text
string | predicted_class
string | confidence
float16 |
|---|---|---|
Having defined the quadrants, the tumor position is then randomly selected inside each quadrant - 3 spatial coordinates (x, y and z) are used to set the center of the tumor - and the excision volume is computed. Briefly, the line between the nearest point of the pectoral muscle and the tumor position sets the normal vector to the muscle, and a predefined cylinder (with a known radius, height and, consequently, volume) is aligned through this direction (Figure 6). Different tumor sizes (volumes) can then be modelled by varying the ratio between cylinder and breast volumes. Once the BCS protocol states that a breast tumor is eligible for BCS only if its removal do not require excision volumes higher than 20% of breast volume , cylinder volumes need to be limited to respect this threshold. In this study, excision volumes of 5%, 7.5% and 10% of the total breast volume were simulated, corresponding to three size categories: small, medium and large tumors, respectively (Figure 7a–c).
|
study
| 100.0 |
Figure 8 shows the main considerations made in the conception of the dataset. This dataset was built using 6 breast PCLs (obtained from MRI data), taking into account a uniform distribution of breast volume (2 small, 2 medium and 2 big breasts) and breast laterality (3 left and 3 right breasts), as described in Table 3. Dataset instances were created by sequentially defining 4 different breast densities for each breast selected before, according to BI-RADS® reporting system (4×6=24 cases), then different quadrants for the tumor location (4×24=96 cases) and, finally, 3 different tumor sizes for each location (3×96=288 cases). In the end, the dataset sums up to a total of 288 cases representing all the possible combinations of the most prominent clinical factors reported to affect breast shape after BCS.
|
study
| 100.0 |
The prediction of the post-surgery shape of the breast after surgical intervention is a complex task that requires modelling the influence of several factors on the aesthetic outcome of surgery. Approaches to model these deformations are typically based on biomechanical models. However, FEM takes longer than expectations, from hours on high-end machines, up to some days on normal computers used in clinics. In this work, an alternative strategy-based on machine learning techniques is proposed which overcomes the timing demands of biomechanical simulation, keeping most of the properties and characteristics of the breast.
|
study
| 99.94 |
Feature extraction and representation is an important step in any machine learning task. Although clinical evidence suggests that a prediction model for breast deformation after BCS should take breast shape (laterality), volume and density into account, considering tumor characteristics such as quadrant (position) or size as inputs, such factors should be inspected in a more systematic way to confirm their influence and effects. Moreover, such analysis is important to suggest the best suited machine learning algorithms to model the problem at hand. For visual purposes, the feature investigation was constrained only to the breast surface.
|
study
| 99.94 |
Regarding breast characteristics, both density and shape are known to affect the extent of breast points displacements. Figure 9 shows the superposition of pre- and post-surgical breasts of the same patient, when different densities are modelled. The resulting plots show that breast density impacts the magnitude of points displacements: the magnitude of displacements decreases as the breast density increases. In fact, this was the expected behaviour, because denser breasts have a higher fraction of glandular tissue, which is less deformable than fat.
|
study
| 100.0 |
Figure 11 shows the effect of the tumor position on the displacements between pre- and post-surgical data. An interesting effect can be noticed: the distribution of displacements on the breast is dependent on the breast quadrant where the tumor is positioned. Larger displacements are centered around the tumor, vanishing as the distance to the tumor center increases. Hence, a feature space transformation might be helpful to describe the displacements distribution as function of the tumor position. In alternative, Euclidean and polar distance to tumor could benefit the modelling of the points displacements.
|
study
| 100.0 |
Finally, the influence of the tumor size on the magnitude of breast displacement is shown in Figure 12. Results evidence that larger tumors cause larger breast displacements after surgery wound healing. In fact, this was the expected behaviour: after tumor removal the remaining breast tissues adapt to fill the left void. This results in breast contraction, which is a function of the excised volume.
