on their helper T counterparts. (For more on two-photon microscopy, see the Primer by David Piston \[[10.1371/journal.pbio.0030207](10.1371/journal.pbio.0030207)\] and "Tracking Migrating T Cells in Real Time" \[DOI: [10.1371/journal.pbio.0030205](10.1371/journal.pbio.0030205)\].) Once a B cell recognizes an antigen via immunoglobulin receptors on its surface, it will not proliferate, differentiate into a plasma cell, and generate mass quantities of antibodies without the go-ahead from a helper T cell. To investigate the dynamics of this process, Okada, Miller, and colleagues transferred fluorescently labeled HEL-specific transgenic B cells, or Ig-tg B cells (engineered to secrete antibodies against the model antigen, hen egg lysozome \[HEL\]), and nonaltered B cells into mice with identical genetic backgrounds. An hour after HEL injections, lymph nodes were removed from the mice for microscopic analysis. The Ig-tg B cells were "fully occupied" by HEL antigen and---unlike the naïve (unengaged) non-Ig-tg cells---had begun to aggregate along the edge of the follicles. After antigen binding, the Ig-tg cells grew sluggish compared to naïve cells, then headed for the B-zone/T-zone (B/T) border. Half of the primed cells reached the B/T boundary, compared to 20% of the naïve cells, and did so by (mostly) taking a path that was closer to a straight line---a sign of directed migration. Ratios of path length plotted against displacement from the path show that antigen-engaged B cells tacked toward the boundary when they got within about 140 microns of it.[](#pbio-0030206-g001){ref-type="fig"} ::: {#pbio-0030206-g001 .fig} ::: {.caption} ###### Two-photon microscopy reveals the three-dimensional dynamics of B cell (red) and T cell (green) conjugate within lymph tissue in real time ::: ![](pbio.0030206.g001) ::: The authors go on to show that antigen-engaged B cells need the chemokine receptor CCR7 to follow directions to the T zone---which contains an abundant supply of CCR7\'s signaling protein, or ligand---CCL21. Besides concentrating in the T zone, CCL21 also showed up in follicles, in an increasing gradient from the follicle periphery to the boundary. Okada, Miller, and colleagues studied interactions between antigen-engaged B cells and activated helper T cells in a transgenic mouse model and found that only B and T cells with cognate antigens formed stable pairs, which moved at the B cells\' discretion. Based on these findings, the authors conclude that once antigen-engaged, B cells follow the long-range chemokine gradient to the B/T boundary. After arrival at the boundary the B cells can undergo multiple, even polygamous, contacts with T cells---which might facilitate optimal pairings---before B cell proliferation and antibody production begins. Whether promiscuous and monogamous liaisons produce different B cell reactions is unclear. And though CCR7 helps B cells find the border, it\'s not clear what keeps them there. But thanks to the T lymphocyte--B lymphocyte dynamics outlined here, immunologists have plenty of avenues for exploring these questions to further elucidate the complex interactions underlying an effective antibody attack.	PubMed Central	2024-06-05T03:55:56.306330	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088283/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e206", "authors": [] }
PMC1088284	The brain is arguably the most complex computing machine on the planet, with billions of individual neurons that process the input they receive from sensors in the body and from other neurons. The composition of channels that regulate electrical conductance across the neuronal membrane is a key determinant of the exact pattern of a neuron\'s response to the stimuli it receives. Bursting---brief epochs of rapid firing---is one such pattern, while single spiking, with longer intervals between firing, is another. In a deep-seated part of the cerebral cortex, the hippocampus receives input from and sends output to widespread regions of the brain; information processing in the hippocampus is thought to underlie aspects of memory. The output of the hippocampus is handled by a subregion, the subiculum, that exhibits response modes that correspond to either bursting or single-spike response patterns. In this issue, Don Cooper, Sungkwon Chung, and Nelson Spruston show that the bursting--spiking switch is controlled by a new and surprisingly simple mechanism, prolonged inactivation (shutting off) of channels that conduct sodium ions through the membrane. Working with rats, the authors measured activity from bursting subicular neurons in vivo, and observed frequent transitions from bursting to single spiking. Bursting tended to occur after long silent periods, while single spiking predominated after short intervals. To determine what drove this change from bursting to single spiking, they examined electric activity in brain slices. The switch was strongest when neurons were stimulated at frequencies between 1 and 10 Hertz, which suggested an inactivation and recovery process, perhaps mediated by prolonged inactivation of a small number of sodium channels. To test this hypothesis, the authors applied a very low concentration of tetrodotoxin, a neurotoxin from puffer fish that specifically blocks sodium channels. Indeed, blocking 16% of the channels with the toxin induced a switch from bursting to single spiking even at very low frequency stimulation, when subicular neurons normally maintain their bursting pattern. Output mode switching by the sustained inactivation of sodium channels is a novel mechanism for controlling the dynamics of neural networks. While the functional significance of the switch remains unexplored, the authors point out that bursting is known to effectively activate target structures. It may be that switching from bursting to single spiking sustains activation of the target once it has been "woken up" by bursting. Conversely, transitions from a powerful burst output to less powerful single-spike mode may serve to initially activate target structures but then allow other inputs to govern the target output. Given the importance of the hippocampus in processing memory and emotion, and its involvement in schizophrenia, epilepsy, and other disorders, these new insights into the regulation of its output may lead to a better understanding of numerous fundamental higher brain processes.[](#pbio-0030210-g001){ref-type="fig"} ::: {#pbio-0030210-g001 .fig} ::: {.caption} ###### An acrylic painting by Don Cooper and Leah Leverich shows the transition zone between the densely packed pyramidal neurons in the CA1 region (right) and the spread-out pyramidal neurons within the subiculum (left) ::: ![](pbio.0030210.g001) :::	PubMed Central	2024-06-05T03:55:56.307230	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088284/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e210", "authors": [] }
PMC1088285	If a novel oxygen-producing metabolic pathway hadn\'t evolved in ancient microbes over 3 billion years ago, it\'s unclear whether humans and other oxygen-dependent species would have either. But evolve we did, adapted to having just the right level of oxygen coursing through our blood---too little oxygen (hypoxia) causes headaches, nausea, and eventually death. Patients with acute respiratory distress syndrome (ARDS) and other serious lung injuries routinely receive oxygenation therapy to facilitate oxygen delivery to deprived tissues. But too much oxygen (hyperoxia) kills, too. Hyperoxia produces free radicals, causing oxidative damage to cells and tissues by disrupting cellular components. And recent evidence suggests that oxygenation therapy might produce dangerous side effects in patients with ARDS who also have severe pulmonary inflammation. In a new study, Manfred Thiel et al., in a team led by Michail Sitkovsky, test the hypothesis that oxygenation weakens a tissue-protecting mechanism triggered by hypoxia. Working with gene-altered mice, the team of immunologists, pathologists, and biochemists finds evidence that clinical oxygenation treatments could aggravate lung injury by inhibiting this protective pathway. But this protective pathway could potentially be restored, they argue, by artificially activating the inhibited pathway with therapeutic activators. Their results have important implications for how patients with ARDS and other serious lung diseases should be treated.[](#pbio-0030211-g001){ref-type="fig"} ::: {#pbio-0030211-g001 .fig} ::: {.caption} ###### Lung tissue under ambient oxygen levels (left) and under 100% oxygen (right), which exacerbates acute inflammatory lung injury ::: ![](pbio.0030211.g001) ::: Hypoxia triggers a signaling pathway mediated by an adenosine receptor (called A~2A~R) that arrests inflammation and tissue damage. It\'s thought that this same hypoxia-driven pathway protects the lungs from the toxic effects of overactive immune cells called neutrophils. Using a mouse model of acute lung injury induced by bacterial infection, Thiel et al. exposed one group of mice to 100% oxygen, mimicking therapeutic oxygenation, and left another group at normal ambient levels (21% oxygen). Five times more mice died after receiving 100% oxygen than died breathing normal oxygen levels. Mice given 60% oxygen---considered clinically safe---got worse, but didn\'t die. To test the hypothesis that oxygen was precipitating these drastic results by exacerbating tissue inflammation, the authors analyzed the neutrophil-mediated immune response. Establishing a correlation between high neutrophil count and increased capillary leakage---indicated by the protein concentrations recovered from the alveolar space, which mediates gas exchange---Thiel et al. confirmed that overactive neutrophils promote lung injury. When otherwise healthy mice were subjected to lung infection and treated with hypoxia (10% oxygen), after 48 hours 90% of the mice showed several signs of improvement associated with an inhibition of neutrophil-mediated inflammation. Thiel et al. went on to show that the adenosine receptor pathway was involved in the oxygen-dependent inflammation. By isolating neutrophils from mice with inflamed lungs and exposing them to high concentrations of a molecule that activates the adenosine receptor, they triggered increased levels of both A~2A~R and cAMP, a molecule that inhibits inflammation. No such increases were seen in mutant mice lacking functional A~2A~R proteins. Hypoxia protects against lung damage, the authors conclude, by working through the A~2A~R signaling pathway to control inflammation. Above-normal oxygen levels interrupt this anti-inflammatory pathway, paving the way for further lung injury. Administering a molecule that jump-starts A~2A~R signaling artificially also significantly reduced the pathological side effects of oxygenation. These results may help explain why some patients with ARDS die following oxygenation therapy. And by identifying the mechanism that is disrupted by oxygenation---A~2A~R signaling---this study suggests that therapies aimed at activating the anti-inflammatory A~2A~R pathway may allow patients to receive the benefits of oxygenation therapy without succumbing to its toxic effects.	PubMed Central	2024-06-05T03:55:56.307786	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088285/", "journal": "PLoS Biol. 2005 Jun 3; 3(6):e211", "authors": [] }
PMC1088286	Most people believe that as societies advance economically they have higher levels of cardiovascular disease (CVD) and other noncommunicable disease (NCD) risks. However, a more detailed analysis of how parameters of economic development are associated with health outcomes as well as NCD risk factors is needed to inform local and global health policies. Such an analysis might dispel prejudices about the "diseases of affluence" and stimulate policy approaches and research that appropriately target emerging risk groups across the globe, regardless of socioeconomic status. From Intuition to Data {#s2} ====================== In a study in this month\'s PLoS Medicine, Ezzati and colleagues have taken a close look at population data available in 80 or so countries on body mass index (BMI), cholesterol, and hypertension \[[@pmed-0020104-b1]\]. They looked at how these CVD risk indicators are predicted by three broad economic parameters: national income, average share of household expenditure spent on food, and proportion of the population living in urban areas. This cross-sectional analysis drew on published studies, reports from ministries of health and the World Health Organization, data from household surveys, demographic data, and centrally available economic indicator data to model the relationship between the selected CVD risk factors and economic/demographic status. The overall results suggest that average BMI and cholesterol increase with national income and then flatten out at higher incomes (or even decline), except in the United States, the home of the Big Mac and of leading practitioners of the sedentary urban lifestyle (in the US, BMI and cholesterol levels do not flatten out with higher income). Not surprisingly, there is an inverse relationship between BMI and proportion of household income needed for buying food in most countries. Urbanization is associated with higher average income, and Ezzatti and colleagues found that urbanization is associated with higher BMI and cholesterol. As urbanization progresses and food availability equalizes among both urban and rural populations, there is less of an association between increasing income and increasing BMI and cholesterol. However, some persistently agricultural economies with large populations (such as Nigeria and Indonesia) tend to retain the inequities that influence diets, leading to protein-calorie deficiencies among those with lower income and increased BMI among those with higher income. Systolic blood pressure did not have as robust a relationship as BMI or cholesterol to urbanization or national income. Looking Ahead {#s3} ============= Despite the limitations of multinational data and the broad brush approach used to interpret these data, there are some important lessons that emerge from this study regarding the population distribution of multiple CVD risk factors. These risk factors are systematically finding their way to low- and middle-income countries and the vulnerable populations therein that still suffer from childhood illness and high communicable disease burdens. This shift is already having an effect on the epidemiology of CVD and other NCDs, particularly in middle-income countries. And we can predict that as low-income countries achieve economic growth, the disease burden of NCDs will be waiting for them as well. Effective policies can prevent some of this impact, if action is taken now; Ezzati and colleagues\' work provides a basis for planning such interventions.[](#pmed-0020104-g001){ref-type="fig"} ::: {#pmed-0020104-g001 .fig} ::: {.caption} ###### A shift towards "Western diets" high in saturated fats is occurring in developing countries (Photo: Renee Comet/National Cancer Institute) ::: ![](pmed.0020104.g001) ::: Planning for the Future {#s4} ======================= To target these interventions effectively, better data are needed on the prevalence of CVD risk factors in low- and middle-income populations, and on the association of these risk factors with NCD health outcomes. CVD risk factor surveillance should be incorporated into national program planning and into best practices for NCD control supported by the World Health Organization and other health development agencies. Multinational agreements, such as the recently activated Framework Convention on Tobacco Control ([www.who.int/tobacco/en](www.who.int/tobacco/en)), can create effective international cooperative efforts to stem the tide of global tobacco use. Global tobacco control should be a health and foreign policy concern even for those countries that have not ratified the treaty. As populations assume more of an urban lifestyle, they should not be limited in their choices for healthy foods, suffer from lack of safe water, or lose opportunities for physical activity. These problems can be reduced through good urban planning, better food policies, improved environmental engineering, and better attention to healthy lifestyle practices in our growing cities. Screening for hypertension, hypercholesterolemia, and nicotine addiction need to become a part of good clinical practices in low- and middle-income countries. Of course, screening for these risks should then also be accompanied by better availability of low-priced secondary prevention therapies such as generic versions of anti-hypertensives, statins, and nicotine replacement therapies. Getting the Balance Right {#s5} ========================= This is not to say that the big infectious disease killers and child health problems should be ignored. Rather, we need to learn from the history of socioeconomic development that it is not simply affluence that permits the increased impact of CVD and other NCDs; it is the risk factors for these diseases that spread across socioeconomic boundaries, causing the same illnesses regardless of the socioeconomic status of the population. Increased attention should be paid to these diseases not just in the developed world, but also in the developing world, where the unfinished agendas on communicable disease and childhood illness have drawn the most attention. Addressing CVD risk factors could best be accomplished through improved international cooperation, better understanding of the risks of globalization, and development of appropriate research and technologies that apply to low- and middle-income populations. Citation: Novotny TE (2005) Why we need to rethink the diseases of affluence. PLoS Med 2(5): e104. BMI : body mass index CVD : cardiovascular disease NCD : noncommunicable disease [^1]: Thomas E. Novotny is Professor in Residence, Epidemiology and Biostatistics, University of California, San Francisco, California, United States of America. E-mail: <tnovotny@psg.ucsf.edu> [^2]: Competing Interests: TEN is on the editorial board of PLoS Medicine.	PubMed Central	2024-06-05T03:55:56.308549	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088286/", "journal": "PLoS Med. 2005 May 3; 2(5):e104", "authors": [ { "first": "Thomas E", "last": "Novotny" } ] }
PMC1088287	Introduction {#s1} ============ Cardiovascular diseases and their nutritional risk factors are among the leading causes of mortality and morbidity globally ([Figure 1](#pmed-0020133-g001){ref-type="fig"}), and have been predicted to rise over the next few decades \[[@pmed-0020133-b01]--[@pmed-0020133-b03]\]. Aging of the world\'s population is a key driver of the expected increase, because cardiovascular disease rates tend to increase with age. In addition to this demographic change, an epidemiological change that involves increases in age-specific rates of cardiovascular diseases in developing countries has also been predicted in some analyses \[[@pmed-0020133-b04]\]. This epidemiological change is a corollary to a predicted population-wide rise in cardiovascular disease risk factors including obesity, blood pressure, cholesterol, and tobacco use with increasing income, originally referred to as the "diseases of affluence" or "Western disease" paradigm \[[@pmed-0020133-b05],[@pmed-0020133-b06]\]. A number of challenges have been made to the diseases of affluence paradigm. For example, it has been observed that cardiovascular diseases and some of their risk factors (e.g., smoking) may decline once they have peaked \[[@pmed-0020133-b07]\]. It has also been documented that within upper-middle-income and high-income countries, cardiovascular diseases and risk factors are increasingly concentrated among the lowest socioeconomic groups \[[@pmed-0020133-b08]--[@pmed-0020133-b11]\]. ::: {#pmed-0020133-g001 .fig} Figure 1 ::: {.caption} ###### Global Mortality and Burden of Disease Attributable to Cardiovascular Diseases and Their Major Risk Factors for People 30 y of Age and Older The size of each circle is proportional to the number of deaths (left) or burden of disease (right; measured in disability-adjusted life years) (in millions). Overweight and obesity affect non-cardiovascular diseases, including diabetes, endometrial and colon cancers, post-menopausal breast cancer, and osteoarthritis, shown as the portions of yellow circles that fall outside the cardiovascular disease circle \[[@pmed-0020133-b57]\]. The mortality estimates exclude osteoarthritis, which results in morbidity but not direct deaths. Disease burden does include nonfatal health outcomes associated with diabetes and osteoarthritis (hence the larger size of the circle for overweight and obesity relative to those for blood pressure and cholesterol). Source: re-analysis of data from Ezzati et al. \[[@pmed-0020133-b57],[@pmed-0020133-b58]\]. ::: ![](pmed.0020133.g001) ::: Despite these challenges in specific populations, at the global level, predictions about rising levels of cardiovascular risk factors with economic development continue to be made \[[@pmed-0020133-b02],[@pmed-0020133-b12],[@pmed-0020133-b13]\]. The global health aspect of the diseases of affluence paradigm is particularly important because it implies that a large proportion of the world\'s population, who live in middle-income countries, will soon face both aging populations and rising age-specific cardiovascular disease rates, and will require increasing focus on policies and interventions to reduce the resulting disease burden \[[@pmed-0020133-b13]\]. Yet the timing of initiating interventions during a society\'s economic development and the specific form of the required interventions have not been addressed based on systematic analyses of risk factor and disease profiles. The diseases of affluence paradigm also implies that cardiovascular disease risk factors are not urgent public health concerns for low-income populations. We systematically examined the population-level relationships between three leading nutritional cardiovascular risk factors---overweight and obesity, elevated blood pressure, and cholesterol---and three economic variables using data for over 100 countries. Analysis of multiple nutritional risks shows more complex economic--epidemiological patterns than those predicted by simple descriptions such as the "diseases of affluence" or "Western disease" paradigms. More importantly, focusing on multiple risk factors helps identify specific intervention and policy options and priorities, with implications for societies at various levels of economic development. Methods {#s2} ======= We examined the cross-sectional relationship between mean population blood pressure, cholesterol, and body mass index (BMI) and three socioeconomic variables: national income, average share of household expenditure spent on food, and proportion of population in urban areas. Blood pressure, cholesterol, and BMI are well-established cardiovascular risk factors and provide aggregate indicators of more complex dietary patterns (e.g., caloric intake, and consumption of salt, fats of different composition, and fruits and vegetables) and physical activity. Further, there are more comparable data from population-based health and nutrition surveys on these physiological indicators than on dietary patterns and physical activity, because these indicators can be more easily defined in a consistent manner and measured using standard techniques. National income is the commonly used indicator for a society\'s material well-being. The share of household expenditure spent on food is a measure of how household economic resources may constrain food purchase. If food forms a large proportion of total household expenditure, households may limit the total amount of food consumed, which should result in lower obesity and possibly lower levels of other risk factors if they are affected by overconsumption, or households may switch to less expensive but lower quality foods that increase various nutritional risk factors (e.g., higher salt, leading to higher blood pressure, or higher sugar and fat, leading to higher obesity) \[[@pmed-0020133-b14]\]. The proportion of the population living in urban areas is a proxy indicator of a number of environmental and lifestyle variables, such as physical activity in occupational and transportation domains, and of access to specific food types. For example, people living in rural areas often have higher levels of physical activity, reflecting their agricultural occupations and the need to walk longer distances for day-to-day activities \[[@pmed-0020133-b15]\]. Similarly, rural and urban populations may have differential access to various food types, possibly with seasonal variations. Urban populations are also likely to have higher access to screening and treatment for risks such as high blood pressure and cholesterol. Data Sources {#s2a} ------------ Data sources for risk factors and economic indicators are provided in [Table 1](#pmed-0020133-t001){ref-type="table"} (see [Table S1](#st001){ref-type="supplementary-material"} for list of countries). Systolic blood pressure (SBP), cholesterol, and BMI were age-standardized so that results could be compared across populations with different age structures. We used the World Health Organization (WHO) standard population \[[@pmed-0020133-b16]\] because it is a better representation of current population age structures than older standard populations (e.g., SEGI). The data used in the analysis were collected by researchers using different instruments. Strict criteria for study selection were applied to ensure methodological rigor and representativeness \[[@pmed-0020133-b17]--[@pmed-0020133-b19]\]. There were, nonetheless, differences among studies in a number of dimensions: sample size, national (versus sub-national) representativeness, year of the study, and age groups included. When restricted to only large-scale, nationally representative studies, the analysis resulted in associations very similar to those presented using the entire dataset; heterogeneity decreased after excluding all countries that did not fulfill these more stringent criteria. ::: {#pmed-0020133-t001 .table-wrap} Table 1 ::: {.caption} ###### Risk and Socioeconomic Variables Used in the Analysis ::: ![](pmed.0020133.t001) ^a^ International dollar (Int\$) is adjusted for purchasing power, and for inflation because the years of data for BMI, SBP, and cholesterol varied across countries. ::: Statistical Analysis {#s2b} -------------------- A local regression model was used to estimate the income--risk relationships \[[@pmed-0020133-b20]\]. A local regression model estimates the association across income levels without assuming a parametric model. Rather, the data determine the fitted curve. This is a preferable approach when there is no theoretical model for the shape of the association. Results {#s3} ======= When considered in relation to national income, mean population BMI and cholesterol increased, then flattened, and eventually declined ([Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"}). Mean BMI increased most rapidly until a national income of about I\$5,000 (see [Table S1](#st001){ref-type="supplementary-material"} for national incomes of individual countries) and peaked at about I\$12,500 for females and I\$17,000 for males. Cholesterol\'s point of inflection and peak (about I\$8,000 and I\$18,000, respectively) were at higher income levels than those for BMI. The BMI decline relative to the peak at high incomes was larger for females than for males. The lower mean female BMI at high-income levels (except in the United States where the female BMI was 28.7 kg/m^2^) is consistent with the evidence on declining female BMI in some high-income countries (e.g., Japan \[[@pmed-0020133-b21]\]) over time, and with the inverse relationship between social class (as measured by education) and female obesity within upper-middle-income countries \[[@pmed-0020133-b11]\]. ::: {#pmed-0020133-g002 .fig} Figure 2 ::: {.caption} ###### Pair-Wise Relationships of Mean Population BMI, SBP, and Total Cholesterol with National Income, Share of Household Expenditure Spent on Food, and Proportion of Population in Urban Areas Data for (A) males and (B) females are shown. National income is measured as per-capita gross domestic product (GDP). BHS, Bahamas; CUB, Cuba; EST, Estonia; ETH, Ethiopia; FIN, Finland; GEO, Georigia; GMB, Gambia; IDN, Indonesia; JOR, Jordan; JPN, Japan; KEN, Kenya; KOR, Korea; KWT, Kuwait; MLT, Malta; MWI, Malawi; NGA, Nigeria; NOR, Norway; NPL, Nepal; PNG, Papua New Guinea; POL, Poland; RUS, Russian Federation; SAU, Saudi Arabia; SLB, Solomon Islands; THA, Thailand; TJK, Tajikistan; TZA, Tanzania; USA, United States; VNM, Viet Nam; WSM, Samoa; ZWE, Zimbabwe. ::: ![](pmed.0020133.g002) ::: ::: {#pmed-0020133-g003 .fig} Figure 3 ::: {.caption} ###### Relationship of Mean Population BMI, SBP, and Total Cholesterol with Average National Income, Food Share of Household Expenditure, and Proportion of Population in Urban Areas Relationships were estimated using local regression models applied to the data in [Figure 2](#pmed-0020133-g002){ref-type="fig"}. Results for (A) males and (B) females are shown. National income was measured as gross domestic product (GDP). The following outlier countries were dropped (see also Results): United States for males and females in the income--BMI relationship, and Russian Federation and Tajikistan for males and females in the food share of household expenditure--BMI relationship. ::: ![](pmed.0020133.g003) ::: [Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"} show that in countries where food is a relatively small proportion of total household expenditure (less than 30%--40%) (i.e., little or no constraints), there was little or no relationship between this factor and mean BMI. Where food is a larger proportion of household expenditure (more than 30%--40%), there was an inverse relationship between mean BMI and the proportion of household expenditure spent on food. Exceptions to this pattern were the Russian Federation and Tajikistan, with a large food share of household expenditure but with relatively high BMI. The departure from the overall pattern by these two countries may reflect the economic consequences of the collapse of the former Soviet Union, which have forced households to devote a large share of their expenditure to food without changing their life style determinants of weight (i.e., diet and physical activity). The relationship between the food share of household expenditure and mean blood cholesterol was similar to that for BMI, but, as in the relationship with income, the flattening of the relationship occurred later (i.e., at lower food shares) for cholesterol than for BMI. The rapid rise in mean population BMI with increasing income indicates higher consumption of energy as food purchase constraints at very low income levels are removed. That cholesterol increases more slowly and continues at higher income levels and lower food share of income than BMI may indicate that additional income is first used to increase caloric intake, followed by dietary change beyond higher calories (e.g., switch from unsaturated vegetable fats and oils to animal fat \[[@pmed-0020133-b22]\]) as income rises further. The proportion of the population in urban areas was positively correlated with mean BMI and cholesterol ([Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"}). The relationships of mean BMI and cholesterol with urbanization also showed some flattening, but this was less noticeable than seen in relation to income, and occurred in countries with more than 60% of the population in urban areas. Urban living---which alters transportation and occupational patterns as well as access to various foods---may affect nutrition and activity, and increase population BMI and cholesterol, over and above its impacts mediated through income. Once a country is primarily urbanized, its infrastructure development reduces the urban--rural differences in access to food and technology \[[@pmed-0020133-b23]\], leading to the observed flattening of the curve at high urbanization levels. Exceptions to the relationship between cholesterol and urbanization were Indonesia and Nigeria (proportion urban 40% or more, but low cholesterol levels of 3.2--3.6 mmol/l). Both countries have large populations and, despite the formation of urban centers, still have predominantly agricultural economies and associated lifestyles. At the aggregate level, the relationships of mean BMI and cholesterol with each other or with the economic variables were qualitatively different from those of mean SBP ([Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"}). Mean population SBP was either not correlated or only very weakly correlated with cholesterol, BMI, and the economic variables. High SBP levels were observed in some developing countries (e.g., Gambia, Ghana, and Nigeria) that had low mean cholesterol or BMI. Similarly, mean population SBP varied by 20 mm Hg in the countries of Western Europe, which have relatively similar income. Individual-level epidemiological studies have established a relationship between BMI and blood pressure \[[@pmed-0020133-b24],[@pmed-0020133-b25]\]. Blood pressure is, however, affected by other aspects of diet, including salt and fruits and vegetables \[[@pmed-0020133-b26]--[@pmed-0020133-b30]\] that may vary across populations independently of BMI, based on cultural and environmental factors such as food preservation techniques and subsistence versus commercial access to agricultural products. Even in within-country studies, the relationships between blood pressure and wealth or the relative levels of rural and urban blood pressure have varied, and even reversed, in different countries \[[@pmed-0020133-b31]--[@pmed-0020133-b37]\]. Multi-variable regression analysis (results not shown) confirmed the absence of a relationship between blood pressure and the economic variables considered (the coefficients for SBP were not statistically significant). BMI and cholesterol both increased significantly with income and as the share of expenditure used for food declined, although the size of the association was attenuated after adjustment for other factors. Countries with a larger urban population had higher BMI even after adjustment for their income. Discussion {#s4} ========== Cardiovascular diseases have multiple well-established behavioral, environmental, and physiological determinants. Less well-established are the patterns of these risks at the population level in relation to one another and to economic variables such as income. Knowledge of these patterns is necessary, however, for better design of long-term policies and interventions that aim to reduce multiple risks that affect a common set of diseases in different populations, and for assessing the global health inequality dimensions of cardiovascular disease risks. Using data from over 100 countries, we found that BMI increased rapidly, then flattened, and eventually declined with increasing national income. Cholesterol had a similar, but more delayed, relationship with national income. Mean population blood pressure was not correlated or only weakly correlated with the economic factors considered, or with cholesterol and BMI. The rapid rise in BMI with income, and the lack of a relationship for blood pressure, illustrate that, currently, major transformations in patterns of cardiovascular risks occur at much earlier stages of economic development than implied by the "diseases of affluence" paradigm. If current income--risk relationships observed in [Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"} hold as economies grow, rapidly increasing BMI, coupled with the presence of elevated blood pressure at all income levels, will increasingly concentrate two major cardiovascular risk factors (blood pressure and obesity) in populations with currently low income levels, and all three risks in currently middle-income countries. The current income--risk relationships shown in [Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"}, however, only partially illustrate the potential future magnitude and global distribution of cardiovascular risks. There are reasons to suspect that the cross-sectional relationships shown in [Figures 2](#pmed-0020133-g002){ref-type="fig"} and [3](#pmed-0020133-g003){ref-type="fig"} may be changing over time, with important implications for the epidemiological transition. Longitudinal data from the United States and a small number of other countries show an upward shift in the entire income--BMI relationship ([Figure 4](#pmed-0020133-g004){ref-type="fig"}). At the same time, a downward shift in the relationship between blood pressure and cholesterol with income occurred during the 1980s, followed by stabilization or a slight increase in the 1990s for blood pressure, and continued decline for cholesterol ([Figure 4](#pmed-0020133-g004){ref-type="fig"}) \[[@pmed-0020133-b38],[@pmed-0020133-b39]\]. Definitive explanations for these trends are not available. Possible reasons for BMI increase include systematic changes in diet and physical activity due to increased access to private transportation, television, and manufactured/packaged foods as a result of technological change, urbanization, and organization of work \[[@pmed-0020133-b40],[@pmed-0020133-b41]\]. For blood pressure and cholesterol, possible contributors to the initial decline include changes in diet (increased access to fresh fruits and vegetables, and lower salt intake) and use of pharmacological interventions (anti-hypertensives and statins); increasing obesity may have been the obstacle to further blood pressure decline. If a similar upward shift in the income--BMI relationship occurs globally, overweight and obesity will play an even larger role in disease burden in developing countries, because these countries will be on an even higher income trajectory of BMI than shown in [Figure 3](#pmed-0020133-g003){ref-type="fig"}. As interventions for blood pressure and cholesterol are adopted, and exposure to these risks thus is lowered, in high-income societies, the three risk factors will become increasingly concentrated in low-income and middle-income nations relative to high-income countries. ::: {#pmed-0020133-g004 .fig} Figure 4 ::: {.caption} ###### Shifting Relationships of BMI, SBP, and Total Cholesterol with Income in the United States, Estimated Using Local Regression Data are from the National Health and Examination Survey, 1976--1980, 1988--1992, and 1999--2000. ::: ![](pmed.0020133.g004) ::: In additional to nutritional risks, tobacco smoking is also an important risk factor for cardiovascular diseases. Currently, an estimated 930 million of the world\'s 1.1 billion smokers live in the developing world \[[@pmed-0020133-b42]\]. Tobacco smoking increased among men, followed by women, in industrialized nations in the last century, and has subsequently declined in some nations (e.g., Canada, the United States, and the United Kingdom) \[[@pmed-0020133-b43]\]. Descriptive models based on historical patterns in the industrialized world \[[@pmed-0020133-b44]\] would predict a decline in male smoking and an increase in female smoking in the developing world over the coming decades. However, there have been major recent transformations in global tobacco trade, marketing, and regulatory control. As a result, tobacco consumption among men and women in most nations is primarily determined by opposing industry efforts and tobacco control measures \[[@pmed-0020133-b45]\], and by the socio-cultural context, rather than national income. Sex-specific data on tobacco consumption are very rare; the available data on smoking prevalence \[[@pmed-0020133-b42],[@pmed-0020133-b46]\] indicate little relationship with national income. The observed patterns of multiple nutritional risk factors also have implications for risk factor measurement and surveillance. Ideally, analysis of cardiovascular disease risks should use longitudinal data to assess the effects of socioeconomic factors on nutrition and nutritional risk factors both within and across populations. Detailed longitudinal data on nutrition (including size and composition of diet and frequency of eating) and physical activity \[[@pmed-0020133-b47],[@pmed-0020133-b48]\], or on associated indicator risk factors like blood pressure and cholesterol, are very scarce, especially in developing countries. Partly because of data limitations, the emphasis of global nutritional surveillance has been on dietary composition using aggregate date sources (e.g., the United Nations Food and Agriculture Organization\'s FAOSTAT; <http://apps.fao.org/>). Analyses of these aggregate data sources indicate that a shift towards "Western diets" high in saturated fat and sugar and low in fiber is occurring \[[@pmed-0020133-b02],[@pmed-0020133-b41]\]. Food composition and total caloric intake, although interrelated \[[@pmed-0020133-b49],[@pmed-0020133-b50]\], would have distinct influences on intermediate risks like BMI and cholesterol. The observed rapid BMI rise with national income indicates that dietary composition, which dominates global nutritional surveillance, should also be explicitly considered in relation to other determinants of weight such as frequency of eating and physical activity in leisure, occupation, and transportation domains \[[@pmed-0020133-b51]\]. Better data on food consumption and physical activity in turn require developing low-cost and valid instruments for large-scale population health surveys. Multi-risk assessment also demonstrates intervention needs and options in societies at different income levels. The observed rapid BMI increase with national income indicates that preventing obesity, which may be more effective than reacting after it has occurred \[[@pmed-0020133-b52]\], should be a priority during economic growth and urbanization. Overweight and obesity are also important because they cause a number of non-cardiovascular outcomes (cancers, diabetes, and osteoarthritis) which cannot be addressed by reducing risk factors such as blood pressure and cholesterol. Current intervention options for obesity in principle include those that reduce caloric intake (e.g., agriculture and food policy and pricing) and those that increase energy expenditure (e.g., urban design and transportation) \[[@pmed-0020133-b14],[@pmed-0020133-b40]\]. There is currently limited evidence on the community effectiveness of these interventions \[[@pmed-0020133-b52],[@pmed-0020133-b53]\]. This limitation highlights the need for research on design and evaluation of interventions for obesity---which in turn requires better data on the relative contributions of nutritional and physical activity to the current trends in weight gain---and on the obesity implications of policies and programs in sectors like transportation and agriculture. Multi-risk assessment also demonstrates that personal and population-level interventions for blood pressure and cholesterol (e.g., salt awareness and regulation, and pharmacological interventions) should be pursued together with attempts to curb or reduce obesity, because these interventions are effective in reducing the cardiovascular consequences of nutritional risk factors \[[@pmed-0020133-b54],[@pmed-0020133-b55]\]. The division between the diseases of poverty and affluence has provided a convenient tool for targeting policies towards risks such as undernutrition that affect the poor \[[@pmed-0020133-b56]\]. Demographic and technological change, however, are increasingly modifying the income patterns of cardiovascular risk factors and shifting their burden to the developing world. As a result, low-income and middle-income countries increasingly face the double burden of infectious disease and cardiovascular risk factors. Unless the research and intervention needs described earlier are pursued, this will create a world in which all major diseases are the diseases of the poor. Supporting Information {#s5} ====================== Table S1 ::: {.caption} ###### Countries Used in the Analysis and Data Availability (38 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Patient Summary {#sb1} --------------- ### {#sb1a} #### Background {#sb1a1} The pattern of cardiovascular diseases is influenced by many lifestyle factors such as diet, physical activity, work and leisure, and smoking. The effects of these factors are partly mediated through intermediate risk factors like overweight and obesity, blood pressure, and cholesterol. As societies grow richer, the patterns of risk-factor exposure change. Understanding these changes is important for health policies and interventions. #### What Did the Researchers Do? {#sb1a2} They looked at cardiovascular disease risks such as being overweight or obese, systolic blood pressure, and total cholesterol, and related them to national income, food purchase constraints, and urbanization. Body mass index (BMI) and cholesterol increased as national income increased, then flattened, and eventually declined. BMI increased until an income of about I\$12,500 (international dollars---a currency adjusted for differences in prices in different countries) for females and I\$17,000 for males, then flattened, and eventually declined. Cholesterol showed the same pattern, but with some delay. As the proportion of household spending devoted to food decreased (i.e., due to changes in income or the price of food), BMI and cholesterol levels increased. Also, as more people lived in cities, the population\'s BMI went up. #### What Do These Findings Mean? {#sb1a3} These findings are from a comparison of countries at one point in time, so they may only partly predict the path of individual countries over time. What they do suggest is that cardiovascular disease risks will increasingly be concentrated in low-income and middle-income countries. Therefore, preventing obesity should now be considered a priority even in these countries, along with measures to control blood pressure, cholesterol, and tobacco use. #### Where Can I Get More Information? {#sb1a4} The World Health Organization has a Web site on nutrition: <http://www.who.int/nut/> MedlinePlus has a selection of topics on obesity: <http://www.nlm.nih.gov/medlineplus/obesity.html> This work was sponsored by the National Institute on Aging (grants PO1-AG17625 and 1-P30-AG024409). The sponsor had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank Xu Ke for data to estimate the food share of household expenditure, Barry Popkin for information about BMI data sources, and Walter Willet for comments on an earlier version of this manuscript. Citation: Ezzati M, Vander Hoorn S, Lawes CMM, Leach R, James WPT, et al. (2005) Rethinking the "diseases of affluence" paradigm: Global patterns of nutritional risks in relation to economic development. PLoS Med 2(5): e133. BMI : body mass index SBP : systolic blood pressure WHO : World Health Organization [^1]: Competing Interests: The authors have declared that no competing interests exist. [^2]: Author Contributions: ME, SV, and CJLM designed the study. SV, CMML, RL, and WPTJ collected data. ME, SV, RL, and WPTJ analyzed the data. ME, SV, CMML, RL, WPTJ, ADL, AR, and CJLM contributed to writing the paper.	PubMed Central	2024-06-05T03:55:56.309341	2005-5-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088287/", "journal": "PLoS Med. 2005 May 3; 2(5):e133", "authors": [ { "first": "Majid", "last": "Ezzati" }, { "first": "Stephen", "last": "Vander Hoorn" }, { "first": "Carlene M. M", "last": "Lawes" }, { "first": "Rachel", "last": "Leach" }, { "first": "W. Philip T", "last": "James" }, { "first": "Alan D", "last": "Lopez" }, { "first": "Anthony", "last": "Rodgers" }, { "first": "Christopher J. L", "last": "Murray" } ] }
PMC1088941	Background ========== DNA microarrays \[[@B1],[@B2]\] have been used to profile gene expression in cancer and other diseases. In cancer, for example, microarray profiling has been applied to classify tumors according to their sites of origin \[[@B3]-[@B5]\], to discover previously unrecognized subtypes of cancer \[[@B6]-[@B11]\], to predict clinical outcome \[[@B12]-[@B14]\] and to suggest targets for therapy \[[@B15],[@B16]\]. However, the identification of improved markers for diagnosis and molecular targets for therapy will depend on knowledge not only of the genes expressed in the diseased tissues of interest, but also on detailed information about the expression of the corresponding genes across the gamut of normal human tissues. At present there is relatively little data on gene expression across the diversity of normal human tissues \[[@B17]-[@B20]\]. Here we report a DNA microarray-based survey of gene expression in a diverse collection of normal human tissues and also present an empirical method for estimating transcript abundance from DNA microarray data. Results ======= Hierarchical clustering of gene expression in normal tissues ------------------------------------------------------------ To survey gene expression across normal human tissues, we analyzed 115 normal tissue specimens representing 35 different human tissue types, using cDNA microarray representing 26,260 different genes (see Materials and methods). To explore the relationship among samples and underlying features of gene expression, we applied an unsupervised two-way (that is, genes against samples) hierarchical clustering method using the 5,592 cDNAs (representing 3,960 different UniGene clusters \[[@B21]\]) whose expression varied most across samples (Figure [1a](#F1){ref-type="fig"}; also see Additional data file 2). Overall, tissue samples clustered in large part according to their anatomic locations, cellular compositions or physiologic functions (Figure [1b](#F1){ref-type="fig"}). For example, lymphoid tissues (lymph node, tonsil, thymus, buffy coat and spleen) clustered together, as did gastrointestinal tissues (stomach, gall bladder, liver, pancreas, small bowel and colon), muscular tissues (heart and skeletal muscle), secretory tissues (parathyroid, thyroid, prostate, seminal vesicle and salivary gland), and female genitourinary tissues (ovary, fallopian tube, uterus, cervix and bladder). Brain and testis were also found to cluster together, largely because genes encoding ribosomal proteins and lymphoid-specific genes were expressed at particularly low levels in both tissues, the latter possibly reflecting immunological privilege \[[@B22]\]. The two-way unsupervised analysis also identified clusters of coexpressed genes (annotated in Figure [1](#F1){ref-type="fig"}), which represented both tissue-specific structures and systems (discussed further below) and coordinately regulated cellular processes. For example, on the basis of the shared characteristics of well annotated genes in the clusters, we identified clusters representing cell proliferation \[[@B23]\], mitochondrial ATP production, mRNA processing, protein translation and endoplasmic reticulum-associated protein modification and secretion. Interestingly, proliferation, mitochondrial ATP production and protein translation were each represented by two distinct clusters of genes, suggesting that subsets of these functions might be differentially regulated among different tissues. One gene cluster corresponded to sequences on the mitochondrial chromosome \[[@B24]\]; we interpret this feature to reflect the relative abundance of mitochondria in each tissue sample. Identifying tissue-specific gene expression ------------------------------------------- While tissue-specific gene expression features were apparent in the hierarchical cluster, in order to identify tissue-specific genes more systematically we performed supervised analyses using the significance analysis of microarrays (SAM) method (\[[@B25]\], see Materials and methods). Tissue-specific genes were identified for all tissues analyzed, and included named genes with known tissue-specific functions, as well as named genes and anonymous expressed sequence tags (ESTs) that had not been previously characterized as having tissue-specific functions. For example, while the set of liver-specific genes (Figure [2](#F2){ref-type="fig"}) included, as expected, genes encoding blood-clotting factors (for example, F2, F7), complement components (C1R, C2), lipid (APOB, APOE) and metal transport proteins (TF, CP), and proteins for detoxification (CYP2D6, CYP3A7), amino acid metabolism (PAH, HAL) and carbohydrate metabolism (G6PT1, GYS2), other intriguing genes, for example, WRNIP1(Werner helicase interacting protein 1), BIRC5(survivin), ANGPTL3(angiopoietin-like 3), and CNTNAP1(contactin associated protein 1), were also identified as selectively expressed in liver. The new connections these results might make between our knowledge of the gene and its product on the one hand, and our knowledge of the physiological functions, cellular characteristics and pathologies of a specific organ, on the other, are a step towards better understanding of both the genes and the organs. Interestingly, we also identified a smaller number of genes displaying selectively decreased expression in some organs, for example, splicing factor SF3B1in the liver (Figure [2b](#F2){ref-type="fig"}): we speculate that the decreased expression of such genes may have a role in regulating cellular/tissue differentiation. Tissue-specific genes characteristically expressed in each of the tissues we examined are viewable in Additional data file 6 (see also Additional data file 3). Recent efforts by the Gene Ontology (GO) Consortium have resulted in the systematic annotation of genes, ascribing genes to specific biological processes, cellular components and molecular functions \[[@B26]\]. This annotation system, while rudimentary, facilitates the systematic exploration of the expression of genes reflecting specific biological processes, cellular components and molecular functions in these normal tissues. For example, the gene sets encoding tyrosine kinase, G-protein-coupled receptor and transcription factor functions, as well as components of the extracellular matrix and the process of programmed cell death, each demonstrate tissue-specific patterns of expression (Figure [3](#F3){ref-type="fig"}; see also Additional data files 4 and 7). Estimating transcript abundance ------------------------------- DNA microarray experiments are often performed as comparative two-color hybridizations, permitting precise quantification of the ratio of each gene\'s expression between two samples. In the experiments reported here, each tissue sample was compared by hybridization to the same \'common reference\' mRNA (see Materials and methods), a standard experimental design permitting the comparison of expression across all samples \[[@B27]\]. Therefore, the primary measurements give us a precise picture of the variation in relative levels of each gene\'s expression among the samples. While this information is sufficient for many purposes, a quantitative comparison of the expression levels of transcripts of different genes is also of interest, for example in selecting especially highly expressed genes for potential diagnostic markers or therapeutic targets. Single-channel fluorescence intensities can provide a crude estimate of the relative transcript abundance of different genes, but do not control for the variable quantities of spotted DNA. To estimate transcript levels for our dataset, we used microarray hybridization to compare the common reference mRNA against normal female genomic DNA. We reasoned that, for each gene on the microarray, the ratio of mRNA to genomic DNA should reflect the relative level of transcript in the common reference compared to normal genomic DNA (for which each gene is present in two copies per cell). For each tissue sample in our study, the ratio of expression for each gene in that sample versus common reference mRNA, multiplied by the ratio for that gene in common reference mRNA versus normal genomic DNA, would then approximate transcript abundance. To test our approach, we compared our estimates of transcript levels for a single prostate specimen, calculated either indirectly using the common reference mRNA versus genomic DNA ratios, or calculated through a direct hybridization comparison of prostate sample mRNA versus normal female genomic DNA. Our results show high concordance for the prostate sample (Figure [4a](#F4){ref-type="fig"}); comparable results were obtained in a similar analysis using liver, breast, heart and kidney specimens (data not shown). The utility of this approach is illustrated for the cluster of prostate-specific genes (derived from the hierarchical cluster in Figure [1](#F1){ref-type="fig"}), and is evident on comparing results depicting the relative level of each gene\'s expression in different samples (Figure [4b](#F4){ref-type="fig"}), and the relative levels of transcripts for different genes (Figure [4c](#F4){ref-type="fig"}). While all genes within the prostate-specific cluster were expressed at relatively increased levels in prostate compared with other tissues, estimates of transcript abundance indicated that only a subset of these genes was highly expressed in the prostate (Figure [4c](#F4){ref-type="fig"}). For example, RDH11was highly expressed in prostate and was expressed at lower levels in other tissues, while STEAP2was expressed at low levels in prostate and displayed very little or no expression in other tissues. For each of the tissue types, transcripts identified as both highly abundant and tissue specific are displayed in Additional data files 5 and 8 (for the transcript levels of all variably expressed genes, see Additional data file 2). Discussion ========== The main objective of our study was to survey variation in gene expression across a diverse set of normal human tissue types. We have reported here a cDNA microarray gene-expression dataset profiling approximately 26,000 human genes across 115 human tissue specimens representing 35 different tissue types. An unsupervised, two-way hierarchical clustering of the genes whose expression varied most across samples showed that at the level of gene expression, the relationship among tissues was in large part based on their anatomic locations, cellular compositions and physiologic functions. Tissue-specific features of gene expression were readily discernable in the hierarchical cluster, as were gene-expression features related to specific cellular processes (as inferred from the named genes within these features). Of particular importance, the function of uncharacterized ESTs might be deduced by virtue of their inclusion in one of these clusters. Supervised analysis also identified genes selectively expressed in each of the tissues types studied, and the analysis of functionally annotated gene sets provided information on the tissue distribution of specific biological processes, cellular components and molecular functions. We have also reported here the application of mRNA versus genomic DNA hybridizations for estimating transcript abundances for expressed genes. Knowledge of transcript abundance should prove useful in prioritizing candidate genes for use as diagnostic markers or therapeutic targets, for which more highly expressed genes might be more tenable candidates. It is worth pointing out that our approach for estimating absolute transcript levels should be applicable to any cDNA microarray study incorporating a common reference mRNA. While many investigators have been using DNA microarrays to profile gene expression in cancer and other human diseases, scant data exist on profiles of gene expression across the diversity of normal human tissues. Our cDNA-microarray-based survey of gene expression in normal human tissues provides a publicly accessible dataset which can be used in future analyses aimed at better understanding the physiology of various normal tissues; developing a baseline for comparison to diseased tissues, including cancer; identifying tissue-specific diagnostic markers that signify tissue injury; discovering tissue-specific therapeutic targets (for example, for treatment of prostate cancer); and identifying tumor-specific diagnostic markers and therapeutic targets, for which minimal expression in the collection of normal adult human tissues is desirable. Conclusions =========== We have used cDNA microarrays to survey gene expression across a diverse set of normal human tissues. Using unsupervised and supervised analyses, we have identified tissue-specific patterns of gene expression. Furthermore, by comparative hybridization to normal genomic DNA, we were able to estimate transcript abundances and identify the subsets of abundantly expressed tissue-specific genes. Our dataset provides a baseline for comparison to diseased tissues, as well as a basis for identifying molecular markers of injury to specific organs and tissues, and for anticancer therapy. Materials and methods ===================== Tissue specimens ---------------- Normal human tissue specimens were obtained from surgery (for example, the uninvolved regions of resected tumors) or from autopsy, with institutional review board approval. Specimens were frozen on dry ice within 30 minutes of surgical removal or procurement and stored at -80°C. Histological evaluations were performed by H&E staining of frozen sections, and a pathologist (J.H. and/or M.vdR.) reviewed all slides to confirm the anatomical site of origin and histological normalcy (that is, to rule out inflammation, infection, necrosis, malignancy). In total, we selected for study 115 tissue samples representing 35 different human tissues (Additional data file 1). Total RNA was isolated from tissues using TRIzol Reagent (Invitrogen) according to the manufacturer\'s instruction, and RNA quality was assessed by the integrity of rRNA bands following gel electrophoresis. The poly(A)^+^mRNA fraction was then isolated from total RNA using FastTrack2.0 kit (Invitrogen), and quantified by UV spectrophotometry. Expression profiling -------------------- Gene-expression profiling was performed essentially as reported previously \[[@B8]\], and detailed protocols for array fabrication and hybridization are available online \[[@B28]\]. Briefly, Cy5-labeled cDNA was prepared using 2 μg mRNA from normal tissue samples, and Cy3-labeled cDNA was prepared using 1.5 μg mRNA common reference, pooled from 11 established human cell lines \[[@B8]\]. For each experimental sample, Cy5- and Cy3-labeled samples were co-hybridized to a cDNA microarray containing 39,711 human cDNAs, representing 26,260 different genes (UniGene clusters \[[@B21]\]). For the common reference mRNA (Cy5) versus genomic DNA (Cy3) comparisons, normal female genomic DNA was labeled as described \[[@B24]\]. Following hybridization, microarrays were imaged using an Axon GenePix 4000 scanner (Axon Instruments). Fluorescence ratios for array elements were extracted using GenePix software, and uploaded onto the Stanford Microarray Database (SMD) \[[@B29]\] for subsequent analysis. The complete microarray dataset is accessible from SMD \[[@B30]\], or from the Gene Expression Omnibus \[[@B31]\] (accession number GSE2193). Data analysis ------------- Fluorescence ratios were normalized by mean-centering genes for each array (that is, \'global\' normalization), and then by mean centering each gene across all arrays. We included for analysis only well-measured genes whose expression varied, as determined by: signal intensity over background more than twofold in either test or reference channels in at least 75% of samples; and a fourfold or more ratio variation from the mean in at least two samples (unless otherwise indicated). Hierarchical clustering was performed and displayed using Cluster and TreeView software \[[@B32]\]. Tissue-selective genes were identified using the two-class (each tissue versus all other tissues) significance analysis of microarrays (SAM) method \[[@B25]\], which utilizes a modified t-test statistic and sample-label permutations to evaluate statistical significance. The false-discovery rate (FDR), an estimate of the fraction of falsely called tissue-selective genes, varied by tissue, but in all cases was less than 5% (specific FDRs are listed in Additional data file 3). For tissue-selective genes, only tissue types with two or more samples were considered for analysis, and we only considered genes that were well-measured in more than 50% of the samples for the selected tissue type analyzed. GO annotations were assigned to arrayed genes using the AmiGO browser \[[@B33]\] to select relevant GO annotations, and the \'loc2go\' file \[[@B34]\] to identify the corresponding sets of genes. Transcript abundance was estimated by multiplying (for each gene) the ratio of tissue sample mRNA versus common reference mRNA by the ratio (average ratio from triplicate experiments) of common reference mRNA versus normal female genomic DNA. Highly-abundant tissue specific transcripts were defined for each tissue type as the top (capped at 50 genes) tissue specific transcripts, identified using the SAM method, from the 1,000 most abundantly expressed transcripts in the full dataset. Additional data files ===================== The following additional data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} is a table listing the normal tissue specimens included in microarray analysis. Additional data file [2](#S2){ref-type="supplementary-material"} is a table listing the variably expressed genes. Additional data file [3](#S3){ref-type="supplementary-material"} is a table listing tissue-specific transcripts. Additional data file [4](#S4){ref-type="supplementary-material"} is a table listing functionally annotated gene sets. Additional data file [5](#S5){ref-type="supplementary-material"} is a table listing highly abundant tissue-specific transcripts. Additional data file [6](#S6){ref-type="supplementary-material"} is a figure showing tissue-specific gene expression. Additional data file [7](#S7){ref-type="supplementary-material"} is a figure showing expression of functionally annotated gene sets. Additional data file [8](#S8){ref-type="supplementary-material"} is a figure showing highly abundant tissue-specific gene expression. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A table listing the normal tissue specimens included in microarray analysis ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A table listing the variably expressed genes. Sheet 1: Dataset represented in Fig. [1](#F1){ref-type="fig"}, which includes well-measured genes in ≥ 75% of samples, and with ≥ 4-fold ratio variation from the mean in at least 2 samples; samples ordered by clustering.Sheet 2: Variably expressed genes which are well-measured in ≥ 75% of samples, with ≥ 2-fold ratio variation from the mean in at least 2 samples; samples ordered by anatomic site. Sheet 3: Variably expressed genes which are well-measured in ≥ 25% of samples, with ≥ 4-fold ratio variation from the mean in at least 2 samples; samples ordered by anatomic site. Sheet 4: Same dataset as sheet 1, but here ratios represent relative transcript abundance (See Materials and methods). Sheet 5: Same dataset as sheet 2, but here ratios represent relative transcript abundance (See Materials and methods). Sheet 6: Same dataset as sheet 3, but here ratios represent relative transcript abundance (See Materials and methods) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 A table listing tissue-specific transcripts ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 A table listing functionally annotated gene sets ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 A table listing highly abundant tissue-specific transcripts ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 A figure showing tissue-specific gene expression. Variably-expressed genes determined to be expressed in a tissue-selective fashion using the SAM method are depicted as described in the legend to manuscript Figure [2](#F2){ref-type="fig"}. a, brain; b, salivary gland; c, esophagus; d, stomach; e, small bowel; f, colon; g, pancreas; h, liver; i, heart; j, skeletal muscle; k, lung; l, kidney; m, bladder;n, prostate; o, seminal vesicle; p, testis; q, ovary; r, fallopian tube; s, uterus; t, cervix, u, thyroid; v, parathyroid; w, adrenal; x, lymph node; y, tonsil; z, thymus; aa, spleen; bb, buffy coat ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 A figure showing expression of functionally annotated gene sets. Hierarchical cluster of 115 normal tissue specimens and annotated gene sets representing examples of specific molecular functions, cellular components, or biological processes. a, tyrosine kinase (activity); b, kinase (activity); c, G-protein coupled receptor (activity); d, transcription factor activity; e, ion channel (activity); f, extracellular matrix (component), g, cell adhesion (process); h, programmed cell death (process) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 A figure showing highly abundant tissue-specific gene expression. Highly-abundant tissue specific transcripts were defined for each tissue type as the top (capped at 50 genes) tissue specific transcripts, identified using the SAM method, from the 1000 most abundantly expressed transcripts in the full dataset. a, brain; b, salivary gland; c, esophagus; d, stomach; e, small bowel; f, colon; g, pancreas; h, liver; i, heart; j, skeletal muscle; k, lung; l, kidney; m, bladder;n, prostate; o, seminal vesicle; p, testis; q, ovary; r, fallopian tube; s, uterus; t, cervix, u, thyroid; v, parathyroid; w, adrenal; x, lymph node; y, tonsil; z, thymus; aa, spleen; bb, buffy coat ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Ash Alizadeh and the members of the Pollack and Brown labs for helpful suggestions. We also thank Janet Mitchell and the Stanford Tissue Bank for collection of tissues, Mike Fero and the staff of the Stanford Functional Genomics Facility for providing high-quality cDNA microarrays, and Gavin Sherlock and Catherine Ball of the Stanford Microarray Database group for providing outstanding database support. This work was supported by a grant from the National Cancer Institute. P.O.B is an investigator of the Howard Hughes Medical Institute. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Hierarchical cluster analysis of normal tissue specimens. (a)Thumbnail overview of the two-way hierarchical cluster of 115 normal tissue specimens (columns) and 5,592 variably-expressed genes (rows). Mean-centered gene expression ratios are depicted by a log~2~pseudocolor scale (ratio fold-change indicated); gray denotes poorly-measured data. Selected gene-expression clusters are annotated. The dataset represented here is available as Additional data file 2. (b)Enlarged view of the sample dendrogram. Terminal branches for samples are color-coded by tissue type. ::: ![](gb-2005-6-3-r22-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Liver-specific gene expression. (a)Thumbnail overview of a hierarchical cluster of 115 normal tissue specimens and 353 variably expressed genes identified using the SAM method (see Materials and methods) as selectively expressed in liver (false discovery rate = 0.12%). Genes are hierarchically clustered, while samples are grouped by tissue type and ordered according to anatomical location/function. Mean-centered gene-expression ratios are depicted by a log~2~pseudocolor scale (indicated); samples are color-coded by tissue type. (b-d)Selected gene-expression clusters (locations indicated by vertical colored bars). Because of space limitations, only named genes (and not expressed sequence tags (ESTs)) are indicated. Tissue-specific genes identified for other tissues are available as Additional data files 3 and 6. ::: ![](gb-2005-6-3-r22-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Brain-selective expression of functionally annotated gene sets. Hierarchical cluster of 115 normal tissue specimens and annotated gene sets representing the following examples of (a-c)specific molecular functions (a) tyrosine kinase, (b) G-protein-coupled receptor, (c) transcription factor, (d)cellular components (extracellular matrix) or (e)biological processes (programmed cell death). Samples are ordered as in Figure 2. Genes are ordered by hierarchical clustering. For gene selection, we considered genes that were well measured in at least 50% of samples; no ratio-fold cutoff was applied. Only features representing brain-specific expression are shown here; the complete clusters are available as Additional data files [4](#S4){ref-type="supplementary-material"} and [7](#S7){ref-type="supplementary-material"}. ::: ![](gb-2005-6-3-r22-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Estimating relative transcript abundance. (a)Comparison of transcript levels estimated either directly by hybridization of prostate sample mRNA versus normal female genomic DNA, or indirectly by multiplying the ratio of prostate sample mRNA vs common reference mRNA by the ratio of common reference mRNA vs normal female genomic DNA. The correlation value (R) is indicated. (b)Prostate-specific gene-expression cluster, extracted from the hierarchical cluster shown in Figure 1a, is displayed as mean-centered relative gene expression (ratio-fold change scale indicated). (c)The same gene-expression feature as in (b), is now displayed as transcript abundance (relative to the average transcript level for all expressed genes), calculated indirectly using the common reference mRNA versus normal female genomic DNA hybridization data. ::: ![](gb-2005-6-3-r22-4) :::	PubMed Central	2024-06-05T03:55:56.311936	2005-2-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088941/", "journal": "Genome Biol. 2005 Feb 14; 6(3):R22", "authors": [ { "first": "Radha", "last": "Shyamsundar" }, { "first": "Young H", "last": "Kim" }, { "first": "John P", "last": "Higgins" }, { "first": "Kelli", "last": "Montgomery" }, { "first": "Michelle", "last": "Jorden" }, { "first": "Anand", "last": "Sethuraman" }, { "first": "Matt", "last": "van de Rijn" }, { "first": "David", "last": "Botstein" }, { "first": "Patrick O", "last": "Brown" }, { "first": "Jonathan R", "last": "Pollack" } ] }
PMC1088942	Background ========== Large-scale sequencing projects continue to generate a growing number of new genomes from an ever-wider range of species. A rarely noted and unappreciated side effect of some projects occurs when the organism being sequenced contains an intracellular endosymbiont. In some cases, the existence of the endosymbiont is unknown to both the sequencing center and the laboratory providing the source DNA. Fortunately, many genome projects deposit all their raw sequence data into a publicly available, unrestricted repository known as the Trace Archive \[[@B1]\]. By conducting large-scale searches of the Trace Archive, one can discover the presence of these endosymbionts and, with the aid of bioinformatics tools including genome assembly algorithms, reconstruct some or most of the endosymbiont genomes. The amount of endosymbiont DNA present in a genome deposited in the Trace Archive depends on several factors: the number of sequences generated by the project, the size of the host genome, the size of the endosymbiont genome, and the number of copies of the endosymbiont present in each cell of the host. Because the copy number varies among cell types, the amount of endosymbiont DNA also depends on the preparation method used to extract host DNA; for example, the use of eggs or early-stage embryos will yield much greater amounts of Wolbachiafrom its hosts, because the bacterium occurs in much higher copy numbers in egg cells than in other cell types \[[@B2]\]. If the host genome is 200 million base-pairs (Mbp) in length, and the endosymbiont is 1 Mbp, and if there is one endosymbiont per host cell, then 0.5% of the sequences from a random sequencing project of the host will derive from the endosymbiont. The critical factor is the copy number per cell: regardless of genome size, if there is one endosymbiont genome per cell, then the endosymbiont will be sequenced to the same depth of coverage as the host, and the genome assembly will, in theory, cover both genomes to the same extent. The search for these hidden genomes is aided greatly by the availability of a complete genome of a related species. Fortunately, the complete genome of Wolbachia pipientis wMel, an endosymbiont of D. melanogaster\[[@B3]\], is available to aid the search. Wolbachiaspecies are common obligate intracellular parasites that infect a wide variety of invertebrates, including not only fruit flies but also mosquitoes, arthropods and nematodes \[[@B4],[@B5]\]. Results and discussion ====================== Using the 1,267,782 bp wMel genome as a probe, we searched the Trace Archive entries of seven recently sequenced Drosophilaspecies, each of which was sequenced to approximately eightfold coverage. For three of these species, we found clear evidence of Wolbachiainfections in the host. From the 2,772,509 traces of Drosophila ananassae\[[@B6]\], we retrieved 32,720 sequences that either matched the wMel strain or were paired with sequences that matched wMel (see Materials and methods). Our assembly of these sequences yielded a new genome, Wolbachia wAna, containing 1,440,650 bp in 329 separate scaffolds, at approximately eightfold coverage. At this coverage depth, we estimate that 98% of the wAna genome is included in the assembly. The alignment of the wAna scaffolds to wMel covers approximately 878 kbp (70%) of the 1.27 Mb wMel genome. A mapping of all the individual wAna reads to wMel gives greater coverage - 1.11 Mbp (87%) of the wMel genome. From the 2,214,248 traces of D. simulans\[[@B7]\], we retrieved and assembled 3,727 sequences. The resulting genome fragments of Wolbachia wSim cover 896,761 bp of wSim at twofold coverage, which we estimate to cover 65-80% of wSim. The comparative assembly (see Materials and methods) resulted in 388 contigs plus 241 singleton sequences, and a separate scaffolding program further grouped 273 of these contigs into 84 scaffolds. The alignment between wSim and wMel covers 861 kbp (65%) of the wMel genome. From the 2,445,065 traces of D. mojavensis\[[@B6]\], we retrieved 101 sequences matching wMel, plus another 13 sequences that did not match wMel but were paired with the matching sequences. The sample is too small for assembly, but even so it represents approximately 87 kb (6-7%) of the Wolbachia wMoj genome. No Wolbachiasequences were found in the other Drosophila species currently available: D. pseudoobscura, D. yakuba, D. virilisand D. melanogaster. Wolbachiahas previously been described to infect multiple strains of D. simulans, and a fragment of the 16S ribosomal RNA gene has been sequenced (GenBank ID AF312372) \[[@B8]\]. It has also been described in D. ananassae\[[@B9]\], but has not been previously reported in D. mojavensis(and no sequences can be found in the Wolbachiadatabase maintained at \[[@B10]\]). Genome organization ------------------- Comparison of the wAna and wMel species indicates extensive rearrangements between the genomes. This is best illustrated with the longest scaffold in wAna, which contains 455,845 bp, approximately one-third of the genome. Figure [1](#F1){ref-type="fig"} shows a map of this scaffold compared to the wMel genome. The scaffold spans more than a dozen rearrangements that have occurred since the divergence of these species. We also found evidence of rearrangements within our wAna sequences (see Materials and methods), indicating that the D. ananassaestrain may have been infected with two or more divergent Wolbachiastrains. The rearrangements shown in Figure [1](#F1){ref-type="fig"} are typical of the interstrain alignments; breakpoints occur even among the very sparsely sampled wMoj sequences. Although only 101 sequences matched wMel, seven of these spanned either insertions or large-scale rearrangements in the wMel genome. Genome comparisons ------------------ In these assemblies, approximately 464, 92 and 6 genes were discovered in the wAna, wSim and wMoj genomes, respectively (see Additional data file 1), that were not found in the previously reported W. pipientis wMel genome. Of these novel genes, 343 were conserved hypothetical proteins, 81 transposases, 13 phage-related proteins and seven ankyrin domain proteins. Of the remaining 118 genes, 34 are proteins from the wAna assembly of insect origin, which are likely to represent Drosophilacontaminants as a result of chimeric inserts in the original sequencing library. Another 51 predicted genes are shorter than 300 bp and may not constitute real genes. The remaining 33 genes have similarity to known genes and include genes that have tentatively been identified to be involved in transport, DNA binding or regulation, and a variety of other functions. Many of the unique genes have anomalous GC content, suggesting horizontal gene transfer (HGT), with 12 genes displaying a GC content greater than 50% as opposed to the typical 35% GC content found in these genomes and wMel (Table [1](#T1){ref-type="table"}). Consistent with the observation that novel genes in the new Wolbachiastrains tend to be hypothetical proteins, genes present in wMel that are absent in the wAna assembly are also predominantly hypothetical proteins. Of the 347 wMel genes not found in wAna, 207 were hypothetical proteins, with the next highest category being mobile elements and extrachromosomal elements, with 37 genes. This suggests that as much as 27% of the predicted genes in wMel could be highly variable. Two large gene clusters in W. pipientis wMel were not identified in the wSim and wAna assemblies (Figure [2](#F2){ref-type="fig"}). This could suggest absence or divergence of these regions. The lack of the recovery of two of the regions (A and B) is interesting as both regions contain genes that have been suggested to affect host-endosymbiont interactions \[[@B3]\]. Region A includes the 3\'-region of the WO-A phage and the region directly downstream. It includes the interval containing genes WD0289-WD0296, which encodes four hypothetical proteins - three ankyrin repeat domain proteins and a conserved hypothetical protein. The absence of WD0289-WD0292 is interesting because it may suggest some variation in the phage 3\'-region. Although WD0289-WD00291 is unique to WO-A, a protein homologous to WD0292 has been found in the previously described Wolbachiaphage \[[@B3],[@B11]\]). Variation in the Wolbachiaphage could facilitate the introduction of novel genes \[[@B12]\]. As ankyrin repeat proteins, WD0291, WD0292, and WD0294 are all of interest as they have been proposed to be involved in host-interaction functions \[[@B3]\]. This could provide a means by which the phage could cause different host-interaction phenotypes. Region B includes WD0509-WD0514, which encodes a DNA mismatch repair protein MutL-2, a degenerate ribonuclease, a conserved hypothetical protein, two hypothetical proteins and an ankyrin repeat domain protein. This region is of further interest since WD0511-WD0514 is found only in W. pipientis wMel and not the related sequenced Anaplasmataceae, Rickettsiaceae or α-Proteobacteria. In W. pipientis wMel, this region is flanked on the 3\'-end by an interrupted reverse transcriptase and an IS5 transposase, supporting the hypothesis that it was acquired horizontally. The absence of MutL-2 might not be functionally important since wMel, wAna, and wSim all have a copy of MutL-1. Evolutionary comparisons ------------------------ We aligned all genomes to one another to find those sequences shared by all four strains. Because W. pipientis wMoj comprises the smallest sample, we used the 114 sequences from that strain as a query to search the other three strains, and found 90 sequences shared among all strains. We then created four-way multi-alignments for each of these 90 sequences (see Materials and methods). Excluding the large insertions and deletions discussed above, the strains are highly similar, as summarized in Table [2](#T2){ref-type="table"}. As the table shows, the two most closely related strains are wAna and wSim, which are nearly identical at the DNA level. Both wMel and wMoj are approximately equidistant from these two strains, at just over 97% identity, but are more distant from one another. Note however that because the wMoj sequences are single reads (that is, single-pass sequencing), the error rate in these sequences is substantially higher than in the assembled genomes of the other strains, which in turn may make it appear that wMoj is more divergent. Ankyrin repeat domain proteins ------------------------------ Ankyrin repeat proteins showed considerable variability among the four Wolbachiastrains. It has been proposed that ankyrin repeat proteins may influence the host by regulating host cell cycle, regulating host cell division, and interacting with the host cytoskeleton \[[@B3]\]. These genes and their relationship to cell cycle, and therefore reproduction, are likely candidates for involvement in host interactions like cytoplasmic incompatibility, male killing, parthenogenesis and feminization. There were four ankyrin repeat proteins absent in wAna and wSim in the Regions A and B above. There were also seven new ankyrin repeat proteins identified in wAna, wSim, and wMoj. In order to infer a relationship between the ankyrin repeat proteins, all the ankyrin repeat-containing proteins greater than 120 amino acids in length were aligned and clustered using ClustalW. The amino-acid sequences were too diverse to permit the construction of a reliable phylogenetic tree. But a tree was drawn that clustered similar proteins and allowed for the classification of families of conserved ankyrin repeat domain proteins within the Wolbachialineage (Figure [3](#F3){ref-type="fig"}). From this tree, several classes of proteins can be determined that are highly conserved between two or more of these Wolbachialineages with greater than 95% similarity at the nucleotide level. In addition, ankyrin repeat domain proteins unique to a particular lineage can also be identified. These differences in the complement of ankyrin repeat domain proteins may affect host-endosymbiont interactions. Comparison with other obligate intracellular bacteria ----------------------------------------------------- The variability of genome content and synteny identified here with Wolbachiais in contrast to that observed for other obligate intracellular bacteria. Comparative analysis of the Chlamydiaceae shows that the genomes of these organisms are highly conserved in terms of content and gene order, with relatively small differences in the genomes \[[@B13]\]. This is despite the fact that the chlamydial genomes sequenced thus far span four distinct species from various hosts and cause different tissue tropism and disease pathology. Similarly, rickettsial genomes have a high degree of synteny and gene conservation with the exception of numerous unique sequences in the genome of Rickettsia conorii\[[@B14]\]. Although R. conoriimaintains synteny with Rickettsia prowazekiiand Rickettsia typhi, it has 560 unique genes relative to the other two. In contrast, the sequencing of R. typhirevealed only 24 novel genes. Wolbachiagenomes seem to have little synteny \[[@B3]\] and large variations in genome size and genome content. This may reflect the levels of intraspecies contact in vivo. Wolbachiaare abundant in nature, are able to co-infect arthropods \[[@B15],[@B16]\], and are propagated by vertical and horizontal transmission \[[@B17]\]. Phylogenetic analysis of the WO-B phage shows that under conditions of co-infection, Wolbachiafrom different supergroups will share the same WO-B phage \[[@B12]\]. These factors may promote genetic exchange between Wolbachiaspecies. In addition, the Wolbachialifestyle of facilitating its own transmission by host reproductive modification may then promote the successful transmission of genetically diverse strains. Other obligate intracellular bacterial genera may find the series of events involving successful co-infection, exchange of genetic information, and then propagation more challenging and therefore less likely. Horizontal gene transfer ------------------------ The presence of endosymbionts within host cells, particularly germline cells, may offer opportunities for HGT, although in general such transfer between prokaryotes and eukaryotes is extremely rare \[[@B18]\]. However, a number of studies have clearly documented cases of transfer of mitochondrial DNA into the nuclear genome \[[@B19]\], in species as diverse as yeast \[[@B20]\], Arabidopsis thaliana\[[@B21]\] and other plants \[[@B22]\], and human \[[@B23]\]. The mitochondrial organelle itself is widely believed to derive from an ancestral endosymbiont \[[@B19],[@B24]\]. Although we do not here provide evidence for HGT from Wolbachiato Drosophila, at least one recent study claims that a Wolbachiaendosymbiont has transferred genes to the X chromosome of an insect, the adzuki bean beetle \[[@B25]\]. The analysis of the wMel genome examined this question, but did not find any evidence for HGT into the D. melanogasterhost \[[@B3]\]. Conclusions =========== The discovery of these three new genomes demonstrates how powerful the public release of raw sequencing data can be. Although none of these projects had as its goal the sequencing of bacterial endosymbionts, we now have as a result three partial genomes - one nearly complete - of this biologically important species. The differences between these genomes and the completed wMel strain demonstrate extensive genome rearrangement and divergence among these Wolbachiaendosymbionts. And although it is a small sample, when taken together the presence of these three new genomes indicates that Wolbachiaendosymbionts appear to be quite common in the Drosophilalineage. Multiple future Drosophilasequencing projects are planned, several of which are already underway, as are projects to sequence other invertebrates, many of which may host Wolbachiaor other endosymbionts. Our results suggest that new screening methods, such as those described here, may yield unexpected discoveries from the data in the Trace Archive. Materials and methods ===================== We downloaded from the Trace Archive at NCBI \[[@B1]\] the following numbers of raw sequences from each Drosophila species: 2,772,509 sequences from D. ananassae; 2,445,065 from D. mojavensis; 2,214,248 from D. simulans; 2,061,010 from D. yakuba; 3,359,782 from D. virilis; 2,590,703 from D. pseudoobscura; and 3,663,352 from D. melanogaster. For each project, we downloaded sequences, quality values, and ancillary data (containing clone-mate information, clone insert lengths, and sometimes trimming parameters), comprising approximately 2-3 gigabytes (GB) of compressed data per genome. For each genome, we used the nucmer program from the MUMmer package \[[@B26]-[@B28]\] to search the complete genome of W. pipientis wMel against the files containing the sequences. We pulled out any single sequence (\'read\') with at least one 30-bp exact match to wMel, and with an extended match that spanned at least 65 bp. We then retrieved the \'clone mates\' of each sequence: most of the reads in whole-genome sequencing projects are obtained via a double-ended shotgun method, meaning that both ends of each clone insert are sequenced. The Trace Archive contains a link to the clone mate for each read; we used this information to extract any mates that were not contained in our original screen. For example, the D. ananassaedata yielded approximately 5,000 additional reads when we pulled in the mates from the original set. We then assembled the Wolbachiareads in two different ways: with the Celera Assembler \[[@B29]\], treating it as a normal (de novo) whole-genome assembly, and with the AMOS-cmp assembler \[[@B30]\], which assembles a genome by mapping it onto a reference. For the reference genome we used wMel. We used Celera Assembler on the relatively well-covered wAna strain; although we ran it on the wSim reads as well, the sequence coverage was too light to yield a good assembly. The high degree of sequence identity, at 95-100% across most regions that are shared between strains, allowed for an excellent comparative assembly of the wSim strain with AMOS-cmp. The AMOS-cmp assembly of wSim contains 388 contigs plus another 241 singleton reads, covering 896,761 bp (see Table [1](#T1){ref-type="table"}). The largest contig contains 16,701 bp. Note that AMOS-cmp produces contigs but not scaffolds. The contigs can easily be aligned to the reference genome to produce scaffolds, with the caveat that any rearrangements will invalidate such scaffolding information. To avoid such problems, we ordered and oriented the contigs separately with Bambus \[[@B31]\], a stand-alone genome scaffolding program, using only the clone-mate information from the original shotgun data. Bambus created 84 multi-contig scaffolds that joined together 273 of the 388 contigs, with the largest scaffold containing 50,851 bp and spanning (including estimated gaps) 54,207 bp. For wAna, when we compared the de novoand comparative assemblies, we observed that there were multiple rearrangements in the wAna genome as compared to wMel. Our conclusion was that a comparative assembly, which relies on the genome structure of the reference, may be less accurate than a de novoassembly in the presence of extensive rearrangements, so we used the latter for our analysis. The wAna assembly presented special challenges because of what appear to be a large number of rearrangements and polymorphisms within the sequences. The number of Wolbachiareads provided very deep coverage, which in principle should have produced a scaffold that covered nearly the entire genome. However, a large number of clone-mate links were inconsistent with one another, indicating that the reads may have been drawn from a population in which many of the individuals had genome rearrangements with respect to one another. We also found locations spanning hundreds of nucleotides where four or five individual reads had one nucleotide and the same number had a different nucleotide. These polymorphisms made it difficult to create many consistent large scaffolds. We created multiple assemblies in which we removed many of the inconsistent links, and eventually settled on the assembly presented here as the best representative of the genome possible given the diversity in the data. The wAna assembly has three large scaffolds of 460 kb, 157 kb, and 121 kb respectively, with all remaining scaffolds less than 20 kb in length. We also include a list of all the individual sequences, including those not incorporated into contigs, in our Additional data files. To annotate the resulting sets of contigs, we used Glimmer \[[@B32],[@B33]\] to make initial gene calls and BLAST \[[@B34]\] to search those calls against a comprehensive protein database. Regions with no gene calls were searched as well in all six reading frames using Blastx. All the predicted genes in wAna, wSim, and wMoj were searched against wMel using Blastn. The results of these searches were used to determine what genes are absent in the wAna, wSim, and wMoj assemblies. DNA sequence matches at 80% identity for 80% length of the smaller of the genes were determined to be conserved and are plotted in Figure [2](#F2){ref-type="fig"}. Regions A and B in Figure [2](#F2){ref-type="fig"} were identified in this manner. To identify the unique genes in the wAna, wSim, and wMoj assemblies, all predicted proteins were searched against the wMel proteins using Blastp. Proteins in the new genomes were considered unique (or highly divergent) when the best match in wMel had an E-value greater than 10^-15^. To create the multiple alignments of the 90 sequences that were shared by all four organisms, we searched the 114 sequences in wMoj against the wMel, wAna, and wSim genome assemblies, again using nucmer. We used the output of nucmer to extract from each genome the appropriate matching sequence, and we fed the results to the overlapper (hash-overlap) from the AMOS assembler \[[@B30]\] to generate all pairwise sequence alignments. All ankyrin repeat domain proteins identified by automated annotation were compiled and an alignment and tree were constructed using ClustalW \[[@B35]\]. The ankyrin repeat domain is a degenerate repeat \[[@B36]\], so no attempt was made to cluster proteins where the ankyrin repeat motifs were removed. The whole-genome shotgun assemblies, with annotation, have been deposited at DDBJ/EMBL/GenBank under the project accession AAGB00000000 (wAna) and AAGC00000000 (wSim). The versions described in this paper are the first versions, AAGB01000000 and AAGC01000000. The sequences and annotation for wMoj have consecutive accessions AY897435 through AY897548. The unassembled wMoj reads are also available from the Trace Archive and from the Additional data files for this paper. Additional data files ===================== The following additional data is available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} contains four tables: the first three list the unique genes in the wAna, wSim and wMoj genomes respectively; the fourth lists the Trace Archive identifiers for the 114 reads comprising the wMoj sequences from the D. mojavensisgenome project. Additional data file [2](#S2){ref-type="supplementary-material"} is a multi-fasta file containing the sequences of the 114 wMoj reads. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Supplementary Tables 1, 2, and 3 listing the unique genes in the wAna, wSim and wMoj genomes respectively and Supplementary Table 4 listing the Trace Archive identifiers for the 114 reads comprising the wMoj sequences from the D. mojavensisgenome project ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 The sequences of the 114 wMoj reads ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Hean Koo for help with genome data management, and Hervé Tettelin and Martin Wu for helpful comments on the manuscript. We also thank Agencourt Bioscience, the Washington University Genome Sequencing Center and the NIH for making sequence data publicly available through the NCBI Trace Archive. S.L.S., A.L.D., and M.P. were supported in part by the NIH under grants R01-LM06845 and R01-LM007938 to SLS. J.D.H. was supported by funds from National Science Foundation Frontiers in Integrative Biological Research under grant EF-0328363. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Alignment of complete wMel genome (horizontal axis) to longest scaffold from the wAna genome assembly. Red points indicate sequences aligned in the forward orientation, green points indicate reverse orientation. The diagonals represent colinear regions, and breaks in the diagonals correspond to inversions and translocations between the two genomes. ::: ![](gb-2005-6-3-r23-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Circular map comparing the wMel genome with the wAna, wSim and wMoj assemblies. Ring 1 (outermost ring): forward strand genes; ring 2: reverse strand genes; ring 3: GC-skew plot; ring 4: X^2^analysis of trinucleotide composition, with peaks indicating atypical regions; ring 5: wMel genes present in wAna assembly; ring 6: wMel genes present in the wSim assembly; ring 7: wMel genes present in wMoj assembly. Large regions on the wMel genome that were not recovered in the wAna or wSim assemblies are marked on the outside (regions A, B). ::: ![](gb-2005-6-3-r23-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Relationship of ankyrin repeat domain proteins between wMel, wAna, wSim and wMoj. All the predicted ankyrin repeat proteins with greater than 120 amino acids were aligned and clustered using ClustalW. Nine predicted ankyrin repeat domain proteins (A-I) were found to be conserved among at least wMel and one other of these Wolbachiaspecies with nucleotide sequence identity \> 95% across the entire length of the gene. ::: ![](gb-2005-6-3-r23-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary statistics for assemblies of the three new Wolbachiagenomes ::: wAna wSim wMoj wMel -------------------------- ----------- --------- -------- ----------- Molecule length (bp) 1,440,650 896,761 86,870 1,267,782 Scaffolds 329 84 114 1 Genes 1837 790 63 1271 Contigs 464 388 114 1 GC content (%) 35.4 35.0 34.5 35.2 Average gene length (bp) 608 916 633 855 The wSim genome was assembled using the comparative assembler, AMOS-Cmp, and scaffolded using Bambus. The wAna genome was assembled using the Celera Assembler, as described in Materials and methods. Note that the high gene count for wAna is likely due to fragmentation of individual genes across separate contigs. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Percent identity between nucleotide sequences of the four sequenced strains of Wolbachia ::: wMel wAna wSim -------- -------- -------- -------- wMel 97.2 97.1 wAna 97.2 99.8 wSim 97.1 99.8 wMoj 94.9 97.5 97.3 :::	PubMed Central	2024-06-05T03:55:56.314177	2005-2-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088942/", "journal": "Genome Biol. 2005 Feb 22; 6(3):R23", "authors": [ { "first": "Steven L", "last": "Salzberg" }, { "first": "Julie C Dunning", "last": "Hotopp" }, { "first": "Arthur L", "last": "Delcher" }, { "first": "Mihai", "last": "Pop" }, { "first": "Douglas R", "last": "Smith" }, { "first": "Michael B", "last": "Eisen" }, { "first": "William C", "last": "Nelson" } ] }
PMC1088943	Background ========== Cell-surface G-protein-coupled receptors (GPCRs) bind exogenous as well as endogenous ligands such as photons, odorants, lipids, nucleotides, hormones, pheromones, peptides and proteins. Interaction with these ligands drives diverse processes such as photoreception, taste and olfactory sensations in animals, mating in fungi and cell-cell communications in slime molds \[[@B1]-[@B3]\]. These receptors are characterized by seven transmembrane α-helices that upon ligand binding relay the signal by bringing about conformational changes in bound G proteins. The extracellular amino terminus in most cases interacts with the ligand and the carboxyl terminus with G proteins. The G proteins in turn activate different signaling pathways, such as those activated by adenylate cyclase and phospholipase C. These GPCRs are of immense importance as they are major targets for drug discovery \[[@B4]\]. A classification scheme that encompasses all GPCRs is the grouping into classes A-E \[[@B5]\]. A-C are the main classes present in animals: class A is the largest and comprises the rhodopsin-like receptors, class B comprises the secretin-like receptors and class C the metabotropic glutamate/pheromone receptors. Class D is unique to fungi and comprises fungal pheromone receptors. Class E contains cAMP receptors, such as the cAMP receptors of Dictyostelium. Other classes include frizzled/smoothened, adhesion receptors and the insect-specific chemosensory receptors \[[@B6],[@B7]\]. Sequence conservation between GPCR classes is limited, however, with each receptor class exhibiting specific identifiable characteristics \[[@B6],[@B8]\]. The secretin and the adhesion receptors are characterized by conserved cysteine residues or by known cysteine-rich domains resembling the epidermal growth factor (EGF) domain at their amino termini. GPCRs form the largest family of receptors in animals, with more than 600 members in the human genome \[[@B9],[@B10]\]. Only a handful of GPCRs have been identified in fungal genomes, however. In Saccharomyces cerevisiaeand Schizosaccharomyces pombeonly three and four receptors, respectively, are well characterized \[[@B1],[@B11]-[@B16]\]. In the Neurospora crassagenome a total of 10 receptors is predicted \[[@B17]\]. A recent report for Aspergillus nidulansidentified GPCRs similar to the yeast pheromone receptors, the glucose-sensing receptor GPR1, the nitrogen-starvation sensing STM1, and the Dictyostelium discoideumcAMP receptors \[[@B18]\]. Given the prevalence and significance of GPCRs in higher eukaryotes, their relative paucity in the kingdom Fungi warranted further investigation. To see if we could find additional families, we searched the predicted proteome of the rice blast fungus Magnaporthe grisea. The fungal plant pathogen M. griseais a powerful model system to study the pathogenicity determinants required for plant cell-surface recognition and production of an appressorium, a specialized structure required to penetrate the plant surface \[[@B19],[@B20]\]. The fungus causes rice blast disease, the most destructive disease of rice worldwide. M. griseais amenable to molecular genetic manipulation and the subject of large-scale genome-wide functional studies following the recent completion of a draft genome sequence \[[@B21]\]. Infection begins when a conidium, attached to the plant surface, sends forth a germ tube that differentiates to form a highly melanized appressorium. Turgor pressure inside the appressorium results in a penetration hypha breaching the cell wall and invasion of the plant tissues. This developmental program, which is accompanied by a number of biochemical and developmental changes, is a result of perception by the fungus of appropriate environmental and plant cell-surface signals and induction of a cascade of signaling pathways. Cell-surface receptors that perceive signals at critical times in the life cycle of M. griseaand other pathogenic fungi are strongly implicated as pathogenicity determinants. Signaling plays a key role in appressorium formation and infection in M. grisea. The cAMP-dependent and pheromone response, as well as other mitogen-activated protein kinase (MAPK)-, phospholipase- and calmodulin-dependent pathways, are essential for pathogenicity and are likely to involve perception of signals through GPCRs \[[@B22]-[@B24]\]. The three identified G-protein alpha subunits, required for different aspects of development and pathogenicity, possibly transduce perceived signals to the above-mentioned pathways \[[@B25]\]. The M. griseaG proteins probably receive signals from receptors such as PTH11, an integral membrane protein required for pathogenicity \[[@B26]\]. As animal GPCRs are important targets for drug discovery, identifying fungal receptors would be equally important for understanding and controlling M. griseaand other fungal pathogens. Identification of new GPCR classes is difficult because of low sequence similarity; even within related classes, sequence conservation is limited to the membrane-spanning regions \[[@B8]\]. There are also large variations in the type and number of receptors in classes that show no sequence or structural similarities to each other. We therefore carried out an exhaustive analysis to mine the proteome of the sequenced genome of the rice blast fungus M. griseafor GPCR-like proteins. Homologs of known fungal GPCRs were found in the M. griseaproteome, including the pheromone receptors STE2 and STE3 and the glucose-sensing receptor GPR1. In total, 76 GPCR-like proteins were identified in the present study of which 61 represent a large novel class related to PTH11, a receptor implicated in fungal development and pathogenicity and proposed to act upstream of the cAMP-dependent pathway. Many of these novel receptors will have roles in known pathways or may define new pathways involved in fungal development. Results ======= Identification of novel classes of GPCR-like proteins in M. grisea -------------------------------------------------------------------- We searched the M. griseaproteome for GPCR-like proteins on the basis of their similarity to known receptors. GPCR sequences including all present in the GPCR database (GPCRDB \[[@B5]\]) were used as a query in a BLASTP search against the M. griseapredicted protein set \[[@B21]\]. The proteins retrieved in this search were used to BLAST the M. griseaproteins again to find all related sequences (Table [1](#T1){ref-type="table"}). A total of 14 GPCR-like proteins were found. These included homologs of characterized fungal GPCRs (GPR1, STM1, and the STE2- and STE3-like pheromone receptors). Other proteins identified were similar to the cAMP receptors and to mPR, a steroid receptor. No homologs of the animal rhodopsin-, secretin- and metabotropic-like receptor classes, which form the majority of the proteins in GPCRDB, could be found. All proteins listed in the table were checked to make sure they had seven transmembrane regions (Additional data file 1). The M. griseaproteins were searched with InterProScan \[[@B27]\] and 16 proteins associated with InterPro entries containing the terms \'GPCR\' or \'G protein-coupled receptors\' were identified. Four were already identified in the above BLAST searches. Of the remaining 12, only one (MG00532.4) had seven transmembrane regions and was added to Table [1](#T1){ref-type="table"}. This protein had weak similarity to rat growth hormone-releasing factor receptor and other GPCRs. A PfamA HMM search revealed that some of the proteins identified above had characteristic GPCR domains (Table [1](#T1){ref-type="table"}, and see \[[@B28]\]). The receptor PTH11 in M. griseais required for development of the appressorium \[[@B26]\]. It is an integral membrane protein and has been localized to the cell membrane. It is proposed to act upstream of the cAMP pathway, which is required for pathogenicity. The PTH11 amino-terminal domain contains an EGF-like cysteine rich CFEM domain, predicted to be extracellular, followed by seven transmembrane regions \[[@B29]\]. Based on the transmembrane topology, with the amino-terminal outside and the carboxy-terminal inside, PTH11 is a novel GPCR-like protein. PTH11 has been reported to have nine transmembrane regions; however, the two putative transmembrane regions at the amino-terminal end are the predicted signal sequence and the hydrophobic region within the extracellular CFEM domain, respectively, and are therefore not membrane spanning \[[@B26],[@B29]\]. The CFEM domain is an EGF-like domain, characteristically present in the extracellular regions of membrane proteins; thus PTH11 is characterized as having an extracellular amino-terminal CFEM domain, followed by seven transmembrane regions. A BLASTP search using PTH11 as query against known M. griseaproteins retrieved a number of proteins with seven transmembrane regions (E-value cutoff of 1e-09). A BLASTP search using these PTH11-related proteins against M. griseapredicted proteins returned additional members within this class (total 61, Table [1](#T1){ref-type="table"}). Only a subset of the retrieved proteins contained the CFEM domain, as indicated in Table [1](#T1){ref-type="table"} (12 CFEM-containing proteins). In total we identified 76 receptors, including members of known classes as well as novel classes. Sixty-one represented a novel class that included PTH11. All other receptors identified were assigned to different classes on the basis of their similarity to known receptors using BLASTP against the GenBank (nonredundant) and Swiss-Prot databases, and their conserved domain characteristics. We found three members of the mPR class and one (MG0532.4) with weak similarity to animal GPCRs. No members of these classes have been reported previously in fungi. Within each class, members were assigned to paralogous families (Table [1](#T1){ref-type="table"}). Many of the genes in Table [1](#T1){ref-type="table"} are expressed, as suggested by representation in expressed sequence tags or microarray experiments. A BLAST search against the GenBank EST databases revealed that some of the predicted open reading frames (ORFs) had matches in the M. griseaESTs (Table [1](#T1){ref-type="table"}, and see \[[@B30]\]). Results from microarray experiments on gene expression during conidia germination and appressorium formation also showed that many of these ORFs are expressed (T.K. Mitchell and R.A.D, unpublished work). Shared and unique GPCR-like protein classes in M. grisea ---------------------------------------------------------- M. griseaGPCRs were compared with published fungal genome sequence databases to identify proteins belonging to the same GPCR classes. A BLASTP search against the genome of the closely related filamentous fungus N. crassa\[[@B17]\], using all the M. griseaGPCR-like proteins as query, revealed the presence of similar proteins in N. crassa, including PTH11 homologs (Table [2](#T2){ref-type="table"} and Additional data file 2). No PTH11 homologs were found in S. cerevisiaeand S. pombe. Further analysis revealed putative homologs of the mPR-1 class in both yeasts, in which they had not previously been identified. In addition, we found no evidence for cAMP receptor-like GPCRs in either yeast, unlike both M. griseaand N. crassa. The cAMP, STM1, and mPR receptors are shared between fungi and other eukaryotic species. However, the fungal pheromone receptors (class D) and GPR1-like receptors appear to be fungus-specific. Members of the large class of PTH11-related receptors were restricted to a fungal subphylum. BLASTP of all the PTH11 class members, and PSI-BLAST using conserved regions, against the GenBank (nonredundant) and Swiss-Prot databases and publicly available fungal genomes retrieved matches in members of the subphylum Pezizomycotina within the Ascomycota, including Podospora anserina, Blumeria graminis, Fusarium graminearumand Aspergillusspecies. Other fungi belonging to the Ascomycota but not to this subphylum, such as S. cerevisiae, S. pombe, Candida albicansand Pneumocystis cariniilacked PTH11-related sequences. Also, no PTH11-related sequences were found in the genomes of the Basidiomycetes Cryptococcus neoformans, Ustilago maydisand Phanerochaete chrysosporium. No matches were found in plant, animal or prokaryotic genomes. Phylogenetic analysis of PTH11-related GPCR-like proteins in M. griseaand N. crassa --------------------------------------------------------------------------------------- PTH11-related receptors from M. griseaand N. crassawere classified into paralogous families (Additional data file 2). We also identified any that were orthologs between these two species. PTH11-related receptors in M. griseaand N. crassaand other sequences from P. anserinaand B. graminiswere aligned to determine any relationships. The region containing the conserved PTH11-domain was used to build a phylogenetic tree (Figures [1](#F1){ref-type="fig"}, [2a](#F2){ref-type="fig"}). Our analysis indicated that PTH11-related proteins form a large and divergent protein family that evolved before the divergence of M. griseaand N. crassa. M. griseaand N. crassaorthologs occurred in the same clades (Figure [1](#F1){ref-type="fig"}). Many different clades on the tree may represent paralogous sequences. The tree supports the putative orthologs and paralogs we identified (see Additional data file 2). Even though only the PTH11 domain was used to build the tree (the amino-terminal CFEM domain seen in a few proteins was not included), the 13 CFEM domain-containing proteins occurred together in one clade, indicating that the sequences are closely related. The phylogeny also revealed that within certain clades there was a marked expansion of the PTH11-related proteins in M. griseacompared to N. crassa. This is particularly notable for the CFEM domain-containing proteins. There were six M. griseamembers containing the CFEM domain in a paralogous family (Table [1](#T1){ref-type="table"} and Figure [1](#F1){ref-type="fig"}; a total of 12 related CFEM and seven-span proteins), but only one from N. crassa. We found 38 PTH11-related proteins in A. nidulanswith an E-value less than 1e-09. Further characterization of these proteins will be required to define the number of seven-span PTH11-related proteins in this genome. Preliminary analysis shows that only two seven-span proteins contain the CFEM domain in A. nidulans. These observations could represent either expansion since speciation of the CFEM-containing PTH11 relatives in M. grisea, or loss of these proteins in the other fungal species. New domain signatures as defined by conserved regions in homologous classes of identified receptors --------------------------------------------------------------------------------------------------- Members of each class of M. griseaGPCR-like proteins described above, for example, cAMP-, STM1-like, PTH11-related receptors, have domains that are conserved within each class. Sequence alignments from the BLASTP searches revealed specific regions containing shared residues for each of these classes of receptors. Figure [2](#F2){ref-type="fig"} shows an alignment for some of the sequences that belong to classes other than the better-studied pheromone and glucose-sensing receptors. In all the PTH11-related members the region towards the amino terminus was conserved (Figure [2a](#F2){ref-type="fig"}, PTH11\_dom). The extreme amino-terminal and the carboxy-terminal sequences flanking this region were divergent. Conserved residues occurred within the seven-span regions for all of these proteins. This is consistent with other observations that sequence conservation is typically limited to the transmembrane regions in GPCRs. The M. griseaprotein MG06738.4, which has similarity to the cAMP receptors, shared conserved amino-acid residues between positions 81-179 with MG06797.4, MG00326.4, MG06257.4, related N. crassaproteins and other cAMP receptors (cAMP\_dom, Figure [2b](#F2){ref-type="fig"}). Other proteins - MG00258.4 and MG10544.4 - with weak similarity to cAMP receptors also shared residues within this domain (data not shown). MG04698.4 shared two domains between amino-acid residues 22-101 and 244-327 with STM1, MG02855.4 and related proteins from different eukaryotic species (stm1\_dom, Figure [2c](#F2){ref-type="fig"}). MG05072.4 shared residues within the region of 56-277 with MG09091.4 (residues 18-228), MG04679.4 (residues 260-497) and other proteins that were retrieved in the BLAST search, including mPR receptors (mPR\_dom, Figure [2d](#F2){ref-type="fig"}). The proteins containing the PFAM GPCR domains are indicated in Table [1](#T1){ref-type="table"}. It is worth noting that the low scores for the PFAM domains that we observed may be due to the need to update these domain alignments by adding many new proteins, including those we discovered. For example, MG06452.4 contains a putative STE3 domain; the alignment score (E-value) is low, however. With the new fungal genomes being sequenced, more STE3 homolog sequences are available and inclusion of these in the seed alignment defining the STE3 domain will make the domain more representative for fungal STE3 domains. Each class of receptors contained specific conserved regions within the membrane-spanning topology. A representative example of each class, showing the location of the conserved region within the membrane topology is illustrated in Figure [3](#F3){ref-type="fig"}. For fungus-specific receptors, the conserved domain spanned almost the entire length of the seven transmembrane regions. When other eukaryotic receptors were included in the class, however, only shorter conserved domains were discerned. These conserved residues may reflect functional constraints and may be valuable for studying the structure-function relationships of these proteins. Discussion ========== Distinct classes of GPCR-like proteins identified in M. grisea ---------------------------------------------------------------- Fungi respond to a variety of signals from the environment that regulate cellular metabolism and development as well as host-pathogen interactions. Cell-surface receptors perceive these signals and relay them to intracellular signaling pathways. We searched the proteome of M. griseafor GPCR-like proteins and identified a total of 76 sequences (Table [1](#T1){ref-type="table"}). This is the largest number of GPCR candidates identified for any fungal species. The identified proteins in M. griseainclude homologs of known fungal receptors and a few other eukaryotic receptors. Putative orthologs of fungal STE2- and STE3-like pheromone receptors required for the mating responses in yeast were identified. A homolog of GPR1, which is involved in pseudohyphal differentiation in S. cerevisiae, and two proteins that share similarities with STM1 from S. pombewere also found \[[@B11],[@B13],[@B16]\]. Six proteins shared similarities with cAMP receptors from Dictyostelium. In Dictyosteliumthe cAMP receptors are involved in establishing polarity during chemotaxis \[[@B3]\]. All the above M. griseaproteins can be annotated as putative GPCRs on the basis of homology to known receptors. It is likely that they respond to similar ligands, such as pheromones, nutrients and cAMP (Table [1](#T1){ref-type="table"}). Response to fungal mating pheromones and the existence of pheromone receptors in M. griseawas first suggested by the observation that M. grisearesponded to S. cerevisiaepheromones in a mating-type-specific manner \[[@B22]\]. Intracellular cAMP, produced by adenylate cyclase, is a critical factor regulating appressorium development in M. grisea. Lee and Dean have found that the fungus will respond to exogenously added cAMP by development of appressoria, although the concentrations required are high \[[@B31]\]. They noted that the cell wall and cell membrane should be relatively impermeable to cAMP, and thus any responses to extracellular cAMP will be due to cAMP receptors. Further research will be required to learn about the mechanism of perception of exogenous cAMP and other ligands and their targets within the cell. PTH11-related proteins share a number of characteristics diagnostic of GPCRs and define a new class of GPCR-like proteins. The predicted membrane topology suggests a seven-span protein with an amino terminus outside the cell, that could respond to extracellular signals, and a cytoplasmic carboxy-terminal domain that could interact with G proteins. All the PTH11-related proteins shared conserved residues within the membrane spans, as observed in other GPCRs classes \[[@B8]\]. A subclass of the PTH11 receptors showed another characteristic that is seen in a few classes of human GPCRs: they have an amino-terminal cysteine-rich EGF-like CFEM domain. The animal secretin receptors are characterized by six conserved cysteines at the amino terminus, with cysteine bridges implicated in ligand binding. Some of the adhesion receptors have cysteine rich-EGF-like domains at their amino termini \[[@B6],[@B8]\]. CFEM-domain-containing proteins, which are smaller in size and lack the seven transmembrane regions, may interact with the CFEM-containing GPCR-like proteins (Additional data file 3 and \[[@B29]\]). The CFEM-containing proteins have a signal peptide and/or a glycosylphosphatidylinositol (GPI) anchor. Thus they are either secreted from the cell or are anchored to the cell membrane. They may be similar to the odorant-binding proteins, which also have cysteine-rich domains and have been proposed to interact with odorant-GPCRs \[[@B32]\]. Unique classes of fungal G-protein-coupled receptors with ancient origins ------------------------------------------------------------------------- Having diverged approximately 1,460 million years ago (Mya) \[[@B33]\], it is clear that fungi have classes of GPCRs that are distinct from those of animals. The class D fungal pheromone receptors define a fungus-specific class of receptors. We found the GPR1-like receptors to be also fungal specific. Classes of receptors specific to a group of species also occur in animals. For example, some of GPCRs in Anopheles gambiaeconstitute an insect-specific class of chemosensory receptors \[[@B7]\]. Insects are estimated to have diverged from other animals nearly 1,000 Mya. Thus, we would expect to find novel fungal GPCRs with no similarities to ones present in other eukaryotic kingdoms. The largest class of M. griseaGPCR-like proteins we identified is the novel PTH11-related class. It is interesting that we only found homologs of PTH11 in fungi belonging to subphylum Pezizomycotina within the Ascomycota (this subphylum has an estimated divergence date of 1,140 Mya). None was found in fungi belonging to other subphyla in Ascomycota or Basidiomycota, estimated to have diverged from each other 1,210 Mya. This indicates that these proteins are extremely ancient in origin, having possibly evolved to serve specialized functions in a specific subgroup of fungi. They are either unique to this fungal group or have evolved sufficiently to be unrecognizable. Relationships between the PTH11-related proteins ------------------------------------------------ The PTH11-related proteins form a large and divergent protein family, as suggested by the similarity between the proteins and the phylogenetic tree (Table [1](#T1){ref-type="table"}, Figure [1](#F1){ref-type="fig"}). This gene family may have evolved before the divergence of M. griseaand N. crassa. There are a few orthologs between these species; however, it is apparent that this family has undergone considerable expansion in M. griseacompared to N. crassa, with the largest subclass in M. griseabeing the CFEM-containing proteins. Many of the PTH11-related genes are located in close proximity to each other on the genome (data not shown), whereas none of the other GPCR-like proteins, except a pair of cAMP-receptor-related proteins, occurs in close proximity. A paralogous pair, MG07553.4 and MG07565.4, occurs close together on linkage group III, indicating that these genes may have arisen as a result of duplication. We blasted these sequences against each other and observed that they show 30% identity with an E-value of 7e-54. This suggests that even if these genes are a result of duplication, they have diverged sufficiently and are not incorrect duplicate predictions of the same gene due to sequencing or assembly errors. Both these genes contain the CFEM domain and also occur in the same clade on the phylogenetic tree (Figure [1](#F1){ref-type="fig"}). Another pair of CFEM-containing proteins is located in close proximity (LGI, group 1). The above examples of relative expansions within the PTH11-related proteins, as compared to N. crassa, are an indication that gene duplication may still be occurring in M. grisea. In N. crassait is believed that because of the phenomenon of repeat induced point mutations (RIP), gene duplications are not maintained \[[@B17]\]. There is evidence of RIP in M. grisea, but the present study provides an example that has escaped the RIP process \[[@B34]\]. Other possibilities are that these genes duplicated before the evolution of RIP or have escaped RIP because M. griseararely undergoes meiosis in the wild. Regulation of the activity of GPCR-like proteins by differential expression and interaction with different signal transducers ----------------------------------------------------------------------------------------------------------------------------- Differential expression and interaction with different signal transducers could be a way to regulate specific signaling pathways. Results from genome-wide microarray experiments suggest different patterns of expression for the GPCR-like receptors during growth and development (T.K. Mitchell and R.A.D, unpublished work). Representation of some of the GPCR-like receptors in the fungal ESTs and microarray experiments suggests that most of these genes are expressed (Table [1](#T1){ref-type="table"}). In addition to differential regulation of the GPCR-like proteins, their interaction with different G proteins could channel various signals to different pathways. As well as the well studied interactions with G proteins, it has been proposed that the seven-span receptors may also interact with other signal transducers and receptor-interacting proteins to transmit the signal to different cellular pathways. Conclusion ========== The number of classes of GPCR-like proteins identified in the present study is the largest reported in fungi. Further research on these receptors will help delineate potentially novel signaling pathways with which they interact. The new class of PTH11-related receptors, specific to an Ascomycota subphylum and relatively numerous in M. grisea, is particularly interesting. PTH11 is an integral membrane protein localized to the cell membrane and is required for pathogenicity \[[@B26]\]. It is proposed to act upstream of the cAMP pathway as a receptor that channels signals to this pathway. PTH11 does not have an ortholog in N. crassa. Also, as discussed earlier, only one CFEM-containing seven-span protein is present in N. crassacompared to 12, including PTH11, in M. grisea. It remains to be determined whether other members of this expanded class of PTH11-related proteins are involved in different aspects of pathogenicity. The subphylum Pezizomycotina includes the majority of known ascomycete species, and includes pathogens and mutualists. Because PTH11-related GPCR-like proteins are present in non-pathogens, many members of this class are likely to be involved in functions not related to pathogenesis. All the seven-span receptors and their characteristic domain signatures we discovered (Figures [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}) will be valuable in the identification and comparative studies of new receptors in the many fungal genomes being sequenced. Materials and methods ===================== Identification of GPCR-like proteins in Magnaporthe grisea ------------------------------------------------------------ Known GPCR sequences, including ones present in the GPCRDB \[[@B5]\], were BLASTed against the predicted M. griseaproteome to identify homologs in M. grisea\[[@B21]\]. The database containing 7,900 GPCR sequences (updated 28 May 2003) was used as a query in a BLASTP search against the M. griseapredicted proteins with an E-value limit of 1e-09. Results from an InterPro scan of the M. griseaproteins were searched for domains containing the following terms: \'GPCR\' and \'G-protein-coupled receptors\' \[[@B27]\]. M. griseaPTH11, a GPCR-like protein (see Results), was also used in a BLASTP search against the M. griseaproteome. BLAST and PfamA searches and related sequence analysis were done using Genomax (Informax (now Invitrogen)). Characterization of the GPCR-like proteins and identification of additional members in M. griseaand other fungi ----------------------------------------------------------------------------------------------------------------- GPCR-like sequences were evaluated for seven transmembrane regions by TMPRED, Phobius and TMHMM \[[@B35]-[@B37]\]. Default settings were used. In nearly all cases at least two of the algorithms predicted the seven-span helix topology (Additional data file 1). A BLASTP search using the seven-span polypeptide sequences as query against the M. griseaprotein set was also done to identify any other similar members. The set of identified seven-span proteins was then subject to BLASTP analysis against GenBank and Swiss-Prot to confirm sequence similarity to GPCRs. This exercise also allowed identification of other members that were similar to these sequences. The M. griseaseven-span proteins identified as above were used as a query in a BLAST search against the N. crassapredicted proteins \[[@B17]\] to identify homologs. The M. griseaand N. crassaproteins were placed into clusters using the blastclust program \[[@B38]\]. All M. griseaand N. crassaproteins that had at least 30% identity and 80% overlap over the length of the proteins were clustered together. Members of the same species within a cluster were considered paralogs. Orthologs were defined as proteins that had bidirectional best BLAST hits. A TBLASTN search using the seven-span containing sequences as query against the GenBank EST database was performed to identify any identical matches in the M. griseaESTs (or other closely related fungal sequences). The GPCR-like sequences identified in M. griseawere used as query in BLASTP searches (cutoff \< 1e-09) against the S. cerevisiaeand S. pombegenomes and other completely sequenced fungal genomes to identify putative homologs in these species. Alignments and phylogenetic relationships between the predicted GPCR sequences ------------------------------------------------------------------------------ The alignment of sequences within related classes in Figure [2](#F2){ref-type="fig"} was done using T\_Coffee and minor editing as per results from the BLAST alignments was done using GenDoc \[[@B39]\]. For phylogenetic analysis, the conserved PTH11-domain that spans the membrane-spanning regions was used. Sequences were aligned using ClustalW version 1.81 \[[@B40]\]. The phylogenetic tree was constructed using PAUP by both neighbor-joining and parsimony methods followed by bootstrap analysis (100 bootstrap replications). A tree was also constructed using the neighbor-joining method implemented in the software package MEGA 2.1 \[[@B41]\]. All methods showed similar relationships between the proteins. Additional data files ===================== The following additional data is available with the online version of this paper: additional data file [1](#S1){ref-type="supplementary-material"} is a table listing M. grisea-GPCR-like protein accession numbers and seven-span predictions; additional data file [2](#S2){ref-type="supplementary-material"} is a table listing M. grisea-GPCR-like protein classes and N. crassahomologs; additional data file [3](#S3){ref-type="supplementary-material"} is a table listing M. griseaCFEM-containing proteins that may be membrane associated or secreted. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 M. grisea-GPCR-like protein accession numbers and seven-span prediction. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 M. grisea-GPCR-like protein classes and N. crassahomologs. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 M. griseaCFEM-containing proteins that may be membrane associated or secreted. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Hemant Kelkar, Center for Bioinformatics, University of North Carolina, for providing helpful comments, and members of the Fungal Genomics Laboratory for valuable discussions. The research was supported by funds from the United States Department of Agriculture (award 2001-52100-11317) and the National Science Foundation (award 0136064). We are grateful to other fungal research communities, particularly Aspergillus nidulansresearchers, for giving us access to unpublished genome sequence data. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Gene phylogeny based on the conserved membrane-spanning PTH11-domain. The tree shown was constructed using parsimony methods. Numbers on branches represent bootstrap values based on 100 random dataset simulations. Open ovals indicate putative paralogs and filled ovals the M. grisea-N. crassaorthologs. For sequences other than the ones predicted from M. griseaand N. crassagenome sequences the GenBank accession numbers are indicated. The abbreviations for species names are indicated in parentheses after the accession numbers as follows: BG, Blumeria graminis; PA, Podospora anserina; NC, N. crassa. The product of the gene PTH11was referred to as Pth11p in the original report. Subsequently it has been referred to as PTH11. We refer to this gene product as PTH11 in this paper and would like to propose revision of its name from Pth11p to PTH11. ::: ![](gb-2005-6-3-r24-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Alignment of GPCR-like proteins. Domains conserved in (a)PTH11-, (b)cAMP-, (c)STM1- and (d)mPR-related classes are shown. Representative sequences from each class were aligned using T\_Coffee \[39\]. The alignment was analyzed using GenDoc. We used the default setting using the conservative shading mode with similarity groups enabled. Black and the dark and light gray represent 80% or greater conserved, 60% or greater conserved, and less than 60% conserved, respectively. Conservative substitutions were counted as a single residue type. The GenBank or Swiss-Prot (SP) accession numbers or the accession numbers of the predicted proteins in the M. griseaor N. crassagenome databases are indicted on the left \[21, 42\]. The boundaries of each sequence used in the alignment are indicated on the right. ::: ![](gb-2005-6-3-r24-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Membrane topology of M. griseaGPCR-like proteins. The figure shows representative examples from different classes with domains that are conserved with respect to other receptors of the same class. Known Pfam domains or domains conserved between the M. griseaprotein and other members of the class, as shown in Figure 2, are shaded in black. The amino-acid residue numbers that mark the boundaries of these domains are given. The location of the domains on the membrane topology shown for the M. griseaprotein is the same for other proteins that share these domains. For GPR1-related proteins, sequence similarity was limited to the membrane-spanning regions and MG00532.4 had sequence similarity with other animal GPCRs between the third and the fifth membrane-spanning regions (not shown in figure). ::: ![](gb-2005-6-3-r24-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Predicted G-protein-coupled receptor-like proteins in M. grisea ::: Known receptors used as query in BLAST against M. griseaproteins or another search method M. griseaproteins retrieved by known receptor (BLASTP) E-value Other proteins homologous to M. griseaproteins retrieved by known receptor PfamA GPCR domains (E-value)/conserved domain identified in the present study --------------------------------------------------------------------------------------------- ---------------------------------------------------------- --------- ------------------------------------------------------------------------------ ------------------------------------------------------------------------------- Pheromone receptor (CAC86431; STE2-like) MG04711.4\* 3e-65 Pfam STE2 (2.1e-04) Pheromone receptor STE3 (STE3\_YEAST) MG06452.4^†^ 2e-14 Pfam STE3 (1.1e-09) cAMP receptor TASA (Q9NDL2) MG06738.4\* ^,†^ 5e-11 Pfam7tm\_2 (1.3e-04)/cAMP\_dom MG06797.4 cAMP\_dom MG06257.4\* cAMP\_dom MG00326.4 Pfam 7tm\_2 (7.9e-05)/cAMP\_dom MG00258.4 cAMP\_dom MG10544.4 cAMP\_dom GPCR GPR1 (GPR1\_YEAST) MG08803.4 4e-18 GPCR STM1 (STM1\_SCHPO) MG04698.4\* 5e-19 STM1\_dom MG02855.4\* ^,†^ 1e-17 STM1\_dom GPCR mPR (NP\_848509) MG05072.4\* 6e-17 mPR\_dom MG09091.4\* mPR\_dom MG04679.4\* ^,†^ mPR\_dom PTH11 receptor (AF119670\_1) MG05871.4 (PTH11) \* ^,†,‡^ 0 PTH11\_dom MG10473.4^‡^ 3e-34 PTH11\_dom MG06755.4^‡^ 1e-33 PTH11\_dom MG07553.4^‡^ 2e-32 PTH11\_dom MG09022.4\* ^,‡^ 2e-27 PTH11\_dom MG07565. \* ^,†,‡^4 6e-23 PTH11\_dom MG07946.4^†,‡^ 3e-21 PTH11\_dom MG11006.4 2e-32 PTH11\_dom MG09070.4\* 2e-29 PTH11\_dom MG07806.4 2e-21 PTH11\_dom MG03584.4^†^ 1e-22 PTH11\_dom MG05214.4\* 4e-31 PTH11\_dom MG09863.4\* ^,‡^ 1e-28 PTH11\_dom MG10407.4\* 3e-26 PTH11\_dom MG10571.4\* ^,†^ 4e-25 PTH11\_dom MG01867.4^‡^ 1e-23 PTH11\_dom MG09455.4^†,‡^ 2e-23 PTH11\_dom MG10050.4^‡^ 1e-14 PTH11\_dom MG09667.4 1E-22 PTH11\_dom MG05352.4\* 2e-22 PTH11\_dom MG07420.4 1e-21 PTH11\_dom MG10442.4 4e-20 PTH11\_dom MG02160.4^†^ 6e-19 PTH11\_dom MG02001.4\* ^,†^ 1e-18 PTH11\_dom MG10257.4 2e-18 PTH11\_dom MG01905.4 2e-17 PTH11\_dom MG07987.4 1e-16 PTH11\_dom MG10438.4\* ^,‡^ 6e-18 PTH11\_dom MG06171.4\* 1e-17 PTH11\_dom MG07851.4 1e-17 PTH11\_dom MG04935.4\* 1e-17 PTH11\_dom MG05386.4 3e-17 PTH11\_dom MG09865.4\* ^,†^ 3e-16 PTH11\_dom MG09061.4 4e-16 PTH11\_dom MG05514.4\* ^,†^ 1e-16 PTH11\_dom MG06535.4\* 3e-14 PTH11\_dom MG01190.4 7e-14 PTH11\_dom MG10581.4\* 7e-14 PTH11\_dom MG03009.4\* 2e-13 PTH11\_dom MG10747.4 8e-13 PTH11\_dom MG03935.4 2e-12 PTH11\_dom MG04682.4\* PTH11\_dom MG09416.4 1e-10 PTH11\_dom MG02692.4\* 2e-10 PTH11\_dom MG07857.4 PTH11\_dom MG00826.4 PTH11\_dom MG06624.4\* ^,†^ PTH11\_dom MG00435.4\* PTH11\_dom MG08653.4\* PTH11\_dom MG10706.4\* ^,†^ PTH11\_dom MG04170.4\* PTH11\_dom MG08525.4\* PTH11\_dom MG00277.4\,† PTH11\_dom MG02365.4\ PTH11\_dom MG06595.4 PTH11\_dom MG06084.4\* PTH11\_dom MG09437.4\* PTH11\_dom MG01890.4 PTH11\_dom MG01871.4 PTH11\_dom MG03794.4 PTH11\_dom MG01884.4\* PTH11\_dom InterProScan MG00532.1 MG00532.4 (weak similarity to animal GPCRs) \* Pfam 7tm\_2 (1.4e-02) Classes of GPCR-like protein in M. griseawere subdivided on the basis of BLASTP analysis and shared domains, as described in Materials and methods. They were clustered into paralogous families if the proteins showed 30% identity and 80% overlap over the complete length of the protein. Paralogous families are separated by a blank line. The GPCR-like proteins in M. griseacould be classified into nine subclasses containing more than one member and 48 containing a single member. Six subclasses contained two members, two contained three and one contained six. \*M. griseaproteins represented by genes expressed in microarray experiments. ^†^M. griseaproteins that are represented in M. griseaESTs. ^‡^Proteins containing the cysteine-rich CFEM domain. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Classes of GPCR-like proteins in fungi ::: Class of receptors M. grisea N. crassa S. cerevisiae S. pombe --------------------------------------------------------------- ------------- ------------- ----------------- ------------ GPCR homologs of known classes Fungal pheromone STE2-like (class D) 1 1 1 1 Fungal pheromone STE3-like (class D) 1 1 1 1 cAMP receptor-like (class E) 6 3 \- \- Other GPCR homologs S. cerevisiaeGPR1-like 1 1 1 1 S. pombeSTM1-like 2 2 3\* 1 H. sapiensmPR-like 3 2 3\* 2\* M. griseaMG00532.4-like (weak similarities to animal GPCRs) 1 1 \- \- Other GPCR-like proteins M. griseaPTH11-related 61 25 \- \- \*Have not been characterized as GPCR in the yeast species but do have seven transmembrane spans. :::	PubMed Central	2024-06-05T03:55:56.316861	2005-3-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088943/", "journal": "Genome Biol. 2005 Mar 2; 6(3):R24", "authors": [ { "first": "Resham D", "last": "Kulkarni" }, { "first": "Michael R", "last": "Thon" }, { "first": "Huaqin", "last": "Pan" }, { "first": "Ralph A", "last": "Dean" } ] }
PMC1088944	Background ========== During transcription, RNA polymerase catalyzes incorporation of nucleotides into growing RNA chains on the basis of complementarity to the DNA template. While transcribing long poly(A) or poly(T) tracts, however, slippage or \'stuttering\' (also known as pseudo-templated transcription) occurs, with the resulting incorporation of one or more extra nucleotides or occasional lack of a base or two corresponding to the run of repeat bases. Transcription slippage was first reported from in vitrostudies \[[@B1]\], and investigated in vivolater \[[@B2]\]. Although sequences that are able to cause efficient transcriptional slippage occur infrequently in genomic DNA, they have been found and a functional role has been assigned to some of them. For example, transcription slippage is utilized for regulation of the Escherichia coli pyrBIand codBAoperons and occurs shortly after transcription initiation when special conditions apply \[[@B3],[@B4]\]. When a transcription slippage-prone (\'slippery\') sequence occurs in a coding sequence, the mRNA products are heterogeneous. In such an mRNA population, the sequence downstream of a slippery pattern generally occurs in all three different reading phases relative to the reading frame 5\' of the slippage-prone sequence. Translation of these mRNAs yields protein products that differ in their amino-acid sequence downstream of the slippery sequence. For genes encoding a single functional protein product, the presence of slippery sequences is expected to be detrimental, as it is likely to squander cellular resources to synthesize unwanted, or in some instances even deleterious, products. Aberrant forms of beta-amyloid precursor protein and ubiquitin B found in Alzheimer\'s and Down syndrome patients are associated with molecular misreading, whose mechanism is likely to be transcriptional slippage \[[@B5]\]. Moreover, this type of molecular misreading was suggested to be relevant to the aging process \[[@B6]-[@B8]\]. Transcriptional slippage in the human APCgene (in addition to replicational slippage \[[@B9]\]) has also been proposed as a cause of colorectal cancer \[[@B10]\]. There are, however, at least two situations where transcriptional slippage inside a coding region can be advantageous. One is where a frameshift mutation occurs in the coding sequence and transcription slippage at a nearby site permits synthesis of a proportion of mRNAs in which a non-templated nucleotide(s) compensates for this mutation, thereby restoring the original framing. An example involving a single nucleotide deletion occurs in apoB, the human gene in which defects cause familial hypobetalipoproteinemia. In addition to encoding the expected truncated dysfunctional product, about a tenth of the product is full length as a result of its mRNA template having an extra A inserted in a run of eight As \[[@B11],[@B12]\]. A similar situation was recently reported for the canine AP3B1gene \[[@B13]\]. A second situation in which transcription slippage has a positive outcome is when it leads to synthesis of more than one useful product from a single gene - during expression of the P gene in paramyxoviruses, for example. The best-studied example is in Sendai virus, where a specific number of untemplated Gs are inserted at the position corresponding to the slippery site (reviewed in \[[@B14]\]). Remarkably, this process depends on a hexanucleotide phasing of the slippery sequence relative to the end of genome and this is modulated by viral protein N \[[@B15]\]. In addition to its involvement in paramyxovirus decoding, transcriptional slippage is used for the synthesis of additional functional proteins in other viruses, such as Ebola virus \[[@B16]-[@B18]\]. Utilization of transcriptional slippage is not limited to viral genes. Highly efficient transcription slippage in the decoding of the cellular dnaXgene of Thermus thermophilusresults in 50% of the product being shorter than the \'standard\' product \[[@B19]\]. This gene has a run of nine As in its sense strand seven-eighths of the way through its coding sequence. During transcription, RNA polymerase synthesizes mRNA that contains poly(A) runs of variable length. When the number of As is equal to the templated 9 or 9 + 3n, the full-length product, the DNA polymerase III tau subunit, is synthesized. When the number of As is anything else, for example 8, 10, 11, 13, the translating ribosomes encounter a 3\' stop codon located close to the poly(A) run. They terminate, resulting in the synthesis of a shorter product (Figure [1](#F1){ref-type="fig"}), the gamma subunit of DNA polymerase III, which has distinctive functional properties \[[@B20],[@B21]\]. In some other bacteria such as E. coli\[[@B22]-[@B24]\] and its close relatives \[[@B25],[@B26]\], dnaXalso encodes both subunits, but the shorter one is synthesized via ribosomal frameshifting instead of transcriptional slippage. The same end result can be achieved by nonstandard events at different levels of readout \[[@B19]\]. Another example of the use of transcriptional slippage was recently reported in the decoding of the Shigella flexneri mxiEgene which encodes a transcription activator \[[@B27]\]. mxiEconsists of two overlapping open reading frames (ORFs), mxiEaand mxiEb. Transcriptional insertion of an additional non-templated nucleotide at the run of Us results in a proportion of the mRNAs having mxiEaand mxiEbin the same reading frame \[[@B27]\]. Therefore, in contrast to T. thermophilus dnaXtranscriptional slippage, where the novel product is shorter than the product of standard decoding, mxiEtranscriptional slippage is required for synthesis of the longer protein product. Transcription slippage-prone sequences are expected to be under-represented in coding regions \[[@B2]\], because functional utilization of such sequences is unlikely to be common. The recent dramatic increase in the number of sequenced bacterial genomes provides an opportunity to perform wide-scale analysis of whole kingdoms of life \[[@B28]\]. The current study explores whether long runs of As or Ts are indeed avoided in the coding regions of 108 sequenced bacterial genomes, and where such runs do occur, whether they play a positive functional role in gene expression. Results ======= Distribution of homopolymeric A and T runs in bacterial genomes --------------------------------------------------------------- If any sequence pattern is randomly distributed in a genomic sequence, the following equation should be satisfied: Pc/Pg ![](gb-2005-6-3-r25-i1.gif) Nc/Ng Where Pc is the number of pattern copies in coding regions, Pg is the number of copies in the whole genome, Nc the number of nucleotides in coding regions and Ng the size of the whole genome. We have analyzed the ratio Pc/Pg for 118 published eubacterial and archaeal genomes for homopolymeric A or T patterns of different lengths (see Additional data file 1). An example of such an analysis for a few representative genomes is illustrated in Figure [2a](#F2){ref-type="fig"}. For several genomes, a sharp reduction in Pc/Pg is evident during transition from the patterns containing nnumber of As or Ts to the patterns containing n+ 1 As or Ts. The position of the transition is different among the genomes analyzed. A sharp transition is evident only for AT-rich bacterial genomes; in GC-rich bacterial genomes the existence of long A/T runs has a low probability (if random) \[[@B29]\]. Therefore, they are more likely to occur if there is positive selection. In some AT-rich genomes, however, there is no transition in the Pc/Pg ratio at any length (for example, Borrelia burgdorferi). This suggests that such organisms have developed a mechanism to suppress transcriptional slippage at long runs of As or Ts. Indeed the frequency of 9 A/T or 10 A/T runs in such genomes is about one per gene. Comparison of poly(A) and poly(T) occurrence in genomic sequences versus coding regions has two disadvantages. First, runs of As cannot be discriminated from runs of Ts at the level of genomic sequences. Second, such runs could have a positive or negative role(s) outside of coding regions. For example, long runs of Ts can serve as parts of transcriptional terminators, although poly(T) runs do not have to be uninterrupted for this purpose \[[@B30]\]. In addition, the occurrence of A and T runs can be affected by dinucleotide bias, codon usage and amino-acid composition of encoding proteins. To minimize the influence of these factors on our analysis, we used another approach to estimate the distribution of such patterns. A thousand random genomes were generated for every genome shown in Figure [2a](#F2){ref-type="fig"} using the following rules: protein sequences from the real genomes were preserved, but the codons encoding the amino acids were randomized, taking into account codon usage. Such random genomes are relieved of selective pressure to avoid slippery sequences. A similar approach was previously used for statistical analysis of frameshift-inducing patterns in E. coli\[[@B31]\] and secondary RNA structures in bacterial genomes \[[@B32]\]. In addition, we used randomization approaches that preserved dinucleotide bias and both dinucleotide bias and codon usage using the DiShuffle and CodonDishuffle programs developed by Katz and Burge \[[@B32]\]. Figure [2b](#F2){ref-type="fig"} shows the distribution of A/T runs in such random genomes compared to the real genomes. If there were no selective pressure on a particular pattern, its occurrence in random genomes would be similar to its occurrence in a corresponding real genome. If there were negative selection against a particular pattern, it would occur more frequently in random genomes than in real ones. This analysis confirmed our general conclusion that runs of As and Ts of a certain length are avoided in some prokaryotic genomes, but the length of the pattern that is likely to be harmful varies among different genomes. Consequently, such patterns are significantly under-represented in AT-rich genomes. Interestingly, in the genome of Wigglesworthia glossinidia, A/T patterns of any length occur with the same frequency in coding and noncoding regions, suggesting that transcriptional slippage is not possible in this species at patterns of any length. However, when the occurrence of such patterns is compared with their occurrence in random genomes, a negative selection is evident for patterns of exceptional length. This suggests that very long patterns have a negative effect not associated with transcriptional slippage. Functional roles of transcriptional slippage -------------------------------------------- The next step was to find occurrences of transcriptional slippage and to investigate, using comparative sequence analysis, whether they are likely to have any functional role. The scheme of this analysis is shown in Figure [3](#F3){ref-type="fig"}. We searched for occurrences of 9As and 9Ts in protein encoding genes. Only those genes were selected where transcriptional slippage would result in synthesis of a protein which is larger than the counterpart generated by standard decoding. When transcriptional slippage results in the synthesis of a truncated product, as in decoding T. thermophilus dnaX, it is difficult to predict functional importance on the basis of comparative sequence analysis, as there is no extensive \'new\' coding sequence suitable for such an analysis. The next filter was the exclusion of genes from bacteria where transcriptional slippage is unlikely to occur on runs of 9As and Ts. Organisms with AT-rich genomes that do not demonstrate selection against 9A and 9T sequences within their coding regions may have evolved to suppress transcriptional slippage on 9A and 9T and are unlikely to exhibit it. To select bacteria in which transcriptional slippage on 9A and 9T is unlikely, we first determined the number of genes containing 9A and 9T. For those bacteria where this number was higher than the threshold number 20 (we assumed that it is unlikely that transcriptional slippage can be utilized by more than 20 genes in the same species) we searched for evidence of negative selection against these sequences. If such sequences were not under-represented, corresponding bacteria were considered as those where transcriptional slippage is unlikely to occur on 9A or 9T runs. Genes from such bacteria were excluded from further analysis. The remaining pool of genes contained some identical genes. Some of these exist in multiple copies inside the same genome whereas others are identical because they derived from genomes of highly related species. Such identical genes were combined to reduce redundancy. In the list of these genes ([2](#S2){ref-type="supplementary-material"}) only one representative is given for each group of identical genes. The products of those genes that can be generated by transcriptional slippage were compared to each other using tBLASTn \[[@B33]\], and to those derived from other sequences present in sequenced bacterial genomes. Genes that produced no significant sequence similarity were considered as ORFans \[[@B34],[@B35]\]. Since ORFans are not suitable for comparative analysis, they were excluded from further analysis (shown in gray in [2](#S2){ref-type="supplementary-material"}). The number of gene groups for which homologs were found is 53. The likelihood of functional utilization of transcriptional slippage was estimated using comparative sequence analysis. According to the scheme utilized (Figure [4](#F4){ref-type="fig"}), we consider transcriptional slippage patterns likely to be functional if the organization of ORFs fused by transcriptional slippage is the same in at least two non-identical sequences sharing significant sequence similarity. We have not found evidence of functional utilization of transcriptional slippage for 40 cases (shown in blue in [2](#S2){ref-type="supplementary-material"}). Most probably, although transcriptional slippage is likely to occur during expression of these genes, it has no significant detrimental effect. This result is consistent with our previous finding that sequences that direct significant levels of frameshifting in the E. coligenome may occur without apparent function \[[@B31]\]. Six cases were found where protein products expressed by transcriptional slippage have homologs encoded in a single ORF in genes from other species. One example is shown in Figure [5](#F5){ref-type="fig"}. Such genes are normally considered as pseudogenes, because their ORF is disrupted. However, transcriptional slippage should result in the synthesis of normal functional protein and consequently such genes should not be treated as inactive as a result of frameshift mutation. These genes are shown in green in [2](#S2){ref-type="supplementary-material"}. In seven cases (red in [2](#S2){ref-type="supplementary-material"}) homologs were found with both a conserved organization of the overlapping ORFs and a conserved pattern of 9As in the overlapping regions. Among them, six cases derive from IS elements whose total number of copies is 27. One group is composed of the mapWgenes from Staphylococcus aureusstrains; mapWis a functional candidate derived from a non-mobile element. Transcriptional slippage was recently found in the S. flexneripathogenicity-encoding plasmid that carries the mxiEgene \[[@B27]\]; it is not included in the 108 sequences of complete genomes downloaded for the present study (even though the chromosomal sequence was included). Discussion ========== We have obtained an initial view of the distribution and functional utilization of simple transcriptional slippage sites in bacterial genomes performed on a multiple-genome scale. The data obtained demonstrate that runs of As and Ts, which result in efficient transcriptional slippage, are significantly underrepresented in coding regions of AT-rich genomes. One likely reason for this underrepresentation is the \'slippery\' nature of such sites. In addition to transcriptional slippage, these sequences are likely to be hypermutable as a result of slippage during replication. This also contributes to negative selection against these sequences. It has previously been shown that in eukaryotes short repetitive sequences of specific length are usually under-represented in coding regions compared to noncoding regions \[[@B36]\]. The implication is that such sequences are susceptible to frameshift errors at the DNA level. We cannot distinguish whether the reason for negative selection against A or T runs is slippage at the replication or transcriptional level or at both. Our approach to finding genes where transcriptional slippage is functionally utilized can, however, discriminate it from replicational slippage in some instances. Since we deal with those cases where sequence extension after a slippery pattern in a shifted reading frame is conserved among several homologs, it is very likely that this extension is expressed. Theoretically, its expression can be achieved as a result of replicational and/or transcriptional slippage. In the first case, the result would be the existence of a population of bacteria with heterogeneous genomes where different members of such a population would have a different number of nucleotides within a repetitive run, as previously described for several occurrences in the Campylobacter jejuni\[[@B37]\]. We have found several such examples for the group of genes that we classified as \'pseudo pseudogenes\' (an example is in Figure [5](#F5){ref-type="fig"}). If a specific run of 9As or 9Ts occurs within a number of homologs and the length of such run is conserved among all homologs, then it is very likely that this specific run is used for purposeful transcriptional slippage to generate a set of heterogeneous mRNAs. Subsequent translation of such mRNAs will result in the synthesis of more than one protein product from the same gene. An example is shown in Figure [6](#F6){ref-type="fig"} for IS elements from D. radiodurans. We have not found homologous IS elements that contain insertions or deletions in the run of As. Those shown on Figure [6](#F6){ref-type="fig"} are the only homologs found. In general, a conserved run of As or Ts in several homologs does not imply that replication slippage is impossible on such a run. For example, when insertion of an additional nucleotide is deleterious, there will be selection against sequences with the additional nucleotide. However, in this case such replicational slippage cannot be referred to as being functional. The comparative sequence analysis of genes with runs of nine As and Ts from genomes where such repeat bases are slippage-prone, revealed S. aureus mapWas a candidate for functional utilization of transcriptional slippage. mapWbelongs to a group of mapgenes encoding MHC class II (major histocompatibility complex class II)-like proteins. mapWconsists of two ORFs and it was proposed earlier that they can be expressed together to produce a full-length \'fusion\' protein \[[@B38]\]. Perhaps the ability of S. aureusto encode MHC-II like proteins with variable length can facilitate survival in mammals of varied genetic backgrounds \[[@B39]\]. However, the presence of mapWgenes with an uninterrupted ORF in some S. aureusstrains suggests that replicational slippage can be also utilized in this case. The largest group of functionally utilized transcriptional slippage sites belongs to mobile IS elements. We have found patterns of 9 As in 27 IS elements from the following organisms - Deinococcus radiodurans, Mesorhizobium loti, Nostocsp. PCC 7120, Streptococcus pyogenesand Sulfolobus solfataricus. Interestingly, some homologous IS elements from D. radioduransand Nostocsp. PCC 7120 have 8As instead of 9As in the same location. This suggests that in these organisms, transcriptional slippage is productive even on eight As. Figure [6](#F6){ref-type="fig"} illustrates codon alignment of homologous IS elements from D. radiodurans. It is clear that the stretch of As is evolutionally preserved among these IS elements (although its length varies, there is no deletion or insertions) and their ORF organization suggests that runs of As are used to produce ORF fusions. (A high-resolution FITC mass spectrometric analysis of numerous tryptic peptides from D. radioduranshas been performed by Smith and colleagues \[[@B40]\]. A preliminary analysis of these data is revealing products of IS element mRNAs synthesized via transcriptional slippage (R. Smith, P.V.B., A.W.H, J.Z, R.F.G. and J.F.A, unpublished results) Alignment of IS elements from Nostocis not shown, as all its elements are identical except for the length of the poly(A) run varying from 8 to 10 As. Many IS elements encode their transposase in two overlapping ORFs, orfAand orfB. Synthesis of a fused ORFA-ORFB product is required for transposition. The most common known mechanism for synthesis of ORFA-ORFB fusion is -1 ribosomal frameshifting (see \[[@B41]-[@B44]\] for reviews). Transcriptional slippage has, however, been proposed previously as an alternative mechanism for one IS element \[[@B19]\]. The present study has identified a number of IS elements utilizing transcriptional slippage for synthesis of their ORFA-ORFB fusion. Therefore transcriptional slippage can be considered as a common mechanism for IS element expression. In addition, we have found a set of pseudo pseudogenes where what is normally considered as a frameshift mutation extends a non-slippery pattern of 8 As to the slippage-prone sequence of 9 As. As a result, such a frameshift mutation does not lead to full inactivation of a gene that normally could be annotated as a pseudogene, as a normal functional product is still produced. The advantage of the unusual decoding of these genes by transcriptional slippage, compared to standard decoding of wild-type counterparts, is uncertain. It is clear that such cases were generated by single mutations and they may, or may not, be present in different isolates from the same species. Transcriptional slippage can, however, be considered as functionally utilized, since if such genes were transcribed, a proportion of the mRNA synthesized should contain the intact coding information. This important consideration needs to be taken in account in genome annotation. Although organism-specific utilization of transcriptional slippage cannot be ruled out, we have identified a large number of genes where, using comparative analysis, no apparent functional role can be assigned for transcriptional slippage. This result is parallel to our previous analysis of frameshift-inducing sequences in the E. coliK12 genome \[[@B31]\]. It was shown that a significant level of frameshifting errors occur in many E. coligenes containing A\_AAA\_AAG sequences (codons are separated by underscoring), but no such sequences were found in highly expressed genes \[[@B31]\]. Similar considerations can be applied here for transcriptional slippage. When erroneous nonstandard decoding occurs in genes that are not highly expressed, the cellular load is modest owing to the low level of aberrant product compared to the total protein mass. Such situations may be easily tolerated. Transcriptional slippage motifs were found in many ORFans, but any functional purpose could not be assessed in the present study. We found runs of 9A or 9T in 48 ORFans. The origin(s) of ORFans is mysterious. While some of them are likely to be \'coincidental ORFs\' or \'junk ORFans\' which do not produce proteins under any conditions \[[@B45]-[@B47]\], many ORFans are likely to be real genes \[[@B36],[@B48],[@B49]\]. The analysis of transcriptional slippage in this study was limited to that occurring on 9As and 9Ts. It is clear, however, that the efficiency of transcriptional slippage on runs of As and Ts is highly organism dependent, and there are a number of bacteria in which transcriptional slippage may occur on runs of shorter length. In addition, transcriptional slippage patterns can occur on other nucleotide repeats. The simplest mechanism that can be proposed for transcriptional slippage is dissociation of the growing RNA chain from its DNA template while inside an open RNA polymerase complex, and subsequent re-association with the DNA template at a new location (Figure [7](#F7){ref-type="fig"}). On this basis, other repeat patterns of low complexity are likely to result in transcriptional slippage. For example, (AT)nmay result in insertion of additional non-templated ATs. Transcriptional slippage sites can be also formed by combination of two relatively short homopolymeric patterns as in paramyxoviruses \[[@B14]\]. Simple sequence repeats (SSR), also known as microsatellites, occur frequently in virulence genes of different pathogenic bacteria \[[@B37],[@B50],[@B51]\]. Because of replicational slippage, they are responsible for hypermutability and phase variations in pathogenic bacteria \[[@B52]\]. The effect of such sequences on transcription and translation has not yet been extensively studied. Such sequences could also result in nonstandard decoding (transcriptional slippage or ribosomal frameshifting) and consequently express more than one protein product. Expression of multiple products encoded by virulence genes may be beneficial for pathogens as a strategy for evading the host immune response. Statistical, experimental and functional analysis of such sequences in relation to transcription and translation will hopefully be the subject of further investigation. Materials and methods ===================== Analysis of A and T repeat distribution in bacterial genomes ------------------------------------------------------------ Fasta files containing nucleotide sequences of entire bacterial genomes and nucleotide sequences of coding regions were downloaded from the National Center for Biotechnology Information ftp site \[[@B53]\] on 25 March, 2003. Occurrences of A and T runs with different lengths were calculated for each genome in the file containing genomic sequences (accession\_number.fna) and in the files containing nucleotide sequences of coding ORFs (accession\_number.ffn). The ratios of occurrences of runs of A and T between .fna files and .ffn files were calculated for every accession number and the data are summarized in [1](#S1){ref-type="supplementary-material"}. Random genomes were generated for representative genomes as described in \[[@B31]\]. In addition we applied DiShuffle and CodonDiShuffle programs provided by C. Burge \[[@B32]\]. The ratios between occurrences of A and T runs in real genomes and the mean values for A and T runs in random genomes were further calculated. Generation of novel protein sequences corresponding to those produced via transcriptional slippage -------------------------------------------------------------------------------------------------- Runs of 9A or 9Ts were sought within coding regions of genomic sequences of completed bacterial genomes. To generate a novel in silicoprotein that can be produced by transcriptional slippage, one and two As or Ts were introduced into the pattern of 9As or 9Ts. The length of the resulting ORF in these sequences was compared to the ORF in the original sequences. Those sequences that contain ORFs longer than the original were selected for further analysis. Additional data files ===================== Additional data is available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} contains numbers of occurrences of A and T runs in bacterial genomes. Additional data file [2](#S2){ref-type="supplementary-material"} contains information about genes where 9A or 9T patterns were found. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Numbers of occurrences of A and T runs in bacterial genomes. Column A is used for the names of analyzed files and row 1 indicates the length of A/T run. Sheet \'whole genomes\' corresponds to occurrences in entire genomes, sheet \'coding sequences\' corresponds to occurrences in coding regions and sheet \'ATRatio\' corresponds to the ratio between these numbers ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Information about genes where 9A or 9T patterns were found. These genes correspond to the pool of genes selected for comparative analysis. The table contains representatives from 98 gene groups selected in step 4 of the scheme in Figure [3](#F3){ref-type="fig"}. Column A is used for accession numbers. B is for coordinates of the corresponding gene. C indicates whether it is a run of A or T in a sense strand. D shows the functional status of a gene. The functional status is annotated by text and by color. The red color is used for genes with potential positive role of transcriptional slippage, blue is for those where there is no positive functional role, green is for genes where transcriptional slippage might restore a disrupted ORF and grey is used for ORFans, where functional status cannot be assessed. Column E contains nucleotide sequences of mRNAs produced viatranscriptional slippage with ORFs longer than those in the original DNA templates. Column F contains the corresponding protein sequences ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We are grateful to Chris Burge for providing us with source codes for the DiShuffle and DiCodonShuffle programs. We thank Norma Wills for her key role in the background work on which this study is based. The salary of J.F.A. was supported by NIH grant GM48152 and an award from Science Foundation Ireland. The salary of P.V.B. was supported by DOE grant DE-FG03-01ER63132 to R.F.G. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A scheme for the nonlinear expression of Thermus thermophilus dnaXvia transcriptional slippage. Transcription of dnaXresults in synthesis of a population of mRNAs in which the sequence 3\' of the slippery AAAAAAAAA is framed in different molecules in all three reading frames relative to sequence 5\' of the slippery motif. ::: ![](gb-2005-6-3-r25-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Analysis of the distribution of runs of As and Ts in selected genomes. Run length is indicated on the x-axis; the ratio of pattern occurrence on the y-axis. (a)Ratio of occurrences in coding regions and in entire genomic sequences. (b)Ratio of occurrences of A runs in real genomes and average occurrence in 1,000 randomized genomes. Biases preserved during the randomization procedure are indicated above each pair of graphs. Accession numbers are as follows: NC\_000913 E. coliK12; NC\_000915 Helicobacter pylori; NC\_000963 Rickettsia prowazekii; NC\_001318 Borrelia burgdorferi; NC\_002163 Campylobacter jejuni; NC\_003450 Corynebacterium glutamicum; NC\_003364 Pseudomonas aeruginosa; NC\_004344 Wigglesworthia glossinidia. ::: ![](gb-2005-6-3-r25-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Scheme for functional analysis of slippery patterns in coding sequences. ::: ![](gb-2005-6-3-r25-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Different types of ORF organization for genes sharing sequence similarities. ::: ![](gb-2005-6-3-r25-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Codon alignments of DNA and mRNA sequences of orthologous genes from two different strains of E. coli. In the DNA, an A causing a frameshift mutation is underlined. In the mRNA, a tandem A inserted by transcriptional slippage which results in ORF restoration is underlined. ::: ![](gb-2005-6-3-r25-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Alignment of a portion of Deinococcus radioduransIS elements containing a run of nine or eight As. Universally conserved residues are in bold, runs of As are in red. The alignment was built using Clustal \[54\]. ::: ![](gb-2005-6-3-r25-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### A model of transcriptional slippage. ::: ![](gb-2005-6-3-r25-7) :::	PubMed Central	2024-06-05T03:55:56.321862	2005-2-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088944/", "journal": "Genome Biol. 2005 Feb 15; 6(3):R25", "authors": [ { "first": "Pavel V", "last": "Baranov" }, { "first": "Andrew W", "last": "Hammer" }, { "first": "Jiadong", "last": "Zhou" }, { "first": "Raymond F", "last": "Gesteland" }, { "first": "John F", "last": "Atkins" } ] }
PMC1088945	Background ========== The malaria parasite (genus Plasmodium) is a unicellular eukaryote which, in the course of its complex life cycle, invades the erythrocytes of its vertebrate host. It is this intraerythrocytic phase of the parasite life cycle that gives rise to all the symptoms of malaria, a disease that is estimated to give rise to almost 5 billion episodes of clinical disease and up to 3 million deaths annually \[[@B1]\]. Plasmodium falciparum, the most virulent of the malaria parasites that infect humans, has developed resistance to most of the antimalarial drugs currently available. There is an urgent need for the development of new antimalarial drug strategies, and for an improved understanding of the mechanisms that underpin the parasite\'s ability to develop resistance to antimalarials. Membrane transport proteins are integral membrane proteins that mediate the translocation of molecules and ions across biological membranes. They serve a diverse range of important physiological roles, including the uptake of nutrients into cells, the removal of unwanted metabolic waste products and xenobiotics (including drugs), and the generation and maintenance of transmembrane electrochemical gradients. These proteins play a key role in the growth and replication of the parasite, as well as in the phenomenon of antimalarial drug resistance. But despite this, and despite the fact that membrane transport proteins have proven to be extremely effective drug targets in other systems \[[@B2]\], all but a few of the membrane transport proteins of the malaria parasite remain very poorly understood, and their potential as antimalarial drug targets remains largely unexplored \[[@B2]\]. The \'permeome\' is a term used here to describe the total complement of proteins involved in membrane permeability in a given organism. It encompasses the full range of channels and transporters encoded in the genome. The original annotation of the P. falciparumgenome, published at the end of 2002, identified \"a very limited repertoire of membrane transporters, particularly for uptake of organic nutrients\" and \"no clear homologs of eukaryotic sodium, potassium or chloride ion channels\" \[[@B3]\]. It is questionable, however, whether this reflects a genuine paucity of such proteins in this organism, or simply shortcomings in the annotation. Despite ongoing improvements in automated gene annotation, it is widely accepted that the routines involved provide a first phase of annotation and that the attainment of a high-quality annotation requires the intervention of manual curation (reviewed in \[[@B4]\]). Errors that are difficult to avoid in automated systems for genome annotation include the incorrect prediction of intron/exon boundaries and the position of the start/stop codons, which can result in incomplete or truncated proteins, or the merging of neighboring proteins \[[@B5],[@B6]\]. The assignment of functional annotations to proteins is hampered by several factors \[[@B7]\], including the non-critical use of annotations from existing database entries, ignoring multidomain organization of the query proteins and/or the database hits, and, in the case of P. falciparum, the considerable divergence that generally exists between the parasite and those organisms for which sequence data is currently available in the databases \[[@B3]\]. Manual curation affords a greater flexibility in handling these problems. For many of the predicted proteins encoded by P. falciparumthe similarity to their closest non-Plasmodiumhomologs is insufficient to permit annotation on the basis of BLAST searches alone. As highlighted in a recent review of the current status of the malaria parasite genome project \[[@B8]\], the annotation of these proteins requires an in-depth assessment by a manual curator using a range of bioinformatic approaches \[[@B7]\] including position-specific iterated BLAST (PSI-BLAST), detection of conserved domains, construction of multiple sequence alignments and comparisons of predicted secondary structure. This process is laborious and time-consuming, but by combining the information gained from these analyses, it is possible to arrive at reliable annotations and to gain significant insight into the function of the proteins of interest. In this paper we report the results of a detailed analysis of the permeome of P. falciparum. The study makes use of a computer program that searches a genome database on the basis of the hydropathy plots of the corresponding proteins \[[@B9]\]. The approach is based on the observation that the polypeptides comprising transporter proteins typically possess multiple hydrophobic transmembrane domains (TMDs) and connecting hydrophilic, extra-membrane loops that are detected as peaks and troughs, respectively, in a plot of the hydrophobicity index of the polypeptide. Many transporters characterized to date have between eight and 14 TMDs \[[@B10]\]. In searching for additional candidate transporters, the P. falciparumgenome was therefore scanned for proteins with seven or more TMDs. Proteins retrieved by this search were subjected to a detailed analysis, involving the application of several different bioinformatic methods. The analysis presented here has doubled the number of candidate membrane transport proteins identified in the genome, as well as attributing putative substrate specificities and/or transport mechanisms to all of those \"transporter, putative\" proteins previously lacking this information. The newly designated proteins include candidate transporters for nutrients such as sugars, amino acids, nucleosides and vitamins. There are also transport proteins predicted to be involved in maintaining the ionic composition of the cell and in the extrusion of metabolic wastes such as lactate. For 34 of the candidate transport proteins of particular interest we have investigated the time-course of expression of mRNA throughout the asexual blood stage of the parasite. The enrichment in the repertoire of P. falciparum-encoded transport proteins reported here indicates that the parasite\'s permeome is not as impoverished as was originally thought. Results ======= Parasite proteins with seven or more putative TMDs -------------------------------------------------- A comprehensive search of the P. falciparumgenome for genes encoding proteins predicted, on the basis of a hydropathy plot analysis, to have seven or more putative TMDs retrieved 167 candidate proteins. These proteins were categorized into three broad classes according to their putative functions (transport, non-transport or no putative function) as predicted by bioinformatic analyses. Known or putative transport functions were assigned to 89 (53%) of the retrieved proteins. A further 50 (30%) proteins were categorized as having functions that are non-transport related; these included various transferases, receptors, and proteins involved in trafficking and secretion (such as protein translocases), many of which have escaped annotation. The remaining 28 (17%) proteins had no non-Plasmodiumsequence homologs or similarities to conserved domains, and did not resemble transporters in structure (see Figure [1](#F1){ref-type="fig"}); they therefore could not be ascribed a putative function. The expanding inventory of P. falciparumtransport proteins ------------------------------------------------------------ Most of the P. falciparumputative transport proteins retrieved by the hydropathy plot analysis used here belong to known transport families and are described in Additional data file 1. They include new additions to the major facilitator superfamily, the drug/metabolite transporter superfamily, and the P-type ATPase superfamily, as well as many others. Several families not previously identified in the genome, such as the voltage-gated ion channel superfamily, the peptide-acetyl-coenzyme A transporter family, the zinc-iron permease family and the multi antimicrobial extrusion family, were also found to have P. falciparum-encoded members. A number of proteins to which we have assigned a putative transport function bear no significant sequence similarity to any functionally characterized proteins (transporters or otherwise) in the current databases. However, they do have hydropathy plots that resemble those of known transport proteins, consistent with the hypothesis that they too are transporters. These proteins fall into two categories: proteins that are related to \'hypothetical proteins\' from other organisms (Additional data file 2); and novel, Plasmodium-specific proteins (Additional data file 3). Transport proteins possessing six or fewer TMDs were not retrieved by our search criteria and for the most part have been omitted from the table in Additional data file 1. A limited number of such candidate transport proteins were identified in the original genome annotation and are as follows: nine members of the mitochondrial carrier family; a cation diffusion facilitator; five members of the ATP-binding cassette (ABC) superfamily; a V-type ATPase; an aquaglyceroporin (PfAQP \[[@B11]\]); and an arsenite-antimonite (ArsAB) effluxer. Two subunits are required to form a functional ArsAB efflux pump - an ATP-hydrolyzing component (ArsA) and a channel-forming integral membrane protein (ArsB). To date, only the ArsA protein has been identified in the P. falciparumgenome and the absence of a parasite ArsB homolog may indicate that either the ArsA protein does not function as part of an ArsAB efflux pump or, alternatively, the ArsB protein is present but remains to be discovered. Likewise, the parasite has genes encoding the α, β, δ, ε and γ subunits of the catalytic F~1~complex of an F-type ATPase as well as the csubunit of the membrane-spanning F~0~component, but genes for the F~0~aand bsubunits have not yet been identified in the genome. Classification of the above proteins can be found at Ian Paulsen\'s TransportDB site \[[@B12]\]. The list of parasite transport proteins possessing six or fewer TMDs has recently been extended by the description of a P. falciparumhomolog of an unusual bifunctional protein that contains an amino-terminal K^+^channel and a carboxy-terminal adenylate cyclase \[[@B13]\]. Only 54 transport proteins were identified in the original genome annotation and many of these are designated with generic descriptions such as \'transporter, putative\', from which no information can be gained about the probable mechanism of transport or substrate specificity. Our analysis has retrieved a further 55 putative transport proteins, as well as attributing putative substrate specificities and/or transport mechanisms to all of those previously without (see Additional data file 1). This brings the total number of putative/proven P. falciparum-encoded transport proteins to 109. Of these, 61 are \'porters\' (that is, uniporters, antiporters or symporters \[[@B10]\]), 29 are primary active transporters (that is, they utilize biochemical energy to pump solutes against an electrochemical gradient), five are channels, and 14 are putative novel transport proteins of unknown classification. Candidate transport proteins with seven or more TMDs (which were the subject of our search criteria) are shown in Figure [2a](#F2){ref-type="fig"}, whereas those with six or fewer TMDs (in the most part sourced from the annotated genome) are shown in Figure [2b](#F2){ref-type="fig"}. Predicting the cellular localization of P. falciparumtransport proteins ------------------------------------------------------------------------- Many transport proteins are located at the surface of the parasite, where they mediate the flux of solutes across the plasma membrane. Other transport proteins are found in the membranes of intracellular compartments such as those of the apicoplast, mitochondrion, digestive vacuole and organelles of the secretory pathway. The likely destination(s) within the cell of a given transporter can often be inferred by signals present in its polypeptide sequence and/or by its close homology to a transport protein of a known cellular localization. For example, the signal peptide required for the targeting of nuclear-encoded proteins to the parasite\'s apicoplast has been elucidated \[[@B14]\] and several P. falciparumtransport proteins contain this type of signal (see Additional data files 1, 2 and 3). These putative apicoplast transporters include the parasite homolog of the plant chloroplast phosphoenolpyruvate:P~i~antiporters (PFE1510c \[[@B3]\]) as well as a putative amino-acid transporter (PFL1515c), several ABC transporters (PFC0125w, PF11\_0466 and PF13\_0271), P-ATPases (PFE0805w and PF07\_0115) and other putative transport proteins of unknown function (PFL2410w, PF13\_0172 and PFE1525w). Likewise, the nine parasite mitochondrial carriers contain putative signals for targeting these transporters to the mitochondrion. One of these, the putative phosphate carrier protein (MPC, PFL0110c), has been cloned and shown experimentally to possess mitochondrial targeting signals \[[@B15]\]. Two putative transporters involved in chloroquine resistance - the P-glycoprotein homolog 1 (Pgh1 \[[@B16]\]) and the \'chloroquine resistance transporter\' (PfCRT \[[@B17]\]) - are localized to the parasite\'s digestive vacuole. PfCRT has recently been shown to be a member of the drug/metabolite transporter superfamily \[[@B18]-[@B20]\] and possesses several putative endosomal-lysosomal targeting signals (R.E.M. and K.K., unpublished work). The parasite V-type H^+^-ATPase is also found at the digestive vacuole membrane \[[@B21]\] and plays the major role in the acidification of the lumen \[[@B22]\]. There is experimental evidence for the presence of another proton pump at the vacuolar membrane, a K^+^-dependent, H^+^-translocating pyrophosphatase (H^+^-PPase) \[[@B22]\], although it is unclear which of the two parasite-encoded H^+^-PPases \[[@B23]\] is responsible for this activity. From its strong homology to Niemann-Pick type-C proteins (implicated in the efflux of lipids and cholesterol from lysosomes \[[@B24],[@B25]\]) the PFA0375c protein is predicted to mediate the H^+^-coupled extrusion of lipids/sterols from the digestive vacuole. Likewise, the PFE1185w protein is predicted to reside at the digestive vacuole, based on its close homology to the endosomal Fe^2+^\'NRAMP2\' transporters (involved in the transferrin cycle \[[@B26]\]), and most probably catalyzes the H^+^-driven efflux of Fe^2+^into the cytoplasm. Several transport proteins are dedicated to performing specialized tasks in the secretory pathway and specific \'retention\' motifs participate in the sorting of these proteins between the membranes of the endoplasmic reticulum (ER) and the various Golgi compartments \[[@B27],[@B28]\]. The nucleoside-sugar transporters are found exclusively at the membranes of the ER and Golgi apparatus of eukaryotes, where they mediate the uptake of nucleotide derivates (for example, UDP-galactose, UDP-glucose and GDP-fucose) from the cytosol in exchange for the corresponding nucleoside monophosphate (reviewed in \[[@B29],[@B30]\]). The nucleotide sugars are then used by specific glycosyl-transferases to add sugar moieties to (glycosylate) proteins and lipids that are transported through the secretory pathway. The parasite\'s UDP-galactose:UMP antiporter homolog (which contains a retention motif) and other putative nucleotide-sugar transporters (such as PFB0535w and PFE0260w) are predicted to be residents of the secretory pathway organelles. In the absence of any targeting signals or sorting motifs, membrane proteins are usually destined to follow the \'default\' pathway and travel through the secretory pathway to the plasma membrane \[[@B31]\]. Misannotation of transport proteins ----------------------------------- Gardner et al. \[[@B3]\] inappropriately assigned a putative transport function to several P. falciparumproteins. The protein encoded by locus PFL0620c is designated as a putative choline transporter, yet it shares strong sequence similarities with known and putative glycerol-3-phosphate acyltransferases from a range of organisms, including the SCT1 protein of Saccharomyces cerevisiae. Indeed, the PFL0620c protein has recently been shown experimentally to be a glycerol-3-phosphate acyltransferase \[[@B32]\]. The annotation of PFL0620c as a transporter mostly probably arose from a misinterpretation of the function of SCT1 (Suppressor of a Choline Transport Mutant). As the name implies, the SCT1 protein was first identified in yeast for its ability to complement a growth defect caused by a deficiency in choline transport \[[@B33]\]. SCT1 was subsequently found to catalyze the acylation of glycerol 3-phosphate in the first step of phospholipid biosynthesis; hence, SCT1 restored growth in the mutant by stimulating the synthesis of phosphatidylcholine, not by increasing choline uptake \[[@B34]\]. The proteins encoded by the genes PF08\_0098, PF11\_0225, PF14\_0133 and PF14\_0321 are all annotated as putative ABC transporters, but none of these proteins contains more than a single putative TMD. Bioinformatic analyses indicate that the PF11\_0225, PF14\_0133 and PF14\_0321 polypeptides are putative soluble ATP-binding proteins. PF11\_0225 encodes a homolog of the S. cerevisiaeGCN20 ATPase, which functions in association with the GCN1 protein to activate the translation initiation factor-2-alpha kinase (GCN2) in amino-acid-deprived cells \[[@B35]\]. The PlasmodiumGCN20 ATPase has been cloned \[[@B36]\] and shown to complement the function of the yeast GCN20 ATPase by participating in the yeast translation regulatory pathway \[[@B37]\]. The PF14\_0133 protein bears strong sequence similarities to the SufC proteins found in archaea, bacteria, cryptomonads, diatoms, dinoflagellates, red algae and plants. SufC is thought to be a versatile ATPase subunit that can interact either with the Suf(ABDSE) proteins to form a cytosolic complex for the assembly of Fe-S cluster-containing proteins, or with (unknown) membrane proteins to form an Fe-S ABC exporter \[[@B38]\]. PF14\_0321 encodes for a short polypeptide (171 residues) which displays a weak homology to other soluble ATPases of unknown function from a wide range of organisms. Finally, the PF08\_0098 protein is a member of the ABC1 family, which is distinct from, and unrelated to, the ATP-binding proteins of the ABC superfamily. ABC1 proteins are novel chaperonins essential for electron transfer in the bc~1~segment of the respiratory chain (S. cerevisiaeABC1 \[[@B39]\]) and for ubiquinone production (Escherichia coliAarF \[[@B40]\]). Gardner et al.\[[@B3]\] reported the presence of 16 P-type ATPases (P-ATPases) in the P. falciparumgenome, although only 15 are listed at TransportDB \[[@B41]\]. Four of these - PFI1205c, PF10\_0096, PF13\_0137 and MAL13P1.352 - have no sequence similarities to known or putative P-ATPases, or to conserved domains of the P-ATPase superfamily. Furthermore, PF10\_0096, PF13\_0137 and MAL13P1.352 do not possess any putative TMDs. The PF13\_0137 and MAL13P1.352 proteins display weak sequence similarities to conserved domains of the asparagine synthase (AsnB) and the nuclear cap-binding protein families, respectively, whereas the PF10\_0096 protein is unrelated to any proteins or conserved domains in the current databases. PFI1205c encodes a large protein (1,249 residues) possessing 12-13 putative TMDs, and while this protein also lacks any similarities to conserved domains, it does appear to be a member of a putative transporter family specific to apicomplexans (see Additional data file 2). Expression of P. falciparumtransport protein genes ---------------------------------------------------- The expression of 34 putative transport genes was analyzed throughout the asexual blood stage of the parasite. In previous studies, comparisons between the levels of transcripts present at different developmental stages of the parasite have been made from samples standardized to total RNA (see, for example \[[@B42]-[@B44]\]). In this study we quantified the amount of total RNA produced by the parasite as it progressed through the intraerythrocytic life cycle. As shown in Figure [3](#F3){ref-type="fig"}, the quantity of RNA in the infected erythrocyte increased significantly as the parasite grew from ring to trophozoite stage. There was 136 ± 19 (n= 2; ± range/2) times more total RNA in late trophozoites/schizonts (around 40 hours old) than in ring-stage parasites (around 8 hours old) and 161 ± 21 (n= 2; ± range/2) times more than in young rings (around 4 hours old). We therefore measured and compared transcript levels at different growth stages of the parasite from samples standardized to cell number rather than to total RNA (see below for further discussion). In the following sections we consider in turn a number of different families of transport proteins, members of which have been identified and their stage-dependent mRNA expression characterized in this study. Members of the major facilitator superfamily -------------------------------------------- The major facilitator superfamily (MFS) is one of the largest classes of transporters; its members are prevalent in organisms from all kingdoms of life and are diverse in both sequence and function \[[@B45]\]. MFS transporters of the same subfamily tend to transport related substrates, and solutes transported by MFS proteins include sugars, metabolites, amino acids, peptides, nucleosides, polyols, drugs and organic and inorganic anions. The mechanism of transport also varies within the superfamily (and sometimes even within a subfamily) with examples of uniport, solute:solute exchange, solute:H^+^antiport, as well as Na^+^or H^+^:solute symport. Our analysis has doubled the parasite\'s complement of MFS transporters from six to 12, and while this still compares very poorly with other eukaryotes such as S. cerevisiae(85 MFS proteins) and Caenorhabditis elegans(137 MFS proteins), it surpasses that found to date in the parasitic eukaryote Encephalitozoon cuniculi(two MFS proteins) \[[@B41]\]. The P. falciparumproteins fall within either the sugar porter, drug:H^+^antiporter-1, monocarboxylate porter or peptide-acetyl-coenzyme A transporter families, although one protein displays only a weak relationship to the MFS and could not be placed reliably within a family (PFL0170w, see Additional data file 4). The P. falciparummembers of the sugar porter family include the hexose transporter (PfHT1/PFB0210c \[[@B46]\]), and the putative transporters PFI0785c and PFI0955w. Of these proteins, PfHT1 (which functions primarily to transport glucose) shows the greatest similarity to glucose transporters from other organisms, including mammals (Additional data file 4). In our expression analysis PfHT1 transcript was found to be present relatively early in the intraerythrocytic life cycle (around 8 hours post-invasion, Figure [4](#F4){ref-type="fig"}) and to increase rapidly in abundance between 16 and 24 hours, after which the level of transcript stabilized temporarily before increasing again to reach a maximum at approximately 36 hours. There is significant sequence homology between PFI0955w and PfHT1 (Additional data file 4); nevertheless, PFI0955w has diverged somewhat from the glucose transporters and may therefore catalyze the transport of other sugars or sugar-related substances. The transcription of PFI0955w was found not to begin until the parasite had spent some 24 hours inside the host cell (Figure [4](#F4){ref-type="fig"}); the level of transcript then very rapidly reached a maximum at around 32 hours and steadily decreased thereafter. PFI0785c bears some similarity to both PfHT1 and PFI0955w, but shows a closer resemblance to two putative MFS transporters from Cryptosporidium parvumand a putative plastid hexose transporter from Olea europaea(Additional data file 4). The PFI0785c transcript was almost undetectable until very late in the cycle, with the greatest increase in transcript level occurring between 32 and 40 hours. Five of the P. falciparum-encoded MFS transporters (the PFB0275w, PFE0825w, PF11\_0059, PF14\_0260 and PF14\_0387 proteins) display a weak relationship with members of the \'drug-H^+^antiporter-1\' family. PFB0275w and PF14\_0260 share extensive amino-acid sequence homology with one another and are related to putative transporters from plants (see Additional data file 1 and 4). The expression profiles of these two genes were strikingly different: the PF14\_0260 transcript was present at a low level very early in parasite development and reached a maximum over the 36-42-hour period, whereas transcription of PFB0275w occurred quite late in the cycle (Figure [5](#F5){ref-type="fig"}). The PF11\_0059 protein is weakly related to putative multidrug resistance transporters but also bears some similarity to transporters of another subfamily of the MFS, the anion:cation symporter family (Additional data file 4). It is therefore possible that the PF11\_0059 protein mediates the transport of organic anions, such as glucarate, biotin, phthalate or pantothenate (substrates of the anion:cation symporter family), rather than the efflux of drugs or metabolites such as polyamines, lactose or arabinose (substrates of the drug-H^+^antiporter-1 family). The level of PF11\_0059 transcript increased rapidly between 16 and 24 hours and reached a maximum between 32 and 36 hours, after which it decreased dramatically (Figure [5](#F5){ref-type="fig"}). The closest BLASTP homolog of PFE0825w is a mouse protein designated as a \'putative organic cation transporter\'. However, the mouse protein is not a member of the organic cation transporter family of the MFS, but does show good homology to a tumour suppressing STF-like protein from C. elegansand a weaker similarity to a putative tetracycline resistance protein from Gloeobacter violaceus(see Additional data file 4; sequences of several organic cation transporters are provided for comparison). The PF14\_0387 protein also displays a weak similarity to the G. violaceusprotein as well as to an Escherichia coliputative arabinose effluxer, and in the sequence alignment shown in Additional data file 4, the PFE0825w and PF14\_0387 proteins are placed within the same cluster. The transcription of PF14\_0387 increases rapidly between 16 and 24 hours, after which the level of transcript plateaued and then began to decrease after 36 hours (Figure [5](#F5){ref-type="fig"}). Expression of the PFE0825w gene was not studied. The PFB0465c and PFI1295c proteins share significant sequence similarities and are related to members of the monocarboxylate porter and oxalate:formate antiporter families. From the alignment shown in Additional data file 4, it appears that the PFB0465c protein resembles oxalate:formate antiporters, such as the OxlT-2 from Archaeoglobus fulgidus, whereas the PFI1295c protein is perhaps more similar to members of the monocarboxylate porter family such as the rat T-type amino-acid transporter and the human MCT-8 protein. The PFB0465c and PFI1295c genes had similar expression profiles (Figure [6](#F6){ref-type="fig"}); in both, the maximum level of transcript occurred at approximately 36 hours post-invasion. However, the transcription of PFI1295c began earlier in the development of the intraerythrocytic parasite. The locus PF10\_0360 appears to contain open reading frames (ORFs) for three different proteins. One of these (amino-acid residues 1,644-2,222) displays strong homology to the acetyl-CoA:CoA antiporters of the ER (Additional data file 4). Expression of the PF10\_0360 gene was not studied. MFS-related families -------------------- The malaria parasite encodes members of the glycoside-pentoside-hexuronide:cation symporter (GPH), organo anion transporter (OAT) and folate-biopterin transporter (FBT) families, which are all relatives of the major facilitator superfamily \[[@B45]\]. The PFE1455w protein is a putative Na^+^- or H^+^-driven sugar symporter of the GPH family and the mRNA transcript of this gene was found to be most abundant between 32 and 40 hours post-invasion (Figure [7](#F7){ref-type="fig"}). The MAL6P1.283 protein belongs to a family of putative transporters from bacteria, plants and animals, members of which exhibit weak similarities to proteins and conserved domains of both the MFS and the OAT family (Additional data file 1 and 5). Members of the OAT family catalyze the transport of organic anion and cations and are found only within the animal kingdom. While the MAL6P1.383 protein and its relatives are only weakly similar to OAT proteins, in the absence of a more appropriate classification we have tentatively placed these proteins within the OAT family. Expression of the MAL6P1.383 gene was not studied. The genes MAL8P1.13, PF11\_0172 and PF10\_0215 encode members of the FBT family. Proteins of this family are found only in cyanobacteria, protozoa and plants, and are thought to function as H^+^symporters. Thus far, only protozoan transporters have been characterized and these are known to mediate the uptake of the vitamins folate and/or biopterin (for example, FT1 \[[@B47]\] and BT1 \[[@B48]\] from Leishmaniaand FT1 from Trypanosoma brucei\[[@B49]\]). The MAL8P1.13 and PF11\_0172 proteins share significant sequence similarities and as they are closely related to known or putative FBT proteins (see Additional data file 1 and 5), it is likely that they too catalyze the uptake of folate and/or biopterin. Both compounds contain the pteridine group, and members of this family may also transport pteridine (not itself a vitamin), though this has not been demonstrated directly. There is significant sequence divergence between the PF10\_0215 protein and members of the FBT family and it is quite feasible that this protein transports other metabolites and/or vitamins. The expression profiles of the MAL8P1.13, PF11\_0172 and PF10\_0215 genes are compared in Figure [7](#F7){ref-type="fig"}. A family of novel putative transporters --------------------------------------- We have assigned a putative transport function to 19 P. falciparumproteins that bear no significant sequence similarities to known or putative transport proteins, but which have hydropathy plots that are similar to those of known transporters. Within this group is a set of five proteins (PFA0240w, PFA0245w, PFC0530w, PFI0720w and PF11\_0310) that share both sequence and structural homology, but which lack sequence similarity to any other proteins in the current databases. Several lines of evidence suggest that these proteins may share a common ancestry with transporters of the MFS, and for this reason they have been included in the table in Additional data file 1, where they are designated as P. falciparumnovel putative transporters (PfNPTs). The PfNPTs share a common topology, consisting of 12 TMDs separated by a hydrophilic loop into two sets of six closely spaced TMDs (Additional data file 6). Such a topology closely resembles that found among transporters of the MFS and, consistent with this observation, one of the PfNPTs (PFA0245w) has a putative match to a conserved domain of the MFS. Furthermore, two or more iterations of a PSI-BLAST search of the National Center for Biotechnology Information (NCBI) database using a PfNPT as the query sequence retrieves, with good significance, several putative MFS proteins. A characteristic of most members of the MFS family is the presence of a conserved amino-acid sequence between TMDs 2 and 3 and a related but less conserved motif in the corresponding loop in the second half of the protein (between TMDs 8 and 9). As shown in Additional data file 6, each PfNPT protein contains a putative MFS-specific motif between TMDs 2 and 3 and between TMDs 8 and 9, consistent with the hypothesis that these proteins are distantly related to the MFS. The PFC0530w and PFI0720w genes were found to share a similar pattern of expression over the asexual blood stage of the parasite, whereas the remaining PfNPT genes exhibited quite different expression profiles (Figure [8](#F8){ref-type="fig"}). Amino-acid transporters ----------------------- We have designated six P. falciparum-encoded proteins as putative amino-acid transporters. Three (MAL6P1.133, PFL0420w and PFL1515c) are members of the amino acid/auxin permease (AAAP) family. The other three (PFB0435c, PFE0775c and PF11\_0334) are members of the neurotransmitter:Na^+^symporter (NSS) family. Proteins of the AAAP family are known to mediate the transport of a specific amino acid (for example, the proline permease of Arabidopsis thaliana\[[@B50]\]), or of a group of similar amino acids (for example, the neutral amino-acid permease of Neurospora crassa\[[@B51]\]), while several members exhibit very broad specificities, transporting all naturally occurring amino acids (for example, the general amino-acid transporter of A. thaliana\[[@B52]\]). AAAP proteins are found in yeast, protozoans, plants and animals, and transport is usually either H^+^- and/or Na^+^-dependent \[[@B53]-[@B55]\]. The MAL6P1.133 protein appears to share the greatest level of sequence similarity with amino-acid transporters from other protozoans, yeast and mammals (Additional data file 7). The PFL0420w and PFL1515c proteins are closely related (Additional data file 1) and appear to be most similar in sequence to amino-acid transporters from plants and insects (Additional data file 7). The PFL1515c protein contains a putative signal for targeting to the apicoplast membrane. Substrates of NSS transporters include amino acids, neurotransmitters and other related nitrogenous compounds such as taurine (a sulfonic amino acid) and creatine. NSS proteins are found only in archaea, bacteria and animals, and most of the transporters characterized so far operate via a solute:Na^+^symport mechanism (for example, the tryptophan:Na^+^symporter of Symbiobacterium thermophilum\[[@B56]\] and the mammalian neutral amino acid:Na^+^symporter \[[@B57]\]). Most are also Cl^-^-dependent, for example the neutral and cationic amino acid: Na^+^:Cl^-^symporter of humans \[[@B58]\]. Two exceptions are the absorptive amino-acid transporters - CAATCH1 \[[@B59]\] and KAAT1 \[[@B60]\] - from the gut epithelium of the insect Manduca sexta. These transporters catalyze the Na^+^-dependent (K~m~(Na^+^) ≈ 6 mM) or K^+^- dependent (K~m~(K^+^) ≈ 32 mM) transport of amino acids when expressed in Xenopusoocytes, but the low Na^+^(less than 5 mM) and high K^+^(aproximately 200 mM) concentrations prevalent in the insect gut lumen ensure that these transporters operate predominately via K^+^symport in vivo\[[@B60]\]. The PlasmodiumNSS proteins, while retaining several of the conserved NSS sequence motifs, have diverged considerably from the other family members (Additional data file 1 and 8), making it difficult to ascertain a putative substrate(s) for each transporter. Nevertheless, it does appear that the parasite proteins may bear more similarities to the NSS members which transport amino acids, than they do to those which transport other neurotransmitters or osmolytes. Figure [9](#F9){ref-type="fig"} shows the stage-dependent gene expression for each of the six Plasmodiumputative amino-acid transporters. Significant levels of PFB0435c, PF11\_0334 or PFL0420w transcript were present only in the second 24-hour period of parasite development and expression of the PFL1515c and PFE0775c genes began in earnest only slightly earlier (at around 20 hours). By contrast, there was a relatively high level of MAL6P1.133 transcript early in parasite development and the expression of this gene continued throughout the intraerythrocytic stage. The equilibrative nucleoside transporter family ----------------------------------------------- Members of the equilibrative nucleoside transporter (ENT) family mediate the uptake of nucleosides and/or nucleobases and are present in yeast, protozoa and animals. Transport via ENT proteins is not usually coupled to the movement of a driving ion (hence the name \'equilibrative\'); the exceptions are three electrogenic nucleoside:H^+^symporters from Leishmania donovani\[[@B61]\]. A P. falciparum-encoded ENT, the PF13\_0252 protein, has been characterized in Xenopusoocytes and shown to transport purine and pyrimidine nucleosides and nucleobases (PfENT1 \[[@B62],[@B63]\]) and a second protein (MAL8P1.32) is annotated in the genome as a putative nucleoside transporter. We have identified two further P. falciparumputative nucleoside/nucleobase transporters, PFA0160c and PF14\_0662. Each parasite ENT protein displays a predicted secondary structure that is characteristic of members of the ENT family - 11 TMDs with a large intracellular loop between domains 6 and 7. However, despite this conservation in structure, the four malaria proteins share limited sequence similarities with each other and are only very weakly related to ENT proteins from other organisms (Additional data file 1). The expression profiles of the PFA0160c, MAL8P1.32 and PF13\_0252 genes were similar; in each there was a significant level of transcript present early in parasite development and a rapid increase in transcript abundance occurred between 16 and 24 hours, after which the level of transcript reached a maximum (at around 32 hours) and then declined slowly (Figure [10](#F10){ref-type="fig"}). By contrast, the PF14\_0662 transcript increased in abundance rapidly between 8-20 hours and peaked at approximately 36 h. Inorganic anion transporters ---------------------------- MAL13P1.206 and PF14\_0679 are candidate inorganic anion transporters. The PF14\_0679 protein bears strong sequence similarity to the bacterial members of the large and ubiquitous sulfate permease (SulP) family (Additional data file 1). None of the bacterial SulP proteins has been characterized functionally, but several of the plant members are known to be SO~4~^2-^:H^+^symporters and different mammalian SulP proteins carry out the following types of transport activities: SO~4~^2-^:HCO~3~^-^antiport; HCO~3~^-^:Cl^-^antiport; and the transport of SO~4~^2-^, formate, oxalate, Cl^-^or HCO~3~^-^in exchange for any one of these anions. As depicted in Figure [11](#F11){ref-type="fig"}, the level of PF14\_0679 transcript is low in the first 16 hours of parasite development, but increased steadily thereafter, peaking at approximately 40 hours. The MAL13P1.306 protein belongs to the family of inorganic phosphate transporters (PiT), members of which catalyze the Na^+^- or H^+^-dependent uptake of inorganic phosphate (P~i~). In the official annotation of the genome, the P. falciparumPiT protein (PfPiT) is designated as a putative P~i~:H^+^symporter. Yet in a BLASTP search of the NCBI database PfPiT retrieves the Na^+^-coupled P~i~transporters from animals and yeast with far greater significance than the H^+^-coupled P~i~transporters of bacteria and plant chloroplasts. This observation has been supported by a detailed phylogenetic analysis in which the PlasmodiumPiT protein was found to cluster within the branch of Na^+^-dependent PiT proteins (R.E.M., K. Saliba, A. Bröer, C. McCarthy, M. Downie, R.I.H., R. Allen, S. Bröer and K.K., unpublished work). Subsequent flux experiments performed with trophozoite-stage parasites revealed the presence of a Na^+^-dependent P~i~transporter at the parasite plasma membrane, and the expression of the PfPiT protein in Xenopusoocytes has verified its function as P~i~:Na^+^symporter (R.E.M., K. Saliba, A. Bröer, C. McCarthy, M. Downie, R.I.H., R. Allen, S. Bröer and K.K., unpublished work). As shown in Figure [11](#F11){ref-type="fig"}, the PfPiT (MAL13P1.206) gene was expressed in the early stages of parasite development and the transcript became increasingly abundant after 16 hours, reaching a maximum at around 36-40 hours. The voltage-gated ion channel superfamily ----------------------------------------- Members of the voltage-gated ion channel (VIC) superfamily are found in all domains of life. The channels characterized thus far are specific for K^+^, Na^+^or Ca^2+^under physiological conditions. Potassium channels of this superfamily are usually homotetrameric structures, assembled from a polypeptide subunit possessing six TMDs, and contain a central ion conduction pore (reviewed in \[[@B64],[@B65]\]). Each subunit contains a highly conserved \'selectivity sequence\' in the loop between TMDs 5 and 6, and in the tetrameric structure these loops are positioned together to form a \'selectivity filter\' which determines the cation specificity of the channel. There is also a \'voltage sensor\' in TMD 4, which consists of three to nine regularly spaced, positively charged amino-acid residues. There are three members of the K^+^channel family in the malaria genome (PFL1315w, PF14\_0342 and PF14\_0622), one of which has recently been cloned (PFL1315w \[[@B66]\]). Unlike most members of the family, the P. falciparumpolypeptides are predicted to possess more than six TMDs; hence they were retrieved by our search criteria (which specified proteins with seven or more TMDs). The PFL1315w and PF14\_0622 proteins display limited sequence similarities to known or putative K^+^channels from other organisms and are also only very weakly related to each other (see Additional data file 1 and 9), but both possess the signature selectivity sequence of the K^+^channel family (Figure [12](#F12){ref-type="fig"} and Additional data file 9). A more extensive bioinformatic study of the PFL1315w and PF14\_0622 proteins will be presented elsewhere (R. Allen and K.K., unpublished work). The PF14\_0342 protein is closely related to the PFL1315w protein and a sequence alignment of these two polypeptides reveals that this similarity extends over most of the lengths of the proteins (Additional data file 1 and 9). However, the PF14\_0342 polypeptide has undergone some remarkable changes in the region of the ion-selectivity sequence. The most noteworthy of these are as follows: the insertion of two alanines, the presence of threonine in a position that, in almost every other member of the family, is occupied by aspartic acid, and the replacement of a neutral amino acid by a lysine two residues to the left of this position (Figure [12](#F12){ref-type="fig"}). Figure [13](#F13){ref-type="fig"} depicts the expression profiles of the PFL1315w, PF14\_0342 and PF14\_0622 genes. The PFL1315w and PF14\_0342 transcripts were present in the early stages of parasite development and increased significantly in abundance between 16-24 hours, after which the level of transcript reached a maximum (at around 36 hours) and then declined. The PF14\_0622 gene had a pattern of expression that was strikingly distinct from any other presented in this study; as the intraerythrocytic parasite matured the level of transcript appeared to rise and fall in successive waves of increasing amplitude. Discussion ========== Enrichment of the P. falciparumpermeome ----------------------------------------- In the original annotation of the P. falciparumgenome, the parasite was described as possessing a very limited complement of transport proteins \[[@B3]\]. The detailed bioinformatic analysis presented here reveals that the parasite permeome is at least twice as large as first reported, and predicts the presence of a range of transport capabilities that were assumed previously to be lacking in the parasite. The newly designated proteins include candidate plasma membrane transporters for nutrients such as sugars, amino acids, nucleosides and vitamins. There are also transport proteins predicted to be involved in maintaining the ionic composition of the cell and in the extrusion of metabolic wastes such as lactate. Several of the new transporters are most probably located on intracellular membranes. Some of these are predicted to catalyze the flux of solutes either into or out of an intracellular compartment (for example, the putative iron effluxer of the digestive vacuole, PFE1185w), whereas others are predicted to mediate the exchange of metabolic intermediates between the cytosol and an organelle lumen (for example, the putative GDP-fucose:GMP antiporter of the Golgi, PFB0535w). A number of the P. falciparumproteins we retrieved with seven or more TMDs bear no significant sequence similarity to any other proteins (transporters or otherwise) characterized previously. Yet they have hydropathy plots that are similar to those of known transport proteins, consistent with the hypothesis that they too are transporters. Within this group is a subset of related proteins that form a novel family of putative transporters, which may be very distantly related to the MFS. These novel putative transporters appear to be specific to plasmodia and are therefore of potential interest as new antimalarial drug targets. This enrichment in the repertoire of P. falciparum-encoded transport proteins indicates that the parasite permeome is not as impoverished as originally thought (although the parasite still cannot be considered to have a transporter-replete genome, see below). For instance, in the original study of the genome data it was suggested, on the basis of the apparent absence of an obvious amino-acid transporter, that the intraerythrocytic parasite must rely almost completely on the ingestion and digestion of host hemoglobin for its supply of amino acids \[[@B3]\]. However, the identification here of several putative amino-acid transporters, along with previous observations that the parasite is capable of both the import \[[@B67]\] and export \[[@B68]\] of amino acids, indicates that this is not the case. The six putative amino-acid transporters we identified display dissimilar mRNA expression patterns, suggesting that they fulfill different roles in the rapid development of the intraerythrocytic parasite. For example, the MAL6P1.133 protein is most closely related to amino-acid transporters from other protozoans, yeast and mammals, and the expression of this gene throughout the intraerythrocytic stage (Figure [9](#F9){ref-type="fig"}) suggests that the transporter has an important role in parasite growth, perhaps as a broad-specificity plasma membrane permease for amino acids. On the other hand, the PFL1515c protein is more similar to plant amino-acid transporters, the gene is expressed slightly later in parasite development (Figure [9](#F9){ref-type="fig"}), and the presence of a putative apicoplast targeting signal indicates that the protein probably mediates the transport of amino acids into and/or out of this organelle. P. falciparum-encoded channels -------------------------------- In their landmark paper Gardner et al.\[[@B3]\] reported \"no clear homologs of eukaryotic sodium, potassium or chloride ion channels could be identified\". However, two putative K^+^channels have since been cloned \[[@B13],[@B66]\], and we have identified an additional putative K^+^channel (PF14\_0622) as well as a novel protein of the K^+^channel family (PF14\_0342). Our analysis of the P. falciparumgenome did not reveal any putative Cl^-^channels, and in this respect, our findings agree with the original annotation. Most organisms, including many other lower eukaryotes (for example, Dictyostelium discoideum, Entamoeba histolyticaand various species of fungi), are known to encode at least one member of the ClC chloride channel family. ClC proteins possess 18 alpha helices \[[@B69]\], 10-12 of which are typically detected as putative TMDs by a TMD prediction program, yet our search criteria, which specified seven or more TMDs, did not retrieve a P. falciparum-encoded ClC protein. To verify this result, we carried out four iterations of a PSI-BLAST search of the NCBI database using the E. coliClC protein (gi:26106498) as the query sequence. The search retrieved 695 ClC proteins from 226 different organisms, including many prokaryotes, unicellular eukaryotes, plants, and a broad range of animals, but a PlasmodiumClC homolog was not amongst the retrieved proteins. Also absent from the list of retrieved organisms were two other lower eukaryotes, the apicomplexan protozoan C. parvumand the microsporidian E. cuniculi, both of which have been reported as lacking a ClC protein \[[@B70],[@B71]\]. P. falciparum, C. parvumand E. cuniculiare all obligate, intracellular parasites that reside in the cytoplasm of the host cell, and to date they are the only eukaryotes that appear to lack a ClC protein. It is tempting, therefore, to speculate that the loss of ClC proteins in these organisms is related to their parasitic, intracellular life style. A primary role of plasma membrane Cl^-^channels in non-excitable cells is in the volume-regulatory response to cell volume perturbation \[[@B72],[@B73]\]. It is likely that within the relatively sheltered environment of the host cytoplasm, the parasite is not exposed to significant permutations in osmolarity and that it therefore does not require Cl^-^channels for this purpose. Furthermore, Cl^-^channels in the parasite plasma membrane would facilitate the distribution of Cl^-^ions in accordance with the membrane potential. The membrane potential across the membrane of the mature trophozoite-stage parasite has been estimated as -95 mV \[[@B74]\]. The \[Cl^-^\] in the erythrocyte cytosol is of the order of 95 mM \[[@B75]\] and if Cl^-^were allowed to distribute between the erythrocyte and parasite cytosols on the basis of the membrane potential the \[Cl^-^\] in the parasite cytoplasm would be of the order of 3 mM (calculated via the Nernst equation). On the basis of X-ray microanalysis data \[[@B76]\] it is likely that the \[Cl^-^\] in the parasite cytoplasm is an order of magnitude higher than this, consistent with Cl^-^being actively accumulated by the parasite, rather than being allowed to equilibrate via Cl^-^channels. The accumulation of Cl^-^by the parasite might be mediated by either a Cl^-^:H^+^symporter (as is found in plants and fungi \[[@B77],[@B78]\]) or perhaps a Cl^-^:anion exchanger (such as the protein we have annotated as a putative inorganic anion exchanger, PF14\_0679). The one role in which Cl^-^channels have been implicated in the physiology of the intraerythrocytic parasite is in the formation of the \'new permeability pathways\' (NPP) that mediate the increased traffic across the host erythrocyte membrane of a wide range of low molecular weight solutes, including polyols, amino acids, sugars, vitamins, and both organic and inorganic (monovalent) anions and cations \[[@B79]\]. The NPP show a marked preference for anions over cations, and for K^+^over Na^+^\[[@B80]\]. Their transport properties are those expected of anion-selective channels \[[@B81]\], and electrophysiological studies have confirmed the presence of such channels in the infected erythrocyte membrane \[[@B82]-[@B84]\]. It has been proposed that these channels are endogenous proteins, activated by stresses or stimuli associated with the parasite\'s invasion of the host cell \[[@B83],[@B84]\]. However, the recent report of strain-specific variations in the properties of a novel inwardly rectifying anion-selective channel induced by the parasite in the host cell membrane is consistent with the hypothesis that this channel (termed PESAC for Plasmodiumerythrocyte surface anion channel \[[@B85]\]) is, instead, parasite-encoded \[[@B86]\]. If this is the case, the lack of an obvious Cl^-^channel in the P. falciparumgenome indicates that it is likely to be a novel type of channel. One protein that warrants further consideration in this context is PF14\_0342, a very unusual member of the K^+^channel family; it is distinguished from all other K^+^-channel family proteins by the presence of several significant mutations in the region of the ion-selectivity sequence. These include the substitution of a highly conserved aspartic acid by threonine, the mutation of a neutral amino acid to a lysine two residues to the amino terminus of this position, and the insertion of two alanines (Figure [12](#F12){ref-type="fig"}). In K^+^channels, the conserved aspartic residue is located at the extracellular edge of the pore \[[@B64]\], where it may create an electrostatic field that \'funnels\' K^+^ions into the channel opening. The replacement of this acidic residue by a neutral amino acid, combined with the appearance of a positively charged lysine residue nearby, raises the possibility that the ion selectivity of the PF14\_0342 channel is no longer strictly cationic, and may even be anionic. The significance of the inserted alanines is difficult to predict, but they may serve to enlarge the diameter of the selectivity filter and thereby permit the transport of larger solutes. The putative ion channel PF14\_0342 might therefore be considered as a candidate for the parasite-induced NPP, and the localization of the protein within the infected cell, and its physiological characteristics, are presently under investigation. If the PF14\_0342 protein is indeed a component of the NPP, it may be anticipated that strain-specific differences in the electrophysiological characteristics of the parasitized erythrocyte \[[@B86]\] will correlate with a difference(s) in the amino-acid sequence of PF14\_0342. Na^+^-dependent transporters and the physiological role of the NPP ------------------------------------------------------------------ Shortly after invasion by the malaria parasite, the concentration of Na^+^in the host erythrocyte cytosol is similar to that in uninfected erythrocytes, and to that in the cytosol of the parasite itself \[[@B76]\]. There is, therefore, little if any Na^+^concentration gradient across the parasite plasma membrane. With the induction of the NPP at around 12-15 hours, however, there is a progressive leakage of Na^+^into, and K^+^out of, the infected erythrocyte \[[@B80]\], resulting in a marked increase in the \[Na^+^\] in the erythrocyte cytosol and, therefore, a substantial inward Na^+^concentration gradient across the parasite\'s plasma membrane \[[@B76]\]. Our identification of several putative Na^+^-coupled transporters in the malaria parasite genome suggests that a significant component of the parasite\'s metabolism might depend on Na^+^-driven transport processes; the influx of Na^+^via the NPP, and the consequent increase in \[Na^+^\] in the infected erythrocyte cytosol may be important for this reason. The Na^+^-dependent P~i~transporter (PfPiT; R.E.M., K. Saliba, A. Bröer, C. McCarthy, M. Downie, R.IH., R. Allen, S. Bröer and K.K., unpublished work) provides one example of how this Na^+^gradient can be used to drive the accumulation of an essential nutrient. Other likely candidates for transporters able to utilize the Na^+^gradient across the parasite plasma membrane to energize solute transport include the three putative amino acid:Na^+^symporters of the NSS family, the putative Na^+^- or H^+^-driven sugar symporter of the GPH family, the putative MATE antiporter, and one or more of the P. falciparumMFS transporters. Standardization of gene-expression levels and stage-specific changes in the total RNA content of intraerythrocytic P. falciparum ---------------------------------------------------------------------------------------------------------------------------------- The level of a target mRNA in a sample is usually standardized to an internal reference, such as the expression of a \'housekeeping gene\', rRNA or total RNA, in order to compare the relative abundance of the transcript between different cell samples. Housekeeping genes, such as that for glyceraldehyde-3-phosphate dehydrogenase (GAPDH), were originally thought to be expressed at constant levels, regardless of cell type, developmental stage or experimental manipulation. However, it has become increasingly evident that the transcripts of housekeeping genes are not always maintained at constant levels in the cell \[[@B44],[@B87],[@B88]\]; nor is it likely that there are mRNA species which are. Likewise, the level of rRNA in the cell is also known to vary in relation to factors such as cell type and age \[[@B88],[@B89]\]. Hence, the use of either a housekeeping gene mRNA or rRNA as an internal standard has lost merit \[[@B88]\], and standardization to total RNA has emerged as the method of choice for comparing mRNA levels between different cell samples \[[@B90],[@B91]\]. Standardization to total RNA may be an appropriate strategy when the (average) amount of total RNA per cell is similar in each of the samples. It is less appropriate for quantifying levels of gene expression between cells of grossly differing transcriptional activities \[[@B88],[@B90],[@B91]\]. In cells that are very transcriptionally active (and hence contain a high total RNA content) a target transcript will appear to be at a disproportionately low level in comparison with that quantified in transcriptionally quiescent cells (containing less total RNA) in which the actual copy number of the target transcript is the same. In this study we measured the total RNA content of malaria parasites as they progressed through the intraerythrocytic lifecycle and found there to be around 160 times more RNA in late trophozoites/schizonts (around 40 hours old) than in young ring-stage parasites (around 4 hours old) (Figure [3](#F3){ref-type="fig"}). This is in agreement with previous findings of a considerably elevated rate of transcription in trophozoites compared with ring-stage parasites \[[@B92],[@B93]\]. The substantial change in the RNA content of the parasite did not occur as a constant, linear increase from invasion through to maturity. Rather, the level of RNA remained very low early in the parasite\'s occupation of the erythrocyte (increasing only slightly in the first 16 hours), then increased rapidly between 20 to 40 hours and declined at 42 hours (Figure [3](#F3){ref-type="fig"}). This pattern correlates well with the onset at 20-30 hours of a broad range of metabolic activities in the parasite \[[@B92],[@B94]-[@B96]\] and of the eventual downregulation of many of these activities in the late stages (schizont/segmenter) of parasite maturation \[[@B97]-[@B99]\]. Two recent large-scale studies have analyzed the transcriptome of the malaria parasite as it progresses through the intraerythrocytic cycle \[[@B42],[@B43]\]. The Le Roch et al.\[[@B43]\] study examined the mRNA levels of approximately 95% of the predicted P. falciparumgenes at six points in the intraerythrocytic cycle, whereas Bozdech and colleagues \[[@B42]\] measured gene expression at 1-hour intervals over the complete 48-hour cycle, thereby providing an impressive and comprehensive investigation of the transcriptome of asexual blood-stage P. falciparum. Both of these studies standardized mRNA levels to total RNA, which hinders gene-wise comparisons with the present work. Nevertheless, it should be noted that the extent of similarity between the mRNA expression profiles obtained when standardizing to cell number (as in this study) and those obtained when standardizing to total RNA (as in the previous studies) depends to a large extent on the amount of transcript present in the ring-stage (the first 16-20 hours of the intraerythrocytic phase). For example, for those genes for which there is a substantial amount of transcript present in the ring stage, and which undergo a further increase during the second half of the intraerythrocytic cycle (for example, PFC0530w, PF11\_0310 and MAL6P1.133), normalization of transcript level relative to total RNA will tend to show the mRNA level to be initially elevated relative to that seen in mature parasites and to then undergo a decrease as the parasite matures, whereas normalization to cell number will show a progressive increase in transcript level as the parasite matures (for example, Figures [8](#F8){ref-type="fig"} and [9](#F9){ref-type="fig"}); that is, the two different types of profiles will look quite different. By contrast, for those genes for which transcript is rare or absent in ring-stage parasites but increases as the parasite matures (for example, PFB0435c, PFL0170w and PFA0245w), then the profiles obtained using the two different normalization methods are likely to share some resemblances, particularly in the second half of the intraerythrocytic cycle, where both modes of normalization will show increases in transcript levels. In summary, the relationship between the two datasets is highly complex, and the apparent differences or similarities observed between the profiles produced using these different methods will depend on such factors as the stage at which the transcription of a given gene begins, the rate at which the level of transcript increases, and the point at which the level of transcript reaches a maximum. Standardization of mRNA levels to cell number, as in the present study, allows the level of a given transcript to be measured independently of the expression of other genes, rRNA or total RNA, all of which are parameters that are likely, or known, to change drastically during the intraerythrocytic cycle. The expression profiles we present reflect the number of copies of the transcript inside the cell and illustrate how this quantity changes as the parasite progresses from early ring stage through to schizont stages. From this it is immediately apparent how the expression of the gene is being regulated - independent of any other variable. Standardization of transcript levels to cell number may also provide more insight into the likely biological role of the encoded protein, than standardization to RNA. This is best illustrated with a specific example. The parasite hexose transporter, PfHT1, provides the major route for the influx of glucose \[[@B46],[@B100],[@B101]\], an essential nutrient required by the intraerythrocytic malaria parasite for the production of ATP (via glycolysis). The rate of glycolysis is highly stage specific; in ring-stage parasites the level of glucose metabolism does not differ greatly from that measured in uninfected erythrocytes, but as the young parasite matures into a trophozoite there is a dramatic increase in the rate of glycolysis and the maximum level of glucose consumption occurs in the schizont stage \[[@B94],[@B102],[@B103]\]. Given the central role of PfHT1 in glucose uptake \[[@B46],[@B100],[@B101]\], it might be expected that the profile of PfHT1expression will reflect the significant increase in the demand for glucose uptake by the maturing parasite. Indeed, when gene expression is standardized to cell number, the greatest increase in the level of PfHT1 mRNA occurs at the transition between the ring and trophozoite stages of the parasite (16-24 hours) and the maximum level is reached early in the schizont stage (Figure [4](#F4){ref-type="fig"}). Yet when standardized to total RNA (see \[[@B104]\]), the level of PfHT1transcript appears to be most abundant in the ring-stage parasite (when relatively little glycolysis is occurring) and decreases rapidly as the parasite develops into a trophozoite, remaining at a low level throughout the parasite\'s most intense period of growth. Trends in the expression profiles of P. falciparumtransport genes ------------------------------------------------------------------- All the genes for which mRNA expression was analyzed were transcriptionally active during the asexual intraerythrocytic stage of P. falciparum, albeit for varying durations. For 13 of the 34 genes investigated there was a readily detectable amount of transcript present from 8 hours through to 42 hours; these included genes encoding proteins known to transport inorganic phosphate (PfPiT), glucose (PfHT1) and nucleosides (PfENT1) and for putative transporters of nucleosides/nucleobases (PFA0160c and MAL8P1.32), amino acids (MAL6P1.133), monocarboxylates (PFI1295c) as well as a putative K^+^channel (PFL1315w) and novel ion channel (PF14\_0342). Three of the 13 genes encode for members of the novel putative transporter family (PFC0530w, PFI0720w and PF11\_0310) and while the likely substrate(s) of each of these transporters are unknown, the expression of the genes from early ring-stage parasites through to late schizonts suggests that these proteins each play an important role in the biochemistry of the intraerythrocytic parasite, presumably by mediating the transport of an important solute(s). By 16 hours post-invasion, the number of genes for which transcript could be detected had increased from 13 to 25, although in many cases the level of mRNA was still quite low (for example, the putative anion exchanger (PF14\_0679) and three amino-acid transporters (PFE0775c, PF11\_0334 and PFL1515c) amongst others). The transcripts of a few genes (PF14\_0387, PF14\_0342, PF14\_0662 and PFL1315w) underwent a rapid increase in abundance between 16 and 20 hours, but for the majority of transport genes the rate of increase in transcript level was highest either between 20 to 24 hours (17 genes) or 24 to 32 hours (seven genes). There is a small subset of transport genes which have expression profiles that depart from this trend. Transcripts of three transport genes (PFB0275w, PFB0435c and PFI0785c) were not detected until very late in the intraerythrocytic cycle (at 32 hours) and reached maximum abundance at 40-42 hours. These genes encode for putative transporters of drugs and metabolites, amino acids and sugars, respectively, and their selective expression in the late trophozoite and schizont stages suggests that the proteins are required for the development of late schizonts and/or merozoite morphogenesis. Perhaps the most intriguing pattern of gene expression presented in this study is that of the putative K^+^channel PF14\_0622. Most transport genes in this study display a monophasic expression profile, with a single maximum and a single minimum, indicating that the gene is activated for transcription only once during the intraerythrocytic stage. This has been reported to be the case for the majority of P. falciparumgenes expressed in the asexual blood cycle \[[@B42]\]. However, the expression of the PF14\_0622 gene is strikingly distinct, in that the level of transcript rose and fell in successive waves of increasing amplitude as the parasite developed. The physiological significance of this expression pattern is unclear. Previous work has revealed that the timing of gene expression can determine the cellular localization of the resulting protein; when the apical membrane antigen-1 gene (AMA-1) is expressed in the late schizont/segmenter stage, the AMA-1 protein is targeted to the rhoptries (its normal destination), but the expression of AMA-1in maturing trophozoites and during early schizogony (when rhoptries are absent) results in AMA-1 being targeted to the parasite plasma membrane as well as to a cytoplasmic location \[[@B105]\]. Therefore, one possibility could be that the PF14\_0622 channel is being targeted to different membranes within the parasitized cell, depending upon the timing of expression. Comparison of the P. falciparumpermeome with that of other organisms ---------------------------------------------------------------------- The 54 transport proteins that were originally identified in the malaria genome account only for approximately 1% of the genes encoded by P. falciparum, and although our analysis has increased this to around 2.1% (109 proteins), this is still low, even when compared to other seemingly transporter-deficient microorganisms. For example, the archeon Methanococcus jannaschiiis currently the prokaryote with the lowest percentage of transport proteins in its genome \[[@B106]\], but at 2.4% this is still higher than that found to date in P. falciparum. The intracellular parasite E. cuniculihas a remarkably reduced genome (around 2.9 Mb \[[@B71]\]) and encodes only 43 transport proteins \[[@B12]\]. However, these account for 2.2% of its genes; this unicellular eukaryote is therefore also slightly more transporter-rich than the malaria parasite. Other eukaryotic genomes have higher proportions of transport proteins; for example, S. cerevisiae(4.2%), A. thaliana(3.2%), Drosophila melanogaster(4.6%) and Homo sapiens(around 3.4%). At the high end of the spectrum is the E. coligenome, with an impressive repertoire of transporters accounting for 7.1% of the total number of genes \[[@B12]\]. The relative abundance of transport proteins can also be measured in terms of the number of transport genes per Mb of DNA. For the P. falciparumgenome this is 4.7 (up from 2.3 estimated from the original annotation) transport genes per megabase, compared with E. cuniculi(17.2 per Mb), S. cerevisiae(22 per Mb) and an average of 36 transport genes per Mb across 18 species of prokaryotes \[[@B107]\]. The Plasmodiumcatalog of transport proteins is even more conspicuously meager when compared to the transporter-rich genomes of E. coli, Bacillus subtilisand Hemophilus influenzae, which have 66, 63 and 52 transport genes per Mb, respectively \[[@B12]\]. However, although the genomes of higher eukaryotes typically encode hundreds of transport proteins, they too have low numbers of transport genes relative to genome size (for example, A. thaliana, 6.7 per megabase and D. melanogaster, 5.3 per Mb). The human genome, which has the largest number of transport proteins (around 1,200), has the lowest relative abundance of transport genes to date (0.37 per Mb \[[@B12]\]). The conclusion that the malaria parasite is \'minimalistic\' with regard to transporters implies that there may well be relatively little redundancy (that is, the parasite tends not to have multiple transporters for particular roles). Compounds that inhibit a single transporter may therefore be highly effective as antimalarials, as the parasite is unlikely to have alternative transporters that it is able to use for the same purpose. Future work ----------- Although the analysis reported here has increased significantly the number of transport proteins predicted to be present in P. falciparumit is highly likely that more parasite-encoded transport proteins remain to be uncovered. Our search criteria were targeted towards identifying transporters with seven or more TMDs; however, many of the polypeptides which form channels possess fewer than six TMDs and several types of transporters also contain fewer than six TMDs. The fact that the parasite has three putative channels (PFL1315w, PF14\_0342 and PF14\_0622) indicates that these types of transport proteins are indeed present in the P. falciparumgenome. Applying the methodology used here to proteins which possess six or fewer TMDs is likely to identify additional transport proteins. The identification of genes and the prediction of intron-exon structures in the P. falciparumgenome are still being refined. For a number of the previously unannotated transport proteins there were inappropriate predictions for the 5\'/3\' ends and/or intron-exon boundaries (for example, PF14\_0387, PFI1295c, PF10\_0360, PF10\_0215) and it is most likely this that led to their being overlooked in the original annotation process. Subsequent revisions of the genome data, using the latest gene-finding tools, comparative genomics and the integration of full-length cDNA sequences and proteomics data, will greatly improve the existing annotation of the genome and this, in turn, is likely to lead to further additions to the P. falciparumpermeome. Materials and methods ===================== Identifying putative membrane transport proteins ------------------------------------------------ A file containing the amino-acid sequences of all predicted proteins from the genome was obtained from the official Plasmodiumgenome database, PlasmoDB \[[@B108],[@B109]\]. Polypeptides with seven or more TMDs were retrieved from this dataset using a computer program described previously \[[@B9]\]. Briefly, in an automated process, each polypeptide sequence was converted to a hydropathy plot and the number of peaks, corresponding to putative TMDs, detected. Proteins which satisfied the search criteria (those having between 250 and 5,000 residues in length and with seven or more TMDs; see Additional data file 10 for full details) were retrieved, and duplicate sequences were removed. The TMD-based search tool at PlasmoDB was used to search for further putative membrane proteins. In these analyses, either the TMHMM2 or TMpred algorithms were used to scan for proteins possessing 7-25 TMDs. The list of proteins retrieved by the TMHMM2 and TMpred programs was compared with that already retrieved using the original program, and while there were a few proteins that the original program had retrieved and PlasmoDB had not, and vice versa, the results were mostly in agreement. The additional proteins identified at PlasmoDB were included in the subsequent annotation studies. Annotation of proteins ---------------------- A combination of bioinformatics tools was used to assign putative functions to the (probable) membrane proteins retrieved from the hydropathy plot analysis. Each sequence was queried against the NCBI nonredundant protein database (using BLASTP \[[@B110]\]) and the Entrez Conserved Domain Database (using reverse position-specific BLAST \[[@B111]\]) in order to determine its relationship to other proteins. The values presented in Additional data file 1 are from analyses carried out in the second half of 2003. In some instances, there was only a weak similarity between the P. falciparumprotein and a Conserved Domain Database entry and/or the closest (non-Plasmodium yoelii) BLASTP homolog. For such cases, the possibility of a common ancestry between the Plasmodiumprotein and proteins from other organisms was explored further by performing several iterations of a PSI-BLAST \[[@B112]\] search of the NCBI database. Comparisons of both protein sequence and predicted secondary structure (see below) were used to evaluate alternate models for each gene; in several cases it was found that a gene prediction other than the \'official\' model provided a more likely candidate protein. Where applicable, P. falciparumproteins were placed into known transport protein families according to the official transporter classification system \[[@B10]\]. Secondary structure predictions ------------------------------- Related transporters (comprising a protein family) show marked similarities in their predicted secondary structures, even when they share a relatively low level of sequence homology \[[@B113]\]. The hydropathy plot of each P. falciparumprotein was compared with that of its BLAST homologs to investigate whether the observed sequence homologies corresponded to a similarity in the predicted structures of the proteins. Such an analysis is of particular assistance when the sequence similarity shared by two proteins is weak; a good agreement between their respective secondary structures adds support to the hypothesis that these proteins share a similar function. It also serves to identify those P. falciparumputative transporters that are truncated or fused to proteins from adjacent genes as a result of errors in the prediction of gene structures. The number and spatial arrangement of putative membrane-spanning domains in a polypeptide was predicted using TMpred \[[@B114]\] and TMMHM v2.0 \[[@B115]\]. Construction of alignments -------------------------- The ClustalW program \[[@B116]\] in MacVector 7.1 was used to generate and edit all alignments. Alignments were converted to PDF form and compiled in Adobe Photoshop 6.0.1. Full-length versions of the alignments presented or mentioned in this paper are available from the authors upon request. Preparation of P. falciparum-infected red blood cells ------------------------------------------------------- Human red blood cells (type O^+^) infected with P. falciparum(strain FAF6) were cultured as described previously \[[@B74]\] using a method adopted from Trager and Jensen \[[@B117]\]. The hematocrit was maintained at 4% and the parasitemia at 13-16%. Tightly synchronized cultures were achieved by repeated applications of sorbitol lysis \[[@B118]\] over a 9-day period immediately before the RNA extractions commenced. Cell counts were made using an improved Neubauer counting chamber and both the culture parasitemia and the parasite growth stage were assessed by methanol-fixed, Giemsa-stained blood smears. Photographs of the blood smears were taken with a SPOT RT colour camera connected to a Leica DML microscope and images of parasite-infected red blood cells were compiled in Adobe Photoshop 6.0.1. Isolation of total RNA ---------------------- Culture samples were collected at nine stages over a single intraerythrocytic growth cycle of the parasite and total RNA was extracted using the NucleoSpin RNA II kit (Macherey-Nagel) in accordance with the \'high yield protocol\' supplied by the manufacturer. This procedure is reported to recover 90-100% of the RNA bound to the column and in control experiments in which as little as 200 ng of RNA was loaded onto a column we confirmed this to be the case (data not shown). A NucleoSpin column can bind a maximum of 100 μg of nucleic acid and the manufacturers recommend that no more than 10^9^cells should be loaded per column; exceeding this quantity may cause clogging of the column, leading to poor RNA yields. However, approximately 85% of the cells in a parasite culture are mature human red blood cells, which lack a nucleus or other organelles and do not produce nucleic acids. We therefore investigated what quantities of parasite-infected red blood cell culture could be loaded onto a column without compromising the RNA yield. Initial experiments revealed that trophozoite-stage parasites contain considerably greater quantities of RNA and DNA than the less mature ring-stage parasites. We determined that for ring-stage parasites, increases in the cell sample from 9 × 10^6^(6 × 10^7^including uninfected red blood cells) to 9 × 10^7^(6 × 10^8^), and even up to 2.7 × 10^8^(1.8 × 10^9^), still gave proportionate increases in the RNA yield, indicating that the binding capacity of the column was not exceeded within this range. For mature trophozoites, however, the relationship between the cell sample size and the amount of RNA extracted became nonlinear when more than 6 × 10^7^(4 × 10^8^) cells were applied to the column. The quantities of parasites from which RNA was extracted (per column) are as follows: ring-stage, 1.3 × 10^8^parasites (8.7 × 10^8^cells in total including uninfected red blood cells); late ring stage/early trophozoite, 8.8 × 10^7^parasites (5.86 × 10^8^cells in total); and mature trophozoite/schizont, 4.4 × 10^7^parasites (2.93 × 10^8^cells in total). The amount of RNA extracted from the first time point (\~4 h post-invasion) was exceedingly low (\< 0.3 μg/10^8^parasites) and insufficient to warrant this sample\'s inclusion in the subsequent gene-expression analyses. Reverse transcription --------------------- cDNA was synthesized from 2 μg of total RNA (DNA-free) using SuperScript II RNase H^-^Reverse Transcriptase (Invitrogen). First-strand synthesis was primed with Oligo-dT (Invitrogen) and the relative dNTP concentrations were adjusted to reflect the AT-bias of P. falciparumDNA (40% dATP, 40% dTTP, 10% dCTP, 10% dGTP). The disaccharide trehalose was added to the reaction (0.6 M) to improve the efficiency of reverse transcription. The action of trehalose is twofold: it stabilizes, and even activates, the reverse transcriptase protein at unusually high temperatures, and regions of secondary structure in the transcript (that otherwise cause the premature dissociation of the enzyme) are reduced or eliminated by both trehalose and the increased temperature of the reaction \[[@B119]\]. Reaction conditions were as follows: 42°C for 30 min, 60°C for 1.5 h, then 70°C for 15 min to inactivate the reverse transcriptase. PCR --- Changes in mRNA levels throughout the asexual blood stage of the parasite of 34 proven/putative transport genes were semi-quantified by two-step RT-PCR using the principles developed by Halford and colleagues \[[@B120],[@B121]\] and Fuster et al.\[[@B122]\]. As described above, the amount of total RNA in the cell increases significantly as the parasite matures (see Figure [3](#F3){ref-type="fig"} for quantitative analysis). For this reason, for each time point the quantity of cDNA added to the PCR was standardized to cell number rather than to total RNA (see Discussion). PCR was performed using the Platinum TaqPCRx DNA polymerase Kit (Invitrogen). Each reaction had a final volume of 50 μl and consisted of: 1x PCRx amplification buffer, 1x PCR enhancer solution, 1.5 mM MgSO~4~, dNTPs (320 μM dATP, 320 μM dTTP, 80 μM dCTP and 80 μM dGTP), 1.25 U Platinum TaqDNA Polymerase, 2 μM of each primer and 5 μl of template cDNA (at an appropriate dilution). Samples were amplified in a MJ Research PTC-200 Peltier thermal cycler and 20 μl of each reaction was loaded onto a 1% (w/v) agarose gel containing 0.5 μg ethidium bromide/ml. Electrophoresis was carried out at 70 V for exactly 90 min in a B2 Owl Separation System. On a separate gel, a duplicate aliquot of 20 μl from each reaction was loaded (in the reverse order) and electrophoresed in the same manner. PCR products were visualized with a UV transilluminator and the gels were photographed using a Gel Doc System camera linked to a computer running NIH Image software. Densitometric analysis of gel images was performed using ImageQuant V3.3 software (Molecular Dynamics). Halford et al.\[[@B120]\] have shown that in a PCR where the concentration of template is limiting and the amplification of primer-dimers is occurring, the yield of product is dependent upon the logarithm of template cDNA input, even after 35 cycles of amplification. Furthermore, this relationship holds from the lowest concentration of template that gives a detectable yield, to amounts that are at least 100-fold (and up to 1,000-fold) greater than this concentration. A preliminary PCR revealed that the amount of product synthesized from trophozoite-stage cDNA varied between the 34 genes - a reflection, perhaps, of differences in transcript levels between these genes, differences in primer efficiency, or differences in the efficiency at which a given transcript species was reverse transcribed into cDNA. A subset of six genes, representative of this range in PCR product yield, was selected for inclusion in an experiment aimed at establishing a set of conditions under which, for each gene and cDNA sample, the final yield of PCR product was proportional to the amount of starting cDNA template. In brief, cDNA from ring (\~20 h old), trophozoite (\~32 h old) and schizont-stage (\~42 h old) parasites was serially diluted and PCR product yields were measured after 15, 20, 25 or 30 cycles of amplification. It was found that using a moderately high dilution of cDNA (equivalent to around 3.8 × 10^4^parasites per 50 μl PCR), and amplification for 25 cycles, resulted in a proportional relationship between the concentration of input cDNA and the yield of PCR product for most genes and cDNA samples (data not shown). However, for a small subset of genes (PFA0245w, PFC0530w, PFE0775c, PFE1455w, PFI0720w, PF11\_0334, PFL0420w, PF14\_0622), these conditions did not result in quantities of product that could be measured reliably by densitometry and these genes required an additional five rounds of amplification before the yields could be measured reliably and the results were reproducible. Polymerase chain reactions were thermal cycled as follows: 94°C for 2 min, then 25 or 30 of 94°C for 30 sec, 55°C for 30 sec and 68°C for 45 sec. The absence of contaminating DNA in the RNA samples was verified by the lack of PCR products formed in reactions (30 cycles) carried out for each primer set using RNA as the template (that is, \'minus RT\' controls). The extraction of RNA from parasites over the intraerythrocytic growth cycle was performed twice (4 months apart) and gene expression was analyzed within a time course. PCR product yields for a given gene are presented as a ratio of the signal measured for the specified time point, divided by that of the time point which gave the greatest yield of product. Ratios from replicate gels from the same PCR were averaged and the data from time course 1 and 2 (each consisting of at least n= 2 polymerase chain reactions) were combined to give the mean ± standard error. Each 50 μl PCR contained an amount of cDNA equivalent to around 3.8 × 10^4^parasites and product yields were measured from 2 × 20 μl aliquots of the reaction. Thus in this study, the relative level of transcript at each growth stage was estimated from around 1.5 × 10^4^parasitized cells. Primers ------- Oligonucleotide primers that would amplify a product of approximately 400 bp from each transcript of interest were designed using Primer3 \[[@B123]\]. These were synthesized by Invitrogen and for each primer pair, the gene identification, primer sequences, and product size are provided in Additional data file 10. Additional data files ===================== The following additional data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"}, [2](#S2){ref-type="supplementary-material"} and [3](#S3){ref-type="supplementary-material"} contain tables that summarize the known and putative transport proteins of P. falciparum. Additional data file [4](#S4){ref-type="supplementary-material"}, [5](#S5){ref-type="supplementary-material"}, [6](#S6){ref-type="supplementary-material"}, [7](#S7){ref-type="supplementary-material"}, [8](#S8){ref-type="supplementary-material"} and [9](#S9){ref-type="supplementary-material"} contain protein sequence alignments; Additional data file [10](#S1){ref-type="supplementary-material"} contains supplementary methods. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Proteins that share sequence similarities with known or putative transport proteins and/or conserved domains of transport protein families ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Proteins that are related to \'hypothetical proteins\' from other organisms and have predicted secondary structures that resemble those of characterized transport proteins, but which do not share sequence similarities with known or putative transport proteins and/or conserved domains of transport protein families ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Plasmodium-specific proteins that have the (predicted) structural characteristics of a transporter, but which, apart from strong sequence similarities to hypothetical proteins from other Plasmodiumspecies, do not display any similarities to proteins or conserved domains in the current databases ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 The region over TMDs 1-5 of the alignment is shown. Sequences are separated into clusters of well-related sequences and are grouped as subfamilies of the MFS. P. falciparumsequences are boxed and the protein designators highlighted. For proteins of other organisms, the NCBI accession (gi) number and the known or putative (p) substrate specificity of the transporter are given. In some proteins, one or more of the extramembrane loop regions have been truncated and this is indicated by a solid black line. Residues are shaded as follows: positively charged, blue; negatively charged, red; hydoxyl, orange; amido, grey; proline, green; cysteine, purple; histidine, mid blue; glycine, light blue; tryptophan and tyrosine, olive green; remaining nonpolar, yellow ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Both transporter families are distantly related to the MFS. The region over TMDs 2-5 and TMD 8 of the organo anion transporter family alignment is shown. The region over TMDs 3-5 and TMD 10 of the folate-biopterin transporter family alignment is shown. Legend as described for Additional data file 4 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 \(A) Hydropathy plots of two representatives of the novel putative transporter family: the PFI0720w and PFC0530w proteins. The two profiles are very similar and in each, there are 12 clear peaks in hydrophobicity, corresponding to 12 TMDs. A topology of 12 TMDs separated, by an extended extramembrane loop, into two sets of 6 closely spaced TMDs is characteristic of transporters of the MFS. (B) The alignment of the five P. falciparumnovel putative transporters. The region over TMDs 2-3 and TMDs 8-9 of the alignment is shown. Members of the MFS family typically possess a conserved amino acid motif between TMDs 2 and 3 and a related but less conserved motif in the corresponding loop in the second half of the protein (between TMDs 8 and 9). The putative novel transport proteins also appear to contain these MFS-specific motifs between TMDs 2 and 3 and, to a lesser extent, between TMDs 8 and 9. For comparison, MFS-specific motifs from a range of known and putative MFS proteins are presented. Legend as described for Additional data file 4 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 The region over TMDs 1-5, TMD 7 and TMD 10 of the alignment is shown. The sequences are separated into two clusters, one containing plant and insect proteins and the other protozoan, yeast and mammalian proteins. The PFL0420w and PFL1515c proteins appear to be more closely related to the former group of transporters, whereas the MAL6P1.133 is more similar to the latter. Legend as described for Additional data file 4 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 The region over TMDs 1-3 and TMDs 6-10 of the alignment is shown. The sequences are separated into two clusters, one containing proteins known to transport neurotransmitters, creatine or taurine, and the other proteins known or hypothesized to transport amino acids. The P. falciparumproteins have diverged considerably from the other family members, but are perhaps most similar in sequence to the amino acid transporters. Legend as described for Additional data file 4 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 9 The region over TMDs 2-6 of the alignment is shown. The locations of the \'voltage sensor\' and \'selectivity sequence\' are indicated. Legend as described for Additional data file 4 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 10 Additional methods ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Richard Allen and Stefan Bröer for helpful discussions and the Canberra Branch of the Australian Red Cross Blood Service for the provision of blood. This work was supported by the Australian Research Council (DP0344425) and by the Australian National Health and Medical Research Council (Grant ID 179804). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Comparison of the hydropathy plots of three P. falciparumproteins designated as putative transport proteins (PF14\_0541, PFI0955w and PF13\_0172) with that of a protein designated as having no putative function (PF14\_0435). The PF14\_0541 protein is a putative V-type H^+^-pumping pyrophosphatase (H^+^-PPase) and its hydropathy plot shows around 15 peaks in the hydrophobicity index, corresponding to 15 predicted transmembrane domains (TMDs) - as is characteristic of H^+^-PPases. The PFI0955w protein is a putative sugar transporter of the major facilitator superfamily (MFS) and its hydropathy plot indicates the presence of 12 TMDs. The PF13\_0172 protein bears no sequence similarities with any known or putative transport proteins but its hydropathy plot shows around 11 peaks in the hydrophobicity index and resembles that of a typical transporter (for example, PF14\_0541 or PFI0955w). The PF14\_0435 protein has no non-Plasmodiumsequence homologs or similarities to conserved domains, and although it is predicted to possess eight or nine putative TMDs, the hydropathy plot of the PF14\_0435 protein does not resemble that of a typical transporter. The predicted TMDs are irregularly spaced (those in typical transporters tend to show more regularity of spacing, as in the first three hydropathy plots shown) and there are several very large extramembrane domains interspersed among the TMDs (many transporters have a single large extramembrane domain in the middle of the protein, but it is unusual for there to be multiple, irregularly spaced extramembrane domains of the type evident in PF14\_0435). The possibility that the PF14\_0435 protein (and others like it) is a transporter can certainly not be excluded; however there is simply not sufficient evidence to warrant its classification as such in the present study. The hydropathy plots were generated using the TMpred server \[114\]. ::: ![](gb-2005-6-3-r26-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Graphical overview of the permeome of P. falciparum. (a)Transport proteins with seven or more transmembrane domains (TMDs). These proteins were retrieved by the analysis of the genome using a computer program that interrogates a genome database on the basis of the hydropathy plots of the corresponding proteins \[9\]. They include all the putative or known transport proteins with seven or more TMDs already identified in the genome, as well as 55 putative transport proteins with seven or more TMDs not previously recognized as such. (b)Transport proteins with six or fewer TMDs. These proteins were sourced in the most part from the annotated genome. Black bars, members of porter families (that is, uniporters, symporters and antiporters); dark-gray bars, members of primary active transporter families (that is, pumps); light-gray bars, members of channel families; white bars, putative transporters of unknown lineage and function. Abbreviations for the families are as follows: MFS, major facilitator superfamily; DMT, drug/metabolite transporter superfamily; ABC, ATP-binding cassette superfamily; P-ATPases, P-type ATPase superfamily; H^+^-PPases, H^+^-translocating pyrophosphatase family; MC, mitochondrial carrier family; CDF, cation diffusion facilitator family; F/V-ATPases, H^+^- or Na^+^-translocating F-type, V-type and A-type ATPase superfamily; ArsAB, arsenite-antimonite efflux family. ::: ![](gb-2005-6-3-r26-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### RNA obtained at different stages of P. falciparumdevelopment in the erythrocyte. (a)Representative Giemsa-stained P. falciparum-infected erythrocytes at the growth stages analyzed in this study. Samples from a tightly synchronized P. falciparumFAF6 culture were collected for the extraction of total RNA at ring (\~4, 8, 16 and 20 h post-invasion), trophozoite (24, 32 and 36 h post-invasion) and schizont stages (40 and 42 h post-invasion). The cells depicted show the morphology of the parasitized cells in the culture at the given time point. The amount of RNA yielded from parasite cultures at around 4 h post-invasion was too low to warrant the inclusion of this time point in the subsequent gene-expression studies. Cells in the top row of boxes are from the first time course; cells in the bottom row are from the repeat time course (performed approximately 4 months later). (b)The quantity of total RNA inside the parasitized cell increases dramatically over the intraerythrocytic cycle. Total RNA was extracted from tightly synchronized P. falciparumFAF6 culture samples collected at nine stages (see above) over a single 48-h growth cycle of the intraerythrocytic parasite. The data are averaged from two different time courses performed approximately 4 months apart and are shown ± range/2. ::: ![](gb-2005-6-3-r26-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Stage-dependent gene expression of transporters throughout the intraerythrocytic cycle of P. falciparum. (a)A putative transporter of the MFS family; (b)the three P. falciparummembers of the sugar porter family (a subfamily of the MFS). The PFB0210c gene encodes the P. falciparumhexose transporter, PfHT1 \[46\]. RT-PCR was conducted to semi-quantify the level of gene expression in \~1.5 × 10^4^parasitized cells at each growth stage. Relative expression (y-axis) is the ratio of the density of the band from the PCR product at each time point in the life cycle relative to that at the time point giving the largest yield of PCR product. Ratios calculated from replicate gels from the same PCR were averaged before the data from the two time courses (carried out approximately 4 months apart and each consisting of ≥ 2 PCRs) were combined to give the mean ± S.E. For comparison, the relative amount of total RNA in the parasitized cell over the same growth stages is also presented (dotted line). ::: ![](gb-2005-6-3-r26-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Stage-dependent gene expression of four putative members of the drug:H^+^antiporters-1 family (a subfamily of the MFS), throughout the intraerythrocytic cycle of P. falciparum. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Stage-dependent gene expression of the two putative members of the monocarboxylate porter and oxalate:formate antiporter families (two closely related subfamilies of the MFS), throughout the intraerythrocytic cycle of P. falciparum. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Stage-dependent gene expression of MFS-related transporters, throughout the intraerythrocytic cycle of P. falciparum. (a)A member of the glycoside-pentoside-hexuronide:cation symporter family; (b)the three P. falciparummembers of the folate-biopterin transporter family. Both transporter families are distantly related to the MFS. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Stage-dependent gene expression of the five members of the novel putative transporter family, throughout the intraerythrocytic cycle of P. falciparum. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Stage-dependent gene expression of putative amino-acid transporters throughout the intraerythrocytic cycle of P. falciparum. (a)Amino acid/auxin permeases; (b)neurotransmitter:Na^+^symporters. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-9) ::: ::: {#F10 .fig} Figure 10 ::: {.caption} ###### Stage-dependent gene expression of the four P. falciparummembers of the equilibrative nucleoside transporter family, throughout the intraerythrocytic cycle of the parasite. The PF13\_0252 gene encodes the P. falciparumnucleoside transporter, PfENT1 \[62,63\]. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-10) ::: ::: {#F11 .fig} Figure 11 ::: {.caption} ###### Stage-dependent gene expression of the putative inorganic anion exchanger (PF14\_0679) of the sulphate permease family and the Pi :Na^+^symporter (MAL13P1.206) of the inorganic phosphate transporter family, throughout the intraerythrocytic cycle of P. falciparum. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-11) ::: ::: {#F12 .fig} Figure 12 ::: {.caption} ###### The alignment over the region of the selectivity sequence of the putative P. falciparumnovel ion channel protein (PF14\_0342) with a representative selection of K^+^channel proteins (known and putative) of the voltage-gated ion channel superfamily. P. falciparumsequences are boxed and the protein designators highlighted. The P. yoeliihomolog of the PF14\_0342 protein is encoded by the chryPy1\_00168 locus, which is available at PlasmoDB. For proteins of other organisms, the NCBI accession number and the known or putative (p) function of the protein are given. A larger alignment of the voltage-gated ion channel superfamily, encompassing transmembrane domains (TMDs) 2-6, is presented in Additional data file 9. Residues are shaded as follows: blue, positively charged,; red, negatively charged; orange, hydoxyl; gray, amido; green, proline; purple, cysteine; mid-blue, histidine,; light blue, glycine; olive green, tryptophan and tyrosine; yellow, remaining nonpolar. ::: ![](gb-2005-6-3-r26-12) ::: ::: {#F13 .fig} Figure 13 ::: {.caption} ###### Stage-dependent gene expression of two P. falciparumputative K^+^channels (PFL1315w and PF14\_0622) and the putative novel ion channel (PF14\_0342) of the voltage-gated ion channel superfamily, throughout the intraerythrocytic cycle of P. falciparum. The analysis was carried out as described in the legend to Figure 4. ::: ![](gb-2005-6-3-r26-13) :::	PubMed Central	2024-06-05T03:55:56.323884	2005-3-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088945/", "journal": "Genome Biol. 2005 Mar 2; 6(3):R26", "authors": [ { "first": "Rowena E", "last": "Martin" }, { "first": "Roselani I", "last": "Henry" }, { "first": "Janice L", "last": "Abbey" }, { "first": "John D", "last": "Clements" }, { "first": "Kiaran", "last": "Kirk" } ] }