|
study
| 100.0 |
A brief look to Figure 9, Figure 10, Figure 11, Figure 12 denotes how different breast and tumor characteristics influence breast deformations after BCS and respective wound healing process. Breast density and tumor size have particular impact on the magnitude of the displacements, while the quadrant where the tumor is positioned and the breast laterality influences the distribution of those displacement. Therefore, one can expect that breast deformations can be modelled using the spatial coordinates of points, the distance of each point to the tumor position (the distance from a point perpendicular to the tumor cylinder), while accounting for categorical features such as breast density, tumor region, breast laterality and tumor size, as described in Table 4. Despite being appointed as an important clinical factor influencing the aesthetics of breasts after BCS, the volume of the breast is not explicitly listed. However, that information is implicitly covered due to the categorization of tumor sizes (expected excision volumes) which are defined as a percentage of the breast volume.
|
study
| 99.94 |
Definition of the features has been conducted aligned with the aforesaid expectations. The constructed feature list comprises three data attributes: points’ coordinates, points’ difference to the excised cylinder (both Euclidean and polar) and points’ distance to the excised cylinder as quantitative continuous, tumor size and breast density with categorical ordinal, and breast laterality and tumor region with categorical nominal attribute. The point coordinates simply expresses the location of each point in the 3D coordinate system, while the point coordinates difference feature reflects the difference between healthy points and the excised cylinder in each axis of the coordinate system. While the distance to the excised cylinder highlights the Euclidean distance from each healthy point, the polar distance to the excised cylinder expresses the same distance, but in polar notation. It is important to note that the coordinate difference and distance features for damaged points (inside the excised cylinder) are considered to be zero.
|
study
| 99.94 |
In this section, it will be explored solution taken into consideration the feature analysis above. The intention of exploiting machine learning in breast shape prediction is to estimate the point coordinates after surgery healing, taking as input the points positions before surgery, as well as breast and tumor characteristics. Since the points coordinates are continuous variables in 3D space, learning techniques providing regression methodologies are taken into consideration; however, prediction of the points coordinates poses a challenge to transfer breasts with different laterality or size, into the same coordinate system. Such circumstance can be prevented by re-formulating the demanded output from the regression. Instead of predicting the exact point coordinate, required displacement to translate a pre-surgery point to its post-surgery location can be predicted, alternatively. The post-surgery PCL is then attainable by applying predicted displacement on the pre-surgery PCL. Described in mathematical notation, the regression model can be expressed as in Equation (1):(1)PpreF⋮⋮→fdisppre→post⋮, where Ppre is pre-surgery PCL, F is the feature list per instances (pre-surgery points), f is the demanded regression model, and finally disppre→post expresses the required displacements to convert pre-surgery PCL to post-surgery. Having predicted the displacement, predicted breast shape (Ppred) is attainable via Equation (2):(2)Ppre+disppre→post=Ppred
|
review
| 76.0 |
Looking back to the objective of predicting breast shape after BCS, the expected prediction should be performed considering the points of both pre- and post-surgery models, together with clinical features. Therefore, the possible learning approach to be proposed must be able to deal with large number of inputs (points), and correlate them with the features (clinical features). In ensemble learning the key idea is that different algorithms explore different search spaces and hypotheses, so composite systems could outperform single ones . The strategy of exploiting ensemble learning methodology assures to gain improvement of not only the robustness, but also the performance of the learners via combining the votes of stronger single regressors combined to build the prediction model. As an ensemble learning approach, tree-based learners provide an appropriate framework (both in time and performance evaluations) to take part in training with large number of inputs. Therefore, within this research, tree-based learning methodologies are taken into consideration to perform the required regression in finding the predicted coordinate of breast shape.
|
study
| 99.94 |
Regression methods also can be categorized based on the number of their outputs. While single output regression are the most used ones, the internal correlation between their voters can be set such that they can generate multiple outputs. Therefore, taking the problem of predicting breast shape into account, regressors can be categorized in the two types of Multiple Input Single Output (MISO), or Multiple Input Multiple Output (MIMO). Both types of the aforesaid regressors are studied and evaluated in this work.
|
study
| 99.8 |
Depending on the goal whether to decrease bias error and overfitting, or to decrease both bias and variance, bagging and boosting are generally used, respectively. Concentrating in the bagging, unstable models, i.e., models whose performance is sensible to small perturbations of the training set, are trained with different replicas of the training set, obtained with replacement to keep the same number of examples (bootstrap aggregation). Then, new examples are predicted by uniform voting between the regressors trained with different dataset replicas. A special variant of bagging applied to decision tress results in the RF method, which operates by constructing multiple decision trees with bagging and random selection of features at each split of each tree. This mechanism assures that the constructed trees become correlated with one or more features which are strong predictors .
|
study
| 97.7 |
The algorithm for Gradient Boosting Regression is a recasted adaptation of AdaBoost that employs boosting methods in regression trees . The general idea is to compute a sequence of simple trees, where each successive tree is constructed for the prediction residuals of the preceding tree . Minimization of the loss (residuals) of the model (or regressor) is pursued by adding weak learners using a gradient descent procedure. Therefore, three elements are considered directly in developing a regressor with gradient boosting: a loss function, a weak learner, and finally an additive model to add weak learners to minimize the loss function. Considering decision trees as the weak learner, they are added one at the time, while the existing trees are kept unchanged. To ensure the simplicity of the learned trees, it is common to assign specific constraints to control the growth of trees, for instance, defining a maximum limit for depth, or the number of leaf nodes.
|
other
| 99.56 |
The capability of optimization is granted to the trees since they are constructed with parameters to be modified in direction of reducing the total loss function of the regressor. A tree which reduces the total loss is added to the existing sequence of trees .
|
other
| 99.9 |
As a greedy algorithm, gradient boosting has overfitting potential in training data, quickly. Therefore, common techniques such as regularization, assists the performance of prediction by penalizing various parts of the algorithm. The discussed constraints on tree construction is a example of the regularization methods to control the greediness of gradient boosting .
|
other
| 99.8 |
Regression models are normally characterized by only one output; however, taken into consideration the problem here presented, it make sense to think in a strategy based in MIMO regressors. In order to predict more than one output, it can be simple considered a regressor for each output, by concatenating several MISO regressors ; however, ignoring possible relationship among the constructed models could result in a drawback for the solution. A smarter solution is also suggested to construct several regressors not only by the input training data, but also by the possible internal relationships between them. In particular, this solution takes the advantage of constructing a MIMO regressor which is smaller than the size of those MISO regressors. It should be noted that the discussed solution leads to better predictions when there is a strong correlation between the features and the targets.
|
study
| 71.1 |
Through a comprehensive study between the pre- and post-surgery PCLs, the influence of each breast and tumor characteristics features were determined. Aligned with the objective of the current research to predict the shape of the breast after surgery, regression was taken into consideration as the main solution. Further discussions unveiled that the tree-based methodologies are capable enough to satisfy the input/output demands of the solution.
|
study
| 100.0 |
The definition of training and testing sets is carried out with a careful approach, in which a patient leave-one-patient-out (LOPO) strategy is advised to obtain test and train subsets. In each split, all example data generated from the same input patient would be used for testing, and the remaining patient’s data would compose the training subset. In the particular case of having 6 patients, the LOPO strategy is followed to prevent overfitting, due to the similarities between the PCLs of a single patient. To assess the model performance through LOPO strategy and tuning parameters at the same time, a cross-validation approach was used in order to find the parameters’ which presents the best configuration of the model.
|
study
| 100.0 |
As there is a deterministic correspondence between the points of the predicted (as the source) and the post-surgery PCL (as the target) for each patient, the evaluation metric can be defined by the Euclidean point-wise distance (p2p). Denoted in Equation (3), the point-wise distance evaluates the performance of the regression model, since it measures the amount of displacement of each pair of points regardless of the total PCL displacements. Less distance means the regression model predicted the coordinate of each point closer to its expected location.
|
study
| 99.94 |
(3)Dp2p=1N∑i=1Nd(Pisource,Pitarget), where N and d denote the number of points and Euclidean distance, respectively, and Pisource is the corresponding point of Pitarget. Not only point-wise, but also global distance can be calculated, as well. Unlike the point-wise, the global distance appraises the displacement between the two comparing sets in whole. That means the discussed metric gives an overview of the similarities between the source set with the target. The global measurement of the distance results in reporting two distances: from the source to target PCL and from the target PCL to the source. Closer reported distances signify the more similarity between the two PCLs.