PMC1088946	Background ========== The steady-state abundance of an RNA species in an organ is, in part, genetically controlled and can be considered a quantitative genetic trait. Microarray methods for estimating RNA sequence abundance \[[@B1]\], combined with genetic methods for identifying loci affecting quantitative traits \[[@B2]-[@B4]\], provide the opportunity to survey tissues for all genetically controlled variation in gene expression. This approach has been called genetical genomics \[[@B5]\], and its feasibility has been demonstrated in experimental crosses and human populations \[[@B6]-[@B10]\]. Genetical genomics is further enhanced by using recombinant inbred lines as a mapping population. The use of recombinant inbred lines allows comparison of gene expression among different tissues and the comparison of gene expression with classical physiological and behavioral traits from the published literature \[[@B11],[@B12]\]. Public datasets and online software at WebQTL \[[@B10],[@B13]\] allow free exploration of the characteristics of this form of analysis \[[@B14]\]. In addition, recombinant inbred lines can provide both replicates from genetically identical individuals and samples from different segregants. Data from these define genetic and non-genetic variation, define a measure of heritability for expression of individual genes, and provide the basis for a new method of data reduction for genetical genomics. Data reduction is an issue because Affymetrix GeneChip oligonucleotide microarrays assay each target mRNA with a set of 11 to 16 pairs of 25-nucleotide DNA probes. Each pair of probes consists of a perfect match (PM) sequence and a mismatch (MM) sequence, the latter intended to estimate nonspecific binding. The Affymetrix software Microarray Suite 4.0 and 5.0 (MAS 4 and MAS 5) estimate expression from the average difference of PM and MM fluorescence. Since the pioneering study of Li and Wong \[[@B15]\], however, it has been clear that MM binding includes target-specific binding as well as nonspecific binding, and the appropriate use of MM fluorescence has been an open question. In fact, a recent publication shows that it may be more useful to use the sum of PM and MM values instead of their difference \[[@B16]\]. In short, the behavior of oligonucleotide microarrays is not adequately explained by models that only consider base complementarity. More realistic models consider nonspecific binding, saturation, the effects of fluorescent labeling and intramolecular folding of target and probe \[[@B15],[@B17]-[@B19]\]. Several alternative methods have been proposed to combine multiple probe-specific values into a single expression estimate. Three widely used alternatives are robust multiarray average (RMA) \[[@B20]\], model-based expression index/intensity (MBEI), implemented in dChip software \[[@B15]\], and positional-dependent nearest-neighbor model (PDNN) \[[@B17]\]. RMA provides statistically robust averaging methods, dChip fits a model that allows probe-specific binding affinities, and PDNN fits a model that allows sequence-specific binding affinities and nearest-neighbor stacking interactions. A weighted-average method is also available, one which weights probe-specific values by a cross-validation procedure \[[@B21]\]; this method, however, does not take advantage of replicate microarrays and the current implementation in Bioconductor \[[@B22]\] is too slow for this application. Finally, a method (SUM) based on the sum of PM and MM values has recently been described \[[@B16]\]. The rationale for this method is that MM probes exhibit probe-specific binding as well as nonspecific binding \[[@B15],[@B17]\] and may therefore be more effective for estimating specific binding than for correcting for nonspecific binding. Indeed, the SUM method outperforms MAS5 in several respects. We describe here a new method, specifically designed for application to genetical genomics. In this method, called heritability-weighted transform version 1 (HWT1), probe-specific data is combined in a weighted average in which the weights are determined by an estimate of the heritability of the data for each probe. Results ======= Figure [1](#F1){ref-type="fig"} provides an overview of the dataset and the data reduction problem for QTL mapping with gene-expression data from recombinant inbred strains. These gene-expression data form a four-dimensional dataset. As shown in Figure [1](#F1){ref-type="fig"}, the first dimension is formed by recombinant inbred strains; the second by replicate samples from each strain; the third by multiple probes of each probe set; and the fourth by multiple probe sets representing different transcripts. For QTL mapping, dimensions 2 and 3 must be collapsed to single values that can be compared with genotypes for each strain (in dimension 1). Normally, dimensions 2 and 3 are collapsed by simple averaging or by averaging probe differences. Heritability is determined by the relative expression variance contributed by dimensions 1 and 2. The HWT1 method described here uses this information from dimensions 1 and 2 to define weights that allow dimension 3 to be collapsed with a weighted average. Dimension 2 is still collapsed with a simple average. The left-hand panels of Figure [2](#F2){ref-type="fig"} show the distribution of estimated heritability of expression for individual PM probes, with frequencies shown on a log scale to make the tails of the distribution visible. Results from three organs or tissues from BXD recombinant inbred lines are shown: brain expression (Brn); hematopoietic stem cell expression (HSC); and cerebellum expression (Cer). Brain and HSC were assayed with Affymetrix U74Av2 microarrays; cerebellum with Affymetrix M430A and B. In all datasets, estimates range from well below 0 to 1 or slightly above. The method used for estimating heritability is known to yield estimates outside the natural range expected for heritability \[[@B23]\]. Indeed, as shown in Figure [2](#F2){ref-type="fig"}, 21%, 45% and 60% of estimates are negative (for brain, HSC and cerebellum, respectively) and a few (\< 0.1%) of brain and cerebellum estimates are above 1.0. Although estimation methods exist that would avoid these values, the current method is simple and serves the intended purpose if negative heritabilities and those above 1 are adjusted by assigning them values of 0 and 1, respectively. When these adjusted heritabilities are normalized by the average (adjusted) heritability of probes in each probe set, the resulting weights are distributed as shown in the right-hand panels of Figure [2](#F2){ref-type="fig"}. About 36%, 49%, and 61% (for brain, HSC, and cerebellum, respectively) of probe weights are zero and 55%, 60%, and 66% are less than 1.0. These probes are fully or partly excluded from any weighted average. A small minority of probes, less than 3%, receive weights above half the maximum possible weight, suggesting that they will dominate the average for the probe set to which they belong. The results of QTL mapping with weighted averages are shown in Figure [3](#F3){ref-type="fig"}, in which sorted P-values from a set of microarrays is plotted against the rank of each P-value \[[@B24]\]. Each P-value represents the significance of the best single QTL, that is, of the best association between expression of one transcript and genotypes at some marker. In this plot, uniformly distributed P-values, from tests in which the null hypothesis is always true, form a straight line along the diagonal. That is, a complete absence of QTLs would yield a straight diagonal line. In each panel, an inset shows the entire range of P-values, most of which do approximately form a diagonal. The main figure shows the smallest values only. In each main figure the line formed by the P-values bends sharply, indicating a local excess of small P-values. Those P-values which fall below the dotted line in each panel form a group in which the false-discovery rate is expected to be no greater than 20%, according to a Benjamini and Hochberg test \[[@B25]\]. This criterion is used throughout this paper to define significant QTLs. The panels of Figure [3](#F3){ref-type="fig"} compare QTLs detected after averaging with Affymetrix MAS 5.0 software and QTLs detected with three variations of heritability-weighted averaging. These variations differ in their use of MM probes. As transcript binding to MM probes seems to include both nonspecific and target-specific binding \[[@B15],[@B17],[@B18]\], we tested both subtracting MM values from PM (to remove nonspecific signal) and adding MM values to PM (to add target-specific signal). Figure [3a](#F3){ref-type="fig"} shows results obtained by calculating heritability from PM probes and averaging only those probes, Figure [3b](#F3){ref-type="fig"} shows results obtained by calculating heritability from and averaging PM - MM differences and Figure [3c](#F3){ref-type="fig"} shows results obtained by calculating heritability from and averaging all probes (PM and MM) together. Using 20% false-discovery rate as a significance cutoff, each of the heritability-weighting methods yields more QTLs than MAS 5.0. With this dataset, using only PM probes yielded more QTLs than the other two weighting methods. When weighted expression averages were randomly permuted among the recombinant inbred (RI) strains before mapping, no QTLs were detected at 20% false-discovery rate (data not shown). Since heritability estimates are unaffected by permutation, permuting data after weighted averaging is equivalent to permuting before averaging. Furthermore, simulation showed that heritable variation alone is not sufficient to define QTLs. Simulated datasets were generated with heritable variation distributed among probes in various ways, including one in which all heritable variation was generated for a single probe of each probe set. In these simulated datasets all variation was independent of marker genotypes. No QTLs were detected from these simulated datasets after heritability-weighting and QTL mapping (data not shown). There is little relationship between the abundance of transcripts and the likelihood of detecting a QTL (data not shown). If anything, strong QTLs tend to be found among transcripts of moderate abundance. This tendency might be explained if apparent interstrain variation, necessary for QTL detection, is reduced when abundance is extreme, either near the lower limit of detection or high enough to saturate some oligonucleotide probes. Probe heritability is a predictor of the existence of a detectable QTL for a probe set. Either average heritability or maximum heritability among probes in a probe set can be used as a predictor. In either case, heritability above a threshold value is taken to predict the existence of a QTL. Figure [4](#F4){ref-type="fig"} shows the receiver operating characteristic (ROC) curves for average or maximum probe heritability used as a predictor of the existence of a significant QTL. The ordinate shows the fraction of transcripts with QTLs that are correctly predicted as such by heritability; the abscissa shows the fraction of transcripts without QTLs that are incorrectly predicted by heritability to have a QTL. The curves are produced by plotting these two quantities for various threshold values for average heritability or maximum heritability. For a perfect predictor, the ROC curve would follow the left and top boundaries of the figure. For a useless predictor, the ROC curve would be a diagonal line between the origin and the upper-right corner. These curves show that maximum heritability is more effective than average in predicting a detectable QTL. Because probe sets that do not define a significant QTL greatly outnumber those that do, probe sets defining a QTL are still a minority among probe sets selected for heritability. This situation is illustrated by three points that are circled in the figure. The right-hand circled point shows that selecting for maximum heritability greater than 0.35 selected 77% of probe sets; 2% of these yielded QTLs composing 99% of all QTLs. The center circled point shows that a threshold of 0.525 selected 17% of probe sets, of which 8% yielded QTLs composing 90% of QTLs. The left-hand circled point shows that a threshold of 0.675 selected 4% of probe sets, of which 32% yielded QTLs composing 75% of QTLs. The availability of RNA from unrelated tissues, brain and HSC, allowed us to consider the question of whether probe heritabilities are specific to the tissue of origin. Raw probe heritabilities for data from brain and HSC have a correlation coefficient of -0.004, but that value means little because most probe heritabilities are close to zero. A more meaningful comparison is between probe heritabilities for probe sets in which at least one probe has significant heritability. Figure [5](#F5){ref-type="fig"} shows scatterplots comparing brain and HSC raw probe heritability and probe weight for 304 PM probes (19 probe sets) in which at least one probe from each organ had heritability greater than 0.90. Even with this degree of selection, the correlation for heritability or weight is only 0.59 or 0.58, respectively. Thus, even with extreme selection, there is little correlation between probe heritabilities from these two sources, suggesting the probe heritabilities are tissue specific. QTLs for gene expression can be classified according to the chromosomal location of the QTL relative to the location of the gene being expressed. Those for which the location of the QTL and gene are tightly linked are characterized as cisQTLs; those for which the locations are different are trans. In this study the location of a QTL is defined by the location of the marker achieving the highest likelihood ratio statistic (LRS), a marker defined by a simple-sequence repeat whose location is known in the mouse sequence. CisQTLs are, somewhat arbitrarily, defined as those for which this marker is within 10 megabases (Mb) of the location of the probe sequence by which the gene expression is measured. QTLs can also be classified according to the direction of the effect on gene expression. We adopt the convention that QTLs are labeled \'+\' if the DBA/2J allele is associated with higher apparent expression and \'-\' if the C57BL/6J allele is associated higher apparent expression. Assuming that Affymetrix probe sequences were largely designed for the C57BL/6 sequence, sequence differences between C57BL/6 and DBA/2 in the sequence recognized by a probe will tend to make DBA/2 hybridize more poorly than C57BL/6. That is, variation in sequences complementary to probe sequences can create artifactual QTLs, reflecting a difference in hybridization rather than a difference in expression. Such artifactual QTLs would be expected to be cis-. Figure [6](#F6){ref-type="fig"} summarizes classification of QTLs detected by heritability-weighting methods. The three panels of the figure show data from brain, HSC and cerebellum. Each dataset confirms previous results that each of the heritability-weighted methods detects more QTLs than MAS 5.0. However, the HSC dataset differs from the other two in that weighted PM - MM differences detected more QTLs than PM probes alone. For all methods in all datasets, cis- QTLs outnumber cis+ QTLs, in some cases by two- or threefold. This excess could be explained by polymorphisms in sequences targeted by Affymetrix probes, polymorphisms reducing the hybridization of DBA/2J RNA. For Brn and HSC the weighting procedure made some attempt to reduce this type of artifact by assigning a weight of 0 to 614 probes having known single-nucleotide polymorphisms (SNPs) in the probe target sequence. The excess of cis- QTLs remaining in Brn and HSC in spite of this procedure suggests that there may be additional effects from polymorphisms not included in our list. The cerebellum dataset yielded a large number of significant QTLs. In part this yield was expected because the number of probe sets for M430 microarrays is 3.6-fold larger than for U74Av2. However, the QTL yield for the cerebellum data is about 10-fold higher than for brain or HSC, or about 2.7-fold higher relative to the number of genes represented on the microarrays. As discussed further below, the cerebellum data were obtained in two unbalanced batches, and a difference between these batches might create artifactual QTLs on chromosome 2. However, although 475 significant QTLs, 16% of the total, appear on chromosome 2, this number is too small to fully explain the large number of the cerebellum QTLs. Figure [7](#F7){ref-type="fig"} shows that the HWT1 method using only PM probes allowed the detection of more QTLs than the dChip, RMA, or PDNN data reduction methods. Compared with these methods, HWT1 detected larger numbers of QTLs in all QTL classes, but the increase in cis- QTLs was disproportionately large. As explained, many of those cis- QTLs could be artifacts caused by polymorphisms. The number of probes that contribute to weighted averages varies considerably between probe sets. The effective number of probes can be defined, as described in Materials and methods, by a measure which is the reciprocal of a weighted average of the weights. The measure varies from 1.0, if all weights but one are zero, to the number of probes (usually 11.0 or 16.0), if all probes are weighted equally. Figure [8](#F8){ref-type="fig"} shows, in boxplot form, the distribution of effective probe number for weighted averages of brain data. Five classes of probe sets are compared, those that do not define QTLs and those that define cis-, cis+, trans-, and trans+ QTLs. In each plot, the central box shows the range between the 25th and 75th percentiles. The line across the box gives the median location, and the shaded area gives the 95% confidence interval for the median. The data in Figure [8](#F8){ref-type="fig"} allow three conclusions. First, a substantial fraction of probes contribute to weighted averages that define QTLs. In each case, the central half of QTLs falls into the 7- to 13-probe interval. Although the groups do not differ significantly, there is a possible tendency for + QTLs to involve more probes than - QTLs. Finally, only the cis- group includes QTLs defined by fewer than four probes. QTLs that depend on so few probes are most likely to be artifactual QTLs caused by polymorphisms in the probe target sequences. Discussion ========== The heritability-weighted averaging method described here successfully summarizes oligonucleotide microarray measurements of gene expression in a way that facilitates detection of QTLs affecting that expression. It is a heuristic method, one that is not derived from an explicit statistical model. Nevertheless, the rationale is simple and rests on three facts: first, heritable variation is necessary (but not sufficient) to define a QTL; second, probes within a probe set differ greatly in the heritability of their expression estimates; and third, probes within a probe set differ greatly in their ability to detect a QTL. These facts suggested that a simple weighted average would summarize probe set data without obscuring the signal of those probes which could detect a QTL. HWT1 is designed specifically for QTL mapping. In its present form, it does not apply to the more common experimental situation designed to estimate expression differences between samples. In that experimental situation, this method would be circular, weighting probes according to an estimate of the quantity to be estimated. QTL mapping, in contrast, does not depend directly on the differences between samples, but on the correlation of those differences with a genetic marker. Indeed, the data of Figure [4](#F4){ref-type="fig"} imply the existence of a few probes with high heritability that nevertheless yield no significant QTL. Although we designed this weighting to reflect heritability, it may, depending on the experimental design, involve more than heritability. The heritability estimate is based on the variance between strains (which includes genetically determined variance) and the variance within strains, as an estimate of non-genetic variance. This estimate is closely related to other size-of-effect measures, such as repeatability, ω^2^, η^2^, or ε^2^\[[@B26]-[@B29]\]. Although we have not tested weighting with these alternative measures, we expect any of them would provide a similar benefit for QTL mapping. However, the optimum weighting for this application is not yet determined. The frequencies of cisQTLs detected in this study (31-77%) fall within the wide range of frequencies detected in other studies. The most closely comparable study is that of mouse liver transcription, in which the frequency of cisQTLs varied from 34% for moderately significant QTLs (log odds score (LOD) \> 4.3) to 71% for more significant QTLs (LOD \> 7.1) \[[@B8]\]. However those results were based on microarrays of 60-nucleotide probes, which would be expected to be less sensitive than Affymetrix probes to the effects of single-nucleotide polymorphisms. The same study reported a frequency of 80% for the more significant QTLs (LOD \> 7.0) for maize leaves. For yeast transcription assayed with cDNA arrays, Brem and co-workers estimated 36% cisQTLs \[[@B7]\], and for a human cell line assayed with Affymetrix arrays Morley and co-workers reported 18% \[[@B9]\]. Variance within strains usually includes non-genetic biological variation, but that was not true for the HSC dataset, for which replicates were derived from a single biological sample. In that dataset, heritability estimates were presumably higher than if replicates had been derived from separate biological samples. Nevertheless, HWT1 weighting was clearly useful for detecting QTLs in this set. Systematic differences among strains can affect weighting in either of two ways. Batch effects that are balanced within strains (partly true in the cerebellum data) will contribute to the within-strain variance and will deflate heritability estimates. This effect may explain why cerebellum raw probe weights include many more negative values than do brain or HSC (Figure [2](#F2){ref-type="fig"}). On the other hand, systematic non-genetic differences between strains (such as the batch effect in HSC data) will inflate heritability estimates. For heritability estimates, the HSC batch effect was avoided by using data from one batch. Such batch effects may also affect QTL mapping, causing a higher frequency of false positives in areas of the genome where a batch effect fortuitously correlates with marker alleles. In fact, if the batch number in cerebellum is treated as a trait, it associates with three areas on chromosome 2 (none of which, however, reaches a suggestive level of significance). These effects could be controlled by using batch as a cofactor, both in the analysis of variance that estimates heritability and in the subsequent QTL mapping. However, these refinements go beyond what is needed to introduce the HWT1 method. Thus, in the cerebellum dataset, QTLs mapping to chromosome 2 may include false positives caused by a difference in microarray processing batch. This batch effect, however, cannot explain the exceptional number of QTLs detected in the cerebellum dataset. The excess number of QTLs detected for cerebellum (compared with brain or HSC) greatly exceeds the total number of QTLs on chromosome 2. The comparison of heritability-weighting with other data reduction methods (Figure [7](#F7){ref-type="fig"}) should be considered as preliminary because they are based on results from only one set of data. More important, that comparison does not imply anything about their suitability for other purposes. In addition, modifications of any of those methods might make them more suitable for QTL mapping. It is not clear why probes of a single probe set should vary so greatly in the heritability of their expression estimates. We suggest three possibilities. First, changes in RNA concentration will result in greatest changes in fluorescence if RNA concentrations are close to the effective binding constant for a probe. Since effective binding constants of probes vary \[[@B17]-[@B19]\], sensitivity to changes will vary. Second, nonspecific hybridization of probes with RNA species that do not vary among strains will reduce specific hybridization that might define a QTL. If probes differ in nonspecific hybridization, they will differ in their ability to define a QTL. Third, since probes assay different parts of the target transcript, alternative splicing and differential degradation will affect probes differently. The QTLs described in this report were detected by fitting a single-QTL model, a statistical model assuming that all QTLs contribute to a trait with independent effects. This model can be misleading if linked and/or interacting QTLs contribute to a trait. Nevertheless, since many traits are largely controlled by one QTL or few unlinked QTLs, these results are reliable and useful. They further suggest that it may be fruitful to adapt the principle of heritability-weighting to QTL searches with multi-QTL models. Conclusion ========== To summarize expression data for individual transcripts, the HWT1 method combines probe-specific data in a weighted average in which weights are determined by the heritability of the probe-specific data. It provides a useful way to summarize datasets for genetical genomics because it places weight on probe-specific data having variation that could define a quantitative trait locus. Materials and methods ===================== Brain RNA --------- Brain RNA was obtained from 32 strains of BXD recombinant inbred mice, the parental strains C57BL/6J and DBA/2J, and (C57BL/6 × DBA/2)F1 hybrid. Data from parental and F1 animals were included in the heritability estimates but were not used for QTL mapping. Each individual array experiment used a pool of brain tissue (forebrain plus the midbrain, but without the olfactory bulb) that was taken from three adult animals usually of the same age. More detailed information is available at WebQTL \[[@B10]\]. All results derive from the 100-array December 2003 data freeze. Hematopoietic stem cell (HSC) RNA --------------------------------- Bone marrow cells were stained with lineage-specific antibodies and purified by flow cytometry. A stem-cell population was defined as the 5% cells showing least lineage-specific fluorescence \[[@B30]\]. Replicate samples of RNA were separately amplified from a single cell preparation for each BXD strain, and these samples were processed in two batches of 22 and eight strains. These data are described at WebQTL \[[@B10]\] as the March 2004 data freeze. Cerebellum RNA -------------- Each individual microarray assay used Affymetrix MOE 430A and MOE430B GeneChip pairs to assay RNA from a pool of intact whole cerebella taken from three adult animals usually of the same age. RNA samples were processed in two large batches. The first batch consisted of single samples from 17 BXD strains. The second batch consisted of biological replicates for 10 strains, additional technical replicates for two strains, single samples for four additional strains, and duplicate samples for five additional strains. RNA was extracted at the University of Tennessee Health Science Center and all samples were processed at the Hartwell Center (St. Jude Children\'s Research Hospital, Memphis). These data are described at WebQTL \[[@B10]\] as the SJUT Cerebellum January 2004 data freeze. Microarrays ----------- Brain and HSC data were obtained from Affymetrix U74Av2 microarrays, which provide more than 12,000 probe sets, almost all of which are represented by 16 PM probes and 16 MM probes. The cerebellum data were obtained from Affymetrix 430A and 430B microarrays, which provide more than 45,000 probe sets, almost all of which are represented by 11 PM probes and 11 MM probes. Microarray data reduction ------------------------- In addition to the HWT1 method, microarray data were processed with Microarray Suite 5.0 (MAS5) software \[[@B31],[@B32]\], RMA \[[@B20]\], PDNN \[[@B17]\] and dChip \[[@B15]\]. HWT1 weighting -------------- Individual probe intensities from Affymetrix U74Av2 microarrays were log~2~-transformed and normalized to a standard array-wide mean and standard deviation. For each probe, mean squared deviations within strains (MS~w~) and between strains (MS~b~) were calculated by analysis of variance of the log-transformed, normalized expression. In the interests of speed, age and sex of animals were not included as cofactors in the analysis of variance. Raw heritability was estimated as (MS~b~- MS~w~)/(nMS~t~), where nis the average number of replicates per strain and MS~t~is total variance (excluding strains without replicates, if any) \[[@B33]\]. Adjusted heritability was derived from raw heritability by assigning values of 0 and 1, respectively, to raw heritability values below 0.0 or above 1.0. Weights for each probe were calculated by dividing the adjusted heritability by the mean adjusted heritability for all probes in the probe set. Finally, expression estimates for each probe set and strain were calculated by an unweighted average of replicates within each strain and a weighted average of those probe-specific means, using the weights just described. To avoid division by zero, and to avoid using weights based on very small heritabilities, probes in a probe set were assigned a weight of 1.0 if the average adjusted heritability of those probes was less than 0.01. That is, expression for those probe sets was calculated from an unweighted average. The number of probe sets affected by this treatment was 5 (0.04%), 33 (.26%) and 4,178 (9.3%), respectively, for the Brn, HSC and Cer datasets. The large number of affected probe sets for cerebellum is consistent with the high number of negative raw heritability estimates for this dataset. As explained under Results, polymorphisms between C57BL/6J and DBA/2J in probe target sequences would be expected to affect hybridization of Affymetrix probes, generating an apparent QTL mapping to the location of the transcript. To reduce the effect of this type of artifact, we prepared, from sequence information for the two strains, a list of 614 probes having polymorphisms in target sequences of probes on the U74Av2 microarray. During the weighting procedure described above, these probes were assigned a weight of 0, removing their contribution from any QTL for their probe set. This procedure was not applied to the cerebellum data, which came from a different microarray. Among the HSC data, a systematic difference between the first and second batches described above would have greatly inflated all heritability estimates. To avoid this problem, heritability estimates were based on the first batch only, but all data were weighted and used for QTL mapping. Among cerebellum data, weighting was necessarily based only on replicated samples, most of which consisted of one sample from each batch. Any systematic batch difference would decrease heritability estimates. As with HSC data, cerebellum data from all strains was included in QTL mapping, weighted according to heritability estimates based on the strains with replicated samples. QTL mapping ----------- Heritability-weighted averages were evaluated by regression against marker genotypes, where alleles at markers were coded as -1 or 1. In the interest of speed, regression was performed only at marker locations, but the limitations of this restriction were minimized by using 779 markers (described as the BXD genotype set at WebQTL \[[@B10]\]). Although WebQTL includes values for parental lines and F1 related to the BXD RI lines, these were not used in QTL mapping \[[@B26]\]. For each microarray trait value, the locus yielding the maximum LRS \[[@B3]\] and the LRS itself were retained. An empirical P-value was then calculated for this LRS by a permutation test \[[@B34]\]. Microarray trait values were permuted randomly among the progeny individuals 1,000 times and the regression analysis is repeated for each permuted dataset. If the original LRS fell within the distribution so that at least 10 values from permuted sets were greater, a P-value was calculated from the rank of the original LRS in the distribution. If a P-value could not be calculated, additional permutations are performed, until a P-value could be calculated or until 1,000,000 permutations had been performed. For each microarray trait, four data values were retained, the locus yielding the highest LRS, the LRS and regression coefficient at that locus, and the P-value of the LRS. To evaluate significance, all results from one microarray experiment were sorted by P-value, and the significance of the smallest P-values was determined by the method of Benjamini and Hochberg \[[@B25]\], using a false-discovery rate of 20%. Mapping was performed with custom software, called QTL Reaper, written in Python and C for Linux. This software will be described fully in a subsequent publication but is currently available from SourceForge \[[@B35]\]. Calculations were performed on an eight-node Linux cluster, which achieved processing rates of about 5,000 genome scans per cpu-second. Most processing time was spent on the small fraction of probe sets requiring more than 10^5^permutations. Effective number of probes -------------------------- Within a probe set, the weight of each probe may vary from 0 to the number of probes in the set, n. The effective number of probes fin a weighted average is defined as ![](gb-2005-6-3-r27-i1.gif) where w~i~is the weight of probe i. This index varies from 1 to n. It is equal to kif kof the probes are weighted equally, and it is less than kif kof the probes are weighted unequally (with zero weight for the n- kremaining probes). Data availability ----------------- The HSC dataset has been placed in GEO. The accession number is GSE2031, and the arrays are GSM36673 to GSM36716. The Brn and Cer datasets are now both accessible from WebQTL \[[@B13]\]. Acknowledgements ================ We gratefully acknowledge the support of The National Institute on Alcohol Abuse and Alcoholism, INIA grants U01AA13499, U24AA13513, and the Human Brain Project P20-MH 62009, funded jointly by the NIMH, NIDA and NSF. Data were generated with funds to R.W.W. from the Dunavant Chair of Excellence, University of Tennessee Health Science Center, Department of Pediatrics. We thank the joint St. Jude Children\'s Research Hospital-UTHSC Cerebellum Consortium and The Hartwell Center for generating the cerebellum (Cer) dataset. We thank Bing Zhang, Cheng Li and Li Zhang, respectively, for performing the RMA, dChip and PDNN transformations for the brain (Brn) dataset. We thank two anonymous reviewers for specific, constructive comments. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The four-dimensional nature of microarray data used for QTL mapping. Recombinant inbred lines (strains) comprise dimension 1; the replicate arrays for each strain, dimension 2. Multiple probes for each probe set comprise dimension 3, and multiple probe sets (transcripts), dimension 4. Green rectangles represent the multiple probe- and replicate-specific expression values that must be collapsed to a single value for QTL mapping. That mapping correlates expression values with genotypes in dimension 1. Heritability-weighted averaging uses information in dimensions 1 and 2 to collapse dimension 3 by weighted averaging. Dimension 2 is collapsed by unweighted averaging. ::: ![](gb-2005-6-3-r27-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distribution of heritability of probe intensities and of probe-specific weights derived from heritability. Frequencies are shown on a log scale to make the tails of the distributions visible. Expression is in BXD RI lines from the tissue indicated; HSC, hematopoietic stem cells. The left-hand panels show the distribution of raw heritability estimates for individual Affymetrix probes. The right-hand panels show the distribution of probe-specific weights derived from those heritability estimates. ::: ![](gb-2005-6-3-r27-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Distribution of P-values from QTL mapping of brain RNA expression. The P-value for the best QTL for each microarray probe set is plotted against the rank of that P-value among all probe sets \[24\]. The four figures show four methods of pre-processing the data before QTL mapping. In each panel the smaller inset shows the entire range of P-values and the larger figure shows the smallest 200 to 300 P-values. The dashed line in the inset shows the expected distribution for random P-values; in the larger figure the dashed line shows the limit for 20% false-discovery rate, according to Benjamini and Hochberg \[25\]. (a)HWT1 weighting of PM values only; (b)HWT1 weighting of PM-MM differences; (c)HWT1 weighting of PM and MM values combined; (d)PM-MM differences averaged by Affymetrix MAS 5.0 software. ::: ![](gb-2005-6-3-r27-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Probe heritability as a predictor of a detectable QTL for a probe set. The figure shows receiver operating characteristic (ROC) curves for prediction of existence of a detectable QTL by either average heritability or maximum heritability among probe-specific data in a probe set. The true-positive fraction on the ordinate is the fraction of probe sets with a significant QTL that are identified as such by selection at a given maximum heritability. The false-positive fraction is the fraction of probe sets without a significant QTL that are selected as having a QTL at the same maximum heritability. Triangle symbols show ROC curve for average heritability; circle symbols show ROC curve for maximum heritability. Circled points are explained in the text ::: ![](gb-2005-6-3-r27-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Poor correlation of heritability and heritability-derived weights. The figures compare the raw heritability (h^2^) and weights of probes from probe sets in which at least one probe had a raw heritability greater than 0.90 in both brain and HSC data. (a)Raw HSC heritability for each probe vs raw brain heritability; (b)probe weight for HSC data vs weight for brain data. ::: ![](gb-2005-6-3-r27-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Number and types of QTLs in the three tissues defined by four methods of data summary. PM, HWT1 weighting of PM values only; Diff, HWT1 weighting of PM-MM differences; All, HWT1 weighting of PM and MM values combined; MAS 5, PM-MM differences averaged by Affymetrix MAS 5.0 software. cis, QTL location within 10 Mb of transcript location; trans, QTL location further than 10 Mb from transcript location; -, B57BL/6 allele associated with higher expression signal; +, DBA/2 allele associated with higher expression signal. ::: ![](gb-2005-6-3-r27-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Number and types of QTLs defined in the brain dataset according to the method of data summary. Methods used were: HWT1, heritability-weighting using only PM probe data; dChipPMMM, dChip method using PM and MM probe data \[15\]; dChipPM, dChip method using only PM data; RMA, robust multiarray averaging using only PM probe data \[20\]; PDNN, PDNN method using PM and MM probe data \[17\]. See legend to Figure 6 for definitions of QTL types. ::: ![](gb-2005-6-3-r27-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Distribution of effective number of probes in heritability-weighted averages. Boxplots show the distribution for probe sets that do not define significant QTLs (QTLs at 20% false-discovery rate) and for those that define QTLs of different types. In each plot, the central box shows the range between the 25th and 75th percentiles. The line across the box gives the median location, and the shaded area gives the 95% confidence interval for the median. Lines above and below the box give the range for all data except outliers, which are plotted singly beyond the range defined by the terminal crossbars. transQTLs are those for which the QTL is more than 10 Mb distant from the location of the transcript whose expression defines it. + QTLs are those for which the DBA2/J allele is associated with higher expression. ::: ![](gb-2005-6-3-r27-8) :::	PubMed Central	2024-06-05T03:55:56.332477	2005-2-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088946/", "journal": "Genome Biol. 2005 Feb 28; 6(3):R27", "authors": [ { "first": "Kenneth F", "last": "Manly" }, { "first": "Jintao", "last": "Wang" }, { "first": "Robert W", "last": "Williams" } ] }
PMC1088947	Background ========== The standard genetic code uses 64 codons for only 22 amino acids, including the amino acids selenocysteine and pyrrolysine whose incorporation into protein requires the reassignment of the UGA and UAG stop codons, respectively \[[@B1],[@B2]\]. This degeneracy of the genetic code has important implications for gene primary structure evolution as it provides nature with a vast array of options for building open reading frame (ORF) sequences for any particular protein. However, the usage of synonymous codons for building ORFs is not random, suggesting the existence of mechanistic or evolutionary constraints that limit the degree of freedom for coding sequence building \[[@B3]-[@B6]\]. In other words, each organism uses a set of rules for building ORF sequences which restrict the total number of options provided by the degeneracy of the genetic code. These rules are only partly understood. Nevertheless, it is becoming increasingly clear that codon usage and context bias reflect the action of two main evolutionary forces: selection for mRNA decoding efficiency and mutational drift acting indiscriminately on coding and noncoding DNA \[[@B7]-[@B10]\]. Codon usage reflects selection for translational efficiency, as highly expressed genes tend to use codons that are decoded by abundant cognate tRNAs \[[@B11]-[@B13]\]. Similarly, the context of a sequential pair of codons (codon-pair) is biased, but this bias is apparently linked more to decoding accuracy than to translational speed \[[@B14]-[@B17]\]. This suggests that the translational machinery is sensitive to the nature of the codon-pair present in the ribosome A and P decoding sites \[[@B16],[@B18]-[@B20]\], raising the possibility that, like codon usage, codon context may also be species specific. This is supported by the fact that tRNA populations diverge in the number and abundance of tRNA isoacceptors for each codon family and also in the pattern of modified nucleosides in the tRNAs, which also affects mRNA decoding accuracy. To shed new light on the overall pattern of codon context at the species level and evaluate how codon-pair context varies between species, we have developed software and statistical methodologies for codon-pair context analysis on all the ORFs in a genome as a whole (the ORFeome). Because our main interest is to evaluate the effect of codon context on mRNA decoding accuracy, this study focuses on the context of codon-pairs and not on long-range context effects. With a few exceptions, long-range context is not relevant for mRNA decoding by the ribosome. These new methodologies were tested using the complete ORFeome sequences of the eukaryotes Saccharomyces cerevisiae, Candida albicansand Schizosaccharomyces pombeand the bacterium Escherichia coli. The methodology developed provides robust and flexible tools for intra- and inter-ORFeome comparative codon-pair context analysis, permits identification of species-specific codon context fingerprints and provides new insight into the role of codon context on mRNA decoding accuracy and ultimately on the pressure imposed by the translational machinery on the evolution of the ORFeome. The software developed, called Anaconda, is available at \[[@B21]\]. Results ======= Global analysis of codon context in yeast ----------------------------------------- The Anaconda bioinformatics system developed in this study identifies the start codon of an ORF and reads it by moving a \'decoding window\' three nucleotides at a time in the 3\' direction until it encounters a stop codon. While doing so it fixes the middle codon of the reading window and memorizes its 5\' and 3\' neighbors. Anaconda creates a table of frequencies of 64 × 64 codons that allows computation of the number of times the complete set of contiguous codon pairs occurs in an ORF or in an ORFeome. The overall architecture of Anaconda is described in Figure [1](#F1){ref-type="fig"}. The codon-pair context frequency table built by Anaconda allows the statistical analysis of contingency tables to be used to test whether the context is significantly biased \[[@B22]-[@B25]\]. These tables allow one to test the existence of association between codon-pairs through the chi-square (χ^2^) test of independence; to identify preferred and rejected pairs of codons in the ORFeome through the analysis of adjusted residuals for contingency tables (Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"}); and to construct a codon context map on an ORFeome scale (Figure [3](#F3){ref-type="fig"}). The Anaconda algorithm, its graphical interface and implemented statistical methodologies were tested using the yeast S. cerevisiaeORFeome. For this, the complete ORFeome was downloaded from the yeast genome database \[[@B26]\], the adjusted residual values for the total number of codon pairs were calculated (see Materials and methods) and each residual value present in a cell of the contingency table (64 lines × 64 columns) was converted into a two-color coded map (Figure [3](#F3){ref-type="fig"}). In the latter, green represents positive values greater than +3 (herein called preferred codon-pairs) and red represents negative values lower than -3 (herein called rejected codon-pairs) according to the color scale indicated in Figure [3a](#F3){ref-type="fig"}. The data clearly show that each codon has a set of preferred 3\' codon neighbors (green) and rejects a set of other codons (red), indicating that codon context is highly biased in S. cerevisiae. However, in a rather large number of cases, the 3\' codon context is not biased or at least strongly rejected or preferred. This is indicated by the black color in the map (Figure [3](#F3){ref-type="fig"}) and in the histogram of the residuals distribution (Figure [4](#F4){ref-type="fig"}). This black color corresponds to residual values that fall within the interval of -3 to +3 and correspond to codon contexts that do not contribute to the bias for a confidence level of 99.73% (Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"}). The overall empirical distribution of residual values for codon context in the yeast ORFeome (Figure [4](#F4){ref-type="fig"}) clearly shows that a large fraction (about 47%) of codon-pair contexts fall within the interval of -3 to +3, indicating that in many cases the context may not be under high selective constraint. Codon clustering unveils unique features of codon context --------------------------------------------------------- The codon-pair context maps shown in Figure [3a,b](#F3){ref-type="fig"} were built using a manually predefined distribution of codons in both lines and columns. To better understand the full extent of the codon-pair context bias in yeast, the data were clustered using the Pearson\'s correlation coefficient \[[@B27]\], which enables grouping of codons with similar context preferences. Using double clustering (that is, clustering both lines and columns) several distinct groups of red and green codon-pair contexts were identified for the S. cerevisiaeORFeome, thus showing that certain groups of codons have similar 3\'-neighbor preferences (Figure [5](#F5){ref-type="fig"}). To identify the codons responsible for defining the subgroups with high bias (red and green clusters) and evaluate whether these could define codon-pair context rules, one zooms in on the context subclusters. Three specific subclusters (one red and two green) were analyzed in this study (Figure [6a-c](#F6){ref-type="fig"}). The red subcluster shown in Figure [6a](#F6){ref-type="fig"} is defined by codon-pairs in which the last nucleotide of the first codon is uridine (U) and the first nucleotide of the next codon (3\' side) is adenosine (A). As no such rule was observed for the other codon positions - that is, positions 1 and 2 or 2 and 3 of codon 1 or positions 1 and 2 or 2 and 3 of codon 2 (data not shown), the codons are clustered based on the following context rejection rule: XXU-AYY. The intensity of rejection (given by the adjusted residual itself) is not identical for all codon combinations within the subcluster. However, with the exception of the asparagine AAU and serine AGU codons, and some others whose residual values fall within the non-statistically significant -3 to +3 interval, all other U-ending codons avoid 3\'-neighbor codons starting with an A. If one assumes that fixed codons in the map (lines) represent P-site codons and 3\' codons (columns) represent A-site codons, then the above rule suggests that the third base of a P-site codon somehow influences the choice of the first base of the A-site codon. In other words, and assuming that context modulates decoding accuracy, S. cerevisiaecodon pairs that end with an U and start with an A are likely to cause some trouble to the ribosome during decoding. The above observations were confirmed by analyzing two green codon-pair context subclusters (good contexts). In these cases, two different clustering rules were identified, namely the XXC-AYY and the XXU-GYY (Figure [6b,c](#F6){ref-type="fig"}). Like the bad context subcluster discussed previously, in these good context subclusters there are exceptions that include red and black context cells. Nevertheless, there is a strong trend for the above rule within each subcluster, indicating once more that the third base of the P-site codon influences the first base of the A-site codon. The fact that these rules cannot be seen for other codon positions, and that there are exceptions to these rules for other codon families in the overall map, excludes the possibility that the third-first base rules identified reflect dinucleotide preferences or rejections arising from DNA replication and repair (\[[@B28]\] and see later). Comparative codon context analysis ---------------------------------- Because the S. cerevisiaecodon-pair context map produced a clear context pattern, we wondered whether this map could represent a species-specific fingerprint, as is the case for the codon-usage fingerprint. For this, maps for S. pombe, C. albicansand E. coliwere also constructed, with the latter being used as an outgroup. Some similarities between the codon-pair context maps were immediately visible, namely a strong green diagonal line in the yeast maps (Figure [7](#F7){ref-type="fig"}). There are, however, important differences that become evident when the negative and positive residual values are ranked for the yeast species studied (Table [2](#T2){ref-type="table"}). These values represent the most negative and positive residuals of the yeast maps and consequently provide a good indication of the differences in codon context present in the three yeast species. Of the 10 most positive residual values ranked in Table [2](#T2){ref-type="table"}, only two are common for the three yeast species, namely GAA-GAA, GGU-GGU and GCU-GCU. A similar result was obtained when the most negative values were ranked (Table [2](#T2){ref-type="table"}). In addition, the C. albicansgenome shows a more biased codon-pair context status. For example, the 10th most positive residual (49,476 for ACA-ACA) is higher than the maximum residual value for S. cerevisiaeand S. pombe: 45,422 for CAG-CAG and 35,086 for UCU-UCU, respectively (Table [2](#T2){ref-type="table"}). An additional approach to identifying codon-pair context differences between S. cerevisiae, S. pombeand C. albicans, was undertaken by overlapping the complete codon context maps displayed in Figure [7](#F7){ref-type="fig"}. For this, the maps built with a predefined order of codons for both the 64 lines and the 64 columns were merged, allowing the construction of a comparison codon-pair context map. We call this a differential codon-pair context map (DCM) and it corresponds to the module of the difference between the residuals of overlapped cells of the 64 × 64 context table (Figure [8](#F8){ref-type="fig"}). A new color scale based on gradation of blue was used for the differential display. Using this methodology, the codon context differences for the three yeast species became self-evident, indicating that codon context - like codon usage - is species specific (Figure [8](#F8){ref-type="fig"}). In all three DCMs shown in Figure [8](#F8){ref-type="fig"} there are common features, which are indicated by the black cells; however, the differences (blue) are clearly visible. As expected from the phylogenetic distance of the various species studied, the DCMs for the pairs E. coli-S. cerevisiaeand E. coli-C. albicansshow many more differences than the DCM for the pair S. cerevisiae-C. albicans. The DCMs also show that codon-pair context is more similar for the pair S. pombe-S. cerevisiae(data not shown) than for the other two yeast pairs, indicating that there are fewer differences between S. pombeand S. cerevisiaethan between C. albicansand S. cerevisiae. This is surprising, considering that S. pombediverged from S. cerevisiae420 million years ago whereas C. albicansdiverged from the latter only 170 million years ago \[[@B29]\]. The effect of the rather strong green diagonal (codon repeats) in the C. albicansmaps is also visible in the DCMs (blue cells) of the C. albicans-S. cerevisiaepairs (Figure [8a](#F8){ref-type="fig"}). In order to shed more light on the differences in the codon context maps for the three yeasts, codon pairs were ranked according to the module of the difference between residuals (Table [3](#T3){ref-type="table"}). Surprisingly, only one codon pair for the three yeast species (CAA-CAA) is present among the 10 highest values that were ranked. Further, the difference between these three species is not only qualitative, as shown above, but is also quantitative. For example, for the S. pombe-S. cerevisiaepair, the highest difference was found for the pair CAG-CAG with a value of 27,798, whereas in the S. pombe-C. albicansmap the CAA-CAA pair showed a difference value of 100,639. In fact, in the latter yeast pair DCM all 10 values related are higher than the highest value (27,798) found for the CAG-CAG codon pair in the S. pombe-S. cerevisiaemap (Table [3](#T3){ref-type="table"}). Therefore, when taken together, DCMs and residuals rankings provide unique insight into the codon-pair context differences, even for phylogenetically related species such as yeasts. Contribution of mutation bias to codon-pair context --------------------------------------------------- An important feature of the codon-pair context map in the yeasts analyzed, but not in E. coli, is the presence of a diagonal green line (Figures [3](#F3){ref-type="fig"}, [7](#F7){ref-type="fig"}). The existence of this green line implies that in those yeasts, most codons prefer to have another identical codon on their 3\' side, indicating a degree of tandem codon duplication in the ORFeome of yeasts. Trinucleotide repeats are characteristic of eukaryotic genomes and have been attributed to DNA polymerase slippage during genome replication \[[@B30]\]. Whether the codon duplication observed in the ORFeome of the yeasts analyzed is a consequence of DNA replication only, or also reflects an evolutionary constraint imposed by the mRNA decoding machinery on those ORFeomes, is not yet clear and we are currently investigating this. In any case, this diagonal line in the codon context maps of yeasts is a strong feature, since the highest residuals of codon pairs (preferred pairs) occur for tandem codon repeats (Table [2](#T2){ref-type="table"}). The above observations prompted us to investigate whether mutational bias also played a part in codon-pair context bias and whether such bias could be extracted from the codon-pair context maps. For this, particular attention was given to GC content because it plays a major role in codon usage \[[@B31]\]. An algorithm was implemented into Anaconda for calculating %GC total, %GC at codon position 1 (GC1), %GC at codon position 2 (GC2) and %GC at codon position 3 (GC3). While scanning an ORFeome, Anaconda divides ORFs into GC-content subgroups and creates groups of ORFs with high and low GC content. It also determines the distribution of ORFs according to their GC total and GC3 (Figure [9a,c](#F9){ref-type="fig"}). Codon-pair codon context maps can be built for each subgroup of codons and the maps compared using the DCM tool (Figures [9b,d](#F9){ref-type="fig"} and [10](#F10){ref-type="fig"}). Because GC bias is better observed at the third codon position as a result of the degeneracy of the genetic code, GC3 was used to evaluate whether mutational bias contributed to the codon-pair context using the S. cerevisiaeand E. coliORFeomes as proof of principle. In the former, the ORF distribution varied from a minimum of 11.9% to a maximum of 76.7%; however, most ORFs fell within a narrow interval between 35-40% GC3 (Figure [9a](#F9){ref-type="fig"}). In the case of E. coli, the ORF distribution is broader, varying from a minimum of 20.0% to a maximum of 89.4%, but most ORFs have a GC3 between 50% and 60% (Figure [9c](#F9){ref-type="fig"}). This distribution made it possible to build codon-pair context maps for the low GC3 and high GC3 subgroups. As differences between these low and high GC3 context maps were expected to allow for evaluation of the importance of the bias introduced by mutational drift into the codon-pair context maps, these maps were overlapped using the DCM tool. As before, the maps were built using a single colour (blue) to aid visualization of the context differences. If mutational drift did not contribute to the context bias, the codon-pair context maps of the GC3 subgroups would be identical, producing a black differential display map. This is because the difference of the module of the residuals would be zero for all cells of the table of residuals. The differential display map for the low and high GC3 ORF subgroups of S. cerevisiaeshowed several differences, indicating that GC bias contributes to the codon-pair context. However, most of these differences corresponded to small deviations in the strength of the rejection or preference of the codon-pair contexts (Figure [9b](#F9){ref-type="fig"} and [10](#F10){ref-type="fig"}, see also Table [4](#T4){ref-type="table"}). In other words, the residual values had the same positive or negative signal in both cases but the value was higher in one GC3 subgroup than the other and vice versa. In some cases, an inversion of signal of the residuals (for example, from positive to negative) was detected, indicating that the residual of the codon-pair was positive in one GC3 subgroup and negative in the other GC3 subgroup (light blue in Figure [9b](#F9){ref-type="fig"}). This inversion of signal provides clear evidence for the influence of GC content bias in the codon-pair context. Similar results were obtained for the E. coliORFeome; however, a much larger number of inversions of the residual signal was observed in this case, indicating that the GC content bias is far stronger in E. colithan in S. cerevisiae(Figures [9d](#F9){ref-type="fig"} and [10](#F10){ref-type="fig"}, see also Table [4](#T4){ref-type="table"}). The reasons for these differences and the quantitative contribution of mutational bias to codon-pair context bias is not yet fully understood and is currently being investigated. However, Anaconda already provides strong evidence for a role for mutational bias on codon-pair context. Discussion ========== Codon context has been extensively studied in prokaryotic, eukaryotic, mitochondrial and viral genomes, and these studies unequivocally showed that codon-pair context is biased \[[@B9],[@B10],[@B32]-[@B35]\]. However, no tool has yet been developed to display codon context data and in particular codon-pair context (short-range context) in a way that would facilitate interpretation of the data and allow inter- or intra-genome context comparisons. This is essential if putative general rules that govern codon-pair context evolution are to be unraveled. The Anaconda bioinformation system has been developed to address this problem. By using statistical methodologies based on contingency tables and residual analysis (see Materials and methods), specific codon-pair context patterns were unveiled and displayed using a color coded ORFeome-context map. The data highlighted codon-pair context bias in yeasts and E. coliand some rules that define codon-pair context patterns in yeast. Forces that shape codon-pair context ------------------------------------ Studies carried out in the 1980 s in E. colihave demonstrated that codon-pair context influences mRNA decoding accuracy and efficiency, indicating that the translational machinery imposes significant constraints on codon-pair context \[[@B17],[@B36],[@B37]\]. For example, in starved E. colicells, the asparagine AAU and AAC codons are misread as lysine at high frequency \[[@B16]\]. Quantification of the level of lysine misincorporation at those codons and determination of the effect of the 3\' nucleotide context on lysine misincorporation showed that the AAU codon is misread up to nine times more frequently than the AAC codon, and that the 3\' nucleotide context (III-I context) influenced the level of misreading by as much as twofold \[[@B16]\]. Additional studies carried out in vitroin E. coli, have also shown that ribosomes discriminate C-ending Phe UUC and Leu CUC codons less well than the U-ending Phe UUU and Leu CUU, showing that synonymous codons differ in translational accuracy \[[@B38]\]. Therefore, a possible role for codon-pair context is minimization of decoding error, in particular in those codons that are poorly discriminated by the ribosome. In E. coli, over-represented codon-pairs are translated more slowly than under-represented codon-pairs, indicating that codon-pair context also influences translational speed \[[@B14]\]. This suggests that codon-pair context in E. coliis under strong selective constraints imposed by the translational machinery. Whether the context patterns now unveiled in yeast reflect similar selective constraints remains unclear. Nevertheless, the codon-pair context maps described here provide a good starting point to address this important biological question in vivoin yeast in a guided manner. Additional evidence for a role for selection on codon-pair context was highlighted by the negligible, or even zero, contribution of GC3 to the context bias in very frequent or very infrequent codon-pairs (strong contexts) in both S. cerevisiaeand E. coli(Figure [9](#F9){ref-type="fig"}, Table [4](#T4){ref-type="table"}) and by a number of exceptions to the context rules that define the subclusters of codon-pairs (Figure [6](#F6){ref-type="fig"}). For example, within the XXU-AYY subcluster of rejected codons (Figure [6a](#F6){ref-type="fig"}), the codon pairs AAU-AGC, AAU-AGU, AAU-AAU, AAU-AAC and the set of AGU-AGC, AGU-AGU, AGU-AAU, AGU-ACA, AGU-AUA have positive residuals, indicating that they are codon pairs preferred by the ORFeome. Similar exceptions are found within the subclusters of preferred codon pairs shown (Figure [6b,c](#F6){ref-type="fig"}). Furthermore, a detailed analysis of the overall ORFeome context map (Figure [5](#F5){ref-type="fig"}) shows that other codon-pairs violate the XXU-AYY rules, namely GGU-AUG, GGU-AUC, GGU-AUU, GGU-ACC, GGU-ACU. This supports the hypothesis that those clusters of the context map are not formed on the basis of particular dinucleotide combinations that may be related to genome mutational drift. This is further confirmed by our observation that the dinucleotide preference in the XXU-AYY, XXC-AYY and XXU-GYY codon pairs is not observed when the various positions within each codon or codon-pair are analyzed. In other words, in the codon pair X~1~X~2~X~3~-Y~1~Y~2~Y~3~, the X~3~-Y~1~preferences are not observed for the dinucleotides X~1~-X~2~, X~2~-X~3~, Y~1~-Y~2~and Y~2~-Y~3~(data not shown). Despite these arguments, mutational bias does influence codon-pair context \[[@B7],[@B39]-[@B41]\]. Observed mutational bias reflects mutational events that act indiscriminately on all DNA sequences (coding and noncoding DNA) and is consequently a property of the genome rather than the result of selection acting within ORFs \[[@B42]-[@B45]\]. The data presented here is in line with those observations. For example, context maps shown in this study indicate that several of the context clusters are formed on the basis of dinucleotide context rules (III-I rule), namely the XXU-AYY, XXC-AYY, XXU-GYY (Figure [6a-c](#F6){ref-type="fig"}). As dinucleotide context is related to DNA repair and replication constraints those clusters reflect mutational bias \[[@B28]\]. An important feature that highlights the influence of mutational bias on codon-pair context is GC content, in particular GC3 content. GC content has a strong influence in codon usage and in extreme cases can even drive certain codons out of ORFeomes \[[@B46],[@B47]\]. The data presented here clearly show that GC3 affects codon-pair context; however, this effect is mainly visible for codon-pairs that have weak residuals (Table [4](#T4){ref-type="table"}, Figure [9](#F9){ref-type="fig"}). As strong residuals (either positive or negative) provide an indirect measure of the strength of the codon-pair association, it is likely that for extreme residuals GC3 bias introduces only noise into the analysis whereas for residuals near the statistically nonsignificant interval (-3, +3), GC3 bias represents a major contribution to the context bias observed (Figure [9](#F9){ref-type="fig"}). Apart from those cases mentioned above, other species-specific genomic features also contribute to codon-pair context bias highlighted by Anaconda. For example, the yeast codon-pair context maps show a feature of eukaryotic genomes which is not related to mRNA translation: trinucleotide repeats which are evident in the diagonal line present in Figures [3](#F3){ref-type="fig"} and [7](#F7){ref-type="fig"}. This strongly suggests that there is a very high degree of tandem codon repeats (trinucleotide repeats), which are likely to arise from biased DNA replication (DNA polymerase slippage, see \[[@B30]\]). Whether these repeated codon-pairs improve mRNA translation efficiency or accuracy in yeast remains to be determined experimentally. As far as we are aware, there is no experimental evidence showing increased decoding accuracy or efficiency at those sites. Finally, constraints imposed by protein sequences and mRNA secondary structure are also thought to influence codon context \[[@B48],[@B49]\]. The context maps seem to exclude the former hypothesis because no cluster is formed as a result of selection or rejection of two adjacent amino acids. In regard to the latter constraint, the Anaconda algorithm was not designed to detect mRNA secondary structures and consequently this question cannot be addressed at this stage. Conclusions =========== The Anaconda algorithm was developed with the aim of studying codon-pair context on an ORFeome scale, define rules that govern codon-pair context, carry out large-scale interspecies codon-pair context comparisons and clarify the effect of selection and mutational drift on codon-pair context. The results provide important new insight on the role of codon-pair context on mRNA decoding accuracy and efficiency, and we expect that it will allow the development of reporter genes for in vivoand in vitroquantification of codon-decoding error and translational speed. Finally, Anaconda will be a valuable tool to redesign ORFs for efficient and accurate heterologous or homologous protein expression in yeast and, eventually, in other suitable host systems. Materials and methods ===================== Statistics ---------- To study the association between contiguous codon-pairs, the coding sequences analyzed by Anaconda are processed in a 64 × 64 contingency table subdivided in mutually exclusive categories. If the 3\' context is being analyzed, the rows of the table correspond to the codons in the P-site and the columns to the codons in the A-site of the ribosome. At the 5\' context analysis the situation is inverted, and so the contingency table built is a transposed version of the one for 3\' analysis. A number of different mathematical methodologies have already been used to study codon context bias (for example \[[@B9],[@B50]-[@B52]\]). In this study, the analysis of contingency tables and residuals (Figure [3](#F3){ref-type="fig"}) was considered appropriate, assuming a multinomial probabilistic model for the contingency table (a detailed discussion of this model in the context of genomic data can be found in \[[@B53]\]). In general, all these methodologies are based on z-score-type tests and give information about preference and rejection. Basically, those methodologies differ in the probabilistic model assumed, leading to statistics whose probability distribution is in most cases unknown. The advantage of the methodology proposed here is that its theory of inference is well known, yielding an analysis that is more sequential, more easily interpretable and with more complementary tools for analysis (for example, measures of association). In other words, this methodology was chosen because the adjusted residual values give direct information about preference and rejection in relation to what would be expected on a random basis. Furthermore, the probability distribution under the hypothesis of independence is determined without data simulations. For analysis of contingency tables and residuals \[[@B22]-[@B25]\], given an r× ccontingency table where a multinomial distribution is assumed (Table [5](#T5){ref-type="table"}), the hypothesis of independence between the variables A and B is tested using the Pearson\'s statistic given by: ![](gb-2005-6-3-r28-i1.gif) where: ![](gb-2005-6-3-r28-i2.gif) It is known that Pearson\'s statistic has an asymptotical chi-square probability distribution with (r- 1)(c- 1) degrees of freedom. To identify cells in the table responsible for the eventual rejections of independence, the adjusted residuals d~ij~are calculated by: ![](gb-2005-6-3-r28-i3.gif) where: ![](gb-2005-6-3-r28-i4.gif) is the variance estimated for r~ij~. Haberman \[[@B54]\] has shown that, under independence between A and B, the adjusted residuals d~ij~have a standardized normal probability distribution, and therefore P(- 3 \<d~ij~\< 3) ≈ 0.9973, as N→ + ∞. This means that, for a 99,73% confidence level, the pair (A~i~, B~j~) is considered responsible for rejection of the hypothesis of independence if \\|d~ij~\\| ≥ 3. In practice, we consider that an adjusted residual is statistically significant if its absolute value is greater then 3. Additionally, to find codon context patterns in the contingency table, lines and columns can be grouped using classifying methodologies such as cluster analysis. These patterns are determined by calculating similarities between two vectors of the contingency table using the centred Pearson correlation coefficient and applying single linkage. The single-linkage method produces groups with \'chaining effect\': that is, any element of a group is more \'similar\' to an element of the same group than to any element of another group. Software -------- The architecture of the Anaconda software is based on three main modules, namely data acquisition, processing and visualization (Figure [1](#F1){ref-type="fig"}). Each module works independently from the others and can easily be replaced or updated. Also, this component-based approach allows for insertion of new modules or new tools in each module, such as new statistical features. The acquisition and processing modules download row data from genome databases, create a local database of usable ORFs and analyze the data using an algorithm that simulates the ribosome during mRNA decoding. It finally constructs a database containing the processed data. This data is then submitted to statistical analysis as described above. The visualization module allows the user to visualize the data matrices and gene sequences and to create filters that permit searching for specific sequence patterns defined by the user. The data-acquisition module deals with genome input files, namely reading and interpreting FASTA sequences of complete or partial sets of ORFs from public or private genome databases. To ensure that the screened sequences have the best possible quality, and hence do not introduce background noise in the following analyses, several quality filters are applied to the reading process. When the filters are activated the data are classified according to the following criteria. Valid data consist of genes whose sequence is a multiple of three; which start with an AUG codon and stop with a UAG, UAA or UGA codon, and which satisfy other user-defined requirements. Rejected data consist of genes whose sequence does not fulfill the above requirements. The result is the separation of valid from rejected ORFs. Other parameters needed by the application, such as reference relative synonymous codon usage (RSCU) values for codon adaptation index (CAI) calculation \[[@B55]\], are also uploaded by this module. The processing module is the core of the application, where the codon context analysis is performed. After prescanning the files, the user can test the existence of significant bias in the codon context and use the residual values to further explore the matrices of residual values (see Statistics, above). The data generated are then converted into a contingency table that includes the corresponding observed values of Pearson\'s statistics, and the matrix of adjusted residuals \[[@B25]\]. After processing, the data become available to the visualization module. This module is the graphical interface. It follows the file manager paradigm in which information is presented in hierarchical views. This module offers a set of tools that enable several tasks to be carried out, namely to search prespecified sequence patterns, to visualize data in histogram form, to cluster codon context data, and to export residual values. It is also possible to visualize other information at the gene level, such as rare codons and their distribution in the ORFs, to determine their ratio relative to the total number of codons, to determine the GC% at the first, second and third codon positions and determine the codon adaptation index (CAI) and the effective number of codons \[[@B55],[@B56]\]. Acknowledgements ================ We thank FCT (Project: POCTI/BME/39030/2001), IEETA and the II-UA (CTS-12) for supporting the development of the Anaconda software. G.M. is funded by FCT grant SFRH/BPD/7195/2001 and M.P. by INFOGENMED (FP-V). M.S. is supported by an EMBO YIP Award. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Architecture of the Anaconda bioinformation system. The Anaconda package contains a data-acquisition module that permits downloading raw data from genome databases and filter it into a local database. This data is then processed using a ribosome simulation algorithm and transferred to a 64 × 64 table that renders itself to statistical analysis. The processed data is then transferred to the visualization module that has a number of different tools that permit different types of data visualization and analysis. RSCU, relative synonymous codon usage values from very highly expressed genes, necessary for codon adaptation index (CAI) calculation (see \[55\]). ::: ![](gb-2005-6-3-r28-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Codon context is highly biased in yeast. The bar chart shows the distribution of the adjusted residual values given in Table 1 for the 3\' context of the S. cerevisiaeCUG codon. See Table 1 legend for details. ::: ![](gb-2005-6-3-r28-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### S. cerevisiaegenome map of codon context. For visualization purposes the values of the residuals of the 64 × 64 codon context table were converted into a color-coded map in which red represents the negative values (bad context) and green the positive values (good context). The values that are not statistically significant are indicated in black (-3 to +3). The color scale represents the full range of values of residuals for yeast codon context. Fixed codons represent the P-site codons and the 3\' context refers to the A-site codons as viewed by the ribosome simulation software module. (a)The yeast complete 3\' codon context map shows a diagonal green line, which indicates that most codons prefer themselves as neighbors on their 3\' side. The map also indicates that without exception, each codon prefers a defined set of neighbors (green) and avoids others (red). The intensity of red and green indicates the extent of the preference or rejection. (b)Codons that are represented in the map can be visualized by zooming into particular areas of the map (boxed in dark blue in (a)). The order of the fixed and 3\' context codons indicated in (b) is predefined in the software module. ::: ![](gb-2005-6-3-r28-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Distribution of the adjusted residuals from the S. cerevisiaecodon context map. Forty-three percent of the residuals fall within the nonsignificant -3 to +3 interval, indicating that a very large number of codon combinations are not significant to the rejection of independence - that is, are not significantly preferred or rejected in this genome. ::: ![](gb-2005-6-3-r28-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Codon context bias is organized in discrete groups. A two-way Pearson clustering by single linkage of the codon context data highlights regions of good and bad codon context, indicating that codon context bias is highly structured. A significant number of codons do not fall into the major clusters, indicating that their preferences and rejections are defined on a one-to-one basis. The 3\' codon contexts whose residual values fall within the nonstatistically significant -3 to +3 interval are also scattered in the map, indicating that there is no cluster of codons that have little or no preference for particular codons as 3\' neighbors. ::: ![](gb-2005-6-3-r28-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Codon clusters define specific codon-context rules in S. cerevisiae. (a)A major cluster of bad context is defined by codon pairs whose wobble base of the first codon is uridine (U) and the first base of the 3\' neighbor is adenosine (A). This cluster defines a XXU-AYY context rule, in which X and Y are any nucleotide. Within this cluster some of the Asn and Ser codons represent exceptions to the above rule as their residual signal is positive (green cells). (b,c)Two of the good context clusters define two distinct codon context rules, namely (b) XXC-AYY and (c) XXU-GYY rules. As before, some of the codons within those clusters are exceptions to the above rules and a number of codons have no particular preferences or rejections (black cells). ::: ![](gb-2005-6-3-r28-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Codon context maps are species specific. Comparison of the genomic codon context maps of S. cerevisiae, C. albicans, S. pombeand E. colishows that they are all different. There are common features between the maps but differences are clearly visible, indicating that each species has a specific set of codon context rules. Among the common features, the green diagonal line in the yeast maps is the most relevant. This diagonal indicates that almost all codons prefer themselves as their 3\' neighbors and is strongly marked in the C. albicanscontext map, suggesting that in this species, tandem codon repetition is very common. ::: ![](gb-2005-6-3-r28-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Differential display maps for comparative analysis of codon context. To compare the codon context maps of different species, the order of the codons displayed in the map was fixed and the maps overlapped using a differential display tool built into the Anaconda bioinformation system. Maps representing the context differences between (a)S. cerevisiaeand C. albicans, (b)E. coliand S. cerevisiaeand (c)C. albicansand S. cerevisiaewere obtained by calculating the module of the difference between the residuals of each map. The differences are represented in blue according to the color scale. The blue cells indicate the highest context difference and the black cells represent pairs of codons that have similar residual values between two species (module of the difference between residuals falls within the 0-15 interval). The maps show rather large differences in codon context between E. coliand S. cerevisiaeor C. albicansand smaller differences between S. cerevisiaeand C. albicans. ::: ![](gb-2005-6-3-r28-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### GC3 distribution in the complete ORFeome of S. cerevisiaeand E. coliand its influence on the overall codon-pair context analysis. In order to study the role of mutational bias upon codon-pair context the ORFeomes of both (a,b)S. cerevisiaeand (c,d)E. coliwere distributed according to the %GC3 of individual ORFs. The GC3 of the S. cerevisiaeand E. coliORFeomes varied between the intervals 11.9-76.7% and 20-89.4%, respectively. For S. cerevisiae, however, most ORFs had a %GC3 between 35 and 40% (light blue bar in (a)), while for E. colithe majority of the ORFs have a %GC3 between 50 and 60% (light blue bars in (c)). Determination of the codon-pair context for the low and high GC3 subgroups permitted identification of their context differences. The computation of the number of residuals that changed their signal (for example, positive to negative) from one subgroup (low GC3) into the other (high GC3) provided a quantitative measure of the role of GC3 on codon-pair context (red bars in (b) and (d)). For both S. cerevisiaeand E. coliGC3 bias has a strong effect on codon-pair context for weak residuals (-3 to +3), but no such effect was observed for contexts with the highest residuals (strong context), indicating that GC3 bias is mainly felt in weak codon-pair contexts. ::: ![](gb-2005-6-3-r28-9) ::: ::: {#F10 .fig} Figure 10 ::: {.caption} ###### ORFs with low and high GC3 have different codon-pair contexts. To highlight the effect of GC3 bias on codon-pair context, the context maps for the subgroups of low GC3 and high GC3 ORFs of both S. cerevisiaeand E. coliwere overlapped using the differential display codon-pair context (DCM) tool. The DCM maps for S. cerevisiaeand E. colishowed significant differences (light blue cells in the DCMs), in particular in E. coli, indicating that GC3 bias influences codon-pair context. ::: ![](gb-2005-6-3-r28-10) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The 3\' codon context of CUG ::: 3\' Codon Residual 3\' Codon Residual 3\' Codon Residual 3\' Codon Residual ----------- ---------- ----------- ---------- ----------- ---------- ----------- ---------- AAA 7.436 ACG 0.644 UCU -10.007 CCA -2.438 AAG 1.927 CGU -1.809 CUU 1.167 CCG 2.895 AAU 0.397 CGC 2.981 CUC 2.18 CAU 2.026 AAC 2.037 CGA 8.258 CUA 5.258 CAC 2.642 ACU -6.947 CGG 5.404 CUG 6.774 CAA 4.049 ACC -5.239 ACG -4.726 CCU -1.769 CAG 7.105 ACA -5.12 AGG -0.666 CCC 8.894 UAA 0.22 Positive values indicate that the 3\' codons appear in the genome more times than expected (good context) while negative values indicate that the 3\' codons appear fewer times than expected assuming a random distribution (bad context). Residual values give a quantitative indication of the context bias, where values falling within the -3 to +3 interval are not statistically significant (no bias). See also Figure 2. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Ranking of the 10 most negative and 10 most positive residual values in S. cerevisiae, S. pombeand C. albicanscontexts ::: S. cerevisiae S. pombe C. albicans ---------------------- ------------- --------------- ------------- --------------- ------------- Context Residual Context Residual Context Residual Most negative values UUU → AAG -24.58 GAA → CCU -24.159 UUU → CCA -32.691 GAU → AAG -22.487 GAU → AAG -24.124 UUC → GAA -31.586 AUU → AAA -21.546 UUU → AAG -23.899 UCA → GAU -28.317 AUU → AAG -21.285 AUU → AAA -22.923 AUU → AAG -28.284 CUU → AAA -20.656 UCU → AAG -22.334 GGU → UUU -27.198 UUU → AAA -20.563 CUU → AAA -21.25 AAC → UUA -26.198 UCC → GAA -20.069 GUU → AAA -21.218 GAC → UUA -25.795 AAG → UCU -19.706 AUU → AAG -21.08 UUU → AAG -25.316 GAU → CAA -19.274 UUU → AAA -20.704 GGA → AAA -25.26 GAA → CCA -19.155 GAA → UCU -20.698 UUC → GAU -24.822 Most positive values GAU → GAU 29.839 CAG → CAA 25.279 ACA → ACA 49.476 AAG → AAG 29.937 GAA → GAG 25.644 CAC → CAC 49.511 UUG → AAA 30.459 AAG → AAG 26.901 CCA → CCA 52.889 GAA → GAA 30.573 CUU → CGU 27.013 GAA → GAA 57.356 AAG → AAA 31.427 GAA → GAA 28.051 AAG → AAA 58.605 CAG → CAA 33.445 AGA → AGA 29.623 GCU → GCU 62.611 AGA → AGA 33.798 AAA → AAG 30.358 ACC → ACC 70.117 GGU → GGU 35.979 GCU → GCU 32.158 GGU → GGU 72.48 GCU → GCU 36.231 GGU → GGU 33.681 AAC → AAC 87.115 CAG → CAG 45.422 UCU → UCU 35.086 CAA → CAA 105.216 Anaconda was used to analyze the codon context of the complete genomes of S. cerevisiae, S. pombeand C. albicans. All possible codon contexts were ranked according to their calculated adjusted residuals, and the 10 most negative and 10 most positive were selected as extreme examples. The results indicate that only a small number of bad or good codon pairs (shown in bold) are shared between all three yeast species. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Ranking of the codon pairs that display the highest residual difference between S. cerevisiae, S. pombeand C. albicans ::: S. pombe-S. cerevisiae S. pombe-C. albicans C. albicans-S. cerevisiae -------------------------- ------------------------ ----------------------------- ------------- --------------- ------------ Context Difference Context Difference Context Difference CAG → CAG 27,798 CAA → CAA 100,639 CAA → CAA 79,38 UUG → AAA 25,266 AAC → AAC 76,716 AAC → AAC 62,939 CUU → CGU 25,168 ACC → ACC 60,208 ACC → ACC 50,735 CAA → CAG 24,507 CCA → CCA 47,603 CCA → CCA 39,196 AAA → AAG 23,593 ACA → ACA 47,359 CAC → CAC 39,032 UUC → AAA 22,86 CAC → CAC 47,175 ACA → ACA 39,029 AAU → AAU 22,021 GGA → AAA 45,043 GGU → GGU 36,501 CAA → CAA 21,259 AAG → AAA 43,994 GGA → UUA 35,81 GUU → CUU 21,194 CAA → CAG 43,927 GGA → AAA 29,786 GAU → GAC 19,483 UCA → UCA 41,533 GUU → GAU 29,753 Anaconda was used to analyze the codon context of the complete genomes of S. cerevisiae, S. pombeand C. albicans. The adjusted residuals of each codon context calculated for each pair of genomes - that is, S. pombe-S. cerevisiae; S. pombe-C. albicans; and C. albicans-S. cerevisiae- were subtracted and the result converted into a positive number by a module calculation. These values were used to rank the respective codon contexts and the 10 highest cases obtained were selected. Among these three yeast species, S. pombeand S. cerevisiaedisplay the lowest differences, with the maximum value of the difference being found for the CAG-CAG pair (27.798). For S. pombeand C. albicansthat value reaches 100.639 for the CAA-CAA codon pair. It is noteworthy that the highest difference value for the former pair is lower than the lowest value for the latter in this ranking of context differences. The only codon pair shared between all three yeast pairs is shown in bold. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### GC3 influences codon-pair context ::: Residuals ----------------- ------------- ------------ ----------- ---------- ------------ ORFeome \[- ∞, -9\] \[-9, -3\] \[-3, 3\] \[3, 9\] \[9, + ∞\] S. cerevisiae 0.0 2.5 94.2 3.3 0.0 E. coli 0.7 15.2 67.1 15.0 2.0 In order to measure the influence of GC bias on codon-pair context, the percentage of adjusted residuals that reversed their residual signals from positive to negative (or vice versa) between low and high GC3 subgroups of ORFs was determined. Most of the residual signal inversions for both species considered fall within the nonstatistically significant interval of the residuals (-3 to +3) indicating that GC3 bias is mainly felt in codon-pairs where the association is very weak or nonexistent (highlighted in bold). ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### A hypothetical r× ccontingency table ::: ---------------- -------- ------ -------- ------ -------- ---------------- B~1~ \... B~j~ \... B~c~ Marginal total A~1~ n~11~ \... n~1j~ \... n~1c~ n~1~\* \... \... \... A~l~ n~l1~ \... n~ij~ \... n~lc~ n~1~\* \... \... \... A~r~ n~r1~ \... n~rl~ \... n~rc~ n~r~\* Marginal total n\~1~ \... n\~i~ \... n\*~c~ N ---------------- -------- ------ -------- ------ -------- ---------------- The table illustrates how contingency tables were constructed and how the statistical methodologies described in methods were implemented. One set of categories is represented by rows, the other by columns. In the present case, if the 3\' context is being analyzed by Anaconda the rows of the table (A) correspond to the 5\' codons and the columns (B) to the 3\' codons of each pair. :::	PubMed Central	2024-06-05T03:55:56.336160	2005-2-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088947/", "journal": "Genome Biol. 2005 Feb 15; 6(3):R28", "authors": [ { "first": "Gabriela", "last": "Moura" }, { "first": "Miguel", "last": "Pinheiro" }, { "first": "Raquel", "last": "Silva" }, { "first": "Isabel", "last": "Miranda" }, { "first": "Vera", "last": "Afreixo" }, { "first": "Gaspar", "last": "Dias" }, { "first": "Adelaide", "last": "Freitas" }, { "first": "José L", "last": "Oliveira" }, { "first": "Manuel AS", "last": "Santos" } ] }
PMC1088948	Rationale ========= An important role for biological databases is the integration of different types of data. Ontologies aim to overcome the semantic differences encountered in data collection and representation, providing common terminology in order to facilitate this integration. An anatomy ontology is a structured vocabulary of anatomical entities in which the terms have unique identities and relate to each other in meaningful ways. For many biological applications, anatomy ontologies are essential for standardized description of data directly related to anatomy, such as gene expression patterns and phenotype information. The Gene Expression Database (GXD) is a resource for gene expression information from the mouse \[[@B1]\]. GXD has been designed as an open-ended system able to store and integrate primary data from many types of expression assays, each of which describe gene expression at different levels of spatial resolution. Currently, both GXD and the Edinburgh Mouse Atlas Gene Expression (EMAGE) database \[[@B2]\] use terms from the Mouse Embryo Anatomy Nomenclature Database \[[@B3]\] developed by the Edinburgh Mouse Atlas Project (EMAP) to describe patterns of gene expression in the developing mouse. However, since GXD also collects gene expression data from mice at postnatal stages, including adult, it became apparent that extension of GXD to fully annotate expression data for adult structures would require the development of a controlled vocabulary beyond the scope of the embryonic mouse anatomy ontology. Therefore, we developed an anatomy ontology for the postnatal mouse. Critical to this effort was the realization that existing sources of controlled vocabularies for anatomy were not sufficient for use with the adult mouse, for several reasons. First, none conforms well to the structure of the embryonic mouse anatomy ontology created by our Edinburgh collaborators, an important factor in enabling planned integration between these ontologies (see below). Human-oriented anatomical ontologies have been developed (for example, the Foundational Model of Anatomy (FMA) \[[@B4]\], OpenGalen \[[@B5]\] and SNOMED CT \[[@B6]\], which covers human and veterinary medicine). In general, the complexity of the concepts represented by these ontologies, issues concerning their accessibility, as well as questions of relevance to the mouse, made it clear that they were neither well suited to nor adequate for our objectives. Thus, one of our goals was to follow the basic framework of the developmental ontology, while taking full advantage of the range of other resources available. Another major consideration involved determination of the hierarchical structure and format of the ontology. Our experience using the developmental ontology made it clear that a mechanism to provide alternative hierarchies would be a critical factor. Consequently, the Adult Mouse Anatomical Dictionary is structured as a directed acyclic graph (DAG) in which an anatomical term can be represented as a child of more than one hierarchical parent term using both is-aand part-ofrelationships. The ontology is organized hierarchically in both spatial and functional ways, and contains more than 2,400 unique anatomical terms for the postnatal mouse. As GXD is part of the larger Mouse Genome Informatics (MGI) system, the ontology will also be used to annotate other types of data pertinent to adult mouse anatomy in order to provide an integrated description of a wide array of biological phenomena in the mouse. Developing an ontology for adult mouse anatomy ============================================== Anatomical terms ---------------- GXD has extensive experience with the Mouse Embryo Anatomy Nomenclature Database, available through Theiler Stage (TS) 26, which is used by GXD and EMAGE to describe developmental gene expression patterns. Based on our annotation work, we continue to contribute to this ontology in the form of extensions and revisions, and by adding synonyms. Consequently, an early objective was to ensure that the anatomy ontology for the postnatal (TS 28) mouse corresponds as much as possible, both in content and in structure, with the developmental ontology. This was done for consistency of nomenclature, because we were familiar with and confident of the utility of this format, and to facilitate the future integration of these ontologies. Eventually, the goal is to combine and integrate the ontologies to generate an anatomy ontology covering the entire lifespan of the laboratory mouse. With the developmental ontology as its framework, the effort was then focused on compiling an extensive list of anatomical terms for the postnatal mouse. The list was based on a number of major sources, including mouse atlases as well as anatomy and histology text resources \[[@B7]-[@B22]\]. For the most part, the preference was to focus on those that were mouse-specific. However, others that were more general were nevertheless extremely valuable. The non-atlas format references were especially useful in the effort to refine anatomic and histological details. Once the basic list of terms had been generated, we confirmed that each term on the initial list represented actual mouse structures. These determinations were usually clear but at times ambiguous. For example, for numerous structures described in anatomy and histology textbooks, no clear documented evidence was found for their existence in the mouse. Consequently, these have not been included in the ontology. Further work is ongoing to ensure accuracy. Careful attention was paid to validating each term, with the requirement for two or more reliable sources whenever possible. Concurrent with the textbook-based identification of terms was the continuing effort to expand the vocabulary using a research data-driven approach. This method included extensive evaluation of published biomedical research literature, as well as data with anatomical attributes that have been collected in scientific databases. For example, several mouse-specific datasets \[[@B23]-[@B26]\] were used as resources to find pertinent anatomical terms. The MGI list of all mouse tissues from which major publicly available cDNA libraries have been generated \[[@B24]\] includes cell types and tumors, as well as gross anatomical concepts. The relevant anatomical structures will eventually be translated using terms from the Adult Mouse Anatomical Dictionary. The data-driven approach was especially useful in determining the level of granularity (that is, level of detail of spatial resolution) expected to be required by users of the ontology. An additional consideration in determining the content of the vocabulary had to do with whether to include cell types. While cell type information is an important component in anatomical descriptions, this also introduces a level of complexity that is difficult to address adequately. We felt that it would be unfeasible to extend the representation to the cellular level owing to the large number of required hierarchical levels and leaf nodes. Therefore, it was concluded that the adult mouse anatomy ontology would not contain cell types, but that cell type terms would eventually be provided by the orthogonal controlled vocabulary for cell types currently being developed as part of the Open Biological Ontologies (OBO) effort \[[@B27]\]. However, to conform to the Edinburgh developmental ontology, we have included tissue type terms such as epitheliumand mesenchyme, as well as defined cell type structures such as purkinje cell layer. In addition, we have also elected to include the term unfertilized eggand its synonyms. Hierarchical organization ------------------------- The anatomy ontology for mouse development is currently structured as a straight hierarchy. In this format, an anatomical term can have only one parent and, thus, one place in the hierarchy. For example, the term femuris placed in the hierarchy according to this limb bone\'s spatial location, as a substructure of the upper leg, rather than as a part of the skeleton. In contrast, the brainis described as being part of the central nervous system, rather than as a part of the head. Based on our experience with the developmental ontology and anticipating planned revisions for it, we decided to represent the adult mouse anatomy ontology as a DAG, in which a given anatomical term is able to have more than one hierarchical parent. This allowed us flexibility in organizing the hierarchies, and provided a mechanism to create a more comprehensive view of the relationships between the anatomical terms. For each of the anatomical terms being evaluated, any one of a number of pathways to that term could be conceptualized. However, it also soon became apparent that two fundamental characteristics could be determined for most of the terms: its spatial location within the animal and its functional contribution as part of a particular organ system. Consequently, we decided to use the distinction between spatial versus organ system representation as an organizational principle. Since \'spatial part\' does not itself represent a unique anatomical entity, it was not included as an independent node in the ontology. However, the initial division of the hierarchy into spatial and organ system components is immediately apparent in the first level of substructures below the root node, TS28. As shown in Figure [1](#F1){ref-type="fig"}, this level is predominantly comprised of spatial parts: for example body, body cavity/lining, head/neck, limband tail. Accordingly, terms defined by these superstructures are primarily organized according to spatial localization. In contrast, another branch of the hierarchy is indicated by the superstructure organ system, where the anatomical terms are organized, as much as possible, according to their respective contribution to a specified functional system. Currently, the distinction between spatial and functional relationships is represented only implicitly. However, based on the parentage of anatomical structures, biologists will be able to intuitively discern both types of relationships. Furthermore, they should be able to perform most of the queries related to expression and phenotype data that are currently envisioned. Explicit representation of both relationship types might be a desirable feature for advanced knowledge representation and computational analysis. On the other hand, it might also introduce unnecessary complexities to a biologist because, for example, many anatomical structures would have both spatial and functional relationships between them. Shielding the user from those complexities would require additional software development. A careful evaluation of the advantages and disadvantages of both approaches will direct our future work in this area. During the construction of the adult mouse anatomy DAG, we had to take into account the fact that terms representing some tissues would logically be spatially located in numerous parts of the ontology. Groups of tissues which meet this criteria include: blood vessel, bone, connective tissue, muscle, nerve, organand skin, which are represented as terms in the organ systempart of the hierarchy. To accommodate the need to represent these tissues in specific body regions, we devised modules (outlined as blocks in Figure [1](#F1){ref-type="fig"}) representing these generic groups. These have been included as subterms, when appropriate, within each spatial region. For nomenclature standardization (more on this below), the subgroup terms are preceded by superstructure name, in noun form (that is, abdomen) rather than as an adjective (for example, abdominal) whenever possible. Consequently, using the DAG format, we have been able to describe adult mouse anatomy from a variety of spatial and organ system perspectives. For example, the heart(Figure [2](#F2){ref-type="fig"}) is represented as a type of thoracic cavity organ, as well as a substructure of the cardiovascular system. As will be discussed below, some of these distinctions are conceptual and by their nature may be somewhat arbitrary. However, from our annotation work we know that the different breakdowns of the anatomy are indeed required to annotate, for example, different types of expression and phenotype data. It should be emphasized that refinements to the hierarchical organization of the ontology will continue to be made. These changes will not affect the identity of the terms themselves. Another issue in constructing the DAG was the use of is-aand part-ofrelationships between the terms. Overall, most of the relationships could be classified intuitively as part-of, indicating that the term is a component of the more general term above it in the tree. For example, the upper bodyis considered to be part-ofthe body, and the heart is part-ofthe cardiovascular system. In contrast, is-arelationships are used to indicate that an anatomical term represents an instance of the certain type or kind of the concept denoted by its parent term. For instance, the cardiovascular system is-aspecific organ system, while cardiac muscle is-atype of muscle. It should be noted that there is no correlation between the is-aand part-ofrelationships and the spatial versus organ systemorganization of the ontology, as shown in Figure [2](#F2){ref-type="fig"}. Further refinement of the relationships will undoubtedly be required, as well as additional types of relationships. For example, it may be useful to distinguish between \'regional\' parts (for instance, head, neck, limb) versus \'systemic\' parts (for instance, body muscle, body organ, body skin). These modifications can be easily accomplished using the DAG-Edit tool (see Software section below). Nomenclature considerations --------------------------- Our experience with the mouse developmental ontology, as well as extensive literature review, provided the primary basis for the naming conventions that were employed. Early in building the ontology, we realized that consistent nomenclature, not only for a given term itself but for related terms and groups of terms, would be a critical requirement. Consequently, whenever possible, the same name was used for a given anatomical structure or concept throughout the ontology. For instance, we have used the term lungrather than \'pulmonary\' to precede each of the terms representing lung substructures. Another consideration regarded the need to clearly distinguish between terms. It is theoretically possible to precisely define an anatomical term based on a combination of the term name and the hierarchical lineage of the term. The term epithelium, for example, is represented as a subterm for many anatomical structures, and a given term\'s precise identity could be defined by its parental lineage. From a practical standpoint, this convention has proved to be problematic; multiple structures with the same term name would be impossible to distinguish in absence of its hierarchical context. This would be complicated further by any additional pathway to a given term. For instance, epitheliumof the lung alveoli is represented both as a part of the alveolusand as a type of lung epithelium. To address this issue, we have attempted to provide sufficient information in the term name (for example alveolus epithelium) so that it becomes easy to interpret and use the term unambiguously. Other factors that were considered were the requirements of the DAG-Edit software (see below), as well as features promoting unambiguous identification of terms. Additional conventions employed for the naming of anatomical terms included: structure names are preceded by superstructure names, in noun form; terms are used in singular form, whenever possible; all term names at the same level in the hierarchy are ordered alpha-numerically; and all characters are in lower case. Nomenclature consistency will also facilitate querying for specific anatomical terms within the ontology. Software issues --------------- An ontology should contain a level of detail appropriate to the data being classified and the level at which queries are likely to be performed, while simultaneously providing sufficient flexibility to enable regular updating without needing to significantly modify the hierarchies. Therefore, we recognized that the adult mouse anatomy ontology would require a format that was both robust and flexible, as well as the tools to accommodate the need for maintenance and updating. The DAG-Edit tool developed by the Gene Ontology (GO) Consortium provides a graphical interface to handle any vocabulary that has a DAG data structure, and has been used by other groups to build ontologies for a wide range of biological subjects, including the GO \[[@B28]\] and Mammalian Phenotype ontology \[[@B29]\]. We have utilized DAG-Edit both for construction of the adult mouse anatomy ontology and for maintenance and editing. Furthermore, the MGI software group has developed a range of tools to handle a DAG-formatted ontology, enabling navigation through the ontology and querying for terms (see below), as well as integration of the ontology with other information stored within the MGI database. Current status and future directions for the Adult Mouse Anatomical Dictionary ============================================================================== We have developed an ontology containing more than 2,400 unique terms to provide standardized nomenclature for anatomical structures in the postnatal mouse. The Adult Mouse Anatomical Dictionary can be accessed at the MGI web site \[[@B30]\]. The MGI Browser page (Figure [3](#F3){ref-type="fig"}) enables one to navigate through the ontology in two ways. Browsing results in the display of progressively lower levels in the hierarchy. Information about individual terms, including its relationship to other terms in the hierarchy, is shown in a \'Term Detail\' page. Alternatively, one can search the ontology by using the \'Query\' field, which accepts any text string and searches for all terms in the vocabulary, including any synonyms, containing that string. The resulting \'Query Results\' page displays all structures that match the query, and also provides links to the appropriate \'Term Detail\' page. The adult mouse anatomy ontology can also be viewed and downloaded at the OBO website \[[@B31]\]. The ontology can be saved in several different formats including GO flat file and OBO formats, as well as XML/RDF and OWL. We will continue to expand and refine the Adult Mouse Anatomical Dictionary in response to additional sources of information, as well as the needs of the scientific community. As part of the ontology\'s ongoing development, we plan to: expand the list of terms, based on additional resources as they become available; further edit the hierarchies when necessary; and provide alternative names for terms as synonyms. A limited number of synonyms have already been included (for example, see \'Term Detail\' page for limbin Figure [3](#F3){ref-type="fig"}). It is envisioned that many more will be added as required, which will also aid in querying for specific terms in the ontology. Precise definitions for each of the terms will also be included as appropriate. Eventually, the adult mouse anatomy ontology will be merged with the Anatomical Dictionary for Mouse Development to generate an anatomy ontology covering the entire lifespan of the laboratory mouse. The proposed effort will include representation of derived-fromtypes of relationships linking anatomical structures at subsequent developmental stages. Such relationships will allow querying for progenitor and derivative tissues. These associations will also enable analysis of differentiation pathways, thus enhancing the ability to explore biological phenomena occurring in the mouse. Anatomy vocabularies are being developed for other organisms and there has been interest in integrating these ontologies at some level. One such effort is the XSPAN project \[[@B32]\], which aims to support cross-species interoperability between developmental anatomy ontologies. On a different scale, Standards and Ontologies for Functional Genomics (SOFG) \[[@B33]\] has set up an international effort to integrate anatomy ontologies of mouse and human. A recent project has been development of the SOFG Anatomy Entry List (SAEL) \[[@B34]\], a list of commonly used anatomical terms that will be directly linked to several major anatomy ontologies, particularly those for human and mouse. It is envisioned that this list will serve as a controlled vocabulary to describe low-resolution anatomical attributes of biological data. For example, the terms included have sufficient resolution to distinguish most samples used for microarray experiments. The Microarray Gene Expression Data (MGED) ontology will use the SAEL for describing anatomical attributes of mouse microarray data. The SAEL and the MGED ontology will also serve as entry points to more comprehensive anatomical resources such as the Adult Mouse Anatomical Dictionary. The Adult Mouse Anatomical Dictionary will be used as a resource to enable standardization and integration of many types of biological data pertinent to postnatal mouse anatomy, including expression, biological process, phenotype and pathology data. GXD currently uses terms from the ontology to annotate expression information at all postnatal stages. While expression results are currently annotated using an abridged version, efforts are underway to map expression data directly to the expanded adult mouse anatomy ontology. GO project curators use terms from the anatomy ontology to describe mouse anatomical concepts. The Mouse Genome Database (MGD) incorporates or associates relevant terms from the adult anatomy ontology into the Mammalian Phenotype Ontology, which is being developed to provide standard terms for annotating mouse phenotype data. Eventually, the standardized anatomy terms will be used to directly link gene expression and phenotype annotations within MGI via the anatomy. The mouse anatomy ontology is also being used to annotate phenotype data for the Eumorphia project \[[@B35]\]. In Pathbase, a database of mutant mouse pathology \[[@B36]\], anatomical attributes of images for mutant postnatal mouse pathology are coded using terms based on the Adult Mouse Anatomical Dictionary. Furthermore, efforts are currently underway to incorporate the adult mouse anatomy ontology into the National Cancer Institute (NCI) Thesaurus, a knowledgebase containing the working vocabularies used in NCI data systems \[[@B37]\]. Anatomy is an important biological integrator. Like expression data, many biological processes and phenotypic observations relate to specific anatomical structures. We have successfully promoted the idea that such data should be described using the same anatomical descriptors. Specifically, we have shown that this can be achieved by describing more complex types of biological information in a modular fashion by combining terms from orthogonal vocabularies \[[@B38]\]. The combinatorial approach takes advantage of existing terms and relationships in the base ontologies. This approach is now being used by most of the resources and projects mentioned above. The use of common anatomical terms will allow for a direct integration of expression, biological process and phenotypic data in the mouse. Links provided with the anatomy terms will, for example, allow display of both expression data and phenotype information associated with specific anatomical structures in the anatomical dictionary browser, as is already the case for developmental expression data \[[@B23]\]. Furthermore, this type of integration will enable complex queries that directly correlate expression and phenotype data. For example, the system will allow queries such as \"Which mouse mutants display phenotypes in a specific anatomical structure?\" and \"How does gene expression in this anatomical structure, or in precursors of this anatomical structure, differ between these mutants and wild type animals?\" Answers to these types of queries hold the promise of providing direct insights into the molecular mechanisms underlying differentiation and disease. Acknowledgements ================ GXD is funded by NIH grant HD33745. M.M. and J.C. were supported by postdoctoral fellowships F32 HD08435-01 and F32 HG00215-01. The authors gratefully acknowledge the contributions of the EMAP, especially Richard Baldock, Jonathan Bard, Duncan Davidson and Matthew Kaufman. We especially thank David P. Hill for input into developing the ontology, Harold Drabkin for assistance and advice on using the DAG-Edit tool, and Constance Smith and Janice Ormsby for critical reading of the manuscript. We also thank colleagues from all the MGI projects for their contributions to an integrated community resource. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Hierarchical organization of the adult mouse anatomy ontology. The hierarchy is divided into spatial and organ system components. Blocks indicate generic group terms appropriate to multiple spatial regions. ::: ![](gb-2005-6-3-r29-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Example showing multiple hierarchical representations for a given anatomical term. The heartis represented both as (a)an organ in the thoracic cavity, and (b)as a part of the cardiovascular organ system. (c)Detail page for the term heartshowing immediate substructures. Note that both spatial and functional representations contain is-aand part-ofrelationships. ::: ![](gb-2005-6-3-r29-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Using the Adult Mouse Anatomical Dictionary Browser. The MGI Browser allows the user to either browse (progressively navigate through the various hierarchies) or search (enter a text string to query for terms, for example limb) within the adult mouse anatomy ontology. \'Term Detail\' pages include the unique numerical identifiers (that is MA ID numbers) for each term, as well as relevant definitions and/or synonyms. ::: ![](gb-2005-6-3-r29-3) :::	PubMed Central	2024-06-05T03:55:56.341092	2005-2-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088948/", "journal": "Genome Biol. 2005 Feb 15; 6(3):R29", "authors": [ { "first": "Terry F", "last": "Hayamizu" }, { "first": "Mary", "last": "Mangan" }, { "first": "John P", "last": "Corradi" }, { "first": "James A", "last": "Kadin" }, { "first": "Martin", "last": "Ringwald" } ] }
PMC1088958	Background ========== In the past few years, several families of regulatory RNA molecules have been shown to be widely expressed in eukaryotes \[[@B1],[@B2]\]. Natural antisense transcripts (NATs) belong to one such family. NATs are endogenous RNA molecules whose partial or entire sequences exhibit complementarity to other transcripts. There are two types of NATs. Cis-NATs are transcribed from the same genomic loci as their sense transcripts but on the opposite DNA strand. By contrast, trans-NATs are expressed from genomic regions distinct from those encoding their sense transcripts \[[@B3]-[@B5]\]. Cis-NATs and their sense RNAs are usually related in a one-to-one fashion, whereas a single trans-NAT may target several sense transcripts; for example, one type of micro RNA (miRNA) could regulate the expression of several distinct target mRNAs \[[@B6]\]. Studies performed in various organisms have suggested that NATs can participate in a broad range of regulatory events, such as transcription occlusion resulting in the reciprocal expression of sense-antisense RNAs \[[@B7],[@B8]\] and RNA interference (RNAi) which leads to the degradation of double-stranded sense-antisense transcript pairs \[[@B9]\]. There is evidence for the involvement of NATs in alternative splicing \[[@B10],[@B11]\], RNA editing \[[@B12],[@B13]\], DNA methylation \[[@B14],[@B15]\], genomic imprinting \[[@B16]-[@B20]\] and X-chromosome inactivation \[[@B21]\]. NATs are also known to regulate expression of some circadian clock genes \[[@B22]\]. However, because each of the above regulatory modes was only observed in a few cases, the general biological functions and regulatory mechanisms of NATs are still unclear. Recent large-scale NAT identifications in several model organisms have revealed the widespread existence of cis-NATs in eukaryotes. Lehner et al.first reported 372 NATs in human by searching for overlapping mRNA sequences in public databases \[[@B23]\]. Using a public expressed sequence tag (EST) database, Shendure and Church also found 144 human NATs and 73 mouse NATs \[[@B24]\]. In a later work, Yelin et al.predicted 2,667 NATs in human and concluded that around 1,600 NAT pairs were transcribed from both strands after experimental validation \[[@B25]\]. The RIKEN group identified 2,481 NAT pairs and 899 non-antisense bidirectional transcript units from 60,770 mouse full-length cDNAs \[[@B26]\]. A similar analysis by the same group uncovered 687 bidirectional transcript pairs from 32,127 rice (Oryza sativa) full-length cDNAs \[[@B27]\]. Antisense expression of about 7,600 annotated genes was observed in a recent work using whole-genome arrays to analyze the transcription activity of the A. thalianagenome. However, a detailed list of these Arabidopsisantisense RNAs and their complete analysis is not yet available \[[@B28]\]. We note that in all previous investigations NAT prediction focused on cis-NATs only. Here, we present results of a genome-wide computational search to predict and identify cis-NATs in Arabidopsis. Combining sequence information of Arabidopsisfull-length cDNAs from the public databases and Arabidopsisannotated genes from the Arabidopsisgenome release, we have identified 1,340 potential cis-NAT pairs. Expression evidence for transcripts derived from both strands of 957 cis-NAT pairs was obtained from the Arabidopsisfull-length cDNA and the public Arabidopsismassively parallel signature sequencing (MPSS) database. Results ======= Prediction and identification of Arabidopsis cis-NAT pairs ------------------------------------------------------------ To search for cis-encoded Arabidopsisnatural antisense transcripts, we aligned all Arabidopsisfull-length cDNA sequences collected in the UniGene and RIKEN datasets with the Arabidopsisgenome sequences. Pairs of transcripts that satisfied the following criteria were selected as cis-encoded natural sense-antisense transcript pairs (referred to as NAT pairs hereafter): first, cDNAs of both transcripts can be uniquely mapped to the Arabidopsisgenome with at least 96% sequence identity; second, the two transcripts are derived from opposite strands of the genome; third, both transcripts are encoded by overlapping genomic loci, and the overlap length is longer than 50 nucleotides; fourth, the sense and antisense transcripts have distinct splicing patterns. Applying all of the above criteria, we identified 332 sense-antisense pairs from Arabidopsisfull-length cDNAs. These NAT pairs are referred to as cDNA-NATs. The 332 pairs of cDNA-NATs can be grouped into two categories. The first category contained 145 NAT pairs in which both the sense and antisense transcripts had nearly perfect annotated gene matches. The second category contained 187 NAT pairs in which at least one transcript had no corresponding annotated gene. This observation led us to hypothesize that additional NAT pairs, whose corresponding cDNAs were not included in the UniGene and RIKEN Arabidopsisfull-length cDNA datasets, could be identified using the Arabidopsisgenome annotation. To identify potential NAT pairs without full-length cDNA evidence, we compared the genomic loci of all Arabidopsisannotated genes to search for gene pairs that overlap in an antiparallel manner. Using the criteria described in Materials and methods, 952 putative NAT pairs were identified from the Arabidopsisgenome and were named genomic-NATs. Among the 952 genomic-NATs, 145 pairs had corresponding full-length cDNA for both the sense and antisense genes, and therefore were also included in the cDNA-NAT set. The remaining 807 new NAT pairs were predicted using the Arabidopsisgenome annotation only and are referred as the unique genomic-NAT set in the following analysis (Figure [1a](#F1){ref-type="fig"}). For most NAT pairs in the second category of the cDNA-NAT set, only one transcript in each pair matched an annotated gene. This indicates that transcripts of some full-length cDNAs could form cis-NAT pairs with other transcripts, although their corresponding genes are not included in the current Arabidopsisgenome annotation. In a search of such NAT pairs, we compared the genomic loci of the UniGene and RIKEN Arabidopsisfull-length cDNAs with those of annotated genes and identified 1,291 full-length cDNAs whose transcripts could form cis-NAT pairs with potential transcripts of annotated genes (see Materials and methods for criteria). The 1,291 genomic-cDNA-NAT pairs included the 332 cDNA-NAT pairs and 758 unique genomic-NAT pairs. Therefore, 201 unique NAT pairs were predicted by the cDNA-genome comparison approach and are referred to as unique genomic-cDNA-NAT pairs hereafter (Figure [1b](#F1){ref-type="fig"}). In total, we have found 1,340 potential NAT pairs from three categories: 332 pairs with cDNA evidence for both sense and antisense transcripts; 807 pairs based on the Arabidopsisgenome annotation (including 758 pairs with full-length cDNA evidence for one strand) and another 201 genomic-cDNA pairs by combining genome annotation with full-length cDNA sequence information. Characterization of ArabidopsisNAT pairs ------------------------------------------ We classified the 1,340 unique NAT pairs according to the exon-intron structures of each transcript and their overlapping patterns (Table [1](#T1){ref-type="table"}). The overlapping patterns of NAT pairs were determined by comparing the exon positions of both transcripts using sim4 \[[@B29]\] alignment results. Consistent with previous reports of NAT pairs in other organisms \[[@B23]-[@B27]\], the majority of ArabidopsisNAT pairs (72.1%) overlapped at their 3\' end. For almost all NAT pairs (99%), the overlapping region included exon sequences, with a few exceptions in which one transcript was transcribed entirely from the intronic sequences of the other. Figure [2](#F2){ref-type="fig"} shows the distribution of overlap lengths of NATs. No obvious chromosomal bias was observed for the genomic distribution of NATs (Table [2](#T2){ref-type="table"}) \[[@B30]\]. The sim4 cDNA alignment results showed that some Arabidopsisfull-length cDNAs are non-spliced transcripts. To assess the quality of full-length cDNAs, we systematically compared the splicing pattern and coding potential of all full-length cDNAs used in this study to all predicted Arabidopsisgenes. Our result showed that the proportion of non-spliced transcripts in UniGene and RIKEN full-length cDNAs was lower than the proportion of non-spliced transcripts in annotated genes, indicating non-spliced cDNAs are likely to be derived from bona fidetranscripts rather than genomic DNA contamination (Table [3](#T3){ref-type="table"}). Expression analysis of NAT pairs using public ArabidopsisMPSS data -------------------------------------------------------------------- To investigate the expression of our predicted NAT pairs, we used the public ArabidopsisMPSS data at the University of Delaware \[[@B31]\]. MPSS is a bead-based sequencing technology that identifies a sequence of 17-20 nucleotides from each transcript. This sequencing technique is capable of identifying new, rarely expressed transcripts. MPSS can also quantitatively measure the expression level of a transcript because the transcripts per million (TMP) value for a transcript in the sequencing results reflect its in vivoabundance \[[@B32],[@B33]\]. The public ArabidopsisMPSS database contains 87,705 \'trusted\' signature sequences from 14 cDNA libraries. By aligning these MPSS sequences to the Arabidopsisgenome and the 1,340 NAT pairs, we identified 455 NAT pairs with unique MPSS matches on both the sense and antisense strands, including 103 cDNA-NAT pairs, 293 genomic-NAT pairs and 59 genomic-cDNA-NAT pairs. Because MPSS signatures are short 17-nucleotide sequences identified from each transcript, sequences with multiple genomic loci were excluded from our analysis to avoid ambiguity with respect to the origin of a MPSS signature and to ensure fidelity of assigning a MPSS signature to its corresponding transcript (see Materials and methods for details). Among the 455 NAT pairs with unambiguous MPSS data for both transcripts, expression of both transcripts of 78 pairs was only found in distinct libraries, indicating these NAT pairs might have an exclusive transcription relationship. For the other 377 NAT pairs, expression of the sense and antisense transcripts was mainly observed in different libraries or one transcript was dominantly expressed when both transcripts could be detected in the same library (Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}). For a pair of NATs found in the same library, if the TPM value of one transcript is at least three times as high as that of the other transcript, we consider that transcript as dominantly expressed. The number of coexpressed and dominantly expressed transcripts in each library was shown in Figure [3](#F3){ref-type="fig"}. On average, coexpression was only observed in two of the 14 tested sample libraries for each of the 377 NAT pairs, whereas dominant expression of one transcript was observed in 9 libraries. No expression was detected in the remaining libraries. We also found additional 222 genomic-NAT pairs and 51 genomic-cDNA-NAT pairs with full-length cDNA evidence for one transcript and MPSS data for the other transcript. Together with the 332 cDNA-NAT pairs, we have obtained either full-length cDNA or MPSS expression evidence for both transcripts of 957 NAT pairs, corresponding to 71.4% of the total 1340 pairs ((455 - 103) + 332 + 222 + 51 = 957). siRNA matches of NAT pairs -------------------------- We compared short interfering RNA (siRNA) sequences collected in the Arabidopsissmall RNA database to investigate the possibility that cis-NAT pairs may generate siRNAs. Similar to the MPSS alignment process, only siRNAs with unique loci on the Arabidopsisgenome were used in the comparison to ensure unambiguous assignment. We found 11 pairs of NATs had siRNA sequences mapped uniquely to their overlapping region (Table [6](#T6){ref-type="table"}). SiRNAs of all but one NAT pairs originated from their overlap region, the only exception being pair At\#S18901030 and At\#S18898439, whose overlap length was only 52 nucleotides long. Conservation of ArabidopsisNAT pairs in rice ---------------------------------------------- To examine whether NAT pairs might be conserved during evolution, we compared the protein sequences of the 1,340 putative ArabidopsisNAT pairs with the protein sequences of the 687 predicted rice NAT pairs \[[@B27]\]. Orthologs of two ArabidopsisNAT pairs were also encoded by antiparallel genes originated from the same locus in rice (Table [7](#T7){ref-type="table"}). In addition, homologs of one transcript of 392 ArabidopsisNAT pairs were also found in the rice NAT set. Discussion ========== Although NATs are often seen in prokaryotes, their prevalence in eukaryotes was not detected until the past few years \[[@B23]-[@B27],[@B34]\]. In this work, we combined sequence information on Arabidopsisfull-length cDNAs with that from the Arabidopsisgenome annotation and identified 1,340 potential cis-NAT pairs in Arabidopsis(Additional data file 1, 2, 3). Assessment of our NAT prediction methods ---------------------------------------- The 1,340 ArabidopsisNAT pairs were identified from three sources. First, by aligning full-length cDNA sequences to the Arabidopsisgenome, we identified 332 cDNA-NAT pairs. However, comparison of these 332 cDNA-NAT pairs with Arabidopsisannotated genes showed that more than half of these NAT pairs had one partner that was not included in the current Arabidopsisgenome annotation. Because traditional genome annotation mainly aims at the identification of protein coding genes within a genome, there is the possibility that non-coding antisense transcripts may be overlooked by currently trained gene finders. A recent report using a genome tiling array to examine the transcription activity of the entire Arabidopsisgenome also supports this notion \[[@B28]\]. To search for potential NAT pairs not included in the current full-length ArabidopsiscDNA library, we compared the genomic coordinates of all annotated genes with each other and with those of full-length cDNAs. This approach uncovered another 807 overlapping genomic-NAT pairs based on the annotation of their corresponding genes, and 201 genomic-cDNA-NAT pairs, each including a transcript derived from an annotated gene on one strand and a transcript represented in the full-length cDNA database on the other strand. The full-length cDNAs included in genomic-cDNA-NAT pairs either had no annotated gene match or their corresponding transcripts cannot form cis-NAT pairs with transcripts of other genes based on their annotation. These results indicate that although the Arabidopsisgenome is currently one of the best annotated eukaryotic genomes, a lot of information is still missing. The identification in eukaryotes of several classes of regulatory RNA genes, such as those encoding natural antisense transcripts, which are the focus here, will not only further our understanding of genome structure and gene regulation, but will also open a new window for improved genome annotation. Most antisense prediction work reported to date has focused on identifying NATs from expressed cDNAs and ESTs \[[@B23]-[@B27]\]. In this work, we avoided using ESTs because of the ambiguous orientation of some sequences. We also included sequence information of annotated Arabidopsisgenes in our NAT prediction in order to provide a more complete picture of antisense transcripts in Arabidopsis. The reliability of our approach is supported by the following lines of evidence: first, the expression of both sense and antisense transcripts of 293 pairs of genomic-NATs (36.3% of a total of 807) was observed in the public MPSS data, and another 222 genomic-NAT pairs (27.5% of a total of 807) have full-length cDNA evidence for one transcript and associated MPSS data for the other transcript; second, the two NAT pairs which were conserved in rice were also identified in our Arabidopsisgenomic-NAT dataset; third, it is known that imprinted genes are likely subject to antisense regulation; three of the six reported Arabidopsisimprinted genes \[[@B35]-[@B39]\], FIE, FIS2and MSI1, are included in our genomic-NAT sets. However, it remains possible that some genomic-NAT pairs are false positives if the lengths of their untranslated regions (UTRs) were annotated inaccurately. In rice, both transcripts of 86% of the NAT pairs have coding sequence (CDS) regions whereas 28% of the predicted ArabidopsisNAT pairs include at least one transcript without coding potential. Non-protein-coding transcripts are more prevalent in cDNA-and genomic-cDNA-NAT pairs in that 170 cDNA NAT pairs and 156 genomic-cDNA-NAT pairs include one non-protein-coding transcript. We used Genescan to evaluate the coding potential of each transcript by screening their corresponding genomic DNA sequence for valid gene structures. Using annotated genes as controls, we estimated the false-negative rate of our definition of coding potential to be 2.3%. Unlike CDS-containing antisense transcripts that may be translated into proteins under certain conditions, transcripts without any protein-coding potential could possess solely regulatory functions. In our work described here, and in all other genome-wide antisense transcript identification papers published so far \[[@B23]-[@B27]\], the investigation was focused on cis-antisense RNAs, which are transcribed from the same genomic loci as their sense RNAs, but on the opposite genome strand. To ensure the cis-antisense relationship of NATs reported here, only cDNAs with unique genomic loci were included in this study. We note that certain number of trans-antisense transcripts also exist in cells. Examples include miRNAs and siRNAs which are widely studied in most model organisms \[[@B6]\]. Genome-wide identification of trans-antisense transcripts in Arabidopsisis being attempted. Evaluation of NAT expression using MPSS data -------------------------------------------- The non-gel-based properties of MPSS technology render it an ideal resource for evaluating the expression profile of NAT pairs for the following reasons: first, because the MPSS technology captures almost all polyadenylated transcripts within cells, this technology is theoretically capable of identifying new, rarely expressed transcripts without prior knowledge of their corresponding genes; second, the digital result of MPSS reflects the expression pattern of a sequenced RNA molecule, and therefore provides a quantitative relationship between the sense and antisense transcript of a NAT pair in different tissues. This information was not available in any of the previous NAT prediction work \[[@B32],[@B33]\]. Using the full-length cDNA and public ArabidopsisMPSS data, we were able to obtain expression evidence for both transcripts of 957 NAT pairs. The digital nature of MPSS data enabled us to evaluate the expression relationship of the sense and antisense transcripts directly. Our results showed that the sense and antisense transcripts of a NAT pair tend to be expressed in different tissues or under different conditions. In addition, in cases where the sense and antisense transcripts of a NAT pair were expressed in the same library, one type of transcript was usually more abundant than the other. On average, transcripts of NAT pairs were found to be coexpressed in only two libraries, whereas dominant expression (the expression level of one transcript was at least three times higher than that of the other transcript) or absolute expression (only one transcript of a NAT pair was expressed) was observed in nine libraries. The tissue-specific expression of sense and antisense transcripts observed in this study is consistent with the Arabidopsisgenome transcription study using a whole genome-tiling array, in which about 7,600 genes were found to have tissue-specific sense and antisense expression \[[@B28]\]. Although a detailed list of these 7,600 genes is not yet available, it is possible that for some genes not included in our list, the antisense transcription activity was contributed by trans-antisense transcripts. This could explain why we predicted fewer NAT pairs than the previous work, as our work only focuses on cis-antisense transcripts. To ensure the MPSS sequences were indeed generated by their matching transcripts, all MPSS data were first aligned with the Arabidopsisgenome and all annotated mRNAs to remove signatures with multiple genomic loci. Therefore, unless an MPSS signature sequence was derived from the joint-exon region of some transcripts that are not included in the current genome annotation, it should originate from its corresponding transcript. Speculation on the function and origin of NATs ---------------------------------------------- One possible function of NATs is to trigger the degradation of their sense transcripts via the RNAi pathway. However, in our study, we found only 11 NAT pairs with known siRNA matches. There are two possible explanations for this observation. First, the current public ArabidopsissiRNA database, which only contains 1,822 unique siRNA sequences, is small and does not cover all siRNAs associated with sequences of the NAT pairs reported here. Second, all NATs identified in this work are cis-antisense transcripts. siRNAs are used to downregulate expression levels of their target mRNA to achieve a low protein concentration. Cis-antisense transcripts can accomplish the same goal by interfering with the transcription of their sense transcripts, and this might be a more energy-efficient mechanism to achieve local gene regulation. This hypothesis predicts that we would expect to find more siRNAs associated with trans-antisense transcripts. For most NAT pairs with associated MPSS data for both transcripts, the expression of sense and antisense transcripts tends to occur in different tissues. In these cases, we could speculate that transcription of genes encoding these NAT transcript pairs may be regulated by similar factors but that the production of antisense transcripts might interfere with the transcription of their sense transcripts, resulting in reciprocal expression patterns. Another possibility is that the two genes of a NAT pair are subject to different transcriptional regulation and consequently they are never expressed in the same tissue at the same time. Functional analysis of all NAT pairs using gene ontology reveals no over-representation of any functional category compared to the Arabidopsisgenome, indicating that cis-antisense regulation might be a global mechanism for all gene families. Further experiments are needed to investigate the validity of these hypotheses. Antiparallel transcription and antisense transcripts are known to be involved in genomic imprinting of Xistgene in mouse and human \[[@B21]\]. There is supporting evidence that the MEAand PHEgenes of Arabidopsisare imprinted \[[@B35]\], and in addition, FIS2, FIE, MSI1and FWAmay also be imprinted, although the evidence for these four other genes is not unequivocal \[[@B36]-[@B39]\]. Nonetheless, we found antisense transcription units for FIS2, FIEand FWA, suggesting that transcription of these three genes might be regulated by antisense transcripts, or their antisense transcripts might be involved in silencing their expression. Genomic imprinting usually involves a chromosomal locus and, in certain cases, may even extend overa chromosomal region. Given the close proximity of the sense-antisense gene transcripts if one member of the pair is imprinted, it is likely that the other would be subject to the same regulation. Unfortunately, because of the absence of data on imprinted genes in rice, we were unable to examine whether imprinted genes were also subject to antisense regulation in rice. We found that two ArabidopsisNAT pairs are conserved in rice. These conserved NAT pairs could be used to study the antisense regulatory mechanism and the origin of NATs in plants. Given over 150 million years of evolutionary distance between Arabidopsisand rice, the gene order on the two genomes has diverged quite significantly. Therefore, the conservation of these two NAT pairs might have some functional relevance. A closer comparison of the Arabidopsisand rice NAT pairs and the identification of additional conserved NAT pairs could help address this issue. Taken together, our results provide the first genome-wide identification and prediction of NATs in Arabidopsis. These results will facilitate functional studies of NATs in this model plant, as well as in other plant species, and help to unravel complex gene regulatory networks in eukaryotes. Materials and methods ===================== Identification of sense-antisense transcript pairs from full-length cDNA datasets --------------------------------------------------------------------------------- The ArabidopsisUniGene (Build 45) dataset (file named At.seq.all) was downloaded from the National Center for Biotechnology Information (NCBI) UniGene Resources \[[@B40],[@B41]\]. A total of 20,683 full-length cDNA sequences were extracted from the UniGene dataset by selecting sequences marked as \'Full-length/full-length cDNA\'. The RIKEN Arabidopsisfull-length cDNA dataset, which contains 13,181 sequences, was downloaded from the RIKEN BioResource Center (BRC) \[[@B42],[@B43]\]. The 20,683 UniGene and 13,181 RIKEN full-length cDNAs were aligned to the Arabidopsisgenome sequences from The Institute for Genomic Research (TIGR) (release version 5) \[[@B44]\] by BLAT. The splicing pattern of the transcript derived from each cDNA was further confirmed using the sim4 sequence alignment program \[[@B29],[@B44],[@B45]\]. cDNAs with at least 96% sequence identity to the Arabidopsisgenome were used in the following analysis. For pairs of cDNAs encoded by opposite strands of the Arabidopsisgenome and sharing overlapping genomic loci, if both their corresponding sense and antisense transcripts had no other genomic locations and exhibited different splicing patterns, they were selected as encoding sense-antisense transcript pairs and are referred to as cDNA-NAT pairs in the text. Prediction of sense-antisense transcript pairs using the Arabidopsisgenome annotation and full-length cDNAs ------------------------------------------------------------------------------------------------------------- We used the A. thalianagenome annotations from TIGR (release version 5) in this study \[[@B44],[@B45]\]. Putative NAT pairs were identified on the basis of annotated genomic loci of Arabidopsisgenes. If a pair of overlapping genes were located on opposite strands of the Arabidopsisgenome and at least one gene had no annotated UTR at the overlap end, their encoded transcripts were selected as a putative NAT pair regardless of the overlap length of the encoded transcripts. Otherwise, if a pair of antiparallel overlapping genes both have annotated UTR regions at the overlap end, the overlap length of their encoded transcripts must be longer than 50 nucleotides to qualify as NAT pairs. NAT pairs from the above two categories are both referred to as genomic-NAT pairs in the text. Genomic-cDNA-NAT pairs were identified by comparing the genomic loci of full-length cDNAs with those of annotated genes. UniGene and RIKEN full-length cDNAs with unique genomic locations and at least 96% sequence identity to the Arabidopsisgenome were used in this step. Using the same criteria for genomic NATs, if an annotated gene had a overlap cDNA match on the opposite strand and the transcript of the annotated gene and that derived from the antisense cDNA had different splicing patterns, the gene and its matching cDNA were selected as a genomic-cDNA-NAT pair. Splicing pattern and coding potential evaluation of full-length cDNAs and annotated genes ----------------------------------------------------------------------------------------- Splicing patterns of transcripts encoded by full-length cDNAs were obtained by aligning the cDNA sequences to the Arabidopsisgenome using the sim4 program \[[@B29]\]. Splicing patterns of transcripts derived from Arabidopsisannotated genes were extracted from the TIGR Arabidopsisgenome annotation (release version 5) \[[@B44]\]. To evaluate the coding potential of full-length cDNAs, their corresponding genomic sequences (determined by BLAT and sim4 result) were extracted and screened by GeneScan \[[@B46]\]. Identification of MPSS evidence for NAT pairs --------------------------------------------- We used the public ArabidopsisMPSS data at the University of Delaware \[[@B31]\] to evaluate the expression of NAT pairs. MPSS sequences from 14 different libraries of ArabidopsisColumbia-0 (Col-0) ecotype were downloaded from \[[@B31]\]. Each MPSS library contained signature sequences identified from the same tissue. The quality of these MPSS sequences was evaluated according to the information provided by the database. Only MPSS sequences with \'reliable\' (present in more than one sequencing run) and \'significant\' (TPM ≥ 4) expression pattern were considered as \'trusted\' signatures and used in this analysis. The public MPSS database contained 87,705 trusted signatures that satisfied the above expression criteria. These signatures were aligned to the sequences of the 1,340 putative NAT pairs to identify MPSS sequences derived from them. Signatures with multiple perfect matches to the Arabidopsisgenome or to cDNAs had ambiguous origins and were not considered further. For a NAT pair, if both the sense and antisense transcripts had associated MPSS data and their expression values were both significant in one or more libraries, transcripts in this NAT pair were considered as coexpressed in the same tissue. On the other hand, if both transcripts had MPSS data but had no significant coexpression in any of the examined libraries, then the transcripts were considered as expressed, but in different libraries. Homology comparison with reported rice NATs ------------------------------------------- Full-length cDNA sequences of the 687 rice NAT pairs were downloaded from the website described in \[[@B27]\]. To facilitate protein sequence comparison, the rice and ArabidopsiscDNAs were mapped to their corresponding genomes by BLAT \[[@B45]\]. Both the A. thalianaand O. sativagenomes were downloaded from TIGR \[[@B44]\]. The corresponding genomic sequences of each cDNA were extracted according to their genomic coordinates from the BLAT results. Protein sequences were obtained by evaluating the genomic sequences of those cDNAs using GENSCAN \[[@B46]\]. The protein sequences of rice NATs were aligned with those of ArabidopsisNATs using blastp \[[@B47]\]. High similarity pairs with E-value less than 10^-30^and alignment coverage greater than 50% of query sequence were considered as homologous sequences. Additional data files ===================== The following additional data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} is a table listing all genomic-NAT pairs. Additional data file [2](#S2){ref-type="supplementary-material"} is a table listing all cDNA-NATs. Additional data file [3](#S3){ref-type="supplementary-material"} is a table listing all genomic-cDNA-NATs. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Classes of overlap patterns: 1. tail to tail (3\' end overlap); 2. head to head (5\' end overlap); 3. one transcript is contained entirely within the other transcript; 4.two transcripts overlap only within introns. Coding potential of a transcript: \'+\' with coding potential; \'- \' without coding potential ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Classes of overlap patterns: 1. tail to tail (3\' end overlap); 2. head to head (5\' end overlap); 3. one transcript is contained entirely within the other transcript; 4.two transcripts overlap only within introns. Coding potential of a transcript: \'+\' with coding potential; \'- \' without coding potential ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Classes of overlap patterns: 1. tail to tail (3\' end overlap); 2. head to head (5\' end overlap); 3. one transcript is contained entirely within the other transcript; 4.two transcripts overlap only within introns. Coding potential of a transcript: \'+\' with coding potential; \'- \' without coding potential ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Takatoshi Kiba and Siripong Thitamadee for fruitful discussions and Peter Hare and Yupu Liang for carefully reading the manuscript. This research was supported by NIH GM44640 to N-H.C. and DBI-9984882 to T.G. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Relationships between NAT pairs from different datasets. (a)Overlap between cDNA-NAT pairs and genomic-NAT pairs. Among the 332 cDNA-NAT pairs, 145 pairs have corresponding annotated genes for both transcripts. For the other 187 cDNA-NAT pairs, at least one transcript has no counterpart in the current Arabidopsisgenome annotation. (b)Overlap between cDNA-, genomic- and genomic-cDNA-NAT pairs. All cDNA-NAT pairs are included in genome-cDNA-NAT pairs. Blue circle, cDNA-NATs; red circle, genomic-NATs; green circle, genomic-cDNA-NATs. ::: ![](gb-2005-6-4-r30-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distribution of genomic overlap lengths of NATs. The overlap length of each NAT pair in exons was calculated. The number of NAT pairs (y-axis) is plotted against the overlap lengths (in nucleotides) of exons in each NAT pair (x-axis). ::: ![](gb-2005-6-4-r30-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Distribution of coexpressed and dominantly expressed NAT pairs in different libraries. The number of coexpressed NAT pairs in each library was shown in blue bar and that of dominantly expressed NAT pairs in red bar. See legend of Table 5 for library information. ::: ![](gb-2005-6-4-r30-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Structure analysis of NAT pairs ::: Category Number of pairs --------------------------------------------------------------- ----------------- ----- ----- -------------- Tail to tail (3\' to 3\') 181 737 48 966 (72.1%) Head to head (5\' to 5\') 97 31 57 185 (13.8%) One transcript contained entirely within the other transcript 51 35 90 176 (13.1%) Two transcripts overlap only within introns 3 4 6 13 (1.0%) Total 332 807 201 1,340 (100%) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Chromosomal distribution of NAT pairs ::: Chromosome Number of NAT pairs Chromosome size (Mb) ------------ --------------------- ---------------------- ----- ------ ------- 1 85 216 55 356 29.1 2 41 120 40 201 19.6 3 69 142 46 257 23.2 4 48 129 29 206 17.5 5 89 200 31 320 26.0 Total 332 807 201 1340 115.4 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Splicing pattern and coding potential of Arabidopsisfull-length cDNAs and annotated genes ::: UniGene cDNAs RIKEN cDNAs The Arabidopsisgenome ------------------------------------------------- --------------- ------------- ------------------------- Total transcripts 20,683 13,181 29,993 Number of transcripts with perfect genome match 17,814 12,877 29,993 Number of transcripts with ORFs 16,621 12,544 26,207 Number of non-spliced transcripts with ORFs 2,534 1,555 4,722 Number of transcripts without ORFs 1,193 333 3,786 Number of non-spliced transcripts without ORFs 466 130 3,786 The splicing pattern of each transcript was obtained by aligning its corresponding cDNA sequences to the Arabidopsisgenome using sim4. The coding potential of the genomic sequence of each transcript was examined by GeneScan. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Summary of MPSS matches for NAT pairs ::: Number of NAT pairs ------------------------------------------------------------------------- --------------------- ----- ----- ------- Total NAT pairs 332 807 201 1,340 Number of pairs with MPSS matches on both strands Total 103 293 59 455 Expressed absolutely in different libraries 14 49 15 78 Expressed mainly in different libraries, occasionally in same libraries 89 244 44 377 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Examples of NAT pairs with MPSS matches on both strands ::: ID Strand Libraries ----------- -------- ----------- ------ ------ ------ ------ --------- ------- ------- ------- ------- ------ ------ ------ ------ Pair A At1g09750 \+ N N N N N 9 N 0 N 1 N N N 1 At1g09760 \- 70 39 32 46 30 240 125 139 208 170 56 48 48 45 Pair B At1g72060 \+ 5 N 31 2 N N N N N 2 N 1 74 79 At1g72070 \- 0 N N 1 N N N N N N 8 N N N Distinct expression of sense and antisense transcripts of NAT pair A was observed in all but one library. In the library where both transcripts of pair A were expressed, the abundance of one transcript was significantly higher than the other. For NAT pair B, the sense and antisense transcripts were expressed differentially in different libraries. Libraries in which both transcripts of a NAT pairs were expressed are shown in bold; libraries in which transcripts of only one gene of a NAT pairs were expressed are shown in italics. Abbreviations for libraries: CAF, callus - actively growing, classic MPSS; INF, infloresence - mixed stage, immature buds, classic MPSS; LEF, leaves - 21 day, untreated, classic MPSS; ROF, root - 21 day, untreated, classic MPSS; SIF, silique - 24-48 h post-fertilization, classic MPSS; AP1, ap1-10 infloresence - mixed stage, immature buds; AP3, ap3-6 infloresence - mixed stage, immature buds; AGM, agamous infloresence - mixed stage, immature buds; INS, infloresence - mixed stage, immature buds; ROS, root - 21 day, untreated; SAP, sup/ap1 infloresence - mixed stage, immature buds; S04, leaves, 4 h after salicylic acid treatment; S52, leaves, 52 h after salicylic acid treatment; LES, leaves - 21 day, untreated. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### siRNA matches of NAT pairs ::: Category of NAT pairs Gene ID Strand Overlap length (nucleotides) Description ----------------------- --------------- -------- ------------------------------ --------------------------------------------------- Genomic-NAT At2g06510 \+ 506 Replication protein, putative At2g06520 \- Membrane protein, putative At4g35850 \+ 360 Pentatricopeptide (PPR) repeat-containing protein At4g35860 \- Ras-related GTP-binding protein, putative At5g20720 \+ 294 Chaperonin, chloroplast At5g20730 \- Auxin-responsive factor At5g41680 \+ 587 Protein kinase family protein At5g41685 \- Mitochondrial import receptor subunit TOM7 At5g48870 \+ 118 Small nuclear ribonucleoprotein, putative At5g48880 \- Acetyl-CoA C-acyltransferase 1 cDNA-NAT RAFL19-56-G17 \+ 1,209 No coding potential RAFL09-70-E21 \- Expressed protein At\#S18901030 \+ 52 Putative transcription factor At\#S18898439 \- Pentatricopeptide (PPR) repeat containing protein At\#S18900150 \+ 884 No coding potential At\#S18898471 \- expressed protein At\#S18912025 \+ 1,149 No coding potential At\#S18898946 \- TCP family transcription factor Genomic-cDNA-NAT At1g07725 \+ 1,640 Exocyst subunit EXO70 family protein At\#S18898556 \- No coding potential At2g16587 \+ 379 expressed protein RAFL19-48-E15 \- No coding potential ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Conserved NAT pairs of Arabidopsisand rice ::: ID Strand Overlap pattern Overlap length (nucleotides) Description ------------ --------------- ------------- -------- ----------------- ------------------------------ --------------------------------- NAT pair 1 Arabidopsis At5g02820 \+ Tail to tail 1,138 DNA topoisomerase VIA At5g02830 \- PPR repeat-containing protein Rice J033010B03 \+ Tail to tail 1 DNA topoisomerase VIA J013135M09 \- PPR repeat-containing protein NAT pair 2 Arabidopsis At5g54270 \+ Tail to tail 1,047 Chlorophyll A-B binding protein At5g54280 \- Myosin heavy chain Rice 006-301-C08 \+ Tail to tail 4,425 Chlorophyll A-B binding protein J013155K02 \- Myosin heavy chain :::	PubMed Central	2024-06-05T03:55:56.343254	2005-3-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088958/", "journal": "Genome Biol. 2005 Mar 15; 6(4):R30", "authors": [ { "first": "Xiu-Jie", "last": "Wang" }, { "first": "Terry", "last": "Gaasterland" }, { "first": "Nam-Hai", "last": "Chua" } ] }
PMC1088959	Background ========== In most organisms, one of three stop codons (UAA, UAG and UGA) signals the termination of protein translation. Occasionally a near-cognate tRNA misreads a stop codon and the ribosome reads through the termination signal. Nichols \[[@B1]\] proposed that a second stop codon following the real termination codon could act as a backup. Since this codon follows the real stop codon, it is called a \'tandem stop codon\'. By giving the translation machinery a second chance to terminate protein translation \[[@B2]\], tandem stop codons provide a \'fail safe mechanism\'. When read-through occurs, extra amino acids are added to the end of the peptide chain. The presence of tandem stop codons downstream of genes reduces the number of extra amino acids, which would influence protein folding; the addition of fewer extra amino acids increases the likelihood that the protein will preserve its three-dimensional structure. In addition, minimizing the number of amino acids added to the polypeptide chain is also beneficial from a purely energetic point of view. These factors suggest that the existence of tandem stop codons would confer a selective advantage, and imply that such codons do not need to follow the real termination signal immediately to be beneficial. So far, the existence of tandem stop codons has remained elusive. In Escherichia coli, for example, a recent study argued that the slightly over-represented tandem stop codons occur only as a consequence of the strong preference for U at the first nucleotide position following the stop codon as part of an efficient stop signal \[[@B3]\]. Tandem stop codons in eukaryotes have not been rigorously examined. Therefore we decided to analyze tandem stop codons in yeasts. We chose yeast as a model system for two main reasons: first, Saccharomyces cerevisiaeis the best-studied eukaryotic genome and the quality of its annotation has been greatly improved by recent comparative analyses \[[@B4]-[@B6]\], making it possible to detect signals of weak selection forces at the genomic level; second, rich experimental data on translation termination efficiency are available to further characterize the nature of these forces. In this study we carried out a computational analysis to address the question of whether tandem stop codons occur at any location downstream of yeast genes with unusually high frequency; whether tandem stop codons are under selection; and what is the primary factor influencing tandem stop codons. Results ======= Are tandem stop codons over-represented in the downstream sequences of yeast genes? ----------------------------------------------------------------------------------- In this study we define tandem stop codons as any stop codon that is downstream from, in the same frame as, and within a relatively short distance from, the real stop codon. We only considered the first nine codon locations downstream of the annotated stop codon, which are all in non-coding regions. The frequency of stop codons in each of these locations following the real stop codon was determined separately for all genes in S. cerevisiaethat use TAA, TAG, and TGA as stop codons, respectively (throughout this paper T is used instead of U where genomic sequences are considered). The frequency of stop codons at the corresponding locations following each of these three nucleotide triplets (TAA, TAG and TGA) in non-coding regions was also calculated as a control. These results are shown in Figure [1a-d](#F1){ref-type="fig"}. Surprisingly, we found that a highly significant excess of stop codons has evolved at the third codon located downstream of the real stop codon TAA, designated here as the UAA+3 codon (we use U instead of T to indicate mRNA) (9.0% versus 6.4%, χ^2^= 24.2, P\< 9 × 10^-7^). No significant excess of stop codons was detected at any other location. The identity of stop codons at the UAA+3 codon location did not matter (Figure [1e](#F1){ref-type="fig"}); we found no statistical significance of any particular stop codon being either over-represented or under-represented at this site. We examined this location in three other closely related yeast species, S. paradoxus, S. mikataeand S. bayanus, and in the distantly related yeast Candida glabrata, which was recently sequenced \[[@B7]\]. Statistically significant excesses of stop codons were confirmed at this location in all species (S. paradoxusχ^2^= 15.1, P\< 1 × 10^-4^,S. mikataeχ^2^= 9.4, P\< 2 × 10^-3^; S. bayanusχ^2^= 20.6, P\< 6 × 10^-6^; C. glabrataχ^2^= 9.0, P\< 3 × 10^-3^) (Figure [2](#F2){ref-type="fig"} and Additional data file 1). As another control, we performed the same analysis in the other two frames (frame+2, with codons beginning at the second nucleotide after the stop codon; and frame+3, with codons beginning at the third nucleotide after the stop codon) in S. cerevisiaeand found several other weakly over-represented locations, such as UAA+1 (frame+2), but none of these trinucleotide locations shows the same tendency in the other species (Additional data files 2 and 3). In-frame codon UAA+3 is the only location significantly over-represented among all the yeast species, indicating that tandem stop codons at UAA+3 are well conserved in yeasts to a large extent. This conservation in all examined species strongly suggests that the observed excess is biologically meaningful, rather than random noise. Therefore, the following analysis will focus only on tandem stop codons at this location. Are tandem stop codons at the UAA+3 codon location under selection? ------------------------------------------------------------------- In order to determine if the excess of tandem stop codons at the UAA+3 codon location are under selection, we examined the third codon following the real stop codon in the 1,029 unambiguous orthologous gene pairs between S. cerevisiaeand S. bayanus(defined in \[[@B4]\]) in which both orthologs use the UAA stop codon. S. bayanuswas chosen because, among the three Saccharomycesspecies, it is the most distantly related to S. cerevisiae\[[@B4]\] and thus the substitutions between these two species provide a better resolution for comparison. The ancestral states of UAA+3 codons were reconstructed using parsimony. Then the number of conserved stop codons, non-conserved stop codons, conserved sense codons and non-conserved sense codons were calculated and are shown in Table [1](#T1){ref-type="table"}. To test whether stop codons at this location are statistically more conserved than sense codons, we used a chi-squared independence test. We found that the conservation of codons at the UAA+3 location strongly depends on whether it is a stop codon (χ^2^= 6.1, P\< 0.01) (Table [1](#T1){ref-type="table"}). Furthermore, we performed the same analysis between S. cerevisiaeand each of the other two remaining Saccharomycesspecies. The tendency of stop codons to be more conserved than sense codons at UAA+3 was the same in all comparisons (between S. cerevisiaeand S. paradoxus, 1,509 orthologous gene pairs, χ^2^= 2.8, P\< 0.1; between S. cerevisiaeand S. mikatae, 1,075 orthologous gene pairs, χ^2^= 4.6, P\< 0.03). With the increase in evolutionary divergence in the comparative analysis, the statistical significance becomes more striking. In addition, among 504 gene groups in which all the orthologous genes in the four Saccharomycesspecies use UAA as stop codons, 10.5% of them have tandem stop codons at UAA+3, which is much higher than random expectation (χ^2^= 18, P\< 2 × 10^-5^). Therefore, tandem stop codons at the UAA+3 codon location appear to be maintained by selection. Does codon bias influence the distribution of tandem stop codons? ----------------------------------------------------------------- We calculated the codon bias (effective number of codons, ENC) \[[@B8]\] of all the genes with UAA stop codons in S. cerevisiae. The number of genes with a tandem stop codon located at the UAA+3 location in the high codon bias quartile (25% of genes with the lowest ENC values) and the low codon bias quartile (25% of genes with the highest ENC values) were determined and then compared. To test whether tandem stop codons tend to follow genes with high codon bias, we used a chi-squared independence test. We found that the proportion of genes with a tandem stop codon in the high codon bias quartile is significantly higher than that in the low codon bias quartile (15% versus 6%; χ^2^= 27.2, P\< 2 × 10^-7^, Table [2](#T2){ref-type="table"}). A Kolmogrov-Smirnov test also indicated that the codon bias distributions were significantly different (P\< 2.7 × 10^-9^, Figure [3](#F3){ref-type="fig"}) between the genes with and without a tandem stop codon. Do tandem stop codons tend to occur in essential genes or genes with shared features? ------------------------------------------------------------------------------------- First, we classified the genes with UAA stop codons into four groups (essential genes, strong fitness effect, moderate fitness effect and weak fitness effect) on the basis of the minimum fitness value across five different growth conditions \[[@B9]\]. The fitness values for each media condition were calculated as the extent of survival and reproduction of the deletion strain relative to the pool of all strains grown and measured collectively \[[@B10]\]. The number of genes with tandem stop codons in the essential gene group and weak effect group were determined and then compared. We tested whether tandem stop codons tend to follow biologically important genes using a chi-squared independence test. We found that the proportion of genes with a tandem stop codon in the essential gene group is not significantly different from that in the weak fitness effect group (9.2% versus 8.0%, χ^2^= 0.621, P= 0.43, Table [3](#T3){ref-type="table"}). Second, we studied the distribution of the genes with tandem stop codons in different biological processes, biological function categories and biological components, respectively (10 Gene Ontology (GO) biological processes, seven GO functional categories and seven GO biological components; see details in Additional data files 4, 5, 6) \[[@B11]\]. In terms of biological processes, genes with tandem stop codons are over-represented in Metabolism (P\< 0.006) and under-represented in Cell Cycle (P\< 0.01). In terms of biological functions, genes with tandem stop codons are over-represented in Structural Molecular Activity (P\< 0.008). In terms of biological components, genes with tandem stop codons are over-represented in the Cytosol (P= 0) and Cytoplasm (P\< 0.01). These observed biases might intrinsically be explained by the difference in expression levels of these specific groups of genes. Third, we studied the distribution of genes with tandem stop codons among different chromosomes. No distribution bias was detected, indicating that the existence of tandem stop codons extends to the whole genome. Fourth, we examined the distribution of genes with tandem stop codons among transcripts with different lengths. No correlation between tandem stop codon frequency and the length of transcripts could be detected, indicating that the length of a transcript has no influence on the presence of a tandem stop codon. Discussion ========== Our results show that a statistically significant excess of stop codons has evolved at the third codon location downstream of UAA stop codons, designated UAA+3, and that this feature is conserved across five distinct yeast species for which data are available. Comparative analysis between closely related species has demonstrated that these stop codons are more conserved than sense codons at the same location, indicating that the tandem stop codons are maintained by selection. While our results support the long-standing hypothesis that tandem stop codons exist, it raises two crucial questions: (i) why is an excess of stop codons observed only in genes using UAA as stop, and not in genes using UAG and UGA? (ii) why do tandem stop codons evolve mainly at the third codon location after UAA, and not the first or second codon locations? There are several possible answers to the first question. One straightforward explanation is that UAA may be a weak stop codon compared to UAG and UGA, thus requiring a backup stop codon more often. This is not the case: experiments have indicated that UAA is the most efficient termination codon in yeast \[[@B12]\]. Since UAA is the most frequently used stop codon in highly expressed yeast genes \[[@B13]\], another explanation is that tandem stop codons may tend to occur in genes whose products are in high abundance. To test this hypothesis, we analyzed the correlation between tandem stop codon distribution and codon bias. Codon bias reflects the propensity of an organism to utilize selectively certain codons. Several studies have shown that codon bias is a good indicator of protein abundance, because highly expressed proteins generally have high codon bias \[[@B14],[@B15]\]. Therefore, codon bias can be used as an approximation for protein expression level, although the protein abundance of a gene cannot be predicted specifically based on its codon bias alone \[[@B16]\]. We found that the distribution of tandem stop codons strongly correlates with codon bias, indicating that protein expression level is an important factor influencing tandem stop codons. We further considered the relationship between tandem stop codon distribution and fitness effects of genes, based on the comparison of tandem stop codon frequency between the essential gene group and weak-effect gene group, and found no correlation. Selection on tandem stop codons is non-negligible in this study only in the highly expressed genes where translation termination occurs many times during the life cycle of yeast. Thus, regardless of the fitness effects of the gene, it seems that tandem stop codons tend to follow frequently used stop codons in the third downstream codon. As a result, it is not surprising that tandem stop codons follow yeast genes that use UAA as a stop and not those that use UAG or UGA, because only a small number of the highly expressed yeast genes use the latter two codons for termination. Regarding the location of tandem stop codons, it is surprising that tandem stop codons mainly evolve at the third codon location after the real stop codons. The biological reason underlying this observation remains unclear. One possible explanation is that the location of tandem stop codons may be related to the termination context of real stops. Numerous results have indicated that the efficiency of translation termination is influenced by the local context surrounding stop codons \[[@B12],[@B17]\]. Recent studies established that the six nucleotides after the stop codon (corresponding to codon locations +1 and +2 in this study) are key determinants of read-through frequency in yeast, and that these nucleotides can influence read-through efficiency more than 10-fold \[[@B18],[@B19]\]. Thus there may be strong evolutionary constraint on the first two downstream codons to maintain an efficient context for the real stop codon, the primary site for signaling termination. A tandem stop codon may not be favorable at either of these locations, as it may not work well as part of the termination signal. For example, at the first nucleotide following the real UAA stop codon, the most common and the most efficient base for termination is G rather than U \[[@B12],[@B17]\], which precludes this nucleotide from being the first position in a stop codon. This may also explain the under-representation of tandem stop codons at the first codon immediately following the actual stop codon in yeasts (Figure [1](#F1){ref-type="fig"}). The same bias was also observed in other eukaryotes (see details in additional data files 7 and 8). As a backup, a tandem stop codon can reduce the negative effects of translation termination errors and would therefore provide a selective advantage. However, this selection should be very weak since translation termination in general is very efficient, and read-through is very rare, with error levels estimated at 0.3% in yeast \[[@B18]\]. Therefore, if the dominant selection on the first two codon locations (the next six nucleotides) following the real stop codon is to maintain a favorable context for efficient translation termination, rather than to accumulate tandem stop codons as backup, the third codon location may become the main site for tandem stop codons. Tandem stop codons are a relatively subtle regulatory mechanism. While the fitness contribution of a single tandem stop codon may be negligible, the whole effect of over-represented tandem stop codons at the genomic level is probably not. It would be desirable to know whether this (or a similar) mechanism operates in other species. However, the answer is not easy to address at this moment. For most bacteria and archaea, the total number of genes is generally very small. Therefore, it is impossible to perform a similar statistical analysis in these genomes, as the absolute percentage of genes with tandem stop codons is low. Regarding other eukaryotes, we performed the same analysis on Drosophila melanogasterand Caenorhabditis elegansand found one or two over-represented downstream codons (see Additional data files 7 and 8), but the statistical significance is very weak. Here it should be emphasized that our success in identifying tandem stop codons in yeasts lies on three necessary factors. First and foremost, recent comparative analysis has greatly improved the quality of Saccharomycesspecies genome annotation, leading to removal of about 500 spurious genes and modification of about 10% of the annotated boundaries of coding regions \[[@B4]\]. This allowed us to observe a strong signal over background. Second, the availability of several closely related yeast genomes permits identification of conserved signals and further filters out background noise. Third, the influence of local contexts on termination and read-through efficiency in yeasts has been the subject of intensive experimental study, which is not true for most eukaryotes. Together, these factors limit the ability to expand this study at the present time. Conclusion ========== Our study demonstrates for the first time the existence of tandem stop codons at the genomic level - a long-standing and intriguing hypothesis. Our results indicate that protein expression level is an important biological factor influencing the presence of tandem stop codons. We hope that our study of yeasts will provide a model for future examinations of other groups of species. Materials and methods ===================== Gene sequences of four yeast species (S. cerevisiae, S. paradoxus, S. mikataeand S. bayanus) were downloaded from \[[@B20]\]. Spurious genes and genes with ambiguous boundaries were excluded from our analysis. Sequences of C. glabratawere downloaded from GenBank (NC\_005967-005968 and NC\_006026-006036). We used PERL scripts to calculate the frequency of stop codons at the nine codon locations downstream of annotated stop codons. As a control, we used non-coding regions of the same genome: first we looked for occurrence of the trinucleotides TAA, TAG and TGA; then we calculated the frequency of occurrence of these trinucleotides in the same frame at the next nine codons. Statistical significance of the difference between the observed tandem stop codon frequency and the corresponding frequency in the control was determined by chi-squared independence tests. Because we used non-coding DNA sequences of the same genome as a control, factors like single and dinucleotide composition are automatically included in the analysis and do not need to be explicitly considered. Similar analyses were also performed on the other two reading frames for comparison. Because here we studied many codon locations simultaneously and used a less restrictive P-value cutoff (0.05), it is very important to examine the signals in all yeast species. We interpret only the codon locations significantly over-represented in all species as biologically meaningful. Information about one-to-one orthologous gene pairs between S. cerevisiaeand S. bayanuswas extracted from \[[@B4]\]. The orthologous gene pairs using UAA as stop codon were used in the conservation analysis. We first reconstructed the ancestral states of UAA+3 codons using a simple parsimony rule. If two orthologous genes share one identical codon at the UAA+3 location, the codon is assumed to be the ancestral state of the UAA+3 codon. If two orthologous genes have different codons at the UAA+3 codon location, each codon is assumed to be the ancestral state with 0.5 probability. Then we compared the inferred ancestral UAA+3 codon with the UAA+3 codon in each of the orthologous genes and decided whether it is conserved in evolution. The number of conserved stop codons, non-conserved stop codons, conserved sense codons and non-conserved sense codons were calculated, respectively. Statistical significance of the conservation difference between stop codons and sense codons was tested by a chi-squared independence test. Here we used a very strict definition of conserved stop codon, requiring it to be identical between the inferred ancestor and its descendant (modern species). Even with this strict criterion, the P-value is significant. If we relax the definition to allow any stop codon, the result would be even more significant. The same analysis was carried out between S. cerevisiaeand S. paradoxus/S. mikatae, respectively. Codon bias (effective number of codons, ENC) of all the genes using UAA as stop codon in S. cerevisiaewas calculated using the CODONW program \[[@B21]\]. Statistical significance of the proportion of genes with and without tandem stop codons in the genes with the highest 25% ENC values versus the lowest 25% was tested by a chi-squared independence test. The difference in codon bias distribution between these two gene groups (with/without a tandem stop codon) was determined by the Kolmogrov-Smirnov test using MATLAB (version 6.5). Fitness measurements were obtained from a high-throughput study \[[@B9]\] that measured the growth of each strain of a nearly complete collection of yeast single-gene-deletion mutants under five growth conditions. We calculated the fitness values for growth in each medium condition and then classified all genes using UAA as the stop codon into four groups based on these fitness values (f). The calculation of the fitness value and gene classification are the same as in \[[@B10]\]. Statistical significance of the difference in tandem stop codon frequency between the weak-effect gene group (f\> 0.95 for all five media conditions) and the essential gene groups (if the deletion is lethal) was determined by a chi-squared independence test. The study of the distribution of genes with tandem stop codons in different biological processes, biological function categories, and biological components was performed by GO Term Mapper at the SaccharomycesGenome Database \[[@B22]\]. The set of all genes with UAA stop codons was used as a control to determine whether the set of genes containing tandem stop codons is statistically over-represented or under-represented in a specific gene category. Additional data files ===================== The following additional data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} gives the frequency of stop codons at each codon location following the real stop codon in other yeast species. Additional data file [2](#S2){ref-type="supplementary-material"} gives the results for stop codons at each codon location following the real stop codons in all three reading frames in S. cerevisiae. Additional data file [3](#S3){ref-type="supplementary-material"} gives the results of over-represented codon locations in other yeast species. Additional data file [4](#S4){ref-type="supplementary-material"} gives the distribution of genes with a tandem stop codon in different biological processes in S. cerevisiae. Additional data file [5](#S5){ref-type="supplementary-material"} gives the distribution of genes with a tandem stop codon in different biological functional categories in S. cerevisiae. Additional data file [6](#S6){ref-type="supplementary-material"} gives the distribution of genes with a tandem stop codon in different biological components in S. cerevisiae. Additional data file [7](#S7){ref-type="supplementary-material"} gives frequency of stop codons at each codon location following the real stop codons in Drosophila melanogaster. Additional data file [8](#S8){ref-type="supplementary-material"} gives frequency of stop codons at each codon location following the real stop codons in Caenorhabditis elegans. Additional data file [9](#S9){ref-type="supplementary-material"} lists the genes with a tandem stop codon in different yeast species. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table 1, Frequency of stop codons at each codon location following the real stop codons in S. bayanus. Table 2, Frequency of stop codons at each codon location following the real stop codons in S. paradoxus. Table 3, Frequency of stop codons at each codon location following the real stop codons in S. mikatae. Table 4, Frequency of stop codons at each codon location following the real stop codons in C. glabrata. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 In-frame UAA+3 is the only codon location significantly over-represented in all the yeast species. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Because here we studied many codon locations simultaneously and used a less restrictive P-value cutoff, it is very important to further examine the biological signals in other yeast species. Therefore, over-represented locations were examined. Only in frame codon location UAA+3showed the same tendency in all other species. 1. Significant P-values are shown in bold (95% confidence level; p \< 0.05). 2. Under-represented positions are shown as \"n.a.\". ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 The blue bars represent the percentages of genes with a tandem stop codon in each biological process and the red bars represent the controls - the percentages of genes with a TAA stop codon in the corresponding groups. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 The red bars represent the percentages of genes with a tandem stop codon in each biological functional category and the red bars represent the controls - the percentages of genes with a TAA stop codon in the corresponding groups. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 The blue bars represent the percentages of genes with a tandem stop codon in each biological component and the red bars represent the controls - the percentages of genes with a TAA stop codon in the corresponding groups. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 The red bars represent the frequency of stop codons at each codon location and the blue bars represent the controls - the frequency of stop codons at the corresponding locations downstream of stop codons in non-coding regions. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 The red bars represent the frequency of stop codons at each codon location and the blue bars represent the controls - the frequency of stop codons at the corresponding locations downstream of stop codons in non-coding regions. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 9 1\. List of genes with a tandem stop codon in S. cerevisiae2. List of genes with a tandem stop codon in S. bayanus3. List of genes with a tandem stop codon in S. mikatae4. List of genes with a tandem stop codon in S. paradoxus5. List of genes with a tandem stop codon in C. glabrata ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Zhenglong Gu, University of Chicago, for kindly providing the calculated fitness values. We also thank two anonymous referees for valuable suggestions. This work was supported by National Institute of General Medical Sciences grant GM59708 and National Science Foundation grant DBI-9875184 to L.F.L. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Frequency of stop codons at the downstream codon locations following the real stop codons (which occur at codon location \'0\') in S. cerevisiae. (a)TAA; (b)TAG; (c)TGA. The red bars represent the frequency of stop codons at each codon location and the blue bars represent the controls - the frequency of stop codons at the corresponding locations downstream of stop codons in non-coding regions (nc). (d)Statistical significance of stop codon frequency at each location following each stop codon. For each codon location in the analysis, the statistical significance (using χ^2^tests) of the difference between the frequencies of tandem stop codons in the yeast genes and in the control sequences is shown. The blue bars represent codon locations following the stop codon TAA; the red bars represent codon locations following the stop codon TAG; the yellow bars represent codon locations following the stop codon TAA. (e)The identity distribution of three stop codons at the UAA+3 codon location. Blue represents the percentage of TAA; red represents TAG; yellow, TGA. ::: ![](gb-2005-6-4-r31-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Statistical significance of the frequency of tandem stop codons. Statistical significance of the frequency of tandem stop codons following a TAA stop codon in the first nine codon locations in four Saccharomycesspecies is shown. ::: ![](gb-2005-6-4-r31-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Distribution of tandem stop codons in different codon bias groups in S. cerevisiae. The blue bars represent the proportion of genes with a tandem stop codon; the red bars represent the proportion of genes without a tandem stop codon. ::: ![](gb-2005-6-4-r31-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Frequency of different codon groups at the UAA+3 location in the conservation analysis ::: Stop codon Sense codon Total --------------- ------------ ------------- ------- Conserved 138 1,089 1,227 Non-conserved 66 765 831 Total 204 1,854 2,058 χ^2^= 6.1; P\< 0.01. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The influence of codon bias on the presence of tandem stop codons ::: Highest codon bias quartile Lowest codon bias quartile Total ----------------------- ----------------------------- ---------------------------- ------- With a tandem stop 87 33 120 Without a tandem stop 480 534 1,014 Total 567 567 1,134 χ^2^= 27.2; P\< 2 × 10^-7^. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The influence of fitness effect of the genes on the presence of tandem stop codons ::: Essential genes Weak-effect genes Total ------------------------- ----------------- ------------------- ------- With the tandem stop 38 97 135 Without the tandem stop 374 1,118 1,492 Total 412 1,215 1,627 χ^2^= 0.621; P= 0.43. :::	PubMed Central	2024-06-05T03:55:56.347647	2005-3-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088959/", "journal": "Genome Biol. 2005 Mar 15; 6(4):R31", "authors": [ { "first": "Han", "last": "Liang" }, { "first": "Andre RO", "last": "Cavalcanti" }, { "first": "Laura F", "last": "Landweber" } ] }
PMC1088960	Background ========== Transcription of mRNA from DNA and subsequent translation of mRNA into protein transform genetic blueprints into cellular functions. This process of gene expression and regulation plays a key role in determining the fitness of the genome, through the production of different proteins in different cells and at different times. Therefore, in addition to genome composition and structure, regulation of gene expression is also a key component in development and evolution \[[@B1]\]. The importance of regulatory genes during evolution is well recognized \[[@B2]\]. For example, major differences in axial morphology consistently correlate with a difference in spatial regulation of Hoxgene expression \[[@B3],[@B4]\]. In addition, a cis-regulatory element has functionally diverged during the course of bird and mammal evolution and has resulted in different gene-expression patterns between these two taxa \[[@B3],[@B4]\]. Recently, many studies have suggested that cis-regulatory regions of regulatory genes and their downstream target genes might be a major driving force behind evolutionary changes in humans \[[@B5]\]. In plants, evidence for the importance of variations in upstream regulatory regions in the evolution of plant form have also been described. Polymorphisms in an upstream regulatory region of the teosinte branched1gene have been implicated in the domestication of maize \[[@B6]\], and changes in the promoter region of ORFXmay associate with increases in fruit size during tomato domestication \[[@B7],[@B8]\]. Despite its potential importance, the genetic basis of cis-regulatory evolution is poorly understood. Stone and Wray \[[@B1]\] suggested the following reasons: first, the lack of information on sequence variations in the regulatory regions, and lack of association between the degree of coding sequence divergence and the change in gene expression \[[@B9]\]; second, the lack of experimental data from gene-expression analyses to support sequence variation analyses; and third, the lack of a conceptual framework for understanding regulatory evolution that could guide empirical studies. Therefore, to better understand cis-regulatory evolution and its implications for genome stability and dynamics, an essential step is to identify sequence variations in the regulatory regions of regulatory genes and downstream target genes on a genome-wide scale, and establish the correlations between gene-expression variations and regulatory sequence divergence. However, few studies have attempted to correlate molecular studies of the evolution of cis-regulatory genotype with that of phenotype \[[@B10]\]. Naturally occurring phenotypic differences such as leaf shape or biomass among different Arabidopsisaccessions \[[@B11]\] have recently become used as resources to study gene function, which traditionally has been studied through mutagenesis and phenotypic characterization of genetic variants \[[@B12]\]. Differences in transcriptional regulation have the potential to contribute substantially to such phenotypic differences among accessions. Thus, it is important to understand the extent to which evolutionary differences between accessions are the result of regulatory polymorphisms causing alterations in transcription, as opposed to coding-region polymorphisms that alter the function of gene products. Although transcriptional profiling has been applied to study the transcriptome differences within or among species using both Affymetrix oligonucleotide GeneChip microarrays and cDNA microarrays \[[@B13]-[@B15]\], a recent study from Hsieh et al. \[[@B16]\] showed a strong species-by-probe interaction effect when using Affymetrix GeneChip microarray for such inter-species transcriptome analysis. Species differences in hybridization signal strength from a probe set can reflect both sequence differences between probes and their hybridizing targets, and differences in abundance of the mRNA. Therefore, comparative transcriptome analysis of different species or accessions is difficult to interpret without controlling for the effect of coding DNA polymorphism before assaying for differences in transcript abundance. The objectives of this study are to develop a reliable method for comparing transcriptomes among samples with different genetic backgrounds, to identify differences in transcriptomes among different genetic lines, and to understand the regulatory mechanisms responsible for gene-expression differences by analyzing their predicted promoters. To accomplish these goals, we have adopted a new analysis strategy to analyze the transcriptome variations in five Arabidopsisaccessions. Our results suggest that genes with functions involved in signal transduction, transcription and stress response are the primary targets for natural selection. This study should shed light on the field of plant evolutionary genomics by furthering our understanding of how the two-way evolutionary interactions between genomic polymorphisms and transcriptional regulatory mechanisms contribute to shaping the evolution of genome. Results ======= Strategy for comparing gene expression among accessions ------------------------------------------------------- The GeneChip microarray used in this study contains approximately 8,700 probe sets for 8,300 Arabidopsisgenes, which covers about one-third of the genome of accession Col-0 (ecotype Columbia) \[[@B17]\]. Both perfect match (PM) and mismatch probes of the majority of the probe sets on this GeneChip microarray are able to cross-hybridize to genomic targets from other accessions; however, the hybridization signals are affected by any sequence polymorphisms between the probes and the targets \[[@B18]\]. With the standard Affymetrix algorithms (MAS4.0 or MAS5.0) polymorphisms between the hybridizing mRNA samples are likely to invalidate the assumptions underlying the perfect-match mismatch signal subtraction step, leading to inaccurate measurements of the transcript levels, and thus preventing accurate comparisons of the transcriptomes among different accessions. To address these issues, we selected for the comparative transcriptome analysis PM probes that hybridize similarly to the genomic targets of test accessions (Figure [1](#F1){ref-type="fig"}). Briefly, genomic DNAs from different accessions were fragmented, labeled and hybridized to the ArabidopsisGeneChip microarrays \[[@B19]\]. The hybridization signals from the PM probes were summarized into genomic DNA hybridization indices (gDHI) using the PM-only model \[[@B20]\] to avoid the complication of the array mismatch probes. The coefficient of variance (CV) of the gDHI among the five accessions used in this study for each probe set was used to determine whether there was sufficient genomic sequence difference among the different accessions to substantially alter hybridization to the oligonucleotide probes. Probe sets were ranked on the basis of their CV and those with the largest CV (CV ≥ 0.20) were eliminated (see Additional data files 1 and 8). The cutoff value was chosen on the basis of the overall mean and standard deviation of the CV from genomic DNA hybridization (mean + standard deviation). For the further comparative transcriptome analysis, 7,736 probe sets with CV less than 0.20 were selected. To measure the consistency of our probe set selection in this procedure, the reproducibility of the comparative genomic hybridization experiments was determined by labeling and hybridizing the same genomic DNA onto two different microarrays in parallel. The results were highly reproducible and only a small fraction of genes showed twofold or greater difference in hybridization signals between the two replicated experiments: 0.1% between the Col-0 replicates, 0.02% between the Lerreplicates, 0.2% between the C24 replicates, 0.01% between the NO-0 replicates, and 0% between the WS-2 replicates. These results are consistent with the average reproducibility for other genomic DNA labeling and hybridization experiments in Arabidopsis, and similar to the results from reproducibility studies for RNA detection using the same GeneChip microarray \[[@B17]\]. Comparative analysis of transcriptome of different accessions and its validation -------------------------------------------------------------------------------- Transcription profiles of different organs at different developmental stages (see Additional data file 2) were compared among the five accessions using the following strategy. First, the PM-only model was used to estimate the raw RNA hybridization index (rRHI), to reduce the complication of the array mismatch probes. Second, gDHIs were used to normalize rRHI to remove contributions from sequence variations due to undetected single feature polymorphisms (SFPs) in probe sets. The normalized RNA hybridization index (nRHI), calculated by dividing the rRHI of each probe set by the corresponding gDHI of a particular accession, is used to represent the relative transcript level of the target gene. Third, all the genes were ranked on the basis of their nRHI values, and the lowest 5% were chosen as the cutoff value for background. Genes with an nRHI value less than the cutoff value across all the RNA samples from at least one accession were eliminated from further analysis. By this method, genes whose transcripts could not be detected or were close to the background level were excluded. Fourth, the nRHI values of the 7,508 genes after step 3 were used for statistical analyses, for calculating the Pearson correlation coefficient between all possible pairs of accessions (10 pairs from pairwise comparison of five different accessions) for each gene, and for cluster analysis \[[@B21]\]. To validate variations in transcript abundance detected by the GeneChip microarray through heterologous hybridization using our strategy, quantitative reverse transcription PCR (RT-PCR) using accession-specific primers and probes was performed. Table [1](#T1){ref-type="table"} compares nRHI of 13903\_at (At3g54050) and 17392\_s\_at (At3g53260), measured by the GeneChip microarray and the quantitative RT-PCR in 18 different samples. In general, the quantitative RT-PCR results agreed with the GeneChip microarray results, and confirmed the expression differences of these two genes between accessions Col-0 and C-24. The correlation coefficient between the results of the GeneChip microarray and quantitative RT-PCR is 0.93 for 13903\_at, and 0.82 for 17392\_s at. As expected, those probe sets with probes cross-hybridizing with genes in a family, such as 17392\_s\_at, correlated less strongly with accession-specific quantitative RT-PCR. In addition, nRHI of 12 randomly selected genes with various expression patterns was also validated by quantitative RT-PCR. Some of them did not show different expression levels, and others did show a difference between the flowers of Col-0 and those of Ler. As shown in Table [2](#T2){ref-type="table"}, the results from the quantitative RT-PCR analysis were generally consistent with the nRHI regarding the trend of the change for each gene between Col-0 flower and Lerflower. There are two exceptions (16892\_at and 20545\_at), which showed slightly reduced expression in Lerflower as compared to Col-0 from the GeneChip microarray experiments, but showed an opposite trend of expression from Taqman data. In addition there are a few examples (14172\_at and 17860\_at), which showed a less than twofold difference from the GeneChip microarray experiments, but slightly higher than twofold differences (14172\_at: 2.05-fold, 17860\_at: 2.26-fold) from RT-PCR. The slight inconsistency between the GeneChip microarray results and the RT-PCR results may result from the difference in detection technology, and associated sensitivities, between the two methods. It also indicates that definition of significance using twofold change is not appropriate for this experiment. Nevertheless, the results from this extensive validation study using accession-specific primers and probes support our analysis strategy used for transcription analysis of different accessions in both sensitivity and specificity aspects. To assess the residual interference from sequence variations between targets and probes within the probe sets used for comparative transcriptome analysis, for each sample, we compared the overall transcriptome profiles by calculating Pearson correlation coefficient between rRHI and nRHI for selected probe sets and all probe sets including those probe sets detecting significant difference in genomic hybridization. A general consistency for each sample was observed (see Additional data files 3 and 9). However, the inclusion of the probe sets detecting difference in genomic hybridization reduces the Pearson correlation coefficients between rRHI and nRHI (see Additional data file 3), demonstrating a greater degree of interference from sequence variation in those probe sets. Data from Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"} also showed examples of high correlation between the rRHI and nRHI. When these data were compared to the data from accession-specific quantitative RT-PCR, the correlation coefficients were slightly different: 0.92 (rRHI) and 0.93 (nRHI) for 13903\_at, and 0.80 (rRHI) and 0.82 (nRHI) for 17392\_at. These results indicate that the probe sets selected for the comparative transcriptome analysis have a low level of interference, and can be utilized to measure the transcript abundance in the five accessions. General similarities of transcriptional profiles among accessions from various organs at different stages --------------------------------------------------------------------------------------------------------- As shown in Table [3](#T3){ref-type="table"} and Figure [2](#F2){ref-type="fig"}, among the 7,508 genes whose expression was above the cutoff value in at least one of the RNA samples, the expression patterns of most of the genes (5,985) were correlated (r\> 0.5) in at least five pairwise comparisons (gray bars), indicating that the expression patterns for most genes from different accessions share some similarity. To test whether the high correlation in expression patterns among different accessions was likely to be obtained by chance, we randomly permuted the RNA samples from the same organs of five different accessions (see Materials and methods for details). The number of genes whose expression did not correlate at r\> 0.5 for any pair of accession comparisons increased significantly (Figure [2](#F2){ref-type="fig"}, white bars) from a total of 65 in the original data to 130 (group 10 in Figure [2](#F2){ref-type="fig"}), and the number of genes whose expression did correlate for all pairs of accession comparisons decreased significantly, from 3,532 in the original data to 1,266 in the permuted data. Because of the close relationship of the five accessions chosen in this study, these data suggest, as expected, that the tissue-specific gene-expression patterns are more consistent between accessions of a single species than any accession-specific patterns between organs. We used by cluster analysis of the nRHI data to further analyze relationships among the accessions on the basis of the transcriptome profiles (Figure [3](#F3){ref-type="fig"}). The overall relationships among all samples confirmed that the expression differences among the accessions were small, as the gene-expression differences were greater across different organs of the same accession than that across different accessions in the same organ (Figure [3](#F3){ref-type="fig"}). Two clusters emerge from the experimental tree: a cluster of axis-origin organs, including roots and young seedlings, and a cluster of auxiliary organs, including vegetative leaves, flowers and siliques (reproductive leaves) and the associated inflorescences (Figure [3](#F3){ref-type="fig"}). The axis cluster consisted of roots from two different developmental stages - 2 weeks and 5 weeks - as well as 4-day-old seedlings, which are mainly composed of root tissues. The cluster of auxiliary organs could be further divided into two subclusters, one for the vegetative leaves, and one composed of organs originating from the reproductive leaves. Within an organ, especially for leaves, however, variations were contributed by both developmental differences and accession differences. These relationships, as illustrated in Figure [3](#F3){ref-type="fig"}, were supported by bootstrap analysis \[[@B22]\]. One hundred datasets, each containing the same number of genes, were generated from the original dataset by random sampling with replacement. The bootstrap results confirmed the robustness of the cluster results at the top two levels of the dendrogram (Figure [3](#F3){ref-type="fig"}). Accession-specific gene expression during development ----------------------------------------------------- Although in general, the gene-expression patterns from the same organs of different accessions were similar, the correlation tends to get worse towards late development (Figure [4](#F4){ref-type="fig"}). The differences observed among the five accessions in late development could be due to the following reasons: biological noise (individual variation) within each accession during the sampling of biological materials; developmental differences among different accessions; and accession-specific differences due to default regulatory programming. It is unlikely that the differences are due to the sampling noise, as these noises will become undetectable by extensive pooling of biological materials in this study. The phenotypic differences, especially during late plant development, such as leaf shape, size and flowering time, prompted us to search for genes whose expression is different among different accessions. To identify genes that represent accession-specific difference, and to differentiate them from the genes which could possibly reflect the developmental differences of these five accession plants at the same age grown under the same conditions, we used the one-way analysis of variance (ANOVA) to analyze nRHI data of 2-, 5-, and 11-week-old leaves from the five accessions. Here we treated samples from 2-, 5-, and 11-week-old leaves as three leaf replicates for each accession, thus the only factor we are analyzing is \'accession\' which has five levels in this study (see Additional data file 4). On the basis of ANOVA, 1,525 genes were found to have p-values less than 0.01 (false discovery rate or FDR = (7,508 × 0.01)/1,525 = 4.9%). Bonferroni correction was further applied for the strong control of family-wise type I error rate (FWER). As shown in Table [4](#T4){ref-type="table"}, 58 genes were thus selected, which potentially represent the genes with differential expression among the leaves from the five accessions (p\< 0.05). These genes were then functionally classified according to the Munich Information Centre for Protein Sequences (MIPS) functional classification. As shown in Figure [5](#F5){ref-type="fig"}, these 58 genes encode products with diverse functions. Besides those proteins with unknown function, the top five categories contained genes with possible functions in transcription (18% vs 9% for all the genes on the chip), subcellular localization (18% vs 11% overall), stress/defense response (15% vs 6% overall), metabolism (9% vs 18% overall) and signal transduction (9% vs 9% overall). Compared to the overall distribution for all the genes on the chip among different functional categories, genes involved in transcription, subcellular localization and stress/defense response are enriched in this group (p≤ 0.008, p≤ 0.018, and p≤ 0.004, respectively). Eight genes encoding putative transcriptional regulators, including Dof zinc-finger transcription factors, HD-zip transcription factor Athb-8, and MADS-box containing proteins, were included within this group of 58 genes. Genes involved in stress/defense responses include ones that encode disease-resistant proteins such as those of the TIR-NBS-LRR class, enzymes involved in secondary metabolism, and proteins involved in detoxification. Organ-specific gene expression in different accessions ------------------------------------------------------ In addition to identifying accession-specific genes, we were also interested in determining if there were genes whose expression is regulated by accession-by-organ interaction. In other words, we tried to test if the accession effect on gene expression is organ/development dependent. To address this question, two-way ANOVA analysis was performed. In one case, two samples from 2- and 5-week-old leaves, and two samples from 2- and 5-week-old roots were treated as replicates. In this two-way ANOVA study, the two factors are \'accessions\' and \'organs\'. For the \'accession\' factor, there are five levels. For the \'organ\' factors, there are eight levels (see Additional data file 4). The total mean squares for all the genes due to organ difference was 13,182.91 (df = 7), much greater than the total mean squares due to accession difference, which was equal to 2,936.21 (df = 4), consistent with our previous observation from the cluster analysis (Figure [3](#F3){ref-type="fig"}). The total mean square due to accession-by-organ interaction was only 436.00 (df = 28), suggesting that the effect of accession-by-organ interaction on gene expression might be small. Among the 296 genes that were found to have p-values less than 0.01 (FDR = 25.36%), 60 were further selected following Bonferroni correction to control the type-I error rate (Table [5](#T5){ref-type="table"}), and subjected to functional classification. As shown in Figure [6](#F6){ref-type="fig"}, the top five categories contained genes with possible functions in plant development/embryonic development, metabolism, seed storage, stress/defense response and biogenesis of cellular components such as cell walls. Compared to the overall distribution for all the genes on the array among different functional categories, genes involved in plant development/embryonic development and in seed storage are enriched in this group (p≤ 0.001 for both categories), suggesting that the differential gene expression in different accession backgrounds might be more profound during late plant development. In contrast to a higher percentage of genes encoding transcription factors, which are differentially expressed in leaves of different accessions, much fewer such genes were found in this group. Genes with expression patterns that vary greatly among accessions ----------------------------------------------------------------- For each gene, the expression pattern reflects the relative abundance of its mRNA in different RNA samples, which is determined by a combination of environmental and developmental factors. Thus the differences in gene-expression patterns from different accessions reflect the different responses of each accession to these factors. To identify genes whose expression is highly sensitive to various environmental and developmental stimuli, and to further understand the differential regulatory mechanisms among accessions, genes with distinct expression patterns in different accessions were identified by their correlation coefficients between every two accessions in the Pearson correlation coefficient matrix (Figure [2](#F2){ref-type="fig"}), using 10 data points from the corresponding 10 organs of each accession (see Additional data file 5 for an example). Of these, 65 genes had correlation coefficients less than 0.5 in all 10 pairs of accession comparisons (Table [6](#T6){ref-type="table"}), 271 genes had correlation coefficients less than 0.5 for nine pairs of comparisons, and 376 genes had correlation coefficients less than 0.5 for eight pairs of comparisons (Figure [2](#F2){ref-type="fig"}). As shown in Figure [7](#F7){ref-type="fig"}, genes belonging to functional categories of signal transduction, transcription, subcellular localization, stress/defense response and protein fate (folding, modification, destination) are among the top five functional categories in this group, whereas the proportion of genes belonging to the transcription functional category is slightly higher (13% for this group and 9% for the overall group). Genes involved in transcription included different types of transcription factor genes, such as bHLH, EREBP-like, and several zinc-finger transcription factor genes. Genes whose products are required for other functions related to the control of mRNA level, such as chromatin remodeling or RNA processing (for example, the mRNA capping enzyme and the chromatin-remodeling factor CHD3 (PICKLE)) were also included in this group (Table [6](#T6){ref-type="table"}). The stress-responsive genes included those for the putative heat-shock protein DnaJ and the α-jacalin-like lectin, a relative of which has been shown to be salt-stress-inducible in rice \[[@B23]\]. A number of genes, whose products are protein kinases and are likely to be involved in cell signaling pathways, were also included in this 65-gene list. Regulatory sequence polymorphisms could account for the gene-expression differences among accessions ---------------------------------------------------------------------------------------------------- To test whether the accession-dependent differences we observed were caused by polymorphisms in regulatory sequence, we sequenced the promoters and coding regions of seven genes selected from genes with Pearson correlation coefficients less than 0.5 in at least five pairwise comparisons among the five accessions discussed here (plus seven additional accessions, RLD-1, Ag-0, Bs-1, Cvi-0, Es-0, Gr-1, Mt-0 and Tsu-0, to obtain a better estimate of relative substitution rates). We identified a total of 167 polymorphic bases in one or more of the five accessions (316 in all 12) across 24.9 kilobases (kb) of promoter and coding sequence. The polymorphism rate among all five accessions in regulatory (promoter) sequence was 8.06 per kilobase, compared to 10.5 per kilobase in introns and 4.08 in exon sequence (Table [7](#T7){ref-type="table"}), indicating that regulatory sequence is the repository for substantially more genetic variation than coding sequence. Details of these polymorphisms are described in Additional data file 6. We then analyzed the promoter sequences of the seven genes selected for further study of sequences matching known plant cis-regulatory elements (see Materials and methods) to determine whether any of the polymorphisms altered sequences corresponding to known cis-regulatory motifs in the promoters. We found that a total of 44 out of the 61 polymorphisms among the seven genes fully sequenced in the five accessions caused alterations in sequences that matched known cis-regulatory motifs (details of all these changes are provided in Additional data file 6). For example, the putative RING-finger protein At4g10160 is one of three genes encoding proteins in this family that we resequenced in the target accessions. In Col-0, the promoter of At4g10160 contains a CAACA element at -164, which is absent in all other accessions as the result of a sequence polymorphism. This element is the binding site for the transcription factor RAV1. RAV1 belongs to the AP2/EREBP transcription factor family, members of which are involved in various aspects of plant development as well as in plant response to environmental stresses \[[@B24]\]. When the expression profiles of this gene were considered, the lowest three correlation coefficients between any of the pairs of accessions were those between Col, Ws, No-0 and Ler(r= -0.045, -0.168 and 0.201 between the pairs Col/C24, Ler/WS and Ler/No-0, respectively). Not all of the transcription difference is associated with altered known cis-elements. For instance, the gene for the PHYB photoreceptor, At2g18790, was also differentially expressed among accessions. There were several polymorphisms in the promoter sequence, most of which were specific to the Ws accession (a natural mutant in another phytochrome gene, PHYD\[[@B25]\]). These polymorphisms included two mutations that both altered cis-regulatory elements (AAAGAA to ATAGAA at -965, and GGTTTATT to GCTTTATT at -445) known to be involved in the regulation of another phytochrome gene \[[@B26]\]. These polymorphisms could not fully account for the different expression patterns, however, as the Col-0 expression pattern correlated quite well to that for Ws (r =0.78), whereas the Ler/Ws pair correlated very poorly (r =0.207). The correlation between Col-0 and C24 was only r =0.341. Because Col-0 and C24 had identical sequence throughout the PHYBpromoter, the difference in expression patterns must be at least partly explained by other factors, such as polymorphisms in enhancers outside the resequenced region, or polymorphisms in the genes encoding regulatory factors that control PHYBmRNA levels. Discussion ========== A number of interspecies or interaccession comparative analyses of transcriptomes using GeneChip microarrays have been attempted recently. Brem et al. \[[@B27]\] conducted a study in yeast to understand the genetic architecture of natural variation in gene expression using GeneChip microarrays. By comparing the transcriptomes of two yeast strains, the study linked 570 differentially expressed genes between the two parental yeast strains to one or more genetic markers, and further grouped these genes into two categories, the cis-acting modulators and trans-acting modulators. More recently, two laboratories independently used the ArabidopsisGeneChip microarrays to detect transcriptional changes in metal homeostasis genes of A. halleri, a closely related species to A. thalianaand a natural metal hyperaccumulator \[[@B28],[@B29]\]. These studies successfully demonstrated the potentials of GeneChip microarrays in the studies of biodiversity among Arabidopsisaccessions and the closely related species, as supported by extensive validations from real-time RT-PCR, and RNA blot experiments. However, these studies were limited to those genes whose mRNAs were expressed at high levels, as they used stringent selection criteria. In addition, the signal differences contributed by the sequence variations between the two species or lines were largely unaddressed. To apply GeneChip microarrays developed for a model species to monitor transcription in other related accessions or species, and to enable the comparisons of transcriptomes among closely related accessions or species with genetic variations, we developed a new strategy for analyzing transcriptome profiles from GeneChip experiments by heterologous probe-target hybridization (Figure [1](#F1){ref-type="fig"}). To minimize the interference from detectable sequence variations between probes selected from one accession and targets from another accession, we identified and selected those probe sets that hybridize similarly to genomic targets from different accessions, and excluded the ones which showed significant difference in their hybridization signals for further analysis. We analyzed the data at the probe set levels using Li Wong\'s PM-only model, as this algorithm takes probe effect into consideration by proper modeling and summarization of probe-level data into probe set indices \[[@B30]\]. We did not perform our analysis at the probe level, because, first, there are substantial single feature polymorphisms (SFPs) among Arabidopsisaccessions, as demonstrated between Col-0 and Ler \[[@B18]\]. If we remove all the probes with SFPs, it will reduce the number of available probes in a probe set, thus compromising the quality of the measurements. Second, comprehensive detection of SFPs is not within the scope of this study. The high correlations observed between the rRHI and nRHI suggest those residual sequence variations between probes and targets from different accessions did not substantially affect the comparisons between mRNA level in the different accessions. Only 986 probe sets (out of 8,722 probe sets) showed substantial difference in genomic DNA hybridization signals from the genomes of the five accessions we investigated (see Additional data file 1). These probe sets, representing the genes with high polymorphism rates, were functionally categorized, and were consistent with the results obtained by the previous study where a number of ArabidopsisSFPs were identified by large-scale comparative genome analysis \[[@B18]\]. For example, among the 127 transposon related genes presented on the array, 88 of them were detected as polymorphic among the five accessions. The molecular mechanism that underlies this observation was not clear, although reduced selection pressure for sequence conservation between transposable elements, combined with the mutations that can result from transposition events, may lead to a higher polymorphism rate. Transposable elements are likely to play an important role in shaping the plant genome \[[@B31]\]. In addition to transposon-related genes, genes encoding disease-resistance proteins and kinases were also found to contain SFPs among different accessions. The specificity of the GeneChip microarray detection was validated experimentally by other methods such as real-time quantitative RT-PCR, using accession-specific primers and probes. Genes for the RT-PCR experiments were selected so that various transcript levels, and various expression patterns during development, were represented, based on the microarray analysis results. The general agreement between the results from GeneChip and the quantitative RT-PCR measurements demonstrate the specificity of the detection in different accessions. Overall, the transcriptome profiles are relatively consistent during development among the Arabidopsisaccessions studied. This is supported by the high degree of Pearson correlation coefficients for each expressed gene from every possible pair of compared accessions. It was also supported by cluster analysis of samples from different organs among the five accessions. Seventy-nine percent of the analyzed genes have correlation coefficients greater than 0.5 in at least five pairs of accessions (Figure [2](#F2){ref-type="fig"}). Interestingly, similarity in gene expression is not consistent with the similarities in the coding sequence among different accessions. Among the pairwise accession comparisons, we found that the C24/Lerpair contained the fewest genes whose expressions did not correlate (data not shown). However, this finding was not consistent with the cluster results based on the coding sequence variations, in which the closest accession to C24 was Col (data not shown). This suggests that transcriptional regulation has a significant role in determining natural variations in gene expression, and there might be more difference in gene-regulation mechanisms between C24 and Col-0 than is suggested by the relative similarity of their genomic sequence. The divergence in transcriptomes and their regulatory mechanism in different accessions become evident from the results of the ANOVA analysis of transcriptomes of 2-, 5- and 11-week-old leaves from the five accessions. It was found that 58 genes showed a statistical difference (p\< 0.05 after Bonferroni correction) in expression among different accessions, and a higher percentage of these differentially expressed genes encode products in transcriptional regulation, and stress responsive proteins (Figure [5](#F5){ref-type="fig"}, Table [4](#T4){ref-type="table"}). The differences in gene expression in leaves of the five accessions are mainly due to the accession differences, because for those genes the differences at different developmental stages of leaves in each accession are not statistically significant compared with the differences among the five accessions. Although we could not correlate the gene-expression difference with any previous reports on these particular accessions, our data suggest that the differential expression of these genes could reflect adaptive responses to the environmental conditions used in this study. It will be interesting to map these genes to their genetic locations to test if any have been previously linked to quantitative trait loci, thus affecting the phenotypes among different accessions. The accession differences in transcriptome programming become more obvious towards late development in an organ-specific manner. Sixty genes whose expression might be affected by accession-by-organ interaction during late development were identified. The top five functional categories contained about 71% of genes whose products might be involved in nutrient storage, stress response and plant, especially reproductive, development (Figure [6](#F6){ref-type="fig"}). As shown in Additional data file 7, the expression of the majority of these genes differed in senescent leaves and mature siliques, suggesting that the transcriptome programs in these organs are more sensitive to different accession backgrounds at late stages, leading to the differential expression of genes involved in late plant development. We could not, however, rule out the possibility that some of these genes might represent the differences in developmental stages for the five accessions around the sample collection time. To further elucidate regulatory mechanisms that are important for the differential gene expression among different accessions, we have identified 65 genes that showed different expression patterns in the five accessions during development by analyzing the Pearson correlation coefficients from the 10 pairs of compared accessions (Figure [2](#F2){ref-type="fig"}). The 65 most plastic genes are predominantly those that function in transcription and in stress and defense responses (Figure [7](#F7){ref-type="fig"}). It has been shown that the expression of many transcription factor genes is sensitive to changes in environmental conditions \[[@B32],[@B33]\]. By examining the expression patterns of these most plastic genes under various environmental conditions \[[@B30]\], such as biotic or abiotic treatments, we found that the expression of a majority of the genes was induced or repressed by various environmental factors, demonstrating their high responsiveness to environmental conditions. These findings suggest that regulatory genes are major targets of natural selection \[[@B34]\], because changes in both the protein structure encoded and gene expression of a limited number of transcription factor genes would result in dramatic phenotypic variations via changes in expression of a large number of downstream genes. The differences in expression of these genes could arise from multiple mechanisms, such as changes in expression or activity of trans-acting regulators, changes in the cis-regulatory regions of the corresponding genes, or even epigenetic modification. Previous studies have shown that both regulatory genes and gene promoter regions are subject to selective forces \[[@B34]\] and that promoters are the primary targets of adaptive evolution relative to coding regions \[[@B35]\]. Here we present one such example, At4g10160, which encodes a RING-finger protein. The change in one of the predicted cis-elements in the promoter of this gene was consistent with the changes in gene expression. This finding is of particular interest as RING-finger proteins are known to be capable of regulating gene expression and altering developmental patterns and cell proliferation \[[@B36],[@B37]\]. Although this finding requires more experimental validation, it represents a clear example of differential gene-expression mechanisms among different accessions. It is recognized, however, that not all the differences in accession-dependent transcription can be explained by regulatory polymorphisms. The difference in PHYBexpression between C24 and Col-0 illustrates the complexity of the regulatory mechanism involved in the adaptation of the transcriptome programs. Changes in expression of this gene might be influenced by other factors, such as alterations in the regulatory sequences of genes encoding controlling factors, for example the RING-finger proteins discussed above. Conclusion ========== Using a GeneChip microarray and a strategy validated experimentally by accession-specific quantitative PCR, we compared the transcriptomes of five Arabidopsisaccessions under identical growth conditions. The detected variations in gene expression among different Arabidopsisaccessions may be caused by a combination of variations in trans-acting factors, or in promoter regions of the variable genes themselves. Using the approach of comparative transcriptome profiling of different accessions, combined with genome sequence information, it is possible to identify polymorphisms putatively associated with the accession-dependent gene-expression patterns, and to link these polymorphisms to the differential expression of genes encoding components of regulatory mechanisms. Mutations of such global consequence are highly likely to have been subject to intense selective pressure during evolution. This could further help in understanding genome and transcriptome dynamics during evolution \[[@B38]\], suggesting that natural selection must not simply act through constantly evaluating the fitness of existing DNA within the genome on a gene-by-gene basis, but also by strongly favoring advantageous polymorphic gene-regulatory mechanisms which arise as a result of rare, but highly significant, genomic mutations that alter the expression patterns of large clusters of genes. Moreover, because phenotypic variation among different accessions probably reflects genetic variation that is important for the plant\'s adaptation to specific environmental conditions, transcriptome analysis, as a powerful tool for molecular phenotyping, should provide a complementary approach to quantitative trait locus (QTL) analysis for studying the interaction between genetic variation and environment. A potential application of this approach to crop breeding is to identify key regulatory mutations conferring desirable, yet highly pleiotropic, traits in commercial cultivars. Regulatory polymorphisms responsible for these variations may then be readily transferred between cultivars as monogenic traits. Materials and methods ===================== Plant materials, growth conditions and sample processing -------------------------------------------------------- Seeds from the five Arabidopsisaccessions Col-0 (Columbia), C24, WS-2, NO-0, and Ler(Landsberg erecta) were obtained from the Arabidopsisstock center (ABRC, Columbus, Ohio). Seeds were geminated in Metro-Mix soil (Scotts-Sierra Horticultural Products) in flats and were grown in controlled-environment chambers CMP4030 (Conviron, Winnipeg, Canada) at 22°C under a 12-hr/12-hr light/dark regime and 80% humidity. Plants received approximately 350 μmol s^-1^m^-2^of light from two light banks emitting 15.069 lux or 45.2 W m^-2^. Ten different RNA samples from 10 different organ samples, including roots, leaves, flowers and siliques, were collected at different plant ages from each accession (Additional data file 2). All samples were collected from at least 10 individual plants between 11 am and 1 pm and were pooled. RNA was extracted from various organs, which were collected. Genomic DNA was extracted from the 4-week-old leaves. DNase I digestion was used to obtain genomic DNA fragments with average sizes ranging from 25 to 150 nucleotides. DNA fragments were end-labeled using terminal transferase according to Winzeler et al. \[[@B19]\]. The ArabidopsisGenome GeneChip array (Affymetrix) was used for this study. Details of array features and performance were described previously \[[@B15]\]. The RNA extraction and GeneChip microarray experiments were exactly performed as described by Zhu et al. \[[@B39]\]. Dataset collection, data processing and data analyses ----------------------------------------------------- The microarray experiments on genomic DNA hybridization were conducted in replicates for all accessions for the reproducibility analysis. Replicate data from Col-0 and Lerwere used for selecting outliers (see below). All statistical analyses were performed using the BioConductor packages \[[@B40]\] in R \[[@B41]\] and S-plus 6.1 (Insightful). The \'.CEL\' files were read directly into R and genomic hybridization intensity indices were computed from the individual probes (16-20 for each gene) using the Li-Wong PM-only model \[[@B20]\], which was implemented in the BioConductor package. The outlier genes from either the Col-0 replicates or the Lerreplicates (false positives) were eliminated. The outliers were defined as those genes whose hybridization intensity indices were at least twofold different between the two replicates. For the rest of the genes, the two Col-0 replicates and the two Lerreplicates were averaged separately to obtain a single value, which represents the signal intensities for Col-0 and Lergenomic DNA hybridization. Then the coefficient of variance (CV) was calculated for each gene on the basis of its genomic hybridization intensity indices from the five accessions. Genes with the highest 11% CV (CV ≥ 0.20) were eliminated from further expression analysis (see Additional data file 1). CV = 0.20 was chosen as the cutoff value on the basis of the following two criteria: it is equal to mean (CV) + 1 standard deviation from genomic DNA hybridization; we tried to exclude as much as possible the genes that could possibly have sequence differences among the five accessions, to ensure less interference when analyzing mRNA expression for the remaining genes. This resulted in 7,736 genes. Genes for the correlation analysis were selected from the 7,736-gene list from genomic DNA hybridization data. The mRNA expression index for each gene was also computed using the Li-Wong PM-only model \[[@B20]\]. The expression values of the selected genes were normalized by dividing the hybridization indices from RNA hybridization from each organ of a particular accession by the indices from genomic hybridization of this particular accession. The relative expression values for all the genes from all the experiments (7,736 × 50 = 386,800 data points) were sorted and the lowest five-percentile value was used as the cutoff value between noise and true signals. Then, genes whose expression value was below the cutoff value across all the RNA samples from at least one accession were further eliminated. This resulted in 7,508 genes. The normalized expression values were log~2~-transformed and used for the correlation analysis. In addition, this dataset of 7,508 genes was used for permutations in which, for a particular organ at a particular developmental stage, we randomly permuted among the five RNA samples from the five accessions (10 organs × (5 × 4 × 3 × 2 × 1 permutations for each organ) = 1,200 potential combinations), thus preserving the organ-age categorization. Then, for each gene, 10 pairwise comparisons, represented by 10 Pearson correlation coefficients, were made from the five different accessions. The Pearson correlation coefficient for each pair was calculated by using the normalized gene expression values from 10 organs (10 data points) of one accession versus the 10 data points from the other accession (see Additional data file 5 for an example). The number of genes that had r\< 0.5 in a given pair of compared accessions was calculated and is shown in Table [3](#T3){ref-type="table"} and Figure [2](#F2){ref-type="fig"}. With the permuted data, the numbers shown in Table [3](#T3){ref-type="table"} and Figure [2](#F2){ref-type="fig"} are the averages of the 10-permuted datasets. Cluster analysis of mRNA expression data was performed with the same list of 7,508 genes used for the correlation analysis. The normalized expression values were then log~2~-transformed, mean centered for each gene across all the samples, and subjected to the self-organizing maps, followed by average linkage hierarchical clustering of both genes and experiments using Cluster and visualized with TreeView to generate Figure [3](#F3){ref-type="fig"}. Analysis of variance (ANOVA) of mRNA expression data was performed with the same list of 7,508 genes used for the correlation analysis with functions in S-PLUS 6.1 (InSightful). The normalized expression values were log~2~-transformed and used for the ANOVA analysis. For one-way ANOVA analysis, the three leaf samples from 2-, 5- and 11-week-old leaves were treated as biological replicates, and the general linear model (GLM) is formulated as: expression = accessions + error. For two-way ANOVA analysis only the two leaf samples from 2- and 5-week-old leaves, and two root samples from 2- and 5-week-old roots were treated as biological replicates, and the GLM is: expression = accessions + organs + accessions × organs + error. We excluded the 11-week-old leaves in two-way ANOVA analysis to take into consideration the effect of age on gene expression. We have estimated the variance for each gene in leaves and roots of different accessions using the local pooled error (LPE) method \[[@B42]\], and found that only a small percentage of genes have different variance in other accessions as compared to one in Col-0. As there is no biological replicate for the rest of the organs, we are assuming that the errors for those organs are at similar levels, as estimated from the two leaf and root samples in the two-way ANOVA analysis. Genes with significant p-value (p\< 0.05) after Bonferroni correction were then selected accordingly. Statistical analysis for enrichment of MIPS functional categories ----------------------------------------------------------------- To test whether genes representing certain MIPs functional categories are over-represented in the list of statistically significant genes identified from either one-way, or two-way ANOVA, bootstrapping was performed by generating 1,000 control lists from all the genes on the array, each of which contains the same number of genes as contained in the list from either one-way, or two-way ANOVA analysis. Genes in each of the control lists were classified on the basis of MIPS functional categories. Then, for each functional category, a distribution of number of occurrences for that particular functional category from 1,000 control lists was generated, and this distribution was compared to the observed occurrence to determine the p-value. Validation of the GeneChip microarray data ------------------------------------------ The genomic sequence for gene 13903\_at (At3g54050) and 17392\_s\_at (At3g53260) from accession C24 was obtained by PCR with genomic DNA from C24, and the following primers based on this gene\'s coding sequence from Col-0. 13903\_at (At3g54050): 5\'-primer: 5\'-GATCCAATGTACGGTGAGTTTG-3\'; 3\'-primer: 5\'-TGCAT-ATACCATGTAGTCAG-3\'. 17392\_s\_at (At3g53260): 5\'-primer: 5\'-CAGTTTCTCAAGTTGCTAAG-3\'; 3\'-primer: 5\'-CATTCC-TTGAGACAATCCAT-3\' The PCR product was then sequenced and these sequences were used for designing gene-specific primers and probes for Taqman assay. The Lersequences of genes 12222\_s\_at (At2g20990), 14097\_at (At2g47770), 20561\_at (At2g46930), 14634\_s\_at (At4g27440), 13483\_at (At2g25650), 15290\_at (At2g20840), 13111\_at (At2g38040), 14072\_at (At1g67480), 14172\_at (At3g54140), 14947\_at (At4g37450), 16892\_at (At5g45890), 17860\_at (At4g27410), 20545\_at (At5g27470) were obtained by BLASTing the full-length cDNA sequences or coding sequences of these genes from Col-0 against the Lersequences available from TIGR \[[@B43]\]. Top BLAST hits were chosen and sequences common for both Col-0 and Lerwere used to design gene-specific primers and probes for Taqman assay. Quantitative RT-PCR (Taqman) assays were performed on an ABI Prism 7700 (Applied Biosystems), as previously described \[[@B44]\], using the following gene-specific primers and probe sets: 13903\_at\_forward primer: 5\'-GGTCCAACTGGGAAGCCTTAC-3\' 13903\_at\_reverse primer: 5\'-CCGTACAACAAAGTCCTGTGAAAA-3\' 13903\_at\_target probe: FAM-CCAACCAAACTTCCAATGTACCTTGCCGTAMRA. 17392\_s\_at\_forward primer: 5\'-GGCTGTGCTTCCAAAGGAAGT-3\' 17392\_s\_at\_reverse primer: 5\'-GTTAGGAATCGGCGCAGTTC-3\' 17392\_s\_at\_target probe: FAM-CTCCCATAAGCTGCTCTAGCCGCTTAMRA. 12222\_s\_at\_forward primer: 5\'-GGCTGTGCTTCCAAAGGAAGT-3\' 12222\_s\_at\_reverse primer: 5\'-GTTAGGAATCGGCGCAGTTC-3\' 12222\_s\_at\_target probe: FAM-CTCCCATAAGCTGCTCTAGCCGCTTAMRA. 14097\_at\_forward primer: 5\'-CAACAAAGGAAAACGCGATCA-3\' 14097\_at\_reverse primer: 5\'-CGCTACCGTCAGAGACTTGAGA-3\' 14097\_at\_target probe: FAM-AGAGGGCGATGGCGAAACGTGTAMRA. 20561\_at\_forward primer: 5\'-TGGTACTTTGACAGAACAACAGTGAA-3\' 20561\_at\_reverse primer: 5\'-TGAAGATGAGATTGTGACATGTTTTG-3\' 20561\_at\_target probe: FAM-CCATTGACTGTCCTTACCCCTGT-TAMRA. 14634\_s\_at\_forward primer: 5\'-CGAATACATTGGCGGGTAATG-3\' 14634\_s\_at\_reverse primer: 5\'-GCCGGCTAAACCCCTCAA-3\' 14634\_s\_at\_target probe: FAM-ACCACCGAAGGCGAATCTCGGTGTAMRA. 15290\_at\_forward primer: 5\'-TCCTGGAGCGTATGTTATGTGGTA-3\' 15290\_at\_reverse primer: 5\'-CACCCAAACTTCAGAGCACTATCA-3\' 15290\_at\_target probe: FAM-CGCCCTCTTTATCGTGCCATGAGGTAMRA. 14072\_at\_forward primer: 5\'-TGTATGACCCGGATGCTTCA-3\' 14072\_at\_reverse primer: 5\'-ACGCAAGAACCAGAGAGTTTGAT-3\' 14072\_at\_target probe: FAM-CAGGCACACAGTGGAAAACGTCTGA-TAMRA. 13111\_at\_forward primer: 5\'-GAGATCAAGAGCATGGTGGAGTT-3\' 13111\_at\_reverse primer: 5\'-GGTGACACCAGGCGTTTTG-3\' 13111\_at\_target probe: FAM-CTGAAAGTGGAAACCGCAAAGGCG-TAMRA. 14172\_at\_forward primer: 5\'-GGGTATAGGTCTTGTGGTCTCCAT-3\' 14172\_at\_reverse primer: 5\'-ATCAAGCCTGACAACCTCCAA-3\' 14172\_at\_target probe: FAM-TTTGCCATGATCACTGCAGGAG-TAMRA. 14947\_at\_forward primer: 5\'-TCCTAACAGTTACATTGATCTGCATTG-3\' 14947\_at\_reverse primer: 5\'-TGGTCGGAGAAGAGATAGGAGATT-3\' 14947\_at\_target probe: FAM-CGTCGCCGGTGTCGGTG-TAMRA. 16892\_at\_forward primer: 5\'-CCGGTTAATGATGAGCAAGCA-3\' 16892\_at\_reverse primer: 5\'-CCTCCTTCAATTCCAACGCTAA-3\' 16892\_at\_target probe: FAM-ATGAAGGCAGTGGCACACCAACC-TAMRA. 17860\_at\_forward primer: 5\'-ACGGTGGTTACGATGCGTTT-3\' 17860\_at\_reverse primer: 5\'-CCGATTCACATGCCCACTCT-3\' 17860\_at\_target probe: FAM-AGCGGCGGAAGGTGAGGCG-TAMRA. 20545\_at\_forward primer: 5\'-GAGCTTGTGTCTTGTTCCAACTGT-3\' 20545\_at\_reverse primer: 5\'-TGCTCTTTTTCTGACCGTATCTGA-3\' 20545\_at\_target probe: FAM-CAGACTACCAGGCTCGCAGGCTTGA-TAMRA. A standard curve consisting of serial 1:5 dilutions was prepared with RNA concentrations of 50 ng/μl, 10 ng/μl, 2 ng/μl, 0.4 ng/μl, and 0.08 ng/μl. Relative expression levels were interpolated by comparison with standard curves with a correlation coefficient of 0.99 or greater. Relative expression levels were normalized to the expression level of the Arabidopsis APX3gene \[[@B44]\], which was expressed at a constant level. All reactions were performed in triplicate. Promoter and polymorphism analysis ---------------------------------- Genomic DNA sequencing was used to analyze the polymorphisms in 12 different Arabidopsisaccessions. Genomic DNA of the accessions Col-0, C24, Ler, Ws-0, No-0, RLD-1, Ag-0, Bs-1, Cvi-0, Es-0, Gr-1, Mt-0 and Tsu-0 was obtained from tissue supplied by the stock center and used as the template for PCR amplification and sequencing. The sequencing strategy was as follows: using the AGI genome annotation as a guide, a region from 1 kb before the annotated translation start of each gene to 300 bp after the stop codon was amplified by LA-PCR (Long and accurate PCR) from each of the accessions. The PCR product was used directly for sequencing of both strands. Several primers were used to complete the sequencing of the whole gene and the 5\' and 3\' regions. Using Sequencher software (GeneCodes) the sequences from each accession were put into contiguous alignment for each gene. Sequence variations between the accessions in the promoter region, open reading frame (ORF), intron, exon and 3\' UTR were confirmed and recorded. The promoter region was defined as the available sequence (1 kb or more) before the translational start codon, while the intron-exon boundaries were defined using the AGI (Arabidopsis Gene Index) gene models, which were obtained from The ArabidopsisInformation Resource (TAIR) \[[@B45]\]. Only those differences confirmed in multiple sequencing were determined as polymorphisms. The polymorphism rate in promoters and exons was calculated as the number of bases substituted in any of the sequenced accession plus the total number of different insertion or deletion (indel) events found in all the accession in that sequence region, divided by the length of the available sequence. Alterations in potential cis-regulatory elements caused by polymorphisms were detected in the following automated way. The mutant and wild-type promoter sequences were searched for all known plant cis-regulatory elements in the databases PLACE \[[@B46]\] and plantCARE \[[@B47]\] using a custom-written PERL script. The lists of cis-regulatory elements were compared to find elements created or destroyed by the polymorphisms. This list was then manually edited to remove unlikely candidates for promoter regulatory sequences, such as potential translation initiation sites that were outside the transcribed region, or putative polyadenylation motifs situated in the promoter region. Additional data files ===================== The following additional data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} is a table showing probe sets representing genes with highly polymorphic coding sequences. Additional data file [2](#S2){ref-type="supplementary-material"} is a table showing samples used in this study. Additional data file [3](#S3){ref-type="supplementary-material"} is a table showing correlations between raw and normalized RNA hybridization indices among all 50 samples. Additional data file [4](#S4){ref-type="supplementary-material"} is a table showing examples of (a) one-way and (b) two-way ANOVA tables from analysis of variance (ANOVA). Additional data file [5](#S5){ref-type="supplementary-material"} is a table showing an example of the Pearson correlation coefficients matrix for a particular gene obtained from 10 pair-wise comparisons among the five accessions. Additional data file [6](#S6){ref-type="supplementary-material"} is a table showing the sequence variation in promoter regions that alters cis-elements. Additional data file [7](#S7){ref-type="supplementary-material"} is a table showing mRNA expression of genes identified from two-way ANOVA. Additional data file [8](#S8){ref-type="supplementary-material"} is a figure showing a histogram of coefficient of variance (CV) based on genomic hybridization intensity indices from the five accessions. Additional data file [9](#S9){ref-type="supplementary-material"} is a QQ-plot showing the effect of using gDHI to normalize rRHI to reduce the residual effect of sequence difference between targets and probes during mRNA hybridization. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 These probe sets were identified based on the coefficients of variance levels of hybridization indices calculated from genomic DNA hybridization data ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A table showing samples used in this study ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 A table showing correlations between raw and normalized RNA hybridization indices among all 50 samples ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 \(a) One-way and (b) two-way ANOVA tables from analysis of variance ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 A table showing an example of the Pearson correlation coefficients matrix for a particular gene obtained from 10 pair-wise comparisons among the five accessions ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 A table showing the sequence variation in promoter regions that alters cis-elements ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 A table showing mRNA expression of genes identified from two-way ANOVA ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 A figure showing a histogram of coefficient of variance (CV) based on genomic hybridization intensity indices from the five accessions ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 9 Two representative samples were shown, the Col-0 4d-seedlings and the NO-0 4d-seedlings before and after genomic DNA normalization. The rest of the 48 samples have similar QQ-profiles ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Bin Han for technical assistance in preparing samples used in the microarray experiments and for help in conducting the microarray experiments, Xun Wang for his support, and Zhen Su for computational analysis. We also thank the anonymous reviewers for constructive suggestions on the statistical analysis of the data. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic diagram of the data analysis process. A genome scan (left panel) was used to identify probe sets corresponding to the genes that were highly polymorphic or less polymorphic in gene coding regions among the five accessions. Genes with polymorphic sequences were functionally categorized. Probe sets corresponding to the less polymorphic genes were used for a transcriptome scan of various accessions (right panel). Genes transcribed at different levels in different accessions were identified and analyzed. ::: ![](gb-2005-6-4-r32-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Correlation analysis of expression patterns of genes among the five accessions. A histogram based on the number of genes in each of the 11 groups in Table 3 that have Pearson correlation coefficients less than 0.5 in a given number of pairwise comparisons (see Table 3 for explanation). The white bars indicate the numbers of genes from the experimental datasets, and the gray bars indicate the average numbers of genes from the 10 permuted datasets, as described in Materials and methods. ::: ![](gb-2005-6-4-r32-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Relationships among the five Arabidopsisaccessions based on their expression patterns in different organs at various developmental stages. The normalized expression values, obtained by dividing the mRNA expression indices of each organ of one accession by the intensity indices in genomic DNA hybridization for that particular accession, were log~2~-transformed and subjected to cluster analysis. The yellow vertical lines separate the whole cluster into three subclusters, the root cluster, the vegetative leaf cluster, and the reproductive organ cluster. ::: ![](gb-2005-6-4-r32-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Correlations in transcription among five accessions during leaf and silique development. (a)The Pearson correlation coefficient for a given sample was calculated with nRHI for all the genes from each accession and the reference accession Col-0. Each bar represents the correlation of a particular accession as compared to Col-0 in the sample group. Note the common trend in reduction of the correlation during leaf and silique development for each organ. (b)The regression coefficient for a given sample was calculated with nRHI for all the genes from each accession (Y-values, regressor) and the reference accession Col-0 (X-values, predictor). Each bar represents the regression coefficient of a particular accession as compared to Col-0 in the sample group. The regression coefficient (b) was calculated as b= (ΣX~i~Y~i~- (ΣX~i~)(ΣY~i~)/n)/(ΣX~i~^2^- (ΣX~i~)^2^/n), where nis the total number of genes in either Col-0 or the sample to be compared (7,508 in this case). The error bar indicates the upper or lower limit of the 95% confidence interval for each of the given regression coefficients. The 95% confidence interval was calculated as b± tα~(2)~, ~(n-2)~Sb, where tα~(2)~, ~(n-2)~is the tcritical value at α = 0.05, two-tail, df = 7,506, and Sbis the standard deviation of b. ::: ![](gb-2005-6-4-r32-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Functional distribution of genes that are differentially regulated in leaves of the five accessions. Fifty-eight genes, identified by one-way ANOVA analysis, were subjected to MIPS functional classification based on their annotations. ::: ![](gb-2005-6-4-r32-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Functional distribution of genes that are differentially regulated by accession-by-organ interactions. Fifty-two genes, identified by two-way ANOVA analysis, were subjected to MIPS functional classification based on their annotations. ::: ![](gb-2005-6-4-r32-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Functional distribution of the 65 most plastic genes. The 65 most plastic genes identified from the expression correlation analysis, whose correlation coefficients are less than 0.5 in all 10 pairwise compared accessions, were subjected to MIPs functional classification based on their annotations. ::: ![](gb-2005-6-4-r32-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Quantitative RT-PCR confirmation of GeneChip Microarray data for genes 13903\_at (At3g54050) and 17392\_s\_at (At3g53260) in Col-0 and C24 ::: --------------------------------------- -------------- -------------- --------------- -------------- -------------- --------------- Samples 13903\_at 17392\_s\_at log~2~(rRHI) log~2~(nRHI) Taqman log~2~(rRHI) log~2~(nRHI) Taqman Col-0-4 day seedlings 10.11940591 0.911529477 1.348 ± 0.262 10.38351776 0.658285681 0.362 ± 0.024 Col-0-2 week leaf 11.80337083 2.595494397 4.652 ± 0.389 10.56878747 0.84355539 0.299 ± 0.050 Col-0-11 week leaf 10.77324577 1.565369327 1.415 ± 0.336 10.33789612 0.612664042 0.163 ± 0.052 Col-0-2 week root 7.674725423 -1.533151014 0.134 ± 0.014 11.26384894 1.538616864 1.313 ± 0.324 Col-0-5 week root 7.873250697 -1.334625741 0.590 ± 0.064 10.99787749 1.272645415 0.648 ± 0.246 Col-0-influorescence 10.09145865 0.883582211 1.320 ± 0.247 11.01034472 1.285112643 0.519 ± 0.104 Col-0-flower 10.42134176 1.213465325 2.093 ± 0.658 10.62442631 0.899194238 0.263 ± 0.053 Col-0-young siliques 10.65287316 1.444996723 1.999 ± 2.885 10.57630495 0.851072873 0.430 ± 0.197 Col-0-mature siliques 9.475076913 0.267200476 1.432 ± 2.345 10.80990555 1.084673476 0.473 ± 0.113 C24-4 day seedlings 10.90593269 1.883371001 3.690 ± 0.482 10.20742445 0.596353845 0.321 ± 0.059 C24-2 week leaf 12.29789156 3.275329874 6.819 ± 3.507 10.65702025 1.04594965 0.299 ± 0.044 C24-11 week leaf 12.09006973 3.067508045 6.073 ± 1.283 9.19787898 -0.413191622 0.071 ± 0.037 C24-2 week root 7.550943148 -1.471618541 0.069 ± 0.022 10.89199181 1.280921209 0.790 ± 0.133 C24-5 week root 7.945743693 -1.076817996 0.317 ± 0.087 11.16598953 1.554918929 1.122 ± 0.324 C24-influorescence 10.72350042 1.700938727 2.397 ± 0.304 11.10540542 1.494334819 0.743 ± 0.105 C24-flower 10.71423996 1.691678266 1.054 ± 0.167 9.761854806 0.150784204 0.153 ± 0.048 C24-young siliques 11.01401689 1.991455197 1.885 ± 0.726 10.61478826 1.00371766 0.365 ± 0.058 C24-mature siliques 11.21144986 2.188888168 3.808 ± 0.569 11.24013223 1.629061624 1.002 ± 0.151 Correlation with log~2~(Taqman assay) 0.925 0.933 0.801 0.821 --------------------------------------- -------------- -------------- --------------- -------------- -------------- --------------- gDHI for 13903\_at is 591.35 and 520.07 for Col-0 and C24, respectively. gDHI for 17392\_s\_at is 846.42 and 782.02 for Col-0 and C24, respectively. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Quantitative RT-PCR confirmation of GeneChip microarray data for genes expressed in Col-0 and Lerflowers ::: Probe set ID Col-flower Ler-flower Fold changes -------------- ------------ -------------- -------------- ------------- --------- --------- ------ ------------- ------ ------- 12222\_s\_at 1407.33 700.57 2.01 0.48 ± 0.16 1440.54 557.60 2.58 0.78 ± 0.13 1.29 1.62 14097\_at 610.06 1822.91 0.34 0.13 ± 0.03 899.39 1762.56 0.51 0.70 ± 0.23 1.52 5.56 20561\_at 760.62 648.27 1.17 0.90 ± 0.14 625.43 719.12 0.87 0.88 ± 0.24 0.74 0.97 14634\_s\_at 2914.65 1050.64 2.77 0.31 ± 0.05 4304.12 871.65 4.94 0.88 ± 0.05 1.78 2.85 15290\_at 701.80 679.74 1.03 0.35 ± 0.03 965.63 583.78 1.65 1.04 ± 0.06 1.60 2.94 14072\_at 2034.34 957.24 1.57 0.85 ± 0.13 2285.99 948.01 1.68 1.08 ± 0.20 1.07 1.27 14172\_at 894.36 1042.33 0.86 0.44 ± 0.06 1114.93 1107.46 1.01 0.91 ± 0.08 1.17 2.04 14947\_at 1888.06 1250.42 1.51 0.98 ± 0.22 1754.25 981.19 1.79 1.24 ± 0.12 1.18 1.26 16892\_at 2688.88 836.69 3.22 0.51 ± 0.05 2798.26 1061.25 2.64 1.10 ± 0.11 0.82 2.17 17860\_at 959.84 1263.46 0.76 0.49 ± 0.06 1209.50 1322.29 0.92 1.11 ± 0.13 1.20 2.26 20545\_at 2183.17 724.58 3.02 0.59 ± 0.09 1971.40 668.92 2.95 0.99 ± 0.09 0.98 1.686 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Correlation analysis of expression patterns of genes among the five accessions ::: Per 1 Per 2 Per 3 Per 4 Per 5 Per 6 Per 7 Per 8 Per 9 Per 10 Average of Per Observed ---- ------- ------- ------- ------- ------- ------- ------- ------- ------- -------- ---------------- ---------- 10 141 133 127 129 129 130 139 121 130 125 130.4 65 9 263 263 246 268 281 285 271 295 273 259 270.4 271 8 324 324 341 323 336 320 307 324 319 359 327.7 376 7 1555 1555 1542 1539 1499 1508 1541 1524 1521 1505 1528.9 399 6 1523 1523 1575 1505 1515 1547 1557 1495 1524 1539 1530.3 412 5 603 603 590 617 607 629 609 642 632 577 610.9 416 4 692 692 661 725 724 660 668 669 715 719 692.5 471 3 441 441 441 436 440 457 461 457 424 441 443.9 438 2 345 345 375 334 341 351 340 355 357 359 350.2 528 1 360 360 365 382 362 329 345 350 358 350 356.1 600 0 1261 1261 1245 1250 1274 1292 1270 1276 1255 1275 1265.9 3532 7508 7500 7508 7508 7508 7508 7508 7508 7508 7508 7508 For each gene, the Pearson correlation coefficient was calculated for all the 10 pairwise comparisons among the five accessions, as described in Materials and methods. Genes were then grouped into 11 groups (0-10) according to the number of comparisons having correlation coefficients less than 0.5 (group 10 corresponds to the genes with r\< 0.5 from all 10 pairwise comparisons, whereas group 0 corresponds to genes with r≥ 0.5 from all 10 pairwise comparisons). These results are given in the Observed column. Columns Per 1 to Per 10 show the numbers of genes from the 10 permuted datasets, as described in Materials and methods. These results are visualized in Figure 2. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Genes whose expression is different in leaves of the five accessions by one-way ANOVA analysis ::: Functional category ATH1 hits rawp Bonferroni correction GenBank ID Description ------------------------------------------------------------------------------------ ------------- ------------- ----------------------- ---------------------------------------------------------------------- ----------------------------------------------------------------------------------------------- 01 METABOLISM 17946\_s\_at At1g03410 2.84132E-06 0.0213326 gb\\|AAB97721.1\\| 2-Oxoglutarate-dependent dioxygenase, putative 19689\_at At5g24140 7.0739E-07 0.0053111 emb\\|CAA06771.1\\| Squalene monooxygenase 2 (squalene epoxidase 2) (SQP2) (SE2) 12277\_at At1g47600 6.47203E-07 0.0048592 gb\\|AAD46026.1\\| Glycosyl hydrolase family 1, similar to thioglucosidase 18836\_at At2g24710 5.09671E-06 0.0382661 gb\\|AAD26894.1\\| Plant glutamate receptor family (GLR2.3) 17620\_s\_at At2g42990 3.79611E-06 0.0285012 gb\\|AAD21711.1\\| GDSL-motif lipase/hydrolase protein similar to family II lipase EXL3 20514\_i\_at At2g15370 1.35384E-08 0.0001016 gb\\|AAD22287.1\\| Similar to xyloglucan fucosyltransferase 02 ENERGY 12277\_at At1g47600 6.47203E-07 0.0048592 gb\\|AAD46026.1\\| Glycosyl hydrolase family 1, similar to thioglucosidase 10 CELL CYCLE AND DNA PROCESSING 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 15785\_g\_at At1g08840 2.79162E-06 0.0209595 gb\\|AAB70418.1\\| Hypothetical protein gene overlaps Sp6 end of F7G19 11 TRANSCRIPTION 12869\_s\_at At4g11880 3.45683E-06 0.0259539 gb\\|AAC49082.1\\| MADS-box protein AGL14 16072\_s\_at At5g65790 2.97265E-06 0.0223187 gb\\|AAC83623.1\\| Identical to putative transcription factor (MYB68) 13575\_at At4g03430 6.50883E-06 0.0488683 gb\\|AAD11585.1\\| Similar to yeast pre-mRNA splicing factors 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| 12366\_s\_at At4g11880 2.5787E-06 0.0193608 emb\\|CAB44326.1\\| MADS-box protein AGL14 14885\_at At4g21340 2.22259E-06 0.0166872 emb\\|CAA20199.1\\| Expressed protein, putative bHLH transcription factor (bHLH103) 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 19244\_s\_at At2g04230 2.84687E-06 0.0213743 gb\\|AAD27915.1\\| F-box protein family, contains F-box domain 19279\_i\_at At4g21040 7.07742E-07 0.0053137 emb\\|CAB45899.1\\| Dof zinc finger protein, finger protein rolB 13306\_at At2g41070 4.38217E-06 0.0329013 gb\\|AAD12004.1\\| bZIP family transcription factor, contains a bZIP transcription factor basic domain signature 13343\_at At1g34310 4.60448E-08 0.0003457 gb\\|AAD39615.1\\| Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related 15224\_at At1g61540 8.21522E-08 0.0006168 gb\\|AAD25554.1\\| Kelch repeat containing F-box protein family low similarity to SKP1 interacting partner 6 15227\_at At2g01280 5.89076E-06 0.0442278 gb\\|AAD14528.1\\| Transcription factor -related, putative transcription factor IIIB 70 KD subunit (TFIIIB) 16263\_at At2g02320 3.12005E-06 0.0234253 gb\\|AAC78515.1\\| F-box protein (SKP1 interacting partner 3-related) 17145\_at At1g10110 6.9462E-08 0.0005215 gb\\|AAC34337.1\\| Contains Pfam PF00646: F-box domain; similar to F-box protein family, AtFBX7 13863\_at At2g21470 9.36899E-07 0.0070342 gb\\|AAD23691.1\\| Nearly identical to SUMO activating enzyme 2 (SAE2) 12599\_at At2g29910 2.33543E-10 0.0000018 gb\\|AAD23631.1\\| F-box protein family contains F-box domain Pfam:PF00646 12913\_at At4g32880 1.90072E-06 0.0142706 emb\\|CAA90703.1\\| Identical to HD-zip transcription factor (athb-8) 13216\_s\_at At1g26310 8.05827E-07 0.0060501 gb\\|AAA64789.1\\| Floral regulatory gene CAULIFLOWER 12863\_r\_at At4g18960 1.05463E-06 0.0079181 emb\\|X53579.1\\| Floral homeotic protein agamous (AGAMOUS) 14 PROTEIN FATE (folding, modification, destination) 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 13863\_at At2g21470 9.36899E-07 0.0070342 gb\\|AAD23691.1\\| Nearly identical to SUMO activating enzyme 2 (SAE2) 16 PROTEIN WITH BINDING FUNCTION OR COFACTOR REQUIREMENT (structural or catalytic) 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| 18836\_at At2g24710 5.09671E-06 0.0382661 gb\\|AAD26894.1\\| Plant glutamate receptor family (GLR2.3) 16262\_at At2g46850 4.75288E-06 0.0356846 gb\\|AAC34215.2\\| Ser/Thr protein kinase -related 20 CELLULAR TRANSPORT, TRANSPORT FACILITATION AND TRANSPORT ROUTES 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 18836\_at At2g24710 5.09671E-06 0.0382661 gb\\|AAD26894.1\\| Plant glutamate receptor family (GLR2.3) 17618\_at At2g31910 3.44193E-06 0.0258420 gb\\|AAD32281.1\\| Similar to monovalent cation:proton antiporter family 2 30 CELLULAR COMMUNICATION/SIGNAL TRANSDUCTION MECHANISM 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 16816\_at At1g19230 5.65137E-06 0.0424305 gb\\|AAC39478.1\\| Respiratory burst oxidase protein E (NADPH oxidase) (RbohE) 18836\_at At2g24710 5.09671E-06 0.0382661 gb\\|AAD26894.1\\| Plant glutamate receptor family (GLR2.3) 19311\_g\_at At2g41210 1.643E-06 0.0123356 gb\\|AAC78530.2\\| Phosphatidylinositol-4-phosphate 5-kinase -related 13343\_at At1g34310 4.60448E-08 0.0003457 gb\\|AAD39615.1\\| Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related 15787\_s\_at At1g09090 3.64297E-07 0.0027351 gb\\|AAB70399.1\\| Respiratory burst oxidase protein B (NADPH oxidase) (RbohB) 16262\_at At2g46850 4.75288E-06 0.0356846 gb\\|AAC34215.2\\| Ser/Thr protein kinase -related 32 CELL RESCUE, DEFENSE AND VIRULENCE 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 12111\_s\_at At4g19240 3.30499E-07 0.0024814 emb\\|CAA18611.1\\| Expressed protein 12258\_s\_at At4g14370 6.60533E-06 0.0495928 emb\\|CAB10216.1\\| Disease resistance protein (TIR-NBS-LRR class) 12277\_at At1g47600 6.47203E-07 0.0048592 gb\\|AAD46026.1\\| Glycosyl hydrolase family 1, similar to thioglucosidase 12956\_i\_at At1g05170 3.64708E-06 0.0273823 gb\\|AAB71461.1\\| Galactosyltransferase family 16375\_at At1g54480 6.28683E-06 0.0472015 gb\\|AAD25626.1\\| Leucine rich repeat protein family contains leucine rich-repeat (LRR) domains 16816\_at At1g19230 5.65137E-06 0.0424305 gb\\|AAC39478.1\\| Respiratory burst oxidase protein E (NADPH oxidase) (RbohE) 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 19244\_s\_at At2g04230 2.84687E-06 0.0213743 gb\\|AAD27915.1\\| F-box protein family, contains F-box domain 15224\_at At1g61540 8.21522E-08 0.0006168 gb\\|AAD25554.1\\| Kelch repeat containing F-box protein family low similarity to SKP1 interacting partner 6 15787\_s\_at At1g09090 3.64297E-07 0.0027351 gb\\|AAB70399.1\\| Respiratory burst oxidase protein B (NADPH oxidase) (RbohB) 16263\_at At2g02320 3.12005E-06 0.0234253 gb\\|AAC78515.1\\| F-box protein (SKP1 interacting partner 3-related) 17145\_at At1g10110 6.9462E-08 0.0005215 gb\\|AAC34337.1\\| Contains Pfam PF00646: F-box domain; similar to F-box protein family, AtFBX7 12599\_at At2g29910 2.33543E-10 0.0000018 gb\\|AAD23631.1\\| F-box protein family contains F-box domain Pfam:PF00646 34 INTERACTION WITH THE CELLULAR ENVIRONMENT 16816\_at At1g19230 5.65137E-06 0.0424305 gb\\|AAC39478.1\\| Respiratory burst oxidase protein E (NADPH oxidase) (RbohE) 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 15787\_s\_at At1g09090 3.64297E-07 0.0027351 gb\\|AAB70399.1\\| Respiratory burst oxidase protein B (NADPH oxidase) (RbohB) 36 INTERACTION WITH THE ENVIRONMENT (systemic) 17946\_s\_at At1g03410 2.84132E-06 0.0213326 gb\\|AAB97721.1\\| 2-Oxoglutarate-dependent dioxygenase, putative 38 TRANSPOSABLE ELEMENTS, VIRAL AND PLASMID PROTEINS 16731\_at At2g11690 1.06284E-06 0.0079798 gb\\|AAD28679.1\\| Pseudogene 18340\_at At4g07700 2.79501E-06 0.0209849 gb\\|AAD29786.1\\| Athila transposon protein -related 40 CELL FATE 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 41 DEVELOPMENT (systemic) 17677\_at At1g03910 1.54447E-06 0.0115959 gb\\|AAD10685.1\\| Hypothetical protein 13216\_s\_at At1g26310 8.05827E-07 0.0060501 gb\\|AAA64789.1\\| Floral regulatory gene CAULIFLOWER 12863\_r\_at At4g18960 1.05463E-06 0.0079181 emb\\|X53579.1\\| Floral homeotic protein agamous (AGAMOUS) 42 BIOGENESIS OF CELLULAR COMPONENTS 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 13343\_at At1g34310 4.60448E-08 0.0003457 gb\\|AAD39615.1\\| Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related 43 CELL TYPE DIFFERENTIATION 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 70 SUBCELLULAR LOCALIZATION 19689\_at At5g24140 7.0739E-07 0.0053111 emb\\|CAA06771.1\\| Squalene monooxygenase 2 (squalene epoxidase 2) (SQP2) (SE2) 20254\_at At2g22390 2.41214E-06 0.0181104 gb\\|AAD22360.1\\| Pseudogene, putative GTP-binding protein 18830\_at At2g32790 1.27302E-06 0.0095579 gb\\|AAC04484.1\\| Ubiquitin-conjugating enzyme 18836\_at At2g24710 5.09671E-06 0.0382661 gb\\|AAD26894.1\\| Plant glutamate receptor family (GLR2.3) 13343\_at At1g34310 4.60448E-08 0.0003457 gb\\|AAD39615.1\\| Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related 13863\_at At2g21470 9.36899E-07 0.0070342 gb\\|AAD23691.1\\| Nearly identical to SUMO activating enzyme 2 (SAE2) 15785\_g\_at At1g08840 2.79162E-06 0.0209595 gb\\|AAB70418.1\\| Hypothetical protein gene overlaps Sp6 end of F7G19 13216\_s\_at At1g26310 8.05827E-07 0.0060501 gb\\|AAA64789.1\\| Floral regulatory gene CAULIFLOWER 14356\_at At5g59370 3.6446E-08 0.0002736 gb\\|AAB39403.1\\| Identical to SP\\|P53494 Actin 4 12863\_r\_at At4g18960 1.05463E-06 0.0079181 emb\\|X53579.1\\| Floral homeotic protein agamous (AGAMOUS) No hits to TIGR gene prediction 20512\_at 4.2379E-07 0.0031818 gb\\|AC002336.3\\| Arabidopsis thalianachromosome 2 clone T2P4 map CIC10A06, complete 18049\_s\_at 5.37206E-07 0.0040333 emb\\|AJ132404.1\\| Arabidopsis thalianaanti-sense transcript, AKL kinase-like gene ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Genes whose expression is affected by accession-by-organ interaction, identified through two-way ANOVA analysis ::: Functional category ATH1 hits Pr(F)-accessions Pr(F)-Organs Pr(F)-accessions: organs Bonferroni corrected Pr(F)-accessions: organs Description --------------------------------------------------------- ----------- ------------------ -------------- -------------------------- ----------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------- 41 DEVELOPMENT (systemic) 18715\_at At1g14930 1.0285E-05 1.3523E-13 1.2253E-08 9.1998E-05 Major latex protein (MLP)-related low similarity to major latex protein 18229\_at At1g14940 5.4018E-07 3.5527E-15 3.3317E-10 2.5014E-06 Major latex protein (MLP)-related low similarity to major latex protein 18717\_at At1g14950 8.6683E-04 2.3537E-14 3.8173E-07 2.8661E-03 Major latex protein (MLP)-related low similarity to major latex protein 17893\_at At2g23110 2.2913E-06 1.7710E-10 3.0508E-07 2.2906E-03 Late embryogenesis abundant proteins -related 12731\_f\_at At2g26960 1.1209E-09 4.2244E-09 1.6659E-09 1.2507E-05 MYB family transcription factor 20004\_s\_at At2g35300 3.0731E-06 1.2479E-13 1.4412E-06 1.0820E-02 Late embryogenesis abundant proteins -related identical to GB:X91917 13674\_s\_at At2g36640 9.1097E-07 1.4619E-11 8.2287E-07 6.1781E-03 Nearly identical to LEA protein in group 3 17038\_s\_at At2g36640 6.4830E-06 1.4944E-10 5.8337E-06 4.3799E-02 Nearly identical to LEA protein in group 3 16896\_s\_at At2g41260 3.9707E-11 1.1213E-14 3.9237E-10 2.9459E-06 Glycine-rich, identical to late-embryogenesis abundant M17 protein GI:3342551 19355\_s\_at At2g41280 3.7790E-09 6.5988E-10 3.4337E-07 2.5780E-03 Late embryogenesis abundant M10 protein identical to GB:AF076979 15747\_at At2g42560 5.1854E-08 6.4206E-12 3.2085E-07 2.4090E-03 Late embryogenesis abundant (LEA) domain-containing protein 15604\_s\_at At3g15400 4.6444E-07 4.5835E-09 3.5477E-06 2.6636E-02 Identical to anther development protein ATA20 GB:AAC50042 19918\_at At3g15670 3.7835E-08 1.5210E-14 4.4004E-08 3.3038E-04 Similar to SP\\|P13934 Late embryogenesis abundant protein 76 (LEA 76) 18872\_at At3g17520 4.2978E-10 1.1102E-16 2.5048E-09 1.8806E-05 Low similarity to PIR\\|S04045\\|S04045 embryonic abundant protein D-29 17282\_s\_at At3g51810 1.4069E-08 1.5599E-13 6.4057E-10 4.8094E-06 Embryonic abundant protein AtEm1 20682\_g\_at At4g26740 6.4621E-04 2.8866E-15 1.0571E-06 7.9364E-03 Embryo-specific protein 1 (ATS1) putative Ca2+-binding EF-hand protein 13675\_s\_at At3g22500 8.1351E-08 7.9450E-09 7.7592E-07 5.8256E-03 LEA protein, putative 04 STORAGE PROTEIN 18295\_s\_at At1g03880 3.5027E-08 0.0000E+00 1.9892E-10 1.4935E-06 12S seed storage protein (CRB) 13200\_s\_at At1g03880 2.0300E-05 2.4425E-15 1.7983E-07 1.3502E-03 12S seed storage protein (CRB) 20221\_at At1g03890 8.7822E-06 5.1070E-15 2.5720E-07 1.9311E-03 Globulin (seed storage protein) family similar to Arabidopsis thaliana12S seed storage proteins SP\\|P15455 20222\_g\_at At1g03890 2.3617E-05 2.7756E-15 2.5729E-07 1.9317E-03 globulin (seed storage protein) family similar to Arabidopsis thaliana12S seed storage proteins SP\\|P15455 20535\_s\_at At2g28490 2.4914E-03 1.1269E-13 3.8367E-06 2.8806E-02 Cupin domain-containing protein similar to preproMP27-MP32 \[Cucurbita cv. Kurokawa Amakuri\] 15983\_s\_at At4g27140 3.1858E-04 1.4433E-15 2.6036E-07 1.9547E-03 2S seed storage protein 1 (NWMU1 - 2S albumin 1) identical to SP\\|P15457 15984\_s\_at At4g27170 8.4937E-06 0.0000E+00 6.1932E-09 4.6498E-05 2S seed storage protein 4 (NWMU2-2S albumin 4) identical to SP\\|P15460 13449\_at At4g36700 1.5016E-05 2.9865E-14 3.3621E-06 2.5242E-02 Cupin domain-containing protein low similarity to preproMP27-MP32 from Cucurbita cv. Kurokawa Amakuri 16025\_s\_at At4g28520 6.5162E-09 0.0000E+00 2.2615E-10 1.6980E-06 12S seed storage protein (cruciferin), putative 16425\_s\_at At5g44120 2.4424E-08 6.1062E-15 3.4512E-07 2.5912E-03 12S seed storage protein (CRA1) 13201\_at At5g54740 3.4456E-08 0.0000E+00 1.8704E-11 1.4043E-07 2S seed storage protein family protein 13194\_at At4g27160 1.0828E-06 5.7732E-15 2.4480E-07 1.8380E-03 NWMU3 - 2S albumin 3 precursor, seed storage protein AT2S3 13198\_i\_at At4g28520 4.1773E-07 4.8295E-14 8.3466E-08 6.2666E-04 12S cruciferin seed storage protein 13199\_r\_at At4g28520 9.8653E-08 1.0880E-14 1.8093E-08 1.3585E-04 12S cruciferin seed storage protein 32 CELL RESCUE, DEFENSE AND VIRULENCE 14789\_at At2g15010 1.0120E-04 1.2166E-12 4.2917E-06 3.2222E-02 Similar to thionin \[Arabidopsis thaliana\] gi\\|1181533\\|gb\\|AAC41679 18715\_at At1g14930 1.0285E-05 1.3523E-13 1.2253E-08 9.1998E-05 Low similarity to major latex protein {Papaver somniferum} 18229\_at At1g14940 5.4018E-07 3.5527E-15 3.3317E-10 2.5014E-06 Low similarity to major latex protein {Papaver somniferum} 18717\_at At1g14950 8.6683E-04 2.3537E-14 3.8173E-07 2.8661E-03 Low similarity to major latex protein {Papaver somniferum} 20375\_at At1g48130 2.0800E-05 3.1086E-15 1.2134E-07 9.1102E-04 Peroxiredoxin identical to SP:O04005 from \[Arabidopsis thaliana\] 18716\_at At1g75830 1.0527E-05 4.7479E-10 1.2692E-06 9.5295E-03 Plant defensin protein, putative (PDF1.1) 16450\_s\_at At3g50980 1.1415E-05 7.6645E-12 8.6448E-07 6.4905E-03 Dehydrin, putative similar to dehydrin Xero 1 17282\_s\_at At3g51810 1.4069E-08 1.5599E-13 6.4057E-10 4.8094E-06 Embryonic abundant protein AtEm1 16892\_at At5g45890 3.2112E-09 0.0000E+00 1.7785E-10 1.3353E-06 Cysteine protease SAG12 identical to senescence-specific protein SAG12 18558\_at At2g21490 3.7353E-07 2.0317E-14 2.1270E-07 1.5969E-03 Putative dehydrin 17310\_at At3g51810 4.4370E-06 4.0301E-14 5.2274E-09 3.9248E-05 Embryonic abundant protein AtEm1 01 METABOLISM 18320\_s\_at At1g02790 5.5345E-07 0.0000E+00 1.1637E-07 8.7372E-04 Similar to polygalacturonase 17316\_at At2g16730 8.3049E-09 1.0945E-12 1.0082E-06 7.5697E-03 Glycosyl hydrolase family 35 (beta-galactosidase) 19003\_at At2g25890 1.3232E-05 1.8763E-13 5.8363E-07 4.3819E-03 Oleosin 20375\_at At1g48130 2.0800E-05 3.1086E-15 1.2134E-07 9.1102E-04 Peroxiredoxin identical to SP:O04005 from \[Arabidopsis thaliana\] 18991\_s\_at At3g27660 1.6605E-04 3.1308E-14 2.4540E-06 1.8425E-02 Identical to oleosin isoform GB:S71286 from \[Arabidopsis thaliana\] 19435\_at At4g00240 3.9561E-08 2.8820E-06 4.0132E-07 3.0131E-03 Phospholipase D -related 16865\_s\_at At3g57510 6.4423E-08 3.7925E-13 6.5117E-06 4.8890E-02 Putative similar to polygalacturonase 20412\_s\_at At4g25140 3.9887E-06 4.4409E-16 1.5210E-07 1.1420E-03 Oleosin 12435\_s\_at At4g34520 1.3485E-05 1.1102E-16 5.6989E-08 4.2788E-04 Fatty acid elongase 1 (FAE1) identical to fatty acid elongase 1 \[GI:881615\] 16575\_s\_at At5g40420 2.6083E-08 0.0000E+00 7.9331E-09 5.9562E-05 Oleosin 20035\_at At5g44440 1.8623E-07 3.5083E-14 4.1535E-07 3.1184E-03 FAD-linked oxidoreductase family similar to SP\\|P30986 reticuline oxidase precursor (Berberine-bridge-forming enzyme) (BBE) 42 BIOGENESIS OF CELLULAR COMPONENTS 18320\_s\_at At1g02790 5.5345E-07 0.0000E+00 1.1637E-07 8.7372E-04 Similar to polygalacturonase GI:288611 from \[Zea mays\] 19003\_at At2g25890 1.3232E-05 1.8763E-13 5.8363E-07 4.3819E-03 oleosin 15604\_s\_at At3g15400 4.6444E-07 4.5835E-09 3.5477E-06 2.6636E-02 Identical to anther development protein ATA20 18716\_at At1g75830 1.0527E-05 4.7479E-10 1.2692E-06 9.5295E-03 Plant defensin protein, putative (PDF1.1) 18991\_s\_at At3g27660 1.6605E-04 3.1308E-14 2.4540E-06 1.8425E-02 Identical to oleosin isoform GB:S71286 from \[Arabidopsis thaliana\] 16865\_s\_at At3g57510 6.4423E-08 3.7925E-13 6.5117E-06 4.8890E-02 Similar to polygalacturonase GI:288611 from \[Zea mays\] 13243\_r\_at At4g37990 2.8137E-07 4.7398E-09 9.4561E-07 7.0996E-03 Mannitol dehydrogenase (ELI3-2), putative 16575\_s\_at At5g40420 2.6083E-08 0.0000E+00 7.9331E-09 5.9562E-05 Oleosin 70 SUBCELLULAR LOCALIZATION 12085\_at At1g04560 7.4897E-04 2.7756E-15 2.3200E-07 1.7418E-03 Expressed protein similar to GB:AAC37469 12731\_f\_at At2g26960 1.1209E-09 4.2244E-09 1.6659E-09 1.2507E-05 MYB family transcription factor 17710\_at At2g28340 7.6288E-08 2.4759E-06 6.9175E-07 5.1936E-03 GATA zinc finger protein and genefinder 20375\_at At1g48130 2.0800E-05 3.1086E-15 1.2134E-07 9.1102E-04 Peroxiredoxin identical to SP:O04005 from \[Arabidopsis thaliana\] 16892\_at At5g45890 3.2112E-09 0.0000E+00 1.7785E-10 1.3353E-06 Cysteine protease SAG12 identical to senescence-specific protein SAG12 14 PROTEIN FATE (folding, modification, destination) 14420\_at At2g31980 1.3121E-03 2.8566E-13 3.1987E-06 2.4016E-02 Cysteine proteinase inhibitor B (cystatin B) -related 17282\_s\_at At3g51810 1.4069E-08 1.5599E-13 6.4057E-10 4.8094E-06 Embryonic abundant protein AtEm1 20682\_g\_at At4g26740 6.4621E-04 2.8866E-15 1.0571E-06 7.9364E-03 Embryo-specific protein 1 (ATS1) putative Ca^2+^-binding EF-hand protein 16892\_at At5g45890 3.2112E-09 0.0000E+00 1.7785E-10 1.3353E-06 Cysteine protease SAG12 identical to senescence-specific protein SAG12 20681\_at At4g26740 1.0968E-05 8.7708E-15 3.7368E-06 2.8056E-02 Embryo-specific protein 1 (ATS1) 17310\_at At3g51810 4.4370E-06 4.0301E-14 5.2274E-09 3.9248E-05 Embryonic abundant protein AtEm1 30 CELLULAR COMMUNICATION/SIGNAL TRANSDUCTION MECHANISM 18958\_s\_at At3g15410 1.0215E-06 1.6373E-08 4.1125E-08 3.0876E-04 Leucine rich repeat protein family contains leucine rich-repeat (LRR) domains 19435\_at At4g00240 3.9561E-08 2.8820E-06 4.0132E-07 3.0131E-03 Phospholipase D -related 18958\_s\_at At3g15410 1.0215E-06 1.6373E-08 4.1125E-08 3.0876E-04 Leucine rich repeat protein family contains leucine rich-repeat (LRR) domains 20682\_g\_at At4g26740 6.4621E-04 2.8866E-15 1.0571E-06 7.9364E-03 Embryo-specific protein 1 (ATS1) putative Ca^2+^-binding EF-hand protein 20681\_at At4g26740 1.0968E-05 8.7708E-15 3.7368E-06 2.8056E-02 Embryo-specific protein 1 (ATS1) 11 TRANSCRIPTION 12731\_f\_at At2g26960 1.1209E-09 4.2244E-09 1.6659E-09 1.2507E-05 MYB family transcription factor 17710\_at At2g28340 7.6288E-08 2.4759E-06 6.9175E-07 5.1936E-03 GATA zinc finger protein and genefinder 20375\_at At1g48130 2.0800E-05 3.1086E-15 1.2134E-07 9.1102E-04 Peroxiredoxin identical to SP:O04005 from \[Arabidopsis thaliana\] 16892\_at At5g45890 3.2112E-09 0.0000E+00 1.7785E-10 1.3353E-06 Cysteine protease SAG12 identical to senescence-specific protein SAG12 02 ENERGY 16892\_at At5g45890 3.2112E-09 0.0000E+00 1.7785E-10 1.3353E-06 Cysteine protease SAG12 identical to senescence-specific protein SAG12 12 PROTEIN SYNTHESIS 17871\_at At2g16360 9.2371E-07 9.4722E-08 8.4690E-09 6.3585E-05 40S ribosomal protein S25 (RPS25A) 34 INTERACTION WITH THE CELLULAR ENVIRONMENT 20375\_at At1g48130 2.0800E-05 3.1086E-15 1.2134E-07 9.1102E-04 Peroxiredoxin identical to SP:O04005 from \[Arabidopsis thaliana\] 36 INTERACTION WITH THE ENVIRONMENT (systemic) 20375\_at At1g48130 2.0800E-05 3.1086E-15 1.2134E-07 9.1102E-04 peroxIredoxin identical to SP:O04005 from \[Arabidopsis thaliana\] ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### The 65 genes with variable expression patterns among the five accessions ::: Functional category ATH1 hits GenBank ID Description ------------------------------------------------------------------------------------ ----------- ------------------- ------------------------------------------------------------------------------------------------------------------------------- 30 CELLULAR COMMUNICATION/SIGNAL TRANSDUCTION MECHANISM 14807\_at At2g17170 gb\\|AAD25145.1\\| Protein kinase family contains protein kinase domain, Pfam:PF00069 12528\_at At2g22200 gb\\|AAD23620.1\\| AP2 domain transcription factor 16848\_at At2g20470 gb\\|AAD25647.1\\| Protein kinase, putative contains protein kinase domain, Pfam:PF00069 15069\_s\_at At2g28060 gb\\|AAC98460.1\\| AKINbeta3 protein, protein kinase-related 12358\_at At1g54610 gb\\|AAC64876.1\\| Similar to CRK1 protein GI:7671528 from \[Beta vulgaris\] 18510\_at At1g60630 gb\\|AAB71975.1\\| Leucine rich repeat protein family, similar to receptor kinase GI:498278 from \[Petunia integrifolia\] 16881\_at At1g69990 gb\\|AAB61113.1\\| Leucine-rich repeat transmembrane protein kinase, putative 18478\_at At1g78530 gb\\|AAD30583.1\\| Protein kinase family contains protein kinase domain, Pfam:PF00069 17223\_at At1g78980 gb\\|AAC17069.1\\| Leucine-rich repeat transmembrane protein kinase, putative 16801\_s\_at At4g29990 emb\\|CAB43834.1\\| Identical to light repressible receptor protein kinase 16849\_at At4g36070 emb\\|CAA18501.1\\| Calcium-dependent serine/threonine protein kinase isoform AK1 11 TRANSCRIPTION 18443\_at At2g03060 gb\\|AAC32924.1\\| MADS-box protein 14963\_at At1g09920 gb\\|AAB60744.1\\| Expressed protein, TRAF-type zinc finger-related 19242\_at At2g13570 gb\\|AAD22680.1\\| CCAAT-box binding trancription factor -related 12528\_at At2g22200 gb\\|AAD23620.1\\| AP2 domain transcription factor 12220\_at At2g20100 gb\\|AAD24387.1\\| Expressed protein, bHLH - like protein (bHLH133) 14313\_at At2g26130 gb\\|AAC31224.1\\| Hypothetical protein, zinc finger (C3HC4-type RING finger) family protein 16175\_g\_at At2g29610 gb\\|AAC35234.1\\| F-box protein family contains Pfam profile PF00646: F-box domain 14370\_at At1g54550 gb\\|AAD25633.1\\| F-box protein family contains Pfam:PF00646 F-box domain 14760\_at At3g46800 emb\\|CAB51185.1\\| CHP-rich zinc finger protein, putative 16209\_s\_at At4g10240 emb\\|CAB39777.1\\| CONSTANS B-box zinc finger family protein 14216\_at At5g01290 gb\\|AAD56326.1\\| mRNA capping enzyme - like protein mRNA capping enzyme (HCE), Homo sapiens 18169\_at At4g31615 emb\\|CAA19761.1\\| Transcriptional factor B3 family low similarity to reproductive meristem gene 1 from \[Brassica oleracea var. botrytis\] 12282\_at At5g44800 gb\\|AAC79140.1\\| Chromodomain-helicase-DNA-binding (CHD) protein family similar to chromatin remodeling factor CHD3 (PICKLE) 20 CELLULAR TRANSPORT, TRANSPORT FACILITATION AND TRANSPORT ROUTES 20248\_at At2g14670 gb\\|AAC69375.1\\| Sucrose transporter (sucrose-proton symporter), putative 18549\_s\_at At2g22950 gb\\|AAF18608.1\\| Potential calcium-transporting ATPase 7, plasma membrane-type 19487\_at At2g25580 gb\\|AAD31361.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 17363\_s\_at At2g32830 dbj\\|BAA24280.1\\| Identical to inorganic phosphate transporter (PHT5) 17242\_at At2g35540 gb\\|AAC36167.1\\| DnaJ domain-containing protein, contains Pfam profile PF00226: DnaJ domain 12389\_at At1g78720 gb\\|AAC83037.1\\| Protein transport protein sec61 alpha subunit -related 18196\_at At4g14820 emb\\|CAB10261.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 19255\_at At4g20770 emb\\|CAB45843.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 16748\_s\_at At4g21300 emb\\|CAA17548.1\\| Pentatricopeptide (PPR) repeat-containing protein contains INTERPRO:IPR002885 PPR repeats 15355\_s\_at At4g21560 emb\\|CAB36800.1\\| Expressed protein hypothetical protein YPL065w yeast, PIR2:S60925 70 SUBCELLULAR LOCALIZATION 18443\_at At2g03060 gb\\|AAC32924.1\\| MADS-box protein 12528\_at At2g22200 gb\\|AAD23620.1\\| AP2 domain transcription factor 12358\_at At1g54610 gb\\|AAC64876.1\\| Similar to CRK1 protein GI:7671528 from \[Beta vulgaris\] 12389\_at At1g78720 gb\\|AAC83037.1\\| Protein transport protein Sec61 alpha subunit -related 15486\_at At4g01880 gb\\|AAD22650.1\\| Expressed protein 12282\_at At5g44800 gb\\|AAC79140.1\\| Chromodomain-helicase-DNA-binding (CHD) protein family similar to chromatin remodeling factor CHD3 (PICKLE) 14 PROTEIN FATE (folding, modification, destination) 19487\_at At2g25580 gb\\|AAD31361.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 12655\_at At2g31780 gb\\|AAD32294.1\\| Ariadne protein from Drosophila -related 17242\_at At2g35540 gb\\|AAC36167.1\\| DnaJ domain-containing protein, contains Pfam profile PF00226: DnaJ domain 19797\_at At1g64030 gb\\|AAC27146.1\\| Serpin family similar to phloem serpin-1 \[Cucurbita maxima\] GI:9937311 12389\_at At1g78720 gb\\|AAC83037.1\\| Protein transport protein sec61 alpha subunit -related 18408\_s\_at At4g03360 gb\\|AAD14465.1\\| Ubiquitin family contains INTERPRO:IPR000626 ubiquitin domain 18196\_at At4g14820 emb\\|CAB10261.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 19255\_at At4g20770 emb\\|CAB45843.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 16748\_s\_at At4g21300 emb\\|CAA17548.1\\| Pentatricopeptide (PPR) repeat-containing protein contains INTERPRO:IPR002885 PPR repeats 32 CELL RESCUE, DEFENSE AND VIRULENCE 16175\_g\_at At2g29610 gb\\|AAC35234.1\\| F-box protein family contains Pfam profile PF00646: F-box domain 17242\_at At2g35540 gb\\|AAC36167.1\\| DnaJ domain-containing protein, contains Pfam profile PF00226: DnaJ domain 14370\_at At1g54550 gb\\|AAD25633.1\\| F-box protein family contains Pfam:PF00646 F-box domain 12358\_at At1g54610 gb\\|AAC64876.1\\| Similar to CRK1 protein GI:7671528 from \[Beta vulgaris\] 16803\_at At1g61230 gb\\|AAB71472.1\\| Jacalin lectin family similar to myrosinase-binding protein homolog 17294\_at At4g19500 emb\\|CAA16927.2\\| Disease resistance protein (TIR-NBS-LRR class), putative 17306\_at At5g35940 gb\\|AAB63636.1\\| Jacalin lectin family similar to myrosinase-binding protein homolog 42 BIOGENESIS OF CELLULAR COMPONENTS 19487\_at At2g25580 gb\\|AAD31361.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 20031\_at At4g14310 emb\\|CAB10210.1\\| Expressed protein, peroxisomal membrane protein-related 18196\_at At4g14820 emb\\|CAB10261.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 19255\_at At4g20770 emb\\|CAB45843.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 16748\_s\_at At4g21300 emb\\|CAA17548.1\\| Pentatricopeptide (PPR) repeat-containing protein contains INTERPRO:IPR002885 PPR repeats 17733\_at At4g28090 emb\\|CAB36778.1\\| Pectinesterase (pectin methylesterase), putative, similar to pollen-specific BP10 protein \[SP\\|Q00624\] \[Brassica napus\] 17586\_at At5g16850 gb\\|AAD54777.1\\| Telomerase reverse transcriptase 12282\_at At5g44800 gb\\|AAC79140.1\\| Chromodomain-helicase-DNA-binding (CHD) protein family similar to chromatin remodeling factor CHD3 (PICKLE) 01 METABOLISM 17817\_at At2g23096 gb\\|AAC17826.1\\| Oxidoreductase -related temporary gene name assignment 18423\_at At1g51260 gb\\|AAD30638.1\\| Acyl-CoA:1-acylglycerol-3-phosphate acyltransferase, putative 12358\_at At1g54610 gb\\|AAC64876.1\\| Similar to CRK1 protein GI:7671528 from \[Beta vulgaris\] 13726\_at At1g74800 gb\\|AAD55296.1\\| Galactosyltransferase family contains Pfam profile: PF01762 galactosyltransferase 19038\_at At3g52160 emb\\|CAB41336.1\\| Beta-ketoacyl-CoA synthase family protein 17646\_at At4g20080 emb\\|CAA16616.1\\| C2 domain-containing protein contains INTERPRO:IPR000008 C2 domain 14274\_at At5g20980 emb\\|CAB38313.1\\| 5-Methyltetrahydropteroyltriglutamate-homocysteine S-methyltransferase - like protein 16 PROTEIN WITH BINDING FUNCTION OR COFACTOR REQUIREMENT (structural or catalytic) 12655\_at At2g31780 gb\\|AAD32294.1\\| Ariadne protein from Drosophila-related 18510\_at At1g60630 gb\\|AAB71975.1\\| Leucine rich repeat protein family, similar to receptor kinase GI:498278 from \[Petunia integrifolia\] 16881\_at At1g69990 gb\\|AAB61113.1\\| Leucine-rich repeat transmembrane protein kinase, putative 17223\_at At1g78980 gb\\|AAC17069.1\\| Leucine-rich repeat transmembrane protein kinase, putative 16801\_s\_at At4g29990 emb\\|CAB43834.1\\| Identical to light repressible receptor protein kinase 12282\_at At5g44800 gb\\|AAC79140.1\\| Chromodomain-helicase-DNA-binding (CHD) protein family similar to chromatin remodeling factor CHD3 (PICKLE) 10 CELL CYCLE AND DNA PROCESSING 12655\_at At2g31780 gb\\|AAD32294.1\\| Ariadne protein from DROSOPHILA -related 17242\_at At2g35540 gb\\|AAC36167.1\\| DnaJ domain-containing protein, contains Pfam profile PF00226: DnaJ domain 12358\_at At1g54610 gb\\|AAC64876.1\\| Similar to CRK1 protein GI:7671528 from \[Beta vulgaris\] 12282\_at At5g44800 gb\\|AAC79140.1\\| Chromodomain-helicase-DNA-binding (CHD) protein family similar to chromatin remodeling factor CHD3 (PICKLE) 02 ENERGY 19487\_at At2g25580 gb\\|AAD31361.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 18196\_at At4g14820 emb\\|CAB10261.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 19255\_at At4g20770 emb\\|CAB45843.1\\| Pentatricopeptide (PPR) repeat-containing protein contains Pfam profile PF01535: PPR repeat 16748\_s\_at At4g21300 emb\\|CAA17548.1\\| Pentatricopeptide (PPR) repeat-containing protein contains INTERPRO:IPR002885 PPR repeats 38 TRANSPOSABLE ELEMENTS, VIRAL AND PLASMID PROTEINS 16879\_at At2g05550 gb\\|AAD24652.1\\| non-LTR retroelement reverse transcriptase -related 15400\_at At4g08110 gb\\|AAD27901.1\\| Expressed protein, CACTA-like transposase family (Ptta/En/Spm) 17201\_at At4g13120 emb\\|CAB41922.1\\| Hypothetical protein 40 CELL FATE 12389\_at At1g78720 gb\\|AAC83037.1\\| Protein transport protein sec61 alpha subunit -related 13058\_s\_at At4g17580 emb\\|CAB10538.2\\| Similar to SP\\|Q9LD45 Bax inhibitor-1 (BI-1) (AtBI-1) 41 DEVELOPMENT (systemic) 18443\_at At2g03060 gb\\|AAC32924.1\\| MADS-box protein 12389\_at At1g78720 gb\\|AAC83037.1\\| Protein transport protein sec61 alpha subunit -related 34 INTERACTION WITH THE CELLULAR ENVIRONMENT 12358\_at At1g54610 gb\\|AAC64876.1\\| Similar to CRK1 protein GI:7671528 from \[Beta vulgaris\] 36 INTERACTION WITH THE ENVIRONMENT (Systemic) 12389\_at At1g78720 gb\\|AAC83037.1\\| Protein transport protein sec61 alpha subunit -related 12 PROTEIN SYNTHESIS 16667\_at At3g48960 emb\\|CAB51060.1\\| 60S ribosomal protein L13 (RPL13C) ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### The combined numbers of polymorphisms and the mutation rates in the promoters, ORFs and exons of seven genes showing high variation in expression ::: Accession ID/polymophisms Description Promoter ORF Exon Promoter ORF Exon --------------------------- --------------------------------------------------------------------------- ---------- ------ ------ ---------- ------- ------- At1g28210 Mitochondrial protein (AtJ1), putative 33 23 4 43 32 4 At2g32930 CCCH Zn-finger protein 1 2 0 1 3 1 At2g34290 Putative protein kinase 1 11 11 10 21 21 At3g13445 Transcription initiation factor TFIID-1 (TATA sequence-binding protein 1) 7 2 0 9 3 3 At4g10160 Putative RING Zn-finger protein 7 46 15 16 57 20 At4g39410 WRKY family transcription factor 3 12 0 6 12 6 At2g18790 Phytochrome B (PHYB) photoreceptor 9 10 7 21 82 71 Total number 61 106 37 106 210 126 Rate per kb 8.06 6.11 4.08 14.00 12.10 13.90 :::	PubMed Central	2024-06-05T03:55:56.350395	2005-3-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088960/", "journal": "Genome Biol. 2005 Mar 15; 6(4):R32", "authors": [ { "first": "Wenqiong J", "last": "Chen" }, { "first": "Sherman H", "last": "Chang" }, { "first": "Matthew E", "last": "Hudson" }, { "first": "Wai-King", "last": "Kwan" }, { "first": "Jingqiu", "last": "Li" }, { "first": "Bram", "last": "Estes" }, { "first": "Daniel", "last": "Knoll" }, { "first": "Liang", "last": "Shi" }, { "first": "Tong", "last": "Zhu" } ] }
PMC1088961	Background ========== The development of an adult from the single cell of a fertilized egg requires a complex orchestration of genes to be expressed at the right time, place, and level. Basic cellular functions require the expression of certain genes in all cells and tissues (that is, in a ubiquitous manner) while specialized functions require restricted expression of other genes in a single or small number of cells and tissues (that is, tissue specific). Both types of genes may be needed for embryonic development as well as for the function of adult cells and tissues. While the details of regulatory mechanisms will vary for individual genes, general features of promoters (and here we will restrict our focus to RNA polymerase II (Pol II) promoters) are likely to facilitate whether a gene will be expressed widely or in a restricted manner. For example, based on the limited number of genes available at the time of the analysis, promoters with CpG islands have been associated with housekeeping genes \[[@B1],[@B2]\]. It is desirable to re-examine this finding in the context of complete genomes for human and mouse and to place it in context with subsequent findings such as the association of CpG islands with embryonic expression \[[@B3]\]. Furthermore, it would also be informative to examine the relationship of CpG islands to the base composition of promoters, and the distribution of motifs thought to be bound by factors closely involved with (or part of) the basal transcription complex. The distribution of major components of the core promoter, the TATA box (TBP/TFIID binding site) and initiator element (Pol II binding site, Inr) \[[@B4]\], and proximal elements such as Yin-Yang 1 (YY1) site \[[@B5]-[@B8]\], among genes is not yet well understood. In addition, the functional correlations with tissue specificity and promoter structure are largely unknown beyond the CpG island association. Our goal is to place these components together in general models for tissue specificity using genome-wide surveys of expression in many tissues. Investigators have searched for combinations of transcription-factor-binding sites that confer tissue-specific expression on particular cell types such as muscle \[[@B9]\] or liver \[[@B10]\] in mammals, or in body plan specification in the fruit fly \[[@B11],[@B12]\] (see \[[@B13]\] for a review). In support of these efforts, analyses of genome-wide expression data have largely focused on identifying common patterns for particular tissues, disease states or signaling inputs. For microarray data, investigators have begun defining these patterns, largely through the application of clustering algorithms \[[@B14],[@B15]\]. Our approach is to rank genes in the spectrum of tissue specificity that runs from expression restricted to one tissue to uniform ubiquitous expression. We can study in detail the distribution of human and mouse genes across the spectrum of tissue specificity and use this to identify commonalities and differences in their promoters with the available complete genome sequences \[[@B16]\], libraries enriched for full-length cDNAs \[[@B17]-[@B19]\] and genome-wide surveys of gene expression using microarrays \[[@B14],[@B20]-[@B24]\], SAGE \[[@B25]\], mRNAs \[[@B18]\] and expressed sequence tags (ESTs) \[[@B26]\]. We validate patterns discovered in human sequence and expression data by comparison to similar mouse data. Measures have been developed for overall tissue specificity \[[@B3],[@B27],[@B28]\] that amount to counting the number of tissues that express a gene. These are really measuring tissue restriction, as they do not consider any bias in the expression levels across the tissues that express the gene. Most specificity measures for a particular tissue are equivalent to the relative expression in a tissue compared to the total expression in all tissues considered, (see, for example \[[@B29]\]). We assert that overall tissue specificity measures should take into account the levels of expression in different tissues, not just presence and absence, and that specificity measures for particular tissues should consider the distribution of expression among all tissues in addition to the tissue of interest. Such measures would enable the correct identification of genes as specific for a tissue when that tissue is not the primary site of expression but there are only a few other tissues where the gene is expressed. A metric for characterizing the breadth and uniformity of the expression pattern of a gene that meets our criteria is the Shannon information theoretic measure entropy. Although entropy has been used previously to identify potential drug targets \[[@B30],[@B31]\] by considering the entropy of the variation of expression levels and to cluster microarray data \[[@B32]\], our direct application of entropy to measuring tissue specificity is unique. Entropy (H) measures the degree of overall tissue specificity of a gene, but does not indicate whether it is specific to a particular tissue. To quantify categorical tissue specificity, we introduce a new statistic (Q) that incorporates overall tissue specificity and relative expression level. We demonstrate that Hand Qare effective metrics for ranking and selecting genes according to tissue specificity and then proceed to use them to investigate promoter features (CpG islands, base composition, transcription factor motifs) that may be used distinguish tissue-specific genes from nonspecific genes. The association of promoter features with a quantitative assessment of tissue specificity using Hand Qis an important step towards developing models for promoter function. Results ======= Defining tissue specificity --------------------------- We begin by defining the measurement of two kinds of tissue specificity, \'overall\' tissue specificity and \'categorical\' tissue specificity. (To avoid confusion we will always use the words \'specificity\' and \'specific\' to refer to the degree of tissue-restricted expression a gene exhibits and never as a synonym for the word \'particular\'.) Overall tissue specificity ranks a gene according to the degree to which its expression pattern differs from ubiquitous uniform expression. We use the term \'ubiquitous\' expression to mean expression at any level above background in all tissues. Categorical tissue specificity places special emphasis on a particular tissue of interest and ranks a gene according to the degree to which its expression pattern is skewed toward expression in only that particular tissue. In both cases, a gene\'s specificity to a tissue, cell type or other condition is decreased as the gene is more uniformly expressed in a wider variety of conditions. In addition, the categorical tissue specificity should decrease as the tissue of interest becomes a smaller component of the overall expression pattern of the gene. Given a static multi-tissue expression profile for a gene, there are at least two dimensions along which we can assess the profile to measure tissue specificity. The first dimension is the number of tissues that express the gene above some background level. It can be argued that this dimension measures tissue restriction, that is, a gene shows restricted expression if it is expressed in only a subset of tissues. The second dimension is the uniformity of expression over all tissues that express the gene. A gene that shows significant non-uniform expression is exhibiting tissue-dependent regulation, in addition to any tissue restriction that may be occurring. We assume that a gene that exhibits no tissue-specific regulation will be expressed at the same level in every tissue. We do not assert that such genes are not regulated, only that they are regulated in a way that is not sensitive to tissue. The term \'most tissue-specific\' will refer to the range of genes that are closer to the extreme of expression in a single tissue than to the extreme of ubiquitous uniform expression. We will refer to genes close to the uniform and ubiquitous end as either \'least tissue-specific\' or \'nonspecific\' though the latter term may not be strictly true. The range in the middle will be termed \'semi-tissue specific\'. The term \'housekeeping\' has been applied to genes that are widely expressed and may show little tissue-specific changes in expression level. We can use such genes as an example of genes that will tend to be ubiquitously and uniformly expressed and thus ought to be nonspecific on average. We will use the phrase \'gene sharing\' to refer to the situation that occurs when a gene is tissue-specific, and is expressed in a small number of tissues that can be said to share the gene. Measuring tissue specificity with entropy ----------------------------------------- We used two gene-expression datasets to evaluate our methods; Affymetrix-based data from the GNF Gene Expression Atlas (GNF-GEA) \[[@B22]\] and the distribution of source tissues for EST libraries in the clusters and assemblies of ESTs in the DoTS mouse and human gene index \[[@B33]\]. As described in Materials and methods, the GNF-GEA data were used as provided; EST counts in the DoTS gene index were adjusted with pseudocounts and normalized to account for the different number of ESTs sampled from each tissue across all libraries. Given expression levels of a gene in Ntissues, we defined the relative expression of a gene gin a tissue tas p~t\\|g~= w~g,t~/∑~1\ ≤\ t≤\ N~w~g,t~where w~g,t~is the expression level of the gene in the tissue. The entropy \[[@B34]\] of a gene\'s expression distribution is H~g~= ∑~1\ ≤\ t≤\ N~- p~t\\|g~log~2~(p~t\\|g~). H~g~has units of bits and ranges from zero for genes expressed in a single tissue to log~2~(N) for genes expressed uniformly in all tissues considered. The maximum value of H~g~depends on the number of tissues considered so we will report this number when appropriate. Because we use relative expression the entropy of a gene is not sensitive to the absolute expression levels. To measure categorical tissue specificity we define Q~g\\|t~= H~g~- log~2~(p~t\\|g~). The quantity -log~2~(p~t\\|g~) also has units of bits and has a minimum of zero that occurs when a gene is expressed in a single tissue and grows unboundedly as the relative expression level drops to zero. Thus Q~g\\|t~is near its minimum of zero bits when a gene is relatively highly expressed in a small number of tissues including the tissue of interest, and becomes higher as either the number of tissues expressing the gene becomes higher, or as the relative contribution of the tissue to the gene\'s overall pattern becomes smaller. By itself, the term -log~2~(p~t\\|g~) is equivalent to p~t\\|g~. Adding the entropy term serves to favor genes that are not expressed highly in the tissue of interest, but are expressed only in a small number of other tissues. As described earlier, we want to consider such genes as categorically tissue-specific since their expression pattern is very restricted. Figure [1](#F1){ref-type="fig"} shows examples of patterns of GNF-GEA expression data for different values of H~g~and Q~g\\|t~. The top five genes specific to mouse amygdala, lymph node, and liver as assessed by this data are listed in Table [1](#T1){ref-type="table"}. Tables of H~g~and Q~g\\|t~values for all genes in all tissues in the GNF-GEA datasets are available in Additional data files 1 and 2. To compare results from microarray and EST-based expression data we mapped the tissues from the GNF-GEA study to the hierarchical controlled vocabulary of anatomical terms used by DoTS and chose a set of 45 tissue terms grouped into 32 groups shown in Table [2](#T2){ref-type="table"}. In both cases, the vast majority of genes are widely expressed as measured by H~g~as shown in Figure [2a](#F2){ref-type="fig"}. Of the 7,714 probe sets in the GNF-GEA data with an average normalized intensity value above 50 arbitrary units (AU), 6,167 (80%) of genes had H~g~≥ 4 bits, which implies expression in at least 16 tissues and typically corresponds to wider, but uneven, expression. Only 87 (2%) of genes had H~g~≤ 1.5 bits, which corresponds to expression in as few as three tissues. Both microarray- and EST-based data yielded similar overall curves. The EST curve peaked at a lower H~g~than the microarray curve. This was due to the small numbers of EST sequences in some of the tissues we considered; EST counts for tissues ranged from 1,933 in the adrenal gland to 331,582 in the central nervous system (CNS). Genes that are ubiquitously expressed may not have ESTs from several of the lightly sequenced tissues, making them appear to have more restricted expression, and hence a lower entropy, than they really do. Figure [2b](#F2){ref-type="fig"} shows the correlation between estimates of H~g~derived from microarray and EST data. Visual inspection of the plot reveals that while there are no strong contradictions between the two methods, quantitative agreement is limited. Detailed analysis shows that the standard deviation of the difference of paired H~g~values is 0.61 bits. Under the null hypothesis that the estimates from the two data sources are totally uncorrelated the average standard deviation was found to be 0.91 bits. We can reject the null hypothesis (P\< 10^-5^as estimated by Monte Carlo methods). The distribution of Q~g\\|t~for selected tissues is shown in Figure [2c](#F2){ref-type="fig"}. These curves can be used to characterize tissues in terms of the number of tissue-specific genes and the amount of gene sharing; for example, liver has a relatively large number of genes shared with a small number of other tissues. In contrast, there were no genes in this set that are uniquely expressed in the amygdala. It is important to determine how well the H~g~and Q~g\\|t~statistics can be estimated from a dataset to determine the smallest meaningful difference in scores and to guide interpretation of gene rankings. To assess the standard deviations of and H~g~and Q~g\\|t~, we sampled from the replicates in the GNF-GEA microarray data to compute a large number of H~g~values for each probe set. We found that the standard deviation for H~g~was less than 0.2 bits for 97% of genes. Q~g\\|t~was not estimated as well; the standard deviation was 1 bit or less for 95% of gene and tissue pairs. This was probably due to the high standard deviation of the -log~2~(p~t\\|g~) term for low expressing gene-tissue pairs. We found much more variation when we measure reproducibility by considering genes that have two or more probe sets (and therefore two or more different transcripts) in the microarray data. In this case, the standard deviation of H~g~estimates was as high as 1 bit for 97% of the genes but less than 0.3 bits for about 70-80% of the genes. We chose a minimum of 1 bit for H~g~bins and 2 bits for Qbins in the rest of the analyses that require binning. This bin size ensured that most of the genes are in the proper bin and thus the bin could be reliably used to determine associations with the tissue specificity of a class of genes. Evaluating a set of housekeeping genes -------------------------------------- A test of the H~g~and Q~g\\|t~statistics is to determine values for a set of nonspecific genes such as housekeeping genes. A list of 797 human housekeeping genes \[[@B35]\] was evaluated using these statistics based on the GNF-GEA dataset using RefSeq accession numbers to identify appropriate probe sets. The housekeeping genes had a mean H~g~= 4.6 ± 0.27 bits in a set of 27 tissues with a maximum H= lg(27) = 4.75 bits; thus they are nonspecific as expected. Interestingly, a small number of these genes did show some degree of tissue specificity yet were ubiquitously expressed. For example, the median expression of NM\_021983 the major histocompatibility complex, class II DR beta 4 gene (32035\_at) is approximately 200 AU, but it shows much higher expression in a small set of tissues (spleen, thymus, lung, heart and whole blood), which lowered its entropy. A more extreme case is NM\_001502 glycoprotein 2 (zymogen granule membrane protein 2), which is expressed between 250 and 1,000 AU in all tissues except pancreas, where it is expressed at 34,183 AU. This is a ubiquitously expressed gene that entropy categorizes as specific since it showed such extreme tissue-specific induction. The housekeeping genes had a mean Q~g\\|t~= 9.5 ± 0.14 bits in the same set of tissues. The expected Qvalue for a uniformly and ubiquitously expressed gene is 2 lg(27) = 9.5 bits. Thus, the H~g~and Q~g\\|t~statistics successfully captured the expected expression properties of housekeeping genes. Most genes are regulated in a tissue-dependent manner ----------------------------------------------------- Although the housekeeping genes assessed above have relatively high entropies, they do show some small degree of overall tissue specificity. We therefore sought to determine how many genes show evidence of tissue-dependent regulation. Since random biological and experimental variation introduce fluctuations in the expression levels of genes, we made a probability model of the effect of these fluctuations on the observed entropy. The experimental variability was estimated from the GNF-GEA data using all normal tissues. The random tissue-to-tissue biological variability was modeled by assuming that each gene has an average expression level across all tissues and that the log base 2 of the tissue-dependent fold changes from the average level follow a normal distribution with mean equal to zero and some unknown, but \'small\', standard deviation(s). We obtain a conservative estimate of the number of genes showing evidence of tissue-dependent regulation by using s= 0.5, which allows for a relatively large amount of variation; up to 1.4-fold tissue-to-tissue variation around the mean expression level in about 63% of tissues and larger changes in the remaining tissues. As a threshold for selecting genes with tissue-dependent expression, we choose H~g~= 4.52 bits which has a p-value of 0.005 under the null hypothesis that all genes are uniform. We then find that 5,837/8,703 (67%) of human genes have entropies less than this and so are probably regulated in a tissue-dependent manner. If we use a more stringent definition of uniform expression that allows half as much variation in tissue-to-tissue expression levels (s= 0.25), then the threshold is H~g~= 4.62 bits and we find that 7,584/8,703 (87%) of human genes show evidence of tissue-dependent regulation. Similar results are found in mouse using all 42 distinct tissues, where the corresponding thresholds are H~g~= 5.24 bits (s= 0.5) and H~g~= 5.35 bits (s= 0.25) and the fractions of genes showing tissue-dependent expression are 5,467/7,913 (69%) and 7,482/7,913 (94%) respectively. Thus we conclude that most genes show evidence of tissue-dependent expression levels. Clustering tissues using Q ---------------------------- A test of Q~g\\|t~with respect to specific genes is to evaluate the tissues in which they rank highly (that is, have low Q) for consistency. This was accomplished by clustering tissues with similar tissue-specific genes and inspecting the clusters formed. We used 27 normal human tissues and, separately, 39 tissues from the GNF-GEA data for mouse and selected the genes (N= 3,768 human and N = 1786 mouse) that express at least 200 AU in at least one tissue and have Q~g\\|t~= 7 in at least one tissue. With these genes, we made a consensus hierarchical clustering of the tissues as shown in Figure [3](#F3){ref-type="fig"}. We found that the tissues in the nervous system, reproductive structures (excluding testis), immune system, and digestive system reliably cluster together in both species. In addition, skeletal muscle and heart clustered in mouse; the human survey did not have skeletal muscle. These results suggest that Q~g\\|t~is correctly identifying tissue-specific genes. Interestingly, testis is an outlier in both trees, indicating that the collection of genes expressed in testis are distinct from any other tissue or organ. Furthermore, H~g~and Q~g\\|t~can also be used in conjunction with a tissue hierarchy to answer more complex questions about the tissue distribution of genes such as \'what genes are specific to the brain but are widely expressed throughout the brain?\' In Table [3](#T3){ref-type="table"} we list the top five mouse genes expressed specifically but uniformly across three of the highlighted groups in Figure [3b](#F3){ref-type="fig"}. CpG islands are associated with the least tissue-specific genes --------------------------------------------------------------- It has been proposed that CpG islands are predominantly associated with promoters of housekeeping genes \[[@B2]\]. We performed a quantitative test of this hypothesis using the GNF-GEA data and determining the frequency of CpG islands in promoters as a function of H~g~. We considered only predicted CpG islands that span the start of transcription (see \[[@B3]\] for a justification of this definition), and genes that expressed at least at the median level of 200 AU (that is, were moderately expressed) in at least one tissue, and were represented by a single probe set on the Affymetrix chip used in the GNF-GEA experiments. Promoter sequences were obtained from DBTSS and were based on the 5\' ends of full-length transcripts \[[@B17]\]. We found that there is a strong, roughly linear, correlation between a gene\'s entropy H~g~and the probability that the gene will have a predicted start CpG island as shown in Figure [4](#F4){ref-type="fig"}. Start CpG islands were associated with only nine of the 100 most tissue-specific human genes as compared to 80% of the least tissue-specific genes. Similar numbers were found for mouse (7% start CpG island frequency for the 100 most tissue-specific genes; about 64% for the least tissue-specific genes). A comparison of CpG islands from the most and least tissue-specific genes did not reveal any significant difference in the overall base composition, or ratio of observed to expected CpG dinucleotides. The distribution of the position of the 5\' end point of CpG islands was also very similar for the most and least tissue-specific genes though CpG islands tend to start further upstream in the least tissue-specific genes (data not shown). Another group of genes observed to be associated with CpG islands are those expressed in the early embryo \[[@B3]\] from the fertilized egg to the blastocyst. The question arises as to whether there is an association of genes having start CpG islands and the developmental stage of expression (that is, embryonic versus adult) in addition to the one for tissue specificity. We investigated this possibility in the mouse using DoTS \[[@B33]\] EST and mRNA assemblies by tabulating the number of DoTS genes that contain at least two ESTs from a mouse early embryo library as shown in Table [4](#T4){ref-type="table"}. We considered 933 genes with start CpG islands (CGI+) and 1,007 genes without start CpG islands (CGI-) that were expressed in the adult. If there were no developmental bias, this distribution of CpG+ and CpG- genes should be maintained in genes expressed in the embryo. However, only 139 (14%) of the CGI- genes were expressed in the early embryo in contrast to 365 (39%) CGI+ genes (P= 3 × 10^-70^exact binomial). Therefore, a gene expressed in the adult was 2.8 (= 0.39/0.14) times more likely to be expressed in the early embryo if it contained a start CpG island. Furthermore, the most tissue-specific genes expressed in the adult were four times more likely to have been expressed in the early embryo if their promoter contained a start CpG island. These results strongly suggest that CpG islands are promoter features for both embryonic and the least tissue-specific genes. Base composition of promoters depends on specificity ---------------------------------------------------- Analysis of base-composition profiles of promoters provides clues to common features, including motifs associated with promoter categories. We examined the base composition profiles of human promoters of high (0 ≤ H~g~≤ 3.5 bits) and low (4.4 ≤ H~g~≤ 4.71 bits) tissue-specificity genes. We considered CGI+ and CGI- genes separately, as it is clear the presence of a CpG island will strongly influence the base composition and that the fraction of start CpG islands varies with entropy. In addition, the presence of a start CpG island may indicate a different regulation mechanism related to either tissue specificity or embryonic expression (or both). The number of promoters from DBTSS in these four classes that were used in the analysis were: 310 CGI- and 129 CGI+ high specificity; 342 CGI- and 1,501 CGI+ low specificity. Genes that have only non-start CpG islands represented a minor component and were not included in this analysis. We used the full set of normal tissues in the first GNF-GEA microarray study for human and mouse. Base composition profiles with 10 base-pair (bp) windows are shown in Figure [5](#F5){ref-type="fig"} for human genes. Each of the features we report were observed in human and mouse (unless noted otherwise) and compare G to C or A to T over spans of at least 10 positional bins; the probability of observing a feature at least this long by chance is less than 0.5^10^which is equivalent to 0.001. Promoters of CGI+ genes (Figure [5a,b](#F5){ref-type="fig"}) shared features but could also be distinguished on the basis of tissue specificity. A common feature of CGI+ promoters was the increase in C+G content that starts at 1,000 bp upstream of the transcription start site and continues at 200 bp downstream. The C+G bias reached p(C+G) = 0.7 at the start of transcription and continued into the 5\' UTR. Nonspecific (Figure [5c](#F5){ref-type="fig"}) and tissue-specific (Figure [5d](#F5){ref-type="fig"}) CGI- genes still showed a C+G bias around the start of transcription, but it was much smaller in magnitude at p(C+G) = 0.54. The low specificity CGI+ genes (Figure [5a](#F5){ref-type="fig"}) showed upstream base composition biases that were not found in any of the other three gene classes. There was a preference for C over G (p(C) \> p(G)) in the (-350, -150) region and also a preference for p(A) \> p(T) in the -600, -200 region in human (this region is located (-400, -150) in mouse). In tissue-specific CGI+ (Figure [5b](#F5){ref-type="fig"}) genes the strong C+G bias held but p(C) = p(G), except for the (+50, +100) region where p(C) \> p(G). These base-composition differences observed between nonspecific and tissue-specific promoters over regions of hundreds of base-pairs, even in the context of a CpG island, suggest different structural features and regulatory mechanisms for these CGI+ classes. Most striking were differences between nonspecific and tissue-specific promoters that are independent of the presence of a CpG island. A sharp spike in the proportion of A and T was seen in the (-50,-1) region for all classes but was most pronounced in the tissue-specific promoters (Figure [5b,d](#F5){ref-type="fig"}). These spikes correspond to the presence of a TATA box and suggest a correlation of this motif with tissue-specific genes (explored more fully later). Conversely, all low-specificity genes (Figure [5a,c](#F5){ref-type="fig"}) shared a common feature in the (+1, +200) region where p(G) \> p(C) and p(T) \> p(A) that was not seen in tissue-specific genes (Figure [5b,d](#F5){ref-type="fig"}). As shown later, this low-specificity feature could be partially explained by the presence of a YY1 motif. These base-composition differences observed between nonspecific and tissue-specific promoters are likely to indicate motifs that distinguish the two classes. Selected transcription factor motifs in the core promoter --------------------------------------------------------- We next examined the distribution of basic core promoter features: the TATA box, the initiator element, and two binding sites for selected ubiquitous transcription factors, Sp1 and YY1, to see if their presence in the proximal promoter was correlated with the tissue specificity of a gene. Two approaches were taken using different datasets and motif-searching methods that gave similar results, providing independent confirmation of results. First, we searched for core motifs using weight matrix hits in promoters of genes selected using H~g~calculated from the GNF-GEA data. Second, we searched for core motif consensus sites in promoters of genes selected using Q~g\\|t~calculated from EST data. TATA boxes are associated with tissue-specific genes ---------------------------------------------------- We grouped the human genes that expressed at least 200 AU (average value) in the GNF-GEA data by entropy and start CpG island status. The number of genes in each category is shown in Table [5](#T5){ref-type="table"} along with a summary of results. We used alignments of position-specific scoring matrices and scoring thresholds included in the Eukaryotic Promoter Database (EPD) \[[@B36]\] to identify the TATA box and initiator element. Matches to these motifs were preferentially located at the expected positions relative to the transcription start site based on the ratio of the number of observed set to the expected number using a set of random sequences with the same position-dependent base composition as each of the promoters. We searched for the TATA box in the (-45, -10) region where the average observed/expected ratio for the TATA box was 3.1. As shown in Table [5](#T5){ref-type="table"}, the most-specific CGI- genes were six times more likely to have a TATA box than the least-specific CGI+ genes (117/215 (54%) versus 183/2072 (9%), P≈ 0 exact binomial). Similar numbers are found in mouse (52%/11% = 4.7) This trend also holds within CGI- genes and CGI+ genes. The most specific CGI- genes were three times more likely to have a TATA box than the least specific CGI- genes (117/215 versus 110/607, P≈ 0 exact binomial). While less common in CGI+ genes, TATA boxes were still almost four times as likely to be found in the most specific CGI+ genes than the least specific CGI+ genes (19/56 versus 183/2,072, P= 2 × 10^-7^exact binomial). Thus TATA boxes are clearly associated with tissue-specific genes and provide a second axis (with CpG islands) for distinguishing between the most and least specific genes. In contrast, the frequency of occurrences of the initiator element (Pol II binding site) was roughly constant across all tissue-specificity classes for both CGI+ and CGI- genes. We searched for the initiator element in the (-10, +10) region. It occurred in 762 of 1,118 (68%) of CGI- genes and 1,273 of 2,434 (52%) of CGI+ genes. Similarly, it occurred in 149 of 215 (69%) of the most specific genes and 388 of 607 (64%) of CGI+ genes. The observed frequency of TATA+/Inr+ promoters was not significantly different from the expected rate assuming independence of the two individual features (data not shown). Sp1-binding sites are weakly associated with the least tissue-specific genes ---------------------------------------------------------------------------- Sp1 \[[@B37],[@B38]\] is a ubiquitous transcription factor with a G-rich binding site with consensus sequence GGGCGGG that might explain the observed G-richness of the 5\' UTR in non-specific genes. We used the GC-box weight matrix and scoring threshold from EPD \[[@B36]\] to identify Sp1 sites. We found that Sp1 sites are preferentially located in the (-150, +1) region in all sets of genes where they occurred on average at twice the expected rate in agreement with previous findings \[[@B36]\]. In both human and mouse, Sp1 sites were rarely found in the 5\' UTR despite the G-richness of this region; they occurred at the expected rate of between 2 and 5%. Thus Sp1 sites were not the cause of the G-richness in the 5\' UTR. Sp1 sites are associated with CpG islands but are an important component of GGI- promoters as well. Considering just the (-150, +1) region, Sp1 sites occurred in 1,105/2,434 (45%) of human CGI+ gene promoters, and 316/1,118 (28%) of CGI- genes at about 2.5 to 3.0 times the expected frequency in both cases. Frequencies in mouse are 927/2075 (45%) of CGI+ promoters and 464/1652 (28%) CGI- promoters. Sp1 sites were also weakly associated with the least specific genes occurring in 1,105/2,679 (41%) of these genes as compared to 94/271 (32%) in the most tissue-specific genes (P= 0.016). Similar numbers are found in the mouse; 38% of the least specific and 26% of the most specific promoters have Sp1 sites. Thus, although Sp1 shows a preference for the least tissue-specific promoters, it is not a strong predictor of the tissue specificity of a gene. YY1 binding sites are associated with low-specificity genes ----------------------------------------------------------- The transcription factor YY1 \[[@B5]-[@B8]\] is also ubiquitously expressed and is thought to bind close to \[[@B39]\] and downstream of the transcription start site. There is evidence that the function of YY1 depends on its orientation \[[@B40]\]. The location and G-richness of the reverse complement consensus sequence (AANATGGCG) make YY1 a candidate for explaining the prominent G \> C feature in the (+1, +200) region of low-specificity genes. We consider YY1 because a YY1-like motif was frequently included among the most statistically significant motifs identified by the motif discovery programs AlignACE \[[@B41]\] and MEME \[[@B42]\] in the (+1, +60) region of nonspecific CGI+ promoters (Figure [6a](#F6){ref-type="fig"}). Our form is most similar to the activating form \[[@B43]\], which may be associated with low-specificity genes. Because of the demonstrated functional sensitivity to the orientation of binding sites we considered each orientation separately. Indeed, as shown in Figure [6b](#F6){ref-type="fig"} we found each orientation exhibits different position preferences. Sites in the reverse orientation (YY1~r~) were preferentially located in the (+1, +25) region but with some elevated levels to +80 bp. Start positions of sites in the forward orientation (YY1~f~) showed a very sharp preference for -3 bp, which probably represents a YY1-like initiator sequence reviewed elsewhere \[[@B44]\]. Both orientations were found predominantly in the least specific genes (Table [5](#T5){ref-type="table"}). YY1~f~initiator sites are rare; only 55/2,679 (2%) were found above background in human low-specificity genes. The rate in mouse, 22/2,832 (0.8%) of low-specificity promoters, is even lower. The YY1~r~sites are more common and were found above background in 217 (8%) of the 2,679 least specific genes. YY1~r~sites were more common in CGI+ genes than in CGI- genes (202/2,072 (10%) versus 15/607 (2%) P= 3.7 × 10^-9^two-population binomial). The corresponding rates in mouse confirm these observations; 178/2,832 (6%) for all low-specificity genes and 152/1,779 (9%) in CGI+ and 26/1,053 (2%) of CGI- low-specificity promoters. These YY1-like sites therefore constitute a feature strongly associated with the least specific genes and may partially explain the observed G \> C ratio in the (+1, +200) region. Q-based analysis of core promoter motifs ------------------------------------------ A second analysis of TATA box and Inr motifs was done to determine if the association of the TATA box with tissue-specific genes is also found in genes ranked by Qand is robust to using EST data as well as promoters that did not specifically rely on full-length cDNA clones. The definition of Qimplies that genes with a particular Q-value can have a variety of H~g~values and thus it may be more difficult to identify features related to tissue specificity. We tabulated all DoTS genes that contained at least two ESTs from an islet-cell library then ranked the genes by Q~pancreas~computed using EST counts. We used Q~pancreas~≤ 7 bits as the criterion for selecting pancreas-specific genes which we grouped into 2-bit Q intervals. For comparison we selected 50 genes with Q~pancreas~= 8.5 bits, and 50 genes with 10 ≤ Q~pancreas~≤ 10.6 bits. Genes with high specificity for the pancreas (0 ≤ Q~pancreas~≤ 2 bits, N= 9) preferentially had TATA boxes (8 of 9) with half of these also having an initiation element (4 of 9; Figure [7a](#F7){ref-type="fig"}). With decreasing specificity, the fraction of genes containing TATA boxes drops with only18 of 81 (2/9) genes with Q\> 6 bits having TATA boxes. Thus, the strong correlation of TATA boxes with specific genes found with H~g~and microarray data was also seen with Qand EST data for pancreas-expressed genes. Also consistent is the observation that initiator elements were found at similar frequencies (around 60%) across all specificity classes (Figure [7b](#F7){ref-type="fig"}). Similar patterns were observed in other tissues (data not shown). The consistency of findings for the TATA box with human islet genes based on Qand ESTs was next tested with orthologous genes in mouse. This test provides a measure for whether the global pattern observed (TATA box with tissue-specific genes) is also found for the same set of genes in another mammal. We also added bins of genes with higher Q-values that represent more widely expressed genes. For each human gene, the orthologous mouse gene was determined (see Materials and methods for details) and analyzed as described above. Overall, 18.8% of the human genes and 22.9% of the mouse genes that were analyzed carry the TATA box motif. Except for the last group (Q\>10 bits) the percentage of the genes with TATA box motifs decreases with the increase in the Q-value. This is to be expected since genes with high Qmay be specific to other tissues and hence are more likely to have a TATA box. Discrepancies between human and mouse promoters were noted for only about 10% of all human-mouse pairs analyzed and may reflect sequence differences and possible annotation discrepancies for the transcription start site. Nevertheless, there is overall excellent agreement for the presence of TATA motifs in human and mouse genes. Thus, our assessment of preferential presence of transcription regulatory motifs in the human pancreas-expressed genes also applies to their mouse orthologs. We conclude that genes expressed with restricted tissue-distribution may be preferentially regulated via TATA-mediated transcription, and that genes with broader expression profiles are more likely to be regulated by non-TATA mediated mechanisms (such as YY1). Promoter classes ---------------- Since the presence or absence of a start CpG island and a TATA box appear to be the primary sequence feature that correlate with tissue specificity, we consider them in more detail. We observe that CpG islands and TATA boxes are not mutually exclusive features of promoters and so we consider all possible combinations of these features. Frequency of promoter classes ----------------------------- Figure [8](#F8){ref-type="fig"} shows the cumulative fraction of each class of promoter as a function of increasing H~g~in human (Figure [8a](#F8){ref-type="fig"}) and mouse (Figure [8b](#F8){ref-type="fig"}). The data from human and mouse follow similar trends even though the mouse has a lower proportion of CGI+ genes. Overall, CGI+/TATA- genes are the most common, at 50-60% depending on the species. Interestingly, the CGI-/TATA- class is the second most common overall, comprising 20-30% of genes, depending on the species. Genes in this promoter class are roughly equally common across the entire entropy range and are the most common promoters in the mid-specificity range in both species. The classes CGI-/TATA+ and CGI+/TATA+ are the least common (8 to 12% overall). CGI-/TATA+ genes are concentrated in the most specific genes. CGI+/TATA+ are found relatively uniformly across all but the most specific genes. Although the TATA box and CpG islands are strongly predictive of a gene\'s entropy, Figure [8](#F8){ref-type="fig"} also illustrates the limitations of the promoter classes as an explanation for expression patterns. First, although the CGI-/TATA+ and CGI+/TATA- classes are strongly associated with the most and least tissue-specific genes (respectively), instances of genes in each class cover virtually the entire range of tissue specificities. Second, the CGI-/TATA- class is the second most common, illustrating that any degree of tissue specificity can be obtained without these sequence features. Functional assessment of promoter classes using Gene Ontology terms ------------------------------------------------------------------- To try to understand the functional correlates of the four promoter classes, we looked for trends in the cellular localization and biological process of the products of genes from each promoter class. We used the DAVID system \[[@B45],[@B46]\], which identifies over-represented Gene Ontology (GO) \[[@B47]\] terms in a set of genes. A summary of the results for human and mouse genes are shown in Table [6](#T6){ref-type="table"}. In each case the set of genes in each promoter class were compared to all genes on the corresponding Affymetrix chip. Products of genes in the CGI-/TATA+ class were often (70/198) located extracellularly. Examples of such genes are the insulin-like growth factor family, serum albumin and chymotrypsin. Some extracellular CGI-/TATA+ genes, such as luteinizing hormone beta (LHB) and bone morphogenetic protein 10 (Bmp10) in the mouse, have a high H~g~because they are not induced in the tissues or at the developmental stages surveyed, but otherwise fit the pattern of secreted proteins. Gene products that are secreted from the cell must be produced at high level to be effective. Indeed we found the maximum expression level of TATA+ genes is higher than TATA- genes; 454/745 (61%) of TATA+ genes express at least 1,000 AU in one or more tissues, whereas only 1,321/3,773 (35%) of TATA- genes express that highly (p-value = 0; two-sample binomial population). A second group of CGI-/TATA+ that is common, but with a p-value just over the p-value cutoff are the muscle contraction-related genes, actin, troponin and members of the myosin family. Products of these genes are also required in large amounts to create the contractile apparatus but are only produced in a few cell types. The biological processes that are enriched in the CGI-/TATA+ class differ between human and mouse, but nearly all of them are descendants of the GO term \'response to stimulus\' (GO:0050896). The CGI+/TATA- promoters produce proteins that are typically located in the cell, especially in the cytoplasm and mitochondrion. These locations are consistent with many housekeeping functions. The human results for biological process suggests a large number of housekeeping processes, but these were not confirmed in the mouse using all CGI+/TATA- genes. When we consider just the least specific CGI+/TATA- mouse genes (4.45 ≤ H~g~≤ 5.57 bits), we find cellular locations (including the nucleus) and biological processes that match the human results. No significant concentrations of cellular locations or biological processes were found among the CGI+/TATA+ genes. A manual examination of genes in both human and mouse identifies a number of heat-shock proteins, histones and ribosomal proteins although these are not statistically significant as a result of the multiple testing correction. Many of these genes fit the expected expression pattern in that they are widely expressed and at high levels. Interestingly, the products of CGI-/TATA- genes are often located in the plasma membrane (244/499 of human genes with a cellular location) and support signaling and response to the environment. Such products, for example, bradykinin receptor B2, prolactin receptor or protocadherin 9, may be expressed in a tissue-specific pattern, but not at the high levels required for secreted proteins. The exact biological process GO terms that are statistically significant vary between mouse and human, but a common core includes defense response (GO:0006952), immune response (GO:0006955) and response to stimulus (GO:0050896). Thus these genes are similar to CGI-/TATA+ genes in that they are involved in response, but are not (typically) required to be expressed at such high levels. Discussion ========== We have applied Shannon entropy as a novel measure of overall tissue specificity of gene expression and have created a new statistic Qto assess the categorical specificity of a gene for a particular tissue. We have evaluated the performance of entropy on microarray-and EST-based estimates of tissue-specific expression and found that it correctly identifies both tissue-specific and housekeeping genes. Ranking and binning genes by entropy allowed us to begin to deconstruct core promoters into features directing when and where the gene will be expressed. We verified and extended previous observations \[[@B2]\] about the correlation of CpG islands with housekeeping genes and embryonic genes. We then identified differences in the base composition profile of promoters of tissue-specific and nonspecific genes. Next, we identified correlations between, on the one hand, the TATA box and tissue-specific genes, and on the other hand, the YY1 site and nonspecific genes. Finally, we identified trends in promoter classes based on CpG island and TATA box status and associated them with common cellular locations and biological processes. Similar observations were also observed for TATA box and Q-selected genes in pancreas. The identification of an association between promoter type and cellular location and biological function, while an important step in a fundamental understanding of biology, also has practical significance, as the genes in the CGI-/TATA+ and CGI-/TATA- classes are enriched for tissue-specific extracellular and cell surface proteins. Such genes are likely to be useful drug targets. Thus entropy H~g~and Qhave allowed us to discover fundamental properties of mammalian Pol II promoters and should allow serve to aid understanding of expression in particular tissues of interest. The validity of our approach is supported by findings in other work and by the fact that they are robust with respect to the algorithm used to process the expression data. Our finding that most genes are regulated in a tissue-dependent manner is consistent with another analysis of gene expression \[[@B14]\], which found that housekeeping genes cluster in a tissue-specific manner. Thus, it appears, even the most basic biological functions are subject to regulation. The tissue trees we produced contain relationships similar to those in an analysis \[[@B48]\] of mid-specificity genes, including the close relation between lung, and the immune system-related organs spleen and thymus. That analysis is based on a different method and a different set of expression data gives us confidence that Q~g\\|t~is properly identifying genes that are specific to a tissue. The GNF-GEA expression data we analyzed was processed with the MAS4 \[[@B49]\] algorithm. We reanalyzed the data from this study after reprocessing it with the more recent Robust Multichip Average (RMA) algorithm \[[@B50]\]. This algorithm tends to suppress low-level signals and we found that most genes appeared to be more tissue specific, that is, had lower H, in the RMA-processed data compared to the reported values. Although this affects some of the precise values of numbers we have reported it does not alter any of the fundamental trends or results. We include tissue specificities based on both analyses in Additional data files 1 and 2. Our analysis focused on only a few sequence features and although we found good correlations, two aspects of our results indicate that there are other regulatory mechanisms not yet identified. First, there is a gradual transition in the frequency of the TATA box and CpG islands between the most and least tissue-specific genes. Second, while these features are strong indicators of high and low specificity, they are far from perfect predictors. Indeed, the middle range of entropies contains a mix of all promoter classes in large numbers, indicating that it is possible to achieve tissue-specific expression with any promoter class. YY1 may be an example of such a supplementary mechanism. While occurring in only 16% of genes, it is very strictly confined to low-specificity genes and is a better indicator of low specificity than CpG islands. We expect that other such signals will be found. Anatomical resolution is an issue with the datasets used in this study. For example, the pancreas consists of exocrine cells, ductal cells and islet cells of several types. The bulk pancreas was used to generate the GNF-GEA data, so the reported expression level is the average mRNA concentrations weighted by the cell-type count. This approximation reduces the maximum possible entropy and, more significantly, can make the apparent entropy different from the true entropy. Genes highly and specifically expressed in a cell type with a small population may currently appear to be ubiquitous with very low overall expression. Genes expressed in a few tissues may be revealed to be less tissue specific as more cell types are measured in detail. Genes that appear to be ubiquitously expressed may turn out to not to be expressed in a few cell types. It will be interesting to see whether data with higher anatomical resolution will help to increase the accuracy of the rules we have identified here for identifying tissue-specific and nonspecific promoters. Our method can be also applied to other sources of expression data including SAGE, reverse transcription PCR (RT-PCR) and in situhybridization data. SAGE has the advantage of sensitivity, as these studies generally sequence to much greater depths than EST libraries \[[@B51]\]. In situhybridization data may increase the anatomical resolution of the data. Qualitative intensities, for example, \'0\', \'+\', or \'+++\', can be converted to representative numeric values as appropriate. Our method can also be applied to other collections of conditions beside normal tissues, for example, different types of cancers or samples of the same cancer from multiple patients. Modification of our method to account for temporal changes in tissue specificity represents another direction for future work. The analysis presented here focuses on genes rather than on transcripts generated from different promoters from the same gene. The rate of the occurrence of alternative transcription start sites is at least 9% \[[@B52]\] and may be as high as 25% \[[@B53]\]. The promoters we used were specified by the DBTSS dataset but there may be alternative promoters with different characteristics and tissue-specific usage patterns. Analyses based on different RNA species can easily be incorporated into our approach and is an area of future research. Our results for CpG island frequency in very tissue-specific genes are lower than recent reports \[[@B3]\] that were based upon present/absent calls, that is, tissue counting, using ESTs to measure tissue specificity. This may be due to two reasons. First, as we described in Results, a significant fraction of genes will show no evidence of expression in poorly sampled tissues. A poorly sampled nonspecific gene will appear therefore more tissue specific than it actually is and this increases the number of apparently tissue-specific genes with CpG islands. Second, when we use microarray data and determine tissue specificity by counting tissues expressing above the median value of 200 AU, we see (data not shown) rates of CpG island occurrence in \'specific\' genes similar to those reported in \[[@B3]\]. Thus, we conclude that including the variation of expression levels rather than mere presence/absence is important for identifying very tissue-specific genes as assessed by start CpG islands. These results present an initial look at the correlation between tissue specificity, CpG islands and binding sites for selected transcription factors that interact with the basal transcription apparatus. Using a novel approach with entropy-based metrics, we have begun to lay out the framework for promoter function by identifying strong correlations between tissue-specific or ubiquitous expression and a number of these sequence features. We plan to extend this work in several ways. First, we plan to identify correlations with other known transcription-factor-binding sites and novel motifs identified as over-represented in promoter regions \[[@B54]\]. Second, these results will help to understand regulation by combinations of multiple upstream transcription factors in genes specific to particular tissues or clusters of tissues. Conclusions =========== We have used Shannon entropy to quantify and rank the tissue specificity of genes using tissue-survey data. First, this has allowed us to assess the prevalence of tissue-specific regulation; we find that most genes show evidence of some degree of tissue-dependent variation in expression levels. It has also allowed us to find and evaluate associations between promoter features and tissue specificity. We have verified and extended understanding of known associations between, on the one hand, CpG islands and the least tissue-specific genes and, on the other hand, the TATA box and the most tissue-specific genes. However, they are not the sole determinants of tissue-specific expression, as indicated by mid-specificity genes that exhibit a mix of all promoter classes. The class of CGI-/TATA- promoters has emerged as the second most common class of promoter overall and the most common promoter class in mid-specificity genes. Therefore, additional determinants of tissue specificity remain to be found. We have identified one potential determinant, a downstream YY1 site, which is very strongly associated with the least tissue-specific genes but is a relatively rare feature of these promoters. Finally, we have also been able to associate trends in the localization and function of protein products of genes according to their promoter class. Many of the CGI-/TATA+ genes code for highly expressed, very tissue specific, extracellular proteins involved in a cell\'s response to the environment. CGI-/TATA- genes are also involved in response to the environment, but are found more uniformly across the spectrum of tissue specificity, are not as highly expressed as CGI-/TATA+ genes, and very often code for membrane-bound proteins. CGI+/TATA- genes are more likely to be located in the cytoplasm or nucleus and, as expected, carry out housekeeping functions. All of the results we report are found in both human and mouse and so may reflect general principles of all mammalian species. Materials and methods ===================== Processing GNF-GEA \[[@B22]\] and DoTS \[[@B33]\] data ------------------------------------------------------ The GNF-GEA data are processed as described \[[@B22]\]. Given a set of Ntissues we define p~t\\|g~= w~g,t~/∑~1\ ≤\ t≤\ N~w~g,t~where w~t~is the expression level of the gene gin tissue t. DoTS, available through the AllGenes \[[@B33]\] site, contains ESTs and mRNAs assembled into transcripts that are then clustered into genes. We did not consider any transcript that contains only one EST as this may represent a spurious sequence and did not consider any gene with fewer than five ESTs because they provide a poor estimate of H~g~. To accommodate the great disparity in sampling depth across tissues we normalized EST counts by tissue. To avoid artificially low entropies for genes that contain relatively few ESTs we used pseudocounts to smooth the data. The expression level of a gene in a tissue is computed as w~g,t~= (n~g,t~+ 1)/(N~t~+ N~g~) where n~g,t~is the number of ESTs from libraries for a tissue included in a gene, N~t~is the total number of ESTs from a tissue assembled into genes, and N~g~is the number of genes. We used different sets of tissues depending on the task. H~g~and Qmeasures in Figure [1](#F1){ref-type="fig"} used the full GNF-GEA mouse set with a few modifications; adipose tissue and brown fat were merged, epidermis and snout epidermis were merged, digits and tongue were not considered as they are both a combination of skeletal muscle and epidermis. The expression level for a set of merged tissues is the median of the individual tissue replicate medians. For comparison of microarray and EST data we used a set of 27 tissues that were common to both datasets and merged the CNS and peripheral nervous system tissues. Estimating variance ------------------- To estimate the variance in Hand Q, we took advantage of tissue replicates in the GNF-GEA data. Using the mouse dataset, we repeatedly sampled one of the measurements from each pair of replicates and computed Hfor each gene. We then computed the variance of the distribution of the estimates of Hfor each gene and show the survivor distribution function in Figure [2](#F2){ref-type="fig"}. The variance of Qwas computed in a similar manner. Clustering tissues ------------------ Clustering was based on the Qscores for the set of mouse genes with Q~g\\|t~≤ 7 for at least one tissue and expressing at least 200 AU in at least one tissue in the GNF-GEA data. There were 1,786 Affymetrix probe sets selected. The tree in Figure [3](#F3){ref-type="fig"} was built by sampling 5,000 sets of 1,000 probe sets and clustering tissues using Pearson correlation and a centered measure using the XCLUSTER \[[@B55]\] program. The consensus tree was built using the program CONSENSE in the PHYLIP \[[@B56]\] package with the default parameters. Identifying genes specific to a set of tissues ---------------------------------------------- The total entropy of all tissues under a node can be computed at each node in the hierarchy using a generalization of the grouping theorem \[[@B57]\]. If the entropy of a gene at a node is close to the maximum possible entropy for the number of tissues under the node, then we select it and compute a Q~g,n~for the gene at the node. Using Q~g,n~we can rank genes by specificity to a cluster of tissues just as we can for an individual tissue. Predicting CpG islands ---------------------- We predicted CpG islands using the program NEWCGREPORT in the EMBOSS \[[@B58]\] package with the default parameters which require a minimum length of 200 bp, C+G fraction of 0.6 and ratio of observed over expected CpG of 0.5. Statistical significance in embryonic expressed genes ----------------------------------------------------- We computed statistical significance of differences between all embryonic-expressed genes and adult-specific rates using a hypergeometric distribution. We start with a collection of NCGI+ genes, n~e~of which are expressed in the embryo, that is, marked as special. The N~A~tissue-specific genes in the adult are considered a random sample from the original Nand we compute the probability of finding that at least (or at most) n~ae~of these were expressed in the embryo. Modeling distribution of entropy from uniform genes --------------------------------------------------- To model the effect of experimental variability, we computed the distribution of the difference between expression levels of individual replicates for each gene and tissue and the mean expression level across replicates as a function of the mean expression level. This distribution was well fit by an exponential distribution with a parameter that depends on the mean expression level. Thus, given an \'ideal\' expression level, we can estimate what the experimental variability will be. To model a uniformly expressed gene, we assume that a gene has some average expression level across all tissues and then allow the expression levels in individual tissues to follow a narrow distribution of random fold changes from that level. Specifically, we assumed that the log base 2 of the fold changes is distributed according to a normal distribution with mean equal to 0 and a standard deviation (s). The standard deviation can be adjusted to control the amount of biological variation a \'uniformly\' expressed gene is allowed to show. For example, setting s= 0.5 means that about 68% of the fold changes between a particular tissue and the nominal level are within 1.4 up or down from the nominal level, that is, a twofold change from the lowest to the highest levels. Larger fold changes are expected to occur in 32% of tissues. This model allows significant variation and so is arguably close to the upper limit of variation allowable for a gene that shows no tissue specificity. We also used s= 0.25 as a more stringent definition of uniform expression. We sampled mean expression levels from the distribution of observed mean expression levels and sampled entropy values from the probability model. An entropy threshold was estimated by sampling approximately 5,000 random expression profiles and determining the value for a p-value of 0.002. This process was repeated 10 times and the corresponding thresholds and fraction of genes were computed. The thresholds spanned a range of less than 0.01 bit. The tissue-dependent gene fractions never varied by more than one percentage point in either direction. Statistical significance of co-occurrence ----------------------------------------- We estimated the statistical significance of the co-occurrence of motifs using the hypergeometric distribution. Given two motifs with occurrence counts n~1~and n~2~, measured in the same set of Npromoters, and a co-occurrence count of n~12~, we compute the significance as the probability of finding no more than (or at least) n~12~hits in a random selection of n~2~promoters from a pool of Npromoters where n~1~of them are \'special\'. Comparison of frequency on independent sets ------------------------------------------- Given two sets of size N~1~and N~2~and positive observations n~1~and n~2~in each, we computed the probability that the underlying rates are different using an exact calculation of the binomial distribution to compute the probability of seeing at least (or no more) than n~i~matches in N~i~trials where the rate is assumed to be r= n~j~/N~j~. We estimated rusing the larger of the two sets. Two binomial populations ------------------------ We used the normal approximation to the difference of the proportions normalized by their variance to compute a z-score. Promoter sequences ------------------ We obtained promoter sequence in two ways. The H-based set of analyses used links from Affymetrix probe sets to RefSeq identifiers to select alignments from the DBTSS promoter sequences covering the (-1000, 200) region downloaded from the DBTSS website \[[@B59]\]. The Q-based analyses of TATA box and initiator elements used genomic locations of DoTS genes on UCSC Golden Path release mm3 \[[@B60],[@B61]\] to identify gene names. Promoter sequences consisting of the 350 bp of the upstream region were then extracted from Ensembl \[[@B62]\]. The mouse homologs were also used as annotated in Ensembl. Core motifs ----------- The H-based analysis used core promoter element models from EPD \[[@B36],[@B63]\]. The fraction of promoters containing each matrix was determined as follows for each set of genes (with and without CpG islands in each entropy bin) individually. Having verified that the positional distribution of each motif was sharply peaked at the appropriate place in the promoter sequences ((-40, -20) region for TATA and (-20, +20) region for the initiator element) we considered only the predictions in these windows from all genes. We used the log-likelihood function to score each subsequence against each matrix using the published score cut-offs. The YY1 motif was found in essentially every run of AlignACE and MEME performed on the downstream regions of ubiquitous CGI+ promoters. We explored different motif widths and other settings and selected version that achieved a combination of good coverage and conservation. In all cases we estimated the background rate of random occurrence of motifs by repeatedly scrambling the individual sequences over a 10 bp window to create approximately 1,000 test sequences for each combination of CpG island status and specificity range. These sequences were scored in the same manner as the unscrambled sequences. We estimated the statistical significance of differences of observed frequencies using exact computation of the binomial distribution. The Q-based analyses of core motifs used the TATA box motif (TATAA) and initiator element (YYANWYY). Motif searches were carried out using the tool patternmatch from the biological workbench 3.2 \[[@B64]\]. Only the TATAA instance located closest to the start of the mRNA\'s alignment to the genome was used. Matches to the initiator element were required to be downstream of the TATAA box when present. YY1 motif --------- We used an AlignACE-derived weight matrix (shown in Figure [6a](#F6){ref-type="fig"}) to assess the occurrence of YY1-like sites as it contained the YY1 consensus and was built using approximately 100 sites which is many more than previously published weight matrices \[[@B43],[@B65]\] also shown in Figure [6a](#F6){ref-type="fig"}. GO association analysis ----------------------- We submitted Affymetrix probe set ids of interest to the DAVID website \[[@B45],[@B46]\] and compared them either to all probe sets on the appropriate Affymetrix chips or to all genes in the selected entropy range. We compensated for multiple testing by requiring the reported p-values be better than either 0.05/1472 = 0.00003 (cellular component) or 0.05/8972 = 0.000006 (biological process) using the number of GO terms for the corresponding GO divisions in a Bonferroni correction. RMA quantification ------------------ We obtained CEL files for the GNF-GEA study from and re-quantified them using the gcrma package \[[@B66]\] in the Bioconductor \[[@B67]\] project for the R statistical analysis program \[[@B68]\]. We use the gcrma options \'type=c(\'fullmodel\')\' and \'fast=T\'. Additional data files ===================== Two additional data files are available with the online version of this article. They contain H and Q values for all normal tissues in the GNF-GEA data set for both human (Additional data file [1](#S1){ref-type="supplementary-material"}) and mouse (Additional data file [2](#S2){ref-type="supplementary-material"}) using both the MAS4 and RMA quantification methods. The RMA data were normalized to yield a common median of 3.75 (human) and 3.22 (mouse) prior to the H and Q calculation. The files are in Excel format. The data for each tissue are placed in separate worksheets. Each worksheet contains H- and Q-values, the expression value of the gene in the worksheet\'s tissue, and its maximum expression across all tissues in the file, the gene symbol, RefSeq, SwissProt, and Unigene ID, and a description. The rows in each worksheet are sorted by increasing values of Q using the RMA data. Thus the top of each worksheet displays the genes most specific to that worksheet\'s tissue. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A table showing H and Q values for all normal human tissues in the GNF-GEA dataset. H and Q values for all normal tissues in the GNF-GEA dataset for human using both the original MAS4 quantification and our RMA re-quantification. The RMA data were normalized to yield common medians of 3.75 prior to the H and Q calculation. The data for each tissue are placed in separate worksheets. Each worksheet contains H- and Q-values, the expression value of the gene in the worksheet\'s tissue, and its maximum expression across all tissues in the file, the gene symbol, RefSeq, SwissProt, and Unigene ID, and a description. The rows in each worksheet are sorted by increasing values of Q using the RMA data. Thus the top of each worksheet displays the genes most specific to that worksheet\'s tissue. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A table showing H and Q values for all normal mouse tissues in the GNF-GEA dataset. H and Q values for all normal tissues in the GNF-GEA dataset for mouse using both the original MAS4 quantification and our RMA re-quantification. The RMA data were normalized to yield common medians of 3.22 prior to the H and Q calculation. The data for each tissue are placed in separate worksheets. Each worksheet contains H- and Q-values, the expression value of the gene in the worksheet\'s tissue, and its maximum expression across all tissues in the file, the gene symbol, RefSeq, SwissProt, and Unigene ID, and a description. The rows in each worksheet are sorted by increasing values of Q using the RMA data. Thus the top of each worksheet displays the genes most specific to that worksheet\'s tissue. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ J.S. thanks J. Mazzarelli, M. Mintz and S. Hannenhalli for many helpful discussions, E. Manduchi and H. He for help with R and RMA, J. Hogenesch and J. Walker at Novartis for providing timely access to the CEL files for the GNF-GEA data, and T. Kadesh for critical readings of the manuscript. C.S. acknowledges support from NIH R01HG001539. J.M.S. and W.-P.S. in C.K.\'s lab were supported by an R01 grant 1R01DK63336. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Examples of GNF-GEA expression patterns for mouse genes at selected H~g~and Q. Liver, indicated in red, is the tissue of interest for Qvalues. (a)Serum albumin (94777\_at Alb1) shows very specific liver expression: H= 1.3 bits and Q~liver~= 2.1 bits. (b)For liver-specific bHLH-Zip transcription factor (99452\_at Lisch7), liver is a strong but not dominant part of the expression pattern: H= 3.7 bits and Q~liver~= 6.8 bits. (c)For chloride channel 7 (104391\_s\_at Clcn7) there is near uniform expression: H= 4.3 bits and Q~liver~= 10.2 bits. (d)Gelsolin (93750\_at Gsn) is an otherwise widely expressed gene but is expressed at a very low level in the liver: H= 4.4 bits and Q~liver~= 15.1 bits. ::: ![](gb-2005-6-4-r33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distributions of Hand Qfor different data sources and tissues. (a)Distribution of Has estimated from GNF-GEA (red line) and DoTS (blue line). The DoTS curve was generated from genes with at least six ESTs. (b)Correlation of Hestimates from GNF-GEA and DoTS. Genes with at least 30 ESTs are shown in red; those with more than 100 ESTs in blue. (c)Cumulative distribution of Qvalues for selected mouse tissues and the average for all 39 tissues. Mammary gland, liver, muscle and the amygdala have decreasing numbers of highly tissue-specific genes. Liver has a very large number of relatively specific genes. All distributions peak at 2 log~2~(39) = 10.6 bits and have a tail at high Q(not shown) that corresponds to genes that are ubiquitously expressed except in the tissue of interest. ::: ![](gb-2005-6-4-r33-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Consensus tissue tree of tissues from human and mouse data. Trees are the consensus of trees created from 5,000 random samples of sets of 1,000 genes from (a)3,768 (human) or (b)1,786 (mouse) genes with Q~g\\|t~≤ 7 bits in at least one tissue. The length of the line leading into a node indicates how many trees did not include the set of tissues to the right of the node. The shortest lines correspond to unanimous subgroups. We have highlighted all maximal subgroups that occurred in at least half of the sampled trees. The nervous system is indicated in red, immune system in blue, reproductive tissue in yellow, digestive organs in purple and magenta, muscle tissue in cyan, and glandular tissue in brown. All maximal subgroups that occurred in at least half of the sampled trees. The tissues not included in a highlighted subgroup typically have statistically significant overlap with many of the highlighted tissues as estimated using the hypergeometric distribution. ::: ![](gb-2005-6-4-r33-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The fraction of start CpG islands in genes ranked by entropy H~g~increases with entropy. Each point represents the fraction of genes in consecutive groups of 100 genes ranked by entropy H~g~computed from GNF-GEA data. Genes in this set are expressed above 200 AU in at least one tissue. The human dataset (diamonds) has 26 tissues (maximum H= 4.7 bits), the mouse dataset (squares) has 42 tissues (maximum H= 5.3 bits). ::: ![](gb-2005-6-4-r33-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Base-composition profiles for ubiquitous and tissue-specific genes with and without start CpG islands. Data is for human genes; similar patterns were observed in mouse. (a)Ubiquitous genes with a CpG island; (b)tissue-specific genes with a CpG island; (c)ubiquitous genes with no CpG island; and (d)tissue-specific genes with no CpG island. Note differences in upstream C+G content, peak sizes at TATA box (-35 bp) and initiator positions, and downstream C versus G differences. ::: ![](gb-2005-6-4-r33-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### YY1 motifs are found downstream of the transcription start site, depending on their orientation. (a)The top image shows a logo \[69\] representation of the YY1 motif in the (+10, +20) region of human CGI+ promoters identified using AlignACE. It is based on 102 sequences. The other two logos are for weight matrices contained in TRANSFAC v7.3 that represent activating and repressing YY1 binding sites. (b)Plot of the positional distribution of predicted YY1 sites and the fraction of genes with a predicted YY1 sites in the (+1, +60) region. YY1 sites were predicted using a weight matrix generated using AlignACE. YY1 sites are more than almost three times (P≤ 2 × 10^-7^) as common in genes with nonspecific CGI+ genes (11%; N= 2,072) than in CGI- genes (4%; N= 607) and occur at more than 10 times the expected rate. Similar trends are observed in genes with 3 ≤ H≤ 4 though with lower absolute and relative rates. The difference between CGI+ and CGI- genes is not statistically significant for genes in the 3 ≤ H≤ 4 bin. Essentially no YY1 sites where observed in specific genes with H≤ 3 bits whether or not they had a CpG island. ::: ![](gb-2005-6-4-r33-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### The distribution of TATA box and initiator element (Inr) in pancreas-specific genes. One hundred and sixty pancreas genes were divided into bins according to their Q-value. Genes that have a TATA box, an initiator with the motif YYANWYY, both, or none of these two motifs, are shown. (a)Absolute numbers of genes with core promoter motifs. Red bars, TATA only; blue bars, TATA and Inr; green bars, Inr only; purple bars, none. The p-values for pairwise comparison of distributions (TATA/total) are given below the graph. P-values were calculated for the sum of genes with TATA box (with and without initiator). (b)Results from (a) plotted as fractions of genes with each motif status within a bin. (c)Number of TATA boxes found in orthologous human and mouse gene pairs. Statistical significance of differences between Qbins are indicated. ::: ![](gb-2005-6-4-r33-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### The cumulative distribution of promoter classes as a function of entropy is similar in human and mouse. The cumulative fractions of genes with all possible combinations of CGI and TATA box features for (a)human and (b)mouse as a function of entropy H~g~as computed from GNF-GEA data is shown. For example, in human about 50% of the genes with H~g~≤ 2.5 have a CGI-/TATA+ promoter. The gray bars indicate the entropy range that is not significantly different from uniform ubiquitous expression. Curves are compiled from genes that express above 200 AU in at least one tissue. As expected, CGI+/TATA- genes are most common in less specific genes and CGI-/TATA+ genes are most common in tissue-specific genes. CGI-/TATA- genes are very common and are found nearly uniformly at every level of specificity. Furthermore, CGI+/TATA- and CGI-/TATA+ genes are both common in mid-specificity (3 ≤ H~g~≤ 4) genes showing that specificity is not determined by these features alone. The trends in human and mouse data are nearly identical despite the lower rate of CpG islands in mouse. The large variations in the graph at low entropy are due to the noise inherent in the small number of genes in this range. ::: ![](gb-2005-6-4-r33-8) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The top five most tissue-specific genes for representative tissues ::: Tissue Probe set ID H Q RefSeq Description ------------ --------------- ----- ----- ------------ ---------------------------------------------------- Amygdala 96055\_at 3.2 5.8 NM\_031161 Cholecystokinin 93178\_at 2.7 5.8 NM\_019867 Neuronal guanine nucleotide exchange factor 93273\_at 3.7 5.8 NM\_009221 Synuclein, alpha 92943\_at 3.5 6.0 NM\_008165 Glutamate receptor, ionotropic, AMPA1 (alpha 1) 95436\_at 3.3 6.1 NM\_009215 Somatostatin Lymph node 98406\_at 2.7 4.0 NM\_013653 Chemokine (C-C motif) ligand 5 98063\_at 1.6 4.1 \- Glycosylation dependent cell adhesion molecule 1 99446\_at 2.5 4.1 NM\_007641 Membrane-spanning 4-domains, subfamily A, member 1 92741\_g\_at 3.3 4.5 \- Immunoglobulin heavy chain 4 (serum IgG1) 102940\_at 2.8 4.6 NM\_008518 Lymphotoxin B Liver 94777\_at 1.3 2.1 \- Albumin 1 101287\_s\_at 1.6 2.2 NM\_010005 Cytochrome P450, 2d10 99269\_g\_at 1.5 2.2 NM\_019911 Tryptophan 2,3-dioxygenase 100329\_at 1.4 2.3 NM\_009246 Serine protease inhibitor 1-4 94318\_at 1.6 2.3 NM\_013475 Apolipoprotein H Genes must express at 200 AU in one or more tissues. A full list of all genes is available in the [Additional data files 1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The list of tissues used in this study ::: GNF+GEA tissues Comparison to EST Hierarchical clustering ----------------------------- ------------------- --------------------------------- DRG PNS Nervous system Trigeminal CNS Hippocampus CNS Amygdala CNS Frontal\_cortex CNS Cortex CNS Striatum CNS Olfactory\_bulb CNS Hypothalamus CNS Spinal\_cord\_lower CNS Spinal\_cord\_upper CNS Cerebellum CNS Eye Eye Spleen Spleen Immune System + trachea Lymph\_node Lymph\_node Trachea Trachea Thymus Thymus Bone\_marrow Bone Bone Bone Lung Lung Uterus Uterus Reproductive organs Umbilical cord Umbilical\_cord Placenta Plancenta Ovary Ovary Epidermis, snout\_epidermis Epidermis Heart Heart Muscle Skeletal\_muscle Skeletal\_muscle Adipose\_tissue, brown\_fat Fat Adrenal\_gland Adrenal\_gland Stomach Stomach Digestive tract Bladder Bladder Small\_intestine Small\_intestine Large\_intestine Large\_intestine Gall bladder Gall\_bladder Gall bladder, liver, and kidney Liver Liver Kidney Kidney Salivary\_gland Salivary\_gland Thyroid Thyroid Mammary\_gland Mammary\_gland Prostate Prostate Testis Testis Tongue Tongue Digits Digits The list of tissues available in the mouse GNF+GEA survey, groupings of tissues used to compare microarray and EST-based entropy estimates, and tissue groups discovered by clustering tissues on the basis of genes expressed in common. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The top five most group-specific mouse genes for selected tissue groups ::: Tissue cluster Probe Set ID H Q RefSeq Description ------------------------ -------------- ------- ------- ------------ ----------------------------------------------------------------------------- Nervous system 100047\_at 3.3 3.4 NM\_011428 Synaptosomal-associated protein, 25 kDa 103030\_at 3.5 3.6 Dynamin 97983\_s\_at 3.7 3.8 NM\_009295 Syntaxin binding protein 1 98339\_at 3.7 3.8 NM\_018804 Synaptotagmin 11 94545\_at 3.7 3.8 NM\_153457 Reticulon 1 Immune system 96648\_at 2.807 2.882 NM\_009898 Coronin, actin binding protein 1a 93584\_at 3.373 3.622 Immunoglobulin heavy chain 6 (heavy chain of IgM) 101048\_at 3.541 3.876 NM\_011210 Protein tyrosine phosphatase, receptor type, C 94278\_at 3.495 3.923 NM\_008879 Lymphocyte cytosolic protein 1 100156\_at 3.609 4.039 NM\_008566 Mini chromosome maintenance deficient 5 Liver and gall bladder 94777\_at 1.280 1.326 Albumin 1 100329\_at 1.394 1.464 NM\_009246 Serine protease inhibitor 1-4 99269\_g\_at 1.471 1.561 NM\_019911 Tryptophan 2,3-dioxygenase 99862\_at 1.503 1.595 NM\_013465 Alpha-2-HS-glycoprotein 96846\_at 1.515 1.607 NM\_080844 Serine (or cysteine) proteinase inhibitor, clade C (antithrombin), member 1 The tissue groups were identified in a consensus clustering of tissues based on common tissue-specific genes. The Qvalue is for the gene and tissue group. To ensure uniform expression across the tissue group, genes were required to have an entropy on the tissue group that was 90% of the maximum possible for the group. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### CpG islands are correlated with embryonic expression even for tissue-specific genes ::: Gene type CpG island state Total genes considered Expressed genes Fraction Fraction ratio ---------------- ------------------ ------------------------ ----------------- ---------- ---------------- Embryo CGI+ 933 365 39% 2.8 CGI- 1007 139 14% Adult-specific CGI+ 29 8 29% 4 CGI- 180 12 7% We determined the fraction of genes with (39%) and without (14%) start CpG islands that are expressed in the early embryo. A gene is 2.8 (= 0.39/0.14) times more likely to be expressed in the early embryo if it has a start CpG island. If we then consider genes that go on to be specific in the adult, we find the ratio of CGI+/CGI- genes is now 4 = 0.28/0.07. The differences in rates between CpG island status within each stage are significant (P\< 0.0005; binomial). Of the between-stage comparisons, only the CGI- adult-specific/embryo change is significant (P= 0.0009; hypergeometric). ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### The most significant indicators of the degree of tissue specificity: start CpG island, TATA box, and YY1 site ::: Features Total fraction H0-3 H3-4 H4-5 ---------- ---------------- -------- ----------- --------- --------- ----------- 3,552 271 602 2679 1.00 0.08 0.17 0.75 CGI+ 2,434 56 306 2072 0.69 0.02 0.13 0.85 0.30 0.74 1.13 CGI- 1,118 215 296 607 0.31 0.19 0.26 0.54 2.52 1.56 0.72 TATA+ 604 136 175 293 0.17 0.23 0.29 0.49 2.95 1.71 0.64 TATA- 2,949 135 427 2,387 0.83 0.05 0.14 0.81 0.60 0.85 1.07 CGI+ TATA+ 284 19 82 183 0.08 0.07 0.29 0.64 0.88 1.70 0.85 CGI- TATA+ 320 117 93 110 0.09 0.37 0.29 0.34 4.79 1.71 0.46 CGI+ TATA- 2,150 37 224 1,889 0.61 0.02 0.10 0.88 0.23 0.61 1.16 CGI- TATA- 798 98 203 497 0.22 0.12 0.25 0.62 1.61 1.50 0.83 YY1+ 293 1 16 276 0.08 0.00 0.05 0.94 0.04 0.32 1.25 CGI+ YY1+ 261 1 10 250 0.07 0.00 0.04 0.96 0.05 0.23 1.27 CGI+ YY1- 2,173 55 296 1,822 0.61 0.03 0.14 0.84 0.33 0.80 1.11 CGI- YY1- 1,086 215 290 581 0.31 0.20 0.27 0.53 2.59 1.58 0.71 CGI- YY1+ 32 0 6 26 0.01 0.00 0.19 0.81 0.00 1.11 1.08 The three columns on the left indicate the combination of features considered; empty cells indicate that the feature is not considered. The \'Total fraction\' column indicates the number of promoters with each feature combination (in bold) and the corresponding fraction of all genes considered. The three columns on the right give statistics for matching genes in three bands of tissue specificity. The top two lines give the number and corresponding fraction of all genes considered for each band. For each feature combination, the numbers indicate the number (top, bold), fraction (middle), and enrichment ratio (bottom) of matching genes. The enrichment ratio is the fraction of promoters of genes in the entropy band that contain a feature divided by the band\'s fraction among all genes considered. For example, specific genes are best recognized by a combination of TATA box (TATA+) and lack of a CpG island (CGI-), which enriches the fraction of such genes from 8% to 37% - a factor of 4.79. Nonspecific genes are most specifically recognized by CpG islands and YY1 sites, which returns a set that is 96% nonspecific genes, but only matches 7%/75% = 10% of the nonspecific genes. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Over-represented Gene Ontology (GO) terms for cellular component and biological process of genes by promoter class ::: Cellular component/biological process Human only Mouse only ------------ ------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------ CGI-/TATA+ Extracellular, extracellular space - Intermediate filament (cytoskeleton) Response to stimulus Cell-cell signaling, organismal physiological process, inflammatory response, innate immune response, response to pest/pathogen/parasite \- CGI+/TATA- Cell, cytoplasm, intracellular, mitochondrion Nucleus, ribonucleoprotein complex - - Nucleobase, nucleoside, nucleotide and nucleic acid metabolism, intracellular transport, metabolism, protein transport, intracellular protein transport, RNA processing, RNA metabolism, cell cycle, mitotic cell cycle - CGI-/TATA- (Integral to) (plasma) membrane \- Extracellular, extracellular space Organismal physiological process, defense response, immune response, response to biotic stimulus, response to stimulus, response to external stimulus Response to pest/pathogen/parasite, cell communication, response to wounding, cellular defense response, signal transduction Complement activation, complement activation (classical pathway), humoral defense mechanism (sensuVertebrata), humoral immune response All terms were selected using a p-value ≤ 0.05 (corrected for multiple testing). Terms common to human and mouse are listed in the second column. The two columns on the right indicate any additional terms found in only one species. The CGI-/TATA+ terms are consistent with a model of strong condition-specific induction, CGI+/TATA- terms are consistent with housekeeping functions. CGI-/TATA- terms indicate support for cell sensing and communication functions. No significant results were found for CGI+/TATA+ genes. :::	PubMed Central	2024-06-05T03:55:56.363343	2005-3-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088961/", "journal": "Genome Biol. 2005 Mar 29; 6(4):R33", "authors": [ { "first": "Jonathan", "last": "Schug" }, { "first": "Winfried-Paul", "last": "Schuller" }, { "first": "Claudia", "last": "Kappen" }, { "first": "J Michael", "last": "Salbaum" }, { "first": "Maja", "last": "Bucan" }, { "first": "Christian J", "last": "Stoeckert" } ] }
PMC1088962	Background ========== The woody tissues of angiosperm trees, the xylem fibers and vessels, are formed from the lateral meristem of the stem, the vascular cambium. In contrast to vessel elements, which differentiate very rapidly close to the vascular cambium, fiber differentiation is a relatively slow process involving initial expansion of the cells in both the radial and longitudinal dimensions, followed by extensive synthesis of the secondary cell walls. The final phase in maturation of both vessel elements and fibers is cell death and autolysis of the cell contents. Xylem-cell death involves a range of morphological and nuclear changes in a strictly spatially and temporally coordinated and programmed fashion \[[@B1],[@B2]\]. The programmed cell death (PCD) of xylem has been analyzed in detail in an in vitrosystem of Zinnia elegans, in which mesophyll cells of Zinniatransdifferentiate into xylem vessels commonly called as tracheary elements in a semi-synchronized manner \[[@B3]\]. In Zinniacells, irreversible differentiation into tracheary elements is marked by the accumulation of hydrolytic enzymes in the vacuole and deposition of the secondary cell walls, followed by tonoplast disruption, release of the vacuolar proteases and nucleases into the cytoplasm, and finally the autolytic loss of cell contents \[[@B2],[@B4]\]. Several different types of proteases have been detected in Zinnia\[[@B5],[@B6]\], and an S1-type nuclease, capable of hydrolyzing both DNA and RNA, seems to control nuclear DNA degradation in the Zinniatracheary elements \[[@B7]\]. Even though the chain of events during tracheary element PCD is well characterized in the Zinniasystem, very little is known about the regulation of this process in intact plants. In addition, the Zinniasystem has not allowed analysis of the different cell types of the xylem, such as the fibers. Programmed cell death also occurs in plants in response to external factors, such as avirulent pathogens, giving rise to the so-called hypersensitive response (HR) and in response to shortening daylength - manifested in the senescence of leaves. HR cell death is usually fast and it shares certain features with the apoptotic death of animal cells, such as nuclear shrinkage and fragmentation of DNA into oligonucleosomal multiples of 180-bp fragments \[[@B8]\]. Senescence-induced cell death is a much slower process, involving nuclear degradation, DNA fragmentation and thorough proteolytic degradation of the cellular contents and controlled remobilization of the nutrients \[[@B9]\]. The death of the xylem elements is different from HR and senescence-related PCD in that the organellar structure remains intact until vacuolar collapse and the oligonucleosomal DNA fragmentation does not precede cell death \[[@B1]\]. Whether these processes are related at the molecular level is unknown, but the differences in temporal and spatial regulation, and in cellular morphology, suggest that there are significant differences not only in the early regulation, but also in the execution of the various plant PCD processes. The genus Populushas emerged as the main model system for trees, because of its amenability for genomic and molecular analyses \[[@B10]\]. Populusis also suitable for analysis of xylem development \[[@B11],[@B12]\]. A Populusexpressed sequence tag (EST) database (POPULUSDB) was created from 19 different cDNA libraries \[[@B13]\]. The database consists of 102,019 ESTs, assembled into a unigene set of 11,885 clusters and 12,759 non-clustered singletons corresponding altogether to 24,644 unique sequences or transcripts \[[@B14]\]. The great diversity of the tissue types giving rise to the different cDNA libraries enables digital analysis of gene expression by comparison of the EST frequencies in the different libraries. One of the libraries was produced from Populuswoody tissues composed of xylem fibers undergoing cell death. In this work, we studied gene expression in the process of fiber death by in silicoanalysis of this \'fiber death library\' and by a microarray analysis with a novel Populus25K cDNA microarray. In addition to its economic importance as one of the processes that regulate wood quality, fiber-cell death is an interesting biological process that as yet is poorly understood. Our analysis identified several novel candidate regulatory genes for xylem PCD. Results and discussion ====================== The unique characteristics of fiber-cell death in Populuswood --------------------------------------------------------------- An analysis of the fiber death cDNA library in the POPULUSDB was undertaken to characterize specific molecular events in Populusxylem fibers approaching cell death. The fiber death library was constructed from xylem tissues in which the fibers had passed the developmental phases of cell expansion and bulk secondary cell wall deposition, and were approaching cell death (see \[[@B13]\], and corresponding to zone B in Figure [1](#F1){ref-type="fig"}). Cell death of the fibers is marked by gradual disappearance of the cytoplasm and finally by complete autolysis of the cells when no cytoplasm can be discerned within the cells (Figure [1](#F1){ref-type="fig"}). Differentiation of xylem vessels differs from that of fibers in that it is much faster, occurring usually within a distance of 100-150 μm from the cambium. Development of the vessels is difficult to study in vivonot only because it is so fast, but also because it takes place in the midst of xylem fibers that are still finishing cell expansion and initiating secondary cell wall deposition. To avoid mixing different processes of xylem development, we decided in this analysis to exclude woody tissues containing differentiating xylem vessels and to focus purely on the late maturation events of xylem fibers. To obtain a broad picture of cell death in xylem fibers, we compared the relative distributions of ESTs with different gene ontology assignations in three cDNA libraries of POPULUSDB: the fiber death library, the tension wood library derived from tension wood-forming xylem, and the leaf senescence library \[[@B13]\]. The leaf senescence library was chosen as it represents another PCD process in plants, and the tension wood library because it represents tissues where fiber death is inhibited but is otherwise comparable to the fiber death library. Tension wood is formed in an asymmetric manner in gravistimulated stems of angiosperm trees. As a part of this process the fibers show delayed cell death due to production of a cellulose-rich layer, the so-called G-layer, inside the secondary cell walls. Remarkably, the fiber death library showed a higher proportion of ESTs (36%) that could not be assigned to any gene ontology term, compared to the tension wood library (23%) and the leaf senescence library (27%) (Figure [2](#F2){ref-type="fig"}). In addition, 6%, 7% and 5% of the ESTs in the fiber death, tension wood and leaf senescence libraries, respectively, represented unknown biological processes. These figures indicate that unique and poorly characterized physiological processes may occur in xylem fibers undergoing cell death. The two cell-death libraries, the fiber death and the leaf senescence, were similar in the sense that they had fewer clones related to biosynthesis and cell communication, and a larger number of clones related to catabolism than the tension wood library (Figure [2](#F2){ref-type="fig"}). However, the leaf senescence library had higher frequencies of clones in the categories of electron transport and development than the other two libraries (Figure [2](#F2){ref-type="fig"}). Altogether, the analysis demonstrated that the distributions of biological processes during fiber-cell death are fairly similar to those during both tension wood formation and leaf senescence. Common features shared by the fiber death and leaf senescence libraries suggest similarities between fiber-cell death and senescence-related PCD. However, fiber-cell death is also expected to involve metabolic and regulatory pathways that have not yet been characterized, based on the high proportion of ESTs with unknown gene ontology or unknown function in the fiber death library. The most abundant transcripts during fiber-cell death ----------------------------------------------------- A high abundance of a transcript suggests that the corresponding protein participates in a process that is important for the cell or tissue. We searched for highly abundant transcripts in the process of fiber-cell death by identifying in the fiber death library the POPULUSDB unigene clusters that had the highest numbers of ESTs. The 28 most abundant transcripts, shown in Figure [3](#F3){ref-type="fig"}, were also enriched in the fiber death library. Assuming a random distribution, the expected EST frequency in the fiber death library is 4.8% (4,867 ESTs in the fiber death library out of the total number of 102,019 ESTs in the POPULUSDB). All of the transcripts shown in Figure [3](#F3){ref-type="fig"} displayed a higher frequency than that. In fact, all clusters except POPLAR.147, POPLAR.58, POPLAR.39, POPLAR.166 and POPLAR.613 had an EST frequency between 10-91% in the fiber death library. The most abundant transcript in the fiber death library was glycine hydroxymethyltransferase (GHMT; POPLAR.161). GHMT was also highly abundant in several other libraries derived from xylem-containing tissues, such as the tension wood and the roots (Figure [3](#F3){ref-type="fig"}). GHMT has also been identified as one of the most abundant proteins in Populusxylem \[[@B15]\], Arabidopsisroots \[[@B16]\] and loblolly pine xylem \[[@B17]\]. GHMT is a key enzyme in one-carbon metabolism, catalyzing reversible conversion between serine and glycine to produce 5,10-methylenetetrahydrofolate, which can be used to recover methionine from 5-methyl-tetrahydrofolate and homocysteine \[[@B15]\]. One-carbon metabolism is known to be active in photorespiration, but its preferential expression in the late maturing fibers suggests that the PopulusGHMT is also involved in some other process(es). Also, three other enzymes that participate in one-carbon metabolism were all highly abundant in the fiber death library. 5-Methyltetrahydropteroyltriglutamate-homocysteine S-methyltransferase (POPLAR.649) catalyzes biosynthesis of methionine, while S-adenosylmethionine synthetase (POPLAR.155) and adenosylhomocysteinase (POPLAR.147) are involved in methionine catabolism (Figure [3](#F3){ref-type="fig"}). It is possible that one-carbon metabolism is required during xylem maturation for the production of glycine, which is abundant in the cell wall proteins. S-adenosylmethionine, which is synthesized from methionine, can also be used for methylation reactions which occur during secondary wall formation. It is, however, unlikely that these enzymes are only needed for fiber-cell death, because they are also highly abundant in libraries derived from the cambial zone and tension wood (Figure [3](#F3){ref-type="fig"}). Cell-wall proteins were highly abundant in the fiber death library (Figure [3](#F3){ref-type="fig"}). Arabinogalactan proteins (AGP) belong to a superfamily of proteoglycans encompassing several subclasses, such as classical AGPs and fasciclin-like AGPs. Both classical AGPs (POPLAR.862 and POPLAR.999) and a fasciclin-like AGP (POPLAR.3203) were highly abundant in the fiber death library (Figure [3](#F3){ref-type="fig"}). AGPs are a major proteinaceous constituent of the cell walls, but their function has remained unclear. A recent report on a hybrid protein containing AGP domains supports the hypothesis that AGPs have a critical function in mediating cell-cell interactions during vascular differentiation \[[@B18]\]. The fasciclin-like AGPs, on the other hand, seem to be involved in the control of cell adhesion \[[@B19],[@B20]\]. AGPs have also been implicated in PCD \[[@B21],[@B22]\]. AGP genes in clusters POPLAR.999 and POPLAR.3203 were also highly expressed in other libraries, especially the library from the tension wood-forming xylem tissues, suggesting that they have a more general function during cell-wall formation (Figure [3](#F3){ref-type="fig"}). However, the POPLAR.862 AGP was enriched in the fiber death library. POPLAR.862 has also been shown to be suppressed in a microarray analysis of tension wood where fiber-cell death is inhibited (S. Andersson-Gunnerås, E. Mellerowicz and B. Sundberg, personal communication), strongly indicating a role for this AGP in the stimulation of fiber-cell death. Two other cell-wall proteins, a glycine-rich protein (POPLAR.1776) and an extensin-like protein (POPLAR.9554), were also highly abundant and overrepresented in the fiber death library (Figure [3](#F3){ref-type="fig"}), suggesting that these proteins also have functions during the late maturation of xylem fibers. A cysteine protease (POPLAR.1250) and a polyubiquitin (POPLAR.58) were highly abundant in the fiber death library (Figure [3](#F3){ref-type="fig"}), as well as in other libraries derived from tissues in which large proportions of cells are dying, such as senescing leaves, the root tissues and petioles. Cysteine proteases are believed to participate in the post mortemevents of xylem elements \[[@B1]\]. The high abundance of polyubiquitin suggests that the ubiquitin-proteosome pathway participates in proteolytic events of xylem cells as well. A further transcript related to proteolysis and cell death was POPLAR.9335, which was highly abundant and also highly enriched in the fiber death library (Figure [3](#F3){ref-type="fig"}). It encodes a protein with an unknown function, but contains a domain found in lipid-transfer proteins, seed storage proteins and protease inhibitors. A similar kind of protein was earlier shown to regulate programmed cell death and plant defense \[[@B23]\]. The expression pattern of POPLAR.9335 was also analyzed by RT-PCR, and the results confirmed the specificity of this transcript in the xylem fibers undergoing cell death (Figure [4](#F4){ref-type="fig"}). The fiber death library-specific transcripts are putatively novel regulators of cell death ------------------------------------------------------------------------------------------ Analysis of the most abundant transcripts in the fiber-cell death library yielded a list of candidate genes with high expression levels. In order to identify fiber-cell death specific transcripts with lower expression levels we identified in POPULUSDB the clusters and singletons (non-clustered transcripts) that were unique to the fiber death library and not present in any of the other 18 EST libraries. In total, 71 clusters and 929 singletons were identified that were unique to the fiber death library (Additional data file 1). Singletons represent either rare transcripts or poorly sequenced regions of transcripts, and are therefore not necessarily indicative of fiber-death specific expression. Of the 71 fiber death library specific clusters, the 12 clusters having the highest abundance of ESTs are shown in Table [1](#T1){ref-type="table"}. First, a microarray experiment was performed to confirm the expression pattern of these transcripts. Two samples were collected from the woody tissues of the stem; one (A) containing the zones where xylem fibers were in the process of cell expansion and secondary cell wall formation, and one (B) containing tissues where the fibers were undergoing cell death (see Figure [1](#F1){ref-type="fig"}). The sample from the fiber-cell death zone corresponded closely to the tissues collected for construction of the fiber death library, and should therefore show high expression of the transcripts unique to this library. The microarray analysis showed that, with the exception of POPLAR.11648, all seven fiber death library-specific clusters that were represented by more than three ESTs were also more highly expressed in the fiber-cell death sample compared to the early developing fibers (Table [1](#T1){ref-type="table"}). Fiber death library-specific clusters with EST abundances below four showed varying results in the microarray analysis (Additional data file 1). None of the transcripts that were shown to be unique to the fiber death library has previously been implicated in the regulation of cell death. The oligopeptide transporter (POPLAR.11639) is involved in amino-acid metabolism related to remobilization of nutrients from dying tissues, but the reason for the specific expression pattern of the other transcripts is not clear. Several of them seem to be membrane proteins or targeted to the endomembrane system, as predicted on the basis of their closest Arabidopsishomologs (Table [1](#T1){ref-type="table"}). A glycosyl hydrolase of family 1 (POPLAR.11628) is the most abundant unique transcript in the fiber death library, but the substrates of this glucosidase and its exact role in fiber-cell death remain to be elucidated. Interestingly, the closest homologs of this protein are cyanogenic in nature, and it is tempting to speculate that cyanide production by this enzyme could participate in regulation of cell death in xylem fibers. Reverse transcription PCR (RT-PCR) analysis of five unique transcripts confirmed that POPLAR.11628 (nine ESTs), POPLAR.11639 (five ESTs) and POPLAR.11646 (four ESTs) were indeed very specifically expressed in xylem fibers undergoing cell death (Figure [4](#F4){ref-type="fig"}). POPLAR.11658 (six ESTs) and POPLAR.11624 (two ESTs) showed highest expression in the dying xylem fibers, but were also expressed in other types of Populustree tissues (Figure [4](#F4){ref-type="fig"}). Combination of the in silicoanalysis with a microarray analysis refines selection of candidate regulatory genes ----------------------------------------------------------------------------------------------------------------- The microarray analysis revealed both singletons and clusters that were represented only by two or three ESTs in the fiber death library, but were highly upregulated in the xylem fibers undergoing cell death (Additional data file 1). To combine the power of the POPULUSDB and the microarray analysis, transcripts were identified that had a high expression level in the dying fibers on the basis of the microarray analysis and that were enriched or unique in the fiber death library within the POPULUSDB. Figure [5](#F5){ref-type="fig"} shows the 50 transcripts that were most upregulated in the dying xylem fibers compared to the early developing xylem on the basis of the microarray analysis. The most upregulated transcripts were generally highly enriched or unique to the fiber death library (Figure [5](#F5){ref-type="fig"}). Of the 20 most upregulated transcripts, nine were unique to the fiber death library and only five were completely absent from it (Additional data file 2). The good correlations between EST abundances in the fiber death library and gene expression in the microarray analysis verify the usefulness of the POPULUSDB in transcript profiling of xylem fiber death. Among the most upregulated transcripts that were unique or highly enriched in the fiber death library, there are transcripts that participate in amino-acid metabolism and transport (X024D04, POPLAR.872), proteolysis (POPLAR.11632, POPLAR.4995, POPLAR.10724, X077D02), and also transcripts, such as kinases (X053F08, POPLAR.9347), nodulin-like proteins (X002G05, POPLAR.4667) and a CACTIN-like protein (POPLAR.11667), that seem to have signaling functions rather than mere cellular disintegration of the dying fibers (Figure [5](#F5){ref-type="fig"}). High expression of nodulin-like proteins in dying xylem fibers suggests that nodulins, which regulate nodule formation in response to Rhizobiuminfection, also have an important function in the maturation of xylem fibers. Interestingly, certain nodulins have been shown to regulate accumulation of reactive oxygen species (ROS) \[[@B24],[@B25]\], and it is possible that nodulin-like proteins regulate ROS accumulation that occurs during the late maturation or cell death of xylem fibers. ROS accumulation is also implied by the high expression levels of peroxidases (POPLAR.11659 and 11669) and a protein kinase (singleton X053F08) that seems to encode the Populusortholog of Arabidopsis thaliana OPEN STOMATA 1(OST1; Figure [5](#F5){ref-type="fig"}). OST1 regulates abscisic acid (ABA)-mediated accumulation of ROS related to stomatal closure in Arabidopsis, and it has also been shown to be expressed in vascular tissues of Arabidopsisleaves and roots \[[@B26]\]. Other important proteins in the early signaling of fiber-cell death could include the basic helix-loop-helix transcription factor V031H02 and the C2H2 class zinc-finger protein POPLAR.10810. Interestingly, POPLAR.11144 is most similar to the Arabidopsisgene VACUOLELESS1, which is required for proper vacuole formation and autophagy \[[@B27]\]. In the Z. eleganscell culture system, the permeability and integrity of the vacuolar membrane regulates initiation of xylem-cell death \[[@B1]\], and our results suggest that VACUOLELESS1could be involved in this regulation during the cell death of fibers. Combining the in silicoexpression analysis in 19 different tissue types with a focused microarray analysis is expected to facilitate the identification of candidate genes better than either method alone. While the microarray analysis allowed selection of genes with high expression levels in xylem fibers undergoing cell death compared to early developing fibers, the in silicoanalysis of POPULUSDB facilitated exclusion of those genes that were highly abundant in other types of Populustree tissues. The candidate genes selected here are potential novel regulators of cell death in xylem fibers. Comparison of proteases in three different cell death processes --------------------------------------------------------------- The expression of serine, cysteine and aspartic proteases was analyzed in detail in the fiber death library, because proteases have well established roles in the control of cell death in both animal cells and plants \[[@B28],[@B29]\]. We also compared their expression in the fiber death library and three other cDNA libraries: the leaf senescence library (library I) and the virus/fungal infected leaf library (library Y), which are expected to be enriched in transcripts related to cell death, and the tension wood library (library G), which should theoretically be devoid of fiber-death-related transcripts. Cysteine proteases (CP) have been identified in xylem elements undergoing cell death \[[@B6],[@B30]\]. Accordingly, inhibitors of cysteine proteases have been shown to impair maturation of xylem elements \[[@B31]\]. Cysteine proteases were highly abundant in the POPULUSDB, and they were usually not restricted to any of the cell death libraries X, I or Y (Table [2](#T2){ref-type="table"}). However, the cysteine protease POPLAR.3310 was somewhat enriched in the fiber death library and not represented in the leaf senescence library. This transcript is most similar to Arabidopsis XYLEM CYSTEINE PEPTIDASE 2(XCP2), which has been shown to be specifically expressed in xylem vessels of leaves and roots \[[@B32]\]. Interestingly, POPLAR.3310 was also present in the virus/fungal infected leaf library, suggesting that this CP is also activated during pathogen-induced cell death. In addition, a VACUOLAR PROCESSING ENZYME(VPE; POPLAR.3027), was enriched in the fiber death library (Table [2](#T2){ref-type="table"}). Two ArabidopsisVPE isozymes, α-VPE and γ-VPE, have been shown to be activated during senescence, but only α-VPE was expressed in the vascular tissues \[[@B33]\]. In tobacco, VPE has been shown to regulate the integrity of the vacuolar membranes and pathogen-induced hypersensitive cell death \[[@B34]\]. Our data suggest that, in addition to hypersensitive cell death, VPE controls developmental cell death in xylem fibers, possibly through regulation of vacuolar integrity. In contrast to the cysteine proteases, the serine proteases displayed higher specificity to given cell death processes (Table [2](#T2){ref-type="table"}). This conforms well to the reported role of serine proteases in the early signaling of apoptotic cell death in virus-infected animal cells \[[@B29]\]. In plant cells, serine proteases have been implicated in pathogen-induced cell death \[[@B35],[@B36]\], and serine protease activities have also been demonstrated during xylem-cell death \[[@B5],[@B6]\]. Groover and Jones \[[@B4]\] showed that serine protease activity stimulated, and was necessary for, the death of Z. eleganstracheary elements grown in vitro. In accordance with these findings, a 40 kDa serine protease was shown to accumulate in the culture medium of in vitrotracheary elements \[[@B4]\]. Our data revealed the identities of two serine proteases possibly related to the regulation of xylem-cell death. The serine proteases POPLAR.10138 and POPLAR.4995 were enriched in the fiber death library, and also highly upregulated in the microarray analysis of the xylem fibers undergoing cell death (Table [2](#T2){ref-type="table"}). A PSORT analysis \[[@B37]\] of protein sequences derived from ESTs and manually assembled genomic sequences supported extracellular locations for both POPLAR.10138 and POPLAR.4995 (data not shown). The targets of these serine proteases are not known. Groover and Jones \[[@B4]\] suggested that the extracellular 40 kDa serine protease was responsible for activation of Ca^2+^channels, which is a prerequisite of xylem-cell death. Other possible targets are membrane-bound leucine-rich-repeat-containing proteins, which have been shown to interact with serine proteases during hypersensitive cell death \[[@B38]\]. One such target could be a plasma-membrane localized leucine-rich-repeat-containing receptor kinase that was unique to the fiber death library (singleton X021F12) and significantly upregulated during fiber death (PU27994; Additional data file 2). However, regardless of their targets, it seems evident from our data that modification of the extracellular matrix occurs during fiber-cell death by two extracellular serine proteases, probably as part of the early signal transduction process. An aspartic protease has been localized specifically in the tracheary elements undergoing cell death \[[@B39]\]. In this analysis, we found no evidence for any fiber-death specific aspartic proteases. All four aspartic proteases identified in POPULUSDB were either present in several cell death libraries or also in the tension wood library, where cell-death-related transcripts should not accumulate (Table [2](#T2){ref-type="table"}). Hormonal control of fiber maturation ------------------------------------ Arabidopsishas been used to analyze transcriptomes during xylem development \[[@B40],[@B41]\]. In Arabidopsis, formation of the vascular cambium, giving rise to the so-called secondary growth, takes place in the hypocotyls after two to three months of growth requiring continuous removal of the inflorescence stems \[[@B42]\]. The early development and maturation of xylem vessels and fibers during the secondary growth of the hypocotyl is similar to Populusexcept for that the fibers in Arabidopsisdo not die in the highly coordinated manner as in Populus, even after an extended growth period of three months (our unpublished work). It is therefore possible to analyze similarities between Populusand Arabidopsisin the process of fiber maturation but not cell death. To identify common patterns in the transcriptomes of Populusand Arabidopsisduring fiber maturation, we identified Arabidopsishomologs to the Populusgenes that were upregulated at least two times (P\< 0.001 and B\> 0) in the fiber cell death sample (late maturing fibers; sample B) compared to the early fiber development (sample A). This dataset, denoted as \'PopulusB/A\', was compared to two previously published Arabidopsisdatasets \[[@B41]\]. The first Arabidopsisdatasets, denoted as \'treatment\', consists of genes that were upregulated at least two times during secondary growth (9 weeks of growth) compared to the primary growth of the hypocotyls (3 weeks of growth), and is therefore expected to enrich transcripts related to secondary growth including fiber maturation. The second Arabidopsisdataset, denoted as \'xylem\', consists of genes that were upregulated more than two times in secondary xylem tissues compared to bark tissues of hypocotyls, and is expected to enrich transcripts related to all aspects of xylem development during secondary growth including death of the xylem vessels and fiber maturation. The common features of these datasets are shown in Additional data file 3. Cell-death-related transcripts were rarely shared by the different datasets due to the sampling method for the comparisons \'treatment\' and \'xylem\'. However, a large number of transcription factors and plant hormone-related transcripts were often shared by the Arabidopsisand the Populusdatasets. We will discuss here the latter group of transcripts as very little previous knowledge exists on the role of plant hormones in the late maturation events of xylem fibers. Indole-3-acetic acid (IAA) is an auxin-type plant hormone which regulates several different developmental processes in plants and also the activity of the vascular cambium and xylem formation \[[@B43]\]. In Populusstems, IAA shows a steep radial concentration gradient across the stem with the highest concentration close to the vascular cambium and a very low concentration in the late-maturing xylem \[[@B11]\]. Several transcriptional regulators of IAA-related gene expression, such as the auxin response factor (ARF) family proteins as well as the auxin/indoleacetic acid (IAA) family proteins, were observed in all the datasets (Additional data file 3), suggesting involvement of these proteins not only in the early development of secondary xylem, as suggested by the high concentration of IAA in these tissues, but also during late maturation of the xylem fibers. The role of IAA during late maturation of fibers is not clear, but it is possible that functional IAA signaling is required for the survival of the cells. At2g21620 encodes a universal stress protein (USP) family member, which is regulated by auxin \[[@B44]\]. USPs have been implicated in the regulation of stress-related metabolic shifts, and activation of the auxin-regulated USP during secondary growth in Arabidopsisand during late fiber maturation in Populus(Additional data file 3), suggests a role for IAA in late maturing xylem fibers that experience some kind of stress. This could be osmotic stress due to condensation of the cytoplasmic contents or nutrient depletion due to the increasing distance from the nutrient-transporting phloem cells. The role of IAA during cellular stress is supported also by the fact that IAA induces expression of enzymes involved in the biosynthesis of ethylene \[[@B45]\], which is a plant hormone that is produced in response to several different stress conditions. Ethylene does not seem to be needed for normal xylem development on the basis of the fact that ethylene-insensitive genotypes in Arabidopsisand in other species grow normally. However, ethylene biosynthesis is activated in gravitationally stimulated Populusstems, that is, when the stem is displaced from its vertical position, resulting in the production of tension wood \[[@B46]\]. By analogy with several other cell-death processes in plants \[[@B47]\], it is expected that ethylene is also involved in regulation of xylem cell death. This is supported by the activation of several ethylene-related transcripts in both Arabidopsisand the late-maturing xylem fibers in Populus(Additional data file 3). ETHYLENE-INSENSITIVE 3(EIN3), EIN3-BINDING F-BOX 1(EBF1) and ETHYLENE RESPONSE SENSOR 1(ERS1) mediate known parts of ethylene signal transduction \[[@B48]\]. EBF1and ERS1are also induced by ethylene. Because both EBF1and ERS1function to suppress ethylene signaling, it seems that the ethylene signal required for xylem maturation needs to be suppressed or only transiently activated. Comparison of the datasets suggests involvement of two additional plant hormones. PATHOGENESIS-RELATED PROTEINand PHENYLALANINE AMMONIA-LYASE 2are related to salicylic acid (SA) signaling and biosynthesis, respectively, and CORONATINE INSENSITIVE 1to jasmonic acid (JA) signaling. Both JA and SA control stress-related processes, such as cell death and other defense responses to pathogens \[[@B47]\]. The transcriptional activation of these genes in the Arabidopsisxylem tissues and in the late maturing Populusfibers suggests that both JA and SA are involved in the regulation of cell death also in xylem fibers (Additional data file 3). Conclusions =========== Even though Arabidopsisis a suitable model system for studying early vascular development and primary growth, it cannot be easily used for studying secondary growth of the stem. Xylem fibers are formed in Arabidopsisonly after two to three months of growth, which is equivalent to the time required to grow Populustrees to a size that allows collection of large amounts of woody tissues from the stem. In addition, because of the larger diameter growth of the tree stem, tissues can be easily collected from the various developmental phases without mixing the different tissue types \[[@B43]\]. Therefore, Populushas several advantages over Arabidopsisfor wood analysis. Development of appropriate genomic and bioinformatic tools has further strengthened Populusas the main model system for wood formation. This analysis fully exploited the advantages of the Populussystem. Combining in silicoanalysis of the POPULUSDB with a microarray analysis using the novel 25K Populusmicroarrays revealed a number of candidate genes that were unique and highly abundant in the late-developing woody tissues, and possibly related to the regulation of cell death in xylem fibers. We found fiber death-specific transcription factors, as well as nodulins and subtilisin-like serine proteases, all of which are expected to play a role in the early signaling of fiber-cell death. Sequences corresponding to the Populushomolog of Arabidopsis OPEN STOMATA 1and peroxidases suggest involvement of ROS and ABA in fiber-cell death signaling as well. Specific expression patterns of downstream components, such as the Populushomolog of Arabidopsis XYLEM CYSTEINE PEPTIDASE 2and the oligopeptide transporters, support the importance of these proteins in protein degradation and remobilization of nutrients in the dying fibers. Interestingly, two proteins that are known to regulate vacuolar assembly and vacuole integrity in other cellular processes, the Populushomolog of Arabidopsis VACUOLELESS1and a VACUOLAR PROCESSING ENZYME, were specifically expressed during fiber-cell death. Permeability of the vacuolar membrane and vacuolar integrity are known to regulate death of the xylem cells, and we can now for the first time suggest candidate regulatory proteins for this process. Taken together, our analyses have identified a number of candidate genes that may be important in the regulation of fiber-cell death. Overexpression and transcriptional suppression experiments should be undertaken to elucidate the function of these genes in the process of fiber-cell death and their impact on economically important traits of woody tissues. Materials and methods ===================== Computational biology and database analyses ------------------------------------------- Construction of the fiber death library and the POPULUSDB \[[@B14]\] is described in \[[@B13]\]. Data were analyzed using mysql, PHP, C++ and Filemaker software. The gene ontology (GO) comparison was made by listing all GO terms \[[@B14]\] in the hierarchy for each clone and selecting clones from appropriate GO hierarchy levels. Around 10% of the clones were assigned to more than one class. The EST clone distribution within POPULUSDB was performed using mysql and Filemaker, and visualized with the R software package \[[@B49]\] according to the program code described in \[[@B13]\]. Annotations were derived from BLASTX analysis against the Arabidopsisproteome or the Swiss-Prot database in cases where no Arabidopsisproteins with sufficient homology were identified according to an annotation pipeline described in \[[@B50]\]. Preparation of the 25K PopuluscDNA microarray ----------------------------------------------- The microarrays used here constitute the second generation of the global PopuluscDNA microarrays and contain in total 24,735 different cDNA fragments. This array is based on the first generation 13K Populusarray \[[@B51]\] with clones from seven cDNA libraries, representing: the cambial zone (AB), young leaves (C), floral buds (F), tension wood (G), senescing leaves (I), dormant cambium (UA), and active cambium (UB). The 25K array contains clones from 12 additional cDNA libraries, representing the apical shoot (K), cold-stressed leaves (L), roots (R), bark (N), shoot meristem (T), male catkins (V), dormant buds (Q), female catkins (M), petioles (P), fiber death (X), imbibed seeds (S) and virus/fungal infected leaves (Y). For a detailed description of the construction and sequencing of the cDNA libraries, see \[[@B13]\]. All clones in the unigene set were resequenced from both the 5\'-end, as were the original EST sequences, and the 3\'-end. Each unigene clone is defined by a PU number on the microarray. Sequence information of the clones can be found in the POPULUSDB \[[@B14]\]. Plasmid preparations were made in a 96-plate format from bacterial cell suspensions with a Montage 96 Plasmid Prep kit (Millipore) using a Biorobot 8000 molecular biology workstation (Qiagen). PCR amplifications were done in 100 μl reaction volumes, and purified in Montage PCR~384~Filter Plates (Millipore) with a Biorobot 8000 workstation (Qiagen). The purified PCR products were dissolved in 40 μl of 30% DMSO and split between a storage plate and a printing plate. The microarrays were printed with a QArray (Genetix) instrument with 24 SMP2.5 pins (Telechem) on Ultra GAPS slides (Corning). The 24,735 cDNA fragments, together with eight copies each of the 23 different Lucidea Universal Scorecard controls (Amersham Biosciences), were spotted with a feature center-to-center distance of 180 μm. The quality of the spotted slides was assessed by staining with Syto61 (Molecular Probes) and by hybridization with random nonamers. The slides were UV cross-linked at 250 mJ/cm^2^followed by baking at 75°C for 2 h. Microarray analysis ------------------- Samples for RNA isolation were collected from the base of the stem of a 6-month-old hybrid aspen tree (Populus tremula x tremuloides) grown in a greenhouse. The bark was peeled off, and the developmental phases of the xylem were identified by light microscopy and by the texture and color of the different tissue types (Figure [1](#F1){ref-type="fig"}). The A sample, consisting of the remains of the cambial zone, expanding xylem and secondary cell wall depositing xylem, was collected by scraping the part of the xylem that was yellowish in color and still relatively soft with a knife. The B sample, consisting of the thick-cell-walled and late maturing xylem fibers approaching cell death, was light yellow in color and relatively difficult to scrape with the knife. The B sample was scraped to the border of the dead wood, which was discernible by its completely white color, dryness and hard texture. Total RNA was prepared from the two samples according to \[[@B52]\], and mRNA was prepared from 1 μg total RNA using paramagnetic oligo(dT) beads (Dynabeads, Dynal Biotech) in a 10 μl elution volume. For cDNA synthesis, the mRNA sample was sheared by repeated suction into a 10 μl syringe with a beveled, non-coring needle point (Hamilton Bonaduz AG). First-strand cDNA was prepared with 1 μg oligo-dT~15~primer and 200 U SuperScript II RNase H^-^reverse transcriptase (Invitrogen). Second-strand cDNA was prepared in a final volume of 150 μl at 16°C for 2 h with 10 U Escherichia coliDNA ligase (Invitrogen), 40 U E. coliDNA polymerase I (MBI Fermentas), 2 U RNase H (USB, GE Healthcare Bio-Sciences) in a buffer containing 0.2 mM of each dNTP, 19 mM Tris-HCl (pH 6.9), 91 mM KCl, 4.5 mM MgCl~2~and 10 mM (NH~4~)~2~SO~4~. A further incubation was performed at 16°C for 20 min with 2 U T4 DNA polymerase (Invitrogen). The reaction was stopped with 10 μl 0.5 M EDTA, and the cDNA was recovered by phenol and chloroform:isoamylalcohol extraction followed by ethanol precipitation. The precipitates were dissolved in 2 mM Tris-HCl (pH 7.5) and purified using Autoseq G-50 columns (Amersham Biosciences). A blunt-end adapter was created by annealing two single-stranded oligonucleotides, (0.5 mmol MarAdLong: 5\'-CTAATACGACTCACTATAGGGCTCGAGCGGCCGCCCGGGCAGGT-3\' and 0.5 mmol MarAdShort 5\'-PO~4~- ACCTGCCC- NH~4~-3\') in a ligase buffer containing 66 mM Tris-HCl (pH 7.6), 5 mM MgCl~2~, 5 mM DTT and 7.5 μg BSA in a total volume of 15 μl with a 0.7°C/min temperature gradient from 50°C to 10°C. The adapter was ligated to the 5\' template strand with 10 U T4 DNA ligase (Invitrogen) in a ligase buffer containing 33.8 mM Tris-HCl (pH 7.6), 2.56 mM MgCl~2~, 2.56 mM DTT and 24 μg BSA in a total volume of 48.8 μl at room temperature. The cDNA was purified with a QIAquick PCR purification kit (Qiagen). The cDNA was amplified by PCR in a 100-μl volume containing 0.2 mM of each dNTP, 0.75 μM MaraAP1 (5\'-CCATCCTAATACGACTCACTATAGGGC-3\'), 0.75 μM oligo-dT~15~, 67 mM Tris-HCl (pH 8.8), 4 mM MgCl~2~, 16 mM (NH~4~)~2~SO~4~, 3 μg BSA and 0.6 μl AmpliTaq DNA polymerase (Applied Biosystems). The PCR procedure was 95°C for 1 min, 72°C for 5 min, addition of the AmpliTaq, and 17 to 29 cycles of 95°C for 1 min, 50°C for 1 min and 72°C for 2 min. The appropriate cycle number was defined as two cycles before saturation of the PCR product as detected by gel electrophoresis. The PCR product was purified using a QIAquick PCR purification kit and its concentration was measured by spectrophotometry (NanoDrop ND-1000, NanoDrop Technologies). Labeling of the amplified cDNA samples was performed by direct incorporation of 3 μl Cy3-dUTP or Cy5-dUTP (Amersham Biosciences) in an asymmetric PCR reaction with 100 ng cDNA, 1 μM MaraAP1 primer, 0.6 μl AmpliTaq, 67 mM Tris-HCl (pH 8.8), 4 mM MgCl~2~, 16 mM (NH~4~)~2~SO~4~, 80 μM of each dATP, dCTP, dGTP and 20 μM dTTP in a total reaction volume of 50 μl. The PCR conditions were 95°C for 1 min followed by nine cycles of 95°C for 30 sec, 50°C for 30 sec and 72°C for 10 min. The PCR product was purified with a QIAquick PCR purification kit and eluted twice in 30 μl 4 mM KPO~4~buffer (pH 8.5). The final volume was decreased to 41 μl. Microarray hybridizations, as well as scanning of the slides and image analysis were done according to \[[@B53]\]. The two samples A and B were hybridized against each other five times (including dye-swaps). The microarray raw data, including tiff and gpr files from scanning and image analysis, is deposited to the UPSC-BASE microarray database \[[@B54]\]. For statistics, B-values based on Bayesian statistics \[[@B55]\] and parametric t-tests were obtained with the program R, version 1.8.1 \[[@B49]\]. The microarray data shown in Additional data files 1 and 2 includes the log~2~differential expression ratio (M) between the two samples B and A, P-value from the t-test and the B-value. RT-PCR analysis --------------- Samples were collected from ten different tissue types of a 6-month-old hybrid aspen (Populus tremula× tremuloides) tree grown in the greenhouse. Stem tissues; cortex, phloem, expanding xylem, secondary cell wall-depositing xylem, and fiber-cell death tissues, were collected by scraping with a knife from the base of the stem (see Figure [1](#F1){ref-type="fig"}). The developmental phase of the different xylem tissues was verified in transverse sections of the stem. The xylem fiber-cell death sample corresponded to the tissues collected for the B sample in the microarray analysis (see above) and the tissues prepared for construction of the fiber death cDNA library analyzed in this study (for a description of the library see \[[@B13]\]). Other tissues collected for the RT-PCR analysis were the apical shoot, root tips, young leaves, old leaves and male catkins. RNA was prepared according to \[[@B52]\]. The samples obtained were treated with RQ1 DNase (Promega), and cDNA was produced using SuperScript II RNase H- (Invitrogen) and random decamers (Ambion), followed by RNase H treatment. Two microliters of cDNA was PCR-amplified using gene-specific primers and an appropriate mixture of the universal 18S gene-specific primers and the 18S competimers (Ambion). The gene-specific primers were ACAGATGAAGCGTTCAGCAA and CAGTGCAGCACACCTAGGAA for cluster POPLAR.11628, CAGGAAAGCTCTCCGTTTCTT and TCAAAGCTCTTCCCTTCTGC for POPLAR.11658, AATCCCATGAATATTACCCCTAGA and TCTCTTGCATGGGTAGACATTTT for POPLAR.11639, ACCTCCATAGCCACCCAAG and CTGCAAGCTGATGCAGAAGT for POPLAR.11646, TGAACAGCAGGAGGTGTGAG and AACAGGTGTCCCCATCTGAG for POPLAR.11624, and TGGCAACTCCAATGAAGAAC and CACCAACAGTTTATTTATTATTCAGATG for POPLAR.9335. Analysis of proteases in the POPULUSDB -------------------------------------- A list of Populusproteases was compiled, based on known protease sequences of A. thalianacollected from the following databases: the ArabidopsisInformation Resource \[[@B56]\], TrEMBL \[[@B57]\], Merops \[[@B58]\], InterPro \[[@B59]\], TIGR \[[@B60]\] and Munich Information Center for Protein Sequences \[[@B61]\]. Sequences of 602 Arabidopsisproteases were found. The contig sequences of the POPULUSDB were then queried with BLASTX against the Arabidopsisproteases. Only the Populussequences that obtained a BLASTX score higher than 100 and were annotated as proteases or as coding for proteins with a biological function related to proteolysis or peptidolysis in the TIGR database were accepted. The corresponding clusters, having three or more ESTs in any of the libraries of interest (tension wood, senescing leaves, fiber death and virus/fungal infected leaves) in POPULUSDB were chosen for the analysis. Additional data files ===================== The following additional data is available with the online version of this article. Additional data file [1](#S1){ref-type="supplementary-material"} is a table containing the complete list of fiber death library-specific transcripts and the corresponding gene expression data from the microarray analysis. Additional data file [2](#S2){ref-type="supplementary-material"} is a table containing the complete gene expression data in xylem fibers undergoing cell death using the 25K PopuluscDNA array and the POPULUSDB. Additional data file [3](#S3){ref-type="supplementary-material"} is a table containing a comparison of the microarray data from xylem tissues of Populusand Arabidopsis. Additional data file [4](#S4){ref-type="supplementary-material"} is a table showing GenBank accession(s) for the EST sequences(s) in each POPULUSDB unigene cluster and singleton. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 The complete list of fiber death library-specific transcripts and the corresponding gene expression data from the microarray analysis. The list shows clusters and singletons unique to the fiber death library within the POPULUSDB, the number of EST sequences in each cluster, annotation, closest Arabidopsisgene and the BLASTX value (E), and the corresponding gene expression from the microarray analysis. The log~2~expression ratio (M) is calculated between gene expression in sample B (fiber-cell death) and sample A (early fiber development). A statistically significant difference in gene expression is predicted for transcripts having P\<0.001 (t-test) and B\>0 (Bayesian statistics \[[@B55]\]). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 The complete gene expression data in xylem fibers undergoing cell death using the 25k PopuluscDNA array and the POPULUSDB. Transcripts and the corresponding library distribution in POPULUSDB are listed in descending order of expression ratio from the microarray analysis. The log~2~expression ratio (M) is calculated between gene expression in sample B (fiber-cell death) and sample A (early fiber development). A statistically significant difference in gene expression is predicted for transcripts having P\<0.001 (t-test) and B\>0 (Bayesian statistics \[[@B55]\]). The annotations and the Evalues were derived from BLASTX analysis against the Arabidopsisproteome or the Swiss-Prot database according to an annotation pipeline described in \[[@B50]\]. The PU numbers define the unigene clone numbers on the array. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Comparison of the microarray data from xylem tissues of Populusand Arabidopsis. The Populusgenes were selected that were statistically significantly more than two times upregulated in the fiber cell death sample (late maturing fibers; B) compared to the early fiber development (A). The automatically assigned Arabidopsishomologs \[[@B14]\] were listed for these Populusgenes (column \'PopulusB/A\'), and compared to Arabidopsisgenes that were previously shown to be upregulated at least two times during secondary growth compared to primary growth of the inflorescence stems (column \'Treatment\') or in xylem tissues compared to bark of the hypocotyls (column \'Xylem\') \[[@B41]\]. The list shows the genes that were shared by the Populusdataset and at least one of the Arabidopsisdatasets. \'1\' denotes the presence, and \'0\' the absence of a particular gene in a dataset. The annotations were derived from the ArabidopsisInformation Resource (TAIR) 28/01/2005. The PU numbers, corresponding to the unigene clones of the selected Populusgenes, are shown for each homologous Arabidopsisgene. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 A table showing GenBank accession(s) for the EST sequences(s) in each POPULUSDB unigene cluster and singleton. A table showing GenBank accession(s) for the EST sequences(s) in each POPULUSDB unigene cluster and singleton. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We are grateful to Andreas Sjödin for statistical analysis of the microarray results and for visualization of the data and to Edouard Pesquet for data analysis. The work was supported by the Wallenberg foundation, the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning, the Swedish Graduate School in Genomics and Bioinformatics and the Kempe foundation. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Sampling of xylem tissues. A transverse section from the base of the stem showing xylem tissues sampled from a Populustree for the microarray and the RT-PCR analysis. The bark was peeled off resulting in a fracture in the cambial zone. For RT-PCR analysis, the different xylem tissues were successively scraped from the surface of the exposed stem to the border with the dead wood. For microarray analysis, the tissues were pooled into two samples: A (early fiber development) and B (fiber-cell death). The fiber-cell death sample corresponded closely to the tissues collected for construction of the fiber death cDNA library \[13\]. V, dead vessel; Fs, developing fibers. Note that the development of vessels is completed within the region of cell expansion, and that the fibers develop at a much slower pace, visualized by the gradual loss of the cytoplasm of the fibers. The asterisks denote fibers close to the moment of death with barely detectable cytoplasm. ::: ![](gb-2005-6-4-r34-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Assignment of gene ontologies to ESTs in the cDNA libraries representing fiber death, tension wood and leaf senescence in the POPULUSDB. ::: ![](gb-2005-6-4-r34-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The most abundant transcripts in the fiber death library. The EST distribution in the different cDNA libraries of the POPULUSDB is shown for the 28 most abundant transcripts of the fiber death library. The transcripts are shown in descending order of EST abundance in the library. For a detailed description of the cDNA libraries, see \[13\]. The color code at the bottom of the figure indicates the number of ESTs in each library for each transcript. ::: ![](gb-2005-6-4-r34-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### RT-PCR analysis of gene expression in different parts of a Populustree. Expression of glycosyl hydrolase family 1 protein (POPLAR.11628), proline-rich protein (POPLAR.11658), oligopeptide transporter (POPLAR.11639), an expressed protein (POPLAR.11646), an F-box protein (POPLAR.11624) and a protease inhibitor/seed storage/lipid transfer protein (POPLAR.9335) is shown in relation to 18S DNA expression. Sampling of the different tissue types is described in Materials and methods. ::: ![](gb-2005-6-4-r34-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Gene expression in the 25K PopuluscDNA array and in the POPULUSDB. The 50 most highly expressed transcripts in the microarray analysis of xylem fibers undergoing cell death are shown together with the corresponding library distribution in POPULUSDB. The transcripts are listed in descending order of expression ratio between the fiber-cell death sample and the early fiber development sample (see Figure 1). The color code at the bottom of the figure indicates the number of ESTs in each library for each transcript. The whole list is given in Additional data file 2. ::: ![](gb-2005-6-4-r34-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The unique transcripts in the fiber death library within the POPULUSDB ::: Cluster Transcript Number of ESTs Most similar Arabidopsisgene E Cellular compartment Expression ratio B/A P B --------- --------------------------------------- ---------------- -------------------------------- ---------- ----------------------- ---------------------- ------------ ------ 11628 Glycosyl hydrolase family 1 protein 9 At3g60130 4.2E-126 Mitochondrion 29.0 0.00000060 11.8 11658 Proline-rich protein family 6 At2g27390 1.6E-18 Membrane 2.5 0.00015 3.1 11639 Oligopeptide transport family protein 5 At2g37900 1.3E-05 Membrane 2.8 0.00076 1.5 11646 Expressed protein 4 At2g22080 6.1E-01 Endomembrane system 7.9 0.0000091 7.2 11619 Expressed protein 4 At3g45160 2.8E-10 Endomembrane system 4.8 0.0000061 7.8 11625 Transferase family protein 4 At2g25150 3.8E-90 1.7 0.00036 1.9 11648 Basic helix-loop-helix family protein 4 At1g61660 2.9E-54 Chloroplast 1.3 0.14 -5.5 11665 Peroxidase 3 At1g05260 5.3E-119 Endoplasmic reticulum 5.5 0.0000079 7.4 11640 Expressed protein 3 At3g19340 1.2E-43 Membrane 1.1 0.83 -7.7 11662 Expressed protein 3 \- \- \- 0.8 0.30 -6.1 11630 PHD-type zinc finger protein 3 At2g36720 1.2E-29 Nucleus 0.7 0.0014 0.2 11653 Ubiquinone reductase 3 At1g76340 9.9E-01 Membrane 0.5 0.010 -1.9 The table shows clusters that are unique to the fiber death library and have an EST abundance above two. The number of EST sequences in each cluster, the most similar Arabidopsisgene and the BLASTX value (E), and the corresponding gene expression from the microarray analysis are shown. The expression ratio is calculated between gene expression in sample B (fiber-cell death) and sample A (early fiber development). The transcripts were each represented on the microarray by a single cDNA clone. A statistically significant difference in gene expression is predicted for transcripts having P\< 0.001 (t-test) and B\> 0 (Bayesian statistics \[55\]). The whole list can be seen in Additional data file 1. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Expression of cysteine, serine and aspartic proteases in POPULUSDB and in a microarray analysis ::: Protease Cluster Transcript Most similar Arabidopsisgene E Number of ESTs in each library Expression ratio B/A ---------- --------- ------------------------------------------- -------------------------------- ---------- -------------------------------- ---------------------- --- ---- ----- -------- ------- ------- ------- ------- ----- Cysteine 11539 Cysteine endopeptidase CysEP At5g50260 7.1e-146 0 4 0 0 30 \14.9 1250 Papain-like cysteine proteinase RD19A* At4g39090 1.4e-133 51 23 2 3 211 1.2 \3.0 1.1 \3.4 \4.1 1.4 289 Papain-like cysteine proteinase At5g43060 1.3e-103 16 1 1 1 32 0.9 1.4 1.4 3027 Vacuolar processing enzyme At4g32940 9.2e-200 4 8 1 1 34 \3.0 \3.8 \3.1 \3.3 3310 Papain-like cysteine endopeptidase XCP2* At1g20850 1.7e-144 0 5 1 2 14 \0.7 \0.2 4102 Papain-like cysteine proteinase SAG12 At5g45890 8.8e-111 13 0 0 0 14 1.0 0.9 5498 Papain-like cysteine proteinase AALP At5g60360 3.4e-129 4 1 0 0 18 1.3 \1.8 \2.3 7212 Papain-like cysteine proteinase RD21A At1g47128 2.6e-169 28 5 2 3 59 2.2 Serine 10138 Subtilisin-like serine protease At1g01900 9.1e-107 0 3 0 0 5 \5.0 4702 Prolyl endopeptidase At1g76140 1.6e-52 4 0 0 0 9 1.1 1.2 4905 Serine carboxypeptidase III-like protein At3g10410 9.8e-194 0 4 0 2 16 1.6 1.3 4995 Subtilisin-like serine protease At1g20160 7.0e-68 0 5 0 0 6 \12.2 8771 DegP protease At3g27925 6.2e-60 9 0 0 0 9 1.6 Aspartic 3648 Aspartic-type endopeptidase (phytepsin) At1g11910 3.5e-208 6 3 0 1 26 0.9 \2.5 4286 Aspartic-type endopeptidase (family A1) At5g43100 3.5e-41 3 1 0 0 4 1.0 1.2 1.2 541 Aspartic-type endopeptidase (family A1) At2g03200 3.0e-124 0 6 0 10 19 1.1 0.7 915 Aspartic-type endopeptidase (phytepsin) At1g11910 1.3e-105 3 3 1 1 35 1.2 1.0 1.3 Distribution of the ESTs is shown in the cDNA libraries of senescing leaves (I), fiber death (X), virus/fungal infected leaves (Y) and tension wood (G) of the POPULUSDB. Tot, the total number of ESTs in the POPULUSDB. Note that the Y library was partially subtracted and only 1,395 ESTs were sequenced from this library. Erepresents BLASTX statistics for the most similar Arabidopsisgene. The expression ratio was calculated between gene expression in sample B (fiber-cell death) and sample A (early fiber development). Several transcripts were represented on the microarray by multiple cDNA clones. The asterisks denote statistically significant differences in gene expression when P\< 0.001 and B*\> 0. The complete microarray data, including statistics, can be found in Additional data file 2. :::	PubMed Central	2024-06-05T03:55:56.372243	2005-3-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088962/", "journal": "Genome Biol. 2005 Mar 22; 6(4):R34", "authors": [ { "first": "Charleen", "last": "Moreau" }, { "first": "Nikolay", "last": "Aksenov" }, { "first": "Maribel García", "last": "Lorenzo" }, { "first": "Bo", "last": "Segerman" }, { "first": "Christiane", "last": "Funk" }, { "first": "Peter", "last": "Nilsson" }, { "first": "Stefan", "last": "Jansson" }, { "first": "Hannele", "last": "Tuominen" } ] }
PMC1088963	Background ========== Global interaction data are synthetically structured as networks, their nodes representing the genes of an organism and their links some, usually indirect, form of interaction among them. This type of schematization is clearly wiping out important aspects of the detailed biological dynamics, such as localization in space and/or time, protein modifications and the formation of multimeric complexes, that have been lumped together in a link. Given these limitations, an important open question is whether the backbone of the interaction network provides any useful hints as to the organization of the web of cellular interactions. A first observation in this direction is that the topology of biological interaction networks strongly differs from that of random graphs \[[@B1]\]. In particular, when transcriptional regulatory networks are compared to randomized versions thereof, some special subgraphs, dubbed motifs, have been shown to be statistically over-represented \[[@B2],[@B3]\]. An example of a motif composed of three units is the feed-forward loop, its name being inherited from neural networks, where this pattern is also abundant. Transcription factors often act in multimeric complexes and the formation of these plays a crucial role in the regulatory dynamics. In order to capture at least part of those effects, transcriptional networks may be integrated with the protein-protein interaction data that have recently become available \[[@B4]-[@B7]\]. An example is provided by the mixed network constructed in \[[@B8]\]. The network is mixed in the sense that it includes both directed and undirected edges, pertaining to transcriptional and protein-protein interactions, respectively. The motifs for the mixed networks were investigated in \[[@B9]\]. The dynamics of motifs has been thoroughly investigated in vitroand in silico, that is, in the absence of the rest of the interaction network and of additional regulatory mechanisms \[[@B10]-[@B12]\]. For instance, the feed-forward loop has remarkable filtering properties, with the downstream-regulated gene activated only if the activation of the most-upstream regulator is sufficiently persistent in time. The motif essentially acts as a low-pass filter, with a time-scale comparable to the delay taken to produce the intermediate protein. Furthermore, the same structure is also found to help in rapidly deactivating genes once the upstream regulator is shut off. Overabundance of motifs and their interpretation as basic information-processing units popularized the hypothesis of an evolutionary selection of motifs \[[@B2],[@B13]\]. In electrical engineering circuits, an abundant structure is likely to correspond to a module that performs a specific functional task and acts in a manner largely independent of the rest of the network. The point is moot for biological networks. A recent remark is that some of the motifs found in transcriptional networks are also encountered in artificial random networks \[[@B14],[@B15]\], where no selection is acting. However, the lists of motifs do not entirely coincide for the two cases \[[@B16]\]. A visually striking fact is that essentially none of the motifs exists in isolation and that there is quite a great deal of edge-sharing with other patterns (see \[[@B17]\] for the network of Escherichia coli). The function of the motifs might then be strongly affected by their context. The use of genetic algorithms to explore the possible structures that perform a given functional task has in fact shown a wide variety of possible solutions \[[@B18]\]. It is therefore of interest to address the issue of the functional role of the motifs in vivo, that is within the whole network, and examine the ensuing evolutionary constraints. In the following, we shall show that the instances of the network motifs are not subject to any particular evolutionary pressure to be preserved and analyze the biological information available on the pathways where some instances of motifs are found. Results ======= List and annotation of network motifs ------------------------------------- The first step in the analysis of network motifs is their identification, as described in detail in Materials and methods. The patterns whose number of counts in the real network is found to significantly deviate from the typical values found in the randomized ensemble of the network are shown in Figure [1](#F1){ref-type="fig"} (a generic representation of all the three-gene patterns independently of their statistical significance is given in Additional data file 1). The order of the patterns which we have examined are n= 2 and n= 3, where nis the number of genes of the pattern (see Materials and methods for the case of self-interactions). The list includes the purely transcriptional feed-forward loop, investigated in \[[@B10]-[@B12]\], and its version augmented with a proteic interaction \[[@B9]\]. The overall list is quite similar to that found in \[[@B9]\], with the only exception of proteic self-interactions, which were not taken into account. General information on the motifs is obtained by looking at the biological processes, molecular functions and cellular components for which the genes found in occurrences of Figure [1](#F1){ref-type="fig"} motifs have been annotated (see Additional data files [1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}). Let us first remark that the various instances of the motifs account for 25% of all the genes annotated as transcription factors in the MIPS/FunCat and GeneOntology (GO) databases. The annotations obtained using the former database indicate that 34% of the genes involved in motifs are annotated as involved in transcriptional regulation and 31% in direct control of transcription; and that 51% of the genes have their products localized within the nucleus. These values should be compared to 5% of all the genes annotated for transcriptional control in either GO or FunCat and 30% of nuclear localization for all annotated genes. Another relevant remark is that transcription factors are found at 93% and 11%, respectively, of the nodes with an outgoing and an ingoing transcriptional link. That is, indeed, the expected behavior for genes in a transcriptional network. These results witness the coherence of the transcription and the protein-protein interaction datasets used for finding the motifs and the published annotations. As for the function of the genes composing the network motifs, the list of the most represented biological processes, as annotated in the MIPS database, is as follows: 50% of the genes are involved in metabolism, 34% in transcription, 21% in cell cycle and DNA processing, 12% in interaction with the cellular environment (10% in cellular sensing and response), 10% in cellular transport and 9% in rescue/defense. As shown clearly in Figure [2](#F2){ref-type="fig"}, motifs are generally combined into larger interaction sub-networks. Among the 504 instances of motifs in Figure [2](#F2){ref-type="fig"}, only four occur in isolation whereas all the others share genes and/or edges. This is also clear when we consider that only 256 different genes compose the 504 motif instances; 1,487 different genes would be possible if the instances were disjoint. Shared edges and/or genes and those forms of interactions not included in our database are likely to strongly affect the function of the motifs, raising the issue of their role in vivo. This will be the subject of the analysis presented in a further paper. Phylogenetic profiles of network motifs --------------------------------------- To ascertain the presence of any special evolutionary pressure acting to preserve over-represented patterns, we have performed a protein comparative analysis between Saccharomyces cerevisiaeand the four hemiascomycetes Candida glabrata, Kluyveromyces lactis, Debaryomyces hanseniiand Yarrowia lipolytica, recently sequenced in \[[@B19]\]. The fact that the four organisms share many functional similarities with S. cerevisiaeand yet span a broad range of evolutionary distances, comparable to the entire phylum of chordates, makes them ideal for protein comparisons. Details of the sequence comparisons are reported in Materials and methods. Previous evolutionary studies on the motifs have explored the presence of common ancestors in different instances of the motifs. The upshot was that the various instances are not likely to have arisen by successive duplications of an ancestral pattern \[[@B20]\]. Here, we consider a different statistic based on the phylogenetic profiles \[[@B21]\] of the genes within the motifs. The profiles are constructed considering an ensemble of organisms and looking at the co-occurrences in the compared organisms of the genes composing the interaction pattern. This is quantified by the evolutionary fragility, F~i~(as defined in Materials and methods), of the interaction pattern i. A small value for the fragility indicates that the genes composing the pattern tend to co-occur in the other compared organisms, hinting at an evolutionary pressure to preserve the pattern and at its functional importance. We shall compare the statistics of the evolutionary fragility for different classes of interaction patterns, thus providing a test of the evolutionary significance of the criterion of overabundance used to identify network motifs. Specifically, in Figure [3](#F3){ref-type="fig"} we report the normalized histograms of the evolutionary fragilities F~i~for three different classes of interaction patterns composed of three nodes: patterns which are instances of the motifs; all the interaction patterns, irrespective of their abundance; and patterns composed of genes taken at random. There are 481 instances of motifs in a total number of 9,962 patterns involving three nodes. Subtracting the 481 from the overall ensemble does not modify the conclusions drawn from Figure [3](#F3){ref-type="fig"}. The histogram for genes taken at random is clearly different from the other two, as expected. The point of interest to us here is that there is no statistically significant difference between the first two classes of patterns, as quantified by a χ^2^test, which gives χ^2^= 4.454 and a one-tailed probability 0.348. This clearly supports the hypothesis that the series of data for the two histograms are drawn from the same distribution. The conclusion of our comparative analysis is that instances of network motifs undergo no special evolutionary pressure as compared to a generic interaction pattern. Function in vivoof realizations of the motifs ----------------------------------------------- Biological information currently available is not sufficient to ascertain the function in vivoof all the occurrences of the motifs previously found. Some of them are, however, placed within well studied pathways and, in particular, a few of them are located at the interface between two blocks, one responsible for conveying a signal and the other for processing it. Two examples are the sub-networks methionine synthesis (MET) and nitrogen catabolite repression (NCR), shown shaded in Figure [2](#F2){ref-type="fig"} and in more detail in Figure [4](#F4){ref-type="fig"}. The former, which is involved in methionine synthesis, receives a signal from the concentration of S-adenosylmethionine (AdoMet), a final metabolite of the sulfur amino acid pathway, and controls genes encoding enzymes involved in the pathway. The sub-network NCR, involved in nitrogen metabolism, receives a signal through the protein Gln3p, which is made available when nitrogen-rich sources are depleted, and controls genes encoding enzymes and transporters able to exploit alternative sources. The importance of these pathways has made detailed biological information on their functions available. The interface location of the identified instances of the motifs raises the hope that they might be implicated in the dynamics of the information processing and, in particular, that the time-filter properties mentioned above might be exploited to control the time-response processing of the external signal. Ascertaining this behavior was our motive for investigating the detailed functioning of each of the pathways. We report here the principles of the core regulatory mechanisms involved in the chosen pathways, referring the reader to the cited literature for a detailed treatment. Here we are interested in identifying the possible role of motifs in biological functions. The methionine pathway ---------------------- Sub-network MET in Figures [2](#F2){ref-type="fig"} and [4a](#F4){ref-type="fig"} shows the interaction graph for the cluster of interacting genes centered on CBF1, MET4and MET28. The graph includes three motifs of type II.2, five of type III.5 and one of type III.7 (see Figure [1](#F1){ref-type="fig"} for motif types). The methionine biosynthesis network has been thoroughly investigated \[[@B22]-[@B25]\] and a detailed biological model of the pathway is now available. Cbf1p, Met4p and Met28p form a heterotrimer that activates target genes of the sulfur pathway (MET genes). Inside the complex, only Met4p has direct transcriptional action, with Cbf1p being involved in chromatin rearrangement and Met28p tethering the complex to the DNA. The MET genes are activated by the complex, but are repressed when one of the final metabolites of the pathway, AdoMet, increases. Two loops drive the dynamics of complex availability, sketched in Figure [4a](#F4){ref-type="fig"}. One is a positive loop: the Met4p complex regulates the transcription of MET28, its product stimulating the tethering of the complex to DNA. This loop is responsible for the increase of the dynamic response when the intracellular AdoMet concentration is low (the transcription of MET4is constitutive). The other is a negative loop: Met4p controls its own fate by regulating the transcription of MET30. The product of the latter is an ubiquitin ligase, which triggers the degradation of Met4p when AdoMet increases. This loop is expected to control high detrimental accumulation of AdoMet. Note that the latter post-transcriptional mechanism is, by definition, not captured by the network, which is limited to transcriptional regulations. Furthermore, an intrinsic limitation of network structures should be noted: the three proteins Cbf1p, Met4p and Met28p always act as a complex. This information does not unambiguously emerge from the topology of the network (Figure [4a](#F4){ref-type="fig"}, left), as the topology is also compatible with the three proteins acting separately. In conclusion, the key features of the methionine synthesis pathway do not seem to hinge on transcriptional regulation via the motifs instances shown in Figure [4a](#F4){ref-type="fig"}. Nitrogen catabolite repression (NCR) system ------------------------------------------- The NCR system shown in Figures [2](#F2){ref-type="fig"} and [4b](#F4){ref-type="fig"} is used by the cell to control the synthesis of proteins capable of handling poor sources of nitrogen. NCR-sensitive genes are not activated when rich sources are available, whereas they get expressed when only poor sources are left. Two II.1 and one II.4 motifs are embedded in this system. DEH1and DAL80are part of the GATA gene family and are known transcriptional repressors, regulating nitrogen catabolite repression via their binding to the GATA sequences upstream of NCR-sensitive genes. For several targets, the two repressors are in competition with Gln3p and Gat1p, which are transcriptional activators binding the same sequences. The accepted mechanisms of NCR are as follows (\[[@B26]-[@B28]\] and see Figure [4b](#F4){ref-type="fig"}). First, in the presence of rich nitrogen sources (ammonia and/or glutamine), Gln3p and Gat1p are sequestered in the cytoplasm and can activate neither NCR-sensitive genes nor DEH1and DAL80. The consequence of the low concentration of Gln3p in the nucleus is a low-level expression of DEH1, DAL80and NCR-sensitive genes. Second, when poor sources only are available (such as urea, prolin, or GABA), Gln3p and Gat1p are released into the nucleus. The former activates GAT1and the two proteins together activate NCR-sensitive genes. After a delay (due to the time taken for transcription and translation), Dal80p and Deh1p are expressed and competitively inhibit these same genes. Interesting dynamic behavior takes place during a transition from rich to poor nitrogen sources, when the cell must cast about for alternative sources, which implies the synthesis of new proteins. The amount of these proteins synthesized must be sufficient to ensure utilization of the new sources but, because of the depletion of nutrient sources, they should not be too high. NCR-sensitive genes are therefore activated only for the limited period of time when Gln3p and Gat1p are present but Dal80p and Deh1p are not. The negative feedback of DAL80on its activator GAT1is the mechanism ensuring that oscillatory behavior. To summarize, the role of the motifs identified in the NCR system is not evident and the entire mechanism of the NCR, within the model currently accepted on the basis of the present knowledge, can be described without any reference to them. Pseudohyphal growth/mating MAPK system -------------------------------------- The sub-network HYPHE in Figure [2](#F2){ref-type="fig"} and Figure [4c](#F4){ref-type="fig"} is formed by one motif of type III.5, involving the two genes STE12and TEC1. These genes both code for a transcription factor and are located downstream of the mitogen-activated protein kinase (MAPK) signal transduction pathway that controls both the pseudohyphal growth of the yeast and its mating response to pheromones. These signal transductions constitute a striking example of a signaling pathway shared by two different signals and yet responding specifically to each of them. It is therefore the object of detailed investigation and much data are available \[[@B29]\]. The phenomenology of the regulatory process is summarized as follows: in response to pheromones, Ste12p binds specifically to the pheromone response elements (PRE) of genes involved in the mating process; under conditions of starvation, a heterodimer composed of Tec1p and Ste12p binds to genes involved in pseudohyphal growth. The fact that STE12regulates TEC1raises the possibility that the switch between the two shared pathways of response to pheromones and pseudohyphal growth be realized by the instance of the feed-forward III.5 motif in the HYPHE sub-network. However, there is quite clear evidence that this is not the case, the most direct indication being provided in \[[@B30]\], where it is shown that the level of expression of TEC1does not correlate with pseudohyphal growth. Recent work indicates that the switch is instead realized via post-transcriptional phosphorylation effects, controlled by the two kinases Fus3p and Kss1p, and affecting the multimerization of Ste12p. Fus3p and Kss1p constitute the final layer of the MAPK system and are differentially activated in the two pathways (see, for example \[[@B31]\]). Regulation of early meiotic genes --------------------------------- The sub-network around IME1in Figure [2](#F2){ref-type="fig"} and Figure [4d](#F4){ref-type="fig"} is made of one II.1, two III.5 and one III.6 motifs and is implicated in the activation of early meiotic genes. The process of regulation of entry into meiosis and the early activation of the relevant genes has been studied in great detail and is summarized in \[[@B32]\]. In short, the meiotic pathway in yeast is initiated by the expression and activation of IME1, which serves as the master regulatory switch for meiosis \[[@B33]\]. Expression of IME1requires the integration of a genetic signal, indicating that the cell is diploid, and a nutritional signal, indicating that the cell is starved. The point of interest here is to ascertain if the processing of these signals takes place at the transcriptional level by the instances of the motifs in the sub-network. This does not seem to be the case. The information processing is rather implemented by alternative routes and the picture of the interactions shown on the sub-network CCYCLE in Figure [2](#F2){ref-type="fig"} and Figure [4d](#F4){ref-type="fig"} (left) appears to be insufficient and misleading. The repression of IME1by RME1has a major role in cell-type control, and IME1expression does not involve the regulation of RME1by the complex Ume6p-Sin3p, as suggested by the sub-network CCYCLE in Figure [2](#F2){ref-type="fig"}. This is realized through the cell-type specific a1 and α2 proteins, which combine in diploid cells and bind specifically to sites in the promoter of RME1to repress its expression \[[@B32],[@B33]\]. The integration of the nutritional signal is processed by both IME1and IME2and is considerably more complex than cell-type regulation, its main steps being reviewed in \[[@B34]\]. For instance, the IME1promoter has at least 10 separate regulatory elements. IME2is also regulated by several distinct signals, integrated at a single regulatory element, the upstream repression site URS1, which is bound by the Ume6p transcription factor under all conditions tested. The activation of IME1and IME2depends on the multimerization of Ume6p with several other proteins regulated either positively or negatively by at least two kinases, Rim11p and Rim15p. Other non-transcriptional mechanisms of gene control (such as targeted degradation) appear also to be involved in the regulation of this process \[[@B35]\]. The motifs in the sub-network CCYCLE fail to capture the complexity of these interwoven interactions. Pleiotropic drug resistance (PDR) system ---------------------------------------- The PDR system is used by the cell to counter the action of a broad spectrum of toxic substances; by activating membrane efflux pumps and modifying the membrane composition, the concentration of these substances is then decreased. Two genes, PDR1and PDR3, encode homologous transcription factors \[[@B36],[@B37]\], which drive multidrug resistance by activating genes involved in active transport and lipid metabolism \[[@B38],[@B39]\]. The corresponding sub-network (named PDR in Figure [2](#F2){ref-type="fig"} and [4e](#F4){ref-type="fig"}) is composed of eight motifs of type III.1 (so-called feed-forward loops) and one of type II.1, showing a star-like configuration with PDR1and PDR3in a central position. In vivo, those two genes have apparent functional redundancy: they target the same genes and the deletion of either PDR1or PDR3does not significantly affect the PDR system; an effect is only shown when both are deleted \[[@B40],[@B41]\]. However, these two factors are used in response of two different cell signals: PDR3is sensitive to mitochondrial activity, whereas PDR1is not \[[@B42]-[@B44]\]. Conversely, PDR1deletion mutants are quite drug-hypersensitive, whereas PDR3mutants are not \[[@B41]\]. In addition to this distinct response of PDR1and PDR3to cellular signals, the regulation link between them is weak, and no proof of cooperativity for the regulation of their targets was highlighted. It the PDR sub-network, the III.1 motifs formed by PDR1, PDR3and their common targets are apparently not exploited by the cell because PDR1and PDR3are not obligatorily active at the same time and the prerequisites for the specific dynamics of feed-forward loops are not fulfilled (sufficient regulation of PDR3by PDR1and cooperativity on the common targets). Discussion ========== The motivating idea behind most discussions on motifs is the possibility of capturing the essential logic of genetic regulation by a small set of interaction circuits performing some specific functional tasks. While this hypothesis is, in principle, experimentally testable, experimental and theoretical work has hitherto considered essentially motifs in isolation, that is, excised from the biological environment in which the motifs\' instances are embedded. We studied in detail the role of motifs in the case of the best-documented genetic sub-networks and biological functions where such motifs are found. In most cases, motifs do not seem to have a central regulatory role in the biological processes associated with each occurrence. The list of examples where enough biological information is available is, of course, limited, and further examples may subvert this picture. At the moment, it is a fact that all the examples studied highlight the high level of integration of different regulatory mechanisms acting altogether. Reception and processing of cellular signals cannot be reduced to transcriptional regulation and protein-protein interaction switches. Other mechanisms such as phosphorylation, triggered degradation, protein sequestration and transport, and higher-order multimerization are central to the logic of the sub-networks. Disentangling information-processing circuits made of transcription reactions and interactions between transcription factors from the whole cellular environment does not seem to be possible for the cases considered. A qualitative impression surmised from the visible aggregation and nesting of the motifs with the rest of the network is that a \'pure\' modular functional behavior is not very likely to occur. This impression is not limited to S. cerevisiae: in previous work \[[@B17]\], other researchers have shown that a similar aggregation of structural motifs occurs for a simpler organism, E. coli, suggesting some degree of generality. Some comments on structuring interaction data in the form of topological networks are worth making. The graph is indeed an abstraction constructed from available databases and its meaning is influenced by several factors. For instance, the graph is a static projection of possible interactions. The analysis of regulatory processes varying in space and time requires additional information not usually included in the topology of biological networks. Indeed, the very representation in the form of a unique network entails the integration in space and time of the interactions taking place during the cellular lifetime. Some of the patterns of interaction might then be spuriously due to a projection effect, whereas they actually take place at different times and/or locations within the cell. This is occurring, for example, in the PDR system: PDR1and PDR3at the base of the eight III.1 motifs respond to different signals and control their outputs independently (no cooperation on the common targets). These motifs appear in the network because different conditions at different times were projected onto the same plane. Furthermore, the patterns in the network may be a direct consequence of the data models in the current databases, and incorrectly represent the biological context. Transitory macromolecular associations like protein complexes and interactions between a whole protein complex and a target are indeed missed, and at most represented as individual links between each component and the target. This is what occurs with the Met4p/Met28p/Cbf1p heterotrimer, which appears in the network as three independent interacting components together with three III.5 motifs that do not actually exist. The NCR system is an interesting example where motifs are clearly identified and seem unambiguous. However, to the best of our knowledge they do not play any significant role. In particular, the role of the mutual interactions between DAL80and DEH1(sustaining a II.4 motif) is not clear. An intriguing hypothesis is that the presence of the interactions might be traced back to the strong sequence similarity between DAL80and DEH1. The products of both these genes form homodimers and inhibit their own expression. The presence of the motif might then be due to a recent duplication event, which has therefore preserved the interactions. Divergent evolution seems also to be the origin of the appearance of motifs in the PDR system. In this case, the two diverging genes PDR1and PDR3have acquired different independent functions. The motif instance that they form together is the apparently unexploited consequence of their common origin. Conclusion ========== The results presented here indicate that the statistical abundance of network motifs has no evident counterpart at the evolutionary and in vivofunctional level. Occurrences of network motifs have indeed been shown to possess the same evolutionary fragility; that is, when different organisms are compared, the genes composing the motif have similar co-occurrence profiles as genes in interaction patterns with a normal abundance. The point seems to be confirmed by the analysis of the functional role of examples of the motifs occurrences. These are located at the interface between two blocks - one responsible for the reception of a signal and the other for its processing - and have been selected because detailed biological information on those pathways is available. The number of cases is limited, but in none of them are the major steps of signal information processing taking place at the transcriptional level through the implementation of the motifs. Alternative routes involving post-transcriptional regulation and intracellular compartmentalization seem to be exploited for this purpose. These results naturally bring up the issue as to the actual role of the motifs. Some occurrences have been shown to arise spuriously from the representation of the interaction data in the form of a network and the ensuing projection effects in space and/or time. It seems, however, fair to assume that those effects should be limited to a few cases. The metabolic costs of producing proteins and the fact that some of the motifs instances examined are active in conditions of starvation make it likely that proteins encoded by genes composing these motifs do play a role. What is however quite clear from Figure [2](#F2){ref-type="fig"} and our analysis is that the great majority of motif occurrences are in fact embedded in larger structures and entangled with the rest of the network. Only a small minority is isolated and likely to perform a specific functional task that does not depend on the context. This clustering is important as it indicates that the choice of the null model used to gauge the statistical importance of the abundance of interaction patterns might be delicate. Indeed, the higher-order context is not taken into account in the randomization process used to generate the null model networks, and we have shown that this is manifestly not a choice ensuring a strong evolutionary and (in vivo) functional significance. Accounting for the various layers of organization of biological networks seems crucial to correctly identify the functional elements responsible for the information processing that allows living cells to cope with their highly variable environmental conditions. Materials and methods ===================== Datasets -------- The transcriptional regulatory network used for the analysis is the one constructed and investigated in \[[@B45]\]. It was preferred to the more extended one derived from ChIP-chips data in \[[@B46]\] as the fraction of links where the regulatory role of the various interactions is documented is higher for the former. The protein-protein interaction data in the Database of Interacting Proteins (DIP \[[@B47]\]) are a large collection of both two-hybrid and TAP-tag data. The resulting network has 476 nodes, 905 directed transcriptional edges and 221 undirected protein-protein edges. Identification of motifs and network randomization -------------------------------------------------- The detection of n-node network motifs is performed along lines similar to those used in \[[@B2]\]. The method exhaustively scans the neighborhood of all the links in the network to search for the motif of interest, and then purges the list for repeated patterns. Randomized versions of the network are generated as follows. Links are swapped as in the Markov-chain algorithm used in \[[@B48]\], that is, two links between the couples of nodes (X~1~Y~1~) and (X~2~Y~2~) are replaced by (X~1~Y~2~) and (X~2~Y~1~). In our case, where the links might be transcriptional or protein-protein interaction, the links that are swapped must be of the same type. This procedure is guaranteed to preserve the single-point connectivity at each node of the network. As for the randomization procedure for n= 3 motifs, we want to avoid the possibility that higher-order motifs spuriously inherit statistical significance from lower orders. In other words, the randomized network ought to have the same statistics for all the patterns of order n= 2 as the real network. This is ensured by converging a simulated annealing, where the elementary steps are the swappings of the links previously described. The transition probabilities are weighted according to the difference: ![](gb-2005-6-4-r35-i1.gif) where the sum runs over all the patterns of order n= 2 and the c~i~values denote the number of patterns in the two types of networks. Statistically significant patterns are those where the number of counts has a low probability to be observed in the ensemble of networks obtained by randomization. Specifically, we require that the observed number of counts ![](gb-2005-6-4-r35-i2.gif), has a one-tailed probability: ![](gb-2005-6-4-r35-i3.gif) \- or the opposite inequality if the pattern is under-represented in the real network - to occur in the randomized ensemble. The probabilities are estimated from a Monte-Carlo sampling of 10,000 trials of the randomized ensemble distribution and the results are sensitive neither to the number of trials nor to the thresholds chosen. The probability distribution functions are often found to deviate from a Gaussian curve and the one-tailed probabilities are therefore directly measured from the normalized histograms without relying on z-scores. Note that patterns involving self-interactions are somewhat special, as their order n, which controls the type of random networks they should be compared to, does not coincide with their number of genes. For example, a single gene self-interacting is treated as an n= 2 pattern. The reason is that a sensible way of assessing the significance for this pattern is by having a fixed number of total proteic links and studying the fraction of them that are self-interactions. In other words, self-interactions are swapped throughout the randomization procedure with proteic links between two distinct proteins and their order is therefore n= 2. Sequence comparisons -------------------- BLAST searches were performed using BLASTP 2.2.6 \[[@B49]\] with the BLOSUM 62 matrix and affine gap penalties of 11 (gap) and 1 (extension). Putative orthologs were inferred from the primary sequence and keeping only bidirectional best hits to reduce the effect of the high number of paralogs in yeast genomes. Tables of bidirectional best hits were constructed by identifying the pairs of proteins in the two organisms compared which are the reciprocal best alignments. The significance of the alignments was quantified by the BLAST e-values and different thresholds were considered, ranging from 10^-1^to 10^-10^. Their choice does not affect the results presented in the body of the paper. Evolutionary fragility of interaction patterns ---------------------------------------------- Let us consider all the interaction patterns, indexed by i, composed of interacting genes of S. cerevisiaeand each one of the other four hemiascomycetes, indexed by α. The boolean variable f~iα~for the pattern iis taken equal to zero if the genes composing the pattern are all present/absent in the other organism αand is unity otherwise. Presence/absence is measured by using the list of bidirectional best hits discussed in the previous section. The selective pressure to preserve the pattern iis quantified by the fragility: ![](gb-2005-6-4-r35-i4.gif) The two extreme cases are F~i~= 0 and F~i~= 4 (the number of organisms compared). The two cases correspond to the genes composing the pattern co-occurring in all or none of the compared organisms, respectively. As an additional example, consider the case where the three genes composing an interaction pattern are all present in C. glabrata, K. lactisand D. hansenii(which are evolutionarily closer to S. cerevisiae) but one (or two) of them is absent in Y. lipolytica. The corresponding value of the fragility is F~i~= 1. Additional data files ===================== Additional data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} is a figure showing general three-gene patterns. Additional data file [2](#S2){ref-type="supplementary-material"} is a table showing motif occurrences. Additional data file [3](#S3){ref-type="supplementary-material"} is a table showing functions of the genes in motif occurrences. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Left: the two possible connectivity topologies between three genes. Each grey line can be any of the seven types of interaction represented on the right. Right: the different types of interaction between two genes and their products. Boxes: genes; green arrows: transcriptional regulation only; dashed lines with circles: protein-protein interaction of the genes products. 1-3: only transcriptional regulation without known ppi interaction, (1 and 2 are distinguished to account for different combinations in the diagrams on the left). 4: protein-protein interaction only. 5-7: transcriptional regulation and interaction between the genes products (without details on the role of the ppi interaction complex). The set of all possible three genes patterns is obtained with all the combinations of the interaction types shown on the right on the topologies shown on the left. The statistically significant patterns form a subset of 8 three genes motifs shown in Figure [1](#F1){ref-type="fig"} ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 The list of the motif instances found for the yeast Saccharomyces cerevisiae. Each line corresponds to a different realization and contains the most used non-ambiguous name of the involved genes, ordered according to their position in the motif. First column contains the motifs type according to Figure [1](#F1){ref-type="fig"}; columns 2 to 4 correspond respectively to the genes positions a, b and c as indicated in figure [1](#F1){ref-type="fig"} ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 The Excel file contains the list of genes found in motif realizations with their biological functions as given by the MIPS database using the FunCat ontology, and different statistics on function occurrences and distribution. The data is presented in three sheets with different viewpoints: First sheet, \"Functions by genes\":gives a list of all genes found in motifs instances with standard, main and alternative names, motifs and positions within motifs where these genes are found (according to types and positions as defined in Figure [1](#F1){ref-type="fig"}), and finally biological functions. Second sheet, \"Functions by positions\":gives motifs and positions within motifs grouped according to functions. For each represented function in FunCat, the first three columns indicate the number, the fraction and the names of the genes found in motifs instances having this function. The following columns indicate the details for each motif type with: the number of genes involved in the given motif with the given function, the fraction of all genes within motifs having this position and this function, the fraction of genes for this function that are at this position, and the fraction of genes at this position having this function. Third sheet, \"Genes by positions\":gives standard and main name of genes found at each position ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We are grateful to B. Dujon, P. Glaser and F. Képès for useful discussions. M.V.\'s research was supported in part by the National Science Foundation under Grant No PHY99-07949. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Types of motifs of order n= 2 and n= 3 for the mixed transcription and protein-protein network. The motifs shown here are those whose abundance patterns in the real network of the yeast Saccharomyces cerevisiaestrongly deviate from the typical values found in randomized versions thereof. The green directed links with arrows represent transcriptional links, while two dashed lines with contacting circles represent an undirected protein-protein interaction. ::: ![](gb-2005-6-4-r35-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Motif occurrence in yeast. The network graph of the occurrences of motifs for S. cerevisiaeillustrates the fact that most of the motifs are not found in isolation and are part of larger aggregates. Green, pure transcriptional regulation of the target gene by the regulatory gene product protein; red, transcriptional regulation and protein-protein interaction of the two partners; dashed line, pure protein-protein interaction. The pathways that will be examined in detail are shaded. ::: ![](gb-2005-6-4-r35-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Phylogenetic profiles of interaction patterns. Normalized histograms of the evolutionary fragility of interaction patterns belonging to the following three classes are shown: instances of network motifs (red); generic patterns of interacting genes, irrespective of their abundance (black); patterns composed of genes taken at random (white). The five possible values (in increasing value 0 to 4) of the evolutionary fragility are reported on the abscissa. A small fragility value indicates that all the genes composing the interaction patterns tend to co-occur in the other genomes compared and point to evolutionary pressure acting to preserve the interaction pattern. ::: ![](gb-2005-6-4-r35-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Outlines of the pathways studied. (a)Methionine (MET); (b)nitrogen catabolite repression (NCR); (c)pseudohypal growth/mating (HYPE); (d)regulation of early meiotic genes (CCYCLE); (e)pleiotropic drug resistance (PDR). The sub-networks enlarged from Figure 2, with the identified motifs within the pathway drawn from the interaction databases, are shown on the left (colors and conventions are the same as in Figure 2). A schematic representation of the regulation mechanisms for the same pathways, based on the present experimental knowledge as discussed in the text, is shown on the right. Full lines represent transcriptional regulation, dashed lines non-transcriptional regulation, and wavy lines transformations and syntheses. Arrowheads, positive regulation; lines ending in a terminal bar, negative regulation. ::: ![](gb-2005-6-4-r35-4) :::	PubMed Central	2024-06-05T03:55:56.377930	2005-3-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088963/", "journal": "Genome Biol. 2005 Mar 24; 6(4):R35", "authors": [ { "first": "Aurélien", "last": "Mazurie" }, { "first": "Samuel", "last": "Bottani" }, { "first": "Massimo", "last": "Vergassola" } ] }
PMC1088964	Background ========== Quantitative traits affecting morphology, physiology, behavior, disease susceptibility and reproductive fitness are controlled by multiple interacting genes whose effects are conditional on the genetic, sexual and external environments \[[@B1]\]. Advances in medicine, agriculture, and an understanding of adaptive evolution depend on discovering the genes that regulate these complex traits, and determining the genetic and molecular properties of alleles at loci that cause segregating genetic variation in natural populations. Assessing subtle effects of induced mutations on quantitative trait phenotypes in model organisms is a straightforward approach to identify genes regulating complex traits \[[@B1]-[@B3]\]. However, the large number of potential mutations to evaluate, the necessity to induce mutations in a common inbred background, and the level of replication required to detect subtle effects \[[@B1]\] all limit the feasibility of systematic whole-genome mutagenesis screens for complex traits in higher eukaryotes. Mapping quantitative trait loci (QTLs) affecting variation in complex traits to broad genomic regions by linkage to polymorphic molecular markers is also straightforward. However, our ability to determine what genes in the QTL regions cause the trait variation is hampered by the large number of recombinants required for high-resolution mapping, and the small and environmentally sensitive effects of QTL alleles \[[@B1],[@B4]\]. There has been great excitement recently about the utility of whole-genome transcriptional profiling to identify candidate genes regulating complex traits, by assessing changes in gene expression in the background of single mutations affecting the trait \[[@B5],[@B6]\], between lines selected for different phenotypic values of the trait \[[@B7]\], and in response to environmental stress and aging \[[@B8]-[@B12]\]. Transcript abundance is also a quantitative trait for which there is considerable variation between wild-type strains \[[@B11],[@B13]-[@B17]\], and for which expression QTLs (eQTLs) \[[@B18]\] have been mapped \[[@B15]-[@B17],[@B19]\]. Thus, candidate genes affecting variation in quantitative trait phenotypes are those for which the map positions of trait QTL and eQTL coincide \[[@B16],[@B20]\]. Transcript profiling typically implicates hundreds to thousands of genes in the regulation of quantitative traits and associated with trait variation between strains; the majority of these genes are computationally predicted genes that have not been experimentally verified. To what extent do changes in transcript abundance predicate effects of induced mutations and allelic variants between strains on quantitative trait phenotypes? It is encouraging that several studies have confirmed the phenotypic effects of mutations in genes implicated by changes in expression \[[@B5]-[@B7]\]. However, limited numbers of genes were tested, and their choice was not unbiased. None of the candidate QTLs nominated by transcriptional profiling has been validated according to the rigorous standards necessary to prove that any candidate gene corresponds to a QTL \[[@B1],[@B4]\]. To begin to answer this question, we need to compare gene-expression data with genes known to affect the trait from independent mutagenesis and QTL mapping studies. This comparison has not been possible to date because there are only a few complex traits for which the genetic architecture is known at this level of detail, one of which is resistance to starvation stress in Drosophila. Previously, we used P-element mutagenesis in an isogenic background to identify 383 candidate genes affecting starvation tolerance in D. melanogaster\[[@B21]\]. Further, we mapped QTLs affecting variation in starvation resistance between two isogenic Drosophilastrains, Oregon-R (Ore) and 2b \[[@B21]\], followed by complementation tests to mutations to identify twelve candidate genes affecting variation in starvation resistance between these strains \[[@B21]\]. Here, we used Affymetrix DrosophilaGeneChips to examine expression profiles of two starvation-resistant and two starvation-sensitive recombinant inbred (RI) lines, as well as parental lines Ore and 2b, under normal and starvation stress conditions. We used a statistically rigorous analysis to identify genes whose expression was altered between the sexes, during starvation stress treatment, between lines, and interactions between these main effects. In the comparison of expression profiling with the P-element mutagenesis performed previously, we found nearly 50% concordance between the effects of 160 P-element mutations on starvation stress resistance and changes in gene expression during starvation - 77 mutations with significant effects also had significant changes in transcript abundance, while 83 mutations did not affect the starvation resistance phenotype, yet had significant changes in transcript level. We identified 153 novel candidate genes for which there was variation in gene expression between the lines and which co-localized with starvation resistance QTLs. However, we did not detect genetic variation in expression for any of the candidate genes identified by complementation tests. Our efforts to associate genetic variation in expression with variation in quantitative trait phenotypes is confounded by the observation of widespread regulation of transcript abundance by unlinked genes, the difficulty in detecting rare transcripts that may be expressed in only a few cell types at a particular period of development, and genetic variation between QTL alleles that is not regulated at the level of transcription. Results ======= The sexually dimorphic transcriptome ------------------------------------ Nearly one-half of the genome (6,569 probe sets) exhibited significantly different transcript levels between the sexes (P(Sex) \< 0.001), with 3,965 probe sets upregulated in females and 2,604 probe sets upregulated in males (the complete list is given in Additional data file 1). The greatest differences in transcript abundance between the sexes were for probe sets implicated in sex-specific functions: chorion, vitelline membrane, and yolk proteins involved in egg production were upregulated in females; and accessory gland peptides, male-specific RNAs, and protein ejaculatory bulb components were upregulated in males. However, the probe sets exhibiting sex dimorphism in expression fell into 28 biological process and 41 molecular function Gene Ontology (GO) categories; for most of these categories, differences in expression between the sexes was unexpected. We determined which GO categories contained significantly different numbers of upregulated probe sets in males and females (Table [1](#T1){ref-type="table"}). Genes involved in the biological process categories of cell communication, cell growth and/or maintenance, development, and cell death were upregulated more often in females than in males. Genes involved in the molecular function categories of binding, most enzymes, signal transduction, structural molecules, and regulation of transcription and translation were upregulated in females more often than in males; however, genes encoding oxidoreductase enzymes, carrier transporters and ion transporters were upregulated in males more often than in females (Table [1](#T1){ref-type="table"}). The genomic distribution of sex-biased genes was not random (Figure [1](#F1){ref-type="fig"}). There was a paucity of male-biased genes on the Xand fourth chromosomes, and an excess on chromosome 2R(χ^2^~5~= 100.77; P\< 0.0001). There was a deficit of female-biased genes on chromosome 4, and an excess on chromosome 2R(χ^2^~5~= 29.18; P\< 0.0001). Transcriptional response to starvation stress --------------------------------------------- We found 3,451 probe sets with significantly different mean transcript levels between the control and starved conditions (P(treatment) \< 0.001): 1,736 were downregulated (some by as much as 40-fold) and 1,715 were upregulated (at most by 7.2-fold) during starvation (the complete list is available as Additional data file 2). These probe sets fell into 24 biological process and 25 molecular function GO categories. We determined which GO categories had a significantly different number of up- and downregulated probe sets in response to starvation stress. Genes affecting the biological processes of protein and nucleic-acid metabolism (protein biosynthesis; protein catabolism, folding, localization, modification, and repair; biosynthesis of nucleic acid macromolecules and lipids) were upregulated during starvation (Table [2](#T2){ref-type="table"}). The expression of genes in three molecular function categories (nucleotide binding, hydrolases binding to acid anhydrides, and ribosome structure) increased during starvation; while defense/immunity proteins, peptidases, cuticle structural proteins, and carrier transport proteins were downregulated (Table [2](#T2){ref-type="table"}). The treatment × sex interaction term was significant (P\< 0.001) for 817 probe sets, of which 715 had significant treatment effects for one or both sexes in the separate sex analyses (Additional data file 3). We categorized these 715 probe sets as sex-specific if significant expression changes in response to starvation occurred in one sex only; as sex-biased if expression levels changed in the same direction in both sexes, but were of different magnitude; or as sex-antagonistic if expression levels significantly changed in both sexes, but in opposite directions (Figure [2a-c](#F2){ref-type="fig"}). Most probe sets exhibited sex-specific or sex-biased expression, with only two genes, CG14095and Rpd3, meeting the sex-antagonistic criterion. More probe sets exhibiting sex-specific or sex-biased expression were downregulated (454) than upregulated (263) during starvation. Starvation stress was accompanied by reduced expression of genes involved in the developmental processes of gametogenesis and sex determination as well as signal transduction in females, and of genes involved in mechanosensory and reproductive behavior in males (Table [2](#T2){ref-type="table"}). Transcript abundance versus mutations ------------------------------------- The genes represented by probe sets with significant treatment and/or treatment × sex effects are candidate genes for starvation resistance. Previously, we screened 933 co-isogenic single P-element insertion lines for their effect on starvation resistance \[[@B21]\]. Of these insertions, 383 had significant effects on starvation resistance, while the remaining 550 did not \[[@B21]\]. Of the 933 lines, we know the locations of the 385 of the inserts and that genes tagged by these inserts are represented on the array. Thus, we can directly compare the extent to which effects of P-element mutations on the starvation phenotype correspond to changes in transcript abundance in response to starvation. This comparison allows us to assess the hypothesis that changes in transcript abundance can be used to identify candidate genes with effects on phenotype, an hypothesis implicit in previous microarray studies \[[@B5]-[@B7]\]. Overall, there was no statistical association between the phenotypic and transcript data (χ^2^~1~= 0.0006, P= 1). For 194 genes, there was agreement between the phenotype and the expression level. Seventy-seven genes had significant differences in both transcript profile and mutant phenotypes, and 117 genes affected neither phenotype nor expression level (Additional data file 4). There was disagreement between the expression and phenotypic analyses for 191 genes (49.6%): 108 of the genes tagged by P-elements affected starvation resistance, but did not display differences in transcript level in response to starvation stress, and P-element insertions in 83 genes that exhibited significant differences in transcription in response to starvation did not have significant phenotypic effects on starvation tolerance (Additional data file 4). The genetic architecture of transcription ----------------------------------------- A total of 706 probe sets exhibited variation in expression among the six lines; 640 probe sets were significant (P\< 0.001) for the main effect of line, 190 for the line × sex interaction, 200 for the line × treatment interaction, and 85 for the three-way interaction of line × sex × treatment (Additional data file 5, and Figure [2d-k](#F2){ref-type="fig"}). Thus, transcript abundance exhibits both genotype by sex and genotype by environment interaction. We used post-hocTukey tests to group lines with similar levels of gene expression, and compared the expression clusters with the Ore and 2b genotype of the six lines. There are three possible scenarios by which genetic variation in transcript abundance could arise. First, genetic variation in regulatory regions of gene Acauses variation in the expression of gene A(cis-acting regulatory variation). Second, genetic variation in regulation of gene Bcauses variation in expression of A, which is itself not genetically variable (trans-acting regulatory variation). Third, genetic variation in both gene Aand gene Baffect the transcript abundance of gene A(cis- and trans-acting regulatory variation). These two-locus interactions could be additive or epistatic. We observe whether or not expression of gene Aco-segregates with markers differentiating the two parental strains. Co-segregation will always be observed in case 1. It could also be observed in cases 2 and 3 if gene Bis tightly linked to gene A, such that it is not separated by recombination from Ain the genotypes tested. However, co-segregation will not be observed if gene Aand gene Bare unlinked. The most prevalent observation was regulation of expression by unlinked genes. For example, there were unambiguous interpretations for 246 probe sets that were significant for the main effect of line only: 65 (26.4%) were regulated by linked genes and 181 (73.5%) were regulated by unlinked genes (Additional data file 6, and Figure [2l-o](#F2){ref-type="fig"}). We also inferred linkage of genes regulating expression levels under control and starved conditions separately. There were unambiguous Tukey interpretations for 277 probe sets under control conditions, of which 32 exhibited linked regulatory variation (11.6%) and 245 were regulated by variation at unlinked genes (88.4%). For 244 probe sets under starved conditions, 46 were regulated by polymorphism at linked loci, (18.9%) and 198 were regulated by variation at unlinked genes (81.1%) (Additional data file 7). Association of genetic variance in transcription with QTLs ---------------------------------------------------------- Probe sets from the three-way ANOVA that are significant for the main effect of line and/or line × sex (P\< 0.001), but not significant for the line × treatment interaction terms, exhibit genetic variation in transcription among the six lines that is independent of the starvation treatment. A total of 489 probe sets met these criteria, and we know the cytological locations of 475 of the corresponding genes. Previously, RI lines derived from Ore and 2b have been used to map QTL affecting variation in life span \[[@B22]-[@B25]\], sensory bristle numbers \[[@B26]\], ovariole number \[[@B27]\], courtship signal \[[@B28]\], olfactory behavior \[[@B29]\], metabolism and flight \[[@B30]\], as well as starvation resistance \[[@B21]\]. Genes that exhibit significant differences for the main effect of line and/or line × sex which are located within QTL regions are putative candidate genes corresponding to the QTL \[[@B16],[@B20]\]. We identified several novel putative candidate genes affecting these traits (Additional data file 5). We examined whether probe sets with significant line and/or line × sex effects tended to cluster within regions containing QTL mapped under standard culture conditions, as would be the case if QTL regions were enriched for genes exhibiting transcriptional variation between the parental lines. We found no evidence for such clustering; indeed, the only trait showing a non-random association of probe sets with QTL that survived a Bonferroni correction for multiple tests was in the direction of a deficiency of probe sets in the QTL region (Table [3](#T3){ref-type="table"}). The 217 probe sets with significant line × treatment and/or line × treatment × sex terms (Additional data file 5) represent genetic differences among the lines in response to the starvation treatment. Are these probe sets enriched in regions to which starvation resistance QTL map? We found that 47 of the probe sets meeting these criteria, representing 45 unique genes, fell within starvation resistance QTL regions; and the remaining 170 probe sets, representing 169 unique genes, fell outside the QTL intervals. These probe sets were not over-represented within starvation resistance QTL (χ^2^~1~= 0.26, P\> 0.05). There is significant variation in starvation half-life among the six lines (P\< 0.0001; Additional data file 8). For those probe sets previously identified as having significant differences in transcript level among the lines, we assessed the extent to which variation in transcript abundance was associated with variation in starvation half-life. We found 281 probe sets with significant correlations (P\< 0.05) between starvation phenotype and transcript level, for 273 of which the cytological location was known (Additional data file 5). However, 66 of the probe sets associated with starvation half-life mapped to starvation resistance QTL, and 207 did not. Again, these probe sets were not over-represented within starvation resistance QTL (χ^2^~1~= 0.45, P\> 0.05). Although there is no tendency for genes exhibiting variation in transcript abundance among lines to cluster within starvation resistance QTLs, those that do co-localize with the QTLs are candidate genes affecting variation in starvation tolerance between Ore and 2b. We found 155 probe sets, corresponding to 153 candidate genes, which met one or more of the above criteria (Additional data file 5). Most (114, 75%) were predicted genes. The remaining genes (Table [4](#T4){ref-type="table"}) are reasonable candidates for starvation resistance QTLs, affecting the processes of protein metabolism, defense/immune response, proteolysis and peptidolysis, and transport. Complementation tests to mutations have implicated several candidate genes affecting variation between Ore and 2b in olfactory behavior \[[@B29]\] (Vanaso), longevity \[[@B31],[@B32]\] (Dopa decarboxylase, shuttle craftand ms(2)35Ci) and starvation resistance \[[@B21]\] (spalt major, Ryanodine receptor 44F, crooked legs, NaCP60E, Phosphoglucose isomerase, bellwether, numb, Punch, l(2)rG270, l(2)k17002, l(2)k00611, and l(2)k03205). None of these genes exhibited significant differences in transcript abundance between lines. Discussion ========== The sexually dimorphic transcriptome ------------------------------------ Consistent with previous reports \[[@B5],[@B11],[@B33],[@B34]\], we observed highly significant differences in transcript abundance between males and females for nearly half the genome. These differences in transcriptional profiles were not confined to stereotypical sex-specific biological processes. Female transcript levels were upregulated for genes involved in protein biosynthesis, metabolism, and transcription regulation, while male transcript levels were higher for probe sets involved in ion and carrier transporters, as in a previous study of sex differences in transcription in Drosophilaheads \[[@B5]\]. Differences in transcript abundance between the sexes may be an underlying mechanism for commonly observed sex-specific effects of QTLs associated with a variety of complex traits in Drosophila\[[@B21]-[@B26],[@B32],[@B35],[@B36]\] and other organisms \[[@B37]\]. Males and females are effectively different environments in which genes act. The chromosomal locations of genes with sex-dependent expression were non-random. We confirmed the apparently general phenomenon that the Drosophila Xchromosome is depauperate for genes that are upregulated in males \[[@B33],[@B34]\]; X-chromosome demasculinization is perhaps attributable to selection against genes that are advantageous in males but deleterious to females \[[@B33]\]. In contrast to previous studies, we observed that chromosome 2Rharbored an excess, and chromosome 4a deficiency, of genes that were upregulated in both males and females. Transcriptional response to starvation stress --------------------------------------------- The transcriptional response to starvation stress involved approximately 25% of the genome. The stress profile indicates upregulation of genes involved in growth and maintenance processes and protein biosynthesis, with increased transcription of genes encoding translation initiation and elongation factors, mitochondrial and cytosolic ribosomal structural proteins, and hydrolases involving acid anhydrides. This increase in protein biosynthesis and hydrolase activity can be interpreted as an attempt to use available proteins for nourishment. A similar phenomenon has been observed in the response of yeast \[[@B38]\] and mammalian cells \[[@B39]\] to starvation, where substantial protein and organelle degradation provides substrate to starving cells \[[@B40]\]. Our observation that peptidases, which catalyze the hydrolysis of peptide bonds, were significantly downregulated in response to starvation, is consistent with the preservation of nascent protein chains. The downregulation of carrier activity and defense/immunity proteins indicates that transport across cell membranes slows and the immune response is compromised in starving flies. We compared our results to those of a previous microarray study investigating gene-expression changes in starved larvae \[[@B41]\]. We found 21 probe sets that were significantly altered in both studies during starvation. Many of these genes have predicted functions that have not been verified experimentally; however, a few of the genes have known functions. Insulin-like Receptor, Serine pyruvate aminotransferase, Amylase distal, and mitochondrial carnitine palmitoyltransferase I, genes known to be involved in metabolism, were common to the two studies. Interestingly, Peroxidasin, a gene involved in oxygen and reactive oxygen species metabolism was upregulated fourfold in larvae, while it was downregulated 1.61-fold in our study. Starvation stress was accompanied by reduced expression of genes affecting gametogenesis, by as much as 66-fold in starved female flies. Egg components such as chorion, yolk, and vitelline membrane proteins were among the most severely restricted transcripts, implicating suppression of female reproductive function during starvation. This depressed reproductive function is not unique to flies, as female mice on a calorically restricted diet experience a cessation in estrous cycle \[[@B42]\] and amenorrhea is one of the hallmarks of anorexia nervosa in human females \[[@B43]\]. Several male accessory gland proteins were also downregulated by as much as 6.5-fold during starvation stress. Oddly, six genes affecting spermatogenesis had significantly different levels of transcript abundance between the control and starved flies in both males and females; we found no male-specific differences in transcript abundance for genes involved in spermatogenesis (Additional data files 1 and 2). Transcription of Rpd3and CG14095was upregulated in females and downregulated in males during starvation. Rpd3is a transcriptional co-repressor, while the function of CG14095is unknown. Sex-antagonistic patterns of expression have been observed in liver tissue studies of ethanol-fed rats \[[@B44]\], suggesting that these expression patterns may not be unique to flies. The large number of transcripts altered during starvation implies massive pleiotropy; even more so when our conservative significance threshold is taken into account. This is consistent with our previous observation that 383 of 933 single P-element insertion lines tested (41%) had direct effects on starvation tolerance \[[@B21]\]. Further, candidate genes identified from the P-element screen and from complementation tests of QTL alleles to mutations at positional candidate genes are pleiotropic, and affect cell fate specification, cell proliferation, oogenesis, metabolism, and feeding behaviors \[[@B21]\]. Transcript abundance versus mutations ------------------------------------- To what extent do candidate genes affecting response to starvation stress identified from changes in transcript abundance coincide with those implicated by assessing quantitative effects of P-element insertions on starvation tolerance? The resounding lack of an overall statistical association between the two methods is somewhat deceptive. While there was no association overall, if we had only tested the 160 P-element mutations corresponding to genes with altered transcript abundance during starvation, we would have found that 77 (48%) actually had phenotypic effects on starvation resistance. The lack of association was caused by 108 genes tagged by P-elements that affected starvation resistance, but did not display differences in transcript level in response to starvation stress, and P-element insertions in 83 genes that exhibited significant differences in transcription in response to starvation but did not have significant phenotypic effects on starvation tolerance. Genes affecting starvation that are regulated post-transcriptionally, or for which differences in transcript abundance that are undetectable on the array have large phenotypic consequences, contribute to the first source of discordance between the two methods. The second source of discordance could arise if the genes exhibiting expression changes during starvation are truly candidate genes affecting starvation resistance, but the particular P-element insertional mutation tested was not in a region affecting the starvation phenotype; a P-element insertion or point mutation in another location might produce a significant effect on starvation tolerance \[[@B45]\]. Another possibility is that the gene is downregulated during starvation; thus, a P-element mutation in the gene might not have an effect on the starvation resistance phenotype. Alternatively, a fraction of these probe sets could be false positives. Therefore, we conclude that assessing the effects of mutations at genes exhibiting changes in transcript abundance in response to an environmental (or genetic \[[@B5],[@B7]\]) perturbation is a highly efficient strategy for identifying networks of pleiotropic genes regulating complex traits. Genetic variation in transcript abundance and quantitative trait phenotypes --------------------------------------------------------------------------- The prospects for easily identifying genes corresponding to QTLs using microarray profiling seem less rosy at present. It has been proposed that candidate genes corresponding to QTLs are those for which expression differs between the parental strains used to construct the QTL mapping population, and which are located in the regions to which the QTLs map \[[@B20]\]. However, differences in expression between lines could be due to polymorphisms between the tested strains and the strain used to construct the probe sets on the array. Further, the lines differ for many traits, and QTLs affecting them overlap; unless the QTLs are mapped with very high resolution, candidate genes chosen by this criterion alone could affect another trait. The issue of polymorphism can be circumvented for traits with environmentally conditional expression by considering probe sets exhibiting a line × treatment environment interaction, and trait specificity can be addressed by correlating expression levels with the trait phenotype. None of these criteria led to an enrichment of candidate genes with variation in expression within QTL regions. A major difficulty in using changes in gene expression between two strains to identify candidate genes corresponding to QTLs arises because variation in transcript abundance for positional candidate genes could arise from several causes. First, variation in transcript abundance is attributable to regulatory polymorphism in the candidate gene itself. Second, the candidate gene is itself not genetically variable, but regulatory variation in a second gene affects variation in its expression. Third, variation in transcript abundance at the candidate gene is attributable to interacting regulatory polymorphisms in both the candidate gene and a second gene. These interactions could be additive or epistatic. Positional candidate genes with variation in transcript abundance arising from the first or third cause could potentially correspond to genetically variable QTLs. However, it is becoming clear that genetic variation in transcript abundance is largely attributable to regulation by unlinked genes (see \[[@B15]-[@B17],[@B19]\] and this paper). Indeed, single P-element insertions can alter the transcript expression of as many as 161 genes compared to a co-isogenic control line \[[@B5]\]. This low signal-to-noise ratio means that choosing positional candidate genes for further study based only on differences in transcript abundance between parental lines does not have a high likelihood of success. In the future, the falling cost of whole-genome expression analysis will facilitate assessing transcriptional variation and variation in trait phenotypes in the same large QTL mapping populations. Co-localization of QTLs with main effects jointly affecting variation in transcription and trait phenotype will help winnow out monomorphic genes that are regulated by unlinked loci, and such data would enable direct tests for epistasis at the level of transcription and the trait. It is unlikely that this approach will completely supplant high-resolution QTL mapping and complementation tests to mutations for elucidating the genetic architecture of complex traits in Drosophila. None of the 12 candidate genes affecting variation in starvation resistance between Ore and 2b \[[@B21]\] exhibited variation in transcript abundance in this study. Possibly any transcriptional differences between Ore and 2b alleles at these loci are rare messages below the threshold of detection, or that are expressed in only a few cell types or at a particular period of development. In addition, not all allelic differences between QTL alleles are necessarily regulated at the level of transcription. Nevertheless, incorporation of knowledge about variation in transcript abundance will greatly inform our choice of candidate genes for confirmation by mutant complementation tests and association studies, which is currently biased by our poor understanding of the pleiotropic and epistatic consequences of variation in positional candidate genes on variation in trait phenotypes. Materials and methods ===================== Drosophilastocks ------------------ We used the isogenic lines 2b \[[@B22],[@B46]\] and Oregon-R \[[@B47]\] (Ore) to establish 98 RI lines for mapping QTLs affecting starvation resistance \[[@B21]\]. Survival times for Oregon-R flies were 36.0 and 51.6 h for males and females, respectively. For 2b, survival times were 29.2 h for males and 40.4 h for females. Here, we assessed transcriptional profiles under control conditions and during starvation for 2b, Ore, two starvation resistant (RI.14, RI.21) and two starvation sensitive (RI.35![](gb-2005-6-4-r36-i1.gif), RI.42![](gb-2005-6-4-r36-i1.gif)) RI lines. Recombination breakpoints for the RI lines have been determined previously \[[@B23]\] and are resolved to the nearest cytological lettered subdivision. We maintained control flies on cornmeal-agar-molasses medium, and starved flies on non-nutritive (1.5% agar and water) medium, under standard culture conditions (25°C, 70% humidity, and a 12-h light: 12-h dark cycle). Starvation half-life -------------------- We assessed survival of all six lines under starvation conditions by placing two replicates of ten flies each per sex on starvation medium, and recording the number of flies alive at 8-h intervals until all were dead. We used these survival curves to infer the starvation half-life for each line/sex combination. We used an analysis of variance (ANOVA) model Y= μ + L+ S+ L× S+ R(L× S) + E, to partition variance in survival times into sources attributable to the cross-classified main effects of lines (L), sex (S), variance between replicate vials (R), and within-vial environmental variance (E). Transcriptional profiling ------------------------- For each of two independent replicates, we collected 300 male and 300 female virgins from all lines, aged 2-5 days post-eclosion. The control treatment consisted of 100 non-starved flies/line/sex. We placed the remaining 200 flies/line/sex on starvation medium, and collected approximately 100 flies/line/sex at the predetermined starvation half-life. Starved flies from all lines should therefore be in roughly the same physiological condition. We extracted whole-body RNA from each of the 48 independent samples (6 lines × 2 treatments × 2 sexes × 2 replicates) with Triazol reagent (Gibco BRL), followed by DNase digestion (RQ1 DNase, Promega,) and a 1:1 phenol (Sigma-Aldrich)-chloroform (Fisher Scientific) extraction. We hybridized biotinylated cRNA probes to single-color whole-genome Affymetrix Drosophila GeneChip arrays as described in the Affymetrix GeneChip Expression Analysis 2000 manual. Data analysis ------------- We normalized the expression data by scaling overall probe set intensity to 100 on each chip using standard reference probe sets on each chip for the normalization procedure. Each probe set on the array consists of 14 perfect match (PM) and single nucleotide mismatch (MM) pairs. We used the average difference (AD) in normalized RNA expression between the 14 perfect match (PM) and mismatch (MM) probe pairs per probe set (Affymetrix Microarray Suite, Version 4.0) as the analysis variable. We calculated the minimum AD threshold value \[[@B5]\] as AD = 30. If the mean AD of a probe set was less than 30, and the maximum AD value was also less than 30, we eliminated the probe set from further consideration. We set all remaining AD scores \< 30, to AD = 30. We performed a three-way factorial ANOVA of AD for each probe set, according to the model: Y= μ + S+ T+ L+ S× T+ S× L+ T×L+ S× T× L+ E, where S, T, and Lrepresent, respectively, the fixed cross-classified effects of sex, treatment (control versus starved), and line, and Eis the replicate variance between arrays. We determined F-ratio tests of significance for each term in the ANOVA, and considered probe sets with Pvalues ≤ 0.001 for any term to be significant. (There are approximately 14,000 probe sets on the array; thus 14 false positives would be expected at this significance threshold.) We computed the female:male ratio of AD values, averaged over all lines and treatments, for probe sets for which the main effect of Swas significant. Similarly, we computed the starved:control ratio of AD values, averaged over lines and sex, for probe sets with significant Tterms. We categorized these probe sets according to their gene ontology (GO) for biological process and molecular function \[[@B48]\]. We assessed significant differences in GO categories between up- and downregulated probe sets using Gtests \[[@B49]\], under the null hypothesis of equal numbers of up- and down- regulated probe sets in each category, and using Bonferroni corrections to account for multiple tests. For probe sets with significant T× Sterms, we ran two-way ANOVAs separately by sex using the reduced model Y= μ + L+ T+ L× T+ E. Probe sets with significant L, L× S, L× Tor L× T× Sterms are candidate QTLs for traits that vary among the lines. We performed post-hocTukey tests for all probe sets for which these terms were significant to determine in which lines transcription was up-or downregulated in response to starvation stress. For probe sets that were significant for the main effect of L, but not any of the interaction terms, we conducted Tukey tests using the expression values pooled across control and starved conditions and both sexes. We computed Tukey tests separately for males and females, averaged over both treatments, for probe sets that were significant for the L× Sinteraction; and separately by treatment, for probe sets significant for the L× Tinteraction. The Tukey analyses separated the lines into groups within which AD values were not significantly different. Since the genotype for each recombinant inbred line at any given location is known, we used the Tukey analyses to classify probe sets as exhibiting linked or unlinked regulation of transcript abundance. We considered linked factors to regulate transcript abundance if Ore and 2b differ in transcript abundance, and this difference is reflected in the RI lines according to their Ore and 2b genotype in the region to which the gene maps. Conversely, we inferred that unlinked factors regulate transcript abundance in cases where there is not a 1:1 correspondence between parental line genotype and Tukey grouping. We determined the fold-change between Tukey groupings by calculating the ratio of the deviant line(s) expression level to the mean expression level of the parental or common group. Most Tukey analyses were unambiguous; where multiple interpretations were possible, we calculated the fold-change for all possibilities. Statistical analyses -------------------- We used SAS procedures for all statistical analyses \[[@B50]\]. Additional data files ===================== The following additional data files are available with the online version of this article. Additional data file [1](#S1){ref-type="supplementary-material"} contains a list of all probe sets with significantly different expression in females and males. Additional data file [2](#S2){ref-type="supplementary-material"} contains a list of all probe sets with significantly different expression under control and starved conditions. Additional data file [3](#S3){ref-type="supplementary-material"} lists the probe sets for which the sex by treatment interaction term is significant. Additional data file [4](#S4){ref-type="supplementary-material"} shows the correspondence between the results of a screen for the effects on resistance to starvation stress for single P-element inserts, in a co-isogenic background \[[@B21]\], and changes in transcript abundance between control and starved treatments. Additional data file [5](#S5){ref-type="supplementary-material"} summarizes probe sets for which there is significant genetic variation in transcript abundance. Additional data file [6](#S6){ref-type="supplementary-material"} shows the probe sets for which the only significant genetic term was the main effect of line. Additional data file [7](#S7){ref-type="supplementary-material"} gives the same information as Additional data file [6](#S6){ref-type="supplementary-material"}, but separately for the control and starved treatments, and with the results of the analyses pooled over sexes, and for males and females separately. Additional data file [8](#S8){ref-type="supplementary-material"} is the ANOVA of starvation half-life for Ore, 2b, RI.14, RI.21, RI.35![](gb-2005-6-4-r36-i1.gif) and RI.42![](gb-2005-6-4-r36-i1.gif). Additional data files [9](#S9){ref-type="supplementary-material"} and [10](#S10){ref-type="supplementary-material"} give the raw expression data and presence/absence calls for the control and starved treatments, respectively. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 The file includes all P-values from ANOVA of expression values; the mean expression in males and females, averaged across treatments and lines; the FlyBase ID, gene name, symbol and synonyms; cytological location; and molecular function, biological process and cellular component gene ontologies ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 The file includes the mean expression levels under control and starved conditions, averaged over sexes and lines, as well as the information given for each probe set in Additional data file 1 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 The file includes the P-values for the treatment term in the reduced analyses for males and females separately; the mean expression values under control and starved conditions for males and females, averaged over lines; whether expression is sex-biased (SB), sex-specific (SS) or sex-antagonistic (SA); plus the information given for each probe set in additional data file 1 ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 The results are grouped into four categories: (1) Transcript abundance is altered between control and starved treatments, and there is a significant effect of the P-element insertion on starvation tolerance; (2) Transcript abundance is altered between control and starved treatments, but the P-element insertion does not significantly affect starvation tolerance; (3) There is no alteration in transcript abundance between control and starved treatments, but the P-element insertion significantly affects starvation tolerance; and (4) There is no alteration in transcript abundance between control and starved treatments, and no significant effect of the P-element insertion on starvation tolerance ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 The file includes all P-values from ANOVA of expression values; the gene name, symbol, synonyms, and cytological location; the molecular function, biological process and cellular location gene ontologies; co-localization of the gene with QTLs affecting life span, sensory bristle numbers, starvation resistance, ovariole number, courtship signal, flight, metabolic rate and triglycerides; and the statistical association of variation among lines in transcript levels and starvation half-life ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 The cells are color-coded: purple for Ore genotype, green for 2b genotype, gray if the genotype is unknown because the gene is between an Ore and a 2b flanking marker, and the exact recombination breakpoint is not determined, and gold if the genotype is unknown because of residual heterozygosity in the RI line. The results of Tukey tests separating the lines into groups within which expression values are not significantly different are given. If more than one interpretation of the Tukey groups are possible, all are given. Linked (L) regulation of variation in transcript abundance is inferred if the Tukey groupings match the genotype. Regulation of variation in transcript abundance is inferred to be linked (IL) if the unknown genotypes could match this interpretation. Unlinked (U) regulation of variation in transcript abundance is inferred if the Tukey groupings do not match the line genotypes ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 A table giving the same information as in Additional data file 6, but separately for the control and starved treatments, and with the results of the analyses pooled over sexes, and for males and females separately ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 A table showing the ANOVA of starvation half-life for Ore, 2b, RI.14, RI.21, RI.35![](gb-2005-6-4-r36-i1.gif) and RI.42![](gb-2005-6-4-r36-i1.gif) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 9 A table showing the raw expression data and presence/absence calls for the control treatments ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 10 A table showing the raw expression data and presence/absence calls for the starved treatments ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank K. Norga for annotating the P-element insertion lines. S. T. H. was the recipient of a W. M. Keck pre-doctoral fellowship. This work was funded by grants from the National Institutes of Health to T. F. C. M. This is a publication of the W. M. Keck Center for Behavioral Biology. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Chromosome locations of genes differentially expressed by sex. (a)Observed (magenta) and expected (blue) number of probe sets upregulated in males. (b)Observed (magenta) and expected (blue) numbers of probe sets upregulated in females. ::: ![](gb-2005-6-4-r36-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Genetic architecture of transcription. (a-c)Sex × treatment interaction for females (magenta)and males (blue): (a) Chorion protein 38; (b) Alkaline phosphatase 4; (c) Phosphogluconate dehydrogenase. (d-k)Interactions with line. Ore (black), 2b (red), RI 14 (green), RI 21 (dark blue), RI 35![](gb-2005-6-4-r36-i1.gif)(magenta), RI 42![](gb-2005-6-4-r36-i1.gif) (light blue). (d, e) Sex × line interaction, averaged over treatments: (d) modulo; (e) l(2) giant larvae. (f-i) line × treatment interaction, averaged over sex: (f) CG11089; (g) Nervana 1; (h) Cyp9b2; (i) Peroxiredoxin 2540. (j, k) Sex × line × treatment interaction. The difference in expression between the starved and control treatments is plotted for females (magenta) and males (blue): (j) sallimus; (k) Esterase 6. (l-o)Regulation of transcript abundance. The same letters denote expression levels that are not significantly different. Magenta indicates 2b and blue indicates Ore genome. (l, m) Linked regulation of variation in transcript abundance: (l) UDP-glycosyltransferase 35b; (m) Signal recognition particle receptor b. (n, o) Unlinked regulation of variation in transcript abundance: (n) Arrestin 2; (o) Klarsicht. ::: ![](gb-2005-6-4-r36-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Gene Ontology categories with sex-biased gene expression ::: Gene Ontology category Number of upregulated probe sets P-value\* ------------------------------------------------ ---------------------------------- ------------- --------------- Biological process Cell communication Signal transduction 135 40 \<0.0001 Cell growth and/or maintenance Cell cycle 184 15 \< 0.0001 Cell organization and biogenesis 207 65 \< 0.0001 Transport 123 49 \< 0.0001 Biosynthesis 238 43 \< 0.0001 Catabolism 71 24 \< 0.0001 Nucleic acid metabolism 374 28 \< 0.0001 Phosphorous metabolism 147 60 \<0.0001 Protein metabolism 495 113 \< 0.0001 Development Cell differentiation 33 11 7.41 × 10^-4^ Embryonic development 126 27 \< 0.0001 Morphogenesis 200 50 \< 0.0001 Pattern specification 76 9 \<0.0001 Post-embryonic 50 11 \< 0.0001 Gametogenesis 164 20 \< 0.0001 Other development 84 17 \< 0.0001 Cell death 25 5 1.54 × 10^-4^ Molecular function Binding DNA binding 310 46 \< 0.0001 Nuclease 31 3 \< 0.0001 RNA binding 180 38 \< 0.0001 Translation factor 40 13 1.58 × 10^-4^ Nucleotide binding 187 68 \< 0.0001 Protein binding Cytoskeletal protein binding 89 43 \< 0.0001 Transcription factor binding 28 3 \< 0.0001 Enzymes Hydrolase enzyme Acting on acid anhydrides 177 94 \< 0.0001 Acting on ester bonds 113 56 \< 0.0001 Kinase enzyme 156 62 \< 0.0001 Ligase enzyme 52 18 \< 0.0001 Oxidoreductase enzyme 69 139 \< 0.0001 Transferase enzyme 327 105 \< 0.0001 Other enzymes 88 16 \< 0.0001 Signal transducer Signal transducer - receptor signaling protein 89 14 \< 0.0001 Structural molecule Ribosome structure 137 8 \< 0.0001 Transcription regulator 199 35 \< 0.0001 Translation regulator 42 13 \< 0.0001 Transporter Carrier transporter 82 143 \< 0.0001 Ion transporter 30 70 \< 0.0001 \Significant after Bonferroni correction. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Gene Ontology categories with increased or decreased gene expression during starvation ::: Gene Ontology category Number of probe sets P-value\ -------------------------------- ---------------------- ------------- ---------------- Biological process Cell growth and/or maintenance Biosynthesis 119 31 \< 0.0001 Protein metabolism 220 95 \< 0.0001 Development 12 35 6.48 × 10^-4†^ Behavior 1 9 8.10 × 10^-3‡^ Molecular function Binding Nucleotide binding 76 38 3.36 × 10^-4^ Defense/immunity protein 3 18 6.55 × 10^-4^ Enzymes Hydrolase Acting on acid anhydrides 77 42 1.25 × 10^-3^ Peptidase 50 104 1.12 × 10^-5^ Structure Cuticle structure 1 14 3.09 × 10^-4^ Ribosome structure 84 3 \< 0.0001 Transporter Carrier 46 84 8.05 × 10^-4^ Signal transducer 2 12 5.67 × 10^-3†^ \Significant after Bonferroni correction; ^†^significant for females only; ^‡^significant for males only. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Association of genetic variation in transcription with genetic variation in quantitative traits ::: Trait QTL^†^ Not QTL χ^2^~1~ ------------------------------------- -------- --------- --------- ----- --------- --------------- Life span \[22\] 5 125 25,351 350 92,625 6.58\ Sternopleural bristle number \[25\] 5 250 54,150 225 63,853 8.70\\ Abdominal bristle number \[25\] 7 154 34,038 321 83,965 2.96 NS Starvation resistance \[21\] 5 110 26,532 365 91,471 0.12 NS Life span \[21\] 4 98 24,305 377 93,698 0.00 NS Life span \[23\] 4 133 32,899 342 85,104 0.00 NS Ovariole number \[26\] 2 70 13,162 405 104,841 6.15\* Life span \[24\] 5 82 19,637 393 98,366 0.13 NS Olfactory behavior \[28\] 1 36 7,944 439 110,059 0.54 NS Courtship signal \[27\] 3 67 15,859 408 102,144 0.18 NS Flight \[29\] 2 119 27,860 356 90,143 0.55 NS Metabolic rate \[29\] 2 41 8,232 434 109,771 2.01 NS Glycogen \[29\] 2 5 4,683 470 113,320 10.60 \\^‡^ Triglycerides \[29\] 2 30 6,044 445 111,959 1.39 NS ^†^Two LOD support intervals. In cases of overlap of support intervals between adjacent QTLs, the two QTLs were merged into a single region spanning both. ^‡^P(line) and/or P(Sex × line) \< 0.001. ^§^Significant after Bonferroni correction. \\\*P\< 0.001; \\0.001 \<P\< 0.01; \0.01 \<P\< 0.05; NS P\> 0.05. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Candidate QTLs for starvation resistance ::: Probe set Significant\ Gene Location Molecular function Biological process Cellular location ----------- ------------------------------------- ------------------------------------------------- ---------- -------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------ 151378 S, L, r mitochondrial ribosomal protein L33 4B6 Structural constituent of ribosome Protein biosynthesis Mitochondrial large ribosomal subunit 151504 L no receptor potential A 4C1 1-phosphatidylinositol-4,5-bisphosphate phosphodiesterase; phospholipase C Olfaction; response to abiotic stimulus inD signaling complex; membrane fraction; rhabdomere 153437 S, T, L, r yippee interacting protein 2 30E4 Acetyl-CoA C-acyltransferase Fatty acid beta oxidation Mitochondrion 146142 S, T, L, r Selenophosphate synthetase 2 31D9 Selenide, water dikinase; purine nucleotide binding Selenocysteine biosynthesis 143984 S, T, S× T, L, L× S Accessory gland-specific peptide 32CD 32D1 Hormone Negative regulation of female receptivity, post-mating Extracellular 141745 S, L, L× S, L× T, r Phosphoethanolamine cytidylyltransferase 34A9 Ethanolamine-phosphate cytidylyltransferase ethanolamine and derivative metabolism; phospholipid metabolism 146347 S, L, L× S, L× T, L× S× T centaurin gamma 1A 34D6-E2 ARF GTPase activator G-protein-coupled receptor protein signaling pathway; small GTPase mediated signal transduction Nucleus 153741 L× T centaurin gamma 1A 34D6-E2 ARF GTPase activator G-protein coupled receptor protein signaling pathway; small GTPase mediated signal transduction Nucleus 143402 S, L, L× S, r vasa 35C1 RNA helicase activity; nucleic acid binding; ATP dependent helicase Dorsal appendage formation; oogenesis; pole plasm RNA localization; pole plasm assembly Polar granule 152721 T, L, L× T, r Imaginal disc growth factor 1 36A1 Imaginal disc growth factor activity; NOT chitinase activity; hydrolase activity, hydrolyzing N-glycosyl compounds Cell-cell signaling;signal transduction Extracellular 154661 S, L midway 36B1-2 Sterol O-acetyltransferase; diacylglycerol O-actyltransferase Cholesterol metabolism; triacylglycerol biosynthesis 152756 S, L, r Arrestin 1 36D3 Metarhodopsin binding G-protein coupled receptor protein signaling pathway; deactivation of rhodopsin mediated signaling; endocytosis; intracellular protein transport; metarhodopsin inactivation Membrane fraction; rhabdomere 143876 S, L Galactose-specific C-type lectin 37D6 Galactose binding; sugar binding; receptor Defense response 146555 S, T, S× T, L, L× S Serine protease inhibitor 3 38F2 Serine-type endopeptidase inhibitor Proteolysis and peptidolysis 146592 S, T, S× T, L× T, L× S× T no mechanoreceptor potential B 39E2 NOT flagellum biogenesis; perception of sound; sensory cilium biogenesis 143709 S, T, L, r Troponin C at 41C 41E5 Calcium ion binding; calmodulin binding Calcium-mediated signaling; muscle contraction 143127 S, T, L, L× T Cytochrome P450-6a2 42C8-9 Electron transporter activity; oxidoreductase Response to insecticide; steroid metabolism Membrane; microsome 146718 S, T, L× T Tetraspanin 42Er 42F1 Receptor signaling protein Ectoderm development; neurogenesis; transmission of nerve impulse Integral to membrane 142222 T, L, L× T Cytochrome P450-9b2 42F3 Electron transporter activity; oxidoreductase Membrane; microsome 143830 S, L Calcineurin B2 43E16 Calmodulin binding; calcium-dependent protein serine/threonine phosphatase, regulator; calcium ion binding Calcium-mediated signaling; cell homeostasis Calcineurin complex 141501 S, T, L, r Proteasome alpha6 subunit 43E18 Proteasome endopeptidase Proteolysis and peptidolysis 20S core proteasome complex 143303 S, T, L, r photorepair 43E18 Deoxyribodipyrimidine photolyase; nucleic acid binding DNA repair 146780 S, L× T, L× S× T, r Sep5 43F8 Structural constituent of cytoskeleton; small monomeric GTPase Cytokinesis; mitosis Septin ring 143780 L, L× S Cytochrome P450-4e1 44D1 Electron transporter activity; oxidoreductase Membrane; microsome 152113 S, T, L× S, r anachronism 45A1 Suppression of neuroblast proliferation Extracellular 143554 S, L trp-like 46B2 Calcium channel; calmodulin binding; light-activated voltage-gated calcium channel; store-operated calcium channel Calcium ion transport Plasma membrane; rhabdomere 146946 S, T, L× T, r Peroxiredoxin 2540 47A7 Antioxidant; peroxidase; non-selenium glutathione peroxidase Defense response; oxygen and reactive oxygen species metabolism 143603 T, L gammaTrypsin 47F4 NOT serine-type endopeptidase Proteolysis and peptidolysis Extracellular 143602 T, L betaTrypsin 47F4 Trypsin Proteolysis and peptidolysis Extracellular 143604 T, L gammaTrypsin 47F4 NOT serine-type endopeptidase Proteolysis and peptidolysis Extracellular 143624 T, L× T epsilonTrypsin 47F4 Trypsin Proteolysis and peptidolysis Extracellular 153279 S, T, L, r Translocon-associated protein d 47F7 Signal sequence receptor Protein-ER retention Signal sequence receptor complex; translocon 141563 L acyl-Coenzyme A oxidase at 57D proximal 57E1 Acyl-CoA oxidase; palmitoyl-CoA oxidase Fatty acid beta-oxidation Peroxisome 151902 S, T, L, r jitterbug 59A3 Actin binding; structural constituent of cytoskeleton Cytoskeleton organization and biogenesis 154177 S, L, L× S, r Cyclin B 59B2 Cyclin-dependent protein kinase, regulator Cytokinesis; mitotic anaphase B; mitotic chromosome movement Nuclear cyclin-dependent protein kinase holoenzyme complex; pole plasm 143203 S, T, L, r inactivation no afterpotential D 59B3 Structural molecule; calmodulin binding; myosin binding; receptor signaling complex scaffold Cell surface receptor linked signal transduction; phototransduction; protein targeting inaD signaling complex; rhabdomere 151517 L Phosphotidylinositol 3 kinase 59F 59E4-F1 Phosphatidylinositol 3-kinase; phosphoinositide 3-kinase Endocytosis; phosphoinositide phosphorylation; protein targeting Phosphoinositide 3-kinase complex, class III 151830 S, T, L, L× T, r lethal (2) essential for life 59F6 Heat shock protein Defense response; protein folding; response to stress 144140 T, L, r Mitochondrial phosphate carrier protein 70E1 Phosphate transporter; carrier Phosphate metabolism; phosphate transport Mitochondrial inner membrane 151748 L, L× T, r Cyclic-AMP response element binding protein A 71E1 DNA binding; RNA polymerase II transcription factor; transcription factor Salivary gland morphogenesis; transcription from Pol II promoter Nucleus 153226 S, T, L Argonaute 2 71E1 Translation initiation factor; protein binding RNA interference; translational initiation RNA-induced silencing complex \Significant (P\< 0.001) for the main effects of Sex (S), treatment (T), line (L) and their interactions from ANOVA of transcript abundance; significant (P\< 0.05) correlation (r*) between starvation half-life and transcript abundance. :::	PubMed Central	2024-06-05T03:55:56.381773	2005-3-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088964/", "journal": "Genome Biol. 2005 Mar 24; 6(4):R36", "authors": [ { "first": "Susan T", "last": "Harbison" }, { "first": "Sherman", "last": "Chang" }, { "first": "Kim P", "last": "Kamdar" }, { "first": "Trudy FC", "last": "Mackay" } ] }