|
study
| 80.3 |
Since the deformation of breast is correlated with both internal and surface tissue (described in Section 2.2.3), both so-called points are considered during the training stage; however, numerical evaluation only comprises the analysis of the surface points.
|
other
| 99.5 |
Note that for the both presented metrics, the mean distance (μ), the standard deviation from the mean (σ) and the maximum distance (Max) between the comparing sets are reported as well. The maximum distance expresses the furthest distance between two corresponding points of the predicted and pre-surgery PCLs. Also, pre and post symbols stand for pre- and post-surgery PCLs, and pred denotes the predicted breast PCL (wherever it is needed the predictions are also compared visually with the post-surgery models).
|
study
| 99.94 |
Finally, the machine learning implementations for RF and GBT were accomplished in python 3, by using scikit-learn package on a machine powered by intel® Core i7® at 3.2 GHz with 128 GB of memory (for cross validation). As long as the scikit-learn package includes multi-output regressor which is based on concatenation of individual regressors, the implementation of MOR was carried out with a package in R, called Multivariate Random Forest .
|
other
| 99.9 |
Following the approach of individual regressor for each axis, three RF regressors were trained. The cross-validation inner loop was set to optimize the number of estimators (trees), the maximum number of features in each tree, and the leaf size are of those parameters through {5,10,…,500}, {2,3,…,23}, and {1,2,…,5}, respectively.
|
study
| 99.94 |
The criterion to select the tuned parameters was chosen considering two objective functions (OF): the average, and the Hausdorff (maximum). Focusing on the average, the best set of tuned parameters are selected such that it minimizes the average distance between the predicted and the post-surgery models. The other OF which is based on the Hausdorff distance, intends to decrease the maximum distance between the points in each set of comparison of predicted and post-surgery PCLs, although the average distance may increase.
|
study
| 100.0 |
Theoretically in RF, inclusion of more points (more sampling rate) is thought to increment the gain of the regressor by declining the average distance between the predictions and the target, though, not only the training time continues its incremental trend, but also the aforesaid slope of the distance decelerates . In this regard, a comprehensive study was designed to find an optimized sampling rate in the range of {5,10,…,100}, according to both distance error and training time. The timing complexity reported in Figure 13 depicts the aforesaid evolution as the training size set increases, as expected. Besides, shown in Figure 14a, considering the average OF, the declining slope of the distance error decreases in defiance of sampling rate, until it reaches to 65%. Although the difference of the reported distances between the rates of 45% and 65% is less than 0.50 mm, to satisfy the condition of the OF, it was decided to use the sampling rate correlated with the global minimum distance (65%).
|
study
| 100.0 |
With the same argument, and in accordance with the Hausdorff OF, the sampling rate of 75% was selected due to the least evaluated maximum distance, though the difference between the rates 75% and 65% is measured around 0.196 mm. Figure 14b depicts the evolution of maximum distance according to the Hausdorff OF. Considering the aforesaid sampling rates, numerical evaluations with respect to average and Hausdorff OFs are calculated and reported in Table 5 and Table 6, respectively. To evaluate the magnitude of the reported distances, an extra comparison is performed with the distance between the two comparing PCLs in case no method is applied (meaning that prediction data is exactly equal to the pre-surgery data). This comparison, so-called baseline evaluation, is reported in the last column of Table 5 and the last two columns of Table 6, for both the average and the Hausdorff OFs.
|
study
| 100.0 |
Deep investigation of the trained RF reveals that the nature of problem demands to define different weights for the points belonging to healthy or damaged tissue. In this work a weight assigning strategy is followed in which the weights are initially assigned to the training instances and then they are updated iteratively in accordance to the distance from their correspondences in the target model. Thus, in each iteration, each point is assigned a weight that is proportional to the distance from its corresponding point of the target. This proposed iterative approach, called adaptive weighting, continues until either a fixed number of number of iteration is reached (in this case 100), or the OF is not satisfied within three consecutive iterations.
|
study
| 99.94 |
A glance to the results obtained in the previous section reveals that the distance for a point is near to 1 mm while the maximum distance is higher than 5 mm. Taken this in consideration we considered a weighting strategy by defining the values for the weights as the ceiling of the point-wise distance (p2p) in a range of 1 to 6, as shown in Figure 15.
|
other
| 51.12 |
Obtained results from evaluation show an interesting trend in decreasing both average and Hausdorff distances. Table 7 and Table 8 express the numerical evaluation based on average and Hausdorff OFs, respectively. Comparing with the best set of results (obtained by using RF with 65% sampling), a slight improvement is observed (1.048 mm for weighted RF vs. 1.052 mm for weightless). Note that both regressors are built using sampled dataset with the rate of 65%. Same decreasing trend is observed while the Hausdorff OF in considered (comparing 4.083 mm for weighted RF vs. 4.100 mm for weightless RF). This improvement certifies the assumption that the provision of a suitable weighting approach can lead the regression to predict post-surgery models with less distance evaluations. It should be noted that the weight assignment strategy has a significant impression to decrease the distance between predicted and post-surgery PCLs. Therefore, a new line of research is opened to investigate appropriate strategies to improve the prediction of breast shape using RF.
|
study
| 99.94 |
Besides the reported numerical evaluation, visual comparisons of three predicted breasts are depicted in Figure 16. The depicted predictions have been evaluated with average pair-wise distance of 1.62 mm (Figure 16a) as a poor prediction, 1.044 mm (Figure 16b) as a fair prediction, and 0.827 mm (Figure 16c) as a good prediction.
|
study
| 100.0 |
Although the displacement of the points plays an important role in finding the regression, the used clinical features contribute in adjusting the amount of prediction required for each point of healthy or damaged region. The construction of trees in RF allows to report the importance of each feature. In this regard, clinical studies can be accompanied with the resulted features’ importance to highlight the ones which contribute more in the prediction. Based on the conducted analysis, the clinical features applied in the training procedure are studied to determine their level of importance to construct the regressor. Table 9 denotes the importance of features in percentage, for the regressors trained in each of the three axes. Additionally, for the features belonging to the same group, not only the individual importance, but also the grouped importance (average of individuals) is reported, as well.
|
study
| 100.0 |
From the obtained results it is observed that the point coordinates has the most significant effect in prediction and as expected, each coordinate point, individually, contribute more in their own axis. The second remarkable feature is the distance to tumor which reflects the Euclidean distance of each point to the tumor (represented as a cylinder). Such trait was also relevant to the features’ behavior observed in clinical analysis, since points in larger distances from the tumor region are deformed less (see Section 3.1.2). Another interesting observed trend is the impression of the Z axis on the importance of the features. While the breast is more affected in the Z axis with respect to the tumor location (see Section 3.1), it is expected that features related with it, are more represented on that axis. Reported analysis of the two features, coordinate difference to the excised cylinder and polar distance to the excised cylinder satisfy the expected hypothesis. Far from expectations, it can be deduced that the tumor size feature not only influences the breast final deformation less than other clinical features, but also shows an irrelevant behaviour compared with the tumor region or the breast density features.
|
study
| 100.0 |
Additionally, to complete the investigation of features on the evaluation results, each clinical feature is assessed individually with respect to its impression of the best obtained prediction. Table 10 reports the results of individual evaluations on the three studied clinical features including breast density, tumor size, and tumor region. By looking to the breast density, it is observed that the average distance error decreases with a direct relation to the amount of the adipose tissue (Table 10), as was previously observed in Section 3.1.1. It can be deduced, breasts with more fibro-glandular tissue, are less affected. The results reported in Table 9 signifies the weakest role of the tumor size among clinical features; however, the outcome presented in Table 10 provide more evidence to evaluate the contribution of this feature. It is observed that large tumors imposes more deformation, which is also an expected result, based on the results obtained in Section 3.1.2. Regarding tumour region, it is difficult to conclude anything relevant, since, by looking to the obtained results, the magnitude of the deformation seem independent on the quadrant.