PMC1088965	Background ========== Selenium is a biological trace element with significant health benefits \[[@B1]\]. This micronutrient is incorporated into several proteins in bacteria, archaea and eukaryotes as selenocysteine (Sec), the 21st amino acid in proteins \[[@B2],[@B3]\]. Sec is encoded by a UGA codon in a process that requires translational recoding, as UGA is normally read as a stop codon \[[@B4]\]. The Sec UGA codon was the first addition to the universal genetic code since the code was deciphered in the mid-1960s \[[@B5]\]. Recently, an additional amino acid, pyrrolysine (Pyl), has been identified, which has expanded the genetic code to 22 amino acids \[[@B6],[@B7]\]. Pyl is inserted in response to a UAG codon in several methanogenic archaea, but the specific mechanism of insertion of this amino acid into protein is not yet known. The mechanism of selenoprotein synthesis in prokaryotes was elucidated extensively by Böck and colleagues \[[@B8],[@B9]\]. Translation of selenoprotein mRNA requires both a selenocysteine insertion sequence (SECIS) element, which is a cis-acting stem-loop structure residing within selenoprotein mRNAs \[[@B4],[@B10]\], and trans-acting factors dedicated to Sec incorporation \[[@B11]\]. In eukaryotes and archaea, SECIS elements are located in 3\'-untranslated regions (3\' UTRs) \[[@B12]\]. Bacterial SECIS elements differ from those in eukaryotes and archaea in terms of sequence and structure and are located immediately downstream of Sec UGA codons in the coding regions of selenoprotein genes \[[@B13],[@B14]\]. As UGA has the dual function of inserting Sec and terminating translation, and only the latter function is recognized by available annotation programs, selenoprotein genes are almost universally misannotated in sequence databases \[[@B15]\]. To address this problem, various computational approaches to predict selenoprotein genes have been developed \[[@B16]-[@B21]\]. These programs successfully identified new selenoproteins in mammalian and Drosophilagenomes and in several EST databases. However, due to lack of bacterial consensus SECIS models, prediction of bacterial selenoproteins in genomic sequences is difficult. Instead, these proteins can be identified through searches for Sec/Cys pairs in homologous sequences \[[@B22]\]. We report here the use of a modified search strategy to characterize the selenoproteome of the largest prokaryotic sequencing project, the 1.045 billion nucleotide whole genome shotgun sequence of the Sargasso Sea microbial populations \[[@B23]\]. This database contains sequences from over 1,800 microbial species, including 148 novel bacterial phylotypes. We detected all known prokaryotic selenoproteins present in this dataset and identified a large number of additional selenoprotein genes. This approach provided a relatively unbiased way to examine the diversity of selenoprotein families and their evolution, and to analyze the composition of the Sargasso Sea microbial selenoproteome as compared with that in the combined set of completely sequenced prokaryotic genomes. Results ======= Identification of selenoprotein genes in the Sargasso Sea environmental genome database --------------------------------------------------------------------------------------- The Sargasso Sea genomic database contains the largest collection of microbial sequences derived from a single study \[[@B23]\]. No genes encoding Sec-containing proteins were previously identified and annotated in this dataset. To identify selenoprotein genes in the Sargasso Sea microbial sequences, we used an algorithm that searches for conserved Sec/Cys pairs in homologous sequences. This approach takes advantage of the fact that almost all selenoproteins have homologs (often in different organisms) in which Cys occupies the position of Sec. The methodology is described in Materials and methods and is shown schematically in Figure [1](#F1){ref-type="fig"}. Briefly, we searched for nucleotide sequences from the Sargasso Sea database which, when translated, aligned with protein sequences from the nonredundant (NR) database such that translated TGA codons aligned with Cys and these pairs were flanked on both sides by conserved sequences. Each TGA-containing sequence in the Sargasso Sea database that was identified in this manner was further screened against a set of filters, which analyzed for possible open reading frames (ORFs), conservation of TGA codons, conservation of Cys in homologs, conservation of TGA-flanking regions in different reading frames and for redundancy. Nonredundant hits were clustered into protein families and a second BLAST search was performed against microbial genomes and NR databases. Finally, all groups of hits were analyzed manually and divided into homologs of previously known selenoproteins, new selenoproteins and selenoprotein candidates. This procedure identified 209 selenoprotein genes, which belonged to ten known selenoprotein families and 101 selenoprotein genes, which belonged to 15 new selenoprotein families (each represented by at least two sequences) (Table [1](#T1){ref-type="table"}). In addition, we detected 28 sequences, which showed homology neither to known and new selenoproteins nor to each other, and these were designated as candidate selenoproteins. Considering that several known selenoproteins were also represented by single sequences (for example, glycine reductase selenoprotein A and glycine reductase selenoprotein B), some of these 28 candidate selenoproteins may be true selenoproteins. However, at present, sequencing errors that generate in-frame TGA codons cannot be excluded and therefore, no definitive conclusions can be made regarding these sequences. Predicted selenoproteins, particularly those represented by a small number of sequences, require future experimental verification. In total, 310 known and new selenoprotein genes and 28 candidate selenoprotein genes were detected. All these genes were misannotated in the Sargasso Sea dataset, because the previously used annotation tools recognized Sec-encoding TGA codons as terminators. Consequently, some selenoprotein ORFs were annotated as truncated proteins lacking either carboxy-terminal or amino-terminal regions containing Sec, whereas other selenoprotein ORFs were missed altogether. Previously known selenoprotein families detected in the Sargasso Sea database ----------------------------------------------------------------------------- Our procedure detected all known prokaryotic selenoprotein genes present in the Sargasso Sea database, which could also be independently identified by homology searches using known selenoprotein sequences as queries. Eight of the ten known selenoprotein families detected in the dataset were represented by 5-48 selenoprotein genes, whereas two families, glycine reductase selenoprotein A (grdA) and glycine reductase selenoprotein B (grdB), were represented by single sequences. Interestingly, although all known selenoproteins present in the dataset were identified, only nine of the ten families had Cys homologs in the NR database. One selenoprotein, grdA, did not have known Cys homologs \[[@B22]\]. Nevertheless, grdA was also identified because of annotation errors, as Sec in this protein was annotated as Cys in some NR database entries. Several selenoprotein families had a particularly high representation in the Sargasso Sea dataset. The most abundant family was SelW-like, which contained 48 genes. Although the function of this protein is unclear, a conserved CXXU motif (Cys separated from Sec by two other residues) suggests a redox function. In addition, this protein was previously found to interact with glutathione, a major redox thiol compound in cells \[[@B24],[@B25]\]. A peroxiredoxin (Prx) family had 43 genes and was the second most abundant selenoprotein family. Peroxiredoxins protect bacterial and eukaryotic cells against oxidative injury \[[@B26]\]. Proline reductase (prdB, 42 genes) and selenophosphate synthetase (28 genes) were the third and fourth most abundant families. The former is involved in amino-acid metabolism and catalyzes the reductive ring cleavage of D-proline to 5-aminovalerate \[[@B27]\]. The latter is a key component in prokaryotic selenoprotein biosynthesis \[[@B2],[@B28]\]. A Prx-like protein family was represented by 22 selenoprotein sequences. It had distant homology to the Prx family, but its predicted active site contained a thioredoxin-like UXXC motif instead of the TXXU motif present in Sec-containing Prx. These five families accounted for 87.6% of known selenoprotein sequences, suggesting importance of their functions in the Sargasso Sea environment. Other detected selenoprotein families included thioredoxin (Trx), formate dehydrogenase alpha chain (fdhA), glutathione peroxidase (GPx), grdA and grdB. New selenoprotein families identified in the Sargasso Sea database ------------------------------------------------------------------ Among 15 new selenoprotein families, 13 contained at least two individual TGA-containing ORFs (Table [1](#T1){ref-type="table"}). Although two selenoprotein families, DsbG-like and NADH:ubiquinone oxidoreductase, were represented by single entries, we placed them in the new selenoprotein category because they had been previously reported as candidate selenoproteins \[[@B22]\]. Of the 15 families, 14 either contained a domain of known function or were homologous to protein families with known functions, including several which were represented by multiple sequences: AhpD-like protein (27 sequences), arsenate reductase (14 sequences), molybdopterin biosynthesis MoeB protein (11 sequences), glutaredoxin (Grx) (ten sequences) and DsbA-like protein (nine sequences). Thus, these findings implicated selenium in arsenate reduction, molybdopterin biosynthesis, disulfide bond formation and other redox-based processes. No functional evidence could be obtained for one family, which was designated as hypothetical protein 1 (represented by four sequences). However, a conserved CXXU motif was present in hypothetical protein 1, suggesting a possible redox function. Multiple alignments of several new selenoproteins and their Cys-containing homologs (Figure [2](#F2){ref-type="fig"}) highlight sequence conservation of Sec/Cys pairs and their flanking regions. All new selenoproteins contained stable stem-loop structures downstream of Sec-encoding TGA codons that resembled bacterial SECIS elements. Representative predicted SECIS elements found in several new selenoprotein families are shown in Figure [3](#F3){ref-type="fig"}. A structural alignment of putative SECIS elements in known and new selenoprotein genes in the Sargasso Sea database (Figure [4](#F4){ref-type="fig"}) showed that they shared the common features of bacterial SECIS elements (for example, a small apical loop containing a guanosine, see Materials and methods). Significant overlap between eukaryotic and prokaryotic selenoproteomes ---------------------------------------------------------------------- Among 25 known and new bacterial selenoprotein families identified in the Sargasso Sea dataset, three families, SelW-like, GPx and deiodinase, were previously thought to be of eukaryotic origin. However, multiple sequence alignments (Figure [5](#F5){ref-type="fig"}) and phylogenetic analyses (Figure [6](#F6){ref-type="fig"}) strongly suggested a bacterial origin of these selenoproteins. Although several eukaryotic sequences in the Sargasso Sea dataset were also detected (for example, GPx homolog, accession number AACY01485942), all SelW and deiodinase sequences and most GPx sequences were bacterial selenoproteins. We based this conclusion on the presence of bacterial and absence of eukaryotic and archaeal SECIS elements in these sequences. In addition, phylogenetic analyses of coding sequences that flanked selenoprotein genes indicated that these contigs were derived from bacteria (data not shown). As information about the species present in the environmental samples is not available, analysis of SECIS elements provides a means of distinguishing selenoprotein sequences in the major domains of life, as SECIS elements are different in eukaryotes, bacteria and archaea in regard to sequence and structure \[[@B29]\]. Representative bacterial SECIS elements of the three bacterial selenoproteins and their eukaryotic counterparts are shown in Figure [7](#F7){ref-type="fig"}. Deiodinase is known to activate or inactivate thyroid hormones via the reaction of reductive deiodination \[[@B30]\]. This protein has previously been described only in animals and only in the selenoprotein form. However, we identified both Cys-containing and Sec-containing homologs of deiodinase in the Sargasso Sea dataset (Figure [5](#F5){ref-type="fig"}). Bacterial deiodinase-like proteins likely serve a different function than animal deiodinases as thyroid hormones are not expected to occur in these organisms. Deiodinases possess a variation of the thioredoxin fold \[[@B31]\], which is known for redox functions. It is possible that bacterial deiodinase-like proteins also serve a redox function. SelW and GPx homologs were recently detected in some bacteria, but the number of these sequences was small and their origin not clear \[[@B22]\]. Detection of a large number of SelW and GPx selenoprotein sequences in the Sargasso Sea allowed us to perform phylogenetic analyses (Figure [6](#F6){ref-type="fig"}), which suggested that at least some members of these families evolved independently in bacteria and eukaryotes. In addition, we identified five eukaryotic selenoproteins: SelM, SelT, SelU, GPx and methionine-S-sulfoxide reductase (MsrA). Except for GPx, these families were represented by single selenoprotein genes. No bacterial SECIS elements were found in these genes. In SelM and SelT sequences, typical eukaryotic SECIS elements were present in 3\' UTRs as detected by SECISearch \[[@B16]\], whereas GPx, MsrA and SelU sequences did not extend enough to test for presence of SECIS elements in 3\' UTRs. However, the MsrA and GPx sequences were most similar to plant proteins, suggesting that the two proteins also were of eukaryotic origin. In addition, eukaryotic GPx sequences could be distinguished by the presence of introns. Previous analyses of selenoprotein sets in the three domains of life revealed that bacterial and archaeal selenoproteomes significantly overlap, whereas eukaryotes had a different set of selenoproteins \[[@B15],[@B20]\]. The only exception was selenophosphate synthetase, but as it is involved in Sec biosynthesis, this protein must be maintained in organisms that utilize Sec. However, our finding of additional selenoproteins in Sargasso Sea organisms revealed a significant overlap between prokaryotic and eukaryotic selenoproteomes. Differences in selenoprotein sets in the Sargasso Sea database and completely sequenced prokaryotic genomes ----------------------------------------------------------------------------------------------------------- An exhaustive search of Sargasso Sea selenoproteins against 260 completely sequenced prokaryotic genomes revealed that these selenoproteins were present in a limited number of genomes, which contrasted with the widespread occurrence of their Cys-containing homologs (Table [2](#T2){ref-type="table"}). Although the size of the Sargasso Sea dataset and the combined set of 260 prokaryotic genomes were similar, the two datasets differed in regard to both number and distribution of selenoprotein genes present in these databases. The Sargasso Sea dataset was three times richer in selenoproteins than the prokaryotic genomes, suggesting that the environment of the Sargasso Sea generally favors evolution and maintenance of selenoproteins. Presumably, the Sargasso Sea organisms take advantage of a relatively constant supply of selenium in sea water and have increased their demand for this trace element, whereas the dependence of the organisms with completely sequenced genomes on selenium is mixed as selenium may be a limiting factor in some environments. Six previously known selenoproteins were not detected in the Sargasso Sea database (Table [2](#T2){ref-type="table"}). This is likely because these selenoproteins primarily occur in archaea. Archaea accounted only for a small fraction of the Sargasso Sea organisms \[[@B23]\]. In addition, the abundance of particular selenoprotein genes in the Sargasso Sea dataset and in the 260 microbial genomes was quite different. Particularly surprising was the small number of formate dehydrogenase genes in the Sargasso Sea database \[[@B32]\]. Previous analyses of completely sequenced prokaryotic genomes found that this protein was present in essentially all organisms that utilized Sec, and its occurrence was by far more common than any other selenoprotein \[[@B22]\]. However, in the Sargasso Sea environment, the utilization of this protein was limited. This might be related to the aerobic nature of microbial species that reside near the surface of the Sargasso Sea (where the environmental samples were collected for sequencing). We also observed that in the previously analyzed prokaryotic genomes, more than half of selenoproteins were metal-binding proteins, in which Sec coordinated molybdenum, tungsten or nickel \[[@B22]\]. In contrast, the Sargasso Sea selenoproteins were primarily thiol-dependent peroxidases and oxidoreductases; metal-coordinating selenoproteins were represented exclusively by formate dehydrogenase and accounted for less than 4% of all detected selenoproteins. These data suggested that the previously characterized genomes did not represent the general composition of prokaryotic selenoproteomes. Although the two sets of selenoproteins (Sargasso Sea and the completely sequenced prokaryotic genomes) were different, the majority of detected selenoproteins showed scattered occurrence. Indeed, the Sec-containing forms of proteins were rare compared to homologous Cys-containing forms, which were widespread. It appears that that most detected selenoproteins evolved recently from Cys-containing homologs in organisms, which already had the system for Sec insertion. It can be predicted that as searches of additional prokaryotic sequence datasets identify new selenoprotein genes, many of these will be present in only a small number of species. At present, Sec evolution is not fully understood, but it is clear that Sec/Cys interchanges are possible in both directions depending on the need for particular redox properties and on the restriction imposed by the dependence of species on the trace element selenium. Most selenoprotein families serve redox functions ------------------------------------------------- Further analysis of both Sargasso Sea and completely sequenced prokaryotic genomes revealed that essentially all selenoproteins with known function were redox proteins, which used Sec either to coordinate redox-active metals or for thiol/disulfide-like redox catalysis. Among 25 selenoprotein families detected in the Sargasso Sea, 14 (194 selenoprotein sequences, 62.6%) were homologs of known thiol-dependent redox proteins (Table [3](#T3){ref-type="table"}), and most other proteins were candidate redox proteins. Many of the Sargasso Sea selenoproteins contained a UXXC redox motif. The analogous CXXC motif is present in a variety of thiol-dependent redox enzymes \[[@B33]-[@B35]\], but it is also common in metal-binding proteins. The catalytic activity of UXXC-containing selenoenzymes is expected to be higher than that of its Cys-containing homologs \[[@B2],[@B36]\]. In addition, several selenoproteins had other candidate redox motifs \[[@B34]\], such as UXXS (arsenate reductase), TXXU (peroxiredoxin and NADH:ubiquinone oxidoreductase), UXXT (glutathione peroxidase) and CXXU (AhpD-like protein \[[@B37]\], SelW-like protein, CMD domain-containing protein and hypothetical protein 1). Discussion ========== Whole-genome shotgun sequencing projects have been applied extensively to determine genomic sequences of a variety of organisms, and recently this approach was used to sequence the microbial community of the Sargasso Sea. Many of the Sargasso Sea organisms represent phyletic groups previously not known or poorly characterized, including organisms that could not be isolated from the microbial community or be cultured \[[@B23]\]. Identification of selenoprotein genes in such a large prokaryotic dataset may help understand the role of selenium in this microbial community and by analogy in other organisms, including humans. Previous functional information on selenoproteins has been derived largely from wet-lab experiments. More recently, several in silicoapproaches that identify full sets of selenoproteins in organisms provided powerful new tools for determining identities of selenoproteins as well as their expression characteristics and functions \[[@B16]-[@B20],[@B38]\]. Most of these methods were based on searches for SECIS elements. As Sec is typically located at enzyme active centers, and most selenoproteins had homologs in which Cys replaced Sec, a SECIS-independent strategy was also developed that allowed searches for Sec/Cys pairs in homologous sequences \[[@B21],[@B22]\]. In the present study, we used a similar procedure, but supplemented it with additional filters to improve performance. All known prokaryotic selenoprotein families present in the Sargasso Sea genomic dataset were identified by this approach (209 genes that clustered into ten prokaryotic selenoprotein families). In addition, 101 sequences that belonged to 15 new selenoprotein families were identified. Thus, our study has approximately doubled the list of known prokaryotic selenoprotein families and generated the largest selenoprotein dataset to date. On the basis of the presence of SECIS elements specific to major domains of life, we could determine the origin of detected selenoproteins (that is, bacterial, archaeal or eukaryotic). All ten known and 15 new prokaryotic selenoprotein families had predicted bacterial SECIS elements. Interestingly, both selenoprotein forms and Cys-containing homologs of thyroid hormone deiodinase, a protein previously thought to be restricted to the animal kingdom and present exclusively in the selenoprotein form, were identified in prokaryotes. The detected deiodinase-like proteins were prokaryotic as they contained bacterial SECIS elements. Detection of prokaryotic deiodinase-like proteins and several other bacterial selenoproteins thought to be restricted to eukaryotes suggests a revision of the view that eukaryotic and prokaryotic selenoproteomes do not overlap. Although this idea was consistent with the previous selenoprotein analyses, at least four selenoprotein families are now known that occur in both prokaryotes and eukaryotes: SelW, GPx, selenophosphate synthetase and deiodinase. We also detected homologs of five additional eukaryotic selenoproteins, but the absence of bacterial SECIS elements, presence of eukaryotic SECIS elements or introns, and homology to eukaryotic proteins argued that these selenoproteins were eukaryotic in origin. Surprisingly, sets of selenoproteins in the Sargasso Sea database and in the combined set of 260 completely sequenced prokaryotic genomes were quite different in regard to both identities and number of selenoprotein genes. The Sargasso Sea dataset was rich in selenoprotein genes, most of which were homologs of known thiol-dependent redox enzymes. In contrast, the proportion of selenoprotein genes in completely sequenced prokaryotic genomes was approximately three times lower, and the majority of detected genes used Sec for metal coordination. Thus, even with the availability of 260 genomes, the roles of selenium in nature are just beginning to be understood. For example, our current analysis of the Sargasso Sea dataset implicated selenium in arsenate reduction, molybdopterin biosynthesis, sulfurtransferase function and other processes, which were not known to be dependent on this trace element. We also observed common features in the two sets of selenoproteins. For example, most of the detected selenoproteins had a large number of Cys homologs. The scattered occurrence of selenoproteins in both datasets suggests a highly dynamic nature of Sec evolution. As long as the system for Sec insertion is maintained, Sec may appear when required by the changing environment and disappear when this requirement recedes. Thus, the analysis of selenoproteomes and the compensatory sets of Cys-containing proteins provides a unique model system to examine evolutionary forces to a changing environment. Materials and methods ===================== Sequence databases and resources -------------------------------- The whole-genome shotgun sequence database of the Sargasso Sea was obtained from the National Center for Biotechnology Information (NCBI) ftp server with the project accession number AACY00000000 \[[@B39]\]. Unlike conventional sequence entries, only the unassembled singletons and individual singletons were deposited in order to accurately reflect the diversity in the sample and to allow searches across the entire sample within a single database. The Sargasso Sea database contained 811,372 genomic sequences, which corresponded to a total of 1.045 billion nucleotides. The NR protein database was downloaded from the NCBI ftp server \[[@B40]\]. This dataset contained 1,990,024 protein sequences (667,623,348 amino acids). Blast programs \[[@B41]\] were also obtained from the NCBI ftp server \[[@B42]\]. We used the 2.2.9 version of this program. To enable selenoprotein searches automatically, we developed a set of programs as discussed below. A UNIX/LINUX platform was used. The entire search process was performed on a Prairiefire 128-node, 256-processor Beowulf cluster supercomputer at the Research Computing Facility of the University of Nebraska - Lincoln. Identification of Cys/TGA pairs in homologous sequences ------------------------------------------------------- Each Cys-containing sequence in the NR protein database was searched against the Sargasso Sea database of nucleotide sequences for possible TGA-containing hits using TBLASTN. E-value cutoff was set to 10.0. TBLASTN output for each protein sequence was parsed and Cys/TAA or Cys/TAG pairs were filtered out. Only local alignments, in which Cys in a query sequence was aligned with TGA in the nucleotide sequence from the target Sargasso Sea database, were further analyzed. As Sec is typically located in enzyme active sites, additional filters were added. Specifically, local alignments were discarded if they contained more than two stop codons (including TGA, TAA and TAG), two stop codons of which one was not TGA, or two TGA codons with one aligned to a non-Cys residue. A total of 38,446 local redundant alignments (also designated as Cys/TGA pairs) were identified which corresponded to 19,410 proteins in the NR protein database. Analyses of ORFs, conservation of TGA-flanking regions and redundancy --------------------------------------------------------------------- For each TGA-containing sequence in the local alignment set, regions upstream and downstream of the TGA were analyzed to identify minimal ORFs with the assumption that in-frame TGA coding for Sec must be inside predicted ORFs. If stop codons were encountered closer to TGA codons than candidate start codons (ATG or GTG), such TGA-containing sequences were discarded. Conservation of TGA-flanking regions in all 6 reading frames was also analyzed with BLASTX and screened against a database of conserved domains using RPS-BLAST. These criteria were also used to filter out false positive hits. Finally, redundant sequences were removed. These filters reduced the set to 2,131 unique TGA-containing candidate ORFs. Clustering of TGA-containing sequences -------------------------------------- To cluster protein sequences into different protein families or groups, the pairwise alignment tool in the BLAST program package, BL2SEQ, was used. 1,045 clusters were obtained with 1 to 63 sequences in each cluster. Cysteine conservation and selenoprotein classification ------------------------------------------------------ Considering that Cys/TGA pairs in most false-positive hits were not expected to be conserved, whereas conservation was expected for true-positive Cys/Sec pairs, all clusters were automatically searched against NCBI NR and microbial databases using BLASTX and TBLASTX. Each predicted ORF containing an in-frame TGA was considered further only if at least two corresponding Cys-containing homologs were detected and the proportion of TGA/Cys pairs in the set of homologs was greater than 50%. This procedure resulted in 331 clusters containing 1,072 ORFs. All 331 clusters were analyzed for the presence of potential bacterial SECIS elements immediately downstream of the TGA codons using mfold \[[@B43]\] or RNAfold \[[@B44]\] programs. In addition, candidate SECIS structures were screened against a bacterial SECIS consensus model \[[@B45]\]. The presence of archaeal or eukaryotic SECIS elements was tested using SECISearch \[[@B20],[@B22]\]. The occurrence of SECIS elements specific for each domain of life was one criterion to determine protein origin. Phylogenetic analyses and the occurrence of introns were also used as criteria for designating proteins as bacterial, archaeal or eukaryotic. A simple classifier was developed to divide clusters that contained bacterial SECIS-like structures into three groups: known selenoproteins, new selenoproteins and selenoprotein candidates. Except for known selenoproteins, clusters containing at least two different sequences with conserved in-frame TGA codons were considered as new selenoproteins. Clusters containing only one sequence were considered selenoprotein candidates because of the possibility of a sequencing error causing an in-frame TGA. Finally, clusters that could be aligned such that their TGAs also aligned, were joined into larger clusters. Additional data files ===================== The complete set of predicted selenoprotein sequences with annotations (accession number, protein name, ORF location and in-frame TGA location) is available as a text file (Additional data file [1](#S1){ref-type="supplementary-material"}) with the online version of this paper and at \[[@B46]\]. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A complete set of predicted selenoprotein sequences with annotations (accession number, protein name, ORF location and in-frame TGA location) ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was supported by NIH grant GM61603. We thank the Research Computing Facility of the University of Nebraska-Lincoln for the use of Prairiefire supercomputer. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A schematic diagram of the search algorithm. Details of the search process are provided in Materials and methods and are discussed in the text. ::: ![](gb-2005-6-4-r37-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Multiple sequence alignments of new selenoproteins and their Cys homologs. The alignments show Sec-flanking regions in detected proteins. Both selenoprotein sequences detected in the Sargasso Sea database and their Cys-containing homologs present in indicated organisms are shown. Conserved residues are highlighted. Predicted Sec (U) and the corresponding Cys (C) residues in other homologs are shown in red and blue background, respectively. Sequence alignments were generated with ClustalW and shaded by BoxShade v3.21. ::: ![](gb-2005-6-4-r37-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Predicted bacterial SECIS elements in representative sequences of some new selenoprotein families. Only sequences downstream of in-frame UGA codons are shown. In-frame UGA codons and conserved guanosines in the apical loop are shown in red. AhpD-like protein, AACY01418594; Arsenate reductase, AACY01238341; Glutaredoxin, AACY01002222; DsbA-like protein, AACY01178397; Hypothetical protein 1, AACY01574522; Rhodanase-related sulfurtransferase, AACY01016424; OsmC-like protein, AACY01145085; DsrE-like protein, AACY01486889. ::: ![](gb-2005-6-4-r37-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Alignment of SECIS elements present in Sargasso Sea selenoproteins. The Sargasso Sea dataset includes 10 known selenoprotein families and 15 new families. SECIS elements in representative members of these families were manually aligned on the basis of primary sequence and secondary structure features. ::: ![](gb-2005-6-4-r37-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Multiple alignments of deiodinase, GPx and SelW. Conserved residues are highlighted. Predicted Sec (U) in selenoproteins and the corresponding Cys (C) residues in homologs are shown in red and blue background, respectively. Sequence alignments were generated with ClustalW and shaded by BoxShade v3.21. ::: ![](gb-2005-6-4-r37-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Phylogenetic analyses of deiodinase, GPx and SelW protein families. Selenoproteins are shown in red and Cys-containing homologs in green. The phylogenetic trees were generated by ClustalW and represented by Treeme. ::: ![](gb-2005-6-4-r37-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Comparison of bacterial and eukaryotic SECIS elements in deiodinase, GPx and SelW protein sequences. (a)In bacterial SECIS elements, only sequences downstream of in-frame UGA codons are shown and the conserved features (in-frame UGA codon and conserved G in the apical loop) are highlighted in red. (b)In eukaryotic SECIS elements, conserved features (quartet, AA in the apical loop or bulge and an A preceding the quartet) are shown in blue. Bacterial elements are from deiodinase-like protein, AACY01143874; GPx, AACY01190440; and SelW-like protein, AACY01475618. Eukaryotic elements are from deiodinase, NM\_000792, Homo sapiens; GPx, X68314, H. sapiens; SelW, AY221261, Danio rerio. ::: ![](gb-2005-6-4-r37-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Selenoprotein families identified in the Sargasso Sea database ::: Prokaryotic selenoprotein family Unique sequences COG/Pfam ID COG/Pfam description ------------------------------------------ ------------------ ------------- ------------------------------------------------------------ Known selenoproteins (209 sequences) SelW-like protein 48 Pfam05169 Selenoprotein W-related Peroxiredoxin (Prx) 43 COG1225 Peroxiredoxin Proline reductase (PrdB) 42 \- Selenophosphate synthetase 28 COG0709 Selenophosphate synthetase Prx-like protein 22 COG0450 Peroxiredoxin-like Thioredoxin (Trx) 11 COG3118 Thioredoxin Formate dehydrogenase alpha chain (fdhA) 8 COG0243 Anaerobic dehydrogenases Glutathione peroxidase (GPx) 5 COG0386 Glutathione peroxidase Glycine reductase selenoprotein A (grdA) 1 \- Glycine reductase selenoprotein B (grdB) 1 Pfam07355 Glycine reductase selenoprotein B New selenoproteins (101 sequences) AhpD-like protein 27 COG2128 Uncharacterized conserved protein Arsenate reductase 14 COG1393 Arsenate reductase and related proteins Molybdopterin biosynthesis MoeB protein 11 COG0476 Dinucleotide-utilizing enzymes, molybdopterin biosynthesis Glutaredoxin (Grx) 10 COG0695 Glutaredoxin and related proteins DsbA-like protein 9 COG2761 DsbA-like Glutathione S-transferase 4 COG0625 Glutathione S-transferase Deiodinase-like protein 4 Pfam00837 Iodothyronine deiodinase Thiol-disulfide isomerase-like protein 4 \- CMD domain-containing protein 4 Pfam02627 Carboxymuconolactone decarboxylase Hypothetical protein 1 4 \- Rhodanese-related sulfurtransferase 3 COG2897 Rhodanese-related sulfurtransferase OsmC-like protein 3 COG1765 Predicted redox protein, OsmC-like DsrE-like protein 2 Pfam02635 DsrE-like DsbG-like protein 1 COG1651 DsbG, Protein-disulfide isomerase NADH:ubiquinone oxidoreductase 1 COG2209 NADH:ubiquinone oxidoreductase Total 310 Classification of selenoproteins (10 previously known and 15 new prokaryotic selenoprotein families) is supported by COG or Pfam sequence clusters (or both) as shown in this table. The number of individual selenoprotein sequences for each family is indicated. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of selenoproteins identified in the Sargasso Sea database and in the combined set of completely sequenced prokaryotic genomes ::: Prokaryotic selenoprotein family Sequences in the Sargasso Sea database Sequences in completely sequenced prokaryotic genomes --------------------------------------------------------------- ---------------------------------------- ------------------------------------------------------- ----- ------------ Known selenoproteins detected in the Sargasso Sea dataset SelW-like protein 48 7 4 20 Peroxiredoxin 43 Widespread\* 1 Widespread Proline reductase PrdB 42 1 1 5 Selenophosphate synthetase 28 23 16 22 Prx-like protein 22 4 2 6 Thioredoxin 11 Widespread 2 Widespread Formate dehydrogenase alpha chain 8 Widespread 40 Widespread Glutathione peroxidase 5 Widespread 1 Widespread Glycine reductase selenoprotein A 1 0 5 0 Glycine reductase selenoprotein B 1 1 5 2 New selenoproteins detected in the Sargasso Sea dataset AhpD-like protein 27 Widespread 0 Widespread Arsenate reductase 14 Widespread 0 Widespread Molybdopterin biosynthesis MoeB protein 11 Widespread 0 Widespread Glutaredoxin 10 17 1 Widespread DsbA-like protein 9 Widespread 0 Widespread Glutathione S-transferase 4 Widespread 0 Widespread Deiodinase-like protein 4 6 0 0 Thiol-disulfide isomerase-like protein 4 Widespread 0 Widespread CMD domain-containing protein 4 14 0 5 Hypothetical protein 1 4 7 0 5 Rhodanase-related sulfurtransferase 3 Widespread 0 Widespread OsmC-like protein 3 10 0 17 DsrE-like protein 2 3 1 9 DsbG-like protein 1 Widespread 1 Widespread NADH:ubiquinone oxidoreductase 1 Widespread 0 Widespread Known selenoproteins not detected in the Sargasso Sea dataset Formylmethanofuran dehydrogenase 0 Widespread 4 Widespread F420-reducing hydrogenase alpha subunit 0 4 4 Widespread F420-reducing hydrogenase, delta subunit 0 1 3 Widespread Methylviologen-reducing hydrogenase 0 3 3 Widespread Heterodisulfide reductase 0 0 4 23 HesB-like protein 0 2 7 2 Total 310 105 \*Widespread, occurrence in more than 40 sequences in the Sargasso Sea database or more than 40 in organisms in the combined set of completely sequenced prokaryotic genomes. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Thiol-dependent redox selenoproteins detected in the Sargasso Sea database and their predicted redox motifs containing Sec (U) ::: Selenoprotein family Redox motif Individual sequences ---------------------------------------- ------------- ---------------------- Known redox proteins Peroxiredoxin TXXU 43 Proline reductase PrdB UXXC 42 Prx-like protein UXXC 21 UXC 1 Thioredoxin UXXC 11 Glutathione peroxidase UXXT 5 AhpD-like protein CXXU 27 Arsenate reductase UXXS 14 Glutaredoxin UXXC 10 DsbA-like protein UXXC 9 Thiol-disulfide isomerase-like protein UXXC 4 OsmC-like protein UXXT 3 DsrE-like protein UXXC 2 DsbG-like protein UXXC 1 NADH:ubiquinone oxidoreductase TXXU 1 Candidate redox proteins SelW-like protein CXXU 48 CMD domain-containing protein CXXU 4 Hypothetical protein 1 CXXU 4 :::	PubMed Central	2024-06-05T03:55:56.389694	2005-3-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088965/", "journal": "Genome Biol. 2005 Mar 29; 6(4):R37", "authors": [ { "first": "Yan", "last": "Zhang" }, { "first": "Dmitri E", "last": "Fomenko" }, { "first": "Vadim N", "last": "Gladyshev" } ] }
PMC1088966	Background ========== Phenotypes are determined by complex interactions among gene variants and environmental factors. In biomedicine, these interacting elements take various forms: inherited and somatic human gene variants and polymorphisms, epigenetic effects on gene activity, environmental agents, and drug therapies including drug combinations. The success of predictive, preventive, and personalized medicine will require not only the ability to determine the genotypes of patients and to classify patients on the basis of molecular fingerprints of tissues. It will require an understanding of how genetic perturbations interact to affect clinical outcome. Recent advances afford the capability to perturb genes and collect phenotype data on a genomic scale \[[@B1]-[@B7]\]. To extract the biological information in these datasets, parallel advances must be made in concepts and computational methods to derive and analyze genetic-interaction networks. We report the development and application of such concepts and methods. Results and discussion ====================== Phenotype data and genetic interaction -------------------------------------- A genetic interaction is the interaction of two genetic perturbations in the determination of a phenotype. Genetic interaction is observed in the relation among the phenotypes of four genotypes: a reference genotype, the \'wild type\'; a perturbed genotype, A, with a single genetic perturbation; a perturbed genotype, B, with a perturbation of a different gene; and a doubly perturbed genotype, AB. Gene perturbations may be of any form (such as null, loss-of-function, gain-of-function, and dominant-negative). Also, two perturbations can interact in different ways for different phenotypes or under different environmental conditions. Geneticists recognize biologically informative modes of genetic interaction, for example, epistasis and synthesis. These two modes can illustrate the general properties of genetic interactions. An epistatic interaction occurs when two single mutants have different deviant (different from wild-type) phenotypes, and the double mutant shows the phenotype of one of the single mutants. Analysis of epistatic interactions can reveal direction of information flow in molecular pathways \[[@B8]\]. If we represent a phenotype of a given genotype, X, as Φ~X~, then we can write a phenotype inequality representing a specific example of epistatic genetic interaction, for example, Φ~A~\< Φ~WT~\< Φ~B~= Φ~AB~. Likewise, a synthetic interaction occurs when two single mutants have a wild-type phenotype and the double mutant shows a deviant phenotype, for example, Φ~WT~= Φ~A~= Φ~B~\< Φ~AB~. Synthetic interactions reveal mechanisms of genetic \'buffering\' \[[@B1],[@B9]\]. Some modes of genetic interaction are symmetric; other modes are asymmetric. This symmetry or asymmetry is evident in phenotype inequalities, and is biologically informative. Epistasis illustrates genetic-interaction asymmetry. If mutation Ais epistatic to B, then Bis hypostatic to A. The asymmetry of epistasis, and the form of the mutant alleles (gain or loss of function), indicates the direction of biological information flow \[[@B8]\]. Conversely, synthetic interactions are symmetric. If mutation Ais synthetic with B, then Bis synthetic with A. The symmetry of genetic synthesis reflects the mutual requirement for phenotype buffering \[[@B1],[@B9]\]. The representation of genetic interactions as phenotype inequalities accommodates all possibilities without assumptions about how genetic perturbations interact. In addition, it demands quantitative (or at least ordered) phenotypes. In principle, all phenotypes are measurable; complex phenotypes (for example, different cell-type identities) are amalgamations of multiple underlying phenotypes. There is a total of 75 possible phenotype inequalities for WT, A, B, and AB. Using a hybrid approach combining the mathematical properties of phenotype inequalities and familiar genetic-interaction concepts and nomenclature, the 75 phenotype inequalities were grouped into nine exclusive modes of genetic interaction, some of which are genetically asymmetric (Additional data file 1). This approach can be extended to the interactions of more than two perturbations as well. The nine interaction modes include familiar ones: noninteractive, epistatic, synthetic, conditional, suppressive, and additive; and modes that certainly occur but, to our knowledge, have not been previously defined: asynthetic, single-nonmonotonic, and double-nonmonotonic. All interaction modes are defined in the Materials and methods; brief descriptions follow for the unfamiliar (previously undefined) modes. In asynthetic interaction, A, B, and ABall have the same deviant phenotype. In single-nonmonotonic interaction, a mutant gene shows opposite effects in the WTbackground and the other mutant background (for example, Φ~WT~\< Φ~A~and Φ~AB~\< Φ~B~). In double-nonmonotonic interaction, both mutant genes show opposite effects. Genetic-interaction networks ---------------------------- Implementation of the foregoing principles renders genetic-interaction-network derivation fully computable from data on any measured cell property with any interacting perturbations. We developed an open-source cross-platform software implementation called PhenotypeGenetics, available at \[[@B10]\], a plug-in for the Cytoscape general-purpose network visualization and analysis platform \[[@B11]\]. PhenotypeGenetics supports an XML specification for loading any dataset, allows user-defined genetic-interaction modes, and supports all of the analyses described in this paper. It was used to derive and analyze a genetic-interaction network from yeast invasion phenotype data. In response to growth on low-ammonium agar, Saccharomyces cerevisiae MATa/αdiploid yeast cells differentiate from the familiar ovoid single-cell growth form to a filamentous form able to invade the agar substrate \[[@B12]\]. Invasive filamentous-form growth is regulated by a mitogen-activated protein kinase (MAPK) kinase cascade, the Ras/cAMP pathway, and multiple other pathways \[[@B13],[@B14]\]. We investigated genetic interaction among genes in these pathways and processes. Quadruplicate sets of homozygous diploid single-mutant and double-mutant yeast strains were constructed (Materials and methods). Two purposes guided the selection of genes and mutant combinations to study: to represent key pathways and processes regulating invasion; and to ensure a diversity of invasion phenotypes (non-invasive, hypo-invasive, wild type, and hyper-invasive) to permit the detection of diverse genetic interactions. A set of 19 mutant alleles of genes in key pathways controlling invasive growth, including 13 plasmid-borne dominant or multicopy wild-type alleles and 6 gene deletions, was crossed against a panel of 119 gene deletions. All mutant alleles used in this study are listed in Additional data file 2. We developed a quantitative invasion-phenotype assay. Yeast agar-substrate invasion can be assessed by growing colonies on low-ammonium agar, removing cells on the agar surface by washing, and observing the remaining growth of cells inside the agar. Replicate quantitative invasion-phenotype data with error ranges were extracted from images of pre-wash and post-wash colonies. Each tested interaction was recorded as an inequality, and assigned a genetic-interaction mode. This process is detailed in the Materials and methods and illustrated in Figures [1a](#F1){ref-type="fig"} and [1b](#F1){ref-type="fig"}, using the example of the epistasis of a deletion of the FLO11gene, a major determinant of invasiveness, to a deletion of the SFL1gene, encoding a repressor of FLO11. Note that the error-bounded intervals (black bars) for the genotypes in Figure [1b](#F1){ref-type="fig"} are representative of the entire dataset. These errors are: flo11, 0.02; flo11 sfl1, 0.01; WT, 0.05; sfl1, 0.06. [Additional data file 3](#S3){ref-type="supplementary-material"} shows a plot of error values for all genotypes sorted by error magnitude. The median error is 0.04. Graphical visualization of the genetic interactions revealed a dense complex network. For clarity, a small part of this network (interactions among transcription factors) is shown in Figure [1c](#F1){ref-type="fig"}, illustrating the diversity of the observed genetic interactions. Perturbed genes are nodes in the network. Each tested allele combination generates an edge representing a genetic interaction. Edge colors and arrow heads (where appropriate) indicate interaction mode and asymmetry as indicated in Figure [1d](#F1){ref-type="fig"}. The entire network of 127 nodes and 1,808 edges is shown in Additional data file 4. All of the underlying data, including tested interactions, genotypes, and quantitative phenotype data with error values, are listed in Additional data file 5. All nine genetic-interaction modes were observed among the 1808 interactions. Other than the noninteractive mode (with 443 occurrences), the most frequent modes were additive (347), epistatic (271), conditional (245), and suppressive (202) interaction. Lower frequencies of asynthetic (111), single-nonmonotonic (74), synthetic (62), and double-nonmonotonic (52) interaction were observed. Note that though the asynthetic, single-nonmonotonic, and double-nonmonotonic modes are not recognized by common genetic nomenclature, they occurred at substantial frequencies. Genetic perturbations interacting with a specific biological process -------------------------------------------------------------------- Because genetic interactions reflect functional interactions, a genetic perturbation may interact in a specific mode with more than one gene in a specific biological process. This conjecture is supported by the finding of \'monochromatic\' interaction among biological-process modules \[[@B15]\]. Table [1](#T1){ref-type="table"} lists 23 interactions in a specific mode between a mutant allele and a biological process. The statistical validation of these interactions is detailed in the Materials and methods. Figure [2](#F2){ref-type="fig"} shows three examples. In Figure [2a](#F2){ref-type="fig"}, a PBS2gene deletion is additive with mutations of small-GTPase-mediated signal transduction genes (P= 0.001). These include genes in the Rho signal transduction/cell polarity pathway (BNI1, CLA4, BUD6) and the Ras/cAMP signaling pathway (RAS2, BMH1, TPK1). These signaling pathways contribute to invasive growth phenotype in concert with the stress response regulated by the Pbs2 MAPK kinase \[[@B16]\]. In Figure [2b](#F2){ref-type="fig"}, deletions of invasive-growth genes DFG16, RIM8, and DIA2are epistatic to overexpression of the invasion-activating Phd1 transcription factor (P= 0.002). The combination of this epistasis with the forms of the interacting alleles (PHD1overexpression is a gain of function, whereas the others are null alleles) leads to the suggestion that DFG16, RIM8, and DIA2may be regulated by Phd1. In Figure [2c](#F2){ref-type="fig"}, a deletion of the ISW1gene suppresses the effects of perturbations of small-GTPase-mediated signal transduction genes CDC42, RAS2, and IRA2(P= 0.005). ISW1encodes an ATP-dependent chromatin-remodeling factor \[[@B17]\]. Halme et al. \[[@B18]\] have shown that invasiveness of yeast cells is controlled epigenetically. High-frequency spontaneous mutations of IRA1and IRA2relieve epigenetic silencing of invasion genes. The suppression of an IRA2mutation by ISW1mutation suggests the possibility that ISW1-dependent chromatin remodeling mediates effects of IRA2mutation. Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"} illustrate local interaction patterns among mutant genes and biological processes. Mutually informative patterns of genetic interaction ---------------------------------------------------- The phenotypic consequences of combinatorial genetic perturbations are complex, in a strict sense; knowing the phenotypes of two single perturbations, there are no simple rules to know the combinatorial phenotype. Counteracting this complexity, large sets of genetic-interaction data may contain large-scale patterns. We examined the possibility that there are pairs of perturbations with mutually informative patterns of genetic interaction with their common interaction partners. In other words, knowing the interactions of one perturbation may allow one to know, to some quantifiable extent, the interactions of another perturbation, and vice versa. Mutual information, and significance thereof, was calculated for all pairs of perturbations sharing tested interactions with other genes. For all 171 pairs of the 19 mutant alleles of genes in key pathways, mutual information was based on their interactions with the panel of 119 gene deletions. Similarly, among all 7,021 pairs of the 119 gene deletions, mutual information was based on their interactions with the 19 mutant alleles of genes in key pathways. Among all possible pairs, 23 showed significant (P\< 0.001) mutual information (Materials and methods and Additional data file 6). The results suggest that the most mutually informative genetic-interaction patterns occur among gene perturbations with similar effects on biological processes. For example, three of the six mutant gene pairs with the most significant mutual information are overexpressers of STE12-STE20, STE12-CDC42, and STE20-CDC42(Additional data file 6). These three genes encode central components of the MAPK signaling pathway promoting invasive filamentous-form growth \[[@B14]\], and they show similar patterns of genetic interaction, as exemplified by STE12and STE20in Figure [3](#F3){ref-type="fig"}. The dominant pattern is one of uniform interaction (A and B interact in the same mode with C), suggesting similar effects of the gene perturbations on the underlying molecular network. In addition, there are frequent occurrences of repeated mixed-mode interaction (A interacts in some mode with C, and B interacts in a different mode with C), suggesting that the molecular effects of gene perturbations may differ yet show consistent differences. Both uniform interaction and consistent mixed-mode interaction contribute to mutual information. Genetic interactions are ultimately a property of a network of biological information flows. The mutual information among pathway co-member genes like STE12and STE20supports this. Figure [4](#F4){ref-type="fig"} shows a mutual-information network of perturbed genes. Each edge indicates significant mutual information (Additional data file 6). Some of these edges connect genes in different cellular processes. For example, an edge connects the GLN3gene, encoding a transcriptional regulator of nitrogen metabolism, and the CDC42gene, encoding a GTPase involved in cell polarity. Such cases of mutual information suggest that in the underlying molecular network, there are important information flows between the different pathways and processes. In addition to pairwise mutual information, there is the possibility that multiple genes may exhibit significant mutual information. The network in Figure [4](#F4){ref-type="fig"} contains multiple n-cliques, subnetworks of ncompletely connected nodes. There is a 3-clique, including two main components (PBS2and HOG1) of the HOG MAP-kinase pathway, and three overlapping 4-cliques (with many subcliques) containing filamentation MAPK pathway components. The STE12-STE20-CDC423-clique is in this cluster of cliques. The cliques and clusters suggest ternary and higher orders of mutual information, reflecting similarities in the global effects of perturbations on molecular information flows. Conclusion ========== The analysis of genetic interactions determining yeast invasion phenotype suggests some prospects for system-level genetics. The gene-process interactions in Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"} suggest that (as noted for epistasis and synthesis) there are characteristic network mechanisms to be found underlying familiar and unfamiliar modes of genetic interaction. Investigation of these mechanisms should provide insight on specific processes and general properties of biological networks. There are several areas for further development of the quantitative analysis of genetic interaction: first, advances in quantitative phenotype measurement and ontologies; second, reinforcement or revision of genetic-interaction mode definitions based on relevance to network mechanisms; third, extension of all genetic-interaction modes beyond phenotype ordering to incorporate parameters derived from phenotype magnitudes; and fourth, comparative genetic-interaction analyses of multiple alleles (with different effects on function) of individual genes to learn how different levels of gene activity impact the network. The global genetic-interaction patterns illustrated in Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"} are readouts of the state of the underlying molecular network. Data relating genotype and phenotype are essential for understanding metabolic and information-flow paths. Genetic data, integrated with gene-activity data and molecular-interaction data, reveal direction of information flow, activations, repressions, and combinatorial controls. The genome-scale integration of molecular-wiring maps, gene-expression data, and genetic-interaction networks will enable the development of biological-network models that explicitly predict the phenotypic consequences of genetic perturbations \[[@B19]\]. Materials and methods ===================== Strain constructions -------------------- A total of 127 genes involved in the regulation of invasion were selected for study from searches of the YPD database \[[@B20]\] and gene-expression profiling experiments \[[@B21],[@B22]\]. 138 mutant alleles of these 127 genes, including 125 deletions and 13 plasmid-borne alleles, were assembled (Additional data file 2). Single-mutant homozygous diploid strains were constructed in the invasion-competent Σ1278b budding-yeast strain background. In quadruplicate constructions, a 19 mutant-allele subset, including the 13 plasmid-borne alleles and six of the gene deletions, was crossed against the other 119 deletions. Homozygous diploid double mutants were generated as follows. Single-gene deletions in the invasion-competent Σ1278b yeast background were constructed. \'Barcode\' gene deletion-insertion alleles \[[@B5]\] were PCR amplified with several hundred base pairs of flanking sequences from their noninvasive strain background. Using the G418 drug-resistance cassette of these alleles, strain G85 (MATa/α ura3Δ0/ura3Δ0 his3Δ0::hisG/his3Δ0::hisG) was transformed with the PCR products. Gene disruption and the presence of the KanMX4 insertion were verified by PCR. The heterozygous diploids were sporulated and the resulting tetrads were dissected and screened to select G418-resistant MATaand MATαhaploids. These were crossed to obtain homozygous diploid gene-deletion strains. Some of the double mutants were generated by transforming the homozygous deletion strains with either low-copy plasmids bearing dominant alleles or multicopy (2 μm-based) plasmids bearing wild-type alleles. All plasmids utilized native gene promoters. Plasmid transformations were performed using an adapted version of a multiwell transformation protocol \[[@B5],[@B23]\]. Four independent transformants were stocked and assayed for each transformation. Strains were also transformed with empty vector plasmids. The high-throughput construction of diploid homozygous double-deletion strains required the use of three drug-resistance markers to be able to select for the desired diploids and intermediate strains. For each deletion, the KanMX4 drug-resistance marker was converted to two other drug-resistance markers, HygMX4 (hygromycin resistance) and NatMX4 (nourseothricin resistance). MATαgene-deletion strains were transformed with the NatMX4 cassette amplified from pAG25 \[[@B24]\]; Nat^R^G418^S^transformants were stocked. MATagene-deletion strains were transformed with the HygMX4 cassette amplified from pAG32 \[[@B24]\]; Hyg^R^G418^S^transformants were stocked. The high-throughput construction of diploid homozygous double-deletion strains required the ability to select haploids of each mating type separately. To accomplish this, we utilized the recessive resistance to canavanine caused by the disruption of the CAN1gene, encoding a transporter, in combination with fusions of the HIS3ORF to the promoters of genes expressed in a specific mating type. A deletion of the CAN1gene was constructed without introducing any marker genes or sequences. A double-stranded 60mer oligonucleotide containing 30 bases from the upstream region fused directly to 30 bases from the downstream region of the CAN1open reading frame (5\'-GTAAAAACAAAAAAAAAAAAAGGCATAGCAATATGACGTTTTATTACCTTTGATCACATT-3\') was amplified with 60mer primers containing additional CAN1flanking sequences (forward primer 5\'-CGAAAGTTTATTTCAGAGTTCTTCA GACTTCTTAACTCCTGTAAAAACAAAAAAAAAAAA-3\', reverse primer 5\'-GTGTATGACTTATGAGGGTGAGAATGCGAAATGGCGTGGAAATGTGATCAAAGGTAATAA-3\'). The resulting PCR product was used to transform two strains to canavanine resistance. Full deletion of the CAN1gene was confirmed by PCR. This generated strains G264 (MATahis3Δ::hisG can1Δ) and G266 (MATα his3Δ::hisG can1Δ). To construct fusions of the HIS3ORF to mating-type specific genes, the S. kluyveri HIS3gene was amplified from pFA6-His3MX6 \[[@B25]\] with primers containing ORF-flanking sequences for the MFA1locus (forward primer 5\'-GTTTCTCGGATA AAACCAAAATAAGTACAAAGCCATCGAATAGAAATGGCAGAACCAGCCCAAAA-3\', reverse primer 5\'-AAGGAAGATAAAGGAGGGAGAACAACGTTTTTGTA CGCAGAAATCACATCAAAACACCTTTGTT-3\') and with primers containing flanking sequences for the MFα 1locus (forward primer 5\'-GATTACAAACTATCAAT TTCATACACAATATAAACGATTAAAAGAATGGCAGAACCAGCCCAAAA-3\', reverse primer 5\'-ACAAAGTCGACTTTGTTACATCTACACTGTTGTTA TCAGTCGGGCTCACATCAAAACACCTTTGGT-3\'). The resulting PCR products were used to transform G264 and G266, respectively, to create strains G544 (MATahis3Δ::hisG can1Δmfa1::HIS3) and G546 (MATα his3Δ::hisG can1Δmfα1::HIS3). Crosses and sporulations were carried out to introduce the canavanine-resistance marker and the mating-type-specific-His^+^markers. MATαNat^R^deletion strains were crossed with G544. Nat^R^Ura^+^diploids were selected; all were Can^S^and His^-^. These diploids were sporulated; from random spore preparations Nat^R^Can^R^His^+^Ura^-^MATahaploids were identified. MATaHyg^R^deletion strains were crossed with G546. These diploids were sporulated; from random spore preparations Hyg^R^Can^R^His^+^Ura^-^MATαhaploids were identified. To make the diploid homozygous double-deletion strains, a series of high-throughput crosses and sporulations, in which all the desired intermediate cell types and deletion genotypes could be selected, was carried out \[[@B1]\]. At all stages, multiple strains were individually verified. Nat^R^MATasingle-deletion strains were crossed to an array of MATαG418^R^single-deletion strains. Hyg^R^MATαsingle-deletion strains were crossed to an array of MATaG418^R^single-deletion strains. From these crosses Nat^R^G418^R^diploids and Hyg^R^G418^R^diploids were selected, respectively. Diploids from each cross were sporulated. Haploid double-deletion strains were selected: His^+^Can^R^(MATahaploid) G418^R^Nat^R^double-deletion segregants, and His^+^Can^R^(MATαhaploid) G418^R^Hyg^R^double-deletion segregants, respectively. The resulting arrays of MATaand MATαhaploid double-deletion strains were mated and subjected to selection for G418^R^, Nat^R^, and Hyg^R^to generate diploid homozygous double-deletion strains. Assay of yeast invasiveness --------------------------- Strains were inoculated from frozen stocks into liquid media in 96-well plates and incubated 18 hours at 30°C. Each plate included at least eight wells containing wild-type controls. Cells were transferred with a 96-Floating-Pin Replicator and colony copier (V & P Scientific) onto SLAD agar \[[@B12]\] in an omnitray. Each 96-well plate was pinned in quadruplicate, resulting in a total of 384 colonies per SLAD-agar plate. Note that each genotype was constructed in quadruplicate and assayed on separate plates. Therefore, each genotype was assayed with a total of 16 replicates. Plates were incubated for 4 days at 30°C. After incubation, cell material was removed from the agar surface while rinsing the plate under running water. A 300 d.p.i. grayscale image of each plate was generated before and after the wash by placing the plate face down on a flatbed scanner and scanning with transmitted light. Images were inverted using Adobe Photoshop 6 and saved as TIFF files for quantitative image analysis. Processing of invasion-assay data --------------------------------- Colony growth and invasion were quantified using Dapple \[[@B26]\], software originally designed for the analysis of DNA microarray images. Each post-wash image was analyzed simultaneously with the corresponding pre-wash image to enable reliable definition of colony boundaries and direct comparison of cell material. Subtraction of local background intensity yielded un-normalized values for growth (G) and invasion (U) and invasiveness ratio (R= U/G). A normalization factor for each plate, N~p~, was obtained from multiple wild-type controls on each plate. For each replicate qon plate p, we obtained the wild-type invasiveness ratio R^wt^~q,p~= U^wt^~q,p~/G^wt^~q,p~and defined the plate normalization factor N~p~as N~p~= median~q~(R^wt^~q,p~)/median~p~(median~q~(R^wt^~q,p~)). For a given genotype g, the normalized invasiveness ratio is given by R^g^~q,p~= (U^g^~q,p~/G^g^~q,p~)/N~p~= R^g^~i~where in the final equality we renumbered into a single ordinal index the Nreplicates i= 1,2,..,N(N≤ 16). We excluded any genotype gfor which N\< 5 due, for example, to deficient growth. From normalized data, we derived phenotype values and measurement errors. We obtained the median ratio R^g^= median~i~(R^g^~i~), and the median absolute deviation MAD^g^= MAD(R^g^~i~) = median~i~(\\|R^g^~i~-R^g^\\|). As a lower bound in error estimates we used MAD^Q=\ 0.1^, the tenth percentile of all MAD^g^. Thus, the phenotype values are reported as R^g^, with error E^g^= max(MAD^g^, MAD^Q=\ 0.1^). The frequencies of genetic-interaction modes were insensitive to increases of the error lower bound to the 50th percentile. Directed checks of individual components of the automated processing were made throughout. These included: visual inspection of each individual image, check of colony morphology, spot checking of well-characterized individual strains from start to finish in the analysis pipeline, screening for systematic errors in assay intensities. We confirmed that the plate-wise normalization did not lead to error amplification due to division by small numbers. Derivation of phenotype inequalities ------------------------------------ The following steps were carried out using PhenotypeGenetics software. Phenotypes and errors of genotypes WT, A, B, and AB\[(R^wt^,E^wt^), (R^A^,E^A^), (R^B^,E^B^), and (R^AB^, E^AB^)\] were assigned a phenotype inequality relation. This was done by first defining the error-bounded interval I^g^= \[R^g^-E^g^, R^g^+E^g^\] for each genotype. All pairs of genotypes were assigned an equality, Φ^g1^= Φ^g2^if interval I^g1^overlapped with I^g2^. Transitivity of equalities (if a = b and b = c, then a = c) was applied to yield disjoint groups of phenotype equalities. Inequalities, greater than (\>) or less than (\<) were assigned for the relations between equality groups. The resulting inequalities were assigned to genetic-interaction modes and asymmetries as described below. The results of all tests of genetic interaction were rendered as a graph as illustrated in Figure [1d](#F1){ref-type="fig"}. The entire resulting network is shown in Additional data file 4. One can obtain PhenotypeGenetics software or use it to analyze the invasion network at \[[@B10]\]. Modes of genetic-interaction ---------------------------- The 75 possible phenotype inequalities were assigned to modes of genetic interaction based on computable criteria. For each mode, we list the criterion for the inclusion of a phenotype inequality. In these criteria, \'background\' refers to a genotype with its complement of wild-type and mutant genes, into which other genetic perturbations are added, and \'effect\' refers to a change in a phenotype, either an increase or a decrease, upon a single genetic perturbation of a background. In the examples below, additional cases may be generated by operations such as exchanging Aand B, or reversing the effect of both alleles (for example reversing the effect of the Amutant gene with Φ~WT~\< Φ~A~gives Φ~A~\< Φ~WT~). Additional data file 1 lists each of the 75 phenotype inequalities and their assigned genetic-interaction mode and asymmetries. Figure [1d](#F1){ref-type="fig"} shows graph visualizations for all nine genetic-interaction modes. ### Noninteractive interaction Ahas no effect in the WTand Bbackgrounds (for example, Φ~WT~= Φ~A~\< Φ~B~= Φ~AB~), or Bhas no effect in the Aand WTbackgrounds, or both hold true (5 inequalities). ### Epistatic interaction Aand Bhave different effects (in terms of direction or magnitude) on the wild-type background and the double mutant has the same phenotype as either Aor B(for example, Φ~A~\< Φ~WT~\< Φ~B~= Φ~AB~) (12 inequalities). ### Conditional interaction Ahas an effect only in the Bbackground, or the Bmutant has an effect only in the Abackground (12 inequalities). ### Suppressive interaction Ahas an effect on WT, but that effect is abolished by adding the suppressor B, which itself shows no single-mutant effect (for example, Φ~WT~= Φ~B~= Φ~AB~\<Φ~A~); or, the corresponding holds under exchange of Aand B(4 inequalities). ### Additive interaction Single-mutant effects combine to give a double-mutant effect as per Φ~WT~\<Φ~A~= Φ~B~\<Φ~AB~, Φ~B~\< Φ~WT~= Φ~AB~\< Φ~A~, Φ~WT~\< Φ~A~\< Φ~B~\< Φ~AB~, Φ~B~\< Φ~WT~\< Φ~AB~\< Φ~A~, and all additional inequalities obtained by interchanging Aand B, or reversing the effect of both Aand B(12 inequalities). ### Synthetic interaction Aand Bhave no effect on the WTbackground, but the ABcombination has an effect (2 inequalities). ### Asynthetic interaction A, B, and the ABcombination all have the same effect on the WTbackground (2 inequalities). ### Single-nonmonotonic interaction Bshows opposing effects in the WTand Abackgrounds (for example, Φ~B~\> Φ~WT~and Φ~AB~\< Φ~A~); or, Ashows opposing effects in the WTand Bbackgrounds, but not both (8 inequalities). ### Double-nonmonotonic interaction Both Aand Bshow opposing effects in the WTbackground and the background with the other mutant gene (18 inequalities). Genetic interaction with biological processes --------------------------------------------- To identify statistically significant correlations between a given allele\'s interaction modes and biological processes, the neighbors of every allele in the network were queried for interaction class and Gene Ontology (GO) Consortium database annotations \[[@B27]\]. Each interaction class is defined by the interaction mode and direction, if any. For example, \'A suppresses B\' and \'A is suppressed by B\' are placed in different interaction classes. There are 13 interaction classes and 9 interaction modes (described above). Likelihood values were computed to find over-represented class-annotation pairings within each set of nearest neighbors, and P-values were assigned relative to a cumulative hypergeometric distribution. The result was a computer-generated list of biological statements relating genes, interaction classes, and target annotations, with entries such as \'A loss-of-function mutation of HSL1is suppressed by mutations of cell wall organization and biogenesis genes (-log~10~P= 2.52).\' These are listed in tabular form in Table [1](#T1){ref-type="table"}. To calibrate the significance of the results, a parallel calculation was performed for every test in the network in which the fractional probabilities of each possible outcome were added to an overall distribution of P-values for the entire network. For example, if a given mutation interacts with Nothers, N~C~of the interactions being of class Cand N~A~of those neighbors carrying annotation A, there is a finite set of outcomes for N~CA~, the number of neighbor mutations with annotation Aconnected via interaction C. The possible values of N~CA~follow a discrete hypergeometric distribution, and summing these distributions over all tests in the network yields a formally randomized distribution of P-values which has been constrained by the topology of the actual network. The distributions, real and theoretical, of -log~10~Pvalues were then compared by performing a chi-square test between comparable histograms. These tests showed a strong excess for -log~10~P\> 1.8. Mutual information of genetic interaction patterns -------------------------------------------------- We calculated the mutual information \[[@B28]\] of pairs of genetic perturbations. Each perturbation, X, has an observed discrete probability distribution of interaction classes (defined by mode and direction) with its tested interaction partners, P(x), where x∈ X, the set of interaction classes of perturbation X, and: ![](gb-2005-6-4-r38-i1.gif) Mutual information, I, of a pair of perturbations, Aand B, is the relative entropy of their joint probability distribution relative to their product probability distribution. Thus: ![](gb-2005-6-4-r38-i2.gif) Significance of mutual information was tested independently for each allele pair by computing the likelihood of obtaining the observed score in randomly permuted data. To remove bias due to our selection of mutant alleles, randomized data were constrained by keeping the wild type and two single-mutant phenotypes fixed and replacing interaction classes only with classes that are consistent with the observed single-mutant phenotypes. The choice among possible replacement classes was weighted by observed frequency in the entire network. Empirical tests showed randomized mutual information scores to be normally distributed, and multiple randomizations were carried out to determine a mean and standard deviation to characterize the distribution for each tested allele pair. P-values were then calculated as the probability of finding a mutual information score at or above the observed score. Allele pairs with probabilities below the cutoff of P\< 0.001 are listed in Additional data file 6, and shown as a graph in Figure [4](#F4){ref-type="fig"}. Additional data files ===================== The following data are available with the online version of this paper. Additional data file [1](#S1){ref-type="supplementary-material"} is a table showing 75 genetic-interaction inequalities in nine modes of genetic interaction. Additional data file [2](#S2){ref-type="supplementary-material"} lists the gene perturbations used in this study. Additional data file [3](#S3){ref-type="supplementary-material"} is a figure plotting phenotype error values in the entire dataset. Additional data file [4](#S4){ref-type="supplementary-material"} shows the entire genetic interaction network derived from yeast invasion-phenotype data. Additional data file [5](#S5){ref-type="supplementary-material"} lists phenotype data for all tested interactions. Additional data file [6](#S6){ref-type="supplementary-material"} lists mutual information in genetic-interaction patterns. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A table showing 75 genetic-interaction inequalities in nine modes of genetic interaction. As described in Materials and methods, all 75 possible phenotype inequalities were classified into nine modes of genetic interaction. The results are listed here. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Gene perturbations used in this study. This file lists all genes, mutant alleles, and allele forms (for example, null, gain-of-function, etc.) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Phenotype error values in the entire dataset. This plot shows the phenotype error values (Materials and methods) plotted against percentile of all genotypes ordered by error magnitude. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 Entire genetic interaction network derived from yeast invasion-phenotype data. Figure [1c](#F1){ref-type="fig"} shows a small part of the genetic-interaction network. This file contains an image including all tested interactions. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Phenotype data for all tested interactions. This file lists all tested genetic interactions as well as the phenotype and error values for all genotypes, WT, A, B, and AB. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 Mutual information in genetic-interaction patterns. This file lists the mutual information, and significance, among pairs of genes connected by edges in Figure [4](#F4){ref-type="fig"}. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank J. Aitchison, C. Aldridge, G. Church, L. Hood, S. Istrail, A. Markiel, S. Prinz, F. Roth, D. Segre, and J. Taylor for their contributions. This work was funded in part by Merck & Co. V.T. was supported by NIH Grant P20 GM64361. T.G. is a recipient of a Burroughs Wellcome Fund Career Award in the Biomedical Sciences. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Application of the method to yeast agar invasion data to derive a genetic-interaction network. (a)Pre-wash and post-wash images of example genotypes in a yeast agar-invasion assay. (b)The invasion data shown on a phenotype axis with replicate-measurement error ranges, as a phenotype inequality, as a genetic-interaction mode, and as a graphical visualization. (c)Part of the network (only transcription factor genes) is shown. Nodes represent perturbed genes; edges represent genetic interactions. A key to the interactions is given in (d). (d)Graphical visualizations of genetic interaction modes and asymmetries, and example phenotype inequalities. ::: ![](gb-2005-6-4-r38-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Gene perturbations show specific modes of genetic interaction with biological processes. (a)PBS2deletion interacts additively with mutations of small-GTPase-mediated signal transduction genes. (b)PHD1overexpression is hypostatic to deletions of invasive-growth genes. (c)ISW1deletion suppresses the effects of perturbations of small-GTPase-mediated signal transduction genes. Key to interactions as in Figure 1d ::: ![](gb-2005-6-4-r38-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Mutually informative genes show large-scale patterns of genetic interaction. Genetic interactions of STE12and STE20overexpressers. Key to interactions as in Figure 1d. ::: ![](gb-2005-6-4-r38-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Networks of mutual information in patterns of genetic interaction show cliques. Nodes represent perturbed genes (see Additional data file 2). gf indicates a gain-of-function allele; lf indicates a loss-of-function allele. Edges connect gene pairs with significant mutual information in their patterns of genetic interaction (see Additional data file 6). ::: ![](gb-2005-6-4-r38-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Genetic interactions of mutant genes with biological processes ::: Gene Form\* Interaction Biological process -log~10~P ----------- -------- ------------------------ ---------------------------------------------------------------- ------------- PBS2 null Additive Signal transduction 2.99 PBS2 null Additive Small gtpase mediated signal transduction 2.96 STE12 gf Single-nonmonotonic to Protein targeting 2.87 STE11 da Noninteractive Cell cycle 2.73 PHD1 gf Hypostatic to Invasive growth 2.68 PDE2 null Noninteractive Protein amino acid phosphorylation 2.56 HSL1 null Suppressed by Cell wall organization and biogenesis 2.52 STE20 gf Single-nonmonotonic to Protein targeting 2.31 EGT2 null Conditioned by Invasive growth 2.30 ISW1 null Suppresses Small gtpase mediated signal transduction 2.30 CLB1 null Noninteractive Protein metabolism 2.30 STE11 da Suppresses Cell surface receptor linked signal transduction 2.28 BEM1 gf Conditioned by Nucleobase, nucleoside, nucleotide and nucleic acid metabolism 2.25 PBS2 null Additive Ras protein signal transduction 2.24 PBS2 null Additive Sporulation 2.24 TEC1 gf Synthetic Intracellular signaling cascade 2.19 IPK1 null Additive M phase 1.95 TEC1 null Epistatic to Metabolism 1.94 TEC1 gf Conditioned by Carbohydrate metabolism 1.94 TEC1 gf Conditioned by Ras protein signal transduction 1.94 BUD4 null Noninteractive Establishment of cell polarity 1.94 HMS1 null Noninteractive Protein amino acid phosphorylation 1.83 YGR045C null Noninteractive Protein amino acid phosphorylation 1.83 \*gf, gain-of-function; da, dominant-active. :::	PubMed Central	2024-06-05T03:55:56.393453	2005-3-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088966/", "journal": "Genome Biol. 2005 Mar 31; 6(4):R38", "authors": [ { "first": "Becky L", "last": "Drees" }, { "first": "Vesteinn", "last": "Thorsson" }, { "first": "Gregory W", "last": "Carter" }, { "first": "Alexander W", "last": "Rives" }, { "first": "Marisa Z", "last": "Raymond" }, { "first": "Iliana", "last": "Avila-Campillo" }, { "first": "Paul", "last": "Shannon" }, { "first": "Timothy", "last": "Galitski" } ] }
PMC1088967	Background ========== Malaria is currently considered to be the most important tropical disease, afflicting 300-500 million people, and killing over 1 million annually \[[@B1]\]. It is caused by infection of the human host with a single-celled parasite belonging to the genus Plasmodiumand relies on female mosquitoes of the genus Anophelesfor its transmission. The recent whole-genome sequencing of Anopheles gambiae, the primary vector in sub-Saharan Africa of Plasmodium falciparum- the agent of the most common and deadly type of malaria - now provides researchers with a vast set of data with which to better understand this insect vector and to develop possible solutions to malaria \[[@B2]\]. Annotation of the A. gambiaegenome by defining genes and other genomic features is the first step in moving from the realm of simply a genome sequence to one of understanding gene function. Extremely important to this effort is the accumulation of high-quality sequence data capable of refining the structural features of known genes and revealing previously unknown genes. Unfortunately, before the completion of the genome sequence, very few Anophelesgenes were well characterized experimentally, with exceptions primarily being genes involved either in olfaction or in host-parasite interactions (for example, innate immunity genes). While the amount and quality of publicly available sequence data is improving, a second complete Anophelesgene build in October of 2003 by Ensembl was able to utilize only 40,000 expressed sequence tag (EST) sequences in the EST gene build, leaving gene predictions heavily reliant on finding regions homologous with Drosophila, an organism that diverged from Anophelesmore than 250 million years ago \[[@B3]-[@B5]\]. A recent preliminary analysis of the Anophelesgenome annotation suggests that this lack of sequence data, combined with potential assembly problems and the absence of a closely related organism to use in homology comparisons, is proving a significant challenge for current attempts at genome annotation \[[@B6]\]. Like other groups \[[@B7],[@B8]\], we have initiated a program to increase the total amount of experimental sequence data and improve current Anophelesgene models. Unlike EST data, full-length cDNA libraries are biased toward complete copies of mRNA transcripts and thus provide significantly more information, including intron-exon structure as well as the first and last exons (often the most difficult to identify in silico\[[@B9]\]), alternative splicing, the correct start codon(s), and the full protein-coding sequence. Additionally, full-length transcripts can be used in the optimization of gene-expression studies and can be used directly as templates for protein synthesis. Here we report the sequencing and preliminary analysis of 67,044 reads from a full-length enriched cDNA library derived from whole-body adult female A. gambiaemosquitoes. These sequences were initially clustered with each other and then aligned to the Anophelesgenome sequence, and correspond to approximately 3,700 genes. Nearly 650 of these genes appear to be novel because they neither overlap nor simply extend previous Ensembl gene models. In addition, clusters that matched previous gene definitions improved those definitions in 85% of cases. These results demonstrate both the usefulness of full-length cDNAs in genome annotation, as well as the degree to which further annotation of the Anophelesgenome is needed. All sequences from this project were submitted to GenBank under the accession numbers BX005485-BX072528 and the physical clones are being submitted to the Malaria Research and Reference Center (MR4) \[[@B10]\]. Results and discussion ====================== We constructed a non-normalized library of enriched full-length cDNAs with RNA extracted from the complete body of adult female mosquitoes (see Materials and methods). Sequencing of clones was carried out from both the 5\' and 3\' ends of the cDNA insert. After sequencing, sequence reads were cleaned, clustered and assembled into consensus sequences using the Paracel Transcript Assembler package. Output from this process results in the creation of either a single consensus sequence or multiple consensus sequences (because of alternative splicing, for example) for each cluster of overlapping cDNAs. Individual reads that could not be initially clustered with any other sequence are referred to as singlets. Together, consensus and singlet sequences were aligned to the genome, and for each strand, overlapping cDNA sequences were grouped into a single cluster representing a putative gene. This process generates three major end products: clusters and singlets that overlap previously predicted gene models; novel clusters or singlets that do not overlap a gene model; and consensus sequences/singlets that do not align anywhere within the genome. A graphical summary of the analysis is shown in Figure [1](#F1){ref-type="fig"}. Note that we use the Ensembl \'unknown\' chromosome as part of this analysis (designated asUNKN). This artificial chromosome contains arbitrarily ordered concatenated scaffolds that currently have not been assigned to a particular chromosomal location. Comparison with previously predicted genes ------------------------------------------ To discern clusters representing known genes from those that would potentially be considered novel, we compared the coordinates on the genome of our cDNA clusters to those of Ensembl transcript models. Specifically, we used transcript model data taken from Ensembl gene build version 16.2.1, which did not have our cDNAs available for the creation of its 14,653 gene models. In this analysis, clusters were categorized as previously predicted if any overlap occurs, by even a single base, between a cluster and an Ensembl transcript. If no overlap occurs, genes were considered to be novel. Note that, as described in \[[@B11]\], Ensembl transcript models were generated from a combination of previously described Anophelesprotein sequences, high homology matches from SwissProt+TrEMBL and AnophelesEST information. In our analysis, we do not consider transcript models generated only from in silicogene-prediction algorithms. In summary, when we find a cluster that does not overlap any predicted Ensembl transcript, the cluster is considered the product of a new gene and is designated as being \'novel\'. If a cluster does overlap an Ensembl transcript, it is classified as known even if the initial evidence for that gene relies, for example, on homology alone. Using this approach we find that 3,032 clusters (86%) correspond to predicted Ensembl genes. Of these, nearly 46% (1,393 sequences) extend both the 5\' and 3\' ends of Ensembl-predicted transcripts. In addition, 9% extend the 5\' end only (271 clusters) and 31% extend only the 3\' end (935 clusters) of the corresponding Ensembl transcript. Just 433 clusters (around 14%) fell entirely within a predicted gene and did not extend either extremity of an Ensembl gene model. In addition, 536 clusters that do not correspond to any previously described Ensembl gene were also identified. The mean length of these novel clusters was 1,303 nucleotides versus 1,615 for Ensembl-predicted genes. As detailed in Table [1](#T1){ref-type="table"}, both Ensembl-predicted and novel clusters appear to be well distributed across the genome. As expected, the majority of clusters are composed of a small number of reads - 37% of clusters have two to three reads and 80% contain fewer than 12. The single cluster with the greatest number of reads (over 2,000) is annotated as a guanine-nucleotide-binding beta subunit. While consistent, this method does require that some qualifying conditions be kept in mind. First, it is possible that a gene that we designate as novel does in fact have some previous transcript information available as supporting evidence (such as EST data). This will happen, for example, if during the process of automatic annotation the existing information did not result in the creation of a new transcript model by Ensembl. In fact, in the initial analysis of the Anophelesgenome, as many as 1,029 genes were believed to have been missed in this manner \[[@B2]\]. Since the initial annotation process, the increased amount of available sequence information has improved coverage considerably. Despite these improvements, however, such misclassifications are unavoidable. In addition, if an Ensembl prediction is incorrect, an overlapping cluster would be classified as previously predicted, although it would, in fact, be new to the annotation. Inspection shows that such instances are rare and generally require additional experimental evidence as well as the manual definition of gene models for complete reconciliation of the data. While difficulties will exist with any such automated comparison, as a whole our approach is consistent, reproducible, and provides realistic estimates of both previously predicted and novel genes. Of the initial set of 10,961 singlets (see Figure [1](#F1){ref-type="fig"}), most (around 80%) not only aligned to the genome with high quality, but also overlapped with Ensembl predictions, while approximately 2,200 singlets were unable to be aligned. This latter group is discussed further in the next section. Additionally, 202 reads or \'singlets\' were found that accurately aligned to the genome but did not overlap with any Ensembl transcript predictions. These singlets are generally shorter in length than clusters, with a mean of 912 nucleotides. Of the 202 sequences, 65 were found through manual examination to be probable 5\' or 3\' extensions of a nearby Ensembl-predicted transcript. Of the remaining 137 singlets, 38 (or 28%) are non-overlapping 5- and 3-prime reads representing 19 genes where additional sequencing must be done to obtain the complete gene sequence. Blastx analysis against a combined SwissProt+TrEMBL database showed that 25 of the novel singlets (around 12%) have limited homology to previously described genes (E-value \< 10^-7^), with the remaining novel singlets having no significant similarity to the database members. Thus singlets provide evidence for 118 additional novel genes, and together with the previously described clusters, support the existence of 654 novel genes. While clusters supported by a singlet provide further opportunities to investigate potentially novel genes, we do not describe them further here. Future work will investigate such transcripts in greater detail. Unalignable sequences --------------------- We note that 2205 sequences (around 3% of all reads) cannot be aligned to the genome. Essentially all of these sequences are singlets, many of which are of low complexity and/or contain repetitive regions. Nearly half (1,066) were eliminated during the alignment process due to their poor quality (identity and/or coverage). It is possible, however, that some unaligned sequences represent genes lying within sequence gaps of the genome assembly. For example, within the unalignable group there are eight clusters having an average length of 1250 nucleotides, composed of from two to four reads, with three of these clusters consisting of overlapping 5\' and 3\' reads. Visual inspection suggests that most of these clusters also contain low-complexity regions. In addition, in two cases Blastx \[[@B12]\] searches against a nonredundant protein database reveal similarity to known proteins. One cluster has high similarity to a receptor for activated protein kinase C (RACK1; E-value \~10^-62^) while the second has similarity to a putative ribosomal protein (S8; E-value \~10^-12^). Comparing the remaining 1,139 reads that could not be initially aligned to any chromosome arm via BLAST, we found that at least 808 reads appear to be bacterial contaminants. Approximately 19% have no similarity to proteins in SwissProt+TrEMBL. Another 10% of the group (118 sequences) have similarity to known proteins (E-value \< 10^-7^). In fact, 32 sequences have similarity to previously identified Anophelesproteins. At this time, it is not clear whether these sequences fall into unsequenced gaps in the genome sequence, are of insufficient quality to align accurately, or are errors or some other artifact. While it is possible that many of these sequences that could encode proteins with similarity to known proteins are actual gene transcripts, we do not consider them further here and do not include them in our group of novel genes. Characterization of Ensembl-predicted and novel cDNA clusters ------------------------------------------------------------- To characterize cDNA clusters in terms of their potential biological role, we compared both Ensembl-predicted as well as novel gene clusters to a Gene Ontology (GO) annotated database (SwissProt+TrEMBL 796,016 sequences) \[[@B13],[@B14]\]. Using Blastx and an E-value of 10^-7^, clusters could be placed into a range of biological processes and functions (Figure [2](#F2){ref-type="fig"}). For the clusters supporting Ensembl-predicted genes, 2,398 of 3,032 (79%) could be assigned to a biological process or function, as compared to the novel clusters where only 123 out of 536 (23%) had at least one qualifying match. Of the deduced proteins of clusters corresponding to predicted genes, approximately 63% could be classified as having catalytic, binding, or nucleic-acid-binding function. Similarly, for deduced proteins of novel gene clusters, these same categories were the most highly populated, representing nearly 80% of classified functions. The processes of cell growth and/or maintenance and protein metabolism and modification were the most highly represented process categories for both Ensembl-predicted and novel cDNA clusters. To better describe the novel genes defined by the cDNAs, we compared consensus sequences from each cDNA cluster to a SwissProt+TrEMBL database and found that approximately 35% (188) of novel clusters had significant hits to known proteins (E-value 10^-7^). Again, these clusters were represented by a single consensus sequence composed of between two and 19 reads. For those transcripts without significant homology results, it is likely that many represent species-specific and/or insect-specific genes, and are thus of particular interest for more detailed experimental study. In addition, we attempted to identify a satisfactory open reading frame (ORF) in each cluster. Of the 536 novel clusters in the final set, 298 contained an ORF of at least 100 amino acids (see Materials and methods). Additional evidence in support of the biological reality of a gene or gene transcript is the existence of protein domains within the ORF as well as multi-exonic structure. As shown in Table [2](#T2){ref-type="table"}, we found 60 ORFs encoding at least one Pfam domain. Most domains are found only once; the zinc finger C2H2 domain is found 18 times, however, distributed over five different proteins. Further evidence in support of these clusters being real biological entities is the observation that nearly half of the clusters (47%) are comprised of two or more exons. GC content of cDNA clusters --------------------------- It has been suggested that, at least in the case of human genome annotation, there is a prediction bias against GC-rich transcripts by current gene-prediction methods \[[@B15]\]. To investigate the possibility that there are obvious biases in sequence properties of novel clusters that would make them more or less difficult to predict computationally, we determined the GC content for novel and predicted cDNA clusters and compared them to all Ensembl-predicted genes. As shown in Figure [3](#F3){ref-type="fig"}, the Ensembl transcript models are largely contained between 35 and 70% GC content with a mean of 54%. The range of GC content for both novel and predicted clusters spans a nearly equivalent range. For the novel clusters, however, there appears to be bias towards more AT-rich transcripts. The mean GC content for novel clusters was 46%, compared to 52% for clusters corresponding to predicted genes. As a whole, the Anophelesgenome has a GC content of 35.2% (Drosophila melanogasteris 41.1%) \[[@B2]\]. As a simple test, we compared novel clusters to geneid \[[@B16]\] predictions and found that 232 clusters (43%) overlap with a geneid prediction, while 311 novel clusters (57%) do not. In contrast, only 9% of Ensembl-predicted genes do not have a corresponding geneid prediction. This result suggests that the majority of novel genes would not be readily discovered without customized gene-finding methods. Currently, newer gene-finding methods specifically trained on AnophelescDNAs are now being developed and implemented (see, for example \[[@B17]\]) into the Ensembl gene prediction and annotation methodology (E. Mongin, personal communication). Examples of Ensembl-predicted and non-predicted clusters -------------------------------------------------------- As discussed earlier, genes represented by full-length cDNA transcripts span a wide range of molecular and cellular roles. Here we highlight a few examples and their relevance to current Anophelesresearch. Note that we have compared these transcripts to a more recent version of the Ensembl database (release 23) that now includes these cDNAs as part of the gene build process. As a result, our cDNA transcripts are identified in this section by their current ENSANGT, ENSANGEST, or name identifier whenever appropriate. While some of the genes described here had previous EST evidence, the availability of full-length enriched cDNAs for these transcripts is particularly valuable for future annotation. One transcript of interest encodes a protein containing both CLIP and serine protease domains. This protein, which we have designated here as Putative\_CLIPA5B, has been incorporated into Ensembl as part of transcript ENSANGT00000027174. In insects, these CLIP-domain serine proteases are involved in a variety of processes, including embryonic development and the innate immune response. For example, in response to malarial infection, CLIP-domain proteases help to initiate the prophenoloxidase cascade which, in \'malaria-resistant\' mosquitoes, results in the generation of reactive oxygen species and the eventual encapsulation of the parasite within a melanin capsule \[[@B18],[@B19]\]. Four subfamilies (A-D) are known within Anopheles, and phylogenetic analysis of the novel protein sequence deduced from our novel cluster suggests that it is a new member of the A subfamily (Figure [4a](#F4){ref-type="fig"}). Ten members of this family were previously described and CLIPA5 appears to be the closest relative of the new protein. The gene for the new protein lies within a cluster of 15 serine protease/CLIP-domain genes located on chromosome arm 3L (between 32.55-32.62 MB). Its similarity and proximity to clipA5would suggest that it arose from a recent duplication event. While the exact function of this new protein is unknown, it is interesting to note that transcription of a related member in the same subfamily, clipA6, is induced by bacterial infection \[[@B20]\]. We also identify a cDNA that encodes a peptidoglycan recognition protein (gene D of the long (L) subfamily - PGRPLD). Members of this protein family play a key role in the response to both bacterial and malarial infection \[[@B21]\]. While PGRPLD was not predicted in the original Anophelesannotation and was not part of the Ensembl 16 annotation, it was predicted without cDNA evidence in the preliminary analysis of immune genes within the genome \[[@B22]\] (Figure [5](#F5){ref-type="fig"}). In Drosophila, PGRPLD is enriched in hemocytes, is probably membrane bound and is actively expressed throughout development. Although its exact role in innate immunity is currently unknown, it is believed to be involved in bacterial recognition \[[@B23]\]. As many as three different gene products may be produced by pgprldin Drosophila, and our full-length cDNAs suggest two alternative start sites for this gene in Anopheles. Interestingly, as described in Drosophila, its untranslated 3\' end overlaps with an ORF on the opposite strand encoding retinaldehyde-binding and alpha-tocopherol transport domains \[[@B23]\]. The cDNAs for pgprldhave been incorporated into the supporting evidence for Ensembl EST transcript models ENSANGESTT00000363407 and ENSANGESTT00000363376. Other transcripts of interest are two previously non-predicted, putative P450 genes, which are of particular interest with regard to insecticide resistance. Currently, the major method for mosquito control within malaria endemic regions is the use of pyrethroid-based insecticides, typically through the impregnation of bednets and application to mosquito breeding sites \[[@B24]\]. While a major tool in the fight against malaria, the continued development of mosquito resistance to these insecticides has become an important problem. One potential mechanism of resistance to insecticides is the oxidative metabolism of insecticides mediated by cytochrome P450 \[[@B25],[@B26]\]. While definitive proof of the involvement of P450 in resistance is limited \[[@B27]\], it has been shown that certain P450 families are expressed at higher levels in various insecticide-resistant strains (see, for example \[[@B28],[@B29]\]). Of the two putative P450 genes discussed here, one (ENSANGT00000029062) has high similarity (E = 10^-146^) to CYP9L1 and the other has similarity to CYP6M4 (E = 10^-149^; Ensembl known transcript AAP76391). Both families are insect-specific, and members of the Cyp6family have been linked to insecticide resistance by elevated P450 activity in insecticide-resistant insects \[[@B25]\]. In total, we retrieved cDNAs representing 23 of the known 111 members of the AnophelesP450 family. We also find examples of interesting novel genes that are currently found only within this cDNA library. For example, our cDNAs identify a 869-base-pair (bp) gene (ENSANGT00000025538) which is most similar to mouse and human members of the MAGE (melanoma antigen-encoding) gene family. This gene was previously unrecognized in A. gambiaeeven though a Drosophilamember of this family does exist \[[@B30]\]. The gene was previously found to be expressed specifically in mammalian tumors and is developmentally regulated in Drosophila\[[@B30]\]. Another example is a transcript of approximately 1,300 bp which is homologous to DrosophilaDIP2 (Disco interacting protein 2, CG9771) which is involved in nervous system development \[[@B31]\]. Conclusion ========== We found that over 85% of previously predicted A. gambiaegenes had their boundaries extended either on the 5\', 3\', or both 5\' and 3\' ends by our initial full-length cDNA collection. While all the consensus models are not complete full-length transcripts, it is particularly encouraging that such a large percentage of previously predicted gene models were extended on both their 5\' and 3\' ends. The use of such full-length data is particularly valuable in the absence of well annotated and evolutionarily close organisms which can be used for sequence comparisons. The sequencing of the Aedes aegyptigenome is much anticipated in this regard. However, even with the availability of a genome from a mosquito relative, species-specific genes, along with the variability inherent in 5\' and 3\' exons, will probably require the use of full-length data for accurate gene characterization. A major result of this study was the finding that approximately 17% of the clusters represent previously unpredicted genes. This is perhaps more significant when considering that this was a non-normalized library constructed from whole mosquitoes. Further extrapolation suggests that at least a similar percentage of genes remains to be found elsewhere in the Anophelesgenome. Additional tissue- and treatment-specific libraries, currently under construction, should help to characterize more undiscovered genes. Note added in proof: A recent status report of the Anophelesannotation effort by Ensembl agrees with our estimates suggesting that around 600 new genes were discovered from the sequences presented in this communication, and that the current transcript set may be under-represented by as much as 20% \[[@B11]\]. Materials and methods ===================== Construction of oligo-capped cDNA libraries ------------------------------------------- Total RNA (cytoplasmic RNA and poly(A)^+^RNA) was isolated from 1,366 adult female A. gambiaestrain 6-9 mosquitoes, collected 24 h after oviposition by homogenization of the insects in TriReagent (Sigma) with an Ultra-Turax T25 homogenizer (IKA-Werke, Germany) as recommended by the suppliers. The isolated total RNA was resuspended in H20 and the poly(A)^+^RNA fraction was obtained from the equivalent of 700 μg total RNA using the Qiagen Oligotex mRNA batch protocol. Oligo-capped libraries were then constructed from the poly(A)^+^RNA fraction as described by Sugano and collaborators \[[@B32],[@B33]\]. Synthesis of the first-strand cDNA was obtained with the SuperScriptII RNase H-Reverse Transcriptase (Invitrogen); subsequently, the template RNA strand was degraded by alkaline hydrolysis and the first-strand cDNA was amplified using the LA Taq polymerase (Takara). After 20 PCR cycles the PCR fragments were digested with SfiI and size-fractionated by agarose gel electrophoresis. Two different size fractions (0.7-1 kilobase (kb), 1 kb-3 kb) were cloned into the vector pME18S-FL3 in an orientation-defined manner, using a DNA ligation kit (Takara). Ligations were electroporated into Escherichia coliDH10B electrocompetent bacteria (Invitrogen). Clones were randomly isolated and subjected to high-throughput single-path sequencing from their 5\' and 3\' ends. Note that this is a female whole-body library created under the constraints of selection for full-length transcripts within a given size range, and as such, does not provide a comprehensive survey of the genes expressed or capable of being expressed within the female Anophelesmosquito. Availability of libraries ------------------------- All libraries/clones are being deposited to MR4 and will be available there \[[@B10]\]. Sequence clustering, assembly and comparison -------------------------------------------- Sequences were cleaned, clustered and assembled using the Paracel TranscriptAssembler software package (Paracel). Cleaning consisted of comparing cDNA sequences against vector and mitochondrial databases, with matching sequences being removed from further analysis. In addition, low-complexity, poly(A/T) regions, and repeat regions (Ensembl repeat library courtesy of E. Mongin, Ensembl) were determined and masked. After sequence cleaning, masking and trimming, sequences with fewer than 200 unmasked bases were removed from further processing. As an aid to the initial clustering process, we used Ensembl release 16 cDNA transcripts as seed clusters. In this process, each cDNA is compared to each Ensembl transcript, and if significant similarity exists between the two, the cDNA is placed into a corresponding seed bin and clustered with all transcripts in this bin. Sequences that did not have high similarity to seed sequences were separately compared and clustered with each other. Next, both seed and non-seed clusters were assembled into one or more consensus sequences. If a sequence could not be assembled into the consensus sequences it was designated as a singlet. Finally, each consensus and singlet sequence was aligned to the Ensembl Anophelesgenome assembly (release 16.2.1) using a combination of BLAST and Spidey \[[@B12],[@B34]\] with minimum identity and coverage of 90% and 75% respectively. In addition, to prevent spurious \'exons\' from being produced from low-quality sequence noise common at read extremities, we trimmed terminal exons separated by over 10 kb and which were less than 50 nucleotides long. We compare the resulting clusters and singlets to Ensembl transcripts (ENSANGT identifiers) from the Anopheles16.2.1 release. Note that database revision numbers between 16.2.1 and 20 contain only one new gene build (ver. 17.2a.1 which incorporates the cDNA sequence data presented in this paper) with the rest primarily representing changes to the underlying database schema. If a cluster did not overlap on the genome with an Ensembl gene, it was classified as \'novel\'; otherwise it was classified as \'Ensembl predicted\'. The protein database used for homology searches was a combined Swiss-Prot (Release 44.2) and TrEMBL (Release 27.2) dataset. Internally, we used the Genome Browser (Gbrowse) \[[@B35]\] developed by the Generic Model Organism Database consortium \[[@B36]\] for display and analysis of clusters as well as the public resources provided by Ensembl \[[@B3]\]. Gene Ontology terms ------------------- We used the following terms and GO IDs in the creation of Figure [2](#F2){ref-type="fig"}: Biological Process-cellular process; GO:0009987, cell communication; GO:0007154, physiological process; GO:0007582, metabolism; GO:0008152, carbohydrate metabolism; GO:0005975, energy pathways; GO:0006091, electron transport; GO:0006118, nucleotide and nucleic acid metabolism; GO:0006139, amino-acid and derivative metabolism; GO:0006519, protein metabolism and modification; GO:0006411, lipid metabolism; GO:0006629, coenzymes and prosthetic group metabolism; GO:0006731, cell growth and/or maintenance; GO:0008151, death; GO:0016265, response to stress; GO:0006950. Biological Function-cell adhesion molecule activity; GO:0005194, chaperone activity; GO:0003754, GO:0003757, GO:0003758, GO:0003760, GO:0003761, defense/immunity protein activity; GO:0003793, catalytic activity; GO:0003824, enzyme regulator activity; GO:0030234, binding; GO:0005488, nucleic acid binding; GO:0003676, motor activity; GO:0003774, signal transducer activity; GO:0004871, structural molecule activity; GO:0005198, transcription regulator activity; GO:0030528, transporter activity; GO:0005215. CLIPA phylogenetic tree ----------------------- The regions containing the CLIP and serine protease domains for each sequences were aligned with ClustalX \[[@B37]\] (default values; version 1.83), manually adjusted in Jalview, and a neighbor-joining tree created, excluding gaps, with PAUP\. The CLIP and serine protease domains were included in the alignment and large insertions were removed before aligning. ORF determination ----------------- For each cluster considered, a representative cDNA sequence was taken (the longest in terms of total concatenated exon length if there were multiple consensus sequences in a cluster) and translated in all six reading frames. An ORF was defined as being at least 100 codons long, starting with a methionine and ending with a stop codon. Acknowledgements ================ We thank Corinne Da Silva, Betina Porcel and Vincent Schachter of Genoscope for helpful discussions. We also thank Emmanuel Mongin and others at Ensembl for their support and assistance to the Anophelesresearch community. Computational resources were provided in part by The AMDeC Bioinformatics Core Facility at the Columbia Genome Center, Columbia University. S.M.G. is supported by a grant from the Pasteur Foundation of New York. P.D. is supported by the Plate-forme Integration et Analyse Genomique, Génopole Institut Pasteur. C.W.R. is supported by the Centre National de la Recherche Scientifique, Sciences de la Vie. This work was supported by the Strategic AnophelesHorizontal Programme, Institut Pasteur. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow chart of sequence processing and categorization. ::: ![](gb-2005-6-4-r39-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Classification of clusters with Gene Ontology. Numbers above bars indicate the number of novel clusters in the given category. ::: ![](gb-2005-6-4-r39-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### GC content of cDNA clusters and Ensembl transcripts. ::: ![](gb-2005-6-4-r39-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Putative novel member of the CLIPA protein subfamily. (a)Phylogenetic tree of CLIPA subfamily proteins and the novel member described here - PUT CLIPA5B. The protein CG5390 is the closest Drosophilarelative to this protein. Bootstrap values are shown as percentages of 1,000 replications (see Materials and methods). (b)Genomic region containing the putative gene. Yellow bars indicate Ensembl 16 transcripts, cDNA evidence is shown in red and cDNA clusters are shown in green. Similarity and proximity suggest that this novel member probably arose through a recent duplication of CLIPA5. ::: ![](gb-2005-6-4-r39-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Peptidoglycan recognition protein LD. Cluster of cDNAs (cluster\_2935 in green) associated with the peptidoglycan recognition protein LD. Note that the current Ensembl definition of PGRPLD (in cyan) is truncated and does not currently reflect available transcript information. ::: ![](gb-2005-6-4-r39-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Distribution of cDNA clusters across the Anophelesgenome ::: 2R 2L 3R 3L X UNKN ----------- ----------- ----------- ----------- ----------- ---------- ---------- Predicted 950 (31%) 676 (22%) 588 (19%) 439 (15%) 229 (8%) 146 (5%) Novel 130 (24%) 141 (26%) 99 (18%) 92 (17%) 50 (9%) 31 (6%) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Pfam domains within novel ORFs ::: Pfam domain Description Number ------------------- ------------------------------------------------------------------- -------- adh\_short Short chain dehydrogenase 1 Aldo\_ket\_red Aldo/keto reductase family 1 Amidase\_2 N*-acetylmuramoyl-L-alanine amidase 1 Ank Ankyrin repeat 3 Bin3 Bicoid-interacting protein 3 (Bin3) 1 CBFD\_NFYB\_HMF Histone-like transcription factor (CBF/NF-Y) and archaeal histone 1 CH Calponin homology (CH) domain 1 CRAL\_TRIO CRAL/TRIO domain 1 Death Death domain 1 DEP Domain found in Dishevelled, Egl-10, and Pleckstrin 1 Dsrm Double-stranded RNA binding motif 2 DUF1395 Protein of unknown function (DUF1395) 1 DUF227 Domain of unknown function (DUF227) 1 DUF783 Protein of unknown function (DUF783) 1 Efhand EF hand 4 Exonuc\_X-T Exonuclease 1 F-box F-box domain 1 FYRC F/Y rich C-terminus 1 G\_glu\_transpept Gamma-glutamyltranspeptidase 1 GST\_C Glutathione S-transferase, C-terminal domain 1 HIT HIT domain 1 Ins\_allergen\_rp Insect allergen related repeat 1 Linker\_histone Linker histone H1 and H5 family 1 LRR Leucine rich repeat 4 LSM LSM domain 1 MtN3\_slv MtN3/saliva family 2 p450 Cytochrome P450 2 Pkinase Protein kinase domain 1 Psf2 Partner of SLD five, PSF2 1 Radical\_SAM Radical SAM superfamily 1 Retrotrans\_gag Retrotransposon gag protein 1 Ribosomal\_L27e Ribosomal L27e protein family 1 Ribosomal\_L36e Ribosomal protein L36e 1 Ribosomal\_L37e Ribosomal protein L37e 1 Ribosomal\_S8 Ribosomal protein S8 1 RRM\_1 RNA recognition motif. (a.k.a. RRM, RBD, or RNP domain) 1 SAM\_1 SAM domain (sterile alpha motif) 1 Serpin Serpin (serine protease inhibitor) 1 Tetraspannin Tetraspanin family 2 THAP THAP domain 2 TIL Trypsin inhibitor like cysteine rich domain 1 TIP49 TIP49 C-terminus 1 TPR TPR Domain 1 TraB TraB family 1 Trypsin Trypsin 1 Tubulin Tubulin/FtsZ family, GTPase domain 1 Tubulin\_C Tubulin/FtsZ family, C-terminal domain 1 UNC-50 UNC-50 family 1 UPF0224 Uncharacterized protein family (UPF0224) 1 WD40 WD domain, G-beta repeat 3 zf-C2H2 Zinc finger, C2H2 type 18 zf-C3HC4 Zinc finger, C3HC4 type (RING finger) 1 :::	PubMed Central	2024-06-05T03:55:56.396779	2005-3-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088967/", "journal": "Genome Biol. 2005 Mar 15; 6(4):R39", "authors": [ { "first": "Shawn M", "last": "Gomez" }, { "first": "Karin", "last": "Eiglmeier" }, { "first": "Beatrice", "last": "Segurens" }, { "first": "Pierre", "last": "Dehoux" }, { "first": "Arnaud", "last": "Couloux" }, { "first": "Claude", "last": "Scarpelli" }, { "first": "Patrick", "last": "Wincker" }, { "first": "Jean", "last": "Weissenbach" }, { "first": "Paul T", "last": "Brey" }, { "first": "Charles W", "last": "Roth" } ] }
PMC1088971	Introduction {#s1} ============ The worldwide emergence of antibiotic-resistant bacteria threatens to undo the dramatic advances in human health that were ushered in with the discovery of these drugs in the mid-1900s. Today, resistance has rendered most of the original antibiotics obsolete for many infections, mandating an increased reliance on synthetic drugs. However, bacteria also evolve resistance to these drugs, typically by acquiring chromosomal mutations \[[@pbio-0030176-b01]\]. Within the classical paradigm that mutations are the inevitable consequence of replicating a large genome with polymerases of finite fidelity, resistance-conferring mutations are unavoidable. However, recent evidence suggests that bacteria may play a more active role in the mutation of their own genomes in response to at least some DNA-damaging agents by inducing proteins that actually promote mutation \[[@pbio-0030176-b07]\]. If the acquisition of antibiotic resistance-conferring mutations also requires the induction of these proteins, then their inhibition would represent a novel approach to combating the growing problem of drug resistance. In an initial effort to examine the role of induced mutation in the evolution of antibiotic resistance, we have focused our studies on the synthetic antibiotic ciprofloxacin. Ciprofloxacin is a member of the quinolone family of antibiotics, which is rapidly becoming the most important family of antibiotics \[[@pbio-0030176-b16]\]. Quinolones function by interfering with the two essential type II DNA topoisomerases in bacteria, gyrase and topoisomerase IV \[[@pbio-0030176-b17]\]. These topoisomerases normally function by forming a protein-bridged DNA double strand break (DSB), manipulating DNA strand topology, and finally rejoining the DNA ends. Ciprofloxacin reversibly binds to the protein-bridged DSB intermediate and inhibits rejoining of the DNA ends. The toxic effects of ciprofloxacin may be the result of topoisomerase subunit dissociation without re-ligation of the DNA ends \[[@pbio-0030176-b17],[@pbio-0030176-b18]\], likely producing free double strand ends (DSEs) when the protein-DNA bond is eventually hydrolyzed or the DNA is processed by a nuclease. In addition, covalently bound topoisomerases may also block DNA replication forks, which after processing will also produce DSEs \[[@pbio-0030176-b17],[@pbio-0030176-b19]\]. Resistance to ciprofloxacin requires mutations in the genes that encode the topoisomerases (gyrA and gyrB, encoding gyrase, and parC and parE, encoding topoisomerase IV) or in the genes that affect cell permeability or drug export \[[@pbio-0030176-b01],[@pbio-0030176-b17]\]. To understand how mutation could be induced by ciprofloxacin, it is essential to consider how bacteria respond to the presence of the antibiotic. Previously it was suggested that ciprofloxacin-mediated DNA damage may be repaired by nucleotide excision repair (NER) or homologous recombination (HR) \[[@pbio-0030176-b17]\]. RecA-single-stranded DNA (ssDNA) filaments play an important role in these processes by catalyzing strand invasion of a DNA end into a homologous sequence. However, RecA-ssDNA filaments also play an important role in induced mutation by binding the SOS \[[@pbio-0030176-b10]\] gene repressor LexA and unmasking its autoproteolytic activity \[[@pbio-0030176-b24]\]. Upon autoproteolysis, LexA no longer represses the approximately 30 genes whose protein products facilitate the repair of DNA damage, delays in cell cycle division, and mutation. In particular, sufficient reduction of the cellular concentration of LexA repressor results in the transcription of the genes encoding Pol II (polB), Pol IV (dinB), and Pol V (umuD and umuC), which are three nonessential DNA polymerases that have been shown to be required for mutation in response to DNA damage \[[@pbio-0030176-b10],[@pbio-0030176-b25]\]. Because ciprofloxacin may induce repair pathways that involve RecA-ssDNA filament formation, the drug itself may act to induce the mutations that confer resistance. This hypothesis is consistent with observations that quinolones can be mutagenic and can induce the SOS response \[[@pbio-0030176-b17]\]. Here, using an in vivo infection model, we show that interfering with LexA autoproteolysis renders pathogenic Escherichia coli unable to mutate and acquire resistance to ciprofloxacin. The same result is demonstrated in vivo for the antibiotic rifampicin, which is a semisynthetic member of the rifamycin class of antibiotics that inhibits the bacterial RNA polymerase. The in vivo results are recapitulated in vitro for ciprofloxacin (here, we use in vitro to refer to bacteria in liquid cultures or on solid media). To understand how LexA cleavage is induced by ciprofloxacin and how this cleavage mediates resistance, we examined the contribution of a variety of different genes to ciprofloxacin tolerance and resistance in vitro. We show that E. coli repairs ciprofloxacin-induced DNA damage primarily by RecBC-dependent HR, and to a lesser extent by illegitimate recombination (IR). Finally, we show that Pol II, Pol IV, and Pol V are all required for the evolution of resistance. Results {#s2} ======= LexA Cleavage Is Required for the Evolution of Resistance to Ciprofloxacin and Rifampicin In Vivo {#s2a} ------------------------------------------------------------------------------------------------- To test whether activation of the SOS response is required to induce the mutations that confer ciprofloxacin resistance during therapy, we used a neutropenic murine thigh infection model \[[@pbio-0030176-b29]\] and the pathogenic E. coli strain ATCC 25922. Mice were infected with 10^6^ colony-forming units (cfu) of either the ΔlacZ (control) or lexA(S119A) 25922 strain ([Table 1](#pbio-0030176-t001){ref-type="table"}). The LexA(S119A) mutant cannot undergo autoproteolysis and thus renders bacteria unable to de-repress the SOS genes \[[@pbio-0030176-b30]\]. After infection, ciprofloxacin (0.5 mg/kg) was administered every 12 h. After 24, 48, and 72 h, mice were sacrificed, thigh tissue was homogenized, and dilutions of the homogenate were plated on media with and without ciprofloxacin to quantify the number of viable and resistant cfu, respectively ([Figure 1](#pbio-0030176-g001){ref-type="fig"}A). Treatment with ciprofloxacin had the same, slow bactericidal effect on both strains, reducing the total cfu/thigh approximately 250-fold over the 72 h of treatment. In mice infected with the ΔlacZ control strain, significant resistance emerged, amounting to approximately 3% of the entire population by 72 h. Individual clones were found to have ciprofloxacin minimum inhibitory concentrations (MICs) as high as 640 ng/ml (ten clones analyzed), significantly increased from the 12 ng/ml MIC observed for the ΔlacZ strain before infection. In dramatic contrast, no resistant mutants were isolated from mice infected with the lexA(S119A) strain. The MICs of the lexA(S119A) clones recovered from the mouse thigh and isolated from the ciprofloxacin-free media remained approximately equal to that of the strain prior to infection (approximately 12 ng/ml, ten clones analyzed). ::: {#pbio-0030176-g001 .fig} Figure 1 ::: {.caption} ###### Survival and Mutation of E. coli Mutants In Vivo after Starting Antibiotic Therapy Survival and mutation of ΔlacZ and lexA(S119A) mutants of E. coli ATCC 25922 in thighs of neutropenic mice at 24-h intervals after starting therapy with (A) ciprofoxacin or (B) rifampicin. Open circles and triangles correspond to the total cfu/thigh of the ΔlacZ and lexA(S119A) strains, respectively. Solid circles and triangles represent the number of drug-resistant ΔlacZ and lexA(S119A) cfu/thigh, respectively. ::: ![](pbio.0030176.g001) ::: ::: {#pbio-0030176-t001 .table-wrap} Table 1 ::: {.caption} ###### Strains Used in This Work ::: ![](pbio.0030176.t001) ::: We next used the same mouse model to characterize the evolution of resistance to rifampicin in vivo. Again, mice were infected with 10^6^ cfu of either the ΔlacZ control or the lexA(S119A) 25922 strain. Rifampicin (100 mg/kg) was administered every 12 h, mice were sacrificed after 24, 48, and 72 h, and the total number of viable and rifampicin-resistant cfu was determined. Treatment with rifampicin also had a slow bactericidal effect on both strains ([Figure 1](#pbio-0030176-g001){ref-type="fig"}B); however, by 72 h all of the ΔlacZ clones recovered from the thigh were resistant to rifampicin. MICs of individual rifampicin-resistant ΔlacZ clones were all higher than 250 μg/ml (ten clones analyzed), which is substantially higher than the 8 μg/ml MIC of the parent strain. Remarkably, despite the rapid acquisition of resistance in the control strain, no resistant clones were isolated from mice infected with the lexA(S119A) strain. The MICs of the lexA(S119A) clones recovered from the mouse thigh remained 8 μg/ml, approximately the same as the parent strain (ten clones analyzed). We conclude that LexA cleavage is absolutely required for the evolution of resistance to both ciprofloxacin and rifampicin during therapy in vivo. Ciprofloxacin Induces Biochemical Pathways That Facilitate Mutation In Vitro {#s2b} ---------------------------------------------------------------------------- In order to further characterize the genetic requirements of these resistance-conferring mutations, we studied bacteria in vitro. We constructed an isogenic series of mutants in the E. coli strain MG1655 (MG1655 is not pathogenic, but it has been fully sequenced \[[@pbio-0030176-b31]\], which facilitated the construction of the required deletion strains) ([Table 1](#pbio-0030176-t001){ref-type="table"}). As with the in vivo studies, a ΔlacZ strain was constructed as a control (the ΔlacZ strain exhibited wild-type growth and mutation). We first determined that the ciprofloxacin MIC for the ΔlacZ control strain is 35 ng/ml in liquid media ([Table 2](#pbio-0030176-t002){ref-type="table"}). On solid media, we found that 40 ng/ml ciprofloxacin killed 99% of the cells within 24 h of plating, while the remaining 1% of the population persisted for several weeks, allowing for the characterization of the bacteria in the presence of the antibiotic ([Figure 2](#pbio-0030176-g002){ref-type="fig"}). ::: {#pbio-0030176-g002 .fig} Figure 2 ::: {.caption} ###### Survival of E. coli Mutants In Vitro Survival on solid media containing 40 ng/ml ciprofloxacin of ΔlacZ control and (A) NER and recombination mutants with wild-type sensitivity, (B) recombination mutants that were hypersensitive to ciprofloxacin, and (C) lexA(S119A) and inducible polymerase mutants. ::: ![](pbio.0030176.g002) ::: ::: {#pbio-0030176-t002 .table-wrap} Table 2 ::: {.caption} ###### Growth and Ciprofloxacin Sensitivity of MG1655-Derived Strains ::: ![](pbio.0030176.t002) ^a^ Complete genotypes are given in [Table 1](#pbio-0030176-t001){ref-type="table"}. ^b^ Not determined. ^c^ Mutants were not viable. ^d^ No mutants isolated. ^e^ Made by P1 transduction. ^f^ Mutant grew too slowly to accurately determine doubling time. ::: We measured the rate at which the ΔlacZ control strain evolves resistance in vitro on solid media containing 40 ng/ml ciprofloxacin; at this concentration, single mutations (typically in gyrA) are sufficient to confer resistance ([Tables 2](#pbio-0030176-t002){ref-type="table"} [and S1](#sg001){ref-type="supplementary-material"}) \[[@pbio-0030176-b01]\]. ΔlacZ cells were grown in permissive liquid culture and then plated onto ciprofloxacin-containing media. Resistant colonies were counted as they arose, in 24 h intervals over 14 d. Colonies that formed immediately (at or before day 2) were attributed to cells that had acquired resistance prior to exposure to ciprofloxacin (via "pre-exposure" mutations), while colonies that formed on day 3 or later were attributed to cells that acquired resistance after exposure to ciprofloxacin (via "post-exposure" mutations). Assignments of pre- and post-exposure mutations were validated by using two different reconstruction assays designed to confirm when the mutations occurred (see [Materials and Methods](#s4){ref-type="sec"}). Mutation rates were defined as the number of resistant colonies that arose per time, per viable cell (it should be noted that this experiment detects only those mutations that confer resistance). We observed a pre-exposure mutation rate of 9.0 (± 9.5) × 10^−10^ mutants/viable cell/d, and a post-exposure mutation rate of 1.8 (± 0.69) × 10^−5^ mutants/viable cell/d ([Table 3](#pbio-0030176-t003){ref-type="table"}). Thus, ciprofloxacin induces resistance by a factor of 10^4^. These rates are in agreement with those reported previously by Hall using a similar in vitro assay \[[@pbio-0030176-b32]\]. ::: {#pbio-0030176-t003 .table-wrap} Table 3 ::: {.caption} ###### Mutation Spectra and Rates ::: ![](pbio.0030176.t003) In parentheses is the ratio of the number of events observed to the total number mutants sequenced. nd, not determined. ::: We next examined the mutation spectrum of the gyrA gene in the resistant clones by sequencing a 1,000-nt region encompassing the quinolone resistance-determining region \[[@pbio-0030176-b33]\]. We found that the spectrum of the post-exposure mutations differed significantly from the pre-exposure mutations ([Table 3](#pbio-0030176-t003){ref-type="table"}). In the pre-exposure mutants (ten clones sequenced) we observed strictly substitution mutations, while in the post-exposure mutants (19 clones sequenced) we observed both substitutions and a three-basepair, in-frame deletion that removed the codon for Ser83. This "codon deletion" occurred despite a lack of flanking direct repeats or palindromic sequences, which are often thought to facilitate spontaneous deletions. LexA Cleavage Is Required for the Acquisition of Resistance In Vitro {#s2c} -------------------------------------------------------------------- To characterize the role of LexA cleavage and the SOS response in the evolution of resistance in vitro, we constructed a lexA(S119A) MG1655 strain. Mutation of LexA did not affect growth or persistence in the presence of 40 ng/ml ciprofloxacin ([Figure 2](#pbio-0030176-g002){ref-type="fig"}C and [Table 2](#pbio-0030176-t002){ref-type="table"}). Previously, another noncleavable LexA mutant, lexA(G85D), was shown to be moderately sensitive to high levels of ciprofloxacin (6× MIC) \[[@pbio-0030176-b34],[@pbio-0030176-b35]\]. The difference in sensitivities associated with these two noncleavable LexA mutants is likely the result of different levels of ciprofloxacin-mediated DNA damage, i.e., an increased reliance on the induction of SOS genes under the more damaging conditions employed in the lexA(G85D) study. We observed the same in vitro pre-exposure mutation rate and spectrum for the lexA(S119A) strain as we did for the control strain ([Table 3](#pbio-0030176-t003){ref-type="table"}). However, the lexA(S119A) strain exhibited a post-exposure mutation rate that was approximately 100-fold lower than that observed for the ΔlacZ control strain ([Table 3](#pbio-0030176-t003){ref-type="table"}). The decreased evolution of resistance did not depend on the concentration of the antibiotic, as virtually identical results were obtained with 60 ng/ml of ciprofloxacin (unpublished data). Given the low mutation rate, only three post-exposure ciprofloxacin-resistant lexA(S119A) mutants were isolated (from more than 10^11^ bacteria plated, overall), but all three acquired resistance by deletion of the Ser83 codon and not by substitution mutation ([Table 3](#pbio-0030176-t003){ref-type="table"}). Similar results were observed in vitro with the ΔlacZ and lexA(S119A) 25922 strains used in the in vivo studies, both in terms of mutation rates ([Figure S1](#sg001){ref-type="supplementary-material"}) and mutation spectrum. To ensure that the reduced ability to evolve resistance observed with the lexA(S119A) strain was not simply because these cells grew slowly with gyrase mutations, we examined the lexA(S119A) gyrA double mutants ([Figure S2](#sg002){ref-type="supplementary-material"} and [Table 2](#pbio-0030176-t002){ref-type="table"}). Relative to the respective control strain, each double mutant showed virtually identical growth in the absence of ciprofloxacin, virtually identical persistence in the presence of ciprofloxacin, and identical ciprofloxacin MICs. In all, the data indicate that the decreased evolution of resistance in the lexA mutant did not result from decreased persistence or the inability to grow upon acquisition of a gyr mutation, but rather from an inability to induce mutations. The evolution of clinically significant resistance requires the stepwise accumulation of several mutations \[[@pbio-0030176-b01]\]. To test whether LexA cleavage is required for these additional mutations, we examined the in vitro evolution of resistance to 650 ng/ml ciprofloxacin in ΔlacZ control and lexA(S119A) MG1655 strains already containing the prototypical "first step" Ser83Leu mutation in gyrA that confers resistance to 40 ng/ml ciprofloxacin (strains RTC0114 and RTC0122; see [Table 1](#pbio-0030176-t001){ref-type="table"}). The "second step" mutation rate was 1.9 (± 0.21) × 10^−4^ mutants/viable cell/d in the control strain and 5.5 (± 4.9) × 10^−7^ mutants/viable cell/d in the lexA(S119A) strain ([Figure S3](#sg003){ref-type="supplementary-material"}). Assuming that the first and second step mutations are independent, the LexA mutant strain evolves resistance to 650 ng/ml ciprofloxacin in vitro with a rate that is approximately 10^4^-fold lower than the control strain. Because clinical resistance typically requires three to five independent mutations \[[@pbio-0030176-b01]\], the data imply that in the absence of efficient LexA cleavage, E. coli would evolve clinical resistance at least 10^6^-fold slower. These in vitro results fully recapitulate the in vivo mouse model studies and demonstrate that LexA cleavage-mediated derepression of one or more genes is essential for the efficient evolution of resistance. RecBC-Mediated Homologous Recombination Likely Induces LexA Cleavage in the Presence of Ciprofloxacin {#s2d} ----------------------------------------------------------------------------------------------------- To understand how the presence of ciprofloxacin leads to LexA cleavage, we characterized a series of MG1655 strains harboring deletions of genes involved in various DNA repair pathways ([Tables 1](#pbio-0030176-t002){ref-type="table"} and [and S2](#sg002){ref-type="supplementary-material"}). We reasoned that a pathway important for mediating ciprofloxacin-induced LexA cleavage would involve the formation of RecA-ssDNA filaments and also would be induced by the drug. Because the ciprofloxacin-topoisomerase complexes act to cross-link DNA, repair may occur by NER, mediated by UvrA, UvrB, and UvrC, followed by recombinational gap repair, mediated by RecF, RecO, RecR, and RecA \[[@pbio-0030176-b36]\]. We tested whether this classical cross-link repair pathway is induced by the antibiotic by examining the sensitivity and the pre- and post-exposure mutation rates for ΔuvrB, ΔrecF, ΔrecO, and ΔrecR strains. No differences in sensitivity (see [Figure 2](#pbio-0030176-g002){ref-type="fig"}A and [Table 2](#pbio-0030176-t002){ref-type="table"}) or mutation rate ([Table 3](#pbio-0030176-t003){ref-type="table"}) were observed compared to the control strain, suggesting that NER and RecFOR-mediated recombinational gap repair are not induced in the presence of ciprofloxacin or required for the efficient evolution of resistance. A previous study of uvrB and recF deletion strains suggested that they were more sensitive to nalidixic acid than was the wild-type strain \[[@pbio-0030176-b37]\]. However, unlike the current study, the previous study involved short incubations at high drug concentrations. Thus, the different experimental conditions are likely responsible for the different results. Because topoisomerase dissociation from the DNA may generate free DSBs, repair may occur by a combination of HR and DNA replication in a process generally referred to as recombination-dependent DNA replication (RDR) \[[@pbio-0030176-b40],[@pbio-0030176-b41]\]. In E. coli, RDR is mediated by RecA filamentation on ssDNA, created by the helicase/nuclease RecBCD, and PriA-dependent replisome assembly. To test whether the RecBCD pathway of HR is induced in the presence of ciprofloxacin, we examined the ΔrecA, ΔrecB, and ΔrecD strains. Although the ΔrecA and ΔrecB strains exhibited virtually wild-type viability, both were hypersensitive to ciprofloxacin (see [Figure 2](#pbio-0030176-g002){ref-type="fig"}B and [Table 2](#pbio-0030176-t002){ref-type="table"}). No pre- or post-exposure mutants were observed in either strain. To investigate whether resistant colonies were not isolated because the gyrA(S83L) ΔrecA and gyrA(S83L) ΔrecB double mutants are not viable, we attempted to construct these strains by P1 transduction of the rec deletion into a gyrA(S83L) strain. No viable double mutants could be constructed in either case. This is consistent with previous reports that certain mutant gyrase proteins cause increased spontaneous fork collapse, which makes the cells dependent on HR \[[@pbio-0030176-b42]\]. Thus, we propose that RecA/RecBC-mediated recombination is important for cell viability in the presence of ciprofloxacin both before and after the acquisition of resistance-conferring mutations. In contrast, deletion of recD had no effect on drug sensitivity (see [Figure 2](#pbio-0030176-g002){ref-type="fig"}A and [Table 2](#pbio-0030176-t002){ref-type="table"}), mutation rate, or mutation spectrum ([Table 3](#pbio-0030176-t003){ref-type="table"}). This result is consistent with the fact that RecBC can process DSEs and load RecA onto ssDNA in the absence of the RecD helicase \[[@pbio-0030176-b46]\]. Because persistent covalently-linked topoisomerase-DNA complexes will eventually block the progression of replication forks, repair may occur by recombination-dependent fork repair, which is a variant of RDR that reestablishes a processive replication fork \[[@pbio-0030176-b22],[@pbio-0030176-b41],[@pbio-0030176-b47]\]. In addition to the proteins that mediate HR (i.e., RecA and RecBC) this process appears to require RecG and RuvABC. To test whether recombination-dependent fork repair is induced in the presence of ciprofloxacin, we examined the ΔrecG, ΔruvB, and ΔruvC strains. Deletion of recG, ruvB, or ruvC did not cause a significant decrease in viability in the absence of ciprofloxacin, but did result in high sensitivity to the drug, although not as high as that observed with deletion of recA and recB (see [Figure 2](#pbio-0030176-g002){ref-type="fig"}B and [Table 2](#pbio-0030176-t002){ref-type="table"}). Like the ΔrecA and ΔrecB strains, no pre- or post-exposure mutants were isolated. However, we were able to delete recG, ruvB, and ruvC in a gyrA(S83L) strain by P1 transduction. These double mutants grew as well as the corresponding single mutants (see [Table 2](#pbio-0030176-t002){ref-type="table"}). While the double mutants each displayed significantly reduced MICs relative to the gyrA(S83L) single mutant (see [Table 2](#pbio-0030176-t002){ref-type="table"}), they were still able to form colonies on media containing 40 ng/ml ciprofloxacin. Thus, selection against resistance-conferring mutants in the recG, ruvB, and ruvC backgrounds does not explain the absence of resistant colonies. These results suggest that the functions of RecG, RuvB, and RuvC are required for viability in the presence of ciprofloxacin (although less so than RecA and RecB) and also for the acquisition of resistance-conferring mutations. Because PriA-dependent reinitiation of DNA synthesis is important for a variety of HR models \[[@pbio-0030176-b49],[@pbio-0030176-b51],[@pbio-0030176-b52]\], it may be required for the repair of ciprofloxacin-mediated DNA damage. Deletion of priA resulted in extreme sensitivity to ciprofloxacin (MIC of less than 1 ng/ml), implying that replication restart is essential in response to the drug (see [Figure 2](#pbio-0030176-g002){ref-type="fig"}B and [Table 2](#pbio-0030176-t002){ref-type="table"}). As with the rec and ruv strains, no mutants were isolated before or after exposure to ciprofloxacin, and as with the ΔrecA and ΔrecB strains, we were unable to construct the gyrA(S83L) ΔpriA double mutant. These results imply that replication restart is essential in response to ciprofloxacin and in tolerating the resistance-conferring gyrase mutations. Induction of the LexA-Repressed Polymerases Does Not Contribute to Survival but Is Critical for the Evolution of Ciprofloxacin Resistance In Vitro {#s2e} -------------------------------------------------------------------------------------------------------------------------------------------------- To characterize how LexA cleavage induces resistance, we examined the role of the three LexA-repressed polymerases, Pol II (polB), Pol IV (dinB), and Pol V (umuD and umuC) in survival and mutation in vitro. In every respect, the effects of deleting a single LexA-repressed polymerase (i.e., ΔpolB, ΔdinB, or ΔumuDC), or any combination of the three, were the same as the effects of preventing LexA cleavage. First, the pol deletion strains showed similar sensitivities to ciprofloxacin as the control strain ([Figure 2](#pbio-0030176-g002){ref-type="fig"}C and [Table 2](#pbio-0030176-t002){ref-type="table"}). (The ΔpolB, ΔpolB ΔdinB, ΔpolB ΔumuDC, and ΔpolB ΔdinB ΔumuDC strains were all slightly sensitive to ciprofloxacin, which is consistent with a role for replication restart in the response to the drug, as replication restart is thought to involve Pol II \[[@pbio-0030176-b01],[@pbio-0030176-b41],[@pbio-0030176-b53]\]). Second, the MIC of each gyrA pol mutant was virtually identical to that of the corresponding gyrA single mutant ([Table 2](#pbio-0030176-t002){ref-type="table"}). Third, the pre-exposure mutation rate and spectrum of the pol deletion strains were the same as the control; however, the post-exposure rates were markedly reduced ([Table 3](#pbio-0030176-t003){ref-type="table"}). Finally, resistance in the post-exposure mutants was acquired strictly through deletion of either Ser83 or Ala84, and not through substitution (29 clones sequenced; [Table 3](#pbio-0030176-t003){ref-type="table"}). These results suggest that it is through the derepression of all three repressed polymerases that LexA cleavage induces substitution mutations. Discussion {#s3} ========== The discovery and development of antibiotics revolutionized medicine, providing easy cures for previously untreatable diseases. However, for every significant infectious disease caused by bacteria, strains resistant to all available antibiotics have been reported \[[@pbio-0030176-b54]\]. We are interested in understanding how bacteria evolve resistance, and have first focused on the antibiotic ciprofloxacin. Ciprofloxacin and the other quinolones are perhaps the most important antibiotics currently available \[[@pbio-0030176-b16]\], partly because of the low levels of resistance currently observed with these newer synthetic drugs. However, clinical resistance to the quinolones is evolving at an alarming rate due to mutations in gyrase, topoisomerase IV, and efflux pumps or their regulators \[[@pbio-0030176-b01]\]. In this study we have shown, in vivo, that preventing LexA cleavage renders bacteria unable to evolve resistance to either ciprofloxacin or rifampicin in a mouse thigh infection model. In vitro, the ability of bacteria to induce mutation and evolve resistance to ciprofloxacin is also dramatically reduced by rendering LexA uncleavable. Thus, our results indicate that the mutations that confer resistance to ciprofloxacin and rifampicin are not simply the result of unavoidable errors accumulated during genome replication, but rather are induced via the derepression of genes whose protein products act to significantly increase mutation rates. In principle, part of the observed increase in mutation rate after exposure to ciprofloxacin could result from selection against resistant mutants during the pre-exposure growth in liquid media (thus underestimating the pre-exposure mutation rates). However, the impact of selection is unlikely to be significant, as gyrA mutations are tolerated without a significant increase in doubling time ([Table 2](#pbio-0030176-t002){ref-type="table"}). Further evidence that the resistance-conferring mutations are induced is provided by the fact that deletion or mutation of several genes, including lexA, renders cells unable to evolve significant levels of resistance. The data suggest that the increase in mutation rate is caused by recombination pathways that are induced to repair antibiotic-mediated DNA damage. A mechanism consistent with our results is illustrated in [Figure 3](#pbio-0030176-g003){ref-type="fig"}. Three recombinational repair pathways appear to be involved that are distinguished by the type of damage they repair and the type of mutation they induce. One pathway is IR-mediated repair of free DSBs where the protein has dissociated from the DNA (pathway A, [Figure 3](#pbio-0030176-g003){ref-type="fig"}). We suggest that this pathway is responsible for the observed codon deletions. The induction of small nucleotide deletions has been observed in bacteria \[[@pbio-0030176-b55]\], and in eukaryotes, it has been suggested that small deletions (including codon deletions) arise during IR after an inhibited topoisomerase aberrantly releases free DSBs, the ends of which are processed by exonucleases and polymerases before being rejoined \[[@pbio-0030176-b58]\]. While we may have observed a similar phenomenon in our in vitro studies, the codon deletion mutants are unlikely to be of much clinical significance on their own, as they have relatively low ciprofloxacin MICs ([Table 2](#pbio-0030176-t002){ref-type="table"}) and have never been observed in the clinic \[[@pbio-0030176-b01],[@pbio-0030176-b33]\]. Isolation of these mutants in the current study was most likely due to the permissive drug concentrations employed. ::: {#pbio-0030176-g003 .fig} Figure 3 ::: {.caption} ###### Proposed Response to Ciprofloxacin In the absence of homologous sequences, free DSBs are repaired by nuclease and polymerase-dependent IR (pathway A). In the presence of a suitable homologous sequence, free DSBs may be repaired by RDR (pathway B). This involves resectioning of the DNA ends by RecBC and loading of RecA onto the ssDNA produced. These RecA-ssDNA filaments catalyze D-loop formation and repair of the DSB. This pathway may also contribute to the repair of replication forks when they encounter the free DSB. Finally, replication forks that encounter topoisomerases that are covalently-bound to the DNA are repaired by recombination-dependent fork repair (pathway C). This involves RecG-mediated fork regression and RuvC cleavage to produce DSEs where RecBC mediates RecA-ssDNA filament formation. These filaments catalyze strand invasion of a homologous sequence where PriA, and possibly Pol II, help to reestablish a processive replication fork. With sufficient accumulation of DSBs and collapsed forks, persistent RecA-ssDNA filaments induce levels of LexA cleavage sufficient to de-repress the error prone polymerases, Pol IV and Pol V, which cooperate to induce mutations (pathway D). Once resistance-conferring mutations are made, DSBs and collapsed forks cease to accumulate and RecA-filaments no longer persist. Subsequently, the cellular concentration of LexA increases, shutting down expression of the pro-mutagenic polymerases. See text for details. ::: ![](pbio.0030176.g003) ::: In addition to IR, RDR and replication fork repair are also induced to repair DSBs in cases where the topoisomerase has dissociated from, or remains bound to, the DNA, respectively (pathways B and C, [Figure 3](#pbio-0030176-g003){ref-type="fig"}). These pathways may be more relevant to clinical resistance as they induce the substitution mutations ubiquitously found in clinically resistant strains. In both pathways, the RecBCD nuclease/helicase loads at DSEs generated (directly or indirectly) by ciprofloxacin and simultaneously degrades and unwinds the duplex while loading RecA onto the ssDNA of the nascent 3′-overhang. (For replication fork repair, RecG and RuvABC are required to prepare the DSE \[[@pbio-0030176-b50],[@pbio-0030176-b59],[@pbio-0030176-b60]\].) RecA forms filaments that promote strand invasion of the ssDNA into a homologous sequence, resulting in the formation of an intermediate known as a displacement-loop structure (D-loop). The invading strand may then prime DNA synthesis, using the homologous sequence as a template, ultimately restoring the genetic information disrupted by the DSE \[[@pbio-0030176-b48]\]. In the case of replication fork repair, the covalently bound topoisomerase must still be displaced from DNA in order to reinitiate processive synthesis. We propose that the topoisomerase is displaced by RuvAB, which has recently been shown to branch migrate D-loop-like structures and simultaneously displace covalently-bound ciprofloxacin-topoisomerase IV complexes \[[@pbio-0030176-b61]\], or perhaps by Rep or UvrD helicase, which have both been shown to displace bound proteins from duplex DNA \[[@pbio-0030176-b62]\]. Following PriA recognition and binding, a processive replication fork is reestablished. However, with the continued presence of ciprofloxacin these processes will continue, resulting in the persistence of the RecA-ssDNA filaments, which eventually degrade enough LexA to derepress the error-prone, SOS-regulated polymerases (pathway D, [Figure 3](#pbio-0030176-g003){ref-type="fig"}). The data suggest that the induction of substitution mutations requires the derepression of all three SOS-regulated polymerases, Pol II, Pol IV, and Pol V. While the evolution of resistance to ciprofloxacin by substitution mutation is to our knowledge the first process found to require all three of the E. coli inducible polymerases, this observation is consistent with previous studies showing that multiple polymerases are required for some mutations \[[@pbio-0030176-b09],[@pbio-0030176-b41],[@pbio-0030176-b63],[@pbio-0030176-b64]\]. It is also consistent with the two-step model of translesion synthesis, wherein one specialized polymerase is required for dNTP misinsertion and another for continued synthesis (mispair extension) \[[@pbio-0030176-b09],[@pbio-0030176-b27],[@pbio-0030176-b65]\]. We propose that the induced mutations conferring antibiotic resistance in vitro and in vivo are the result of Pol V mispair synthesis \[[@pbio-0030176-b66],[@pbio-0030176-b67]\] and Pol IV mispair extension \[[@pbio-0030176-b10],[@pbio-0030176-b68]\], while Pol II may be required to initiate replication restart \[[@pbio-0030176-b41], [@pbio-0030176-b53]\] (after which it may be replaced by Pol V and then Pol IV), or to fix the nascent mutation by extending the primer terminus sufficiently to avoid exonucleolytic proofreading upon reloading of Pol III \[[@pbio-0030176-b69]\]. This process continues until mutations are made that allow for the resumption of normal DNA synthesis. The key signal that links the cellular response to the antibiotic with the evolution of resistance appears to be the RecA-ssDNA filaments that are formed to facilitate the repair of antibiotic-mediated DNA damage. These RecA-ssDNA filaments also induce LexA cleavage and derepression of the mutagenic polymerases. We suggest that a similar mechanism might also serve to induce mutation and evolution in response to other antibiotics, or other forms of cellular stress, where DNA damage per se is not involved. For example, the ratio of ATP to ADP determines the level of supercoiling in the bacterial genome \[[@pbio-0030176-b17]\], and both increased and decreased levels of supercoiling inhibit replication fork progression \[[@pbio-0030176-b70]\]. Thus, different stresses that perturb metabolism (i.e., alter ATP/ADP ratios) might also alter DNA topology and result in stalled replisomes; recombination-based rescue and RecA-ssDNA filament formation; and the induction of mutations required to reestablish a normal cellular environment. Interestingly, it has recently been shown that β-lactams can induce the SOS response via a two-component signal transduction system \[[@pbio-0030176-b71]\]. The traditional paradigms of DNA replication and mutation suggest that resistance-conferring mutations are the inevitable consequence of polymerase errors, and offer no obvious means for intervention. In stark contrast, the model described above suggests that bacteria play an active role in the mutation of their own genomes by inducing the production of proteins that facilitate mutation, including Pol IV and Pol V, as has been suggested with other forms of mutation \[[@pbio-0030176-b07]\]. In turn, this suggests that inhibition of these proteins, or the prevention of their derepression by inhibition of LexA cleavage, with suitably designed drugs, might represent a fundamentally new approach to combating the emerging threat of antibiotic-resistant bacteria. Future efforts will focus on determining the generality of the observations, in terms of both other pathogenic bacteria and other antibiotics. Materials and Methods {#s4} ===================== {#s4a} ### Bacterial strains and growth {#s4a1} Strains of E. coli used in this study are listed in [Table 1](#pbio-0030176-t001){ref-type="table"}. Solid media was Lennox LB \[[@pbio-0030176-b72]\] plus 1.6% agar; liquid media was Miller LB \[[@pbio-0030176-b72]\]. For selection in MG1655- and ATCC 25922-derived strains, antibiotics were used as follows: kanamycin, 30 and 50 μg/ml; spectinomycin, 100 μg/ml; and chloramphenicol, 20 μg/ml. All bacteria were grown at 37 °C unless otherwise indicated. Ciprofloxacin and rifampicin were purchased from MP Biomedicals (Aurora, Ohio, United States). For strain construction and additional experimental details, see [Protocol S1](#sd001){ref-type="supplementary-material"}. A standard mouse infection model was employed \[[@pbio-0030176-b29]\]. Briefly, 6-wk-old, specific-pathogen-free, female CD-1 mice (weight, 23--27 g; Harlan Sprague Dawley) were used. Mice were rendered neutropenic (neutrophil counts less than 100/mm^3^) by intraperitoneal injection with 150 mg/kg cylcophosphamide (Mead Johnson Pharmaceuticals, Evansville, Indiana, United States) 4 d before infection and 100 mg/kg cyclophosphamide 24 h before infection. Previous studies have shown that this regimen produces neutropenia in this model for 5 d \[[@pbio-0030176-b73]\]. Mueller-Hinton (MH) (Difco) broth cultures inoculated from freshly plated bacteria were grown to logarithmic phase (OD~580~ of approximately 0.3), and diluted 1:10 in MH broth. Thigh infections were produced by injecting 0.1 ml volumes (approximately 10^6^ cfu) of the diluted broth cultures into halothane-anesthetized mice. Starting 2 h after infection (defined as time zero), mice were administered subcutaneous injections of either 0.5 mg/kg ciprofloxacin or 100 mg/kg rifampicin every 12 h for 3 d. After 24, 48, and 72 h, both thighs from two sacrificed animals were removed and homogenized. Serial dilutions of homogenates of each thigh were plated on MH agar (MHA) and MHA containing either 80 ng/ml ciprofloxacin or 128 μg/ml rifampicin (lower limit of detection was 100 organisms/thigh). MICs for both ciprofloxacin and rifampicin were determined by standard microdilution methods of the National Committee for Clinical Laboratory Standards. MICs prior to drug exposure for both the LexA mutant and control strains were determined by examining ten clones isolated from MHA plates. MICs of the post-exposure isolates from both the LexA mutant and control strains were determined by examining ten clones isolated from the MHA plates at the 72-h time point. ### Mutation assay. {#s4a2} For each strain, five independent cultures were grown for 25 h without ciprofloxacin. Viable cell counts in these cultures were determined by plating serial dilutions onto permissive media. For assaying mutation in MG1655-derived strains, 150 μl from each culture (approximately 10^8^ cells) was plated in duplicate on LB/agar containing 40 ng/ml ciprofloxacin. Five additional 150 μl aliquots from two cultures of each strain were also plated on the same media for use in the "survival" assay (see below). ATCC 25922-derived strains were assayed similarly with 12 ng/ml ciprofloxacin, except cultures were concentrated approximately 3-fold before plating to assay the same number of cells as in the MG1655 experiments. At 24 h intervals, visible colonies were counted, their location on the plate was marked, and they were stocked at −80 °C for later use in the reconstruction assay (see below). ### Survival assay. {#s4a3} Every 24 h, in parallel with the mutation assay, all visible colonies were excised from plates designated for assaying survival (see above), the remaining agar was homogenized in saline, and dilutions were plated in duplicate on LB/agar to determine the total number of viable, ciprofloxacin-sensitive cells present as a function of time, and LB/agar containing 40 ng/ml ciprofloxacin to determine if any ciprofloxacin-resistant colonies remained after excision. An experimental validation of this method is described in [Protocol S1](#sd001){ref-type="supplementary-material"} and [Table S3](#st003){ref-type="supplementary-material"}. ### Reconstruction assay. {#s4a4} We determined whether colonies isolated after plating on ciprofloxacin formed as a result of post-exposure mutation or as a result of mutation prior to exposure to the drug. Liquid cultures of permissive media were inoculated with ciprofloxacin-resistant clones stocked during the mutation assay (see above) and grown to saturation overnight. Cultures were diluted and plated in duplicate on both LB/agar, to confirm viability, and LB/agar containing 40 ng/ml ciprofloxacin, to confirm resistance. Clones that were resistant before exposure were defined as those that formed colonies on the ciprofloxacin-containing media in the same number of days in the reconstruction assay as they did in the original mutation assay. Conversely, clones that mutated after exposure to ciprofloxacin were defined as those that formed colonies at least 2 d faster in the reconstruction assay. To further confirm our assignment of the pre- and post-exposure mutants, an alternative method was also employed. Ciprofloxacin-resistant clones isolated during the mutation assay were suspended in 1 ml of 9 mg/ml NaCl. A series of dilutions was plated onto permissive media (LB/agar) and plates were incubated at 37 °C for 24--48 h. From each set of dilution plates, a plate was chosen that contained approximately 50--300 colonies. This plate was replica-plated onto permissive and nonpermissive media (LB/agar with 40 ng/ml ciprofloxacin), and replica plates were incubated at 37 °C. Replica plates were analyzed at 24 h intervals for the appearance of colonies ([Figure S4](#sg004){ref-type="supplementary-material"}). At the same time, the ciprofloxacin-resistant clones were also assayed using the reconstruction method described above. The two reconstruction assays gave identical results. ### Calculation of the rate at which cells become resistant to ciprofloxacin (mutation rate). {#s4a5} Mutation rate was defined as the number of ciprofloxacin-resistant mutants per viable cell that evolve as a function of time. We emphasize that it reflects only those mutations that both allow cells to survive and confer resistance to the drug. The mutation rate before exposure to ciprofloxacin (pre-exposure rate) was determined by fluctuation analysis and application of the p~0~ method \[[@pbio-0030176-b74]\]. The mutations after exposure to ciprofloxacin (post-exposure rate) exhibited the expected Poisson distribution \[[@pbio-0030176-b07]\] and the associated rate was therefore determined as the ratio of colonies on a particular day to the number of cells present at the time the cells became resistant, which we approximated as the viable cell count 2 d prior. The assumption that a colony takes 2 d to form accounts for both the actual time required for colony growth and the time required to turn over any residual ciprofloxacin-sensitive protein, i.e., phenotypic lag. After determining the post-exposure mutation rate for each day from days 3--13, rates were averaged over days 3--4, 5--8, and 9--13. Error bars reflect standard deviation in rate determinations from at least three independent sets of experiments. Supporting Information {#s5} ====================== Figure S1 ::: {.caption} ###### In Vitro Mutation Rate of ATCC 25922-Derived Strains Mutation rate of ATCC 25922 (1), ATCC 25922-ΔlacZ (2), and ATCC 25922-lexA(S119A) (3). Bars represent total mutation rate (base substitution and codon deletion). (59 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S2 ::: {.caption} ###### gyrA Mutant Growth on Ciprofloxacin Is Not Affected by Genetic Background Growth of gyrA(S83L) mutant clones in the (A) ΔlacZ, (B) ΔpolB ΔdinB ΔumuDC, and (C) lexA(S119A) genetic backgrounds on LB/agar containing 40 ng/ml ciprofloxacin. Plates were photographed after incubation at 37 °C for 24 h. (945 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S3 ::: {.caption} ###### Stepwise Mutation Rate Stepwise mutation rate of ΔlacZ, gyrA(S83L) (1) and lexA(S119A), gyrA(S83L) (2) to 650 ng/ml ciprofloxacin. (48 KB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S4 ::: {.caption} ###### Reconstruction of Mutants by Replica Plating Re-growth of eight ΔlacZ ciprofloxacin-resistant colonies isolated during the mutation assay on (A) Day 5, (B and C) Day 6, (D and E) Day 7, (F) Day 8, (G) Day 9, and (H) Day 10, 48 h after replica-plating onto LB/agar (left) and LB/agar containing 40 ng/ml ciprofloxacin (right). This reconstruction method demonstrates that all of the cells from a colony isolated during the mutation assay are ciprofloxacin resistant. See [Materials and Methods](#s4){ref-type="sec"} for details. (2.4 MB TIF). ::: ::: {.caption} ###### Click here for additional data file. ::: Protocol S1 ::: {.caption} ###### Supplementary Methods (195 KB DOC). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S1 ::: {.caption} ###### Effect of Different Ciprofloxacin-Resistance Mutations (28 KB DOC). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S2 ::: {.caption} ###### Primers Used in This Work (72 KB DOC). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S3 ::: {.caption} ###### Validation of Survival Assay (42 KB DOC). ::: ::: {.caption} ###### Click here for additional data file. ::: Accession Numbers {#s5a9} ----------------- The TIGR (<http://www.tigr.org>) Locus Names for the genes discussed in this paper are dinB (NT01EC0267), gyrA (NT01EC2662), lacZ (NT01EC0408), lexA(NT01EC4829), polB (NT01EC0071), priA (NT01EC4700), recA (NT01EC3205), recB (NT01EC3353), recD (NT01EC3352), recF (NT01EC4430), recG (NT01EC4373), recO (NT01EC3056), recR (NT01EC0568), ruvB (NT01EC2229), ruvC (NT01EC2233), umuC (NT01EC1406), umuD (NT01EC1405), and uvrB (NT01EC0934). We thank S. Rosenberg and A. Segall for helpful discussion. This research was supported by the Office of Naval Research (Award No. N00014--03-1--0126). Competing interests. The authors have declared that no competing interests exist. Author contributions. RTC, JKC, and FER conceived and designed the experiments. RTC, DRA, and WAC performed the experiments. RTC, JKC, DRA, WAC, and FER analyzed the data. VdCL contributed reagents/materials/analysis tools. RTC, JKC, and FER wrote the paper. Citation: Cirz RT, Chin JK, Andes DR, de Crécy-Lagard V, Craig WA, et al. (2005) Inhibition of mutation and combating the evolution of antibiotic resistance. PLoS Biol 3(6): e176. cfu : colony-forming units DSB : double strand break DSE : double strand end HR : homologous recombination IR : illegitimate recombination MIC : minimum inhibitory concentration NER : nucleotide excision and repair RDR : recombination-dependent replication ssDNA : single-stranded DNA	PubMed Central	2024-06-05T03:55:56.400404	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088971/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e176", "authors": [ { "first": "Ryan T", "last": "Cirz" }, { "first": "Jodie K", "last": "Chin" }, { "first": "David R", "last": "Andes" }, { "first": "Valérie", "last": "de Crécy-Lagard" }, { "first": "William A", "last": "Craig" }, { "first": "Floyd E", "last": "Romesberg" } ] }
PMC1088972	Introduction {#s1} ============ Apoptosis plays a central role in animal development and tissue homeostasis. The essential machinery responsible for the execution of apoptosis is the caspases, a family of closely related cysteine proteases \[[@pbio-0030183-b01]--[@pbio-0030183-b06]\]. Caspases comprise two classes: the initiator caspases, such as caspase-8 and -9, and the effector caspases, such as caspases-3 and -7. Due to their deleterious roles, all caspases are synthesized as relatively inactive zymogens in the cell and, in response to apoptotic stimuli, undergo proteolytic activation. The activation of effector caspases is performed by the initiator caspases. The autoactivation of the initiator caspases is facilitated by other auxiliary factors. For example, the autoactivation of caspase-9 is mediated by the assembly of a heptameric complex involving Apaf-1 and cytochrome c, dubbed the apoptosome \[[@pbio-0030183-b07]--[@pbio-0030183-b09]\]. Once activated, caspase-9 cleaves and activates caspases-3 and -7. The molecular mechanism for the activation of the effector caspases has been elucidated \[[@pbio-0030183-b06]\]. The caspase zymogen, such as procaspase-7, exists as a homodimer in solution but exhibits poor catalytic activity because its active sites exist in an unproductive conformation. The activation cleavage of procaspase-7 allows the relocation of a surface loop from one monomer to critically support the active site of the adjacent monomer, hence allowing catalysis to proceed \[[@pbio-0030183-b10],[@pbio-0030183-b11]\]. In contrast, the mechanism for the activation of the initiator caspases, which generally involves the formation of an oligomeric complex \[[@pbio-0030183-b09]\], has remained elusive \[[@pbio-0030183-b12]\]. The prevailing hypothesis has been the "induced proximity" model, which summarized a set of elegant experiments \[[@pbio-0030183-b13]--[@pbio-0030183-b17]\] to state that caspase zymogens are autoprocessed once they are brought into proximity with each other \[[@pbio-0030183-b18]\]. However, this initial explanation of the induced proximity model contained little mechanistic information until it was enhanced by the proposition that the oligomeric complex (such as the apoptosome) serves to activate the monomeric initiator caspases through induced dimerization via their intrinsic dimerization interface \[[@pbio-0030183-b19]--[@pbio-0030183-b21]\]. Supporting this hypothesis, caspase-9 exists predominantly as a monomer in solution and exhibits a basal level of catalytic activity \[[@pbio-0030183-b19],[@pbio-0030183-b22]\]. Induced proximity, interpreted as dimerization-driven activation of caspases, is perceived as a dominant model of initiator caspase activation \[[@pbio-0030183-b19]--[@pbio-0030183-b21],[@pbio-0030183-b23]\]. Despite its popularity, the notion of dimerization-driven activation of initiator caspases is at odds with some well-known facts. For example, if the activation of the initiator caspases requires merely dimerization, why would the apoptosome exhibit a 7-fold symmetry? Along this line, the activation of caspase-8 requires the assembly of a multi-component death-inducing signaling complex, which has a 3-fold symmetry. The induced proximity model states that the dimerization of caspase-9, via its intrinsic dimerization interface, is the central and only step in its activation, and the assembly of the apoptosome serves merely to facilitate the dimerization of caspase-9 \[[@pbio-0030183-b19],[@pbio-0030183-b20],[@pbio-0030183-b23]\]. Thus, caspase-9 in its dimeric form is expected to exhibit the same catalytic activity as caspase-9 when it is activated by the apoptosome. An essential experiment in assessing the correctness of this prediction is to directly compare the catalytic activity of the dimeric caspase-9 with that of the apoptosome-activated caspase-9. Unfortunately, the wild-type (WT) caspase-9 dimer defies isolation, because it does not represent a stable state in solution \[[@pbio-0030183-b22]\]. The only approach is to engineer a constitutive caspase-9 homodimer that will dimerize via its intrinsic dimerization interface. In this report, we revisit the induced proximity model through the engineering of a constitutively dimeric caspase-9. We demonstrate that, despite the fact that this dimer variant exhibits a higher level of catalytic activity in vitro and induces more efficient cell death than the WT caspase-9, it is qualitatively different from the Apaf-1-activated caspase-9. Importantly, the structure of the engineered caspase-9 closely resembles that of the WT caspase-9, including all relevant structural details and the asymmetric nature of the two monomers. Our data, in conjunction with other evidence, suggest a refined model of the induced proximity hypothesis. At the center of this model is an induced conformation of the active site, hence the name induced conformation model. Results {#s2} ======= Rationale for the Design of a Dimeric Caspase-9 {#s2a} ----------------------------------------------- While all members of the caspase family share a similar overall structure, small differences in their primary sequences lead to their individuality \[[@pbio-0030183-b01],[@pbio-0030183-b04]\]. An effector caspase, such as caspase-3, exists exclusively as a stable homodimer in solution. In contrast, biochemical and structural analyses revealed that dimerization of the initiator caspase-9, although possible, was energetically unfavorable \[[@pbio-0030183-b01],[@pbio-0030183-b19],[@pbio-0030183-b22]\]. To facilitate the design of a constitutively dimeric caspase-9, we examined the dimerization interface of several representative caspases. In caspase-3, the dimerization interface is mediated primarily by two β-strands (β6 and β6′), one from each monomer ([Figure 1](#pbio-0030183-g001){ref-type="fig"}A). Together, 63% of the buried surface area at the dimeric interface is contributed by these two centrally located β-strands and their immediate flanking residues. Consistent with a critical role in dimerization, residues on these two β-strands exhibit relatively low temperature factors. Importantly, sequence alignment among four representative caspases revealed an extremely varied section of five consecutive amino acids on strand β6, flanked on either side by stretches of conserved residues ([Figure 1](#pbio-0030183-g001){ref-type="fig"}B). For example, the five residues Cys^264^-Ile-Val-Ser-Met^268^ in caspase-3 are replaced by Gly^402^-Cys-Phe-Asn-Phe^406^ in caspase-9, while five of the eight flanking residues are invariant between these two caspases. Compared to caspase-3, caspases-6 and -7, also known to exist as homodimers in solution, contain similarly conserved residues on the β6 strand ([Figure 1](#pbio-0030183-g001){ref-type="fig"}B). This observation further supports a critical role of the β6 strand in dimerization and suggests a conserved pattern required for dimerization. In agreement with this analysis, previous structural investigation showed that Phe^404^ and Phe^404^′, one of the five residues on strands β6 and β6′, respectively, appear to impede the dimerization of caspase-9 by way of their incompatible side-chain configuration at the interface ([Figure 1](#pbio-0030183-g001){ref-type="fig"}A) \[[@pbio-0030183-b19]\]. ::: {#pbio-0030183-g001 .fig} Figure 1 ::: {.caption} ###### The β6 Strand Is the Major Determinant for the Dimerization of Caspases \(A) Comparison of the dimerization interfaces of caspase-3 and caspase-9. Caspases-3 and -9 exist primarily as a dimer and a monomer, respectively, in solution. However, inhibitor-bound caspase-9 was crystallized in its dimeric form \[[@pbio-0030183-b19]\]. The overall structures of caspases-3 and -9 are similar. A close-up view of the dimerization interfaces reveals sharp variation of residues on the β6 strand between caspase-3 and -9, which likely contributes to their different propensity for dimer formation. For caspase-9, Phe^404^ on strand β6 appears to impede dimerization as it sterically clashes with Phe^404^′ of the adjacent monomer \[[@pbio-0030183-b19]\]. \(B) A sequence alignment of the residues on and surrounding the strand β6 in four representative caspases. Similar to caspase-3, caspases-6 and -7 also exist as homodimers in solution. Residues on β6 are highlighted in blue and orange for the monomeric and dimeric caspases, respectively. Note that the variation of residues on β6 strand appears to correlate with the propensity for dimerization. ::: ![](pbio.0030183.g001) ::: In addition to strand β6, two surface loops also appear to contribute to the dimerization of both caspases-3 and -9. Although residues from these surface loops also differ between the two caspases, they are involved in a coordinated change. For example, Asp^231^ on chain A of a caspase-3 dimer hydrogen bonds to His^234^ on chain B, while the same hydrogen bond is spatially conserved between Lys^410^ on chain A and Ser^382^ on chain B in caspase-9 \[[@pbio-0030183-b19]\]. Thus we concluded that these surface loops in caspase-9 play a similar role as in caspase-3 and are unlikely to be the impediment for the dimerization of caspase-9. The Engineered Caspase-9: A Constitutive Homodimer {#s2b} -------------------------------------------------- The drastic difference in sequences as well as in bonding arrangements of strand β6 between caspases-3 and -9 led us to conclude that the higher propensity of caspase-3 to dimerize than caspase-9 is a direct result of the sequence variation in strand β6. To generate a potentially dimeric caspase-9, we replaced five residues in the β6 strand of caspase-9 (Gly^402^-Cys-Phe-Asn-Phe^406^) with those in caspase-3 (Cys^264^-Ile-Val-Ser-Met^268^), using standard mutagenesis methods. This engineered caspase-9 was overexpressed in bacteria and purified to homogeneity (see [Materials and Methods](#s4){ref-type="sec"} for details). First, we compared the oligomeric states of the engineered and WT caspase-9 using size exclusion chromatography. The elution volume for the WT protein (residues 1--416), which contains a prodomain and a flexible linker segment between the prodomain and the protease domain, corresponds to a molecular mass of approximately 60 kDa ([Figure 2](#pbio-0030183-g002){ref-type="fig"}), slightly larger than that expected for a monomer (approximately 46 kilodaltons \[kDa\]). This is consistent with the fact that the prodomain and the flexible linker segment of caspase-9 may increase its hydrodynamic radius in solution. In contrast, the elution volume of the engineered caspase-9 (residues 1--416) corresponds to a molecular mass of about 120 kDa ([Figure 2](#pbio-0030183-g002){ref-type="fig"}), approximately twice that observed for the WT protein. To better correlate elution volume with molecular mass, WT and engineered proteins were also prepared in which the prodomain and the flexible linker were removed (Δ139). In this case, the elution volumes for the WT and engineered caspase-9 (Δ139, residues 140--416) corresponded to 30 kDa and 60 kDa, respectively, consistent with a monomer and a dimer of caspase-9 (Δ139) (unpublished data). ::: {#pbio-0030183-g002 .fig} Figure 2 ::: {.caption} ###### The Dimeric Caspase-9 Exists as a Monomer in Solution The apparent molecular masses for the WT and dimeric caspase-9 (full-length, residues 1--416) were analyzed by gel filtration. Relevant peak fractions were visualized by SDS-PAGE followed by Coomassie staining. The dimeric caspase-9 was eluted from the column with a molecular mass about twice that of the WT protein. ::: ![](pbio.0030183.g002) ::: To precisely determine the molecular mass of the engineered caspase-9, we performed sedimentation equilibrium experiments using analytical ultracentrifugation. Analysis of the data for the engineered caspase-9 (residues 1--416), at three rotor speeds, reveals a molecular mass of 91,030 ± 2,100 daltons (Da) ([Table 1](#pbio-0030183-t001){ref-type="table"}). This is very close to the expected molecular weight of a homodimer (94,610 Da). In contrast, the WT caspase-9 exhibited a molecular mass of 50,550 ± 2,550 Da ([Table 1](#pbio-0030183-t001){ref-type="table"}), slightly larger than that of a monomer (47,357 Da). Data fitting indicated that the slightly larger molecular mass is not due to a monomer-dimer equilibrium. Rather, it is likely due to a small degree of higher-order nonspecific association among the monomers. Similarly, analysis of the data for the WT and engineered caspase-9 (Δ139) reveals molecular masses of 36,450 ± 3,300 and 61,490 ± 2,350 Da, respectively. These values are close to the predicted molecular masses of monomeric (31,426 Da) and dimeric (62,746 Da) caspase-9 (Δ139). ::: {#pbio-0030183-t001 .table-wrap} Table 1 ::: {.caption} ###### Analytical Ultracentrifugation Analysis of Caspase-9 Proteins ::: ![](pbio.0030183.t001) ::: Thus, results derived from gel filtration and analytical ultracentrifugation analysis unequivocally show that our engineered caspase-9, which contains a five-residue replacement at the dimeric interface, exists predominantly as a homodimer in solution. For the sake of discussion, the engineered caspase-9 is hereafter referred to as the dimeric caspase-9. Structure of the Dimeric Caspase-9 {#s2c} ---------------------------------- One important assumption from our engineering effort is that the replacement of five amino acids in strand β6 would not significantly alter the local structure of the dimeric caspase-9 surrounding strand β6, as compared to the WT caspase-9. Although this assumption is supported by the highly conserved nature of caspase structures, it must be proved experimentally. To obtain a structure of the dimeric caspase-9, we mounted a vigorous crystallization effort. However, the enhanced enzymatic activity (see below) of the dimeric caspase-9 led to time-dependent degradation of the protein under a variety of conditions (unpublished data) that impeded crystallization. To solve this problem, we mutated the catalytic residue Cys^287^ to Ser and were able to crystallize this catalytically inactive form of the dimeric caspase-9. The structure was determined at 2.8 Å ([Figure 3](#pbio-0030183-g003){ref-type="fig"}A and [Table 2](#pbio-0030183-t002){ref-type="table"}). ::: {#pbio-0030183-g003 .fig} Figure 3 ::: {.caption} ###### The Dimeric Caspase-9 Closely Resembles the WT Caspase-9 \(A) Overall structure of the dimeric caspase-9 (C287S). The structural core is shown in blue; the solvent-exposed active site loops are shown in magenta. The β6 and β6′ strands are highlighted in red. Note the asymmetry of the dimer. \(B) Structural superposition of the dimeric caspase-9 (blue and magenta) and the WT caspase-9 (grey and green). The only significant structural difference is in the solvent-exposed active site loops, which is due to the inactive nature of the dimeric caspase-9 (C287S). \(C) A stereo comparison of the region surrounding strands β6 and β6′ in WT and dimeric (engineered) caspase-9. The side chains of the WT and dimeric caspase-9 are shown in orange and yellow, respectively. To avoid congestion in the graphic, residues from only one caspase-9 monomer are labeled. The Cys^287^ in the upper left corner is the catalytic residue in the active site of the asymmetric caspase-9. \(D) A stereo comparison of the region surrounding strands β6 and β6′ in caspase-3 and dimeric (engineered) caspase-9. The side chains of the caspase-3 and dimeric caspase-9 are shown in green and yellow, respectively. The labels refer to amino acids in caspase-3. ::: ![](pbio.0030183.g003) ::: ::: {#pbio-0030183-t002 .table-wrap} Table 2 ::: {.caption} ###### Data Collection and Statistics from Crystallographic Analysis ::: ![](pbio.0030183.t002) R~sym~ = ∑~h~∑~i~ \\| I~h,i~ − I~h~ \\| / ∑~h~∑~i~ I~h,i~ , where I~h~ is the mean intensity of the i obervations of symmetry related reflections of h. R = ∑ \\| F ~obs~ − F ~calc~ \\| / ∑F ~obs~, where F ~obs~ = F ~P~, and F ~calc~ is the calculated protein structure factor from the atomic model (R~free~ was calculated with 5% of the reflections). RMSD in bond lengths and angles are the deviations from ideal values, and the RMSD in B factors is calculated between bonded atoms. ::: This structure offers a number of interesting findings, all of which support the assumption that the five-residue replacement does not alter the local structure of caspase-9 surrounding strand β6. First, the dimeric caspase-9 exists as an asymmetric homodimer in an identical manner to the WT caspase-9 \[[@pbio-0030183-b19]\]. In fact, the dimeric caspase-9 was crystallized in the same space group as the WT caspase-9 with essentially identical unit cell parameters. Just like the WT caspase-9, there are two molecules of the dimeric caspase-9 in each asymmetric unit, and the crystal packing interaction is identical. This observation also indicates that Phe^404^ is not a contributing reason for the observed asymmetry in caspase-9 as previously suggested \[[@pbio-0030183-b19]\], as Phe^404^ is replaced by Val in the dimeric caspase-9. Rather, the asymmetry appears to be an intrinsic property of caspase-9 determined by other sequence elements. It should be pointed out that, in theory, crystal-packing interactions could be the reason for both WT and dimeric caspase-9 to adopt a similar conformation in the crystals. However, such packing interactions are usually extremely weak compared to the forces that govern protein stability and are generally unable to perturb any well-formed structural elements. Second, despite being an inactive zymogen, the dimeric caspase-9 is nearly identical to the WT caspase-9 for the core structural elements, with a root mean square deviation (RMSD) of 0.4 Å over 380 aligned Cα atoms ([Figure 3](#pbio-0030183-g003){ref-type="fig"}B). The only significant structural difference occurs at the solvent-exposed surface loops L2, L4, L2′, and L4′ ([Figure 3](#pbio-0030183-g003){ref-type="fig"}B). These differences reflect the active and inactive nature of the WT caspase-9 and dimeric caspase-9, respectively. As is the case for all other active caspases \[[@pbio-0030183-b01]\], the conformation of these four loops in the inhibitor-bound dimeric caspase-9 is expected to be identical to that of the inhibitor-bound WT caspase-9. Third and most importantly, the conformation of the amino acids surrounding strands β6 and β6′ in the dimeric caspase-9 remains extremely similar or identical to that in the WT caspase-9 ([Figure 3](#pbio-0030183-g003){ref-type="fig"}C). For example, all amino acids in the neighboring strands β5 and β5′ in the dimeric caspase-9 retain the same side chain rotamer conformation as in the WT caspase-9 ([Figure 3](#pbio-0030183-g003){ref-type="fig"}C). There is no significant conformational change in any part of the local structure. Thus we conclude that the impact of the five-residue replacement is naturally absorbed without any significant structural rearrangement, and that the dimeric caspase-9 faithfully mimics the dimerized state of the WT caspase-9. We also compared the dimeric caspase-9 to caspase-3 ([Figure 3](#pbio-0030183-g003){ref-type="fig"}D). The ten Cα atoms on strands β6 and β6′ of the dimeric caspase-9 superimpose very well with the corresponding atoms in caspase-3, exhibiting an overall RMSD of 0.46 Å ([Figure 3](#pbio-0030183-g003){ref-type="fig"}D). The surrounding residues, however, show significant deviation in both main chain and side chain atoms ([Figure 3](#pbio-0030183-g003){ref-type="fig"}D). Structural alignment of the entire dimeric caspase-9 and caspase-3 results in an RMSD of 1.74 Å over 396 Cα atoms. The Dimeric Caspase-9: More Potent than the Wild Type {#s2d} ----------------------------------------------------- Previous structural studies on procaspase-7 demonstrate that the active site conformation in one monomer exists in an unproductive conformation until it is supported by the critical loop L2′ from the adjacent monomer \[[@pbio-0030183-b10],[@pbio-0030183-b11]\]. This conclusion is thought to be generally applicable to other caspases \[[@pbio-0030183-b01]\]. Therefore, the dimeric caspase-9 is predicted to exhibit a higher level of catalytic activity than its monomeric WT counterpart, because the dimeric caspase-9 is poised to provide the L2′ loop. To examine this hypothesis, we reconstituted an in vitro assay in which the ability of caspase-9 to cleave its physiological substrate, procaspase-3 (C163A), was measured in a time-course experiment ([Figure 4](#pbio-0030183-g004){ref-type="fig"}A). As anticipated, both the full-length and the Δ139 dimeric caspase-9 (unpublished data) exhibited a significantly higher activity than their WT counterparts. To accurately determine the differences in activity, we repeated these experiments using the fluorescent substrate specific for caspase-9, LEHD-AFC ([Figure 4](#pbio-0030183-g004){ref-type="fig"}B). These results revealed that the full-length and the Δ139 dimeric caspase-9 were approximately 2-fold and 5-fold more active than their WT counterparts, respectively. ::: {#pbio-0030183-g004 .fig} Figure 4 ::: {.caption} ###### The Dimeric Caspase-9 Exhibits Higher Catalytic Activity and Stronger Cell-Killing Activity than the WT Protein \(A) A time course experiment of caspase-9 activity using procaspase-3 (C163A) as the substrate revealed that the dimeric caspase-9 (residues 1--416) exhibited an approximately 5-fold higher level of catalytic activity than the monomeric WT caspase-9. The concentrations were: WT and engineered caspase-9, 0.5 μM; procaspase-3 (C163A), 33 μM. \(B) Comparison of catalytic activity for the WT and dimeric caspase-9 using fluorescent substrate LEHD-AFC. Both the full-length and Δ139 dimeric caspase-9 displayed higher activities than their WT counterparts. \(C) The dimeric caspase-9 induced apoptosis more effectively than the WT protein. Mammalian expression vector pcDNA3 constructs encoding WT or dimeric caspase-9 were transfected into HeLa or 293 cells. The extent of cell death induced by each construct was examined. \(D) The dimeric but not the WT caspase-9 promoted the activation of caspase-3. The expression of caspase-9 variants (upper panel) and the processing of caspase-3 (lower panel) from the cell extracts were detected by antibodies against caspase-9 and -3, respectively. No endogenous caspase-9 band was visible in the vector-transfected cells at lower exposure. ::: ![](pbio.0030183.g004) ::: Increased catalytic activity for the dimeric caspase-9 is predicted to correlate with stronger ability in promoting apoptosis. To test this prediction, the WT and dimeric caspase-9 were expressed in HeLa and HEK 293 cells; indeed, the dimeric caspase-9 induced extensive cell death in both HeLa ([Figure 4](#pbio-0030183-g004){ref-type="fig"}C) and HEK 293 cells (unpublished data), whereas overexpression of the WT caspase-9 induced little apoptosis. The induction and extent of apoptosis were confirmed by the processing of caspase-3 in cells expressing the dimeric but not the WT caspase-9 ([Figure 4](#pbio-0030183-g004){ref-type="fig"}D). Despite its strong ability in inducing apoptosis, the dimeric caspase-9 was expressed at a much lower level than the WT enzyme ([Figure 4](#pbio-0030183-g004){ref-type="fig"}D). This likely reflects the cytotoxic effect of the dimeric caspase-9 once expressed, or a shorter half-life of the processed dimeric caspase-9. The activation of caspase-3 and induction of apoptosis by the dimeric caspase-9 no longer require the caspase-9-activating factor Apaf-1, because the dimeric caspase-9 (Δ139), which lacks the Apaf-1-binding domain, CARD, was still able to induce efficient cell death and caspase-3 activation ([Figure 4](#pbio-0030183-g004){ref-type="fig"}C and [4](#pbio-0030183-g004){ref-type="fig"}D). These data confirm that the dimeric caspase-9 is active in mammalian cells and that it induces apoptosis by activating caspase-3. The Dimeric Caspase-9 Is Qualitatively Different from the Apaf-1-Activated Caspase-9 {#s2e} ------------------------------------------------------------------------------------ If induced proximity, interpreted as dimerization-driven activation of caspases \[[@pbio-0030183-b23]\], models the correct mechanism for the activation of caspase-9, then the dimeric caspase-9 should exhibit a similar level of catalytic activity to the apoptosome-activated caspase-9. To examine this scenario, we reconstituted an apoptosome-activated caspase-9 assay using an Apaf-1 fragment (residues 1--570) that was known to activate caspase-9 in a cytochrome c-independent manner \[[@pbio-0030183-b16],[@pbio-0030183-b24],[@pbio-0030183-b25]\]. Compared to the dimeric caspase-9, the Apaf-1-activated WT caspase-9 exhibited an approximately 35-fold higher activity using LEHD-AFC as the substrate ([Figure 5](#pbio-0030183-g005){ref-type="fig"}A). We also confirmed this result by performing a time course experiment ([Figure 5](#pbio-0030183-g005){ref-type="fig"}B), and obtained the same conclusion using procaspase-3 (C163A) as the substrate (unpublished data). These results suggest that the dimerization of caspase-9 may be qualitatively different from the way in which Apaf-1 activates caspase-9. ::: {#pbio-0030183-g005 .fig} Figure 5 ::: {.caption} ###### The Dimeric Caspase-9 Exhibits a Much Lower Activity Than the Apaf-1-Activated WT Caspase-9 \(A) Using fluorescent substrate LEHD-AFC, the catalytic activity of the Apaf-1-activated WT caspase-9 was approximately 35-fold greater than that of the dimeric caspase-9. \(B) A time course comparison of the catalytic activities between the Apaf-1-activated WT caspase-9 and the dimeric caspase-9. WT and dimeric caspase-9, 0.5 μM; Apaf-1 (residues 1--570), 2 μM; LEHD-AFC, 200 μM. ::: ![](pbio.0030183.g005) ::: Next, we examined whether the dimeric caspase-9 can be activated by Apaf-1 the same way as the WT caspase-9. Using procaspase-3 (C163A) as the substrate, the catalytic activity of the dimeric caspase-9 does not change appreciably in the presence of increasing amounts of Apaf-1 ([Figure 6](#pbio-0030183-g006){ref-type="fig"}A, lanes 5--8). In contrast, the WT caspase-9 was activated by Apaf-1 in a concentration-dependent manner ([Figure 6](#pbio-0030183-g006){ref-type="fig"}A, lanes 1--4). Subsequently, in the presence of Apaf-1, the catalytic activity of WT caspase-9 significantly exceeded that of the dimeric caspase-9 ([Figure 6](#pbio-0030183-g006){ref-type="fig"}B). ::: {#pbio-0030183-g006 .fig} Figure 6 ::: {.caption} ###### The Dimeric Caspase-9 Can No Longer Be Activated the Same Way as the WT Caspase-9 \(A) Increasing amounts of Apaf-1 (residues 1--570) led to a corresponding increase in the catalytic activity of the WT caspase-9, but had no effect on the activity of the dimeric caspase-9. WT and dimeric caspase-9, 0.5 μM; procaspase-3 (C163A), 33 μM; Apaf-1, 0.24, 0.6, and 1.2 μM. \(B) A time course of procaspase-3 cleavage by the WT and dimeric caspase-9 in the presence of Apaf-1. WT and dimeric caspase-9, 0.3 μM; procaspase-3 (C163A), 33 μM; Apaf-1, 0.3 μM. ::: ![](pbio.0030183.g006) ::: Discussion {#s3} ========== Although induced proximity through high levels of protein expression can lead to caspase activation, it remains questionable whether this is indeed how the initiator caspases are activated under physiological conditions \[[@pbio-0030183-b12]\]. In fact, much of the evidence supporting the induced proximity model can also be used to argue against it. For example, effector caspases can be autoactivated through induced proximity when overexpressed in bacteria; yet in mammalian cells the endogenous effector caspases are activated by the initiator caspases. Thus, merely showing the processing of caspases due to forced oligomerization does not constitute strong support of the induced proximity model and does not explain how initiator caspases are activated in cells. More importantly, the initial proposition of the induced proximity model gives little or no consideration to the most fundamental aspect of cellular biochemistry: specificity, the specific protein-protein interactions that are required for the precise positioning and activation of the initiator caspases. It should be noted that the concept of caspase activation is fundamentally different between the effector and the initiator caspases. The effector caspases, including the zymogens, exist as constitutive homodimers in solution. Their activation requires an interdomain cleavage that facilitates the formation of a productive active site conformation. However, for the initiator caspases, activation has a quite different meaning. For example, the processed caspase-9, similar to the unprocessed procaspase-9, is only marginally active \[[@pbio-0030183-b26],[@pbio-0030183-b27]\]; the primary function of the apoptosome is to up-regulate caspase-9 activity rather than to facilitate its autoprocessing \[[@pbio-0030183-b26],[@pbio-0030183-b28]\]. In fact, the unprocessed caspase-9 can be activated to the same level by the apoptosome as the processed caspase-9 \[[@pbio-0030183-b27]\]. Thus, the activation of caspase-9 is reflected by its association with the apoptosome and not by the interdomain cleavage. It remains to be seen whether caspase-9 activation represents an isolated example or a general theme among the initiator caspases. The reported experimental evidence supporting the initial induced proximity model employed means to facilitate the oligomerization of caspases using heterologous domains that dimerize constitutively or upon binding to ligands \[[@pbio-0030183-b13]--[@pbio-0030183-b15],[@pbio-0030183-b17]\]. The design of these experiments allowed caspases to be brought close to each other via their attached dimerization domains; however, it was not shown that the caspases themselves had dimerized via their intrinsic dimerization interface. The initial induced proximity model was given greater mechanistic meaning when it was explicitly proposed that dimerization via the intrinsic dimerization interface of initiator caspases drives their activation \[[@pbio-0030183-b19]\]. This improved interpretation of the induced proximity model impinges upon specific protein conformation and was perceived as the dominant model to explain the mechanism of initiator caspase activation \[[@pbio-0030183-b20],[@pbio-0030183-b21],[@pbio-0030183-b23]\]. To validate this induced proximity model, the activity of the dimerized caspase, such as caspase-9, must be compared with that of the apoptosome-activated caspase-9. However, a WT caspase-9 homodimer cannot be isolated because caspase-9 exists predominantly as a monomer in solution \[[@pbio-0030183-b22]\]. If there is any tendency for the WT caspase-9 to dimerize, the kinetics must be exceedingly fast, since dimeric caspase-9 has eluded detection by all biochemical means in our hands. It should also not noted that, although a stable caspase-9 homodimer was reported to exist in solution \[[@pbio-0030183-b19]\], rigorous effort in several laboratories, including ours, to reproduce this result have not been successful. Using protein engineering, we generated a stable caspase-9 homodimer by changing residues exclusively at the dimerization interface. It should be noted that our dimeric caspase-9 is different from other heterologous caspase constructs reported in the literature, in which the caspases were fused to heterologous dimerization domains \[[@pbio-0030183-b13]--[@pbio-0030183-b17]\]. Whether the caspases can dimerize via their intrinsic dimerization interface in those circumstances remains undetermined. In contrast, our design relies on the assumptions that the engineered caspase-9 would dimerize via its intrinsic dimerization interface and would closely resemble the WT protein except at the buried dimerization interface. Importantly, these assumptions have been proved correct by our structural analysis of the dimeric caspase-9. However, the engineered dimeric caspase-9 exhibits a catalytic activity that is only a small fraction of that of the WT caspase-9 activated by Apaf-1. The discrepancy in activity suggests that dimerization of caspase-9 may be qualitatively different from the Apaf-1-mediated activation of caspase-9 and is unlikely to be responsible for the activation of caspase-9 in cells. Interestingly, the dimeric caspase-9 exhibits an activity that is only 2- to 5-fold higher than that of the WT caspase-9 (see [Figure 4](#pbio-0030183-g004){ref-type="fig"}B). This observation exactly argues against the prevailing hypothesis that dimerization drives activation of caspase-9, because if dimerization of caspase-9 were the mechanism for its activation, the dimeric caspase-9 should exhibit a much high level of activity---similar to that of the apoptosome-activated caspase-9. This analysis is further supported by our structural observation, which reveals that the five-residue mutation at the interface of the dimeric caspase-9 does not result in any significant conformational changes in the local structure surrounding strands β6 and β6′. If dimerization of caspase-9 is not the major mechanism for its activation, then how is caspase-9 activated by the apoptosome? We note that the Apaf-1-mediated apoptosome has a 7-fold symmetry. The activation of another initiator caspase, caspase-8, is facilitated by the death-inducing signaling complex, which involves a homotrimeric (3-fold symmetry) assembly of the death receptor and other associated factors. One possibility is that the apoptosome may directly activate monomeric caspase-9 through modification of its active site conformation \[[@pbio-0030183-b08],[@pbio-0030183-b12]\]. An alternative model is that the apoptosome assembles the dimeric caspase-9 into a higher-order complex, which results in the modification of the active site conformation for an enhanced activity \[[@pbio-0030183-b12]\]. In addition, it is possible that the dimerized caspase-9 in the context of the apoptosome exhibits a perturbed interface relative to the crystallographically observed interface, which may greatly facilitate the catalytic activity of caspase-9. Although available data are insufficient to differentiate among these models, exquisite conformational change of caspase-9 must be induced upon binding to the apoptosome \[[@pbio-0030183-b29]\]. This conformational change, most likely at the level of active site conformation, is the prerequisite for the activation of caspase-9. We propose that this induced conformation model is the mechanism for the activation of initiator caspases. The induced conformation model for the activation of initiator caspases is different from the dimerization-driven induced proximity model, but these two models may not be mutually exclusive. In some cases, they emphasize different aspects, and initiator caspases may exist in several distinct classes. For example, for some initiator caspases, dimerization might be sufficient for inducing the correct conformation needed for its activation. In this case, the two models are in agreement with each other. However, for caspase-9, dimerization itself is unlikely to be the sole mechanism of activation. Finally, regardless of the semantics, the activation of any initiator caspase must require the formation of a productive active site conformation. Materials and Methods {#s4} ===================== {#s4a} ### Protein preparation. {#s4a1} All constructs were generated using a standard PCR-based cloning strategy, and the identities of individual clones were verified through double-stranded plasmid sequencing. All caspase-9 constructs were expressed from the vector pET-21b in the Escherichia coli strain BL21(DE3). The protein was purified using a Ni-NTA (Qiagen, Valencia, California, United States) column, and further fractionated by anion-exchange (Source-15Q, Pharmacia, Uppsala, Sweden) and gel-filtration chromatography (Superdex-200, Pharmacia). Caspase-3 (C163A) and Apaf-1 (1--570) were overexpressed and purified as described \[[@pbio-0030183-b22]\]. ### Crystallization and data collection. {#s4a2} Crystals were grown by the hanging-drop vapor-diffusion method by mixing dimeric caspase-9 (residues 140--416, C287S) with an equal volume of reservoir solution containing 100 mM MES (pH 6.0), 5% (w/v) PEG 5000 monomethyl ether, and 3% tacsimate. Crystals appeared overnight and grew to a typical size of 0.1 × 0.5 × 0.5 mm^3^ in 3 d. The crystals were in the C2 space group, with the cell parameters a = 144.6 Å, b = 79.0 Å, c = 125.9 Å, and β = 112.5 degrees. There are two complete molecules of dimeric caspase-9 in each asymmetric unit. Diffraction data were collected using an R-AXISIV imaging plate detector mounted on a Rigaku 200HB generator (Rigaku, Tokyo, Japan). For data collection, crystals were equilibrated in a cryoprotectant buffer containing well buffer plus 20% glycerol, and were flash frozen in a −170 °C nitrogen stream. All datasets were processed using the software Denzo and Scalepack \[[@pbio-0030183-b30]\]. ### Structure determination of the dimeric caspase-9. {#s4a3} The structure was determined by molecular replacement, using the software AmoRe \[[@pbio-0030183-b31]\]. The atomic coordinates of caspase-9 \[[@pbio-0030183-b19]\] were used for rotational search against a 15−3.2 Å dataset. The top solutions from the rotational search were individually used for a subsequent translational search, which yielded the correct solutions with high correlation factors. A model was built using the program O \[[@pbio-0030183-b32]\] and refined using CNS \[[@pbio-0030183-b33]\]. The final refined atomic model for the dimeric caspase-9 contains residues 58--196 and 212--303, and 111 ordered water molecules at 2.6 Å resolution. The final refined atomic model contains residues 140--288 and 335--416 for chains A and C, and 140--289 and 337--416 for chains B and D, respectively. ### Caspase-9 assay. {#s4a4} The reaction was performed at 37 °C under the following buffer conditions: 25 mM HEPES (pH 7.5), 100 mM KCl, 4 mM MgCl~2~, and 2 mM dithiothreitol (DTT). The substrate (procaspase-3 \[C163A\]) concentration was approximately 33 μM. Caspase-9 variants were diluted to the same concentration (0.3 or 0.5 μM) with the assay buffer. Reactions were stopped with the addition of an equivolume of 2× SDS loading buffer and boiled for 3 min. The samples were size-fractionated by SDS-PAGE, and the results were visualized by Coomassie staining. The assays using the fluorogenic substrate Ac-LEHD-AFC were performed under similar conditions, where the substrate concentration was 200 μM. Reactions were stopped by placing the samples on ice. Fluorescence emission at 440 nm was measured using an excitation wavelength of 380 nm. ### Analytical ultracentrifugation {#s4a5} Protein samples were prepared in 25 mM HEPES (pH 7.5), 100 mM NaCl, and 0.5 mm DTT. Protein loading concentrations were all 20 μM except for caspase-9 (Δ139) which was studied at 10 μM. All sedimentation equilibrium experiments were carried out at 4 °C using a Beckman Optima XL-A analytical ultracentrifuge (Beckman Instruments, Fullerton, California, United States) equipped with an An60 Ti rotor and six-channel, 12 mm path length, charcoal-filled Epon centerpieces and quartz windows. Data were collected at three rotor speeds (8,000, 11,000, and 14,000 rpm) and represent the average of 20 scans using a scan step-size of 0.001 cm. Partial specific volumes and solution density were calculated using the Sednterp program ([www.bbri.org/RASMB/rasmb.html](www.bbri.org/RASMB/rasmb.html)). Data were analyzed using the WinNONLIN program from the Analytical Ultracentrifugation Facility at the University of Connecticut (Storrs, Connecticut, United States). ### Apoptosis assays. {#s4a6} The ability of caspase-9 variants to induce apoptosis was assayed as described earlier \[[@pbio-0030183-b16]\]. Human HeLa cells or HEK 293 cells were transfected in 12-well plates (10^5^ cells/well) with 0.3 μg of pEGFP-N1 reporter plasmid (Clontech, Palo Alto, California, United States) and 1.0 μg of vector, or a construct encoding WT or dimeric caspase-9 using the LipofectAMINE (Invitrogen, Carlsbad, California, United States) as per the manufacturer\'s recommendations. After 24 h of transfection normal and apoptotic GFP-expressing cells were counted using fluorescence microscopy. The percentage of apoptotic cells in each experiment was expressed as the mean percentage of apoptotic cells as a fraction of the total number of GFP-expressing cells. Apoptosis in the cells transfected with caspase-9 constructs was further confirmed by immunoblotting these cellular extracts with antibodies (Cell Signaling Technology, Beverly, Massachusetts, United States) raised against caspase-3 and caspase-9, respectively. The data represent the average of four independent experiments. We thank N. Hunt for administrative assistance, S. Riedl for Apaf-1 protein, and members of the Shi laboratory for discussion. This work was supported by National Institutes of Health grant CA90269 (to YS), National Science Foundation grant MCB-0211754 (to RF), and a pre-doctoral fellowship from the New Jersey Commission on Cancer Research (to ENS). SMS is a Kimmel Scholar. Competing interests. The authors have declared that no competing interests exist. Author contributions. YC, ENS, and YS conceived and designed the experiments. YC, ENS, SMS, DJR, and RF performed the experiments. SMS, DJR, and RF analyzed the data. YC, ENS, SMS, RF, and YS discussed the results. YC, ENS, and YS wrote the paper. Citation: Chao Y, Shiozaki EN, Srinivasula SM, Rigotti DJ, Fairman R, et al. (2005) Engineering a dimeric caspase-9: A re-evaluation of the induced proximity model for caspase activation. PLoS Biol 3(6): e183. Da : dalton(s) kDa : kilodalton(s) RMSD : root mean square deviation WT : wild-type	PubMed Central	2024-06-05T03:55:56.404403	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088972/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e183", "authors": [ { "first": "Yang", "last": "Chao" }, { "first": "Eric N", "last": "Shiozaki" }, { "first": "Srinivasa M", "last": "Srinivasula" }, { "first": "Daniel J", "last": "Rigotti" }, { "first": "Robert", "last": "Fairman" }, { "first": "Yigong", "last": "Shi" } ] }
PMC1088973	Introduction {#s1} ============ Numerous soft tissue tumors demonstrate specific differentiation toward connective tissue \[[@pbio-0030187-b01]\]. This may be represented in cytoplasmic organelles or extracellular matrix deposition, or defined by immunohistochemical features. Some soft tissue tumors have features of smooth muscle cells (leiomyomas, leiomyosarcomas) or adipocytes (lipoma, liposarcoma). Other soft tissue tumors exhibit features of rarer cell types such as the interstitial cell of Cajal (gastrointestinal stromal tumor) and glomus cells (glomus tumor). There are numerous tumors with fibroblastic and myofibroblastic features, but their corresponding normal counterparts are not well delineated by available markers. We examined two fibroblastic tumors: solitary fibrous tumor (SFT) and desmoid-type fibromatosis (DTF). Both tumors are composed of spindled cells, typically have low-grade nuclear morphology, and can occur throughout the body. Most SFTs occur on the pleural surface, but they have been recognized in a wide range of anatomic locations. Although they were initially thought to be associated with mesothelial differentiation, a number of studies have indicated that SFTs are derived from fibroblasts \[[@pbio-0030187-b02]\]. The vast majority of SFTs are CD34 immunoreactive \[[@pbio-0030187-b05]\]. SFTs do not generally infiltrate into surrounding soft tissue, recur after excision, or metastasize. However, a minority of cases exhibit malignant features \[[@pbio-0030187-b06]\] and these are associated with chromosomal alterations \[[@pbio-0030187-b07]\]. DTF is widely assumed to be derived from fibroblasts of the deep soft tissue. DTFs occur both sporadically or as part of a syndrome due to germline APC mutations in familial adenomatous polyposis coli. These tumors are often found in the deep soft tissue of the trunk or abdomen. The sporadic DTFs also often have mutations in APC or b-catenin \[[@pbio-0030187-b08]\], suggesting that abnormal activation of the canonical Wnt pathway plays a role in their pathogenesis. Sporadic and familial DTFs have been found to be composed of a monoclonal population \[[@pbio-0030187-b09],[@pbio-0030187-b10]\]. DTFs are locally aggressive and are difficult to resect completely: local recurrences in anatomically critical sites can be fatal. Thus SFT and DTF show significant differences in clinical behavior. Although the histologic growth patterns are distinct, with DTF showing a more aggressive infiltrative growth than SFT, the individual cells that comprise these tumors are histologically very similar and hard to distinguish. As such, these two tumors form a good model system to use for discovery of novel connective tissue markers. In this study, we used DNA microarrays to profile gene expression of two fibroblastic tumors, DTF and SFT. The gene expression profiles define two different fibroblastic neoplasms that may correspond to two physiologic fibroblastic phenotypes or fibroblastic response patterns. We demonstrate that several genes differentially expressed in DTF and SFT are also differentially expressed in characteristic patterns in conditions from inflammatory and reparative tissue to neoplasia. The interaction between tumor cells and surrounding stroma has been the subject of many studies. Here we show that gene sets discovered in fibroblastic tumors can be used to recognize prognostically distinct subsets of breast carcinomas. Results {#s2} ======= Expression Profiling Comparison of SFT and DTF {#s2a} ---------------------------------------------- The ten cases of DTF and 13 cases of benign SFT were compared to 35 other previously examined soft tissue tumors \[[@pbio-0030187-b11],[@pbio-0030187-b12]\] with expression profiling on 42,000-element cDNA microarrays, corresponding to approximately 36,000 unique gene sequences. Unsupervised hierarchical cluster analysis organized the 58 tumors and the 3,778 gene spots that demonstrate at least 4-fold variation from the mean in at least two tumors (see [Materials and Methods](#s4){ref-type="sec"}). Based on gene expression, all the DTF and SFT cases can be separated into two groups according to the pathologic diagnosis. The two fibroblastic tumors did not group together. Instead, the SFTs clustered on the same branch as synovial sarcoma and gastrointestinal stromal tumor, whereas the DTF cases clustered on the same branch as the majority of leiomyosarcomas, dermatofibrosarcoma protuberans, and malignant fibrous histiocytomas ([Figure 1](#pbio-0030187-g001){ref-type="fig"}). ::: {#pbio-0030187-g001 .fig} Figure 1 ::: {.caption} ###### Soft Tissue Tumor Gene Expression Unsupervised hierarchical clustering of ten cases of DTF (blue), 13 cases of SFT (orange), and 35 other previously examined soft tissue tumors (black) based on expression profiling on 42,000-element cDNA microarrays. Red represents high expression, black represents median expression, green represents low expression, and grey represents no data. Gene array data are available at <http://microarray-pubs.stanford.edu/tma-portal/DTF_SFTbreast>. DFSP, dermatofibrosarcoma protuberans; GIST, gastrointestinal stromal tumor; LMS, leiomyosarcomas, MFH, malignant fibrous histiocytomas; SS, synovial sarcoma. ::: ![](pbio.0030187.g001) ::: Comparison of Expression Patterns in SFT and DTF {#s2b} ------------------------------------------------ To directly compare the expression patterns, the ten cases of DTF and 13 cases of SFT were analyzed without the other soft tissue tumors. Using the same filtering criteria as above, the 23 tumors were clustered based on 1,010 gene spots. Again, the tumors clustered according to pathologic diagnosis (see [Figure S1](#sg001){ref-type="supplementary-material"}). The dataset was analyzed using the significance analysis of microarray (SAM) method \[[@pbio-0030187-b13]\] to create two lists. The two lists included genes significantly more highly expressed in either SFT or DTF. A total of 786 gene spots, differentially expressed between the two tumor types, had a false discovery rate of one in 786 (0.13%). The SFT-specific gene list shared 64% identity with a list of genes selected using SAM for specific expression in SFT compared to all other soft tissue tumors in the initial set of 58 soft tissue tumors. Likewise, the DTF-specific gene list shared 65% identity with a list selected by SAM based on differential expression in DTF compared with the 58 soft tissue tumors. The two tumor types differed in their patterns of expression in a number of different functional categories of genes ([Tables 1](#pbio-0030187-t001){ref-type="table"} and [S1](#st001){ref-type="supplementary-material"}). On the basis of these differences in expression, we hypothesize that the cells of origin for each lesion may perform different functions in normal connective tissue. One of the more striking differences is in the variation of genes involved in fibrotic response and basement membrane synthesis between the two tumors. DTF has high expression of genes involved in the fibrotic response. These include numerous collagen genes, such as COL1A1 and COL3A1, involved in fibrosis and contraction and a number of growth factors that stimulate the classic fibrotic response. DTFs also highly express numerous genes that remodel the extracellular matrix, including ADAM and MMP family members, consistent with its infiltrative behavior. In contrast, SFTs highly express collagen genes and other genes involved in basement membrane formation and maintenance, such as COL4A5 and COL17A1. In contrast to DTF, no metalloproteinase family members were especially highly expressed in SFTs. Possible exceptions were ADAM22 and ADAM23, which were highly expressed in SFT. But the metalloprotease domain is inactive in these proteins, and these proteins are more likely involved in cell adhesion than in matrix remodeling. SFTs highly express a number of signaling pathways involved in growth and survival, including BCL2 and IGF1. DTF and SFT also differed in other pathways, including WNT signaling and THY1 expression. Thus, although SFT and DTF both express genes typically expressed in fibroblasts, they express genes that belong to very different functional groups. ::: {#pbio-0030187-t001 .table-wrap} Table 1 ::: {.caption} ###### Selected Genes in DTF (Group A) and SFT (Group B) ::: ![](pbio.0030187.t001) DTF and SFT were analyzed by SAM (see [Materials and Methods](#s4){ref-type="sec"}) resulting in 786 genes with fewer than 0.1% false positive genes. Entire gene list is available at <http://microarray-pubs.stanford.edu/tma-portal/DTF_SFTbreast>. ::: Histologic Patterns of Expression of Genes Characteristic of SFT and DTF {#s2c} ------------------------------------------------------------------------ To confirm, localize, and extend our observations on the expression of DTF- and SFT-specific genes, we constructed a tissue microarray (TMA) and measured expression using immunohistochemistry (IHC) and in situ hybridization (ISH; see [Materials and Methods](#s4){ref-type="sec"}). The TMA contained representative cores of five DTFs and SFTs, in addition to cores of scar and keloid. In addition, the TMA included well-oriented embedded pieces of normal skin, lung, and breast tissue ([Figure S2](#sg002){ref-type="supplementary-material"}). The array also contained 11 fibroadenomas, as well as five colorectal and 24 breast carcinomas. SFTs, fibroadenomas, and a subset of normal fibroblasts in the skin and breast specimens demonstrated expression of SFT-specific genes ([Figures 2](#pbio-0030187-g002){ref-type="fig"}, [3](#pbio-0030187-g003){ref-type="fig"}, and [S3](#sg003){ref-type="supplementary-material"}). Normal fibroblasts that reacted for SFT markers, APOD and CD34, included those associated with adnexal glands and dermal fat. The reactivity of so-called dendritic interstitial cells for CD34 in a number of locations was previously reported \[[@pbio-0030187-b14]\]. These tissues were rarely positive for DTF-specific gene probes. DTF-specific probes, for OSF2 and CTHRC1, were positive in DTF, keloid, scar, granulation tissue, and fistula tract ([Figures 2](#pbio-0030187-g002){ref-type="fig"} and [3](#pbio-0030187-g003){ref-type="fig"}). In the granulation tissue and fistula tract tissue, a gradient of expression dependent on location of the cells within the tissue could be identified in some hybridizations. There was no staining of fibroblast-like cells by probes for OSF2 and CTHRC1 in the normal tissues. ::: {#pbio-0030187-g002 .fig} Figure 2 ::: {.caption} ###### Localization of Fibroblastic Gene Expression Comparison of expression of two SFT markers APOD (ISH) and CD34 (IHC), and two DTF markers CTHRC1 (ISH) and OSF2 (ISH) in SFT and DTF. SFTs express ApoD and CD34 whereas DTFs express CTHRC1 and OSF2. H&E, hematoxylin-eosin. Magnification = 600×. ::: ![](pbio.0030187.g002) ::: ::: {#pbio-0030187-g003 .fig} Figure 3 ::: {.caption} ###### Fibroblastic Markers in Non-Neoplastic Tissue \(A) Skin adnexa, (B) breast, (C) dermis, (D) reactive, and (E) keloid tissue arranged in rows. Fibroblastic markers: CD34 (IHC), APOD (ISH), CTHRC1 (ISH) and OSF2 (ISH) arranged in columns. SFTs express APOD and CD34 whereas DTFs express CTHRC1 and OSF2. Magnification = 600×. (A magnification of 300× is shown in [Figure S3](#sg003){ref-type="supplementary-material"}.) ::: ![](pbio.0030187.g003) ::: A similar pattern of differential expression of SFT and DTF markers was observed in breast carcinoma. With the exception of APOD, only stromal staining was observed with these markers whereas the neoplastic epithelial cells did not react. For breast carcinoma, 24 cases were scored for stromal staining (see [Materials and Methods](#s4){ref-type="sec"}) and clustered by hierarchical clustering. The resulting dendrogram and heatmap are shown in [Figure 4](#pbio-0030187-g004){ref-type="fig"}. A subset of cases was positive for the SFT markers, CD34 and APOD, another for the DTF markers, OSF2 and CTHRC1. ::: {#pbio-0030187-g004 .fig} Figure 4 ::: {.caption} ###### Fibroblast Markers in Breast Carcinoma \(A) Examples of SFT (APOD \[ISH\] and CD34 ) and DTF (CTHRC1 \[ISH\] and OSF2 \[ISH\]) expression in breast carcinoma stroma. Each panel shows expression of the marker that is restricted to the fibroblasts between neoplastic cells. Magnification = 600×. \(B) Hierarchical clustering of 24 breast carcinomas based on TMA staining with fibroblast markers: CD34 (IHC), APOD (ISH), CTHRC1 (ISH), and OSF2 (ISH). Bright red represents high expression, dull red represents intermediate high expression, green represents negative expression, and white represents no data. The DTF-associated cluster is highlighted in blue. The SFT-associated cluster is highlighted in orange. Most breast carcinomas express either a DTF or SFT gene in the stromal fibroblasts. However, some breast carcinomas express a combination of DTF and SFT genes, and some express neither. ::: ![](pbio.0030187.g004) ::: Variable Expression of Genes Characteristic of Fibroblastic Tumors in Breast Carcinoma {#s2d} -------------------------------------------------------------------------------------- To further investigate the implication of the variation in expression of these fibroblastic tumor-related genes in breast cancer, we analyzed their expression in 295 breast carcinomas using a previously published dataset. We focused on the genes selected by SAM for differential expression in DTF versus SFT, and investigated their expression levels in the published breast cancer dataset (see [Materials and Methods](#s4){ref-type="sec"}). When clustering the breast carcinomas with the fibroblastic tumor-related genes only, the resulting dendrogram of the tumors/samples showed several high-order branches of correlation between distinct tumor groups. Two of these groups ([Figure 5](#pbio-0030187-g005){ref-type="fig"}, groups A and B) showed remarkable differences in the expression of DTF versus SFT genes. Tumor group A, composed of 120 breast carcinomas, showed high levels of expression of a gene cluster (gene cluster 1, left sidebar) highly enriched for genes that are found in DTF (see right sidebar: genes highly expressed in DTF are represented by purple). This gene cluster was predominately composed of genes whose protein products interact with the extracellular matrix, including collagens, cadherins, and remodeling enzymes. Moreover, two key growth factors in the fibrotic response were also identified, TGFB3 and CTGF. The second tumor group (group B), composed of 59 breast carcinomas, showed expression of a mixture of genes (gene cluster 2, left sidebar) that were enriched for those genes that positively identified SFT (see right sidebar: genes highly expressed in SFT are represented by pink). This gene cluster contained extracellular matrix-interacting genes, such as COL9A3 and ADAMTS1. An additional cluster (gene cluster 3, left sidebar), containing a mixture of SFT and DTF genes, was predominately highly expressed across all tumors except for the tumor group B. ::: {#pbio-0030187-g005 .fig} Figure 5 ::: {.caption} ###### Hierarchical Clustering of 295 Breast Carcinomas with 471 SFT and DTF Genes Within the heatmap, red represents high expression, black represents median expression, and green represents low expression. Sidebar on right indicates which tumor the gene is positively associated with: pink is SFT and purple is DTF. Sidebar on left indicates gene cluster. Gene array data are available at <http://microarray-pubs.stanford.edu/tma-portal/DTF_SFTbreast>. ::: ![](pbio.0030187.g005) ::: The prognosis of these two tumor groups, (A and B), was assessed by distant metastasis-free survival and overall survival ([Figure 6](#pbio-0030187-g006){ref-type="fig"}). Group A demonstrated significantly better outcomes in both overall survival (80% at 10 y vs. 63%; p = 0.0009) and metastasis-free survival (77% at 10 y vs. 58%; p = 0.002) as compared to the all tumors. In contrast, group B demonstrated significantly poorer outcome in overall survival (45% at 10 y vs. 76%; p \< 0.00001) and distant metastasis-free survival (50% at 10 y vs. 69%; p = 0.002) compared to all other tumors. ::: {#pbio-0030187-g006 .fig} Figure 6 ::: {.caption} ###### Outcome Data Statistical method of the y-axis is Kaplan-Meier survival curves compared by the Cox-Mantel log-rank test. The x-axis unit of measure is years. \(A) Time to first recurrence for tumor group A versus all other tumors. \(B) Time to first recurrence for tumor group B versus all other tumors. \(C) Survival outcome for tumor group A versus all others. \(D) Survival outcome for tumor group B versus all others ::: ![](pbio.0030187.g006) ::: For both tumor groups A and B, prognostic performance was independent in multivariate analysis for clinical risk factors including tumor size, lymph node status, and tumor grade (see [Table 2](#pbio-0030187-t002){ref-type="table"}). The hazard ratio for death was 2.6 (1.6--4.4, 95% confidence interval \[CI\]) for group B and 0.55 (0.33--0.92, 95% CI) for group A. Group B also retained independent prognostic relevance when the previously described 70-gene prognosis profile \[[@pbio-0030187-b15]\] is considered in the model. ::: {#pbio-0030187-t002 .table-wrap} Table 2 ::: {.caption} ###### Multivariate Analysis for Tumor Group Status versus Clinical Risk Factors including Treatment with Chemotherapy, Tumor Size (\<2 cm), Lymph Node Status, Tumor Grade (Low and Intermediate versus High), Age (\<40 y old), Vascular Invasion ::: ![](pbio.0030187.t002) The hazard ratio for death, CI, and statistical significance are included. The "70 genes" factor refers to the 70 genes previously published to be predictive in the 295 breast carcinomas dataset \[[@pbio-0030187-b15]\]. ChemoTx, chemotherapy; LN, lymph node. DOI: [10.1371/journal.pbio.0030187.g002](10.1371/journal.pbio.0030187.g002) ::: Discussion {#s3} ========== Expression patterns among fibroblasts in tumors/carcinomas in vivo are difficult to assess due to tissue heterogeneity, which includes the relative content of epithelial cells, vascular structures, and inflammatory cells, and the diversity of fibroblastic and myofibroblastic cells that may be present. We have attempted to gain insight into the possible variation in expression patterns in fibroblastic cells by examining two fibroblastic neoplasms, SFT and DTF. Soft tissue tumors are comprised of relatively pure populations of cells in comparison with other tissue types, including normal tissues and other neoplasms \[[@pbio-0030187-b16]\]. Thus, the gene expression profile of a soft tissue tumor represents primarily a single cell type. To a degree, many soft tissue tumors recapitulate normal tissue components both morphologically and by protein expression, and this is the basis for much of the diagnostic nomenclature in surgical pathology. Interactions between carcinoma and host tissue have long been recognized. Many studies have demonstrated the importance of vascular recruitment and inflammatory response in tumorigenesis. The role that fibroblastic cells play in carcinoma has been less well defined. In part, this problem arises from our limited understanding of fibroblast subtypes and/or fibroblast activation states. Past studies have noted the presence of a "fibroblast signature" in carcinoma \[[@pbio-0030187-b17]\] and other studies have demonstrated topographical variation in fibroblast gene expression in vitro \[[@pbio-0030187-b18]\]. Two previous studies have examined the gene expression profiles for stromal cells in the context of carcinoma. One study examined the gene expression progression in cultured primary fibroblasts in response to serum exposure \[[@pbio-0030187-b19]\]. This expression program included many features suggestive of a wound response \[[@pbio-0030187-b20]\]. Tissue localization studies demonstrated that in carcinomas, most of these "wound-response" genes were expressed by the tumor and stromal cells, although some were expressed by tumor cells, and some by stromal cells alone. The wound-response signature was strongly predictive of metastasis and progression for a variety of carcinomas. There is no significant overlap between the genes in the "serum-response" signature and the genes we report here to be associated with either SFT or DTF. A follow-up study \[[@pbio-0030187-b21]\] demonstrated that the serum-response signature was an independent predictor of outcome in the same dataset of 295 breast carcinomas currently studied. When compared to the 509 unique gene sequences of the serum-response signature applied to the NKI (Nederlands Kanker Instituut \[Netherlands Cancer Institute\]) breast carcinoma dataset \[[@pbio-0030187-b21]\], there are only 15 matches to the SFT/DTF gene list. The lack of overlap makes sense as the experimental approach between the two lists is fundamentally different. The serum-response signature looks at the effects of serum on cells and whether the resulting gene expression pattern could be seen in cancer. The Chang et al. study \[[@pbio-0030187-b19]\] used cultured fibroblasts as a detection system for serum response, but in breast carcinomas most of the genes thus identified were expressed in both tumor and stromal cells. In this study we searched for genes expressed by "fibroblastic tumors" with the aim of gaining insight into stromal cells within tumors. A second study used serial analysis of gene expression on sorted components of the breast cancer microenvironment \[[@pbio-0030187-b22]\]. The authors used antibody beads to separate the cancer tissue into five categories: "epithelial cells," "leukocytes," "myoepithelium/myofibroblasts," "endothelium," and "stroma." Interestingly, a number of genes were found to be highly expressed in their "myoepithelial/myofibroblast" cell population that are also present in our fibromatosis gene list, including COL1A1, MMP11, and CTHRC1. However, that study only examined three invasive breast carcinomas and did not report on prognostic significance. We hypothesized that tumors with different fibroblastic features might represent different activation states or different subtypes of normal fibroblasts or stromal cells. Thus, we examined two tumors with fibroblastic differentiation: SFT and DTF. These two tumors have been extensively studied by morphology, IHC, and electron microscopy and are known to share features with non-neoplastic fibroblasts \[[@pbio-0030187-b01]\]. In this study we demonstrate that the gene expression patterns of these two tumors are distinguished by differences in expression of a variety of functional groups of genes. DTF expresses numerous collagens that are present in a fibrotic response. Numerous myofibroblastic genes are also expressed by DTF. In contrast, SFTs express collagens and other extracellular matrix proteins that are typically found in the basement membrane. DTF tumors express several genes in the ADAM and MMP families involved in extracellular matrix remodeling, which might be relevant to the more infiltrative behavior of these tumors. SFTs expresses few of these genes, and the ADAMs that are expressed in SFT (ADAM22 and ADAM23) are probably involved more in cell adhesion than in extracellular matrix remodeling. In addition, DTF tumors express growth factors involved in the profibrotic response, such as TGFB and CTGF. By IHC and ISH, markers representative of the separate DTF and SFT gene sets highlighted at least two groups of normal connective tissue "fibroblasts" or stromal cells. The cells positive for DTF markers are found in a variety of reactive tissues, ranging from inflammatory granulation tissue to scar tissue. In contrast, cells positive for SFT markers tend to be found in normal tissue. The stromal cells surrounding breast lobules and eccrine lobules of the skin were strongly reactive for SFT markers and negative for DTF genes. These findings are consistent with the gene expression data in which SFTs highly express many genes that help create basement membrane. We created two gene sets consisting of genes that are positively identified either as DTF or SFT. For four genes we determined the expression patterns in breast carcinoma samples and showed that they were restricted to connective tissue cells and were not expressed by tumor cells. With these gene sets, we can evaluate for the presence of an expression signature of either SFT or DTF in other gene array datasets. In this study, we examined a previously published breast carcinoma dataset that contains 295 tumors with a median follow-up of 7.8 y \[[@pbio-0030187-b15]\]. These gene sets highlight a minor expression pattern within a gene expression dataset that may not be readily apparent when the entire dataset is examined. In this case, the expression pattern is putatively associated with stromal fibroblast-like cells, a cell population that is often the minority in breast carcinoma and may not have as much RNA expression. Thus, we might expect the expression signature of stromal cells to be obscured in the hierarchical clustering of the entire dataset. When the breast carcinoma dataset was analyzed with the SFT and DTF gene sets, three main gene clusters were apparent, one more tightly correlated than the other two. The first gene cluster (see [Figure 5](#pbio-0030187-g005){ref-type="fig"}, gene cluster 1) was composed almost entirely of DTF genes. Most of these genes are involved in stimulating or interacting with the extracellular matrix in a pro-fibrotic manner. This gene cluster identified a tumor cluster of 120 cases (tumor group A). Tumor group B showed a less-obvious relationship to either of the soft tissue tumors. However, it was defined by two gene clusters enriched for SFT genes, either by high expression for the genes (gene cluster 2) or relatively low expression for these genes (gene cluster 3). Interestingly, the two tumor groups had very different clinical behaviors. Tumor group A had a statistically significant better overall survival and metastasis-free survival when compared to the rest of the dataset. In contrast, tumor group B had a statistically significant worse overall survival and metastasis-free survival when compared to the rest of the dataset. In multivariate analysis this predictive value is independent of clinico-pathological risk factors. These findings show that stromal expression patterns can vary amongst breast carcinomas and may be clinically significant. In summary, analysis of gene expression patterns in two soft tissue tumors, DTF and SFT, has allowed identification of at least two different nonneoplastic subtypes of stromal cells. Furthermore, analysis of the gene expression signatures of these soft tissue tumors in a breast carcinoma expression dataset has suggested that there may be molecularly distinct patterns of stromal reaction in breast cancer. These stromal reaction patterns appear to be correlated with differences in the biology of the tumors that are reflected in clinical outcome. Materials and Methods {#s4} ===================== {#s4a} ### Tumor samples for DTF and SFT cDNA microarray analysis {#s4a1} Tumors were collected from four academic institutions (Vancouver General Hospital, Cleveland Clinic Foundation, University of Washington Medical Center, and Stanford University Medical Center) with IRB approval. After resection, a representative sample was quickly frozen and stored at −80 °C. Prior to processing, frozen sections of the tissue were cut and histologically examined to ensure that the tissue represented the diagnostic entity. The DTFs were all sporadic cases, including five cases from the extremities, two cases from the abdomen, two cases from the sacrum, and one case from the chest wall. The SFTs included 13 cases with benign features; all but one were derived from the chest cavity. SFT cases with malignant pathologic or clinical features were excluded. The diagnoses were based on clinical data, morphologic data, and IHC, including CD34 ([Table S2](#st002){ref-type="supplementary-material"}). ### DTF and SFT cDNA microarray procedures {#s4a2} We used 42,000-spot cDNA microarrays to measure the relative mRNA expression levels in the tumors. The details of isolating mRNA, labeling, and hybridizing are described elsewhere \[[@pbio-0030187-b11]\]. The raw data files are available at Stanford Microarray Database (<http://genome-www5.stanford.edu/>; the filtered data used for the paper are available at the accompanying Web site (<http://microarray-pubs.stanford.edu/tma-portal/DTF_SFTbreast>). Data were filtered using the following criteria: Only cDNA spots with a ratio of signal over background of at least 1.5 in both the Cy3 and the Cy5 channel were included; only cDNAs were selected that had an absolute value at least four times greater in at least two arrays than the geometric mean; and only cDNA spots that fulfill these criteria on at least 70% of the arrays were included. Data were evaluated with unsupervised hierarchical clustering and SAM \[[@pbio-0030187-b13]\]. ### Analysis of breast carcinoma dataset {#s4a3} The gene array dataset for breast carcinoma contained 295 tumors arrayed on 25,000-spot oligo nucleotide arrays as described elsewhere \[[@pbio-0030187-b15]\]. In short, patients were all diagnosed and treated in the Netherlands Cancer Institute for early breast cancer (Stage I and II) between 1984 and 1995. The median follow-up for living patients is 7.8 y. Additional clinical data can be found in [Table S3](#st003){ref-type="supplementary-material"}. For DTF and SFT, genes were identified that were highly expressed in either of the two tumor types by using SAM \[[@pbio-0030187-b13]\]. A total of 1,010 spots satisfied the gene-filtering criteria mentioned above in the clustering of the DTF and SFT tumors. The criterion for SAM was set to yield 0.1% false-positive data. A list of 786 clones was obtained that consisted of 493 genes positively identifying fibromatosis and 293 genes positively identifying SFT. Equal numbers of DTF and SFT clones were chosen for breast carcinoma analysis, and clones having the same Unigene locus were removed, resulting in 237 unique gene sequences identifying DTF and 246 unique gene sequences identifying SFT. These gene sequences were mapped to spots on the NKI array using Unigene build 172 (release date 17 July 2004) to give 471 unique spots. Gene measurements were mean centered. The resulting dataset was subjected to hierarchical clustering with average linkage clustering. Overall survival (OS) was defined by death from any cause. In this cohort of young breast cancer patients, only six patients died of causes other than breast cancer (five second primaries and one cardiovascular). Distant metastasis-free survival (DMFS) was defined by a distant metastasis as a first recurrence event; data on all patients were censored on the date of the last follow-up visit, death from causes other than breast cancer, the recurrence of local or regional disease, or the development of a second primary cancer, including contra-lateral breast cancer. Kaplan-Meier survival curves were compared by the Cox-Mantel log-rank test in Winstat for Microsoft Excel (R. Fitch Software, Germany). Multivariate analysis by the Cox proportional hazard method was performed using the software package SPSS® 11.5 (SPSS, Inc.). ### TMA construction {#s4a4} A TMA of fibroblastic conditions was constructed using a manual tissue arrayer (Beecher Instruments, Silver Spring, Maryland, United States) following previously described techniques \[[@pbio-0030187-b23]\] with modifications. Briefly, certain specimens, such as skin and fistula tract, contained tissues whose positional orientation was important for analysis. Coring of these tissues could lose orientation of the cells within the core. Therefore, orientation-sensitive material was dissected from the original blocks and re-embedded into the paraffin block used for tissue arraying. Tissues thus embedded included skin, lung, breast, granulation tissue, and fistula tract (see [Figure S2](#sg002){ref-type="supplementary-material"}). After the embedding process was completed, construction of the tissue array was performed using single 2-mm cores. In addition, the TMA contained 0.6-mm cores of lobular (n = 14) and ductal (n = 10) breast carcinomas, fibroadenomas (n = 11), SFT (n = 5), DTF (n = 5), and colorectal carcinomas (n = 2), scar (n = 1), and keloid (n = 1). All samples were obtained from archived material at the Stanford University Medical Center Department of Pathology between 2001 and 2004 with IRB approval. The cores were taken from areas in the paraffin block that were representative of the diagnostic tissue. ### IHC {#s4a5} Serial sections of 4 μm were cut from the TMA blocks, deparaffinized in xylene, and hydrated in a graded series of alcohol. The slides were pretreated with citrate buffer and a microwave step. Staining was then performed using the DAKO EnVision+ System, Peroxidase (DAB), (DAKO, Cambridgeshire, United Kingdom) for APOD (Clone 36C6, 1:40 dilution, Novocastra, Newcastle, United Kingdom), CD34 (1:20 dilution, BD Biosciences, San Diego, California, United States), and BCL2 (1:800 dilution, DAKO Cytomation, Carpinteria, California, United States) stains. Results were interpreted as follows: Staining was interpreted as negative when no more than 5% of the spindled stromal cells showed light staining. A score of "weak positive" was given for light-brown staining in more than 5% of the spindled stromal cells. A score of "strong positive" was given for staining in more than 50% of the spindled stromal cells. Cores in which no diagnostic material was present were omitted from further analysis. The cores were initially reviewed independently by two pathologists (RW and MvdR), and disagreements were reviewed together to achieve a consensus score. Scoring of the arrays was analyzed using the Deconvoluter software as previously described \[[@pbio-0030187-b24]\], with each sample receiving the highest score for either of the two cores. ### In situ hybridization (ISH) {#s4a6} ISH of TMA sections was performed based on a protocol published previously \[[@pbio-0030187-b23],[@pbio-0030187-b25]\]. Briefly, digoxigenin (DIG)-labeled sense and anti-sense RNA probes are generated by PCR amplification of 400 to 600 bp products with the T7 promoter incorporated into the primers. In vitro transcription was performed with a DIG RNA-labeling kit and T7 polymerase according to the manufacturer\'s protocol (Roche Diagnostics, Indianapolis, Indiana, United States). We cut sections 4 μm thick from the paraffin blocks, deparaffinized them in xylene, and hydrated them in graded concentrations of ethanol for 5 min each. Sections were then incubated with 3% hydrogen peroxide, followed by digestion in 10 μg/ml of proteinase K at 37 ^o^C for 30 min. Sections were hybridized overnight at 55 ^o^C with either sense or anti-sense riboprobes at 150 ng/ml dilution in mRNA hybridization buffer (DAKO). The following day, sections were washed in 2× SSC and incubated with a 1:35 dilution of RNase A cocktail (Ambion, Austin, Texas, United States) in 2× SSC for 30 min at 37 ^o^C. Next, sections were stringently washed in 2× SSC/50% formamide twice, followed by one wash at 0.08× SSC at 50 ^o^C. Biotin blocking reagents (DAKO) were applied to the section to block the endogenous biotin. For signal amplification, a HRP-conjugated rabbit anti-DIG antibody (DAKO) was used to catalyze the deposition of biotinyl tyramide, followed by secondary streptavidin complex (GenPoint kit; DAKO). The final signal was developed with DAB (GenPoint kit; DAKO), and the tissues were counterstained in hematoxylin for 15 s. The primer sequences used for the amplification of probes for OSF2, CTHRC1, and APOD are given in [Table S4](#st004){ref-type="supplementary-material"}. Supporting Information {#s5} ====================== Figure S1 ::: {.caption} ###### Soft Tissue Tumor Gene Expression of Ten Cases of DTF and 13 Cases of SFT The DTF cases (blue) and the SFT cases (orange) are based on expression profiling on 42,000-element cDNA microarrays. (1.6 MB JPG). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S2 ::: {.caption} ###### Low-Power Image of TMA with Oriented Fragments of Tissue (and Cores The oriented fragments of tissue are show at the bottom of the figure; the cores are shown at the top (683 KB JPG). ::: ::: {.caption} ###### Click here for additional data file. ::: Figure S3 ::: {.caption} ###### Fibroblastic Markers in Non-Neoplastic Tissue \(A) Skin adnexa, (B) breast, (C) dermis, (D) reactive, and (E) keloid tissue arranged in rows. Fibroblastic markers: CD34 (IHC), APOD (ISH), CTHRC1 (ISH), and OSF2 (ISH) arranged in columns. SFTs express APOD and CD34 whereas DTFs express CTHRC1 and OSF2. Magnification = 300×. (7.0 MB JPG). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S1 ::: {.caption} ###### Significance Analysis of Microarray (SAM) SFT and DTF are analyzed with a false discovery rate of 0.1%. (1.0 MB XLS). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S2 ::: {.caption} ###### Clinical Information for DTF and SFT Cases (17 KB XLS). ::: ::: {.caption} ###### Click here for additional data file. ::: Table S3 ::: {.caption} ###### Clinical Data on the 295 Breast Carcinoma Case Set from the Netherlands Cancer Institute (131 KB XLS) ::: ::: {.caption} ###### Click here for additional data file. ::: Table S4 ::: {.caption} ###### Primer Sequences Used for the Amplification of Probes for ISH (17 KB PDF). ::: ::: {.caption} ###### Click here for additional data file. ::: Accession Numbers {#s5a8} ----------------- The Entrez Gene (<http://www.ncbi.nlm.nih.gov/entrez/>) GeneID accession numbers for the genes and gene products discussed in this paper are, APOD (GeneID 347), BCL2 (GeneID 596), CD34 (GeneID 947), COL9A3 (GeneID 1299), CTHRC1 (GeneID 115908), OSF2 (GeneID 10631), and THY1 (GeneID 7070). This work was supported by the National Institutes of Health (grant CA85129) and the Howard Hughes Medical Institute. POB is an Associate Investigator of the Howard Hughes Medical Institute. TON is a scholar of the Michael Smith Foundation for Health Research. Competing interests. The authors have declared that no competing interests exist. Author contributions. RBW, DSAN, POB, MvdV, and MvdR conceived and designed the experiments. RBW, DSAN, POB, MvdV, and MvdR analyzed the data. RBW, DSAN, SS, KM, and SZ performed the experiments. RBW, DSAN, SS, TON, CLC, BPR, RP, TH-B, JRG, POB, MvdV, and MvdR contributed reagents/materials/analysis tools. RBW, DSAN, SS, TON, CLC, BPR, POB, MvdV, and MvdR wrote the paper. Citation: West RB, Nuyten SA, Subramanian S, Nielsen TO, Corless CL, et al. (2005) Determination of stromal signatures in breast carcinoma. PLoS Biol 3(6): e187. CI : confidence interval DTF : desmoid-type fibromatosis IHC : immunohistochemistry ISH : in situ hybridization SAM : significance analysis of microarray SFT : solitary fibrous tumor TMA : tissue microarray	PubMed Central	2024-06-05T03:55:56.408005	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088973/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e187", "authors": [ { "first": "Robert B", "last": "West" }, { "first": "Dimitry S. A", "last": "Nuyten" }, { "first": "Subbaya", "last": "Subramanian" }, { "first": "Torsten O", "last": "Nielsen" }, { "first": "Christopher L", "last": "Corless" }, { "first": "Brian P", "last": "Rubin" }, { "first": "Kelli", "last": "Montgomery" }, { "first": "Shirley", "last": "Zhu" }, { "first": "Rajiv", "last": "Patel" }, { "first": "Tina", "last": "Hernandez-Boussard" }, { "first": "John R", "last": "Goldblum" }, { "first": "Patrick O", "last": "Brown" }, { "first": "Marc", "last": "van de Vijver" }, { "first": "Matt", "last": "van de Rijn" } ] }
PMC1088974	Introduction {#s1} ============ In 1861, Paul Broca \[[@pbio-0030190-b01]\] proclaimed that the "majesty of the human" could be attributed to its superior faculties, such as abstraction and judgement, and that these superior faculties lie within the province of the "anterior lobes of the brain." Today the picture has changed little, and the frontal lobes, in particular the prefrontal cortex (PFC), are viewed as the repository of a host of higher-order faculties such as control of action, temporal organization of behaviour, sequencing, goal-directed behaviour, abstract and conceptual thinking, creativity, reasoning, and cognitive flexibility, to name a few \[[@pbio-0030190-b02]--[@pbio-0030190-b05]\]. Often these faculties are subsumed under two broad headings that define PFC function: working memory and executive control. There is ample evidence that the PFC is important for both working memory, the ability to store information for short periods of time, and executive control, processes that operate on the contents of stored information. With respect to working memory, numerous studies in humans have shown that the PFC is activated during tasks that require retention of information \[[@pbio-0030190-b04],[@pbio-0030190-b06],[@pbio-0030190-b07]\]. Similarly, studies with monkeys have shown that certain neurons in the PFC display increased and sustained activation during the retention period of working memory tasks \[[@pbio-0030190-b08]--[@pbio-0030190-b11]\]. This sustained activation, also commonly referred to as delay activity because it occurs during the delay (memory) portion of a working memory task, is believed to represent a neural correlate of working memory \[[@pbio-0030190-b12]--[@pbio-0030190-b16]\]. In contrast to the evidence regarding the PFC and working memory, the evidence that the PFC is involved in executive control processes is considerably more varied in nature. In humans, tasks that require conflict resolution, that is, tasks that pit attending to task-relevant information and inhibiting task-irrelevant information, such as the Stroop Interference test or the Wisconsin Card Sorting Task, result in activation of the frontal cortex \[[@pbio-0030190-b17],[@pbio-0030190-b18]\]. In monkeys, findings that PFC neurons are robust to distracting events \[[@pbio-0030190-b11]\], are modulated by attentional demands \[[@pbio-0030190-b19]\], encode abstract rules \[[@pbio-0030190-b20]\], and integrate information across different senses \[[@pbio-0030190-b21]--[@pbio-0030190-b23]\] have all been taken as evidence for its role in executive control processes. Part of the reason for the varied evidence for executive control processes in both human and monkey studies is that the term executive control is itself somewhat poorly defined \[[@pbio-0030190-b24]\]. While the taxonomy of executive control still remains vague, some have argued that of all the processes that fall under the rubric of executive control, one of the fundamental processes is the ability to selectively remember relevant information and discard irrelevant information \[[@pbio-0030190-b18]\]. This seems reasonable, for although our memory capacity is impressive, it is not limitless \[[@pbio-0030190-b19],[@pbio-0030190-b25],[@pbio-0030190-b26]\], and we must possess and engage mechanisms that allow us to selectively filter information. In the current study, we provide evidence for neural correlates of this fundamental form of executive control in the avian brain. We recorded from the nidopallium caudolaterale (NCL) of homing pigeons. On the basis of behavioural/lesion \[[@pbio-0030190-b27],[@pbio-0030190-b28]\], neurochemical \[[@pbio-0030190-b29]\], and anatomical \[[@pbio-0030190-b30],[@pbio-0030190-b31]\] studies, the NCL is considered, much like the PFC for mammals, to be the main integrative and executive area of the avian brain. The pigeons were trained on a directed forgetting \[[@pbio-0030190-b32],[@pbio-0030190-b33]\] version of the delayed matching-to-sample task, a standard test of working memory used across a number of species, including birds, rats, monkeys, and humans. The birds were first shown a sample stimulus and then presented with either a remember cue or a forget cue ([Figure 1](#pbio-0030190-g001){ref-type="fig"}A and [1](#pbio-0030190-g001){ref-type="fig"}B; see [Materials and Methods](#s4){ref-type="sec"}). Following the remember or forget cue was a delay (memory) period. What followed the delay period was a function of whether the remember cue or the forget cue had been presented. If the remember cue was presented, then the delay was followed by a test period in which two stimuli were displayed, and a response to the stimulus that had appeared as the sample was rewarded. On the other hand, if the forget cue was presented, then following the delay there was no test period, and the trial terminated. Effectively, the remember cue instructed the bird that its memory for the sample would be tested after the delay and that it should therefore remember the sample stimulus, whereas the forget cue instructed the bird that there would be no test following the delay, and hence that it could forget the sample stimulus. Whether the cues were indeed instructing the birds to remember and forget the sample stimulus was tested with forget-probe trials ([Figure 1](#pbio-0030190-g001){ref-type="fig"}C), which will be discussed shortly. ::: {#pbio-0030190-g001 .fig} Figure 1 ::: {.caption} ###### Behavioural Task Sequence of events on (A) remember trials, (B) forget trials, and (C) forget-probe trials. On remember and forget-probe trials the birds were presented with a test period, whereas on forget trials the test period was absent. The three horizontally arranged circles represent the projectors on which the stimuli, in this case a circle and dot, were displayed. During the cue and delay periods, the projectors were turned off. ITI, intertrial interval; R, remember cue, a high-frequency tone; F, forget cue, a low-frequency tone. ::: ![](pbio.0030190.g001) ::: To the extent that sustained activation is a neural correlate of memory, and that the avian NCL is involved in executive control processes, we predicted that the activity of NCL neurons would be sensitive to cues to remember and forget. We found that neurons in the avian NCL showed sustained activation when the subject was instructed to remember, and the sustained activation was abolished when the subject was instructed to forget. Our findings indicate that neurons in the avian NCL participate in one of the fundamental forms of executive control, the control of what should be remembered and what should be forgotten. Results {#s2} ======= Histology {#s2a} --------- In two birds we recorded from both left and right NCL. For the three remaining birds, units were only recorded from either the left or right NCL. We recorded from a total of 124 NCL neurons. [Figure 2](#pbio-0030190-g002){ref-type="fig"} shows the electrode track reconstructions for the five subjects. With respect to the placements of the electrodes, the tracks were within the boundaries of NCL as defined by Kröner and Güntürkün \[[@pbio-0030190-b31]\]. All tracks were within 0.75 mm of the desired AP + 5.5 location (range AP + 5.25 to AP + 6.25), and all tracks were within 0.5 mm to the desired ML ± 7.5 location (range ML ± 7.0 to ML ± 8.0). We found no evidence that the characteristics of the neurons differed between the left and right hemispheres, along the dorsal and ventral extent of the NCL, or from one bird to the other. We have therefore collapsed the results across these variables. ::: {#pbio-0030190-g002 .fig} Figure 2 ::: {.caption} ###### Histology \(A) Lateral view of the pigeon brain. The NCL is shaded in red. The black line represents the intended electrode trajectory. \(B) Histological reconstruction of the electrode tracks for the five pigeons. The NCL is shaded in red. The black lines represent the electrode track. All tracks were within the boundaries of the NCL. APH, area parahippocampalis; CDL, area corticoidea dorsolateralis; Hp, hippocampus; NC, nidopallium caudale; SGC, stratum griseum centrale; TrO: tractus opticus; V: ventricle. ::: ![](pbio.0030190.g002) ::: Incidence and Type of Delay Activity {#s2b} ------------------------------------ A neuron was defined as a delay neuron if the level of activity during the delay period, when memory was required, was significantly different from the level of activity during the intertrial interval period, when memory was not required (paired t-test, see [Materials and Methods](#s4){ref-type="sec"}, Data analysis). Of the 124 NCL cells, 83 (66.9%) were classified as delay neurons. For 30 of the 83 cells, delay activity occurred after only one of the two to-be-remembered (sample) stimuli used on the memory task; these cells therefore contributed one instance each of delay activity for subsequent analysis. The remaining 53 cells exhibited delay activity after both of the to-be-remembered stimuli; these cells therefore contributed two instances each of delay activity for subsequent analysis. In all, across the 83 delay cells there was a total of 136 instances of delay activity. Modulation of Delay Activity by Remember and Forget Cues {#s2c} -------------------------------------------------------- What effect did the remember and forget cues have on delay activity? In 120 of the 136 (88.2%) instances of delay activity, sustained activation was found only on remember trials, when memory was needed to solve the task. On forget trials, when no memory was required, the sustained activation was abolished. Examples of three neurons that exhibit this effect are shown in [Figure 3](#pbio-0030190-g003){ref-type="fig"}. In each case, the remember cue was followed by a high level of activity that persisted beyond the cue period into and throughout the delay period. In contrast, the forget cue triggered an immediate decrease in activity during the cue period to intertrial interval (baseline) levels, and this decrease in activity persisted throughout the delay period. In short, following instructions to remember the sample stimulus, the cells exhibited sustained activation throughout the cue and delay periods, whereas following instructions to forget the sample stimulus, the sustained activation was abolished. Again, this effect was observed in the vast majority of instances of delay activity. Note that the pattern of data is not affected if we average the delay activity across the two sample stimuli and think of each cell as contributing only one instance of delay activity. In this case, of the 83 delay neurons, 63 (75.9%) showed the effect, that is, sustained activation on remember trials but not on forget trials. ::: {#pbio-0030190-g003 .fig} Figure 3 ::: {.caption} ###### Modulation by Remember and Forget Cues Examples of remember and forget cues modulating neural activity in three delay neurons from three different birds. The cue and delay periods are shaded in grey. On remember trials, there is sustained activation in the cue and delay periods. On forget trials, the sustained activation is abolished. The binwidth is 50 ms. The vertical dashed lines separate the different periods of the task. ITI, intertrial interval; S, sample period. ::: ![](pbio.0030190.g003) ::: Population Response {#s2d} ------------------- Delay activity was further classified as excitatory or inhibitory, referring to an increase or decrease in activity relative to intertrial interval levels. Of the 83 delay neurons, 22 were inhibitory (10 contributing one instance of delay activity and 12 contributing two instances of delay activity) and 61 were excitatory (20 contributing one instance of delay activity and 41 contributing two instances of delay activity). In total, across the 136 instances of delay activity we encountered 102 instances of excitatory delay activity and 34 instances of inhibitory delay activity. The average response profile across all 102 instances of excitatory delay activity is shown in [Figure 4](#pbio-0030190-g004){ref-type="fig"}. After a brief response to the remember cue, the population maintained high levels of activity for the rest of the cue and delay periods. In contrast, on forget trials, the response to the forget cue was followed by a drop in activity to baseline levels in the cue period, which persisted into the delay period. A repeated-measures ANOVA applied to all 102 instances of excitatory delay activity, with cue (remember vs. forget) and bins (100, 6, 40, and 60 bins of 50 ms in the intertrial interval, sample, cue, and delay periods, respectively) as factors, with repeated measures over bins, confirmed a significant difference in activity between remember and forget trials during the cue period, F(1, 202) = 24.3, p \< 0.001, and the delay period, F(1, 202) = 16.4, p \< 0.001, but not during the intertrial interval period, F(1, 202) = 0.07, p = 0.80, and sample period, F(1, 202) = 0.25, p = 0.62. A significant difference in activity in the delay period following remember and forget cues was also observed for the 34 instances of inhibitory delay activity, F(1, 66) = 5.4, p \< 0.05. In these cases of inhibitory delay activity, the activity remained depressed after the remember cue, but was elevated to baseline levels after the forget cue. ::: {#pbio-0030190-g004 .fig} Figure 4 ::: {.caption} ###### The Response Profile of Excitatory Delay Neurons The response profile of all 102 instances of excitatory delay activity on remember and forget trials. To account for differences in firing rates between the neurons, each neuron\'s firing rate was normalized against its baseline firing rate. The cue and the delay periods are shaded in grey. The vertical dashed lines separate the different periods of the task. ITI, intertrial interval; S, sample period. ::: ![](pbio.0030190.g004) ::: Behavioural Evidence of Forgetting {#s2e} ---------------------------------- If sustained activation represents a neural correlate of active remembering, then the lack of such activity after the instruction to forget must be a neural correlate of active forgetting. The forget cue, however, acquires its function because it predicts the absence of the test period. If the birds are not tested after the forget cue, then how is it possible to know that the forget cue has directed them to forget the sample stimulus? We examined this issue by testing the birds occasionally with probe trials in which, against its prediction, the forget cue was followed by a test period. An example of such a forget-probe trial is shown in [Figure 1](#pbio-0030190-g001){ref-type="fig"}C (see [Materials and Methods](#s4){ref-type="sec"}). The birds were tested only rarely with these forget-probe trials. The reason for not testing more often with forget-probe trials is that, if we had done so, the ability of the forget cue to predict the absence of a test period would have been jeopardized. Performance on sessions with forget-probe trials is shown in [Figure 5](#pbio-0030190-g005){ref-type="fig"}A. On every one of the sessions, performance on the forget-probe trials was lower than performance on the remember trials. [Figure 5](#pbio-0030190-g005){ref-type="fig"}B shows the performance on remember and forget-probe trials averaged across the four birds that were tested with probe trials. The average performance after the forget cue (43.8%) was significantly lower than the average performance after the remember cue (78.5%), t(3) = 3.31, p \< 0.05. In addition, a one-sample t-test evaluated against 50% indicated that performance on forget-probe trials was not significantly different from chance, t(3) = 1.00, p = 0.39. The chance levels of performance on forget-probe trials indicates that the forget cue is indeed directing the subject to forget the sample stimulus, a finding in line with other studies using the directed forgetting paradigm \[[@pbio-0030190-b32],[@pbio-0030190-b33]\]. ::: {#pbio-0030190-g005 .fig} Figure 5 ::: {.caption} ###### Performance on Probe Trial Sessions \(A) The performance on each of the eight sessions with forget-probe trials. \(B) The performance on the probe sessions (±1 SE) averaged across all four birds. The dotted line represents chance levels of performance. ::: ![](pbio.0030190.g005) ::: Discussion {#s3} ========== Summary of Findings {#s3a} ------------------- To summarize our findings, approximately 67% of the neurons we sampled in the NCL exhibited sustained activation during the delay period of the working memory task when the animal was required to engage memory processes. For the vast majority (88%) of the instances of delay activity, the sustained activation was modulated by instructions to either remember or forget. Instructions to remember resulted in sustained activation during the cue and delay portions of the memory task. In contrast, following instructions to forget, the sustained activation was rapidly abolished, and the neural activity returned to baseline levels. The high levels of performance following remember trials confirmed that the remember cue was directing the subject to remember the sample stimulus. In addition, the chance levels of performance on the forget-probe trials, in which the birds were cued to forget but then actually given a retention test, confirmed that the forget cue was directing the subject to forget the sample stimulus. Significance of Current Findings {#s3b} -------------------------------- The neurons in our study fired when the birds were told to remember, and they stopped firing when the birds were told to forget. We conclude that neurons in the avian NCL participate in one of the fundamental forms of executive control, the control of what should be remembered and what should be forgotten. Naturally, it is hard to know whether it is the NCL neurons that are performing the executive control function, or whether we are observing the effects on NCL neurons of executive functions that lie elsewhere in the brain. Given the evidence that we will review shortly that NCL may be an analogue of the mammalian PFC, and given the role of the PFC in executive control, we believe the former is the more likely case. This cause-and-effect issue aside, the need for such a filtering mechanism should be obvious. Although our memory capacity is impressive, it is not limitless \[[@pbio-0030190-b19],[@pbio-0030190-b25],[@pbio-0030190-b26]\]. We must have the ability, therefore, to filter information, allowing access to memory or retaining in memory that which is relevant, while restricting access to memory or discarding from memory that which is not. Our data are the first example of neural correlates of executive control in a nonmammalian species. We would also argue that they are the most straightforward example of neural correlates of executive control reported in any species thus far. Using a delayed matching-to-sample task in which monkeys were cued to attend to certain visual information in an array of visually presented stimuli, Rainer et al. \[[@pbio-0030190-b19]\] found that the delay activity of PFC neurons was dominated by information from the attended rather than the unattended target. On the basis of these findings they concluded that "we hold in working memory that to which we attend" (p. 578). We take these findings one step further by showing that even after attending to a stimulus we can still make the executive decision to allow information to remain in working memory, or to discard that information from working memory. Our findings show that remembering and forgetting are both well controlled and active processes that can be engaged and disengaged at any time \[[@pbio-0030190-b16],[@pbio-0030190-b34]\]. Avian NCL and Mammalian PFC: Homologue or Analogue? {#s3c} --------------------------------------------------- The NCL is a multimodal telencephalic region located in the posterior forebrain of birds. Divac and colleagues \[[@pbio-0030190-b28],[@pbio-0030190-b29]\] were the first to suggest that the avian NCL might correspond to the mammalian PFC. Naturally, given almost 320 million years of independent evolution \[[@pbio-0030190-b35]\], there are some differences between the two structures. Most notably, the NCL is neither cortex nor in the frontal (i.e., anterior) part of the brain, but then topographical location is a poor criterion for comparing structures \[[@pbio-0030190-b36]\]. Although the NCL and PFC are not homologous structures \[[@pbio-0030190-b36]\], the evidence we review below does suggest that they are analogous structures. Both NCL and PFC are ideally situated to integrate sensory information and translate that information into action. Anatomically, the NCL and PFC have similar patterns of afferent and efferent connections \[[@pbio-0030190-b31]\]. Both receive projections from modality-specific secondary visual, auditory, and somatosensory areas \[[@pbio-0030190-b31],[@pbio-0030190-b37]--[@pbio-0030190-b39]\], and both project to motor and limbic areas of the brain \[[@pbio-0030190-b31],[@pbio-0030190-b37]\]. In addition, both the NCL and PFC receive dense dopaminergic innervation from midbrain structures \[[@pbio-0030190-b29],[@pbio-0030190-b40],[@pbio-0030190-b41]\]. There are, of course, anatomical differences between the NCL and the PFC. One of the hallmark characteristics of PFC in primates is that it receives projections from the mediodorsal (MD) nucleus of the thalamus \[[@pbio-0030190-b42],[@pbio-0030190-b43]\]. In birds, the main thalamic projection to the NCL is the nucleus dorsolateralis posterior thalami (DLP) \[[@pbio-0030190-b44]\]. The DLP, however, is not generally considered homologous to the MD, nor is it identical to the MD in all of its connections \[[@pbio-0030190-b27]\]. Nevertheless, there is evidence that the DLP may serve the same functions as the MD \[[@pbio-0030190-b27]\], and that therefore the DLP--NCL system may be analogous to the MD--PFC system. There is also considerable behavioural/lesion data to support the contention that NCL and PFC perform similar functions. Damage to the NCL and PFC results in impairments on delayed alternation and pattern-reversal tasks while having little or no effect on simultaneous visual discriminations and basic sensory processes \[[@pbio-0030190-b02],[@pbio-0030190-b27],[@pbio-0030190-b28],[@pbio-0030190-b45]--[@pbio-0030190-b47]\]. In addition, blockade of D1 receptors in NCL and PFC both cause impairments on tasks sensitive to PFC and NCL damage \[[@pbio-0030190-b48]--[@pbio-0030190-b50]\]. Finally, the response profiles of NCL neurons are similar to those found in PFC \[[@pbio-0030190-b51],[@pbio-0030190-b52]\]. In summary, despite the anatomical and functional differences, which are not limited to comparisons between birds and mammals, but also include comparisons between one mammal and another \[[@pbio-0030190-b53]\], we believe there is considerable evidence to support the view that the avian NCL is analogous to the mammalian PFC. What Is Being Coded by Delay Activity? {#s3d} -------------------------------------- Early studies of delay activity tended to be of the view that it represented a neural correlate of the to-be-remembered sample stimulus \[[@pbio-0030190-b12]--[@pbio-0030190-b14]\]. We have cast our findings along similar lines, and viewed the presence and absence of delay activity on remember and forget trials as neural correlates of the birds\' remembering and forgetting the sample stimulus. The view that delay activity represents memory of the sample stimulus, however, has not gone unchallenged \[[@pbio-0030190-b54]--[@pbio-0030190-b57]\]. Recently, Lebedev et al. \[[@pbio-0030190-b58]\] have provided compelling evidence that delay activity, at least in PFC, represents more than just "maintenance memory." By pitting attention and memory against each other, they reported that PFC delay activity more likely represents the location that the subject was attending to, rather than the location that the subject was remembering. It is important to bear in mind that although we discuss our delay activity in terms of the subjects\' remembering and forgetting the sample stimulus, there are other, equally likely, interpretations of the delay activity. For example, the remember and forget cues are also informing the subject about the prospect of obtaining a reward \[[@pbio-0030190-b59]\]: the remember cue tells the subject that there is the opportunity for a reward (assuming a correct response is made), and the forget cue tells the subject that no reward is forthcoming. The presence of sustained activation during remember trials, and the lack of such activity on forget trials, could therefore also reflect the neuron\'s coding for the possibility of a reward, and the absence of a reward, respectively. Similarly, along the lines suggested by Lebedev et al. \[[@pbio-0030190-b58]\], the remember and forget cues could also be informing the subject about where to direct attention in the chamber. Thus the presence and absence of sustained activation on remember and forget trials could therefore also represent a neural correlate of the subject attending to the spatial position where the choice stimuli will appear, and not attending to any particular spatial location, respectively. The fact that sustained activation may serve as a code for the sample stimulus \[[@pbio-0030190-b52]\], aspects of the impending reward \[[@pbio-0030190-b60]--[@pbio-0030190-b62]\], or spatial attention \[[@pbio-0030190-b58]\] raises another important issue about PFC function. Many studies show that it is a combination of features that activate PFC neurons \[[@pbio-0030190-b61],[@pbio-0030190-b62]\], a finding that may not be too surprising, given the known multimodal nature of the PFC \[[@pbio-0030190-b37],[@pbio-0030190-b63]\], and a point that fits well with the established view that memory consists of a collection of attributes \[[@pbio-0030190-b64],[@pbio-0030190-b65]\]. The point we are trying to emphasize is that even though delay activity correlates with aspects of the reward, attentional factors, and the memorandum, this does not weaken the notion that delay activity is a neural correlate of working memory, nor does it weaken the notion that the remember and forget cues are modulating the contents of working memory. Against this backdrop of PFC neurons coding for attributes of working memory other than the sample stimulus, it might be interesting to speculate on our observation that the majority of NCL delay activity was nonselective in that it tended to occur after both (63.9%) rather than just one (36.1%) of the stimuli used on the memory task. A similar preponderance of nonselective delay activity has been observed in the hippocampus \[[@pbio-0030190-b15]\], which like the PFC, is also a zone of multimodal convergence in the brain \[[@pbio-0030190-b37]\]. In contrast, the majority of delay activity in the inferior temporal cortex, a higher-order area of the primate brain dedicated exclusively to visual processing, is selective, in that it tends to occur after only one rather than both of the stimuli used on the memory task \[[@pbio-0030190-b15]\]. It may be the case that delay activity in modality-specific visual areas may reflect more a code of the sample stimulus, whereas delay activity in multimodal areas like the PFC and NCL may reflect the other attributes of working memory, such as spatial attention and reward anticipation. The Value of Comparative Research {#s3e} --------------------------------- Is it necessary to invoke higher-order executive control mechanisms to explain the behaviour of pigeons in the current study? And in doing so, have we not dispensed with Occam\'s razor? We understand that it might be uncomfortable for some to accept that pigeons have executive control mechanisms. But if studies of animal cognition have shown us anything it is that behaviours once thought to be the exclusive domain of humans can be seen in many nonhuman (and nonmammalian) species \[[@pbio-0030190-b66]\]. The often heard retort to these examples of cognitive equivalencies is that just because you can mimic the behaviour of one species in another does not mean that the behaviour of both species is governed by the same mechanism. While true, the fact that you can get a bird to exhibit behaviour like that of a human could just as easily make us question how humans are solving the task, as make us question how birds are solving the task. The bottom line is that one would be hard pressed to argue against the view of mental continuity across species \[[@pbio-0030190-b66]\], and so if we accept executive processes in humans, it is very likely that executive processes exist in nonmammalian species as well. What then are the implications of showing neural correlates of executive control in the avian brain? For one, it would appear that the ability to filter relevant and irrelevant information, what some have argued is a fundamental form of executive control \[[@pbio-0030190-b18]\], is clearly not an exclusive feature of the mammalian brain. However, we believe our data speak to another issue, one that emphasizes the importance of conducting comparative work. There is little doubt that it is commonplace to richly interpret human behaviour. But when you show that a pigeon can recognize itself in a mirror \[[@pbio-0030190-b67]\], or that a crow can manufacture a tool \[[@pbio-0030190-b68]\], it is not just an interesting demonstration, but it also constrains and refines our views regarding the neural mechanisms that underlie such behaviour. When you show that the avian hippocampus is in every respect an analogue of the mammalian hippocampus \[[@pbio-0030190-b69]\], yet at the same time structurally totally different \[[@pbio-0030190-b70]\], it is not just idle curiosity, but it also requires that we take this into consideration when we devise computational models of hippocampal function. And when you show that neurons in the avian brain engage in executive control processes, one must then wonder how much of what makes us such majestic creatures is present in an organism whose behaviour rarely inspires the attribution of high cognitive skills. Materials and Methods {#s4} ===================== {#s4a} ### Subjects. {#s4a1} The subjects were 5 homing pigeons (Columba livia) weighing approximately 400--650 g. The animals were housed individually in wire-mesh cages inside a colony room, had free access to water and grit, and were maintained on a 12:12-h light:dark cycle with lights on at 0700. The pigeons were fed a mixture of wheat, peas, and corn in an amount adjusted to maintain them at 80% of their free-feeding body weight. The experiments were approved by the University of Otago Animal Ethics Committee and conducted in accordance with the University of Otago\'s Code of Ethical Conduct for the Manipulation of Animals. ### Apparatus and stimuli. {#s4a2} All training and testing was conducted in standard sound attenuated operant chambers. Situated on the front panel of each chamber were three horizontally-arranged clear plastic circular response keys, each 2.5 cm in diameter. All three keys were mounted 22 cm above the floor and were 10 cm apart from centre to centre. Situated behind each key was a stimulus projector (IEE model 1071; Institution of Electrical Engineers, London, United Kingdom) used to deliver the visual stimuli. Food reward (wheat) was delivered via an illuminated magazine situated below the centre key. The stimuli consisted of standard geometric shapes (circle and dot) and two colours (red and white). The geometric shapes appeared as white forms against a black background. The circle was composed of a line that was 1.5 mm thick and was 17 mm in diameter, whereas the dot was solid and 7 mm in diameter. The colour stimuli consisted of illumination of the entire 25-mm diameter response key with the colour red or white. The presentation of the stimuli, reward, and punishment contingencies, and all data recording were controlled by Pentium II computers attached to the chamber. ### Behavioural task. {#s4a3} The birds were trained on a directed forgetting version of the delayed matching-to-sample task. The sequence of events on a typical trial is shown in [Figure 1](#pbio-0030190-g001){ref-type="fig"}. Each bird was trained with two stimuli, either a circle and a dot, or the colours red and white. All trials began with a 15-s intertrial interval, followed by the sample period, during which one of two stimuli was displayed on the centre of three keys. Following three pecks to the sample stimulus, the sample was turned off. This was followed by a cue period in which either the remember or forget cue was presented for 2 s. A delay period of 3 s followed the cue period. If the remember cue had been presented ([Figure 1](#pbio-0030190-g001){ref-type="fig"}A), the delay was followed by a test period in which both stimuli were displayed. A single response to the stimulus that had been presented in the sample period was rewarded with wheat, whereas a single response to the other stimulus was punished with a 30-s time-out. If the forget cue had been presented ([Figure 1](#pbio-0030190-g001){ref-type="fig"}B), the trial ended after the delay period without a test period. A session consisted of 96 trials, 48 remember trials and 48 forget trials, randomly mixed. With two stimuli (A and B), there are four possible trial types: sample A can be followed by A on the left and B on the right, or B on the left and A on the right, likewise for sample B. Which of the two stimuli (e.g., A or B) was presented during the sample period, the position of the two stimuli on the side keys during the test period (e.g., A B or B A), as well as the sequence of remember and forget cues, was balanced within a session. For different birds, the remember and forget cues were either a high-frequency tone (5,000 Hz) and a low-frequency tone (500 Hz), or a four-lobed visual pattern (two figure eights superimposed on each other) and a 16-lobed visual pattern (resembling a sun). The four-lobed and 16-lobed patterns appeared as white shapes against a black background and were each 17 mm in diameter. Approximately after every 15th recording session the birds were tested with forget-probe trials ([Figure 1](#pbio-0030190-g001){ref-type="fig"}C). These forget-probe trials were administered to four of the five birds; bird T18 did not receive a probe trial session because we lost the ability to record from this bird after the 11th recording session. These probe trials allowed us to examine the effectiveness of the forget cue as an instruction to forget the sample stimulus. Recall that on standard forget trials, following the delay period there was no test period, so it is impossible to know whether the subject had indeed forgotten the sample stimulus. On a forget-probe trial, against its usual prediction, the forget cue was followed by a test period. A probe trial session contained four forget-probe trials in addition to the 48 remember and 48 forget trials. All aspects of the stimulus and cue presentations were balanced within a session. ### Training protocol. {#s4a4} Although all of the birds had been trained on a simultaneous matching-to-sample task, none had received any delayed matching-to-sample experience at the start of this experiment. Delayed matching-to-sample training was divided into two phases. In phase 1, the birds were first trained with a 0-s delay followed by training with a 0.5-s delay. The birds were required to achieve a performance level of two consecutive sessions at or better than 90% correct at each delay. To complete phase 1 the birds required an average of 43 sessions (range 24--63 sessions) and accumulated an average of 1,987 trials (range 1,297--3,264 trials). In phase 2 the birds were introduced to the directed forgetting procedure, which used the same delayed matching-to-sample task but with remember and forget cues. The birds were gradually trained with increasing delays and increasing session lengths until they were able to perform the task with a 5-s delay (2-s cue period and 3-s delay period) and 96 trials per session. To complete phase 2 the birds required an average of 214 sessions (range 148--236 sessions) and accumulated an average of 14,545 trials (range 11,088--16,268 trials) on the task. Note that half of the trials accumulated during phase 2 were forget trials. ### Surgery. {#s4a5} Upon completion of behavioural training, the birds were prepared for alert recording by implanting a miniature movable microdrive \[[@pbio-0030190-b71]\]. Surgery was conducted under ketamine (225 mg/kg) and xylazine (5 mg/kg) anaesthesia. The head was immobilized using a Revzin stereotaxic adapter \[[@pbio-0030190-b72]\]. A topical anaesthetic (10% Xylocaine) was applied to the scalp, which was then cut and retracted to expose the skull. A small hole above the NCL was drilled through the skull at AP + 5.5 and ML ± 7.5 \[[@pbio-0030190-b72]\], and the microdrive then lowered so that the tips of the electrodes were positioned just above the NCL. Stainless steel skull screws, one serving as a ground screw, were placed into the skull, and the entire microdrive was attached to the skull using dental acrylic. The incision was then sutured, Xylocaine applied to the wound margin, and the animal allowed to recover in a heated, padded cage until alert and mobile, at which point it was returned to its home cage. All animals were allowed to recover for 7--10 d prior to the start of recording. ### Neuronal recording. {#s4a6} The microdrive housed eight 25-μm Formvar-coated nichrome wires that were used to measure the extracellular activity of single neurons. All signals were first impedance matched through a FET headstage and then amplified and filtered to remove 50 Hz noise using Grass P511K preamplifiers (Grass Instruments, Quincy, Massachusetts, United States). A separate electrode with minimal activity served as the indifferent electrode. The signals were monitored with an oscilloscope and speaker. Behavioural time-tagging of all events and analysis of the spike data was accomplished using a CED 1401 plus system (Cambridge Electronic Design Limited, Cambridge, United Kingdom) and CED Spike 2 software. The only criterion for the selection of a neuron was that it was well isolated with a signal-to-noise ratio of at least 2:1. After isolating a neuron, the delayed matching-to-sample task was started. A typical session lasted approximately 45 m to 1 h. The pigeons were tested once a day. At the end of the recording session, the electrodes were advanced at least 40 μm and the animal returned to its home cage. ### Histology and electrode track reconstruction. {#s4a7} Upon completion of the experiment, the final electrode position was marked by passing a current through each electrode, thus creating a small electrolytic lesion. The pigeons were then deeply anaesthetised with halothane and perfused through the heart with physiological saline followed by 10% formalin. The brains were blocked, removed, placed in 10% formalin for 5 d, placed in 30% sucrose and 10% formalin, and allowed to sink twice. The brains were then frozen and sectioned at 50 μm, with every section mounted and stained with cresyl violet. The positions of the recorded neurons were calculated from the electrode track reconstructions, position of the electrolytic lesion, and depth records. ### Data analysis {#s4a8} Only correct trial data were analyzed. For initial assessment of delay activity, for each neuron the average activity in the middle 5 s of the intertrial interval was compared to the average activity in the delay period after each of the two sample stimuli using paired t-tests with a modified \[[@pbio-0030190-b73]\] Bonferroni correction (p \< 0.025). The population data were subjected to a repeated-measures ANOVA with Greenhouse--Geisser correction. We thank Nicole Frost for assistance with data collection, Onur Güntürkün for providing the lateral view of the pigeon brain, and Neil McNaughton, Geoff White, David Bilkey, Olivier Pascalis, Charlie Gross, Harlene Hayne, and Andrew Iwaniuk for commenting on drafts of the manuscript. This research was supported by a Royal Society of New Zealand Marsden Grant UOO012 to M. Colombo. Competing interests. The authors have declared that no competing interests exist. Author contributions. MC conceived and designed the experiments. JR performed the experiments. MC and JR analyzed the data and wrote the paper. Citation: Rose J, Colombo M (2005) Neural correlates of executive control in the avian brain. PLoS Biol 3(6): e190. DLP : dorsolateralis posterior thalami MD : mediodorsal NCL : nidopallium caudolaterale PFC : prefrontal cortex	PubMed Central	2024-06-05T03:55:56.411387	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088974/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e190", "authors": [ { "first": "Jonas", "last": "Rose" }, { "first": "Michael", "last": "Colombo" } ] }
PMC1088975	In 1972, John Kerr, Andrew Wyllie, and Alistair Currie introduced the word "apoptosis" (from the Greek for apples falling off trees) to describe a special form of cell death. Apoptosis, or programmed cell death, is a normal physiological process in which specific signals trigger cells to self-destruct in a carefully choreographed manner. Without apoptosis, we would have paddles for hands, rather than individual fingers. And without apoptosis, we would be at much greater risk of developing cancer, since apoptotic mechanisms destroy cells that have taken the first steps toward tumor formation. Cells apoptose when the positive signals needed for their survival are withdrawn or when negative signals tell the cell to commit suicide. A cell undergoing apoptosis shrinks and develops "blebs"---small bubbles---on its surface, its mitochondria break down, and its genomic DNA breaks into fragments. Finally, the dead cell is engulfed by phagocytic cells, thus avoiding any inflammation in surrounding tissues. Apoptosis can be triggered by external and internal signals, but in both the extrinsic and intrinsic pathway of apoptosis, once a signal has been received, the next molecular step is caspase activation. These proteases, which cleave proteins at specific sites, fall into two classes. Initiator caspases activate themselves before proteolytically activating the effector (or executioner) caspases, which degrade numerous cellular proteins, thus causing cell death. All the caspases are made as inactive enzymes (zymogens); they are much too dangerous to be stored as active enzymes.[](#pbio-0030198-g001){ref-type="fig"} ::: {#pbio-0030198-g001 .fig} ::: {.caption} ###### Experiments with artificially joined molecules of caspase 9---which normally exists as monomers (above)---suggest that caspase 9 can\'t activate cell death pathways by dimerization alone ::: ![](pbio.0030198.g001) ::: Yigong Shi and colleagues are studying the mysterious process of initiator caspase activation---the molecular mechanism of effector caspase activation is fairly well understood. The autocatalytic activation of caspase-9, an initiator caspase in the intrinsic pathway of apoptosis, is mediated by the assembly of the apoptosome, a heptameric complex containing Apaf-1 (apoptotic protease activating factor 1) and cytochrome c. Caspase-9 activation is currently explained by the induced proximity model in which the apoptosome, by increasing the local concentration of caspase-9, promotes its homodimerization (the formation of a complex containing two caspase-9 molecules) and subsequent autoactivation. To test this model, Shi and coworkers engineered caspase-9 so that it exists in the cell all the time as a homodimer---wild-type caspase-9 usually exists as monomer---and then examined its catalytic activity. They found that although the engineered, dimeric caspase-9 was more active in in vitro assays and induced more cell death when expressed in cells than the wild-type enzyme, its activity was only a fraction of that of Apaf-1-activated wild-type caspase-9. Furthermore, its activity was not stimulated by Apaf-1, unlike that of the wild-type enzyme. Importantly, they also show that the crystal structure of their engineered caspase-9 closely resembled that of wild-type caspase-9, indicating that the changes they made did not cause any significant changes in the protein. Overall, their results led the researchers to suggest that the dimerization of caspase-9 may be qualitatively different from the Apaf-1-mediated activation of caspase-9, and that dimerization may not be the major mechanism behind the activation of caspase-9. Instead, they suggest that a shape (conformational) change is induced in caspase-9 when it binds to the apoptosome and that this change drives its activation. This "induced conformation" model provides an alternative model to the induced proximity model for initiator caspase activation, but, as the researchers note, the two models need not be mutually exclusive. Indeed, only the determination of high-resolution structure of the apoptosome will unravel exactly how caspase-9 and other initiator caspases are activated.	PubMed Central	2024-06-05T03:55:56.415429	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088975/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e198", "authors": [] }
PMC1088976	After the wounded soldier in Dalton Trumbo\'s Johnny Got His Gun slowly realizes he has no limbs, no face, ears, eyes, or mouth---but isn\'t dead because his mind continues to think---he desperately tries to find a way to reclaim his humanity. He decides to track time: he counts to sixty, files that minute "in one side of his mind ... and \[begins\] counting from one to sixty again." When he starts to wonder whether he\'s counting at the right speed, his mind slips "off track" and his figures disappear. People have been known to exhibit enhanced memory in extraordinary circumstances, but human memory has its limits. Short-term retention of information, called working memory, involves doing at least two things simultaneously: paying attention to the task at hand---in the soldier\'s case, counting---while processing that information and deciding how to handle it---adding the numbers and keeping a running total. The secondary processing of stored information is referred to as executive control. Many studies suggest that the neural seat of both working memory and executive control---which together encompass planning, creativity, reasoning, abstraction, and most of the other higher-order cognitive properties humans like to claim as their own---lies within the prefrontal cortex. Teasing out the neural components of these overlapping processes has proven challenging. In a new study, Jonas Rose and Michael Colombo investigate the neural basis of executive control by training homing pigeons to remember or forget a visual stimulus. Recording from the nidopallium caudolaterale (NCL), a region of the avian brain considered analogous to the mammalian prefrontal cortex, the authors show that neurons in the NCL selectively fire when the birds are told to remember and stop firing when they are told to forget. To test the hypothesis that the NCL plays a role in executive control, the authors trained five pigeons on a directed forgetting test, a variation on the classic match-to-sample test. After viewing sample stimuli consisting of one of two shapes (a circle or dot) or colors (red or white), the birds were cued to remember or forget the sample (signaled by either a high- or low-frequency tone or one of two distinct patterns). A delay period followed these cues. If a forget cue was presented, the trial ended after the delay, and no memory test was given. If the remember cue was presented, the birds were given a memory test in which they saw two stimuli after the delay; if they responded to the sample stimulus (by pecking on a key), they were rewarded with wheat. The pigeons\' NCL activity was recorded during the trials. Eighty-three of the 124 recorded neurons were classified as delay neurons because they showed significantly different activity during the delay period, when memory was required, than during the intervals between trials, when it was not. During the remember trials, neurons showed sustained activation throughout the cue and delay periods; during the forget trials, sustained activation disappeared. To make sure the forget cue was indeed directing the birds to forget the sample stimulus, the authors ran the forget trials again, but this time, tricked the birds and gave them a memory test. The birds consistently performed worse on the forget trials than on the remember trials, confirming the forget cue\'s effect. These results suggest that sustained NCL neuronal activation reflects working memory or at least some type of cognitive activity associated with a working memory task. Either way, these findings support the notion that NCL neurons play a role in executive control---what to remember and what to forget---by linking the presence or absence of neuronal activity with remembering and forgetting. And though the avian NCL and mammalian prefrontal cortex clearly differ after 320 million years of divergent evolution, Rose and Colombo make a strong case that they are similar enough to support the NCL\'s likely contribution to executive control in mammals as well. And they suggest that seeing such similarities between the bird and human brain forces us to reexamine not only our notions of how these structures operate but also our hubris in thinking our biology and nature is unique.[](#pbio-0030216-g001){ref-type="fig"} ::: {#pbio-0030216-g001 .fig} ::: {.caption} ###### Neurons in the avian version of the prefrontal cortex fire when pigeons perform memory tasks and may play a role in higher cognitive processes such as decision making and reasoning ::: ![](pbio.0030216.g001) :::	PubMed Central	2024-06-05T03:55:56.416051	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088976/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e216", "authors": [] }
PMC1088977	Of the 200 or so diseases collectively known as cancer, tumors of the soft tissue include some of the rarest and most diverse. Malignant soft tissue tumors, or sarcomas, and benign soft tissue tumors---which together comprise over 100 subtypes---take root in immature nerve and mesenchymal cells, which give rise to muscle, fat, cartilage, and other connective tissues. Reflecting the ubiquitous nature of connective tissue, soft tissue sarcomas can occur nearly anywhere in the body. Pathologists classify soft tumor sarcomas based on the proteins they express and on their resemblance to normal connective tissue cells (such as adipocytes, smooth muscle cells, and fibroblasts): tumors consisting of cells with cytologic features of fat cells, for example, are called liposarcomas while those forming the spindle-shaped, organized sheets typical of smooth muscle cells are called leiomyosarcomas. But many connective cells appear too similar and express too many of the same proteins for traditional screens to distinguish among them. Assays are further complicated by the admixture of non-malignant cells, including inflammatory cells and those recruited to form new blood vessels, into the soft tissue tumor landscape. In a new study, Robert West, Matt van de Rijn, and their colleagues investigate the notion that different types of fibroblastic tumors mirror the features of normal fibroblasts and search for tissue markers that might distinguish them. Using DNA microarrays, the authors profiled gene expression patterns in two types of fibroblastic tumors and found significant differences in the expression of functionally related genes, confirming that each tumor carries a unique genetic signature. These gene sets also appear in the matrix of normal connective tissue, or stroma, leading to the identification of two noncancerous fibroblast subtypes. The tumors analyzed in this study---solitary fibrous tumor (SFT) and desmoid-type fibromatosis (DTF)---behave differently but consist of cells that look similar under the microscope, making them well suited to the task of identifying novel connective tissue markers. Typically benign, SFT tumors respond well to surgical excision and are thought to arise either from fibroblasts, most frequently in the thoracic cavity. DTFs, aggressive tumors found deep within the soft tissue of the trunk, abdomen, or extremities, are difficult to excise completely and are also thought to arise from fibroblasts. Gene arrays taken from ten DTF tumors and 13 SFT tumors showed that each tumor type had distinct gene expression patterns reflecting different gene functions. For example, DTF gene profiles included many genes involved in fibrosis (scarring) and extracellular matrix remodeling, a prerequisite for the invasive behavior of aggressive DTF tumors. SFTs, on the other hand, express many genes involved in synthesis and maintenance of the basement membrane that surrounds muscle cells, blood vessels, and other specialized cells. Based on these variable expression patterns, the authors hypothesized that the tumors\' cells of origin might perform different functions in normal tissue.[](#pbio-0030220-g001){ref-type="fig"} ::: {#pbio-0030220-g001 .fig} ::: {.caption} ###### Tissue samples at 600× magnification show that these two soft tissue tumors express different protein markers: solitary fibrous tumors express the APOD protein marker (top), and desmoid-type fibromatosis tumors express OSF2 ::: ![](pbio.0030220.g001) ::: Next, West et al. examined gene expression patterns in fibroblasts found in normal tissue samples to look for tumor-specific markers. As predicted, DTF markers were found in one set of tissues---related to scarring and inflammation---and SFT markers were found in another---breast and skin fibroblasts, and benign breast growths. The authors\' use of soft tissue tumors to define different subsets of stromal cells is similar to past studies that used lymphomas to discover novel subsets of normal lymphoid cells. Because normal fibroblasts can contribute to cancer progression---by providing the cellular matrix, or stroma, that supports tumor growth---the authors looked for DTF- or SFT-specific markers in the stroma of breast cancer patients. Two groups, the authors note, showed differences in the expression of DTF and SFT genes. Patients with tumors showing high DTF gene expression had the best prognosis, suggesting that tumor--stromal interactions might affect disease progression. The difficulty of classifying fibroblasts has complicated efforts to understand how fibroblasts aid tumor progression. This study suggests that breast cancers with different stromal signatures may have different clinical outcomes, which raises many questions for future study. If the same tumor type grows on a different stromal background, will it progress differently? If so, how do tumor--stromal interactions influence progression? Thanks to the fibroblast markers identified here, scientists have new tools for exploring these questions.	PubMed Central	2024-06-05T03:55:56.416631	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088977/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e220", "authors": [] }
PMC1088978	The development of antibiotic drugs changed the face of clinical medicine forever, and, for a short while at least, it seemed that a perfect cure for bacterial infections had been discovered. But widespread overuse and misuse of antibiotics over the last several decades has proved the Achilles heel of this once seemingly invincible class of drugs and has fostered bacterial resistance to conventional antibiotic therapies. New drugs are continually being developed to replace those crippled by resistance, but despite scientists\' best efforts, new "superbugs" are evolving faster than the drugs required to control them. For some time now, drug resistance has been considered a consequence of errors (called mutations) that accumulate spontaneously during replication of the bacterial genome. In many cases those mutations are either inconsequential or harmful to the bacteria, but on rare occasion, they provide an accidental benefit: resistance to the drugs that kill them. Because the mutations were assumed to be spontaneous, there was no obvious way to prevent them and thus antibiotic resistance appeared inevitable. But some researchers are taking a pro-active approach by testing the assumptions about how mutations are made. For one group of bacterial researchers, this approach may have paid off. Chemists at the Scripps Research Institute and the University of Wisconsin have uncovered evidence that spontaneous mutations are not the only way in which bacteria acquire resistance to antibiotics. It appears that the bacteria, rather than passively waiting around for a lucky break, may play an active role in their own evolution. The key is in the way antibiotics interact with their bacterial targets. Quinolone antibiotics are a relatively new class of antibiotics that work by interfering with proteins called topoisomerases, which assist DNA replication by loosening tightly wound DNA and making it accessible. In order to do this, the topoisomerase must break the DNA strands and fill in the gap with a temporary protein bridge. Under normal circumstances, the bridge is removed and the DNA is reconnected after the topoisomerase has done its job, but quinolones bind to this protein bridge and prevent the DNA from resealing. The freed double-strand ends signal that DNA damage has occurred and activate the cell\'s repair pathway. According to Ryan Cirz et al., DNA damage, induced by antibiotics or other stressors, sets off a bacterium\'s emergency repair mechanism: the SOS DNA damage response. Under normal conditions, the genes are turned off by a special repressor protein called LexA. In response to the damaged DNA, the LexA repressor is cleaved and no longer inhibits transcription of the SOS response genes. Cirz et al. propose that antibiotic-mediated DNA damage generates a reduction in the concentration of LexA that is sufficient to increase the expression of three nonessential DNA polymerases shown to be required for mutation: Pol II (encoded by the gene polB), Pol IV (encoded by dinB), and Pol V(encoded by umuD and umuC). Together these polymerases promote DNA repair---and cause mutations in bacterial DNA that can lead to antibiotic resistance. This suggests that quinolone antibiotics (and other antibiotics that cause similar kinds of DNA damage) may increase the likelihood that bacteria will evolve resistance and that new generations of drugs will have little chance of succeeding where today\'s drugs have failed. But all hope is not lost. After showing that the evolution of quinolone resistance depends on activating the SOS response genes gated by LexA, Cirz et al. go on to demonstrate that blocking LexA cleavage, in vitro and in a mouse model, prevents mutation and results in bacteria that are unable to evolve antibiotic resistance. Thus, developing novel therapeutic agents that target LexA or the associated SOS pathway may prove a promising strategy for controlling the spread of the superbugs.[](#pbio-0030221-g001){ref-type="fig"} ::: {#pbio-0030221-g001 .fig} ::: {.caption} ###### Inhibiting a protein involved in DNA repair in bacteria might prevent mutations that promote antibiotic resistance, prolonging the effectiveness of ciprofloxacin and other quinolone antibiotics ::: ![](pbio.0030221.g001) :::	PubMed Central	2024-06-05T03:55:56.417389	2005-5-10	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1088978/", "journal": "PLoS Biol. 2005 Jun 10; 3(6):e221", "authors": [] }
PMC1090442	Background ========== It is generally accepted that early experiences have profound influences on brain development and thus on adult brain function and behavior. However the neurobiological mechanisms involved still remain elusive. An animal model employed in experiments aiming to elucidate such mechanisms is \"neonatal handling\" \[[@B1]\]. This manipulation alters hypothalamic-pituitary-adrenal (HPA) axis function and the ability of the organism to respond to stressful stimuli \[[@B1]\]. Thus, as adults, neonatally handled rats are less emotionally reactive, synthesize and secrete less corticotropin-releasing factor, adrenocorticotropin hormone (ACTH) and corticosterone following a variety of stressors \[[@B2]\], and their stress-induced secretion is more short-lived \[[@B3]\]. These differences in HPA axis reactivity have been attributed to an enhanced sensitivity of the negative-feedback loop \[[@B2]\], due to a handling-induced increase in the number of type II glucocorticoid receptors (GR) in the hippocampus \[[@B2]\]. In addition to GR, glucocorticoids also bind to type I (MR) receptors, and the hippocampus is rich in both these types of receptors \[[@B4]\]. GR and MR receptors are the molecules mediating the negative feedback control exerted by glucocorticoids on HPA axis function \[[@B5]\]. Furthermore GRs and MRs influence spatial learning, a process controlled by the hippocampus \[[@B6]\]. MRs have a role in behavioral reactivity during novel situations \[[@B7]\], whereas GRs are involved in consolidation of learned information. In addition to GRs and MRs, glucocorticoid levels also play a determinant role in the ability for learning and memory. The effect of corticosteroid levels on cognition exhibits a U-shaped dose-response dependency \[[@B8]\]. Interestingly, as mentioned above, handled animals have lower corticosterone levels following stress \[[@B9]\], which could alter their ability for learning and memory. Another molecule that has been shown to play a key-role in the cellular processes underlying learning and memory is Brain Derived Neurotrophic Factor (BDNF), a member of the neurotrophin family \[[@B10]\]. BDNF mRNA is increased during LTP, indicating that BDNF is involved in plastic changes of neuronal function \[[@B11]\]. Memory acquisition is also associated with increased BDNF mRNA and activation of its receptor TrkB \[[@B12],[@B13]\]. On the other hand, LTP is markedly impaired in BDNF mutant mice and the deficit is restored by the re-expression of BDNF \[[@B14],[@B15]\]. Moreover, BDNF mutant mice show learning deficits \[[@B16]\]. Similarly, the pharmacologic deprivation of BDNF or its receptor TrkB, results in severe impairment of learning and memory in mice, rats and chicks \[[@B15]\]. BDNF mutant mice develop enhanced aggressiveness, and hyperphagia, accompanied with weight gain in early adulthood, findings reminiscent of dysfunction of the serotoninergic system \[[@B17]\]. Indeed BDNF is known to have trophic effects on serotoninergic neurons \[[@B18]\]. It is well known that depression is associated with hypofunctioning of the serotoninergic system. Recently BDNF has emerged as a major factor in the pathophysiology of depression: BDNF mRNA is increased in the rat brain following chronic anti-depressant or electro-convulsive shock treatment \[[@B19],[@B20]\]. Administration of BDNF in the hippocampus has been shown to have an anti-depressant effect in the forced swimming and learned helplessness paradigm \[[@B21]\]. Furthermore, in patients with major depression, serum BDNF levels were decreased, while hippocampal BDNF immunoreactivity was increased in post-mortem tissues from subjects treated with anti-depressants \[[@B22]\]. Previous results from our laboratory have shown that handled males exhibit decreased expression of \"depressive\" behavior \[[@B26]\]. Recent evidence indicates that among the serotonin receptors, the type 1A are involved in the etiopathogenesis of certain types of depression \[[@B23],[@B24]\] and is the one through which the therapeutic effects of the Selective Serotonin Re-uptake Inhibitors (SSRIs), a major class of antidepressants, are mediated \[[@B25]\]. Results from our laboratory have shown that handled male rats show increased 5-HT1A receptor sensitivity as assessed by the hypothermic response to 8-OH-DPAT compared to the non-handled \[[@B26]\]. Depression and the response to chronic stress are often associated with disorders in food-intake behavior, which is influenced by serotonin and, as mentioned above, by BDNF. A key hormone regulating food-intake behavior is leptin, the product of the obgene \[[@B27]\]. Leptin, whose levels reflect the organism\'s current energy balance, is secreted from adipose tissue proportionally to body fat mass and acts on the CNS to limit food intake, and thus promote body weight loss \[[@B28]\]. Recent evidence indicates that glucocorticoids induce leptin synthesis and secretion and that, conversely, leptin participates in the regulation of HPA axis function \[[@B29]\]. Thus, we investigated the effects of \"neonatal handling\" on factors influencing cognitive abilities and affective states of the adult rat. Specifically, we determined the \"neonatal handling\" effects on A. the ability for spatial learning and memory -in the Morris water maze- when a short-term restrain stress has preceded the learning process, B. GR and MR levels in the hippocampus after the completion of the Morris water maze test, C. BDNF levels in the hippocampus, D. hippocampal 5-HT1A receptor density and E. plasma leptin levels, food intake and body weight change during long term forced swimming stress. Methods ======= Animals ------- Male Wistar rats reared in our laboratory were kept under standard conditions (24°C; 12:12 h light/dark cycle; food and water ad libitum). All animal experimentations were carried out in agreement with ethical recommendation of the European Communities Council Directive of 24 November 1986 (86/609/EEC). In total, 43 handled and 45 non-handled (control) animals were used in this study. Neonatal handling ----------------- Pups were removed from their mothers and placed for 15 min in a plastic container lined with paper towel, daily from postnatal day 1 until weaning (postnatal day 22). The non-handled animals were left completely undisturbed until weaning. Restraint stress ---------------- Adult, handled and non-handled males were placed in a cylinder 15 cm in length and 5 cm in diameter for 30 min. Spatial learning and memory test -------------------------------- The Morris water maze (MWM) apparatus was a circular galvanized tank (1.38 m in diameter, 0.5 m in height), filled to a depth of 28 cm with water (24°C), made opaque with milk. The training session took place 90 min after the completion of the restraint stress. For this session a 2 cm submerged platform (13 × 13 cm) was placed in a fixed position. The single training session consisted of 8 trials with 4 different starting positions. After finding the platform, the animals were allowed to remain on it for 20 sec and were then placed in a holding cage for 30 sec until the beginning of the next trial. The testing trial was performed 24 hours later. It consisted of a 60 sec free swim period without a platform and was recorded on videotape. The rat was placed in the tank at a position directly opposite to the imaginary platform quadrant. Animals were sacrificed upon termination of the testing session and their brains were used for GR and MR immunocytochemistry. Immunocytochemistry ------------------- For the GR and MR immunocytochemistry the same animals were used, whereas for the BDNF immunocytochemistry a different set of animals was employed. All animals were deeply anesthetized with ether and perfused transcardially with 4% paraformaldehyde in 0.1 M phosphate buffer (PB). Immunocytochemistry was performed as previously reported \[[@B30]\] on paraffin, sagittal brain sections (6 μm). The primary antibodies used were an anti-BDNF rabbit polyclonal antibody (Santa Cruz) or an anti-MR goat polyclonal antibody (Santa-Cruz) or an anti-GR moloclonal antibody (kindly provided by Dr. Alexis, NHRF). The secondary antibodies were biotinylated goat anti-rabbit or rabbit anti-goat or rabbit anti-mouse antibody respectively (DAKO). Staining of the immunopositive cells was performed using the DAKO ABC reagent followed by the 3,3\'-diaminobenzidine (DAB) reaction. The number of immunopositive cells was evaluated using Image-Pro Plus program (Media Cybernetics, USA), in 3--5 sections from each brain, and an average value was calculated for each of the areas studied per animal. In vitro binding ---------------- Animals used for 5-HT1A receptor autoradiogarphy were killed by decapitation under ether anesthesia. Their brains were frozen at -40°C in dry-ice cooled isopentane and subsequently cut coronally (10 μm) in a cryostat (-17°C). The sections were processed using standard autoradiographic procedures \[[@B31],[@B32]\]. Briefly, the localization of 5-HT1A receptors was performed using 4 nM ^3^H-8-OH-DPAT (129Ci/mmol, NEN) and non-specific binding was determined in the presence of 10 μM serotonin. Bound ^3^H-8-OH-DPAT was visualized by exposing the labeled sections to tritium-sensitive film (Biomax, KODAK) (4oC, 1 month) along with ^3^H-standards (^3^H-microscales, ARC). Quantitative image analysis of the autoradiograms was performed using SCION-Image for Windows. Specific binding, \>95% of the total binding, was expressed as fmol/mgr tissue. Long term forced swimming ------------------------- On each of 15 consecutive days adult handled and non-handled male animals were placed for 5 min in a glass cylinder 33 cm in height and 20 cm in diameter containing tap water at 24°C. Body weight measurement ----------------------- During the period of the long term forced swimming handled and non-handled animals were weighed daily prior to the exposure to the stressful stimulus. Moreover, the amount of food consumed daily was determined for each one of these animals. Determination of plasma leptin levels ------------------------------------- Immediately after the last exposure to long term forced swimming (day 15) blood samples from all animals were collected by cardiac puncture under ether anesthesia, using heparinized syringes, and centrifuged to obtain plasma. Leptin concentrations were determined by RIA (Linco\'s™ rat leptin \[^125^I\] assay system). Statistical Analysis -------------------- Data were analyzed by a one-way analysis of variance (ANOVA) with handling as the independent factor. Data on learning, body weight and food intake were analyzed by a one-way ANOVA with repeated measures (handling served as the independent factor and days of training served as the repeated factor). All tests were performed with the software SPSS for Windows (10.0.1, SPSS Inc.). Differences were considered as significant if p \< 0.05. Results ======= Following exposure to a short term restraint stress handled animals displayed a greater ability for spatial learning in the Morris water maze, as shown by the lower mean escape latencies (time to find the submerged platform) of the handled animals during the acquisition of the task (F~1,15~= 4.565, p = 0.05) (Fig. [1A](#F1){ref-type="fig"}). Furthermore, handled animals spent more time in the target, and less in the opposing quadrant during the probe trial (F~1,15~= 6.320, p = 0.024) (Fig. [1B](#F1){ref-type="fig"}), indicating superior mnemonic function (better consolidation of information). The effects of \"neonatal handling\" on cognition were accompanied by changes in GR and MR hippocampal levels: Higher GR and lower MR levels were found in the CA2 region of the hippocampus of handled, compared to the non-handled animals, following their exposure to the Morris water maze (F~1,13~= 14.632, p = 0.002 and F~1,13~= 5.268, p = 0.042, respectively) (Fig. [2](#F2){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Effects of handling on spatial learning and memory in the Morris water maze following an acute restraint stress.A. Mean escape latencies-Learning: handled animals took less time to find the submerged platform during the 8 learning trials compared to the non handled (p = 0.05, one way ANOVA with repeated measures). Values represent mean escape latencies ± S.E.M.B. Memory: handled animals spent more time in the target and less in the opposing quadrant compared to the non handled (p = 0.024, one way ANOVA with repeated measures). Values represent the mean time spent in each quadrant ± S.E.M. ::: ![](1744-859X-4-8-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Effects of handling on MR and GR immunoreactivity in the CA2 region of the hippocampus.Handling decreased the number of MR positive cells (p = 0.042, one way ANOVA) but increased the number of GR positive cells (p = 0.002, one way ANOVA) in the CA2 region of the hippocampus. The arrow points to a GR positive cell. Values represent means ± S.E.M. ::: ![](1744-859X-4-8-2) ::: \"Neonatal handling\" resulted in increased number of BDNF immunopositive cells, in the CA4 region of the hippocampus (F~1,13~= 35.388, p \< 0.001) (Fig. [3](#F3){ref-type="fig"}). BDNF immunoreactivity was clearly localized in the cytoplasm. The BDNF positive cells were large, with typical neuronal morphology, including processes (see arrow). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Effect of handling on BDNF immunoreactivity in the CA4 region of the hippocampus.Handling resulted in increased number of BDNF positive cells in the CA4 (p \< 0.001, one way ANOVA) region of the hippocampus. The arrow points to a neuronal process. Values represent the mean number of BDNF positive cells ± S.E.M ::: ![](1744-859X-4-8-3) ::: \"Neonatal handling\" increased the density of 5-HT1A receptors in the hippocampus (areas CA1, CA2, CA4 and DG) as revealed by ^3^H-8-OH-DPAT binding (F~1,13~= 9.170, p = 0.027). Notably, the CA3 region was devoid of any detectable labeling (Fig. [4](#F4){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Effects of handling on the density of 5-HT1A receptors in the hippocampus.Neonatal handling increased the number of ^3^H-8-OH-DPAT binding sites in the hippocampus (p = 0.027, one way ANOVA), indicating an increased density of 5-HT1A receptors in this area. Values represent the mean ± S.E.M. of 5-HT1A receptor density in fmoles/mgr tissue. ::: ![](1744-859X-4-8-4) ::: Handled animals had lower plasma leptin levels (F~1,45~= 4.163 p = 0.047), (Fig. [5](#F5){ref-type="fig"}), consumed less food (F~1,15~= 4.580, p = 0.05), (Fig. [6](#F6){ref-type="fig"}), and gained less weight (F~1,15~= 7.392, p = 0.017) during long-term forced swimming stress, compared to the non-handled (Fig. [7](#F7){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Effect of handling on leptin secretion following long term forced swimming stress.Handled animals had lower plasma leptin levels after long term forced swimming, (p = 0.047, one way ANOVA). Values represent mean leptinlevels ± S.E.M ::: ![](1744-859X-4-8-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Effect of handling on food consumption during long term forced swimming stress.Handled animals consumed less food during long term forcedswimming (p = 0.05, one way ANOVA). Values represent the mean of food consumed in gr/100 gr body weight ± S.E.M ::: ![](1744-859X-4-8-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Effect of handling in body weight change during long term forced swimming stress.Handled animal gained less weight during long-term forced swimming stress compared to the non-handled (p = 0.017, one way ANOVA). Values represent the mean % change in body weight ± S.E.M. ::: ![](1744-859X-4-8-7) ::: Discussion ========== Neonatal handling has beneficial effects in the male rats. In addition to its well-documented effects in increasing their ability to cope with stress \[[@B2],[@B3]\], our present results show that it also improves their cognitive abilities. Furthermore, handling resulted in increased hippocampal GR and decreased MR levels. The observed increase in the GR/MR ratio reflects prevalence of GR-mediated effects and implies an increased HPA axis sensitivity. It is noteworthy that the handling-induced increase in basal GR levels, shown by others, \[[@B33]\] persists after exposure to a short-term restraint stress, followed by the Morris water maze as shown by the present results. The superior mnemonic performance of the handled animals could be attributed to the increased levels of GR, since they are involved in the consolidation of learned information and their activation is a prerequisite for optimal memory \[[@B5]\]. Furthermore, our results show that handling increases BDNF. BDNF levels are known to be positively related to learning as well as to have anti-depressive effects. This is particularly interesting in relation both to the present data regarding the effects of handling on learning and memory and our previous results showing that handled males show less \"depressive\" behavior as assessed by shorter immobility times in the chronic forced swimming stress \[[@B26]\]. It thus appears that handling protects males from chronic stress-induced \"depressive\" behavior, possibly by increasing basal BDNF levels. Another pathway underlying the protective effects of handling against stress could involve the serotoninergic system, since our results show that handling increases 5-HT1A receptors, which are directly involved in the action of anti-depressants. Furthermore, results from our laboratory have shown that handling also increases serotonin levels \[[@B34]\]. Interestingly, BDNF has been shown to have a trophic effect on serotoninergic neurons \[[@B18]\] and in general to interact with the serotoninergic system \[[@B17]\]. Among its actions presumed to be mediated through such mechanisms are the effects on appetite, body weight and plasma leptin levels \[[@B17],[@B35]\]. It is noteworthy, that there is an inverse relationship between BDNF and leptin levels: BDNF conditional knockout mice exhibit hyperphagia \[[@B17]\] and over 15-fold higher leptin levels \[[@B35]\]. According to the results of the present work, during chronic forced-swimming stress non-handled males, consume more food and gain more weight compared to the handled. Furthermore, after the last exposure to the stressor, they have higher plasma leptin concentrations. These findings may be relevant to the human condition of stress-induced obesity \[[@B36],[@B37]\], which is believed to be associated with glucocorticoid-induced resistance to leptin \[[@B38]\] accompanied by elevated leptin levels \[[@B39]\]. In addition to increased food intake, non-handled males showed decreased energy expenditure, as revealed by longer immobility times, during the last exposure to our chronic forced-swimming paradigm \[[@B26]\]. Both decreased energy expenditure and increased appetite push energy balance towards energy storage and weight gain. This could explain our results showing that during chronic-forced swimming, non-handled males gain more weight than handled males. It has been proposed that the beneficial effects of neonatal handling are the outcome of the increased maternal care, which the handled animals receive \[[@B33]\]. Thus, our work provides evidence that alterations in maternal care can lead to long lasting changes in brain function affecting cognitive abilities and affective states. Conclusion ========== Handling has a beneficial effect on males, improving their cognitive abilities and reducing their propensity to express maladaptive behavior following chronic stressors. The molecular basis of these effects on behavior could involve the observed handling-induced increase in hippocampal GR/MR, BDNF, and 5-HT1A receptor levels, as well as the decrease in circulating leptinlevels. List of abbreviations ===================== 5-HT1A type 1A serotonin receptors ANOVA analysis of variance BDNF brain derived neurotrophic factor CA1-4 fields 1--4 of Ammon\'s horn DG hippocampal dentate gyrus GR glucocorticoid receptors HPA axis hypothalamic-pituitary-adrenal axis MR mineralocorticoid receptors MWM Morris watter maze RIA radio-immuno-assay Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= EG carried out the BDNF immunocytochemistry. TP carried out the body weight and food consumption measurements as well as the plasma leptin levels determination. SP carried out the spatial learning and memory tests as well as the GR and MR immunocytochemistry. AS carried out the in vitro binding for the 5HT1A receptors. EF and FS conceived, designed and coordinated the study. All authors participated in the statistical analysis of the data as well as to draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This research was supported by a grant from the University of Athens (Special Account for Research Grants) and scholarships by the Hellenic State Scholarship Foundation to SP and AS.	PubMed Central	2024-06-05T03:55:56.418172	2005-4-6	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090442/", "journal": "Ann Gen Psychiatry. 2005 Apr 6; 4:8", "authors": [ { "first": "Efstathios", "last": "Garoflos" }, { "first": "Theofanis", "last": "Panagiotaropoulos" }, { "first": "Stavroula", "last": "Pondiki" }, { "first": "Antonios", "last": "Stamatakis" }, { "first": "Eleni", "last": "Philippidis" }, { "first": "Fotini", "last": "Stylianopoulou" } ] }
PMC1090443	Background ========== Diseases of the brain are frequently manifested by psychiatric symptomatology, a condition conventionally termed \'Organic Brain Syndrome\'. Given the complexity of the nervous system and the vast range of pathological processes that can affect it, a broader view that there exist a number of different and distinct organic brain syndromes seems more likely. OBS is not a specific neurological diagnosis although it remains a standard diagnostic category. One justification for the use of the term is as a kind of abbreviated phrase to refer to the full range of abnormal mental symptoms commonly associated with definable neurological disease \[[@B1]\]. It should be stressed that OBS are defined in psychiatric terms and not in neurologic terms. They are purely descriptive and carry no specific aetiologic implications \[[@B2]\]. Considering the variety of pathological processes that fall under this heading, it is not surprising that no one particular agent has proven to be of significant benefit to date \[[@B1]\]. Symptoms suggestive of cognitive impairment may even persist in a proportion of cases long after the initial episode, especially when the cerebral insult is irreversible \[[@B3]\]. The aims of this study were: \(1) to measure the efficiency of medical personnel in detecting patients suffering from the psychiatric effects of organic states, \(2) to compare the various patterns of clinical presentation between those with the Acute and Chronic varieties of OBS, \(3) to assess the mortality of these neuropsychiatric episodes after a 2 year period, and, \(4) to determine the various ways psychotropic medications were used Methods ======= Sample ------ A total of 196 patients were referred to the C-L Psychiatry services of the Department of Psychological Medicine, University of Malaya Medical Centre (UMMC) between 1^st^March and 30^th^September, 1998. Of this number, 59 patients were diagnosed to have OBS and this sample constituted the focus of this study. Being a cross-sectional follow-up study, the 3 patients whose case notes were not traceable were excluded from the sample. Materials --------- The data were collected from the referral records and further information was obtained from the patient\'s case notes. Based on the case notes, all cases were assessed during the index admission by a Trainee Psychiatrist, the Principal Investigator (PI i.e. author PK), and a Consultant Psychiatrist within 3 hours of receiving the referral form. Patients who were diagnosed to have Acute or Chronic OBS were selected for this study. Their demographic data, psychiatric history (which included clinical presentation and premorbid personality), medical history, mental status examination, physical examination, laboratory investigations, treatment and the progress, in terms of symptom resolution, were recorded. The data were used for specific sub-diagnoses according to the Diagnostic and Statistical Manual of Mental Disorders -- 4^th^Edition (DSM-4) \[[@B4]\] and the fatality rates were determined. Determining the number of days that elapsed from the onset of symptoms to the time the C-L referral was made gave a crude assessment of efficiency of medical personnel in detecting OBS. The case notes were examined further to see the follow-up progress of patients after 2 years. The PI then called the patients who had defaulted follow-up for enquiries about their condition and treatment. All data entered into the proforma was validated by the Lecturer involved, the Second Investigator (author ST). Statistical analysis -------------------- With the Consultant involved, the Third Investigator (author NZ), overseeing the progress, the data were analyzed using the Statistical Package for the Social Sciences (SPSS) 7.5. Descriptive statistics were presented as mean plus or minus standard deviation (SD) and the differences between groups were assessed by the independent samples t-test for equality of means (2-tailed). Categorical data were analyzed using the Pearson\'s Chi-square test (2-sided) for differences between the Acute and Chronic groups or the Fisher\'s exact test (2-sided), where appropriate. The level of significance is p = \< 0.05. Results ======= A. Demographic data ------------------- 44 of the total number of patients were below the age of 65 (78.6%) and 12 were above 65 (21.4%). 37 were male (66.1%) and 19 were female (33.9%). B. Descriptive data ------------------- ### 1) Duration of symptoms before referral The minimum number of days elapsed from onset of symptoms to the C-L referral was 0 days and the maximum was 16 days. The mean value was 3.52 days and the SD was 3.29. ### 2) Underlying psychiatric disorder Listed below are 17 of the 49 patients who had premorbid psychiatric illnesses and all of them with functional diagnoses were in remission at the time of this study. • Alzheimer\'s dementia -- 5 • Major depression -- 3 • Alcoholic dementia -- 2 • Post-ictal psychosis -- 1 • Alcoholic hallucinosis -- 1 • Mental retardation with Bipolar affective disorder -- 1 • Post-concussional dementia -- 1 • Brief reactive psychosis -- 1 • SLE-induced psychosis -- 1 • Simple deteriorative disorder -- 1 ### 3) Perceptual disturbances and thought disorder 17 of the patients (30.4%) experienced visual hallucinations, 15 (26.8%) of them had auditory hallucinations and only 12 (21.4%) were deluded. ### 4) Cognitive functions 17 patients had global disorientation to time, place and person. Furthermore, all those disorientated had disorientation to time. All patients had impairment of attention and concentration (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Cognitive functions ::: Disorientation to: Frequency % ----------------------------- --------------- ------- Time 48 85.7 Place 25 44.6 Person 23 41.1 Impairment of: Frequency % Recent memory 54 96.4 Remote memory 41 73.2 Attention and concentration 56 100.0 ::: ### 5) Liaison psychiatry diagnosis A clear organic triggering factor could be found for all patients. 49 (87.5%) of them had Acute OBS and only 7 (12.5%) had the Chronic variety. The respective coded DSM-4 diagnoses, with specific coding, were given (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Liaison psychiatry diagnosis ::: Acute: -------------------------------------------------------------------------------------------------------------- 293.0-- There were 44 with delirium due to various causes: • Head trauma -- 6 • Uremia -- 4 • Post-ictal state -- 4 • Post-operative state -- 2 • Brain metastasis -- 2 • Hyperglycaemia -- 2 • Burn trauma -- 2 • Anaemia -- 2 • Cerebral infarction -- 2 (1 with alcohol-induced persisting dementia -- 291.2) • Hepatic encephalopathy -- 1 • Septicaemia -- 1 • Multiple myeloma -- 1 • Cerebral lupus -- 1 • Cerebral hypoxia -- 1 • Hyponatremia -- 1 (with co-existing thyrotoxicosis) 291.0-- Alcohol withdrawal delirium -- 6 (1 with co-existing delirium due to hypoglycaemia -- 293.0) 292.81-- Steroid-withdrawal delirium -- 2 290.11-- Dementia of Alzheimer\'s type, early onset, with delirium due to post- operative state -- 1 290.11-- Dementia of Alzheimer\'s type, early onset, with delirium due to non- convulsive status -- 1 290.3-- Dementia of Alzheimer\'s type, late onset, with delirium due to carcinoma -- 1 292.81-- Opioid intoxication delirium -- 1 293.81-- Psychotic disorder due to Cushing\'s disease, with delusions -- 1 293.82-- Psychotic disorder due to end stage renal failure -- 1 293.83-- Mood disorder due to acute myocardial infarction -- 1 293.83-- Mood disorder due to post-operative state -- 1 293.83-- Mood disorder due to cerebral lupus -- 1 Chronic: 290.40-- Uncomplicated vascular dementia -- 2 290.42-- Vascular dementia with delusions -- 1 290.43-- Vascular dementia with depressed mood -- 1 290.20-- Dementia of Alzheimer\'s type, late onset, with delusions -- 1 290.0-- Dementia of Alzheimer\'s type, late onset, uncomplicated -- 1 294.0-- Alcohol-induced amnestic disorder, chronic -- 1 ::: ### 6) Psychiatric treatment Only 2 did not require psychiatric treatment and they were those with vascular dementia and morphine intoxication delirium. Of the 54 that required it, 21 of them required relatively high doses i.e. 12 with delirious states, including all 5 alcohol withdrawals, 2 with vascular dementias, 1 with the organic psychotic disorder and 1 with the organic mood syndrome. 3 of them had a previous history of a mental illness. Only 9 of them were agitated. 32 of the 49 (65.3%) with acute syndromes required relatively low doses of medication. 27 of the 33 (81.8%) that required low doses were in a delirium (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Psychiatric treatment. The category axis (y) represents the types of treatment used and the value axis (x) represents the frequency of use. ::: ![](1744-859X-4-9-1) ::: ### 7) Total symptom resolution (upon discharge) 34 (60.7%) of the total had symptom resolution on discharge and 22 of them (39.3%) did not. Below is the breakdown of symptom resolution for the specific subgroups. • Delirium -- 33 of the 45 (73.3%) had total symptom resolution • Dementia -- 5 of the 10 (50.0%) with dementia had no symptom resolution • Organic psychotic disorder -- 1 with post-ictal state and 1 with Cushing\'s disease had symptom resolution • Organic mood syndrome -- only the 1 with post-operative state recovered • Transient amnestic disorder -- the 1 with this disorder had total resolution of symptoms ### 8) Mortality 19 of these patients (33.9%) had passed away during the 2-year period and another 19 had defaulted follow-up. There were only 18 (33.9%) alive at the end of this study. ### 9) Continuing treatment At the end of this study period, of the 18 that could be traced, 14 were not on treatment and of the 4 who were still on treatment, 1 of them was in the Chronic group -- vascular dementia -- and 3 in the Acute group -- cerebral hypoxic delirium, organic psychotic disorder and organic mood syndrome. 3 of them had a past history of a mental disorder and all of them were on Chlorpromazine, Thioridazine, Sulpiride or Risperidone. 3 had been on Haloperidol and 1 on Mianserin during the index admission. C. Difference in clinical presentation between those with the Acute and Chronic varieties ----------------------------------------------------------------------------------------- None of these analyses proved to be of any significance. D. Influence of previous psychiatric history on hallucinations and delusions in OBS ----------------------------------------------------------------------------------- Again, none of these associations proved significant. Cross-tabulations reported p = 0.919, p = 0.770, p = 0.336 respectively for visual hallucinations, auditory hallucinations and delusions. E. Association between psychiatric diagnosis and symptom resolution upon discharge ---------------------------------------------------------------------------------- Those patients with the Acute syndrome had significant symptom resolution as compared with those having the Chronic syndrome (p = 0.001). However, the elderly patients had no significant decline towards symptom resolution as compared to the younger age group (p = 0.127). F. Presence of previous psychiatric history and symptom resolution upon discharge --------------------------------------------------------------------------------- In 1 patient with OBS and a history of a mental illness, symptom resolution after commencing treatment was not prolonged as indicated by an index of p = 0.167. G. Effect of psychiatric diagnosis and mortality ------------------------------------------------ There was no difference in terms of mortality between those with the Acute or Chronic varieties of OBS. Even in older patients with OBS, a value of p = 0.124 showed that there was no significant difference in mortality as compared to those younger than 65 years old. H. The association between symptom resolution upon discharge and mortality -------------------------------------------------------------------------- There was no significant association between these 2 variables. I. Influence of presence of previous psychiatric history in those with OBS on mortality --------------------------------------------------------------------------------------- This association proved to be of statistical significance (p = 0.029)indicating that patients with a premorbid mental disorder had lower mortality rates. J. The need for continued treatment in the subgroups of OBS ----------------------------------------------------------- This was not significant (p = 0.405) showing that those with the Chronic syndrome required no more maintenance treatment as compared with the Acute group. And in those with a previous psychiatric history, the need for maintenance treatment was no different from those without (p = 0.275). Even in the elderly patients, there was no increased need for continued treatment, as evidenced by a value of p = 0.405. Discussion ========== Medical records provide a useful source of information and diagnoses based on medical records are acceptable as long as they are considered a substitute of diagnoses obtained from a direct interview. Telephone interviewing is also considered an acceptable alternative method and it has been reported that comparable diagnostic information is obtained through face-to-face and telephone interviews \[[@B5]\]. We had used both modalities to a certain extent and they had their limitations, as would be discussed later. In this study, the geriatric group made up less than a quarter of the sample, and on the whole, males predominated the sample by two-thirds. The mean duration of time elapsed from onset of symptoms in comparison with the SD proved that detection of these syndromes has been rather inefficient in this center (3.52 days). The association between elapsed time and symptom resolution was not significant. Although almost a third of them had a previous history of a mental illness, it had no bearing on the presence of hallucinations and delusions, nor did it on symptom resolution or the need for continued treatment. Those with the previous history did, however, require higher doses of medication as compared to the rest because of their underlying psychiatric illness. Oddly, there were significantly lower mortality rates (p = 0.029) in those who had a previous history of psychiatric disorder, possible reasons being that those cases may not have been OBS in the first place but misdiagnosed instead, and also the small number in that category. The above findings suggest that premorbid functional disorders do not affect the clinical presentation of patients during the course of an OBS. However, since all the patients with premorbid mental illnesses involved in the study were in remission, the above suggestion cannot be concluded. When there are severe perceptual disturbances in the visual modality, acute cerebral disorder is more implicated than the chronic type \[[@B6]\]. Visual hallucinations predominated the clinical picture in contrast to auditory hallucinations and delusions, but again did not vary in their occurrence between both varieties of OBS. In a study by Hirono \[[@B7]\], it was found that half of their patients with Alzheimer\'s disease showed evidence of delusions or hallucinations. Independent factors associated with psychosis were older age, female sex, longer duration of illness and more severe cognitive impairment. Orientation to time is labile and quickly disrupted by organic causes. Orientation to place is disturbed later in the disease process. When established, disorientation to time and place are evidence of an organic state and may be the earliest signs in a dementing process. Disorientation to person occurs at a very late stage. It was found that a very high number of patients experienced disorientation to time and less than half were disorientated to place and person. This points to the early detection of these cases before their condition deteriorated and produced global disorientation. Memory disturbances associated with brain disease is referred as organic or true amnesia and manifest as impairments of registration, retention, retrieval, recall and recognition. In organic states, attention may be profoundly decreased and usually accompanied by lowering of consciousness \[[@B8]\]. Almost all patients had impairment of recent memory and only just over a quarter of them had remote memory impairment. Attention and concentration was, however, impaired in all of them. We tackled the confusion surrounding the Acute-Chronic dichotomy by carrying on the initial diagnosis given by the PI and Consultant Psychiatrist during the index admission and going by the possible reversibility of a particular condition instead of the rapidity of its development or resolution. Put simply, the primary cause of the acute impairment is usually \'outside the brain\' and that of the chronic syndrome normally \'within the brain\'. The distinction between these two organic conditions is most clearly derived from the history of the mode of onset of the disorder. A short history and firm knowledge of an acute onset will make a chronic reaction unlikely and onset in association with a physical illness is strongly suggestive of an acute organic reaction \[[@B6]\]. The use of specific diagnoses is helpful as although most chronic organic disorders cannot be reversed, a small number are potentially treatable \[[@B9]\]. Acute disturbances of cerebral function may, in time, progress to the development of irreversible structural pathology with an admixture of features specific to both. The two may co-exist when a chronic dementing process is complicated by another concomitant or superimposed disease \[[@B6]\]. Those with delirium superimposed on dementia were designated as Acute as their symptoms in their index admission were those of a delirious nature. As expected, it was found that symptom resolution occurred with significance in the Acute group as compared to the Chronic group (p = 0.001). However, the younger age group did not show any statistical significance toward symptom resolution as compared to the older group. Delirium has poor outcomes in hospitalized older patients \[[@B10]\]. It has multiple aetiologies and a poor long-term prognosis \[[@B11]\]. Advancing age increases the risk and those over 60 years are at highest \[[@B12]\]. The older the patient and the longer the delirium, the outcome is a longer resolution of symptoms. A complete resolution of confusional symptoms is not usually achievable in prolonged confusional states that are superimposed on dementia. Improvement from severe to mild confusion or merely a reduction of symptoms would be a more realistic goal \[[@B13]\]. However, in this study, it was found that those who had delirium on dementia had resolution of their confusional symptoms. Even with treatment, there was no improvement in their dementing features, as may be expected. It was found that there was no significant difference in mortality rates between the Acute and Chronic groups, possibly due to the small number of patients assigned to the latter group. This was in keeping with observations made by Inovye \[[@B14]\] that there were no significant associations between delirium and mortality and between delirium and length of hospital stay. That study, however, found delirium to be a significant predictor of functional decline at both hospital discharge and at follow-up, therefore making it an important independent prognostic determinant of hospitalization outcomes. Our findings disagreed with the generally held concept that the occurrence of delirium was associated with a high mortality rate in the following year, mainly because of the serious nature of the provoking medical conditions. Even the mortality in the elderly within the sample showed no statistical significance as compared to those who were younger than 65 and the finding was not in keeping with related literature. Huang \[[@B15]\] investigated the rate of delirium, reasons for admission, clinical features, aetiologies and mortality during a 2-year follow-up and found that the incidence of delirium was higher in their geriatric group. However, the older patients had a higher mortality rate during the 2-year follow-up period and that stressed the importance of after-discharge care in those patients. Higher death rates have also been found among the cognitively impaired elderly patients than those aged-matched patients with functional psychiatric illnesses and the cognitively intact elderly. Koponen\[[@B16]\] was of the same school of thought and associated delirium with a significant rate of mortality. These results, however, were not in line with findings by Rabins & Folstein \[[@B17]\] that cognitively impaired individuals have higher fatality rates than cognitively intact individuals. There was also no significant association between symptom resolution upon discharge during the index admission and mortality. As observed earlier, the most popularly used treatment in our setting was a combination of a neuroleptic and a benzodiazepine, usually Haloperidol and Lorazepam. This combination accounted for the treatment of over a third of patients and the use of a neuroleptic alone came second, amounting to just under a third of the patients. Adams \[[@B18]\] showed that parenteral Haloperidol offered the first hope for treating delirium and the addition of Lorazepam quickened the onset of sedation. Delirium is a common component of dementia and may produce considerable morbidity. In addition to psychotic features, it may produce considerable agitation, which may be unresponsive to conventional medications. The main approach is to treat any underlying medical condition that could cause the delirium. It is, however, not always reversible and there is no specific treatment for persistent delirium \[[@B19]\]. Cole, Primeau and Elie \[[@B20]\] found Haloperidol, Chlorpromazine and Mianserin to be useful in controlling the symptoms of delirium and high levels of premorbid functioning were related to better outcomes. The use of this selection of drugs was similarly practiced in our setting although Chlorpromazine is now less widely used and usually reserved more for its sedative-hypnotic effects. This is because we have had experiences with its propensity to lower the seizure threshold and to cause hypotension. Finally, there was no significant difference in the need for continued treatment at 2 years in the Chronic group as compared to the Acute group. Even in those with a previous psychiatric history or in those who were in the elderly age group, there appeared to be no difference. Only 15 patients afflicted with these conditions were compliant to follow-ups. There were only another 3 of those who defaulted follow-up and whose conditions were documented in their case notes when they were subsequently admitted for other problems unrelated to that of their index admission. Thus, there were still 19 of them whose status at 2 years was unknown. The problem was mainly with those having alcohol-related disorders and it has been found that patients with alcohol delirium have been known to have higher mortalities and have been known to be more difficult to follow-up \[[@B16]\]. Of the 8 with these disorders, 2 had passed away, 4 were not contactable and the 2 that were eventually contacted had not turned up for follow-ups. This large number of dropouts where follow-ups were concerned caused difficulties in assessing the mortality rate after 2 years. It proves to be a major issue in C-L Psychiatry and needs to be addressed to ensure more comprehensive post-discharge care to this group of patients. Although the methods by which data were obtained in this study have been validated previously \[[@B17]\], the questionable reliability of the data collected from the medical records forms the first limitation. There was also little information on the outcome of these patients in the records and as earlier mentioned, telephone calls revealed no new information. The second limitation was that assessment scales had not been incorporated. Another limitation to this study was the small sample size and confined only to the UMMC thus, we were not able to apply the results as representing a whole region. Also, the relatively small number of patients with a diagnosis that suited the criteria for the Chronic syndrome had caused difficulties in statistical analysis, as did the high rate of dropouts on establishing the mortality rate after 2 years. This study was intended to promote practical awareness and possibly, improve the understanding and treatment, of patients afflicted with organically-induced psychiatric conditions. Its implications for clinical practice raise several questions. We hope this report will stimulate renewed interest in this field and although the findings do not contribute to a new conceptual understanding of OBS, they do suggest directions for further research on their management.	PubMed Central	2024-06-05T03:55:56.420328	2005-4-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090443/", "journal": "Ann Gen Psychiatry. 2005 Apr 15; 4:9", "authors": [ { "first": "Prem K", "last": "Chandrasekaran" }, { "first": "Stephen T", "last": "Jambunathan" }, { "first": "Nor Z", "last": "Zainal" } ] }
PMC1090548	Introduction ============ Candida albicansis a yeast pathogen that causes mild, chronic, superficial systemic infections in immunocompromised and cancer patients \[[@B1]-[@B3]\]. Research, worldwide is broadly focused on 1) search of new targets in C. alibcans, 2) Screening inhibitory activities with therapeutic agents, and 3) designing or modifying novel agents, to overcome treatment resistance \[[@B4]\]. C. alibcanshas been shown to harbor a self splicing group-I intron in the nuclear 25s rRNA genes \[[@B5]\]. Presence of this gene in C. alibcans, but not in the human genome, suggests that splicing inhibitors could potentially act as alternate targets in controlling the proliferation of fungal pathogens. Recent screening studies using disk diffusion have shown that cis-diamminedichloro platinum (Cisplatin), a well-established antitumor agent, inhibits the growth of C. alibcansat lower concentrations \[[@B6]\]. Despite its prevalent use in cancer treatment, very little work has been done \[[@B7]-[@B10]\] to document the effects of Cisplatin and related metal complexes in controlling the proliferation behavior of C. alibcans. With the molecular mechanisms remaining largely unknown, this provided us a basis to study the effects of Cisplatin treatment on the morphology and growth curve of C. alibcans. Morphological changes in Candidacan be followed by direct fluorescence staining with suitable dyes to selectively visualize different parts/aspects of cellular activities \[[@B11]-[@B13]\]. Though the use of Ethidium Bromide for fluorescence staining is well known, there is no report to show the use of this technique in studying the morphology of Candida, especially in response to treatment with anticancer agents such as Cisplatin. Herein, we report the morphological changes of C. alibcansin response to cisplatin using fluorescence microscopy with ethidium bromide staining. Materials and methods ===================== Media and Inoculums ------------------- Yeast isolates (clinical strains) were obtained from patients, Royapettah Hospital, Chennai. Sub-cultured isolates were maintained in Sabouraud dextrose agar (B.B.L, Cockeysville, MD) at 1--5 × 10^6^CFU/10 ml media. 30 μl of each sample (log phase) was used for slide preparation to study the morphological changes. Drug preparation ---------------- A 2 mg/ml stock of Cisplatin (TNDPL, Chennai) was prepared in sterile Millipore water. Stock was diluted to 60 μg/10 ml culture for treatment purposes. Staining -------- Ethidium bromide staining was performed using a freshly prepared stock of ethidium bromide (1 mg/ml water) from which 50 μl/100 ml concentration was used for staining the slides. Gram staining was performed from commercially available kits as per instructions. Morphological studies using Fluorescence Microscopy --------------------------------------------------- 30 μl of cultured C. alibcans(Drug treated/non-treated cells) was taken and spread over the clean glass slide and dried at 37°C for 5 minutes. Slides were dipped into the EtBr staining solution for 30--40 seconds followed by immersion in water for 15--20 seconds to remove excess stain. Gram staining was performed in parallel on a similar set of slides to validate the results. Both the set of slides were viewed using NFM under the 100× magnification. Results and Discussion ====================== Treatment of C. alibcanswith Cisplatin was found to markedly inhibit hyphae and ovoid growth (Fig. [2](#F2){ref-type="fig"}) as revealed by ethidium bromide staining of drug treated cells. This was in contrast to the controls showing markedly branched hyphae, with budding characteristics (Fig. [1](#F1){ref-type="fig"}). The ovoid cells of Cisplatin treated C. alibcansshowed an increased uptake of the ethidium bromide stain (red color) in contrast to the untreated counterparts (yellowish orange), thus pointing to drug induced poisoning and death of treated cells. These changes were concomitant with inhibitory effects on the growth curve with respect to untreated cells (Fig. [3](#F3){ref-type="fig"}). Presence of Cisplatin not only caused suppression in the limiting growth values but caused a slight left shift in the EC50 values, which remains to be studied further. Some of the ovoid cells undergoing poisoning with cisplatin were found to be unusually enlarged before undergoing their natural fate (Fig. [2](#F2){ref-type="fig"}). This is known to be one of the typical effects of poisoning with cisplatin in case of tumor cell lines as well thus suggesting formation of similar cytotoxic end products \[[@B14]\]. Cisplatin in water forms a diaqua species of Pt^+2^due to the fast leaving Cl^-^ions. This could lead further to host of reactions in the cytoplasm before forming end products with DNA. It is also possible that cisplatin exerts its activity through other secondary response mechanisms. Further studies using RT-PCR will elucidate 1) the role of the binding affinity of cis-DDP to certain RNA species in cytotoxicity against C. albicans, for example, group-I intron ribozyme, known in many human diseases causing pathogens and a potential target site for therapeutic agents. 2) Whether cis-DDP inhibits the C. alibcansgrowth through penetration into the cell or some secondary response. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Photograph showing morphology of Candida albicansin the absence of cisplatin (6 ug/ml), stained with ethidium bromide and gram stain. ::: ![](1476-0711-4-7-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Photograph showing morphology of Candida albicansin the presence or absence of cisplatin (6 ug/ml), stained with ethidium bromide and gram stain. ::: ![](1476-0711-4-7-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Growth curve of [Candida albicans]{.underline}in the presence/absence of Cisplatin (6 μg/ml). Curve fitting of data points was performed using non-linear regression(sigmoidal curve): GraphPad Prism Version 3.02. ::: ![](1476-0711-4-7-3) ::: Acknowledgements ================ K.C thanks the Lady Tata Memorial Trust for the fellowship in pursuing this research work.	PubMed Central	2024-06-05T03:55:56.422929	2005-4-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090548/", "journal": "Ann Clin Microbiol Antimicrob. 2005 Apr 25; 4:7", "authors": [ { "first": "Chandrasekhar", "last": "Kesavan" }, { "first": "Malathi", "last": "Raghunathan" }, { "first": "Natarajan", "last": "Ganesan" } ] }
PMC1090549	Background ========== Anesthetics like halothane are rarely used in most nations except in developing countries, which still widely use this method of anesthesia \[[@B1]\]. Also on the rise is alcohol consumption in developing countries \[[@B2]\]. Patients consuming alcohol who are anesthetized with halothane could potentially have inadequate metabolism or adduct formation leading to problems such as cardiovascular disease or liver injury. Hepatic metabolism of halothane and ingested ethanol (ethyl alcohol, alcohol) yields the highly reactive metabolites: trifluoroacetyl chloride (TFA) from halothane and acetaldehyde (AA) and malondialdehyde (MDA) from the oxidation of ethanol \[[@B3],[@B4]\]. MDA and AA react together with endogenous proteins (most likely the ε-amino group of lysine residues) to form distinctive new adducted proteins \[[@B3],[@B4]\]. The adduct formed by the combination of MDA and AA has been termed the MAA-adduct by Tuma et al\[[@B4]\] and has been detected in humans and rats chronically consuming ethanol \[[@B5],[@B6]\]. In fact, Slatter et al\[[@B7]\] have recently confirmed that MDA, AA, and lysine react to form a dihydropyridine derivative structurally identical to the MAA-adduct. Similarly, trifluoroacetyl chloride (TFA) will react with amine groups to form a distinctive protein termed the TFA-adduct \[[@B8]\]. Recent reports by our laboratory have demonstrated that the MAA-adduct will induce the release of the proinflammatory cytokine tumor necrosis factor-alpha (TNF-α) in a purified rat heart endothelial cell culture (HEC) \[[@B9]\]. Trudell et al\[[@B3]\] reported data suggesting that the TFA-adduct may cause cell injury by inducing a similar inflammatory response. Importantly, it has been demonstrated that the TFA-adduct is present in heart tissue obtained from halothane pre-treated rats \[[@B10],[@B11]\]. If both adducts share a similar mechanism of cell activation, receiving halothane anesthesia while intoxicated with alcohol could exacerbate the inflammatory response. Also of interest is that both halothane and ethanol are metabolized through cytochrome P450 2E1 (CYP2E1) \[[@B12]\], possible providing a shared mechanism. In support of this is data suggesting that acetaldehyde effects ventricular myocyte contraction through mechanisms related to CYP oxidase and lipid peroxidation \[[@B13]\]. This could help explain how ethanol consumption and halothane anesthesia could enhance the sensitization of an individual to halothane and MAA adducts, thereby increasing their risk to cardiovascular disease. Therefore, we hypothesize that halothane pre-treatment may potentiate the inflammatory response induced by the MAA-adduct as determined by TNF-α release. Thus, this rat-model study evaluates the effects of halothane pre-treatment in combination with an alcohol diet on in vitroHEC TNF-α release following stimulation with the MAA-adduct. Methods ======= Chemicals and proteins ---------------------- Bovine serum albumin (Alb) was purchased from CalBiochem (La Jolla, CA). Acetaldehyde (AA) was obtained from Aldrich Chemical Co. (Milwaukee, WI). Malondialdehyde (MDA) was obtained as the sodium salt (MDA\~Na) by treatment of tetramethoxypropane (Aldrich Chemical Co.) with NaOH, according to the method of Kikugawa and Ido \[[@B14]\]. Lipopolysaccharide and Eschericahia coli 0111:B4 (LPS) was purchased from Sigma Chemical Co. (St. Louis, MO). Halothane was purchased from Halocarbon Laboratories (River Edge, NJ). Production of the malondialdehyde-acetaldehyde adduct ----------------------------------------------------- MAA-Alb was prepared as described by Tuma et al. \[[@B15]\] Briefly, Alb was modified with 1 mM malondialdehyde (MDA) and 1 mM acetaldehyde (AA) by incubating at 37 degrees for 72 hours. Following incubation, free and reversibly-bound MDA or AA was separated by exhaustive dialysis against a phosphate buffer for 24 hours at 4 degrees. Fluorescence measurements were obtained on post dialysis samples using a Perkin Elmer (Norwalk, CT) LS-5B spectrophotofluorometer attached to a Perkin Elmer GP-100 graphics printer as previously described \[[@B15]\]. Protein concentrations were measured as described by Bradford \[[@B16]\]. Animal preparation ------------------ Male Wistar rats purchased from Charles River Laboratories (Willmington, MA) were maintained on a Purina rat chow diet, until they reached a weight of 140--150 grams, and were divided into three groups. These three groups were housed individually and acclimated to the Lieber-DeCarli liquid control diet from Dyets, Inc. (Bethlehem, PA) for 3 days \[[@B17]\]. The rats were paired by weight, one rat was given the ethanol-containing diet ad libitum, and the other rat was fed an isocaloric amount of the control liquid diet as determined by the pair-fed rat from the day before. Pair feeding was continued for 6 weeks. Finally, the ethanol-containing diet consisted of 18% of the total calories as protein, 35% as fat, and 36% as ethanol. In the control liquid diet, ethanol was replaced isocalorically with carbohydrates. The final group was given free access to standard laboratory chow and water. For adduct immunization rats were injected once per week for 3 consecutive weeks beginning on day fourteen with one of the following protocols: (1) An injection of Alb only (25 μg/ml) subcutaneously plus an i.p. injection of an equal volume of sesame oil were given to control animals.; (2) halothane as a 21.5% solution in sesame oil at a dose of 10 mmol/kg intraperitoneally (i.p.);. (3) MAA-Alb (25 μg/ml) subcutaneously (s.c.); or (4) MAA-Alb and halothane combined in the previously mentioned doses. After one month (day twenty nine) on their respective diets, and 24 hours following the final injection of Alb, halothane (i.p.), MAA-Alb (s.c.), or MAA-Alb (s.c.) and halothane (i.p.) combined, the rats were sacrificed, and hearts removed for use in in vitrostudies as described below. All animals were allowed free access to their food and/or water up to 1 hour before sacrifice. All procedures were approved by the animal subcommittee of the Omaha VA Medical Center, and are in accordance with the National Institutes of Health Guidelines on the Use of Laboratory Animals. Transaminase assay ------------------ Animals injected with the above ligands were bleed prior to sacrifice and serum transaminase enzymes determined using an (ALT/GPT and AST/GOT) assay kit purchased from Sigma Diagnostics (St. Louis, MO). Isolation and culture of heart endothelial cells (HECs) ------------------------------------------------------- Male Wistar rats were anesthetized intraperitoneal with phenobarbital (100 mg/kg) and the intact beating heart was immediately removed under sterile conditions. After mincing and dispase digestion, heart endothelial cells (HECs) were isolated and grown to confluency as previously described \[[@B9],[@B18]\]. In brief, HECs were separated by centrifugation at 400 × g for 10 min followed by three washes with M199-F12 (GIBCO, Grand Island, NY) containing 10% fetal bovine serum (GIBCO). Cells collected were \>90% HECs, verified by staining with mouse anti-rat RECA-1 (Harlan Bioproducts for Science, Indianapolis, IN) and mouse anti-Factor VIII-von Willebrand\'s Factor (Cedar Lane Laboratories Limited, Hornby, Ontario, Canada) \[[@B9]\]. Cells were seeded into 24 well tissue culture plates (Becton-Dickinson Labware, Franklin Lakes, NJ) containing fibronectin (20 μg/well) (Sigma Chemical Company, St. Louis, MO) and grown to confluency at 37°C for 48--72 hours. Percentage of cell necrosis determinations ------------------------------------------ The percentage of cell necrosis (death) of HECs during exposure to MAA-Alb was determined by an enzyme (lactic acid dehydrogenase, LDH) release assay of the HEC supernatant as described by Korzeniewski and Callewaert \[[@B19]\]. Briefly, following stimulation of HECs with 1,5,10, and 25 μg/ml MAA-Alb or media only (control) for 3 and 24 hours, the HECs were centrifuged (200 × g, 10 min) and a 100 μl aliquot of the HEC supernatant was transferred to the corresponding wells of flat-bottomed microtiter plates. Subsequently, 100 μl of a freshly prepared lactic acid dehydrogenase substrate mixture \[5.4 × 10^-2^lactate (Acros Organics, New Jersey, USA), 6.6 × 10^-4^M 2p-iodophenyl-3p-nitrophenyl tetrazolium chloride (Acros), 2.8 × 10^-4^M phenazine methosulfate (Acros), and 1.3 × 10^-3^M nicotineamide nucleotide NAD in 0.2 M Tris buffer, pH 8.2 (Sigma)\] was added to each well. The plates were incubated in the dark at room temperature for 10 min and the reaction stopped by the addition of 50 μl/well of a 1 M HCl solution. A microtiter plate reader (MR 7000, Dynatech Labs, Inc., Chantilly, VA) was used to monitor the resultant light absorbance at 490 nm while 630 nm was used as reference. LDH activity, expressed as change in absorbance/min, was calculated with Biolinx 2.21 software (Dynatech) on an IBM compatible computer. Percentage necrosis of the HECs was determined by the following formula: % Necrosis = (E-S)/(M-S) × 100\[[@B19]\], where E is the optical density (OD) of the experimentally induced release of LDH activity from the HECs incubated in the presence of the various concentrations of MAA-Alb, S is the spontaneous release of LDH activity (OD) from HECs incubated with media only, and M is the maximal release of LDH activity (OD) determined by total HEC necrosis induced by exposure to 10% Triton X-100 (Fisher Scientific, Fair Lawn, NJ) \[[@B19]\]. Endotoxin assay for LPS contamination ------------------------------------- Prior to any stimulation all ligands, buffers, and media were tested for LPS content, which could influence the levels of background cytokine secretion. Samples were monitored for endotoxin using a Limulus Amebocyte Lysate assay from BioWhittaker (Walkersville, MD). Those samples contaminated with LPS at concentrations greater than 0.1 ng/ml were not utilized in these studies. MAA-Alb stimulation of HECs --------------------------- HECs were washed on the day of the experiment with M199-F12 without serum and allowed to incubate for 1 hour to remove excess serum components. Following this incubation period, cells were stimulated with: 5 μg/ml Alb, MAA-Alb, LPS, and 10 μM Hexyl-MAA (a synthetic analog to the MAA-adduct) in serum free M199-F12 for 3 hours. Supernatant was collected and frozen at -70 degrees until assayed using a commercially available TNF-α ELISA kit. TNF-alpha ELISA --------------- Quantification of TNF-α levels of the HEC supernatants was performed with a Factor-Test-X ™ rat TNF-α ELISA kit (Genzyme Diagnostics, Cambridge, MA), which employs a multiple antibody sandwich principle. The ELISA kit was developed, stopped and read at 450 nm on a MR7000 plate reader using BIOLINX ™ software. Final concentrations of TNF-α is expressed in pg/ml. Statistical analysis -------------------- All results are reported as means ± Standard Deviation (SD). Analysis of variance (ANOVA) was used to compare means between treatment groups. Dunnett\'s two-tailed t-test was used to determine if any pre-treatment was significantly different when compared to the unpretreated (control) group of similar diet and in vitrostimulant conditions. P values of 0.05 or less were regarded as statistically significance. Results ======= Transaminase release -------------------- In an effort to determine liver damage from the administration of halothane and the MAA-adduct, serum from these animals were collected and assayed for the release of the serum transaminases, ALT and AST. Results indicated no difference between the animals injected with Alb, halothane, MAA-Alb, or halothane + MAA-Alb. There was a slight increase in ALT/AST levels in the ethanol-fed animals, yet these results were determined to be insignificant. Effects of increasing concentrations of MAA-Alb on in vitroHEC cell death --------------------------------------------------------------------------- In order to determine what concentrations of MAA-Alb would result in cell death of HECs, cells were isolated from chow-fed rats and stimulated with increasing doses of the antigen. As shown in Table [1](#T1){ref-type="table"}, HECs incubated with media alone, 1 and 5 μg/ml of MAA-Alb had little effect on % cell death after 3 and 24 hours of incubation. However, both 10 and 25 μg/ml of MAA-Alb had a significant increase in cell death over the control and lower concentrations of the same antigen. There was a statistical difference in the amount of cell death observed when HECs were exposed to10 μg/ml for 24 hours as compared to the 3 hour stimulation period. However, these differences were not observed with 25 μg/ml of MAA-Alb. For these reasons, 5 μg/ml of MAA-Alb was chosen as the optimum concentration for use in the remainder of the experiments in this manuscript. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The percentage of cell death of heart endothelial cells (HECs) after stimulation with MAA Alb as determined by LDH release. ::: Time 10 μg/ml Alb 1 μg/ml MAA-Alb 5 μg/ml MAA-Alb 10 μg/ml MAA-Alb 25 μg/ml MAA-Alb ---------- -------------- ----------------- ----------------- ------------------ ------------------ 3 hours 1.8 ± 0.37 2.8 ± 0.55 2.6 ± 0.68 7.5 ± 0.70\* 11.2 ± 0.58\* 24 hours 2.6 ± 0.67 3.2 ± 0.58 3.0 ± 0.89 13.3 ± 0.53\* 13.3 ± 0.71\* Results are expressed as means +/- SD for 6 determinations in each group. Values different from the media control are indicated (\) at P \< 0.05. ::: Effects of pre-treatment with halothane on TNF-α release by HECs ---------------------------------------------------------------- In order to determine the effects of halothane pre-treatment on the release of TNF-α by HECs, the cells were isolated from pair and ethanol-fed rats that had been injected as previously described in the Materials and Methods with one of the following; Alb, halothane, MAA-Alb, or both halothane and MAA-Alb. The isolated HECs were stimulated in vitrowith Alb, MAA-Alb, LPS, or Hexyl-MAA (a synthetic analog to MAA). As shown in Figure [1](#F1){ref-type="fig"}, HECs from ethanol-fed rats injected with MAA-Alb + halothane and stimulated with 5 μg/ml of Alb significantly (P \< 0.01) increased the amount of TNF-α release when compared to animals injected with Alb. Increases in TNF-α release were observed in Alb, halothane, and MAA-Alb + halothane injected ethanol-fed animals in comparison to the pair-fed controls (P \< 0.05). The most significant increase was demonstrated in the MAA + halothane injected ethanol rats when compared to the halothane or MAA-Alb injected animals (P \< 0.001). As a positive control, LPS was used as the stimulating antigen, (Figure [2](#F2){ref-type="fig"}) and found to increase TNF-α secretion in the ethanol-fed animals as previously shown by others \[[@B20]\]. There was an increase over the Alb control in animals injected with halothane, MAA-Alb, and MAA + halothane (P \< 0.05). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Alb-stimulated release of TNF-α by HECs isolated from pair and ethanol-fed rats immunized with Alb, halothane, MAA-Alb, or halothane + MAA-Alb. Results are expressed as means +/- SD for 5 experiments. Values different from Alb are indicated (\) at P \< 0.01. Values different from the pair-fed control group are indicated (+) at P \< 0.05. Values different from Halothane or MAA-Alb injected animals are indicated at (\#) P \< 0.001. ::: ![](1471-2253-5-3-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### LPS-stimulated release of TNF-α by HECs isolated from pair and ethanol-fed rats immunized with Alb, halothane, MAA-Alb, or halothane + MAA-Alb. Results are expressed as means +/- SD for 5 experiments. Values different from Alb are indicated (\) at P \< 0.05. Values different from the pair-fed control group are indicated (+) at P \< 0.05. ::: ![](1471-2253-5-3-2) ::: Pair and ethanol-fed rats injected with the above antigens increased TNF-α release 3 fold in response to stimulation with MAA-Alb in comparison to the Alb or LPS stimulated HEC. As shown in Figure [3](#F3){ref-type="fig"}, increases in TNF-α release was demonstrated in the halothane, MAA-Alb, and halothane + MAA-Alb rat HECs after stimulation with MAA-Alb as compared to the Alb injected rat HECs (P \< 0.001). When comparing the four groups of rats, TNF-α release was increased in HECs from ethanol-fed rats over the pair-fed controls (P \< 0.001). Ethanol-fed rats injected with halothane or MAA-Alb had similar effects on HEC secretion of TNF-α, while MAA-Alb + halothane together had a 3 fold increase over halothane or MAA-Alb alone (P \< 0.001). This synergistic response was also observed in the pair-fed controls for the MAA-Alb + halothane injected rat HECs. As witnessed in Figure [4](#F4){ref-type="fig"}, Hexyl-MAA (the synthetic analog of MAA) stimulated HECs similar to that of MAA-Alb. Increases in TNF-α secretion is witnessed in the ethanol-fed rats for all the experimental conditions (P \< 0.001). There is a 2--3 fold increase in groups injected with halothane, MAA-Alb, or halothane + MAA-Alb over the Alb injected control (P \< 0.001). The most significant increase is the 2 fold increase in the MAA-Alb + halothane injected ethanol-fed rats (P \< 0.001). These data show an additive effect using the combination of both antigens. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### MAA-Alb-stimulated release of TNF-α by HECs isolated from pair and ethanol-fed rats immunized with Alb, halothane, MAA-Alb, or halothane + MAA-Alb. Results are expressed as means +/- SD for 5 experiments. Values different from the Alb injected group are indicated at (\) P \< 0.001. Values different from pair-fed control group are indicated at (+) P \< 0.001. Values different from Halothane or MAA-Alb injected animals are indicated at (\#) P \< 0.001. ::: ![](1471-2253-5-3-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Hexyl-MAA-Alb-stimulated release of TNF-α by HECs isolated from pair and ethanol-fed rats immunized with Alb, halothane, MAA-Alb, or halothane + MAA-Alb. Results are expressed as means +/- SD for 6 experiments. Values different from the Alb injected group are indicated at (\) P \< 0.001. Values different from pair-fed control group are indicated at (+) P \< 0.001. Values different from Halothane or MAA-Alb injected animals are indicated at (\#) P \< 0.001. ::: ![](1471-2253-5-3-4) ::: Discussion ========== Tumor necrosis factor-alpha (TNF-α), the most proximal pro-inflammatory cytokine mediator released following LPS (endotoxin) stimulation, induces an inflammatory response through several mechanisms, including increased neutrophil-endothelial cell adherence, increased endogenous nitric oxide production \[[@B21]\], and the stimulation of the production and release of other pro-inflammatory cytokines, including interleukin-8 (IL-8) \[[@B22]\]. Recently, the role of TNF-α in the process of apoptosis (cell death) has been demonstrated \[[@B23]\]. For example, while apoptosis occurs naturally in the liver at a low and controlled rate, an increased rate of apoptosis is observed following increased cellular TNF-α concentration. The direct correlation between TNF-α concentrations and the rate of apoptosis has been described in several types of liver diseases characterized by cell necrosis and death \[[@B24]\]. Apoptosis has also been demonstrated in heart endothelial cells in response to TNF-α and Interleukin-18 \[[@B25],[@B26]\]. Cell death experiments demonstrate that the MAA-adduct caused cell death only at concentrations of 10 μg/ml or greater (Table [1](#T1){ref-type="table"}). Since TNF-α may cause cell death, \[[@B24]\] the significant increase in the percentage of HEC cell death following stimulation with 10 and 25 μg/ml of MAA-Alb may be due to the direct toxicity of the MAA-adduct or indirectly due to MAA-adduct-induced TNF-α release. Further studies will be required to clarify this issue. Ohki et al\[[@B27]\] reported, consistent with the results of our study, that ethanol fed rats demonstrate greater TNF-α release when exposed to LPS, with the increased TNF-α release following ethanol feeding caused increased neutrophil-endothelial adherence. Other investigators have similarly demonstrated increased neutrophil-endothelial adherence induced by ethanol, suggesting ethanol ingestion induces an inflammatory injury \[[@B28]\]. Since the MAA-adduct is a primary metabolic end product of ethanol metabolism, our demonstration that this adduct induces TNF-α release and that alcohol feeding potentiates endotoxin-induced TNF-α release is consistent with the understanding that alcohol adduct induced TNF-α release may play a significant role in alcohol induced solid organ injury \[[@B20]\]. MDA has been detected in guinea pig heart tissue following exposure to halothane \[[@B29]\]. Also interesting is that circulating antibodies to cardiac protein-acetaldehyde adducts have been found in alcoholic heart muscle disease \[[@B30]\]. If AA from alcohol metabolism and MDA from halothane are present in heart tissue, the possibility of MAA-adduct formation is likely. The majority of adducts formed when acetaldehyde reacts with proteins for short time periods are unstable AA-protein adducts. With time, unstable AA-adducts stabilize and can form an irreversible adduct. This irreversible, stable adduct has been demonstrated to be the MAA-adduct \[[@B15]\]. These MAA-adducts have also been found in atherosclerotic human aortic heart tissue \[[@B9]\], providing a mechanism of heart tissue damage. Heavy alcohol consumption can accelerate human atherosclerotic heart disease \[[@B31]\], making MAA-adducts a possible candidate for this process. Increased TNF-α levels have similarly been implicated in other disease states, like end-stage heart disease and intractable heart failure, by the promotion of monocyte dysfunction and death \[[@B32]\]. Since TFA-adduct production is induced with a single dose of halothane and persists in measurable concentrations in rat heart for greater than 90 hours (but less than 10 days), \[[@B8],[@B33]\] the possibility for cross-reaction with MAA-adducts is plausible. These experiments demonstrated that halothane and the MAA-adduct-induced HEC TNF-α release in a synergistic manner. In support of this data, Trudell et al\[[@B3]\] demonstrated that antibodies raised against AA-adducts and TFA-adducts cross-react, suggesting that the immunologic properties of both adducts may be similar. This data suggests that similar immunologic mechanisms may be shared by both halothane hepatitis and ethanol-induced liver injury. The data reported in this study demonstrates that halothane pre-treatment will potentiate the MAA-adduct induced TNF-α release in vitroof HECs. This gives support to the Trudell et al*\[[@B3]\] demonstration that the AA-adducts and TFA-adducts induce organ injury by the release of chemoattractants during the metabolism of ethanol and halothane, resulting in the recruitment of inflammatory cells as the initial step in the initiation of organ injury. TNF-α production and release is a primary biologic mechanism in inflammatory cell recruitment since TNF-α induces the production of IL-8 \[[@B34]\]. Interleukin-8 is the primary cytokine responsible for the promotion of inflammatory cell (neutrophils and monocytes) chemotaxis (migration) and recruitment toward an inflammatory site \[[@B34]\]. Further studies need to be done in order to prove that the TFA-adduct and MAA-adduct cross-react. Conclusion ========== In conclusion, the current study demonstrates that halothane or MAA-Alb pre-treatment potentiates the HEC TNF-α release following MAA-adduct stimulation of ethanol-fed rats when compared to control pair-fed rats. This suggests that the TFA-adduct resulting from the metabolism of halothane increases the inflammatory response, as measured by TNF-α release, following LPS and alcohol (MAA-Alb) adduct stimulation, and that this TNF-α release may contribute to post-halothane exposure solid organ injury. The data also suggest that solid organ injury following halothane administration may be enhanced by prior ethanol consumption. This could help explain the increased risk of cardiovascular disease following excessive alcohol consumption. Finally, these results demonstrate that the combination of ethanol consumption and halothane exposure may enhance the possibility of the development of the sepsis syndrome, since increased systemic concentrations of TNF-α induced by endotoxin is a primary cause of that clinical condition. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= GMT was the originator of the concept, wrote the article, and participated in the design, coordination, and implementation of the study. GEH participated in the design, performed technical work, and participated in the writing. JAP participated in the design and performed technical work. TLF participated in the writing of the manuscript. DJT participated in the design of the study. MJD participated in the drafting of the manuscript and performed technical work. LWK participated in the design of the study and coordinated the experiments. All the authors approved the final draft of this manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2253/5/3/prepub> Acknowledgements ================ We thank the members of the Experimental Immunology Laboratory at the Omaha VA Medical Center including; Carlos D. Hunter, Bartlett C. Hamilton III, and Karen C. Easterling for their valuable help with this project.	PubMed Central	2024-06-05T03:55:56.423655	2005-4-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090549/", "journal": "BMC Anesthesiol. 2005 Apr 12; 5:3", "authors": [ { "first": "Geoffrey M", "last": "Thiele" }, { "first": "Gary E", "last": "Hill" }, { "first": "Jacqueline A", "last": "Pavlik" }, { "first": "Thomas L", "last": "Freeman" }, { "first": "Dean J", "last": "Tuma" }, { "first": "Michael J", "last": "Duryee" }, { "first": "Lynell W", "last": "Klassen" } ] }
PMC1090550	Background ========== Propofol is widely used for the induction and maintenance of anesthesia and as a sedative in intensive care units where it is given as a constant intravenous infusion for periods of many days. In addition to its clinical importance, propofol provides a valuable model for understanding the human pharmacokinetics of agents that are concentrated in fat. Propofol has an oil/water partition coefficient (K~oil~) of about 4700 \[[@B1]\], one of the largest of any pharmacological agent. In comparison, the highly lipophilic volatile anesthetics, such as halothane, have a K~oil~of less than 300 \[[@B2]\]. Because of this large fat partition, propofol is highly concentrated in adipose tissue where it has slow uptake and release kinetics. This paper presents the first detailed physiologically based pharmacokinetic (PBPK) description of human propofol pharmacokinetics. The model describes the pharmacokinetics in terms of realistic human parameters, such as the organ blood flow and the tissue/blood partition. The PBPK model is implemented in PKQuest, a new software routine that has now been applied to more than 25 different solutes with a wide range of pharmacokinetic properties \[[@B3]-[@B9]\]. The major limitation in most human PBPK models is the uncertainty in the values used for the tissue/blood partition coefficients, which cannot be directly measured and are usually based on uncertain extrapolations from animal measurements. In general, the tissue partition of solutes has a complex dependence on protein and lipid binding and can vary markedly from tissue to tissue \[[@B10]-[@B13]\]. This means that the PBPK model is dependent on a large number of individual tissue partition coefficients that are not well characterized and, effectively, become adjustable model parameters. However, the highly fat soluble non-polar solutes, such as the volatile anesthetics, are a special case. Their tissue/blood partition is dominated by simple, non-specific partition into the tissue lipid. In a previous application of PKQuest to the volatile anesthetics \[[@B4]\], it was shown that the water/tissue partition could be directly determined just from a knowledge of the fraction of lipid in the different tissues and the value of lipid/water partition coefficient (K~oil~). This means that once the tissue lipid fractions are known (which are not solute dependent), the tissue/blood partition coefficient for any solute of this type is completely characterized by knowledge of the just the one physical parameter, the K~oil~. The PBPK modeling of solutes of this type is not only greatly simplified, but one can have more confidence in the model predictions because of the elimination of most of the adjustable parameters. It is assumed here that this same approach can be applied to propofol. The tissue/blood partition coefficient is equal to the ratio of the tissue/water and blood/water partition: ![](1471-2253-5-4-i1.gif) The tissue/water partition is determined from the fraction of lipid in the tissue and the K~oil~of propofol. The blood/water partition is determined from experimental measurements of the fraction of propofol that is free (unbound) in blood -- defined by the parameter freepl.. Since there is large individual variation in the value of freepl \[[@B14]-[@B19]\] one might regard it as an adjustable parameter that varied from subject to subject. However, it was found that the PBPK model adequately predicted the individual results using one average value of freepl for all subjects. Only one parameter was adjusted for each subject -- the intrinsic liver clearance. All the other PBPK parameters are identical to those that have been used previously in the application of PKQuest to a variety of solutes. A detailed description of the propofol PBPK model is provided in the Methods Section. This PBPK model is evaluated by applying it to the experimental data of Schnider et. al. \[[@B20]\]. This data describes the arterial pharmacokinetics of propofol in 24 healthy volunteers with ages varying from 24--81 years, and at 4 different doses. Each subject was given an initial bolus dose, followed 60 minutes later by a constant 60-minute infusion. In the original publication, the pharmacokinetics for the constant infusion phase was interpreted in terms of a 6 parameter compartmental model using NONMEM. Surprisingly, the kinetic parameters from the constant infusion phase were poor predictors of the kinetics following the bolus injection in the same subject. This suggested that there was some systematic difference between the bolus and infusion kinetics, and effects such as early recirculation, a propofol induced change in liver blood flow or pulmonary sequestration were listed as possible explanations. This new PBPK analysis suggests that pulmonary sequestration is the major factor responsible for this discrepancy. The propofol is formulated as a lipid emulsion because it has a low aqueous solubility. In some subjects, a significant fraction of the bolus emulsion (0 -- 60%) is sequestered in its first pass through the lung, and then slowly released. When a quantitative model (with two additional adjustable parameters) of this sequestration is incorporated into the PBPK model, a single set of PBPK parameters provides a good description of both the bolus and constant infusion phases. Methods ======= Experimental data ----------------- The methodology for the data collection and analysis was described in detail in the previous publication of Schnider et. al. \[[@B20]\]. Briefly, 24 volunteers in 3 age groups (18--34; 35--65; and \>65 years) of 8 each were selected. The individual subjects will be identified as, eg, Subject \# 1--5 where the first number refers to the age group (1: 18--34 years, etc.) and the second number is the individual number for that group. Each subject was given an initial bolus (≈ 20 second) dose (2 mg/kg for subjects \< 65; 1 mg/kg for subjects \>65) followed 60 minutes later by a constant 60 minute infusion of 25, 50, 100 or 200 μg/kg/min. Arterial blood samples were taken at 0, 1, 2, 4, 8, 16, 30, 60, 62, 64, 68, 76, 90, 120, 122, 124, 136, 150, 180, 240, 300 and 600 minutes and the arterial plasma propofol concentration was determined. Each subject was studied on two separate visits, using either propofol with or without EDTA under otherwise identical infusion conditions. Since there was no significant difference with or without EDTA, the data from these two experiments were averaged and were used as the individual data that was fitted with the PBPK model. The percent body fat for each subject was determined using the regression equation of Gallagher et. al. \[[@B21]\] with an additional correction for Asians \[[@B22]\]: ![](1471-2253-5-4-i2.gif) where Sex is 0 for females and 1 for males, Asian is 1 for Asians and 0 for others, Age is in years and BMI = weight/height^2^(kg/m^2^). Although the only value that is directly used in the PBPK calculation for each subject is the percent fat, the age, weight, sex, height and ethnicity enter as covariate parameters through eq. (2). Description of the PBPK tissue model for propofol ------------------------------------------------- The PBPK model is identical to the one that has been used in previous applications of PKQuest (fig. [1](#F1){ref-type="fig"}). The model parameters (organ blood flow, volume, etc.) are identical to those used in previous applications of PKQuest \[[@B3]-[@B9]\]. The connective tissue is divided between two organs: \"tendon\" with a relatively low blood flow, and \"other\" with a higher blood flow \[[@B8]\]. The PBPK values for the tissue weight, blood flow and fraction lipid are listed in Table [1](#T1){ref-type="table"}. A major limitation of this PBPK analysis is that a constant set of \"standard human\" resting organ blood flow is assumed and any hemodynamic changes associated with the uptake and washout of propofol are ignored. The experimental observation that cardiac output is either unchanged \[[@B23],[@B24]\] or only slightly decreased \[[@B25],[@B26]\] with short term propofol supports this assumption. However, Sellgren et. al. \[[@B25],[@B26]\] reported that propofol produced large changes in peripheral blood flow which would suggest that there may be a significant redistribution in the organ blood flow that is ignored in this PBPK model. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic diagram of the PBPK model. The organ \"portal\" refers to all the organs drained by the portal vein. The connective tissue is divided between two organs: \"tendon\" with a relatively low blood flow and \"other\" with a higher blood flow. ::: ![](1471-2253-5-4-1) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Standard human PBPK organ weights, blood flows and fraction lipid ::: Organ Weight (Kg) Fraction Lipid Perfusion (L/Kg) Total Flow (L/min) ------------- ----------------- -------------------- ---------------------- ------------------------ Blood 5.5 ≈0.01 liver 1.8 0.25 0.25 0.45 portal 1.5 0.02 0.75 1.125 muscle 26 0.017 0.0225 0.585 kidney 0.31 0.017 4 1.24 brain 1.4 0.022 0.56 0.784 heart 0.33 0.017 0.8 0.264 lung 0.536 0.017 10.482 5.6184 skin 2.6 0.017 0.1 0.26 tendon 3 0.017 0.01 0.03 other 5.524 0.017 0.02 0.1104 bone 4 0 0 0 adipose 17.5 0.8 0.0422 0.7385 Total 70 5.5877 ::: The only PBPK features that must be uniquely specified for propofol are the tissue/blood partition coefficients. The free (unbound) water concentration (c, amount/liter water) plays a central role in the PBPK calculations (see \[[@B6]\] for details). For example, at equilibrium, c will have the same value in all tissues and blood and the total (measured) concentration (C, amount/kg) is related to c by: ![](1471-2253-5-4-i3.gif) where v~w~^i^(liters/kg) is the water fraction in tissue i and fw~i~is the fraction of the total solute that is free in the water phase. The parameter fw~i~characterizes the equilibrium tissue/blood partition coefficients: ![](1471-2253-5-4-i4.gif) The tissue/blood ratio determines the \"effective\" volume of the different tissues and the overall volume of distribution. As discussed in the Background section, it is assumed that the propofol tissue concentration depends simply on the propofol partition in the tissue water and fat: ![](1471-2253-5-4-i5.gif) where v~w~^i^and v~f~^i^are water and fat fractions in tissue i, c^i^is the free water concentration, C~f~is the fat concentration and K~oil~is the oil/water partition coefficient (= C~f~/c). The values used for the tissue fat fraction (v~f~^i^, Table [1](#T1){ref-type="table"}) are identical to those used in the previous application of PKQuest to the pharmacokinetics of the volatile anesthetics \[[@B4]\]. They were determined from in vitro measurements of human tissue/air partition coefficients for a number of volatile anesthetics with a wide range of oil/air and water/air partition. \[See [Additional file 1](#S1){ref-type="supplementary-material"} for a description of the experimental basis for this set of values of v~f~\]. The value used for K~oil~of propofol was 4715, which was determined by Weaver et. al. \[[@B1]\] from measurements of the water/Diprivan partition and the triglyceride concentration in Diprivan. This is similar to the value of the octanol/water partition of 4300 determined by Tonner et. al. \[[@B27]\] and smaller than the octanol/water partition of 6165 reported by Hansch et. al. \[[@B28]\]. Triglyceride should provide the best model for tissue lipid/water partition. The value that is used in eq. (4) for the propofol fraction that is free in the blood (fw~blood~) is based on experimental measurements of the fraction free in human plasma (freepl), which is related to fw~blood~by: ![](1471-2253-5-4-i6.gif) where rblpl is the blood/plasma concentration ratio. In normal subjects, freepl varies from subject to subject by about ± 40% of the mean \[[@B14]-[@B19]\], primarily because of individual variation in plasma lipids. It is assumed that only the plasma lipid varies from subject to subject, while the red cell lipid is constant, so that the value of the blood/plasma ratio (rblpl) also varies for each subject as a function of freepl: ![](1471-2253-5-4-i7.gif) where rblpl~st~and freepl~st~are the mean experimental human propofol values for the blood/plasma ratio and the free plasma concentration and hmt is the standard hematocrit. To summarize the procedure that is used to determine the tissue blood/partition coefficients for the PBPK model as a function of the PBPK parameter freepl: 1) The blood/plasma ratio (rblpl) is determined from eq. (7) using the experimental values for freepl~st~and rblpl~st~(see below). 2) fw~i~for each tissue is determined using eq. (5) and the standard values of K~oil~and v~f~^i^. 3) fw~blood~is determined from eq. (6). 4) Finally, the tissue/blood partition is determined from these values of fw~i~and fw~blood~using eq. (4). Since the plasma propofol concentration was measured in the experiments of Schnider et. al \[[@B20]\], the whole blood model concentrations were first converted to the equivalent plasma concentration before the model results were output in the plots of plasma model concentration versus experimental data. Standard value for human propofol free plasma concentration fraction (freepl~st~) and blood/plasma ratio (rblpl~st~) -------------------------------------------------------------------------------------------------------------------- As discussed above, the tissue/blood partition depends on the value of the fraction free in plasma (freepl). Experimental measurements of the normal human free plasma propofol fraction (freepl) fall into two different ranges. The smaller value of about 0.01 comes from a series of publications by Suarez and colleagues using ultrafiltration \[[@B14]-[@B16]\]. A larger value of about 0.02 has been reported by several other research groups: Using equilibrium dialysis, Altmayer et. al \[[@B17]\] reported values ranging from 0.014 to 0.026 and Servin et. al. \[[@B18]\] reported an average value of 0.022; while Mazoit and Samii \[[@B19]\] report an average value of about 0.02 using a charcoal co-binding technique. All reports agree that the plasma binding is independent of propofol concentration in the clinical concentration range and that the normal individual variation is about ± 40% of the mean. Although the difference in these measurements of the freepl~st~(0.01 versus 0.02) is small in absolute value, it has a dramatic effect on the pharmacokinetics -- producing a roughly two-fold difference in the tissue/blood partition coefficient (eq. (4)). Since the PBPK analysis (see below) is consistent with a value of about 0.022, this was the value that was assumed for standard values for freepl (freepl~st~). The reported values for the normal human blood/plasma ratio (rblpl~st~) vary from 1.1 to 1.3: \[[@B18],[@B19],[@B29],[@B30]\] and a standard value of the blood/plasma ratio (rblpl~st~) of 1.0 was assumed. Because of the large (40%) individual variation in freepl, it might be regarded as an adjustable parameter that varied from subject to subject. However, since it was found that there was no significant improvement of the PBPK model fit to the individual data when freepl was allowed to vary, in the following analysis it is assumed that all subjects had a freepl of 0.022. Using eqs. (4) -- (7), the corresponding normal model values for the tissue/blood partition coefficients using the value of v~f~^i^in Table [1](#T1){ref-type="table"} are: adipose 84; brain 1.87; liver 2.12; intestine 1.7 and the rest 1.45. These values are in the same range as the experimental measurements of Weaver et. al. \[[@B1]\] after a 2 hour constant infusion in sheep: brain 1.8 -- 2.4; kidney 1.36 -- 1.85; skeletal muscle 0.68 -- 1.4 (the range corresponds to the values for 2 different infusion rates). Assuming that the propofol binding in blood is produced entirely by partitioning between the blood lipid and water, the freepl value of 0.022 corresponds to a value for the fraction of lipid in whole blood (v~f~^blood^) of about 0.0093 (using eq. (5)), which is in the range of the normal human blood total lipid (total plasma lipid of 0.0082 gm/ml \[[@B31]\] and blood/plasma lipid ratio of 1). PBPK description of propofol metabolism --------------------------------------- It is assumed that the kinetics of all the tissues are described by a flow limited, well mixed model, except for the liver, which is described using the \"Dispersion\" model of Rowland and colleagues \[[@B32],[@B33]\] with a dispersion coefficient of 0.3. Measurements of brain, arterial and venous propofol concentrations during infusions in sheep are consistent with a flow limited, well mixed model for the brain \[[@B34]\]. This well-mixed tissue assumption is only correct as a first approximation and there is some evidence that it may not be rigorously correct for muscle \[[@B35]\]. Since the kinetic analysis of Schnider et. al. \[[@B20]\] of the data analyzed in this paper indicated that the kinetics were linear over the constant infusion range (25 to 200 μg/kg/min) that was investigated, a linear PBPK model is used. It is assumed that the propofol removal is entirely the result of liver metabolism and is described by the intrinsic liver clearance (Tclr), defined by: \(8) Q(x)dx = Tclr c(x) dx where Q is the rate of liver metabolism (amount/min) and c is the free liver tissue concentration. Since a dispersion model is used for the liver \[[@B32],[@B33]\], the metabolism varies as a function of distance (x) from the start of the liver sinusoid. For the subjects investigated in this paper, Tclr varies from about 65 to 400 liters/min/70 Kg human. Although Tclr has units of clearance it is not equal to the actual liver whole blood clearance for two reasons. First, the concentration term in eq. (8) is the free liver concentration, not the whole blood concentration. The effective clearance from the blood (Tclr~blood~) is related to Tclr approximately by: \(9) Tclr~blood~= (water / blood partition) Tclr≈ 0.02 Tclr A second, more complicated, correction arises from the position dependence in eq. (8) and the fact that as Tclr increases to infinity; the actual liver clearance rises to a maximum equal to the liver blood flow. A more direct measure of liver metabolism is the steady state \"Fractional Liver Clearance\" defined by: ![](1471-2253-5-4-i8.gif) Table [2](#T2){ref-type="table"} provides a useful conversion between Tclr and Fraction Liver Clearance for the standard human. In the description of the PBPK model fits to the individual subjects (figs. [5](#F5){ref-type="fig"}, [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}), both Tclr and Fraction Liver Clearance are indicated. In PKQuest, either Tclr~liver~or the equivalent Fraction Liver Clearance can be input. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Individual PBPK model fits to the experimental data for the young subjects. The values of the 3 parameters that provided the best fit for each individual subject are listed: Tclr = intrinsic liver clearance; frdose = fraction of the bolus dose sequestered in first passage through the lung; T = time constant for release of sequestered propofol. The value in parenthesis following Tclr is the fraction of the total liver blood flow that is cleared (eq. (10)). ::: ![](1471-2253-5-4-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Individual PBPK model fits to the experimental data for the middle aged subjects. The values of the 3 parameters that provided the best fit for each individual subject are listed (see fig. 5). For 6 of the 8 subjects there was no pulmonary sequestration of the bolus dose. In order to fit the data for subject 2--4, a small amount of extra-hepatic metabolism was required (kidney clearance = Tclrkid = 35). ::: ![](1471-2253-5-4-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Individual PBPK model fits to the experimental data for the old subjects. No pulmonary sequestration of the bolus dose was required in any subject in this age group. Therefore, there is only 1 adjustable parameter (intrinsic liver clearance, Tclr). ::: ![](1471-2253-5-4-7) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Relationship between the intrinsic liver metabolic clearance (Tclr) and the fractional clearance of the total liver blood flow or the absolute steady state liver blood clearance for a standard male (20% fat, fraction free in plasma = 0.02). ::: Tclr (liters/min) Fraction Liver Blood Flow Liver Clearance (liters/min) ----------------------- ------------------------------- ---------------------------------- 10 0.13 0.23 50 0.47 0.846 100 0.68 1.224 200 0.86 1.548 300 0.93 1.674 400 0.962 1.731 500 0.977 1.76 ::: Although there is evidence of extrahepatic human propofol metabolism \[[@B36]\], it has been assumed in the PBPK model that all the clearance results from liver metabolism. With the exception of subject 2--4, the clearance could be modeled by this assumption for all the subjects in the study- that is, the clearance did not exceed the total liver blood flow. For subject 2--4 (see fig. [6](#F6){ref-type="fig"}) an additional extrahepatic (about 10%, assigned arbitrarily to the kidney) metabolism was required to fit the data. It has been assumed in the PBPK model that there is no pulmonary propofol metabolism. This is supported by the experiments of He et. al. \[[@B37]\] who found no significant pulmonary artery -- radial artery propofol concentration difference during a constant infusion. Also, Gray et. al.\[[@B36]\] found no arterial -- venous difference in propofol or propofol metabolites during the anhepatic phase of liver transplantation. However, in opposition to these experiments, Dawidowicz et. al. \[[@B38]\] found a significant pulmonary arterial -- venous difference for both propofol and propofol metabolites, indicating some pulmonary metabolism. Pulmonary sequestration ----------------------- As discussed in the Background section, the PBPK analysis suggests that in some subjects a fraction of the bolus propofol injection is sequestered in the lung and then slowly released. This sequestration is described by two parameters: 1) frdose -- the fraction of the dose that is sequestered; and 2) T -- the time constant of the exponential release from the lung of the sequestered fraction. The rate of release from the lung (R(t)) of the sequestered propofol after the bolus input is described by: \(11) R~bolus~(t) = (1/T) Dose frdoseexp(-t / T) This sequestration was incorporated into the PBPK model simply by dividing the bolus input into two non-sequestered components: 1) A bolus input of Dose × (1-frdose); and 2) An exponential input of total amount = Dose × frdose. The data was also analyzed to check whether there was sequestration of the constant infusion dose. This required a more complicated modification. The constant infusion was again divided into two components. The first, non-sequestered, component was just a constant infusion of Dose × (1-frdose). For the second, sequestered component, it was assumed that a fraction frdose of the constant infusion was accumulated in a well-mixed compartment in the lung that was released with the same exponential time course as the bolus dose. The release for the bolus dose (eq. (11)) corresponds to the bolus response function of this sequestering lung compartment. Thus, the rate of release (R~I~(t)) for an arbitrary input I(t) to this sequestering compartment is equal to the convolution of I(t) and R~bolus~(t): ![](1471-2253-5-4-i9.gif) This input is pre-programmed into PKQuest as one of the input options and is invoked simply by calling this input option. Determination of PBPK parameters -------------------------------- Each subject in the study was given a bolus propofol injection followed 60 minutes later by a constant 60 minute infusion and the PBPK model was used to fit the individual concentration curves. The previously determined \"standard human\" PBPK parameter set (organ blood flow, volume, etc., see Table [1](#T1){ref-type="table"}) was used for all subjects. The tissue/blood partition was determined from eqs. (4)--(7) using a freepl = 0.022, a blood/plasma ratio of 1 and the tissue lipid fractions determined previously (Table [1](#T1){ref-type="table"}) \[[@B4]\]. The values of these standard parameters depend on the percent body fat, which is determined using eq. (2). For each subject the intrinsic liver clearance was adjusted to fit the data. For some subjects, in order to fit the bolus phase it was also necessary to choose the two parameters describing the pulmonary sequestration (frdose and T). Several steps were used to determine these parameters for each subject: 1) Subjectively adjust the liver clearance (Tclr) to fit the constant infusion phase. 2) If both the bolus and constant infusion phase could be adequately fit by one value of the liver clearance, than it was assumed that there was no pulmonary sequestration and step 3 was implemented. In 14 of the 24 subjects, no pulmonary sequestration was required. In the other 10 subjects it was necessary to adjust frdose and T to fit the bolus phase. Depending on the value of T, the slow release from the sequestered compartment can extend into the constant infusion phase, and it was necessary to repeat this cycle of parameter estimation. 3) Finally, a 1-parameter Powel non-linear minimization routine in PKQuest was used to find the value of Tclr that provided the best fit to the entire experimental concentration curve (bolus plus constant infusion) for that subject. This used a weighted least square minimization technique with weights determined by assuming that the standard deviation of the measurement was proportional to the model concentration. The experimental plasma concentration was collected over a time course of 0 to 600 minutes, with the first data points at 1, 2, 4 \... minutes and the last two samples at 300 and 600 minutes. Because of mixing and circulation time effects \[[@B39]\], the PBPK model is not accurate at times \< 2 minutes, so that the first point that was used in the analysis was at 2 minutes. The fits to the 600 minute data was less accurate than for the earlier data points (see figs. [5](#F5){ref-type="fig"}, [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). The 600 minute data point has the most scatter because the status of the subjects was not controlled during the 300 to 600 minute time period and ambulation or food intake during this period could produce large shifts in the PBPK parameters, particularly muscle, liver and fat blood flow. The quality of the fits were quantitated and compared with the 6 parameter NONMEM fit by the use of the \"weighted residual error\" (WRE) defined as the average value of the absolute (measured -- model)/model concentration ratio All the calculations and the graphical output were implemented using the PKQuest software routine. The procedures involved in using PKQuest have been described previously \[[@B3]-[@B8]\]. All of the standard human PBPK parameters (blood flows, organ weights, etc) and the equations for, e.g., converting freepl to tissue partition values, are pre-programmed and do not have to be entered. The only parameters that the user must enter are those that are unique to the propofol study: the bolus input and constant infusion rate; the experimental arterial plasma concentration data; the PBPK parameters weight, height, sex, age, and Tclr, and, for subjects with pulmonary sequestration, frdose and T. All of the figures used in this paper represent standard PKQuest graphical output. \[See [Additional file 1](#S1){ref-type="supplementary-material"} for a sample PKQuest Maple worksheet for one subject\]. Model simulation of normal weight and obese subjects ---------------------------------------------------- The PBPK model parameters were chosen to simulate the experiments of Servin et. al. \[[@B40]\] in which the pharmacokinetics of normal weight and morbidly obese subjects were compared. The average values were used for the fraction free in plasma (freepl = 0.022) and intrinsic liver clearance (Tclr = 162 liters/min/70 kg). This value of Tclr corresponds to a steady state liver clearance of 1.49 liters/min or 83% of the total liver flow for the standard 70 kg man (see Table [2](#T2){ref-type="table"}). Results were compared for normal weight subjects (fat fraction = 20%) and for the average obese subject studied by Servin et. al. Using the reported body weight and ideal body weight of the obese subjects, it was estimated that they had a fat fraction of about 50% based on the regression relationship of Rhode et. al \[[@B41]\] that 69% of the weight greater than ideal weight is fat. One significant change was made in the PBPK parameters for the obese subjects. Adipose blood flow is heterogeneous, with the highest values in subcutaneous tissue and lower values in visceral and perirenal fat \[[@B42]\]. It is probable that morbidly obese subjects have a lower average adipose blood flow then normal weight subjects. An estimate of the adipose blood flow can be obtained from the relationship between cardiac output and excess body weight. The cardiac output measurements in morbidly obese subjects of Alexander et. al. \[[@B43]\] were used to estimate an average adipose tissue blood flow of 0.03 liters/kg, 28% less than the standard value of 0.042 liters/kg (see Table [1](#T1){ref-type="table"}). All the other PBPK parameters are identical for the normal and obese subjects. Since it is assumed in PKQuest that liver weight is a constant fraction of non-fat body weight, the obese subjects have a lower relative liver weight and, therefore, a lower rate of propofol clearance per kg body weight. Servin et. al. \[[@B40]\] used a stepwise infusion regimen: 0.35 mg/kg/min for 5 minutes, 0.2 mg/kg/min for 10 minutes, and 0.1 mg/kg/min for the remainder of the 180 minute infusion. In the simulation, a constant 0.1 mg/kg/min infusion for the entire 180 minutes was assumed. The same propofol infusion rate/kg was used in both normal and obese subjects. PKQuest is freely distributed at <http://www.pkquest.com> Results ======= Pulmonary sequestration ----------------------- Figure [2](#F2){ref-type="fig"} shows a comparison of the PBPK model versus the experimental data for the arterial plasma concentration for subject 1--3. The figures on the left are plotted on an absolute scale, and those on the right on a semi-log scale. The bottom row shows the data for just the bolus phase (0 to 60 minutes). All subjects in the first two age groups received the same initial bolus propofol dose (2 mg/kg), followed 60 minutes later by a 60 minute constant infusion at rates varying from 25 to 200 μg/kg/min. Subject 1--3 had a constant infusion rate of 100 μg/kg/min, which is large enough to swamp out most of the residual from the initial bolus. In fig. [2](#F2){ref-type="fig"} the liver clearance (Tclr) has been adjusted to optimize the fit to the constant infusion phase (60 to 600 minutes). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Error in fitting the kinetics following the bolus injection (0 to 60 minutes) using the PBPK parameters determined by fitting the constant infusion kinetics (\>60 minutes) assuming no pulmonary sequestration. The solid lines are the PBPK model predictions and the squares are the experimental results of Schnider et. al. \[20\]. The top row shows the entire time course and the bottom shows the first 60 minutes. Absolute concentration on left and semi-log plot of concentration versus time on right. ::: ![](1471-2253-5-4-2) ::: It can be seen in fig. [2](#F2){ref-type="fig"} that, using a value of Tclr that accurately described the constant infusion phase, the model prediction for the bolus phase is poor. The deviation between the model and bolus data is unusual -- with too high a model concentration at early times (2, 4, 8, and 16 minutes) and too low a concentration at the 30 and 60 minutes time points. This same discrepancy in the model prediction was noted in the original analysis of this data by Schnider et. al.\[[@B20]\] using a NONMEM compartmental model. It is impossible to explain this deviation using the standard PBPK model. Any variation that improves the fit to the bolus phase significantly worsens the fit to the constant infusion phase. This deviation between theory and experiment could be explained if part of the bolus dose was sequestered in the lung and then slowly released. The initial sequestering would reduce the blood concentration at the early times, while the later release would increase the concentration at later times. There is direct experimental support for human pulmonary sequestration of the lipid emulsion that is used in the propofol formulation \[[@B37],[@B44]-[@B46]\] see Discussion). Figure [3](#F3){ref-type="fig"} shows that the experimental data for the entire time course for subject 1--3 can be accurately fit by the PBPK model if it is assumed that 60% of the bolus dose is sequestered (= frdose) and then released with an eighty minute time constant (T, see, eq. (11)). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Same experimental data as in figure 2 except that 60% of the bolus dose has been sequestered in the lung and released with a time constant of 80 minutes. ::: ![](1471-2253-5-4-3) ::: The PBPK model was modified to look at whether the constant infusion phase dose was also sequestered with the same time constant (see Methods, eq. 9). Figure [4](#F4){ref-type="fig"} shows the fit for this same subject with a sequestered fraction of 0.5 (black line), 0.25 (red line) or 0.1 (green line) during the constant infusion phase. (The sequestered fraction during the bolus phase is 0.6). It can be seen that sequestering as little of 10% of the constant infusion dose significantly worsens the prediction of the PBPK model -- suggesting that there is no sequestration during the constant infusion phase. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Effect of pulmonary sequestration of 0, 10, 25 or 50% of the continuously infused dose. Same experimental data as in figure 2 with 60% sequestration of the bolus dose. The sequestered propofol is released with a time constant of 80 minutes. ::: ![](1471-2253-5-4-4) ::: This subject (\#1--3) was chosen as an illustration because he had the largest value of sequestration. However, all the other subjects in the younger age groups had some sequestration (20 to 40%, fig. [5](#F5){ref-type="fig"}), as indicated by an improvement in the fit to the bolus phase. In the middle aged group, only two of the subjects had significant sequestration (fig. [6](#F6){ref-type="fig"}). The subjects in the oldest group received a bolus dose one half that of the two younger age groups, and none of the subjects in the oldest age group had significant sequestration (fig. [7](#F7){ref-type="fig"}). PBPK analysis for 18--34 year old subjects ------------------------------------------ Figure [5](#F5){ref-type="fig"} shows the semi-log plots of the PBPK arterial plasma concentrations for the 8 young subjects at the 4 different constant infusion rates (all subjects received the same bolus dose). It can be seen that the PBPK model provides a good description of the experimental results for the 8 subjects over the entire range of infusion rates. The values of the 3 adjustable PBPK parameters are listed in the figure. The value of the intrinsic liver clearance (Tclr, eq. (8)) ranges from 85 to 478 liters/min. As discussed above (see Table [2](#T2){ref-type="table"}), the actual liver clearance has a very non-linear dependence on Tclr. The value in parenthesis after Tclr in fig. [5](#F5){ref-type="fig"} is the steady state fractional liver clearance. This clearance provides a better indication of the individual variation in liver clearance. The value of the fraction of the bolus dose that is sequestered (frdose) varies from 0.1 to 0.6, with 5 of the 8 subjects having a frdose 0.2 -- 0.3. The value of the time constant for release of the sequestered propofol (T) varies from 40 to 500 minutes with 5 of the 8 subjects having a value of 80 minutes. The fits are not very sensitive to the value of T, and differences of ± 30 minutes are not significant. The average weighted residual error for these subjects is 13.5% (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of the \"average weighted residual\" percent error for the PBPK model (this paper) and the 6 parameter NONMEM model of Schnider et. al. \[20\]. The PBPK model fits were for both the bolus and the constant infusion input phases, while the NONMEM model fit was for just the constant infusion phase. The PBPK model results are subdivided into the different age groups. The column labeled \"Individual Fits\" is the error when the model parameters were adjusted for each subject (see figs. 4--6). The column labeled \"Average Fits\" is the error when one parameter set with a linear age dependence for the fraction of pulmonary sequestration (eq. (13)) was used for all subjects (see figs. 9--11). The \"Average Fit\" error for the NONMEM model is listed for the case where the same set of 6 parameters were used for all subjects (\"No Covariates\") or an additional 5 covariate parameters (e.g. weight, height, age, etc) were used (\"Covariates\"). ::: PBPK Model Bolus + Infusion Age Group Individual Fits Average Fits ----------------------------------- --------------- ----------------------------------------- ------------------ 18--34 years 13.5% 18.5% 35--65 years 16.1% 23.8% \>65 15.4% 17.6% Average 15.0% 20.0% NONMEM Model Infusion Only \[20\] 14.18% 23%No Covariates 17.39%5 Covariates ::: PBPK analysis for 35--65 year old subjects ------------------------------------------ Figure [6](#F6){ref-type="fig"} shows the semi-log plots of the PBPK model for the 8 middle-aged subjects at the 4 different constant infusion rates. Pulmonary sequestration of the bolus dose was decreased for this age group. In 6 of the 8 subjects, the fits were only slightly improved by adding sequestration, and, to minimize the number of adjustable parameters, it was assumed that sequestration was negligible. A sequestered component is clearly present only for the two subjects 2--6 and 2--8. The propofol kinetics for the other 6 subjects in fig. [6](#F6){ref-type="fig"} can be satisfactorily fit by a PBPK model with just one adjustable parameter -- Tclr. The range of values for both these parameters are similar to those in the younger age group (fig. [5](#F5){ref-type="fig"}). For one subject in this group (\# 2--4) it was necessary to add an additional renal clearance in order to fit the data. The weighted residual error is 16.1% (Table [3](#T3){ref-type="table"}). PBPK analysis for subjects older then 65 years ---------------------------------------------- Figure [7](#F7){ref-type="fig"} shows the semi-log plots of the PBPK model for the 8 subjects in the \>65 year old group at the 4 different constant infusion rates. These subjects were given a bolus dose of 1 mg/kg, half the value of the bolus dose for the younger subjects. Since there was no evidence of pulmonary sequestration in any of these 8 older subjects, the PBPK model required only one adjustable parameter (Tclr). The weighted residual error was 15.4% (Table [3](#T3){ref-type="table"}). Age dependence of PBPK parameters and model predictions using \"averaged\" parameters ------------------------------------------------------------------------------------- Table [4](#T4){ref-type="table"} lists, for the 3 age groups, the average values of the age, steady state fractional liver clearance, the intrinsic liver clearance (Tclr) and the fraction of the bolus dose sequestered in the first pass through the lung (frdose). There is no significant dependence of liver clearance on age. The fraction sequestered (frdose) decreases with age and a linear dependence was determined (see fig. [8](#F8){ref-type="fig"}). ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Age dependence of the fraction sequestered in lung. The dashed line is the linear regression for the data (eq. (13)). ::: ![](1471-2253-5-4-8) ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Age dependence of steady state fractional liver blood flow clearance (Liver Clearance), the intrinsic liver clearance (Tclr), and fraction of bolus dose sequestered in lung (frdose). ::: Age (years) Liver Clearance Tclr (liters/min) frdose ----------------- --------------------- ----------------------- ------------ 29.125 0.772 165 0.3 52.5 0.781 208 0.06875 74.75 0.762 104 0 ::: ![](1471-2253-5-4-i10.gif) The oldest subject group (\>65) was not used in this frdose correlation calculation because they received only half the bolus dose of the other subjects and there was no significant sequestration (see Discussion). Figures [9](#F9){ref-type="fig"}, [10](#F10){ref-type="fig"}, [11](#F11){ref-type="fig"} show the accuracy of the PBPK model predictions when just one set of \"averaged\" PBPK parameters is applied to all the subjects. The \"averaged\" parameters are: 1) Fractional liver clearance = 0.76; 2) Sequestration time constant = 80 minutes; and 3) fraction pulmonary sequestration described by eq. (13). The weighted residual error using these \"averaged\" values is listed in Table [3](#T3){ref-type="table"}. ::: {#F9 .fig} Figure 9 ::: {.caption} ###### PBPK model fits to the experimental data for the young subjects using the same age dependent parameters for all subjects: fractional liver clearance = 0.76; fraction sequestered described by eq. (13), and T = 80 minutes. ::: ![](1471-2253-5-4-9) ::: ::: {#F10 .fig} Figure 10 ::: {.caption} ###### PBPK model fits to the experimental data for the middle-aged subjects using the same age dependent parameters for all subjects (see fig. 9). ::: ![](1471-2253-5-4-10) ::: ::: {#F11 .fig} Figure 11 ::: {.caption} ###### PBPK model fits to the experimental data for the old subjects using the same age dependent parameters for all subjects (see fig. 9). ::: ![](1471-2253-5-4-11) ::: Discussion and conclusions ========================== Pulmonary sequestration ----------------------- The pharmacokinetic evidence for this sequestration comes from a comparison of the kinetics after the bolus injection versus a constant infusion. It was clearly recognized in the original analysis of this data by Schnider et. al. \[[@B20]\] that there was some systematic difference between the kinetics for these two different inputs. Schnider et. al. fit the constant infusion phase kinetics for each subject with a 6 parameter NONMEM model. When these kinetic parameters for the infusion phase were used to fit the initial bolus kinetics, the predicted concentrations for the early time points (2 to 4 minutes) were about 50% greater then the experimental values, while the predicted concentration for the later data points (\>8 minutes) were about 50% less than the experimental values. The same systematic difference is seen in the younger subjects when the PBPK parameters for the constant infusion phase are used to predict the concentration following the bolus dose (fig. [2](#F2){ref-type="fig"}). Pulmonary sequestration of the bolus dose with a slow release provides an explanation of this systematic difference. The sequestration will reduce the concentration of the early time points and, as the sequestered dose is slowly released, increase the concentration of the later time points. A simple model in which a fraction (frdose) of the bolus dose is sequestered and is then released as a single exponential (eq. (11)) provides a good fit to the experimental data (fig. [3](#F3){ref-type="fig"}, and figs [5](#F5){ref-type="fig"}, [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). The fraction of the dose that was sequestered was largest in the younger age group, ranging from 0.2 to 0.6 (fig [5](#F5){ref-type="fig"}). As described in fig. [4](#F4){ref-type="fig"}, there does not seem to be any significant sequestration when the propofol is given as a constant infusion because its subsequent slow release would produce a significant deviation of the model predictions from the observed experimental data. This suggests that the sequestration depends on the concentration of the emulsion when it is mixed with the venous blood at the injection site. The bolus injection rate of 140 mg/70 kg/20 sec propofol in Diprivan corresponds to an injection of about 4.2 gm/min of lipid into the vein of a 70 kg man. In contrast, the highest rate of constant infusion (14 mg/70 kg/min) represents a 30-fold lower infusion rate. There is direct experimental evidence for pulmonary sequestration of propofol following a bolus injection in humans \[[@B37],[@B45]\]. He et. al.\[[@B37]\] simultaneously injected propofol and indocyanine green (ICG) in a central vein and sampled radial arterial blood at 1 second intervals for a period of 1 minute -- the first pass time. They injected 5 mg/70 kg propofol within 1 second -- about twice the rate of the bolus injection of Schnider et. al. \[[@B20]\] that was fit with the PBPK model. A first pass pulmonary clearance of propofol of about 28% was estimated from the difference in the first pass AUC of propofol and ICG. This analysis cannot distinguish between pulmonary metabolism versus sequestration. However, in the same study, He et. al. \[[@B37]\] demonstrated that there was no significant difference in the radial and pulmonary artery concentrations during a 60 minute constant infusion, indicating that propofol is neither metabolized nor sequestered during a constant infusion. In the standard propofol formulation (Diprivan), the propofol is dissolved in a lipid emulsion that is identical to the emulsion (Intralipid) that is used for parenteral feeding. There is also direct evidence that Intralipid is sequestered in the human pulmonary circulation. Zauner et. al. \[[@B44]\] measured the brachial artery -- pulmonary artery concentration difference of Intralipid during and after a constant infusion. At high rates of Intralipid infusion (about 6.5 times the maximum rate used by Schnider et. al. \[[@B20]\]) they observed a significant pulmonary sequestration of about 20% during the infusion, with a subsequent release of the sequestered Intralipid that persisted for at least 15 minutes after stopping the infusion. In addition, Gigon et. al. \[[@B46]\] found fat in the lumen of pulmonary capillaries in human lung biopsies a few minutes after beginning Intralipid infusion. There are large individual variations in the model predictions of the fraction of the dose that is sequestered, varying from 0 to 60%. All of the younger subjects had some sequestration (fig. [5](#F5){ref-type="fig"}) with a median value of about 30%. Only two of the eight middle-aged subjects (fig. [6](#F6){ref-type="fig"}) and none of the subjects in the oldest group (fig. [7](#F7){ref-type="fig"}) had significant sequestration. This variation cannot be explained by experimental variations in the bolus infusion rate, which was carefully controlled. The lack of sequestration in the oldest subjects might be explained, in part, because they received only half the bolus dose of the other two groups. Variation in sequestration could result in significant variations in the pharmacodynamic effect of a bolus propofol injection. This new evidence supporting the concept of pulmonary sequestration is indirect since it is based just on an analysis of the PBPK model. There clearly is a need for more direct experimental measurements to either confirm or rule out this effect. In addition, the predicted differences in the magnitude of the sequestration between the young and middle-aged subjects is a surprising result and one that requires further documentation. Validity of the PBPK propofol model ----------------------------------- The PBPK model (fig. [1](#F1){ref-type="fig"}) requires assumptions about a large number of physiological parameters, such as tissue volumes and flows, and is limited by many simplifying assumptions, such as that the tissue regions are well stirred (except for the liver) and flow limited. The usual criticism of the PBPK approach is that, since it uses a large number of adjustable \"physiological\" parameters that cannot be directly measured, it becomes, in effect, a glorified compartmental model. This criticism applies particularly to the tissue/blood partition coefficients of the different organs, which are impossible to directly measure in humans. The major goal in the development of PKQuest is to address this criticism by the development of a \"standard human\" PBPK data set whose values have been refined by application to more than 25 different solutes with a wide range of pharmacokinetic properties \[[@B3]-[@B6],[@B8]\]. This propofol PBPK analysis uses the previously determined \"standard human\" data set for the organ volumes and blood flows so that these parameters are no longer \"adjustable\". Because of the very high propofol fat partition, the tissue/blood partition is dominated by the partition into the tissue and blood fat. In this PBPK analysis, the tissue/blood partition was directly determined using the tissue fat fractions that were derived in a previous PBPK analysis of the volatile anesthetics \[[@B4]\] along with the experimental value for the propofol fraction unbound in blood (freepl = 0.022), and no other assumptions or adjustable parameters were required. Thus, in this PBPK model there is only one new adjustable parameter required to completely describe the blood concentration curves -- the intrinsic liver clearance. (Two additional parameters are required to describe the sequestration of the bolus dose in some subjects). The very high fat solubility of propofol makes it an ideal candidate for a PBPK model. The partition of propofol in tissue fat dominates the tissue/blood partition coefficient, allowing one to estimate the tissue/blood partition simply from knowledge of the tissue fat fraction. For other less fat soluble solutes, the tissue/blood partition cannot be predicted by this a priori approach, and the number of poorly characterized adjustable PBPK parameters is markedly increased. Comparison of PBPK and compartment models ----------------------------------------- Although compartmental and PBPK models are often regarded as competitors, they are actually complementary and serve different purposes. Compartment models, as implemented in NONMEM, provide an unbiased parametric description of a data set using a using a minimal number of model assumptions. If one is only interested in a parametric description of a given clinical data set, then this is all that is needed. The limitation of the compartment model is that the parameters are only weakly related to physiological variables. One cannot use a compartment model to predict the pharmacokinetics under varying physiological conditions, such as changes in portal, muscle or fat blood flow, or variations in plasma protein binding or body fat content. This paper describes one of the few quantitative comparisons of PBPK and compartmental models that is available in the literature. Table [3](#T3){ref-type="table"} compares the average weighted residual percent error (WRE) for each subject using the 6 parameter NONMEM compartment model versus the 1 parameter PBPK model (for some subjects 2 additional parameters are required to fit the bolus phase). The WRE for the compartment model are for the fits to just the constant infusion phase, while the WRE for the PBPK model is for both the bolus and infusion phase. The quality of the fits for the PBPK model is slightly poorer than that for the NONMEM compartmental model (compartment model: 14.18% error; PBPK model: 15.0% error). The clinically important measure of the quality of the fit is how well it predicts the kinetics using \"averaged\" parameters without any prior information about the individual model parameters. Two different \"averaged\" errors were reported for the NONMEM model (Table [3](#T3){ref-type="table"}). The error was 23% when an identical \"optimal\" set of 6 parameters was used for all subjects. The error was reduced to 17.39% when 5 \"covariate\" parameters were used that were based on the subjects\' sex, weight, height and age. The ability to calculate these covariate parameters is one of the major strengths of NONMEM compartmental modeling. Although some PBPK models have been developed that allow Bayesian population modeling \[[@B47]\], this is not possible with the current version of PKQUEST. Thus, it is not possible to determine the covariate dependence of the PBPK model parameter (Tclr, frdose, T) on subject\'s age, weight, etc. The only correlation that was investigated for the PBPK model was the age dependence of the parameters (Table [4](#T4){ref-type="table"}, fig. [8](#F8){ref-type="fig"}). The only parameter with significant age dependence was the pulmonary sequestration fraction (fig. [8](#F8){ref-type="fig"}). Using this linear age dependent correlation (eq. (13)) an \"averaged\" PBPK model fit was obtained for all subjects (figs. [9](#F9){ref-type="fig"}, [10](#F10){ref-type="fig"}, [11](#F11){ref-type="fig"}). These calculations also include a limited correlation for sex, age, weight and height through the use of an estimate of each subject\'s percent body fat using eq. (2). For the PBPK model, the \"average\" error was 20.0%, significantly worse than the covariate NONMEM model error of 17.39%. It should be emphasized that the PBPK model error is for both the bolus and constant infusion phases, while the NONMEM error is for just the constant infusion phase. An important limitation of the NONMEM compartmental approach is that, since it is only a parametric description of a particular data set, errors may occur when the model is extrapolated beyond this data set. This is clearly illustrated for propofol if one tries to extrapolate the compartmental model to times longer than those used in the experimental data set (10 hours). The time constant (T) for equilibration with adipose tissue is described by: ![](1471-2253-5-4-i11.gif) For an adipose blood flow of 0.042 ml/gm/min and an adipose/blood partition coefficient of 84, T is 2000 minutes, or 33.3 hours. Accurate estimates of steady state clearance (Cl~ss~) and volume of distribution (V~ss~) require that the kinetics be determined for times of about twice that of the longest time constant, eg. about 50 hours. Most kinetic studies are for periods much shorter than this and have led to misleading values for these parameters. This is clearly illustrated by the NONMEM compartmental analysis of Schnider et. al. \[[@B20]\]. Based on measurements for a 10 hour sample period, they found a V~ss~of about 260 liters, about 10 times less than the V~ss~of 1200 to 3940 liters obtained by Campbell et. al. \[[@B48]\], Morgan et. al. \[[@B49]\] and Albanese et. al. \[[@B50]\] using plasma values sampled for times varying from 40 to 100 hours. In addition, Schnider et. al. \[[@B20]\] estimated a Cl~ss~of 1.89 liters/min, significantly larger than the experimental estimates of Campbell et. al. \[[@B48]\] and Morgan et. al. \[[@B49]\] of 1.02 to 1.6 liters \[[@B48],[@B49]\]. In contrast, a much better estimate of V~ss~and Cl~ss~is obtained using the PBPK model. For the standard human (20% fat) the V~ss~for the PBPK model is about 1500 liters, of which about 97% is contributed by the adipose tissue. For body fat varying from 12% to 40%, the PBPK model V~ss~varies from 980 to 3000 liters, in good agreement with the experimental measurements of Campbell et. al. \[[@B48]\] and Morgan et. al. \[[@B49]\]. The intrinsic clearance, Tclr, (eq. (8)), varies from about 70 to 500 liters/min with a median value of about 160 liters/min (Table [4](#T4){ref-type="table"}). This corresponds to a median value for the steady state liver clearance of about 1.49 liters/min/70 kg, with a range of 1 to 1.8 liters/min (Table [2](#T2){ref-type="table"}), which, again, is similar to what is observed experimentally \[[@B48],[@B49]\]. A direct comparison of the PBPK and NONMEM model predictions is shown in fig. [12](#F12){ref-type="fig"} (standard male, 21% body fat) and fig [13](#F13){ref-type="fig"} (obese female, 47% body fat). The PBPK model used the average Tclr value for the young subjects (Tclr = 165) with the body fat determined using the subjects age, weight and height (eq. (2)). The top row compares the model predictions for a 1-hour constant infusion followed by a 9-hour washout. Both models give very similar results over this limited washout time. This is expected since this is the time period for which the NONMEM and PBPK parameters were optimized. The differences in the two models become dramatic if the washout is extended to 49 hours (fig. [12](#F12){ref-type="fig"} and [13](#F13){ref-type="fig"}, middle row), long enough for differences in the value of V~ss~to become important. This underestimate of V~ss~has important clinical implications, which are illustrated in the bottom row of figs. [12](#F12){ref-type="fig"} and [13](#F13){ref-type="fig"} for the case of a constant 5-day infusion, mimicking the use of propofol for long term sedation. As can be seen in fig. [12](#F12){ref-type="fig"}, for the standard male (21% fat) the predicted steady state plasma value using the NONMEM model is about 19% less than that for the PBPK model using the \"correct\" value of V~ss~. This difference becomes significantly greater (53%) in obese subjects (fig. [13](#F13){ref-type="fig"}) with a larger adipose compartment. ::: {#F12 .fig} Figure 12 ::: {.caption} ###### Comparison of NONMEM (Schnider et. al., Table 2 \[20\]) and PBPK model predictions for a standard male (21% body fat, age 53, weight 77 kg and height 177 cm). The top row shows the model predictions for a 1-hour constant infusion of 100 μg/kg/min followed by a 9 hour washout (left, absolute plot; right, semi-log plot). The middle row shows the model predictions when the washout period is extended to 49 hours. The bottom row compares the model predictions for a 5 day constant infusion of 100 μg/kg/min. ::: ![](1471-2253-5-4-12) ::: ::: {#F13 .fig} Figure 13 ::: {.caption} ###### Same as fig. 12, except for an obese female with 47% body fat (age 53, weight 77 kg and height 150 cm). ::: ![](1471-2253-5-4-13) ::: Physiological implications -- body fat fraction ----------------------------------------------- The major advantage of an accurate PBPK model is that it can be used to predict the pharmacokinetics for varying physiological conditions. The primary application of the PBPK approach has been in the field of toxicokinetics where the ability to predict the kinetics under varying physiological conditions is critical (for a recent review, see \[[@B51]\]). The usefulness of these PBPK predictions is critically dependent on having confidence that the PBPK model is not just a set of adjustable parameters but, instead, represents a true \"physiologically based model\" that can be extrapolated to conditions that are different from those that were used to derive the model parameters. The fact that this propofol PBPK model has just one new adjustable parameter (liver clearance), helps to justify this confidence. This section will focus on one physiological variable -- the body fat fraction. The property that most distinguishes the propofol kinetics is its very high oil/water partition coefficient of about 4715. The corresponding adipose/blood partition coefficient is about 84 (eq. (4) which results in a huge volume of distribution, about 2000 liters/70 kg, of which about 97% is in the adipose tissue. One might expect that the body fat content should have a large influence on the kinetics. However, Servin et. al. \[[@B40]\] compared the propofol kinetics for normal weight and morbidly obese subjects during standard surgical anesthesia (180 minute constant infusion of 0.1 mg/kg/min followed by an 8 hour washout) and showed that the differences in the two groups of subjects was small. Figure [14](#F14){ref-type="fig"} shows a PBPK model simulation of the experiments of Servin et. al. (see Methods for details). The black line is for a normal subject (20% body fat) and the red is for the average morbidly obese subject (50% body fat). Servin et. al. followed the washout for 8 hours after stopping the infusion (fig. [14](#F14){ref-type="fig"}, top panel). It can be seen that over most of the infusion and washout period, the concentrations in the obese subjects differs by less than 30% from that for the normal subjects, which is small compared to the inter-individual variation in the experimental study of Servin et. al. Thus, this model simulation is consistent with the main conclusion of Servin et. al. that variations in body fat content have relatively small effects on the propofol kinetics during standard surgical procedures. However, if the washout is followed for 5 days, a period that is long compared to the 33 hour adipose time constant, the difference in kinetics for the obese subjects becomes dramatic (fig. [14](#F14){ref-type="fig"}, bottom panel). ::: {#F14 .fig} Figure 14 ::: {.caption} ###### Comparison of arterial blood concentration in normal weight (black) and morbidly obese subjects (red). The propofol was infused at a constant rate of 0.1 mg/kg/min for 180 minutes and the washout was followed for 8 hours (top panel) or 5 days (bottom panel). ::: ![](1471-2253-5-4-14) ::: The most significant difference that was observed by Servin et. al. \[[@B40]\] for the obese subjects is that they woke up significantly faster than the normal weight subjects after stopping the propofol infusion. Figure [15](#F15){ref-type="fig"} shows the first 20 minutes of the washout period after the 180-minute propofol infusion. Servin et. al. recorded the time and arterial blood concentration of first eye opening. The eye opening arterial blood concentration was the same for obese and normal subjects (about 1 mg/l; fig. [15](#F15){ref-type="fig"}, dashed line). The model simulation in fig. [15](#F15){ref-type="fig"} predicts an eye opening time of about 5 minutes for the obese subjects and 13 minutes for the normal subjects. This is qualitatively similar to what was observed by Servin et. al. (obese: 10.3 ± 6.3 minutes; normal: 18.4 ± 5.7 minutes). ::: {#F15 .fig} Figure 15 ::: {.caption} ###### Comparison of time for eye opening after a 180 minute constant infusion in normal weight (black) and morbidly obese subjects (red). Same conditions as fig. 14, but only the first 20 minutes of the washout is plotted. The dashed line indicates the arterial concentration associated with the first eye opening. ::: ![](1471-2253-5-4-15) ::: Propofol is routinely used for long-term sedation. Figure [16](#F16){ref-type="fig"} shows the model simulation of the arterial concentration for a 10-day constant infusion of 0.1 mg/kg/min in normal weight and morbidly obese subjects. It can be seen that for these long-term infusions, the pharmacokinetics in obese subjects differs significantly from normal subjects. In normal weight subjects the arterial concentration reaches a steady state value after about 3 days, while in the obese subjects the arterial concentration is still rising at the end of the 10 day infusion. The arterial propofol rises to a higher concentration in the obese subjects because the same dose/kg was given to both subjects, while the obese subject has a lower rate of liver metabolism/kg because of the assumption of a constant liver weight per lean body mass. Probably the most important clinical implication of this model analysis is the prediction the kinetics of the washout phase after long-term sedation. Figure [17](#F17){ref-type="fig"} shows the model simulation of the first 60 minutes of washout after a 10 day constant infusion. The infusion rate has been adjusted so that the normal and obese subjects have identical arterial concentrations just before the infusion is stopped. Sixty minutes after stopping the infusion the arterial concentration in obese subjects has dropped to a value of only 65% of the 10-day value, versus 45% in normal subjects. This suggests that blood concentrations during long-term infusions in obese subjects must be carefully monitored to assure rapid awakening when the infusion is terminated. ::: {#F16 .fig} Figure 16 ::: {.caption} ###### Comparison of arterial blood concentration in normal weight (black) and morbidly obese subjects (red). Ten-day constant infusion of 0.1 mg/kg/min. ::: ![](1471-2253-5-4-16) ::: ::: {#F17 .fig} Figure 17 ::: {.caption} ###### Comparison of washout kinetics in normal weight (black) and morbidly obese subjects (red). The 10 day constant infusion rate has been adjusted so that the concentration at the end of 10 days is identical for the normal and obese subjects (normal: 0.1 mg/kg/min; obese: 0.058 mg/kg/min). Only the first 60 minutes of the washout is plotted. ::: ![](1471-2253-5-4-17) ::: Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= D. G. L. performed the PBPK model development; the fitting of the PBPK parameters; and the analysis of the results. T. W. S provided the experimental data and critically evaluated and significantly revised the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2253/5/4/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A\) A sample Maple worksheet for one subject. This worksheet can be used with the freely available PKQuest program to generate all the data and figures that were used for this subject in this paper. B) A detailed analysis and description of the use of the pharmacokinetics of the volatile anesthetics to determine the lipid fraction in different tissues. ::: ::: {.caption} ###### Click here for file :::	PubMed Central	2024-06-05T03:55:56.426107	2005-4-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090550/", "journal": "BMC Anesthesiol. 2005 Apr 22; 5:4", "authors": [ { "first": "David G", "last": "Levitt" }, { "first": "Thomas W", "last": "Schnider" } ] }
PMC1090551	Background ========== Microarray analysis has become a widely used technique for the study of gene-expression patterns on a genomic scale \[[@B1],[@B2]\]. Oligonucleotide and cDNA arrays have been utilized to study mRNA \[[@B3]\] and protein levels \[[@B4]\], to decipher protein-DNA interactions \[[@B5]\], to analyze the DNA copy number \[[@B6]\], to detect methylated sequences \[[@B7]\], and to analyze gene phenotypes in living mammalian cells \[[@B8]\]. Microarrays represent a very complex, multi step technique involving array fabrication, labeling, hybridization, and data analysis. Currently, most laboratories are using either one labeled sample (Affymetrix microarrays) or two labeled samples (cDNA microarrays) for hybridizations, but several applications have been established were three color microarrays are used \[[@B9],[@B10]\]. State-of-the-art microarrays can have from several hundred up to tens of thousands of elements annotated by dozens of parameters. Information on details of the bench work, typically kept in lab notebooks or scattered files, as well as information regarding spotting, reliable tracking of the spotted molecules, scanning, and image quantification settings, is important for the computational analysis and reproducibility of experiments. Every step generates a wealth of data spanning tens of megabytes and in each of them errors may occur or protocols might need optimization to improve results. Moreover, all these information must be archived according to accepted scientific standards, which allow scientists to share common information and to make valid comparisons among experiments. For this reason the Microarray Gene Expression Data Society (MGED) \[[@B11]\] is focusing on establishing standards for microarray data annotation and exchange, facilitating the creation of microarray databases and related software implementing these standards. MGED is heavily promoting the sharing of high quality, well annotated data within the life sciences community. Their initiatives -- MIAME (Minimum Information About a Microarray Experiment) \[[@B12]\], MGED Ontology \[[@B13]\], and MAGE-ML (MicroArray Gene Expression Markup Language) \[[@B14]\] -- maximize the value of microarray data by permitting greater opportunities for sharing information within scientific groups and thus for discovery. These will ultimately affect the description, analysis, and management of all high throughput biological data. The \'list of genes\' resulting from microarray analysis is not the end of a microarray experiment. The major challenge is to assign biological function and to generate new hypotheses. The simplest way to find genes of potential biological interest is to search the normalized data for the highly expressed ones. Additionally, identifying patterns of gene expression and grouping genes into expression classes can provide greater insight into their biological relevance. For this purpose several supervised or unsupervised clustering algorithms like support vector machines (SVM), hierarchical clustering, k-means, self organizing map (SOM), or principal component analysis (PCA) are in use. The annotation of genes or gene clusters can be achieved by mapping them to the Gene Ontology (GO) \[[@B15]\] in order to provide insights into relevant molecular functions, biological processes, and cellular components \[[@B16]\]. Another way to identify genes of biological interest is to map the normalized data or gene expression clusters \[[@B17]\] to known metabolic pathways as provided e.g. by KEGG \[[@B18]\] or BioCarta \[[@B19]\]. Several academic as well as commercial systems are available that address at least some of the needs such as laboratory information management systems (LIMS) \[[@B20]\], microarray databases \[[@B21]-[@B24]\] and repositories, normalization, clustering, pathway or GO mapping tools or expression analysis platforms \[[@B25]\]. However, freely available systems which integrate all the aspects mentioned above are rare and may lack important issues like usability, scalability, or standardized interfaces. Furthermore, for such integrated systems it is desireable to use a uniform and state-of-the-art software architecture in order to enhance setup, maintenance and further development. We have therefore developed a Microarray Analysis and Retrieval System (MARS) using latest Java 2 Platform, Enterprise Edtition (J2EE) software technology. MARS provides modules mandatory for microarray databases: • a laboratory information management system (LIMS) to keep track of information that accrues during the microarray production and biomaterial manipulation • MAGE-ML export of data for depositing to public repositories e.g. ArrayExpress \[[@B26]\], GEO \[[@B27]\] For these components already existing projects \[[@B21],[@B23],[@B26]\] have been evaluated. Their advantages as well as disadvantages have been taken into account for the design of MARS. Widely used concepts have been taken into consideration and accepted standard libraries like MAGE-STK \[[@B11]\]have been used whenever possible. Additionally, we extented this solid foundation and added novel features which can be highlighted as distinct advantages of the MARS system. • a quality management application storing necessary quality control parameters indispensable for high-quality microarray data • Web services to connect several well established tools such as normalization, clustering and pathway annotation applications • applications for microarray normalization, gene expression clustering, and pathway exploration that are tightly integrated into the microarray analysis pipeline • a novel, comprehensive, and Web based user management system to administrate institutes, groups, users, and their corresponding access rights Implementation ============== Software architecture --------------------- MARS is based on a three tier architecture (Figure [1](#F1){ref-type="fig"}) using the Java 2 Platform, Enterprise Edition (J2EE), which defines a standard for developing multi tier enterprise applications. The J2EE platform simplifies the development of enterprise applications using on standardized, modular components like Enterprise JavaBeans (EJB), Java Servlets, Java Server Pages (JSP), and XML technology. A relational database (Oracle or PostgreSQL) builds the data- or Enterprise Information System tier. In the middle tier the J2EE compliant application server JBoss \[[@B28]\] is situated. It manages the access to the relational database as well as the interaction with the data. The Web server in conjunction with a servlet-container is responsible for the presentation tier. All the servlets and JSPs are executed to enable input and output of an application and to manage the applications workflow logic. An advantage of a multi tier architecture is that different tiers can be deployed to different servers, enabling load distribution as well as scalability. Systems ------- The database schema, the business logic, and the Web interface can be subdivided into five major groups: ### 1. Microarray production To address the needs of many laboratories which produce their own microarrays, MARS includes a generic array production LIMS. It manages data regarding the substances (clones) and their localization in microtiter plates, the array design spotted on the support, as well as single arrays and array batches. The flexible and generic database design facilitates mapping of the steadily changing laboratory workflow. Additionally, each plate can be assigned to a library, which designates the organism and contains details about the cloning vector, forward and reverse primer and standard molecule annotations including gene name, accession number, UniGene number, and sequence. Substances stored in microtiter plates may undergo certain manipulations such as PCR amplification. Therefore a PCR amplification event can be assigned to a plasmid plate in order to generate a PCR plate in the database. After entering the information necessary for spotting, a file is generated and prepared for download. This file is used by the spotting robot software to generate an array design file. After the spotting run has been completed, the array design file has to be uploaded into MARS. For each spotting run an array batch has to be created in MARS, and all slides spotted by this spotting run have to be assigned to this array batch. Additionally, important parameters regarding the spotting run such as temperature, duration, or humidity can be assigned to this array batch. Barcode tracking is employed for plates as well as for arrays to reduce possible input errors. Laboratories using commercial arrays have to upload the array design instead and define an array batch afterwards. ### 2. Sample preparation Samples can be annotated in a user-customizable manner. MARS allows the annotation of biological descriptions such as the source and characteristics of a sample (e.g. tissue and disease), any genetic and chemical manipulation and stimulation. Performing such annotations in free text fields leads to large undefined vocabularies and makes them difficult to query. Thus, three different annotation types are provided: 1) enumeration enabling the usage of defined vocabularies or ontologies, 2) numbers to allow scoring and counting and 3) free text. Annotated samples will be linked to an extract, enabling a lab worker to annotate the extraction method, protocol, concentration, purity, and quantity. The labeled extract stores information on used extract quantity, the label and the labeling protocol. ### 3. Hybridization and raw data management The hybridization page archives parameters regarding the hybridization tool and method and is linked to the used labeled extracts. In contrast to several other microarray databases MARS can handle any number of labeled extracts and thus allows the storage of multi color experiments. Resulting images from hybridized scanned slides can be uploaded to MARS and added to a hybridization record. It is noteworthy that a hybridization can have several image sets associated with images of different scanner settings. After analyzing the images several different raw datasets analyzed with different program settings can be uploaded and added to the appropriate image set. ### 4. Experiment annotation A set of hybridizations forms an experiment. To store the experimental design these hybridizations can be divided into classes, paired, and flagged as a dyeswap hybridization. Additionally, an experiment can be annotated using MGED Ontology definitions (Figure [2](#F2){ref-type="fig"}) to specify the perturbational, methodological, and epidemiological design, as well as the biological properties. Transformed datasets can be added to classes and their corresponding raw dataset. ### 5. Quality management To ensure high quality data and to allow the detection of possible sources of errors, a powerful quality management system has been integrated into MARS. This system is based on standard quality control procedures conducted during microarray production as well as during sample preparation, extraction and hybridization. In order to control the quality of PCR and purified PCR products generated during probe production, authorized users can upload gel images and analyze the bands according to a predefined schema (Figure [3](#F3){ref-type="fig"}). Based on this schema, PCR products can be identified later as a source of bad or missing spots on a slide. Quality annotation can be viewed by any user. Slides can be scanned after fixation and/or after staining and parameters like spot walking or the number of missing spots are used to determine slide quality. In addition to array production quality controls, it is also necessary to check the quality of samples and its extracts. Data gained from an Agilent Bioanalyzer or gel images can be uploaded and analyzed either automatically (Bioanalyzer file) or manually (gel images) (Figure [4](#F4){ref-type="fig"}). Labeled extracts can be measured with a spectrophotometer to assess the efficiency of dye incorporation. Results of these measurements can be entered into MARS and the corresponding efficiency is calculated automatically. Finally, the quality of a hybridized slide is analyzed by extracting and displaying several statistical parameters from the raw data result file and by examining positive and negative controls printed onto a slide. Data interfaces --------------- One of the most important parts for the acceptance of a database is the data import interface. To allow the import of generic file formats, we have implemented a user definable parser that allows to read any tab delimited text file. The user has to define a file format where file columns are assigned to appropriate database fields. MARS allows to define file formats for importing plates, raw datasets, transformed datasets, and array designs. Any file that has to be imported, linked, or used has to be uploaded to MARS at first. Afterwards these data can be analyzed by the users at their office desk without having to use another central storage system. Uploaded files are stored on the servers file system where MARS has been installed. Additionally, links to these files are maintained in the relational database to prevent the deletion of already imported, linked, or used files. The implementation of other Web based applications and more important, the usage and correct linkage of their stored data have been addressed by an External Application Connector Interface. Additional applications like supplementary quality checks can be added without any additional coding in MARS. The MARS user interface is dynamically displaying links to all former registered applications. The Microarray Gene Expression Markup Language (MAGE-ML) has emerged as a language to describe and exchange information about microarray based experiments \[[@B29]\]. MAGE-ML is based on XML (eXtensible Markup Language) and can describe microarray designs, microarray manufacturing information, microarray experiment setup and execution information, gene expression data, and data analysis results. By using the Java MAGE-STK (Mage Software Toolkit) \[[@B11]\] MARS is able to export samples, extracts, labeled extracts, arraydesigns, raw datasets, or whole experiments including several hybridizations. Web service ----------- In order to grant users access to MARS with software they are familiar with (e.g. BioConductor \[[@B30]\] or Matlab \[[@B31]\]), MARS provides a well defined Simple Object Access Protocol (SOAP) interface. SOAP is an XML-based communication protocol and encoding format for inter-application communication. After minor software adaptions these interfaces allow to authenticate against MARS, to browse own and shared datasets, to download raw data, to filter the data, and to insert transformed datasets into MARS. To take advantage of the SOAP Web service we provide a Java library called MARSExplorer, that allows software developers to extend their programs with data access functionality to MARS. Additionally, if no firewall is located between the client software and MARS, the MARS API (Application Programming Interface) can be used to access public accessible methods via the RMI (Remote Method Invocation) interface. Access control -------------- To avoid unauthorized database access in a multi user environment the control of user access is a crucial criterion for the acceptance of any database managing functional genomic data. Furthermore, the definition of several fine grained user access levels that allow to visualize, edit or delete data (e.g. expression and sample data, protocols) based on the user rights is mandatory. Therefore we have developed an extensible and easy to use authentication and authorization system (AAS) which rests upon the same technology as MARS. In addition to its Web based management interface, the AAS provides software libraries that enable existing and new applications the integration of highly sophisticated authentication and authorization mechanisms. Moreover, the AAS provides single-sign-on to all its connected Web based applications. Since this AAS can also be used in various projects or institutions relying upon freely available software, MySQL has been choosen as database management system. If desired, this AAS can also manage Windows and Unix accounts using SAMBA \[[@B32]\] and LDAP (Lightweight Directory Access Protocol) \[[@B33]\]. For instance, at the Insitute for Genomics and Bioinformatics all Web based applications and user accounts are administrated by one single instance of the AAS. Results ======= Database -------- All MARS user interfaces are providing a consistent look and feel and are very intuitive to use. In general, the Web based user interface can be divided into two types of user interaction pages: The first one is an input form, where a user can record required and optional data according to the MIAME standard. Required fields are marked in magenta and are validated for correct input. The second allows to list all stored records. To keep the information on a page simple, a user can hide unnecessary datafields. Furthermore it is possible to query for specific records (Figure [5](#F5){ref-type="fig"}) using the MARS report query tool. Because all Web pages are linked together, MARS permits to follow all conducted steps from the transformed data back to the corresponding well in a microtiter plate and to visualize the results of quality controls. The description of an experiment including hybridizations and their raw datasets is typically the starting point for further analysis. Analytical pipeline ------------------- The usability of MARS and the functionality of the provided interfaces and APIs (Figure [6](#F6){ref-type="fig"}) are revealed by the integration of MARS into an analytical pipeline of microarray analysis, beginning with image analysis, normalization, gene expression clustering, and finally mapping of gene expression data onto biological pathways. After entering all required information into MARS, the first step is to normalize the raw data gathered from the image analysis software in order to remove systematic and random errors inherent in the data. ArrayNorm \[[@B34]\], an application for visualization, normalization and analysis of two-color microarray data facilitates these essential steps. Raw data including the definition of experiment classes (biological conditions) and pairs (replicated or dye swapped slides) from whole experiments can be loaded from MARS into ArrayNorm. After visualization and applying different normalization methods like linear regression, LOWESS, or self-normalization, the transformed intensities can be written back to MARS, including the history of the applied methods. The next step in the analytical pipeline is usually gene expression cluster analysis to extract the fundamental patterns inherent in the data and to organize genes with similar expression patterns into biological relevant clusters. Normalized gene expression data can be loaded into Genesis \[[@B35]\]. Genesis allows to cluster the dataset using various similarity distance measurements and different clustering algorithms like hierarchical clustering, k-means, self-organizing maps, principal component analysis, correspondence analysis, and support vector machines. Moreover it is possible to perform one-way ANOVA to identify differentially expressed genes and to incorporate the Gene Ontology (GO) to map gene expression clusters to GO terms. Results can be written back into MARS. Finally, the Pathway Editor \[[@B36]\] provides the opportunity to access MARS and to map data either from whole experiments or from gene expression clusters to specified pathways in order to get an overview of gene expression changes and their influencing factors. All aforementioned applications have integrated MARSExplorer to connect to MARS and to query, up- and download datasets. Discussion ========== The database design, state-of-the-art software technology, well designed user interface, and its application interfaces make MARS a powerful tool for storing, retrieving, and analyzing multi color microarray data. The fusion of Web based and standalone applications provides researchers with an unique set of computational tools for genomic and transriptomic data. The main strengths of MARS are: 1. Data interfaces ------------------ Fundamental for the acceptance of a database are the data interfaces. In principle two types of data interfaces for human computer interactions can be distinguished. Standalone applications allow better program-user interactions while having the drawback that several or even very old versions are in use. On the other hand Web based applications can be easily used on every computer without any installation effort and they provide the same and newest version to all users with the cost of limited user interaction. To ensure data integration and good usability we have developed the core data manipulation and storing functions using Web based technology and for data analysis we are using robust applications. 2. Application interfaces ------------------------- Excellent usability does not only account for primely data interfaces. The ability to easily import data and the availability of well defined application interfaces are also crucial. Different institutions use diverse, mostly self tailored applications with proprietary and varying data formats. MARS provides several data and application interfaces. To import data we provide user definable and manageable parsers. When a user is uploading data, MARS tries to find an appropriate parser based on the file data or format header. Once the data is uploaded and stored, the data can be analyzed using the provided applications. For scientists who would like to analyze their data with other software, MARS provides also a Web service data interface. After some slight adaptations, users can authenticated and down- or upload data. Providing a Web service interface allows through its wide spread and platform independence to be implemented in all well-established programming languages and in tools like Matlab or BioConductor. Existing Web applications can be plugged-in using the EACI that enables the linkage between data provided by the plugged-in application and data stored in MARS. Moreover it is possible to extend MARS without having to amend the MARS source code. 3. Quality management --------------------- In order to assure high-quality data and to understand or optimize lower value data it is important to be able to trace back all conducted quality control steps. MARS traces several quality measurements performed during the microarray production as well as during the sample preparation, extraction, and hybridization process. These quality checks are implemented as an additional application called MARS-QM, which is tightly integrated into MARS. 4. Data sharing and export -------------------------- MARS enables users to share their datasets with other users. Supplementary to the user oriented data management an institution oriented level has been introduced. This amelioration allows several institutes to store their data into one data repository without having to share common settings and resources such as scanners, but offering the possibility to share the data among them. Besides the sharing of microarray experiment data we provide the possibility to export hybridizations and experiments using the common exchange format MAGE-ML. This feature facilitates the easy sharing and publishing of high quality, well annotated data within the life sciences community by uploading the generated files to public repositories like ArrayExpress \[[@B26]\]. 5. User management ------------------ Since microarray- as well as the corresponding quality control data may contain highly sensitive data, we have integrated our AAS into MARS to provide authentication and fine grained authorization mechanisms. The combination of AAS and External Application Connector Interface provides through a single-sign-on mechanisms and dynamic linkage of data the possibility to assemble heterogeneous Web applications to one powerful suite. Because information attached to molecules is changing quickly, we are currently implementing the possibility to update and enhance the information tagged to a molecule. Changing this information on the molecule level may affect already existing results. In order to avoid such precarious alterations, a user should be able to update the molecule information for each experiment separately instead of replacing the initial molecule information. Further ongoing projects concentrate on the integration of Affymetrix GeneChip arrays into MARS and the improvement of MAGE-ML export capabilities in order to obtain approval from the ArrayExpress annotation team. Both features will be made available to the public in the next major release. Conclusion ========== In summary, we have developed an integrated system consisting of a microarray database and a microarray quality control database, that has been tailored to serve the specific needs of microarray based research projects. Due to the unique fusion of using Web based and standalone applications connected to the latest J2EE application server technology, bioinformatics researchers receive the benefits of standards-based software engineering. The system can provide a model how to build up a similar platform for other emerging functional genomics technologies. Availability and requirements ============================= • Project name: MARS • Project home page: <http://genome.tugraz.at/Software/MARS/MARS.html> • Operating system: Solaris, Linux, Windows • Programming language: Java, HTML • Other requirements: Java JDK 1.4.x, Oracle 9i, MySQL 4.0.xx, Server with at least 1 GBytes of main memory • License: IGB-TUG Software License • Any restrictions to use by non-academics: no Installation of MARS is not complicated and should be manageable within a few hours if necessary access rights especially to Oracle and MySQL are granted. Step-by-step instructions are provided at the projects Web site together with the files and scripts necessary for installation. The reference installation of MARS is running on a Sun Fire V880 server under Solaris 9 using Oracle 9i as Database Management System. Attached is a Storage Area Network (SAN) with 2 TBytes. The production instance of MARS contains information from more than 1000 microtiter plates, 24 array batches, 232 hybridizations, and 312 rawbioassays with about 9,170,000 datapoints. Authors\' contributions ======================= MM, RM, and AS designed and implemented the current version of MARS. They were responsible for the database design, the development of the business- as well as presentation logic. JH developed the quality management system and incorporated it into MARS. MM and JH were the lead developers of the AAS. HH, AP, GS, and MS have been involved in the compilation of the user requirements document and contributed to the conception and design of the system. ZT was responsible for the overall conception and project coordination. All authors gave final approval of the version to be published. Acknowledgements ================ The authors thank the staff of the Institute for Genomics and Bioinformatics for valuable comments and contributions. This work was supported by the Austrian Science Fund (Grant SFB Biomembranes F718) and the bm:bwk, GEN-AU BIN (Bioinformatics Integration Network) and GEN-AU GOLD (Genomics of Lipid-Associated Disorders). Michael Maurer, Robert Molidor and Juergen Hartler were supported by a grant from the Austrian Academy of Sciences. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Three tier Java 2 Enterprise Edition software architecture. The J2EE platform simplifies the development of enterprise applications by providing standardized modular components like EJBs, JSP and Servlets. Furthermore it is providing a complete set of services to those components. ::: ![](1471-2105-6-101-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Experiment annotation. Web interface to define microarray experiments according to the MGED Ontology. ::: ![](1471-2105-6-101-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Quality control. A gel image from PCR products can be scored and associated to a plate. ::: ![](1471-2105-6-101-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Quality control. Bioanalyzer analysis to check the RNA quality for a given RNA extract. ::: ![](1471-2105-6-101-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Typical MARS interface listing stored records. It allows to query for specific records using the user friendly query tool. ::: ![](1471-2105-6-101-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### MARS system interactions. MARS and MARS-QM are deployed in a J2EE compliant application server. Interaction is possible either with a standard Web browser or an application supporting the SOAP or RMI protocol. The External Application Connector Interface (EACI) facilitates to connect to data from additional Web applications. SOAP and http/https enable MARS access also through firewalls. ::: ![](1471-2105-6-101-6) :::	PubMed Central	2024-06-05T03:55:56.431538	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090551/", "journal": "BMC Bioinformatics. 2005 Apr 18; 6:101", "authors": [ { "first": "Michael", "last": "Maurer" }, { "first": "Robert", "last": "Molidor" }, { "first": "Alexander", "last": "Sturn" }, { "first": "Juergen", "last": "Hartler" }, { "first": "Hubert", "last": "Hackl" }, { "first": "Gernot", "last": "Stocker" }, { "first": "Andreas", "last": "Prokesch" }, { "first": "Marcel", "last": "Scheideler" }, { "first": "Zlatko", "last": "Trajanoski" } ] }