|
study
| 100.0 |
As in RF, in GBR a cross validation is performed with LOPO strategy, the following range for each parameter of the model:number of estimators (trees), in the rage of {5,10,…,500}; the maximum number of features in each tree in {2,3,…,23}; the leaf size in {1,2,…,5}; and the learning rate in the range of {0.01,0.02,…,1}. Besides, the criterion which measures the quality of a split is set friedman mse. The numerical evaluation are reported in Table 11 and Table 12, for both the average and Hausdorff OFs.
|
study
| 99.94 |
Numerical comparisons indicate that the average pair-wise distance obtained by GBR was 1.326 mm and 1.285 for the average global distance. The performance of GBR should be investigated through the definition of the learners. To keep the learners weak, constraints were imposed on the number of leaves and the depth of the trees. The assigned constraints to maintain the trees small, have led to propagate error in the whole sequence of model, which is noticeable through the increase of the errors in comparison with RF; however, more research could be performed in future in order to improve the understanding of this behaviour.
|
study
| 98.6 |
Discussed in Section 4.1.3, some of the most important features, presents a high correlation of each individual axis with each specific direction. This fact might influence the MOR to present poor predictions. More experiments should be performed in the future, but by the obtained results we can say that the point coordinates could play an independent role in predicting the displacement of each axis.
|
other
| 98.44 |
In this section two machine learning methodologies were studied; RF and GBR. The influence of sampling the data was studied taken into account the performance of the algorithm in terms of computational time and error of the prediction. Additionally, the use of adaptive weights were shown to have positive influence in the prediction. In any case, it might be difficult to understand the magnitude of the error, due to the lack of comparative approach, which could lead to the weak understanding of the obtained results. For this reason, a Heuristic Model (HM) was designed, taken into account the feature analysis performed in Section 3.1. With this simple and heuristic method, we could understand how difficult is to achieve a good result without using complex models, taken into account only the knowledge of the problem.
|
study
| 100.0 |
(5)disp¯i∈x,y,zq∈quadrants=1n∑i=1N(Pipost−Pipre) where Pipre and Pipost are corresponding points of pre- and post-surgery PCL, respectively, and N is the number of points in each PCL. The proposed HM follows the following strategy: the displacement of each point belonging to the healthy quadrants is computed based on the average displacement of each specific quadrant; for the points belonging to the unhealthy quadrant, the displacement is computed taken into account the average displacement of that quadrant, but multiplied both by {4,3,2,1} for the breast density ({A,B,C,D}, respectively), and by {3,2,1} for tumor size ({L,M,S}, respectively), following the knowledge obtained in Section 3.1 (see Equation (6)):(6)Pinew=Pi+dispi¯qPi∈healthybreastquadrantsPi+b×s×dispi¯qPi∈quadrantswhichcomprisesthetumor, where b, s, and dispi¯q are breast density, tumor size, and displacement of the equivalent breast quadrant, respectively, and finally, pinew denotes the calculated post-surgery PCL. The presented results of Table 15 are obtained by this this approach.
|
study
| 100.0 |
By comparing this simple approach with the learning models presented previously, we can state that the magnitude of the error obtained with RF is acceptable. The framework designed in this work, composed by data, features and models, presented some simplification regarding the reference framework from Vavourakis et al. , which could lead to some errors on the prediction, but with not so high influence in the visual aspect. Some new directions for this line of research are already planned trying to improve the obtained results, which will be presented in Conclusions Section.
|
study
| 99.75 |
Considering the characteristics of the problem, such as the type of features, and the demanded output, RF-based methodologies were primarily chosen as learning models for this problem. We started by studying the influence of data sampling for the performance of the model, and it was observed that less sampling generally leads to less errors on the prediction, but took to high computational time. Focusing on the pair-wise distance, the declining trend of errors stops when the sampling rate reaches to 65%. More sampling rates result in the errors oscillating in a the range of 0.1 mm. Although the difference errors between the distances in the range of 45% to 100% remains about 0.1 mm, the sampling rate with the minimum distance (1.048 mm) was chosen as the final sampling rate (65%). Same argument has been carried out to select the sampling rate of 75% when the OF is set to be Hausdorff distance. The minimum distance with respect to the Hausdorff OF is reported near to 4.022 mm.