PMC1090552	Background ========== The volume of published biomedical research, and therefore the underlying biomedical knowledge base, continues to grow. The MEDLINE 2004 database is currently growing at the rate of about 500,000 new citations each year \[[@B1]\]. With such growth, it is challenging to keep up-to-date with all of the new discoveries and theories even within one\'s own field of research. Methods must be established to aid biomedical researchers in making better use of the existing published research and helping them put new discoveries into practical use \[[@B2]\]. Text mining and knowledge extraction are ways to aid biomedical researchers in identifying important connections within information in the biomedical knowledge base. A subset of natural language processing (NLP), text mining and knowledge extraction concentrate on solving a specific problem in a specific domain identified a priori. For example, literature searching may be improved by identifying all of the names and symbols used in the literature to identify a particular gene \[[@B3]\], or potential new treatments for migraine may be determined by looking for pharmacological substances that regulate biological processes associated with migraine \[[@B4],[@B5]\]. Similar to acronym and abbreviation extraction, which has been studied by several groups \[[@B6]-[@B8]\], the problem of gene and protein name synonymy is one that can be addressed with the aid of text mining. Many genes and proteins have multiple names with several orthographic and lexical variants. Gene names are often not used consistently, and new names continue to be created \[[@B9],[@B10]\]. Many attributes of a gene, such as its phenotypes and polymorphisms, may lead to it being given several names over time. Also, genes may have names that are later retracted when new information becomes available \[[@B11]\]. While databases of gene names exist, they have several limitations. Gene name databases such as FlyBase \[[@B12]\] and Genew \[[@B13]\] are restricted to a single species (fruit flies and humans, respectively). LocusLink includes genes and names for several species, but does not attempt to include all names, symbols, and lexical variations that refer to a gene. The Genew database was created by the Human Genome Organisation for the purpose of establishing an approved set of unique gene names and symbols for every gene in the human genome \[[@B14]\]. However, Genew is focused on creating the set of gene names recommended for use in biomedical writing. It is not intended to be a complete collection of the gene names and symbols actually used in the biomedical literature \[[@B15]\]. Since the gene names and symbols used in a journal article are fixed once published, later correction of improper names does not affect the prior published literature. Therefore, the name space representing a gene can become quite large between the time a gene is first suspected and when it is well studied and has a universally agreed upon name. In addition, gene and protein names overlap. They are often used in place of one another within the literature, with the intended gene or protein being dependent upon context. When conducting a literature review, it is useful to search for both gene and protein names simultaneously \[[@B9]\]. Therefore in this work we make no distinction between names of genes and the names of the proteins for which they encode. An automatically generated list of synonyms would be a useful aid in searching the biomedical literature. These could then be used to improve the recall of genomics investigators trying to find all known information on a gene or protein, regardless of the name or names used in a specific article, although a decrease in precision may result in cases where some of the symbols are shared by multiple genes. An automatically generated list of name synonyms would also be useful in further work on extracting other genomics information from textual sources \[[@B16]\]. To make efficient use of the available data when mining the biomedical literature for relationships, it is important to recognize differing names for identical concepts and treat these as a single concept \[[@B17]\]. The basic idea of name synonym extraction is to automatically extract synonymous names for a given concept from natural language text. In this case, the goal is to extract the names and symbols referring to an individual gene from MEDLINE abstracts. There is significant prior work in this area, done over the last five years by Yu and Agichtein. Yu \[[@B18]\] first worked on gene name synonym extraction with a system that extracted gene name synonyms based on manually identified patterns in which gene name synonyms commonly occur. Domain experts were used to identify common patterns. Yu et al. estimated the precision of their system to be approximately 71%. Recall measurements were not published. Yu and Agichtein \[[@B3]\] then worked together to combine several gene and protein name synonym text-mining approaches. Their best single system, a pattern-based system named Snowball, was based on Brin\'s Dual Iterative Pattern Expansion (DIPRE) system for the Web \[[@B19]\], which had previously been adapted for extracting relationships from large text collections \[[@B20]\]. A small set of initially known facts is used to find the patterns in which these facts occur within a large corpus. Then these patterns were used to extract more facts, which in turn were used to find more patterns. Yu and Agichtein combined four approaches, including Snowball, and GPE, a system based on labor-intensive manually created patterns and heuristic rules, into a single system, by computing the overall system confidence in each synonym pair. The overall confidence measure for the Combinedsystems was defined as one minus the probability that all of the other systems are incorrect, which is the product of one minus the individual confidences. They found that the Combinedapproach worked better than any individual approach, producing a recall of about 80% with a precision of about 9%. Automatic gene and protein synonym extraction systems have not been put into general use, perhaps because the current level of performance is inadequate for many purposes. It is therefore important to investigate alternative and complementary approaches. Additionally, since the primary work in this area has been done by a single group of investigators, it is essential that other researchers investigate this problem to verify the reproducibility of the results. Results ======= Running our system on the test collection for 9 iterations took approximately 14.5 hours on a 1.7 GHz Pentium 4 with 512 M of RAM. For rapid prototyping the system was implemented in Python, an interpreted language. It is expected that recoding in a compiled language could substantially reduce the execution time. The experiment produced two kinds of results: performance measures and error analysis. The performance measures summarize the quality of the extracted information. Error analysis provides insight into the strengths and weaknesses of the approach. Performance measures -------------------- System performance was measured using the precision, recall, and F-score of the extracted set of synonym pairs, as well the absolute and relative number of correct pairs extracted, cumulative for each iteration. Precision is defined as the number of correct pairs, divided by the number of pairs extracted. Recall is defined as the number of pairs extracted that are also present in the recall gold standard, divided by the number of pairs in the recall gold standard. The F-score is the harmonic mean of precision and recall, defined as 2\precision\recall / (precision + recall) \[[@B9]\]. Figure [1](#F1){ref-type="fig"} shows precision versus recall of the extracted synonym pairs, starting with the first iteration at the left-most point and continuing to the 25th iteration at the right-most point. The graph includes plots of both FOUND pairs (synonym pairs explicitly found in the text by the patterns), as well as FOUND plus INFERRED synonyms (pairs inferred by the graph traversal algorithm). The first iteration achieved a precision of about 25.0%, at a recall of about 6.2%. Precision declines and recall increases practically monotonically over the 24 following iterations to a high recall of about 27.3%, and a precision low of 5.9%. Figure [2](#F2){ref-type="fig"} presents the F-score at each iteration, and again the graph includes plots of both FOUND synonyms as well as FOUND+INFERRED. The maximum F-score of 18.35% for FOUND+INFERRED occurs at iteration 9 (precision 16.18%, recall 21.33%), gradually falling off during subsequent iterations. The use of inference does not greatly impair the algorithm\'s overall accuracy (as measured by the F-score) until approximately iteration 15. The absolute number of correct pairs extracted is presented in Figure [3](#F3){ref-type="fig"}. Including pairs identified using the inference capability of the network consistently found more pairs than not using the inference capability. At the maximum F-score the system using FOUND+INFERRED synonyms extracted 539 correct synonym pairs, including only the FOUND pairs yielded 479 synonym pairs. The approximately 10% (12.5% at iteration 9) difference in extracted pairs is fairly consistent across all iterations after the initial iteration. Figure [4](#F4){ref-type="fig"} compares the results of our system with those of Yu and Agichtein\'s Snowball(their best automated pattern-based approach) and Combined(their best overall approach) systems, interpolated from published graphs. The maximum F-score we obtained is comparable with that of Snowball(16.77%, precision 52%, recall 10%), but less than that of the Combinedsystem (30.24%, precision 62%, recall 20%). The combined system of Yu and Agichtein had superior performance to any single method. Another useful measure of system performance is the amount of knowledge extracted per unit of instance knowledge input to the system. This can be interpreted as a measure of how efficiently the algorithm uses the seed data. Figure [5](#F5){ref-type="fig"} compares the number of correct extracted pairs to the number of seeds used by our system and those of Yu and Agichtein. Results are shown at the point of maximum F-score in order to provide a consistent comparison. Our system used 8 seed pairs, and 539 correct synonym pairs were extracted. The Snowballand Combinedsystems used 650 seed pairs and extracted 700 and 950 correct synonym pairs respectively. The number of correct pairs divided by number of seeds used gives a ratio of 67.38 for our method, with the other systems having much smaller ratios of 1.08 and 1.46 respectively. The Snowballand Combinedsystems may not have actually required all 650 seed pairs given as input. However, peak performance of these systems was achieved after only two iterations, implying that the large number of seeds had a substantial influence on the reported results. Further study on the Snowballand Combinedsystems is needed is determine how many seed pairs are actually required. Error analysis -------------- Two kinds of errors were studied, precision errors and recall errors. Precision errors occurred when the algorithm extracted symbol pairs that were later not verified as synonyms by the precision gold standard data set. These are false positives. Recall errors occurred when the algorithm failed to extract symbol pairs present in the recall gold standard data set. These are false negatives. Errors were studied at the point of maximum F-score, iteration 9. ### Recall error analysis Recall errors were categorized into two pre-defined and mutually exclusive categories, No matching pattern, and Pattern not accepted. The No matching patternerror category included all recall errors for which the pattern generation routines failed to identity a pattern that matched the given pair in the abstract text. Pattern not acceptederrors included those recall errors for which a matching pattern was found, but the matching pattern or patterns were not accepted during the pattern selection optimization step. Using a random sample of 100 false negatives, the majority, 65%, were attributed to the system failing to generate a pattern that matched the recall synonym pair. The remaining 35% of recall errors were due to matching patterns not being accepted by the pattern optimization step. Some of the recall errors identified as No matching patternmay be fixable using a more flexible matching algorithm. In the current system, exact word matching is required of surrounding and intervening words. Small variations in contextual words may have made the algorithm fail to extract a synonym pair that could have been found with more flexible pattern matching. For example, our system treats the patterns \"\$GENE\$(\$GENE) gene\" and \"\$GENE\$/\$GENE gene\" as completely separate patterns. A more flexible \"fuzzy\" matching system could allow a pair of gene names followed by the word \"gene\" to be treated as variants of a single pattern. This approach requires additional tuning to determine how close is \"close enough\" for a fuzzy match. Other recall errors identified as No matching patternmay be unavoidable in an approach such as ours if the error arises from a synonym pair that only occurs within text pattern not used by any other synonym pair, that is, the text surrounding the synonym pair is unique. This can happen when the synonyms are close together in the text, or when many words separate the synonyms in the text. Because they are long, these patterns will most often be unique. For example, the pair (AHC, NR0B1) is only found in two places in the test collection, in both cases separated by many unique words: DAX1 encoded by NR0B1, when mutated, is responsible for X-linked adrenal hypoplasia congenita (AHC). \[[@B21]\] Mutations in DAX1 \[dosage-sensitive sex reversal-adrenal hypoplasia congenita (AHC) critical region on the X chromosome gene 1; NR0B1\] cause X-linked AHC, a disease characterized by primary adrenal failure in infancy or childhood and reproductive abnormalities later in life. \[[@B22]\] ### Precision error analysis Precision errors were categorized by first reviewing a small random sample of 20 errors. From this pilot set of errors, a set of mutually exclusive precision error categories was determined by inspection. The resulting set of six error categories was then applied to an additional random sample of 100 precision errors. The six categories of precision error and the proportions found were: (1) Not a gene name(28%). One or both symbols were not the name of a gene, allele, mutation, or gene family. For example, in the pair (GLN, HBD-2) GLN is an abbreviation for the amino acid glycine. (2) Partial gene name(9%). One or both symbols were part of an incompletely extracted gene name pair. For example in the pair (MAPK, P38), MAPRK1, and MAPK2 are synonyms of P38. (3) Biochemically related(48%). Two different genes that have been reported to interact within the context of a biochemical mechanism, or, names for two distinct genes from the same functionally related family. For example in the pair (IGFI, IGFII) the genes are both members of the family of insulin-like growth factors, and in the pair (BCR, ABL1) the fusion of the BCR and ABL1 genes has been found to be a \"recurrent aberration in B cell precursor leukemia cells\" \[[@B23]\]. (4) Unrelated genes(3%). Two complete gene names but we were unable to establish family or biochemical relationship by reviewing the test data set or MEDLINE. For example in the pair (ARC, CH3) the names are both genes, and were not found to co-occur in the literature. This error is most likely caused by the inference of synonyms from other synonym pairs. (5) Mutation variants(5%). Two mutation names for the same gene but nonspecific for that gene. These are allele or mutation names that are generic and/or only used within a single abstract. For example, the allele A1, or the pair (CYS106ALA, CYS7ALA). (6) Correct(7%). A correct gene synonym pair not included in the gold standard dataset, found later during error analysis by abstract review. By far, the most commonly occurring error was a pair of gene symbols being chemically or biologically related but distinct, non-synonymous entities. These errors accounted for 48% of the total. The next most common error, occurring 28% of the time, resulted when one or both of the extracted pair of symbols were not a gene or protein name or symbol. The remaining errors were much less common. ### Incorporation of error analysis into performance results About 7% of precision errors were later determined to be false negatives, that is, the synonym pair was determined by manual inspection of the literature to be correct but was not part of the gold standard data set. Incorporating this proportion of additional correct synonym pairs back into the performance measures previously shown results in an estimated precision of 23.18% and an estimated peak F-score of 22.21%. A comparison of this performance estimate with prior work is shown in Figure [6](#F6){ref-type="fig"}. Discussion ========== Our results demonstrate that this method compares well to other automated methods of synonym extraction and is a useful general approach to knowledge extraction. The method is highly efficient in its use of seed pairs. This may be an advantage in situations where large numbers of seed pairs are difficult or expensive to collect. During training, it was determined that using eight initial seed pairs was adequate. It was observed that the performance was largely stable for different initial numbers of seed pairs between 8 and 32. This suggests that an initial \"critical mass\" of seed pairs was necessary to get the process started. Beyond the critical number, the algorithm automatically found additional common seeds. Including additional common synonym pairs as seeds simply gave as input high confidence pairs that the algorithm could find on its own. Optimizing the network structure based on the quality metric of the overall network MCC/MNCC (see Methods section) ratio was an effective way to pick the best text patterns for gene synonym pair extraction. Using the symbolic network to support inference of synonym pairs improved both the recall as well as the absolute number of synonym pairs discovered, consistently finding approximately 10% more verified pairs. While there was some loss in precision for these additional pairs, the cost was modest until well past the peak F-score iteration. The inference capability added to its utility as a tool in knowledge discovery, and helped extract additional synonym pairs beyond those found strictly in the text. One way to improve system performance would be to reduce the very common Biochemically relatederrors by filtering the results to remove known associated gene pairs. There are several on-line databases of gene relationship networks \[[@B24],[@B25]\], and the information in these databases could be used as evidence of the genes being distinct and non-synonymous. While it is unlikely that this filtering could remove all of the false positives from this large source of error, the improvement is likely to be significant. The relative frequencies of the two types of recall errors present evidence suggesting a general observation about pattern-based text relationship mining systems. Two-thirds of the recall errors were due to the system not having discovered a pattern that matched the non-recalled pair, and only one-third of errors were due to the system having found a matching pattern, rejecting it based on the network metric criteria. The current system used a large number of very specific patterns based on the text surrounding high confidence gene symbol pairs. The Snowballsystem used more flexible patterns, allowing \"fuzzy\" matching based on the relative importance of word in a pattern. The two different systems performed similarly, which may be due to some inherent limitation of the pattern-based approach to uncovering gene synonym relationships. The textual context of interesting biological relationships may not be specific enough to significantly improve performance. Certainly, more work is needed in this area before drawing strong conclusions. Since there is no standard test collection for gene symbol synonym extraction research and no absolute gold standard for recall, the recall standard used was an approximation. The method of constructing a recall standard used in this work facilitated comparison with prior work in the field. However, it was by nature a biased sampling method, and does not completely characterize the recall capabilities of current knowledge extraction systems as compared to manual expert review. The full text test collection previously used by Yu and Agichtein was not publicly available. Major limitations of our study include the lack of a widely available full text test collection of adequate size and the inability to use the same test collection as previous investigators. MEDLINE abstracts were used because they are plentiful and readily available. While prior investigators have stated that full text articles are better sources data for the extraction of gene name synonyms \[[@B18]\], it was encouraging to find that applying our method only to the article abstracts produced comparable results. The performance of the current system is limited somewhat by the simple orthographic approach used for named-entity recognition (NER). Gene names and symbols were required to be a single string delimited by spaces and other punctuation characters. Not all gene names fit this description, although the gene name pairs extracted for the recall gold standard from SWISSPROT met this requirement. Precision error analysis showed that approximately 28% of precision errors were due to a non-gene or protein symbol being treated as a gene or protein. Another 9% of precision errors were due to an incomplete portion of a gene symbol being identified as a gene symbol. These two categories together represent failure of named entity recognition (NER) and account for 37% of precision errors. Current state-of-the-art F-score performance of biological named-entity recognition is approximately 80% \[[@B26]\]. Using this number as the measure of performance, it can be estimated that the maximum improvement that could be obtained by incorporating a state-of-the-art gene and protein named-entity recognizer into the system would decrease these errors 20%, and increase the precision at the peak F-score to 27%. The actual improvement is likely to be less than the maximum if the NER system makes use of the same contextual information used by the synonym extraction system. There are many other potential applications of our general approach to mining the biomedical literature. Many inter-entity relationships, such as enhance/inhibit relations between drugs, biological substances, and diseases, and the promoter/suppressor relationships between genes could be modeled as graph structures and appropriate metrics created to measure the relevant network properties. Multiple separate networks can be created simultaneously and then used together during the logical inference step to extend the approach to multiple types of entities and multiple types of relationships between those entities. Further work is necessary to determine whether extracting enhance/inhibit and other functional relations from biomedical text is amenable to our approach. Automatic extraction of complex functional relationships is likely to be more complex than the extraction of synonyms. Perhaps the most exciting application for the network-based approach is in mining the biomedical literature for hypothesis generation, such as that done manually by Swanson \[[@B27]\], and automatically by others \[[@B28],[@B29]\]. While the Swanson approach is limited to relations between three entities, the network approach can support practically limitless intermediate inferences, limited largely by the confidence in the individual relationships. Future refinements will have to go beyond the simple method used in the current work to determine which relationships are strong enough to support inference. The chain of inference can be modeled as a confidence path with each link reducing the confidence in the entire path by a fraction based on the uncertainty of the relationship. Having the ability to infer useful hypotheses across several intermediate relationships has the exciting potential to accelerate the rate of medical progress and focus efforts on the most promising prospects. With the biomedical knowledge and the corresponding bibliomegrowing at an exponential rate, the raw material exists for computer assisted hypothesis generation. Further work on text mining and knowledge extraction will be necessary in order to better understand the problems to which it can be most usefully applied, as well as the means to evaluate these systems in order for text mining and knowledge extraction to realize its full potential. Conclusion ========== These results support the conclusion that our method is useful in extracting gene and protein name synonym relationships from biomedical literature abstracts. The current system could be improved by incorporating state-of-the-art NER, and by including additional domain knowledge from richer data sources such as full text articles, and gene network databases which could provide data for negative examples. Use of negative examples could be incorporated into our approach by adding a penalty for extracting negative examples to the genetic optimizer evaluation function. While performance is not as good as the best combined approach of other investigators, it is as good as the best of the individual methods. With more accurate NER, as well as post-filtering using knowledge contained in online databases, the system may perform even better. Data sets and gold-standard files used in this work are available for download at \[[@B30]\]. Methods ======= In this section we present our gene and protein synonym extraction algorithm, and our evaluation methods. Algorithm --------- We approached the problem of gene and protein name synonym extraction as a problem in mathematical network analysis. In the network, nodes are gene and protein names and symbols, and edges are labeled with the number of times the connected names have occurred in a text source together (i.e., the co-occurrence count). An initial set of synonym pair \"seeds\" is used to search through the text corpus for text patterns in which those synonym pairs occur. Occurrences of gene and protein names are replaced with a regular expression that matches a wide variety of possible gene and protein names. This regular expression is designed to have high recall for single word gene and protein names and symbols, at the expense of low precision. Then these patterns are matched against the corpus, extracting text patterns that include co-occurrences between pairs of names that are potential synonyms. The name co-occurrences extracted by the patterns are used to construct a gene name synonym network, and this network is mathematically analyzed to determine the combination of patterns that produces the strongest set of synonyms. The new synonyms with the highest confidence are then used as seeds in the next iteration of the algorithm. This process can be repeated for a set number of iterations, or until no new high confidence synonym pairs are found. The regular expression used to identify gene and protein names is very non-specific: (\[\^\\s,/%\<\>;+&()=\\\[\\\]\\?\\\$\\\'\\\"\]{3,14}). The pattern excludes some punctuation and other special characters, but allows letters, numbers, as well as the period and colon characters. Gene and protein names are required to be between 3 and 14 characters long. The system then applies a set of heuristic rules to further screen out non-gene names. The name must not be in a stop list of words and patterns found during system development to be confused with gene and protein names (e.g., \"RNA\", \"DNA\", \".com\"). The name may not begin with a digit, dash, colon, period, or asterisk, and may not end with a dash, period, or colon. Furthermore, the name may not contain only lowercase characters. All uppercase, a mix of upper and lowercase characters, or a combination of letters and numbers is required. These rules favor recall over precision. The synonym text patterns are extracted from the text surrounding a pair identified synonyms. The system requires the synonym pairs to be within 4 words of each other, and includes zero or one words to either side of the synonym pair. For example, if (CIP1, WAF1)is an initial seed pair, and the text corpus includes the sentence: Two percent or greater nuclear staining with WAF1/CIP1 monoclonal antibody was determined by hazard ratio analysis to constitute positive p21 expression. \[[@B31]\] Then the system will extract the following patterns, where \$GENE\$stands for the gene and protein name matching regular expression: • \$GENE\$/\$GENE\$ • with \$GENE\$/\$GENE\$ • \$GENE\$/\$GENE\$ monoclonal • with \$GENE\$/\$GENE\$ monoclonal These patterns can then be applied to the text corpus to find name co-occurrences. For example, using the pattern with \$GENE\$/\$GENE\$, the system will extract the co-occurrence pairs (CARD15, NOD2)and (MMAC, PTEN)from the following sentence fragments found in the corpus respectively: Of the children with NOD2/CARD15 variants\...\[[@B32]\] Human glioma xenografts treated with PTEN/MMAC gene transfer exhibited\...\[[@B33]\] Given a set of patterns and the set of co-occurrences found by each pattern, the algorithm selects the best combination of patterns by evaluating the structure of the network created by the co-occurrences. The metric used to compare network structures is based on clustering coefficient measures \[[@B34]\]. A pattern is required to occur in the text a minimum of four times. The assumption is made that good synonym co-occurrence networks will have many separate, internally tightly linked clusters, since synonyms of synonyms should also have co-occurrences in the network. Figure [7](#F7){ref-type="fig"} pictorially shows high versus low clustering co-occurrence networks. The quality of a co-occurrence network is taken to be the ratio of the mean clustering coefficient (MCC) over the mean non-clustering coefficient (MNCC), and is computed as: ![](1471-2105-6-103-i1.gif) ![](1471-2105-6-103-i2.gif) quality= MCC/ MNCC (3) where Cis the number of nodes in the network, n(c)is the list of neighbors for node c, w(a,b)is the number of co-occurrences seen between aand b, and cmb(m, n)is the standard combination function giving the number of combinations of mitems taken nat a time. The minimum quality score is zero, the maximum is open ended and depends upon the number of nodes in the network and how interconnected they are. It is possible that a simpler measure could also work, however using MCC alone was considered but rejected because it favors larger lightly connected networks over smaller highly connected networks. The MNCC takes into account the size of the network and the number of nodes not connected to a given node. Note that MCC is only defined for nodes with two or more neighbors. A simpler measure of summing the weights for all the shared neighbors was considered, but not implemented. Analytically, it appears to give too much weight to a single very common synonym pair that is falsely connected to the node being measured. Computing the MCC/MNCC of the node pair averages the inter-connectivity across all the nodes connected to a pair of nodes, and therefore should be more accurate for groups of pairs synonymous to each other. Finding the set of patterns which produce the network with the highest quality measure is a combinatorial optimization problem; the co-occurrences found by each pattern can either be included in the network or not. One of the best methods of solving this type of problem uses a genetic algorithm to optimize the combination of patterns chosen. We have chosen a variation of the canonical genetic algorithm that uses rank-order-based selection pressure \[[@B35],[@B36]\]. It is used simply as a combinatorial optimizer. This variation was chosen because it works well and is easy to implement. Other genetic algorithm variants likely would perform just as well. Once the set of patterns and their associated co-occurrences are chosen, the algorithm extracts synonym pairs from the co-occurrence network. This is done using a graph traversal algorithm much like Dijkstra\'s shortest path algorithm \[[@B37]\], and extracts synonym pairs explicitly found in the text as well as those that can be inferred by following the synonym relationships represented by the edges in the network. For example, if A is a synonym of B, and B is a synonym of C, then A is likely a synonym of C. During system training it was found to be best to restrict inference to network edges that had co-occurrence counts of 2 or greater. In order to proceed with another iteration of the algorithm, the best synonym pairs must be chosen to use as seeds in the next iteration. Confidence in individual synonym pairs is determined using two network-based metrics. First, the overall confidence in a synonym pair with a given co-occurrence count nis estimated by computing the probability of seeing less than that occurrence count in a random graph with the same number of nodes and edges. This is computed as: ![](1471-2105-6-103-i3.gif) where Mis the total number of co-occurrences, Nis the number of nodes in the network, and μ= N/M. During training it was found that a confidence threshold of 0.999 gave the best results. Synonym pairs with confidence greater then the threshold are then ordered by another network-based metric that measures the local clustering for the pair of nodes representing the synonym pair. The individual node clustering (CC) non-clustering (NCC) coefficients are computed, resulting in a local clustering metric (LCM) for each synonym pair: ![](1471-2105-6-103-i4.gif) ![](1471-2105-6-103-i5.gif) ![](1471-2105-6-103-i6.gif) Patterns are then extracted from the text using the high confidence synonym pairs as seeds, choosing the highest local clusterings first. The number of patterns to evaluate at each iteration was limited to 150, which was found to balance the quality of the results with the need to make the combinatorial optimization step solvable in a reasonable amount of time. Figure [8](#F8){ref-type="fig"} illustrates the overall algorithm. The iterative pattern matching part of the algorithm is, like Snowball, based on the DIPRE approach of Brin. The novel parts of the algorithm presented here include the use of network-based metrics for evaluating the quality of patterns and synonym pairs, the use of a genetic optimization algorithm to determine the optimal set of patterns to use in extracting gene name synonyms, and the use of graph-based inference to infer synonym pairs not found explicitly in the text corpus. Experimental design ------------------- The experiment was performed in three steps. The first step was to develop and refine the algorithm detailed in the previous section on the training and validation data sets. Next, the algorithm was run on the MEDLINE records in the test set. Lastly, the quality of the extracted synonym list was evaluated by validating the synonymy of each extracted pair against a gold standard and then computing performance metrics. ### Data sets The training, validation, and test data sets used in this experiment consist of sentences extracted from approximately 50,000 abstracts from a year\'s worth of MEDLINE records containing the word \"gene\" for each set. Abstracts from 2001, 2002, and 2003 served as training, validation, and test sets, respectively. After downloading the MEDLINE records from PubMed, the records were parsed to extract the abstract field. The abstracts were then separated into sentences using a simple, lexically based sentence boundary detection algorithm. Finally, the sentences were screened to remove non-contributing sentences. These were sentences that did not contain at least two words that matched the gene and protein name regular expression discussed previously. This resulted in the three data sets used in this experiment each consisting of about 145,000 sentences each. The training set was used for system development, debugging, and parameter tuning, as well as for choosing the initial set of seed synonym pairs. The validation set was used to verify the system and ensure that the chosen parameters worked as expected on multiple data sets. The test set was used to produce the experimental results. ### Gold standards Calculation of performance metrics for the experiment required gold standards for both precision and recall. Gene name synonyms available in on-line genomics databases served as the basis for both gold standards. The approach used to create the precision gold standard consisted of downloading several genomics databases available on-line, extracting out the name, alias, and synonym fields, and combining them into a single gold standard for use in the computation of precision. The database snapshots that we used to construct our gold standard consisted of: SWISSPROT downloaded on 12/10/2003, FlyBase, Genew, and LocusLink, downloaded on 1/12/2004, and the MGI, and SGD databases downloaded on 1/22/2004. The online databases do not contain all synonyms in common use. Orthographic variations (e.g., \"WAF1\" and \"WAF-1\") are often missing. Therefore during training and validation extracted candidate synonym pairs that were not found in the precision gold standard were manually reviewed for pairs that were likely to be correct (e.g. \"CONNEXIN32\" and \"CX-32\"), and these pairs were checked by reviewing MEDLINE for supporting information in the titles and abstracts. Manually verified pairs were added to the precision gold standard for use in scoring the results from the test data. Creation of the recall gold standard was more challenging. Typically an accurate gold standard requires multiple experts to agree on definitions and then manually review the literature for the information in question, comparing multiple expert opinions and computing inter- and intra-rater agreement. Considering the large amount of text used, the expert resources were not available to use this method. Instead a simpler method was employed, based on the approach used by Yu and Agichtein. To approximate a recall gold standard, all of the synonym pairs extracted from the SWISSPROT database \[[@B38]\] were compared to all of the sentences in the test collection. If both symbols of a synonym pair given in SWISSPROT were present together in at least one sentence in the test collection, that synonym pair was included in the recall gold standard. This resulted in a recall gold standard set of 483 synonym pairs for the test collection. While this may bias the recall gold standard towards the gene and protein names present in SWISSPROT, the bias is independent of any feature of the algorithm. Additionally, using a recall gold standard construction method like that of Yu and Agichtein facilitates later comparison of results. Note that even though a pair of gene synonymous names from SWISSPROT may be present in a single sentence of the test set, it may be impossible for this or any other pattern-based algorithm to extract the pair. The synonyms could be separated by too many words, or the synonym pair may not occur in a repeated pattern. Nevertheless, a recall gold standard constructed by this method provides a useful benchmark. Abbreviations ============= clustering coefficient (CC) local clustering metric (LCM) mean clustering coefficient (MCC) mean non-clustering coefficient (MNCC) non-clustering coefficient (NCC) named entity recognition (NER) Authors\' contributions ======================= AC wrote the software, ran the experiments, performed the data analysis, and drafted the manuscript. WH helped to design the study, participated in its coordination, and assisted in drafting the manuscript. CS provided support on gene nomenclature and databases. KS participated in the design of the algorithms and the evaluation methodology. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by NIH Grant number 2 T15 LM07088-11 from the National Library of Medicine. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Precision versus recall over all iterations. ::: ![](1471-2105-6-103-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### F-score versus iterations. ::: ![](1471-2105-6-103-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Number verified pairs versus iterations. ::: ![](1471-2105-6-103-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Verified F-score comparison with work of Yu and Agichtein. ::: ![](1471-2105-6-103-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Number seeds, verified pairs extracted, and extraction efficiency. ::: ![](1471-2105-6-103-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Estimated F-score comparison with work of Yu and Agichtein. ::: ![](1471-2105-6-103-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Representation of high and low clustering co-occurrence networks. ::: ![](1471-2105-6-103-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Synonym extraction algorithm. ::: ![](1471-2105-6-103-8) :::	PubMed Central	2024-06-05T03:55:56.435050	2005-4-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090552/", "journal": "BMC Bioinformatics. 2005 Apr 22; 6:103", "authors": [ { "first": "AM", "last": "Cohen" }, { "first": "WR", "last": "Hersh" }, { "first": "C", "last": "Dubay" }, { "first": "K", "last": "Spackman" } ] }
PMC1090553	Background ========== As our acquaintance with RNA\'s diverse functional repertoire develops \[[@B1]-[@B5]\], so does demand for faster and more accurate tools for RNA sequence analysis. In particular, comparative genomics approaches hold great promise for RNA, due to the well-behaved basepairing correlations in an RNA gene family with conserved secondary structure (at least, well-behaved compared to protein structures). Whereas the structural signal encoded in a single RNA gene is rather weak and may be barely (if at all) distinguishable from the secondary structure of a random sequence \[[@B6]\], the covariation signal increases with every additional sequence considered. Many programs for comparative analysis of RNA require the sequences to be prealigned \[[@B7]-[@B9]\]. This can be a source of error, since misaligned bases can add noise that swamps the covariation signal. The most recent of these methods allows for some uncertainty in the alignment \[[@B7]\]. More generally, one can view the alignment and structure prediction as a combined problem, to be solved simultaneously. This is the approach taken in this paper, and by earlier programs such as FOLDALIGN \[[@B10]\], DYNALIGN \[[@B11]\], CARNAC \[[@B12]\], QRNA \[[@B9]\] and our dart library, introduced in a previous paper \[[@B13]\] and extended here. In this framework, fixing of the alignment can be viewed as a partial constraint on the simultaneous alignment/folding problem. A powerful, general dynamic programming algorithm for simultaneously aligning and predicting the structure of multiple RNA sequences was developed by David Sankoff \[[@B14]\]. The energy-based folding of Zuker et al\[[@B15]\] and recent approaches based on Stochastic Context-Free Grammars (SCFGs) \[[@B9],[@B13],[@B16]-[@B20]\] are both closely related to Sankoff\'s algorithm. The method takes time O(L^3N^) and memory O(L^2N^) for Nsequences of length L. This is prohibitively expensive at the time of writing, except for fairly short sequences, which has motivated the development of various constrained versions of these algorithms \[[@B9]-[@B11],[@B13],[@B21]\]. The purpose of this paper is to report our progress on general pairwise constrained versions of Sankoff\'s algorithm (or, more precisely, constrained versions of some related dynamic programming algorithms for SCFGs). The overall aim is the simultaneous alignment and structure prediction of two RNA sequences, Xand Y, subject to an SCFG-based scoring scheme and user-supplied constraints. Additionally, we wish to be able to parameterize the model automatically from training data. Without constraints, the above tasks are addressed by the resource-intensive CYK and Inside-Outside algorithms; here, we present constrained versions of these algorithms that work in reduced space and time (the exact complexity depends nontrivially on the constraints). Our system of constraints is quite general. Previous constrained versions of Sankoff-like algorithms, such as the programs DYNALIGN \[[@B11]\] and FOLDALIGN \[[@B10]\], have been restricted to \"banding\" the algorithm e.g. by constraining the maximum insertion/deletion distance between the two sequences or the maximum separation between paired bases. Alternately, constraints on the accessible structures \[[@B13]\] or alignments \[[@B9]\] have been described. The algorithms described here can reproduce nearly all such banding constraints and, further, can take advantage of more flexible sequence-tailored constraints. Specifically, the fold envelopesdetermine the subsequencesof Xand Ythat can be considered by the algorithm, while the alignment envelopedetermines the permissible cutpointsin the pairwise alignment of Xand Y. The fold envelopes can be used to prune the search over secondary structures (e.g. by including/excluding specific hydrogen-bonded base-pairings), while the alignment envelopes can be used to prune the search over alignments (e.g. by including/excluding specific residue-level homologies). The fold envelopes can be precalculated for each sequence individually (e.g. by an energy-based folding or a single-sequence SCFG), and the alignment envelope by comparing the two sequences without regard for secondary structure (e.g. using a pairwise Hidden Markov Model); both types of pre-comparison are much more resource-friendly than the unconstrained Sankoff-like algorithms. The design of the constrained algorithms is discussed using concepts from object-oriented programming: the dynamic programming matrix can be viewed as a sparsely populated container, whereas the main loop that fills the matrix is a complex iterator\[[@B22]\]. The algorithms have been implemented in a freely available program for RNA sequence alignment, stemloc, which also includes algorithms to determine appropriate constraints in an automatic fashion. Results demonstrating the program\'s efficient resource usage are presented. The stemloc program also implements various familiar extensions to pairwise alignment, including local alignment \[[@B23]\], Waterman-Eggert N-best suboptimal alignments \[[@B24]\] and progressive multiple alignment \[[@B25]\]. Although the envelope framework, rather than these extensions, is the main focus of this paper, implementation of the extensions is straightforward within this framework, and is briefly described. Results ======= To investigate the comparative resource usage of the various different kinds of constraint that can be applied using fold and alignment envelopes, stemloc was tested on 22 pairwise alignments taken from version 6.1 of RFAM \[[@B37]\], spanning 7 different families of functional noncoding RNA. Each chosen test family had a consensus secondary structure published independently in the literature, and no two sequences in the test set had higher than 60% identity. The EMBL accession numbers and co-ordinates of all sequences are listed in Table [5](#T5){ref-type="table"}. The table shows the performance of stemloc using the 1000-best fold envelope and the 100-best alignment envelope. The various RFAM families are S15, the ribosomal S15 leader sequence; the U3 and U5 spliceosomal small nucleolar RNAs; IRE, the iron response element from UTRs of genes involved in vertebrate iron metabolism; glmS, the glucosamine-6-phosphate activated mRNA-cleaving ribozyme; Purine, the prokaryotic purine-binding riboswitch; and 6S, the E.colipolymerase-associated transcriptional repressor. The following three test regimes were used, each representing a different combination of fold and alignment envelopes: N-best alignments, all folds ------------------------------ The alignment envelope containing the Nbest primary sequence alignments, with the unconstrained fold envelopes (stemloc options: \'\--nalign N \--nfold -1\'). This is the red curve in Figures [8](#F8){ref-type="fig"}, [9](#F9){ref-type="fig"}, [10](#F10){ref-type="fig"}, [11](#F11){ref-type="fig"}, [12](#F12){ref-type="fig"}, [13](#F13){ref-type="fig"} N-best folds, all alignments ------------------------------ The unconstrained alignment envelope, with the fold envelopes containing the Nbest single-sequence structure predictions (stemloc options: \'\--nalign -1 \--nfold N\'). This is the green curve in Figures [8](#F8){ref-type="fig"}, [9](#F9){ref-type="fig"}, [10](#F10){ref-type="fig"}, [11](#F11){ref-type="fig"}, [12](#F12){ref-type="fig"}, [13](#F13){ref-type="fig"} N-best folds, 100-best alignments ----------------------------------- The alignment envelope containing the 100 best primary sequence alignments, with the fold envelopes containing the Nbest single-sequence structure predictions (stemloc options: \'\--nalign 100 \--nfold N\'). This is the blue curve in Figures [8](#F8){ref-type="fig"}, [9](#F9){ref-type="fig"}, [10](#F10){ref-type="fig"}, [11](#F11){ref-type="fig"}, [12](#F12){ref-type="fig"}, [13](#F13){ref-type="fig"} In the first two tests, Nwas varied from 1 to 100; in the latter test, Nwas varied from 1 to 10000. The lower ceiling for Nin the first two tests was imposed by resource limitations. Note that the endpoint of the red curve (\"N-best alignments, all folds\"), which occurs at N= 100, coincides with the asymptotic limit of the blue curve (\"N-best folds, 100-best alignments\") at high N. A range of different values for the parameter Nwas used to test the above three strategies. As Nwas increased over the range, the size of the corresponding fold or alignment envelopes was found to be strongly correlated (Figures [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). However, the actual size of the fold/alignment envelopes in each particular test case varies widely (see large error bars in Figures [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}), perhaps due to variable factors such as the sequence lengths, compositions and/or identities. Since it is easier to control the envelope construction parameter Nthan to control the envelope sizes directly, the following section will report performance indicators as a direct function of N, rather than as a function of the strongly-correlated but widely-varying envelope sizes. We report performance indicators for stemloc as follows. Let Aand Bbe alignments of a given pair of sequences, each represented as a set of aligned residue-pairs {(i, k)}. Suppose that Ais the alignment according to RFAM, and Bis the alignment predicted by stemloc. Then define the alignment sensitivityto be \\|A∩ B\\|/\\|A\\| and the alignment specificityto be \\|A∩ B\\|/\\|B\\|. Further, let Sand T be possible secondary structures for a given sequence, each represented as a set of base-pairs {(i, j)}. Suppose that Sis the published structure, and T is the structure predicted by stemloc. Then the basepair sensitivityis \\|S∩ T\\|/\\|S\\| and the basepair specificityis \\|S∩ T\\|/\\|T\\|. These performance indicators are averaged over all 22 pairwise alignments and plotted for the three test regimes in Figure [8](#F8){ref-type="fig"} (alignment sensitivity), Figure [9](#F9){ref-type="fig"} (alignment specificity), Figure [10](#F10){ref-type="fig"} (basepair sensitivity) and Figure [11](#F11){ref-type="fig"} (basepair specificity). As can be seen, the N-best alignment regime empirically seems to achieve an asymptotic maximum performance around N≃ 100 (possibly even around N≃ 10), while the N-best fold envelope underperforms compared to the unconstrained fold envelope up to around N≃ 1000. The tests were performed on a 2.3 GHz Apple PowerPC G5. The resource usages of the test regimes are plotted in Figure [12](#F12){ref-type="fig"} (user-mode running time) and Figure [13](#F13){ref-type="fig"} (memory usage). The resource usage of the constrained algorithms is substantially reduced when the envelopes are smaller (i.e. at lower N). This is especially notable when contrasting the resource usage of the \"N-best folds, all alignments\" test with the more constrained \"N-best folds, 100-best alignments\" test. Three main conclusions can be drawn from these data. First, allowing the search to consider more than a single alignment greatly improves structure prediction (the red curve). Second, constraining the alignment search while exhaustively scanning fold space (the red curve) outperforms constraining the fold search while exhaustively scanning alignment space (the green curve). Third, the hybrid strategy (the blue curve), which partially constrains both searches, approaches the alignment-constrained, fold-unconstrained strategy (the red curve) in performance, with a significant saving in CPU and memory resources. Memory is the limiting factor in pairwise RNA alignment, and the primary motivation for constraints. For example, without constraints, alignment of two 16S ribosomal subunits using the stemloc grammar would take approximately 500 terabytes. (Using fold envelope constraints with structures fully specified, it can be done in under 5 gigabytes.) Based on the results of these tests, the default envelope options for stemloc were chosen to be the 100-best alignment envelope and the 1000-best fold envelope. The performance of stemloc with these envelopes on each of the pairwise test alignments is given in Table [5](#T5){ref-type="table"}. Discussion ========== The algorithms presented here include constrained versions of Pair-SCFG dynamic programming algorithms that run in significantly reduced space and time. The primary advance over previous work is the simultaenous imposition of fold and alignment constraints, including alignment constraints that are more general than others previously described. Thes constraints lead to significant reductions in requirements for processor and memory usage, which will increase the length of RNA sequences that can be analyzed on mainstream computer hardware. These algorithms have been used to implement stemloc, a fast, efficient software tool for multiple RNA sequence alignment implementing numerous extra features such as local alignment, Waterman-Eggert N-best suboptimal alignment and progressive multiple alignment. The source code for the program is freely available from <http://www.biowiki.org>. The results given here should be regarded as preliminary. For example, we have only tested the pairwise alignment functionality; full evaluation/optimisation of the multiple alignment algorithm remains. Rather than using the CYK algorithm, one could use the Inside-Outside algorithm with a decision-theoretic dynamic programming step to maximize expected performance \[[@B38],[@B39]\]. As noted in the Parameterization section, it might also be possible to improve on the training procedure. We are also considering ways of elaborating the grammar to include basepair stacking terms. These and other improvements we hope to address in future work. Conclusion ========== RNA sequence analysis has generated considerable interest over recent years, as many new roles for RNA in the cell have come to light. RNA genes and regulatory elements are components of many molecular systems and comparative genomics is a powerful way to probe this function, perhaps even more so for RNA than for protein (due to the \"well-behaved\" statistical correlations found in RNAs with conserved secondary structure). Furthermore, statistical modeling of RNA evolution continues to play a fundamental role in the phylogenetic classification of new forms of life. These biological motives have driven a demand for RNA sequence analysis tools that are faster, slimmer and more scaleable. It is hoped that the algorithms and approaches described here, together with development and analysis of RNA evolutionary models \[[@B36]\], may expand the applications of RNA informatics. Methods ======= We begin our description of the envelope method with an explanatory note regarding our decision to present these constraints in terms of SCFGs, rather than other scoring schemes such as those based solely on energies \[[@B15]\] or on energy/information-theoretic hybrids \[[@B11]\]. The reason for our choice of SCFGs is simple: stochastic grammars are, in our opinion, the most theoretically well-developed of the scoring schemes used for RNA. They come with well-documented algorithms for sequence alignment, structure prediction, parameterization by supervised learning from various kinds of training data, and calculation of posterior probabilities \[[@B20]\]. Discussion of these algorithms is facilitated by a well-developed and widely-understood probabilistic vocabulary. Stochastic grammars are actively researched outside bioinformatics, principally in natural language processing \[[@B26]\]. Crucially, SCFGs are sufficiently general to express virtually all of the features offered by other scoring schemes \[[@B19]\]. We also acknowledge the appeal of free energy-based scoring schemes, which have the advantage that the parameters can be determined experimentally. Energy-based scores can also be used to find posterior probabilities of base-pairings using a partition function \[[@B27]\]. However, to extend energy-based methods to two or more sequences, one must incorporate substitution scores. These are information-theoretic in nature and so are measured in bits, rather than kilocalories-per-mole \[[@B28]\]. Reconciling these two units of score (in a principled way) is an open problem. However, in an SCFG framework, all scores are information-theoretic and so there is no conflict of units. Despite these arguments, many people continue to find calories preferable to bits as a unit of score. For such readers, we note that the system of constraints described here is entirely applicable to the general score-attributed grammar. This includes energy-based and heuristic scoring schemes as well as (for example) grammars whose rule \"probabilities\" actually represent log-odds ratios, or which are conditionally normalized with respect to one sequence. Notation -------- To implement SCFG dynamic programming algorithms efficiently for RNA, it is convenient to define a simplified (but universal) template for grammars, similar in principle to \"Chomsky normal form\" \[[@B29],[@B30]\]. Our \"RNA normal form\" preserves the RNA-optimized efficiency of the Pair SCFG form presented in an earlier paper \[[@B13]\] (based on a single-sequence form due to Durbin et al\[[@B20]\]) by introducing different types of production rule to minimize bifurcations and collect emissions. The form defined here is slightly different from the above forms, in that it classifies only production rules, and not nonterminals, into different types. Let ![](1471-2105-6-73-i1.gif) be the \"ungapped RNA alphabet\", i.e. the set of four possible nucleotides in RNA. Let ![](1471-2105-6-73-i2.gif) be the \"gapped RNA alphabet\", i.e. the ungapped RNA alphabet Ω plus the gap symbol ![](1471-2105-6-73-i3.gif). Finally, let Ψ = Ω\' × Ω\' be the \"gapped-pair RNA alphabet\", i.e. a Cartesian product of two gapped RNA alphabets. We write Ψ-symbols by vertically stacking pairs of Ω\'-symbols, like this ![](1471-2105-6-73-i4.gif) or this ![](1471-2105-6-73-i5.gif). A pairwise stochastic context-free grammar ![](1471-2105-6-73-i6.gif) in RNA normal form consists of a nonterminal symbol alphabet Φ, a terminal symbol alphabet that is the gapped-pair RNA alphabet Ψ, and (for each nonterminal L) a probability distribution over a set of transformation rules (or production rules), ![](1471-2105-6-73-i7.gif)(L→ R), where R= R~1~\... R~K~is a sequence of nonterminal or terminal symbols, taking one of several stereotypical forms (see below). The nonterminal Lis referred to as the left-hand side (LHS) of the production rule and the symbol sequence Ras the right hand side (RHS). Let S, U, V, W∈ Φ denote nonterminal symbols. The allowable forms for production rules include terminations, transitions, bifurcationsand emissions. These are defined as follows Terminations: rules of the form L→ ε. Transitions: rules of the form L→ V. The directed graph formed by transition rules on nonterminals must be acyclic, and the list of nonterminals Φ must be topologically reverse-sorted with respect to this graph; i.e. if ![](1471-2105-6-73-i7.gif)(U→ V) \> 0, then Vappears before Uin Φ. Bifurcations: rules of the form L→ VW. There must be no transition-termination path from Vor Wto ε, i.e. neither Vor Wcan have completely empty inside sequence-pairs (see next section for a formal definition of the \"inside sequence-pair\"). Emissions: rules of the form ![](1471-2105-6-73-i8.gif) where A\', B\', C\', D\'∈ Ω\' are gapped RNA symbols, at least one of which is a non-gap symbol. For convenience, we also define A, B, C, D∈ Ω\* to be the corresponding ungapped RNA sequences, as follows: Ais the empty string if and only if A\'is the gap character; otherwise, A= A\'. Similar definitions apply for B, Cand D. The particular RNA normal form described in this section is chosen for ease of presentation. The implementation in the dart library uses the slightly more restrictive form for Pair SCFGs defined in an earlier paper \[[@B13]\]. For presentational purposes, we will generally omit all-gap columns from the pairwise alignment and the grammar. For example, an emission rule having the form ![](1471-2105-6-73-i9.gif) would be written as ![](1471-2105-6-73-i10.gif) instead. All-gap columns are not very interesting to a sequence analyst, and only arise in our formalism because all emission rules have the same form. Table [1](#T1){ref-type="table"} is an example of an RNA normal form grammar with two nonterminals, Stem and Loop. The grammar generates simple alignments of stems and loops, using two nonterminals (Stem and Loop); the starting nonterminal is Stem. The rule probabilities are functions of five scalar probability parameters (stemExtend, stemGap, bifurcate, loopExtend and loopGap) and four arrays of probability parameters (baseIndel\[4\], baseSubstitution\[16\], basepairIndel\[16\] and basepairSubstitution \[256\]). Here we introduce the notation X \[N\] for an array of Nprobability parameters normalized so that ![](1471-2105-6-73-i11.gif). It is also convenient to introduce some notation for ungapped sequences at this stage. Let X, Y∈ Ω\* denote ungapped RNA sequences, including (possibly) the empty string ε. Let X~i~be the i\'th symbol of X(counting from 1, so e.g. X~3~is the third symbol), let X~ij~denote the subsequence from X~i+1~to X~j~inclusive, or the empty string if i= j(so e.g. X~0,3~contains the first three symbols of X, while X~3,3~= ε) and let \\|X\\| denote the length of X. The parse tree and the sequence likelihood ------------------------------------------ The grammar ![](1471-2105-6-73-i6.gif) is a probabilistic model for deriving sequences X, Yfrom a single nonterminal. This derivation proceeds as follows: start with an initial sequence containing one starting nonterminal, S, then repeatedly apply probabilistically-sampled transformations to the nonterminals in the sequence. Eventually the sequence will contain only terminals from Ψ. This process generates a parse tree, rooted at node S, in which internal nodes are labeled with nonterminals and leaf nodes with terminals, with children of each node ordered left-to-right (Figure [1](#F1){ref-type="fig"}). Sequence Xcan be obtained by reading off ungapped RNA symbols from the top row of the output, and sequence Yby reading off the bottom row. Note that the subtree rooted at any internal W-labeled node describes a sub-process that generates some pair of subsequences (X~ij~, Y~kl~) starting from nonterminal W. We will refer to this subsequence-pair (X~ij~, Y~kl~) as the inside sequence-pairof W. The parse tree likelihoodis the product of all the rule probabilities corresponding to the internal nodes. Summing the likelihoods of all parse trees rooted in state Sand generating sequences X, Y, one obtains P(X, Y\\| S,![](1471-2105-6-73-i6.gif)), the sequence likelihood. This sum can be performed efficiently by the Inside algorithm, as will be described below. Dynamic programming algorithms for Pair SCFGs --------------------------------------------- The following section describes the constrained and unconstrained dynamic programming (DP) algorithms used for Pair SCFGs. ### The Inside algorithm The Inside algorithm \[[@B26]\] computes P(X, Y\\| S,![](1471-2105-6-73-i6.gif)) by recursive decomposition via intermediate probabilities ![](1471-2105-6-73-i18.gif)(i, j, k, l) ≡ P(X~ij~, Y~kl~\\|U,![](1471-2105-6-73-i6.gif)). In RNA normal form, the time-limiting step in the Inside algorithm involves summing contributions to ![](1471-2105-6-73-i18.gif)(i, j, k, l) from bifurcation rules of the form U→ VW, such that the bifurcation splits the two sequences (X, Y) between bases (X~m~, Y~n~) and (X~m+1~, Y~n+1~) ![](1471-2105-6-73-i19.gif) An asymptotically faster step involves summing contributions from matching emission rules of the form ![](1471-2105-6-73-i20.gif) (transition rules are represented as the special case A\'= B\'= C\'= D\'= ![](1471-2105-6-73-i3.gif)) ![](1471-2105-6-73-i21.gif) (Recall that Ais the ungapped version of A\'. Thus \\|A\\| = 0 ⇔ A= ε⇔ A\'= ![](1471-2105-6-73-i3.gif) and similarly for B, Cand D.) The intermediate probabilities of the Inside algorithm can then be expressed as ![](1471-2105-6-73-i18.gif)(i, j, k, l) = Q~1~+ Q~2~ Termination of the recursion is provided by matching end rules, U→ ε, if and only if the subsequences (X~ij~, Y~kl~) are empty ![](1471-2105-6-73-i18.gif)(i, i, k, k) = ![](1471-2105-6-73-i7.gif)(U→ ε) The sequence likelihood is obtained as P(X, Y\\| S,![](1471-2105-6-73-i6.gif)) ≡ ![](1471-2105-6-73-i22.gif)(0,\\|X\\|,0,\\|Y\\|) In pseudocode, the Inside algorithm is • Inputs: X, Y, S, ![](1471-2105-6-73-i6.gif) • For i= \\|X\\| to 0 (descending) • For j= ito \\|X\\| • For k= \\|Y\\| to 0 (descending) • For l= kto \\|Y\\| • For each nonterminal U∈ ε • Set Q~1~← 0; calculate Q~2~ • For m= ito j • For n= kto l • Calculate Q~B~(m, n) and add to Q~1~ • Calculate ![](1471-2105-6-73-i18.gif)(i, j, k, l) and store • Return ![](1471-2105-6-73-i22.gif)(0, \\|X\\|, 0, \\|Y\\|) The time-limiting step of the Inside algorithm (computing the Q~B~) involves six indices (i, j, k, l, m, n) and the time complexity of the full recursion is O(\\|X\\|^3^\\|Y\\|^3^). However, the stored intermediate probabilities involve only four indices (i, j, k, l) and so the memory complexity is O(\\|X\\|^2^\\|Y\\|^2^). In RNA normal form, the emission rules (Q~2~) account for homologous base-pairings between residues (X~i~, X~j~) and (Y~k~, Y~l~), or unpaired residues at X~i~, X~j~, Y~k~or Y~l~. This may also imply that X~i~and Y~k~are aligned, or that X~j~and Y~l~are (Figure [2](#F2){ref-type="fig"}). The bifurcation rules (Q~B~(m, n)) account for conserved multiloop structures in the RNA, i.e. one homology between substructures X~im~and Y~kn~and another between substructures X~mj~and Y~nl~(Figure [3](#F3){ref-type="fig"}). ### Imposing constraints The high time and memory cost of the Inside and related algorithms motivate the development of slimmer, faster versions. To begin with, we impose constraints that narrow the search space. For example, we might want to pre-parse the sequences individually (using a single-sequence SCFG, or other O(\\|X\\|^3^) RNA-folding method) and identify likely base-pairs (X~i~, X~j~) or (Y~k~, Y~l~), and/or likely unpaired nucleotides X~i~or Y~k~. Even simpler, we could simply throw out basepairs (X~i~, X~j~) between distant residues (i.e. where j- iexceeds some cutoff) Alternatively, we might want to pre-align the sequences (using a pairwise hidden Markov model, or other O(\\|X\\|\\|Y\\|) alignment method) and identify likely alignment columns (X~i~, Y~k~) and/or likely indels X~i~or Y~k~. Again, more simply, we could simply exclude columns (X~i~, Y~k~) for which \\|k- i\\| is too large. We can combine these various strategies into a generalized constraint on base-pairs, alignment-columns or both. We stipulate that ![](1471-2105-6-73-i18.gif)(i, j, k, l) = P(X~ij~, Y~kl~\\|U, ![](1471-2105-6-73-i6.gif)) = 0 unless the following conditions are satisfied ![](1471-2105-6-73-i23.gif) Here ![](1471-2105-6-73-i24.gif) and ![](1471-2105-6-73-i25.gif) are sets of permissible co-ordinates for structurally discrete subsequences in Xand Y. Fold-related features (basepairs and unpaired residues) can be included or excluded by this set, and so we refer to it as a fold envelope\[[@B13]\]. The set ![](1471-2105-6-73-i26.gif) is a set of possible cut-points in the alignment of Xand Y, and is referred to as an alignment envelope\[[@B31]\]. Both types of envelope are illustrated in Figure [2](#F2){ref-type="fig"}. The fold and alignment envelopes satisfy the following set relations ![](1471-2105-6-73-i27.gif) If equality holds in all three cases, then we recover the unconstrained Inside algorithm. Note that the co-ordinates (i, j, k, l) for the cutpoints and subsequences lie betweenresidues of Xand Y. There are (\\|X\\| + 1)(\\|Y\\| + 1) cutpoints in the maximal alignment envelope ![](1471-2105-6-73-i26.gif) and ![](1471-2105-6-73-i28.gif)(\\|X\\| + 1)(\\|X\\| + 2) subsequences in the maximal fold envelope ![](1471-2105-6-73-i24.gif). As an alternative to the unconstrained Inside algorithm, we can partially initialize the envelopes to limit the maximum subsequence length and/or the maximum deviation of the alignment from the main diagonal (Figure [4](#F4){ref-type="fig"}). More flexibly, we can limit the recursion to a single alignment, a single structure, or a broadly-specified set of alignments or structures (Figure [5](#F5){ref-type="fig"}). Applications such as alignment of two known structures \[[@B13],[@B32]\], alignment of an unstructured sequence to a known structure \[[@B33]\] or structure prediction from a known alignment \[[@B9]\] all reduce to simple application of the appropriate constraints. ### Further possible constraints The constraints given here allow the independent imposition of alignment or fold constraints. One can imagine further, even more general constraints. For example, one could exclude subsequence-pairs (X~ij~, Y~kl~) of radically different lengths, i.e. for which \\|(j- i) - (k- l)\\| exceeds some cutoff. This constraint is employed by the FOLDALIGN program. It is not expressible as a combination of independent alignment and fold constraints, and has not been implemented for the present work, though it would be relatively straightforward to combine it with the other constraints described here \[[@B10]\]. ### Accelerating the iteration Simply setting some intermediate probabilities to zero is not sufficient to accelerate the Inside algorithm. We also need to redesign the iteration to avoid visiting zero-probability subsequence-pairs (X~ij~, Y~kl~). This is achieved by pre-indexing the fold envelopes ![](1471-2105-6-73-i24.gif), ![](1471-2105-6-73-i25.gif) and the alignment envelope ![](1471-2105-6-73-i26.gif) so that we can quickly locate valid co-ordinates (i, j, k, l). The following is pseudocode for the algorithm with the redesigned iterator • Inputs: X, Y, S, ![](1471-2105-6-73-i6.gif) • For i= \\|X\\| to 0 (descending) • For each jsatisfying (i, j) ∈ ![](1471-2105-6-73-i24.gif) (ascending) (†) • For each ksatisfying (i, k) ∈ ![](1471-2105-6-73-i26.gif) (descending) (†) • For each lsatisfying (k, l) ∈ ![](1471-2105-6-73-i25.gif) (ascending) (†) • If (j, l) ∈ ![](1471-2105-6-73-i26.gif) then • For each nonterminal U∈ Φ • Set Q~1~← 0; calculate Q~2~ • For each msatisfying {(i, m), (m, j)} ⊂ ![](1471-2105-6-73-i24.gif) (†) • For each nsatisfying {(k, n), (n, l)} ⊂ ![](1471-2105-6-73-i25.gif) (†) • If (m, n) ∈ ![](1471-2105-6-73-i26.gif) then • Calculate Q~B~(m, n) and add to Q~1~ • Calculate ![](1471-2105-6-73-i18.gif)(i, j, k, l) and store • Return ![](1471-2105-6-73-i22.gif)(0, \\|X\\|, 0, \\|Y\\|) (†) These ordered subsets of ![](1471-2105-6-73-i24.gif), ![](1471-2105-6-73-i25.gif) and ![](1471-2105-6-73-i26.gif) can be precomputed for speed. Alternative designs for the algorithm are possible, and indeed different circumstances may affect the choice of optimal design (e.g. depending on which envelopes are most constrained). ### Slimming the container Memory is the most prohibitively expensive resource demand of the Inside algorithm. In its simplest form, the algorithm stores the intermediate probabilities ![](1471-2105-6-73-i18.gif)(i, j, k, l) using a five-dimensional array indexed by U, i, j, kand l. To get the most benefit out of imposing constraints, it is necessary to replace this multidimensional array with an efficiently-indexed reduced-space container. This design decision involves a close trade-off between CPU and memory usage. Initially, we tested various combinations of generic containers with O(N)-storage and O(Nlog N) access-times, such as balanced search trees \[[@B22]\]. These were found to be unbearably slow, and so we settled on a configuration of nested preindexed arrays. This configuration wastes some space, but has the important advantage of constant access time for given indices (U, i, j, k, l). For fold envelope ![](1471-2105-6-73-i24.gif), we precompute and sort the list ![](1471-2105-6-73-i29.gif) = {(i, j)} ⊆ ![](1471-2105-6-73-i24.gif) of subsequences starting at each position i. For each subsequence (i, j) ∈ ![](1471-2105-6-73-i24.gif), let ![](1471-2105-6-73-i30.gif) be its rank in ![](1471-2105-6-73-i29.gif). These ![](1471-2105-6-73-i30.gif) are also precomputed and stored. Similar precomputed sets ![](1471-2105-6-73-i31.gif) and ranks ![](1471-2105-6-73-i32.gif) are stored for ![](1471-2105-6-73-i25.gif). Our DP matrix then uses an inner two-dimensional array nested inside an outer two-dimensional array. The outer array has dimensions (\\|X\\|, \\|Y\\|) and is indexed by (i, k). The inner array has dimensions (\\|![](1471-2105-6-73-i29.gif)\\|, \\|![](1471-2105-6-73-i31.gif)\\|) and is indexed by (![](1471-2105-6-73-i30.gif), ![](1471-2105-6-73-i32.gif)). Cells of this inner array are further sub-indexed by nonterminal Uusing a standard fixed-length array, yielding ![](1471-2105-6-73-i18.gif)(i, j, k, l). This particular configuration is efficient when the alignment envelope is densely populated and the fold envelopes are sparsely populated. As with the redesigned iterator, there may be alternative designs that are resource-optimal under various different circumstances, depending on the nature of the envelopes. ### The CYK algorithm The Cocke-Younger-Kasami (CYK) dynamic programming algorithm \[[@B20]\] is related to the Inside algorithm, but replaces \"p+ q\" with \"max(p, q)\" in all formulae (that is, instead of summing probabilities, the maximum probability is always taken). The entries of the DP matrix, ![](1471-2105-6-73-i33.gif)(i, j, k, l), represent the maximum likelihood of any parse tree for (X~ij~, Y~kl~). A recursive/stack-based traceback algorithm can be used to recover this parse tree, beginning from subsequence (i, j, k, l) = (0, \\|X\\|, 0, \\|Y\\|) (for global alignment) or (i, j, k, l) = argmax ![](1471-2105-6-73-i22.gif)(i, j, k, l) (for local alignment) \[[@B13]\]. ### The Outside and KYC algorithms The Outside and KYC algorithms widen the applications of probabilistic inference with SCFGs. The Outside algorithm, together with the Inside, can be used to recover posterior probabilities of given basepairs/columns, which can be used as alignment reliability indicators or as update counts in Expectation Maximization parameter training \[[@B20],[@B26]\]. The KYC algorithm can be used with CYK to recover maximum-likelihood tracebacks from given co-ordinates. (We introduce the name \"KYC\" as a simple reversal of \"CYK\", reflecting the fact that KYC is to CYK as Outside is to Inside, i.e. the \"reverse\" version of the algorithm.) These algorithms use dynamic programming recursions that are related to Inside and CYK. The Outside algorithm calculates intermediate probabilities of the form ![](1471-2105-6-73-i34.gif)(i, j, k, l) = P(X~0,i~, Y~0,k~, V, Y~l,\\|Y\\|~, X~j,\\|X\\|~\\|S) representing the sum-over-probabilities of all partial parse trees rooted at Sand ending in Vwithout having yet generated sequences X~ij~and Y~kl~. Then, for example, the posterior probability that some node Vin the parse tree has inside sequence-pair (X~ij~, Y~kl~) is ![](1471-2105-6-73-i35.gif) As CYK is related to Inside, the KYC algorithm is related to the Outside algorithm: the intermediate probabilities ![](1471-2105-6-73-i36.gif)(i, j, k, l) are found using similar recursions (see below), but with all sums \"p+ q\" replaced by maxima \"max(p, q)\". The CYK and KYC DP matrices can be used to find the maximum likelihood of any parse tree containing a node Vwith inside sequence-pair (X~ij~, Y~kl~): this maximum likelihood is ![](1471-2105-6-73-i36.gif)(i, j, k, l) ![](1471-2105-6-73-i33.gif)(i, j, k, l). A recursive/stack-based traceback algorithm can be used to find the parse tree with this maximum likelihood. As with the Inside algorithm, we sum contributions to ![](1471-2105-6-73-i34.gif)(i, j, k, l) from various matching production rules. In contrast to the Inside algorithm, the nonterminal Vthat indexes ![](1471-2105-6-73-i34.gif) (\...) must now be matched on the right-hand-side, not the left-hand-side, of these production rules. We first consider bifurcations from a source nonterminal Uthat adjoin adjacent subsequences to left (U→ WV) or right (U→ VW), so that the inside sequence-pairs for Uare (respectively) (X~mj~, Y~nl~) or (X~im~,Y~kn~) ![](1471-2105-6-73-i37.gif) Next we consider emission rules, ![](1471-2105-6-73-i20.gif), again representing transitions as the special case A\'= B\'= C\'= D\'= ![](1471-2105-6-73-i3.gif) ![](1471-2105-6-73-i38.gif) The intermediate Outside probabilities are thus ![](1471-2105-6-73-i34.gif)(i, j, k, l) = ![](1471-2105-6-73-i39.gif) + ![](1471-2105-6-73-i40.gif) ![](1471-2105-6-73-i41.gif)(0, \\|X\\|, 0, \\|Y\\|) = 1 (Termination condition) Note that the Inside probabilities ![](1471-2105-6-73-i42.gif)(i, j, k, l) are needed to compute the Outside probabilities. We supply the Inside matrix, I, as an input to the Outside algorithm (they are usually calculated at the same time anyway). In terms of the underlying iteration, the key difference between the Inside and Outside algorithms is as follows. Suppose subsequence-pair INNER = (X~ij~, Y~kl~) is enclosed by subsequence-pair OUTER = (X~i\'j\'~, Y~k\'l\'~) (that is, 0 ≤ i\'≤ i≤ j≤ j\'≤ \\|X\\| and 0 ≤ k\'≤ k≤ l≤ l\'≤ \\|Y\\|, and INNER ≠ OUTER). Then the Inside iterator visits INNER before OUTER, whereas the Outside iterator visits OUTER before INNER. The order in which the Outside iterator visits nonterminals is topologically forward-sorted with respect to the grammar\'s transition-rule graph (i.e. the reverse of the order used by the Inside iterator). • Inputs: X, Y, S, ![](1471-2105-6-73-i6.gif), I • Initialize ![](1471-2105-6-73-i41.gif)(0, \\|X\\|, 0, \\|Y\\|) • For i= 0 to \\|X\\| (ascending) • For each jsatisfying (i, j) ∈ ![](1471-2105-6-73-i24.gif) (descending) (†) • For each ksatisfying (i, k) ∈ ![](1471-2105-6-73-i26.gif) (ascending) (†) • For each lsatisfying (k, l) ∈ ![](1471-2105-6-73-i25.gif) (descending) (†) • If (j, l) ∈ ![](1471-2105-6-73-i26.gif) then • For each nonterminal U∈ Φ (reverse order) • Set ![](1471-2105-6-73-i39.gif) ← 0; calculate ![](1471-2105-6-73-i40.gif) • For each msatisfying {(m, j), (m, i)} ⊂ ![](1471-2105-6-73-i24.gif) (†) • For each nsatisfying {(n, l), (n, k)} ⊂ ![](1471-2105-6-73-i25.gif) (†) • If (m, n) ∈ ![](1471-2105-6-73-i26.gif) then • Calculate ![](1471-2105-6-73-i43.gif)(m, n) and add to ![](1471-2105-6-73-i39.gif) • For each msatisfying {(i, m), (j, m)} ⊂ ![](1471-2105-6-73-i24.gif) (†) • For each nsatisfying {(k, n), (l, n)} ⊂ ![](1471-2105-6-73-i25.gif) (†) • If (m, n) ∈ ![](1471-2105-6-73-i26.gif) then • Calculate ![](1471-2105-6-73-i44.gif)(m, n) and add to ![](1471-2105-6-73-i40.gif) • Calculate ![](1471-2105-6-73-i45.gif)(i, j, k, l) and store (†) These ordered subsets of ![](1471-2105-6-73-i24.gif), ![](1471-2105-6-73-i25.gif) and ![](1471-2105-6-73-i26.gif) can be precomputed for speed. The reduced-space dynamic programming matrix that was developed above for the constrained Inside algorithm can be re-used for the constrained Outside algorithm. Implementation -------------- The above-described algorithms were implemented in the C++ dart library. One dart program in particular, stemloc, is an efficient general-purpose RNA multiple-sequence alignment program that can be flexibly controlled by the user from the Unix command line, including re-estimation of parameters from training data as well as a broad range of alignment functions. The dart libraries provide Inside, Outside, CYK, KYC, traceback and training algorithms for any pairwise SCFG in RNA normal form, whose rule probabilities can be expressed as algebraic expressions of some set of probability parameters (with associated normalization constraints). The operator-overloading features of C++ are utilized in full, so that the syntax of initializing a grammar object involves very few function calls and is essentially declarative. dart source code releases can be downloaded under the terms of the GNU Public License, from the following URL (which also gives access to the latest development code in the CVS repository) <http://www.biowiki.org/> The grammars and algorithms described in this paper specifically refer to release 0.2 of the dart package, dated October 2004 (although the algorithms are also implemented in release 0.1, dated 10/2003). ### Selecting appropriate fold and alignment envelopes This section offers a non-exhaustive list of possible strategies for choosing appropriate fold/alignmentenvelopes. (Italicized termsapply to fold envelopes, and bold termsto alignment envelopes.) • Choose some appropriately simplified grammar, such as a single-sequence SCFG/pair HMMthat models RNA folding/primary sequence alignment.Compute posterior probabilities of subsequences/cutpointsusing the Inside-Outside/Forward-Backwardalgorithm for some single-sequence SCFG/pairHMMthat models RNA folding/primary sequence alignment.Select all subsequences/cutpointswith posterior probability above some threshold (or select e.g. the top 10 percentile of the posterior probabability distribution). Ensure that each of these subsequences/cutpointsis on a valid traceback path, e.g. by running the CYK-KYC/Viterbi-forward-backwardalgorithm to find the maximum-likelihood traceback path from any given subsequence/cutpoint. Take the union of all subsequences/cutpointson these traceback paths to obtain the required envelope. • As above, choose some appropriate single-sequence SCFG/pair HMMthat models RNA folding/primary sequence alignment. Compute the maximum traceback path-likelihood from all subsequences/cutpointsusing the CYK-KYC/Viterbi-forward-backwardalgorithm for this grammar. Select all subsequences/cutpointswith maximum traceback likelihood above some threshold (or select e.g. the top 10 percentile of the max-traceback likelihood distribution) and do a traceback from each such cell. Take the union of all subsequences/cutpointson these traceback paths to obtain the required envelope. • As above, choose some appropriate single-sequence SCFG/pair HMMthat models RNA folding/primary sequence alignment. Sample some number of RNA structures/pairwise alignmentsusing the Inside/Forwardalgorithm with stochastic traceback. Take the union of all subsequences/cutpointson these traceback paths to obtain the required envelope. The latter two strategies have been implemented in the stemloc package described below. Empirically, the stochastic strategy appears to be less reliable than the deterministic strategies (although in theory the stochastic strategy will eventually find the globally optimal alignment given sufficiently many random repetitions, which may be a useful property). ### Multiple sequence alignment A heuristic algorithm for performing multiple alignment-and-folding of RNA sequences with a pairwise SCFG by progressive single-linkage clustering runs as follows • Start by making pairwise alignments (with predicted secondary structures) for all pairs of input sequences. • Mark the highest-scoring pair, and extract the two marked sequences with their predicted secondary structures. This highest-scoring pair is called the seed alignment. • While some sequences remain unmarked: • For each newly-marked sequence: • Align the marked sequence, with a fold envelope constrained by its predicted structure, to each unmarked sequence in turn. (The fold envelope can be tailored to allow e.g. extension of local alignments.) • Select the highest-scoring of the pairwise (marked-to-unmarked) alignments. Use this alignment to merge the unmarked sequence into the seed alignment, and mark this sequence as newly aligned. • Return the seed alignment. The above algorithms have been implemented in stemloc. The multiple alignments produced by this algorithm lack well-defined probabilistic scores unless the pair SCFG is conditionally normalized. It is also straightforward to retrieve the Nbest non-overlapping alignments by repeatedly applying an incremental Waterman-Eggert-style mask to the alignment envelope \[[@B24]\]. This is implemented in stemloc for the pairwise case. ### A grammar for pairwise RNA alignment and structure prediction After some empirical experimentation, we developed the grammar of Tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"} for the stemloc program. The grammar is split over three tables due to its considerable number of rules. Table [2](#T2){ref-type="table"} contains rules describing the connectivity of stems, loops and multiloops, and contains the only bifurcation rule. Table [3](#T3){ref-type="table"} describes the connectivity of bulges and Table [4](#T4){ref-type="table"} handles emissions (basepaired, unpaired, aligned or gapped). The starting nonterminal is Start. The nonterminals representing higher-level units of RNA structure are Loop, Stem, LBulge, RBulge and LRBulge. Each of these has associated Match, Ins and Del states (e.g. StemMatch, StemIns and StemDel) and each of these states has an associated emission state, prefixed with x, yor xysuperscripts (e.g. ^xy^StemMatch, ^y^StemIns and ^x^StemDel). The Multi state models multiloops, using a bifurcation to LMulti and RMulti. Tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"} also refer to probabilistic parameters used by the models. Free probability parameters (allowed to range from 0 to 1) include startInStem (determining the probability of ending the alignment with a basepair), loopExtend, stemExtend and multiExtend (determining the geometric distribution over loop and stem lengths, or the number of branches in a multiloop), multiBulgeOpen (determining the probability of bulges in multiloops) and stemGapOpen, stemGapExtend, stemGapSwap, loopGapOpen, loopGapExtend and loopGapSwap (determining the geometric distribution over gap lengths in stems and loops). There are also five parameter arrays: postStem\[[@B4]\] (determining whether a stem is followed by a loop, a bulge, a double-stranded bulge or a multiloop); baseIndel\[4\] (a probability distribution for a single unaligned, unpaired nucleotide); baseSubstitution\[16\] (a joint probability distribution for two aligned, unpaired nucleotides); basepairIndel\[16\] (a joint probability distribution for two unaligned, basepaired nucleotides); and basepairSubstitution \[256\] (a joint probability distribution for four aligned, basepaired nucleotides). All of the above parameters were automatically estimated from training data by the dart software. To summarize, the grammar models homologous stems, loops, multiloops and bulges in pairwise RNA alignments, with covariant substitution scores and affine gap penalties (geometric indel length distributions). It has the property that any combined alignment and structure prediction for two RNA sequences has a single, unambiguous parse tree. In our investigations, this unambiguity appeared to improve the accuracy of alignment and structure prediction substantially; see also writings on this topic by Giegerich \[[@B34]\] and Dowell, Eddy et al\[[@B35]\]. The grammar is also designed to minimize the number of bifurcation rules (the only bifurcation is Multi → LMulti RMulti). The stemloc grammar does not model basepair stacking effects due to π-orbital overlap, nor does it allow low-cost insertion or deletion of whole stems or substructures. Since the algorithms are implemented for any SCFG, it is straightforward to modify the program to experiment with grammars that model such phenomena. An example of a grammar that models the latter type of mutation (whole-substructure indels), and is also fully derived from an evolutionary rate-based model, is presented in a companion paper \[[@B36]\]. ### Parameterization Under the SCFG framework, the probability parameters for the grammar can be estimated directly from data using the Inside-Outside algorithm with appropriate constraints, which are easy to supply (e.g. to sum over all parses consistent with a given alignment, one simply uses an appropriate alignment envelope). The parameters were trained from 56376 (non-independent) pairwise alignments from RFAM \[[@B37]\], with each pairwise alignment making a weighted contribution of 1/Nto the counts computed during Expectation Maximization training, where Nis the number of sequences in the multiple alignment from which the pairwise alignments were derived. Furthermore, the pairwise alignments were binned according to sequence identity, providing four alternative parameterisations; the bin ranges were 30--40%, 50--60%, 70--80% and 90--100%. stemloc allows the user to re-estimate all parameters from their own personal training set of trusted alignments. This may be a useful feature, since the training procedure described above is probably biased. Since training was performed using all kinds of sequence available in RFAM, including RNA sequences with computationally predicted secondary structure as well as those for which structures were experimentally confirmed, it is possible that the stemloc parameters may be skewed by the parameters of the computational methods used by the RFAM curators to predict structure. These include the homology modeling program INFERNAL \[[@B37]\] and the de novostructure prediction program PFOLD \[[@B8]\]. Authors\' contributions ======================= IH designed, programmed, tested and documented the algorithms. Acknowledgements ================ The author thanks Sean Eddy for inspiring discussions and three anonymous reviewers for their helpful suggestions. The work was conceived during an NIH-funded workshop on RNA informatics organised by Elena Rivas and Eric Westhof in Benasque, Spain, 2003. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A parse tree for the grammar of Table 1. Each internal node is labeled with a nonterminal (Stem or Loop); additionally, the subsequences (X~ij~, Y~kl~) generated by each internal node are shown. The parse tree determines both the structure and alignment of the two sequences. The cut-points of the alignment are the sequence co-ordinates at which the alignment can be split, i.e. {(0, 0), (1, 1), (2, 2) \... (15, 12), (16, 13), (17, 14)}. ::: ![](1471-2105-6-73-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Parsing a pair of sequences (X, Y) using the Inside algorithm involves iterating over subsequence-pairs (X~ij~, Y~kl~) specified by four indices (i, j, k, l). In the constrained Inside algorithm, these indices are only valid if the fold envelopes(triangular grids) include the respective subsequences (i, j) and (k, l) (shown as black circles) and the alignment envelope(rectangular grid) includes both cutpoints (i, k) and (j, l) (shown as short diagonal lines). The filled cells in the rectangular grid show the aligned nucleotides. Note that the co-ordinates (i, j, k, l) lie on the grid-lines betweenthe nucleotides. ::: ![](1471-2105-6-73-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Bifurcation rules allow a subsequence-pair (X~ij~, Y~kl~) to be composed from two adjoining subsequence-pairs (X~im~, Y~kn~) and (X~nj~, Y~ni~). For this to be permitted by the constraints, the X-fold envelope (upper triangular grid) must contain subsequences (i, m), (m, j) and (i, j) (black dots), the Y-fold envelope (rightmost triangular grid) must contain subsequences (k, n), (n, l) and (k, l) (black dots) and the alignment envelope (rectangular grid) must contain cutpoints (i, k), (m, n) and (j, l) (short diagonal lines). The filled cells in the rectangular grid show the nucleotide homologies highlighted in the alignment. Note that all co-ordinates (i, j, k, l, m, n) lie on the grid-lines between nucleotides. ::: ![](1471-2105-6-73-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### These fold envelopes (triangular grids) limit the maximum length of subsequences (black dots), while the alignment envelope (rectangular grid) limits the maximum deviation of cutpoints (short diagonal lines) from the main diagonal. ::: ![](1471-2105-6-73-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### These fold envelopes (triangular grids) and alignment envelope (rectangular grid) limit the subsequences (black dots) and cutpoints (short diagonal lines) to those consistent with a given alignment and consensus secondary structure (shown). The alignment path is also shown on the alignment envelope as a solid black line, broken by cutpoints. ::: ![](1471-2105-6-73-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Fold envelope size is highly correlated with Nin the N-best fold test, although the variance is large due to the diversity of alignments in the test. ::: ![](1471-2105-6-73-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Alignment envelope size is highly correlated with Nin the N-best alignment test, although the variance is large due to the diversity of alignments in the test. ::: ![](1471-2105-6-73-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Alignment sensitivity as a function of envelope size parameter Nfor three different test regimes. ::: ![](1471-2105-6-73-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Alignment specificity as a function of envelope size parameter Nfor three different test regimes. ::: ![](1471-2105-6-73-9) ::: ::: {#F10 .fig} Figure 10 ::: {.caption} ###### Fold sensitivity as a function of envelope size parameter Nfor three different test regimes. ::: ![](1471-2105-6-73-10) ::: ::: {#F11 .fig} Figure 11 ::: {.caption} ###### Fold specificity as a function of envelope size parameter Nfor three different test regimes. ::: ![](1471-2105-6-73-11) ::: ::: {#F12 .fig} Figure 12 ::: {.caption} ###### Total running time of stemloc (including envelope generation phases) as a function of envelope size parameter Nfor three different test regimes. ::: ![](1471-2105-6-73-12) ::: ::: {#F13 .fig} Figure 13 ::: {.caption} ###### Peak memory usage of stemloc (i.e. the size of the principal CYK matrix) as a function of envelope size parameter Nfor three different test regimes. ::: ![](1471-2105-6-73-13) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### A stochastic context-free grammar for generating pairwise alignments of RNA structures. ::: L → R (L→ R) ------ ---- ----------- -------------------------------------------------------------- Stem → stemExtend (1 - stemGap) basepairSubstitution \[AC, BD\] \\| stemExtend (stemGap/2) basepairIndel \[AC\] \\| stemExtend (stemGap/2) basepairIndel \[BD\] \\| (1 - stemExtend)(1 - bifurcate) baseSubstitution \[AB\] \\| Stem Stem (1 - stemExtend) bifurcate Loop → loopExtend (1 - loopGap) baseSubstitution \[AB\] \\| loopExtend (loopGap/2) baseIndel \[A\] \\| loopExtend (loopGap/2) baseIndel \[B\] \\| ε 1 - loopExtend ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The stemloc grammar, part 1 of 3: stem and loop structures. ::: L → R (L→ R) ----------- ---- ----------------- ------------------------------------------------------- Start → Stem startInStem \\| LBulge (1 - startInStem) postStem \[2\]/ (2 postStem\[i\]) \\| RBulge (1 - startInStem) postStem\[2\]/ (2 postStem\[i\]) \\| LRBulge (1 - startInStem) postStem \[3\]/ ( postStem\[i\]) \\| Multi (1 - startInStem) postStem \[4\]/ ( postStem\[i\]) Stem → ^xy^StemMatch 1 - stemGapOpen \\| ^y^StemIns stemGapOpen/2 \\| ^x^StemDel stemGapOpen/2 StemMatch → ^xy^StemMatch (1 - stemGapOpen) stemExtend \\| ^y^StemIns stemGapOpen/2 \\| ^x^StemDel stemGapOpen/2 \\| StemExit (1 - stemGapOpen)(1 - stemExtend) StemIns → ^xy^StemMatch (1 - stemGapExtend)(1 - stemGapSwap) stemExtend \\| ^y^StemIns stemGapExtend \\| ^x^StemDel (1 - stemGapExtend) stemGapSwap \\| StemExit (1 - stemGapExtend) (1 - stemGapSwap)(1 - stemExtend) StemDel → ^xy^StemMatch (1 - stemGapExtend)(1 - stemGapSwap) stemExtend \\| ^x^StemDel stemGapExtend \\| ^y^StemIns (1 - stemGapExtend) stemGapSwap \\| StemExit (1 - stemGapExtend) stemGapSwap (1 - stemExtend) StemExit → Loop postStem \[1\] \\| LBulge postStem \[2\]/2 \\| RBulge postStem \[2\]/2 \\| LRBulge postStem \[3\] \\| Multi postStem \[4\] Multi → LMulti RMulti 1 LMulti → LBulge multiBulgeOpen \\| Stem (1 - multiBulgeOpen) RMulti → Multi multiExtend \\| Stem (1 - multiExtend)(1 - multiBulgeOpen)^2^ \\| LBulge (1 - multiExtend)(1 - multiBulgeOpen) multiBulgeOpen \\| RBulge (1 - multiExtend)(1 - multiBulgeOpen) multiBulgeOpen \\| LRBulge (1 - multiExtend) multiBulgeOpen^2^ Loop → ^xy^LoopMatch (1 - loopGapOpen) \\| ^y^LoopIns loopGapOpen/2 \\| ^x^LoopDel loopGapOpen/2 LoopMatch → ^xy^LoopMatch (1 - loopGapOpen) loopExtend \\| ^y^LoopIns loopGapOpen/2 \\| ^x^LoopDel loopGapOpen/2 \\| ε (1 - loopGapOpen) (1 - loopExtend) LoopIns → ^xy^LoopMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^y^LoopIns loopGapExtend \\| ^x^LoopDel (1 - loopGapExtend) loopGapSwap \\| ε (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) LoopDel → ^xy^LoopMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^x^LoopDel loopGapExtend \\| ^y^LoopIns (1 - loopGapExtend) loopGapSwap \\| ε (1 - loopGapExtend)(1 - loopGapSwap)(1 - loopExtend) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The stemloc grammar, part 2 of 3: bulges. ::: L → R (L→ R) -------------- ---- -------------------- ------------------------------------------------------- LBulge → ^xy^LBulgeMatch (1 - loopGapOpen) \\| ^y^LBulgeIns loopGapOpen/2 \\| ^x^LBulgeDel loopGapOpen/2 LBulgeMatch → ^xy^LBulgeMatch (1 - loopGapOpen) loopExtend \\| ^y^LBulgeIns loopGapOpen/2 \\| ^x^LBulgeDel loopGapOpen/2 \\| Stem (1 - loopGapOpen)(1 - loopExtend) LBulgeIns → ^xy^LBulgeMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^y^LBulgeIns loopGapExtend \\| ^x^LBulgeDel (1 - loopGapExtend) loopGapSwap \\| Stem (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) LBulgeDel → ^xy^LBulgeMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^x^LBulgeDel loopGapExtend \\| ^y^LBulgeIns (1 - loopGapExtend) loopGapSwap \\| Stem (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) RBulge → ^xy^RBulgeMatch (1 - loopGapOpen) \\| ^y^RBulgeIns loopGapOpen/2 \\| ^x^RBulgeDel loopGapOpen/2 RBulgeMatch → ^xy^RBulgeMatch (1 - loopGapOpen) loopExtend \\| ^y^RBulgeIns loopGapOpen/2 \\| ^x^RBulgeDel loopGapOpen/2 \\| Stem (1 - loopGapOpen) (1 - loopExtend) RBulgeIns → ^xy^RBulgeMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^y^RBulgeIns loopGapExtend \\| ^x^RBulgeDel (1 - loopGapExtend) loopGapSwap \\| Stem (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) RBulgeDel → ^xy^RBulgeMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^x^RBulgeDel loopGapExtend \\| ^y^RBulgeIns (1 - loopGapExtend) loopGapSwap \\| Stem (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) LRBulge → ^xy^LRBulgeMatch (1 - loopGapOpen) \\| ^y^LRBulgeIns loopGapOpen/2 \\| ^x^LRBulgeDel loopGapOpen/2 LRBulgeMatch → ^xy^LRBulgeMatch (1 - loopGapOpen) loopExtend \\| ^y^LRBulgeIns loopGapOpen/2 \\| ^x^LRBulgeDel loopGapOpen/2 \\| RBulge (1 - loopGapOpen) (1 - loopExtend) LRBulgeIns → ^xy^LRBulgeMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^y^LRBulgeIns loopGapExtend \\| ^x^LRBulgeDel (1 - loopGapExtend) loopGapSwap \\| RBulge (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) LRBulgeDel → ^xy^LRBulgeMatch (1 - loopGapExtend)(1 - loopGapSwap) loopExtend \\| ^x^LRBulgeDel loopGapExtend \\| ^y^LRBulgeIns (1 - loopGapExtend) loopGapSwap \\| RBulge (1 - loopGapExtend)(1 - loopGapSwap) (1 - loopExtend) ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### The Stemloc grammar, part 3 pf 3: emissions. ::: L → R (L→ R) -------------------- --- ------- ------------------------------------- ^xy^LoopMatch → baseSubstitution \[A, B\] ^y^LoopIns → baseIndel \[B\] ^x^LoopDel → baseIndel \[A\] ^xy^LBulgeMatch → baseSubstitution \[A, B\] ^y^LBulgeIns → baseIndel \[B\] ^x^LBulgeDel → baseIndel \[A\] ^xy^RBulgeMatch → baseSubstitution \[C, D\] ^y^RbulgeIns → baseIndel \[D\] ^x^RBulgeDel → baseIndel \[C\] ^xy^LRBulgeMatch → baseSubstitution \[A, B\] ^y^LRBulgeIns → baseIndel \[B\] ^x^LRBulgeDel → baseIndel \[A\] ^xy^StemMatch → basepairSubstitution \[AC, BD\] ^y^StemIns → basepairIndel \[BD\] ^x^StemDel → basepairIndel \[AC\] ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### The subset of RFAM used to test the constrained SCFG algorithms. ::: RFAM family Seauence (EMBL.ID / startpoint-endpoint) Alignment Basepair ------------- ------------------------------------------ -------------------------- ---------- ------- -------- ------- S15 AE004150.1/7123-7243 AE004888.1/2785-2659 0.65 0.752 0.462 0.353 S15 AE005545.1/3797-3683 AE004888.1/2785-2659 0.652 0.701 0.615 0.4 U3 U27297.1/2-180 AF277396.1/3-126 0.252 0.248 0.0833 0.087 glmS AL935254.1/94449-94600 AE010557.1/24-169 0.603 0.599 0.667 0.595 glmS AE010557.1/24-169 AE013165.1/2616-2459 0.532 0.587 0.545 0.667 glmS AL596166.1/50734-50929 AE013165.1/2616-2459 0.873 0.873 0.757 0.7 glmS AC078934.3/32621-32405 AE010557.1/24-169 0.869 0.863 0.756 0.689 glmS AL935254.1/94449-94600 AE013165.1/2616-2459 0.715 0.715 0.811 0.769 Purine AE007775.1/3558-3459 AL591981.1/205922-205823 0.869 0.869 0.773 0.81 Purine AL591981.1/205922-205823 AP004595.1/160373-160472 0.838 0.838 0.591 0.5 Purine AE007775.1/3558-3459 AE010606.1/4680-4581 0.67 0.699 0.636 0.875 Purine AP003194.2/163700-163601 AE016809.1/202496-202595 0.84 0.866 0.857 0.75 U5 M16510.1/245-451 AF095839.1/890-777 0.584 0.579 0.667 0.8 U5 X63789.1/2236-2349 AF095839.1/890-777 0.716 0.69 0.8 0.8 IRE AY112742.1/12-41 S57280.1/391-417 0.667 0.667 0.6 0.75 IRE AF266195.1/14-43 X01060.1/3950-3976 0.963 0.963 0.9 0.9 IRE S57280.1/391-417 X13753.1/1434-1460 1 1 0.6 0.6 IRE AY112742.1/12-41 X13753.1/1434-1460 0.778 0.778 0.8 0.727 IRE AF266195.1/14-43 AF171078.1/1416-1442 0.963 0.963 0.7 0.7 IRE AF171078.1/1416-1442 X01060.1/3950-3976 1 1 0.7 0.875 6S Y00334.1/77-254 AL627277.1/108623-108805 0.869 0.869 0.811 0.754 6S AE004317.1/5626-5807 AL627277.1/108623-108805 0.777 0.777 0.736 0.709 :::	PubMed Central	2024-06-05T03:55:56.439537	2005-3-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090553/", "journal": "BMC Bioinformatics. 2005 Mar 24; 6:73", "authors": [ { "first": "Ian", "last": "Holmes" } ] }
PMC1090554	Background ========== Computational approaches to problem solving need to interleave information accessand algorithm executionin a problem-specific workflow. In complex domains like molecular biosciences, workflows usually involve iterative steps of querying, analysis and optimization. Bioinformatic experiments are often workflows; they link analytical methods that typically accept an input file, compute a result, and present an output file. Most tool-driven integration approaches have so far addressed the problem of providing a single GUI for a set of analytical methods. Combining methods into a flexible framework is usually not considered. Analytical workflows provide a path to discover information beyond the capacities of simple query statements, but are much less easy to implement within a common environment. Workflow management systems (WFMS) are basically systems that control the sequence of activities in a given process \[[@B1]\]. In molecular bioscience, these activities can be divided among those that address query formulation, and those that focus more on analysis. At this abstract level, WFMS could serve to control the execution of both query and analytical procedures. All of these procedures involve the execution of activities, some of them manual, some automatic. Dependency relationships among them can be complex, making the synchronization of their execution a difficult problem. One dimension of the complexity of workflows in molecular biosciences is given by the various transformations performed on the data. Syntactic (operational) interoperability establishes the possibility for data to be piped from one method into another. Semantic issues (another dimension) arise from the fact that we need to separate domain knowledge from operational knowledge. We should be able to describe a task of configuring a workflow from its primary components according to a required specification, and implement a program that realizes this configuration independently of the workflow and components themselves. Biologists provide rich descriptions of their experiments (materials and methods) so they can be easily replicated. Once techniques have been standardized, usually this knowledge is encapsulated in the form of an analytical protocol. With in silicoexperiments as well, analytical protocols make it possible for experiments to be replicated and shared, and (viameta-information) for the knowledge behind these workflows to be captured. These protocols should be reproducible, ontology-driven, curated internally, and annotated externally. Systems such as W2H/W3H \[[@B2]\] and PISE \[[@B3]\] provide some tools that allow methods to be combined. W3H is a task framework that allows the methods available under W2H \[[@B4]\] to be integrated; however, those tasks have to be hard-coded. In the case of PISE, the user can either define a macro using Bioperl <http://www.bioperl.org>, or use the interface provided and register the resulting macro. In either case, it is assumed that the user can program, or script in Perl. Macros cannot be exchanged between PISE and W2H, although these two systems provide GUIs for more or less the same set of methods (EMBOSS: \[[@B5]\]). Indeed, macros cannot be easily shared even among PISE users. Biopipe <http://www.biopipe.org>, on the other hand, provides integration for some analytical tools using Bioperl API (Application Programming Interface) using MySQL to store results as well as the workflow definition; in this way, users are able to store results in MySQL and monitor the execution of the pre-defined workflow. The TAVERNA <http://taverna.sourceforge.net> project provides similar capabilities to those offered by GPIPE. However, on one hand inclusion of new analytical methods is not currently possible since no GUI generator is provided, and on the other hand as TAVERNA is part of myGrid \[[@B6]\] it follows a different integrative approach (Web Services). Pegasys \[[@B7]\] is a similar approach, going beyond analytical requirements and providing database capacities. GPIPE provides a real capacity for users to define and share complete analytical workflows (methods, parameters, and meta-information), substantially mitigating the syntactic complexity that this process involves. Our approach addresses overall collaborative issues as well as the physical integration of tools. GPIPE provides an implementation that builds on a flexible syntactic structure and a set of algebraic operations for analytical workflows. For testing purposes we provide a simple example of a workflow (inference of a phylogeny of rodents) that involves piping among three methods. Although here their execution takes place on a common server, it is equally possible to distribute the process over a grid using GPIPE. Results ======= Our workflow follows a task-flow model; in bioinformatics, tasks can be understood as analytical methods. If workflow models are represented as a directed acyclic graph (DAG), analytical methods then appear as nodes, and state information is represented as conditions attached to the edges. Our syntactic structure and algebraic operators can be used to represent a large number of analytical workflows in bioinformatics; surprisingly, there are no other algebraic operators reported in the literature capable of symbolizing the different operations required for analytical workflows in bioinformatics (or, indeed, more broadly in e-science, although they are widely used in the analysis of business processes). Different groups have developed a great diversity of GUIs for EMBOSS and GCG, but a meta-analysis of the processes within which these analytical implementations are immersed is not yet fully available. Some of the existing GUIs have been developed to make use of grammatical descriptions of the analytical methods, but there exists no standard meta-data framework for GUI and workflow representation in bioinformatics. Syntactic and algebraic components ---------------------------------- Our workflow conceptualization (Figure [1](#F1){ref-type="fig"}) closely follows those of Lei and Singh \[[@B8]\] and Stevens et al.\[[@B9]\]. We have adapted these meta-models to processes in bioinformatic analysis. We consider an input/output data objectas a collection of input/output data. For us a transformeris the atomic work item in a workflow. In analytical workflows, it is an implementation of an analytical algorithm (analytical method). A pipe componentis the entity that contains the required input-output relation (e.g.information about the previous and subsequent tasks); it assures syntactic coherence. Our workflow representation has tasks, stages, and experimental conditions (parameters). In our view, protocolsare sets of information that describe an experiment. A protocol contains workflows, annotations, and information about the raw data; therefore we understand a workflowto be a group of stages with interdependencies. It is a process bound to a particular resource that fulfils the analytical necessities. We identify needs common to analytical workflows in bioinformatics: • Flexibility in structuring and modelling (open-ended, sometimes ad hocworkflow definition, allowing decision-making whilst a workflow is being executed). • Support for workflows with a complex (or nested) inner structure of individual steps (such that multi-level modelling becomes appropriate). Biological workflows may be complex not simply because of the discrete number of steps, but due to the highly nested structure of iteration, recursion and conditional statements that, moreover, may involve interaction with non-workflow systems. • Distribution of workflow execution over grid environments. • Management of failures. This particular requirement is related to conditional statements: wherethe service will be executed should be evaluated based on considerations of availability and efficiency made previous to the execution of the workflow. In situations where a failure halts the process, the system should either recover it, or dispatch it somewhere else without requiring intervention by the user. • System functionality features such as browsing and visualization, documentation, or coupling with external tools, e.g.for analysis. • A semantic layer for collaborative purposes. This semantic layer has many other features, and may be the foundation for intelligent agents that facilitate collaborative research. Executing these bioinformatic workflows further requires: • Support for long-running activities with or without user interaction. • Application-dependent correctness criteria for execution of individual and concurrent workflows. • Integration with other systems (e.g. file managers, database management systems, product data managers) that have their own execution/correctness requirements. • Reliability and recoverability with respect to data. • Reliable communication between workflow components and processing entities. Among these types of requirements, we focus our analysis only on those closely related to workflow design issues, more specifically (a) the piping of data, (b) the availability of conditional statements, (c) the need to iterate one method over a set of different inputs, (d) the possibility of recursion over a parameter space for a method, (e) and the need for stop/break management. Algebraic operators can accurately capture the meaning of these functional requirements. To describe an analytical workflow, it is necessary to consider both algebraic operators and syntactic components. In Table [1](#T1){ref-type="table"} we present the definition of those algebraic operators we propose, and in Figure [2](#F2){ref-type="fig"} we illustrate how these operators and syntactic elements together can describe an analytical workflow. Iterationis the operator that enables processes in which one transformer is applied over a multiple set of inputs. A special case for this operator occurs when it is applied over a blank transformer; this case results in replicates of the input collection. Consider an analytical method, or a workflow, in which the same input is to be used several times; the first step would be to use as many replicates of the input collection as needed. The recursionoperation takes place when one transformer is applied with parameters defined not as a single value, but as a range or as a set of values. The conditional operator has to do with the conditioned execution of transformers. This operation can be attached to a function evaluated over the application of a recursion, or of an iteration; if the stated condition is true, then the workflow executes a certain path. Conditional statementsmay also be applied to cases where an argument is evaluated on the input; the result affects not a path, but the parameter space of the next stage. The suspension/resumptionoperation stands for the capacity of the workflow to stop and re-capture the jobs. Formal Concept analysis (FCA) is a mathematical theory based on ordered sets and complete lattices. Numerous investigations have shown the usefulness of concept lattices for information retrieval combining query and navigation, learning and data-mining, visual constructors and visual programming \[[@B10]\]. FCA helps one to define valid objects, and identify behaviours for them. We are currently working on a complete FCA for biological data types and operations (database and analytical). Here we define operators in terms of pre- and post-conditions, as a step toward eventual logical formalization. We focus on those components of the discovery process not directly related to database operations; a good integration system will \"hide\" the underlying heterogeneity, so that one can query using a simple language (which views all data as if they are already in the same memory space). Selection of the query language depends only on the data model. For the XML \"data model\", XML-QL, XQL, and other XML query languages are available. For the nested relational model there are nested relational calculi and nested relational algebras. For the relational model SQL, relational algebras and so on are available. For database operations, the issues that arise are lower-level (e.g.expression of disk layout, latency cost, etc. in the context of query optimisation), and it is not clear that any particular algebra offers a significant advantage. Operator: Iteration(I): I\[Transformer, (CC1, CC2, \..., CCn)\]: (CC1\', CC2\', \..., CCn\') Pre-condition: T = Transformer ∧ T ≠ *blank* C = {CC1, CC2, \..., CCn} such that CCi∈ { Biological data types } Post-condition: C\' = {CC\'1, CC\'2, \..., CC\'n} such that CC\'i = T(CCi) ∧ 1 ≤ i ≤ n Operator: Iteration(I): I\[*blank: num, (CC1, CC2, \..., CCn)\]: (CC1, CC2, \..., CCn)1, (CC1, CC2, \..., CCn)2, \..., (CC1, CC2, \..., CCn)num* Pre-condition: num∈ ![](1471-2105-6-87-i1.gif), num = number of replicates. C = {CC1, CC2, \..., CCn} such that CCi∈ { Biological data types } Post-condition: C\' = {CC\'1, CC\'2, \..., CC\'n} such that CC\'i = CCi∧ 1 ≤ i ≤ n Operator: Recursion(R): R\[Transformer: Parameter, (Parm\_Space)\]: Parm\_Space\' Pre-condition: P = Parametersuch that P ∈ Parm\_Space∨ (Parm\_Space= { Parm\_Values} ⋁ Parm\_Space= { Parm\_Values}) T = Transformer Post-condition: Parm\_Space\' = T (Parm\_Space) Operator: Condition(C): C\[Functional\_Condition\]: PATH Pre-condition: FC = Functional\_Condition Post-condition: PATH = true ∨ false ∨ value Operator: Suspension/Resumption(S): S\[re-take, jobs\]: Execution Pre-condition: (Re-take= true) ∨ (Re-take= false ∧ jobs= Set of jobs which should be suspended) Post-condition: (Re-take= true ∧ ((Execution = true ∧ Previously suspended jobs are re-taken)∨ (Execution = false ∧ There were no suspended jobs))) ∨ (Re-take= false ∧ (Execution = true ∧ ∀ j such that j ∈ jobs, j is suspended)) A more-detailed example involves the inference of molecular phylogenetic trees by executing software that implements three main phylogenetic inference methods: distance, parsimony and maximum likelihood. Figure [3](#F3){ref-type="fig"} illustrates how our algebraic operators and syntactic components define the structure of this workflow. In collaboration with CIAT (Center for International Tropical Agriculture, Cali, Colombia) we have implemented an annotation workflow using standard technology (GPIPE/PISE) and web services (TAVERNA). Our case workflow is detailed in Figure [4](#F4){ref-type="fig"}. Implementation of both of these workflows was a manual process. GUI generation was facilitated by using PISE as our GUI generator, and this simplified the inclusion of new analytical methods as needed. Database calls had to be manually coded in both cases. Choreographing the execution of the workflow was not simple, as neither has a real workflow engine. It proved easier to give users the ability to manipulate parameters and data with PISE/GPIPE, partly due to wider range of methods within BioPerl partly because algebraic operators were readily available as part of PISE/GPIPE. From this experience we have concluded that, due to the immaturity of current available web service engines, it is still most practical to implement simple XML workflows that allow users to manipulate parameters, use conditional operators, and carry out write and read operations over databases. This balance will, of course, presumably shift as web services mature in the bioinformatics applications domain. Workflow generation, an implementation -------------------------------------- We have developed GPIPE, a flexible workflow generator for PISE. GPIPE extends the capabilities of PISE to allow the creation and sharing of customised, reusable and shareable analytical workflows. So far we have implemented and tested GPIPE over only the EMBOSS package, although extension to other algorithmic implementations is possible where there is an XML file describing the command-line user interface. Workflows automate businesses procedures in which information or tasks are passed between conforming entities according to a defined set of rules; some of these business rules are defined by the user, and in our implementation are managed viaGPIPE. For our purposes, the conforming entities are analytical methods (Clustal, Protpars, etc.). Syntactic rules drive the interaction between these entities (e.g.to ensure syntactic coherence between heterogeneous file formats). GPIPE also assures the execution of the workflow, and makes it possible to distribute different jobs over a grid of servers. GPIPE addresses these requirements using mostly Bioperl. In GPIPE, each analysis protocol (including any annotations, i.e.meta-data) is defined within an XML file. A Java applet provides the user with an exploratory tool for browsing and displaying methods and protocols. Synchronization is maintained between client-side display and server-side storage using Javascript. Server-side persistence is maintained through serialized Perl objects that manage the workflow execution. GPIPE supports independent branched tasks in parallel, and reports errors and results into an HTML file. The user selects the methods, sets parameters, defines the chaining of different methods, and selects the server(s) on which these will be executed. GPIPE creates an XML file and a Perl script, each of which describes the experiment. The Perl file may later be used on a command-line basis, and customized to address specific needs. The user can monitor the status of workflow execution, and access intermediary results. A workflow built with GPIPE can distribute its analyses onto different, geographically disperse GPIPE/PISE servers. Discussion ========== The syntactic and algebraic components we introduce above make it possible to describe analytical workflows in bioinformatics precisely yet flexibly. Detailed algebraic representation for these kinds of processes have not previously been used in this domain, although they are commonly used to represent business processes. Since open projects such as Bioperl or Biopipe contain the rules and logic for bioinformatic tasks, we believe that having an algebraic representation could contribute importantly to the development of a biological \"language\" that allows developers to avoid the tedious parsing of data and analytical methods so common in bioinformatics. The schematic representation for workflows in bioinformatics that we present here could evolve to cover other tool-driven integrative approaches such as those based on web services. Workflows in which concrete executions take place over a grid of web services involve basically the same syntactic structure and algebraic operators; however, a clear business logic needs to be defined beforehand for those web services in order to deepen the integration beyond simply the fact of remote execution. A higher level of sophistication for the pipe component as well as for the conditional operator may be needed, since remote execution requires (for example) assessment and availability of the service for the job to be successfully dispatched and processed. For our implementation we use two agents, one on the client side and the other on the server side, with the queue handled by PBS (Portable Batch System). It is possible to add a semantic layer, thereby allowing conceptual selection of the transformers; clear separation between the operational domain and the knowledge domain be would then be achieved naturally. Semantic issues are particularly important with these kinds of workflows. An example may be derived from Figure [3](#F3){ref-type="fig"}, where three different phylogenetic analysis workflows are executed. These may be grouped as equivalent, but are syntactically different. Selection should be left in the hands of the user, but the system should at least inform about this similarity. Despite agreement on the importance of semantic layers for integrative systems, such a level of sophistication is far from being achieved. Lack of awareness of the practical applications of such technologies is well illustrated with a traditional and well-studied product: Microsoft Word^®^. With Word, syntactic verification can take place as the user composes text, but no semantic corroboration is done. For two words like \"purpose\" and \"propose\", Word advises on syntactic issues, but gives no guidance concerning the context of the words. Semantic issues in bioinformatic workflows are more complex, and it is not clear if existing technologies can effectively overcome these problems. Transformers and grid components are intrinsically related because the services are de factolinked to a grid component. It has been demonstrated that the use of ontologies facilitates interoperability and the deployment of software agents \[[@B11]\]; correspondingly, we envision semantic technology supporting the agents to form the foundation of future workflow systems in bioinformatics. The semantic layer should make the agents more aware of the information. More and more GUIs are available in bioinformatics; this can be seen in the number of GUIs for EMBOSS and GCG alone. Some of them incorporate a degree of workflow capability, more typically a simple chaining of analytical methods rather than flexible workflow operations. A unified metadata model for GUI generation is lacking in the bioinformatics domain. Web services are relatively easy to implement, and are becoming increasingly available as GUI systems are published as web services. However, web services were initially developed to support processes for which the business logic is widely agreed upon, well-defined and properly structured, and the extension of this paradigm to bioinformatics may not be straightforward. Automatic service discovery is an intrinsic feature of web services. The accuracy of the discovery process necessarily depends on the ontology supporting this service. Systems such as BioMoby and TAVERNA make extensive use of service discovery; however, due to the difficulty in describing biological data types, service discovery is not yet accurate. It is not yet clear whether languages such as OWL can be developed to describe relations between biological concepts with the required accuracy. Integrating information is as much a syntactic as a semantic problem, and in bioinformatics these boundaries are particularly ill-defined. Semantic and syntactic problems were also identified from the case workflow described in Figure [4](#F4){ref-type="fig"}. There, we saw that to support the extraction of meaningful information and its presentation to the user, formats should be ontology-based and machine-readable, e.g.in XML format. Lack of these functional features makes manipulation of the output a difficult task that is usually addressed by use of parsers specific to each individual case. For workflow development, human readability can be just as important. Consider, for example, a ClustalW output where valid elements could be identified by the machine and presented to the user together with contextual menus including different options over the different data types. In this way the user would be able to decide what to do next, where to split a workflow, and over which part of the output to continue or extend the analysis. Inclusion of this functionality would allow the workflow to become more concretely defined as it is used. Failure management is an area in which we can see a clear difference between the business world and bioinformatics. In the former, processes rarely take longer than an hour and are not so computationally intensive, whereas in bioinformatics, processes tend to be computationally intensive and may take weeks or months to complete. How failures can be managed to minimize losses will clearly differ between the two domains. Due to the immaturity of both web services and workflows in bioinformatics, it is still in most cases more practical to hard-code analytical processes. Improved failure management is one of the domain-specific challenges that faces the application of workflows in bioinformatics. So far we have intentionally referred to GUIs and workflows as more-or-less independent. A glimpse into the corresponding metadata reveals that GUIs are themselves components of workflow systems. In the bioinformatics domain this relationship is particularly attractive, since algebraic operations are usually highly nested. The interface system should therefore provide a programming environment for non-programmers. The language as such is not complex, but makes extensive use of statements such as while\...do, if\...then\...else, and for\...each. The representation should be natural to the researcher, separating the knowledge domain from the operational domain. Conclusion ========== We have developed GPIPE, a flexible workflow generator that makes it possible to export workflow definitions either as XML or Perl files (which can later be handled viathe Bioperl API). Our XML workflow representation is reusable, execution and edition of those generated workflows is possible either via the BioPerl API or the provided GUI. Each analysis is configurable, as users are presented with options to manipulate all available parameters supported by the underlying algorithms. Integration of new algorithms, and Grid execution of workflows, are also possible. Most available integrative environments rely on parsers or syntactic objects, making it difficult to integrate new analytical methods into workflow systems. We are planning to develop a more wide-ranging algebra that includes query operations over biological databases as well as different ontological layers that facilitate data interoperability and integration of information where possible for the user. We do not envision GPIPE to be a complete virtual laboratory environment; future releases will provide a content management system for bioinformatics with workbench capacities developed on top of ZOPE <http://www.zope.org>. We have tested our implementation over SUSE and Debian Linux, and over Solaris 8. Authors\' contributions ======================= AGC was responsible for design and conceptualization, took part in implementation, and wrote a first draft of the manuscript. ST was the main developer of GPIPE. LJG assisted with server issues and FCA. MAR supervised the project and participated in writing the manuscript. Acknowledgements ================ We gratefully acknowledge the collaboration of Dr Fernando Rodrigues (CIAT) in developing the case study outlined in Figure [4](#F4){ref-type="fig"}, and Dr Lindsay Hood (IMB) for valuable discussions. ST thanks Université Montpellier II for travel support. This work was supported by ARC grants DP0344488 and CE0348221. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Syntactic components describing bioinformatic analysis workflows. ::: ![](1471-2105-6-87-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Syntactic components and algebraic operators. ::: ![](1471-2105-6-87-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Phylogenetic analysis workflow. ::: ![](1471-2105-6-87-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Case workflow. ::: ![](1471-2105-6-87-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### GPIPE. <http://if-web.imb.uq.edu.au/Pise/5.a/gpipe.html> ::: ![](1471-2105-6-87-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Algebraic operators ::: ------------------------------------------- ------------------------------------------------------------------ Operator: Iteration (*I) I\[Transformer\](CC1, CC2\...CCn) Operator: Recursion (R) R\[Transformer: Parameter\](Parm\_Space) Operator: Condition (C) C\[functional condition(true:PATH;false:PATH;value:PATH)\] Operator: Suspension/resumption (S) S*\[re-take jobs: execution\] ------------------------------------------- ------------------------------------------------------------------ :::	PubMed Central	2024-06-05T03:55:56.445823	2005-4-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090554/", "journal": "BMC Bioinformatics. 2005 Apr 7; 6:87", "authors": [ { "first": "Alexander", "last": "Garcia Castro" }, { "first": "Samuel", "last": "Thoraval" }, { "first": "Leyla J", "last": "Garcia" }, { "first": "Mark A", "last": "Ragan" } ] }
PMC1090555	Background ========== Knowledge discovery and data mining in the biological literature have been attracting more and more interest \[[@B1],[@B2]\]. Automated text mining can facilitate the efforts of both biological database curators \[[@B2]\], and of biologists who consult the literature to acquire novel information both within and outside of their immediate expertise. Text mining applications come in various styles. Some rely on statistical methods to detect unusually strong co-occurrences between genes or gene products (e.g., PubGene \[[@B3]\] and as described by Wilkinson et al. \[[@B4]\]). Other applications aim to extract precise information from the text, for instance protein mutations \[[@B5]\] or interactions \[[@B6],[@B7]\]. A new promising type of application, pioneered by Textpresso \[[@B8]\], consists of portals that help end-users locate information more effectively. Most text mining applications require the ability to identify and classify words, or multi-word terms, that authors use in an article to refer to biological entities (biological entities include, but are not limited to, genes and their products, cell types, and biological processes). This task is called named entity recognition and has been well studied in computer science for problems such as recognizing names of places, people and organizations in news articles \[[@B9]\]. Efforts to automate the named entity recognition task in the news domain have been very successful, with accuracies of automated methods that compete with accuracies of human efforts (\~95% \[[@B9]\]). However, adapting these methods to the biomedical literature has been challenging \[[@B10],[@B11]\], and recently published methods on text from abstracts report accuracies around 75% for protein names \[[@B11],[@B12]\]. Several strategies have been tried to recognize biological entity names in articles. Some methods rely on protein and gene databases to assemble dictionaries of protein names. A key disadvantage of these methods is that they critically depend on the database used as source of names, and will not recognize new protein names before the name is entered in the database. Another problem is that databases may not contain each name variant used to refer to a protein, so that partial or fuzzy matching of the names in the database to the text in the article is needed \[[@B13]\]. Fuzzy matching is difficult to control for and may introduce spurious matches. Another class of methods attempt to recognize protein names using the morphology of the term: whether a term contains mixed case, or includes a prefix or suffix next to a protein name \[[@B14]\]. The main advantage of these methods is that they can detect protein names that authors have just created, if they follow the morphological clues that the method recognizes. The main disadvantage is that authors do not use morphology consistently, and that certain terms of the language that have been used to refer to proteins (e.g., MAD and Eye, \[FlybaseIds FBgn0011648 and FBgn0000624\]) cannot be recognized based on morphology \[[@B10]\]. Most of these methods were developed for abstracts, because abstracts are readily available for millions of articles (e.g., PubMed). However, it is clear that information content in text from full length articles is much greater than in abstracts, even if the information density (information content divided by document length) is higher in abstracts \[[@B15],[@B16]\]. Specifically, as noted by Horn et al., information about protein mutations is much more likely to be found in full length articles than in abstracts \[[@B5]\]. This is most likely to be true of other types of information as well, such as experimental techniques and protocols, cell types, species names, and interactions. Since full texts are becoming more accessible (with open access efforts such as the Public Library of Science, PubMed Central or BioMed Central), developing methods to extract information from full text is becoming more practical. Compared to abstracts, however, full length articles are a complex source of text. They usually require pre-processing to convert the document from the HTML or PDF format to plain text streams. The result of the conversion will contain all the sections of the article and separating sections is a hard problem in itself because formatting styles and conventions vary across journals, and change periodically even for a given journal. In addition, journal presentation styles are interleaved with the text of the article and require specialized processing so that the tokenization of the document matches the way the document looks when rendered in a web browser. For instance, removing all HTML tags when converting an HTML document to text is inappropriate, as this may create tokens that do not exist in the article (for instance, when removing all HTML tags, \"morphology\<P\>Browser\" could be interpreted as \"morphologyBrowser\"). Although some measures can be taken to limit these problems, in practice, conversion procedures are difficult to fine tune for each style and journal, and converted full length articles are noisy sources of text. Therefore, methods designed to extract named entities from full text must be robust to types and amounts of noise which are not found in abstracts. In this article, we present a method that extracts biological named entities directly from full length articles. This method is used to process a corpus of 80,528 full text articles and assemble a catalog of protein name references indexed by an article PubMed identifier (PMID) with high precision. Based on this catalog, we construct a dictionary of 59,990 protein names. We also present a method that uses this dictionary to identify the set of protein names a full text article refers to. We evaluated the performance of this method with a set of full text articles that was not used for the construction of the catalog. The method evaluated on a test set of 14 articles achieves an average precision of 75%. Most other published methods have been trained and tested on abstracts only (e.g., GAPScore, YAGI and NLProt \[[@B12],[@B14],[@B17]\]). While these methods were not developed for full text, they are representative of the state of the art for protein name extraction. We thus chose to compare the results obtained with our approach to the results obtained with NLProt and YAGI, when used on the same text material. We found that the performance of our method compares favorably to the 42% average precision that NLProt achieves on the same test set, at similar levels of recalls. Tested on another set, our approach has better average precision than YAGI, but fails to detect long protein name variants that are often detected by YAGI. These results demonstrate the usefulness of direct name extraction from full text articles. We have implemented the approaches described here in the Textractor framework <http://icb.med.cornell.edu/crt/textractor/index.xml>. Results ======= High frequency terms in biological journal articles are enriched in biological entity names ------------------------------------------------------------------------------------------- Many biological research journal articles focus on one or a few genes or gene product(s). Typical examples are articles that describe the cloning of a particular gene. Other articles that describe the discovery of a new interaction will often focus on the two gene products involved in the interaction. Thus, the high frequency terms in these articles often refer to a gene/protein, the cell line in which the gene/protein is being studied, or the biological process in which the gene/protein is involved. Setting a threshold on the frequency of terms in an article allow us to define the set of terms that are most frequent in the article. In this study, we used a threshold of 30, and thus we considered \"most frequent\" a term that occurred at least 30 times in a given article (see Table [1](#T1){ref-type="table"}). Table [1](#T1){ref-type="table"} lists the most frequent terms found in one example article, and how we sort them into two categories. Category (a) includes names of biological entities. Category (b) includes terms that may have a high frequency in an article, but are not biological entity names. Terms in category (b) include common words or combinations of common words (e.g., \"the\", \"1\", \"figure\", \"the protein\", \"binding\"). It is possible to build an exhaustive list of terms in (b) because most of them tend to be repeated from article to article. We have created a filter that can remove most terms from category (b) from the list of frequent terms in an article. Figure [1](#F1){ref-type="fig"} presents the algorithm and information that this filter uses. An exhaustive list, however, cannot be built for terms in category (a) because, as with protein names, this list is potentially infinite: the rules that authors use to define new names are only bounded by their imagination. We applied a machine learning strategy to differentiate and classify terms that belong to category (a). Learning and classifying with the neighboring features ------------------------------------------------------ Three support vector machine (SVM) models were constructed to predict if a term occurs in a context that indicates (1) a protein versusa cell, (2) a protein versusa process (3) a protein versusan interaction (See Materials and Methods). Since each of the most frequent terms occurs at least 30 times in a given article, we can make at least 30 predictions for one term in one article. We expect that some occurrences of a term will occur in contexts that do not clearly indicate if the term is a protein or not. For instance, in a sentence such as \"The exact \[stoichiometry of this \> multienzyme complex \< is unknown but\] its molecular mass in insect cells\...\", the context included in the window offers little clue to decide if the term refers to a protein or not (The occurrence of the term under inspection is marked by \> \< and terms considered as features in the window around the occurrence are enclosed in \[ \]). In such cases, the SVM score is expected to be small. In comparison, occurrences like \"\...is restricted to skeletal muscle and does not affect \[expression of the \>Glut4\< isoform\].\", are very informative since the features \"expression\" and \"isoform\" are present together in the local context (note: if \"isoform\" always occurs after \"Glut4\" in that article, in our algorithm, \"Glut4 isoform\" may also be extracted by our system). We used a simple heuristic expression to combine the classification results of these three models (See Materials and Methods) and created a protein name catalog for all the articles in the JBC2000 dataset. Quality of the protein name catalog ----------------------------------- For the JBC2000 dataset, we predicted a catalog of 21,501 protein names and ranked them by their combined scores (Sc). The Sc distribution was evenly divided into 10 buckets. One continuous score range was then randomly selected in each of the 10 buckets, so that each range includes 50 protein names. We evaluated the quality of this catalog by manually verifying this random sample of 500 names in the corresponding articles. Figure [2](#F2){ref-type="fig"} summarizes this evaluation and provides precision measures. From JBC2000, we estimate that, on average only 3 or 4 protein names will be extracted from a given article. Since this number of names extracted is very low compared to the complete set of names that an average article contains, we have not evaluated the coverage of the prediction. However, because protein names are reused in different articles, if the method misses a name in one article, this name may still find its way into the protein name dictionary through another article (see below). A more interesting evaluation is thus how well would this catalog do in a specific application, such as extracting unique protein names in a full text article. This is the evaluation that we present in the rest of this article. Quality of the protein dictionary --------------------------------- To address this question, we created a protein dictionary from a more comprehensive protein catalog, which is based on 80,528 full text articles (from JBC, EMBO, PNAS) (see Figure [3](#F3){ref-type="fig"}). We used names in the dictionary to lookup protein names in a set of 14 full length articles from Nature Cell Biology(NCB14). The tool used to lookup protein names is provided in Additional File [1](#S1){ref-type="supplementary-material"}.We used the relative recall measure to compare our dictionary-based method to NLProt on NCB14 (See Materials and Methods). Table [2](#T2){ref-type="table"} summarizes these precision and relative recall measurements. These data indicate that the lookup approach that we implemented in Textractor outperforms NLProt on full text articles in precision (75% compared to 42%) at the same level of relative recall (See Materials and Methods for our definition of \"relative recall\'). During the evaluation on NCB14, we noticed that NLProt made systematic errors on tokens that appeared in the text of these articles because of journal style formats (e.g., mistakes were made for instance on terms such as \"NPG\" for Nature Publishing Group, or \"getObject(name)\" which was part of the NPG Javascript header that our pre-processing protocol failed to remove). These terms were mistaken for gene/protein names. To rule out any possible effect due to the selection of the articles in NCB14 (all selected from the same journal), we performed a second evaluation on 15 randomly selected articles from PubMedCentral (published in 2003 and not part of our training set). Data in Table [3](#T3){ref-type="table"} confirm that Textractor outperforms NLProt on full text (see relative F-1 measure). Table [3](#T3){ref-type="table"} also shows the results of the comparison between Textractor and YAGI \[[@B17]\]. While Textractor outperforms YAGI in term of precision, the recall of YAGI is more than double the recall of Textractor. This is summarized by a relative F1-measure of 51% for YAGI, and 42% for Textractor. Inspection of the validation data suggests that YAGI is very good at identifying protein names variants that are longer (word length) than the ones that Textractor typically detects (e.g., in article with PMCID 150636, when Textractor identified NS4B as a protein name, YAGI also identified \"hepatitis C virus nonstructural protein NS4B\"). This point also illustrates a difficulty of comparing protein name extraction methods, since the total set of correct names may include redundancy. In our evaluation, we have chosen to count as correct names those that refer to proteins by themselves, even if such names are shorter versions of longer names. These shorter versions are not partial names, however, because their occurrence in the text is sufficient to recognize a reference to a specific protein. Since YAGI and NLProt use similar sources of information, including protein names derived from biological databases, but differ in their machine learning approach, these results also suggest that Conditional Random Fields (see \[[@B18]\], used by YAGI), were able to better capture the distribution of protein names in full text, when learning on abstracts, than the learning approach used by NLProt. To corroborate this conjecture, many errors made by YAGI seem to appear in the reference section of articles, which has a different structure than the rest of a full text article. Since both YAGI and NLProt use a database derived dictionary, it would be interesting to see if the precision of these approaches on full text can be increased when using a dictionary built directly from full text. Limited overlap with SwissProt derived names -------------------------------------------- We matched terms in the dictionary to description lines in SwissProt (release 45), as described for matching terms in the dictionary to full text articles. We found that only 8.3% of the terms in our dictionary are included in SwissProt descriptions. Examples of terms present in the dictionary that do not match SwissProt description lines include: \"collagen XI\", \"Collagen alpha1 I\", \"TRHRs\", \"TRIM8 GERP\". This result indicates that our method is strong at identifying protein name variants that are not found in SwissProt descriptions. SwissProt is a main source of protein names used for building protein name recognition methods \[[@B12],[@B13],[@B17]\]. Therefore, combining our dictionary with names derived from SwissProt may lead to improved performance for a variety of name recognition methods. Implementation and performance ------------------------------ We implemented the methods described in this manuscript in the textractor framework (L. Shi and F. Campagne, unpublished). Briefly, this framework provides support to parse full length articles into sentences; create inverted indices where sentences are indexed by words; store sentences, articles and other information in a database; calculate features from sentences, articles or part of the above; and import results of predictions made with the features into the database for integration with other types of information. The Textractor framework uses JDO \[[@B19]\] for data persistence, MG4J \[[@B20]\] for inverted-index support, and SVMLight \[[@B21]\] for machine learning. The framework is implemented in Java 1.4+. The distribution includes a version of the lookup tool that functions independently of a database and uses MG4J for fast term lookups. This tool is distributed as Additional File [1](#S1){ref-type="supplementary-material"}. With a dictionary containing about 59,000 names, processing a set of 30 articles (133,675 words total) took about 16 seconds on a Red Hat Linux dual Xeon 3 GHz processor server, while it took NLProt 19 minutes, and YAGI 29 seconds. Similarly, YAGI processed a set of 693 articles in 753 seconds, while Textractor needed 494 seconds (65% of the time required by YAGI). We distribute the implementation of our method and other relevant information, e.g., the list of regular expressions used to extract the terms for SVM training, under the Gnu General Public License to maximize their use in the biomedical community <http://icb.med.cornell.edu/crt/textractor/index.xml>. Discussion ========== Information in full text ------------------------ Compared to full text, abstracts have the advantages of convenient access and uniformity of format. Based on recent studies, however, the information content in abstracts is less than half the information content in full text \[[@B15],[@B16]\]. In addition, most experimentalists would consider full text a more informative source of information than abstracts. Indeed, an abstract usually focuses on the general idea of a biological issue per se, but the details, such as how the issue was studied in that article, are very unlikely to be fully described in an abstract. For an experimentalist, the question is more often \"how\" than \"what\" and is always about details. Thus, automatic text mining applications that target experimentalists should ideally also be able to extract relevant information from full text articles. This study addresses an initial step in this direction -- to collect and classify several categories of biological entity names from full text that are essential for a detailed understanding of an issue, including how it was studied, from a biological article. These categories include the terms that refer to protein/gene, cell type, biological process, and interactions. Specifically, we compared the recognition of the protein/gene names with other systems and emphasized the strength of our approach in identifying term variants. Term variants and disambiguation -------------------------------- Our method draws on several concepts introduced in \[[@B22]\], but differs in the following ways. (1) We use only features derived from the context of the term. This makes our method insensitive to the morphology of the term and allow us to collect terms as diverse as \"d2-dopamine receptor\", \"dopamine D2 receptor\", \"D2 dopamine receptor\", \"D2R\", \"D2DR\", \"DRD2\", \"d2s receptor\", \"d2l receptor\" and their plural forms for the same protein in the catalog. (2) We applied term disambiguation to terms extracted from full text without the help of a dictionary of protein names. In contrast, \[[@B22]\] used Genbank as the source of its dictionary and extracted terms by fuzzy-matching the text of the article. (3) We combine information from the multiple sentences in which a term occurs to make a prediction if the term is a protein or not in a given article. Since we disambiguate only terms that appear more than 30 times in an article, information from at least 30 sentences is considered for each term in an article. (4) The precision measures that we report benchmark the ability of our method to precisely identify that a term is a protein in a given article. Performance measures given in \[[@B22]\] measure the performance of a related, but distinct task: assigning a term to one of three classes (protein, mRNA, gene) when it is already known that the term belongs to one of these classes. Extraction of long protein names -------------------------------- We have described that our method may fail to detect long protein names (see Results). Such names appear with a lower frequency in articles and are often abbreviated in full text. Approaches to extract acronym definitions from full-text may complement our approach and help extract long protein names (see \[[@B23]\] for a review of such approaches). Specific evaluation protocol for full-text ------------------------------------------ Protein name marking is a task where a program tries to mark the occurrences of all the protein names in a text. In full text, some protein names are repeated very often. If we followed the name marking benchmark standards, we would count correct extractions several times for such repeated terms. While this is not a problem for abstracts, we noticed that this practice would artificially inflate the precision measures on full text. Instead of benchmarking the performance of gene/protein name tagging in the text, we thus benchmark the performance of extracting the names of gene/proteins mentioned in a given article. We evaluate this by counting errors and success at the level of unique terms. Since this protocol is different from the ones used in other published studies, performance measures that we report cannot be compared to other published measures. Another factor that renders such comparison meaningless is the difference in the input material. We show in Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"} that the performance of a given method varies widely from article to article. Our validation protocol, however, compares the performance of several methods on the same input material, using the same evaluation measures and therefore supports objective comparisons between the methods benchmarked. Relative recall helps compare several methods --------------------------------------------- Obtaining accurate measures of coverage is challenging because coverage requires counting the number of correct protein names in an article. Several factors make obtaining these counts more difficult for full text than for abstracts. (1) Full texts refer to a large number of protein names in all the sections of the paper (in the first test set, NCB14, we recognize on average a total of 58 unique names per article). (2) It is difficult for human annotators to identify all spelling variants of protein names that automated methods may identify. To alleviate the impact of these factors on the evaluation, we presented annotators with names identified by each prediction method. In practice, this procedure guarantees that annotators consider each name variant predicted by one of the methods and determine if the name refers to a protein. The protocol thus helps avoid omissions that can occur when annotators are not familiar with the subject of the article, and directly provides the annotation counts (Ci) required to calculate the relative recall measure (see methods). Building the catalog -------------------- Our method creates a catalog of protein and gene name references directly from full length noisy article materials. Tanabe and Wilbur recently described a method to assemble a gene/protein lexicon from the text of abstracts \[[@B24]\]. While there is certainly an overlap between protein names used in abstract and in full text, it is not clear what the extent of the overlap is. Thus, our method and the one presented by Tanabe and Wilbur should be complementary. Finally, our study of the quality of the protein catalog in a common application (protein name recognition in full text material), demonstrates the utility of direct name extraction from full text articles. Conclusion ========== Our results show that a pure dictionary-based lookup method can outperform NLProt \[[@B12]\] on full text articles, when using a dictionary built directly from the same type of source material (full length articles). We have presented a method that can build such a focused dictionary from a large corpus (\> 80,000) of articles. The method is computationally efficient. We freely distribute the dictionary that we have built to carry out our evaluations and a program that can extract names of proteins from full length articles. Our method is robust to various sources of noise found in full length articles and achieves a state of the art level of precision on this material. A key element that contributes to the robustness of our method seems to be that we never extract a name from an article as a protein name based only on the morphology of the term, but instead require that the term is predicted several times as a protein name in other articles. This robustness may be the result of considering an array of evidence, found in the sentences from several articles, to determine if a term is likely to refer to a gene or gene product, and reusing this knowledge again and again. In its current state, an outstanding limitation of our approach is its inability to deal with certain types of term ambiguity (i.e., the same term referring to a protein (\"TnT\"/Troponin T or \"TnT\"/Translation and Transduction kit). This is an area for improvement that will require further research. Methods ======= Document pre-processing and extraction of terms ----------------------------------------------- We represent all documents with the Unicode standard \[[@B25]\]. We define a wordas a contiguous sequence of letter, digits, dashes and dot characters (we use the classification of Unicode characters established for the Java language and implemented in the java.lang.Character class of this language, so that letters include special characters such as α or β). We split documents into sentences of at least mcharacters, with heuristics that consider the types or identity of characters in a window of 6 characters around potential sentence terminators (characters .;?!). Parameter mwas set to 40 in this experiment, a value that rejects sentence splits if they would create a sentence shorter than about a quarter of a column of this article. Full details of the sentence splitting procedure are given in the source code in the supplementary material (see class SentenceSplitterIterator). We define a termas a word or a contiguous sequence of words, which appears in an article, but does not span sentence boundaries. The term may or may not refer to a protein name or to any grammatical class. We call n-grama term that contains nwords. \"The\" is thus considered a term, and a one-gram, while \"The method\" can be called a term, a 2-gram. The frequencyof a term in a document or set of documents is the number of times the term occurs in the document or set of documents. Here, we identify n-grams in an article by finding all the unique word sequences of length 1 to 5 words that have a frequency greater than one. (We observed that most protein names with high frequency are shorter than 5 words, but this parameter could be varied.) Disambiguating biological entity names -------------------------------------- In an approach similar to \[[@B22]\], we convert terms to features and use a machine learning approach to disambiguate protein names from other names. In contrast to \[[@B22]\], where Hatzivassiloglou et al. used morphology and positional information, we only use the context of the terms in the sentences of the article to make a decision. Furthermore, in our approach, only the terms that are the most frequent in an article are considered for disambiguation. We treat each occurrence of a term in an article separately and then aggregate the predictions to determine one prediction per term and article. This process is illustrated in Figure [4](#F4){ref-type="fig"}. Calculating features for each occurrence of a term in the article ----------------------------------------------------------------- We call featurea real-valued number that is used as input to a machine learning algorithm. Let p(a,t,o)denote the position of one occurrence oof term tin article a, and i(w)be a function that maps each word of the corpus to an integer (a number that uniquely identifies a feature). Such a function could map word \"the\" to 12 and word \"The\" to 105, or map both words to the same integer, to wrap cases. The features for term tat p(a,t,o)are the set of words W(p(a,t,o))that exist in a window of words from p(a,t,o)- lto p(a,t,o)+ length(t)+ l-1, excluding the term word(s), with lthe length of the window on either side of the term occurrence and length(t)the number of words of term t. We define one feature, identified by its index: i(w)for each word wthat occurs in the training corpus. For each word wthat occurs in this window, and does not belong to the term we set the value of feature i(w)to 1. The value of feature i(w)is set to zero if the word does not occur within the window. The feature thus gets the same value if the term occurs once or multiple times in the window around a term occurrence. For experiments described in this article, we used a window size of 3. Training data set ----------------- Our training data set (JBC99) consisted of 1,814 articles (about 520,000 sentences) published in the Journal of Biological Chemistry (JBC) during the last quarter of year 1999. Full length articles were obtained as HTML files and converted to text using the HtmlParser package \[[@B25]\]. Image tags were replaced by their ALT tag, when available, or by whitespace. For most journals that present Greek letters as images, the ALT tag contains a textual representation of the symbol, in these cases, β is replaced by \"beta\". Presentation tags, such as \<p\> and \<br\>, were replaced by whitespace. Training SVM models ------------------- We trained three support vector machines \[[@B26]\] (also called SVM models). Training SVM models requires training data sets with positive and negative examples. To construct training data sets, we created lists of terms in several categories. To create these lists, we filtered the most frequent terms obtained for each article of the training corpus. We built four lists of non-ambiguous terms: protein/gene names (PG), cell names (C), process names (Pr), and interaction keywords (IK). We made sure that the names in these lists were non-ambiguous, that is, that the name, in any sentence context would be a true instance of its class. To construct PG, for instance, we included n-grams that match the regular expressions \".+ receptor\" or \".+ kinase\" (thus, n\> = 2), because n-grams that end in \"receptor\" or \"kinase\" are very unlikely to be used in a context where they do not refer to proteins. Other terms that could be ambiguous -- e.g., SNF, which could be a gene/protein name, or a funding agency (Swiss National Science Foundation) -- were not used for training. Regular expressions were used to facilitate the assembly of drafts of these lists, but the lists were carefully inspected and edited manually before training. Table [4](#T4){ref-type="table"} describes the composition of the training sets built from the non-ambiguous lists described above and indicates the number of terms used for training. Table [5](#T5){ref-type="table"} presents \"ξα estimates\" after training. The ξα values are conservative estimates of the leave-one-out error that can be computed efficiently after training an SVM \[[@B21]\]. We created three SVM model training sets: PG+/C-, where PGterms are labelled in the positive class, and Cterms in the negative class. The other sets used for training were PG+/Pr- and PG+/IK-, with the same naming conventions. These training set compositions are chosen so that the three SVM models trained from these datasets will give positive scores to terms that are predicted to be in the category PG. Training was performed with the RBF kernel and parameters γ = 0.005 and C = 19.4433 (PG+/C-), C = 19.0005 (PG+/IK-), C = 19.1587(PG+/Pr-). Binary SVM classifiers assign a predicted class to a term by considering the sign of the output of a trained support vector machine. The output of the SVM is un-calibrated and is not the prior probability of the class given the features \[[@B27]\]. However, the SVM output is correlated with the probability \[[@B28]\] and thus the magnitude of the output can be used as a measure of confidence in the prediction (small absolute values indicate smaller confidence, while larger absolute values indicate stronger confidence). Combining predictions for several occurrences of a term in one article ---------------------------------------------------------------------- For a SVM model, the individual classification scores for each occurrence of a term in an article were summed, so as to produce three values: Sum~PG+/C-~, Sum~PG+/IK-~and Sum~PG+/Pr-~. We used a simple heuristic expression to combine the three scores into the final Sc score (we summed the three Sum scores and adversely weighted a negative sum in any of the three classifications by multiplying each negative Sum by 50 before adding them). The parameter 50 was chosen empirically to give more weight to negative individual Sum scores. The combined score, Sc, is such that greater, positive values have a higher possibility of referring to protein names, and negative values are unlikely to refer to protein names. JBC2000 Test set ---------------- We assessed the quality of the biological entity name disambiguation with a test set derived from articles published in JBC during the year 2000. The test set is mostly independent from the training set, as the only common points between the articles in the two sets is that they were published in the same journal and formatted with the same conventions. Creation of a dictionary of gene and gene product names ------------------------------------------------------- Not each term in the protein catalog can be used directly to lookup protein names in articles. A key problem is ambiguity, terms that refer to proteins in certain articles, but refer to other concepts in other articles. In an attempt to reduce ambiguity, we filtered the protein catalog with several heuristics. These heuristics are presented in Figure [3](#F3){ref-type="fig"}. When applied to the catalog of protein references produced from the \"most frequent\" terms obtained from 80,528 articles from JBC, EMBO, and PNAS, these filters produce a dictionary of 59,990 terms. Evaluation of the dictionary to lookup gene and gene product names in full text ------------------------------------------------------------------------------- ### Dictionary Test Set construction We built the NCB14 set with 14 articles selected randomly from articles published in Nature Cell Biology in 2003. Since this journal was not used for the construction of the protein catalog, the style, formatting and names present in these test articles had never been used to develop our method. Furthermore, we did not refine or tune any of the parameters of our method after we started this evaluation. Therefore, the performance values that we report here should be representative of what can be expected when the method is used on new, unseen, but similar full length article materials. ### Matching names of the dictionary to articles with textractor Names found in the dictionary were matched to the text of each article in the test sets. When matching n-grams, we match letters and digits and ignore punctuation and special characters except dashes (\'-\') and dots (\'.\'). Using this strategy, if A, B and C are words, an \"A B-C\" in the dictionary will match \"A,B-C\" and \"A\#B-C\" in the text of an article, but not \"A B,C\". Using this matching procedure, the name \"PKC-zeta lambda\" will match \"PKC-zeta/lambda\" in article PMID 10749857. Matching of words shorter than 6 characters is case-sensitive (i.e., \"TnT\" will not match \"TNT\"), while matching of longer words is case insensitive \"human topoisomerase II\" will match \"human TOPOISOMERASE II\"). In contrast to other dictionary-based approaches (e.g., \[[@B13]\]), our matching procedure does not allow variations on the words that constitute the name: \"receptor\" will not match \"receptors\". Since this matching method is fairly strict, plurals and other variations of protein names must appear explicitly in the dictionary for the name to be matched to the full text. ### How to define gene/gene product/protein names The definition of protein that we used to train Textractor differs from the definition used by the human annotators who created the YAPEX corpus \[[@B29]\]. For Textractor, protein names refer to a protein or to the part of a protein, while for YAPEX, proteins must be single entities. The Textractor definition allows for protein domains (e.g., \"SH2 domain\") and parts of proteins, such as \"histone H3 at serine 10\", or protein complexes (\"the proteasome\"). Our definition was chosen pragmatically, since parts of proteins are often mentioned when describing interactions and a long term goal of our project is to extract protein names to support the extraction of information about interactions. We count partial matches as errors when the partial match does not refer to a protein name in itself. For instance, \"proteasome\" and \"the proteasome\" are both correct identifications, while \"endothelin-converting\" is incorrect even if \"endothelin-converting enzyme\" is a correct match in the same article. ### Comparing with NLProt Articles in the HTML format were converted to Unicode text as described under document preprocessing. Both NLProt and the Textractor lookup tool were given the same text material as input. The output produced by NLProt was parsed to extract the protein names recognized by the method. We sorted names to be unique (following the term matching criteria described above), and produced a tab delimited file (first column: PMID; second column: term extracted from this article). This file was merged with the Textractor predictions using the term as unique key. The numbers of occurrences found by each method are also listed. These files were annotated by the authors during the evaluation and are provided in supplementary material to allow comparison with future methods. ### Evaluation measures Various performance measures can be used when comparing prediction or extraction methods. Accuracy of a prediction is defined as the percentage of correct predictions, over all the classes of terms predicted. For instance, when predicting protein and non-protein names, accuracy measures how well both protein and non-protein names are predicted. However, accuracy can be misleading when the test set contains one class in a larger proportion than the other. For instance, if the test set contains 10% of terms that are proteins and no proteins names are predicted, the prediction has an accuracy of 90%. For this reason, we prefer the precision and recall measures. The precision is the accuracy measured over one class, for instance measured over proteins. It is calculated as the ratio of correct predictions over the number of predictions made. Recall measures how many elements of one class are predicted in this class. Recall is calculated as the ratio of correct predictions in one class over the total number of instances of this class that could have been predicted. Intuitively, \"precision\" measures how specific a prediction is when a class is predicted, and \"recall\" measures how much the prediction method has missed in a class. To combine and summarize performance values obtained for several articles, we use micro- and macro-evaluation \[[@B21],[@B25]\]. Macro-evaluation averages the values of precision and relative recall calculated in individual articles, while micro-evaluation sums the counts of correct and incorrect predictions over all the articles before calculating global measures. ### Relative recall When comparing several prediction methods, we define the relative recallof method i(among npossible methods)as: ![](1471-2105-6-88-i1.gif) where Ci represent the set of all the correct predictions made by method i, Card(set) represents the number of elements in a set, and Union(set1, set2,\...) represents the set that is the union of several sets. It can be seen that if any of the methods has perfect recall, the relative recall of each method matches the traditional recall. Furthermore, when the measure is applied with more than two methods, the values of the relative recall will converge towards the recall as the number of methods increases (each method contributes to the union the true positives that is detects, until the union matches the complete set of positives in the data set). The advantage of the relative recall is that only predicted terms need to be evaluated for each prediction method, so that reading the whole article is not required. Given the observed variability of precision in the articles of our test sets, we assume that the recall of an extraction method will also vary from one article to another. Therefore, it is clear that a measure of recall, however consistent and careful the evaluation, will not be informative if evaluated for a single article. Measuring relative recall lowers the cost of evaluation and makes it possible to compare the recall of several methods over larger article samples. Authors\' contributions ======================= LS and FC designed, conducted this study, and wrote the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Full-text lookup tool with dictionary.This tool implements the lookup approach that is used for evaluation of the protein dictionary in the manuscript. The Java Archive (Zip format) contains a copy of the protein dictionary that is also available from our web site. (Additional instructions will be printed to the console.) ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Thorsten Joachims for stimulating discussions, and for distributing SVMLight to the research community, and Lucy Skrabanek for technical assistance in the early stages of the project. We thank Sebastiano Vigna and Paolo Boldi for MG4J and for sharing version 0.9 with us before its official release, and Marko Srdanovic for integrating the new release with Textractor. We thank Mika Sven and Burkhard Rost for distributing NLProt and Burr Settles and Mark Craven for distributing YAGI/Abner to the research community. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Filtering process and exclusion lists.Oval boxes on the left of the figure present the exclusions lists used as input to the filtering process. An exclusion list is connected to the step of the filtering process that uses it. Each step filters out terms that match the exclusion list in sequence based on the rules described in the boxes of the second column. Some steps perform matches by considering the entire term (e.g., the first step on the top), other steps use only specific words in a term. These lists have been built and are being maintained manually. The rectangular boxes on the right show the number of terms that have been excluded at each step, when processing the most frequent terms from JBC2000 (numbers in parentheses indicate counts of unique terms). ::: ![](1471-2105-6-88-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Relation between the precision of the disambiguation and score values. Ten score values were chosen randomly. For each of these values, we considered the 50 terms that had scores immediately greater than the value, and evaluated if the term referred to a gene or gene product in the article where the prediction was made. Precision of the prediction was calculated as the number of correct predictions over the total number of predictions (50). The 500 names which were checked represent a random sample of 21,501 names in the catalog. The Figure shows that predictions made with higher scores have a greater probability to be correct. The label shown for each evaluation point indicates the percentage of terms found in the evaluation corpus with a score above the score of this specific point. ::: ![](1471-2105-6-88-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Heuristics filtering of the protein catalog to produce the protein dictionary.The dictionary was constructed by considering the classification results of a particular term in different articles. Step 1: we filtered out terms that were predicted to be a protein in less than 75% of the articles where a prediction was made. For example, if term A appears in 4 articles and is classified as a protein name in 3 of them, term A is accepted in the dictionary. This process collected 61,312 terms. Step 2: we removed terms with two characters or less. Step 3: to remove ambiguity with protein names that are also common nouns, we filter the dictionary against the Webster\'s Revised Unabridged Dictionary (G & C. Merriam Co., 1913, edited by Noah Porter, provided by Patrick Cassidy of MICRA, Inc, and retrieved from <http://www.dict.org>). We estimate that this edition contains about 80 common protein names (e.g., amylase). Step 4: we filter the dictionary against species names from the NCBI taxonomy database \[30\]. ::: ![](1471-2105-6-88-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Steps involved in constructing the catalog of protein references.Terms are shown enclosed in rectangular boxes. Terms may occur in the context of sentences (when on a horizontal line, left), or in an article (right). Step 1: Articles are split into sentences, and sentences are split into tokens. Tokens roughly correspond to words (see text for details). Tokens with high frequency that are not eliminated by the exclusion lists (see Figure 1) are grouped into n-grams. On the figure, APE1/ref-1 is a n-gram that consists of two tokens: APE1 and ref-1, and can be recognized if the two terms co-occur frequently in sequence in a full length article. When the terms are recognized, each occurrence of a term in sentences of the article is identified. Step 2: Machine learning features are calculated from the context of the term in the article (see text for details) and the support vector machine (SVM) model classifies the context of the term. We obtain the score for each context of a term. In our experimental setup, smaller scores suggest that the context provides little evidence that the term refers to a protein, while larger scores (in absolute values) indicate more support. Step 3: We calculate the combined score (Sc) as the sum of the scores for each occurrence of a given term in a given article. The final catalog consists of a table with one row per term and article. Each row has three columns: PubMedID, term, and Sc. ::: ![](1471-2105-6-88-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Example of terms that occur most frequently in an article (selected from article with PMID: 10506131). ::: Term Frequency Category ------------------- --------------- -------------- GnRH 79 a side chain 60 a the GnRH receptor 30 a d7.49 318 38 a the 503 b of 334 b and 212 b expression 49 b d7.49 38 a the GnRH 30 a Terms in italic are not considered for further analysis because their frequency in the document is the same as the frequency of a longer term that exactly contains them (\"d7.49\" is contained in \"d7.49 318\" and their frequencies are the same). This indicates that the shorter term never occurs alone in the document. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Evaluation results for the NCB14 dataset. ::: Textractor NLProt Union ---------------------- ---------------- --------------- --------------- ----------------- ------------- --------------- --------------- ----------------- ------------- PubMedID Correct Incorrect Precision Rel. Recall Correct Incorrect Precision Rel. Recall Correct 12629548 41 19 68% 67% 37 77 32% 59% 63 12629549 38 3 93% 84% 32 79 29% 40% 50 12640462 23 17 58% 100% 1 13 7% 4% 23 12640463 18 17 51% 36% 46 77 37% 87% 53 12640464 67 18 79% 72% 50 26 66% 50% 100 12669071 14 8 64% 54% 19 35 35% 73% 26 12669072 14 3 82% 65% 18 27 40% 51% 26 12669073 27 1 96% 81% 25 31 45% 53% 39 12669075 19 7 73% 73% 16 26 38% 47% 26 12669077 19 5 79% 73% 16 38 30% 39% 28 12669082 86 26 77% 68% 82 87 49% 65% 126 12669083 64 13 83% 58% 81 74 52% 72% 112 12679784 25 7 78% 58% 26 37 41% 60% 43 12692559 54 26 68% 60% 67 77 47% 55% 99 micro-evaluation 509 170 75% 63% 516 704 42% 63% 814 macro-evaluation 75% 67% 39% 54% ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Evaluation results for the PMC15 data set. ::: Evaluation Performance Measure YAGI Textractor NLProt ------------------ ------------------------- ---------- ---------------- ------------ micro-evaluation precision 43% 66% 34% micro-evaluation rel. recall 63% 31% 41% micro-evaluation rel F-1 51% 42% 37% macro-evaluation precision 40% 53% 29% macro-evaluation rel. recall 66% 30% 37% macro-evaluation rel F-1 52% 40% 34% ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Composition of the training sets ::: Training lists PG C IK Pr ------------------------------------------------------------------ -------- -------- -------- -------- \# n-grams, n\>=2 304 193 111 254 \# occurrences in articles where the n-gram is most frequent 16,543 10,862 5,853 12,547 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Performance estimators after training. ::: Training sets PG+/C- PG+/IK- PG+/Pr- ------------------- --------------- -------- --------- --------- ξα estimates recall \>= 73.96% 78.62% 58.48% precision \>= 72.82% 75.52% 57.75% error \<= 32.37% 34.61% 47.93% :::	PubMed Central	2024-06-05T03:55:56.449036	2005-4-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090555/", "journal": "BMC Bioinformatics. 2005 Apr 7; 6:88", "authors": [ { "first": "Lei", "last": "Shi" }, { "first": "Fabien", "last": "Campagne" } ] }
PMC1090556	Background ========== Ribonucleic acid (RNA) plays various roles in the cell. Many functions of RNA are attributable to their structural particularities (herein called RNA motifs). RNA motifs have been extensively studied for noncoding RNAs (ncRNAs), such as transfer RNA (tRNA), ribosomal RNA (rRNA), small nuclear RNA (snRNA), small nucleolar RNA (snoRNA), etc. \[[@B1]\]. More recently, small interfering RNA (siRNA) and microRNA (miRNA) have been under intensive studies \[[@B2]\]. Less well characterized are the structures in the un-translated regions (UTRs) of messenger RNAs (mRNAs) \[[@B3]\]. However, biochemical and genetic studies have demonstrated a myriad of functions associated with the UTRs in mRNA metabolism, including RNA translocation, translation, and RNA stability \[[@B4]-[@B6]\]. RNA structure determination via biochemical experiments is laborious and costly. Predictive approaches are valuable in providing guide information for wet lab experiments. RNA structure prediction is usually based on thermodynamics of RNA folding or phylogenetic conservation of base-paired regions. The former uses thermodynamic properties of various RNA local structures, such as base pair stacking, hairpin loop, and bulge, to derive thermodynamically favourable secondary structures. A dynamic programming algorithm is used to find optimal or suboptimal structures. The most well-known tools belonging to this group are MFOLD \[[@B7]\] and RNAFold in the Vienna RNA package \[[@B8],[@B9]\]. Similar tools have been developed in recent years to predict higher order structures, such as pseudoknots \[[@B10]\]. On the other hand, RNA structure prediction using phylogenetic information infers RNA structures based on covariation of base-paired nucleotides \[[@B11]-[@B14]\]. It is generally believed that methods using phylogenetic information are more accurate. However, their performance critically depends on the high quality alignment of a large number of structurally related sequences. Tools that align biosequences (DNA, RNA, protein), such as FASTA and BLAST, are valuable in identifying homologous regions, which can lead to the discovery of functional units, such as protein domains, DNA ciselements, etc. \[[@B15],[@B16]\]. However, their success is more evident in the study of DNAs and proteins than of RNAs. This is mainly because the sequence similarity among DNAs and proteins can usually faithfully reflect their functional relationship, whereas additional structure information is needed to study the functional conservation among RNAs. Therefore, it is necessary to take into account both structural and sequential information in comparing RNA sequences. Several tools are available that carry out RNA alignment and folding at the same time (Table [1](#T1){ref-type="table"}). The pioneer work by Sankoff \[[@B17]\] involves simultaneous folding and aligning of two RNA sequences, and has huge time and space complexity (Table [1](#T1){ref-type="table"}). FOLDALIGN \[[@B18]\] improves the Sankoff\'s method by (1) scoring the structure solely based on the number of base pairs, instead of the stacking energies; and (2) disallowing branch structures (junctions). Dynalign \[[@B19]\] reduces the time complexity by restricting the maximum distance allowed between aligned nucleotides in two structures. By taking into account local similarity, stem energy and covariations, Perriquet et al. \[[@B20]\] proposed CARNAC for pairwise folding of RNA sequences. Ji et al. \[[@B21]\] applied a graph-theoretical approach, called comRNA, to detect the common RNA secondary structure motifs from a group of functionally or evolutionally related RNA sequences. One noticeable advantage of comRNA is its capability to detect pseudoknot structures. In addition, algorithms using derivative-free optimization techniques, such as genetic algorithms and simulated annealing, have been proposed to increase the accuracy in structure-based RNA alignment \[[@B22]-[@B24]\]. For example, Notredame et al. \[[@B22]\] presented RAGA to conduct alignment of two homologous RNA sequences when the secondary structure of one of them was known. As shown in Table [1](#T1){ref-type="table"}, most of these methods suffer from high time complexities, making the structure-based RNA alignment tools much less efficient than sequence-based alignment tools. Tools that search for optimal alignment for given structures include RNAdistance \[[@B25]\], rna\_align \[[@B26]\], and RNAforester \[[@B27]\]. RNAdistance uses a tree-based model to coarsely represent RNA secondary structures, and compares RNA structures based on edit distance. In a similar vein, rna\_align \[[@B26]\] models RNA secondary structures by nested and/or crossing arcs that connect bonded nucleotides. With the crossing arcs, rna\_align is able to align two RNA secondary structures, one of which could contain pseudoknots. RNAforester extends the tree model to forest model, which significantly improves both time and space complexities (Table [1](#T1){ref-type="table"}). In addition, methods using Stochastic Context Free Grammars (SCFGs) have been developed to compare two RNA structures. Original SCFG models \[[@B28],[@B29]\] require a prior multiple sequence alignment (with structure annotation) for the training purpose, thus their applicability is limited to RNA types for which structures of a large number of sequences are available, such as snoRNA and tRNA \[[@B28],[@B30]\]. However, Rsearch \[[@B31]\] and stemloc \[[@B32]\], both based on SCFG, are capable of conducting pair-wise structure comparisons with no requirement for pre-alignment. Rsearch uses RIBOSUM substitution matrices derived from ribosomal RNAs to score the matches in single-stranded (ss) and double-stranded (ds) regions. stemloc uses \"fold envelope\" to improve efficiency by confining the search space involved in calculations. The time and space complexities of these two tools are also listed in Table [1](#T1){ref-type="table"}. Furthermore, pattern-based techniques such as RNAmotif, RNAmot and PatSearch \[[@B3],[@B33],[@B34]\] have been used in database searches to detect similar RNA substructures. These tools represent RNA structures by a consensus pattern containing both sequence and structure information. One important advantage of these pattern-based tools is the ability of dealing with pseudoknots. We present here a computationally efficient tool, called RSmatch, capable of both globally and locally aligning two RNA secondary structures. RSmatch does not require any prior knowledge of structures of interest. It can uncover structural similarities by means of direct aligning at the structure level. We demonstrate its application to database search and multiple alignment. We compared RSmatch with three widely used tools, PatSearch \[[@B35]\], stemloc \[[@B32]\] and Rsearch \[[@B31]\], demonstrating that RSmatch is faster or achieves comparable or higher accuracy than the existing tools when applied to a number of known RNA structures, including simple stem-loops and complex structures containing junctions. Implementation ============== Secondary structure decomposition --------------------------------- RSmatch models RNAs by a structure decomposition scheme similar to the loop model commonly used in the algorithms for RNA structure prediction \[[@B36],[@B37]\]. With this model, pseudoknots are not allowed. Our method differs from the loop decomposition methods in that it completely decomposes an RNA secondary structure into units called circles (Figure [1A](#F1){ref-type="fig"}). When the secondary structure is depicted on a plane, a circle is defined as a set of nucleotides that are reachable from one another without crossing any base pair. As shown in Figure [1A](#F1){ref-type="fig"}, all circles are closed or ended by a base pair except the first circle (circle one in the Figure [1A](#F1){ref-type="fig"}), which always contains the 5\'-most and the 3\'-most bases. Various types of RNA structures, such as bulge, loop, and junction can be represented by circles, as shown in Figure [1A](#F1){ref-type="fig"}. Circles of an RNA structure can be organized as a hierarchical tree according to their relative positions in the secondary structure, where each tree node corresponds to a circle (Figure [1B](#F1){ref-type="fig"}). This tree organization is informative to deduce the structural relationship among circles and reflects the structure particularities of the given RNA secondary structure. If two circles reside on the same lineage (path) in the tree, the circle appearing higher in the tree is called an ancestor of the other, and the latter is a descendent of the former. As a result, in the context of the hierarchical tree, two distinct circles fall into one of the following two categories, in the order of decreasing closeness: (i) the two circles maintain an ancestor/descendent relationship, or (ii) they share a common ancestor in the tree. For example, in Figure [1B](#F1){ref-type="fig"}, circle 2 is an ancestor of circle 5, whereas circle 6 does not have ancestor/descendent relationship with circle 5 since they are not on the same lineage. The double-stranded region or stem of a structure is decomposed into a set of \"degenerated\" circles, each containing only two base pairs. As such, a stem of nbases in length will result in n- 1 consecutive degenerated circles. Since a base pair may have two associated circles; we name one circle \"the parent circle\" and the other \"the child circle\" according to their positions in the hierarchical tree. For example, for the boxed C-G base pair in Figure [1A](#F1){ref-type="fig"}, circle 2 is its parent circle and circle 6 is its child circle. Structure alignment formalization --------------------------------- Given an RNA secondary structure, we consider two types of structure components, single bases and base pairs, in the secondary structure. To integrate both sequence and structure information, we introduce two constraints among the structure components: precedence constraintand hierarchy constraint. The precedence constraint is defined based on the precedence order among structure components and the hierarchy constraint specifies the inter-component relationship in the context of the hierarchical tree described above. The precedence order is determined by the 3\' bases of individual structure components: the one with its 3\' base closer to the RNA sequence\'s 5\'-end precedes the other. For example, in Figure [1A](#F1){ref-type="fig"}, the single base component U (marked as the 11^th^nucleotide) in circle 5 precedes the base pair component C-G (boxed) in circle 6. To capture the inter-component relationship within the hierarchical tree context, we need to map each structure component to a circle in the tree. It is obvious that each single base can be mapped to a unique circle. However, a base pair could be mapped to two alternate circles: one parent circle and one child circle. To resolve this ambiguity, we always require mapping to the parent circle. The inter-component relationship is then reduced to the inter-circle relationship of three types: (i) ancestor/descendent, (ii) common ancestor, and (iii) identical circle. Given two RNA secondary structures Aand B, where A, referred to as the query structure, has mstructure components {A^1^, A^2^, \..., A^m^} and B, referred to as the subject structure, has nstructure components {B^1^, B^2^, \..., B^n^}, the structure alignment between Aand Bis formalized as a conditioned optimization problem based on the above two constraints: given a scoring scheme consisting of two matrices, one for matching two single bases and the other for matching two base pairs, find an optimal alignment between the two sets of structure components such that the aforementioned precedence and hierarchy constraints are preserved for any two matched component pairs (A^i^, B^i^) and (A^j^, B^j^). In other words, the two structure constraints between A^i^and A^j^must be respectively equivalent to that between B^i^and B^j^. This formalization has an implicit biological significance in that a single stranded region in one structure, if not aligned to a gap as a whole, will always align with a single stranded region in the other structure. This alignment requirement is important because single stranded regions are usually treated as functional units in binding to specific proteins. Algorithmic framework --------------------- A dynamic programming algorithm is employed in RSmatch. As with sequence alignment, the structure alignment could be either global or local. The difference lies only in the setup of initialization conditions; the algorithmic framework is the same since both global and local alignments must preserve the two constraints described above. A scoring table is established with its rows/columns corresponding to the structure components of the two given RNA secondary structures. We organize the rows/columns in such a way that the precedence and hierarchy constraints are combined and easy to follow in the course of alignment computation. Specifically, we sort the structure components of each structure according to the precedence order defined above. It is straightforward that this arrangement of rows/columns makes the precedence constraint automatically preserved. However, preservation of the hierarchy constraint is much more complicated and can only be accomplished in the derivative analysis for each cell (entry) in the scoring table. We will discuss the derivation when filling in the scoring table. Each cell of the scoring table represents an intermediate comparison between two partial structurescorresponding to the cell\'s row and column components (either single base or base pair) respectively. The partial structure with respect to a structure component c(single base or base pair) is a set of structure components S~c~such that for any component a∈ S~c~, the following three structure constraints between cand amust be satisfied: (i) aprecedes c; (ii) by the hierarchy constraint, ais not an ancestor of c; and (iii) citself is included in S~c~. Furthermore, since a base pair could appear in two circles, its corresponding partial structure could be divided into two smaller substructures: parent structureand child structure. Formally, given a base pair component c, the parent structureof cis the set of structure components P~c~⊆ S~c~(excluding citself) such that for any component a∈ P~c~, a\'s 3\'-base is always 5\' upstream of c\'s 5\'-base; the child structureof cis the set of structure components L~c~⊆ S~c~(including c) such that for any component a∈ L~c~, a\'s 5\'-base is always 3\' downstream of c\'s 5\'-base. It can be shown that P~c~∪ L~c~= S~c~and P~c~∩ L~c~= φ. Examples of partial structures are given in Figure [1C--1E](#F1){ref-type="fig"}. As shown in Figure [1C](#F1){ref-type="fig"}, for a base pair, its child and parent structures together constitute the whole partial structure for the base pair. As we will see in the following discussions, the concept of a partial structure and its byproducts (parent structure and child structure) form the kernel of our algorithmic framework. We can solve the RNA structure alignment problem progressively by aligning small structures and expanding each of them one structure component at a time until all structure components are covered. Preliminaries ------------- Cells in the scoring table are processed row by row from top to bottom and from left to right within each row. By considering the row/column components, we have three types of cells: (i) a cell corresponding to two single bases; (ii) a cell corresponding to one single base and one base pair; and (iii) a cell corresponding to two base pairs. For (i), each cell stores the score of aligning the partial structures corresponding to the cell\'s row and column components respectively. For (ii) and (iii), we need to consider alignments involving the partial and child structures induced by the base pair components. Notice that the parent structures of the base pair components are excluded. It can be shown that each parent structure P~c~of component ccan always be considered as the partial structure S~x~of some other component x, which means we only need to consider child and partial structures in the alignment computation. Consequently, the above three types of cells have one, two and four alignment scores respectively. A scoring scheme is required to score the match of two structure components. We define the scoring scheme as a function g(a, b) where aand brepresent two structure components that are matched with each other. Another important aspect of the alignment algorithm is to penalize the match involving gap(s). In the course of computation, one structure component (single base or base pair) could match with a gap or a whole small structure (parent or child structure) could match with a large gap. Intuitively, the larger the gap is, the heavier the penalty will be. In our implementation, we set an atomic penalty value, denoted as u, for the smallest gap equivalent to a single base. The penalty value for a large gap is proportional to its size in terms of the number of bases matched with the gap. Let A\* be a small structure in the query RNA structure A and B\* a small structure in the subject RNA structure B. The score obtained by aligning the two structures A\* and B\, denoted as f(A\, B\), is ![](1471-2105-6-89-i1.gif), where Grepresents the total number of gaps in aligning A\ and B\. Initialization -------------- We assume that the row components (a\'s) are from the query RNA structure A and the column components (b\'s) from the subject RNA structure B. We focus on global alignment here; initializations for local alignment can be derived similarly. The initialization conditions deal with the cases where at least one of the structures under alignment is an empty structure φ. This is equivalent to setting up the 0^th^row/column in the scoring table. As discussed above, each base pair component has two small structures to be considered: a child structure and a partial structure. Thus, the aforementioned three types of cells have one, two and four initialization scores respectively. For a given structure component x(single base or base pair), let S~x~represent its partial structure. If xis a base pair, we also use L~x~to represent its child structure. We have f(φ, φ) = 0. Furthermore, for any structure components aand b, f(S~a~, φ) = \\|S~a~\\|·u, f(φ, S~b~) = \\|S~b~\\|·u, if aand bare base pairs, f(L~a~, φ) = \\|L~a~\\|·uand f(φ, L~b~) = \\|L~b~\\|·uwhere \\|·\\| represents the cardinality of the respective set. Filling in the scoring table ---------------------------- The simplest cell type is the one whose row (column, respectively) component is a single base a(single base b, respectively). Let a^p^denote the structure component that precedes aby precedence order established before. Formally, in matching the partial structure S~a~with the partial structure S~b~there are only three possibilities: (i) ais aligned with b; (ii) ais aligned with a gap; and (iii) bis aligned with a gap. Thus the score of matching S~a~with S~b~can be calculated by Equation (1). ![](1471-2105-6-89-i2.gif) The second cell type is the one formed by one single base and one base pair. There are actually two symmetric subtypes where either aor bis a base pair. Since the analysis is identical, we only focus on the former case where ais a base pair. As discussed before, besides the partial structure S~a~we have to consider the child structure L~a~for the base pair a. Thus, for this type of cells, we have to compute two alignment scores. By the principle of dynamic programming, the smaller size problem needs to be solved before the larger size problem. Thus we first find the structure alignment between the child structure L~a~and the partial structure S~b.~There are only two possibilities: (i) the single base component bis aligned with a gap; and (ii) the base pair ais aligned with a gap (see Figure [2A](#F2){ref-type="fig"}). Therefore we have ![](1471-2105-6-89-i3.gif) In aligning the partial structure S~a~with the partial structure S~b,~to preserve precedence and hierarchy constraints simultaneously, there are only three possibilities: (i) the single base bmatches with a gap; (ii) the partial structure S~b~matches with the child structure L~a~; (iii) the partial structure S~b~matches with the parent structure P~a~(see Figure [2B](#F2){ref-type="fig"}). Thus, ![](1471-2105-6-89-i4.gif) For the third cell type, ais a base pair and bis a base pair. We need to compute four alignment scores because each base pair corresponds to two structures: one child structure and one partial structure. While aligning the child structure L~a~with the child structure L~b~, it is clear that ![](1471-2105-6-89-i5.gif) since both aand bare the last components in the respective child structures by precedence order. Equation (5) gives the alignment score between the partial structure S~a~and the child structure L~b~. ![](1471-2105-6-89-i6.gif) The first case corresponds to that bis aligned with a gap. If bdoes not match with a gap, it can be shown that, to preserve both precedence and hierarchy constraints, the second and third cases in Equation (5) cover all possible situations. Similarly, we can calculate the score of aligning the child structure L~a~and the partial structure S~b~as shown in Equation (6). ![](1471-2105-6-89-i7.gif) In aligning the partial structure S~a~with the partial structure S~b~, there are five possibilities: (i) the parent structure P~a~is matched with the parent structure P~b~and the child structure L~a~is matched with the child structure L~b~; (ii) the child structure L~a~is matched with gaps; (iii) the child structure L~b~is matched with gaps; (iv) the parent structure P~a~is matched with gaps; and (v) the parent structure P~b~is matched with gaps. Therefore ![](1471-2105-6-89-i8.gif) Data sets --------- All experiments (unless otherwise specified) were carried out on a Linux system with two 2.4 GHz Intel processors and 3 GB memory. A human UTR structure database was constructed as follows. We downloaded 19,986 human RefSeq mRNA sequences (January 2004 version) from National Center for Biotechnology Information (NCBI). Each RefSeq sequence containing UTR regions, as indicated by RefSeq\'s GenBank annotation, was processed to extract its 5\'UTR and 3\'UTR sequences. For each UTR sequence, we took a 100 nt subsequence at every 50th nucleotide position from 5\' to 3\', making consecutive subsequences overlap with one another on a 50 nt segment. Subsequences shorter than 100 nt, e.g. at the 3\' end, were also kept. Using the Vienna RNA package\'s RNAsubopt function with setting \"-e 0\", we then folded all obtained sequences to form the structure database. For any given RNA sequence, the setting \"-e 0\" resulted in multiple RNA structures all having the minimum free energy. The final database contained \~575,000 RNA secondary structures. The structural patterns of a histone 3\'UTR stem-loop structure (HSL3) and an iron responsive element (IRE) were used in this study, based on their specifications in the UTRdb database \[[@B3]\]. Three tools, PatSearch \[[@B38]\], stemloc \[[@B39]\] and Rsearch \[[@B40]\], were employed for comparison purposes. The efficiency of these tools was measured by CPU running time. The performance of each program was assessed by specificity and sensitivity. Specificity was calculated as TP/(TP + FP) and sensitivity as TP/(TP + FN), where TP was the number of true positives, FP the number of false positives, and FN the number of false negatives. To test the applicability of RSmatch to complex structures, we downloaded RNA families from Rfam \[[@B1]\]. We only chose those families that had more than 10 seed RNAs and its consensus sequence length is no longer than 250 nucleotides. We had 64 families in the final data set. For each family, we randomly selected one member RNA as the query RNA and obtained its structure from Rfam. We then randomly chose 10 subject RNAs in the same family. Here we intentionally introduced noise by extending each subject RNA sequence with its adjacent sequences at both 3\' and 5\' ends to make the total length three times its original one. Results ======= Studies of stem-loop structures in UTRs --------------------------------------- Using our proposed algorithm, we first studied RNA motifs in UTR regions of human mRNA sequences. A well-known fact is that the accuracy and efficiency of RNA folding programs will decrease significantly when the sequences to be folded become very long. Satisfactory performance is usually obtained when the sequences have moderate lengths, i.e. one hundred nucleotides. Thus, we used a moving window scheme to get subsequences of 100 nt and folded them using the Vienna RNA package (see Implementation). In the RSmatch package, this subsequence length is a user-defined parameter. Since the nucleotide conservation in the single-stranded region of an RNA sequence may differ from that in the double-stranded region, we used two scoring matrices, one for substitutions among single bases and the other among base pairs. This type of scoring scheme was also used in other studies \[[@B31],[@B41]\]. Theoretically, the scoring matrix for single bases is a 4 × 4 table for all types of substitutions of single nucleotides, and the one for base pairs is a 16 × 16 table for all types of substitutions of base pairs. However since we used the Vienna RNA package, only six types of base pairs were observed in our studies, i.e. Watson-Crick base pairs A-U, U-A, G-C, C-G, and wobble base pairs G-U and U-G. Values used in the two matrices were empirically chosen so as to conform to the general understanding of the sequence and structure conservation of RNA motifs, as follows. (1) Mutations in the double-stranded region may not be detrimental to RNA\'s function if the mutated sequence still preserves the same secondary structure. Therefore base pair substitutions were rewarded with a positive score, instead of a penalty. (2) A sequence in the single-stranded region may be important for RNA\'s function, such as binding to proteins, and thus mismatches were penalized. To process gaps we used an arbitrary function u× l, where uwas the atomic penalty value for a gap that is one single base long and lis the length of the gap in terms of the number of bases matched with the gap. In our experiments otherwise stated explicitly, the uwas empirically set to -6 and changing the uvalue did not change the qualitative conclusion made in the paper provided that the absolute value of uwas greater than any positive score in the scoring matrices. Users can freely change the uvalue when applying RSmatch to their own data set. We tested our program with a query sequence containing an iron response element (IRE). The IRE motif is a bipartite stem-loop structure containing \~30 nucleotides. Two alternative types of IREs have been found, which differ in the middle region \[[@B3]\]. Type I has a bulge, whereas type II has a small internal loop. IREs have been found in both 5\' and 3\' UTRs of genes that are involved in iron homeostasis in higher eukaryotic species. They interact with iron regulatory proteins (IRPs) and play a role in RNA stability and translation. Using a subsequence in the 3\'UTR of transferrin receptor (NM\_003234) that contains an IRE motif, we searched the UTR structure database described in Implementation. A list of top hits is shown in Figure [3](#F3){ref-type="fig"}. The best hit of the search is the query structure itself, as expected. Other regions of the same mRNA and regions of other RNAs are also found to have homologous structures with the query. As clearly shown in the result, the region containing the IRE motif, which is from about the 30th nucleotide to about the 60th nucleotide of the query structure, has been located by the RSmatch program, indicating that a local optimal alignment has been achieved. Among the top 10 hits, several sequences are known to have IREs, such as several regions in the 3\'UTR of transferrin receptor (NM\_003234) and the 5\'UTR of solute carrier family 40 protein (NM\_014585). Other top hits have not been shown so far to have IREs. It is not known if some of them are novel IRE-containing RNAs and the definitive answer will await wet lab validation. The output shows detailed alignment and related information, including the numbers of bases in the single-stranded and double-stranded regions, and the percentages of identity in single-stranded and double-stranded regions. RSmatch can also accept pattern-based RNA structures (sometimes called descriptors) to search a structure database. Since a pattern-based search method has an intrinsic primitive scoring scheme by using degenerate bases, we used simplified binary matrices as the equivalent to score an alignment. In the matrices, the match of a pair of structure components (single bases or base pairs), including those containing degenerated bases, was given a score of 1, a mismatch was penalized by a score of -1, and the atomic gap penalty uwas set to -3. To allow variability in single-stranded and/or double-stranded regions for a structure pattern, we introduced a wildcard \"n (lower case)\" to represent optional single base component (\"n\") and base pair component (\"n-n\"). The meaning of \"n\" is identical to the IUB code \"N\" except that the matching score for both structure components \"n\" and \"n-n\" is always zero regardless of whether they are aligned with a structure component or a gap. Two RNA motifs were used to test our method, namely a histone 3\'UTR stem-loop structure (HSL3) and IRE. HSL3, which resides in the 3\'UTR region of histone mRNAs, has a typical stem loop structure with two flanking tails (Figure [4A](#F4){ref-type="fig"}). Both the stem and the flanking sequences are important to bind with a stem-loop binding protein (SLBP), which controls the pre-mRNA processing and stability of histone mRNAs \[[@B42]\]. In contrast to the HSL3 motif, IRE is relatively flexible in length and in nucleotide composition in its stem region (Figure [4B](#F4){ref-type="fig"}). We compared our program with PatSearch \[[@B35]\], a widely used tool that searches a sequence database for sequence and structure patterns. Using the HSL3 motif and UTR sequence database, PatSearch found 55 hits whose locations were presented in Table [2](#T2){ref-type="table"}. Among them, one is a false positive (NM\_014372, ring finger protein 11, Table [2](#T2){ref-type="table"}). Therefore the specificity (98.2%) of PatSearch is very high. This is attributable to the precise specification of the HSL3 pattern. However, if a pattern description is too precise, it may lead to the \"overfitting\" problem. This problem prevents the tool from finding slightly divergent structures, thus lowering the tool\'s sensitivity. Indeed, several histone genes were not detected by PatSearch, including two histone genes (histone H4c NM\_003542 and histone H4 NM\_003548) which were found by RSmatch among its top 33 hits. Several other histone genes appeared among the top 184 hits of RSmatch (Table [2](#T2){ref-type="table"}). This indicates that by gaining specificity, PatSearch loses sensitivity for HSL3. Since RSmatch gives a score to each alignment, different cutoffs can be used for selecting top hits (Table [3](#T3){ref-type="table"}). It seems that newly detected true positives are heavily outnumbered by false positives as RSmatch relaxes its cutoff value. However, with some properly chosen cutoff, i.e. 12, RSmatch could still achieve a comparable specificity with PatSearch. One possible explanation of getting high false positives for RSmatch could be that, with respect to the particular case of the HSL3 motif, its secondary structure conformation might be too pervasive in RNA sequences to be used as a discriminative feature. This could point out a problem concerning RSmatch\'s current scoring matrices, which need to be fine tuned to improve the tool\'s specificity. Good tuning could be achieved by setting up the scoring matrices through learning from a training data set. One interesting observation, however, was that RSmatch and PatSearch agreed perfectly upon the HSL3 locations for almost all of the true positives they found. Using the IRE motif, we performed further comparisons between RSmatch and three other tools: PatSearch \[[@B43]\], stemloc \[[@B32]\] and Rsearch \[[@B31]\]. We used default parameters for Rsearch; for stemloc, the fold envelope was set to 1000. Instead of using the large UTR structure database described in Implementation we constructed a small test data set to expedite the comparison process. First, we used PatSearch to search human UTR sequences for IRE motifs. Then for each hit sequence we selected its corresponding mRNA\'s 3\' or 5\' UTR sequence. Following the same folding process as discussed in Implementation, we folded these UTR sequences to form the test data set. Totally, PatSearch found 27 hits, among which 9 were known true positives. Therefore PatSearch\'s specificity was \~33%. These hits were from 23 distinct mRNA sequences. We assumed that PatSearch had a 100% sensitivity. We extracted the 5\' and 3\' UTR sequences from the 23 distinct mRNAs and obtained 46 UTR sequences. We then folded the 46 UTR sequences to get a small test data set, which contained 1196 structures. Using a known IRE-containing structure (NM\_000032), which was one of the 9 true positives found by PatSearch, as the query, we searched the small test data set. Table [4](#T4){ref-type="table"} shows the results we obtained. Since Rsearch accepts sequences only, it was tested using only the primary sequence information in the test data set. Except for the IRE-containing structure NM\_001098, which was one of the 9 true positives found by PatSearch, and the query itself (NM\_000032), all tools agreed on the IRE locations for the other seven true positives without salient discrepancy. It was found that NM\_001098 was not properly folded to exhibit the existence of the IRE motif. RSmatch has the best specificity by ranking all seven true positives within its top 8 hits with only one false positive (NM\_032484). Rsearch is close to RSmatch by ranking all seven true positives within its top 8 hits with one false positive (NM\_003672). In contrast, stemloc gives five false positives within its top 10 hits. Setting different cutoff values yields different specificity and sensitivity for each tool. The point of balanced specificity and sensitivity appears at the cutoff value of 8 for all three tools. With this cutoff value, the specificity of RSmatch and Rsearch tied at 7/8 × 100% = 87.5%. This is better than the specificity of PatSearch (33%) and the specificity of stemloc (\~50%). The sensitivity of RSmatch, Rsearch and stemloc is 87.5%, 87.5% and 50% respectively. It is worth noting that RSmatch runs \~30% faster than Rsearch; it took Rsearch 34 seconds to search the whole data set of 1196 structures while RSmatch used only 23 seconds. Consequently, RSmatch would be suitable for analyzing large data sets. It should also be pointed out that RSmatch permits wildcards in database searching and structure matching, which are not supported by Rsearch or stemloc. Performance on complex structures --------------------------------- We further tested how accurate RSmatch is for complex structures. To this end, we downloaded RNA structures and sequences from the Rfam database (see Implementation). We used 64 RNA structure families, each of which has more than 10 seed sequences and has the consensus sequence length less than 250 nucleotides. For each RNA structure family, we randomly selected a structure and searched against 10 randomly selected sequences belonging to the same family. To reflect real world scenarios, we extended RNA sequences at both 5\' and 3\' ends so that the length of a subject sequence is three times that of the original one. To ensure that the folded structures are long enough to fully contain the structure being investigated, we required the moving window size to be 1.5 times the length of the query RNA sequence. Furthermore, to include suboptimal structures, we used all structures with free energy within 1.5 kcal/mol above the minimum one. Compared with HSL3 and IRE, the 64 query structures we used were much more complex, with average length of \~120 nt and more than 70% of them comprised of nested loops and conjunctions. To assess the accuracy, we used a measure called structure coverage, denoted as p, which is calculated by the following formula: p= \\|Q~align~\\|/max(\\|Q\\|, \\|S~align~\\|), where \\|Q~align~\\| and \\|S~align~\\| are the lengths of aligned portion of query RNA and subject RNA, respectively, and \\|Q\\| is the length of query RNA sequence. As shown in Figure [5A](#F5){ref-type="fig"}, even though Rsearch has slightly more points clustered around high coverage (90--100%), the overall difference between RSmatch and Rsearch is not significant. In addition, the difference between RSmatch and Rsearch do not seem to be related to structure size or complexity. This result indicates that RSmatch has the ability to process complex structures. We also selected 5S rRNA for further detailed tests. 5S rRNA has a length of \~120 nt, which contains several types of RNA structures, including hairpin, internal loop, bulge, and junction. There are 602 sequences in the 5S rRNA family, allowing us to carry out a thorough analysis. We randomly chose one 5S rRNA as query structure and ten others as subject sequences for alignment. This process was repeated 100 times. The performance comparison of Rsearch and RSmatch is shown in Figure [5B](#F5){ref-type="fig"}. For 5S rRNA, RSmatch outperforms Rsearch in discovering the complete structure more frequently. An exemplary alignment is shown in Figure [5C--5E](#F5){ref-type="fig"}. Running efficiency ------------------ By dynamic programming, the running time of computing an alignment equals the number of writing operations needed to fill the scoring table. Thus the time complexity of RSmatch is O(mn), where m(n, respectively) is the number of structure components in the query (subject, respectively) RNA structure. To test the scalability, we downloaded the seed sequences for 5S rRNA family from Rfam and randomly selected one annotated structure as the query while folding the rest sequences to prepare the structure database as discussed above (Figure [6](#F6){ref-type="fig"}). We plotted the RSmatch running time versus the database size. The program was run 10 times and the result is shown in Figure [6](#F6){ref-type="fig"}. The nearly perfect linear growth of the running time gives an empirical proof that the algorithm\'s time complexity is bounded by O(nm). Multiple structure alignment and iterative database search ---------------------------------------------------------- We also extended RSmatch algorithm to conduct multiple structure alignment. An example using IRE is shown in Figure [7](#F7){ref-type="fig"}. While the alignment algorithm is the same, the multiple alignment function uses a position-specific scoring matrix (PSSM, Figure [7C](#F7){ref-type="fig"}). For a given set of structures, the multiple alignment function first identifies the best alignment of two structures, and builds a PSSM. The PSSM is then used to search for the closest structure in the rest of the set. A flowchart of multiple structure alignment is shown in Figure [7A](#F7){ref-type="fig"}. If the alignment score of a structure to the PSSM is above a cutoff (user-defined), it is selected and its structure is used to update the PSSM. This step is iteratively conducted until no structures have alignment score above the cutoff. In a sense, this method employs an implicit guided hierarchical tree using the average value for joining nodes. As an example, from our human UTR database we selected 6 IRE-containing structures and randomly chose other 6 none-IRE structures to form a small dataset and run RSmatch against it. The output is shown in Figure [7B](#F7){ref-type="fig"}. The final result is in Stockholm format for multiple structure alignment. Conceivably, when the given set of structures is a large database, the multiple structure alignment function of RSmatch in effect conducts iterative search for finding similar structures. Discussion and conclusions ========================== The work presented here is intended to provide an efficient tool to directly perform structure alignment and search of RNA secondary structure databases. Its capability to carry out multiple structure alignment and iterative database search can potentially be used to uncover RNA motifs ab initio. For example, one can use an RNA structure of interest to search an RNA structure database, and build PSSM iteratively to build an RNA motif, as demonstrated for IRE in this study (Figure [7](#F7){ref-type="fig"}). RSmatch bears similarities to rna\_align and RNAforester in that the structural particularities are either explicitly captured using hierarchical tree/forest structures or implicitly represented using arc-annotated structures. However, RSmatch differs from rna\_align and RNAforester in two major aspects. First, RSmatch keeps structural consistence by only allowing single bases matched with single bases and base pairs matched with base pairs whereas rna\_align and RNAforest do not impose this restriction. Second, RSmatch keeps the integrity of single-stranded regions by matching one with another, instead of breaking a single-stranded region into pieces and aligning them with different single-stranded regions. In addition, RSmatch has less time and space complexities than the other two tools. The concept of circles introduced in this paper is reminiscent of the \"k-loop\" described in the classic RNA structure prediction paper \[[@B44]\]. The difference is that the circles can reflect the inter-base-pair relationship by focusing on two base-pairs at a time while the \"k-loop\" cannot. By organizing all circles into a hierarchy tree, we can capture the overall structural particularity. It should also be pointed that there is a major difference between the hierarchy tree introduced here and the parse trees of SCFG \[[@B28]\]. The hierarchy tree is constructed from circles and aims to obtain the panorama of the secondary structure of RNA at a higher level than that of the SCFG parse tree, while detailed information is still available within each circle in the tree. With the introduction of partial structures, this two-level structure modeling (intra- and inter- circles) allows us to develop an efficient algorithm that runs at time O(mn)as we have shown in the paper. Our program takes full advantage of structure prediction techniques. It separates RNA folding from structure alignment. Simultaneous RNA folding and alignment is believed to be the optimal solution for both finding the right structure and locating homologous sub-structures of RNAs \[[@B17]\]. Unfortunately, it is computationally prohibitive for even a moderate number of RNAs. Some improvements have been proposed, but extensive computing time still makes them infeasible for database searches \[[@B18],[@B19]\]. By separating the process into two steps, we greatly enhanced the computing efficiency, making it possible to process a large-scale pre-folded RNA structure database for homologous motifs. However, a drawback of using pre-folded RNAs is that the prediction tools may not produce correct RNA structures, as observed in our experiments. It is estimated that the RNA folding programs solely based on thermodynamic properties of RNA can correctly predict RNA structures with about 70% of chance \[[@B45]\]. Secondly, higher complex structures, such as pseudoknots, cannot be predicted in most commonly used programs, including the Vienna RNA package used in this study. A solution to removing the first drawback is to choose suboptimal structures in addition to the optimal one to increase the chance of obtaining correct structures. It has been reported that using suboptimal structures whose thermodynamic free energies are within 2% of that of the optimal one can greatly improve the structure prediction of RNA \[[@B44]\]. In our IRE experiments, we found that the predicted structure for NM\_001098/1--23 did not exhibit the existence of an IRE motif. By relaxing the free energy range, we finally detected the IRE motif from one suboptimal structure whose free energy was 1.7 kcal/mol higher than the optimal one. Because of the computing efficiency of our program, an increase of the number of RNA structures does not impose big burden on database searching (data not shown). The cost will be at the database building stage, which is however done only once. The moving window approach we used to extract and fold subsequences was aimed to make the folding process more accurate and efficient. This is because RNA folding programs are known to have pronounced difficulties in correctly predicting large RNA structures. Furthermore, predicting the structure for a long sequence takes much longer time than predicting structures for its subsequences. Another advantage of using the moving window method is that small motifs falling in the overlapped subsequences could be folded twice, increasing the chance of their being detected. Pattern-based tools, such as PatSearch and RNAmotif, use descriptions of an RNA structure as queries to search a sequence database for similar structures. This type of search does not take into consideration the context of a hit sequence, which could influence the (sub)structure of the sequence. For example, as shown in our experimental results, PatSearch can achieve a satisfactorily high specificity when the structure of a pattern is not flexible and its description is relatively precise, such as the HSL3 motif. However, the sensitivity of PatSearch is low with rigid pattern descriptions. For relatively flexible structures, such as IREs, the specificity of PatSearch drops because it does not take into account the context in which a motif is located. On the other hand, using folded RNA structures, the proposed RSmatch tool overcomes these shortcomings with a high specificity, thus complementing the pattern-based tool. However, as also shown in our experimental results, the error existing in folding an RNA sequence (NM\_001098) can lower the sensitivity of RSmatch. We suspect that the inaccuracy introduced by RNA folding could be a bottleneck for our technique in achieving a very high sensitivity. Our scoring matrices for single-stranded and double-stranded regions and the gap penalty assignment are very primitive in the sense that they are not based on any probabilistic model or learned from any training data set. One interesting observation in our HSL3 experiment was that RSmatch did find most HSL3 sites correctly. However, the scoring scheme seemed not acute enough to filter out many false positives. Part of the problem is that there are not enough motifs that can be used to construct optimal scoring matrices. In fact, we also tested the matrices (RIBOSUM) proposed by Klein and Eddy, which were built upon small subunit ribosomal RNAs. We did not find any discernible difference in our HSL3 experiment, in which both matrices were used (data not shown). Another related question is whether different types of RNA, such as tRNA, rRNA, and UTRs, need their own scoring matrices. It is conceivable that large highly structured RNAs, such as rRNA, may be able to tolerate more mutations than short RNA motifs that occur in UTR regions. If so, using different scoring matrices for different types of RNAs will be warranted. Furthermore, it is possible that the mutation rate is different for nucleotides in different regions of an RNA motif. Therefore, PSSM might be more suitable in these cases. To this end, the iterative search function of RSmatch, which searches a database using PSSM, can be applied. Motivated by the statistical methods of assessing results in sequence alignment \[[@B46]\], we tried to develop scores of our database search with known probabilistic distributions. The score distribution seemed close to be normal (data not shown). However since our scoring scheme is still at its preliminary stage and much is to be learned about the RNA structure database presented in the paper, we only presented search results in terms of ranking. More elaborate statistical assessment of the search results will be developed in the future. Availability and requirements ============================= The RSmatch package has been implemented in Java and Perl and is freely available for academic use at <http://exon.umdnj.edu/software/RSmatch> or <http://aria.njit.edu/rnacenter/RSmatch>. Authors\' contributions ======================= JL, JTW, and BT participate in the design of the algorithm. JL developed the software. JL and BT did the study with various RNA structures. JH tested the software and participated in the study of HSL3. JL, JTW and BT wrote the manuscript. Acknowledgements ================ We thank Kaizhong Zhang and other members of BT lab for helpful discussions. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### RNA structure decomposition (A-B) and Partial structure determination (C-E).(A) A hypothetical RNA secondary structure is decomposed into a set of circles. (B) The circles are organized into a hierarchical tree. As shown, circle 8 contains only one pair of bases that are bonded with each other; therefore it corresponds to a loop. Circle 7 contains two pairs of bases that are bonded with each other respectively and also contains a single base (nucleotide C); therefore circle 7 corresponds to a bulge. Circle 6 corresponds to a stem of length two since it does not contain any single base. Circle 2 contains more than two pairs of bonded bases; therefore it corresponds to a junction. (C) A hypothetical RNA secondary structure is used to illustrate how partial structures are determined. (D) The partial structure for the single base G in boldface is shown. (E) The partial structure for the base pair C-G in boldface consists of two parts, a parent structure and a child structure. The base pair itself is included in the child structure. ::: ![](1471-2105-6-89-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Optimal structure alignment derivation.(A) Structure alignment between the child structure L~a~in the query and the partial structure S~b~in the subject. The substructures enclosed by dashed lines are to be inserted/deleted and the substructures enclosed by solid lines are to be matched. (B) Structure alignment between the partial structure S~a~in the query and the partial structure S~b~in the subject. The substructures enclosed by dashed lines are to be inserted/deleted and the substructures enclosed by solid lines are to be matched. ::: ![](1471-2105-6-89-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Database search with an RNA structure containing an IRE motif.A structure element (from base 3,451 to base 3,550) in the 3\'UTR of human transferrin receptor (NM\_003234) was used as a query to search the UTR structure database. (A) The output from RSmatch showing the top 11 hits. The six columns in the \'\'Hits\'\' section are, from left to right, rank, alignment score, region in the query, name of the hit, region in the hit, and annotation of the hit respectively. (B) A pairwise alignment of the query structure and a hit structure (NM\_003234:3401-3500), which is the region from base 3,401 to base 3,500 of transferrin receptor (NM\_003234). The sequence length is shown after \"Query\" on the first line: a 31 nt long query sequence containing 7 nt in ss region and 24 nt in ds region. Numbers after \"Identity\" on the second line are percentages of identity of secondary structure (100%), and primary sequence (54%). The latter is further decomposed into two numbers indicating the sequence identity in ss region (71%) and ds region (50%) respectively. The number of gaps in the overall alignment is shown after \"Gap\", followed by the number of gaps in ss region and ds region, both shown in parenthesis. The same format is used for nucleotide mismatches. Alignments of both structure and sequence are given, where \"\\|\" indicates identical nucleotides in either ss region or ds region, and \":\" indicates identical secondary structures with different sequences. RNA structures are presented as follows: nested parentheses are used for base pairs and dots are used for nucleotides in ss regions. (C) The RNA structures corresponding to the query and the subject (hit) structure in (B). (D) Scoring matrices and the gap penalty used in the search. T and U are used interchangeably in this study. ::: ![](1471-2105-6-89-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The two pattern-based RNA structures used in this study.(A) Histone 3\'-UTR (HSL3) motif. (B) Iron Response Element (IRE) motif. A wildcard, represented by a lowercase letter n, is allowed to appear in a motif. When matching the motif with an RNA secondary structure, the wildcard in the motif can be instantiated into zero or one nucleotide in the secondary structure at no cost. Wildcards are used in places where the length of a region, either single-stranded or double-stranded, is variable. For example, the 5\' flanking tail of HSL3 can be 4 or 5 nt long, and the lower part of the stem region of IRE can be 2 to 8 nt long. ::: ![](1471-2105-6-89-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Performance comparison of Rsearch and RSmatch and an alignment of two 5S rRNAs.(A) Performance comparison for 64 RNA families. Different colors are applied to represent structures of different sizes. Each point corresponds to one alignment between a query structure and a subject structure. The x-axis is the percent of coverage by Rsearch and y-axis is the percent of coverage by RSmatch. (B) Performance comparison of 5S rRNA. A 5S rRNA was randomly chosen as the query structure and ten others as the subject sequences. The median value of the ten structure coverage values was then calculated. This process was repeated 100 times to generate 100 points for the graph. Therefore, each point represents one particular query structure. An example alignment of two 5S rRNA was shown: (C) the query structure is X07545/505-619; (D) the subject RNA is X02729; and (E) the detailed alignment by RSmatch. ::: ![](1471-2105-6-89-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### CPU time versus database size.From the 5S rRNA family, a randomly picked 5S rRNA was used as the query to search a structure database obtained by folding the rest seed sequences in the family. The program was run 10 times, and the average running time of each time is shown as a circle in the graph. ::: ![](1471-2105-6-89-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Multiple structure alignment and iterative database search.(A) Flowchart of multiple structure alignment and iterative database search. Step (1a) accepts a query structure to start an iterative database search; step (1b) processes a small database for multiple structure alignment; step (2) derives a profile from the seed alignment; step (3) uses the profile to conduct search; and step (4) updates the profile with new alignment. (B) Multiple structure alignment of several IRE structures. (C) PSSM of the multiple alignment of IRE in (B). Each column in the PSSM corresponds to the position of a structure component, either single base or base pair. Position of a single base is represented by the nucleotide number and position of a base pair is represented by two nucleotide numbers connected by a dash. For each column, the scores of individual structure components in that position are listed in rows where \"-\" means not applicable. ::: ![](1471-2105-6-89-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Performance comparison of RNA secondary structure tools ::: Tool Name Running Time Space Requirement Reference ---------------- ---------------------------------------------------------------- ----------------------------------------- ----------- Sankoff^a^ O(N^6^) O(N^4^) \[17\] FOLDALIGN^b^ O(N^4^) O(N^4^) \[18\] RAGA^c^ O(M^2^N^3^) O(M^2^N^2^) \[22\] rna\_align^d^ min{O(MN^3^), O(M^3^N)} O(MN^2^) \[26\] Dynalign^e^ O(M^3^N^3^) O(M^2^N^2^) \[19\] stemloc^f^ O(LM) N/A \[32\] Rsearch^g^ O(M^3^N) O(M^3^) \[31\] RNAforester^h^ O(\\|F~1~\\|\\|F~2~\\|deg(F~1~)deg(F~2~)(deg(F~1~) + deg(F~2~))) O(\\|F~1~\\|\\|F~2~\\|deg(F~1~)deg(F~2~)) \[27\] CARNAC^i^ O(N^6^), O(N^2^) O(N^4^), O(N^2^) \[20\] comRNA^j^ O(M^N^) N/A \[21\] ^a^N is the average sequence length; ^b^N is the average length of a given set of RNAs; ^c^M and N are the lengths of the two given sequences; ^d^M and N are the two sequence lengths; ^e^M is the maximum distance allowed to match two nucleotides and N is the length of the shorter sequence; ^f^L and M are the two RNA sequence lengths; only valid in extreme cases; ^g^M is the query length and N is the subject sequence length; ^h^\\|F~i~\\| is the number of nodes in forest F~i~and deg(F~i~) is the degree of F~i~; ^i^N is the sequence length, theoretical time complexity of O(N^6^) could be significantly reduced to around O*(N^2^) by pre-processing of the sequences, as noted by the authors \[20\]. ^j^M is the maximum number of stems examined and N is the number of total sequences under analysis. The comRNA\'s average run-time can be significantly improved by carefully chosen parameters, as noted by the authors \[21\]. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### HSL3 motifs found by RSmatch and PatSearch^a,b^ ::: RefSeq ID Location by PatSearch^c^ Score of RSmatch Location by RSmatch Annotation ------------ -------------------------- ------------------ --------------------- ----------------------------------- NM\_002105 551--572 16 549--574 Hs H2A histone family, (H2AFX) NM\_003493 454--475 16 452--478 Hs histone 3, H3 (HIST3H3) NM\_003495 342--363 16 341--366 Hs histone 1, H4i (HIST1H4I) NM\_003509 445--466 16 444--469 Hs histone 1, H2ai (HIST1H2AI) NM\_003512 521--542 16 520--542 Hs histone 1, H2ac (HIST1H2AC) NM\_003517 413--434 16 412--437 Hs histone 2, H2ac (HIST2H2AC) NM\_003518 408--429 16 407--432 Hs histone 1, H2bg (HIST1H2BG) NM\_003519 429--450 16 428--450 Hs histone 1, H2bl (HIST1H2BL) NM\_003520 425--446 16 424--449 Hs histone 1, H2bn (HIST1H2BN) NM\_003522 406--427 16 405--427 Hs histone 1, H2bf (HIST1H2BF) NM\_003525 413--434 16 413--434 Hs histone 1, H2bi (HIST1H2BI) NM\_003526 414--435 16 413--435 Hs histone 1, H2bc (HIST1H2BC) NM\_003527 442--463 16 441--466 Hs histone 1, H2bo (HIST1H2BO) NM\_003528 476--497 16 475--500 Hs histone 2, H2be (HIST2H2BE) NM\_003530 443--464 16 442--467 Hs histone 1, H3d (HIST1H3D) NM\_003535 454--475 16 453--478 Hs histone 1, H3j (HIST1H3J) NM\_003537 445--466 16 444--469 Hs histone 1, H3b (HIST1H3B) NM\_003539 343--364 16 342--367 Hs histone 1, H4d (HIST1H4D) NM\_003546 340--361 16 339--364 Hs histone 1, H4l (HIST1H4L) NM\_005320 753--774 16 752--777 Hs histone 1, H1d (HIST1H1D) NM\_005325 733--754 16 732--757 Hs histone 1, H1a (HIST1H1A) NM\_021052 494--515 16 493--516 Hs histone 1, H2ae (HIST1H2AE) NM\_021059 483--504 16 483--504 Hs histone 2, H3c (HIST2H3C) NM\_021062 407--428 16 406--431 Hs histone 1, H2bb (HIST1H2BB) NM\_021063 463--484 16 462--484 Hs histone 1, H2bd (HIST1H2BD) NM\_021064 470--491 16 469--494 Hs histone 1, H2ag (HIST1H2AG) NM\_021066 414--435 16 413--435 Hs histone 1, H2aj (HIST1H2AJ) NM\_021968 331--352 16 330--355 Hs histone 1, H4j (HIST1H4J) NM\_170610 413--434 16 412--437 Hs histone 1, H2ba (HIST1H2BA) NM\_175055 428--449 16 427--450 Hs histone 3, H2bb (HIST3H2BB) NM\_003542 N/A^c^ 14 365--390 Hs histone 1, H4c (HIST1H4C) NM\_003548 N/A 14 371--396 Hs histone 2, H4 (HIST2H4) NM\_021058 457--478 14 455--481 Hs histone 1, H2bj (HIST1H2BJ) NM\_003510 436--457 12 435--456 Hs histone 1, H2ak (HIST1H2AK) NM\_003511 446--467 12 445--466 Hs histone 1, H2al (HIST1H2AL) NM\_003514 463--484 12 462--483 Hs histone 1, H2am (HIST1H2AM) NM\_003516 510--531 12 509--530 Hs histone 2, H2aa (HIST2H2AA) NM\_003523 411--432 12 412--435 Hs histone 1, H2be (HIST1H2BE) NM\_003529 439--460 12 440--462 Hs histone 1, H3a (HIST1H3A) NM\_003536 449--470 12 448--469 Hs histone 1, H3h (HIST1H3H) NM\_005319 709--730 12 708--729 Hs histone 1, H1c (HIST1H1C) NM\_005322 766--787 12 767--787 Hs histone 1, H1b (HIST1H1B) NM\_021018 444--465 12 445--466 Hs histone 1, H3f (HIST1H3F) NM\_175054 389--410 12 388--409 Hs histone 4, H4 (HIST4H4) NM\_175065 425--446 12 424--445 Hs histone 2, H2ab (HIST2H2AB) NM\_033445 472--493 10 471--492 Hs histone 3, H2a (HIST3H2A) NM\_003513 452--473 8 454--476 Hs histone 1, H2ab (HIST1H2AB) NM\_003521 N/A 8 421--441 Hs histone 1, H2bm (HIST1H2BM) NM\_003524 401--422 8 400--420 Hs histone 1, H2bh (HIST1H2BH) NM\_003533 453--474 8 452--472 Hs histone 1, H3i (HIST1H3I) NM\_003534 N/A 8 442--462 Hs histone 1, H3g (HIST1H3G) NM\_003540 N/A 8 348--368 Hs histone 1, H4f (HIST1H4F) NM\_003541 331--352 8 330--350 Hs histone 1, H4k (HIST1H4K) NM\_003543 N/A 8 349--369 Hs histone 1, H4h (HIST1H4H) NM\_003545 N/A 8 352--372 Hs histone 1, H4e (HIST1H4E) NM\_170745 441--462 8 440--460 Hs histone 1, H2aa (HIST1H2AA) NM\_003531 435--456 4 438--459 Hs histone 1, H3c (HIST1H3C) NM\_003532 438--459 4 441--459 Hs histone 1, H3e (HIST1H3E) NM\_005323 701--722 -4 705--721 Hs histone 1, H1t (HIST1H1T) NM\_021065 436--457 -10 314--335 Hs histone 1, H2ad (HIST1H2AD) NM\_005321 761--782 -41 85--116 Hs histone 1, H1e (HIST1H1E) NM\_014372 1345--1366 -42 1381--1389 Hs ring finger protein 11 (RNF11) ^a^Items listed here include those found by PatSearch and those found by RSmatch using cutoff value of 8 that are related to histone genes. ^b^RSmatch gets 33 hits at cutoff value of 14 and gets 184 hits at cutoff value of 8. ^c^mRNAs that are not detected to have the HSL3 motif by PatSearch are marked with \"N/A\". ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Performance of RSmatch in the HSL3 experiment^a^ ::: Cutoff Score Selected Hits^b^ True Positives Specificity Sensitivity^c^ -------------- ------------------ ---------------- ------------- ---------------- 14 33 33 100.0% 53.2% 12 47 45 95.7% 72.6% 10 69 46 66.7% 74.2% 8 184 56 30.4% 90.3% ^a^PatSearch has a specificity of 98.2% and sensitivity of 87.1%. ^b^Hits whose scores are greater than or equal to the cutoff value used in this study are selected. ^c^Assume there are 62 mRNA structures containing the HSL3 motif, which include all histone mRNAs found by RSmatch and PatSearch. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### IRE experiment results ::: True Positive RefSeq ID Location by PatSearch RSmatch Rsearch stemloc --------------- --------------- ----------------------- ------------ --------- --------- ------------ ------- ---- ------------- -------- ---- x NM\_000032^a^ 13--35 \- \- \- \- \- \- \- \- \- x NM\_014585 203--229 202--231 21 1 202--231 34.11 1 202--226 13.021 5 x NM\_003234 3479--3511 3484--3506 19 2 3480--3510 31.42 2 3486--3503 15.936 2 x NM\_003234 3883--3913 3887--3909 17 3 3876--3925 27.80 6 3889--3906 10.914 7 x NM\_003234 3950--3976 3952--3974 17 3 3952--3974 25.50 10 3954--3971 10.476 9 x NM\_003234 3996--4024 3999--4021 17 3 3999--4022 28.53 4 4042--4048 1.149 25 x NM\_000146 19--41 20--40 16 6 7--51 27.84 5 17--41 8.574 12 NM\_032484 2353--2376 2358--2373 13 7 2355--2377 22.26 11 2354--2375 16.411 1 x NM\_003234 3429--3461 3434--3456 13 7 3433--3458 26.40 8 3436--3453 6.218 15 NM\_018992 2182--2205 2186--2202 12 9 2186--2202 18.43 19 2186--2202 11.459 6 NM\_003449 2160--2180 2163--2178 11 10 2160--2180 26.27 9 2161--2181 13.198 4 NM\_002081 3449--3469 3452--3467 11 10 3446--3472 20.47 17 3450--3470 8.290 14 NM\_173649 1371--1398 1431--1446 7 12 1372--1398 18.83 18 1376--1396 8.493 13 NM\_033337 1202--1226 1253--1257 5 13 1202--1227 21.52 14 1202--1227 4.540 18 NM\_001234 1106--1130 1157--1161 5 13 1106--1131 21.52 15 1106--11331 4.540 19 NM\_153706 174--194 108--119 5 13 171--198 17.49 20 219--234 4.400 20 NM\_003607 6892--6914 6851--6854 4 16 6892--6914 21.98 12 6930--6950 10.827 8 NM\_002086 82--102 126--129 4 16 94--117 16.48 22 101--125 2.833 22 NM\_012256 2594--2617 2571--2574 4 16 2536--2570 20.76 16 2590--2606 1.770 24 x NM\_001098 1--23 17--19 3 19 1--23 30.67 3 3--20 14.185 3 NM\_006731 4439--4465 4487--4489 3 19 4442--4462 21.65 13 4443--4460 9.920 10 NM\_003672 2556--2576 2592--2594 3 19 2547--2587 26.44 7 2558--2574 9.049 11 NM\_018234 2038--2058 2176--2178 3 19 2035--2061 14.11 24 2046--2058 4.986 16 NM\_024076 1799--1822 1816--1818 3 19 1800--1821 15.00 23 1832--1850 4.876 17 NM\_000877 3274--3294 3336--3338 3 19 3275--3293 13.81 25 3302--3321 3.182 21 NM\_003675 2--27 27--31 3 19 1--29 16.74 21 21--31 1.980 23 NM\_032323 1924--1944 1990--1992 3 19 1925--1943 12.43 26 1928--1948 0.678 26 ^a^NM\_000032 is used as the query structure for RSmatch, Rsearch, and stemloc. Thus there is no value (shown as \"-\"). :::	PubMed Central	2024-06-05T03:55:56.455038	2005-4-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090556/", "journal": "BMC Bioinformatics. 2005 Apr 7; 6:89", "authors": [ { "first": "Jianghui", "last": "Liu" }, { "first": "Jason TL", "last": "Wang" }, { "first": "Jun", "last": "Hu" }, { "first": "Bin", "last": "Tian" } ] }
PMC1090557	Background ========== In contemporary biological research, organisms are often identified by firstly sequencing one or more of their genes and then comparing the sequences with those of known species, either by inferring phylogenies or by database searches \[[@B1]\]. Once a sequence is available it may be used to design oligonucleotide probes, and these are used for most routine DNA diagnostic work, because probe hybridisation tests are far less expensive and less technically complex than sequence analysis. Specific oligonucleotide probes are used that are complementary in sequence to, and hence hybridise with, selected regions of the DNA, RNA or cDNA of the target species or genotype. Most such routine tests aim to identify specimens of a single species, or only a very few. Each probe is at least 18 nucleotides long and often twice as long, and is chosen so that it is unique and only hybridises with a single target. As a result, at least one specific probe is required for every target, although usually several different probes are used for each. In some tests, sets of species- or genotype-specific probes are deposited as arrays on solid supports, so that it is possible to check simultaneously if an unknown organism belongs to one or other of many different taxa or genotypes; this strategy is widely used for gene expression analysis. \'High density arrays\' of such probes have been used on occasion for identifying pathogens \[[@B2]\], but they are not used routinely because, like sequence analysis, they are costly and technically complex \[[@B3]\], nonetheless the potential market for identifying pathogens in this way is very large (see Discussion). By contrast, in traditional taxonomy, specimens are rarely identified using characters specific for an individual target, but, instead, by using combinations of characters shared by different members of a set of target organisms. In practice the characters are used to devise a series of presence/absence questions arranged as a \'dichotomous taxonomic key\', so that answering these questions sequentially leads to the identification of a specimen. The main advantage of this strategy is that far fewer characters and questions are required to uniquely identify an individual target. The strategy is most efficient when each character bisects the targets into two equal groups, i.e. it is bisectionally distributed, and when different characters bisect the targets differently, ideally in a progressively binary way. In theory the minimum number of characters required to distinguish a finite number of targets by this method is defined by the binary logarithm X = log~2~Y, where X is the number of characters and Y is the number of targets. For example, ten ideal characters would, in theory, identify each of a set of 1024 targets, and only 20 ideal characters could identify more than a million targets; 1013 and 1,048,555 fewer tests respectively than using target-specific characters. Using target-specific characters (i.e. one specific character for each target) is the least efficient strategy for identification when efficiency is measured as the number of characters required to identify a target. Using combinations of shared bisectionally-distributed characters can be much more efficient. The use of such shared characters is most efficient when they separate the targets in a progressively binary way. In this paper we report that gene sequences contain sub-sequences that are present in quasi-randomly distributed sets of around half of the sequences, and hence their presence or absence could be used like the questions in a traditional taxonomic key. These sub-sequences could be detected by sets of probes with complementary sequences. A suitable set of such \'combinatorial probes\' could be used to uniquely identify different individual DNAs as these would give different patterns of hybridisation, \'fingerprints\', with different individual DNAs Sub-sequences that are suitable targets for probe combinations are most commonly 6--30 nts long. Sub-sequences of such lengths are not unique to the set of target genes, and so the target genes must first be separated from other \'contaminating\' DNAs. Various physical or chemical techniques could be used to isolate the target sequences, but perhaps the most convenient would be by PCR using target region-specific primer mixtures (i.e. \'redundant\' primers). Attempts have been made to use algorithms based on suffix trees to find sub-sequences that could be used in combinations to distinguish between gene sequences \[[@B4]\], and others have used selection algorithms based on entropy maximisation \[[@B5],[@B6]\] or on Lagrangian relaxation \[[@B6]\] to optimise probe selection. These studies focussed on the algorithmics of probe selection and demonstrated that sets of sub-sequences 5--8 nts long could distinguish individual sequences. Probes that are only 5 to 8 nts long are not widely used because they usually require unusual hybridisation conditions. In the work reported here we looked at a range of sub-sequence lengths and used a simple greedy algorithm, where sub-sequences were successively chosen that merely maximised the number of pairs of gene-sequences that were distinguished; the algorithm was based on suffix arrays because they use less computer memory than suffix trees to manipulate as large sets. Our study focussed on understanding the effect of gene-sequence diversity on the number and diversity of sub-sequences of different lengths that might be targeted by probes, as these factors will affect their use in practical applications. Here we report a study of three published sets of cytochrome oxidase c subunit 1 (CO-1) genes from representative groups of animal species \[[@B7],[@B8]\]. These data were chosen because each set is consistent in length and composition, but differs greatly from the others in phylogenetic range and diversity. We have also studied, in less detail, several sets of sequences of plant and animal viruses and the ribosomal genes of bacteria. Results ======= Three sets of CO1 sequences were used; the details of which are available via the Internet \[[@B7],[@B8]\]. The \"CO1-animal\" data was from 96 species of animals representing the seven dominant phyla of animals, the \"CO1-insect\" data was from 92 species of insects representing eight of the largest orders of insects and the \"CO1-moth\" data was from 201 species from three superfamilies of moths found near Guelph, Canada. After being aligned using ClustalX \[[@B9]\] with default parameters, the same region was selected from all three datasets for analysis; see Methods. The selected regions, the \"test-sequences\", were 604, 603 and 595 nucleotides long respectively in the three sets. Comparable random test-sequence datasets were constructed. The three sets differed greatly in diversity: the CO1-animal set was the most diverse with an average nucleotide difference, ignoring positions that had gaps for alignment, between all pairs of sequences of 35.2% (S.D. 8.3%) with a range from 12.4% to 57%; the CO1-insect set had a mean nucleotide difference of 22.2% (S.D. 4.2%) with a range from 7.4% to 39.5%, and the CO1-moth set had a mean difference of 12.9% (S.D. 1.9%) with a range from 1.0% to 19.4%. Random sequence datasets were constructed that matched the length and average nucleotide composition of each test-sequence dataset, and had mean nucleotide differences of 72.1% to 73.3% (S.D. 1.8% to 1.9%). The CO1-animal sequences yielded a pool of 112,800 sub-sequences 6 nts long that included all replicates. Pools of sub-sequences up to 31 nts long were also produced and these were, of course, a little smaller. The CO1-insect sequences produced pools that were almost the same size, whereas those from the larger CO1-moth dataset were about twice as large. Distinguishing sub-sequences ---------------------------- It was assumed that two test sequences could be distinguished if one of them contained a sub-sequence and the other did not, even if the second contained a sub-sequence that differed from one in the first at only one position, as hybridisation methods to distinguish such sequences are well established for the assay of single nucleotide polymorphisms \[[@B10],[@B11]\]. To find sub-sequences that could be used like questions in a taxonomic key we searched among those that were shared and bisectionally distributed. We excluded specific sub-sequences, namely sub-sequences that were singletons, and also all sub-sequences found in all the test sequence set. Therefore, the search was confined to \"distinguishing sub-sequences\" (DSSs), namely those that were present in at least two test-sequences but not present in all the test-sequences. Distinguishing sub-sequences (DSSs) constituted, at most, 15% of the sub-sequences in the pools from each CO1 sequence set (Fig. [1](#F1){ref-type="fig"}). Almost all nucleotide combinations up to 6 nucleotides (nts) long were present in all the sequences and, as they were uninformative and therefore eliminated, the percentage of DSSs tended towards zero for lengths less than 6 nts. The proportion of DSSs increased in pools of longer sub-sequences, but the number of singletons also increased with length, and so as sub-sequence length increased the percentage of DSSs in the sub-sequence pool peaked and then decreased. The position of the peak depended on sequence variation; the peak was found at lengths of 8 or 9 nts in pools from the random sequences, and at 9, 10 and 20 nts in those from the CO1-animal, CO1-insect and CO1-moth sequences respectively. Only a few short sub-sequences were repeated within any one sequence and so these had only a minor effect on the size of the DSS pool. Plots of the number of DSSs in each \'occupancy\' category, namely the percentage of test-sequences in which each DSS occurred, showed large variations between the datasets (Fig. [2A](#F2){ref-type="fig"} &[2B](#F2){ref-type="fig"}), and this mostly reflected the diversity of the sequences. Whereas the CO1-moth set yielded DSSs with 50% occupancy over the complete range of lengths tested, the CO1-insect sequences yielded none longer than 17 nts, and the CO1-animal sequences yielded none longer than 7 nts. In general, as the length increased so the number of DSSs in each occupancy category declined at approximately a negatively exponential rate, but there were large variations between the datasets. For all pools, most DSSs were present in fewer than 10% of the sequences and singletons were most common in pools of the longest DSSs, especially from the diverse CO1-animal data. Minimum complete sets (MC-sets) of CO1 sequences ------------------------------------------------ Sets of DSSs that, in combination, would distinguish between test sequences were selected. A set of DSSs that could distinguish all the test-sequences in a dataset, in a manner like a taxonomic key, was considered a \"complete set\". A minimum complete set (MC-set) was defined as a set that contained the fewest DSS found by a random trajectory method (see Methods). Table [1](#T1){ref-type="table"} gives a MC-set for the CO1-moth sequences, and Table [2](#T2){ref-type="table"} gives the \'DSS signatures\', binary barcodes or \'fingerprints\' for some representative moths. MC-sets obtained from twenty searches each of the CO1-animal, CO1-insect and CO1-moth data consisted of only 9, 8 and 11 sub-sequences respectively (Fig. [3](#F3){ref-type="fig"}). In theory, 7 DSSs behaving in a perfectly dichotomous way would be required to distinguish all the sequences in the CO1-animal and CO1-insect data, and the CO1-moth data should require 8 DSSs. Thus, the MC-sets of the shortest DSSs were close to the theoretically predicted size. However, as DSS length and sequence diversity increased, so too did the sizes of the MC-sets. The increase was smoothly curvilinear with the random data, but more variable with real sequences. The more diverse sequences usually required larger MC-sets, although the greater size of the CO1-moth dataset also increased the MC-set size. Each of the DSS pools was shown to contain several independent equally parsimonious MC-sets by successively excluding MC-sets from the pools and searching the depleted pools for new MC-sets. When this was done using the CO1-animal data and DSSs 8 nts long, the first MC-set was of 11 DSSs, but it was not until eight MC-sets had been successively removed that the MC-set size increased to 12. During the removal process the average occupancy of the DSSs in the MC-sets steadily declined from a mean of 40.2% (range 48% to 18%) to 29.7% (range 40% to 3%) When several MC-sets were obtained using the random DSS choice method and compared, it was clear that many DSSs from different MC-sets were interchangeable. Sub-sequence efficiency ----------------------- The relative efficiency of each DSS within a complete set was assessed by calculating the percentage of sequence pairs it distinguished, from among those remaining to be distinguished when it was chosen. In this way, it was found that relative efficiency depended on whether suitable DSSs were available for selection, so whereas the first DSS selected from the CO1-animal sub-sequences 6 nts long was able to distinguish 50% of the sequences, the first DSSs that were 10 nts and 14 nts long only distinguished 41% and 28% of the sequences respectively (Fig. [4A](#F4){ref-type="fig"} &[4B](#F4){ref-type="fig"}). Sequence groups --------------- Our search method also allows groups of the sequences to be defined, so that the resulting MC-sets only contain DSSs that distinguished between members of different groups of sequences, but not necessarily between sequences of the same group. This enabled, for example, the 96 different CO1-animal sequences to be grouped into seven phyla (e.g. Chordata, Annelida, Nematoda, etc) but this only decreased the size of the MC-set for DSSs six nts long from 9 to 8 DSSs, and for DSS 12 nts long from 26 to 23. However grouping was more valuable for sequence sets containing many nearly identical variant sequences. For example a set of sequences from 240 isolates of Potyvirus, a genus of plant viruses, gave MC-sets of 22, 38 and 50 DSSs with sub-sequences 7, 10 and 12 nts long respectively, but when the sequences were grouped as the 62 recognized species the MC-sets were less than half the size; only 10, 14 and 19 DSSs respectively. Speed ----- The search method took 54 seconds to select an MC-set of 16 DSS 20 nts long from the 201 CO1-moth sequences when using one processor on a dual Opteron 242 processor machine running at 1.6 GHz. The same system took 13 seconds to select an MC-set of 17 DSSs 10 nts long from the 96 CO1-animal sequences. These tasks took 8 minutes 14 seconds and 85 seconds respectively in a PC with a Pentium CPU at 2.4 GHz. A version of the program is available for use for research purposes over the Internet, contact the corresponding author (MJG) for details. Discussion ========== All the studies described above in which the three sets of CO1 sequences were compared, illustrate the fact that the number of DSSs in a set of sequences is mostly determined by its diversity and by the length of the sub-sequences being sought. Short sub-sequences for probe targeting could readily be found, but longer sub-sequences that would be more useful for identification in standard hybridisation reactions were less common and more likely to be found among closely related, well conserved, gene sequences. The most useful sub-sequences for identification were, as predicted, those that were present in about half of the targets (i.e. those with occupancy scores of about 50%). Most gene sub-sequences less than 18 nts long are not unique to particular genes. Therefore they can only be used as targets for diagnostic tests when the target nucleic acids that contain them have been preselected in some way. This could be accomplished most conveniently by PCR using region-specific primers or primer mixtures. One advantage of combining region-specific amplification with identification using combinatorial probes is that related but previously unrecognised or uncharacterised species or subtypes may be found. The chosen region, even from unknown species or subtypes, is likely to be amplified using the region-specific primers or primer mixture, and it is then also likely that the combinatorial probes will hybridise with at least some of the target sub-sequences, but will give DSS \'signatures\' that have not been seen before. This is because each MC-set that we have found is many-fold redundant, and has the potential to generate many more different signatures than would be generated from the known test-sequences. For example, the MC-sets 18 nucleotides long that distinguished the 201 CO1-moth sequences were of 16 DSSs. Sixteen DSSs could, if they behaved in a perfectly dichotomous way, uniquely identify 65,536 different gene sequences or species (i.e. 2^16^). Thus the MC-sets we found were 99.7% redundant, and the combinations of DSSs not represented among the target sequences would be available to distinguish previously unknown variants of the selected gene region. The aim of the work reported in this paper was to investigate the factors that influenced the numbers of sub-sequences that, in combination, could distinguish sequences or groups thereof. We therefore tested our selection algorithm using three published sets of CO1 sequences that were consistent in length and composition, but differed greatly from one another in phylogenetic range and diversity. We have also examined, but in less detail, a set of ribosomal RNA genes from 17 bacterial species representing 12 genera and also gene sequences from several groups of animal and plant viruses, namely flaviviruses, orthomyxoviruses, potyviruses and tobamoviruses (unpublished results). The results obtained with the bacterial and viral sequences did not differ in any significant way from those obtained with CO1 sequences, which suggests that there is no a priorireason to believe that DSSs for targeting by probe combinations are not present in all genes. The design of practical diagnostic tests, based on the principles outlined in this paper, would involve several stages. First, known sequences of potential targets would be examined to find regions of convenient length and variability bracketed by conserved sites for PCR primers. The region-specific primers would be tested and optimised using a range of variant sequences. Then all known sequences of the region would be used to identify MC-sets of DSSs, whose complements could be used as probes in hybridisation-based tests to identify individual variants. However an iterative process will be required to design a working set of combinatorial probes as it is well known that a significant proportion of sub-sequences selected as hybridisation probes fail to behave as expected because of secondary structures in the target nucleic acid or the probe \[[@B12]\]. First an initial MC-set would be selected bioinformatically, then tested biochemically, and the probes that performed correctly used as a \'starter set\' for further rounds of bioinformatic and biochemical selection, until a working MC-set was obtained. When this DSS set is used in practice, variant sequences giving unknown DSS signatures are likely to be found. These would then be sequenced and added to the trainer set, and the MC-set might have to be redesigned. The value of target-specific \'high-density microarrays\' of DNA probes was most spectacularly demonstrated when the pathogen causing SARS was shown to be a coronavirus. It was detected using an array of about 10,000 different oligonucleotides from some of the most conserved regions of about 1,000 reference viral genomes \[[@B2],[@B13],[@B14]\]. However, the microarrays used for SARS were not standard diagnostic tools, and high-density microarrays are also not used routinely in infectious disease diagnostics because of their cost and complexity \[[@B3]\]. Nonetheless multiplexing offers clear benefits \[[@B15]\] as more information is provided by each test. At present non-multiplexed tests or tests that use just a few specific probes are the standard. These tests are used routinely for screening donor blood for viruses, including human immunodeficiency lentiviruses and hepatitis C hepaciviruses, and as the primary or confirmatory diagnostic tests for sexually transmitted pathogens and pathogens that cause meningitis \[[@B3],[@B16]-[@B20]\]. These nucleic acid probe-based medical diagnostics have a very large market value \[[@B21]\]. Probes, which identify by being used in combinations, could be most usefully used in low-density DNA microarrays. Low-density microarrays typically comprise fewer than 100 probes and often fewer than 40 probes, and it seems likely that such microarrays could outperform high-density microarrays for routine diagnostic applications because of their reliability, simpler data analysis and much lesser cost \[[@B22]-[@B24]\]. Different combinatorial probe sets could be combined in each low-density array to achieve greater redundancy and accuracy; they might not merely replicate one another but could optimally target different major organism groups or different epidemiologically important strains with each replicate MC-set \[[@B19],[@B25]\]. Conclusion ========== This paper reports a method that finds sub-sequences which, in combinations, distinguish the individual gene sequences or groups of gene sequences from which they came, and that could be used as targets for DNA probes. Sequence diversity and sub-sequence length were found to be the major factors influencing the number of sub-sequences available as probe targets. Methods ======= DATA ---- Three previously described datasets of CO1 sequences \[[@B7],[@B8]\] were used, although certain sequences were not included either because they could not be retrieved from GenBank, or they were incomplete. The CO1-animal data was from 96 species of animals and lacks sequences AF310721, AJ271612, NC\_002767 and AF370851 in the reported set; the CO1-insect data was from 92 species of insects and lacks sequences NC\_003372, AY165779, AF146683, AB010925, NC\_001566, NC\_002084, NC\_000857 and NC\_001322, and the CO1-moth data was from 201 species of Lepidoptera. The sequences were aligned using Clustal X \[[@B9]\] with the default parameters, and the region providing the \'test sequences\' was that bounded by the semi-conserved sequences 5\'-GTNGGNACNGCNNT-3\' and 5\'-GGNGGNGGNGAYCC-3\', which are potential gene specific PCR primer sites. Random sequence datasets were constructed that matched the length and average nucleotide composition of each test-sequence dataset. Sets of sub-sequences that could distinguish the test-sequences were found using research programs written in Lahey Fortran 95. Test-sequences were degapped, and then every test-sequence in a dataset was initially converted into a pool of all the possible overlapping sub-sequences of a chosen length that it contained. Pools of sub-sequences of different lengths, ranging from 6 to 31 nucleotides (nts), were analysed separately. The uninformative sub-sequences that were discarded were singletons, replicates and sub-sequences found in all the test-sequences. Sets of DSSs that, in combination, would distinguish between test sequences were selected by a \"greedy algorithm\". First, an array was constructed that recorded the DSSs in each test-sequence. A \"distinguishing array\" was then constructed that recorded for every pair of test-sequences, the DSSs that distinguished the pair. A \"distinguishing-score\" was then calculated for each DSS by summing the number of pairs of test-sequences that it distinguished. The DSS with the largest distinguishing-score was chosen. This DSS and the pairs that it distinguished were then eliminated from the distinguishing array. The process of DSS selection was then repeated either until a set of DSSs had been found that, in combination, distinguished all the test-sequences, or until no DSS could be found that would distinguish the remaining test-sequences. The set of DSSs that could distinguish all the test-sequences, was considered a \"complete set\". The ability of a complete set to distinguish the test-sequences in a dataset was independently confirmed by using a separate program to search the test-sequences for every DSS in the set, and by checking that the resulting pattern of its presence/absence, its \"DSS signature\", was unique. During most searches, the greatest distinguishing score at each step of the search was achieved by more than one DSS, so one was chosen at random from among those with the greatest score at each step of the search for a complete set. This allowed a search to have a random trajectory through a succession of DSS choices, and often produced MC-sets of different sizes for the DSSs of the same length; the smallest were sometimes 3 DSSs smaller than the largest. To aid the discovery of probe sets for different applications, options were included in the programs that permitted: (i) the exclusion of particular DSSs from the minimum set, (ii) inclusion of particular DSSs in the minimum set, (iii) exclusion of DSSs that, as double-stranded DNA, would \'melt\' outside a chosen temperature range \[[@B26],[@B27]\], and (iv) the exclusion of DSSs with runs of more than a defined number of consecutive residues of the same nucleotide. Authors\' contributions ======================= The authors contributed equally to this project. It was devised by MJG and AJG, all contributed equally to its development, JSA did all the programming, and AJG all the data testing. Acknowledgements ================ We thank PANBIO Pty Ltd of Brisbane, Australia for generous support of this project. Provisional Patents (60/226,212, PQ9026/00, PQ9483/00 and WO02/10443) have been lodged. The Australian Research Council funded a small part of MJG\'s work on this project. We thank Roger Brown of the Australian National University for suggesting the use of a greedy algorithm. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The percentage of DSSs of different lengths in the CO1 sequences, and in random sequences of the same length and composition. ::: ![](1471-2105-6-90-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### DSS occupancy; the number (log~10~) of DSSs of different lengths shared by different percentages of the test sequences in: A) the CO1-animal sequences; B) the CO1-moth sequences ::: ![](1471-2105-6-90-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The minimum number of DSSs of different lengths that distinguish all sequences in each of the three CO1 datasets and in datasets of random sequences of the same length, number and average base composition. ::: ![](1471-2105-6-90-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The cumulative and relative percentages of pairs of (A) the CO1-animal sequences and (B) the CO1-moth sequences, distinguished by successively selected DSSs. The \'relative efficiency\' of each DSS is the number of pairs it distinguishes as a percentage of the pairs remaining to be distinguished. ::: ![](1471-2105-6-90-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### A minimum complete (MC) set of DSSs that distinguish 201 CO1-moth sequences; the DSSs 18 nts long have predicted T~m~s in the range 37°--47°C and no consecutive \'runs\' of more than three residues of the same nucleotide. ::: DSS sequence ----- ---------------------- 1 -ATAAAGGTATTTGATCAA- 2 -ATCCTCCAATTATAATAG- 3 -TCAAGAAGAATTGTAGAA- 4 -CTAATTCAGCTCGAATTA- 5 -TCATCTCCAATTAAAGAT- 6 -AAATTAATAGCTCCTAAA- 7 -GGAGGATTTGGAAATTGA- 8 -ATAAATTTGATCATCTCC- 9 -TCGAAATTTAAATACATC- 10 -GCAGGAACAGGATGAACA- 11 -TTTAGCTGGAGCTATTAC- 12 -AACAGATCGAAATTTAAA- 13 -ATTCGAGCAGAATTAGGA- 14 -AATTCTGCTCGAATTAGT- 15 -AAATGCAGTAATCCCTAC- 16 -AGAAGTATTTAAATTACG- ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Species representing various superfamilies of moths, together with the Accession Codes of their CO1 gene sequences and their \'DSS signatures\', namely the presence/absence of the sub-sequences listed in Table 1 in the selected region of their CO1 gene sequences. ::: Species Superfamily Accession Code DSS signature --------------------------- -------------- ---------------- ------------------ Ennomos subsignaria; Geometroidea AF549628 1010110110110000 Lobophora nivigerata; Geometroidea AF549636 0001000000000000 Lomographa vestaliata; Geometroidea AF549637 1010000011001000 Euchaetes egle; Noctuoidea AF549609 1010000111011000 Acronicta morula; Noctuoidea AF549731 1000000001001000 Hypena humuli; Noctuoidea AF549743 0100010000000000 Idia concise; Noctuoidea AF549761 1101000001010000 Orthosia alurina; Noctuoidea AF549703 0111110011000000 Orthosia hibisci; Noctuoidea AF549725 0111100111000000 Zale unilineata; Noctuoidea AF549715 0111000000000000 Gluphisia lintneri; Noctuoidea AF549780 1100001101100001 Ceratomia undulosa; Sphingiodea AF549807 1100010110011000 Smerinthus jamaicensis; Sphingiodea AF549797 0010100100100000 Sphecodina abbottii; Sphingiodea AF549804 1100101010001000 :::	PubMed Central	2024-06-05T03:55:56.463326	2005-4-8	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090557/", "journal": "BMC Bioinformatics. 2005 Apr 8; 6:90", "authors": [ { "first": "Mark J", "last": "Gibbs" }, { "first": "John S", "last": "Armstrong" }, { "first": "Adrian J", "last": "Gibbs" } ] }
PMC1090558	Background ========== Genome sequencing is the first step towards understanding the complex interplay between pathways and networks that determine the biology of living organisms. The next important step in these analyses is to perform genome-wide investigations to identify the functions of individual genes. While hybridization techniques such as DNA-based microarrays can provide insight into groups of genes that potentially operate in common pathways, validation is required before final functional assignment \[[@B1]\]. Furthermore, many genes are regulated in a post-transcriptional manner, thus their function would not be definable by microarrays \[[@B2]\]. Genome-wide screens of mutants created by targeted and random mutagenesis, as well as the method of gene silencing, are particularly powerful for ascribing phenotypes to individual genes and gene families and can potentially validate predictions from sequence and microarray data \[[@B3]-[@B7]\]. In many cases, taking a genome-wide approach to functional gene analysis requires the combined skills and resources of several research groups working with a semi-automated, rapid-throughput pipeline. To facilitate our goal of a comprehensive functional gene analysis in the fungus Magnaporthe grisea, we have developed a platform for high-throughput mutagenesis and phenotypic characterization. Using this platform, we are seeking to elucidate the functions of the approximately 11,000 genes in the thirty-eight megabase genome of this fungus \[[@B8]\]. M. griseais the causal agent of rice blast disease, the most devastating disease of rice worldwide \[[@B9]\]. The economic importance of this pathogen and its genetic tractability make it a model system for understanding fungal biology, as well as plant-pathogen interactions \[[@B10]\]. One of the strategies that we have adopted to determine the functions of individual genes is to create 50,000 M. griseastrains, each carrying a single random mutation within the genome. The mutant strains are generated by introducing a disruption cassette into the fungus, which consists of a DNA fragment that confers resistance to the antibiotic, hygromycin B \[[@B11]\]. Transformed M. griseacells that incorporate the cassette into their chromosomal DNA are then able to grow on media containing the antibiotic. During the process, the cassette will often insert into an open reading frame or regulatory region, resulting in a loss of gene function and thus a biochemical or structural deficiency. Identification and characterization of phenotypic changes in each mutant provides information about the normal biological role(s) of the disrupted gene, whose identity is established by taking advantage of the fact that it has been \"tagged\" by the inserted antibiotic resistance marker \[[@B12],[@B13]\]. Research groups from two universities, University of Arizona (UA) and University of Kentucky (UKY), are cooperating to create the tagged M. grisealines and to characterize any phenotypic changes. The mutant strains are then shipped to North Carolina State University (NCSU), where they are screened for changes in pathogenicity using susceptible rice varieties. Finally, all mutant strains are sent to the Fungal Genetics Stock Center (Kansas City, MO), a fungal strain repository, where they will be archived and made available to the public. The distribution of research efforts and pooling of the resources and data generated dramatically increases the necessity of having a system for each research laboratory to enter and access the information being produced. From creation to final analysis, each mutant is processed through a total of eight barcoded steps and four phenotypic assays resulting in the capture of a dozen individual pieces of data over a period of 3--6 months. The ability to log, process and archive information in an efficient and secure manner is vital to the success of this project. To record data and track these mutants, we have developed a minimal Laboratory Information Management System (LIMS), called PACLIMS ([P]{.underline}henotype [A]{.underline}ssay [C]{.underline}omponent LIMS) that is described in this report. This system was designed to be flexible in order to accommodate the experimental protocol as it evolved. The software fulfills the role of process control by enforcing the steps of our protocol, and reduces laboratory and data entry errors while allowing the data generated at the three universities to be entered from separate locations. Many LIMS are implemented using expensive commercial products or are integrated systems that provide a complete solution and utilization of commercial database systems \[[@B14]-[@B16]\]. A primary goal in the creation of PACLIMS was to design a system that simplified data entry, and was inexpensive yet flexible to allow modification based on user experience. PACLIMS utilizes the freely-available, standards-based SQL, HTML and SSL technologies, and adheres to common web practicesthroughout. Data is entered into PACLIMS by researchers working at each site (Figure [1](#F1){ref-type="fig"}), and the results from assays performed at each university are made available and updated on a daily basis through a publicly-accessible database called MGOS (M. grisea-Oryza sativa)\[[@B17]\]. In this paper, we describe the conception, creation and implementation of the PACLIMS database, as well as the experimental procedure and data it was designed to manage. Within the project website we provide access to a publicly available \'demo\' database, documentation and the PACLIMS software which can be downloaded and modified to suit other researchers\' needs. Implementation ============== The PACLIMS system was implemented with Open Source, freely available software. The server machine runs Red Hat linux (RH), which runs on a large variety of commodity PC hardware. The RH distribution includes most of the software components that are required to construct PACLIMS. The Postgresql relational database system was used for data storage \[[@B18]\]. This allows the utilization of transactions for data integrity, network based access, and supports numerous interface technologies. The Apache web server was employed for the user interface and for interconnecting the database and control programs, via a simple CGI oriented mechanism that follows normal web practices \[[@B19]\]. Implementation was performed using Perl, a common bioinformatics language, allowing the system to be readily modified \[[@B20],[@B21]\]. Distributed operations and client/server web interface ------------------------------------------------------ A centralized, web-based client/server paradigm was chosen to reduce the management burden presented by the system. All server-based processing occurs on a single computer. Web server dependence was minimized by using a simple CGI interface between the server and the PACLIMS control programs. Secure access is ensured by employing the SSL-based HTTPS protocol. Secure user-access and presentation of security credentials occurs through a web browser such as Netscape and Internet Explorer, so that when a user logs into the system, their identity is associated with all subsequent actions. Results ======= PACLIMS is composed of nine modules that facilitate the management of three basic components of this project: barcoding for tracking the progress of mutants through the pipeline, mutant production and initial characterization, and pathogenicity screening (Figure [2](#F2){ref-type="fig"}). The role of PACLIMS in managing these processes is described below. Barcode management ------------------ Due to the high-throughput nature of this project, all stages of mutant processing and analysis are performed in either 24- or 96-well microtiter plate format, with each plate being assigned a barcode. Thus, each mutant is identified by its barcode-assigned plate number and by its coordinates within the plate. The researcher uses a PACLIMS web-link to request sheets of barcode labels, which can be printed locally. To ensure that each plate has a unique identifier, PACLIMS controls the generation of barcode images, so that each barcode is printed only once. The researcher affixes a barcode to each microtiter plate and then scans it into PACLIMS (Figure [3A](#F3){ref-type="fig"}), whereupon the barcode identifier is permanently associated with that plate (Figure [3B](#F3){ref-type="fig"} and [3C](#F3){ref-type="fig"}). By separating barcode label generation and the association of a barcode with a plate, issues such as lost, misapplied and damaged labels, are avoided. If a previously used label is erroneously affixed to a new plate, the system recognizes that the barcode has already been assigned to a previous plate, and instructs the researcher to choose another barcode and re-enter the plate identifier. All copies of the parent plate and the derived (replicate) plates also receive barcodes and are scanned into the database. The barcode of any replicate plate can be re-scanned at any stage, including mutant production or pathogenicity screening, to trace its history back to the corresponding parent plate. Mutant production pipeline -------------------------- The initial stages of mutant production and morphological characterization are performed at UKY and UA. After the creation of mutants and genetic purification each mutant is transferred to a well of a 24-well plate containing complete medium agar plus hygromycin with three cellulose paper disks on the agar surface. This \"parent plate\" marks the entry point for PACLIMS. All subsequent daughter plates can be tracked back to their parent. A barcode is attached to the plate and scanned into PACLIMS, which then directs the user through web forms, in order to record details about the plate\'s contents (Figure [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). The parent plate is incubated for a defined period of time at which point the user collects phenotype data such as growth rate and enters it into the system (Figure [2](#F2){ref-type="fig"}, Module 9). PACLIMS also directs the user to create other copies of the mutants for sporulation and auxotrophy analyses. Permanent stocks are created in triplicate (Figure [2](#F2){ref-type="fig"}, Module 5), with one replicate being retained at the site of origin, one being shipped to NCSU for pathogenicity screening, and the final replicate going to the Fungal Genetics Stock Center (Kansas City, MO) for public request (Figure [2](#F2){ref-type="fig"}, Module 6). PACLIMS is used to direct the creation of these stocks, and to record the receipt of their shipment. All phenotype data generated are recorded in PACLIMS (Figure [2](#F2){ref-type="fig"}, Module 4, Module 9). Pathogenicity screening ----------------------- Upon receipt of mutant plates by NCSU the barcode on the 96-well plate is scanned and PACLIMS logs the plates\' arrival (Figure [3A](#F3){ref-type="fig"}) and provides a screen to \"create\" 24-well plates for \"activation\" of the cultures in the 96-well storage plate (Figure [2](#F2){ref-type="fig"}; Module 7). These 24-well plates are then used to generate conidia for pathogenicity assays and mycelia for DNA extractions. Each of these plates receives a barcode, and when they are scanned into the database, the user is automatically transferred to the corresponding stage of the experimental procedure. Mutants are screened for pathogenicity and each result is recorded in the PACLIMS database. Data entry is facilitated by scanning the barcode for the rack of inoculated plants, at which stage the user is presented a display of data columns set to the default value of wild-type for the individual wells (Figure [4](#F4){ref-type="fig"}). Mutants with aberrant phenotypes are re-tested in a secondary assay to reduce isolation of false positives after being transferred to a new 24-well plate consisting of only reduced pathogenicity mutants by the LIMS. Report generation ----------------- Sufficient reporting functionality is built into the system to support the data entry process. Contextual information is supplied to the user to allow review of the entered information prior to permanently committing it to the database. Robust reporting is provided by third party software such as Microsoft Access database communications protocol or database systems like MGOS by using Postgresql\'s own network communications protocol \[[@B17],[@B18]\]. Separating and relegating reporting to an external component increases the reusability and component nature of the implementation. PACLIMS can be readily modified to account for different research protocols without disrupting the reporting mechanism. Moreover, specialized third party reporting tools provide a ready means of creating custom reports, as need dictates. Availability ============ The current version of PACLIMS is freely available to academic and non-profit users at <http://paclims.sourceforge.net>. Furthermore, the system is modular and readily customized to suit a laboratory\'s specific needs for a high-throughput screen. There is no need for purchasing additional software to use the system. Laboratory personnel who have introductory level experience with Perl can readily adapt the software to different protocols. Please contact <Ralph_Dean@ncsu.edu> for further details. Authors\' contributions ======================= RR, DB, DW and VP coded the software, ND and SD wrote and edited the manuscript, SN, AF, NG, ST, and GP provided testing and feedback, RD, YL, CS, MF, MO, and TM developed the concept and provided guidance. Acknowledgements ================ This project is funded by a grant from the National Science Foundation Plant Genome Program award number DBI \#0115642. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic of data acquisition and transfer to PACLIMS from multiple universities. ::: ![](1471-2105-6-94-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A schematic overview of the flow in information and materials managed by PACLIMS. Module 1:entry point for PACLIMS, accessed by scanning or entering a plate or plant rack\'s barcode; the user is directed to the appropriate web form. Modules 2and 3:displays web forms for data entry on a new, or \"parent\" plate, including fields for information on mutants in each of the 24 wells. Module 4:provides access to a web form for entry of growth rate data that is collected from the parent plate and allows the user to create \"copies\" of the parent plate for assaying other phenotypes. Module 5:guides and documents the transfer of paper disks from four 24-well \"parent\" plates into the four quadrants of three 96-well plates, used for permanent storage and shipping, the latter process being recorded with Module 6. Module 7:records the revival of cultures from permanent storage, specifically the reversion of the 96-well format into four 24-well plates for pathogenicity assays. Module 8:directs the inoculation process for each 24-well plate of spores, beginning with entering a barcode for a rack of plants to be inoculated, and culminating in entering pathological data for each mutant isolate. Module 9:records phenotypic data via four different web forms, each of which records specific phenotypes; module 1 controls the particular data entry form that is accessed. ::: ![](1471-2105-6-94-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### \(A) Front page for barcode scanning into PACLIMS. (B-D) Depending upon which barcode has been scanned, the user will be transferred by Module 1 to the next step in the experimental process. ::: ![](1471-2105-6-94-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Data entry page for pathogenicity screen results. ::: ![](1471-2105-6-94-4) :::	PubMed Central	2024-06-05T03:55:56.466007	2005-4-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090558/", "journal": "BMC Bioinformatics. 2005 Apr 12; 6:94", "authors": [ { "first": "Nicole", "last": "Donofrio" }, { "first": "Ravi", "last": "Rajagopalon" }, { "first": "Douglas", "last": "Brown" }, { "first": "Stephen", "last": "Diener" }, { "first": "Donald", "last": "Windham" }, { "first": "Shelly", "last": "Nolin" }, { "first": "Anna", "last": "Floyd" }, { "first": "Thomas", "last": "Mitchell" }, { "first": "Natalia", "last": "Galadima" }, { "first": "Sara", "last": "Tucker" }, { "first": "Marc J", "last": "Orbach" }, { "first": "Gayatri", "last": "Patel" }, { "first": "Mark", "last": "Farman" }, { "first": "Vishal", "last": "Pampanwar" }, { "first": "Cari", "last": "Soderlund" }, { "first": "Yong-Hwan", "last": "Lee" }, { "first": "Ralph A", "last": "Dean" } ] }
PMC1090559	Background ========== Transcriptional profiling studies can produce data in the form of abundance measurements for genes in samples assigned to one of two classes. A recent exemplar employed cDNA microarrays to assay 6605 clones from normal liver and liver cancer (hepatocellular carcinoma) tissues \[[@B1]\]. Given such two-class high-dimensional data, one analytical task is identifying a \"small\" subset of features able to discriminate between the classes. Tools that solve this problem would accelerate development of novel and/or improved molecular targets for diagnosis, prognosis, and therapy \[[@B2]\]. For example, enunciating genes able to distinguish liver cancer from normal samples could assist investigations into the etiology and treatment of liver cancer. Existing classification and feature selection techniques can be employed to ascertain the cardinality of a feature subset yielding a classifier that generalizes well, i.e., one which makes zero (or few) errors in assigning the class of an unseen data point. Frequently, application of these approaches to a data set results in the definition of one discriminatory subset with tens to hundreds of features and requiring similar numbers of free parameters. This work focuses on subsets smaller than those produced by existing algorithms: all subsets of one-, two-, (and sometimes three-) features that can be separated by a linear surface without error. A multiplicity of error-free linear classifiers constructed from few features could facilitate the creation of cost-effective clinical tests and guide further basic research. Here, an m-feature classifier is defined as a decision surface for m-dimensional data points where the mfeatures are a subset of P a priorifeatures, m≪ P. The potential number of these classifiers is equivalent to choosing m items out P, i.e., ![](1471-2105-6-97-i1.gif). This number increases when different types of decision boundaries are permissible for each value of m. The scope of the problem can be reduced and simplified if only m-feature linearclassifiers (m-LCs) are considered. This restriction of neglecting non-linear decision surfaces is reasonable because hyperplanes can be calculated efficiently, and Support Vector Machines with linear kernels are sufficient for classification problems associated with profiling data (see for example \[[@B3]-[@B6]\]). Recent work by Bo \[[@B7]\] and Kim \[[@B8]\] demonstrate the utility of looking for small feature sets. Bo and Jonassen surveyed a number of classifier discovery methods including linear hyperplanes. They showed that accurate two-gene classifiers exist in real world data sets and that they perform well. They only analyzed 2 data sets, did not report computer runtimes nor consider single genes or gene triples in their analysis. Kim et alemployed a heuristic, Monte Carlo-based strategy to discover 2- and 3-LCs for a real-world, 3226-dimensional, two-class transcriptional profiling data set \[[@B8]\]. This sophisticated method computes noise tolerant hyperplanes using an analytic spherical model. However, 140 hours on a supercomputer cluster were required to identify at least 11 pairs of genes, each of which separates the data. Thus, although brute force exhaustive search provides a comprehensive and systematic method for finding all small discriminatory feature subsets, the strategy is expensive computationally and largely untenable for m≥ 3 (the problem size grows combinatorially). The high dimensionality of transcriptional profiles and the logistical issues associated with exhaustive enumeration of all 1-, 2- and 3-LCs have lead to the prevailing assumption that such searches are both too expensive and unlikely to be informative. Thus, while some recent studies have made use of Kim\'s method \[[@B9]-[@B12]\], most profiling studies neither consider nor report small sized discriminatory feature subsets. Here, a relatively inexpensive method for calculating maximal margin hyperplanes, LIKNON \[[@B6],[@B13]\], is utilized to rapidly find error-free m-LCs in ten published transcriptional profiling data sets that assayed samples from liver, human breast, ovary, lung, skin, gastrointestinal tract, bone marrow, brain, and prostate. The number of free parameters in this method is m+ 1, hence is very small relative to the size of the data, greatly reducing the problem of over-fitting to the training data (see Random Data section). It seems plausible that the existence of single genes and gene pairs with the ability to form perfect linear classifiers may be a widespread phenomenon. To demonstrate the biological utility of the strategy, the gene pairs and triples discovered in the aforementioned LiverCancer data set were examined and found to yield new and unanticipated scientific insights. Overall, the results indicate the importance of ascertaining, as a matter of routine, the presence (or absence) of small distinguishing feature subsets. Results ======= All results are available through the web site \[[@B14]\]. The 10 published cancer data sets examined here are listed in Table [1](#T1){ref-type="table"}. They range in size from small (few genes and/or a small class) to large (many genes with large classes), using a variety of microarray technologies. For each data set, all single and pairs of genes were tested (for some data sets gene triples were tested) using the LIKNON technique. All gene sets that formed zero training error classifiers were saved and are available via the web site. Table [2](#T2){ref-type="table"} lists the number of such gene sets. Many data sets have single genes or pairs that form such perfect linear classifiers which is interesting as most original reports did not note their presence. Computer runtimes are given in Table [3](#T3){ref-type="table"}. A rough time estimate is 1 second per million pairs per sample. Evaluating all pairs requires checking about 0.5 \* n^2^pairs (a triangular matrix). So a 2000 gene, 30 sample set would need about ![](1471-2105-6-97-i2.gif) seconds. As expected, small data sets are found to have many thousands of gene pairs while large data sets have few. The gene sets discovered in the two large data sets (LiverCancer and YeohALL) are likely biologically relevant and are discussed later. The thousands of pairs found on small data sets are mostly due to the small sample size of the data, and would likely not maintain their perfect classification upon addition of further patient samples. Generalization performance was estimated with a LOO methodology (see Methods) and is often reasonably good. However, even gene sets found in small data sets can be interesting depending on the end use. For use in a medical diagnostic it is desirable that the gene set be highly accurate on large number of patient samples so the test result error rate is low. If used to guide basic research, even a poor error rate could indicate a productive research direction. The results are easily visualized for single and pairs of genes by a scatter plot. For each sample the expression values for the gene(s) set the x-y position and the point is labeled with the sample class as shown in the example plot Figure [1](#F1){ref-type="fig"}. The separating plane is drawn between the two clouds of points and one can see the amount of separation between the two classes (which is twice the \"margin\") a particular set of genes pair yields. A larger margin means the data are more well separated hence is more resilient to noise and more likely biologically relevant. Discussion ========== Many of the genes sets found accurately classify the data with large separation between the classes. It is exciting to consider the possibilities for medical diagnostics if some small gene set is found to accurately and reproducibly indicate a disease state. While general machine learning principles suggest that having more features (genes in this case) is desirable in order to make more noise resistant classifiers, this is data dependent and gene pairs with a training error rate of 1/90 as found in the LiverCancer data could be perfectly acceptable. Small gene sets, even if not accurate enough for medical purposes, can indicate fruitful new research directions. Like other classification methods that produce large numbers of genes, considering the corpus of all genes found can provide insight to the underlying biology. Experimental design and construction of a data set profoundly influences the presence of small sized classifiers. For example, the BRCA sets are labeled according to their known BRCA1/BRCA2 mutation status. If some of the measured genes reflect this mutation status we would apriori expect to find some (possibly many) small feature sets, and not finding any could indicate errors in the class labels (a sample is mislabeled). In data where the class sizes are small, most of the small gene sets will prove sensitive to noise. Thus they are not likely to perform well as predictors on new samples, or in different experiment setups. Classifiers made from large data sets are more likely to be reproducible and perform well in other situations. Classifiers with large margin and zero LOO error are more likely to indicate real biological effects that would hold true on new patient samples. The GIST and BreastBRCA experiments are examples of both of the above conditions, and both lead to very large numbers of pairs. Both have small class sizes and have pre-disposed differences between the classes. The GIST experiment compares cancers from different tissue types which means there will be a very strong signal from just the tissue differences rather than just the cancers alone. The BreastBRCA experiment has the pre-disposition of being split along BRCA status lines. In both cases the number of patient samples and class sizes is small and 100,000+ pairs are found. We suspect there are some biologically real pairs hidden in the large background noise due to small sample sizes. Random data tests (see Methods) indicate that 30+ samples with more even class sizes are needed in order to reduce the random chance noise to a very small level (Table [4](#T4){ref-type="table"}). Gene sets that perform well across independent experimental data sets also likely indicate real biological effects. However, cross-experiment array comparisons are difficult and would be much easier and more broad if experimenters used more common clones and references. Is the cost of such computation worth it? Certainly it is for single genes and pairs. Modern computers are powerful enough to solve these size problems in a few minutes. Triples needs tens to hundreds of hours on a single computer, which is still tractable. Quadruples are beyond single computer tractability. However, this type of algorithm is trivially parallelized over a standard network of computers leading to linear speed up. Each computer would be instructed to examine a given part of the search space and thereafter be independent from all the rest. A super computer or dedicated computer cluster is not required. It seems possible that the occurrence of such small sized classifiers is a common characteristic of microarray data, thus making the effort of searching for small gene sets worth the computational cost. The technique is not restricted to RNA/DNA transcript microarray data as used here. It can be applied to protein microarray, mass-spectra, or any data with similar characteristics. Data set discussion ------------------- We can\'t discuss all the gene sets found in all data sets: there are too many. Here we discuss results from the 2 large data sets that we think produce highly biologically relevant results. The supplement \[see [Additional file 1](#S1){ref-type="supplementary-material"}\] contains further discussion and the web site \[[@B14]\] provides access to all the results and plots. Liver cancer ------------ The large liver cancer data set contains 2 classes (tumor and normal) with 181 patient samples and measurements for 6605 genes. We examined this data set in more detail than the others as it is large and any results found are likely to be biologically relevant. The original data \[[@B1]\] was re-normalized using the Intensity/local then Spatial/local methods as implemented in the BioConductor R package \[[@B15]\] (this is the best performing method as outlined in, Wei Wu, unpublished 2004). With the original data example labelings (105 tumor, 76 normal), there are no pairs found. However, normal LIKNON \[[@B6]\] discovers a 23 gene classifier with 2 of the tumor examples strongly mis-classified (patient samples 108 and 109 in the raw data table from \[[@B1]\]). This suggests these 2 samples might have some sort of problem with them (contamination with too much normal tissue, a very different type of cancer, different tumor stage, etc) or are simply mislabeled. Relabeling these 2 tumor samples to be normal results in 4 gene pairs being found. Going further we wished to look for gene triples, yet using all 6605 genes would lead to excessive runtimes, so we applied a variation filter to reduce the number of genes. This variation filter (requiring a variation of at least 3.6 in log~2~values for each gene) reduced the number of genes examined from 6605 down to 1956, which yielded 43 gene triples (using the original labelings) and 9496 gene triples (with the 2 \"outlier\" genes relabeled). Only 291 of these triples were saved for further analysis. All 4 gene pairs, 35 of the 43 and 229 of the 291 triples include the recently annotated gene PLAC8 (IMAGE:491644) a \"placenta specific gene of unknown function\". The 4 genes in pairs with PLAC8 are • IMAGE:669379 GLCCI1 glucocorticoid induced transcript 1 • IMAGE:590591 ADCY6 adenylate cyclase 6 • IMAGE:260259 Transcribed sequence with moderate similarity to protein sp:P39188 (H.sapiens) • IMAGE:756490 BCAT2 branched chain aminotransferase 2, mitochondrial Figure [2](#F2){ref-type="fig"} shows the plot of the pair PLAC8 and BCAT2. There is good separation between the tumor and normal samples, except for the 2 \"outlier\" examples. Most of the triples found when the \"outlier\" examples are relabeled have larger margins than those found with the original data labels. The top triple with the original labeling is • IMAGE:1472735 MT1E metallothionein 1E (functional) Hs74170 • IMAGE:784593 ARHE ras homolog gene family, member E Hs6838 • IMAGE:878564 GPC3 glypican 3 Hs119651 and is shown in Figure [3](#F3){ref-type="fig"}. GPC3 is a recently noted HCC cancer marker \[[@B16]\], where it was elevated in 6 out of 7 patient samples. Here this gene triple makes no errors on all 181 patient samples. The top triple using the relabeled examples is • IMAGE:78353 RNAHP RNA helicase-related protein Hs8765 • IMAGE:491644 PLAC8 • IMAGE:667883 PHLDA1 pleckstrin homology-like domain, family A, member 1 Hs82101 In all, the 291 triples make use of 168 of the genes. The fact that triples exist when using the original labelings argues that the \"outlier\" examples are correctly labeled as tumor, albeit maybe a different type of tumor (or in a different development stage). That these small gene sets exist in such a large data set argues that they are biologically relevant. These 4 gene pairs only make 2 errors out of the 181 patient samples (the two \"outlier\" samples are the errors), which is an error rate of 2/181 = l/90. The triples found with the original labelings make no errors. Based on such data, one can imagine a simple few-gene diagnostic test based on these pairs and triples. It is perhaps not surprising that a placental gene is associated with liver cancer. Both are blood organs, and cancers often recapitulate early development stages, of which the fast growing placenta might be an example. In addition, mitochondria related genes have been associated with cancer progression. For this data, normal LIKNON was a useful aid in identifying outlying data examples. The two outlying tumor samples in this data set could represent a rare tumor type or development state. Using such aids duringthe experiments would allow such samples to be identified in a timely manner for further investigation. ### Modeling of PLAC8 PLAC8 is noted to have a match to the PFAM model pfam04749.5 DUF614. A search of the PDB database using a SAM HMM model \[[@B17]\] created from the PFAM alignment finds a hit to 1P59 (gi\\|34811270\\|pdb\\|1P59\\|A) which is a Non-Covalent Endonuclease Iii-Dna Complex from bacillus stearothermophilus. The PFAM model locates to the C terminus of the 1P59 crystal structure. The sequence of 1P59 is some 85 amino acids longer than PLAC8, and the alignment hit aligns the last 110 amino acids or so. Visualizing the 3D structure of 1P59 with the alignment hit in PLAC8 colored silver in the RASMOL tool (Figure [4](#F4){ref-type="fig"}), shows that this hit forms a distinct mostly helical domain at the C terminus. YeohALL ------- The YeohALL data set is a large multi-class pediatric acute lymphoblastic leukemia cancer set. The original data contains 7 classes, we use only 6 of them (we do not use their \"other\" class). For each of the 6 classes we compare each class against all others combined, thus asking the question can each class be distinguished from all the others. There are 248 patient samples and the 6 classes are T, E2A, BCR, TEL, MLL and Hyperdiploid \> 50. Both the T vs the rest and E2A vs the rest splittings have single genes that perfectly separate T or E2A from the others. Five out of the 6 splittings have gene pairs, only Hyperdiploid vs the rest does not have any small gene sets. We applied a variation filter to reduce the original 12625 Affymetrix probes down to 4196 genes (the filter level used requires a variation of at least 10000 within each gene). We only discuss the top results for the T and E2A splittings here, see the supplement \[see [Additional file 1](#S1){ref-type="supplementary-material"}\] for more details. Our results reinforce many of the findings in Yeoh\'s work that there are single genes that accurately classify the T and E2A leukemia subtypes and extends it by identifying accurate 2 gene classifiers. ### T vs the rest The only single gene separating T vs the rest is the same one identified by Yeoh, \"38319\_at CD3D antigen, delta polypeptide (TiT3 complex) Hs95327\". This gene has a high value in T and a low value otherwise (Figure [5](#F5){ref-type="fig"}) and clearly separates the data. There are 1169 pairs making use of 681 different genes, the top pair is \"37039\_at HLA-DRA major histocompatibility complex, class II, DR alpha Hs409805\", and \"1105\_s\_at M12886 HUMTCBYY Human T-cell receptor active beta-chain mRNA\" (Figure [6](#F6){ref-type="fig"}). In T, HUMTCBYY is generally high and HLA-DRA is low. These and many other of the genes in the top pairs are identified by Yeoh as being significantly differentially expressed in T. However they did not identify any gene pairs as accurate classifiers. ### E2A vs the rest The 4 single probe sets for E2A are • \"33355\_at PBX1 pre-B-cell leukemia transcription factor 1 Hs408222\" • \"1287\_at ADPRT ADP-ribosyltransferase (NAD+; poly (ADP-ribose) polymerase) Hs177766\" • \"430\_at NP nucleoside phosphorylase Hs75514\" • \"32063\_at PBX1 pre-B-cell leukemia transcription factor 1 Hs408222\" (note that there are two probes for PBX1 both giving the same result, so there are really only 3 genes) All of these are high in E2A and lower in the rest. Probe 33355\_at for PBX1 is shown in Figure [7](#F7){ref-type="fig"} where it clearly separates the classes and was the only single gene identified in Yeoh\'s original work. The other 3 probes barely separate the data and likely failed Yeoh\'s stringent cross-validation criteria. These 4 probes are in Yeoh\'s significantly differentially expressed list. There are 386 pairs, the best pair is \"35125\_at RPS6 ribosomal protein S6 Hs408073\" and \"35974\_at LRMP lymphoid-restricted membrane protein Hs124922\". Figure [8](#F8){ref-type="fig"} shows that this pair separates the data well. LRMP is identified in Yeoh\'s lists (but not RPS6), as are many of the genes in the other top pairs. These and the other top pairs often highly separate the data indicating they might be biologically relevant and resilient to noise. These and all the rest of the results are available through the web site. Conclusion ========== Small sets of genes (single genes, pairs of genes, and triples of genes) able to accurately classify two-class microarray data occur in many real-world data sets. These small sets could portend simple medical diagnostics and point to important research targets. The many small sized gene sets found here were not noted previously in the literature, and seemingly went unnoticed. Many members of the pairs discovered here have known associations with cancer and indicate the possibility of simple, accurate medical diagnostic tests based on such results. Exhaustively examining all pairs in thousand gene size datasets is easily tractable on modern computer hardware. All triples is harder, needing a few days, but this is only computer time, and powerful computers are cheap and plentiful. Given that the compute time is small enough and the results possibly important, we conclude that examining microarray data for single genes, gene pairs and maybe gene triples should be done routinely. When performed along with acquiring the biological data, the results can be used as a quality check on the experimental process. The gene of unknown function, PLAC8 appears to have a role in Liver cancer, and based on our HMM modeling might have a domain similar to part of the crystal structure of 1P59. We find that there are 4 genes that when paired with PLAC8 form a classifier with the low error rate of 1/90. In addition there are many gene triples, often including PLAC8, that form zero error classifiers and might be good biomarkers. We find the previously identified HCC biomarker gene, glypican 3 (GPC3), is part of an accurate gene triple involving MT1E and ARHE. We also find small gene sets able to accurately distinguish leukemia subtypes in the large pediatric acute lymphoblastic leukemia cancer set of Yeoh et al. Methods ======= Transcriptional profiling data sets ----------------------------------- The existing transcriptional profiling data sets investigated here are summarized in Table [1](#T1){ref-type="table"}. We generally chose to minimally process the data. cDNA microarray data were logtransformed; Affymetrix data were used as is. Only the LiverCancer data set was subjected to advanced re-normalization procedures. The one data set with a few missing values (LungStanford) had them set to the appropriate class average value. If there were too many genes (operationally defined as more than about 5000), a variation filter was used to remove genes that didn\'t vary enough across all samples. The filterings (if used) were applied once, before LIKNON analysis, solely to reduce the number of genes and hence the runtimes. They were not used to adjust for \"good\" results. When three or more categories of samples had been defined by the original authors, two-class data sets were produced by partitioning these categories. Table [1](#T1){ref-type="table"} provides statistics for the final data matrices used as input to LIKNON for determining m-LCs. m-feature linear classifier (m-LC) -------------------------------------- Consider a two-class data set composed of Ndata points, ![](1471-2105-6-97-i3.gif). Each data point is a P-dimensional vector of features, x~n~∈ ℝ^P^, assigned to one of two classes, y~n~∈ {+1, -1}. Assume that the classes are linearly separable. A classifier for such data is a hyperplane parameterized by a weight vector, w∈ ℝ^p^, and an offset from the origin, b∈ ℝ. A hyperplane, ![](1471-2105-6-97-i4.gif)(w, b), can be used to predict the class of a data point x∈ ℝ^P^by computing sign(w^T^x+ b). If this value is positive, xis identified with the +1 class, otherwise it belongs to the -1 class. Data points that define the hyperplane, positive half-space (+1 class), and negative half-space (-1 class) are the sets {x\\|w^T^x= b}, {x\\|w^T^x\>b}, and {x\\|w^T^x\<b} respectively. For two-class profiling data, an error free m-LC is a maximal margin hyperplane based on mof these Pfeatures, which assigns the class of every data point correctly. The number of free parameters in such models is m+ 1, so in this setting is quite low (2--4), which is much smaller than the number of samples (for the smallest data sets is 14, on the largest data this is 248). By optimizing of the choice of mthis might be over fitting the data, but in the Random Data section we show that we are finding many more features than chance alone would account for. Given data points specified by Pfeatures, the potential number of m-LCs is equivalent to the combinatorial problem of choosing mitems out of P, i.e., ![](1471-2105-6-97-i1.gif). In this work all single genes (m= 1) and pairs (m= 2) and in some cases triples (m= 3) were evaluated using a linear sparse hyperplane method that has been described elsewhere (LIKNON) \[[@B6]\]. LIKNON determines a maximal margin hyperplane (for example in Figure [1](#F1){ref-type="fig"}) that separates the data classes. Maximal margin means that the hyperplane is positioned halfway between the two classes. Gene sets that linearly separate the data are recorded, otherwise they are rejected. If a single gene is a classifier, it is not used during the pair checks as it would always form a classifier with any other gene. Acceptance, generalization and error performance ------------------------------------------------ This work accepts only perfect (no training error) linear hyperplane classifiers from the above method. This criteria for accepting only perfect classification is very stringent. We first thought that this would lead to few result sets being found, and that allowing non-perfect classification would need be done to find more classifiers. But this loosening turned out not to be necessary. These models have only 2--4 free parameters. Thus the problem of over-fitting the model to the data during training is not a large issue and we don\'t perform any stringent generalization tests such as multi-round cross validation. Others interested in evaluating particular gene sets should perform such tests. Generalization performance of classifiers was evaluated using a Leave-One-Out procedure. In LOO testing, one data point is removed and a classifier re-learned using the rest of the data. The resultant classifier is tested on the held out data point and if it is in error, a LOO error is counted. However for these m-LC\'s where mis small, LOO error is not a good performance measure. In general, only somewhat isolated points lying close to the classifier decision boundary will be found in error during LOO testing. Thus LOO error rates tend to be very low, 0 -- 3 out of the total number of points. The small magnitude of this error count allows one to be misled thinking the classifier is \"good because the LOO error is low\". The LOO error is included in the results available from the web site. In the end, classifiers with a larger separation (margin) between the classes are able to tolerate more noise without errors. Thus larger margin classifiers are more desirable for use on future data points. Implementation -------------- The program is an adaptation of the lp\_solve version of LIKNON \[[@B6]\] implemented in C. A modern desktop workstation (1.8 Ghz Athlon) is able to evaluate many thousands of pairs per second using this LIKNON method. Table [3](#T3){ref-type="table"} lists run times for some of the data sets. Evaluating all pairs requires about n^2^/2 (equivalent to filling in a triangular matrix), and all triples about n^3^/6. A data set with P =3000 features necessitates the evaluation of 4.5 × 10^6^pairs and 4.5 × 10^9^triples. An approximate average time is 1 second/10^6^pairs/sample. Thus, evaluating all pairs for these data sizes requires only a few minutes whereas all triples needs many hours. Random data ----------- How many such pairs occur by chance? Theory suggests \[[@B18]\] that the probability of a 2 class data set of Nitems with ain one class, being linearly separable in 1 dimension is ![](1471-2105-6-97-i5.gif) and in 2 dimensions is approximately ![](1471-2105-6-97-i6.gif). Experiments were performed where a data matrix containing random numbers from a uniform distribution was analyzed by LIKNON. The results are shown in Table [4](#T4){ref-type="table"}. When the number of samples is lower than about 30, or one class is very small relative to the other class, then the chance of finding a pair of random genes that form a perfect classifier is large enough to easily measure. The smallest real data set examined in this work (the GIST set) has 1987 genes in 19 samples with 13 in one class and 6 in the other. Experiments with 10 random data sets of this size shows that 1500 pairs would be expected on average with a maximum found of 2706, and no single genes, Table [5](#T5){ref-type="table"}. Also in Table [5](#T5){ref-type="table"} are results of experiments where the real data and class labels are used, but the class labels are randomly shuffled. The label shuffling results are worse than random data only for the smallest two data sets (GIST and BreastBRCA BRCA1 vs BRCA2). The real GIST data has more than 137000 pairs, some 50 times more than found in random data and 30 times more than when labels are randomly shuffled. It also has 74 single genes, where the random data yields none. The three result sets that are closest to these random results are Cutaneous (596 vs 62), BRCA Breast BRCA1 & BRCA2 verses Sporadic splitting (2114 vs 1286) and BRCA Breast BRCA1 verses BRCA2 (143574 vs 53900). Web site ======== The web site \[[@B14]\] contains the data and results for this work. Authors\' contributions ======================= LG carried out all work outlined in this article. LG wrote and approved the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 The expanded discussion section. A discussion of the top results for all data sets. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The author acknowledges I Saira Mian for her support of this work. This work was supported by the National Institute on Aging, National Institute of Environmental Health Sciences, U.S. Department of Energy and California Breast Cancer Research Program. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Example plot using two hypothetical genes. Each data point is labeled with the class, and the separating plane is computed to be positioned halfway between the two classes. In this example there is a large separation between the two classes and perfect separation is achieved and no data point is close to the plane. ::: ![](1471-2105-6-97-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Liver cancer pair PLAC8 verses BCAT2. The two misclassified samples 108 and 109 are shown as squares. There are 3 other genes that form such pairs with PLAC8. ::: ![](1471-2105-6-97-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Liver cancer 3D plot of MT1E, ARHE and GPC3. These 3 genes form a perfect classifier although the margin is small. Red are cancer samples. The web site contains an interactive plot. ::: ![](1471-2105-6-97-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### 1P59 crystal structure. Shown with the alignment hit to the liver cancer possible biomarker PLAC8 highlighted in strands at the top. Alignment generated from PFAM model pfam04749.5DUF614 using the SAM HMM system and displayed in RASMOL. ::: ![](1471-2105-6-97-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### From the YeohALL data, T vs the rest, the best single gene CD3D. This gene perfectly separates the classes. Plus signs are T subtype samples. ::: ![](1471-2105-6-97-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### From the YeohALL data, T vs the rest, the best pair HLA-DRA and HUMTCBYY. Each gene alone provides some classification power, but when linearly combined form a perfect classifier, albeit with a small margin. Plus signs are the T subtype samples. ::: ![](1471-2105-6-97-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### From the YeohALL data, E2A vs the rest, the best single gene PBX1. This gene perfectly separates the classes with a wide margin and has higher values in E2A. Plus signs are E2A subtype samples. ::: ![](1471-2105-6-97-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### From the YeohALL data, E2A vs the rest, the best gene pair RPS6 and LRMP. Plus signs are the E2A subtype samples. LRMP by itself is a reasonable indicator of E2A status, but when combined with RPS6 can perfectly separate the data. ::: ![](1471-2105-6-97-8) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Published transcriptional profiling data sets reexamined in this study. For each set, the Table gives the abbreviation; dimensionality of the data points investigated after pre-processing of the features assayed in the original study; number, description and abbreviation for samples assigned to a category. ::: ------------------------------------------------------------------------------------------------------------------------------------------------------ Name Description -------------- --------------------------------------------------------------------------------------------------------------------------------------- BreastER Node-negative breast carcinomas, P= 3398 (3,398 cDNA clones) \[19\]\ 23 Estrogen receptor αpositive (ER^+^)\ 24 Estrogen receptor αnegative (ER-)\ <http://www.nhgri.nih.gov/DIR/Microarray/ER_data.txt> BreastBRCA Primary breast tumors, P= 3226 (6,512 cDNA clones) \[20\]\ 7 BRCA1 mutation (BRCA1)\ 8 BRCA2 mutation (BRCA2)\ 7 Sporadic (Sporadic)\ <http://www.nhgri.nih.gov/DIR/Microarray/NEJM_Supplement> OvarianBRCA Primary epithelial ovarian adenocarcinomas; P= 6445 (7,651 cDNA clones) \[21\]\ 18 BRCA1 mutation (BRCA1)\ 16 BRCA2 mutation (BRCA2)\ 27 Sporadic (Sporadic)\ <http://jncicancerspectrum.oupjournals.org/cgi/content/full/jnci;94/13/990/DC1> LungStanford Lung tumors; P= 918 (blanks set to average, 24,000 cDNA clones) \[22\]\ 41 Adenocarcinomas (AC)\ 16 Squamous cell carcinomas (SCC)\ <http://genome-www.Stanford.edu/lung_cancer/adeno> LungBeer Lung tissue samples; P= 4966 (4,966 cDNA clones) \[23\]\ 10 Non-neoplastic (Normal)\ 86 Adenocarcinomas (AC)\ <http://www.nature.com/nm/journal/v8/n8/suppinfo/nm733-S1.html> Cutaneous Cutaneous melanomas; P= 3613 (the \"detected\" set, 8,150 cDNA clones) \[24\]\ 31 Melanoma biopsies (Melanoma)\ 7 Tumor cell lines (Cell line)\ <http://research.nhgri.nih.gov/microarray/Melanoma_Supplement> GIST Tumors; P= 1987 (1,987 cDNA clones) \[25\]\ 13 KIT-mutation positive gastrointestinal stromal tumors (KIT^+^)\ 6 Spindle cell tumors from locations outside gastrointestinal tract (Spindle)\ <http://research.nhgri.nih.gov/microarray/gist_data.txt> YeohALL Pediatric, acute lymphoblastic leukemia bone marrows; P= 4196 (10000 variation filter, 12,625 Affymetrix HG\_U95Av2 probes) \[26\]\ 43 T-lineage ALL (T)\ 27 E2A-PBX1 (E2A)\ 15 BCR-ABL (BCR)\ 79 TEL-AML1 (TEL)\ 20 MLL (MLL)\ 64 Hyperdiploid\>50 (Hyperdip50)\ <http://www.stjuderesearch.org/data/ALL1> Prostate Prostate tissue samples; P= 3958 (12,626 Affymetrix U95a probes) \[27\]\ 25 Primary cancer tissue (Tumor)\ 9 Non-neoplastic tissue (Normal)\ <http://carrier.gnf.org/welsh/prostate/> Liver Liver Cancer (Hepatocellular carcinoma); P= 6605 (cDNA clones) \[1\]\ 105 Tumor (Tumor)\ 76 Normal (Normal)\ <http://genome-www.stanford.edu/hcc/> ------------------------------------------------------------------------------------------------------------------------------------------------------ ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Number of 1-, 2- and 3-LCs for the data sets described in Table 1. For the two-class partitionings shown, the Table gives the number of perfect linear classifiers (given Pinitial features) that can be constructed using one, two and for some sets 3, genes. The 3 gene results report the total number of triples, and in parenthesis the number saved for later analysis. ::: Data set Class +1 Class -1 1-LCs 2-LCs 3-LC -------------- ------------- ------------------------- ------- --------- -------------------- BreastER ER^+^ ER^-^ 0 16 108 k(2045/180^1^) BreastBRCA BRCA1 BRCA2 18 143,574 BRCA1/BRCA2 Sporadic 0 2,114 BRCA1 BRCA2/Sporadic 0 12,729 BRCA1 Sporadic 4 66,754 BRCA2 BRCA1/Sporadic 0 10,027 BRCA2 Sporadic 7 78,901 OvarianBRCA BRCA1 BRCA2 0 1,612 BRCA1/BRCA2 Sporadic 0 0 BRCA1 BRCA2/Sporadic 0 0 BRCA1 Sporadic 0 0 249^2^ BRCA2 BRCA1/Sporadic 0 0 BRCA2 Sporadic 0 0 2^3^ LungStanford AC SCC 2 484 565 k(65) LungBeer AC Normal 5 22,102 Cutaneous Melanoma Cell line 0 596 4.2 m(386) GIST KIT^+^ Spindle 74 137,981 YeohALL T E2A/BCR/TEL/MLL/Hdip50 1 1169 E2A T/BCR/TEL/MLL/Hdip50 4 386 BCR T/E2A/TEL/MLL/Hdip50 0 1 TEL T/E2A/BCR/MLL/Hdip50 0 3 MLL T/E2A/BCR/TEL/Hdip50 0 2 Hdip50 T/E2A/BCR/TEL/MLL 0 0 Prostate Tumor Normal 52 249,665 Liver Tumor Normal, Original Labels 0 0 44^4^ Tumor Normal, Relabeled 0 4 9.5 k(291^4^) (^1^Only 180 of the estrogen triples do not contain the Estrogen Receptor 1 gene. ^2^Ovarian BRCA B1 vs Sporadic triples were run with a reduced set of genes, variation filtered to 2109. ^3^Ovarian BRCA B2 vs Sporadic triples were run with a reduced set of genes, variation filtered to 2097. ^4^Liver Cancer triples use only 1956 genes. Relabeled means the the 2 \"outlier\" samples are re-labeled as normal.) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Wallclock run times for 2-LCs (pairs) for some of the data sets listed in Table. The P-LC program is written in C and uses double precision arithmetic. \"Total time\" (seconds) and \"Time estimator\" (seconds per 10^6^evaluations per sample) are execution times for the software on a 1.8 GHz AMD Athlon computer running Linux in an unloaded network configuration. An average time estimate is roughly 1 second per 10^6^evaluations per sample. ::: Data set Number samples Number genes Number pairs Total time Time estimator ------------ ---------------- -------------- -------------- ------------ ---------------- BreastER 47 3,389 5,700,000 251 0.94 BreastBRCA 22 3,226 5,200,000 161 1.41 LungBeer 96 4,966 12,300,000 1,500 1.27 Cutaneous 38 3,613 6,500,000 260 1.05 YeohALL 248 4,169 8,700,000 1,800 0.83 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Random data tests. The experimentally determined largest number of pairs found from 20 runs on random data. Each class has half the number of samples (number positives = number of negatives). The total number of genes is 2000. The second column is the number of pairs found for this 2000 gene set size, third column is the observed probability a single pair will be a perfect classifier (\"observed\" / 2000^2^/2). ::: Samples observed Chance for a single pair --------- ---------- -------------------------- 16 4625 0.002312 18 1822 0.000911 20 965 0.0004825 22 510 0.000255 24 30 0.000015 26 10 0.000005 28 8 0.000004 30 1 0.0000005 32 1 0.0000005 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Random data simulations of real data sets. This table compares the results found from the real data (Real column) to two different types of random data. The Random column contains the experimentally determined largest number of pairs found from 10 simulation runs using a random data matrix (drawn from a uniform distribution) where the number of genes and class sizes is the same as the indicated for the real data. The Label Shuffled column contains the experimentally determined largest number of pairs found from 30 simulation runs where the class labels were randomly shuffled. In the samples column, the number in parenthesis is the number of positive samples. The numbers after the slash are the number of single genes found. Label shuffling leads to more pairs found \"by chance\" only for the smaller data sets. The small data sets have large numbers of pairs expected \"by chance\". ::: Data set Samples Genes Real Random Label Shuffled --------------------------------------- --------- ------- ----------- --------- ---------------- GIST 19(6) 1987 137981/74 2706/0 4622/2 BreastBRCA(brca1 vs brca2) 15(7) 3226 143574/18 20563/2 53900/11 BreastBRCA(brca1 & brca2 vs Sporadic) 22(7) 3226 2114/0 1286/1 0/0 Cutaneous 38(7) 3613 596/0 62/0 24/0 LungStanford 52(13) 918 486/2 0/0 0/0 LungBeer 96(10) 4966 22102/5 0/0 0/0 Prostate 34(9) 3958 249662/52 57/0 13/0 :::	PubMed Central	2024-06-05T03:55:56.467638	2005-4-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090559/", "journal": "BMC Bioinformatics. 2005 Apr 13; 6:97", "authors": [ { "first": "Leslie R", "last": "Grate" } ] }
PMC1090560	Background ========== Uromodulin is the most abundant protein in the urine of all placental mammals, with approximately 50--200 mg released per day. It is an 85-kD glycosylphosphati-dylinositol (GPI)-anchored glycoprotein secreted from the epithelial cells of the thick ascending limbs (TAL) of Henle\'s loop and the early distal convoluted tubule in kidney \[[@B1]\]. Uromodulin has an identical amino acid sequence, immunologic cross-reactivity and tissue localization as Tamm-Horsfall protein (THP) \[[@B2]\]. Physiological functions of uromodulin have remained elusive, but recent knock-out studies have suggested that it plays a role in defense against urinary tract infection \[[@B3],[@B4]\]. It may also have an immuno-suppressive role \[[@B5]\]. The uromodulin gene promoter has been cloned from human, bovine, rat and mice species \[[@B6]-[@B8]\]. The abundance of the uromodulin protein in urine makes the uromodulin promoter a good candidate for driving the production of recombinant (rc)-proteins in the kidney and eventual excretion into the urine of transgenic animals (mice, goats, etc.). The potential of targeting rc-proteins in the urine may have advantages over the more widely used mammary system. Rc-protein may be harvested right after birth for both sexes from a rather simple medium. However, for such a system to be effective, expression levels should be in the range of 0.5--1 g/L to satisfy commercial applications. To date hGH is targeted into mouse urine using mouse uroplakin II at \~0.5 μg/mL \[[@B9]\]. The same promoter has been used to target the human granulocyte macrophage-colony stimulating factor (hGM-CSF) to the kidney of transgenic mice with the urine secretion level up to 180 ng/mL \[[@B10]\]. Secretion of rc-alpha1-antitrypsin in mouse urine at levels as high as 65 μg/mL has been achieved using the human uromodulin promoter \[[@B11]\]. As a first step in further exploring the production of rc-proteins in the urine of transgenic animals, we cloned a goat uromodulin gene promoter fragment, fused it to a GFP reporter gene and studied GFP targeting and distribution in the kidney and more specifically in the epithelial cells of the TAL of Henle\'s loop and the early distal convoluted tubule. GFP is an important tool in molecular and cellular biology as a transcriptional reporter, fusion tag, or biosensor. It is considered as an almost ideal in vivoreporter gene, because it does not interfere with cell vitality. It is highly sensitive and it can be easily detected using fluorescence microscopy. Results ======= Cloning and characterization of the goat uromodulin gene promoter and its partial 3\' end ----------------------------------------------------------------------------------------- A 3.7 kb fragment of the goat uromodulin gene fragment containing a 1.5 kb 5\' flanking region, exon 1, intron 1, exon 2 and part of intron 2 was cloned by PCR genomic walking based on the bovine sequence \[[@B7]\]. The goat uromodulin promoter proximal end shares 95 % identity with bovine uromodulin gene promoter sequence \[[@B7]\]. The potential transcription initiation site as well as DNA transcription factor binding sites of the goat uromodulin promoter fragment was deduced by comparison with the bovine promoter sequence (Figure [1](#F1){ref-type="fig"}). The proximal 1.5 kb 5\'-flanking region contains typical eukaryotic promoter elements including two CCAAT boxes at position -65 and position -558, a TATA box at position -30, and the conserved sequence TGTAAAAGG (nucleotides -3 to +6). The proximal 5\'-flanking region also contains several putative binding sites for known transcription factors such as CEBPB, NFAT, MZF1, SOX5, TCF11, GATA1, IK2 and DELTAEF1, etc., as analyzed with MatInspector V2.2 \[[@B12]\]. In addition, several consensus binding sites for activator protein-1 (AP-1) are present. A 2.7 kb fragment of the goat uromodulin gene 3\' end was also cloned using PCR genomic walking and degenerate primers designed based on conserved nucleotides between bovine and human \[[@B2]\] uromodulin sequences. The cloned 3\' end fragment contains part of exon 10, intron 10 and exon 11 including the 3\' untranslated region and a potential polyadenylation site (AATAAT). Assignment of exon/intron boundaries was achieved by comparison to the bovine sequence \[[@B7]\]. FISH analysis, performed using the 1.5 kb fragment of the goat uromodulin promoter as a probe, revealed that the gene was localized on goat chromosome No. 25 (25q14-17). Chromosome assignment was performed using a known BAC goat genomic clone as a marker \[[@B13]\] (Figure [2](#F2){ref-type="fig"}). Characterization of mice transgenic for goat uromodulin-GFP construct --------------------------------------------------------------------- In order to test the functionality of the cloned goat uromodulin promoter we generated transgenic mice using expression DNA cassette GUM-GFP (Figure [3](#F3){ref-type="fig"}) in which 2 copies of the chicken β-globin insulator sequences were fused 5\' upstream to the 1.5 kb fragment of the GUM promoter. Three transgenic founder mice were mated with wild type mice and tail DNA samples from the F1 and F2 offspring were analyzed by PCR and Southern blot (Figure [4](#F4){ref-type="fig"}). Digestion with AflII resulted in a 2.8 kb fragment which includes 500 bp of the insulator sequence, the promoter region and the full length of the GFP cDNA. The transgene may be truncated in 99-122-1M-1B1 mouse as the major hybridizing band appeared smaller in size (Figure [4](#F4){ref-type="fig"}). Approximately 50 % of the F1 littermates were identified as transgenic. Thus, the gene construct was transmitted according to the Mendelian rules of inheritance. The three transgenic founder lines were viable and fertile. The production of the GFP transgenic mice was through germline random integration. In this way the transgenic expression pattern would likely be reproducibly observed in F1 and subsequent generations. This approach may eliminate the problem of mosaic expression and, in most cases, the transgenic lines will have correct patterns of expression, mimicking those of endogenous uromodulin gene from which the promoter was derived. However, variable transgenic expression could still occur among different transgenic lines, attributed to a chromosomal position effect \[[@B14]\]. To address this problem, efforts have been made in transgenic animals by using certain chromosomal controlling elements, such as locus control region from the globin gene cluster \[[@B15],[@B16]\] and matrix attachment region from the chicken lysozyme gene \[[@B17]\]. The inclusion of the two copies of the upstream sequence of the chicken β-globin gene, used as an insulator \[[@B15]\], might have stabilized the GFP expression in kidney of the transgenic mice developed in this study. Expression of the GFP transgene in tissues of transgenic mice ------------------------------------------------------------- In order to analyze expression of GFP, extracts from various tissues (lung, heart, kidney and liver, etc.), dissected from a transgenic mouse as well as a negative control, were subjected to Western blotting analysis using a polyclonal anti-GFP antibody (Figure [5](#F5){ref-type="fig"}). The result demonstrated that the GUM promoter directed kidney specific expression of the reporter gene since GFP expression was identified only in kidney extracts but not in lung, heart and liver. Cellular localization of GFP reporter gene in the kidney cells of mice transgenic for GUM-GFP --------------------------------------------------------------------------------------------- It is established that uromodulin protein is expressed exclusively in the epithelial cells of the TAL of Henle\'s loop and the early distal convoluted tubule in the kidney \[[@B18]\]. It can, therefore, be used as an ideal marker for identifying these epithelial cells. Since GFP expression in the transgenic mice generated was under the control of the goat uromodulin promoter we reasoned that GFP would co-localize within the same epithelial cells that normally express endogenous uromodulin. Immunohistochemistry analysis of cryo-sections of kidneys dissected from the transgenic mice using an anti-human uromodulin antibody confirmed that expression of the GFP marker protein co-localized with the endogenous uromodulin (red-color stained) in the medulla region of the kidney (Figure [6B, C](#F6){ref-type="fig"}). It was noteworthy, however, that only a small proportion of the cells within the medulla region of the kidney co-expressed both uromodulin and GFP (Figure [6D--F](#F6){ref-type="fig"}) with the rest of the cells forming a mosaic pattern between cells expressing either endogenous uromodulin or GFP (Figure [6C](#F6){ref-type="fig"}). The cellular localization of endogenous uromodulin and transgenic GFP was in the plasma membrane and cytoplasmic space, respectively (Figure [6B, C](#F6){ref-type="fig"}). The two proteins were not detected in macula densa and from glomeruli, proximal convoluted tubules, thin limbs of the Henle\'s loops, collecting duct, blood vessels, and the interstitium (data not shown). Onset of uromodulin expression during embryonic life ---------------------------------------------------- We investigated the onset of the endogenous uromodulin expression during embryonic life and compared it with the expression of the GFP transgene driven by the GUM promoter as it has been reported that in human the uromodulin protein can be detected after 20 weeks of gestation \[[@B19]\]. Two GUM-GFP transgenic F1 mice were mated with wild type females. Pregnant females were euthanized at days 10--15 of gestation and fetuses were recovered. DNA extracted from the head of each fetus was used to identify by PCR the fetuses as transgenic. The transgenic fetuses as well as negative ones were cryo-sectioned and whole embryo mounts were examined for the presence of the GFP transgene and by immunohistochemistry for uromodulin positive cells. No endogenous uromodulin or GFP positive staining was observed on the embryo sections analyzed, indicating that uromodulin was not detectable during embryonic stage tested (data not shown). However, Western blot analysis performed on the homogenized kidney samples of day 0 pups from a transgenic mouse line demonstrated that the GFP was expressed at birth (Figure [7](#F7){ref-type="fig"}). Discussion ========== Transgenic animal technology is a valuable tool for understanding gene function. Moreover, the use of transgenic farm animals for the production of pharmaceutically important rc-proteins, such as antibodies, anti-clotting factors, and growth factors, in the mammary gland is also very well documented \[[@B20]\]. A transgenic animal that secretes rc-proteins in its urine offers certain advantages and compliments mammary-gland based rc-protein production system. We have chosen to use the goat uromodulin promoter driving the GFP kidney-specific expression in transgenic mice as a pilot feasibility study for the goat. Compared with other farm animal species, the goat is one of the most promising models for commercial production for rc-proteins and it is used more frequently because of relatively low cost of maintenance and faster breeding times as compared with cattle \[[@B21],[@B22]\]. The feasibility of targeting proteins in the uro-epithelium of bladder and kidney has been demonstrated \[[@B9],[@B23]-[@B25]\]. The present study further confirms these studies. The practical outcome of these studies would be to use urine as a bioreactor system in which rc-proteins could be expressed and excreted under the control of the goat uromodulin promoter. Characterization of targeting the uro-epithelium of bladder and the kidney by the use of the uroplakin II or uromodulin promoters has been recently initiated as mentioned above. Promoters from erythropoietin \[[@B26]\] and rennin \[[@B27],[@B28]\], have been used as kidney specific genetic elements. We report on the tissue specificity of a goat uromodulin promoter in transgenic mice. The reporter GFP gene used was co-expressed in the epithelial cells in the Henle\'s loop together with the endogenous uromodulin. Using the same GUM-GFP transgene we established the concomitant developmental control of the uromodulin expression and identified that the earliest stage of GFP expression was detected at birth. The tissue specificity observed demonstrate that the goat uromodulin promoter fragment cloned is capable of controlling kidney specific expression of GFP. It has been reported that small quantities of uromodulin/THP can be detected in blood using sensitive immunoassays \[[@B29],[@B30]\], but it is possible that some proteins in the serum simply cross-react with the antibody raised against the uromodulin/THP proteins \[[@B31]\]. The renal environment appears not to be essential for transcription of the uromodulin/THP gene or synthesis of the protein. Support of this stems from the observation that tissue of rat renal cortex and medulla, when transplanted to the anterior chamber of the eye, proliferated to form tubule-like structures \[[@B32]\]. Sections of these implants reacted with monoclonal antibodies specific for rat uromodulin/THP by immunohistochemistry \[[@B18]\]. The uromodulin/THP immuno-cross-reactive material in liver and cerebrospinal fluid is suggestive that this protein may be present extrarenally in all tissues involved in active chloride transport (such as intestinal and liver, etc), however this finding has not been confirmed either with the use of monoclonal antibodies or the mRNA presence in these tissues \[[@B2]\]. In agreement with this, our data, as well as other recent reports \[[@B11],[@B23]-[@B25]\], clearly demonstrate that the uromodulin promoter is an excellent candidate for driving foreign proteins expressed specifically in the kidney. GFP fluorescence provides an easily visualized marker for the epithelial cells in whole kidneys dissected in frozen sections. This eliminates the need to stain with lectins or antisera to visualize the cells, and therefore allows the cells to be visualized simultaneously with other structures in the developing kidney by combined immunohistochemistry/GFP detection. The arrays of the GFP expression in the kidney cells are of interest (Figure [6](#F6){ref-type="fig"}). Normally one would assume that since the goat uromodulin promoter was used to drive the GFP expression the same cells should express both uromodulin and GFP all together, however this was only observed in a small proportion of the cells. Due to the copy numbers of the uromodulin promoter integrated in the genome of the transgenic mice it is possible that it competes with the endogenous uromodulin promoter for the transcriptional regulatory factors in the same cells and as a result it may compromise endogenous uromodulin expression. A similar example has been reported for expression of a Bombyxcytoplasmic actin gene in cultured Drosophilacells to show that expressions of endogenous and recombinant actin genes are not independent \[[@B33]\]. When the Bombyxcytoplasmic actin A3 genes were introduced into Drosophilacells, the amounts of transcripts from the endogenous cytoplasmic Act 5C and Act 42A actin genes decreased proportionally. Furthermore, in cell lines with stably integrated A3 genes, the increased accumulation of endogenous cytoplasmic actin mRNAs is accompanied by a decrease in the A3 mRNA accumulation: thus when the transcription of resident genes is stimulated, the expression of the transgene is reduced. This balance suggested a competition for the factors involved in the transcription of the cytoplasmic actin genes. The appearance of uromodulin/THP coincides developmentally with maturation of the TAL of Henle\'s loop \[[@B18]\]. It has been reported that it is present at eight weeks in the human fetal kidney, two days pre-term in rat \[[@B34]\], and three days pre-term in hamster kidney \[[@B35]\]. Urine from fetal kidney contributes to amniotic fluid from 12 to 14 weeks of gestation \[[@B36]\]. While uromodulin/THP has been described in human amniotic fluid near term by several researchers, studies in mid-term amniotic fluid have been more equivocal \[[@B37]-[@B39]\]. Kumar et al. have tested amniotic fluid from four patients at 16 weeks and could not detect uromodulin/THP by an enzyme-linked immunosorbent assay in which the lower limit of detection was 20 ng/mL \[[@B18]\]. The earliest stage in which we were able to detect GFP expression driven by uromodulin promoter in transgenic mice was at birth (Figure [7](#F7){ref-type="fig"}). Although immunohistochemistry was attempted to examine the expression of both uromodulin and GFP at earlier embryonic stages neither could be detected (data not shown). It is likely that the expression level was not high enough to be visualized by the method employed, or that there is a species difference in the ontogeny of the uromodulin protein. Conclusion ========== The goat uromodulin gene promoter efficiently targeted expression of GFP in the kidney of transgenic mice and more specifically in the epithelial cells of the TAL of Henle\'s loop. Expression of GFP was co-localized in cells expressing endogenous uromodulin. Expression appeared to be specific to kidney at least among the tissues tested. Methods ======= Cloning of the goat uromodulin gene promoter and its partial 3\' end -------------------------------------------------------------------- PCR was performed using genomic DNA from a standard Saanen goat as template with two primers deduced from the conserved regions of human and bovine uromodulin gene promoters \[[@B7]\]. The sense primer was 5\' CAT TCT CAG CTC CTY TCY TGC 3\' and the antisense primer was 5\' AGA GAC CCC AAA TGA TTG ACA 3\'. A 350 bp PCR fragment was obtained and sequenced, showing 95 % identity with the known bovine uromodulin promoter sequence \[[@B7]\]. The gene-specific primers were designed from the PCR product to conduct PCR genomic walking using a goat (standard Saanen breed) genomic DNA library, created with the Universal Genome Walker Kit (Clontech). A 3.7 kb PCR fragment was obtained both from the genome-walking library and the goat genomic DNA with newly designed gene-specific primers. This fragment contains a 1.5 kb 5\' flanking region, exon 1, intron 1, exon 2 and part of intron 2. Similar strategy was employed for cloning a 2.7 kb fragment of the 3\' end. The initial set of degenerate primers was designed based on the conserved regions of bovine uromodulin gene. GUM/GFP plasmid construction ---------------------------- A 1.5 kb fragment of the goat uromodulin promoter was subcloned into the pGEM-T easy vector (Promega) and a HindIII-PstI digested promoter insert was ligated with HindIII-PstI digested CMV-IE and EF-1alpha-less pCEEGFP \[[@B40]\] to generate pGUMGFP3. Two copies of a 1.2 kb fragment of the chicken β-globin insulator \[[@B15]\], was digested with XhoI and SalI, and ligated at the XhoI site, located upstream of the goat promoter fragment in pGUMGFP, resulting in the final construct, pING32. The cloned 2.7 kb fragment of the goat uromodulin gene 3\'end was not part in the DNA construct used in the generation of transgenic mice. Production of uromodulin-GFP transgenic mice -------------------------------------------- The plasmid backbone of pING32 was removed by XhoI-SphI digestion and the 5.5 kb uromodulin-GFP transgene fragment (Figure [3](#F3){ref-type="fig"}) was gel-purified and microinjected into the pronuclei of FVB mouse zygotes using standard techniques \[[@B41]\]. The production and maintenance of transgenic mice were conducted at the McIntyre Transgenic Core Facility of McGill University. Animal studies were carried out in accordance with the Guide for the Care and Use of Laboratory Animals as adopted by the U.S. National Institutes of Health. PCR analysis ------------ Transgenic mice were detected by PCR analysis of tail DNA with primers amplifying sequences of the insulator, the junction sequence between the uromodulin promoter and GFP. The sense and antisense primers used to amplify the insulator fragment were 5\' AGG AGC ACA GTG CTC ATC CAG ATC 3\' (Acb347) and 5\' GAC GCC CCA TCC TCA CTG ACT 3\' (Acb478). The sense and antisense primers for amplifying the uromodulin promoter-GFP were 5\' GAT CAT TGG AGG AGA GAT TGC CAG TG 3\' (Acb504) and 5\' GTC TTG TAG TTG CCG TCG TCC TT 3\' (Acb412) (Figure [3](#F3){ref-type="fig"}). PCR was performed using the Ready-to-Go™ PCR beads (Amersham-Pharmacia Biotech) under the following conditions: 94°C/2 min, 36 cycles of 94°C/30 s, 60°C/45 s, 72°C/45 s and finally 72°C/5 min in a PTC-100™ Programmable Thermal Controller (MJ Research, Inc.). The PCR products were visualized on a 1 % agarose gel and analyzed by FluorChem™ 8000 Advanced Fluorescence, Chemiluminescence and Visible Light Imaging System (Alpha Innotech Corporation). Southern blot analysis ---------------------- Genomic DNA extracted from mouse tail was digested with AflII, subjected to electrophoresis on a 1 % agarose gel and transferred to a Nylon membrane (Roche), positively charged. The membrane was hybridized using a DIG Easy Hyb buffer (Roche) at 42°C overnight containing a PCR-generated probe, labeled by the PCR DIG probe synthesis kit (Roche). This fragment was amplified using the Acb504/412 set of primers amplifying a 490 bp fragment. The membrane was washed once at room temperature with 2 × SSC, 0.1 % SDS for 5 min, three times with 0.5 × SSC, 0.1 % SDS at 68°C for 15 min per wash. Detection of hybridization signals was by using the CDP-Star™ substrate (Roche). The membrane was analyzed by the FluorChem™ 8000 System (Alpha Innotech Corporation). Fluorescent In Situ Hybridization (FISH) ---------------------------------------- The method used essentially was as described \[[@B42],[@B43]\]. Briefly, R-banded chromosomes spreads were prepared by synchronization of goat peripheral blood lymphocytes cultured in RDG medium with thymidine for 18 h. A DNA probe including the 1.5 kb goat uromodulin promoter fragment was labeled with a Bionick labeling kit (Life Technologies Inc). Hybridization medium containing biotinylated probe and genomic goat DNA, was placed on each slide and covered with a plastic film. The slides were incubated overnight at 37°C in a humidified chamber with 50 % (v/v) formamide/2 × SSC followed by wash steps. The slides were incubated for 45 min at 37°C with rabbit anti-biotin Enzo (1 % in modified PBS: PBS with 0.1 % Tween™ 20 and 0.15 % BSA added), and rinsed twice in modified PBS. The samples were incubated with a mouse anti-rabbit biotin-conjugated antibody (0.75 % in modified PBS, Pierce) and rinsed. Slides were stained with Propidium Iodide, mounted and covered with a coverslip. Samples were observed under an Olympus microscope B × 40 using an appropriate combination of filters for fluorescein labeling. Western blot analysis --------------------- Tissue extracts from kidney, lung, heart, liver and bladder dissected from uromodulin-GFP transgene positive mice as well as negative controls were prepared by homogenization in lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1 mM DTT, pH 7.5) containing a tablet of a proteinase inhibitor cocktail (Roche). Proteins recovered in the supernatants following a centrifugation step were separated by SDS-PAGE on a 4--20 % Tris-Glycine polyacrylamide gel (precasted, Invitrogen) and electrophoretically transferred to an ECL nylon membrane (Amersham). After pre-incubation of the membrane for 1 h in TBST \[0.24 % (w/v) Tris, 0.8 % (w/v) NaCl, pH 7.6, 0.05 % (v/v) Tween™ 20\] containing 5 % nonfat milk, it was incubated in TBST containing a polyclonal anti-GFP antibody (Clontech) at a 1:100 dilution, followed by incubation in TBST containing a HRP-conjugated secondary anti-rabbit IgG antibody (Promega) at 1: 5,000 dilution. Immunocomplexes were detected with the ECL chemiluminescence detection kit (Amersham). Analysis of the immunoreactive bands on the blots was performed with FluorChem™ 8000 System (Alpha Innotech Corporation). Immunohistochemistry -------------------- Mice transgenic for the uromodulin-GFP transgene, as well as a negative control, were perfused with 4 % paraformaldehyde for fixing the GFP in situ. Kidney and other tissues were dissected from the perfused mice and cryo-sectioned with a thickness of 5 μm. Following incubation of the tissue sections with 5 % goat serum in PBST \[PBS, pH 7.4, 0.05 % (v/v) Tween™\] they were incubated in PBST containing a polyclonal anti-human uromodulin antibody (BTI) at 1:200 dilution, followed by treatment with a Texas-Red conjugated secondary anti rabbit IgG antibody (Molecular Probe). The sections were mounted with DAKO fluorescence mounting medium (DAKO), and observed with an Olympus UV-microscope with a FITC filter (for GFP) and rhodamine filter (for Texas-red). Confocal fluorescence microscopy -------------------------------- Immuno-stained slides were imaged using a Zeiss LSM 410 confocal scanning laser microscope equipped with argon-krypton and helium-neon lasers and appropriate filter sets for independent detection of FITC and Texas Red. Superimposition of serial sections was used to compare the distribution of each of the labels. Authors\' contributions ======================= YJH carried out the molecular cloning studies, participated in the sequence alignment, analysis of tissues and extracts, and drafted the manuscript. NC participated in the production of the transgenic mice. ASB conducted the FISH studies. JFZ and AL participated in the cloning and the production of the transgenic mice. CNK envisioned and supervised all the studies, as well as drafting the manuscript. Acknowledgements ================ We thank Dr. M. Tremblay and J. Penney for the generation and maintenance of the transgenic mice at the Transgenic Core Facility of McGill University; Drs. Y. De Koninck and T. P. Wong for use of their Confocal Microscope at the Department of Pharmacology and Therapeutics, McGill University; and S. Tibbit for technical assistance. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Sequence comparison of the proximal 5\' -flanking regions of goat and bovine uromodulin genes.Identical nucleotides are indicated dashed. The putative transcriptional start site of the goat promoter (nt 1428) is in bold, deducted from the known bovine start site (nt 591, bold). TATA, CCAAT boxes and other potential binding sites of transcription factors are underlined. The nucleotide sequences of the goat uromodulin gene promoter and the goat uromodulin gene partial 3\' end are available in the GenBank Database under the accession numbers AY702660 and AY702661, respectively. ::: ![](1472-6750-5-9-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### FISH analysis of R-banded metaphase spreads from peripheral blood with a goat uromodulin promoter probe.The arrows indicate the localization of the uromodulin gene on goat chromosome \#25, whereas the signal of the BAC clone (EPO) was used for chromosome assignment. ::: ![](1472-6750-5-9-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Schematic drawing of the pING32 transgene, excised from pING32 by digestion with XhoI-SphI.The lower bars indicate the lengths of the PCR products and the primer pairs used for the screening of the transgenic mice. ::: ![](1472-6750-5-9-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Southern blot analysis of AflII digested mouse genomic DNA.Lane 1: Dig-labeled Marker VII (Roche). Lane 2 & 11: GFP transgene plasmid DNA. Lane 3--9: tail DNA derived from GFP-positive mice. Lane 10: tail DNA from a non-transgenic mouse. 7 μg of DNA was loaded in each lane (lanes 3--10). The membrane was hybridized with a Dig-labeled PCR probe (Acb504-412, see Figure 3). The arrow indicates the GFP transgene. ::: ![](1472-6750-5-9-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Western blot analysis of protein extracts from tissues of GUE-GFP transgenic mice.Tissue extracts from a transgenic mouse (99-120-2F) and a negative control mouse (99-120-6F) were separated on 4--20 % SDS-PAGE and transferred onto the membrane. The GFP positive control was from a cell line with GFP expression. 12 μg of total protein was loaded in each lane. A.Immunoreacting bands were detected using a polyclonal anti-GFP antibody (Clontech). B. The same blot was reprobed with a mouse monoclonal anti-actin antibody (Chemicon) to show that equal amount of protein was loaded in each lane. ::: ![](1472-6750-5-9-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Demonstration of expression of the goat uromodulin -- GFP transgene and endogenous uromodulin in kidney sections by immunostaining.(A). H & E kidney TAL region staining of a control non-transgenic mouse. The arrows show microstructures with thick walls indicative of the thick segment of the ascending limb of Henle\'s loop within the TAL region. Fixed kidney sections from a negative control mouse (B) and the 99-122-1-A5M transgenic mouse (C) were stained with a polyclonal anti-human uromodulin antibody, followed by treatment with Texas-Red conjugated secondary anti rabbit IgG antibody. Expression of endogenous uromodulin in the cells (red) appears in a punctuated pattern, whereas the GFP expression is cytoplasmic (green). Immunostaining of similar sections for uromodulin confirmed that expression is restricted to tubular epithelial cells of these structures (B, C) indicated by red. The same sections observed for GFP expression by confocal microscopy indicated that expression was restricted to similar structures as in B&Cbut not co-expressed with the endogenous uromodulin staining. As expected GFP was absent in the renal sections obtained from a non-transgenic mouse. Both GFP (D, F) (green) and uromodulin (E, F) (red) were co-expressed in cells in the TAL segment (transgenic mouse, 99-122-1-A5M). ::: ![](1472-6750-5-9-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Western blot analysis of kidney extracts from Day 0 pups, offspring of a GUM-GFP transgenic F2 mouse.Kidney extracts (12 μg of total protein loaded in each lane) from 99-120-2 and 99-122-1A3 were used as positive controls, whereas 99-120-6 was used as a negative control. GFP immunoreacting protein was detected using the polyclonal anti-GFP antibody (Clontech) in 2 out 4 Day 0 pups, offspring of a transgenic F2 mouse, 00-074-1A2B8M, bred with a wild type mouse. ::: ![](1472-6750-5-9-7) :::	PubMed Central	2024-06-05T03:55:56.471845	2005-4-11	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090560/", "journal": "BMC Biotechnol. 2005 Apr 11; 5:9", "authors": [ { "first": "Yue-Jin", "last": "Huang" }, { "first": "Nathalie", "last": "Chretien" }, { "first": "Annie S", "last": "Bilodeau" }, { "first": "Jiang Feng", "last": "Zhou" }, { "first": "Anthoula", "last": "Lazaris" }, { "first": "Costas N", "last": "Karatzas" } ] }
PMC1090561	Background ========== Leukemias are the most frequent type of cancer in childhood \[[@B1]\], the incidence of which varies depending on the area of the world where they are studied \[[@B2]\]. An elevated frequency of acute leukemias has been reported in populations of Hispanic origin \[[@B3]-[@B5]\]. For example, in Costa Rica, the highest incidence rate in the world for acute leukemias has been reported: 56 per million of children under the age of 15 years \[[@B5]\]. In California and Texas, two states of the U.S. which have a Hispanic component predominately of Mexican origin \[[@B3]\], the incidence rates for ALL and for AL in general have been found to be greater than those previously reported for Mexico City \[[@B6]-[@B8]\]. However, because the studies carried out in Mexico City have been retrospective, it is possible that a under-register of the infirmity may have existed in Mexico City. In Florida, the state with the highest predominance of Hispanics of Caribbean and Central American origin, the incidence rate of ALL was higher than that reported for the Caucasian population and similarly elevated to that reported for Costa Rica. For the Central American country of El Salvador, there is no prior information on the incidence of acute leukemias. In developing countries, there are few data on the frequency of leukemias during childhood \[[@B5],[@B9]\] because these countries have not had the infrastructure required to keep reliable records. Some international studies have produced information from some developing countries, particularly in Latin America \[[@B10]\]; however, this information has certain constraints with respect to its validity and coverage in representing the populations \[[@B11]-[@B15]\] Of the studies carried out in Mexico City, an important increase in the incidence of ALL, but not of AML, has been reported. In 1991, an incidence of 22.2 per million of children under 15 years of age was reported \[[@B6]\]. From the data from the Instituto Mexicano del Seguro Social, the Medical Center having the greatest coverage of the population in the whole of the country, an incidence rate was reported of 29.1 per million of children under 15 years of age who were residents of Mexico City and who had the right to receive attention from this Institution \[[@B7]\]. Given the retrospective nature of these studies, it is possible that there was underestimation of the incidence rate of ALL. On the other hand, when taken into account that, in the different states in the U.S. which have a high component of Hispanics of Mexican origin, the incidence rate of acute leukemias is the highest in the world, one may consider it important to perform a prospective population-based study in Mexico City that would determine if Mexican children, residing in this City, had an incidence rate similar to those reported for Texas and California \[[@B3]\]. Prospective records of those children treated for leukemia was initiated in El Salvador in 1994 and in Mexico City in 1996 \[[@B16],[@B17]\]. In Mexico City, the record was kept in the hospitals of the Instituto Mexicano del Seguro Social (IMSS), an institution that attends about 50% of the population inhabiting Mexico City \[[@B7]\]. Because the IMSS keeps a register of the population that has a right to receive medical attention at its facilities, the base population is known and it is feasible to obtain the incidence rate for the population under 15 years of age. In El Salvador, because the record was kept by the Hospital Nacional de Niños \"Benjamín Bloom\" (BB), the only center in the country which attends all children, younger than 12 years with oncologic disease, data on the population base which was necessary for calculating the incidence rate was available. Treatment of children with leukemia is completely free of charge both in El Salvador and at the IMSS in Mexico City, thus allowing a greater coverage for children with this disease \[[@B16],[@B17]\]. This work presents the incidence of leukemia in children younger than 12 years, because this is the age range admitted in the BB and data for children younger than 15 is used for Mexico City. In this report, for purposes of analysis, the data from 1996--2000 were used, with 1996 being the first year of record keeping in Mexico City and 2000 being the most recent year for data reported from El Salvador. Methods ======= Design ------ Population-based data. Hospitals --------- The Hospital Nacional de Niños \"Benjamín Bloom\" (BB), located in San Salvador, capital of El Salvador; is a tertiary care center and is the only hospital in that country that has a pediatric hematology-oncology service with pediatric hematology staff. Being the only domestic reference center for pediatric oncologic diseases, it is the only hospital to which children with presumed leukemia diagnosis are referred for treatment. For this reason and because treatment is completely free of charge \[[@B16]\], no case of presumed childhood leukemia should be missing from the records. Children from other countries (Nicaragua, Honduras, etc) were excluded from the numerator. The hospitals included in the Mexico City study were the Pediatric Hospital of the Centro Médico Nacional \"Siglo XXI\" (HP) and the General Hospital of the Centro Médico Nacional \"La Raza\" (HG). Children who were attended in the pediatric hematology service and who resided in Mexico City were included in both hospitals. There are accurate records on the population that has a right to this service because the population attended by this institution in Mexico City is formed by workers whose personal data are registered. Only those children who were Mexican nationals and whose parents were residents of Mexico City were included in the numerator. Diagnosis --------- Once diagnosed with presumed leukemia, a child was referred to one of the hospitals where trained staff did a comprehensive follow-up of the case to either discard or confirm the diagnosis of leukemia. In the BB, infirmary staff trained for that activity carried out this work. In all cases, the diagnosis was confirmed with bone marrow smear, and histochemical tests (myeloperoxidase, sudan black B reaction, esterases, periodic acid Schiff (PAS) reaction, and acid phosphatase) were performed to differentiate the types of leukemia. In the cases in El Salvador, the staff of St. Jude Hospital confirmed some diagnostic tests \[[@B16]\]. The types of leukemia were divided according to the five-group morphological classification of the international classification of childhood cancer. Only four of the types were found in this study: a) Acute lymphoid leukemia (ALL) (9820--9827, 9850); b) acute myeloid leukemia (AML), the preferred terminology instead of \"acute nonlymphocytic leukemia\" \[[@B18]\] (9840, 9841, 9861, 9864, 9866, 9867, 9891, 9894, 9910); c) chronic myeloid leukemia (CML) (9863, 9868); and e) unspecified leukemias (UL) (9800--9804) (International Classification of Disease for Oncology, ICD-O2; 1990) \[[@B19]\]. The codes were reviewed with the Child-Check program \[[@B19]\] to corroborate that there were no duplicate records or inconsistent data. A database was generated to record age, sex, residency, year of diagnosis, and clinical manifestation of the patients. Populations ----------- Because, in El Salvador, only children under the age of 12 are attended at the BB, our analysis was limited to that cohort, instead of that conventionally used in a study of this type, i.e., children younger than 15 years \[[@B2],[@B10]\]. The denominators to calculate the rates were obtained by every year from the General Direction of Statistics and Census (Dirección General de Estadística y Censos), which is the only government source that calculates the estimates of population in El Salvador. The population younger than 12 was 1,756,513; 1,764,422; 1,787,064; 1,808,946, and 1,829,146 for the years 1996 to 2000, respectively. The last available census in El Salvador dates from 2000. In Mexico City, the denominators were obtained from the Coordination of Planning and Medical Information of the IMSS, which is the only government source that calculates all the populations with access to medical attention provided by the IMSS \[[@B20]\]. This information is updated every year by the IMSS. The recorded population younger than 15 years in Mexico City was 786,754; 832,199; 765,268; 806,043; and 881,887 for the years 1996 to 2000, respectively. The recorded population of children in Mexico City appeared to fluctuate from year to year because only that population of workers who are under contract with a company are entitled to the care provided by the IMSS. Thus, the denominator reflects the change in the number of jobs created or lost each year. Analysis -------- Annual incidence rates (AIR) were calculated by age group, sex, and each type of leukemia. Also, the average annual incidence rate (AAIR) was calculated by sex and total for the whole study period. These rates were standardized by age through the direct method with the world standard population \[[@B21]\], with the AIR and AAIR reported per million of children under 15 years old. The AAIRs were calculated by using the total of cases found in the study period as numerator and the sum of the populations found in each year of the study as denominator. Cases were stratified by age and assigned to one of four age strata: 1) \<1 year; 2) 1--4 years; 3) 5--9 years, or 4) 10--14 or 10--11 years, for Mexico City and El Salvador, respectively. Therefore, for El Salvador, it was not possible to calculate the AAIR accurately for the 0--14 year-old age group. In the study, an average rate over the whole period and for the 0--11 year-old age group was calculated. These rates were standardized by age by direct method and also with the world standard population. Results ======= During the study period, there were 375 children diagnosed with leukemia in El Salvador and 238 in Mexico City. Of total leukemias in El Salvador and in Mexico City, ALL represented 80.5% and 76.9%; AML, 16.8% and 18.1%; CML, 1.6% and 4.2%; and UL, 1.1 and 0.8%, respectively. The standardized rate by age of all leukemias in both countries was higher in boys than in girls, with a more marked male:female ratio in El Salvador than in Mexico City: 1.18 (46.8 per million children:39.5 per million children) vs. 1.01 (58.8 per million children:58.1 per million children). Whereas standardized rates by age in El Salvador revealed that all types of leukemias were more frequent in boys, this was not observed in AML or CML in Mexico City, where the frequency of both leukemias was higher for girls than that reported for boys (Tables [I](#T1){ref-type="table"} and [II](#T2){ref-type="table"}). The low incidence of ALL in youngsters less than 1 year old found in El Salvador is notable, as is the fact that no case of AML in this age group was found in Mexico City. The data for Mexico City, as those for El Salvador, showed that ALL had a peak between the ages of 1--4 years (Table [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). It is important to emphasize that, for the children in Mexico City, the age group with the highest AIR after that of the 1--4 year old group was the group of less than one year of age. Among the subtypes of AML, the most frequent in Mexico City was AML M3 (01B9866) with nine cases, giving a frequency of 20.9% and an AAIR of 2.21 per million children. It is noteworthy that all these AML M3 cases were females. Unfortunately, in El Salvador, the frequency of AML M3 could not be assessed, as the major part of AML was classified only in the general category 01B9861. Discussion ========== Because there are only two important reviews that summarize the high incidence rates of leukemias in children reported in the world up to 1999 \[[@B1],[@B10]\], it has been very difficult to obtain reliable data on the incidence of cancer in children, including leukemias, in developing countries. In fact, the review of Parkin et al., carried out in 1999, did not include data from either El Salvador or Mexico City because there were no reliable data from these areas at the time (1980 to 1989) at which they performed their study \[[@B10]\]. There are some issues of the present study that must be emphasized. In this study, statistics were not performed to compare the rates in Mexico City with the rates in El Salvador over time, because this was not the aim. The inclusion of the aforementioned institutions helped guarantee that all children diagnosed with leukemia during the studied period were included. This is due to the fact that, because BB is the only hospital in El Salvador that attends children with leukemia, any child with presumed leukemia is referred to that institution \[[@B16]\]. Similarly, in Mexico City, HP and HG are the only hospitals of the IMSS that attend the children with leukemia whose parents have rights to receive medical services from IMSS \[[@B17]\]. Even if there had been an under-registration of cases because some could not be diagnosed (e.g., patients dying before leukemia diagnosis is confirmed), the incidence rates in children of 0--11 years old found in El Salvador are similar to those reported in North America, which range from 36 to 45 <http://seer.cancer.gov/csr/1975_2001/>, and with rates from South and Central America, which range from 29.0 to 57.9 per million children \[[@B10]\]. In Mexico City, there are various institutions outside the IMSS system that can attend children with cancer, particularly leukemia. However, such private institutions represent an economic burden for the family of the patient, in that the treatment for patients with leukemia lasts several years, with the possibility of expenses totaling almost \$65,000 USD \[[@B22]\]. Therefore, it is less probable that the family of a child with leukemia would turn down access to the free service offered by HP and HG, opting instead for a costly private institution. For this reason, we think that cases in Mexico would not be underestimated due to patients being treated outside the IMSS system. Mexico is a developing country with a high rate of child mortality in general. The absence of AML cases in infants of less than 1 year in Mexico City is notable because this is the peak age for AML in most developed countries \[[@B1],[@B23]\], and suggests the probability that children younger than 1 year with AML die before being diagnosed. In contrast to ALL, AML seems to have no worse prognosis in children younger than 1 year \[[@B24],[@B25]\], indicating that all children with AML, regardless of age, run the same risk of not being diagnosed. This increases the possibility of an apparently different pattern of age of AML onset in Mexico City. Little can be argued about the delay of diagnosis in these diseases, due to fast evolution of the disease and short-term mortality when no prompt management is provided; thus, acute leukemias are considered as the reference point for time of diagnosis \[[@B26]\]. As to the validity of measurements, recommended international morphological criteria were used in this study, and all measurements were performed by staff highly specialized in diagnosis and management of children with leukemia \[[@B7],[@B16]\]. This study did not aim to assess incidence trend over time, a parameter that is affected by change(s) in diagnostic criteria \[[@B27]\], which were not a variable in this study. Since the period under study was too short to assess the incidence trend of leukemia, no analysis was done in this respect. However, this study evidences once again that choosing only one year to report a disease incidence may result in an over- or underestimation of the incidence of that disease \[[@B2]\]. For example, the observed incidence rate in Mexico City for the year 1996 in this study was much higher than those in the following years, suggesting that this might have been a completely random condition. Although the standardized rates reported in this study were higher than those previously reported for Mexico City, the prior studies \[[@B6]-[@B8],[@B26],[@B28]\] were retrospective and therefore are not comparable to the present study. However, the possibility that the incidence of ALL is increasing, particularly in Mexico City as reported in these studies, must be considered \[[@B6],[@B8]\]. This condition has been observed in various areas of Mexico City \[[@B8]\] and is associated with the more polluted areas and with farming areas \[[@B6],[@B8]\]. The elevated incidence of ALL in children of Mexico City agrees with those found in other reports that show that ALL are more frequent in populations of Hispanics \[[@B3]-[@B5]\]. Table [3](#T3){ref-type="table"} shows that the AAIR of ALL is the highest reported for the population of Hispanic origin, with the value for the Hispanic populations of Texas being the highest, followed by that for Florida. Note that the majority of Hispanics in Texas are of Mexican origin\[[@B4]\]. The AAIR for ALL reported by SEER for Hispanics for the period 1992--1998 is very similar to that reported for Mexico City and agrees with the fact that, of the Hispanic children included in this report of SEER, 66% come from the Los Angeles register \[[@B29]\], where the great majority of the Hispanic population are of Mexican origin \[[@B4]\]. It is noteworthy that the AAIR of the age peak of ALL in boys in Mexico City is very high (82.4 per million in boys), an is higher than that reported for Costa Rica (76 per million in boys) and by SEER (76.7 per million in both sexs) \[[@B5],[@B30]\]. However, the AAIR reported for Hispanics in general in Florida for the 0--4 year-old age group is 87.6 per million, which to our knowledge is the highest AAIR reported in the literature \[[@B4]\]. This finding is important because this peak in age is related with the higher frequency of ALL with the B-cell precursor immunophenotype. Although such data were not reported in this study, this immunophenotype is thought to be more frequent in developed countries \[[@B1]\] and, in addition, is related with genetic rearrangements such as TEL/AML1 that occur in the child during the intrauterine stage \[[@B31]\]. On the other hand, this age peak had been observed in relation to high socioeconomic status and population mixing \[[@B1]\]. The mix of urban and rural populations has been associated with the development of ALL \[[@B32],[@B33]\] and especially with the cases that occurred in children under the age of 5 years \[[@B1]\]. Those states in the U.S.A. that have reported an elevated AAIR for ALL have an elevated level of immigration. The greater portion of Hispanics that emigrate to these zones come from rural regions. Mexico City has a similar situation, in that it constantly receives an influx of people from rural zones. Although, in this study, patients were determined to be residents of Mexico City, neither the place from which their parents came nor the length of time that the children resided in Mexico City was investigated. It will be important to investigate in the future if the mix between urban and rural populations is related with the very high incidence of ALL in Mexico City and in Hispanics in the U.S.A. Another factor to be considered is the role of pesticides, whether occupational exposure of parents or exposure of children in the home. Also, we must not lose sight of the activities that Hispanic immigrants in the U.S.A. may perform, such as those related to the cultivation of crops, especially in areas of Texas where a higher frequency of ALL has been reported \[[@B34]\]. In Mexico City, the major zones with highest incidence rate of ALL are the zones in the South of the City where a portion of the population still carries out agricultural activities \[[@B8]\]. In a study carried out in Mexico City, an important relation was found between exposure to pesticides and the risk of developing AL \[[@B35]\]. In an ecological study realized in California, in which 36% of the cases with cancer were Hispanic children, it was found that the use of \"Propargite\", a pesticide used primarily in orchards and vineyards to control mites, was related to the higher rates of childhood leukemia \[[@B36]\]. The standardized incidence rate of ALL for children under 12 years of age in El Salvador was found to be less than the AAIR reported for Costa Rica or for Florida, where the highest percentage of Hispanics are of Caribbean and Central American origin, but this rate is relatively higher than that reported by SEER for all races \[[@B30]\]. This comparison has to be taken with some reservation because the AAIRs of SEER are reported for children under 15 years of age, whereas that of El Salvador is for only those under 12 years of age. The frequency of AML M3 in the Hispanic population has been reported to be higher than those for other populations \[[@B37]\]. This has been observed both for children and for adults \[[@B3],[@B37]-[@B39]\] (Table [4](#T4){ref-type="table"}). In the present study, it was found that, in Mexico City, the frequency was 21%, a value consistent with the data that the frequency of AML M3 is higher in a population of Hispanic origin. The present study has the advantage over other studies that were carried out in Latin America because it is population-based. One study has proposed that the elevated frequency of AML in the Hispanic population may be related not only genetic factors but also nutritional ones \[[@B38]\]. It is important to emphasize that, despite recent data published on the incidence of leukemias in Latin America \[[@B5],[@B40]\] or in Hispanics from the USA \[[@B29]\], those study periods were prior to those analyzed here. Finally, it can be stated that the rates here presented are similar to those reported for Latin America, thereby supporting the reliability of these results. This is the start of a project that intends to join efforts in Mexico City, El Salvador, Honduras, Guatemala, Costa Rica, and Nicaragua to keep records of cancer in children, which will be highly reliable and which, therefore, can provide valid and timely data on the great concern that child cancer represents in the Latin American countries. Conclusion ========== The AAIRs reported for Mexico City and the rate in children from 0--11 years old in El Salvador were similar to those that have been reported by other Latin American countries. The AAIR of ALL in Mexico City is one of the highest in the world. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= JMMA conceived and designed the study, analyzed the data and wrote the first draft of the manuscript. MB and AFG designed the study, analyzed the data, and provided guidance to all aspects of this project. RL, SJO, GR, MLPS, GGM, RBR, AOF, MOA and MCMG registered, recorded, and analyzed the data. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/33/prepub> Acknowledgements ================ This work was partially financed by the Instituto Mexicano del Seguro Social, through its Strategic and International Exchange Projects (Proyectos Estratégicos y de Intercambio Internacional) programs (FP-00038/218/415/459; FP-2002/089 and FP-2003/162), by the Consejo Nacional de Ciencia y Tecnología (CONACyT Fondos Sectoriales SALUD-2003-C01-102) and by the Fundación Ayúdame a Vivir pro Niños con Cáncer of El Salvador. We thank Veronica Yakoleff and Yolanda Castelazo for editing the manuscript. Figures and Tables ================== ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The incidence of leukemia in children aged 0--14 years in Mexico City from 1996 to 2000. ::: ------------------ --------- ----------------------------- ---------- --------------- ---------------- ---------------- ------------------ -------- ---------- ---------- ---------- ---------- ---------- 1996 to 2000 Number of cases by year Sex n AAIR \< 1 year 1--4 years 5--9 years 10--14 years 1996 1997 1998 1999 2000 ALL M 103 49.6\* 60.2 82.4 49.9 19.5 n 49 33 33 42 26 F 80 40.1\* 36.7 45.9 41.8 33.9 AAIR 62.3 39.6 43.1 52.1 29.5 T 183 44.9\* 48.8 64.6 45.9 26.6 AML M 13 6.3\* 0 10.5 6.9 3.0 n 9 4 11 7 12 F 30 15.0\* 0 16.5 14.4 17.0 AAIR 11.4 4.8 14.4 8.7 13.6 T 43 10.6\* 0 13.4 10.6 9.9 CML M 5 2.4\* 8.6 0 1.4 4.5 n 2 3 0 4 1 F 5 2.5\* 0 5.5 1.4 1.5 AAIR 2.5 3.6 0 5.0 1.1 T 10 2.5\* 4.4 2.7 1.4 3.0 UL M 1 0.5\* 0 0 1.4 0 n 1 0 0 1 0 F 1 0.5\* 0 0 0 1.5 AAIR 1.3 0 0 1.2 0 T 2 0.5\* 0 0 0.7 0.8 ------------------ --------- ----------------------------- ---------- --------------- ---------------- ---------------- ------------------ -------- ---------- ---------- ---------- ---------- ---------- AAIR: Average Annual Incidence Rate M = male; F = female; T = total; AAIR = average annual incidence rate per million. AML = Acute myeloid leukemia; ALL = Acute Lymphoid leukemia; CML: Chronic Myeloid Leukemia; UL = Unspecified leukemias.\* Age standardized rate per million children. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The incidence of the leukemias in children younger than 12 years in El Salvador from 1996 to 2000. ::: ------------------ --------- ----------------------------- --------------------------------- --------------- ---------------- ---------------- ------------------ ------------------------------- ---------- ---------- ---------- ---------- ---------- 1996 to 2000 Number of cases by year Sex n Rate over the entire period \< 1 year 1--4 years 5--9 years 10--11 years 1996 1997 1998 1999 2000 ALL M 162 36.1\* 4.9 49.6 36.3 20.1 n 54 64 61 60 63 F 140 32.3\* 15.1 47.3 23.0 29.7 Rate over the entire period 30.7 36.3 34.1 33.2 34.4 T 302 34.2\* 9.9 48.4 29.7 24.8 AML M 37 8.1\* 0 9.4 9.2 7.2 n 19 12 12 9 11 F 26 6.0\* 7.6 9.1 3.4 4.4 Rate over the entire period 10.8 6.8 6.7 5.0 6.0 T 63 7.1\* 3.7 9.2 6.3 5.8 CML M 3 0.6\* 0 0 1.1 1.4 n 0 0 1 3 2 F 3 0.6\* 0 0 1.1 1.5 Rate over the entire period 0 0 0.6 1.7 1.1 T 6 0.6\* 0 0 1.1 1.5 UL M 3 0.7\* 2.4 1.2 0 0 n 0 0 1 1 2 F 1 0.2\* 2.5 0 0 0 Rate over the entire period 0 0 0.6 0.5 1.1 T 4 0.5\* 2.5 0.6 0 0 ------------------ --------- ----------------------------- --------------------------------- --------------- ---------------- ---------------- ------------------ ------------------------------- ---------- ---------- ---------- ---------- ---------- AAIR: Average Annual Incidence Rate. M = male; F = female; T = total. AML = Acute myeloid leukemia; ALL = Acute Lymphoid leukemia; CML: Chronic Myeloid Leukemia UL = Unspecified leukemias. \* Age standardized rate per million children. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of annual age-adjusted rates of lymphoid leukemias per million for Mexican and Salvadoran children with those for Hispanic children from three U.S. Cancer Registries and for children from Costa Rica ::: *Leukemia type* *SEER 2001 All races(30) Mexico City, IMSS* *Texas(34) California(3) SEER 1992--1998 Hispanics(29) El Salvador* *Florida(4) Costa Rica(5) --------------------- ------------------------------- ------------------------- ----------------- --------------------- ------------------------------------- ------------------- ------------------ --------------------- Lymphoid leukemia 33.2 44.9 52.0 44.0 43.0 34.2\ 49.7 43.1 IMSS = Instituto Mexicano del Seguro Social. \* This rate is only for children 0--11 years old. (References) ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Comparison of the proportion of promyelocytic leukemias for Mexican children with Hispanic children from other countries and cities. ::: Adults Children ------------------------ ------------------- ----------------------------------- ---------------- -------------------------- ------------------------ -------------------------- ------------------------- -------------------------------------------------- *Leukemia type* *LAC-USC(37) LA County and AML study(37) Peru(38) Puebla, Mexico(39) California\ (3) *Puebla, Mexico(39) Mexico City, IMSS* *Range in non-Hispanic populations\\* (37) Promyelocytic leukemia 37.5% 24.3% 22.0% 20.0% 8.3% 30.0% 21.0% 3--13% LAC-USC = Los Angeles County-University of Southern California; LA: Los Angeles; AML = acute myeloid leukemia; IMSS = Instituto Mexicano del Seguro Social. \* This is a population-based study in which a frequency of 3.8%was found for non-Hispanic children. \\ This range is based on proportions found in series of cases. (References) :::	PubMed Central	2024-06-05T03:55:56.474387	2005-4-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090561/", "journal": "BMC Cancer. 2005 Apr 4; 5:33", "authors": [ { "first": "Juan Manuel", "last": "Mejía-Aranguré" }, { "first": "Miguel", "last": "Bonilla" }, { "first": "Rodolpho", "last": "Lorenzana" }, { "first": "Servando", "last": "Juárez-Ocaña" }, { "first": "Gladys", "last": "de Reyes" }, { "first": "María Luisa", "last": "Pérez-Saldivar" }, { "first": "Guadalupe", "last": "González-Miranda" }, { "first": "Roberto", "last": "Bernáldez-Ríos" }, { "first": "Antonio", "last": "Ortiz-Fernández" }, { "first": "Manuel", "last": "Ortega-Alvarez" }, { "first": "María del Carmen", "last": "Martínez-García" }, { "first": "Arturo", "last": "Fajardo-Gutiérrez" } ] }
PMC1090562	Background ========== Brain metastases represent a significant health care problem. An estimated 20--40% of cancer patients will develop multiple brain metastases \[[@B1]\], and 30--40% will develop a single metastasis \[[@B2]\] during the course of their illness. The prognosis for patients is generally poor, and treatment decisions are based on a combination of factors, including survival, quality of life, intracranial progression-free duration, response of brain metastases to treatment, symptom control, neurological function, and toxicity. These outcomes were considered in the systematic review that informed this provincial clinical practice guideline, which was initiated to summarize the evidence and to provide recommendations on the optimal management of brain metastases. The systematic review and meta-analyses, conducted as the initial step in formulating this practice guideline, are described in a companion document that has been submitted elsewhere for publication \[[@B3]\]. It is also currently undergoing review with the Cochrane Collaboration. Briefly, the evidence was broadly divided into studies aimed at management of a single brain metastasis \[[@B4]-[@B7]\] versus those aimed at management of multiple brain metastases \[[@B8]-[@B33]\] arising from cancer of any histology. The following interventions were compared in the randomized controlled trials included in the systematic review: for single brain metastasis, whole brain radiotherapy (WBRT) with or without surgery \[[@B4]-[@B6]\] and surgery with or without WBRT \[[@B7]\]; for multiple brain metastases, supportive care with or without WBRT \[[@B8]\], various altered dose-fractionation schedules \[[@B9]-[@B17]\], WBRT with or without radiosensitizers \[[@B18]-[@B23]\], chemotherapy with WBRT \[[@B24]-[@B28]\], and WBRT with or without radiosurgery \[[@B29]-[@B33]\]. Methods ======= Clinical practice guideline development --------------------------------------- This practice guideline was developed by Cancer Care Ontario\'s Practice Guidelines Initiative (PGI), using the methods of the Practice Guidelines Development Cycle \[[@B34]\]. The practice guideline report is a convenient and up-to-date source of the best available evidence on the role of radiation therapy in adult patients with brain metastases, developed through systematic reviews, evidence synthesis, and input from practitioners in Ontario. The report is intended to promote evidence-based practice. The PGI is editorially independent of Cancer Care Ontario and the Ontario Ministry of Health and Long-term Care. The PGI has a formal standardized process to ensure the currency of each guideline report. This process consists of the periodic review and evaluation of the scientific literature and, where appropriate, integration of this literature with the original guideline information. Evidence was selected and summarized by two members of the Supportive Care Guidelines Group (SCGG) and methodologists. Members of the SCGG disclosed potential conflict of interest information, reviewed the analysis of the evidence, and prepared draft recommendations. The membership of the SCGG includes palliative care physicians, radiation and medical oncologists, radiation therapists, psychiatrists, nurses, psychologists, an anesthetist, a surgeon, and methodologists. After reviewing the evidence, the SCGG reached consensus on draft recommendations. External review by Ontario practitioners was obtained through a mailed survey consisting of items that address the quality of the draft practice guideline report and recommendations and whether the recommendations should serve as a practice guideline. The efficacy of the practitioner feedback survey process has been previously described \[[@B35]\]. Final approval of the original guideline report was obtained from the Practice Guidelines Coordinating Committee (PGCC). Interpretation of the evidence ------------------------------ ### Single brain metastasis Two of the three trials using WBRT with or without surgical excision of a single brain metastasis detected an overall survival benefit favouring the addition of surgery. The trial that did not detect a benefit \[[@B4]\], however, included more patients with poorer performance status and a higher proportion of patients with extracranial disease as compared to the other two trials. The randomized trial by Patchell et al. \[[@B7]\] reported on the use of surgery with or without WBRT. A significant reduction in brain recurrence rates was detected in the surgery and WBRT arm, but there was no significant difference in overall survival. The methodologic quality of the studies was similar. However, the description of withdrawals and dropouts was variable. Only the Patchell trials \[[@B6],[@B7]\] required magnetic resonance imaging (MRI)-confirmed single metastasis. As such, those trials which relied on brain computed tomography (CT) may have included patients with multiple brain metastases rather than single. The benefit of adding surgery in these patients with truly multiple brain metastases may have been diminished. In the trials examining the use of surgery and WBRT for single brain metastasis, the WBRT doses were 3000 cGy/10 fractions daily \[[@B4]\], 4000 cGy/20 fractions given twice a day \[[@B5]\], 3600 cGy/12 fractions daily \[[@B6]\] and 5040 cGy/28 fractions daily \[[@B7]\]. The Radiation Therapy Oncology Group (RTOG) trial \[[@B31]-[@B33]\] randomized 164 patients to WBRT and radiosurgery boost versus 167 patients to WBRT alone. Overall, there was no improvement in overall survival. An improvement in one-year brain control rates was observed in the radiosurgery boost arm. That trial included a predefined hypothesis to detect a 75% median survival time improvement (80% statistical power) in patients with single brain metastasis. Median survival was 6.5 months in patients with single brain metastasis treated with radiosurgery boost as compared to 4.9 months in patients with single brain metastasis treated with WBRT alone, p = 0.0393. The evidence provided in the systematic review \[[@B3]\] suggests that surgical resection of a single brain metastasis in a patient with good performance status (Karnofsky Performance Status \[KPS\] ≥ 70) and stable or no extracranial disease improves overall survival. The addition of WBRT after surgical resection of a single brain metastasis decreases brain recurrence rates. Based on one randomized trial, the use of radiosurgery boost with WBRT was reported to improve survival as compared to WBRT alone in selected patients with single brain metastasis. ### Multiple brain metastases One randomized trial \[[@B8]\] examined the use of prednisone with or without WBRT. This was an older trial, with a small sample size of 48 patients, reported in the era prior to CT scanning. The diagnosis of brain metastases was based on outdated criteria; not contemporary CT or MRI criteria. The proportion of patients with improved performance status was similar in the steroid-alone and combined WBRT and steroid arms (63% and 61% respectively). The median survival of the steroid-alone arm was 10 weeks as compared to 14 weeks in the combined arm (p-value not stated). The methodologic quality of that study was poor. Sample size calculations were not described a priori, and a description of dropouts and withdrawals was not provided. Statistical analyses were not performed and therefore, the magnitude of benefit with the use of WBRT over supportive care alone remains unclear, particularly in patients with poor performance status and/or active extracranial disease. In several randomized controlled trials included in the systematic review \[[@B3]\], a significant benefit in terms of overall survival or symptom control was not detected with altered dose-fractionation schedules as compared with a standard dose-fractionation schedule of 3000 cGy in 10 fractions. The included studies were similar in methodologic quality. Details of randomization (e.g., blinding of randomization) were rarely provided, and complete follow-up was variable among the studies. None of the trials reported on the blinding of outcomes. Furthermore, none of the negative trials commented on confidence intervals or power calculations. A lack of sufficient high-quality evidence precludes recommendations on which treatment regimen(s) provide the greatest improvement in symptom control. In an attempt to improve the response of brain metastases to treatment, radiosensitizers have been added to WBRT. However, none of the five randomized trials \[[@B18]-[@B21],[@B23]\] detected a significant benefit in overall survival or brain metastases response (CR + PR). None of the trials examining the use of radiosensitizers were double-blind. However, the events review committee (ERC) in the gadolinium trial \[[@B22],[@B23]\] was blinded to treatment assignment and reviewed baseline and follow-up data. Based on subgroup analysis, there was a suggestion that recursive partitioning analysis (RPA) Class II lung cancer patients with brain metastases may benefit from the use of motexafin gadolinium and WBRT. This is being further studied in a phase III trial where patients with metastatic non-small cell lung cancer are randomized to WBRT with or without motexafin gadolinium. In a non-blinded study, Ushio \[[@B24]\] randomized patients with metastatic lung cancer to the brain to one of three groups (WBRT alone, WBRT + chloroethylnitrosoureas, or WBRT + chloroethylnitrosoureas + tegafur). No significant difference in overall survival was seen among the three groups. Brain response rates were significantly different between the WBRT-alone arm and the WBRT + chloroethylnitrosoureas + tegafur arm. However, 12 patients were excluded from the evaluation due to protocol violations, which may have skewed the results of the study given the small number of patients. Two patients died of probable side effects of chemotherapy. For metastatic small-cell lung cancer, Postmus \[[@B25]\] found no difference in overall survival in patients treated with teniposide alone versus teniposide and WBRT. Although the combined arm had higher brain response rates, there is no comparison with WBRT alone. That study showed that chemotherapy alone is inferior to the use of WBRT and chemotherapy for improved brain metastases response rates. However, it does not address the question as to whether WBRT alone is superior or equivalent to WBRT and chemotherapy for brain response and neuropsychological outcomes. For metastatic non-small cell lung cancer, Robinet \[[@B26]\] found no difference in overall survival with early versus delayed WBRT when given with chemotherapy. Delayed WBRT was given to intracranial non-responders to chemotherapy. That non-blinded study was powered to detect a 25% improvement in the six-month survival rate. Approximately 13% of patients were inevaluable for intracranial or extracranial response. However, withdrawals and drop-outs were described in terms of numbers and reasons per group. There was a 21% overall response (CR + PR) after two cycles of chemotherapy alone and 20% overall response to chemotherapy and early WBRT. Six-month survival was no different between the two arms. The results confirmed that chemotherapy alone may reduce the size of brain metastases from metastatic non-small cell lung cancer. The timing of WBRT in relation to chemotherapy did not affect survival. However, it was not possible to establish the optimal timing of WBRT when given concurrently with chemotherapy from the results of the Robinet trial \[[@B26]\]. Mornex \[[@B27]\] found no difference in cerebral response rates between combined fotemustine and WBRT versus fotemustine alone in patients with metastatic melanoma to brain. However, there was a significant difference in favour of the combined arm for time to cerebral progression. The most severe side effect was myelosuppression. Delayed grade 3--4 neutropenia occurred in 46% of patients in the fotemustine alone arm and 35% in the combined arm. Delayed grade 3--4 thrombocytopenia occurred in 44% of patients in the fotemustine-alone arm and 38% in the combined arm. That trial did not address the question of whether WBRT alone is superior or equivalent to WBRT and fotemustine in terms of therapeutic benefit and toxicity in patients with metastatic melanoma to the brain. Antonadou \[[@B28]\] found no difference in overall survival for patients treated with WBRT and temozolamide chemotherapy versus WBRT alone. However, an improved brain response rate was seen in the combined arm. Those results were published in abstract form. Further trials are needed to confirm a benefit in brain control with the addition of chemotherapy to WBRT. Three trials \[[@B29],[@B30],[@B33]\] reported on the use of radiosurgery in addition to WBRT. Only one of those trials found a benefit to the use of WBRT in addition to radiosurgery for selected patients with 2--4 brain metastases; however, the trial was small (n = 27 patients), and the results were reported early at 60% accrual. The rate of local brain failure was 100% after WBRT and 8% in those treated with boost radiosurgery. Furthermore, the 100% recurrence rate in the WBRT arm was unusually high. There was no significant difference in overall survival, 7.5 months for WBRT and 11 months for patients in the WBRT and boost radiosurgery arm, p = 0.22. As previously mentioned, the RTOG trial \[[@B31]-[@B33]\] randomized 164 patients to WBRT and radiosurgery boost versus 167 patients to WBRT alone. No improvement in overall survival was detected. In patients with single brain metastasis treated with radiosurgery boost median survival was 6.5 months as compared to 4.9 months in patients with single brain metastasis treated with WBRT alone, p = 0.0393. Another trial published in abstract form \[[@B30]\] examined the use of Gamma knife radiosurgery (GK RS), WBRT, or both in the treatment of 1--3 brain metastases. There was no difference in overall survival. Local control rates were superior for the GK RS and GK RS + WBRT arms. Subgroup analysis for patients with single brain metastasis in this latter study \[[@B30]\] was not reported. Thus, the use of radiosurgery appears to improve 1-year local control of brain metastases when used in conjunction with WBRT in selected patients. There is Level 1 evidence (three trials) that overall survival is not improved with the addition of radiosurgery boost to WBRT as compared to WBRT. The optimal timing of radiosurgery has not been elucidated. The question of whether radiosurgery should be used as a boost treatment with WBRT, at the time of relapse after WBRT, or used alone, reserving WBRT for future extensive brain relapse, remains unanswered. Supportive Care Guidelines Group consensus ------------------------------------------ Originally proposed as a guideline topic for Cancer Care Ontario\'s Neuro-Oncology Disease Site Group (NDSG), in 2002 it was decided that the guideline would be developed under the auspices of the SCGG since the view was to maintain a palliative focus. A separate practice guideline on the management of single brain metastases was developed by the NDSG and is consistent with the current guideline. Both the SCGG and NDSG reviewed all draft versions of the guideline. Modifications were made at various stages as per the groups\' feedback and the final version was approved in February 2004. Results ======= Draft recommendations --------------------- Based on the evidence described above, the SCGG, with the opinions of the NDSG, formulated the following [draft]{.underline}recommendations, which were subsequently sent out for external review: ### Target population These recommendations apply to adult patients with a clinical and radiographic diagnosis of brain metastases (single or multiple) arising from cancer of any histology. #### Radiotherapy and surgery for single brain metastasis • Surgical excision is recommended, in addition to WBRT, for patients with good performance status, minimal or no evidence of extracranial disease, and a surgically accessible single brain metastasis (single or multiple) arising from cancer of any histology. • Postoperative WBRT should be used to improve brain control for patients who have undergone resection of a single brain metastasis. #### Radiotherapy for multiple brain metastases • Whole brain radiotherapy is the recommended volume of treatment for multiple brain metastases. Commonly used dose-fractionation schedules are 3000 cGy in 10 fractions or 2000 cGy in 5 fractions. • There are no advantages of other altered dose-fractionation WBRT schedules in terms of overall survival or neurologic function. • The use of radiosensitizers is not recommended outside research studies. • The optimal use of radiosurgery in the treatment of brain metastases remains to be defined. In patients with one to three brain metastases (less than 3 cm in size) and limited or controlled extracranial disease, radiosurgery may be considered to improve local control either as boost therapy with WBRT or at the time of relapse after WBRT failure. #### Chemotherapy and whole brain radiotherapy • The use of chemotherapy as primary therapy for brain metastases (with WBRT used for intracranial non-responders) or the use of chemotherapy with WBRT to treat brain metastases remains experimental. #### Supportive care and whole brain radiotherapy • Supportive care alone without WBRT is an option for patients with poor performance status or widely disseminated progressive cancer. ### Qualifying statements • The number of patients included in the two trials comparing 3000 cGy in 10 fractions versus 2000 cGy in 5 fractions for multiple brain metastases was small. • In the trials examining the use of surgery and WBRT for single brain metastasis, the WBRT doses were 3000 cGy/10 fractions daily, 4000 cGy/20 fractions given twice daily, 3600 cGy/12 fractions daily, and 5040 cGy/28 fractions daily. As such, the use of 2000 cGy/5 fractions of WBRT has not been studied directly in this scenario. External review process -- Ontario practitioner feedback -------------------------------------------------------- Feedback on the draft practice guideline report was obtained through a mailed survey of 246 practitioners in Ontario (26 neurosurgeons, 137 medical oncologists, and 83 radiation oncologists). The survey consisted of items evaluating the methods, results, and interpretation of the evidence and whether the draft recommendations should be approved as a practice guideline. Written comments were invited. The SCGG reviewed the results of the survey. Results of practitioner feedback -------------------------------- One hundred nine responses were received out of the 246 surveys sent (44% response rate). A summary of the results is provided in Table [1](#T1){ref-type="table"}. Of the practitioners who responded, 85 indicated that the report was relevant to their clinical practice and completed the survey. The survey results indicated that 96% of respondents agreed with the interpretation of the evidence and 94% agreed with the draft recommendations as stated. Ninety-two percent of respondents agreed that the report should be approved as a practice guideline. Twenty-three respondents (27%) also provided written comments. The final guideline recommendations, which appear at the end of this report, were modified in accordance with the suggestions from the external reviewers and were subsequently approved by Cancer Care Ontario\'s Practice Guidelines Coordinating Committee. Practice Guidelines Coordinating Committee approval process ----------------------------------------------------------- The practice guideline report was circulated to the PGCC for review and approval. Four of eight members of the PGCC completed and returned ballots. Three of these members approved the practice guideline report as written, while one member approved the guideline and provided a suggestion for consideration by the SCGG. The suggestion was to revise the wording of the recommendation for single brain metastasis to \"considered\" rather than \"recommended\" as the evidence for benefit is not compelling. The SCGG agreed with the suggestion and modified the guideline accordingly. Conclusions =========== Guideline recommendations ------------------------- For adult patients with a clinical and radiographic diagnosis of brain metastases (single or multiple) arising from cancer of any histology (except for choriocarcinoma and other germ cell tumours, and hematologic malignancies), we recommend that: ### Radiotherapy and surgery for single brain metastasis • Surgical excision should be considered for patients with good performance status, minimal or no evidence of extracranial disease, and a surgically accessible single brain metastasis amenable to complete excision. • Postoperative WBRT should be considered to reduce the risk of tumour recurrence for patients who have undergone resection of a single brain metastasis. • Radiosurgery boost with WBRT may also improve survival in select patients with unresectable single brain metastases. ### Radiotherapy for multiple brain metastases • The whole brain should be irradiated for multiple brain metastases. Commonly used standard dose-fractionation schedules are 3000 cGy in 10 fractions or 2000 cGy in 5 fractions. • Altered dose-fractionation WBRT schedules have not demonstrated any advantages in terms of overall survival or neurologic function relative to more commonly used fractionation schedules. • The use of radiosensitizers is not recommended outside research studies. • In select patients with up to four brain metastases (up to 4 cm in size) and limited or controlled extracranial disease, radiosurgery may be considered as a boost therapy with WBRT to improve local tumour control. Radiosurgery boost may also improve survival in select patients with unresectable single brain metastases. ### Chemotherapy and whole brain radiotherapy • The use of chemotherapy as the primary therapy for brain metastases (with WBRT used for those whose intracranial metastases fail to respond) or the use of chemotherapy with WBRT to treat brain metastases remains experimental. ### Supportive care and whole brain radiotherapy • Supportive care alone without WBRT is an option (for example, in patients with poor performance status and progressive extracranial disease). However, there is a lack of Level 1 evidence to guide practitioners as to which subsets of patients with brain metastases should be managed with supportive care alone without WBRT. To support the application of these recommendations in clinical practice, the following qualifying statements should be considered: The number of patients included in the two trials comparing 3000 cGy in 10 fractions versus 2000 cGy in 5 fractions for multiple brain metastases was small. In the trials examining the use of surgery and WBRT for single brain metastasis, the WBRT doses were 3000 cGy in 10 fractions daily, 4000 cGy in 20 fractions given twice daily, 3600 cGy in 12 fractions daily, and 5040 cGy in 28 fractions daily. As such, the use of 2000 cGy in 5 fractions of WBRT has not been studied directly in this scenario. The results of the studies may not be generalizable to all tumour types. The majority of the patients in the studies (except the chemotherapy studies) had lung, breast, or colorectal cancer primaries. List of abbreviations used ========================== cGy, centigray(s); cm, centimeter(s); CR, complete response; CT, computed tomography; ERC, events review committee; GK RS, gamma knife radiosurgery; Gy, gray(s); KPS, Karnofsky performance status; met, metastasis(es); MRI, magnetic resonance imaging; NDSG, Neuro-Oncology disease site group; PGCC, Practice Guidelines Coordinating Committee; PGI, Practice Guidelines Initiative; PR, partial response; RPA, recursive partitioning analysis; RTOG, Radiation Therapy Oncology Group; SCGG, Supportive Care Guidelines Group; WBRT, whole brain radiotherapy. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MT was the lead author responsible for designing and conducting the systematic review of the literature and the meta-analyses that informed the practice guideline, and for drafting and modifying the practice guideline report. MT is a member of the Supportive Care Guidelines Group and a Radiation Oncologist at the Toronto-Sunnybrook Regional Cancer Centre. NL conducted literature searches and drafted and edited the guideline report during its various stages of development. NL conducted duplicate data extraction and meta-analyses and coordinated input from members of the SCGG. NL updated the literature search, incorporated new data, conducted the practitioner feedback survey, and coordinated approval of the guideline by the PGCC. RW reviewed all drafts of the guideline report and made major contributions to performing the meta-analyses that informed the practice guideline, and provided extensive input to the guideline as a radiation oncologist and methodologist. RW is co-Chair of the Supportive Care Guidelines Group. Members of the SCGG provided feedback on all draft guideline reports. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/34/prepub> Acknowledgements ================ The Supportive Care Guidelines Group would like to thank Drs. Tsao and Wong and Ms. Lloyd for taking the lead in drafting and revising this practice guideline report. Many thanks also to Drs. Normand Laperriere, Eileen Rakovitch, and Edward Chow, for their contributions to the systematic review that informed this practice guideline and to the members of Cancer Care Ontario\'s Neuro-Oncology Disease Site Group for reviewing draft versions of the practice guideline report. For a complete list of Supportive Care Guidelines Group members, please visit the Cancer Care Ontario Web site at <http://www.cancercare.on.ca/access_PEBC.htm>. Figures and Tables ================== ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Practitioner responses to eight items on the practitioner feedback survey. ::: Item Number (%) ----------------------------------------------------------------------------------------------------------------------------------- --------------------------- ------------ ----------------------------------- The rationale for developing a clinical practice guideline, as stated in the \"Choice of Topic\" section of the report, is clear. 82 (98) 1 (1) 1 (1) There is a need for a clinical practice guideline on this topic. 70 (83) 12 (14) 2 (2) The literature search is relevant and complete. 77 (94) 5 (6) 0 The results of the trials described in the report are interpreted according to my understanding of the data. 81 (96) 3 (4) 0 The draft recommendations in this report are clear. 81 (96) 1 (1) 2 (2) I agree with the draft recommendations as stated. 79 (94) 2 (2) 3 (4) This report should be approved as a practice guideline. 77 (92) 6 (7) 1 (1) If this report were to become a practice guideline, how likely would you be to make use of it in your own practice? Very likely or likely Unsure Not at all likely or unlikely 57 (68) 6 (7) 21 (25) :::	PubMed Central	2024-06-05T03:55:56.478558	2005-4-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090562/", "journal": "BMC Cancer. 2005 Apr 4; 5:34", "authors": [ { "first": "May N", "last": "Tsao" }, { "first": "Nancy S", "last": "Lloyd" }, { "first": "Rebecca KS", "last": "Wong" } ] }
PMC1090563	Background ========== During normal organ development apoptosis provides an efficient mechanism whereby unwanted cells are \"discretely\" eliminated. The functions that ensure proper activation of apoptosis are somehow lost during tumorigenesis, allowing cancer cells to proliferate indefinitely and in an uncontrolled fashion\[[@B1],[@B2]\]. Apoptosis is triggered by several stimuli including hypoxia, radiation-induced DNA damage, oxidative stress, and lack of attachment. The term anoikis defines the type of apoptosis induced after proper adherence to the extracellular matrix (ECM) is denied\[[@B3],[@B4]\]. Attachment to the ECM is mainly mediated by integrins; a family of heterodimeric transmembrane receptors composed of an alpha and a beta chain. In response to physiological clues, bidirectional integrin signaling mediates cell differentiation, proliferation, homing, migration and survival\[[@B5],[@B6]\]. Integrins lack kinase domains; they signal by associating in complexes with other mediators such as FAK, ILK, Src, Shc, Syk, and paxillin \[[@B7]-[@B13]\]. Anoikis resistant tumor cells have circumvented the death signals generated by the lack of attachment affording them increased survival times while migrating to secondary sites. Thus resistance to anoikis has been regarded as a crucial step during tumorigenesis \[[@B14]-[@B16]\]. Our previous work has shown that human osteosarcoma cells, SAOS, are sensitive to anoikis\[[@B17]\]. However, anoikis resistance can be driven in originally sensitive clones by alternating culture cycles under adhered and suspended conditions. This resistant phenotype is stable and indicates that the processes of de-adhesion or exposure to a non-adhesive environment acts as a driving force towards anoikis resistance\[[@B18]\]. While the precise role of anoikis resistance in osteosarcoma progression is still unclear, chemoresistance continues to be an important problem in the clinic. Despite significant advances in the treatment of osteosarcoma, the prognosis of patients with metastasis at presentation remains poor, with an overall survival of 55% after aggressive chemotherapy and surgery\[[@B19],[@B20]\]. Historically, resistance to chemotherapy has been attributed to the overexpression of genes encoding cellular efflux pumps\[[@B21]\]. Recent studies have shown that the action of many anti-cancer agents results in apoptosis, therefore alterations in the apoptotic pathway may also confer multidrug resistance \[[@B22]-[@B24]\]. Since the general acquisition of apoptosis resistance would affect both de-adhesion and chemotherapy-induced cell death, we investigated whether acquisition of anoikis resistance conferred general resistance to other apoptotic inducers or was independent of these other apoptotic pathways. Methods ======= Cell culture and reagents ------------------------- The parental human osteosarcoma cell lines SAOS-2 (SAOSp) and TE-85 (TE-85p) were obtained from the American Type Culture Collection (Manassas, VA). SAOSp and TE-85p cells were maintained in Eagle\'s MEM (BioWhittaker, Walkersville, MD), supplemented with 10% fetal bovine serum (BioWhittaker, Walkersville, MD), 2 mM L-glutamine, 1 mM sodium pyruvate and non-essential amino acids (Sigma, St. Louis, MO). Anoikis resistant SAOS (SAOSar) and TE-85 (TE-85ar) cells were generated by sequential cycles of culture on untreated (adhered) and poly-HEMA treated cell culture wells (suspended)\[[@B18]\]. The resulting variants were maintained in culture under adhered conditions. Poly-HEMA was prepared by dissolving it in 95% ethanol to a concentration of 50 mg/ml. Poly-HEMA was added to cell culture wells at a density of 5 mg/cm^2^and allowed to dry overnight, under sterile conditions in a laminar flow hood. Etoposide, vinblastine and paclitaxel were purchased from Sigma, St. Louis, MO. Cisplatin was purchased from Bristol-Myers Squibb Company, Princeton, NJ. Adriamycin was purchased from GensiaSicor™ Pharmaceuticals, Irvine, CA. In vitroLD~50~for each agent was determined. One in vitroLD~50~is defined as the dose required to induce apoptosis in approximately 50% of the cells in 24 hr of culture. Live/Dead cytotoxicity assay ---------------------------- SAOSp and SAOSar were cultured under suspended conditions (poly-HEMA treated cell culture wells) for 24 h. After culture cells were washed with PBS, and pellets resuspended in 250 μl of PBS containing 2 μM Calcein-AM and 8 μM ethidium homodimer-1 (Molecular Probes, Eugene, OR). Cells were incubated at room temperature for 15 minutes and visualized using a fluorescent inverted microscope. Apoptosis analyses ------------------ Cell cycle/apoptosis analyses were performed using propidiumiodide (PI) staining with subsequent FACS analysis. 5 × 10^5^cells/well were cultured either on plastic or poly-HEMA treated 6-well tissue culture plates with or without the metabolic inhibitors and drugs for 24 hrs at 37°C in a 5% CO2 atmosphere. After incubation, adherent cells were detached with trypsin (0.5% trypsin/0.1% EDTA in PBS). Detached and suspended cells were harvested in complete EMEM medium and centrifuged at 500 g for 10 min. Pellets were washed with PBS and fixed with ice cold 75% ethanol overnight at 4°C. After fixation, cells were washed with PBS and stained with 500 μl of PI solution (50 μg/ml in PBS) containing 25 μg/ml of RNase. Cells were incubated at 37°C for 30 min and analyzed by flow cytometry on an Epics Profile flow cytometer (Coulter, Miami, FL). Apoptosis was also assayed by Annexin-FITC/PI staining following manufacturer instructions (Trevigen, Inc. Gaithersburg, MD). Briefly, treated or untreated cells were collected and washed in cold PBS. Cells were incubated for 15 min at room temperature in the presence of 1 μl Annexin V-FITC, 1 μl of propidium iodide and 98 μl of 1x binding buffer (all reagents provided by the manufacturer). After incubation, 400 μl of 1X binding buffer was added to each tube, and cells were analyzed by flow cytometry. Results ======= Parental human osteosarcoma SAOS-2 cells (SAOSp) undergo apoptosis after adherence to the ECM is denied (anoikis) by culture in poly-HEMA treated cell culture wells. An anoikis resistant subline (SAOSar) has been generated after sequential cycles of culture under suspended and adhered conditions. This stable phenotype is not the result of mere selection of pre-existing anoikis resistant sub-populations, since anoikis resistant cells can be derived from anoikis sensitive clonal populations\[[@B18]\]. Fig [1](#F1){ref-type="fig"} panel A shows the results of a Live/Dead™ assay of SAOSp and SAOSar cells cultured in poly-HEMA coated wells for 24 hr. After incubation, cells were stained with a mixture of calcein-AM and ethidium homodimer-1 (see Materials section). Intracellular esterases in live cells convert non-fluorescent cell-permeant calcein-AM to green fluorescent calcein. By contrast, in non-viable cells with damaged membranes, non-permeant ethidium homodimer-1 enters the cells, and by binding to nucleic acids the ethidium homodimer-1 produces a bright red fluorescence. Cells were viewed under a fluorescent inverted microscope using a longpass filter in order to simultaneously visualize both green and red fluorescence. Panel A shows a larger fraction of the parental SAOSp cells stained red (dead), in comparison with the anoikis resistant SAOSar cells in which the majority of the cells stained green (live). Under control adherent conditions both SAOS populations were uniformly alive and stained green (data not shown). For the quantitative analysis of apoptosis, two assays were used, cell cycle analysis and a membrane quality assay. Cell cycle was analyzed by staining DNA with PI and determining by flow cytometry the percentage cells with sub-G/0 content. The membrane quality was assessed by utilizing Annexin staining to identify phosphatidylserine and PI to monitor membrane integrity. Figure [1B](#F1){ref-type="fig"} shows a higher percentage of SAOSp cells in sub-G/0 phase, indicative of apoptosis (33.3%) than of SAOSar cells (8.37%) after culture in poly-HEMA treated plates for 24 h. The same was observed after staining with Annexin V-FITC/PI and flow cytometry. Figure [1C](#F1){ref-type="fig"} shows higher percentage of SAOSp cells (32.6%) than of SAOSar cells (6.09%) stained positive for Annexin V (top and bottom right quadrants) representative of apoptosis as well. Thus, SAOSar cells resist apoptosis after attachment to ECM is denied (anoikis). We have previously shown that SAOSp and SAOSar cells attached to the ECM are equally sensitive to induction of apoptosis and die after treatment with staurosporine, cycloheximide and hydrogen peroxide\[[@B18]\]. We hypothesized that even though the apoptotic machinery was intact while cultured under adhered conditions, once the anoikis resistant SAOSar cells were detached from the ECM, anti-apoptotic mediators could be activated resulting in a more generalized resistance to apoptosis. Therefore, to reexamine the sensitivity of anoikis resistant cells to other apoptotic stimuli, SAOSar cells were cultured under non-adherent conditions and exposed to staurosporine, cycloheximide or hydrogen peroxide. As shown in Fig. [2](#F2){ref-type="fig"}, untreated SAOSar cells are resistant to apoptosis when placed in nonadherent conditions, whereas staurosporine, cycloheximide, or hydrogen peroxide treatment of SAOSar cells results in apoptosis. These data suggest that the mechanisms conferring resistance to anoikis do not protect SAOSar cells from apoptosis induced by other stimuli. Since the mechanism of action of certain chemotherapy agents can result in apoptosis we tested whether anoikis resistant SAOS cells were more resistant to chemotherapy-induced apoptosis. In vitroLD~50~for chemotherapy agents etoposide, adriamycin, vinblastine, cisplatin and paclitaxel was determined for both SAOSp and SAOSar cultured under adhered conditions. Similar doses of the agents used were required to induce apoptosis in ≈ 50% of SAOSp and SAOSar cells cultured under adhered conditions for 24 h (Fig [3](#F3){ref-type="fig"}). We then tested whether culture under suspended conditions would have a chemoprotective effect in anoikis resistant SAOSar cells. SAOSp and SAOSar cells placed in suspended conditions (poly-HEMA coated cultured wells) were treated with in vitroLD~50~of etoposide, adriamycin, vinblastine, cisplatin or paclitaxel. Apoptosis was assayed by PI and Annexin V-FITC/PI staining followed by flow cytometry analyses. Figure [4A](#F4){ref-type="fig"} shows that the percent of untreated SAOSp cells in the sub-G/0 phase representative of apoptosis is significantly increased in comparison to untreated SAOSar when cells are placed in suspension conditions. The chemotherapeutic agents do not appear to have an additive effect. By contrast, the untreated anoikis resistant SAOSar cells remain viable when placed in suspension but the cells retain their sensitivity to the chemotherapeutic agents and undergo apoptosis. This was corroborated by Annexin V-FITC/PI staining and flow cytometry analyses. Figure [4B](#F4){ref-type="fig"} shows a higher percentage of untreated SAOSp cells positive for Annexin V/FITC staining (top and bottom right quadrants) than of untreated SAOSar cells after culture under suspended conditions for 24 h. Chemotherapy-treated suspended SAOSp and SAOSar showed similar percentages of cells stained with Annexin V-FITC indicative of apoptosis. A second osteosarcoma cell line, TE-85, was used to determine if chemotherapy sensitivity despite anoikis resistance was unique to SAOS-2 cells or represented a more generalized phenomenon among osteosarcoma. Anoikis resistant TE-85 cells (TE-85ar) were generated following the same procedure used for generating the anoikis resistant SAOSar cells. Suspended anoikis sensitive (TE-85p) and anoikis resistant (TE-85ar) cells were treated with the same doses of the same drugs and apoptosis was measured 24 hr later by PI and Annexin V-FITC/PI staining followed by flow cytometry analyses. As with SAOS-2 cells, untreated TE-85ar cells remained viable after culture in suspension for 24 hr but retained their sensitivity to chemotherapy-induced apoptosis at similar levels that of TE-85p cells. No significant differences among the percentages of apoptotic TE-85p or TE-85ar suspended cells were found after chemotherapy treatment, either by PI (figure [5A](#F5){ref-type="fig"}) or by Annexin V-FITC/PI (figure [5B](#F5){ref-type="fig"}) staining. These results indicate that resistance to anoikis does not confer resistance to these chemotherapeutic agents, and that this trend is not unique to SAOS-2 osteosarcoma cells but also applies to TE-85 osteosarcoma cells. Discussion ========== During tumorigenesis the delicate balance between survival and cell death is altered. Thus, cancer cells are able to survive under adverse conditions that normally would trigger apoptosis such as hypoxia, low glucose, and lack of attachment. Until recently it was thought that resistance to chemotherapy was due to mechanisms that prevented the intake of the drug or the presence of intracellular detoxificants. Since the discovery that drug-mediated cell death can be the result of physiological processes such as apoptosis, mitotic catastrophe and cellular senescence, resistance to chemotherapy has been linked to alterations in the pathways mediating such processes. \[[@B22]-[@B26]\]. Resistance to detachment-induced apoptosis (anoikis) is known as a important step during metastasis by affording tumor cells increased survival times while migrating to secondary sites. However, the relationship between anoikis resistance and chemotherapy response remains to be elucidated. Previously, we have shown that anoikis resistance can be induced in anoikis sensitive human osteosarcoma cells, SAOS-2, by exposure to culture in suspension (poly-HEMA treated culture wells). We also demonstrated that oxidative damage (H~2~O~2~), inhibition of protein synthesis (cycloheximide) or inhibition of calcium-dependent protein kinases (staurosporine) resulted in apoptosis of adherent SAOS-2 cells regardless of their anoikis resistant phenotype\[[@B18]\]. This suggested that under adhered conditions the apoptotic machinery was intact. In this study, we tested whether the anti-apoptotic mechanisms that rendered the cells anoikis resistant would be activated upon detachment from the ECM, resulting in a more generalized resistance to apoptosis and hence to chemotherapy. For instance, in acute myelogeneous leukemia interactions between α4β1 integrins and fibronectin activate the PI3-K/Akt pathway resulting in resistance to both anoikis and to treatment with daunorubicin or AraC\[[@B27]\]. By contrast, our data suggested that despite the resistant phenotype and the suspended conditions, apoptosis can still be induced by oxidative damage, inhibition of protein synthesis or inhibition of calcium-dependent protein kinases in anoikis resistant SAOSar cells. Furthermore, anoikis resistant SAOSar cells are equally sensitive to chemotherapy-induced apoptosis when compared to anoikis sensitive SAOSp cells under either suspended or adhered culture conditions. Similar results were obtained after anoikis sensitive and anoikis resistant TE-85 cells were treated with the same agents. The chemotherapeutic agents tested vary widely in their mode of action; etoposide, adriamycin and cisplatin cause DNA damage by forming DNA adducts or by inhibiting topoisomerase II resulting in DNA breaks. Vinblastine and paclitaxel target the microtubules and are known as \"spindle poisons\", however their mode of action is different. Vinblastine binds to tubulin dimers preventing the formation of microtubules and paclitaxel binds to the microtubules inducing mitotic arrest by excessively stabilizing them. Regardless of their mode of action, under adhered conditions the in vitroLD~50~for etoposide, adriamycin, vinblastine, cisplatin or paclitaxel was similar for both SAOSp and SAOSar cells. Similar levels of apoptosis were found after suspended SAOSp and SAOSar cells were treated with the same doses of the different agents. The same was observed after suspended TE-85p and TE-85ar cells were treated with the same agents. These data suggest that acquisition of anoikis resistance does not necessarily render osteosarcoma cells resistant to other apoptotic stimuli including chemotherapy. The specific mechanisms involved in anoikis resistance are not completely understood. Overexpression of oncogenes such as ras, raf, racand srcas well as the deletion of tumor suppressor genes such as PTENand p53have been associated with resistance to anoikis \[[@B28]-[@B30]\]. Recently it was reported that TrkB, a nerurotrophic tyrosine kinase receptor, is able to suppress anoikis of non-malignant epithelial cells by activating the PI3-K/Akt pathway\[[@B31]\]. Activated Akt exerts its anti-apoptotic effect by modulating the activity of mediators that are directly involved in the apoptotic cascade or by regulating the transcription of pro- and anti-apoptotic genes \[[@B32]-[@B34]\]. Normal breast epithelial cells expressing constitutively active Akt1 lose their sensitivity to anoikis and become resistant to apoptosis after treatment with cisplatin and mitoxantrone\[[@B35]\]. Likewise, in pancreatic adenocarcinoma cells increased activity of Akt in response to overexpression of carcinoembryionic antigen-related cell adhesion molecule (CEACAM)6 results in resistance to both anoikis and gemcitabine treatment\[[@B36]\]. In these systems, Akt protects cells against death induced after DNA damage as well as death induced by anoikis. In our osteosarcoma model, it is clear that resistance to anoikis is independent of resistance to other apoptotic inducers. We have recently found that activation of the PI3-K/Akt pathway is important during anoikis resistance (Díaz-Montero CM and McIntyre BW, unpublished data). Since these cells are still sensitive to other apoptosis inducing agents, upregulation of Akt is not sufficient to confer resistance to all apoptotic stimuli. Furthermore, in breast cancer cell lines SKBR-3 and MDA-MB-453, anoikis resistance can be restored by induction of ILK, independently of Akt activity\[[@B37]\]. These different studies suggest that the mechanisms that confer resistance to anoikis and/or chemotherapy may be unique to each type of malignancy. Thus it can be argued that anti-cancer agents that target apoptosis will be less effective against malignancies in which the pathways for resistance to both apoptosis and anoikis overlap. Likewise, therapies that can inhibit a common apoptosis and anoikis resistance pathway could be a potent new anti-cancer treatment. Our work suggests that in at least two human osteosarcoma cell lines, resistance to anoikis and apoptosis are regulated by different mediators, and might explain why anoikis resistant cells are still vulnerable to other apoptosis-inducing stimuli. This is a fortunate situation except in those cases where the tumor cells have detached from the primary sites and survive because of the induction of anoikis resistance. In many cases, tumor cells that lose normal adhesive constraints will enter cell cycle arrest and thus be refractory to many chemotherapeutic reagents. In this scenario, the ability to target the anoikis resistance pathway may provide a new approach for chemotherapy. In conclusion, in order to effectively target anoikis resistant/migrating metastatic tumor cells the apoptotic pathways that are altered and the ones that remain normal need to be identified. Only then, agents with specific action against the mediators involved during the relevant disease stage, i.e. primary vs. metastatic, will become available as more efficient treatment strategies. Conclusion ========== The development of newer and more efficacious treatments against cancer will require the understanding of the mechanisms behind apoptosis and/or anoikis resistance in a disease-specific way. We have shown that acquisition of resistance to anoikis in human osteosarcoma cells does not necessarily result in a generalized resistance to all apoptotic stimuli. Thus in this particular system, targeting the pathways involved might control the spread of anoikis resistant/migrating metastatic cells. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= CMDM designed and executed the experiments, collected and analyzed the data and drafted the manuscript. BWM designed experiments, interpreted the results, and edited the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/39/prepub> Acknowledgements ================ The authors are grateful to Karen Ramirez for flow cytometry analyses and to Ricky Rojas and James Wygant for technical assistance. This work was supported by the NIH grants T32DE015355-01, CA 62596 and CA 166672 Cancer Center Support Core Grant, a Kleberg Fund for Innovative Research Institutional Grant and by the Onstead Foundation for Osteosarcoma Research. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### SAOSp cells undergo anoikis after culture under suspended conditions.After culture under suspended conditions for 24 h SAOSp and SAOSar cell viability was assayed using a Live/Dead™ assay. Panel A shows a larger fraction of SAOSp cells stained red (dead) than of SAOSar cells (majority stained green). Apoptosis was quantitated by PI (panel B) or Annexin V-FITC/PI (panel C) staining followed by flow cytometry analyses. Panel B shows a higher percentage of SAOSp (33.3%) cells than of SAOSar (8.37%) in the sub-G/0 phase representative of apoptotic cells. Similar results are shown in panel C, in which a higher percentage of SAOSp cells (32.6%) than of SAOSar (6.09%) stained positive for Annexin V (top and bottom right quadrants) representative of apoptotic cells. Results shown are representative of three independent experiments. ::: ![](1471-2407-5-39-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Apoptosis can be induced in SAOSar cells while cultured under suspended conditions.Apoptosis was triggered in suspended anoikis-resistant SAOSar cells by staurosporine-induced inhibition of calcium-dependent protein kinases, by cycloheximide-induced inhibition of protein synthesis and by H~2~O~2~-induced oxidative damage. Percentage apoptosis was determined by flow cytometry analyzes after staining with propidium iodide. The percentage of cells in the sub-G/0 phase representative of apoptotic cells is marked on each histogram. Results shown are representative of three independent experiments. ::: ![](1471-2407-5-39-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Chemosensitivity of SAOSp and SAOSar cells.In vitroLD~50~for chemotherapy agents etoposide (50 μM), adriamycin (1 μg/ml), vinblastine (1 μg/ml), cisplatin (10 μg/ml) and paclitaxel (100 ug/ml) were determined for SAOSp and SAOSar cells cultured under adhered conditions. Similar chemosensitivity was observed for all agents tested by adhered SAOSp and SAOSar cells. Apoptosis was determined by PI staining followed by flow cytometry analyses. Error bars indicate the standard deviation of three independent experiments. ::: ![](1471-2407-5-39-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Resistance to anoikis does not confer chemoresistance in SAOS cells.Suspended SAOSp and SAOSar cells were treated with 1 in vitroLD~50~of etoposide, adriamycin, vinblastine, cisplatin or paclitaxel. In Panel A, similar percentages of SAOSp or SAOSar cells in the sub-G/0 phase representative of apoptotic cells was found after incubation with the indicated agents for 24 h under suspended conditions. Error bars indicate the standard deviation of three independent experiments. Panel B shows similar percentages of Annexin V-FITC positive SAOSp or SAOSar suspended cells (top and bottom right quadrants) after the same treatments. Data shown is representative of three independent experiments. ::: ![](1471-2407-5-39-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Resistance to anoikis does not confer chemoresistance in TE-85 cells.Suspended TE-85p and TE-85ar cells were treated with etoposide, adriamycin, vinblastine, cisplatin or paclitaxel for 24 h and apoptosis was assayed by PI (panel A) or Annexin V-FITC/PI (panel B) staining followed by flow cytometry analyses. In panel A similar percentages of TE-85p or TE-85ar cells in the sub-G/0 phase representative of apoptotic cells were found after incubation with the indicated agents. Error bars indicate the standard deviation of three independent experiments. Panel B shows similar percentages of Annexin V-FITC positive TE-85p or TE-85ar suspended cells (top and bottom right quadrants) after the same treatments. Data shown is representative of three independent experiments. ::: ![](1471-2407-5-39-5) :::	PubMed Central	2024-06-05T03:55:56.480727	2005-4-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090563/", "journal": "BMC Cancer. 2005 Apr 13; 5:39", "authors": [ { "first": "C Marcela", "last": "Díaz-Montero" }, { "first": "Bradley W", "last": "McIntyre" } ] }
PMC1090564	Background ========== High blood pressure or hypertension, \'a silent-killer\' condition, is now the most common chronic condition, affecting 20--30% of the adult population. It is rapidly becoming a major problem in developing countries, including Malaysia. About 90--95% of hypertension (HTN) is idiopathic and apparently primary or essential hypertension (EH). Of the remaining 5--10%, most are secondary to renal and adrenal diseases. EH is a multifactorial disorder arising from the influence of several susceptibility genes and environmental stimuli. Evidence suggests that genes may contribute to 30% of the variation of blood pressure. However, the number of genes involved or the model of interaction with other genes or environmental risk factors is unknown. The angiotensiongen (AGT) gene regulates the expression of angiotensinogen, a polypeptide primarily produced by the liver. Cleavage of the angiotensinogen molecule by renin, liberating angiotensin I, and then converted into angiotensin II by angiotensin-converting enzyme. This product binds to its receptor, exerting physiologic effects on the sodium homoeostasis and vascular resistance, thus regulates the blood pressure \[[@B1]\]. The plasma concentration of AGT is correlated with blood pressure \[[@B1],[@B2]\]. Mice with the AGT gene duplicated have blood pressure and plasma AGT levels positively correlated with the number of gene copies \[[@B3]\]. The human AGT gene has been cloned and sequenced \[[@B4]\]. Fifteen molecular variants have been identified and only three have so far been reported to have a possible genetic association with hypertension \[[@B5]\]. One variant encodes threonine instead of methionine at position 235(T235) \[[@B5],[@B6]\], the others encode methionine instead of threonine at position 174 \[[@B5],[@B6]\] and the microsatellite (a GT-repeat sequence, varies highly as 11 different allelic variants among individuals) \[[@B7]\]. Association studies of the M235T variant in essential hypertension have yielded conflicting results. Some found linkage or association in Caucasian \[[@B5]-[@B7]\], African-Carribean \[[@B8]\], Japanese \[[@B9],[@B10]\] and Taiwanese populations \[[@B11]\], while others did not \[[@B12]-[@B14]\]. A molecular variant of AGT has also been reported to be associated with preeclampsia \[[@B15]\]. The study addressed the question as to whether there is an association between the AGT M235T gene variant and essential hypertension in the Malaysian subjects since genetic diversity exists among different ethnic populations and realizing the fact that the association in one population could not be extrapolated to another population. Methods ======= Study subjects -------------- Approval and permission were obtained from the ethics committees of the Faculty of Medicine and Health Sciences, Universiti Putra Malaysia, the Federal Territory Kuala Lumpur Health Department and the Ministry of Health, Malaysia to meet the ethics guidelines. The permission allowed the study to be carried out in the Health Clinic, Kuala Lumpur Hospital. Upon the approval, subjects were recruited consecutively from patients attending the Health Clinic from 1^st^October to 31^st^October, 2002. The patients referred to the clinic were residents of the Klang Valley, consisting of the Federal Territory Kuala Lumpur and parts of the Selangor state. The subjects can be categorized into three main ethnic groups: Malay, Chinese and Indian. Hypertensive subjects were defined as those with systolic blood pressure (SBP) of greater than or equal to 140 mmHg, with a diastolic blood pressure (DBP) of greater than or equal to 90 mmHg, or are currently administered at least one hypertensive medication. Any subjects with the possibility of secondary hypertension were excluded. Hypertensive subjects whose parents both had hypertension were considered to have a positive family history of hypertension. Normotensive was defined as those with a blood pressure of less than 140/90 mmHg; those with a positive family history of hypertension were excluded. Both groups with subjects under the influence of estrogen, thyroid and cortisol hormones were excluded. The subjects were selected by medical officers and also approached by the field team. Informed consent was obtained from the subjects and a total of 101 hypertensives (23 males and 68 females) and 87 normotensives (23 males and 64 females) were recruited. Questionnaire, blood pressure and body mass index measurements -------------------------------------------------------------- A three-page pre-tested questionnaire in both the Malay and English language were developed to assess the socio-demographic background, socio-demographic data, dietary habits, physical activity, smoking habits, alcohol consumption and family history for hypertension. The blood pressure (BP) was measured with the subject sitting, using an automated sphygmomanometer (Colin Press-Mate BP-8800C^®^) after at least 10 minutes of resting. Height and weight of subjects were obtained by using the TANITA digital weighing scale and the SECA Bodymeter 208 respectively. The body mass index (BMI) of subjects was calculated as weight (kg) / height^2^(m^2^). Blood collection ---------------- Four to five mililitres of peripheral venous blood were collected into two separate K~2~EDTA vacutainer test tubes from each subject for radioimmunoassay (RIA) and AGT M235T variant genotyping. Overnight fasting blood samples were collected into tubes containing sodium fluoride. Blood was collected using a 21-Gauge needle with a 5.0 ml syringe by a qualified phlebotomist. The collected blood in test tubes were kept at 4°C and centrifuged at 1200 g for 15 minutes in order to separate the plasma from whole blood. The plasma samples were frozen at -20°C until fasting blood glucose (FBG), triglycerides (TG), total cholesterol (TC), low-density lipoprotein (LDL-C), high density lipoprotein (HDL-C) and its percentage, plasma sodium and potassium levels were determined within 3 days. Detection of the AGT genotypes ------------------------------ Genomic DNA extraction was carried out using the QIAamp^®^DNA Blood Mini Kit by QIAGEN^®^. DNA fragments including the M235T variant were amplified by allele-specific polymerase chain reaction (PCR). The AGT M235T polymorphism was typed by the previously described mismatch priming method with some modifications \[[@B12]\]. The forward primer sequence from +921 to +941 in exon 2 of the AGT gene is 5\' GAT GCG CAC AAG GTC CTG TC 3\' whereas the reverse primer sequence from +1202 to +1224 is 5\' GGT GCT GTC CAC ACT GGA CCC C 3\'. The reverse primer was designed to contain a base substitution T→C at the fourth last nucleotide from its 3\' end \[[@B12]\]. The individual PCR reaction vial contains a final volume of 50 μl solution. One-hundred nanograms of DNA samples was added to 14 μl of PCR master mix consisting of 5.0 μl of Promega^®^10X Mg Free PCR Buffer, 3.0 μl of Promega^®^25 mM MgCl~2~, 2.5 μl of 10 pM forward primer, 2.5 μl of 10 pM reverse primer and 1.0 ml of Promega^®^10 mM dNTP. An appropriate amount of sterile ultrapure water (which totals up to 50 μl) was added to each of the microfuge tube. One microlitre of 5 units/μl TaqDNA polymerase was added to the reaction vial only after 5 minutes of pre-denaturation process prior to performing \'hot start\' PCR. The 10X Mg-free PCR Buffer has a composition of 50 mM KCl, 10 mM Tris-HCl, (pH 9.0 at 25°C) and 0.1% Triton^®^X-100. The PCR was performed using the Mastercycler Gradient (Eppendorf^®^) for 35 cycles. The temperature for the initial denaturation of DNA was 95°C for 1 minute, annealing at 71°C for 1 minute and extension 72°C for 1 minute and a final extension at 72°C for 7 minutes following the last cycle. The PCR product was subjected to PsyI (isoschizomere for Tth111I restriction enzyme) digestion for 3 hours at 37°C and electrophoresed on a 5.0% agarose gel with ethidium bromide staining. Plasma renin activity radioimmunoassay -------------------------------------- Plasma renin activity (PRA) was determined using the Angiotensin I \[^125^I\] RIA kit (PerkinElmer Life Sciences, Inc.). It is a two-step protocol: the generation of Angiotensin I and the RIA to quantitate the amount of Angiotensin I generated. Plasma was incubated at 37°C, at pH 6, with protease inhibitors (EDTA, dimercaprol (BAL) and 8-hydroxyquinoline) for 1 hour to allow the generation of angiotensin I by the endogenous renin and substrate reaction. The total concentration of the Angiotensin I was determined by radioimmunoassay. Plasma renin activity was expressed as ng/ml/hr of angiotensin I generated. The intra-assay coefficient of variation for this assay was 4.0 % (n = 12) and the inter-assay coefficient of variation was 10.4% (n = 84). Statistical analysis -------------------- The SPSS (previously known as Statistical Package for Social Sciences) for Windows^®^Version 11.0 was used to statistically analyze the data obtained. Descriptive statistics were used to analyze all variable studies such as the socio-demographic characteristics, dietary patterns, physical activity, smoking practices and alcohol consumption, anthropometric measurements, biological parameters and AGT genotypes of the subjects. Genotype and allele frequencies in control and hypertensive groups were compared by Chi-square (χ^2^) analysis. Continuous variables were compared between hypertensive and control groups by Student\'s ttest (or the Mann-Whitney Utest for non-normally distributed variables). The influence of AGT genotype on continuous variable was investigated by one-way ANOVA. In addition, the effect of AGT genotype on BP was investigated with the General Linear Model ANOVA with adjustment for age, sex, race and BMI. Multiple regression analysis was also performed with SBP or DBP as the dependent variable and sex, age, BMI, total cholesterol and Plasma Renin Activity and AGT genotype (coded 0, 1, or 2 according to the number of T235 alleles) as independent variables. P \< 0.05 was considered statistically significant. Results ======= Demographic data ---------------- There were a total of 188 subjects recruited in the study, consisting of 101 hypertensives (33 males and 68 females) and 87 normotensives (23 males and 64 females). Of the 150 subjects approached, the respondence rate was around 72% due to some who choose not to participate or was excluded due to non-compliance of the inclusion criteria. The majority of the subjects were females. The Malays (n = 97, 51.6%) comprises more than half of the subjects, followed by Chinese (n = 56, 29.8%) and Indians (n = 35, 18.6%). The hypertensive subjects ranged from 30 to 78 years old, with a mean age of 54.7 years, while the normotensives ranged from 25 to 78 years old, with a mean 49.3 years, indicating that the normotensives are younger in the age group. Sixty-three people or 62.4% of the hypertensive subjects have a family history of hypertension while all of the normotensives do not have a family history of hypertension in order to be included in the study. Genotypes and allele frequencies -------------------------------- Figure [1](#F1){ref-type="fig"} shows the results of the PsyIdigestion on PCR products. AGT +704 T→C missense mutation (cause amino acid substitution of AGT M235T) created a new restriction site with the sequence recognition: GACN NNGT↓C for PsyI. PsyI digested the fragment into 2 parts, the longer fragment, 279 bp and the shorter 24 bp. However, the 5.0% agarose gel was unable to retain the shorter fragment and it was suspected to have migrated out of the gel. Therefore, a band at 303 bp indicates homozygous wild-type (MM), a band at 279 bp indicates homozygous mutated (MT) and two bands at 303 bp and 279 bp indicates heterozygous mutation (TT). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### 5.0% agarose gel electrophoresis.Lanes 1, 2 and 3 correspond to RFLP pattern of homozygous mutant (TT), heterozygous (MT) and homozygous wild-type (MM), respectively. M is a 100 bp linear DNA ladder (Promega). ::: ![](1471-2261-5-7-1) ::: According to Table [1](#T1){ref-type="table"}, the prevalence of AGT M235T missense mutation in all the subjects was 17% for homozygous mutation (21% for hypertensives and 11% for normotensives), 34% for heterozygous mutation (46% for hypertensives and 21% for normotensives respectively) and 49% for homozygous wild-type (33% for hypertensives and 68% for normotensives respectively). The allele frequencies and genotype distribution of the M235T variant were in the Hardy-Weinberg equilibrium in either data set for cases (χ^2^= 0.634, df = 2, p \> 0.5) and for control (χ^2^= 917.1, df = 2, p \> 0.5) subjects. There was significant difference in genotype and allele frequencies between hypertensive and normotensive groups. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Genotype and Allele Frequencies of M235T variant of the AGT gene for both cases and controls. ::: Group Genotypes Alleles ------------------- --------------- ------------- ---- ----- ----- Hypertensives 33 46 22 112 90 Normotensives 59 18 10 136 38 Total 92 64 32 248 128 χ^2^value 23.184 21.482 **P\* value \<0.001 \<0.001 Odds ratio** 1.36 1.98 \Significant values were obtained through the Chi-square test. ::: When the allele frequencies are categorized based on ethnic groups (Table [2](#T2){ref-type="table"}), there were significant differences of the prevalence of T allele between hypertensives and normotensives in the Malays (45 controls and 52 cases, χ^2^= 12.765, df = 2, p = 0.001) and Indian (20 controls and 15 cases, χ^2^= 12.519, df = 2, p = 0.001) but not the Chinese (22 controls and 34 cases, χ^2^= 3.083, df = 2, p = 0.212). In addition, there was significance difference between the prevalence of the T allele of hypertensive and normotensive groups in female (23 controls and 33 cases, χ^2^= 20.828, df = 2, p = 0.001) but no significant differences between groups in male (64 controls and 68 cases, χ^2^= 3.836, df = 2, p = 0.140). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Frequencies of M235T variant of the AGT gene according to ethnic background for both cases and controls. ::: Group* Male Female ------------------- ---------- ------------ ----- ---- Hypertensives Malay 28 14 36 26 Chinese 10 10 28 20 Indian 1 3 9 17 Total 39 27 73 63 Normotensives Malay 25 5 48 12 Chinese 4 4 28 8 Indian 7 1 24 8 Total 36 10 100 28 χ^2^value 3.836 20.828 **P\* value** 0.140 0.001 \Significant values were obtained through the Chi-square test. ::: PRA levels ---------- According to Table [3](#T3){ref-type="table"}, PRA was significantly higher (p \< 0.001) in hypertensive subjects (n = 101, 3.8 ± 2.5 ngAI/ml/hr) compared to normotensive subjects (n = 87, 2.6 ± 1.3 ngAI/ml/hr,). However, the PRA was higher in hypertensives group among all the genotypes but not significantly different between genotypes classes (p = 0.687, hypertensives and p = 0.252, normotensives) as shown in Figure [2](#F2){ref-type="fig"}. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Plasma renin activity (PRA) levels according to genotypes and genders in both the cases and controls. Values are expressed in ngAI/ml/hr ± standard deviation. Values in parentheses represent the number of subjects. ::: Genotypes* Normotensives Hypertensives Total **p\* --------------- ------------------- ------------------- ------------------- ------------------ ------------------ ------------------- ---------- MM 2.77 ± 1.69 (15) 2.54 ± 1.24 (44) 3.68 ± 2.14 (13) 3.29 ± 2.97 (20) 3.19 ± 1.93 (28) 2.77 ± 1.96 (64) 0.118^a^ MT 3.16 ± 1.29 (6) 2.65 ± 1.24 (12) 3.88 ± 2.90 (13) 3.95 ± 2.51 (33) 3.65 ± 2.49 (19) 3.61 ± 2.31 (45) 0.292^e^ TT 1.47 ± 1.38 (2) 2.08 ± 1.28 (8) 2.56 ± 1.47 (7) 4.45 ± 2.34 (15) 2.32 ± 1.45 (9) 3.63 ± 2.31 (23) 0.129^f^ Total 2.76 ± 1.57 (23) 2.50 ± 1.24 (64) 3.52 ± 2.35 (33) 3.87 ± 2.62 (68) 3.21 ± 2.09 (56) 3.21 ± 2.17 (132) 2.57 ± 1.32 (87) 3.75 ± 2.53 (101) 3.21 ± 2.14 (188) p^§^** 0.434^b^ 0.521^c^ 0.997^d^ \<0.001^g^ \Significant values were obtained through one-way ANOVA between ^a^groups of genotypes, ^e^among males between groups of genotypes and ^f^among females between groups of genotypes. ^§^Significant values were obtained through Student T-test between ^b^normotensive males and females, ^c^hypertensive males and females, ^d^males and females, and ^g^normotensive and hypertensive subjects. ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Plasma renin activity (PRA), systolic blood pressure (SBP) and diastolic blood pressure (DBP) for normotensive and hypertensive subjects as grouped according to AGT M235T genotypes.The values were expressed as mean (as indicated by the red dots) whereas the bars represent the 95.0% confidence interval of standard deviations. N represents the number of subjects. MM, MT and TT correspond to homozygous wild-type, heterozygous and homozygous mutant, respectively. ::: ![](1471-2261-5-7-2) ::: The means of the PRA were not significantly different between female and male among hypertensives and normotensives. One way ANOVA also showed that there was no significant difference of PRA between age groups in both hypertensives and normotensives (p = 0.611 and p = 0.119, respectively). BP variations ------------- Table [4](#T4){ref-type="table"} shows BP and other variables according to AGT genotypes in hypertensive and normotensive data sets. There was no significant difference between genotype classes for both normotensives and hypertensives for unadjusted BP and any other measured variable. After adjustment was done to BP, SBP was significantly different between genotype classes in hypertensives. BP was significantly lowered (p \< 0.001) in normotensives (124.0 ± 17.6 / 74.1 ± 11.2 mmHg) as compared to hypertensives (150.3 ± 21.9 / 84.3 ± 12.6 mmHg). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Blood pressure and other variables according to AGT genotypes in normotensive and hypertensive subjects. Values are expressed as mean. Values in parentheses represent standard deviation unless stated otherwise. ::: Variables* Hypertensives Normotensives ------------------------ ------------------- ------------------- -------------- ---------- -------------- -------------- -------------- ---------- Age 55.5 (8.9) 55.1 (9.0) 52.7 (9.0) 0.498 48.2 (10.2) 53.7 (9.0) 48.5 (12.4) 0.136 SBP, mmHg 151.5 (16.2) 150.8 (25.9) 147.6 (20.6) 0.795 122.0 (18.0) 129.1 (15.8) 126.9 (17.8) 0.283 Adjusted SBP, mmHg 150.9 (3.7) 149.6 (3.1) 150.8 (4.6) 0.014^ψ^ 122.6 (1.9) 126.5 (3.5) 128.1 (4.5) 0.123^ψ^ DBP, mmHg 81.7 (14.7) 85.4 (12.3) 85.9 (9.3) 0.352 74.4 (12.0) 73.7 (11.4) 73.0 (4.6) 0.930 Adjusted DBP, mmHg 81.6 (2.2) 85.1 (1.9) 86.7 (2.8) 0.382^ψ^ 74.4 (1.4) 73.0 (2.6) 73.8 (3.4) 0.106^ψ^ BMI 27.9 (4.7) 30.0 (4.9) 27.6 (6.3) 0.098 25.9 (3.4) 25.7 (3.3) 25.3 (4.9) 0.860 \Significant values were obtained through one-way ANOVA unless stated otherwise. ^ψ^Significant values were obtained through ANOVA using a general linear model with adjustment for age, sex, race and BMI. ::: By multiple regression analysis, age was only the predictor of SBP in the hypertensive group (P \< 0.001) but not AGT genotype (R^2^= 0.147, p = 0.019 for SBP, R^2^= 0.064, p = 0.380 for DBP). BMI was the predictor for both SBP and DBP in the control group (P = 0.037 for SBP, P \< 0.001 for DBP). Meanwhile, age and sex (P \< 0.001 and 0.027 respectively) were the predictors for SBP in the control group (R^2^= 0.376, p \< 0.001 for SBP, R^2^= 0.104, p = 0.172 for DBP). Sex, age, BMI, total cholesterol and Plasma Renin Activity and AGT genotype were all not predictors for SBP and DBP in hypertensive subjects (R^2^= 0.147, p \< 0.019 for SBP, R^2^= 0.064, p = 0.380 for DBP). Associated risk factors ----------------------- Although normotensive subjects were younger, there was no significant difference between the two groups with respect to vegetarian practices, exercise, smoking practices, alcohol drinking, glucose level (6.2 ± 2.7 mmol/L for cases and 6.7 ± 3.3 mmol/L for control, p= 0.198) TG (1.7 ± 0.9 mmol/L for cases and 1.6 ± 1.1 mmol/L for control, p= 0.558), TTLC (5.5 ± 1.1 mmol/L for cases and 5.3 ± 1.1 mmol/L for control, p= 0.084), LDLC (3.5 ± 1.0 mmol/L for cases and 3.3 ± 0.9 mmol/L for control, p= 0.162), HDLC (1.3 ± 0.5 mmol/L for cases and 1.3 ± 0.3 mmol/L for control, p= 0.875) and plasma sodium (139 ± 5 mmol/L for cases and 139 ± 3 mmol/L for control, p= 0.285) and potassium levels (4.0 ± 0.5 mmol/L for cases and 4.0 ± 0.4 mmol/L for control, p= 0.921). However, using the Mann-Whitney U test shows that coffee consumption habit (69 cases and 71 controls, p= 0.038) whereas Student\'s T-test shows that BMI (28.8 ± 5.2 kg/m^2^for cases and 25.8 ± 3.6 kg/m^2^for control, p= 0.001) were lower in control subjects. The Pearson\'s Correlation test showed no correlation between exercise frequency and PRA in both hypertensives (p= 0.369, r = -0.098) and normotensives (p= 0.088, r = -0.170). No association was observed between years of smoking and years of drinking habits with PRA in both hypertensives (p= 0.994, r = -0.007 and p= 0.691, r = 0.040 respectively) and normotensives (p= 0.223, r = -0.123 and p= 0.174, r = -0.147 respectively). Discussion ========== In this study, it was found that the M235T polymorphism of the AGT gene is associated with essential hypertension. The relative risks for hypertension are 1.36 for subjects carrying the TT phenotype and 1.98 for those having allele T of the M235T variants. These results saw some agreement with some studies, but not with others. Jeunemaitre and associates \[[@B5]\] were the first to report the linkage of the molecular variants M235T with hypertension in the Whites/Caucasians. Subsequent studies among the Whites/Caucasians supported the former finding \[[@B16],[@B17]\] while others did not \[[@B7],[@B18],[@B19]\]. The association studies in the Africans /African-Americans mostly found a negative association \[[@B8],[@B13],[@B20]\], but the T allele is associated with increased plasma AGT \[[@B13],[@B20]\]. Studies on other Asian populations like the three studies in the Japanese population, found a positive association. \[[@B9],[@B10],[@B21]\] but not in others \[[@B14],[@B22]\]. The Chinese \[[@B23]\] and the Taiwanese \[[@B11]\] population reported a positive association. The frequency of the T235 variant among hypertensives this study, which was 0.45, is similar to the French cohort by the initial report by Jeunemaitre and associates \[[@B5]\]. Subsequent studies of the White populations reported frequencies of T235 allele in control groups of approximate 0.40, with the range from 0.31 \[[@B24]\] to 0.49 \[[@B7]\]. The frequency of T235 allele among Africans and African Americans is much higher than in whites, with the frequency as high as 0.92 \[[@B25]\]. Among the Japanese, the frequency is similarly high, around 0.75 \[[@B21]\]. Our study showed a higher Odds Ratio (O.R.) of 1.98 (95% CI, 1.46--2.67) compared to a recent meta-analysis \[[@B26]\] of 12 studies in the Whites which indicated that T235 is associated with a 20% increase risk of hypertension (O.R. = 1.22, 95% CI, 1.10--1.29). In this study, the significantly higher prevalence of T allele in hypertensive females is in agreement with the study by Jeunemaitre et al., \[[@B5]\] which reported that the T allele was significantly more prevalent among female hypertensives (0.51) than in controls (0.37) (χ^2^= 16.9, p \< 0.001). In contrast, Freire and associates \[[@B27]\], found that the AGT M235T homozygous mutation genotype was significantly higher in male compared to female, and Pereira et al. \[[@B28]\] reported that no association between gender and T allele in a cross-sectional study involving 647 females and 776 males. However, both of the studies were not confined to essential hypertensive patients. Although the T235 allele is associated with increased plasma AGT in Blacks \[[@B20],[@B25]\], the measurement of plasma AGT was not conducted in this study. However, the Plasma Renin Activity (PRA) test, which is the indicator of hyperaldosteronism, was carried out. Decreased PRA indicates primary hyperaldosteronism (adrenal-origin), while increased PRA indicates secondary hyperaldosteronism, an extra-adrenal cause. The concentration of AGT in blood is rate-limiting, and a change in its concentration can affect PRA \[[@B29]\]. The normal circulation level of AGT is close to Km, which means that a rise in plasma AGT could cause close to linear increase in the rate of angiotensin formation \[[@B29]\]. However, in vivo, an increase in plasma AGT will not increase the rate of angiotensin production, since the secretion of renin normally has a feedback relationship with angiotensin II so as to maintain a physiologically appropriate rate of formation of angiotensin II. \[[@B30]\] Exceptions of this condition are in individuals under estrogen influence \[[@B31]\] and in Cushing\'s syndrome \[[@B32]\]. An exclusively AGT-dependent hypertension is thus theoretically impossible, although two exceptional cases of hypertension associated with hepatic cell tumors producing large amounts of AGT have been reported \[[@B33],[@B34]\]. Thus, this justifies that the measurement of PRA being done, instead of plasma AGT levels. PRA has an advantage of estimating the extent of aldosterone in response to the activation of the Renin-Angiotensin System (RAS). In this study, the PRA was significantly higher in hypertensive subjects (P \< 0.0001), thus suggesting that the over activity of the RAS as a whole, thus contributing to hypertension. However, the PRA was not significantly different between groups of genotypes (P = 0.118). There are several possible reasons for the discrepancies found between previous studies and this study. It might be due to ethnic differences due to the heterogenous population or sampling bias. The background of the study subjects recruited from a hypertensive clinic in this study might differ from that of subjects selected from the general population. Racial differences, including diverse social and cultural factors may have contributed to the different results. Discrepancies may also be related to different methodologies and study designs used. The mechanism by which the molecular variant M235T of the AGT gene is related to hypertension is poorly understood. The AGT 235T variant has been found to be in complete linkage disequilibrium with a guanine-to-adenosine transition at -6 bp upstream of the initiation site of transcription \[[@B35]\]. In vitro*tests of promoter activity and DNA-binding studies with nuclear proteins show that this nucleotide substitution affects the basal transcription rate of this gene in various cell lines, thereby the AGT T235 variant and increased plasma AGT levels \[[@B5]\] and hence might contribute to the elevation of blood pressure. There were some limitations in this study. First, the normotensive subjects were relatively young compared with the hypertensives, although a 5-year difference in age might not have caused significant blood pressure variation. Secondly, the case-control design used to investigate the influence of the AGT 235T variant is known to be prone to selection bias and confounding especially when applied to the investigation of complex genetic traits, such as BP. The population studied was also not homogenous, as Malaysians consists of different ethnic groups, with different genetic make-up. While a positive association does not necessarily prove a causal relationship, it can provide useful information regarding the clinical importance of a genetic marker. Therefore, linkage and association studies are complementary, each providing a different type of information \[[@B36]\]. The mechanism by which the M235T variant contributes to the pathogenesis of hypertension needs to be elucidated further. It is also not possible to determine at the present whether the observed molecular variants of the AGT gene directly affect angiotensinogen function or whether they are markers of functional variants that have not yet been detected. If indeed the T235 variant directly affects plasma angiotensinogen concentration, it will be necessary to look for a possible difference in clearance rate of Km for renin between the two angiotensinogen isoforms. The response to antihypertensive agents, especially those blocking the RAS will have to be evaluated in patients classified according to their AGT genotype (pharmacogenomics). With the establishment of the association of the M235T variant with EH, further association studies of other cardiovascular diseases and diabetes involving this variant or other candidate genes involving the RAS system can be done. Hence, further genotyping of the Malaysian population could predict the risk of getting hypertension or other related cardiovascular disease. Conclusion ========== In this study, the T235 variant of the AGT gene is associated with essential hypertension in Malaysian subjects. The T235 variant is a risk factor or possibly a potential genetics marker for hypertension. Plasma renin activity is also significantly higher in hypertensive subjects but was not significantly different between groups of genotypes. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= YHS carried out the radioimmunoassay. Both YHS and KHL carried out the mutation studies, statistical analyses, developing questionnaires and drafted the manuscript. Both GD and SI carried out the anthropometry measurements and reviewed the patients\' history and making decision regarding patient eligibility in the study. RR conceived the study, and participated in its design and coordination. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/7/prepub> Acknowledgements ================ The authors would like to extend their gratitude to the Ministry of Health for granting the permission to conduct the study at the Health Clinic, Kuala Lumpur Hospital. The authors are also grateful to all the respondents who volunteered in this study and the staff of the Health Clinic of Kuala Lumpur Hospital for providing excellent technical assistance.	PubMed Central	2024-06-05T03:55:56.483330	2005-4-5	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090564/", "journal": "BMC Cardiovasc Disord. 2005 Apr 5; 5:7", "authors": [ { "first": "Yee-How", "last": "Say" }, { "first": "King-Hwa", "last": "Ling" }, { "first": "Gnanasothie", "last": "Duraisamy" }, { "first": "Suzanne", "last": "Isaac" }, { "first": "Rozita", "last": "Rosli" } ] }
PMC1090565	Background ========== More then 25 years ago, it was reported that ultrasound (US) may enhance the fibrinolytic process \[[@B1]\]. Throughout the 1980s and 2000s, several in-vitro and in-vivo experiments verified and further explored this effect, using US alone \[[@B2]\] or as a thrombolytic adjuvant \[[@B3]-[@B13]\]. The concept has recently been explored in myocardial ischemia \[[@B14]\] and infarction \[[@B15]\] in humans. The physical properties of US fields that may account for the observed profibrinolytic effects include thermal effects, the cavitation effect and micro-streaming \[[@B8],[@B11],[@B16],[@B17]\]. It is, however, still unclear as to how these contribute to the profibrinolytic mechanism. Although these effects of US are beneficial in US enhanced fibrinolysis, they might be harmful to biological tissue in other circumstances, for instance, in already injured tissue. The mechanisms by which US may be potentially harmful to biological tissue are in fact similar to those described in connection with US enhanced fibrinolysis \[[@B18]-[@B24]\]. The possible net benefit of successful US enhanced thrombolysis in the setting of myocardial infarction is the result of an earlier reperfusion, this could be diminished by the possible unfavourable effect of US on the ischemic myocardium. We have therefore undertaken this safety study exploring the effects of exposing ischemic myocardium to US. For this purpose a porcine model was developed in which blinded histopathological examination technique was used. Methods ======= Transducer calibration and measurements --------------------------------------- Before the animal experiments, calibration and measurements of US fields were performed on the two unfocused piezoelectric transducers used (Ceram AB, Lund, Sweden). The transducers had a resonance frequency of 1.0 MHz and diameter 16 mm. The transducers were excited by an electronic system consisting of a function generator (HP 3314A, Hewlett-Packard, Washington, USA) and a RF power amplifier (ENI 240L, ENI, Rochester, New York, USA). Electronic scale measurements ----------------------------- Measurements to determine the total radiation force were performed, using an electronic scale (Model UPT-DT-1, OHMIC Instrumental Co, St Michaels, Maryland, USA). US of 1 MHz and a spatial-averaged, temporal-averaged intensity (I~SATA~) of 1 W/cm^2^was sent as continuous wave and US of 1 MHz and a intensity of 1 W/cm^2^(I~SATA~) was sent as pulsed wave (one burst of 100 pulses per millisecond). The continuous US exposure constituted a reference measurement and was only used in the electronic scale measurements. Hydrophone measurements ----------------------- US of 1 MHz and an intensity of 1 W/cm^2^(I~SATA~) was sent as pulsed wave (one burst of 100 pulses per millisecond). Measurements were performed by scanning with a polyvinylidene fluoride membrane hydrophone, (GEC-Marconi Hydrophone Type Y-34-3598, Calibrated at National Physical Laboratory, Teddington, England) to determine the Mechanical Index (MI), the peak compressional pressure, rarefactional pressure and the maximal spatial peak temporal average Intensity (I~SPTA~). The signal was registered on an oscilloscope (Tektronix TDS 360, Tektronix UK, Ltd. Berkshire, United Kingdom). Measurements were also performed of the distribution of the US field yielded by a comparable transducer as used in the study. Scanning was performed with a 0.5 mm diameter needle hydrophone and amplifier (Precision Acoustic LTD. United Kingdom). From the oscilloscope, digitised signals were transferred into a computer program based on Lab-View software (Department of Electrical Measurements, Lund Institute of Technology, Lund, Sweden). The computer-controlled scanning-system enabled the hydrophone to be translated along three orthogonal axes (X, Y and Z). Scanning was performed over an area of 80 × 30 mm^2^in the Y and Z-directions starting close to the transducer surface. Temperature measurements ------------------------ Control measurement was performed of pulsed US exposure effects on temperature rise on non-circulated pig myocardium. One 0.5 mm temperature probe was placed 1.5 cm inside an extracted pig myocardial muscle (3.0 cm thick) with no circulation \[[@B25]\]. The myocardial muscle was then placed in a degassed water bath that was heated to 37°C. The US transducer was placed 1.5 cm perpendicular to the myocardial muscle surface and centred to the temperature probe. Pulsed US exposure started when water bath and muscle reach equivalent temperature. US of 1 MHz and an intensity of 1 W/cm^2^(I~SATA~) was sent as pulsed wave (one burst of 100 pulses per millisecond) during one hour. Simultaneous temperature measurement was performed inside the myocardial muscle and in the surrounding water bath once every half-second during the one hour of pulsed US exposure. Animal preparation ------------------ The study was approved by the Ethical Committee of the University of Lund (approval M246/91). Seventeen 25--30 kg Swedish landscape pigs were used in the study. Anaesthesia was induced with 5--10 ml (25 mg/ml) sodium pentothal (Pentothal Natrium, Abbot Scandinavia AB, Sweden) intravenously (I.V.) before tracheotomy. The pigs were mechanically ventilated (Serviventilator 900 B, Siemens Elema, Sweden). Access to circulation was maintained through one arterial entrance and at least two venous lines. Blood pressure was continuously monitored through the arterial line. Anaesthesia was maintained with ketamine (Ketalar, Parke-Davis, Division of Warener Lambert Nordic AB, Sweden) at a dose of 5 mg/Kg/min I.V., and pancurone (Pavulone, Organon Teknika AB, Sweden) at a dose of 0.3 mg/Kg/min I.V. Following sternotomi, the pericardium was incised and its borders along the incision line sutured to the skin overlying the sternal edges. A large proximal diagonal branch of the left anterior descending artery (LAD) was ligated to induce myocardial infarction. Transducer and US exposure -------------------------- Following 1 hour of coronary ligation, the US transducers were applied. Each of the two transducers was fixed to a universal joint attached to a small steel pipe on a stand. The transducers were thus in a fixed position, and placed approximately 1.5 cm from the epicardium. The transducer at the non-infarcted myocardium was placed to radiate part of the anterior/apical free wall of the left ventricle, while the transducer radiating part of the infarcted myocardium was placed in the mid/basal region of the anterior portion of left ventricle, corresponding to the myocardial region perfused by the ligated large diagonal branch of the LAD (Figure [1](#F1){ref-type="fig"}). US gel (Clinical, Diagramm Halbach AG, Germany) was then applied to the entire anterior portion of the heart to ensure adequate sound wave transmission to the epicardium. Due to loss of gel during the procedure, additional gel was deposited repeatedly during the one hour transducer and pulsed US exposure period. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Experimental setup. Schematic depiction of the open chest porcine model with the ligated diagonal branch of the left anterior descending artery. The shading indicates the extent of ischemic tissue. Ultrasound transducers are applied over ischemic as well as non-ischemic tissue. ::: ![](1471-2261-5-8-1) ::: The US radiation applied over the myocardial areas was, pulsed US of frequency 1 MHz and intensity of 1 W/cm^2^. Each millisecond, a burst of one hundred cycles was sent, equivalent to a duty cycle of 10% and a resulting intensity of 0.1 W/cm^2^(I~SATA~). The experiment was designed to illustrate possible injury effects of pulsed US exposure, mechanical handling of the hearts and application of transducers alone. The altogether 68 tissue samples from the 17 pig hearts were thus used as follows: A\) 17 samples of non-infarcted myocardium without any exposure. B\) 4 samples of non-infarcted myocardium exposed to transducer alone. C\) 13 samples of non-infarcted myocardium exposed to pulsed US. D\) 17 samples of infarcted myocardium without any exposure. E\) 8 samples of infarcted myocardium exposed to transducer alone. F\) 9 samples of infarcted myocardium exposed to pulsed US. Tissue preparation and histopathological evaluation --------------------------------------------------- After one hour of applied pulsed US radiation or transducer exposure, epicardial sutures were placed at two locations under the transducers, to indicate the diameter of the exposed area and indicate areas were tissue samples should be removed. The pigs were then immediately given 10 ml of potassium (2 mmol/ml, Addex-Kalium™, Pharmacia & Upjohn AB, Sweden) intravenously to induce ventricular fibrillation, which occurred momentarily following administration. The great vessels were then clamped and cut by scalpel, while the veins were ligated and cut by scissors. The entire heart was then extracted intact followed by immediate removal of tissue samples. During the procedure, care was taken to minimize traumatic handling of the heart. Transmural samples, 5--10 mm in diameter, were then carefully removed with a scalpel. Following excision, the tissue samples were prepared for histopathological evaluation. From each sample, three slides were cut from the formalin fixed and paraffin embedded blocks and stained with Hematoxylin-Eosin (Van Gieson, and Phosphor Tungistic Acid Hematoxylin, PTAH) respectively. The samples were then examined by routine light microscopy in 10×, 20× and 40× enlargement. An experienced pathologist examined all slides in a blinded fashion unaware of whether the tissue samples were from infarcted, or non-infarcted tissue, or if it had, or had not been exposed to US. In the microscopic evaluation only the common signs of tissue damage, seen during the early phase of myocardial infarction, were observed \[[@B26]\]. The parameters evaluated for damage score were as follows: 1\) Eosinophilic changes in the myocyte, (Ischemia) 2\) Reduction and/or loss of cross striation, (Loss of striation) 3\) Coagulation necrosis, (Necrosis) 4\) Infiltration of polymorphonuclear cells, (PMN) All parameters were scored from 0 to 3, where 0 indicated no damage, 1 indicated minimal change, 2 intermediate changes and 3 extensive changes. Analysis was made of the damage score for each of the parameters (Ischemia, Loss of striation, Necrosis and PMN) and the total sum of damage scores (TDS) of the four parameters. Thus, the TDS had a minimum possible total damage score of 0 and a maximum possible total damage score of 12. Statistical analysis -------------------- The TDS of all groups were analysed to estimate the normal distribution by Chi-square test for goodness of fit. The statistical analysis compared the TDS scores in two ways: by paired Students t-test for the comparison of effects in individual animals and Student t-test for group comparison. A p value of less than 0.05 was considered statistically significant. Results ======= Transducer calibration and measurements --------------------------------------- ### Electronic scale measurements In the balance measurements, using continuous wave US, the US power supply was set to yield an acoustic power value of 2.0 W, corresponding to an intensity of 1 W/cm^2^(I~SATA~) adjusted to the area of the transducers. Using pulsed wave (10% duty cycle) US at the same power supply the intensity was 0.19 W, which correspond to an intensity of 0.1 W/cm^2^(I~SATA~). ### Hydrophone measurements In the membrane hydrophone measurements the I~SPTA~was measured to 460 mW/cm^2^and the MI was calculated \[[@B27]\] to be 0.41 at a distance of 3.0 -- 3.5 cm from the surface of the transducer. At the same distance the peak compressional pressure was measured to a value of 0.41 Mega Pascal (MPa) and the peak rarefactional pressure was 0.41 MPa. The needle hydrophone measurement was performed in degassed water to explore field distribution for the used transducers, the distribution was determined but exact values of intensity were not measured (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Ultrasound field distribution. The field distribution for the transducers is shown, but no exact values of intensity was measured. Scanning was performed over an area of 80 × 30 mm^2^in the Y and Z-directions starting close to the transducer surface. The transducer surface and the start of myocardium is mark. ::: ![](1471-2261-5-8-2) ::: ### Temperature measurements During the first 20 min of pulsed US exposure to the extracted myocardial muscle an increasing temperature difference was seen between the non-circulated pig myocardium and the surrounding water bath. After 20 min of exposure the temperature reached a steady state difference of 0.5°C (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Temperature measurements The temperature increase for non-circulated pig myocardium during pulsed ultrasound exposure. Temperature was measured every 0.5-second during 1 hour and 27 min. Temperature at different time interval is presented as mean ± standard deviation °C. Differences in temperature between the exposed myocardium and surrounding water bath are also shown. ::: Inside water bath Inside heart muscle Difference degrees C ------------------------------------- ----------------------- ------------------------- -------------------------- 5 min steady state 36.8 ± 0.12 36.8 ± 0.03 0.0 During first 2 min US 36.7 ± 0.12 36.9 ± 0.05 0.2 During first 5 min US 36.8 ± 0.12 37.0 ± 0.07 0.2 During first 10 min US 36.8 ± 0.12 37.1 ± 0.08 0.3 10--20 min US 36.7 ± 0.11 37.2 ± 0.03 0.4 20--40 min US 36.7 ± 0.10 37.2 ± 0.03 0.5 40--60 min US 36.7 ± 0.10 37.1 ± 0.03 0.5 20 min after US termination (2 min) 36.6 ± 0.12 36.7 ± 0.03 0.1 ::: Physiologic monitoring ---------------------- All animals were stable in circulation during the experiments and were under supervision of both anaesthesia and cardiology expertise during the whole experiment period. An estimation of the infarcted area is shown in figure [1](#F1){ref-type="fig"}. No exact measurement of the size of infarcted area was however performed. Histopathological evaluation ---------------------------- Examples of the different score grades are shown in figure [3](#F3){ref-type="fig"}. Signs of myocardial damage were noted already in all 17 perfused tissue specimen untouched by an US transducer, TDS being 4.3 ± 1.5 (mean ± standard deviation). In comparison, the infarcted tissue, untouched by any US transducer, had a significantly higher damage score, irrespective if comparison used the paired difference technique (p \< 0.001) or Student t-test for group comparison, 6.2 ± 2.0 vs. 4.3 ± 1.5 (p = 0.004). There was a further significant augmentation of the tissue injury in the infarcted myocardium exposed to US, also evidenced by paired difference (p = 0.026) and Student t-test for group comparison 8.1 ± 1.7 vs. 6.2 ± 2.0 (p = 0.027). The individual damage scores and statistical analyses for all groups are shown in figure [4](#F4){ref-type="fig"}. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Examples of histopathological indicators. Examples of the various histopathological indicators of tissue damage at different damage scores: A = Eosinophilic changes in the myocyte (Ischemia), B = Reduction and/or loss of cross striation (Loss of striation), C = Coagulation necrosis (Necrosis), D = Infiltration of poly-morphonuclear cells (PMN). The estimated degree of damage in each sample is graded in a 0--3 scale. For further information see text. ::: ![](1471-2261-5-8-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Individual total damage score and statistical analysis. The dot-plot shows the total damage scores obtained in 64 individual tissue samples from the 6 different myocardial areas. Statistical comparisons of the total damage scores (TDS) were performed using the Students t-test for group comparison. A p value of less than 0.05 was considered statistically significant. ::: ![](1471-2261-5-8-4) ::: Control experiments ------------------- Following US exposure of non-infarcted myocardium, there was a significant increase in TDS, from 4.3 ± 1.5 to 5.8 ± 1.7 (p = 0.015, Student t-test for group comparison). The application of transducer exposure alone did not significantly affect the damage caused by myocardial infarction, evidenced by the TDS 6.2 ± 2.0 vs. 6.6 ± 2.1 (p = 0.662, Student t-test for group comparison and p = 0.244 if paired Student t-test was used). Discussion ========== The theoretical possibility of exaggeration of myocardial damage in non-perfused tissue following pulsed US exposure prompted the present analysis of the effect of US on the ischemic myocardium. We chose an open chest porcine model with an induced myocardial infarction and applied US to test our hypothesis that lengthy US exposure on ischemic myocardial tissues could potentiallybe harmful, even when used within limits hitherto considered safe in cardiac exposures \[[@B28],[@B29]\] The histopathological diagnosis of tissue damage during the first hour of coronary artery occlusion is based on subtle findings. The assessment of cardiac damage used is therefore not specific as a in a fully developed infarction. Furthermore, we found a notable amount of damage already in the samples from the non-infarcted myocardium that not were exposed to transducer or US, a finding which could be explained by stress-induced damage \[[@B30]-[@B32]\] as well as mechanical handling of the heart during removal. The genuine effect of the 2-hour long lasting ischemia shows however a significant increase in the total damage score by 45 %. US exposure of the infarcted myocardium resulted in a significant further increase of the total damage score by another 30 %. Possible mechanisms of ultrasonic injuries ------------------------------------------ High Intensity US causes damage in biological tissues through three mechanisms; thermal, mechanical and cavitational injuries \[[@B18]-[@B24]\]. The physical properties of US fields which may account for the profibrinolytic effects observed are thermal effects, the cavitation effect and micro-streaming \[[@B8],[@B11],[@B16],[@B17]\]. As earlier stated, the US-induced damage in biological tissues may develop by the same mechanisms that enhance fibrinolysis. Heat produced in perfused tissues exposed to US within safe levels would under ordinary circumstances mainly be lost to the circulation. In ischemic tissues, the reduced circulation may thus decrease the heat-loss ability during US exposure resulting in undesirable heating and creating a thermal injury \[[@B22]\]. However, in the present study, temperature measurements in the non-circulated myocardial tissue illustrated only a small increase (0.5°C), well within in the limits of now used safety rules. High frequency US exposure of circulated tissue has been shown to induce a mild increase in interstitial oedema, an accumulation of polymorphonuclear cells \[[@B33]\], oxidative stress in endothelial cells \[[@B34]\] and increased cell lysis and apoptosis in human myelomonocytic leukaemia cells \[[@B35]\]. It has been hypothesized that the effects may be caused by radiation force \[[@B35]\] and other non-thermal effects \[[@B34]\]. US at similar exposure settings as used in the present study have been shown to produce necrotic and cellular damage in epidermis layers in goldfish \[[@B25]\]. Increasing sonication duration resulted in a progressively greater damage score. Damage was also shown to gradually propagate inwards the cell layers with increased sonication duration. It was concluded that the damage produced was caused by cavitational injury \[[@B25]\]. Anatomical structural differences and the age of the animal have also been shown to affect the sensitivity for US exposure \[[@B21]\], although the role of the age dependence remains unclear\[[@B24]\]. In conclusion, available data are unable to disclose the true mechanism of the myocardial injury induced by pulsed US. Finally, the development of ischemic myocardial damage over time is evidenced by different and unspecific indicators of damage, some being reversible \[[@B36]\]. It is therefore not possible to estimate how the myocardial tissue in the present study would be affected after a fully developed infarction. Relevance of study results -------------------------- Multiple studies of the beneficial effects of US used to enhance thrombolysis have been conducted in different animal models, both in the venous and arterial systems \[[@B7],[@B8],[@B12],[@B33],[@B37]-[@B45]\]. The progress in the area has also reached humans in the clinical setting. Thus, a prospective controlled study of US augmentation of thrombolysis in myocardial ischemia failed to verify a beneficial effect \[[@B14]\]. In fact, the number of ischemic complications increased following US exposure. In contrast, no undesirable effects were however noticed in 25 patients with myocardial infarction, who received traditional thrombolytic treatment with adjunctive US exposure \[[@B15]\]. Interestingly, early and complete arterial recanalisation was noted in selected patients with stroke, who received thrombolytic treatment and whose cerebral arterial blood flow was followed with transcranial Doppler technique \[[@B46]\]. The result of our study clearly obviates the importance of evaluating the potentially non-beneficial effects of US aiming at enhancement of thrombolysis. In the setting of US exposure during cerebral ischemia, and at US energy levels verified to be beneficial in experimental studies of thrombolysis \[[@B7]\], no detectable additive damage was however verified following exposure of pulsed US \[[@B47]\]. Study limitations ----------------- Two control experiments were carried out. Firstly, we verified that fixation of the transducer and application of US gel close to the infarcted myocardium did not significantly affect the TDS. Secondly, we explored the effect of US on perfused myocardium. Interestingly there were significant signs of myocardial damage also in this presumed healthy tissue following US exposure. The sensitivity for stress-induced damage in the pig limits this study considerably and makes the interpretation of data more difficult. Still, the total damage score after US exposure significantly exceed damage scores in the unexposed infarcted tissue samples. No threshold measurements were performed, however the total US energy we used is in the order of 10 times higher than required to enhance the thrombolysis in-vitro\[[@B6]\]. Neither was the temperature monitored in the areas during exposure to US. The low number of samples (n = 4) in the group of non-infarcted myocardium exposed to transducer alone is the result of protocol misinterpretation, unfortunately necessitating the exclusion of these data from the statistical comparison. Conclusion ========== Lengthy pulsed US exposure of low intensity exerting beneficial pro-fibrinolytic effects in the setting of thrombolysis may increase instantaneous myocardial damage. List of abbreviations used ========================== US: Ultrasound I~SATA~: Spatial-Averaged, Temporal-Averaged Intensity MI: Mechanical Index I~SPTA~: Spatial-Peak Temporal Average Intensity LAD: Left anterior descending artery TDS: Total damage score Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= GO designed the investigation, did the surgical and ultrasound experiments, performed the statistical analysis and interpretation of the results, as well as the preparation of the manuscript. BMH performed the statistical analysis and interpretation of the results, as well as the preparation of the manuscript. AR was involved in the design of the investigation and interpretation of the results and approved the final manuscript. MB was involved in the histopathological evaluation and interpretation of the results and approved the final manuscript. EG did the surgical and ultrasound experiments and approved the final manuscript. HWP was involved in the designed the investigation and interpretation of the results, as well as the preparation of the manuscript. LJ was involved in the histopathological evaluation and interpretation of the results and approved the final manuscript. SBO supervised and designed the investigation as well as participated in the preparation of the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/8/prepub> Acknowledgements ================ The project has been supported with grants from The Swedish Medical Research Council (K98-19X), Swedish Heart and Lung Foundation (Hjärt-Lung Fonden), the Thorsten Westerström Foundation (Thorsten Westerströms stiftelse) and Franke and Margareta Bergqvist foundation (Franke och Margareta Bergqvist stiftelse). We appreciate the invaluable advice and recommendations from Professor Karel Maršál, Department of Gynecology and Obstetrics, Lund University Hospital, Sweden. We will also want to thank Johan Persson at the Department of Orthopedics, Lund University, Sweden for professional advise and help with the temperature measurements.	PubMed Central	2024-06-05T03:55:56.486409	2005-4-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090565/", "journal": "BMC Cardiovasc Disord. 2005 Apr 15; 5:8", "authors": [ { "first": "Göran K", "last": "Olivecrona" }, { "first": "Bjarne Madsen", "last": "Härdig" }, { "first": "Anders", "last": "Roijer" }, { "first": "Mattias", "last": "Block" }, { "first": "Edgars", "last": "Grins" }, { "first": "Hans W", "last": "Persson" }, { "first": "Leif", "last": "Johansson" }, { "first": "Bertil", "last": "Olsson" } ] }
PMC1090566	Background ========== EB1 is the prototypical member of a highly conserved family of proteins that localise to centrosomes and growing microtubule tips \[[@B1]-[@B3]\]. EB1 has been shown to directly interact with microtubules, the adenomatous polyposis coli (APC) tumour suppressor protein and the p150glued subunit of the dynein/dynactin microtubule motor complex \[[@B3]-[@B9]\]. In a previous study of a series of EB1 deletion mutants we noted that an EB1 protein lacking its N-terminal 100aa and fused at its C-terminus to GFP (EB1-ΔN2-GFP) spontaneously formed perinuclear aggregates in transfected COS-7 cells fixed and examined by immunostaining \[[@B3]\]. The present study represents a further characterisation of these aggregates. This phenomenon was considered worthy of investigation for a number of reasons. Removal of the N-terminal 100aa of EB1 to generate EB1-ΔN2-GFP neatly removes one of the major structural features of the protein, a calponin homology (CH) domain implicated in the microtubule-binding ability of EB1 \[[@B3],[@B9]-[@B11]\], while leaving the region that mediates the interactions with APC and p150glued intact. Consistent with this, EB1-ΔN2-GFP does not localise to microtubules in transfected cells whereas a GST-EB1-ΔN2 recombinant fusion protein binds to both APC and p150glued in vitro\[[@B3]\]. The behaviour of EB1-ΔN2-GFP in transfected cells might therefore reveal new information about EB1 folding and function. In addition, EB1-ΔN2-GFP aggregation resembled a cellular response to the presence of misfolded, proteolytically resistant proteins termed aggresome formation \[see ref \[[@B12]\] for a recent review\]. Aggresomes assemble around centrosomes in a process that requires microtubules and dynein/dynactin-mediated transport \[[@B13]-[@B15]\]. As EB1-ΔN2-GFP retains the ability to directly interact with p150glued, an examination of the aggregation of this protein in transfected cells might yield further insight into the process of aggresome formation. Furthermore, it seemed possible that the perinuclear EB1-ΔN2-GFP aggregates might represent a structure unrelated to aggresomes and instead arise from dominant-negative effects on the normal EB1/p150glued interaction. We therefore reasoned that detailed examination of EB1-ΔN2-GFP aggregation might shed further light on the normal function of this interaction within cells. Results ======= EB1-ΔN2-GFP aggregate dynamics in living cells ---------------------------------------------- We first examined the behaviour of EB1-ΔN2-GFP using time-lapse fluorescence microscopy in living COS-7 cells. In a minority of cells EB1-ΔN2-GFP displayed a diffuse cytoplasmic distribution with specific labelling of the centrosome (not shown), as described previously in fixed cells \[[@B3]\]. The remaining cells contained single large perinuclear aggregates of the fusion protein against a background of diffuse cytoplasmic fluorescence, with smaller, motile, non-perinuclear structures sometimes apparent (ie see [additional file 2](#S2){ref-type="supplementary-material"}). Perinuclear aggregates at this time (14--18 h post-transfection) typically had a compact morphology and were less than half the size of the nucleus in the transfected cell (Fig. [1](#F1){ref-type="fig"}; additional files [1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Live imaging of EB1-ΔN2-GFP aggregates in transfected COS-7 cells. Panel A. Single frame from [additional file 1](#S1){ref-type="supplementary-material"} showing the structure of EB1-ΔN2-GFP aggregates. Bar = 5 μm. Panel B. Single frame from [additional file 2](#S2){ref-type="supplementary-material"}, sequence 1 showing aggresome (arrows) behaviour in two transfected cells. Panel C. Single frame from [additional file 2](#S2){ref-type="supplementary-material"}, sequence 2 that shows the cells from sequence 1 imaged using phase-contrast microscopy to reveal adjacent untransfected cells. This sequence was begun 30s after the completion of sequence 1. Arrows indicate aggresome location in the transfected cells. Bar = 10 μm. Panels D-F. Three frames from [additional file 3](#S3){ref-type="supplementary-material"}, sequence 1 showing the extension of a linear structure from an EB1-ΔN2-GFP aggregate. Panels G-I. Three frames from [additional file 3](#S3){ref-type="supplementary-material"}, sequence 2 showing the retraction of this structure. Bar for D-I = 5 μm. Times shown are relative to the first frame in each sequence. Panels A and D-I were obtained using a 63X oil immersion lens. Panels B and C were obtained using a 40X dry lens. ::: ![](1471-2121-6-17-1) ::: Time-lapse imaging of transfected cells revealed that the EB1-ΔN2-GFP aggregates consisted of variably sized, brightly fluorescent foci closely associated with structures of lower fluorescence intensity (Fig. [1A](#F1){ref-type="fig"}; [additional file 1](#S1){ref-type="supplementary-material"}; [additional file 2](#S2){ref-type="supplementary-material"}, sequence 1), with a dense appearance in phase contrast images (Fig. [1B](#F1){ref-type="fig"} and [1C](#F1){ref-type="fig"}, arrows; [additional file 2](#S2){ref-type="supplementary-material"}, sequence 2). Aside from this, phase contrast imaging did not reveal any obvious differences in cellular morphology when transfected cells were compared with adjacent, untransfected cells (Fig. [1C](#F1){ref-type="fig"}, [additional file 2](#S2){ref-type="supplementary-material"}, sequence 2). EB1-ΔN2-GFP aggregates exhibited constant minor changes in shape while maintaining a relatively cohesive overall structure. However, occasional larger changes in morphology were also observed, typically a rapid extrusion of ribbon-like structures away from the main body of the aggregate (Fig [1](#F1){ref-type="fig"}. panels D-I, arrows; [additional file 3](#S3){ref-type="supplementary-material"}, sequence 1; see also [additional file 1](#S1){ref-type="supplementary-material"}). This behaviour suggested that the aggregates consisted of brighter particles linked by a less fluorescently intense matrix. These extrusions appeared relatively stable, but re-incorporation into the main body of the aggregate was also seen (Fig. [1](#F1){ref-type="fig"} panels G-I; [additional file 3](#S3){ref-type="supplementary-material"}, sequence 2). The speeds of extrusion and retraction were consistent with previous observations of microtubule motor-mediated transport in living cells. For example, in the sequences shown in [additional file 3](#S3){ref-type="supplementary-material"} particle tracking analysis indicated that extrusion occurred at an average speed of 0.14 μm/s whereas retraction occurred at a speed of around 0.1 μm/s, although peak speeds of movement in other time lapse sequences sometimes approached 2.5 μm/s. Addition of the microtubule depolymerising drug nocodazole to the cell culture medium at concentrations known to induce microtubule disassembly in COS-7 cells followed by imaging every 30 min over a time course of 90 min after drug addition revealed the appearance of small spots of fusion protein scattered throughout the cytoplasm and an increase in diffuse cytoplasmic fluorescence. No evidence of shrinkage or dispersal of the perinuclear aggregates was seen over this timescale (Fig. [2](#F2){ref-type="fig"} panels A and B, arrows). However, both perinuclear and dispersed aggregates were completely immobile in the presence of nocodazole (Fig. [3](#F3){ref-type="fig"}, panels A and B; [additional file 4](#S4){ref-type="supplementary-material"}). When nocodazole was removed and fresh medium added these aggregates gained motility (Fig. [3](#F3){ref-type="fig"}, [additional file 5](#S5){ref-type="supplementary-material"}) and perinuclear accumulations rapidly assembled, progressively increasing in size and intensity (Fig. [2](#F2){ref-type="fig"}, panels C-F). Time-lapse imaging revealed smaller structures moving rapidly towards and incorporating into the perinuclear aggregates (Fig. [3](#F3){ref-type="fig"}; [additional file 5](#S5){ref-type="supplementary-material"}). Coalescence of cytoplasmic aggregates before their transport to the perinuclear accumulations was also evident ([additional file 5](#S5){ref-type="supplementary-material"}). Particle movement after nocodazole wash out was discontinuous and occasionally bidirectional, but with an obvious bias towards retrograde movement (in both frequency and persistence) as would be expected for the assembly of a perinuclear aggregate. Again, average speeds of particle movement were consistent with microtubule motor-mediated translocation at around 0.1 μm/s. In sum, this data indicated that that the formation and dynamic behaviour of the perinuclear EB1-ΔN2-GFP aggregates was critically dependent upon the presence of an intact microtubule cytoskeleton. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Perinuclear EB1-ΔN2-GFP aggregate formation is microtubule-dependent. Panel A. Still image from living cells containing EB1-ΔN2-GFP aggregates. Panel B. The cells shown in panel A following 90 min incubation with 5 μg/ml nocodazole. Arrows indicate the presence of small peripheral aggregates. Panel C. Still image from a living cell expressing EB1-ΔN2-GFP following 4 h incubation in 5 μg/ml nocodazole. Panels D-F. Images of the cell shown in panel C at 30 min intervals following nocodazole wash out. The formation of a single perinuclear aggregate is seen. Bar = 10 μm. ::: ![](1471-2121-6-17-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Microtubule-dependent movement of peripheral EB1-ΔN2-GFP aggregates. Panels A and B. Single frames from [additional file 4](#S4){ref-type="supplementary-material"} showing a living cell expressing EB1-ΔN2-GFP after 2 h incubation in nocodazole. All EB1-ΔN2-GFP aggregates are immobile. Bar = 10 μm. Panels C-F. Single frames from [additional file 5](#S5){ref-type="supplementary-material"} showing a living cell expressing EB1-ΔN2-GFP during the recovery phase following nocodazole wash out. The retrograde transport of a peripheral aggregate and its incorporation into a large perinuclear aggregate is arrowed. Times shown are relative to the first frame in the sequence. Bar = 10 μm. ::: ![](1471-2121-6-17-3) ::: EB1-ΔN2-GFP aggregates are aggresomes ------------------------------------- The behaviour and morphology of the EB1-ΔN2-GFP structures in living cells was reminiscent of the previously reported characteristics of aggresomes \[[@B12]-[@B14]\]. Aggresomes form around the centrosome when the accumulation of an aberrantly folded protein exceeds the ability of the cells protein degradation apparatus to dispose of it \[[@B18]\], and they typically share certain common features. For example, like the EB1-ΔN2-GFP aggregates examined here, aggresome formation but not maintenance is microtubule dependent and in the presence of microtubule poisons newly synthesised protein is found in aggregates dispersed throughout the cytoplasm \[[@B12]\]. We therefore fixed and immunostained EB1-ΔN2-GFP transfected COS-7 cells for other aggresomal markers. Co-immunostaining for GFP and either α-tubulin (Fig. [4](#F4){ref-type="fig"} panels A-C) or γ-tubulin (not shown) confirmed that EB1-ΔN2-GFP aggregates formed around the centrosome in transfected cells. No γ-tubulin immunoreactivity was seen in the aggregates themselves, unlike some aggresomes observed in different systems \[[@B19],[@B20]\]. Co-staining for α-tubulin also indicated that in cells with small aggregates the radial microtubule cytoskeleton was either slightly deformed in the region of the aggregation or unaffected by the presence of the aggregates (Fig. [4](#F4){ref-type="fig"} panels A-C). In some cells, particularly those harbouring large aggregates, evidence of extensive microtubule disorganisation was seen (not shown). This is not a widely reported consequence of aggresome formation and in this case may reflect a dominant-negative inhibition of endogenous EB1 function, since we have previously noted the same effect in COS-7 cells expressing different EB1 N-terminal deletion mutants \[[@B3]\]. Further co-staining experiments demonstrated that the aggregates were immunopositive for both the 20s subunit of the 26S proteasome (Fig. [5](#F5){ref-type="fig"} panels A-C) and ubiquitin (Fig. [5](#F5){ref-type="fig"} panels D-F; refs \[[@B18],[@B21]\]), and indicated that they induced the collapse of the vimentin cytoskeleton (Fig. [5](#F5){ref-type="fig"} panels G-I; ref \[[@B13]\]). However, the ER resident chaperone BiP/GRP78 was not detected in EB1-ΔN2-GFP aggregates (not shown; ref \[[@B18]\]). These data indicate that the EB1-ΔN2-GFP aggregates bear many of the characteristic hallmarks of an aggresome formed by a misfolded cytosolic protein. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### EB1-ΔN2-GFP aggregates are associated with the centrosome. Panels A-C. Cells expressing EB1-ΔN2-GFP were fixed and co-immunostained for GFP (green, panel A) and α-tubulin (red, panel B). A merged image is shown in panel C. EB1-ΔN2-GFP aggregates are clustered around the centrosome. Bar = 10 μm. ::: ![](1471-2121-6-17-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### EB1-ΔN2-GFP aggregates are aggresomes. Cells expressing EB1-ΔN2-GFP were fixed and co-immunostained for GFP (green, panels A, D and G) and either the 20s proteosomal subunit (red, panel B), ubiquitin (red, panel E) or vimentin (red, panel H). Merged images are shown in panels C, F and I. All three aggresome markers were present in EB1-ΔN2-GFP aggregates. Bar = 10 μm. ::: ![](1471-2121-6-17-5) ::: EB1-ΔN2-GFP aggregates contain APC ---------------------------------- EB1-ΔN2-GFP retains the ability to interact with APC \[[@B3]\]. To examine whether this interaction might target APC to EB1-ΔN2-GFP aggresomes we co-immunostained transfected cells with antibodies to APC and GFP and examined them using confocal microscopy. Robust APC immunostaining was detected in the EB1-ΔN2-GFP aggregates (Fig. [6](#F6){ref-type="fig"} panels A-F). Closer examination of single confocal sections indicated that APC immunostaining in the aggresomes was more heterogeneous that that obtained for GFP, but where present completely overlapped with GFP-positive structures (Fig. [6](#F6){ref-type="fig"} panels D-F). This heterogeneity may reflect APC antibody accessibility problems in the centre of the dense aggresomal matrix. We were unable to investigate whether endogenous EB1 was present in the aggresomes as EB1-ΔN2-GFP contains the epitope recognised by the EB1 antibody used here \[[@B3]\]. However, another member of the conserved EB1 protein family, EB3, is also known to interact with APC and p150glued \[[@B9]\] and to localise to growing microtubule tips \[[@B22]\]. COS-7 cells co-express EB1 and EB3 (JMA, unpublished observations). We therefore examined EB3 distribution in transfected cells and found no evidence for EB3 in the EB1-ΔN2-GFP aggregates (Fig. [6](#F6){ref-type="fig"} panels G-I). This indicates that EB3 does not associate with EB1-ΔN2-GFP and suggests that it cannot interact with APC or p150glued in the aggresomes, possibly because the EB1/EB3 binding sites in these proteins are already filled by EB1-ΔN2-GFP. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### APC is present in EB1-ΔN2-GFP aggresomes. Panels A-C. Cells expressing EB1-ΔN2-GFP were fixed and co-immunostained for GFP (green, panel A) and APC (red, panel B). A merged image is shown in panel C. APC is seen in aggresomes in transfected cells and at peripheral sites in adjacent untransfected cells. Bar = 10 μm. Panels D-F. A single 0.365 μm confocal section through a cell expressing EB1-ΔN2-GFP. APC immunoreactivity (panel E) in aggresomes is heterogeneous but only seen in association with GFP immunoreactivity (panels D and F). Bar = 5 μm. Panels G-I. Cells expressing EB1-ΔN2-GFP were fixed and co-immunostained for GFP (green, panel G) and EB3 (red, panel H). EB3 is seen at microtubule tips but is not present in EB1-ΔN2-GFP aggresomes. Bar = 20 μm. Images in panels A-C and G-I are projections of confocal image stacks. ::: ![](1471-2121-6-17-6) ::: Dynein/dynactin components are present in EB1-ΔN2-GFP aggresomes ---------------------------------------------------------------- EB1-ΔN2-GFP can also directly interact with the p150glued subunit of dynactin \[[@B3]\]. Co-immunostaining consistently revealed strong p150glued immunostaining in EB1-ΔN2-GFP aggregates (Fig. [7](#F7){ref-type="fig"} panels A-F). As with APC, close examination of single confocal sections indicated that p150glued immunoreactivity overlapped with but was more heterogeneous than that seen for GFP in these structures (Fig. [7](#F7){ref-type="fig"} panels D-F). Immunostaining for p150glued was also prominent at centrosomes (a normal intracellular localisation for this protein) surrounded by EB1-ΔN2-GFP aggregates (Fig. [7](#F7){ref-type="fig"} panels D-F, arrows). Co-immunostaining also revealed the presence of both the CDIC and p50dynamitin subunits of dynein/dynactin in EB1-ΔN2-GFP aggresomes (Fig. [7](#F7){ref-type="fig"} panels G-L). As with p150glued both of these proteins displayed a heterogeneous distribution in the aggresome and were also present at the centrosome. In general, and bearing in mind the problems involved in trying to compare staining intensities for different proteins, aggresomal immunostaining for these proteins appeared to be weaker and more variable than that obtained for p150glued. For example, strong immunostaining for both CDIC and p50dynamitin was observed most consistently in larger aggregates but was less apparent in smaller aggregates. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Dynein/dynactin subunits are present in EB1-ΔN2-GFP aggresomes. Panels A-C. Cells expressing EB1-ΔN2-GFP were fixed and co-immunostained for GFP (green, panel A) and p150glued (red, panel B). A merged image is shown in panel C. p150glued is seen in aggresomes. Bar = 10 μm. Panels D-F. A single 0.365 μm confocal section through a cell expressing EB1-ΔN2-GFP. p150glued immunoreactivity in aggresomes (panel E) is heterogeneous but only seen in association with GFP immunoreactivity (panels D and F). Centrosomal immunostaining is also observed (arrow). Bar = 5 μm. Panels G-L. Cells expressing EB1-ΔN2-GFP were fixed and co-immunostained for GFP (green, panel G) and either CDIC (red, panel H) or p50dynamitin (red, panel K). Merged images are shown in panels I and K. Heterogeneous CDIC and p50dynamitin immunostaining is seen in aggresomes and at centrosomes. Bars = 10 μm. Images in panels A-C and G-L are projections of confocal image stacks. ::: ![](1471-2121-6-17-7) ::: Fractionation and Western blotting analysis of EB1-ΔN2-GFP expressing cells --------------------------------------------------------------------------- A characteristic of aggresomes is their insolubility in non-ionic detergent solutions \[[@B13],[@B14]\]. We therefore performed a simple fractionation by extracting cells in 0.1% Triton X-100 in PBS and separating soluble and insoluble fractions by centrifugation. The resulting fractions were examined using SDS-PAGE and Western blotting. As shown in Fig. [8A](#F8){ref-type="fig"}, immunoblotting with a GFP-specific antibody revealed the presence of single bands representing GFP, EB1-GFP or EB1-ΔN2-GFP in the soluble fraction of the relevant transfected cell extract (arrowed). Immunoblotting with an EB1-specific antibody gave a similar result for EB1-GFP and EB1-ΔN2-GFP and also revealed that the levels of endogenous soluble EB1 were unchanged by transfection. Immunoblotting with an EB3-specific antibody detected a single immunoreactive band. We also probed soluble cell fractions with antibodies specific for the p150glued, p50dynamitin and CDIC subunits of the dynein/dynactin complex. This indicated that the levels of these proteins in the soluble fraction were unaffected by the expression of GFP or either of the EB1-derived fusion proteins (Fig. [8B](#F8){ref-type="fig"}). ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Fractionation and Western blotting analysis of EB1-ΔN2-GFP aggregates. COS-7 cells were transfected with GFP, EB1-GFP or EB1-ΔN2-GFP for 24 h before extraction in detergent-containing buffer and separation into soluble (panels A and B) and insoluble (panels C and D) fractions and analysis by SDS-PAGE and Western blotting. Panel A.GFP immunoblotting of soluble fractions revealed the presence of single immunoreactive bands for GFP, EB1-GFP and EB1-ΔN2-GFP (arrows). Non-specific cross-reacting bands are evident using this polyclonal antibody. EB1 immunoblotting revealed single bands for EB1-GFP, EB1-ΔN2-GFP and endogenous EB1 (arrows). EB3 immunoblotting revealed a single immunoreactive band in all soluble fractions. Panel B.Immunoblotting for p150glued, p50dynamitin and CDIC indicated that the level of these proteins in soluble fractions were similar in all transfections. Panel C.GFP immunoblotting revealed the presence of two immunoreactive bands in the insoluble fraction of GFP transfected cells (arrow and arrowhead), at least three bands in EB1-ΔN2-GFP transfected cells (arrow and arrowheads) and a single weak band for EB1-GFP (arrow). An overexposed image is shown to highlight the accessory bands. EB1 immunoblotting revealed the presence of two immunoreactive bands in EB1-ΔN2-GFP insoluble fractions (arrow and arrowhead). This blot is also shown overexposed. Panel D.Immunoblotting detected p150glued in the insoluble fraction of all extracts (arrow). Two lower molecular weight bands were specifically detected in the insoluble fraction of EB1-ΔN2-GFP transfected cells (arrowheads). The 43 kDa non-specific band present in the GFP and p150glued panels was present in immunoblots performed on insoluble cellular fractions regardless of the primary antibody used. ::: ![](1471-2121-6-17-8) ::: These analyses were then repeated on the detergent-insoluble fraction from transfected cells. GFP immunoblotting revealed two immunoreactive bands in the insoluble fraction of cells expressing GFP (Fig. [8C](#F8){ref-type="fig"}). The smaller of these (arrowed) corresponded to the expected molecular weight of GFP whereas the larger (arrowhead) is consistent with the size of a monoubiquitylated GFP molecule, suggesting that a proportion of overexpressed GFP in COS-7 cells might be turned over by normal cellular mechanisms of protein degradation. Only weak immunolabelling was detected for EB1-GFP in overexposed blots (Fig. [8C](#F8){ref-type="fig"}, arrowed), consistent with previous observations showing endogenous EB1 to be highly soluble (1). In contrast, EB1-ΔN2-GFP was highly insoluble in transfected cell extracts (Fig. [8C](#F8){ref-type="fig"}). Furthermore, in addition to a major band corresponding to the expected size of EB1-ΔN2-GFP (arrowed), at least two higher molecular weight bands were also detected (arrowheads). The equal spacing between these bands suggests that these represent ubiquitylated forms of the insoluble protein, consistent with the positive ubiquitin immunoreactivity observed in EB1-ΔN2-GFP aggregates by immunostaining of transfected cells (Fig. [5](#F5){ref-type="fig"} panels D-F). The presence of at least one higher molecular weight species of EB1-ΔN2-GFP was confirmed when blots were probed with the EB1 antibody and examined after overexposure (Fig. [8D](#F8){ref-type="fig"}, arrowhead). EB3 and the dynein/dynactin components p50dynamitin and CDIC were essentially undetectable in the insoluble fraction of any of the transfected cell extracts (not shown). p150glued was detected in the insoluble fraction of all extracts (Fig. [8C](#F8){ref-type="fig"}, arrowed) but no evidence for an enrichment of insoluble p150glued in cells expressing EB1-ΔN2-GFP was found. However, two lower molecular weight species of approximately 45 and 15 kDa were specifically detected by the p150glued monoclonal antibody in the insoluble fraction of cells expressing EB1-ΔN2-GFP (Fig. [8C](#F8){ref-type="fig"}, arrowheads). Discussion ========== The data presented in this study indicate that the perinuclear aggregates formed in cells expressing EB1-ΔN2-GFP are aggresomes, although the underlying reason for their formation by this fusion protein remains unclear. Two mechanisms seem possible. First, the initial aggregation occurs as a direct consequence of the interaction between EB1-ΔN2-GFP and p150glued, perhaps coupled to aberrant retrograde transport of this complex, and aggresome formation occurs subsequent to this. A second possibility is that EB1-ΔN2-GFP misfolds, becomes refractive to proteolytic degradation and thereby triggers the formation of an aggresome. Neither EB1-GFP nor any of the other GFP-tagged EB1 deletion mutants generated in our laboratory, including those that do not interact with microtubules but retain an ability to bind APC and p150glued, form aggresomes in transfected cells \[[@B3]\]. This suggests that the first possibility is less likely. Furthermore, our Western blotting analyses indicated that cell extracts from cultures transfected with the EB1-ΔN2-GFP construct contained significantly more fusion protein than those from cultures transfected with the EB1-GFP construct (Fig. [8](#F8){ref-type="fig"}). Since the transfection efficiency for the EB1-GFP and EB1-ΔN2-GFP constructs was similar, equal amounts of transfected cell extracts were analysed and the base expression vector used for both constructs was identical, the best explanation for the higher levels of EB1-ΔN2-GFP relative to EB1-GFP in our experiments is that EB1-GFP was turned over normally within the cell whereas EB1-ΔN2-GFP was not. This data therefore supports the second mechanism suggested above. However, if aggregation is a response to EB1-ΔN2-GFP misfolding, then the presence of both p150glued and APC in the aggregates suggests that this misfolding is partial since the binding sites for both of these proteins appears to be functional (Figs. [6](#F6){ref-type="fig"} and [7](#F7){ref-type="fig"}; ref \[[@B3]\]). We therefore propose that EB1-ΔN2-GFP in COS-7 cells adopts a conformation where APC and p150glued binding are maintained but the fusion protein is recognised by the cellular stress response pathway for misfolded proteins, perhaps as a result of a disordered N-terminal region arising from the loss of the CH domain from the EB1 N-terminus. As this partially misfolded protein is resistant to degradation it accumulates within the cytoplasm and an aggresome is formed. Other investigators have previously described the formation of aggresomes by a cytosolic GFP fusion protein (GFP-250) in living cells \[[@B14]\]. EB1-ΔN2-GFP aggresome formation appeared to correlate well with that observed for GFP-250, particularly during aggresome assembly following nocodazole treatment and wash out. However, previous ultrastructural studies have suggested that aggresomes represent an \"aggregate of aggregates\" formed by an accumulation of individual particles without further coalescence into a cohesive structure \[[@B12]-[@B14]\]. Our time-lapse studies suggest that this is not the case with EB1-ΔN2-GFP. In our system aggresomes exhibited dynamic microtubule-dependent linear extensions. These typically resembled beads on a string with one end remaining attached to the main structure of the aggresome. Detachment and anterograde movement of individual particles from aggresomes was never seen. These observations indicate that the brighter structures in EB1-ΔN2-GFP aggresomes must be linked in some way and, since extensions were never seen to fully detach from aggresomes, this linkage is strong enough to resist the forces generated by microtubule motors. Together with our data showing that microtubules are not necessary to maintain the pericentrosomal location of pre-existing EB1-ΔN2-GFP aggresomes, this suggests that these structures possess an inherent cohesiveness and are not simply a collection of individual elements maintained in close proximity to the centrosome by dynein motor activity. At present it remains unclear whether this dynamic behaviour is a universal feature of aggresomes or is restricted to those formed by EB1-ΔN2-GFP. However, we note that associations with kinesin II have recently been described for both p150glued \[[@B23]\] and APC \[[@B24]\]. Since both proteins are clearly present in EB1-ΔN2-GFP aggregates these interactions could potentially contribute to the dynamic behaviour of these structures. Recent studies of aggresome formation by mutant CFTR and SOD proteins in cells treated with proteosome inhibitors indicated that the dynein/dynactin components CDIC, p50dynamitin and p150glued were all recruited to aggresomes \[[@B15],[@B25]\]. This raises the possibility that the p150glued observed in EB1-ΔN2-GFP-induced aggresomes was present as a normal cofactor for aggresome formation rather than as a specific EB1-ΔN2-GFP ligand. Furthermore, an increase in the amounts of p150glued and p50dynamitin in the detergent insoluble fraction of cells containing mutant SOD-induced aggresomes was reported \[[@B25]\]. In our experiments no increase in the amount of insoluble full-length forms of these proteins was observed. However, in contrast to previous studies we found lower molecular weight p150glued species in the detergent-insoluble fraction of cells expressing EB1-ΔN2-GFP. The p150glued antibody used in this study recognises an epitope at the N-terminus of the protein \[[@B26]\], the same region that mediates the EB1-p150glued interaction \[[@B3]\] but distinct from the regions responsible for the interactions with other dynein/dynactin subunits \[[@B8]\]. It seems possible that the bands detected in our immunoblots represent N-terminal p150glued fragments tightly bound to insoluble EB1-ΔN2-GFP and inaccessible to the proteosome. An analogous mechanism has been proposed to explain the limited proteolytic processing of NF-κB precursors, where it is suggested that a close association with a partner molecule inhibits the processive degradation of the NF-κB p50 domain \[[@B27]\]. To explain the discrepancies between our data and that presented in the study of mutant SOD-induced aggresomes we therefore propose that in the latter case the dynactin complex is recruited to but not trapped within aggresomes as part of a normal cellular response, whereas in our system the aggresome-associated p150glued is tightly bound to EB1-ΔN2-GFP. Subsequent degradation of exposed regions of the molecule destroys the binding sites for p150glued-associated dynein/dynactin subunits, precluding their stable co-incorporation into the aggresome. However, binding to proteolytically-resistant EB1-ΔN2-GFP molecules preserves the p150glued fragments detected by the monoclonal antibody used in our study. Conclusion ========== In this work we were fortunate in being able to examine aggresome formation by a defined cytosolic protein that possesses direct interactions with two other well-characterised proteins, p150glued and APC. This allowed us to show that these proteins were present in the aggresomes arising from EB1-ΔN2-GFP expression in transfected cells. Our data therefore suggests that a misfolded protein can trap its normal endogenous cellular ligands in an aggresome, potentially resulting in the degradation of these partners. This could have implications for our understanding of human diseases thought to involve protein misfolding and the formation of aggresome-like structures, such as Parkinsons disease and other neurodegenerative disorders \[[@B12]\]. Our data provides a proof of principle that the pathologies characteristic of these diseases could arise in part from the inappropriate trapping and degradation of the normal physiological partners of misfolded proteins in aggresomes and inclusion bodies. Methods ======= Cells ----- COS-7 cells were cultured and drug treatments performed as described previously \[[@B1]\]. Transfections were performed using GeneJuice (Novagen) according to the manufacturers instructions. Manual counting of GFP immunostained transfected cell populations indicated that transfection efficiencies at 18 h post-transfection were consistently in the region of 60% for the GFP expression plasmid and somewhat lower at around 25% for the EB1-GFP and EB1-ΔN2-GFP expression plasmids. Antibodies and reagents ----------------------- Monoclonal antibodies specific for EB1, EB3, p150glued and p50dynamitin were obtained from Transduction Laboratories. Monoclonal antibodies specific for γ-tubulin and cytoplasmic dynein intermediate chain (CDIC) were obtained from Sigma. A mouse monoclonal antibody specific for BiP was obtained from BD Biosciences. A rat anti α-tubulin antibody was obtained from Serotec. Rabbit polyclonal antibodies against ubiquitin and the 20S proteosomal subunit were obtained from DAKO and Affiniti Research Products respectively. Rabbit polyclonal and mouse monoclonal anti-GFP antibodies were obtained from Clontech. A rabbit polyclonal antibody specific for APC (M-APC; ref \[[@B16]\]) was a kind gift from Dr Inke Nathke, University of Dundee, UK. All secondary antibodies were highly cross-adsorbed Alexa 488 and 594 conjugates obtained from Molecular Probes. Nocodazole was obtained from Sigma. Plasmids -------- The GFP, EB1-GFP and EB1-ΔN2-GFP expression plasmids used in this work have been described previously \[[@B3],[@B7],[@B17]\]. Immunofluorescence ------------------ COS-7 cells were cultured and transfected on glass coverslips. Cultures were processed for immunocytochemistry using methanol fixation 18 h after transfection and imaged using a Leica TCS-SP confocal microscope as described previously \[[@B7]\]. Alternatively, cells were imaged by fluorescence microscopy using a Zeiss Axiovert 200 inverted microscope coupled to an Orca II ER CCD camera controlled by AQM6 software (Kinetic Imaging, Nottingham, UK). Figures were assembled using Adobe Photoshop 7. Time-lapse fluorescence imaging ------------------------------- Cells were grown and transfected in 35 mm glass-bottomed culture dishes (Iwaki brand; Asahi Techno Glass Corporation, Japan) obtained from Bibby Sterilin. 14--18 h after transfection the cell culture medium was replaced by 1.5 ml of pre-warmed medium containing 20 mM HEPES. The cells were then transferred to a Zeiss Axiovert 200 inverted microscope with the stage enclosed in a heated chamber (Solent Scientific, UK) maintained at 37°C. After 15 min equilibration cells were examined by fluorescence microscopy using a Zeiss Plan Apochromat 63X/1.4NA oil immersion lens or by fluorescence and phase contrast microscopy using a Zeiss A-Plan 40X/0.65NA dry lens. An excitation/emission filterset optimised for eGFP imaging was used for fluorescence microscopy (Chroma Technology Corp., Brattleboro, USA; filterset ID 86007). Time-lapse images were obtained using Ludl shutters and a Hamamatsu Orca II ER camera. Microscope, camera, filterwheels and shutters were controlled by AQM 6 software (Kinetic Imaging, Nottingham, UK). Typically, images were obtained using 1 × 1 binning and exposure times of less than 250 ms/frame with time-lapse intervals ranging between 5s and 30s. Time-lapse image series were saved as uncompressed AVI files then cropped, compressed and converted into Quicktime movies using Adobe ImageReady 7. Some movies are presented using an inverted greyscale colour look-up table to enhance the visibility of small structures. Particle tracking analyses were performed using Motion Analysis software from Kinetic Imaging. Cell extractions, SDS-PAGE and Western blotting ----------------------------------------------- Cells were harvested by scraping and collected by centrifugation 18 h after transfection. Cell pellets were resuspended in ice-cold PBS containing 0.1% Triton X-100, a mixture of protease inhibitors (Complete EDTA-free tablets, Roche Diagnostics, Germany), 2 mM EDTA, 50 mM sodium fluoride and 100 μM sodium orthovanadate (PBS/TX100) then incubated on ice for 5 min with occasional mixing. Insoluble and soluble fractions were separated by centrifugation at 12000 g for 5 min in a benchtop microcentrifuge. Once the supernatant was removed to a fresh tube the pellet was resuspended in a volume of PBS/TX100 equivalent to that of the removed supernatant. An equal volume of 2x concentrated SDS-PAGE sample loading buffer containing 5 mM DTT was added to both the soluble and insoluble fractions, followed by boiling for 5 min. To reduce viscosity pellet samples were passed repeatedly through a narrow gauge needle attached to a syringe before gel loading. Unused sample was snap frozen in liquid nitrogen and stored at -80°C until needed. SDS-PAGE and Western blotting were performed essentially as described previously \[[@B1]\]. Blots were probed using fluorescently conjugated secondary antibodies and visualised using a Li-Cor Odyssey quantitative Western Blotting system. Equal loading of different cell extracts onto SDS-PAGE gels and subsequent transfer on to Western blotting membranes was confirmed by quantitative immunoblotting with a monoclonal β-actin antibody. Gel images were processed and converted into greyscale images using Adobe Photoshop 7. The images of some blots shown in this study have been cropped due to size considerations; only regions where no specific immunoreactivity was observed after overexposure have been removed. Authors\' contributions ======================= NPR contributed to the live cell imaging, immunostaining and Western blotting studies and helped draft the manuscript. KM contributed to the live cell imaging, Western blotting and immunostaining studies and helped to draft and revise the manuscript. TL maintained cell cultures, participated in the Western blotting studies and helped revise the manuscript. MA contributed to the live cell imaging and immunostaining studies and helped to revise the manuscript. JMA and EEM conceived the study, participated in its design and execution, contributed reagents and helped to both draft and revise the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 2 EB1-ΔN2-GFP aggregates in living COS-7 cells. Sequence 1: image capture rate was 1 frame/5s using a 40X lens, playback rate is 20 frames/s. The dynamic behaviour of EB1-ΔN2-GFP aggregates is again evident. Sequence 2: image capture rate was 1 frame/5s using a 40X lens, playback rate is 20 frames/s. This movie shows the cells in sequence 1 imaged by phase-contrast microscopy and allows the behaviour of these cells to be compared with two adjacent untransfected cells. The sequence begins 30s after the end of sequence 1. EB1-ΔN2-GFP aggregates are visible as phase-dense structures but their presence has no obvious effect upon cell behaviour. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 1 EB1-ΔN2-GFP aggregate morphology and dynamics. Image capture rate was 1 frame/5s using a 63X oil immersion lens, playback rate is 20 frames/s. Aggregates consist of brighter puncta in a matrix of lower fluorescence intensity. The dynamic nature of the aggregate can be seen. Structures can be observed extending away from the aggregate while remaining attached to it; on obvious example of this can be seen around the nuclear periphery at the bottom left of the aggregate. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Extrusion and retraction of a linear structure from an EB1-ΔN2-GFP aggregate. Image capture rate was 1 frame/5s using a 63X oil immersion lens, playback rate is 20 frames/s. Sequence 1: the extension of a linear fluorescent structure from an EB1-ΔN2-GFP aggregate is shown; the aggregate itself is overexposed to allow better resolution of this structure, which consists of brighter puncta linked to each other and the aggregate by a ribbon of less fluorescent material. Sequence 2: imaging was initiated approximately 1 min after the end of sequence 1. Retraction of the structure extended by the EB1-ΔN2-GFP aggregate can be seen. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 Movement of EB1-ΔN2-GFP aggregates is microtubule-dependent. Image capture rate was 1 frame/5s using a 40X lens, playback rate is 20 frames/s. This movie shows aggregate behaviour in cells after 2 h incubation in nocodazole. In the presence of the drug the aggregates are completely immobile. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Perinuclear EB1-ΔN2-GFP aggregate assembly is microtubule-dependent. Image capture rate was 1 frame/10s using a 63X oil immersion lens, playback rate is 20 frames/s. Cells were incubated in nocodazole for 4 h followed by a wash and the addition of fresh imaging medium without drug. Imaging was initiated 30 min after nocodazole wash out. In the absence of drug EB1-ΔN2-GFP aggregates regain motility. Movement is intermittent and with an overall retrograde bias leading to the growth of perinuclear aggregates. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This research was supported by the Medical Research Council (UK), Yorkshire Cancer Research and Cancer Research UK. We wish to thank Drs Phil Robinson and Helen Ardley of the Molecular Medicine Unit at St James\'s University Hospital, Leeds for useful discussions and the gift of antibody reagents.	PubMed Central	2024-06-05T03:55:56.488858	2005-4-6	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090566/", "journal": "BMC Cell Biol. 2005 Apr 6; 6:17", "authors": [ { "first": "Nick P", "last": "Riess" }, { "first": "Kelly", "last": "Milward" }, { "first": "Tracy", "last": "Lee" }, { "first": "Matthew", "last": "Adams" }, { "first": "Jon M", "last": "Askham" }, { "first": "Ewan E", "last": "Morrison" } ] }
PMC1090567	Background ========== G protein-coupled receptors (GPCRs) represent the largest group of integral membrane proteins involved in signal transduction and are the most important targets of clinically marketed drugs \[[@B1]-[@B3]\]. The known GPCRs mediate messages from ligands as diverse as neurotransmitters, lipid mediators, hormones, and sensory stimuli \[[@B4]\]. The classical scheme of GPCR signaling implies that agonist-induced conformational changes in receptor molecule will result in activation of cognate G proteins and subsequently in the regulation of downstream effectors, second messengers, and the activation of protein kinases, for example \[[@B4]\]. However, recent work has indicated that GPCR signaling is subject to complex, cell-type specific regulation, involving a plethora of kinases, as well as newly-identified signaling partners, such as regulators of G protein signaling (RGS) \[[@B5]\], and activators of G protein signaling (AGS) \[[@B6]\]. Nitric oxide (NO) is a unique gaseous messenger generated in vivo by three isoforms of NO synthases (NOS). The established mode of NO signaling is through the activation of the hemoprotein, soluble guanylyl cyclase, resulting in increased production of the second messenger cGMP. However, accumulating evidence points towards cGMP-independent mechanisms by which NO can react with proteins, forming covalent post-translational modifications \[[@B7]\]. S-nitrosothiols (RSNOs) are biological metabolites of NO, that may prolong and spatially extend the in vivo actions of locally produced NO \[[@B8]\]. NO and RSNOs can reversibly react with free SH-groups of target cysteine (Cys) residues, including those in proteins, leading to S-nitrosylation and/or S-thiolation (disulfide linkage of low-molecular weight thiols to proteins) \[[@B8]-[@B17]\]. A broad functional spectrum of potential S-nitrosylation target proteins is currently recognized. A growing list of targets include ion channels, transporters, transcription factors, signaling proteins, metabolic enzymes, as well as respiratory proteins \[[@B7],[@B14],[@B18]-[@B20]\]. Although individual components of the GPCR signaling machinery are implicated as potential targets of reactive oxygen species (ROS), including NO \[[@B21]-[@B35]\], a broader view on how NO, and RSNOs in particular, might modulate GPCR signaling, has not been established. To begin to address these issues, we have studied how exogenous RSNOs affect receptor-mediated G protein activity -- a very proximal step of GPCR signal transduction -- by studying the signaling of several receptors that couple to the G~i~family of heterotrimeric G proteins. This family consists of both pertussis toxin sensitive (Gα~i1-3~, Gα~o~, transducin, gustducin) and insensitive (Gα~z~) members. We applied the powerful functional approach of \[^35^S\]GTPγS autoradiography in brain cryostat sections, as this technique allows selective detection of receptor-dependent G protein activity simultaneously in multiple brain regions with minimal disturbance of the GPCR microenvironment \[[@B36]\]. We anticipated that accessibility of target thiols would be minimally disturbed in cryostat sections. Moreover, it is increasingly recognized that specialized plasma membrane microdomains (variously described as detergent-resistant fractions, low-density fractions, lipid rafts, or caveolae) act as unique signaling platforms with specific enrichment of GPCRs, their cognate G proteins, as well as effectors \[[@B37]-[@B40]\]. Such an enrichment is thought to be well-preserved in cryostat sections, but might be compromised to a variable extent \[[@B38]\], or even lost in bulk membrane preparations obtained using traditional protocols. To complement the autoradiography approach, we used membrane and lysate \[^35^S\]GTPγS binding assays to more systematically study the effects of RSNOs on a panel of G~i~-coupled receptors in native tissues. Signaling of selected receptors was further studied after their heterologous expression in Chinese hamster ovary (CHO) cells. Our studies reveal highly receptor-specific modulation of GPCR signaling by RSNOs, as signaling of some receptors can be amplified, or inhibited, whereas for others, the activity is only marginally affected by comparable treatments. The GPCR itself and/or its native signaling partners seem to act as potential targets of RSNO action, and therefore their modulation may be diminished, or even totally masked under heterologous expression. Results ======= Exogenous RSNOs modulate GPCR signaling via mechanisms likely involving S-nitrosylation --------------------------------------------------------------------------------------- We used the functional approach of \[^35^S\]GTPγS autoradiography, as this technique allows selective detection of receptor-stimulated G~i~protein activity simultaneously in multiple brain structures with minimal disturbance of the GPCR microenvironment \[[@B36]\]. We focused the initial experiments on three G~i~-linked receptors, namely M2/M4 AChRs, the P2Y~12~purinoceptor, and the LPA~1~receptor, as G protein activity upon stimulation of these receptors has been previously characterized using the autoradiography approach and each receptor shows a unique anatomical distribution pattern in the developing rat brain \[[@B36],[@B41]-[@B43]\]. As depicted in Figure [1](#F1){ref-type="fig"}, pretreatment of brain sections with freshly prepared GSNO (0.5 mM) had distinct effects on basal and receptor-stimulated \[^35^S\]GTPγS binding responses. In GSNO-treated sections, basal G protein activity was increased throughout the gray matter areas and this effect was fully reversed in the presence of excess thiol, either in the form of dithiotreitol (DTT) or reduced glutathione (GSH). The autoradiography images were quantified for selected brain regions and these results are shown in Supplementary Figures 1 and 2 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]. Consistent with the anatomical distribution of atropine-sensitive and M2/M4 AChR-dependent G protein activity \[[@B36],[@B44],[@B45]\], the cholinergic agonist carbachol (CCh) stimulated \[^35^S\]GTPγS binding to multiple gray matter regions, including the striatum (Str), the thalamic structures, with the most intense responses in the superficial gray layer of the superior colliculus (SuG), as well as various brainstem (bs) nuclei. In all visible regions, CCh-stimulated G protein activity was robustly amplified by GSNO. This was particularly evident in the above-mentioned M2/M4 receptor-enriched anatomical loci. It is noteworthy that the GSNO effects were fully reversed in all regions in the presence of DTT or GSH (Figure [1](#F1){ref-type="fig"}, Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). As further illustrated in Figure [1](#F1){ref-type="fig"}, the P2Y receptor agonist 2-methylthio-ADP (2MeSADP) activated G proteins both in gray and white matter regions, producing a heterogeneous activity pattern with characteristic \"hot spot\" appearance, as described earlier \[[@B41],[@B42]\]. Previous studies have established that the 2MeSADP-stimulated G protein activity in rat brain sections is mediated by a P2Y receptor subtype that pharmacologically corresponds to P2Y~12~\[[@B41],[@B42]\]. In contrast to the robust amplification of M2/M4 receptor-dependent G protein activity, GSNO clearly inhibited P2Y~12~receptor signaling (Figure [1](#F1){ref-type="fig"}, Supplementary Figures 1 and 2 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). This inhibition was evident throughout the effective agonist concentration range (10^-7^-10^-4^M 2MeSADP), and responses to the endogenous agonist ADP (5 × 10^-5^- 10^-3^M) were similarly blunted (Supplementary Figure 2 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). The GSNO-effect was fully reversible upon addition of DTT or GSH (Figure [1](#F1){ref-type="fig"}, Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). In the developing rat brain, LPA-stimulated G~i~protein activity is largely restricted to the myelinating white matter tracts \[[@B36],[@B41],[@B43]\] (Figure [1](#F1){ref-type="fig"}), closely reflecting the anatomical distribution of LPA~1~receptor subtype \[Ref. 36, and references therein\]. Similar to the P2Y~12~responses, LPA~1~receptor responses were suppressed in GSNO-treated sections throughout the white matter tracts. Also this effect was fully reversed in the presence of thiols (Figure [1](#F1){ref-type="fig"}, Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### S-nitrosoglutathione (GSNO) reversibly modulates basal and receptor-dependent G protein activity in rat brain cryostat sections. \[^35^S\]GTPγS autoradiography of sagittal brain sections was conducted using a 3-step protocol with DPCPX (10^-6^M) present throughout steps 2 and 3, as detailed in the Methodssection. Where indicated, GSNO (0.5 mM) was present for 60 min during the GDP loading (step 2). When used, DTT (1 mM) or GSH (1 mM) were present during the \[^35^S\]GTPγS labeling (step 3). The muscarinic agonist, carbachol (CCh, 10^-4^M), the P2Y receptor agonist 2-methylthio-ADP (2MeSADP, 10^-5^M) or lysophosphatidic acid (LPA, 5 × 10^-5^M in 0.1% fatty acid free BSA) were present in step 3. In the control panel (left), the anatomical loci where receptor agonists typically activate G proteins are indicated. Note GSNO-dependent overall increase in basal G protein activity, as well as robust amplification of CCh-stimulated G protein activity in several gray matter regions visible at this sagittal plane, most notably the brain stem (bs) nuclei, the striatum (Str), and the superficial gray layer of the superior colliculus (SuG). Note also clear attenuation of 2MeSADP-stimulated responses in all brain regions, and blunting of LPA-stimulated responses, especially in the white matter areas, including the corpus callosum (cc), the fimbria of the hippocampus (fi) and the cerebellar white matter (Cbw). Scale bar = 2 mm. For quantitative data on selected brain regions, see Supplementary Figs. 1 and 2 in [additional file 1](#S1){ref-type="supplementary-material"}. ::: ![](1471-2121-6-21-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### GSNO modulates GPCR signaling in a dose-dependent manner. \[^35^S\]GTPγS autoradiography was conducted using a 3-step protocol with DPCPX (10^-6^M) present throughout steps 2 and 3, as detailed in the Methodssection. GSNO was present at the indicated concentrations for 60 min during the GDP loading (step 2). Carbachol (CCh, 10^-4^M), 2MeSADP (10^-6^M) or LPA (5 × 10^-5^M in 0.1% fatty acid free BSA) were present in step 3. Note dose-dependent amplification of CCh-stimulated G protein activity, most evident at this coronal plane in the cerebral cortex (Cx), the striatum (Str), and the thalamus (Thal). Note also dose-related attenuation of 2MeSADP- and LPA-stimulated responses, especially in the white matter regions, including the corpus callosum (cc), the fimbria of the hippocampus (fi), and the striatal white matter (Sw). Scale bar = 2 mm. For quantitative data on selected brain regions, see Supplementary Figs. 1 and 2 in [additional file 1](#S1){ref-type="supplementary-material"}. ::: ![](1471-2121-6-21-2) ::: GSNO is present in significant amounts (\~15 pmol/mg protein) in the brain tissue and it is thought to act as a physiological carrier of NO for S-nitrosylation reactions \[[@B46],[@B47]\]. As shown in Figure [2](#F2){ref-type="fig"} and Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\], GSNO modulated GPCR responses in a dose-dependent manner, being effective at the submillimolar concentration range. However, the threshold and maximal concentrations needed for the modulation of distinct receptors slightly varied. Potentiation of M2/M4 receptor signaling was evident already with 0.05 mM GSNO, but statistically significant responses were obtained using 0.2 -- 1 mM GSNO (Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). P2Y~12~receptor responses were inhibited with 0.2 mM GSNO and higher concentrations (Figure [2](#F2){ref-type="fig"}, Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). LPA~1~receptor signaling was only marginally affected with 0.05 mM GSNO, but was severely blunted with 0.2 mM GSNO and higher concentrations (Figure [2](#F2){ref-type="fig"}, Supplementary Figure 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). Collectively, these experiments demonstrate that GSNO, a physiologically relevant RSNO, can modulate GPCR signaling in discrete brain regions in a receptor-specific and fully reversible manner. Various RSNOs, including S-nitrosocysteine (CysNO) (Figure [3](#F3){ref-type="fig"}, Supplementary Table 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]), S-nitrosocysteamine, S-nitroso-L-cysteinylglycine (CysNO-Gly), L-γ-glutamyl-S-nitrosocysteine (Glu-CysNO), and S-nitroso-N-acetyl-D,L-penicillamine (SNAP) -- a compound with a sterically hindered SNO group -- mimicked the effects of GSNO on the three studied receptors (Figure [3](#F3){ref-type="fig"}, Supplementary Figure 2 \[see [additional file 1](#S1){ref-type="supplementary-material"}\], and data not shown). When equimolar concentrations (0.5 mM) of GSNO and SNAP were compared, SNAP equally well suppressed P2Y~12~receptor responses, whereas GSNO more efficiently modulated M2/M4 and LPA~1~receptor responses (Supplementary Figure 3 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). The modulatory effect of CysNO on the three receptors was reversed by addition of DTT or cysteine (Supplementary Figure 4 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). However, when \"aged\" CysNO was used (the stock solution was left to stand for 2 days in ambient light, oxygen and temperature), receptor-stimulated G protein activity was almost completely abolished (Supplementary Figure 5 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### S-nitrosocysteine (CysNO) mimics the effects of GSNO in modulating GPCR signaling, whereas sodium nitroprusside (SNP) and 8-bromo cyclic GMP (8Br-cGMP) do not. \[^35^S\]GTPγS autoradiography was conducted using a 3-step protocol with DPCPX (10^-6^M) present throughout steps 2 and 3, as detailed in the Methodssection. Where indicated, CysNO (1 mM), SNP (0.5 mM), or 8Br-cGMP (0.25 mM) were present for 60 min during the GDP loading (step 2). Carbachol (10^-4^M), 2MeSADP (10^-6^M), or LPA (5 × 10^-5^M in 0.1% fatty acid free BSA) were present in step 3. Note amplification of CCh-stimulated G protein activity by CySNO, most evident at this coronal plane in the cerebral cortex (Cx), the thalamus (Thal), including the superficial gray layer of the superior colliculus (SuG). Note also CysNO-dependent attenuation of 2MeSADP- and LPA-stimulated responses, most evident in the corpus callosum (cc) and the fimbria of the hippocampus (fi). Scale bar = 2 mm. For quantitative data on selected brain regions, see Supplementary Table 1 in [additional file 1](#S1){ref-type="supplementary-material"}. ::: ![](1471-2121-6-21-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Effects of GSNO treatment on agonist dose-response parameters in \[^35^S\]GTPγS binding assays of various G~i~-coupled receptors in their native cellular environment. Membranes or lysates were preincubated in control conditions or in the presence of 0.5 mM GSNO for 30 min. Values are mean ± SE from three to four independent experiments performed in duplicate. E~max~is expressed in percentage over basal with nonspecific binding subtracted. ::: Control GSNO -------------------------------- -------------- --------- ------------------ --------------- *Rat forebrain membranes* M2/M4 mAChRs (CCh) -4.94 ± 0.05 176 ± 2 -5.55 ± 0.11\\ 207 ± 4\\ LPA~1~(LPA) -6.84 ± 0.26 140 ± 7 -6.53 ± 0.40 116 ± 3\* Cannabinoid CB~1~(CP55940) -7.84 ± 0.07 223 ± 3 -7.69 ± 0.08 183 ± 2\\\* Adenosine A~1~(2ClAdo) -6.90 ± 0.04 256 ± 3 -6.87 ± 0.12 234 ± 7\* μ-opiate (DAMGO) -6.85 ± 0.09 167 ± 3 -7.31 ± 0.19 164 ± 5 ORL1 (Nociceptin) -8.66 ± 0.12 173 ± 3 -8.78 ± 0.10 161 ± 2\* *CHO cell lysates* LPA (LPA) -7.37 ± 0.11 202 ± 4 -7.05 ± 0.12 176 ± 4\\ *Human platelet membranes* P2Y~12~(MeSADP) -8.19 ± 0.03 279 ± 2 -8.30 ± 0.08 195 ± 3\\\* α~2A~-adrenoceptor (NA) -5.33 ± 0.08 161 ± 2 -5.49 ± 0.09 182 ± 3\\ \* Statistically different from control (P \< 0.05) \\ Statistically different from control (P \< 0.01) \\\* Statistically different from control (P \< 0.001) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### GSNO only marginally affect opiate related receptor (ORL1), μ opioid receptor (MOR) and adenosine A~1~(Ado A~1~) receptors signaling in brain sections. \[^35^S\]GTPγS autoradiography was conducted using a 3-step protocol with adenosine deaminase (ADA, 1 U/ml) present throughout steps 2 and 3, as detailed in the Methodssection. Where indicated, GSNO (1 mM) was present for 60 min during the GDP loading (step 2). Protease inhibitor cocktail was included in step 2 for brain sections used for testing ORL1 and MOR responses. Receptor agonists nociceptin (ORL1), DAMGO (MOR), and 2-chloroadenosine (2ClAdo) (adenosine A~1~receptor) were present at submaximal concentrations during step 3. Note wide distribution of nociceptin-responsive brain regions, including the cerebral cortex (Cx). Note also robust response to DAMGO in the MOR-enriched striatal patches (Sp), as well as the relatively GSNO-resistant, and widely distributed adenosine A~1~receptor-dependent signal throughout the sagittal plane. Scale bar = 2 mm. For quantitative data on selected brain regions, see Supplementary Table 2 in [additional file 1](#S1){ref-type="supplementary-material"}. ::: ![](1471-2121-6-21-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### The P2Y~12~receptor is not a direct target of RSNO action. Human P2Y~12~receptor was stably expressed in CHO cells and agonist-stimulated G protein activity was determined in control conditions and in membranes pretreated for 30 min with 0.5 mM GSNO or CysNO, as detailed in the Methodssection. There were no statistical differences in the agonist potency \[log (EC~50~): control -9.20 ± 0.09; GSNO -9.14 ± 0.07; CysNO -9.27 ± 0.14) or efficacy \[E~max~(% basal): control 182 ± 3 %; GSNO 180 ± 2; CysNO 179 ± 5%\]. Values are mean ± SE from three independent experiments performed in duplicate. ::: ![](1471-2121-6-21-5) ::: Previous studies have suggested that RSNOs can act as NO^+^, NO^.^, and NO^-^donors under physiological conditions \[[@B9]\]. The RSNO effects on GPCR responses were not mimicked by the NO donor sodium nitroprusside (SNP), when tested at similar (0.5 mM) concentrations (Figure [3](#F3){ref-type="fig"}, Supplementary Table 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). Furthermore, the NO^+^-releasing NO donor, nitrosodium tetrafluoroborate (NOBF~4~, 0.5 mM) (data not shown), hydrogen peroxide (1 mM H~2~O~2~+ 0.2 mM FeSO~4~) (data not shown), or the cyclic GMP analog 8-bromo-cyclic GMP (0.25 mM) (Figure [3](#F3){ref-type="fig"}, Supplementary Table 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]) were largely ineffective. For selected brain regions, quantitative autoradiography data on CysNO, SNP and 8-bromo-cyclic GMP are presented in the Supplementary Table 1 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]. The above results demonstrate that the effects of NO-related species was shared by -and restricted to -- different classes of RSNO compounds, suggesting that S-nitrosylation rather than other types of NO reactions, or cGMP-dependent mechanisms, were involved. According to the S-nitrosylation scheme, treatment with exogenous RSNOs should result in transnitrosylation of potential protein thiols (R-SNO + Protein-SH ↔ R-SH + Protein-SNO). To demonstrate the presence of SNO moieties in GSNO-treated brain section, we used the indirect approach where heterolytic cleavage of S-NO bond with HgCl~2~generates nitrite which can be measured by a colorimetric method. To this end, brain sections were treated with GSNO (0.5 mM), and after thorough washes, the sections were incubated further in the absence or presence of HgCl~2~(10^-4^M). These experiments (shown in Supplementary Figure 6 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]) revealed that HgCl~2~-catalyzed nitrite formation was significantly higher in GSNO treated sections than that in control conditions (2.46 ± 0.13 vs. 0.61 ± 0.14 nmol NO~2~^-^per four coronal brain sections, mean ± SE, n = 3, P \< 0.001). Collectively these experiments suggest that the RSNO-evoked and receptor-specific modulation of GPCR signaling in brain sections most likely involves S-nitrosylation mechanisms. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### GSNO accelerates the rate of M2/M4 AChR-stimulated \[^35^S\]GTPγS binding to rat forebrain membranes but has no effect on basal guanine nucleotide exchange rate. Membranes were preincubated for 30 min in control conditions or in the presence of 0.5 mM GSNO, and time-response for basal and CCh-stimulated (10^-4^M\] \[^35^S\]GTPγS binding was determined, as detailed in the Methodssection. Values represent specific binding (mean ± SD of duplicate determinations) from one representative experiment that was replicated three times with similar outcome. ::: ![](1471-2121-6-21-6) ::: RSNOs modulate GPCR signaling in native tissues in a highly receptor-specific manner ------------------------------------------------------------------------------------ Although \[^35^S\]GTPγS autoradiography offers the advantage of monitoring G protein activity simultaneously in multiple brain regions with minimal disturbance of the GPCR microenvironment, generating quantitative data from the autoradiography images is relatively tedious. As a complementary approach, we tested the effect of RSNOs on agonist potency and efficacy for several additional G~i~-coupled receptors using classical membrane and lysate \[^35^S\]GTPγS binding assays. The results of these experiments are summarized in Table [1](#T1){ref-type="table"}. There was one major difference from the situation with brain sections; in the rat forebrain membrane preparations GSNO did not significantly affect basal G protein activity (102 ± 2 % control, mean ± SE, n = 26, see also Figures [6](#F6){ref-type="fig"} and [7](#F7){ref-type="fig"}). However, GSNO modulation of receptor-mediated responses was found to be highly receptor-specific. In line with the autoradiography data, M2/M4 receptor signaling was markedly amplified (both the potency and the efficacy of agonist increased) whereas LPA-evoked signaling efficacy (but not agonist potency) was significantly decreased in the rat forebrain membrane preparations. Moreover, the efficacy of cannabinoid CB~1~receptor signaling was significantly inhibited with no concomitant change in agonist potency (Table [1](#T1){ref-type="table"}). \[^35^S\]GTPγS autoradiography studies further indicated that CB~1~receptor signaling was similarly inhibited in various CB~1~receptor-enriched brain regions, including the cerebral cortex, the hippocampus and the globus pallidus (Supplementary Figure 7 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]). On the other hand, signaling via other widely distributed receptors, such as adenosine A~1~, μ-opioid (MOR) and opiate-related receptor (ORL1), was only marginally (A~1~and ORL1), or not detectably (MOR), altered in bulk membrane preparations (Table [1](#T1){ref-type="table"}) or in brain cryostat sections (Figure [4](#F4){ref-type="fig"}). For the three receptors, quantitative autoradiography data on selected brain regions are presented in Supplementary Table 2 \[see [additional file 1](#S1){ref-type="supplementary-material"}\]. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### GSNO increases the number of M2/M4 AChR interacting high-affinity \[^35^S\]GTPγS binding sites in rat forebrain membranes. Membranes were preincubated for 30 min in control conditions or in the presence of 0.5 mM GSNO, and incubated thereafter for 90 min with 0.15 nM \[^35^S\]GTPγS, 10^-5^M GDP and indicated concentrations of unlabeled GTPγS in the presence and absence of CCh (10^-4^M), as detailed in the Methodssection. Statistical comparison of one- versus two-site competition curves (nonlinear regression) indicated that the one-site model best described GTPγS displacement in agonist-stimulated conditions. Note that GSNO significantly increased the number of high-affinity \[^35^S\]GTPγS binding sites in CCh-treated membranes (CCh-control 1.49 ± 0.02; CCh-GSNO 1.75 ± 0.02 pmol/mg protein, P \< 0.001, unpaired T-test). There were no statistical differences in the potency for GTPγS in displacing radioligand in any condition \[log (EC~50~): basal-control -8.10 ± 0.07; basal-GSNO -8.11 ± 0.12; CCh-control -8.06 ± 0.06; CCh-GSNO -8.10 ± 0.04). Values represent specific binding (mean ± SE) from three independent experiments performed in duplicate. ::: ![](1471-2121-6-21-7) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Effects of GSNO treatment on agonist (CCh) dose-response parameters in \[^35^S\]GTPγS binding assays of hM4 cell line membranes. Membranes were preincubated in control conditions or in the presence of 0.5 mM GSNO for 30 min. Values are mean ± SE from three to four independent experiments performed in duplicate. E~max~is expressed in percentage over basal with nonspecific binding subtracted. ::: Control GSNO -------------- -------------- ---------- -------------- -------------- hM4-WT-A1 -4.87 ± 0.14 586 ± 31 -4.94 ± 0.09 674 ± 23 hM4-WT-E5 -4.87 ± 0.05 213 ± 3 -4.98 ± 0.13 276 ± 10\\ hM4-WT-C2 -5.21 ± 0.20 134 ± 3 -5.27 ± 0.20 154 ± 8\* hM4-C133S-H2 -5.13 ± 0.10 198 ± 4 -5.12 ± 0.11 235 ± 6\\ \* Statistically different from control (P \< 0.05) \\ Statistically different from control (P \< 0.01) ::: CHO cells express endogenous G~i~-coupled LPA receptors \[[@B48]-[@B50]\], and similarly to the situation in brain membranes and cryostat sections, GSNO significantly inhibited LPA signaling in CHO cell lysates without affecting the agonist potency (Table [1](#T1){ref-type="table"}). Platelets offer another readily accessible model to study P2Y~12~-G~i~(specifically Gα~i2~) signaling in native cellular environments \[[@B42],[@B51],[@B52]\]. In human platelet membranes, GSNO inhibited basal G protein activity by 22 ± 4 % (mean ± SE, n = 3). This effect was statistically significant. Similar to the situation with brain P2Y~12~receptor signaling, GSNO inhibited P2Y~12~receptor-dependent G protein activity in human platelet membranes, but had no effect on the agonist potency (Table [1](#T1){ref-type="table"}). To examine whether the receptor protein serves as a direct target of this action, human P2Y~12~receptor was stably transfected into CHO cells. Several cell lines responding to 2MeSADP in \[^35^S\]GTPγS binding assays were established (our unpublished observations). However, neither GSNO nor CysNO affected the 2MeSADP dose-response curves of any of the hP2Y~12~-expressing cell lines (Figure [5](#F5){ref-type="fig"} and data not shown). These data rule out the P2Y~12~receptor protein as a direct target of the RSNO action. The inhibitory effect of GSNO on basal and P2Y~12~receptor-dependent G protein activity was not due to a nonspecific action on platelet membranes, since the signaling of another G~i~-linked platelet receptor, the α~2A~-adrenoceptor, was significantly amplified in GSNO-treated membranes (Table [1](#T1){ref-type="table"}). This effect was not restricted to platelets, nor was it unique to the α~2A~-subtype, as signaling of the three human α~2~-adrenoceptor subtypes (α~2A~, α~2B~, and α~2C~) was potentiated by GSNO in CHO cell lines stably expressing these receptors (our unpublished observations). RSNOs amplify muscarinic responses by increasing the rate of GDP/GTP exchange and the number of high-affinity GTP binding sites ------------------------------------------------------------------------------------------------------------------------------- The amplification of M2/M4 responses by RSNOs was clearly evident in native brain tissue. Further experiments were designed to address the mechanism of this action. Results of these studies are presented in Figures [6](#F6){ref-type="fig"} and [7](#F7){ref-type="fig"}. A time-response study on basal and CCh-stimulated \[^35^S\]GTPγS binding responses in forebrain membranes revealed that GSNO accelerated the rate of \[^35^S\]GTPγS binding in agonist-stimulated conditions (Figure [6](#F6){ref-type="fig"}). In contrast, GSNO had no effect on basal guanine nucleotide exchange. Given that GDP release is generally thought to be the rate-limiting step in receptor-driven G protein activation, these data indicate that the amplifying effect of GSNO on M2/M4 receptor-stimulated G protein activity is due to an accelerated rate of GDP/GTP exchange at the receptor-activated G protein α subunits. In line with this, GSNO significantly increased the number of high-affinity \[^35^S\]GTPγS binding sites available for M2/M4 receptor activation under agonist-stimulated conditions (Figure [7](#F7){ref-type="fig"}) (Mean ± SE: CCh-control 1.49 ± 0.02 vs. CCh-GSNO 1.75 ± 0.02 pmol/mg protein, P \< 0.001, unpaired T-test). In contrast, there were no statistical differences in the potency of GTPγS to displace \[^35^S\]GTPγS in any of the tested conditions (Figure [7](#F7){ref-type="fig"}) \[log (EC~50~) ± SE: basal-control -8.10 ± 0.07; basal-GSNO -8.11 ± 0.12; CCh-control -8.06 ± 0.06; CCh-GSNO -8.10 ± 0.04). These experiments indicate that RSNOs amplify muscarinic receptor-stimulated G protein activity in native brain tissue by accelerating the rate by which agonist-occupied receptors can activate their cognate G proteins. RSNO amplification of M4 receptor responses are preserved under heterologous expression but the amplification is lost with constitutive receptor activity --------------------------------------------------------------------------------------------------------------------------------------------------------- To investigate the signaling of the human M4 receptor (hM4) under heterologous expression system, the receptor was stably transfected into CHO cells. The effects of GSNO on agonist-stimulated G protein activity were compared in rat forebrain membranes and three cell lines expressing the wild-type (WT) hM4 receptor with increasing capability to activate G proteins. The results of these experiments are shown in Figure [8](#F8){ref-type="fig"}, and the potency and efficacy values are summarized in Table [2](#T2){ref-type="table"}. In control conditions, CCh stimulated \[^35^S\]GTPγS binding in the three WT-hM4 cell lines at a similar potency but varying efficacy (E~max~ranging from 134 to 586 %) (Figure [8](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). GSNO treatment significantly increased the E~max~in the two WT-hM4 cell lines (E5 and C2) that showed comparable maximal responses to the values obtained in rat forebrain membranes (Figure [8](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). In contrast, the potentiating effect of GSNO was lost in the WT-hM4-A1 cell line, where CCh robustly activated G proteins (Figure [8](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). In contrast to the situation in brain membranes and other hM4 cell lines of this study, WT-hM4-A1 cell line exhibited constitutive activity, i.e. significant G protein activation was evident in the absence of added agonist. In this cell line, the mAChR antagonist, atropine, significantly inhibited basal G protein activity by 13 ± 3 % (mean ± SE, n = 3, P \< 0.05) (Figure [8](#F8){ref-type="fig"}, bottom panel middle). The effect was dose-dependent \[log(IC~50~) -8.48 ± 0.23 (mean ± SE, n = 3)\], indicating increased constitutive activation of WT-hM4 receptor in this cell line. Collectively these experiments indicate that the potentiating effect of GSNO on hM4 receptor responses was preserved under heterologous expression but that the effect was diminished with constitutive receptor activity. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### GSNO-evoked potentiation of muscarinic signaling is preserved under heterologus expression but the effect diminishes with increasing constitutive activity. The human M4 (hM4) receptor was stably transfected into CHO cells and wild type (WT) or mutant (C133S) cell lines with differential G protein activation capacity were compared with that in rat forebrain membranes. CCh-stimulated G protein activity was determined in control conditions and in membranes pretreated for 30 min with 0.5 mM GSNO, as detailed in the Methodssection. Note that cell lines WT-C2, WT-E5 and C133S-H2 have maximal responses comparable to that in native brain tissue. Note also robust G protein activation in WT-A1 cell line, as well as its constitutive activity, as evidenced by the ability of the inverse agonist atropine to inhibit basal G protein activity in this cell line. Potency and efficacy values are summarized in Tables 1 (rat brain) and 2 (hM4 clones). Values are mean ± SE from at least three independent experiments performed in duplicate. ::: ![](1471-2121-6-21-8) ::: The final experiments were intended to clarify whether cysteine 133 (C133), located adjacent to the G protein-interacting DRY-motif in the intracellular end of transmembrane helix 3 of the G~i~-coupled muscarinic receptors (M2/M4), could serve as the molecular target of RSNO action. To this end, C133 was mutated into serine to reveal if the C133S mutation would abolish the effect of GSNO on the efficacy of CCh. However, GSNO treatment significantly increased E~max~also in the mutant C133S-hM4 cell line (Figure [8](#F8){ref-type="fig"}, bottom panel right, Table [2](#T2){ref-type="table"}), indicating that cysteine C133 of the hM4 receptor is not the specific target of RSNOs. The mutation did not affect the potency of CCh to activate G proteins via the hM4 receptor (Table [2](#T2){ref-type="table"}). Discussion ========== \[^35^S\]GTPγS autoradiography of brain cryostat sections revealed a highly receptor-specific modulation of GPCR signaling by RSNOs in several receptor-enriched anatomical structures. This modulation was fully reversible upon addition of excess thiols. We have provided evidence indicating that S-nitrosylation, rather than other types of NO reactions or the NO -- guanylyl cyclase -- cGMP signaling pathway, was responsible for the observed effects. The RSNOs effects were receptor-specific, as signaling of some receptors was markedly potentiated (M2/M4 AChRs and α~2~-adrenoceptors), whereas that of others was clearly inhibited (P2Y~12~, LPA and cannabinoid CB~1~receptors), while signaling of other receptors was only marginally affected (adenosine A~1~, MOR, and ORL1 receptors) by comparable treatments. We further demonstrated that RSNOs can amplify M2/M4 receptor responses by increasing the rate of GDP/GTP exchange as well as the number of high-affinity G protein α subunits capable of interacting with the agonist-activated receptors. The potentiating effect of RSNOs on hM4 receptor responses was preserved when this was studied in a heterologous expression system but was diminished in constitutively active hM4 receptors. We also demonstrated that the GPCR itself or its native signaling partners serve as potential targets of this modulation, as it was attenuated, or even lost, when receptor signaling was studied under heterologous expression. Our study suggests that GPCR signaling is subject to a highly receptor-specific modulation by NO-derived RSNOs. The GPCRs and their proximal signaling partners as likely targets of RSNO action -------------------------------------------------------------------------------- Since \[^35^S\]GTPγS binding assays monitor G protein activation, one of the earliest measurable steps in GPCR signal transduction, it is obvious that the molecular targets of RSNO action are the receptors, their cognate G proteins and/or additional signaling partners, whose thiol modification can directly regulate guanine nucleotide binding and hence G protein activation. It is interesting that RSNO treatment of brain sections consistently resulted in thiol-sensitive increases in basal G~i~protein activity throughout the gray matter regions. However, no such effect was present in brain membrane \[^35^S\]GTPγS binding assays, nor was it detected in CHO cell membranes but in platelet membranes, RSNOs inhibited basal G protein activity by \~20%. As receptor input should be minimal in basal conditions, the differential behavior of RSNOs in cryostat sections and various membrane preparations likely reflects direct action on the G~i~proteins and/or their proximal regulatory partners. It has been known for some time that G proteins can serve as direct targets of ROS, including NO \[[@B21]-[@B24],[@B28],[@B29]\]. Exogenous NO donors stimulate the monomeric G protein p21^ras^via S-nitrosylation to a single cysteine residue \[[@B22],[@B53]\]. Furthermore, Gα~i~and Gα~o~serve as direct protein targets of ROS, and they can be activated in the absence of input from the GPCRs \[[@B29],[@B33]\]. Specifically, ROS were shown to modify two cysteine residues of Gα~i/o~and this modification accelerated GDP release from Gα with a concomitant increase in the formation of the GTP-bound form of Gα without receptor activation \[[@B29],[@B33]\]. Furthermore, GSNO was reported to stimulate basal G protein activity in bovine aortic endothelial cells and human peripheral blood mononuclear cells \[[@B22],[@B26]\]. The RSNO-elicited increase in basal G protein activity in brain sections is consistent with these findings. However, it is not clear why no such stimulation was detected in bulk membrane preparations. One explanation could be the differential accessibility of target thiol(s) in cryostat sections as compared to membrane preparations. It is reasonable to assume that thiol accessibility in the former situation is close to natural. It is also possible that some crucial regulatory component of the signaling machinery is lost in bulk membrane preparations. Since several high-speed centrifugation steps were employed to obtain the relatively pure membrane preparation used in our study, this possibility should not be underestimated. This could also explain why RSNOs had no effect on basal G protein activity in CHO cell membranes, although several members of the G~i~family, including Gα~i1/2~, Gα~i3~, and Gα~o~, are endogenously present in these cells \[[@B54]\]. In platelet membranes, RSNOs inhibited both basal and P2Y~12~receptor-dependent G protein activity, but clearly potentiated α~2A~-adrenoceptor responses. Platelet P2Y~12~receptors couple to Gα~i2~\[[@B51],[@B52]\], whereas α~2A~-adrenoceptors can communicate via Gα~z~, at least in the mouse \[[@B55]\]. Collectively these data suggest that RSNOs inhibit both basal and receptor-stimulated Gα~i2~activity in the platelets. It is not yet known whether brain P2Y~12~receptors couple to this particular G protein subtype. Further, it is unclear whether LPA receptors in brain and CHO cells communicate via Gα~i2~. It is interesting that RSNOs inhibited the signaling of all three receptors in native tissues. In addition, we found that brain cannabinoid CB~1~receptor signaling was inhibited by RSNOs. A previous study reported that pulmonary vasoconstriction by serotonin was also inhibited by GSNO \[[@B34]\]. Although the basic module of GPCR signaling is traditionally considered to be the receptor, its cognate G protein, and the effector, recent studies have identified a wide range of proteins that can directly interact with the receptor and/or G proteins. These can modulate signaling efficiency, cellular localization, or the regulation of the GPCRs or G proteins \[[@B6]\]. One such recently-identified protein is the brain-enriched, Ras-related monomeric G protein Dexras1 (human counterpart is termed activator of G protein signaling 1, AGS1). Dexras1/AGS1 is physiologically activated upon NMDA receptor-stimulated NO synthesis and S-nitrosylation on cysteine C11 \[[@B56],[@B57]\]. Dexras1/AGS-1 also interacts with Gα~i~/Gα~o~, and can activate GPCR signaling systems independently of receptor input \[[@B6],[@B58]\]. Interestingly, Dexras1/AGS-1 was shown to proximally antagonize the signaling of M2 AChRs and formyl peptide receptors, possibly by altering the pool of G proteins available for receptor coupling and/or disruption of a preformed signaling complex \[[@B59],[@B60]\]. It is currently unknown whether RSNOs and/or S-nitrosylation could alter the ability of Dexras1/AGS-1 to modify G protein function and/or input from the GPCRs. In light of the present findings, this should be an attractive hypothesis for future studies. Most Gα proteins are palmitoylated at a cysteine near the amino terminus and this modification is required for G protein targeting to lipid rafts \[[@B61]\] and/or normal signaling \[[@B62]\]. Addition and removal of the palmitoyl group appear to be dynamic receptor-mediated processes that may contribute to recycling of Gα between the membrane and cytosolic compartments \[[@B63]\]. Since our experiments used only nonliving tissue, it is unlikely that the RSNO actions would achieve such extensive lipid modifications, although such alterations reportedly occur after exposure of living cells to NO \[[@B64]\]. Functional implications ----------------------- Specialized plasma membrane microdomains act as unique platforms with specific enrichment of GPCRs, their signaling partners, and the enzymatic machinery for NO biosynthesis \[[@B35],[@B37]-[@B40]\]. Such close proximity of the GPCR signaling complex and NO source is particularly relevant for many of the receptors whose signaling was shown here to be reversibly modulated by RSNOs. In the heart, endothelial NOS (eNOS) is localized n caveolin-enriched myocyte membrane fractions and it has been shown that lipid draft-disrupting agents severely compromise NO-dependent inhibition of adenylyl cyclase types 5 and 6 \[[@B35]\]. M2-mediated parasympathetic cardiac signaling also critically involves eNOS activation and NO production \[[@B65]\]. On the other hand, caveolar sequestration of M2 receptors and NO signaling was shown to be finely tuned in the myocytes \[[@B66]\], suggesting a dynamic interplay between the M2 receptor and NO. In hM4 expressing CHO cells, SNP induced agonist-independent internalization of the receptor via atropine- and thiol-sensitive mechanisms \[[@B67]\]. Furthermore, ROS including NO, potentiate cardiac M2 receptor signaling via poorly defined mechanisms \[[@B32],[@B68]\]. Consistent with these findings, our study revealed robust amplification of M2/M4 receptor signaling by RSNOs both in the brain and in CHO cells expressing the hM4 receptor. In CHO cells, however, the RSNO effect clearly diminished with increasing constitutive receptor activity, suggesting that RSNO action and constitutive receptor activity likely share common mechanisms, including accelerated GDP/GTP exchange in cognate Gα subunits. As far as we are aware, this is the first study to show robust, and highly region-specific amplification of M2/M4 receptor signaling in discrete anatomical loci of the central nervous system. The functional consequences of these findings remain to be established. The P2Y~12~receptor plays a central role in platelet activation and aggregation \[[@B69]\]. Previous studies have indicated that endothelial and platelet-derived NO, as well as exogenous RSNOs, are potent inhibitors of platelet aggregation \[[@B70]-[@B73]\]. Both cGMP-dependent and -independent mechanisms and several potential molecular targets have been implicated in these effects \[[@B72],[@B74]\]. A previous report provided evidence for cGMP-mediated signaling in the inhibition of platelet G~i~signaling \[[@B74]\]. The present study adds further dimensions to this scheme by demonstrating that RSNOs can inhibit platelet (and brain) P2Y~12~receptor function via cGMP-independent mechanisms, likely involving S-nitrosylation. However, since the effect was lost in CHO cells stably expressing the hP2Y~12~receptor, the native signaling partners, rather than the P2Y~12~receptor, serve as obvious targets of this action. One of the novel findings in this study was that RSNOs strongly inhibited G~i~-mediated LPA receptor signaling in the brain and in CHO cells. The relevance of this finding with respect to brain LPA~1~receptor signaling remains to be established. In vivo, peripheral LPA receptor signaling is closely associated with NO. In bovine aortic endothelial cells, LPA stimulates endothelial NOS via G~i~-coupled LPA receptors \[[@B75]\]. LPA is released from activated platelets and this stimulates other platelets to activate aggregation processes \[[@B76],[@B77]\]. Inhibition of LPA signaling via endogenous RSNOs could be one important mechanism by which endothelium-derived NO can suppress LPA-mediated athero- and thrombogenic signaling. Conclusion ========== In conclusion, this study revealed that G protein activation, an early step of GPCR signal transduction, is subject to a reversible and highly receptor-specific modulation by exogenous RSNOs at physiologically relevant concentrations. Since NOS synthases (and thus NO production) have been shown to reside in close proximity with the GPCR signaling machinery, especially for many of the receptors whose signaling is subject to modulation by exogenous RSNOs, these findings suggest that GPCR signaling in vivo is likely to be finely tuned by NO-derived RSNO species. Future studies should aim at pinpointing the precise molecular targets of these actions, and at understanding the specific modifications (S-nitrosylation and/or S-thiolation) involved, as well as revealing the physiological and/or pathophysiological relevance in vivo. Methods ======= Materials --------- All drugs and chemicals were from Sigma (St. Louis, MO) or Merck (Darmstadt, Germany), unless otherwise stated. Cell culture media, sera, and antibiotics were from Euroclone (Pero, Italy). Protein concentrations were determined with Bio-Rad protein assay (Bio-Rad, Hercules, CA, USA). Adenosine deaminase (ADA) was purchased from Roche (Mannheim, Germany) and guanosine-5\'-O-(3-\[^35^S\]-thio)-triphosphate (\[^35^S\]GTPγS; initial specific activity 1250 Ci/mmol) from NEN (Boston, MA). CP55940, DAMGO, nociceptin, and SNAP were purchased from Tocris Cookson Ltd. (Bristol, UK). DNA constructs -------------- Human M4 muscarinic receptor (hM4, gift from Dr. Johnny Näsman, University of Kuopio) was subcloned from pBluescript into pcDNA3 mammalian expression vector and a triple hemagglutinin (HA) epitope tag was subcloned after the initiating Met codon of the hM4 gene. This construct was used to create a C133S mutant hM4 receptor with QuickChange Site Directed Mutagenesis Kit (Stratagene, La Jolla, CA). Human P2Y~12~purinergic receptor (hP2Y~12~) was amplified from QuickClone human brain cDNA (Stratagene) using RT-PCR with gene-specific primers. The PCR product was ligated into pcDNA3 and a N-terminal hemagglutinin (HA) epitope tag was inserted in a PCR reaction with 5\' primer containing the HA tag DNA sequence. All receptor constructs were confirmed by restriction analyses and DNA sequencing prior to transfections. Cell culture and transfection ----------------------------- Recombinant plasmids were introduced into Chinese hamster ovary (CHO) cells with Lipofectamine 2000 transfection reagent (Gibco, Paisley, UK). Transfected cells were placed under G-418 selection (600 μg/ml) and several cell lines originating from single G-418 resistant cells were isolated. The G-418 resistant cell lines were cultured as monolayers with 100 μg/ml G-418 in Ham\'s F-12 nutrient mixture, containing 10% fetal calf serum, 100 U/ml penicillin and 100 μg/ml streptomycin at 37°C in a humidified atmosphere of 5% CO~2~/ 95% air. Stable cell lines were analyzed for HA tag (and thus receptor) expression using receptor ELISA with mouse anti-HA primary antibody \[[@B78]\]. The cell lines that showed high receptor expression levels in receptor ELISA were maintained for subsequent experiments. From these, several hM4 and hP2Y~12~cell lines were established that responded to carbachol (CCh) or 2-methylthio-ADP (2MeSADP), respectively. Nontransfected CHO cells did not respond to CCh (data not shown). CHO cells endogenously express G~q~-coupled P2Y receptors (P2Y~1~and P2Y~2~). However, CHO cells that had not been transfected with the hP2Y~12~receptor construct did not respond to 2MeSADP, indicating that activation of G~q~was not detected using \[^35^S\]GTPγS binding assays (data not shown). Preparation of cryostat sections, membranes, and cell lysates ------------------------------------------------------------- Naïve, four-week-old male Wistar rats were used for the preparation of brain cryostat sections essentially as described earlier \[[@B41],[@B79]\]. All animal protocols were approved by the local ethics committee. Platelet membranes were prepared from expired human platelets (Red Cross, Helsinki, Finland) as previously described \[[@B42]\], except that protease inhibitor cocktail was omitted. CHO cells were harvested by scraping in PBS containing 5 mM EDTA. Membrane fractions (P2) from rat brains and hM4 cell lines were isolated with differential centrifugation using previously published protocols \[[@B80],[@B81]\]. For membrane preparation from hP2Y~12~-expressing cell lines, the centrifugation method used for platelet membrane preparation \[[@B42]\] was used, with the modification of omitting the protease inhibitor cocktail. For the preparation of whole-cell lysates, the cells were harvested by trypsinization in normal growth medium. Cell density was counted in a hemocytometer and the cells were pelleted by centrifugation for 10 min at 250 × g at room temperature. The cell pellets were washed with PBS and centrifuged twice as above, after which the dry pellets were snap-frozen on dry ice. The frozen pellets were thawed for 1 min in a water bath at room temperature, after which snap-freezing was repeated. The final membrane and lysate preparations (1--5 mg protein/ml) were stored as single-use aliquots in -75°C. Preparation of S-nitrosothiols ------------------------------ S-nitroso-N-acetyl-D,L-penicillamine (SNAP) was purchased from Tocris Cookson Ltd. (Bristol, UK). All other RSNOs were synthesized from the respective thiols using acidified NaNO~2~. For example, S-nitrosoglutathione (GSNO) was prepared by mixing 100 μl sodium nitrite (100 mM) with 100 μl HCl (150 mM) and adding 100 μl reduced glutathione (100 mM). Reactions were allowed to proceed for 10 min at room temperature, protected from light. Reaction mixtures were neutralized with 150 μl NaOH (100 mM) and used immediately in the experiments. Millipore-quality water was used throughout and the assay buffer routinely contained 1 mM EDTA. The concentrations of RSNOs were determined by UV spectroscopy using previously published \[[@B70]\] values for the molar absorption coefficients (ε) and absorption maxima (λ~max~). \[^35^S\]GTPγS autoradiography ------------------------------ The assay was conducted under optimized conditions, where basal noise due to tonic adenosine A~1~receptor activity has been eliminated \[[@B79]\]. Experiments were conducted in light-protected chambers and in the absence of dithiotreitol (DTT), unless indicated otherwise. Briefly, the assay consisted of preincubation for 20 min at 20°C in buffer A (50 mM Tris-HCl, pH 7.4, 1 mM EDTA, 100 mM NaCl, 5 mM MgCl~2~), followed by GDP loading and RSNO treatment for 1 h at 20°C in buffer A, containing additionally 2 mM GDP and 8-cyclopentyl-1,3-dipropylxanthine (DPCPX, 10^-6^M) or adenosine deaminase (ADA, 1 U/ml) to eliminate tonic adenosine A~1~receptor activity. When peptide agonists were used, protease inhibitor cocktail (Sigma P-2714) was included in step 2 at the concentrations recommended by the manufacturer. For \[^35^S\]GTPγS binding, sections were incubated for 90 min at 20°C in buffer A, containing additionally 80 pM \[^35^S\]GTPγS, 2 mM GDP, DPCPX (10^-6^M) or ADA (1 U/ml), and the receptor agonists and/or reduced thiols (DTT, GSH or cysteine), as detailed in the results section. Nonspecific binding (Nsb) was determined in the presence of 10 μM GTPγS. The sections were washed twice at 0°C for 5 min in washing buffer (50 mM Tris-HCl, 5 mM MgCl~2~, pH 7.4), rinsed in ice-cold deionized water for 30 s, air dried and apposed to Biomax™ MR film (Kodak) for 6--11 days. Autoradiography images were digitized and processed for figures, as previously described \[[@B41]\]. \[^35^S\]GTPγS membrane binding assays -------------------------------------- The incubations were carried with slight modifications to previously published protocols \[[@B80],[@B81]\]. Membranes or lysates were preincubated for 30 min at room temperature in 50 mM Tris-HCl (pH 7.4), 1 mM EDTA, 100 mM NaCl, 5 mM MgCl~2~, 10 μM GDP and 0.5 U/ml ADA, under constant shaking and protected from light. RSNOs (usually GSNO or CysNO) were included in the preincubation at final concentrations of 0.5 mM. The assay was performed in duplicate in a final assay volume of 400 μl. The reaction was initiated by adding 40 μl of membrane or lysate preparation (5 μg protein or 50,000 cells / tube) to incubation tubes containing drug dilutions and binding cocktail. The final concentrations of the components in binding reaction were 50 mM Tris-HCl (pH 7.4), 1 mM EDTA, 100 mM NaCl, 5 mM MgCl~2~, 0.5% BSA, 10 μM GDP, 0.5 U/ml ADA and 150 pM \[^35^S\]GTPγS. Nsb was defined using 10 μM GTPγS. Reaction tubes were incubated for 90 min at 25°C under constant shaking. The reaction was quickly terminated by the addition of 4 ml ice-cold wash buffer (50 mM Tris-HCl (pH 7.4), 5 mM MgCl~2~) followed by rapid filtration through Whatman GF/B glass fiber filters (Whatman, Maidstone, UK) and two additional 4 ml washes with the buffer. In the time-response study (Figure [6](#F6){ref-type="fig"}), initial reaction volume was 2 ml and aliquots (400 μl) from duplicate samples were draw at different time points (15, 30, 60 and 90 min), and the reaction was terminated as described above. Radioactivity in filters was counted with Wallac Rackbeta liquid scintillation counter (Wallac, Turku, Finland). It should be noted that RSNO treatment caused a small, yet consistent increase in Nsb (25 ± 4 %, mean ± SEM, n = 13). However, as Nsb represented \<0.3 % of total radioactivity, this effect was considered negligible and did not contribute to the results and conclusions thereof. Data analysis ------------- \[^35^S\]GTPγS-membrane binding data were analyzed with GraphPad Prism software (GraphPad, San Diego, CA) using non-linear fitting for sigmoid dose-response curves. Statistical analyses were made with one-way analysis of variance (ANOVA) followed by Tukey\'s multiple comparison test. When comparison was made between only two groups, unpaired T-test was used. List of abbreviations ===================== 2MeSADP, 2-methylthio-ADP; \[^35^S\]GTPγS, guanosine-5\'-O-(3-\[^35^S\]-thio)-triphosphate; 5-HT, 5-hydroxytryptamine; ADA, adenosine deaminase; AGS, activator of G protein signaling; CCh, carbacholine; CHO, Chinese hamster ovary; CP-55940, (-)-3-\[2-hydroxy-4-(1,1-dimethylheptyl)-phenyl\]-4-\[3-hydroxypropyl\]cyclohexan-1-ol; CysNO, S-nitrosocysteine; CysNO-Gly, S-nitroso-cysteinyl-glycine; DAMGO, \[D-Ala^2^, N-Me-Phe^4^, Gly^5^-ol\]-enkephalin; DPCPX, 8-cyclopentyl-1,3-dipropylxanthine; DTT, dithiotreitol; Glu-CysNO, L-γ-glutamyl-S-nitrosocysteine; GSH, glutathione; GPCR(s), G protein-coupled receptor(s); GSNO, S-nitrosoglutathione; HA, hemagglutinin; hM4, human muscarinic receptor subtype 4; hP2Y~12~, human P2Y~12~purinergic receptor; LPA, lysophosphatidic acid; NA, noradrenaline; NO, nitric oxide; NOBF~4~, nitrosodium tetrafluoroborate; NOS, NO synthase; RGS, regulator of G protein signaling; RT, reverse transcriptase; SNAP, S-nitroso-N-acetyl-D,L-penicillamine; RSNO, S-nitrosothiol; SNP, sodium nitroprusside Authors\' contributions ======================= TK carried out the cell culture, molecular biology and mutagenesis studies, participated in the membrane and lysate \[^35^S\]GTPγS binding assays, participated in the design of the study and drafted the manuscript. JRS and MDR carried out most of the membrane \[^35^S\]GTPγS binding assays and performed the statistical analyses for these. KSM carried out G protein activation assays with human platelets. JTL conceived of the study, its design and coordination and conducted \[^35^S\]GTPγS autoradiography experiments. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Supplementary material (Supplementary Figures 1, 2, 3, 4, 5, 6, 7 and Supplementary Tables 1, 2) is provided as a single file. This pdf-file (size 0.75 MB) is readable using Adobe Acrobat. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ MDR is a medical student from Madrid, Spain and participated in this study as an IFMSA (International Federation of Medical Students\' Association) exchange student. We wish to thank Dr. Johny Näsman (University of Kuopio) for the hM4 receptor cDNA, and Prof. Mika Scheinin (University of Turku) for the CHO cell lines expressing α~2~-adrenoceptor subtypes. Reija Heikkinen (M.Sc.) is acknowledged for her contribution in the generation and functional testing of hM4 cell lines. Dr. Ewen MacDonald is acknowledged for the revision of the language of this paper. We are thankful to Mrs. Taina Vihavainen, Mrs. Tiina Räsänen and Mrs. Taija Vaarala for skillful technical assistance and to Mrs. Merja Saastamoinen for secretarial help.	PubMed Central	2024-06-05T03:55:56.492747	2005-4-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090567/", "journal": "BMC Cell Biol. 2005 Apr 25; 6:21", "authors": [ { "first": "Tarja", "last": "Kokkola" }, { "first": "Juha R", "last": "Savinainen" }, { "first": "Kati S", "last": "Mönkkönen" }, { "first": "Montse Durán", "last": "Retamal" }, { "first": "Jarmo T", "last": "Laitinen" } ] }
PMC1090568	Background ========== The words selectivity, specificity, and sensitivity (derived from Latin seligere, specificus, sensitivus), can be confusing terms as they are often used synonymously in the medical literature. However, they should not be used interchangeably as each represents a different phenomenon. For the sake of consistency and clarity, this paper will use the terms as defined below: Selectivity will be used to describe the ability of a drug to affect a particular population, i.e., gene, protein, signaling pathway, or cell, in preference to others. For example a selective drug would have the ability to discriminate between, and so affect only one cell population, and thereby produce an event. Specificity, a term most often confused with selectivity, will be used to describe the capacity of a drug to cause a particular action in a population. For example, a drug of absolute specificity of action might decrease or increase, a specific function of a given gene or protein or cell type, but it must do either, not both. Sensitivity will be used to describe the capacity of a population, to respond to a drug\'s ability, to stimulate that entity at a specified dose. The smaller the dose required producing an effect, the more sensitive is the responding system. (The word used to describe this activity in the drug which is the cause of the population sensitivity, is potency). It can be seen therefore, that a drug\'s activity may involve all the above attributes-it may be selective to one cell population, and also be specific to one kind of action on that cell population, and the population in turn, may be sensitive to the drug\'s influence at a lower dose than would other responding systems. As part of the current state of art in the search for new therapeutic agents, the property of selectivity is a mode of action thought to have a high degree of desirability and there is a great deal of activity in this area of research. It is the growing understanding of the cellular mechanisms that control etiology of diseases together with modern technology that enables and motivates the development of targeted therapies. This search for selective drugs has led to the development of high-throughput, virtual screening, and rational drug design techniques that are widely used to discover leads for drug candidates. Successes have lead to small molecular drugs that can target specific proteins involved in signal-transduction pathways leading to pathogenesis. Indeed, a drug having high selectivity may have a dramatic effect when there is a single agent that can be targeted against the appropriate molecular-driver involved in the pathogenesis of a disease. An example is chronic myeloid leukemia that has a specific chromosomal abnormality called the Philadelphia chromosome that results in a gene that produces an abnormal protein. However a protein being targeted may share a similar active domain with other proteins having normal physiologic functions; hence this would cause undesirable side effects. Long term experience with drugs supposedly designed for only a single target, but which unavoidably involve other functional effects, is uncovering the fact that molecular targeting is not medically flawless. The recent discovery related to Cox -2 inhibition is the most striking example \[[@B2]\]. Discussion ========== Selectivity is generally a worthy property in a drug, e.g., it is desirable to have a chemotherapy drug to affect prostate cancer cells and not affect nearby healthy prostate cells and other normal tissues. Similarly an anti-bacterial for germs or parasites should be suitably potent so that it can be used in a small dose sufficient to kill the infectious agents but not to affect host cells. Sometimes, however, selectivity is undesirable as in a case where a drug strongly targets pleiotropic cytokines or other proteins that have multifunctional effects and so can influence multiple cellular pathways simultaneously. Important examples are TNF-α and cyclic AMP, both of which have a variety of effects in a cell, pathogenetically and physiologically. TNF-α has a dual nature in cell death and cAMP acts to control a protein kinase that in turn affects activities of a variety of cellular proteins \[[@B3]-[@B5]\]. The strategy for selective drug development is based on the abundant progress made in the last decade in human genomic and proteomic projects. For example, a major related effort is structure-based drug design. Here the three-dimensional structure of a drug target interacting with small molecules is used to guide drug discovery. Structure-based design enables a researcher to \"see\" exactly how a molecule interacts with its target protein and so bind a selective agent to the target. Specifically, cancer treatment is rapidly evolving from systemic, non-specific, high-dose chemotherapy to a wide variety of targeted therapies. Advances in molecular genetics, and immunology, along with improved laboratory techniques, have led to the discovery of unique targets integral to the growth and proliferation of malignant cells. These revolutionary discoveries provide a foundation for the development of a new generation of anti-tumor agents. They include such new targeted, non-cytotoxic anticancer agents, as small-protein kinase inhibitors. Examples are the FDA approved tyrosine kinase inhibitors, Gleevec \[Gleevec (imatinib mesylate); Novartis Pharmaceuticals Corp.\], Iressa (gefitinib); Astra Zeneca Pharmaceuticals Inc.\], and others that are in various stages of clinical development. \[[@B6]\] Another example, with a different targeting mechanism, is the humanized monoclonal antibody bevacizumab (Avastin; Genentech). This agent targets and inhibits the function of the vascular endothelial growth factor (VEGF) that stimulates new blood vessel formation; and is therefore an anti-angiogenesis agent. It is currently approved as a first-line treatment for patients with metastatic colorectal cancer \[[@B7]\]. In addition to treatment of cancer there are new targeting agents known as biological response modifiers (BRMs). These target specific cytokines such as TNF-α and IL-1. These are being used for the treatment of diseases having an inflammatory component such as inflammatory bowel disease (IBD) and rheumatoid arthritis (RA). In the case of IBD they are the drugs of choice and for RA they are proving to be more effective than traditional disease-modifying antirheumatic drugs (DMARDs), especially when used in combination treatment such as with methotrexate. Four targeted drug therapies are approved for use in IBD and/or RA; three target TNF-α (1) etanercept; Enbrel, Amgen Inc. (A TNF-α decoy receptor). (2) infliximab; Remicade, Centocor Inc. and (3) adalimumab; Humira, Abbott Laboratories Inc. (both TNF-α monoclonal antibodies) and one that targets IL-1 (4) anakinra; Kineret, Amgen Inc. A reason why this problem is arising for the above three strong TNF-α inhibitors may be because TNF-α has a dual nature. It triggers the JNK-dependent pathway required for TNF-α-induced apoptosis and it also activates the protein. PI3-kinase, associated with cell survival that can block this very pathway. This dual nature thus sets up a \"delicate life-death balance\" in the cell. This finding originally brought with it increased hope for the use of TNF-α as a possible treatment against cancer. \[[@B8]-[@B10]\]. Strategies for targeting a single genes or proteins ignore a very important fact that the most, if not all of diseases involve a sophisticated network \"system\" \[[@B11]\]. For example, chemokines, a family of immune molecules related to IL-8 contains approximately 50 ligands and 20 receptors, often acting with redundancy, thus making selection of appropriate specific antagonists not only difficult, but lacking in long-term efficacy \[[@B12]\]. This argument is supported by the fact that many agents recently developed by targeting a specific molecule for the treatment of IBD are proving to be either insufficiently effective or totally ineffective. The examples of insufficient efficacy include p55-TNF binding protein, interferon α, β-a, interferon γ antibody, IL-12 antibody, P65 antisense oligo, G-CSF, GM-CSF, EGF, hGH, keratinocyte GF-2, CD4 antibody and α4β7-intergrin antibody whereas ineffectiveness includes IL-10, IL-11, ICAM-1 antisense, TNFR2 fusion protein Enbrel. Many of single-targeted drugs mentioned above have been clinically proven effective in short term. For example, the treatment with biologic DMARDs relieves symptoms, inhibits the progression of structural damage, and improves physical function in patients with moderate to severe active RA. The 3 marketed TNF-α blocking agents have similar efficacy when combined with MTX, a widely used DMARD, in the treatment of patients with RA \[[@B13]\]. While providing significant efficacy and a good overall safety profile in the short and medium term in many patients with RA, these biologic treatments, however, may create serious problems and long-term side effects, such as on the liver, and still need to be evaluated. There has been a disturbing association between the use of both of Enbrel or Remicade and the development of lymphoma \[[@B14]\]. As described above, several reports have shown that patients treated with Enbrel or Remicade worsen their congestive heart failure and develop serious infection and sepsis, and increase exacerbations of multiple sclerosis and other central nervous system problems \[[@B15],[@B16]\]. It is because many pathogenetic targets also have their multiple physiological functions and so can influence multiple cellular pathways simultaneously. Nevertheless short-term side effects of all of the above drug treatments have been thought to be generally manageable. However, as they are relatively new agents, extended follow-ups are revealing unanticipated longer term results. As one example, imatinib, the first major drug in its class of specific inhibitors of tyrosine kinase receptors, has been found to be far less effective in patients who relapsed with accelerated and blast phases of CML \[[@B1]\]. It is therefore being recommended for use either as an alternative, or as an adjunct to donor lymphocyte infusions for patients with stable phase myeloid leukemia who relapse after allogeneic stem cell transplantation. It is not surprising that after a particular protein is targeted, resistance to a drug can evolve when cancer cells create a by-pass to the targeted activity. As a result, the emergence of resistance to imatinib has been recognized as a major problem in the treatment of CML. Therefore, regimens that combine imatinib with conventional chemotherapeutic agents, or with inhibitors of other signal transduction proteins that may be preferentially activated in CML cells are being pursued \[[@B17],[@B18]\]. Two other very recent examples (2005) of drugs having a high degree of selectivity but nevertheless failed to live up to expectations or had unanticipated adverse events, are Iressa, a specific epidermal growth factor inhibitor which has a variety of side effects originally thought to be acceptable in light of its anti-cancer activity. However, Iressa failed to significantly prolong survival in comparison to placebo or in patients with adenocarcinoma \[[@B19]\]. The other, Tysabri, is a laboratory-produced monoclonal antibody which is the first of a new class of agents known as selective adhesion molecule (SAM) inhibitors. Tysabri was pulled from the market, and clinical trials of Crohn\'s diseases and rheumatoid arthritis were suspended because of two cases of progressive multifocal leukoencephalopathy (PML) \[[@B20]\]. An interesting and to the point of this paper, is the publication by Roth et al., in \"Nature Reviews\". The authors discuss the concept of using selective versus non-selective drugs for central nervous system (CNS) disorders. Since in most cases, multiple molecular lesions or signaling pathways are involved in pathogenesis of CNS disorders, the authors conclude that attempts to develop more effective treatments for diseases such as schizophrenia and depression by discovering drugs selective for single molecular target that is, \"magic bullets\" have, not surprisingly, been largely unsuccessful. They propose that \"designing selectively non-selective drugs (that is, \'magic shotguns\') that interact with several molecular targets will lead to new and more effective medications for a variety of central nervous system disorders\" \[[@B21]\]. At the 10^th^anniversary conference of The Society for Biomolecular Screening a presentation on single vs. combination drugs concluded that \"During the last decade, the industry has followed an assumption that a single drug hitting a single target was the \'rational\' way to design drugs. Now post-genomics biology is teaching us the fundamental limitations of the single target philosophy. Ironically many drugs on the market, discovered in \'black box\' phenotype screens, are observed to bind potently to multiple targets and more so, this poly-pharmacology is key to their effect (personal communication; Andrew Hopkins, Pfizer Global Research & Development -United Kingdom; The Society for Biomolecular Screening, 10^th^anniversary conference, September 14, 2004, Point/ Counterpoint -- Polypharmacology: Single vs. Combination Drug) A case in point is found in another recent study that combined an anti-EGFR monoclonal antibody and tyrosine kinase inhibitor, which target extra-cellular and intracellular domains of the receptor, respectively. Specifically, the combination of cetuximab (Erbitux, ImClone Systems, with either gefitinib (Iressa, AstraZeneca, or erlotinib (Tarceva, Genentech) across a variety of human cancer cells. The combination of cetuximab plus gefitinib or erlotinib enhanced growth inhibition over that observed with either agent alone. The study concludes that \"together, these data suggest that combined treatment with distinct EGFR inhibitory agents can augment the potency of EGFR signaling inhibition. This approach suggests potential new strategies to maximize effective target inhibition, which may improve the therapeutic ratio for anti-EGFR-targeted therapies in developing clinical trials\" \[[@B22]\]. At the American Society of Clinical Oncology annual meeting held in June, 2004, scientists discussed the concept of drug promiscuity. The focus here was in some cases, a drug that attacks multiple but limited targets may keep cancer cells from developing resistance \[[@B23]\]. The examination of the role of deregulated cell cycle progression and uncontrolled cellular proliferation can also elucidate the advisability of hitting multiple targets simultaneously in order to show superior efficacy. Uncontrolled proliferation is a condition found in a variety of diseased cells. It is a feature of cellular transformation, accompanied by the deregulation of Cyclin dependent Kinases (Cdks) involved in the control of the cell cycle, check points, and apoptosis which has a crucial role in the growth of both normal and malignant cells \[[@B24],[@B25]\]. Apoptosis is an energy-dependent, normal process of cell death, based on morphological and biochemical changes in the cell that occurs in many biological conditions, but without concurrent pathological necrosis and inflammation \[[@B26]\]. Since immune cells in bone marrow (B-cells) or in the thymus (T-cells) undergo repeated cycling as part of their development, cell proliferation is a fundamental processwithin the immune system function \[[@B27]-[@B29]\]. Immune based diseases with a component of inflammation are risk factors for the development of several diseases including cancer, and it is known that anti-inflammatory agents markedly inhibit the development of cancer in humans; e.g., colon cancer \[[@B30]\]. Although the specific intracellular pathways and networks involved in these processes are not completely understood, at the molecular level, this is thought to be based on the combination of cellular transformation and cellular proliferation. The normal cell cycle and its intricate mechanism when deregulated (by germ or somatic factors) can lead to uncontrolled cellular proliferation, defective apoptosis, autoimmunity and inflammation and uncontrolled proliferation and defective apoptosis, can be viewed as both cause and consequence ofinflammation, cancer, and autoimmunity \[[@B31]-[@B34]\]. Multiple target drug screening approaches are being developed. As indicated, the pathogenesis of a disease is usually multi-factorial involving numerous risk factors and defective genes or proteins or signaling pathways out of balance with each other. There may be one major or most easily definable defective target (or pathway)externalized or typifiedfor a given disease, but collateral proteins which can act in a network rather than single pathway is likely to be involved and may lead to the emergence of backup (or redundant) systems to sustain the disease, or cause undesirable side effects \[[@B35]-[@B37]\]. This is a fundamental defect in the single target reasoning for therapeutic development. To overcome this problem, a network approach has been described \[[@B20],[@B38],[@B39]\]. In this network model of pharmacological actions elements of the network represents various targets (proteins, RNA-s or DNA sequences), and each link corresponds to an interaction between proteins of the cell. Interestingly, application of this model revealed that multi-target drugs affect their targets only partially, which corresponds well with the presumed low-affinity interactions of these drugs with several of their targets. Low-affinity, multi-target drugs might have another advantage -- weak links stabilize the systems buffering the changes after system-perturbations \[[@B39],[@B40]\], thus lower side effects. Therefore, multi-target drugs and the network approach might become a useful mean of novel drug discovery. Many successful drugs are promiscuous. The best known is aspirin or acetylsalicylic acid known to target any area where inflammation is present. In recent years, aspirin has surpassed the area of pain relief to also include activity as blood thinner, to reduce platelet aggregation in the prevention of cardiovascular disease, prevention of preeclampsia (an hypertensive disorders in pregnancy), and prevention of cancer. Aspirin\'s antiarthritic effect requires chronic or long-term therapy for pain and/or inflammation, e.g., rheumatoid arthritis, juvenile rheumatoid arthritis, systemic lupus erythematosus, osteoarthritis, ankylosing spondylitis, psoriatic arthritis, Reiter\'s syndrome, and fibrositis. Aspirin\'s major mode of activity is associated with its cyclooxygenase (COX) inhibitory activity. Cyclooxygenase enzymes are required for the conversion of arachidonic acid to prostaglandins. COX-2 mediates the inflammatory effects, and is induced by a wide spectrum of growth factors and proinflammatory cytokines. It is over-expressed in numerous pre-malignant and malignant lesions, including colorectal and prostate cancer. Recent papers suggest aspirin and salicylate at therapeutic concentrations inhibit COX-2 protein expression through interference with binding of CCAAT/enhancer binding protein beta (C/EBPbeta) to its cognate site on COX-2 promoter/enhancer. COX-2, is not normally produced in most tissues but is induced by a wide spectrum of growth factors and pro-inflammatory cytokines such as IL-1 and TNF-α, observed in such cell types as synoviocytes, endothelial cells, chondrocytes, osteoblasts, and monocytes/macrophages. Expression of genes, such as inducible nitric oxide synthase and interleukin-4, may be inhibited by aspirin and salicylate by a C/EBP-dependent mechanism Aspirin at supra-pharmacological concentrations inhibits NF-kappaB-mediated gene transcription and protects tissue from injury. Other pathways yielding other effects are likely but as yet, uncovered \[[@B41]-[@B46]\]. Summary ======= As discussed, although the basic mechanistic activity of a drug in a given cell may be relatively simple, the resultantactivity it produces in the human body can be highly complex. An activity can involve the balance and interplay of multiple signaling networks and result in unintended consequences. The pathogenesis of a disease is usually multi-factorial involving numerous risk factors and defective proteins or proteins out of balance with each other. There may be one major or most easily definable defective target (or pathway)externalized or typifiedfor a given disease, but collateral proteins which can act in a network rather than single pathway are likely to be involved and may lead to the emergence of backup (or redundant) systems to sustain the disease, or cause undesirable side effects \[[@B35]-[@B37]\]. This is a fundamental defect in the single target reasoning for therapeutic development. Consequently, it is conceivable that a perfect drug is not one that has selectivity for one protein, nor one molecular mechanism. Such super-selectivity gives a drug maximal efficacy and minimal adverse effects or toxicity only if the target of drug is the only one involved in the pathogenesis of a disease or the target is presented only in the targeted tissues. This, situation apparently occurs only rarely. Perhaps, therefore, an ideal drug may be one whose efficacy is based not on the inhibition of a single target, but rather on the rebalancing of the several proteins, or events, that contribute to the etiology, pathogeneses, and progression of diseases, i.e., in effect a promiscuous drug Based on the above discussion, one may infer that for control of some diseases, an \"ideal\" drug could very well be one that can hit more than one target, and stimulate/inhibit more than one molecular activity but do so at a concentration sufficiently low not to induce undesirable side effects. When a disease is based on an imbalance of several proteins or is genetically, physiologically, and ultimately pharmacologically heterogeneous, the logic for promiscuity in drugs is intuitive. However, it remains to be seen if drug promiscuity will be superior to targeted drugs when a disease may be homogeneous. The ability to detect true homogeneity is the crux of this dilemma. Nevertheless, when considering all the ramifications of a rare, ideal case of a single agent affecting a single molecular-driver, andeven then, having concomitant serious side effects, one wonders if a broad based and very expensive screening process used to search for truly selective drugs is a beguiling, but distracting and perhaps a deluding, application of large amounts of research money Briefly, a promiscuous drug may be advantageous because: • Different diseases may have related etiology or similar pathological alteration. • Multiple factors contribute to the pathogenesis of a particular disease. • Redundancy widely exists in biologically critical pathways • Being promiscuous is not necessarily more toxic. • Promiscuous drugs do not necessarily completely shut down or excessively activate a pathway or network Such drugs are the result of the emerging use of Network Biology. Network Biology is based on the understanding of how cellular molecules and their interactions determine the function of complex cellular machinery, both by themselves in isolation, as well as with other nearby cells. Various types of molecular interaction webs (including protein-protein interaction, metabolic, signaling and transcription-regulatory networks) emerge from the sum of these interactions that together are principal determinants of the system-scale behavior of the cell \[[@B47]\]. Further, new technology is being developed related to screening for the discovery of promiscuous drugs. An example is the new cell-based high-throughput technology for screening chemical libraries against several potential cancer target genes in parallel. This multiplex gene expression (MGE) analysis provides direct and quantitative measurement of multiple endogenous mRNAs using a multiplexed detection system coupled to reverse transcription-PCR \[[@B48]\]. Several brief examples of promiscuous drugs currently under development include: 1). SU11248 (Pfizer) -------------------- SU11248 is a small protein receptor tyrosine kinase inhibitor that interferes with several cellular signaling pathways. It has with direct anti-tumor as well as antiangiogenic activity via its multi-targeting process; the vascular endothelial growth factor (VEGF), platelet-derived growth factor (PDGF), KIT, and FLT3 receptor tyrosine kinases. 2). Bay 43-9006 (Bayer) ----------------------- BAY 43-9006 inhibits a variety of kinase receptors, including VEGF and PDGF. It is the first agent to target both the RAF/MEK/ERK signaling pathway to inhibit cell proliferation and the VEGFR-2/PDGFR-β signaling cascade to inhibit tumor angiogenesis. 3). NTI-2001 (Natrogen Therapeutics) ------------------------------------ NTI-2001 regulates two very different avenues associated with disease pathogeneses. It modulates several cytokines, e.g., TNF-α only moderately and so should not do so to the degree of being a cancer causative, but its modulation is effective for restoring the cytokine balance, key for the treatment of inflammatory based diseases. In addition to regulating such cytokines as Il-1β, Il-6, and Il-10 it can inhibit Cdks involved in cellular transformation and cellular proliferation. The restoration of cytokine balance and Cdks inhibition, are both important factors not only in cancer but in the pathogenesis of other inflammatory diseases. Competing interests =================== The authors have a financial interest in Natrogen Therapeutics Corporation that is developing the drug described in the last paragraph of this paper. The company is a small, private organization however it is conceivable that in the future, they can gain or lose financially from the publication of this manuscript. Authors\' contributions ======================= Both authors of the paper, SK Mencher and LG Wang have contributed equally to the paper.	PubMed Central	2024-06-05T03:55:56.496879	2005-4-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090568/", "journal": "BMC Clin Pharmacol. 2005 Apr 26; 5:3", "authors": [ { "first": "Simon K", "last": "Mencher" }, { "first": "Long G", "last": "Wang" } ] }
PMC1090569	Background ========== Concerns about both the nature and the adequacy of the provision of mental health services in primary care settings have prompted exploration of the respective roles of psychiatrists and family practitioners in such activities. The interface between physician groups has been examined from the perspective of family doctors gatekeeping access to specialized services in Canada \[[@B1]\], the United States \[[@B2]\], Britain \[[@B3]\], and Australia \[[@B4]\]. Attention has focussed on the specific purpose, role and outcome of the consultation process. While some motivation for these reviews was economic, the process and quality of communication between referring physician and consultant also began to be explored.\[[@B5]\] Goldberg and Huxley \[[@B6],[@B7]\] have described potential obstacles along a pathway to mental health consultation, indicating the importance of integration of services from family doctor to specialty care. \[[@B8]\] Within the domain of mental health services a shared care model between psychiatrist and family physician has been described \[[@B9],[@B10]\] and advocated for the care of elderly.\[[@B11]\] For example, while the prevalence of major depression in an older primary care population is 6%, \[[@B12]\] less than 20% of these depressed elderly are diagnosed and adequately treated.\[[@B13]\] Interventions may be designed to improve the process of out-patient psychogeriatric care.\[[@B14]\] The PROSPECT \[[@B15],[@B16]\] and IMPACT \[[@B17]-[@B20]\] studies have recently demonstrated, for example, the particular benefit of psychiatric care managers in managed health care settings. Elderly with mental health problems usually present to primary care doctors \[[@B21]\] and the actual pathway to consultative specialized psychiatric services is complex.\[[@B13]\] Nutting et al distinguish between consultation, referral, and transfer of care.\[[@B22]\] The consultation involves one physician asking another to perform a specific diagnostic or therapeutic task, and to provide specific impressions and recommendations for care. In a referral the patient will see another physician for diagnosis, investigation and possible short term management of a specific problem, but continue to see the referring doctor for other issues. If a problem is complex there may be complete transfer for on-going care of the specific problem under review. Problems in communication between family practitioners and psychiatrists may exist within the aforementioned options for care.\[[@B23]\] A referring note may not succinctly indicate why the input of a specialist is desired and what is specifically expected from the consultation. The consultation report may not respond with sufficient content to provide the family doctor with the necessary information to permit comfort to assume the care of the patient. In this regard Craven and Bland\'s overview of shared mental health services suggests that more information is required about why family doctors refer to or consult psychiatrists and whether the process is effective and relevant for these primary care physicians.\[[@B24]\] The goal of our project was therefore to explore aspects of the consultation process/referral process between family physicians and geriatric psychiatrists. Specific objectives were to: \(1) Characterize the reasons seniors undergo psychogeriatric consultations. \(2) Describe expectations of the consultation visit for the patient/family, the referring family doctor, and the consultant. \(3) Assess patients\' and family practitioners\' level of satisfaction with the consultation process. Methods ======= St. Mary\'s Hospital Centre is a 300 bed, McGill University affiliated, Montreal community hospital well-suited to study seniors because it serves a catchment area 23% of whose population is 65 or older. Over an 18 month period all individuals 65 and older appearing at its Psychogeriatric Outpatient Clinic with family doctor-initiated consultations or referrals were approached by a research assistant to participate in this research ethics approved study. Patients had to speak either English or French, or if they were unable, to be accompanied by a family member or informant who could. Verbal consent to screen for cognitive functioning was obtained since evidence of mental competency was required for written consent to participate in the study. On the SPMSQ scale \[[@B25]\] ranging from 0 (no impairment) to 10 (severe impairment) scores of 4 or less were satisfactory for informed consent for participation since they would be indicative, at most, of only mild cognitive impairment. Non-patient related exclusion criteria included referral from a specialist or a resident physician. The 3 members (with 3,12, and 25 years of practice respectively) of the Division of Psychogeriatrics of the Department of Psychiatry all participated. Prior to the actual consultation the patient or family member/informant completed a pre-tested questionnaire on patient\'s age, gender, education, present or past occupation, marital status and place of residence. Immediately following the consultation the participants completed the Consultation Satisfaction Scale \[[@B26]\], an 18 item questionnaire that evaluates overall satisfaction with a medical visit, satisfaction with professional care, amount of time spent, and depth of the doctor-patient relationship. Following each patient encounter the psychiatrist completed a short questionnaire on clinical diagnosis, recommendations made verbally to the patient, the written response to the referring family doctor, and perceptions on the reason for the consultation and expected process of care. These impressions were based on a composite of what was written on the consultation/ referral request and what was learned from the patient. Following the consultation the research assistant notified the family doctor it had taken place and that the patient had given permission for the research team to contact him/her for additional information. To allow sufficient time for the referring physician to receive and evaluate the consultation report a subsequent 10 minute telephone interview with the family practitioner was scheduled to take place 4 weeks after the consultation. The family practitioner was asked about the presence or absence of academic affiliation, type of practice (solo, group, hospital), frequency of referrals for psychogeriatric assessment; patient\'s diagnosis and treatment prior to the consultation; criteria for referral (e.g. diagnosis, treatment, investigation, reassurance, legal concern), and degree of pressure exerted by patient/family (on a 3 point scale) for the consultation \[[@B27]\]); whether the consult report had been received; patient\'s treatment at time of interview; whether the consult report met the expectations for which it had been made (ie was it for consultation, referral, or transfer of care); overall satisfaction with the consultation process; whether the physician would consult again; and suggestions for improvement in the overall consultation process. Additional physician demographic data was obtained from the directory of the College des Médecins du Québec. After data collection from patients, psychiatrists, and family doctors was complete, the original consultation /referral notes from the latter were reviewed independently by a panel of one family physician (MJY) and two geriatric psychiatrists (FP,MC) to assess the reasons for the referrals. Since such notes tended to range from a few non-specific words to a detailed clinical history or specific queries (and sometimes legibility was an issue), the panel agreed in advance to categorize referrals according to clearly stated diagnoses (DSM IV) or constellations of symptoms. Results ======= Patient participant characteristics ----------------------------------- Over 18 months 207 people were referred by family practitioners for psychogeriatric consultation. As shown in Figure [1](#F1){ref-type="fig"}, 173 actually presented for the assessment and a research assistant was able to approach 163 of them about the study. With the application of exclusion and consent criteria and 5 eligible individuals refusing to participate, this number was reduced to 101. 82.2% (83/101) of them had SPMSQ scores suggestive of sufficient mental competency to complete the questionnaire while the remaining seniors required it be completed by another informant. The mean age of participants was 78.0, with 73.3% (74/101) female, just under half had a college or university education, and 79.3% having a past or current marriage. 71.3% of the sample were living in their own homes, 49.5% were living alone, 2/3 required assistance with housework and just under half needed help for food shopping or meal preparation. When patients were stratified according to the psychiatrist that they saw, no socio-demographic differences were found between the 3 groups. (data not shown) No demographic information was available for those patients who did not keep the consultation appointment. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Identification of Patient/Family and Family Physician Samples ::: ![](1471-2296-6-15-1) ::: Reasons for consultation ------------------------ The physician panel\'s blinded and independent review of reasons for the consultations showed full consensus for 76% (77/101) 44.6% (45/101) of the total were for depression, 31.7% (32 /101) for non-depressive problems such as anxiety or cognitive deficits, and in the remaining 24 % there was no consensus on 19 and insufficient information on which to judge 6. By comparison the frequency of psychiatrists\' diagnoses were 47.5% (48/101) depression, 26.7% (27/101) dementia, anxiety 4% (4/101), bipolar disease 1% (1/101,)and 20.8 % (21/101) received any of 15 other diagnoses. For these 101 cases conditions the psychiatrists recommended medication in 85.1%, psychotherapy 19.8%, social work referral 10.9%, medical assessment/laboratory tests 14.9%, and other options in 9.9%. Participating family physician characteristics ---------------------------------------------- The 101 consultation requests came from 63 family doctors, for whom follow-up was possible for 41 (65.1%) (Fig [1](#F1){ref-type="fig"}). The latter group was 55% male, had 18.3 mean years in practice, just over 40% were affiliated with our hospital, and were predominantly McGill University -- trained. No statistically significant differences were found between these doctors and those who were not successfully contacted when compared for gender, years in practice, date and school of graduation, or hospital affiliation. As well, the patients coming from either group of physicians showed no clinically significant differences. When contacted doctors were asked whether they believed that they had less, the same, or greater interest in psychiatry as compared to the average family doctor they were equally divided between the latter two options. To support this perception 60% had attended a CME event on a psychiatry topic in the 3 years preceding the study. In the 12 months preceding the study the frequency these doctors initiated consultations with psychiatrists was 65.0% (26/41) for 1--5 patients, 27.5% (11/41) for 6 -- 10, and 7.5% (3/41) for more than 10. These doctors indicated however that prior to initiating their referrals they attempted in 75.8% (50/67) of cases to diagnose and manage the patients on their own and in 46% this took place for greater than 6 months. Consultation for management strategy ------------------------------------ For the sample of 67 patients (Fig [1](#F1){ref-type="fig"}) Table [1](#T1){ref-type="table"} suggests that the primary motivations behind the consultations for both referring and consulting doctors were diagnosis and management concerns. At face value the differences in the proportions for the latter appear to be wide between the two groups of doctors (62.7% vs. 48.5% respectively). However, grouping potential options with possible overlapping meanings -- e.g. management strategy, failed treatment, dealing with treatment side effects, and reassurance -- psychiatrists and family doctors actually had similar perceptions on management concerns (65.7% vs. 60.6% respectively). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Perceived reason for consultation ::: (N = 67) ----------------------------------- ---------- ------ ------ ------ Diagnosis 15 22.4 20 30.3 Management strategy 42 62.7 32 48.5 Patient/family request 2 3 3 4.6 Lack of skill/facilities to treat 2 3 2 3 Failed treatment 2 3 5 7.6 Reassurance 0 0 2 3 Medico-legal 1 1.5 0 0 Treatment side-effect 0 0 1 1.5 Follow-up 2 3 0 0 Other 0 0 1 1.5 Don\'t know 1 1.5 0 0 Total 67 100 66\* 100 \excludes missing values ::: Perceptions about process of care --------------------------------- For the sample of 67 patients (Fig [1](#F1){ref-type="fig"}) the family practitioners reported that for 72.3% of referrals they indicated to the patient or family member what to expect as a result of the consultation visit. Nonetheless, there was wide variation between the family doctors, psychiatrists, and patients as to perception of the process of assessment and follow-up (Table [2](#T2){ref-type="table"}). Psychiatrists saw the referral as a request for only an assessment for 80.6% of the patients, while family doctors expected short to long-term care by the psychiatrists in 55.4% of cases. Patients/families showed a large range of expectation, and about half actually had no expectation or opinion. When such analyses were done for the pairing of 101 patients/families with psychiatrists comparable findings were found. Finally, looking at manpower utilization appropriate to implement consultation recommendations, little difference between the 2 doctor groups was found, and use of nurses or community resources seemed to not be considered often. (Table [3](#T3){ref-type="table"}) ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Expectations of Process of Care ::: (N = 67) -------------------------------------------- ---------- ------ -------- ------ -------- ------ Assess and return patient to FP 54 80.6 28 43.1 16 24.2 Assess, treat short term, and return to FP 7 10.4 34 52.3 4 6.1 Assess, transfer care to psychiatrist 3 4.5 2 3.1 12 18.2 Other 1 1.5 1 1.5 2 3 No expectation 0 0 0 0 21 31.8 Don\'t know 2 3 0 0 11 16.7 Total 67 100 65\\* 100 66\\ 100 \FP = Family physician \\excludes missing values ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Expectations of who would implement recommendations ::: (N = 67) ---------------------------- ---------- ------ ------ ------ Referring physician 35 53.9 32 54.2 Consulting psychiatrist 24 36.9 18 30.5 Both 3 4.6 1 1.7 Geriatric psychiatry nurse 1 1.5 0 0 Community Resource 1 1.5 0 0 Unclear 1 1.5 8 13.6 Total 65\ 100 59\* 100 \Excludes missing values ::: Table [4](#T4){ref-type="table"} examines concordance between pairings of patients, family doctors and psychiatrists for 3 aspects of care. On \"expectations of process of care\" family doctors and psychiatrists agreed as often as they disagreed; patients and psychiatrists disagreed 3/4 of the time, and patients and family doctors disagreed almost 90% of the time. On \"reason for the consultation\" family doctors and psychiatrists once again agreed as often as they disagreed, and only for \"identification of the professional responsible for follow-up\" was the tendency for agreement found to be strongest. Interestingly, despite such differences Table [5](#T5){ref-type="table"} suggests that patients/ informants were nonetheless generally satisfied with the care, depth, and length of the consultation. Patients\' / informants\' satisfaction was found not to be associated with whether they and their family doctors were concordant on the process of care. Similarly, family physicians\' satisfaction was not associated with whether they and the psychiatrists had concordance on the process of care, reason for the consultation, or responsibility for follow-up. (Table [6](#T6){ref-type="table"}) ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Concordance: Family physician, psychiatrist, patient ::: Total cases = 67 ------------------------------------------------------------------------- ------------------ ------ ---- ------ [Reason for consultation]{.underline} Family physician and psychiatrist 33 49.3 34 50.7 [Expectations of process of care]{.underline} Patient and Family Physician 9 13.4 58 86.6 Patient and Psychiatrist 15 22.4 52 77.6 Family physician and psychiatrist 32 47.8 35 52.2 [Professional responsible for implementing recommendations]{.underline} Family physician and psychiatrist 41 61.2 26 38.8 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Patient/Informant Satisfaction\ with Consultation ::: N = 101\\ ------------------------ ------------- -------- General Satisfaction\* 4.11 (0.77) Professional Care 4.06 (0.53) Depth of relationship 3.64 (0.69) Length of Consultation 3.67 (0.97) \* Satisfaction ranged from 1 (very dissatisfied) to 5 (very satisfied) \\ 7 missing ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Impact of Expectations and Concordance on Satisfaction\* ::: ---------------------------------------------------- -------------------------------------- ------------------------------------------ ------ ----------- ----------------- [Patients\'/Informants\' Expectations]{.underline} [FP/Patient Concordance]{.underline} [FP/Patient Non-concordance]{.underline} Mean (std dev) Mean (std dev) t-test(p-value) On Process of care 4 (0.5) 3.8 (0.6) 0.300 [Physicians\' Expectations]{.underline} [FP/Patient Concordance]{.underline} [FP/Patient Non-concordance]{.underline} Mean (std dev) Mean (std dev) t-test(p-value) On Process of care 4.1 (1.0) 4.2 (0.8) 0.616 On reason for consultation 4.1 (1.0) 4.2 (0.8) 0.787 On follow-up after consultation 4.3 (0.8) 4.0 (1.1) 0.178 ---------------------------------------------------- -------------------------------------- ------------------------------------------ ------ ----------- ----------------- \Satisfaction ranged from 1 (very unsatisfied) to 5 (very satisfied) \\* FP = family physiciana ::: Family doctors\' assessment of consultation ------------------------------------------- The family doctors acknowledged receiving psychiatrists\' consultation reports from the psychiatrists in 91 % (61/67) of cases. They rated the overall consultation process as satisfactory to very satisfactory 83.3% of the time. 67.2% (41/61) of consult reports were felt to be very useful, compared with 27.9% (17/61) perceived as somewhat useful. The reports\' information was rated good to very good in 74.3% of cases. Consultants\' recommendations were felt to be clear in just over 83%. Two thirds of FPs had no comments or recommendations on how to improve the consultation/referral process. Among those making suggestions the most frequent was a request for typed consultation reports. Discussion ========== Depression as a common problem ------------------------------ This study examined aspects of the consultation/referral process for seniors from family practitioners to geriatric psychiatrists. The predominant diagnosis for which help was sought was depression, a diagnosis not unexpected given the relatively high prevalence and multi-faceted etiology of depression in the elderly (medication, co-morbidity, loss/grieving, disability). Uncertainty about management ---------------------------- When one looks at referrals to sub-specialized mental health services (e.g. a cultural consultation service), the vast majority are reported to be made for diagnostic purposes.\[[@B28]\] Pressure from patients and/or families has been hypothesized as a cause for initiating consultation in internal medicine care.\[[@B29]\] Our data, however, would suggest that within the specific sample studied this does not seem to be an important factor. A recent study looking at referral patterns from generalists to specialists (that did not include psychiatrists) found that therapy management was the most common reason for referral.\[[@B29]\] In our study, from the perspectives of both family doctors and psychiatrists the predominant reason for referral was for assistance with management strategies. Since 3/4 of the cases had management attempts by the family practitioner prior to referral (almost half for greater than 6 months) one wonders if this reflects specific challenges in the care of mental health problems in seniors. Alternatively, since the consulting psychiatrists felt 85.1% of referrals needed pharmacotherapy, family physicians may be having trouble in this domain -- either in knowledge base, in how to choose the right drug and evaluate its efficacy, or in how to changes drugs. Post-consultation care ---------------------- About 1/2 as many patients as family doctors, and similarly 1/2 as many of family doctors as psychiatrists had an expectation for the consultant to assess and return the patient to the referring physician for care. The psychiatrists\' viewpoint may derive from a desire to limit their roles specifically to that of consultants (not to primary therapists), consistent with evolving consensus on roles for psychiatrists \[[@B9],[@B10]\]. Alternatively, practicing in a catchment area with a high proportion of seniors, they may have felt a practical necessity to rapidly return patients to the family physicians. On the other hand, since 75.8 % of family physicians indicated that they had tried diagnosis and management prior to referral, they may be indicating insufficient comfort or experience with their roles as the major care providers for seniors with mental health problems. This may generate a preference for the security of having the psychiatrists initiate treatment and follow for at least a short time. This observation may be important since, at least on the basis of years in practice, the responding cohort of physicians was an experienced one. Nonetheless, the proportionately equal agreement between the two professions as to who should implement recommendations from consultations suggests a willingness by family doctors to assume subsequent care of agreed upon patients. Communication problems ---------------------- The study does suggest that family doctors likely have to do a better job of explaining their consultation expectations to both patients/families and consultants, and that psychiatrists likely need to examine their communication on this, as well. Since the majority of referring doctors had sent fewer than 5 patients in the preceding 12 months for psychiatric consultation one wonders if the low concordance between the psychiatrists and the family doctors reflects the need for a threshold number of consultations to promote different or more specific communication between referring physician and consultant. In just under 3/4 of the referrals the family doctors indicated that they told their patients what the process of care likely would be. Nonetheless 48.5% of patients/families indicated they had no expectation or did not know what to expect from the consultation visit. Interestingly, a larger proportion of patients, when compared to either group of doctors, expected transfer of care to the psychiatrists. Again this raises questions of how well the referring physician prepared the patient for the consultation, or were some patients so distressed by symptoms or the idea of having to see a psychiatrist that they did not hear or retain what they were told? Alternatively, is there something in the community mindset of mental health care that suggests seeing a psychiatrist naturally implies engaging in some form of on-going therapy? Study limitations ----------------- Conclusions are restricted to those seniors with mental health problems who kept their appointments for a psychogeriatric consultation. The study may also be limited by its predominantly descriptive nature and by the relatively low family doctor participation rate in the follow-up survey. Although general characteristics of doctors were similar, it is possible that those who were less satisfied and perhaps less concordant, did not participate. Further, since the interview with family practitioners was retrospective, the information on, for example, motive for consulting, might have been distorted by recall or wishful thinking. Conclusion ========== This study appears to have identified some concerns about the process and possible outcome of consultation/referral for mental health care for seniors. Our findings support Canadian recommendations towards the development of new models of shared care for mental health \[[@B9],[@B10]\] and parallel observations made in internal medicine about the need for different communication patterns for consultations.\[[@B29]\] One solution may lie in a pre-consultation orientation to patient /family (in verbal, written, or videotape form) about what the visit may entail and aim to achieve. Another sees an expanded role for nurses or other appropriately trained individuals to function as comprehensive case managers. \[[@B15]-[@B20]\] Further, the consultation and referral process might be improved through the use of consultation request and response forms that specifically structure written content according to mutually pre-determined headings. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= FP, JM, MJY, MC, and JL made contributions to the study conception and design; MJY, FP, MC, ME, and JL were involved in acquisition of data; ND, EB, MJY, JM and FP were involved in data analysis; MJY, FP, JM, and MC were involved in drafting the article; and all authors have had critical input and have read and approved the article. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2296/6/15/prepub> Acknowledgements ================ Research was funded by Health Canada Health Transition Fund grant \# QC428 and by Fonds de Recherche en Sante du Quebec (FRSQ) grant \# 25004-2560. Appreciation to Dominique Proulx for assistance in the study implementation.	PubMed Central	2024-06-05T03:55:56.499658	2005-4-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090569/", "journal": "BMC Fam Pract. 2005 Apr 19; 6:15", "authors": [ { "first": "Mark J", "last": "Yaffe" }, { "first": "Francois", "last": "Primeau" }, { "first": "Jane", "last": "McCusker" }, { "first": "Martin G", "last": "Cole" }, { "first": "Eric", "last": "Belzile" }, { "first": "Nandini", "last": "Dendukuri" }, { "first": "Michel", "last": "Elie" }, { "first": "Johanne", "last": "Laplante" } ] }
PMC1090570	Background ========== In the last decades, childhood asthma has risen dramatically and now forms a major and still increasing health problem, notably but not exclusively in the affluent parts over the world \[[@B1]-[@B4]\]. Large research efforts have therefore been directed to it, and especially the multinational International Study of Asthma and Allergies in Childhood (ISAAC) \[[@B4]\] has only in the last few years provided a wealth of additional data, e.g., on prevalence \[[@B5]-[@B12]\] and symptoms \[[@B8],[@B13],[@B14]\], which further document the worsening of the situation \[[@B15]\] in spite of increasing community awareness and number of children receiving inhaled steroids\[[@B11]\] and other treatment. In an international study of 12-year old children in 1994, questionnaire-reported asthma-prevalence was found to range from 16.8 % in New Zealand, 12 % in Wales and 11.5 % in South Africa to only 4% in Sweden \[[@B16]\]. Support of these data and that questionnaires may not grossly over-estimate prevalence is found by another questionnaire study from Sweden in 1989 where a frequency of 5.1 % in 9000 rural children of age between 4--14 years was reported \[[@B17]\], whereas in 1988 the percentage of 7--8-year-old children in Northern Sweden with asthma diagnosed by a physician was found to be 6% \[[@B18]\]. A previous Swedish cumulative incidence investigation was reported for a defined region in 1992, where up to the age of 12--14 years, 5.3 % of an n = 1654 birth-year cohort were seen at the University Hospital with a clinically confirmed diagnosis of asthma \[[@B19]\]. Further examination of the epidemiology and natural history \[[@B20]\] of childhood asthma is therefore warranted, especially in infancy since it starts before the age of six in about 80--90 per cent of the cases. Hereby, comprehensive coverage in a well-defined area might complement the mainly large-scale surveys that are so far available. Also the etiological aspects, i.e. the associations and determinants of the disorder are of high interest. Since before, factors like changes in housing conditions, broad-spectrum antibiotic use, viral infections, dietary habits, less out-door activities, young maternal age, season of birth, living in urban or rural areas, and social, cultural and economic conditions \[[@B21]-[@B28]\] have been discussed, but not considered sufficiently clarified \[[@B29]\]. The new, mainly ISAAC findings here provide a host of extra, though as a rule cross-sectional and scattered information, e.g., on food allergy \[[@B30]\], rhinitis \[[@B31]-[@B33]\], wheezing \[[@B33]-[@B37]\], reduced pulmonary function \[[@B34],[@B35]\], atopy \[[@B31],[@B34],[@B36],[@B38],[@B39]\], virus infections \[[@B40]\], smoking \[[@B38],[@B39],[@B41]\], socioeconomic status \[[@B42]\], diet \[[@B31]\], behaviour problems \[[@B43]\], exposure to pets \[[@B38],[@B44],[@B45]\], indoor chlorinated swimming pools \[[@B46]\], per capita gross national product \[[@B47]\], and Caesarean section \[[@B48]\]. However, a more overall inventory of risk factors \[[@B49],[@B50]\], especially according to the standards of a regular case-control study \[[@B51],[@B52]\] is still rare. The present account of the findings in a defined and representative geographic area, involving childhood asthma and related conditions, hospital and primary care services, and diagnosis and questionnaire assessments, should therefore be of value, especially since it emanates from a country with previously low childhood asthma prevalence in an international comparison. The aims were to follow up at the age of ten the patterns of asthma symptoms and associations among children with a clinically diagnosed asthma in a sizeable urban-rural community and to compare them with demographic controls using a standardised questionnaire. Methods ======= In the representative, mixed urban-rural geographical region of Linköping in the county of Östergötland in Southern Sweden, with a population of about 150 000, all children (n= 2 104) born in 1990 at the University Hospital (which is the only somatic hospital in Linköping and where all births occurred) except those suffering neonatal death and those living outside the region, were recorded. For all of them still living in the region at the age of seven, the computerised medical records of the Department of Paediatrics at the University Hospital and at all 14 Primary Health Care (PHC) Units and at the two existing private Paediatrician Offices in the region were examined for the occurrence of the principal diagnosis asthma (ICD-9: 493). Secondary asthma-suspected or asthma-like symptoms were disqualified. Data of perinatal and obstetric factors as well as some social factors at baseline (1990) were obtained by investigations of the mothers\' medical records at the University Hospital including check-ups in the PHC organisation of statements made by the mid-wife in the medical records during the pregnancy. Children born 1990 in the region and still living in the region at the age of seven, was the inclusion criteria. At the age of seven n = 1752 children (n = 845 girls and n = 907 boys) met these criteria\'s as shown in figure [1](#F1){ref-type="fig"}. Practically all of the missing children at this follow-up had moved out of the region, only a few had deceased. Children born in 1990 that had moved in to the region after 1990 were excluded from the present study. In the follow-up at the age of seven, n = 191 children of this birth cohort were found in the medical records with a documented asthma diagnosis. In all, n = 41 cases of asthma (8.5 %) were identified in the urban girls and n = 31 (8.5 %) in the rural, whereas the corresponding figures for the urban and rural boys were n = 73 (14.2 %, relative risk (RR) to all girls 1.67, p \< 0.005) and n = 46 (11.7%, RR to all girls 1.37, p = 0.051), respectively \[[@B53]\]. Together, these 191 cases gives a total cumulative asthma incidence at the age of seven of 10.9 %, overall significantly (p \< 0.05) higher among the boys (13.1 %, n = 119) than the girls (8.5%, n = 72). Increased relative risks were also noted in children born in the autumn and winter, and in children born of the 392 mothers in the youngest maternal age bracket, \<25 years \[[@B53],[@B54]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Participants in the study. ::: ![](1471-2296-6-16-1) ::: To each of the identified asthma cases at the age of seven, a matched control (a child of the same sex with the nearest birth date i.e. \"a demographic twin\") was identified forming an equally large control group (n = 191). When the children with a documented asthma diagnosis were ten years old, their parents and their matched controls (total n = 382) were sent the International Study of Asthma and Allergies in Childhood (ISAAC) questionnaire \[[@B11]\] concerning asthma history, symptoms, related conditions, heredity, socio-economic factors and environmental exposure. The response rate to this postal questionnaire was 82.7 % (n = 158) in the asthma cases and 89.5 % (n = 171) in the controls, in total a response rate of 86.6 %. The number of matched twins gained with both a case and a control respondent was limited to n = 144 cases and n = 144 controls. However, the findings of the ISAAC questionnaire in this report are presented for all the responders from the case group (n = 158) and the control group (n = 171), respectively, in total n = 329 responders. All data were stored and computerised in a common database and statistically analysed using the SPSS-program. In the statistical analysis, differences were assessed by the chi^2^-method. The estimation of cumulative asthma incidence at the age of ten in the cohort was based on an assumption that the same proportion of new cases occurred in the whole cohort between the age of seven and ten as among the controls. Results ======= There were no significant differences concerning social factors like; having younger or older siblings, social class and living conditions between the cases and controls, as presented in Table [1](#T1){ref-type="table"}. The proportion of smoking mothers was higher (p = 0.007) among the cases (33.6 % smoking mothers) than among the controls where 20.5 % of the mothers were smokers. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Comparison of background factors between cases and controls. ::: Case (n = 158) Control (n = 171) --------------------------------------------------------- ---------------- ------------------- ----- ------ ------- Gender Male 102 64.5 107 62.5 Female 56 35.4 64 37.4 0.71 Having older siblings 99 62.6 108 63.1 0.37 Having younger siblings 92 58.2 93 54.3 0.45 Social class of father 1 21 13.2 20 11.7 0.85 2 51 32.2 57 33.3 3 75 47.5 87 50.8 Social class of mother 1 8 5.0 7 4.0 0.35 2 69 43.6 62 36.2 3 81 51.2 100 58.4 Living conditions during the child\'s first 3 years Countryside 61 38.6 79 46.2 0.16 City 92 58.2 92 53.8 Mixed 5 3.2 0 0.0 Villa/house 102 64.5 116 67.8 0.46 Apartment 53 33.5 49 28.6 Mixed 3 1.8 6 3.5 Living conditions at the age of 10 Countryside 71 44.9 89 52.0 0.16 City 83 52.5 82 47.9 Mixed 4 2.5 0 0.0 Villa /house 118 74.7 135 78.9 0.29 Apartment 36 22.7 35 20.4 Mixed 4 2.5 1 0.5 Smoking habits of father Non-smoker 117 74.1 140 81.8 0.22 Smokes 0--9 cig/day 13 8.2 15 8.8 Smokes 10--20 cig/day 19 12.0 11 6.4 Smokes \>20 Cig /day 2 1.3 5 2.9 Smoking habits of mother Non-smoker 105 66.4 136 79.5 0.007 Smokes 0--9 cig/day 24 15.2 13 7.6 Smokes 10--20 cig/day 27 17.1 20 11.7 Smokes \>20 Cig /day 2 1.3 2 1.2 Smoking habits of other members of the household Non-smoker 152 96.2 164 95.9 0.34 Smokes 0--9 cig/day: 4 2.5 6 3.5 Smokes 10--20 cig/day: 0 0.0 1 0.5 Smokes \>20 Cig /day 2 1.2 0 0.0 Smoking at home during the first 3 years No: 128 81.0 143 83.6 0.53 Yes: 30 18.9 28 16.3 ::: There were no differences between cases and controls in self-reported in-door exposure of mist, mould or dry air inside the house. Exposure to pet animals like cat, dog, and animals with furs or cage birds was almost the same between cases and controls (not shown). Among the cases, 39.3 % reported asthma heredity in the family and this was significantly lower 26.4% (p = 0.007) among the controls, as presented in Table [2](#T2){ref-type="table"}. Heredity for allergic rhinitis in the family was reported by 57.6 % among the cases and 46.8 % among the controls. Almost 50 % in both cases and controls reported heredity for eczema as a child in the family. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of heredity between cases and controls. ::: Case (n = 158) Control (n = 171) ------------------------------ ---------------- ------------------- ----- ------ ------- Asthma No 96 60.7 126 73.6 0.007 Father 12 7.6 7 4.1 Mother 12 7.6 7 4.1 More than one in the family 18 11.4 5 2.9 Sibling 20 12.6 26 15.2 Allergic rhinitis No 67 42.4 91 53.2 0.10 Father 16 10.1 20 11.7 Mother 29 18.4 17 9.9 More than one in the family 34 21.5 28 16.3 Sibling 12 7.6 15 8.8 Eczema as a Child No 80 50.6 87 50.8 0.97 Father 8 5.0 6 2.9 Mother 15 9.5 17 9.9 More than one in the family 22 13.9 25 14.6 Sibling 33 20.8 36 21.0 ::: Table [3](#T3){ref-type="table"} gives an overview of past and present symptoms and signs of asthma. It is seen that they are as such highly significantly over-represented in the cases. Wheezing or whistling in the chest at any time in the past was reported by around 75 % of the asthma cases. Even among the controls almost one fourth reported this symptom, while its frequency during last year was 26.6 versus 5.8 %. In around one fourth of the cases it was reported that the chest of the child had sounded wheezy after or during exercise during the last year, in comparison with 5% of the controls. Sleep and speech disturbances were exceptional and more frequent among the cases, but not significantly so. However, dry cough at night during the last 12 months, not associated with a cold or chest infection, was significantly (p \< 0.0001) more common among the cases (33%) in comparison with the controls (11%). Only in a few children, but significantly more among the cases (p = 0.007), it was reported that they had been blocked in the chest or experienced mucous cough ≥ 4 days per week during a period of at least 3 months per year. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of symptoms between cases and controls. ::: Case (n = 158) Control (n = 171) ------------------------------------------------------------------------------------------- ---------------- ------------------- ----- ------ ----------- A. Wheezing or whistling in the chest at any time in the past Yes 116 73.4 40 23.4 \<0.0001 No 42 26.6 131 76.6 B. Wheezing or whistling in the chest in the last 12 months Yes 42 26.6 10 5.8 \<0.0001 No 116 73.4 161 94.1 C. How many attacks of wheezing in the last 12 months None 113 71.5 161 94.1 \<0.0001 1--3 25 15.8 5 2.9 4--12 18 11.3 5 2.9 More than 12 2 1.3 0 0.0 D. Disturbed sleep due to wheezing in the last 12 months Never woken with wheezing 137 86.7 163 95.3 0.08 Less than one night per week 20 12.6 6 3.5 One or more nights per week 1 0.6 2 1.1 E. Limiting speech to only one or two words at a time between breaths last 12 months Yes 5 3.1 3 1.7 0.40 No 153 96.8 168 98.2 F. Has the child\'s chest sounded wheezy after or during exercise last 12 months Yes 42 26.6 9 5.2 \<0.0001 No 116 73.4 162 94.7 G. Dry cough at night (not associated with a cold or chest infection) last 12 months Yes 52 32.9 19 11.1 \<0.0001 No 106 67.1 152 88.9 H. Blocked in chest or mucous cough ≥ 4 days/week during in total ≥ 3 months per year Yes 11 6.9 2 1.1 0.007 No 147 93.0 169 98.9 I. Problem with sneezing or a runny or blocked nose without a cold or the flu Yes 70 44.3 37 21.6 \< 0.0001 No 88 55.7 134 78.4 J. Itchy rash coming and going for at least 6 months Yes 49 31.0 35 20.5 0.03 No 109 68.9 136 79.5 ::: Problem last year with sneezing or a runny or blocked nose without a cold or flu was reported by 44 % of the asthma cases and by just over 20 % of the controls. Coming and going itchy rash during last 6 months was significantly (p = 0.03) more frequently reported in the cases (31%) in comparison with the controls (20%). In the follow-up ISAAC-questionnaire the parents were asked if their child ever had asthma, eczema or hay fever. The results are presented in Table [4](#T4){ref-type="table"}. Among the asthma cases 46.8 % also reported eczema, while the corresponding figure for the controls was 39.2 %. Hay fever was reported for 21.5 % of the asthma cases and 14 % for the controls. In the cases with a well-documented asthma diagnosis, 39.2 % of the parents answered that their child never had asthma. On the other hand, among the control group 4.7 % (8 children; 3 girls and 5 boys) of the parents reported that their child actually had asthma. These 8 children are possibly new asthma cases that had occurred between the age of seven and ten. If so, one can estimate an incidence rate at the age of ten for the studied cohort. The rate at the age of seven in the cohort was 10.9 %. The eight potentially new cases represent 4.7 % of the control group (n = 171) and if the same proportion (4.7 %) is applied for the whole cohort (4.7 % of n = 1 561 children) it gives an estimated additional number of n = 73 new asthma cases between the age of seven and ten, which gives an estimated cumulative incidence rate at the age of ten of 15.1 %. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Comparison of previous diseases between cases and controls. ::: Case (n = 158) Control (n = 171) ---------------------------------------- ---------------- ------------------- ----- ------ ---------- Has your child ever had asthma? Yes: 96 60.7 8 4.7 \<0.0001 No: 62 39.2 163 95.3 Has your child ever had eczema? Yes: 74 46.8 67 39.2 0.16 No: 84 53.2 104 60.8 Has your child ever had hay fever? Yes: 34 21.5 24 14.0 0.075 No: 124 78.5 147 85.9 ::: A comparison of symptoms at the age of ten between the eight potentially new asthma cases and the previously confirmed asthma cases and controls is shown in Table [5](#T5){ref-type="table"}. It is seen that in terms of symptoms and associations they appear as genuine cases and hence also diagnostically qualify as potentially new cases between the age of seven and ten from the part of the initial non-asthma cohort. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Symptoms (A. to J. as in Table 3) among potentially new asthmacases in comparison with previously identified asthma cases and controls. ::: --------------------------------------------------------------------------------------------------------- New asthma cases\"\ Asthma cases\ Controls\ (n = 8) (n = 158) (n = 163) -------- ---------------- --------------------- --------------- ----------- ------ ---- ------ ---------- A. Yes 8 100.0 117 73.6 33 20.1 \<0.0001 B. Yes 4 50.0 42 26.4 7 4.3 \<0.0001 C. Yes, 1 or more 4 50.0 45 28.3 7 4.2 \<0.0001 D. Yes 3 37.5 21 13.2 5 3.0 \<0.0001 E. Yes 1 12.5 5 3.1 2 1.2 0.09 F. Yes 4 50.0 42 26.4 6 3.7 \<0.0001 G. Yes 2 25.0 53 33.3 17 10.4 \<0.0001 H. Yes 0 0.0 11 6.9 2 1.2 0.03 I. Yes 4 50.0 70 44.0 34 20.7 \<0.0001 J. Yes 1 12.5 49 30.8 34 20.7 0.08 --------------------------------------------------------------------------------------------------------- ::: Discussion ========== When the diagnosis of childhood asthma in the large majority of recent studies is based upon the ISAAC questionnaire, which has been tested with a sensitivity in relation to various standards between 64--76 \[[@B5]\] and 75--87 \[[@B6]\] but generally low specificity and predictive values, the obtained prevalence is remarkably high; from about 6.7 -- 10.2 in Brazilian schoolchildren \[[@B8]\], 9 % in Texas \[[@B5]\], 7.2 -- 9.6% in Palestine \[[@B7]\], 16.3 % in Australian indigenous children \[[@B6]\], and 22.3 % in the northeast of England \[[@B9]\]. This is clearly higher than corresponding data from, e.g., 1994 \[[@B16]\], and calls for further study including clinical and questionnaire comparison, where we think that our project is of value both in terms of scale, relevance and representativity. It covers a whole birth-year cohort in a sizeable and epidemiologically well defined, mixed urban-rural region. In the initial study group, it is based on the totality of principal asthma diagnosis clinically confirmed and documented in the health services network of Linköping, i.e., not only at the University hospital but also in the private paediatrician and public PHC organisation. In the follow-up it is complemented by the thoroughly validated ISAAC questionnaire, allowing assessment and cross-comparison of symptoms and some relevant proxy associations in both cases and controls, as well as transfer of by these means identified new cases from the latter to the former group leading to a complete 0--10-year cumulative incidence estimation in the order of 15.1 %. We believe that strength of our study is its case-control design in which the ISAAC questionnaire was complementary and could be followed up clinically. We therefore believe that the 0--7-year incidence for childhood asthma of around 11 % and 0--10-year around 15 % reflects the contemporary situation in the studied Swedish region. Since only principal asthma diagnosis was included, there are few false positives; in fact, because the clinical diagnosis of asthma is regularly used as standard for sensitivity and specificity calculations, these expressions are not appropriate here. However, even if it is as such methodologically acceptable to estimate asthma incidence at age 10 by adding to the rate at age seven the new cases that developed in the control group, the further approximation in this based upon the data in the questionnaire cases presents a certain limitation to the findings. A more detailed clinical and therapeutic assessment of the material including also the cases reported to be free of asthma at the follow-up should be warranted in coming studies. This apparent remission in a proportion of the cases is interesting per se and somewhat at variance with the stereotype of asthma as a permanent chronic illness with, in analogy with other long-term childhood diseases, implicitly more serious prognosis the earlier the onset. The children in the ten-year follow-up reported to be free of asthma, in consequence represent an interesting category. As such, parental reports of childhood asthma have been found to be reliable \[[@B55],[@B56]\], so this question clearly warrants separate examination because elucidation of the patterns in that group might give additional preventive and therapeutic clues. However, also the patterns in the persisting cases provide several but still often quite puzzling hints to that end. One must again ask what the reasons are: of the disorder itself as well as of its steep world-wide rise of late. It is apparent, that expanded health services have not served to alleviate the problem. The studied Swedish region has a well developed infant and school health organization -- yet childhood asthma escalates almost uncontrollably as judged from the obtained incidence data; from 5.3 % in 1992 \[[@B19]\] to the current estimated 15.1 %. In addition there are high and rising figures for symptoms like rhinitis and wheezing which, as well known in previous studies \[[@B31],[@B37]\], might be an associated but also an extra problem as inferred from the overlap between cases and controls in table [3](#T3){ref-type="table"}. The higher frequency in boys is well known \[[@B5]-[@B12],[@B53]\], too, and again one might ask: Why? Implicitly it somewhat goes against the tacit assumption that overly cleanliness predisposes to asthma. On the other hand, the seasonal variation with autumnal, in the Northern Hemisphere September, October and November incidence peaks is likewise established \[[@B53]\], and in that connection, boys might be more exposed to outdoor, plausibly physiological strains, notably cold, which from earlier studies showing higher childhood asthma prevalence in Northern than in Southern parts of Sweden \[[@B57]\] would seem to be of importance. Urban dwelling were other factors that showed weakly significant differences between cases and controls in this study, whereas we could not verify associations with socio-economic class or pet animal or indoor allergen exposure. Smoking exposure in the family, especially from the mother, is confirmed to be of importance in study as in previous follow-up \[[@B54]\]. A relation to asthma and to a lesser extent hay fever heredity was confirmed, however, and stemming from such a complete and sizeable material indicates that one important line of future research, which we are following also in our project, goes into a more specific constitutional and genetic direction \[[@B52],[@B54]\]. As expected, established symptoms of asthma were significantly over-represented in the clinically diagnosed cases, but there was a strong overlap to the controls in this study. This supports that questionnaires alone may serve well as screening instruments but constitute somewhat of a circle argument in establishing an asthma diagnosis. It is true that one can discern highly specific constellations of symptoms and signs, but they are rare so do not contribute much to the operative strength. It has been pointed out in the literature, that such shortcomings strongly call for more stringent case-control studies of the type here reported \[[@B51],[@B52]\], and from which further investigations on the individual level are then of equally large interest. Reciprocally, similar applies to the non-correlation to common cold, and, more intriguingly, eczema that we noted in this study. We think that the further study potential is strengthened by the completeness of the material. Instrumental for this was the engagement also of the non-hospital services in its recruitment. Even when the University Hospital provided the bulk of cases, the virtually total population coverage of the PHC organisation provided the otherwise missing. Moreover, the high response frequency to the ISAAC questionnaire that we used in a confirming sense, both shows the high parental concern understandably given to asthma in their child and ensures a maximal degree of completeness of the data. Conclusion ========== A combination of medical verified asthma diagnosis through medical records and the use of self-reported symptom through the ISAAC questionnaire seem to be valid and reliable measures to follow-up childhood asthma in the local community. The asthma prevalence at the age of ten in the studied birth cohort is considerably higher than previous reports for Sweden. Both the high prevalence figure and allowing the three-year lag phase for further settling of events in the community point at the complementary roles of both hospital and primary care in the comprehensive coverage and control of childhood asthma in the community. As inferred from the symptom spectrum presented in this report, especially the therapeutic aspects call for increased study and the findings support that equal alert is warranted during the whole childhood period. The follow-up indicated an almost five percent additional incidence in the period 8--10 years of age. It is thus tempting to conclude that there is a continuous interplay between external and constitutional factors, which are so far only quite scarcely known and accordingly warrant extensive further research. Also applying to the considerable fraction of cases, as judged from the parental report in this study, that are cured from the disease, such studies are under way. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= ER, ÅF and TF conceived and designed the study, participated in the collection, statistical analysis and interpretation of data and drafted the manuscript. ET and OZ participated in the analysis and interpretation of data and draft of the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2296/6/16/prepub>	PubMed Central	2024-06-05T03:55:56.502724	2005-4-21	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090570/", "journal": "BMC Fam Pract. 2005 Apr 21; 6:16", "authors": [ { "first": "Eduardo", "last": "Roel" }, { "first": "Åshild", "last": "Faresjö" }, { "first": "Olle", "last": "Zetterström" }, { "first": "Erik", "last": "Trell" }, { "first": "Tomas", "last": "Faresjö" } ] }
PMC1090571	Background ========== The SPR technique is suitable for studying biomolecular interactions on, or close to, a surface \[[@B1]\]. SPR enables rapid detection in real time without any labeling of samples and can be utilised for concentration determination, kinetics studies and epitope mapping. In brief, light incident on a metal surface at a given angle of incidence can excite a surface-bound electromagnetic wave, a surface plasmon, which propagates along the interface between the metal and the ambient medium. Associated with the surface plasmon is an evanescent field that probes local changes in the refractive index of the ambient medium, induced, for example, by binding of a biomolecule to the surface. The change in refractive index will shift the angle of incidence at which the SPR excitation occurs. This shift in the angle is tracked by monitoring the movement of intensity minima of the reflected light with time using the Kretschmann configuration, and the binding event is presented as a sensorgram \[[@B1]\]. The sensor surface of the SPR apparatus consists of a carboxy-methylated dextran matrix, which is a hydrogel providing a solution-like environment in which the biospecific interactions occur. The carboxyl groups in the dextran matrix enable covalent coupling of ligands (proteins, receptors, DNA etc) to the surface. Hepatocyte growth factor (HGF) is a growth factor that is produced by mesenchymal cells during injuries in various organs. It stimulates cell division \[[@B2]\], motility \[[@B3]\], and a normal morphogenic structure \[[@B4]\] in the epithelial cells that have survived an injury (adjacent to the injured area). HGF is translated as a single-chain precursor and activated at the site of injury by proteolytic cleavage, resulting in a double-chained active HGF \[[@B5]\]. Interaction between the active HGF and its specific receptor (c-Met) \[[@B6]\] initiates intracellular signal pathways that result in regeneration and repair of damaged tissue \[[@B7]\]. High amounts of HGF have been determined systemically during injuries caused by infection \[[@B8]\]. At the site of infection, a local production of HGF has been found during bacterial meningitis and pneumonia \[[@B9],[@B10]\]. Using a commercially available ELISA kit, we have shown that the concentration of HGF in faeces increases significantly during infectious gastroenteritis \[[@B11]\]. The levels decrease to normal after recovery. We have further studied the presence and stability of HGF in faeces samples during infectious processes \[[@B12]\]. Since traditional immunoassays (e.g., ELISA) are time-consuming and laborious, there is a need for quicker and simpler methods. In this study we present a method based on surface plasmon resonance, SPR, which allow us to detect and measure the levels of HGF in faeces in individual samples with the same accuracy but more rapidly than traditional methods. Methods ======= Patients -------- Nine patients admitted to hospital with signs of acute untreated infectious diarrhoea were included. These patients had participated in a previous study \[[@B11]\]. Faeces samples were taken at admittance and cultured (Campylobacter, Salmonellaand Shigella), and laboratory diagnostic tests for Rotavirus(antigen detection) and Calicivirus(electron microscopy) as well as cultures and cytotoxin analysis for Clostridium difficilewere conducted. Cultures and cytotoxin analysis (toxin A) were positive for Clostridium difficilein 1 case. Cultures revealed growth of Campylobacter jejuniin 4 cases. Salmonellaart (Salmonella DO) was found in 3 cases. Faeces culture was negative in one patient. This patient had diarrhoea and fever and his wife had gastroenteritis caused by Campylobacter jejuni. A second group of eleven patients with Clostridium difficile(positive culture and toxin A) were also analysed. This gave a total of 20 samples of faeces from 20 patients (11 women, 9 men, range 20--85, median 51.5 and mean 52 year). Faeces haemoglobin (Actim Fecal Blood test; Orion Diagnostica) was examined and was positive in 10/13 samples. Controls -------- Faeces samples were obtained from 10 healthy vaccination volunteers (six women and four men, 20--60 years) without signs of infection or diarrhoea. Faeces culture and diagnostic tests for Rotavirus and Caliciviruswere negative in all of these cases. These patients had negative Actim Fecal Blood test results (Orion Diagnostica) Standardizing faeces volume and reconstitution procedure -------------------------------------------------------- All faeces samples were stored at -20°C. Prior to handling, the samples were thawed at room temperature and mixed using a Vortex (Vortex-Genie, Scientific Industries Inc., Bohemia, NY, USA). The samples were then placed at -70°C for 15 minutes, followed by room temperature for 2 minutes and dissolved in distilled water at a dilution of 1:6. The suspension was centrifuged at 1000--3000 Gfor 15 minutes and the supernatant stored at -70°C until analysed. To avoid the effects of digestive as well as bacterial enzymes, protease inhibitor (1--5%) (Sigma Aldrich) with specific inhibition of serine, cysteine, aspartic proteases and aminopeptidases, containing 4-(2-aminoethyl) benzenesulfonyl fluoride (AEBSF), pepstatinA, E-64, bestatin, leupeptin, and aprotinin (but no metal chelators) was added to the thawed samples (room temperature) 30 minutes prior to analysis. SPR measurements and ligand immobilisation procedures ----------------------------------------------------- SPR measurements were conducted at 760 nm in a fully automatic Biacore 2000 instrument and a semi-automatic Bicaore X instrument (Biacore AB, Uppsala, Sweden) equipped with four and two flow cells respectively. The flow cell temperature was 25°C in all experiments. The sample surfaces used were carboxy-methylated dextran CM5 chips (Biacore AB, Uppsala, Sweden). Coupling of ligands to the carboxylic acid groups of the dextran hydrogel was carried out by conventional carbodiimide chemistry using 200 mM EDC (N-ethyl-N\' -(3-diethylaminopropyl) carbodiimide) and 50 mM NHS (N-hydroxysuccinimide). The activation time was 7 min, followed by a 2--7 min ligand injection. Deactivation of remaining active esters was performed by a 7 min injection of ethanolamine/hydrochloride at pH 8.5. A flow rate of 5 μl/min was used during immobilisation. All ligands were diluted in 10 mM acetate buffer pH 4.5, i.e., below the protein isoelectric point, thus enhancing the electrostatic interactions between the dextran matrix and the ligands. The monoclonal anti-HGF (500 μg/ml) was diluted 1:10, the recombinant Met proto-oncogene receptor (100 μg/ml) 1:5, and the HGF recombinant (5 ug/ml) 1:3. The contact time varied between two and seven minutes resulting in levels of immobilisation between 8000 and 30000 RU (response units). After deactivation, the surfaces were washed with five subsequent one-minute injections of 5 mM glycine buffer pH 2.0 with 1 M NaCl. One of the flow cells was used as a reference to monitor the response due to buffer and unspecific interactions. This flow cell was treated in the same way as the other during the immobilisation procedure, but omitting the ligand immobilisation step. Determination of faeces HGF by ELISA ------------------------------------ After storage all samples were thawed and centrifuged at 1000 Gfor 15 minutes prior to analysis. Immunoreactive HGF was determined by ELISA using a commercially available kit (Quantikine HGF Immunoassay, R&D Systems Inc., Minneapolis, USA). Determination of faeces haemoglobin ----------------------------------- Actim Fecal Blood (Orion Diagnostica) is an immunochromography technique that is specific for the determination of human haemoglobin using two monoclonal antibodies. The detection limit in faeces is 50 μg haemoglobin/L or 25--50 μg haemoglobin/g. The method detects the intact haemoglobin molecule, but neither haemoglobin that is influenced by the enzymes during gastrointestinal passage nor animal haemoglobin is detected. Purification of faecal samples by affinity chromatography, ultrafiltration, and SDS-PAGE ---------------------------------------------------------------------------------------- The affinity chromatography columns (Hi-trap Amersham Biosciences) were immobilised with monoclonal anti-human HGF and recombinant human HGF receptor (c-MET)/Fc chimera (R&D Systems) respectively. Faeces samples were reconstituted in distilled water. The samples were applied to the 500 μl centrifugal tubes (Amicon Ultra, Millipore, S.A.S. Molsheim, France) and centrifuged at 4000 G for one hour. SDS-PAGE was performed with the faeces samples before and after purification and filtration, using a 4% stacking gel and a 12% running acrylamide gel \[[@B13]\]. For size estimation prestained protein standard (Sigma Aldrich) was electrophoresed simultaneously. Statistics ---------- Non-parametric Friedman and Wilcoxon signed rank tests (absolute values, Statview & SPSS Base 11.0) were used. The Spearman rank correlation coefficient was used for analysis of correlation between parameters. A p-value ≤ 0.05 was regarded as statistically significant. Results ======= Patients with infectious gastroenteritis had significantly higher signal responses in flow cells immobilised with monoclonal anti-HGF antibody compared to the healthy controls (determined in at least six independent experiments)(Friedman p = 0.006, Wilcoxen Signed Rank Tests p = 0.008) (Fig [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}). The signal responses in the flow cells immobilised with c-Met receptor were also significantly higher in the group with infectious gastroenteritis (Wilcoxen Signed Rank Tests, p = 0.02). Correlation with ELISA ---------------------- Determination of HGF in faeces (patients with infectious gastroenteritis) by utilising SPR correlated significantly (n = 20, r = 72%, p \< 0.001) with the levels measured by ELISA (range 0.14--8.24, median 0.83 ng/mL) (Fig [3](#F3){ref-type="fig"}). HGF levels determined by ELISA in healthy controls (n = 10) were low (range 0.01--0.26, median 0.06) Correlation with age -------------------- We found a negative correlation between the SPR signal intensity and age in patients with infectious gastroenteritis (n = 20, r = - 0.55, p = 0.010) (Fig [4](#F4){ref-type="fig"}). The patients with acute gastroenteritis were older than the healthy controls (median 51.5 respective 35.0 years). Effects of altering the immobilisation levels --------------------------------------------- The effect on signal responses due to the amount of monoclonal anti-HGF antibody immobilised was determined in different samples (n = 12). Increasing the immobilisation levels by increasing the contact time during immobilisation (1, 5, and 10 minutes) caused a significantly higher response of all samples (Friedman p \< 0.01, Wilcoxon signed ranks test between 1 and 5 minutes p = 0.058, between 5 and 10 minutes p = 0.003, and between 1 and 10 minutes p = 0.003) (Fig [5](#F5){ref-type="fig"}). Altering the flow rate (5, 10, and 15 μl/min) did not have a significant effect on the response of the different samples (Friedman p = 0.48) (data not shown). When altering levels of immobilisation of the c-Met receptor (n = 15), no effect in responses in the case of infection (P = 0.85) was observed. Changing of the flow rate (5, 10, and 15 μl/min) did not influence the signal responses (P = 0.36) (data not shown). Binding of HGF to dextran ------------------------- Addition of dextran (0.05%) to the samples resulted in a significant (p \< 0.05) decrease in signal responses in flow cells with immobilised anti-HGF but did not affect the HGF binding to c-met (p = 0.11). Diluting recombinant HGF with dextran (0.05%) also resulted in a decrease of binding to the monoclonal anti-HGF antibody (n = 3). The test of significance was not performed. SDS-PAGE -------- We investigated the presence of HGF in the samples by SDS-PAGE of faeces samples as well as SDS-PAGE of purified faeces samples in an affinity chromatography column immobilised by either mouse anti-human HGF monoclonal antibody or recombinant human c-met receptor. Bands corresponding to the HGF were detected with apparent molecular masses of 75--90 kDa (Fig [6](#F6){ref-type="fig"}). Protease Inhibitor ------------------ Faeces destroyed the proteins immobilised in the flow cells of the CM5 chip (Fig [7](#F7){ref-type="fig"}). Adding protease inhibitor (1--5%) to the faeces samples within 30 minutes prior to experiments could inhibit this effect completely. Discussion ========== In the present work the Surface Plasmon Resonance (SPR) based method has been used for the first time to evaluate faecal samples and monitor the relative amount of HGF in the samples. HGF has been investigated widely over the past decade. The unique properties of this cytokine, makes it likely to be involved in the recovery process after injuries \[[@B3]\]. Recent studies have strengthened the notion that HGF plays an important role in the regeneration of an injured organ \[[@B2]\]. Several researchers have attempted to treat induced hazardous injuries in animal models by HGF \[[@B14],[@B15]\]. We have studied HGF during infectious diseases and found that HGF was produced in high amounts both systemically and locally during injuries caused by infection \[[@B8]\]. Low amounts of serum HGF in patients with pneumonia correlated significantly to poor prognosis \[[@B16]\]. Application of HGF locally at the site of an injury such as a chronic ulcer resulted in an accelerated healing process \[[@B17]\]. The gastrointestinal mucosa has a remarkable ability to repair damage, and growth factors play an important role in the regeneration of injured cells in gastrointestinal organs \[[@B18]\]. Nishimura et al. (1998) \[[@B19]\] showed that HGF was the most potent of the cytokines (HGF, TGF-α, TGF-β, and keratinocyte growth factor) in accelerating repair of the damaged monolayer of an epithelial cell line derived from normal rat small intestine. We studied the amounts of HGF in faeces of patients with infectious gastroenteritis and found that the levels of HGF were significantly elevated during infection \[[@B11]\]. Although this observation might indicate healing of injury caused by an infection, we could discriminate with high specificity and sensitivity between infectious gastroenteritis and other disorders that present with diarrhoea \[[@B11]\]. In a previous study \[[@B12]\] the concentration and stability of HGF in fecal samples was investigated using ELISA. Although the ELISA method was very reliable, it was laborious and gave no information about the form of HGF that was present in the samples. In the present study we found that there is good correlation between the results obtained from SPR and ELISA measurements of faecal HGF levels. In some cases differences between results obtained by ELISA and SPR are found and this phenomenon is to be further investigated. Using SPR we were able to monitor the presence of HGF in the samples and the interactions between HGF and different ligands with different binding specificity to the HGF molecule. First, a recombinant HGF that showed affinity to the monoclonal anti-HGF antibody as well as c-met receptor was examined (data not shown). The affinity of recombinant HGF disappeared within one week after reconstitution of the lyophilised form (data not shown). On the other hand, the signal responses after analysis of faeces samples in the patients with infectious gastroenteritis were stable after reconstitution. We have tested widely the stability of HGF in faeces samples by ELISA \[[@B12]\] and previously in serum \[[@B20]\]. Unlike our previous observations \[[@B11]\] that showed no significant correlation between age of patients and HGF levels determined by ELISA, in this study we found a negative correlation between age and SPR signal in the group with infectious gastroenteritis. This might strengthen the notion that the ability of HGF to interact with its ligand or possibly its activity might change by age. Moreover, in spite of the fact that patients with infectious gastroenteritis were older than healthy controls they had higher SPR signals. Presence of blood was shown in ten out of thirteen cases with infectious gastroenteritis while it was negative in healthy controls. However we have previously studied the correlation between presence of blood in faeces and HGF levels (determined by ELISA) and did not find any significant correlation \[[@B12]\]. The immoblised layer in the flow cells of the chip were destroyed after analysis of faeces samples (Fig [7](#F7){ref-type="fig"}). Adding protease inhibitor to the samples before analysis inhibited this effect which might indicate the presence of proteases in faeces. We have previously shown that adding protease inhibitor to faeces does not influence the ability to detect HGF by ELISA \[[@B12]\]. At least two binding sites for HGF on liver cell surfaces have been demonstrated in rat: the heparin-resistant and acid-washable site (HGF receptor, binding site) and the heparin-washable site (low-affinity, cell-surface heparan sulfate proteoglycan binding site) \[[@B21]\]. HGF has also been reported to interact with the extracellular matrix \[[@B22]\]. The low affinity sites may serve as a reservoir for the regulation of endogenous HGF levels and provide a matrix for conversion of promitogen to the two-chain form \[[@B21]\]. The HGF detected during infectious gastroenteritis had binding affinity to dextran (proteoglycan). By adding carboxy-methyl dextran (0.05%) to the samples, the levels of HGF bound to the immobilised monoclonal anti-HGF decreased significantly. Adding the same amount of dextran to recombinant HGF gave the same result in signal responses (data not shown). This might indicate a binding affinity of this form of HGF to dextran, which might resemble such affinity to cell-surface heparan sulfate proteoglycan binding sites. These binding sites either overlap with the antibody-binding epitope or induce conformation shift in the HGF upon binding to the dextran, resulting in sterical hindrance and thus preventing binding to the antibody. Conclusion ========== Infectious gastroenteritis is a common disease in all societies. Determination of HGF by SPR is a reliable method to evaluate the relative amounts of HGF in faeces. This might be beneficial in diagnosis of acute situations that present with symptoms of gastroenteritis and may, possibly, guide appropriate medical treatments. Competing interests =================== The corresponding author is applying a patent that might be indirectly connected to the manuscript: -Rapid determination of hepatocyte growth factor (HGF) in the body fluids. There are no other financial competing interests. Authors\' contributions ======================= 1- FN:Participated in design of method, design of study and collecting of samples and drafted the manuscript. 2- DI:Participated in the design of method and in performing the statistical analysis. 3- TN:Performed the reconstitution procedure of faeces samples. Participated in design of method. Performed the SPR analysis. 4- JX:Carried out the affinity chromatography and SDS-PAGE studies for HGF in faeces. 5- SA:Participated in design of study, collecting of samples and statistical analysis. 6- BÅ:Performed the microbiological assessment of samples. 7- IL:Participated in design and co-ordination of the study 8- BL: Participated in design and co-ordination of the study and evaluation of reliability of the new method as an expert. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-230X/5/13/prepub> Acknowledgements ================ We are grateful to Lise-Lott Lindvall and Barbro Furuendal for nursing assistance. This study has been supported by a Pharmacia&Upjohn grant and by FORSS (the Research Council in Southeast Sweden). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Histogram showing signal responses determined by SPR (channel immobilised by mouse anti-human HGF monoclonal antibodies) in infectious gastroenteritis (n = 20, median 1895 ± 429 RU, range 123--6660 RU) and healthy controls (n = 10, median 171 ± 38 RU, range 14--459 RU). The response is defined as the difference in response signal before and after the injection of the sample minus the difference observed in the reference channel. ::: ![](1471-230X-5-13-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### SPR sensogram showing the signal obtained from analysis of faeces from a patient with infectious gastroenteritis. Note the difference in signal intensity between the control channel (dextran) and the channel immobilised by mouse anti-human HGF monoclonal antibody. ::: ![](1471-230X-5-13-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Correlation between the concentration of HGF in faeces obtained by ELISA and signal responses by SPR (Spearman rank correlation coefficient n = 20, r = 0.72, p \< 0.001). The shadow points along the regression line are not data points. ::: ![](1471-230X-5-13-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Correlation between age of the patients with infectious gastroenteritis and signal responses by SPR (Spearman rank correlation coefficient r = - 0.55, p = 0.01). The shadow points along the regression line are not data points. ::: ![](1471-230X-5-13-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Histogram showing the effects of increasing the immobilisation levels by increasing the contact time during immobilisation (1--3 representing 1, 5, and 10 minutes) (Friedman p \< 0.01, Wilcoxon signed ranks test between 1 and 5 minutes p = 0.058, between 5 and 10 minutes p = 0.003, and between 1 and 10 minutes p = 0.003). The monoclonal anti-HGF (500 μg/ml) was diluted 1:10 in 10 mM acetate buffer pH 4.5. The activation time was 7 min, followed by a 1,5 and 10 min ligand injection respectively. Deactivation of remaining active esters was performed by a 7 min injection of ethanolamine/hydrochloride at pH 8.5. A flow rate of 5 μl/min was used during immobilisation. ::: ![](1471-230X-5-13-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### SPS-PAGE of faeces. A: before purification; B: after purification in an affinity chromatography column immobilised by mouse anti-human HGF monoclonal antibody; C: after purification in an affinity chromatography column immobilised by recombinant human c-met/fc chimera. ::: ![](1471-230X-5-13-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### SPR sensogram from faeces analysis. The flow cells were immobilised by recombinant Met proto-oncogene receptor (20 μg/ml), monoclonal anti-HGF (50 μg/ml) and HGF recombinant (3 μg/ml), respectively. The last channel (lowest line) was used as a reference to monitor the response due to buffer and unspecific interactions. Every other signal is the regeneration signal (1:1 combination of 1M Nacl and Glycine pH 2.0). The signals at time limits 12000, 12500, 14100 and 14600 seconds belong to patients with infectious gastroenteritis. The signals at time limits 13000 and 13500 seconds belong to faeces from healthy control. Please note the decline in the base line of the flow cells at the time limit 13000 when a faeces sample was injected that did not contain protease inhibitor. ::: ![](1471-230X-5-13-7) :::	PubMed Central	2024-06-05T03:55:56.507585	2005-4-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090571/", "journal": "BMC Gastroenterol. 2005 Apr 12; 5:13", "authors": [ { "first": "Fariba", "last": "Nayeri" }, { "first": "Daniel", "last": "Aili" }, { "first": "Tayeb", "last": "Nayeri" }, { "first": "Junyang", "last": "Xu" }, { "first": "Sven", "last": "Almer" }, { "first": "Ingemar", "last": "Lundström" }, { "first": "Britt", "last": "Åkerlind" }, { "first": "Bo", "last": "Liedberg" } ] }
PMC1090572	Background ========== Recent advances in high-throughput genotyping technologies have realised the possibility of performing large-scale, high-resolution genetic studies in human complex diseases. Single nucleotide polymorphisms (SNPs) have become the markers of choice due to their frequent occurrence, simple mutational dynamics and the fact that they lend themselves to automated allele calling \[[@B1],[@B2]\]. The number of SNPs with allele frequencies higher than 10% has been estimated to exceed 5,000,000 \[[@B3]\]. Exhaustive genome-wide association studies, thereby, reach prohibitive costs and require ultra-high throughput technologies. Comprehensive SNP screening of regions or genes of interest is both inefficient and unnecessary, as information redundancy can arise from linkage disequilibrium (LD). A common strategy in complex disease association studies is the selection and genotyping of a subset of SNPs, assumed to be in LD with the untested polymorphisms. In the past, association study designs have not selected markers on a strong scientific basis, due to restricted comprehension of LD patterns. Gaining a better understanding of the LD blueprint of the human genome can now facilitate disease gene mapping, as sets of non-redundant SNPs can be employed to design cost-effective strategies \[[@B4]-[@B6]\]. SNP maps utilised by current genetic studies concentrating on chromosomal regions cover a wide spectrum of marker spacing intervals, ranging from \~50 kb \[[@B7],[@B8]\] to 15--20 kb \[[@B9],[@B10]\] to high resolution maps of approximately 1 SNP every kb \[[@B11],[@B12]\]. Patterns of LD across the genome have been shown to be variable and found to be a property of individual chromosomal regions rather than a simple monotonic function of physical distance between markers \[[@B6],[@B9],[@B13]-[@B15]\]. Regions of low haplotype diversity interspersed by regions of low LD (termed haplotype blocks) have been empirically identified and proposed to constitute a ubiquitous feature of the genome \[[@B16]-[@B18]\]. Their presence has triggered funding of the Haplotype Map project, leading to the generation of a genome-wide index of common blocks. Characterisation of haplotype blocks can provide association studies with a shortcut to screening chromosomal regions for the presence of disease variants through the identification of haplotype-tagging SNPs (htSNPs) and can additionally aid in interpreting the results of initial scans through knowledge of the underlying genetic architecture \[[@B5],[@B19]-[@B22]\]. The potential benefits of utilising haplotype blocks may, however, be challenged by concerns regarding their consistency and, hence, applicability to different populations, the information loss incurred by examining common variation and the arbitrary choice of block definition \[[@B23]\]. Different studies investigating the structure of haplotype blocks have used distinct definitions based on various subjective criteria. Block definition methods can be broadly classified into three categories: those based on measures of LD \[[@B16],[@B24]\], those based on haplotype diversity \[[@B11],[@B25],[@B26]\] and those combining both approaches \[[@B9],[@B27]\]. Methods based on LD measures generally define blocks as regions in which all pairwise LD coefficients exceed a subjective threshold. Methods based on haplotype diversity generally define blocks as regions in which a small, arbitrary number of haplotypes accounts for a predefined percentage of the observed variation. The consensus finding is that denser marker maps, larger sample sizes and use of common variants lead to shorter blocks \[[@B23],[@B24]\]. However, the extent of difference in block structure, to which distinct haplotype block definitions and thresholds may result in, remains unclear. The size and number of generated blocks could have an impact on the downstream analysis of association studies and could, therefore, influence the design of fine mapping strategies to identify disease-causing variants. In the present study, we address these issues by applying different haplotype block definition criteria to 137 SNPs, in order to describe the genetic architecture of a 6.1 Mb region of 17q in a set of 189 unrelated healthy individuals. Employing methods based on both LD measures and haplotype diversity, we evaluate their relative merits and limitations, given our median marker spacing of 15.5 kb. Comparing the generated underlying block structures, we assess the usefulness and applicability of distinct methods in genetic association studies of complex human diseases. Methods ======= Subjects -------- DNA from a cohort of 189 healthy, unrelated, UK individuals of Euroepan ancestry was studied. Individuals were recruited from general practice or were blood donors. The collection was approved by the regional ethics committee. Markers and genotyping ---------------------- One hundred and thirty seven SNPs dispersed over 6.1 Mb of the human chromosomal 17q region were examined. We are currently investigating these markers as part of a fine mapping study for the identification of rheumatoid arthritis susceptibility genes. SNPs were selected from the SNP Consortium database \[[@B28]\] to span the region in equally spaced intervals. The SNP map of successfully genotyped markers was constructed based on the November 2002 Freeze of the Human Genome Sequencing Project, available through the UCSC Genome Browser \[[@B29]\]. Methodological details are available upon request from the authors and SNP IDs can be found in Additional file 1 \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Briefly, SNPs were genotyped using either the primer extension SNaPshot™ method (Applied Biosystems, CA, USA) through use of an ABI Prism 3100 DNA Analyzer and GeneScan^®^analysis software (Applied Biosystems, CA, USA), or the allelic discrimination 5\' nuclease assay (TaqMan^®^, Applied Biosystems, CA, USA) through use of an ABI Prism 7700 platform (Applied Biosystems, CA, USA). All SNP genotype calls were independently checked by two individuals. Haplotyping ----------- Departure from Hardy Weinberg equilibrium was initially assessed for each SNP. None of the SNPs were found to deviate from Hardy-Weinberg equilibrium significantly. Haplotypes were then inferred using the expectation-maximisation (EM) algorithm, either through the HelixTree™ (Golden Helix, Inc, Montana, USA) or the snphap (David Clayton, Cambridge, UK) software packages. Convergence of the algorithm was checked by repeating the haplotype estimation process 3 times, ensuring that identical results were generated. Pairwise LD ----------- Using SNP genotypes, the pairwise LD measure of D\' was calculated. As values of D\' can be overestimated with rare allele frequencies \[[@B20]\], the LD correlation coefficient r^2^was additionally calculated for all pairs of SNPs. Observed D\' and r^2^values were sorted according to distance between the corresponding marker pairs. Running average D\' and r^2^values for sliding windows of 2 consecutive observations were estimated and plotted. Haplotype block definitions --------------------------- Haplotype block definitions were applied to the total set of 137 SNPs, as well as to the set of SNPs with allele frequencies exceeding 0.2 separately, in order to assess the effects of variant frequency on block structure. Definition 1: The block definition method based on the D\' measure of LD, employed by Gabriel et al. 2002 \[[@B16]\], was applied to the SNP genotype data through the HaploView software package (MJ Daly and JC Barrett, Whitehead Institute, MA, USA). Briefly, a block was defined as a region in which less than 5% of SNP pairs had a D\' upper confidence bound less than 0.9. In addition, blocks consisting of 2 SNPs could span up to 20 kb and blocks of 3 or 4 SNPs could span up to 30 kb. Blocks were not allowed to overlap. Definition 2: a; A simplified block definition method, also based on LD measures, was used. A haplotype block was defined as a region in which over 95% of all pairwise r^2^LD correlation values exceeded 0.4. The same block length constraints as in Definition 1 were imposed, but a less rigid threshold was employed for stringency evaluation purposes. Blocks were allowed to overlap.b; For the subset of common SNPs, an additional method, based on D\' values that tend to be overestimated for rare allele frequencies, was also employed. A haplotype block was defined as a region in which over 95% of all pairwise D\' values exceeded 0.4. Blocks were allowed to overlap. Definition 3: The block definition method proposed by Wang et al. 2002 \[[@B24]\] was applied to the dataset. Briefly, a block was defined as a region in which, for all possible pairs of markers, less than four gametes were observed (D\' = 1). Blocks were allowed to overlap. Definition 4: A block definition method based on haplotype diversity was developed. For a set of n SNPs, the maximum number of haplotypes observed in the absence of recurrent mutation and / or recombination is n+1. Therefore, a haplotype block was defined as a region consisting of n SNPs, in which n+1 haplotypes could account for at least 95% of the observed variation. Taking each SNP as a seed, blocks were expanded or contracted to find the optimal window. Haplotype blocks were allowed to overlap. Definition 5: A further block definition method based on LD measures, as applied in the HaploView software package (MJ Daly and JC Barrett, Whitehead Institute, MA, USA), was employed. A haplotype block was defined as a region in which all of the pairwise D\' values exceeded 0.8. Blocks were not allowed to overlap. htSNP identification -------------------- The minimum number of SNPs that capture the maximum number of haplotypes (htSNPs) \[[@B30]\] were determined for each resulting block of each definition method. The htSNP2 programme (David Clayton, Cambridge, UK) implemented in Stata and the HaploView software package (MJ Daly and JC Barrett, Whitehead Institute, MA, USA), both making use of the EM algorithm, were employed to identify htSNPs. The r^2^correlation measure, calculating the ability to predict frequencies at a series of loci using just the subset of htSNPs, was set to the stringent threshold of 0.95 for the htSNP2 programme. Good correspondence was observed between the two methods. HtSNPs were additionally identified with a set htSNP2 r^2^threshold of 0.80, in order to check consistency under varying degrees of stringency. Results ======= Minor allele frequencies of the 137 SNPs studied ranged from 0.06 to 0.5 (Figure [1a](#F1){ref-type="fig"}), with an average frequency of 0.29. The observation that 100 (73%) of the SNPs were common (frequencies greater than 0.2) could be explained by ascertainment bias, as all SNPs were selected from publicly available databases \[[@B31]\]. Inter-SNP distances ranged from 55 bp to 951 kb (median spacing 15.5 kb). The marker map contained 4 gaps longer than 200 kb (Figure [1b](#F1){ref-type="fig"}). Plotting the moving average of r^2^exhibited an overall negative correlation between LD and physical distance, with some variability observed for distant SNPs exhibiting evidence for association (Figure [2a](#F2){ref-type="fig"}). The moving average of D\' demonstrated extreme variability in the distribution of LD and its decay with distance, an artefact stemming from low allele frequencies (Figure [2b](#F2){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### a; The distribution of minor allele frequencies for the 137 SNPs used in this study. The bias toward common alleles is inherent to the sampling of markers from publicly available databases. b; The distribution of physical gaps between the 137 SNPs used in this study (median spacing 15.5 kb). ::: ![](1471-2156-6-21-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Running average values of LD measures for sliding windows of 2 SNPs for the 137 markers studied. a; Variability of r^2^. Patterns of decay of LD in this dataset correlate well with observations in different regions of the human genome. b; Variability of D\'. ::: ![](1471-2156-6-21-2) ::: To characterise and compare block patterns, 5 distinct haplotype block definitions were applied to the SNP genotype data. The same sets of parameters reflecting on underlying block structure were determined for each method (Table [1](#T1){ref-type="table"}). Therefore, in order to gain an understanding of how each definition portrayed the region\'s genetic architecture, the number of resulting haplotype blocks, the average length and SNP content of blocks, as well as the proportion of sequence and markers covered by blocks were evaluated. The process was repeated for the subset of common SNPs only (Table [2](#T2){ref-type="table"}). In general, methods based on LD measures (Definitions 1, 2, 3 and 5) resulted in fewer, shorter blocks, while the haplotype diversity-based method (Definition 4) provided an overall greater coverage of the region (Figure [3](#F3){ref-type="fig"}). Although map density was sparser in the group of common SNPs (median spacing of 1 SNP / 29.5 kb), the inclusion of markers with minor allele frequencies less than 0.2 appeared to have an overwhelming effect for the majority of definitions and generally resulted in reduced coverage of the sequence examined. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Haplotype block characteristics according to different definition methods, applied to the total group of 137 SNPs. ::: Definition^a^ n blocks Average block length (kb) Average n SNPs/block \% of sequence covered n SNPs in blocks (%) --------------- ---------- --------------------------- ---------------------- ------------------------ ---------------------- Definition 1 20 28.3 2.6 9.3 52 (38) Definition 2 19 16.5 2.6 5 46 (33.6) Definition 3 32 24.2 2.1 12.7 62 (45.3) Definition 4 60 130.7 4.3 85.8 130 (95) Definition 5 38 42.0 2.9 26.2 111 (81) ^a^Definition 1 \[16\], Definition 2 (modification of \[16\]), Definition 3 \[24\] and Definition 5 (D\' high threshold method) were based on measures of LD, whereas Definition 4 (n+1 method) was based on haplotype diversity. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Haplotype block characteristics according to different definition methods, applied to the subset of 100 common SNPs (minor allele frequency \>0.2). ::: Definition^a^ n blocks Average block length (kb) Average n SNPs/block \% of sequence covered n SNPs in blocks --------------- ---------- --------------------------- ---------------------- ------------------------ ------------------ Definition 1 17 31.1 2.6 8.7 44 Definition 2a 14 10.9 2.6 2.4 32 Definition 2b 21 10.8 2.5 3.3 53 Definition 3 19 18.4 2 5.7 33 Definition 4 39 107.0 3.8 55.4 85 Definition 5 27 41.1 2.7 18.2 73 ^a^Definition 1 \[16\], Definitions 2a and 2b (modifications of \[16\]), Definition 3 \[24\] and Definition 5 (D\' high threshold method) were based on measures of LD, whereas Definition 4 (n+1 method) was based on haplotype diversity. ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Snapshot of haplotype block organisation on 17q. Blocks identified by each of the 5 Definitions for the first 3 Mb of the region are depicted. SNPs are shown as triangles according to their relative spacing. Genes in the region are shown in pink (circles denote the start and end points of genes). Haplotype blocks are colour-coded according to the Definition used to characterise them: Definition 1 \[16\] is in purple; Definition 2 (modification of \[16\]) is in orange; Definition 3 \[24\] is in red; Definition 4 (n+1 method) is in blue; Definition 5 (D\' high threshold method) is in green. Squares represent the SNPs that fall within the defined blocks and lines extend across each haplotype block. Adjacent and overlapping blocks are depicted in consecutive rows. ::: ![](1471-2156-6-21-3) ::: Figure [4](#F4){ref-type="fig"} depicts the total number of markers that were necessary to capture most variation in this chromosomal region, as derived from each definition, for the total group of SNPs and for the subset of common SNPs. To calculate this parameter, the number of htSNPs identified for each haplotype block was added to the number of SNPs that were not encompassed within blocks. Genotyping of a similar proportion of markers appeared to be necessary across the different definitions for all markers (90.5% to 96.4%) and for the common SNPs (88% to 97%) at the stringent r^2^correlation measure threshold of 0.95. The observation that the vast majority of SNPs needed to be typed in order to capture most of the chromosomal variation was confirmed when the r^2^threshold was decreased to 0.80. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Number of SNPs that need to be genotyped, in order to capture the majority of variation in the region, according to the different haplotype block definition methods. a; In the total group of 137 SNPs. b; In the subset of 100 common SNPs (minor allele frequency \>0.2). The number of htSNPs falling within haplotype blocks is denoted by black, while the number of SNPs that need to be typed but are not included in blocks, is depicted in grey. ::: ![](1471-2156-6-21-4) ::: Discussion ========== Recently, numerous groups have studied the presence and distribution of haplotype blocks in the human genome, each proposing and utilising distinct block definition methods. Each study has examined different numbers of SNPs, dispersed throughout differently sized chromosomal regions, at varying minor allele frequency and map spacing, making use of diverse sample sizes \[[@B9],[@B11],[@B16],[@B25],[@B27]\]. The underlying design of this study reflects a realistic scenario, in which a region of several Mbs has been implicated in susceptibility to a human complex disease and is being refined through LD mapping. The ascertainment of SNPs through publicly available databases additionally represents practically favoured selection processes, giving rise to a well-recognised bias toward common polymorphisms and leading to, in this case, a median marker spacing of 15.5 kb (equivalent to that used by Dawson et al. 2002 \[[@B9]\]). This SNP density would be expected to give rise to apparently longer haplotype blocks compared to denser maps, such as the HapMap. The selection of unrelated individuals is in keeping with the current trend toward population-based, rather than family-based, studies and the pragmatic sample size of 378 chromosomes allows effective in silicohaplotype inference. The extent and variability of inter-marker LD in this study corroborates recent findings of distribution irregularity \[[@B6],[@B9],[@B13]-[@B15]\]. The observation that LD does not decay uniformly with physical distance exemplifies the need for haplotype block structure determination. Evidence for disequilibrium has been detected between SNPs as far as 1.7 Mb apart, extending much further than previously indicated through simulation \[[@B32]\]. The discrepant LD patterns derived from distinct LD measures highlight the need for caution in interpreting and comparing studies, especially when using D\', for which there is an upward bias with small sample sizes and rare allele frequencies. Haplotype block definition methods employed in this study have been based both on measures of LD and on haplotype diversity, and applied to the same dataset, thus enabling direct comparison of their performance. Definitions 1 \[[@B16]\] and 3 \[[@B24]\] have been proposed in recent studies of block structure in human genomic regions. Definition 2, based on measures of LD, was developed as a simplified modification of Definition 1 \[[@B16]\] to accommodate less stringent thresholds and criteria. Definition 5 was used to reflect block structure based on criteria setting a high threshold of D\', but no length constraints. In addition, a novel diversity-based method was developed (Definition 4), which does not impose strict block boundaries and incorporates the notion of recombination events and recurrent mutation, factors known to diminish inter-marker LD. All methods provided evidence for a block-like organisation of the genetic variation in the chromosomal region under investigation on 17q, characterised by marked differences among the distinct definitions, in accordance with previous observations \[[@B33],[@B34]\]. Overall, the haplotype diversity-based method (Definition 4) gave a more comprehensive coverage of the sequence, resulting in a higher number of blocks with a longer average physical size, compared to definitions based on measures of LD. These findings are in agreement with a recent study, comparing the performance of one LD-based with one haplotype-based block definition \[[@B34]\]. Such differences in characterising the underlying genetic architecture of a region could have implications in the interpretation of association studies and the design of subsequent strategies. Inclusion of a greater proportion of the region into blocks maximises the chances that a significant association observed through a first scan will be encompassed within a haplotype block, thus delineating the interval on which further fine mapping attempts can be focused. Localisation of a positive result outside the boundaries of defined blocks would necessitate more intensive genotyping efforts targeted to the surrounding region. Although extended coverage of a sequence interval may prove useful, it could be artificial, stemming from methodological inadequacies, thereby leading to a false representation of the underlying genomic structure. Although the newly developed Definition 4 (n+1 method) resulted in higher sequence coverage, the lack of any LD constraints in this definition could lead to a falsely inflated detection of short haplotype blocks in cases of SNPs with rare minor allele frequencies. Of the LD-based methods, Definition 5 (D\' high threshold method) provided the highest coverage of the sequence studied, although approximately 74% of the region did not fall into blocks. The observed inconsistency among methods illustrates the subjectivity of haplotype block definition and prevents the conclusive characterisation of the region\'s block structure. Haplotype block assignment was found to change not only due to inter-method differences, but also as a result of altering parameters within the same set of definition criteria. These observations corroborate previous findings \[[@B34]\]. Application of the same methods to the subset of common SNPs led to an overall reduction in the proportion of the sequence covered. Although marker spacing was sparser and the number of polymorphisms examined smaller, high minor allele frequency had an overwhelming effect on haplotype block size, generally resulting in shorter blocks. Definition 1 \[[@B16]\], however, appeared to be robust to such changes, therefore offering a possible mechanism to achieve consistency in block structure between SNP subgroups of varying allele frequencies. Among LD-based definitions, use of D\' rather than r^2^resulted in the generation of more haplotype blocks and in an increased coverage of the region. Similarly to when applied to the total set of markers, the newly introduced Definition 4 (n+1 method) produced the highest sequence coverage when examined in the subset of common SNPs only, although a proportion of relatively infrequent SNP pairs in low LD could have been falsely categorised as blocks. Comparison of different haplotype block definitions in characterising the genomic organisation of the human chromosomal region 17q revealed discrepancies among methods and could, therefore, raise concerns about both the suitability of ad hoc approaches for the crude identification of block structure, as well as the validity of the notion of haplotype blocks as a genomic feature. However, the observed overlap in SNPs encompassed within blocks across all definitions used, indicated an underlying genetic architecture captured by all methods. Concordance among all definitions was additionally exhibited in calculating the subset of SNPs necessary to encapsulate the vast majority of genetic variation in the region. Selection of block definition method appeared to be irrelevant when genotyping a sample subset for all markers in order to identify haplotype tagging SNPs. In this study, the percentage of markers that needed to be typed was extremely high (over 90%), indicating that the marker map density employed was not suited to achieving significant cost-effectiveness through htSNP characterisation. Reassuringly, as all methods suggested typing the same number of markers, they probably also carry equal chances of detecting a possible association due to LD. The differences, however, would arise in interpreting downstream results and developing follow-up strategies. The proposal of taking advantage of haplotype blocks to inform strategic designs in genetic association studies constitutes a welcome step forward, rather than a panacea, for the field of human complex disease genetics. In a realistic study design, the choice of block definition method could be of consequence in designing and interpreting genetic association scans. In addition, the inclusion of SNPs with rare minor allele frequencies appears to convolute, rather than clarify, the underlying genomic structure. Given the marker density of 15.5 kb, a whole genome scan by association would require approximately 100,000 SNPs to be genotyped. The findings of this study indicate that such a spacing would not be adequate for characterising the genomic architecture in sufficient detail through a haplotype block definition approach. Further issues inherent to the characterisation and utilisation of chromosomal underlying block structure need to be addressed in both real and simulated datasets, in order to clarify the settings in which haplotype blocks may prove useful. Authors\' contributions ======================= EZ participated in study design, carried out the statistical analyses and drafted the manuscript. AB participated in study design and coordinated genotyping efforts. SE and DW carried out the SNP genotyping. JW and WO participated in study design. SJ participated in study design and coordination and helped draft the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 SNP IDs and chromosomal locations at the time of SNP selection. Short description of the data: Additional file 1 contains a full list of SNPs used for the analyses. The SNPs are identified by their rs number. Where rs numbers are not available, the ABI assay-on-demand ID has instead been included. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We are grateful to Golden Helix Inc, for making the HelixTree™ software package available to us and to Neil Shephard for his valuable advice. This work was supported by funding from the MRC and arc, UK.	PubMed Central	2024-06-05T03:55:56.509551	2005-4-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090572/", "journal": "BMC Genet. 2005 Apr 25; 6:21", "authors": [ { "first": "Eleftheria", "last": "Zeggini" }, { "first": "Anne", "last": "Barton" }, { "first": "Stephen", "last": "Eyre" }, { "first": "Daniel", "last": "Ward" }, { "first": "William", "last": "Ollier" }, { "first": "Jane", "last": "Worthington" }, { "first": "Sally", "last": "John" } ] }
PMC1090573	Background ========== Genome research in agriculturally important species is facilitated by the availability of species-specific molecular genetic tools and resources such as chromosome maps and large volumes of sequence data. Recently such resources have been developed for important aquaculture species including rainbow trout, which are also widely used as a model system for carcinogenesis, toxicological, and comparative immunological research \[[@B1]\]. Several genetic maps \[[@B2]-[@B4]\] consisting primarily of type II markers \[[@B5]\] (amplified fragment length polymorphism simple sequence repeats) have been utilized in the identification of qualitative and quantitative trait loci (QTL) \[[@B6]\] associated with rainbow trout production traits. This includes QTL for natural killer cell-like activity, temperature tolerance, spawning date, body weight, resistance to infectious pancreatic necrosis virus (IPNV), resistance to infectious hematopoietic necrosis virus (IHNV), embryonic development rate, and albinism \[[@B7]-[@B19]\]. Although the genetic improvement of these traits through selective breeding would benefit the aquaculture industry, these QTL span large chromosomal intervals and will not be practical for marker assisted selection \[[@B20]\] without additional mapping. The current rainbow trout genetic maps lack marker densities and comparative information necessary to conduct fine mapping aimed at reducing QTL interval sizes, developing practical marker assisted selection schemes, and selection of comparative positional candidates \[[@B21]\] to specifically identify the gene(s) affecting traits of interest. The recent evolutionary divergence of the salmonids \[[@B22]\] and the importance of many of these species to aquaculture will allow for comparative QTL mapping. For example, the development of genetic linkage maps for Atlantic salmon and Arctic char \[[@B23]-[@B25]\] has enabled the identification of QTLs for growth characteristics, disease resistance, and temperature tolerance in those species \[[@B18],[@B26],[@B27]\]. The development of microsatellites markers from EST sequences will facilitate the use of genome information in salmonids species by 1) increasing Type II \[[@B5]\] marker densities on genetic maps; 2) integrating physical and genetic maps; 3) developing comparative genetic maps among salmonids; and 4) developing comparative maps with aquatic model organisms such as zebrafish, fugu, and tetraodon and with better studied avian and mammalian species. This comparative information will aid in the identification of positional candidate genes \[[@B28]\] for production traits in salmonid aquaculture and for basic research which utilizes rainbow trout as model organism. An expressed sequence tag (EST) \[[@B29]\] project was initiated for rainbow trout with the following aims: 1) identify as many unique transcribed sequences as possible; 2) annotate sequence data with information from other species; 3) develop functional genome tools for rainbow trout; and 4) identify microsatellite and single nucleotide polymorphism (SNP) genetic markers for the construction of high-density chromosome maps \[[@B30]\]. Sequences from a normalized cDNA library (NCCCWA 1RT) constructed from brain, gill, liver, muscle, kidney, and spleen tissue resulted in the creation of the Rainbow Trout Gene Index (RTGI) \[[@B31]\]. Microsatellite marker development was conducted simultaneously with the sequencing phase of the project through hybridization of (GT)~11~and (GA)~11~probes to high-density filters representing 27,648 clones from the library. Positive clones were selected for further analyses resulting in 89 polymorphic microsatellite markers derived from ESTs, 30 which were informative in mapping reference families, 55 were associated with functional annotation, and 20 for which comparative mapping assignments were determined. Results ======= Marker development ------------------ Hybridization of high-density filters representing 27,648 cDNA clones from a normalized cDNA library with (GA)~11~and (GT)~11~oligonucleotide probes identified 415 clones potentially containing microsatellite repeats. Forward and reverse sequencing for 384 of these clones resulted in 755 sequences of good quality (PHRED score \> 20 over 100 bp \[[@B32]\]). Dinucleotide microsatellite repeat were identified from 181 clone sequences. Analysis of redundancy identified 161 unique sequences. PCR primer design was possible for 128 of the 161 sequences which were assigned locus names using OMM5000 nomenclature (in-house terminology for microsatellite markers derived from ESTs). PCR optimization was successful for 93 of the 128 primer pairs. Testing for polymorphism in three reference parents and five doubled haploids resulted in the development of 89 polymorphic microsatellites markers with an average of 4.52 alleles (range 2--7), 40% of which were duplicated as determined by the observance of multiple alleles in clonal lines (see Additional File [1](#S1){ref-type="supplementary-material"}). Cross-amplification in other salmonid species using PCR conditions that were optimized for rainbow trout was determined (Table [1](#T1){ref-type="table"}) to be similar to markers from previous publications \[[@B33]\]. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Cross-species amplification. Cross-species amplification allele size range information (bp) for microsatellite markers generated from rainbow trout ESTs ::: Locus Artic Char Brook Trout Atlantic Salmon Brown Trout Chinook Salmon Coho Salmon Sockeye Salmon Cutthroat Trout ----------- ---------------- ----------------- --------------------- ----------------- -------------------- ----------------- -------------------- --------------------- OMM5000 253--255 260--294 256 239--254 250--260 256 251--263 240--252 OMM5001 87 89 87 85 97 89 97--109 123--151 OMM5002 142--155 \- 165--265 336--346 274--282 139--151 139--151 277 OMM5003 177 176--179 173 179--183 179 168--184 173--186 176--181 OMM5004 187 189 185--187 187 187 150--164 193 189--193 OMM5005 201--203 195--202 219 196--207 186--187 189 191 200--204 OMM5006 \- 225--231 \- 205--245 203 207--209 201 211--236 OMM5007 162--167 182--187 181--192 166 170--176 180--199 156--170 147--163 OMM5008 259 247 223--255 253--261 246--95 227--252 232 236--254 OMM5009 247 335--363 \- 266 279--283 410 \- 303--331 OMM5010 288--334 \- 346--370 350--358 305--332 341--346 294--297 362 OMM5011 214--248 217--248 213--248 214--233 224--244 227--247 228--249 224--246 OMM5012 170--188 202--208 186 199--201 174--184 175--190 196--223 169--187 OMM5013 \- 107 98 98--111 111--135 96--220 126 102--187 OMM5014 \- \- \- 201--208 185--202 181--198 \- 230--262 OMM5015 228 228 228 228 228 228 228 228 OMM5016 239 \- 239 239 198 249--311 189 231--236 OMM5017 203--234 195--200 188--200 208--256 217--237 190--201 193 184--209 OMM5018 192 215--231 182--199 198 184--192 \- \- 186--215 OMM5019 298 256--268 269--335 298--321 269--279 272--275 \- 272--282 OMM5020 262 262 262--275 262 256--259 255--258 256--261 261--263 OMM5023 131 131 122 122 130--136 122 122 126--142 OMM5024 198 202 170 194--195 209--230 \- \- 212--214 OMM5025 154 152 160 160 186--188 158 164 160 OMM5029 210--214 209--227 192--211 193--215 208 207--213 210 203--228 OMM5030 135--187 135--137 140--187 137--139 129--141 150--164 129--141 141--155 OMM5031 145 143 102--144 129--144 142--155 136 142--144 140 OMM5032 198 \- 178 216--218 191--201 159 175--179 179--187 OMM5033 284 \- 274--280 225 \- 243 254--279 260--295 OMM5034 239--287 264 238--240 238--263 239--275 236--269 236--267 246--269 OMM5037 \- 266 262--268 260--293 251 254--281 264--272 260--285 OMM5039 284 280 274--280 268 248--286 243--308 282--286 260 OMM5041 168 185--187 172 170 170--173 183--187 132 132--189 OMM5042 133 140 \- 127--137 122 \- \- \- OMM5043 122 122 126 114--128 -- 112--114 112--122 112--131 OMM5044 206 199 \- 230 242 \- \- 199--217 OMM5047 \- \- 317--328 245--251 257--258 261 190 257--263 OMM5050 242 247--249 251 240 243 245--247 246--248 251--261 OMM5051 174--192 190 179 201--218 212--214 190--191 \- 195--206 OMM5053 134--198 \- \- \- 122 248 202 228--237 OMM5054 172--240 171--265 161--240 172--241 171--240 171--265 172--241 162--278 OMM5055 217--219 190--220 190--212 190--212 221 225--243 221 219 OMM5056 254--280 268--299 \- 196 186 282--319 \- 218--253 OMM5058 \- \- 216--219 239--244 198--204 192 231--235 194--209 OMM5059 151 \- 157 145--172 134 124--135 121--127 126--136 OMM5060 105--164 165 \- 160 164 164 164 105--164 OMM5061 400 354--358 291--293 274--282 275 274 262 274--278 OMM5062 229 225 200 192--212 223--233 212--231 221--241 222--240 OMM5063 \- 148--195 207--243 203--237 150--154 237--282 154--182 178--220 OMM5064 272--274 92--110 276--283 274 290--319 283--285 279 95--286 OMM5067 153--186 153--164 185--187 171--185 153--186 153--164 171--192 153--187 OMM5072 160--164 161--167 171 170--187 146--158 158--164 158--167 164--170 OMM5074 306--344 247--260 242 238--244 245--253 232 228--233 241--244 OMM5075 214 206--208 \- 189--191 186 192--198 194 208--229 OMM5077 368 372 377 383 373 377--380 365 334--354 OMM5088 174 168 \- 153 159--161 153 168 159--174 OMM5089 \- \- 134--167 \- 154--161 132--140 134--146 \- OMM5090 153 152--255 153 249--255 153--269 249--255 153--255 239--248 OMM5091 276 201--210 178 201--205 221--262 368 223--244 265--283 OMM5092 161 161 186 186 202--208 192--217 \- \- OMM5093 285 285 285 285 285 285 285 285 OMM5099 244 243 228 219--234 278--296 260--268 214--254 213--260 OMM5100 137--185 \- 182 138--143 173 \- 160--173 167--201 OMM5106 358--388 328--353 260--271 261 306--322 361--395 257--274 273--318 OMM5107 258--264 255 250--254 264 \- 254 255 \- OMM5108 265 251 265--271 256 263--292 260--262 251--271 256--267 OMM5109 256 254--256 256--271 260--263 256 254--256 260--262 262--271 OMM5112 194 198 194--218 196 193--202 189--196 189 193--206 OMM5113 \- 320--368 \- 274--320 \- 286--304 \- 243--288 OMM5117 135--142 138 142 125--140 138 125--140 138 137--154 OMM5121 230 228--230 156--173 166--176 166--267 178--230 173--175 197--222 OMM5124 271--281 271--277 \- 272--273 258 266 269 280 OMM5125 256 262--264 254 250--252 256--277 250--260 256--260 256--260 OMM5126 295--299 286 295--299 286--290 286 286--291 286 286--307 \% Amp. 87 83 83 97 94 93 85 94 ::: Functional annotation --------------------- Functional annotations were associated with ESTs by BLAST analyses of the RTGI which previously included EST sequence data for the clones described in this manuscript. The highest scoring matches all had E-values ranging from 0 to 10^-40^and percent identities ranging from 91--100 % (see Additional File [2](#S2){ref-type="supplementary-material"}). TIGR gene index annotation for tentative consensus sequences (TCs) includes three levels of significance based on percent identity: matches in the range of 90 to 100% are categorized as \"homologues,\" matches in the range of 70--90% are categorized as \"similar,\" and matches less than 70% are categorized as \"weakly similar.\" Annotation of ESTs in this manuscript resulted in 10 highly significant matches to genome sequences, 8 categorized as homologues, 28 as similar, 9 as weakly similar, and 41 for which no associations were determined Locus or gene symbols from Locus Link \[[@B34]\] or UniProt \[[@B35]\] were added to 8 loci designated as homologues. Genetic and comparative mapping ------------------------------- Linkage analyses of 33 informative markers resulted in the assignment of 30 markers to linkage groups (see Additional File [3](#S3){ref-type="supplementary-material"}). Twenty-three markers were informative in the reference families of Sakamoto et al. \[[@B3]\] and 7 markers were placed on the map of Nichols et al. \[[@B2]\] in addition to 3 which were not included into previous linkage groups (Table [2](#T2){ref-type="table"}). Comparisons to zebrafish and fugu databases identified homologous assignments for 16 ESTs each (see Additional File [4](#S4){ref-type="supplementary-material"} and Additional File [5](#S5){ref-type="supplementary-material"}), however, the chromosomal assignments in these 2 species are not yet available. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Identification of homologous segments between rainbow trout, human, mouse and tetraodaon chromosomes. Rainbow trout linkage group nomenclature is from Nichols et al. (2003a) ::: Locus Rainbow Trout Linkage Group Tetraodon (TNI) Human (HSA) Mouse (MMU) ----------- --------------------------------- --------------------- ----------------- ----------------- OMM5000 27 8 19 7 OMM5002 21 6 10 OMM5003 23 OMM5005 11 2 13 14 OMM5012 23 OMM5017 20 3 OMM5019 9 17 11 OMM5023 22 OMM5025 8 OMM5026 29 OMM5029 12 OMM5033 16 OMM5034 19 8 OMM5041 12 10 3 3 OMM5045 19 12 12 16 OMM5051 ? 2 OMM5056 ? 10 14 OMM5057 9 OMM5059 13 5 OMM5062 27 OMM5065 25 OMM5077 25 X OMM5088 19 OMM5090 21 OMM5093 ? 4 5 OMM5099 7 6 8 15 OMM5100 15 19 OMM5106 14 OMM5107 22 9 OMM5108 20 OMM5109 31 OMM5112 23 OMM5113 6 10 OMM5117 10 14 OMM5121 31 6 OMM5126 21 OMM5127 9 16 7 ::: Discussion ========== Microsatellite marker development --------------------------------- Marker development strategies for the construction of high-density genetic maps typically utilize random or targeted approaches. Random approaches are commonly employed in the early phases of the map construction and are characterized by the use of sequence data not associated with mapping or functional annotation for marker development. In targeted approaches, commonly employed to increase marker density in a specific chromosome region or to map genes of interest, only sequence data meeting specified parameters with respect to mapping or function are utilized for marker development. Our approach for increasing the marker densities of rainbow trout genetic maps was a hybrid of random and targeted approaches. Although clones for marker development were not chosen based on functional annotation, the sequence data utilized were known to be transcribed. The benefit of this approach is that these microsatellites are Type I and II markers \[[@B5]\], serving to increase marker densities on both genetic and comparative maps. Similar strategies have been employed in the development of microsatellite markers for other agriculturally important animals including sheep, turkey, cattle, catfish, and pig \[[@B36]-[@B40]\]. Cross amplification within the salmonidae ----------------------------------------- Salmonids are believed to have diverged from a common tetraploid ancestor some 25 million years ago \[[@B22]\]. As a result of this evolutionarily recent divergence, microsatellite markers can be used in the development of comparative genetic maps among the salmonidae. Cross-species amplification was obtained for 74 markers and ranged between 83% and 97% per species, with observed polymorphism that ranged between 36% and 82% per marker. Sampling additional individuals from multiple populations is likely to increase observations of polymorphism. This high level of cross-amplification and polymorphism should facilitate the development of comparative and genetic maps for the salmonids. Functional annotation --------------------- The RTGI was used to associate ESTs with functional annotation as their sequence data was previously included in RTGI Version 4.0. Unfortunately, 42% of the markers were not associated with any annotation, demonstrating an overall lack of functional annotation of the rainbow trout transcriptome. Genetic and comparative mapping ------------------------------- The goal of the activities outlined in this manuscript was to identify homologous regions of chromosomes between rainbow trout and species for which there is an abundance of genome information including whole genome sequence. Eight regions of homology were identified between trout and tetraodon, seven with human, and 10 with mouse (Table [2](#T2){ref-type="table"}). Although mapping single loci does not identify segments of conserved synteny, the homologies reported in this paper are supported by the examination of direct comparative information between tetraodon and human and mouse. For instance, OMM5000 was observed to be homologous with TNI 8, HSA19, and MMU7. The NCBI human/mouse comparative map \[[@B41]\] reveals a homologous region between HSA 19 and MMU 7, and the tetraodon comparative map \[[@B42],[@B43]\] reveals regions of homology between TNI8 and both HSA19 and MMU7. Similar analyses of comparative assignments in two or more species supported our findings for every marker reported. Conclusion ========== This project was initiated at a time where very little sequence data was publicly available for salmonid species. Now the RTGI contains over 150,000 ESTs which represent \~ 50,000 unique sequences. Current methods to develop new microsatellite markers from EST sequences would most likely replace hybridization with an in silicostrategy on the RTGI data set. Therefore, the continuation of microsatellite marker development from expressed sequence tag data is feasible and will be useful for developing comparative maps with other salmonids and with better studied species. Methods ======= Identification of cDNA clones with microsatellites -------------------------------------------------- A rainbow trout normalized cDNA library was constructed using mRNA from brain, gill, liver, spleen, kidney, and muscle tissues. The library was plated, picked, and arrayed into 384-well plates. Sets of 72 plates were gridded onto single 20 cm^2^positively charged nylon membranes for hybridization experiments. One high-density membrane (representing 27,648 clones) was hybridized overnight at 65°C with radioactively (^32^P) labeled (GA)~11~and (GT)~11~oligonucleotide probes using standard protocols \[[@B44]\]. Membranes were removed from hybridization solution, washed, and exposed to storage phosphor screens for 1 hour. The phosphor screens were scanned on a Storm (Amersham Biosciences Corp, Piscataway, NJ) and positive clones identified. Sequencing and primer design ---------------------------- Positive clones were re-arrayed into 96-well plates and grown overnight. DNA was isolated for each clone using manufacturer\'s standard miniprep protocols for the BioRobot 8000 (QIAGEN, Valencia, CA, USA). Sequencing reactions were carried out using ABI Dye Terminator Chemistry (Applied Biosystems, Foster City, CA, USA) using SP6 and T7 primers. Sequencing reactions were purified and electrophoresed on an ABI3700. Sequences were trimmed for quality and vector using PHRED and Cross\_match \[[@B32]\]. Consensus sequences were constructed for clones having multiple sequence data files. Those containing microsatellites were analyzed for redundancy within the dataset and previously discovered salmonid microsatellites using Vector NTI Suite 6.0 (InforMax, Bethesda, MD). PCR primer pairs were designed to amplify unique microsatellite sequences using Oligo 6.0 \[[@B45]\]. PCR and genotyping ------------------ PCR primer pairs were obtained from commercial sources with the forward primers labeled with FAM, HEX, or NED. Primer pairs were optimized by varying annealing temperatures and MgCl~2~concentrations to amplify in rainbow trout (Kamloop strain), the clone of origin, and a negative control with no DNA. Reactions (11 μl total volume) included 25 ng DNA, 1.5--2.5 mM MgCl~2~, 2.0 μM of each primer, 200 μM dNTPs, 1 × manufacturer\'s reaction buffer, and 0.5 unit AmpliTaq Gold Polymerase (ABI, Foster City, CA). Amplifications were conducted in an MJ Research PTC 200 DNA Engine thermal cycler (MJ Research, Waltham, MA) as follows: an initial denaturation at 94°C for 10 min, 36 cycles consisting of 94°C for 30 s, annealing temperature for 30 s, 72°C extension for 30 s; followed by a final extension of 72°C for 10 min. Successfully optimized primer pairs were used to amplify DNA from the three reference family parents \[[@B3]\] and five doubled haploid clonal lines (OSU, Arlee, Swanson, Hot Creek, and Clearwater \[[@B46]\]). Cross-species amplifications were attempted in two samples representing various other salmonids including cutthroat, Sockeye, Kokanee, Chinook, Atlantic salmon, brown trout, brook trout, and Artic char. PCR products were electrophoresed and verified by visualization in 3% agarose gels. PCR reactions were then combined according to label and size. Typical combinations of markers for capillary electrophoresis were made by combining PCR reactions for markers having alleles of at least 100 bp (based on agarose results) difference in size and different fluorescent labels. One microliter of each PCR product was added to 20 microliters of water, of which one microliter was added to 12 microliters of HiDi formamide and 0.5 microliters of ROX standard for genotyping for electrophoresis on an ABI PRISM3700 DNA Analyzer or an ABI PRISM 3100 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). Genescan output files were analyzed using Genotyper 3.5 software (Applied Biosystems, Foster City, CA, USA). Markers for which the parents of the reference families were informative were genotyped on the offspring. Markers not informative on the Sakamoto et al. \[[@B3]\] map having been associated with mapping annotation were genotyped on the reference families of Nichols et al. \[[@B2]\]. Annotation ---------- A FASTA file was generated containing clone sequence data for use in standalone BLAST with the goal of obtaining functional and mapping annotation. Functional annotation was associated by comparison to the RTGI Version 4.0 (Appendix 2) \[[@B31]\]. Mapping annotation was obtained by comparisons to sequence data from the Tetraodon Genome Browser \[[@B47]\] and zebrafish, fugu, human and mouse genome sequences from NCBI (Appendices 4 and 5) \[[@B48]\]. Authors\' contributions ======================= CER conceived the study and participated in its design and coordination and drafted the manuscript. MFR, IC and YP assisted in genotyping analyses, RGD and KG conducted linkage analysis on the Sakamoto et al. \[[@B3]\] reference families and JD and RP conducted mapping in the doubled haploid crosses \[[@B2]\]. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Appendix 1. Microsatellite marker information including GenBank accessions, duplication status, allele size ranges, repeat motif, primer sequences, and optimized PCR conditions. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Appendix 2. Functional Annotation. Tentative annotation assigned to marker ESTs acquired via BLAST of the rainbow trout gene index version 4.0. Markers identified as homologues are annotated with gene or locus name symbols from UniProt \[[@B35]\] or NCBI. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Appendix 3. Mapping information for rainbow trout microsatellites. Each marker which was informative for mapping is included with cross, closest marker locus name, linkage group, and map position. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 Appendix 4. In silicoderived comparative mapping information I. BLAST was used to identify similar sequences between mouse, human, zebrafish, and pufferfish. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Appendix 5. In silicoderived comparative mapping information II. BLAST was used to identify similar sequences with tetraodon. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The authors thank Roseanna Athey, Renee Fincham, Ashley Gustafson, and Connie Briggs for their technical contributions to this manuscript and Krista Nichols and Robert Drew for their assistance in linkage analyses with doubled haploid crosses. Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S Department of Agriculture.	PubMed Central	2024-06-05T03:55:56.512206	2005-4-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090573/", "journal": "BMC Genomics. 2005 Apr 18; 6:54", "authors": [ { "first": "Caird E", "last": "Rexroad" }, { "first": "Maria F", "last": "Rodriguez" }, { "first": "Issa", "last": "Coulibaly" }, { "first": "Karim", "last": "Gharbi" }, { "first": "Roy G", "last": "Danzmann" }, { "first": "Jenefer", "last": "DeKoning" }, { "first": "Ruth", "last": "Phillips" }, { "first": "Yniv", "last": "Palti" } ] }
PMC1090574	Background ========== DNA microarrays are a powerful tool for global analysis of gene transcript expression. The initial studies using arrays required large amounts of starting material in order to reliably detect sample signals. Since that time, improvements in sample preparation, amplification and labeling methods \[[@B1]-[@B5]\] have reduced the starting material requirement to \~1--5 μg of total RNA \[[@B6]\]. Efforts to use smaller amounts of starting material have focused on PCR \[[@B7],[@B8]\] and multiple rounds of T7 RNA polymerase in vitrotranscription \[IVT\] \[[@B9]-[@B12]\]. PCR based methods have been successfully used to amplify subnanogram quantities of RNA from as little as a single cell \[[@B13],[@B14]\], but these approaches have not been widely adopted. Most attempts to perform arrays using submicrogram amounts of RNA have relied on 2 or 3 rounds of linear amplification using IVT, but this approach has proven to be time consuming and technically demanding. In our hands, two round IVT is necessary to prepare samples from 5--50 ng total RNA and the amplification typically takes 4--5 days to complete. Others have reported a 10% decrease in sensitivity in detection of differentially expressed genes with the addition of a second IVT round \[[@B15]\]. A new single primer, isothermal linear amplification method (ribo-SPIA) has been specifically developed for amplification of very small samples for use on DNA microarrays \[[@B16],[@B17]\]. With this method (Figure [1](#F1){ref-type="fig"}), small amounts of total RNA are reverse transcribed into cDNA using a chimeric RNA/DNA primer containing oligo(dT) and a unique RNA sequence tag at the 5\' end. Linear amplification requires the addition of RNase H, DNA polymerase and excess chimeric primer. The RNase H digests RNA from RNA/DNA hybrids thus exposing a single stranded binding site where a new copy of the primer anneals and the DNA polymerase initiates synthesis of a fresh copy of cDNA, displacing the original antisense strand of the cDNA. A single isothermal linear amplification reaction rapidly generates sufficient single-stranded cDNA (sscDNA) for multiple hybridization reactions. sscDNA samples are fragmented to provide sscDNA fragments of \~50--200 bp, end labeled with biotin and used for microarray hybridization. Approximately 100,000-fold amplification is typical for a single amplification step. The ribo-SPIA method is potentially attractive because the amplification can be completed in a single day and there are no purification steps until after completion of amplification, thus reducing the risk of losing sample during handling. In this study, we have investigated the utility of ribo-SPIA generated sscDNA for DNA microarray analysis of small starting samples. We assessed yield and reproducibility of sscDNA, and compared sscDNA-based microarray results with microarray results obtained using IVT amplification and with results of quantitative PCR in order to assess if this method conferred new advantages for use with DNA microarrays. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Diagram of the ribo-SPIA process for synthesis of sscDNA. ::: ![](1471-2164-6-57-1) ::: Results and discussion ====================== Yield of sscDNA --------------- We used the ribo-SPIA method to amplify several different total RNA samples. RNA was obtained from a pool of human tissues (Clontech Universal Human Reference RNA, or cUHR, Experiment 1), mouse liver (Experiment 2), a second pool of human cells (Stratagene Universal Human Reference RNA, or sUHR, Experiment 3), and K562 human erythroleukemia cells (also Experiment 3). The amount of starting total RNA ranged from 5--100 ng and yields were in the range of \~6--12 μg of sscDNA (Table [I](#T1){ref-type="table"}). The somewhat lower yields seen in Experiment 1 are likely attributable to our unfamiliarity with the protocol, and yields improved in subsequent experiments. There was no clear relationship between the amount of starting RNA and the sscDNA yield. To determine how much sscDNA product was produced in the absence of any template, we performed two additional amplifications with no input RNA (Experiment 4). Some sscDNA was produced, although the amount was substantially less than that seen when input RNA was present (Table [1](#T1){ref-type="table"}). To determine the possible impact of this template-independent product on microarray results, we hybridized the entire sscDNA product from a template-independent reaction to a U95Av2 microarray. This resulted in low overall signal intensity with only 0.6% of probe sets yielding \"present\" calls. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### sscDNA yield from ribo-SPIA experiments ::: Sample Input RNA sscDNA Total Yield (μg) ---------------- --------------- ----------------------------- Experiment 1 cUHR 1 20 ng 5.7 cUHR 2 20 ng 7.3 cUHR 3 20 ng 4.6\* Experiment 2 Mouse liver 1 5 ng 9.4 Mouse liver 2 5 ng 11.8 Mouse liver 3 5 ng 10.0 Mouse liver 4 100 ng 11.4 Mouse liver 5 100 ng 10.5 Mouse liver 6 100 ng 8.7 Experiment 3 K562 1 10 ng 6.6 K562 2 10 ng 7.8 K562 3 10 ng 8.3 sUHR 1 10 ng 8.4 sUHR 2 10 ng 9.4 sUHR 3 10 ng 7.8 Experiment 4 No Input RNA 1 0 2.7 No Input RNA 2 0 3.2 \Part of sample lost during handling and was removed from later calculations. ::: Size of sscDNA products ----------------------- sscDNA preparations were analyzed by electrophoresis using an Agilent 2100 BioAnalyzer. sscDNAs ranged widely in size and the median size was typically slightly greater than 1 kb (data not shown). sscDNAs were fragmented in preparation for hybridization resulting in fragments of \~50--200 bp. These results are similar to those previously obtained using this method \[[@B16],[@B17]\]. Reproducibility of microarray hybridization results --------------------------------------------------- Each of the experiments included replicate amplifications (independent amplifications of aliquots of the same starting material). We hybridized each replicate to a separate microarray and calculated intensities for each probe set. Fig. [2A](#F2){ref-type="fig"} shows an example of one pair of replicate hybridizations from Experiment 1, each performed using 20 ng of cUHR RNA as starting material. Pairwise comparisons of probe set intensities for replicate hybridizations (Table [2](#T2){ref-type="table"}) produced very high correlations (r = 0.983--0.996) across the entire range of starting RNA amounts used in the three experiments (5--100 ng). One previous study also showed high correlations (r \~ 0.97--99) between replicate array data produced using the ribo-SPIA method \[[@B16]\]. To assess how the amount of input RNA affects microarray results, we compared intensities found using 5 versus 100 ng of murine liver RNA (Experiment 2). When a 5 ng sample was compared to a 100 ng sample, there was a similar strong correlation (r = 0.987), indicating that large differences (20-fold) in starting material between reactions had small effects on measurements of gene expression (Fig. [2B](#F2){ref-type="fig"}). Previous reports have used correlation values between replicates to assess the reproducibility of other amplification methods. Those studies involve different laboratories and a wide range of microarray platforms, which makes direct comparison challenging. However, the correlations that we obtained using the ribo-SPIA method compare favorably with those reported for two rounds of IVT amplification (r = 0.92--0.98) \[[@B18]-[@B21]\], SMART amplification (r = 0.85--0.97) \[[@B22]-[@B24]\], and a PCR-based form of amplification (r = 0.97) \[[@B19]\]. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Correlation of gene expression measurements for technical replicates prepared using the ribo-SPIA protocol.(A) Data shown here are from two independent amplifications and hybridizations performed using the same starting RNA (cUHR, Experiment 1) and are representative of the pairwise correlations obtained for all replicate hybridizations in Experiments 1 -- 3. (B) Data shown compare two independent amplifications using 20-fold different mass of starting RNA (5 ng versus 100 ng). ::: ![](1471-2164-6-57-2) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Correlations between signal intensities for replicate hybridizations ::: Sample* Input RNA Range of correlations^1^ ------------------ --------------- ------------------------------ Experiment 1 cUHR 1--2^2^ 20 ng 0.996 Experiment 2 Mouse liver 1--3 5 ng 0.983--0.993 Mouse liver 4--6 100 ng 0.986--0.991 Experiment 3 K562 1--3 10 ng 0.985--0.990 sUHR 1--3 10 ng 0.983--0.991 ^1^Correlation coefficients (r values) were calculated by examining all three possible pairwise comparisons between replicates. ^2^The third replicate from this set was removed from all calculations due to losses of material during sample preparation. ::: Differential gene expression measurement ---------------------------------------- To test the ability to detect differential gene expression using amplified sscDNAs, we compared gene expression in two different RNA samples. We chose to compare K562 cell and sUHR RNAs since we have previously used these two RNAs to compare the performance of single round IVT-based amplification with other methods \[[@B25]\]. In Experiment 3, we did three separate sscDNA amplifications of K562 and sUHR RNAs. We used the same RNA preparations as for the previously reported IVT-based amplification experiments, but started with 1000-fold less material (10 ng instead of 10 μg). For each probe set, we used intensity values from replicate arrays to calculate relative gene expression (M, defined as log~2~\[mean K562 intensity/mean sUHR intensity\]) and average signal intensity (A, defined as 1/2 log~2~K562 mean intensity + 1/2 log~2~mean sUHR intensity). M and A values obtained using sscDNA are shown in Fig. [3B](#F3){ref-type="fig"} and those obtained using IVT-generated cRNA are shown in Fig. [3A](#F3){ref-type="fig"}. The range of signal intensities was similar for the two methods, although the mean intensity was lower for sscDNA hybridizations (A = 5.22 for sscDNA, A = 6.12 for cRNA). In contrast, the range of M values was somewhat larger with sscDNA. In particular, the sscDNA method identified several transcripts that were more than 2^7^-fold higher in sUHR than K562 cell RNA (M \< -7), but no differences of this magnitude were identified using cRNA. The number of probe sets associated with greater than 2-fold differences in expression (\\|M\\| \> 1) was 1518 for sscDNA and 1043 for cRNA. 51% of the genes with \>2-fold differences in expression on sscDNA arrays were not detected as \>2-fold on cRNA arrays, but only 25% of the genes that were \>2-fold different on cRNAs were not detected as \>2-fold different on sscDNA arrays (Fig. [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Differential gene expression measurements made using IVT- and ribo-SPIA-based amplification methods.Differential gene expression (M) and average intensity (A) were calculated by averaging results from replicate hybridizations performed using cRNA prepared by IVT (A) or sscDNA prepared by ribo-SPIA (B). Points outside the horizontal lines indicate probe sets with more than a 2-fold change in expression level for that sample preparation method. Blue points in (A) indicate probe sets with more than a 2-fold change as determined using sscDNAand red points in (B) indicate probe sets with more than a 2-fold change as determined using cRNA. ::: ![](1471-2164-6-57-3) ::: The new observation that sscDNA gave a wider range of relative expression (M) values despite lower average intensity (A) values could be explained by improved hybridization specificity under the conditions used in this study. This is plausible because the binding energy for DNA-DNA interactions is more sensitive to base pair mismatching than the binding energy for DNA-RNA interactions \[[@B26],[@B27]\]. To look for further evidence about specificity of hybridization, we took advantage of the mismatched (MM) probes included on the arrays. For each perfect match (PM) GeneChip 25 mer probe, there is a corresponding MM probe with a single base mismatch at base 13. The MM probes were included in the probe set design to allow adjustments for nonspecific hybridization. Under ideal conditions, MM probes would never give signals higher than PM probes, although in practice this does sometimes occur. MM probes would be more likely to give stronger signals than PM probes if there was more non-specific hybridization of off-target sequences to the probes. We found that MM intensities exceeded PM intensities less frequently when we used sscDNA as compared to cRNA. When sUHR RNA was used as starting material, the average number of probe sets where MM intensity exceeded PM intensity was 2247 for cRNA versus 1671 for sscDNA (34% higher, p = 0.008). MM intensity also exceeded PM intensity more frequently with cRNA probes for K562 RNA arrays (2903 vs. 2482 probe sets, 17% higher, p = 0.017). When we looked at raw signal intensity for all MM probes, we found that the cRNA MM intensity distributions were skewed compared to the sscDNA MM distribution (Fig. [4](#F4){ref-type="fig"}). A closer examination of these distributions revealed that the use of sscDNA instead of cRNA resulted in a substantial reduction in the number of MM probes that gave relatively high intensity signals (Table [3](#T3){ref-type="table"}). These findings strongly suggest that hybridization specificity is better for sscDNA than for cRNA. In a related study, Gingeras and coworkers \[[@B28]\] observed that increased nonspecific hybridization was observed when using directly labeled E. coliRNA as compared to cDNA. The increased nonspecificity was attributed to the presence of large amounts of rRNA in the samples. In our study however, both target preparations were prepared using oligo(dT) primers for the synthesis of first strand cDNA, so this explanation is less likely. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Density plot of mismatch probe signal from cRNA and sscDNA targets.The raw intensity distribution of all mismatch probes are plotted for three sUHR cRNA and three sUHR sscDNA arrays. ::: ![](1471-2164-6-57-4) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Mismatch probe signal intensities from sscDNA and cRNA hybridizations. ::: Target \> 2× median\* \> 4× median\* \> 8× median\* ------------ -------------------- -------------------- -------------------- sscDNA 14.1 ± 1.8% 5.0 ± 0.8% 2.0 ± 0.3% cRNA 26.6 ± 1.4% 12.9 ± 0.9% 5.4 ± 0.6% The proportion of mismatch probes with intensities greater than 2, 4, or 8 times the median for all mismatch probes on the same array. Values are mean ± standard deviation for triplicate arrays hybridized with sscDNA or cRNA prepared from sUHR RNA. Each array had 201,800 mismatch probes. ::: Comparison of expression measurements made with sscDNA, cRNA, and qPCR ---------------------------------------------------------------------- We wished to compare how measurements made using amplified sscDNA and microarrays compared with measurements made using other approaches. We began by comparing results obtained using sscDNA and cRNA microarray hybridizations for all 12,625 probe sets on the arrays. Since the sscDNA and cRNA methods would be expected to introduce different systematic biases, we were not surprised that direct correlations between signal intensities obtained with the two different methods showed show relatively poor agreement (r = 0.72--0.75 for K562 and r = 0.68--0.70 for sUHR, as opposed to r = 0.98--0.99 between replicates performed using the same sample preparation method). The finding indicates that it will not be useful to directly compare one array hybridized with sscDNA to another one hybridized with cRNA. We next compared differential gene expression measures (M values) determined using sscDNA with those determined using cRNA. There was a clear correlation (r = 0.83, Fig. [5A](#F5){ref-type="fig"}). We expected that probe sets associated with low intensity signals would give less reliable measures of gene expression and when we removed these probe sets from the analysis the correlation improved (r = 0.90, Fig. [5B](#F5){ref-type="fig"}). On average, the estimated M values were slightly larger (\~1.2 times higher) when sscDNA was used instead of cRNA. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparison of gene expression measurements between sample preparation methods.(A) Differential gene expression measurements for all U95Av2 probe sets. (B) Differential gene expression measurements for 4179 probe sets after removal of signal less than the median intensity from both cRNA and sscDNA samples (A\<5.485). ::: ![](1471-2164-6-57-5) ::: Next we generated another set of expression measurements that could be used as a basis for comparison for the sscDNA and cRNA array results. qPCR is typically used as \"gold standard\" to confirm putative differentially expressed genes detected with microarrays. Since we saw a subset of genes for which expression differed between sscDNA and cRNA targets, we next assessed if either method tracked more closely to qPCR. We chose qPCR primers and probe sets from a large group of \>1000 sets that have been developed for various studies. From these, primers and probes for four subsets of genes were selected for qPCR. The first set included all genes with \>4 fold difference in expression between K562 and sUHR samples as determined using the sscDNA method, the cRNA method, or both methods (53 primer/probe sets). The second set included all other genes in which the two methods disagreed by more than 2-fold (29 primer probe sets). The third set consisted of a group of 33 empirically-derived \'housekeeping genes.\' These were all genes that were nearly equally expressed (\\|M\\| \< 0.1) in K562 and sUHR samples according to both the sscDNA and cRNA methods and gave strong signals (A \> 5 for both methods). The fourth set included 8 housekeeping genes that had been previously validated as controls for qPCR in other experiments. We determined the gene copy number for each qPCR primer and probe set and then calculated a measure of relative expression, M = log~2~(K562 copy number)/(sUHR copy number), that could be directly compared to M values from arrays. 37 putative duplicate probe sets from 17 genes probe sets were hand-curated to confirm that they would correspond to the predicted qPCR product. In two cases probe sets were found to be misidentified in the GeneChip annotation and were removed from the analysis. In the remaining cases of duplication, the qPCR and microarray values were averaged across the duplicates. The final set of 106 curated genes and the associated data can be found at <http://asthmagenomics.ucsf.edu>. There were clear correlations between qPCR M values and array M values obtained using sscDNA (Fig. [6A](#F6){ref-type="fig"}) or cRNA (Fig. [6B](#F6){ref-type="fig"}). When all 106 genes were included, qPCR results correlated slightly better with sscDNA (r = 0.72) than with cRNA (r = 0.66). When we included only the 29 genes for which sscDNA and cRNA methods disagreed by more than 2-fold, the difference between the two sample preparation methods became more pronounced (r = 0.75 for sscDNA vs. r = 0.57 for cRNA). Not surprisingly, both array-based methods tended to give smaller estimates of M than qPCR; this relative underestimation was somewhat less marked for the sscDNA than the cRNA method. In summary, results from arrays hybridized with sscDNA samples amplified using the ribo-SPIA method tracked with qRT-PCR more closely than did results from arrays hybridized with cRNA samples prepared using the traditional IVT method. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Comparison of differential gene expression measurements between qRT-PCR and microarrays.Differential gene expression measurements for 106 genes made using qRT-PCR were compared to array measurements made using sscDNA targets (A) or cRNA targets (B). Red points indicate genes for which cRNA and sscDNA samples varied by more than 2-fold in differential gene expression. ::: ![](1471-2164-6-57-6) ::: Conclusion ========== We examined the suitability of a new isothermal linear amplification method for application to Affymetrix GeneChip microarrays. We performed a series of tests using starting amounts of RNA ranging from 5 to 100 ng for amplification yield and reproducibility. The amplification reactions consistently produced sufficient sscDNA for multiple array hybridizations. Pairwise comparison of technical replicates hybridized to microarrays by regression analysis showed excellent consistency. When we used sscDNA to analyze differential gene expression between two samples, we found a larger dynamic range than that obtained with cRNA hybridizations. The improved performance appears to be related to increased sscDNA hybridization specificity. The data obtained using this new method also more closely matched the results from qRT-PCR than data obtained using standard IVT reactions, even though the amount of starting RNA used was 1000-fold less. This new amplification method is a useful alternative approach for preparing targets that is especially well-suited for experiments involving small amounts of starting material. Methods ======= Test samples ------------ Clontech Human Universal Reference Pool total RNA (cUHR), derived by pooling RNA from a variety of human tissues, was purchased from BD Biosciences and used in Experiment 1. Mouse total liver RNA was isolated by standard methods from C57/BL6 mice according to procedures approved by the UCSF Committee on Animal Research and used in Experiment 2. For Experiment 3, we used Stratagene Human Universal Reference Pool and K562 erythroleukemia total RNAs from the same batches used in a previous study \[[@B25]\]. All samples were assessed for size and integrity using the Agilent 2100 BioAnalyzer RNA 6000 Nano LabChip assay. RNA and DNA samples were quantified using a NanoDrop ND-1000 spectrophotometer. Sample preparation ------------------ sscDNA samples were prepared using the NuGEN Technologies Ovation RNA amplification and Biotin Labeling system (Version 1.0) according to the manufacturer\'s directions from the indicated amount of starting RNA (5--100 ng). All reactions were performed in 0.2 ml strip PCR tubes in an MJ GeneWorks PTC-100 thermocycler using recommended programs. Since the seal for PCR tubes and caps tends to deteriorate with repeated use, we replaced the caps for each tube before each resealing step in the protocol. Following amplification, sscDNA product was purified using QIAquick PCR purification kits (Qiagen). Samples were fragmented and end labeled with biotin. After stopping, each reaction was concentrated in a Microcon YM-3 column to a final volume of \~20 μl. The concentrated material was purified using a Centri-Sep 100 spin column (Princeton Separations). Negative control reactions were prepared by replacing input RNA with the appropriate volume of RNase free water. DNA microarrays --------------- All samples were placed in standard Affymetrix hybridization buffer. The sample denaturation time for the sscDNA samples was reduced from 5 to 2 minutes and hybridization time increased from 16 to 20 hours as recommended by NuGEN Technologies. cUHR gene expression was assayed using Affymetrix Human Genome U133A GeneChip arrays (Experiment 1). Mouse liver RNA was assayed using Murine Genome Mu6500A arrays (Experiment 2). K562 and sUHR RNAs were assayed using Human Genome U95Av2 arrays (Experiment 3). One template independent sample was also analyzed using a Human Genome U95Av2 array (Experiment 4). Arrays were stained with phycoerythrin-streptavidin according to the manufacturer\'s instructions. Metrics for all sample hybridizations including scaling factors, mean background intensities, and percent present calls have been provided (see Additional File [1](#S1){ref-type="supplementary-material"}). Each set of data was normalized independently using RMAExpress software <http://stat-www.berkeley.edu/~bolstad/RMAExpress/RMAExpress.html>. K562 and sUHR microarray data were also analyzed using Microarray Suite 4.0 in order to calculate PM-MM values for each transcript probe set. Probe level analyses were performed using the BioConductor \[[@B29]\] affy analysis package \[[@B30]\]. All microarray data have been deposited in the Gene Expression Omnibus (GEO) database under the accession numbers GSM41384 -- GSM41393, GSM41433 -- GSM41438 and GSM4843 -- GSM4847. Real-time PCR ------------- Real-time (RT) PCR was used to measure the expression of selected genes in sUHR and K562 cells. Gene-specific primers for multiplex real time RT-PCR were designed for each gene of interest using \"Primer Express\" software (Perkin-Elmer) and purchased from Biosearch Technologies. Sequence data for all oligonucleotides primers has been provided (see Additional File [2](#S2){ref-type="supplementary-material"}). First strand cDNA synthesis was performed using total RNA, Powerscript reverse transcriptase (BD Biosciences), and random hexamer primers. Real time amplification was performed using an ABI Prizm7900 and Invitrogen Universal Master Mix. Relative gene copy numbers (GCN) were calculated as described previously \[[@B31]\]. GeNorm \[[@B32]\] was used to select the two most stable housekeeping genes across all specimens for normalization. Authors\' contributions ======================= CSB conceived of the study, participated in the design and coordination, and wrote the manuscript. CG participated in the design of the study and both CG and KH performed the microarray experiments. GMD carried out the real time PCR studies. JYHY performed the statistical analyses. DJE participated in the study design and analysis and revised the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Hybridization Metrics is a .txt file suitable for opening in Microsoft Excel containing information about each individual hybridization. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Taqman Primers provides sequence information for oligonucleotides used for qRT-PCR. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This study was supported in part in the General Clinical Research Center at San Francisco General Hospital and supported by Grant 5-MO1-RR00083 from the Division of Research Resources, National Institutes of Health, the UCSF NHLBI Shared Microarray Facility (NIH grant HL072301) and the UCSF Sandler Center for Basic Research in Asthma. We thank Yanxia Hao, Jennifer Gregg, Agnes Paquet and Michael Salazar for excellent technical assistance, Ron Davis for helpful discussions and Karena Essex for assistance in preparation of the manuscript.	PubMed Central	2024-06-05T03:55:56.517951	2005-4-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090574/", "journal": "BMC Genomics. 2005 Apr 22; 6:57", "authors": [ { "first": "Christopher S", "last": "Barker" }, { "first": "Chandi", "last": "Griffin" }, { "first": "Gregory M", "last": "Dolganov" }, { "first": "Kristina", "last": "Hanspers" }, { "first": "Jean Yee Hwa", "last": "Yang" }, { "first": "David J", "last": "Erle" } ] }
PMC1090575	Background ========== Invasive fungal infections caused by Candidaspecies have increased significantly. They now rank fourth as the most common cause of nosocomial bloodstream infections in the United Stated \[[@B1]\] and the most common one at one major hospital in Taiwan \[[@B2]\]. Candidemia is not only associated with a mortality of about 30% to 40% but also extends the duration of hospital stay \[[@B3],[@B4]\] and increases the cost for medical care \[[@B5]\]. In recent years, Candidaspecies associated with candidemia have shifted from Candida albicansto non-albicans Candidaspecies (NAC) \[[@B6]\]. Approximately half of the reported cases of candidemia are now caused by NAC \[[@B7]-[@B10]\]. Patients with candidemia usually present an acute septic syndrome that is indistinguishable from bacteremia, but they may also exhibit a more indolent course manifested by fever of unknown origin. Major risk factors for candidemia include intravascular catheters, parenteral hyperalimentation, and broad-spectrum antibiotics. Empirical antifungal agents may be administrated to patients manifested with fever of unknown origin and have above mentioned risk factors, particularly those who have been treated with broad-spectrum antibiotics. The clinical presentation of the patients with sepsis caused by C. albicansand NAC are indistinguishable. However, NAC are often less susceptible to fluconazole than C. albicansis \[[@B7],[@B11],[@B12]\] and may require greater dosage to cure clinically \[[@B13],[@B14]\]. We have conducted a retrospective chart review of patients whose death was associated with candidemia. The goal was to assist decisions to select the most appropriate empirical therapy for patients suspected to have candidemia. The objective was to determine whether specific risk factors could be identified to help selecting those patients who are more likely infected with C. albicansversus NAC. Methods ======= This study was conducted at the Veterans General Hospital-Taipei. This is a 2800-bed teaching hospital with about 69,000 admissions annually \[[@B7]\]. A total of 415 blood isolates of Candidaspecies (one strain per species per patient) were collected from April 1996 to December 1999. During this period 130 patients met the inclusion criteria of having at least one positive blood culture for Candidaspecies within one month prior to the fatal outcome. There were 68 patients infected with C. albicansand 62 with NAC. Candidaspecies were isolated from blood using BACTEC 860 system (Becton Dickinson, INC, Sparks, MD). The species were identified using API-32C system (bioMerieux Vitek, Inc, St. Louis, MI). Clinical data ------------- Clinical data of 130 patients were recorded on standardized forms and analyzed according to age at the time of diagnosis, gender and underlying illnesses. These included cancer, diabetes, immunosuppression accounted to prior use of steroids, systemic antifungal agents, cancer chemotherapy, parenteral nutrition, the presence of central or arterial line, endotracheal tube, and urinary catheter, admission to an intensive care unit (ICU), history of infection or gastrointestinal surgery, hemogram, antifungal management, and cause of death. The predisposing events within one month prior to diagnosis and the laboratory data within seven days prior to obtaining the first positive blood culture were analyzed. To decide that a patient died directly of candidemia or underlying illness, we recoded the causes of death according to the death certificate of the patients. Statistical analysis -------------------- The statistical significance of association between categorical variables was assessed using the Fisher\'s Exact Test. The risk factors with p values \< 0.1 in the univariate analysis were placed in a multivariate analysis using the multiple logistic regression package in the SAS System for Windows V8. Confidence limit of 95% means that the confidence limit contains the true value of odds ratio with probability of 0.95. The narrower the confidence limit is, the more accurate the estimate is. Results ======= The population consisted of 130 patients whose death was associated with candidemia. Fifty-five (42.3 %) of these deaths were considered to be directly caused by Candidaspecies. The remaining died of complications of their underlying illness. The distribution of species of the total deaths and those causally related to candidemia is shown in Table [1](#T1){ref-type="table"}. There were no significant differences in causation of death according to species. Interestingly, of the four major species related to candidemia, 61.5% of the isolated Candida glabratacontributed to mortality, followed by 50% of Candida parapsilosis, 41.2% of C. albicans, and 31.3% of Candida tropicalis. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Distribution of Candida speciesof fatal outcome attributed to disseminated candidemia ::: Blood isolates (N = 130) Death directly related to candidemia (N = 55) ------------------------------- -------------------------- ----------------------------------------------- ---- ------ ------- C. albicans 68 52.3 28 50.9 41.2 non-albicans Candidaspecies 62 47.7 27 49.1 43.5 C. tropicalis 32 24.7 10 18.2 31.3 C. glabrata 13 10.0 8 14.5 61.5 C. parapsilosis 12 9.2 6 10.9 50.0 C. guilliermondii 2 1.5 1 1.8 50.0 C. peniculosa 2 1.5 2 3.6 100.0 C. famata 1 0.8 0 0 0 ::: The demographic and clinical characteristics of the study population categorized according to the presence of C. albicansor NAC in the blood are summarized in Table [2](#T2){ref-type="table"}. The significant (p≤ 0.05) risk factors on univariate analysis for C. albicanscandidemia were age ≧ 65 years, prior bacterial urinary tract infection, central venous catheter, parenteral nutrition and leucocytosis with white blood cell counts (WBC) ≧ 15000/mm^3^, without significant neutropenia with absolute neutrophil counts (ANC) ≧ 100000/mm^3^. The significant risk factors for NAC were age \< 65 years, cancer chemotherapy, neutropenia (WBC \< 3000/mm^3^) and severe thrombocytopenia (platelet count ≦ 20000 /mm^3^). NAC patients were more likely to have cancers and to be located on the medical wards (p = 0.08). When multivariate analysis was employed, the significant risk factors for C. albicanswere age ≧ 65 years, immunosuppressive therapy, and leucocytosis. Significant risk factors for NAC were age \< 65 years and location on medical wards, which are shown in Table [3](#T3){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Characteristics of 130 patients who died in association with candidemia ::: ----------------------------------- ----------------------- -------------- ---------------------- ------ ------------------- Characteristic C. albicans(N = 68) NAC (N = 62) Fisher\'s Exact Test No. \% No. \% Two-sided P-value Age ≧ 65 years old 56 82.4 36 58.1 0.004\* MICU 6 8.8 10 16.1 0.286 Medical ward stay 24 35.3 32 51.6 0.077\* Cancer 24 35.3 32 51.6 0.077\* Immunosuppressive therapy 17 25 7 11.3 0.069\* Bacteremia 25 36.8 15 24.2 0.133 Bacterial urinary tract infection 23 33.8 10 16.1 0.026\* Prior antifungal administration 5 7.4 9 14.5 0.259 Catheters within 30 days 65 95.6 56 90.3 0.309 ETT 35 51.5 23 37.1 0.114 CVC 57 83.8 41 66.1 0.025\* AL 21 30.9 14 22.6 0.326 Urinary catheter 49 72.1 39 62.9 0.348 PPN 50 73.5 34 54.8 0.029\* Cancer chemotherapy 5 7.4 12 19.4 0.066\* Corticosteroids 26 38.2 18 29 0.354 ANC (N = 44) (N = 40) 1 ANC ≦ 1500/mm^3^ 2 4.5 2 5 WBC (N = 67) (N = 61) 0.0004\* ≦ 3000/mm^3^ 4 6 14 23 3000/mm^3^to 5000/mm3 35 52.2 38 62.3 \> 15000/mm^3^ 28 41.8 9 14.8 Platelet (N = 66) (N = 61) 0.011\* ≦ 20000/mm^3^ 5 7.6 10 16.4 20000/mm^3^to 100000/mm^3^ 20 30.3 29 47.5 \> 100000/mm^3^ 41 62.1 22 36.1 ----------------------------------- ----------------------- -------------- ---------------------- ------ ------------------- \* Factors selected for multivariate analysis. AL, arterial line; ANC, absolute neutrophil count; CVC, central venous catheter; ETT, endo trachial tube; ICU, Intensive care unit; PN, parenteral nutrition; WBC, white blood cell ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Multiple regression analysis of risk factors for candidemia caused by Candida albicansvs. non-albicans Candida species ::: ----------------------------------------- ------------ ----------------------- --------- -------- Effect Odds Ratio 95% confidence Limits P-value White Blood Cell (WBC) 0.0202 \> 15000/mm^3^vs. ≦ 3000/ mm^3^ 5.89 1.35 25.7 0.0184 3001 -- 15000/mm^3^vs. ≦ 3000/ mm^3^ 1.66 0.44 6.26 0.4549 \> 15000/mm^3^vs. 3001 -- 15000/mm^3^ 3.55 1.29 9.79 0.0144 Age65 (≧ 65 vs. \< 65) 6.73 2.37 19.1 0.0003 Immunosuppressive therapy 4.39 1.26 15.32 0.0201 Located on a medical ward 0.37 0.16 0.88 0.025 ----------------------------------------- ------------ ----------------------- --------- -------- ::: Discussion ========== Candida albicansremains the most common species causing candidemia, but the proportion caused by NAC is increasing \[[@B7]-[@B10]\]. Patients with NAC are more likely to require greater dosage of fluconazole to cure clinically \[[@B13],[@B14]\]. Thus there is a need to identify patients at risk of NAC candidemia to initiate empirical amphotericin B therapy or high-dose fluconazole. Krcmery and Barnes have identified the following risk factors for NAC \[[@B15]\]. These include prophylaxis with azole compounds as a risk factor for C. kruseiand C. glabrata, neutropenia and bone marrow transplantation for C. tropicalis, and insertion of foreign bodies, neonates, and hyperalimentation for C. parapsilosis\[[@B15]-[@B18]\]. In this study we have identified additional risk factors that help distinguish candidemia caused by C. albicansand NAC. The most important risk factors for C. albicansare the old age, procedures associated with intensive care, and an acute sepsis. The most important risk factors for NAC are cancer chemotherapy in association with leukopenia and thrombocytopenia. It has been reported that Candidaspecies exhibit a spectrum in the extent of adherence to tissues, which correlates with the pathogenicity in humans and animals \[[@B19]\]. Candida albicansexhibit the greatest capacity to adhere to gingival epithelial cells, followed by C. tropicalisand C. glabrata\[[@B19]\]. The capacity of yeasts to attach to a wide range inanimate surfaces appears to protect them from immune responses and antimicrobial agents \[[@B20]\]. This may explain why C. albicansis more likely to be associated with central venous catheters in this study. Conventional methods to eradicate C. albicansrely upon the use of antifungal drugs designed to kill the yeast or arrest its growth. However, removal of intravascular catheters is in fact the most simple and effective method \[[@B21],[@B22]\]. Another approach is to design new agents that disrupt adherence of the yeasts to host tissues and catheters. Conclusion ========== It is customary to administer amphotericin B for critically ill patients suspected of having candidemia. More often than not, azoles are used for relatively stable patients. NAC tend to be less susceptible to fluconazole than C. albicansis \[[@B7],[@B11],[@B23]\]. Thus, greater dosage is needed till the causative microorganism is isolated and identified, particularly with patients at high risk for NAC candidemia. Though, a large, prospective study is needed to validate this concept, the risk factors identified in this study may help, in clinical practice, to differentiate fatal candidemia caused by C. albicansversus NAC. Abbreviations used ================== NAC, non-albicans Candidaspecies; VGH-TPE, Veterans General Hospital-Taipei; ICU, intensive care unit; AL, arterial line; ANC, absolute neutrophil count; CVC, central venous catheter; ETT, endo trachial tube; WBC, white blood cell. Competing interests =================== The author(s) declared that they have no competing interests. Authors\' contributions ======================= MFC conceived the study and designed it together with, YLY, MH, and HJL. MFC conducted the experiments with contribution from RBT, YHF, and KSH. KWY collected clinical isolates. JSL performed the statistical study with contributions from CYL and TJY. MFC drafted the manuscript with contribution from YLY and HJL. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2334/5/22/prepub> Acknowledgements ================ We would like to thank Dr. Calvin M. Kunin and Dr. Clifford L. McDonald for their expert discussion. We would like to thank Mrs. Ing-Ming Liu at VGH-TPE for her help in collecting the isolates. We are also indebted to Mr. Hsiao-Hsu Cheng and Dr. Feng-Jui Chen at the National Health Research Institutes, Miaoli County, Taiwan, for their help in data processing.	PubMed Central	2024-06-05T03:55:56.520521	2005-4-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1090575/", "journal": "BMC Infect Dis. 2005 Apr 7; 5:22", "authors": [ { "first": "Ming-Fang", "last": "Cheng" }, { "first": "Yun-Liang", "last": "Yang" }, { "first": "Tzy-Jyun", "last": "Yao" }, { "first": "Chin-Yu", "last": "Lin" }, { "first": "Jih-Shin", "last": "Liu" }, { "first": "Ran-Bin", "last": "Tang" }, { "first": "Kwok-Woon", "last": "Yu" }, { "first": "Yu-Hua", "last": "Fan" }, { "first": "Kai-Sheng", "last": "Hsieh" }, { "first": "Monto", "last": "Ho" }, { "first": "Hsiu-Jung", "last": "Lo" } ] }

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.