|
study
| 100.0 |
The influence of weighting was also taken into account. Following the strategy of adaptive weight assignment, the points on training data PCLs were weighted based on the errors to the corresponding point in the target PCL. It should be noted that the trained model in each iteration is evaluated to determine the new weights for the next iteration. The obtained results highlight the effectiveness of using an adaptive weighting function, obtaining better results, even they are not so significant. Additional investigation should be performed in this part, in order to find the most suitable weighting function for this problem.
|
study
| 99.9 |
RF were compared with GBR methodology. As previously discussed, learners of the GBR was kept weak intentionally to refrain the greedy growth of the methodology. The imposed constraints kept the learner leaves less (or equal) than 3, and the learners depth less (or equal) than 5. Therefore, in both lines of OF, an increase of the distances were observed. However, it should be mentioned that the current strategy to consider the aforementioned constraints, limited the range of reported distances, in which the average and maximum distance in the two OFs approached to each other.
|
study
| 99.94 |
A MOR approach was also taken into consideration. Although the results showed that it fell behind the single output RF, this weak result might lay in the facts that not only the MOR trainer was unable to find a correlation between the points’ displacements of multiple coordinates, but also the features of different coordinates intruded the learner such that they cancel the impression of each others. The aforementioned hypothesis needs to be investigated more to determine the weak performance of MIMO methodology.
|
study
| 98.1 |
Finally, to close the discussion, it should be noted that all the regressors were design with a the same sampling rate. The RF with adaptive weights outperforms the other regressors with 1.048 mm and 4.083 mm for the average and the Hausdorff optimization, respectively. In the second rank, RF without consideration of adaptive weights stands with an average distance of 1.052 mm and maximum distance of 4.101 mm. Third and forth ranks were obtained by MOR (with average of 1.173 mm, and Hausdorff of 4.738 mm), and GBT(with average of 1.326 mm, and Hausdorff of 5.564 mm).
|
study
| 99.94 |
To overcome the nonexistence of dataset suited for learning breast healing deformations, an in-house dataset was generated using MRI data from real patients combined with a multiscale biomechanical and biochemical models to simulate post-surgical breast shape. Several clinical features that might impact breast healing deformations, including breast density, tumor region, and tumor size, were considered in the creation of the dataset that is representative of the population distributions.
|
study
| 100.0 |
The learning process was divided into two main tasks: feature inspection and model selection. In the first task, the complex interplay of clinical features for conditioning the breast shape after surgery healing was untangled by comparing the influence of different sets of features in the deformed breast after resulted from BCS. Here, it was concluded that breast density had the highest impact on the breast displacements while the quadrant where comprises the tumor was determining to predict the directions of shape adjustments. Further analyses of the importance of features resulted from learning process reinforced the experimental findings. In the model selection task, several regression methodologies, including Random Forest, Gradient Boosting Regression, and Multi-Output Regression were studied. Beside the learning model, a heuristic model was also proposed to validate the veracity of learning methodologies and understand the magnitude of the distances. Additional investigations were conducted with respect to sample PCL data, as well as a complementary study to assign different weights to the training instances, which resulted in predictions with slightly better evaluations. The numerical evaluations with the pair-wise and global distances indicated that the RF regressor constructed with the adaptive weighted training set outperformed MOR and GBR in both lines of average and Hausdorff OFs.
|
study
| 100.0 |
Improvement of results by assigning adaptive weights opens a new line of investigation for the future work, in order to define better mechanisms to improve the performance of regressor. Additionally, regression based on GBR can be improved by re-defining new learners and putting better constraints to keep the learners weak enough to avoid overfitting. Besides, more investigations are needed to study the correlations between the features in order to provide more evidence to the hypothesis of poor performance of MOR. As an alternative solution, it is worth studying a combined methodology of predicting breast deformation with machine learning techniques followed by some final steps of the biomechanical models; hence the required computational time for the biomechanical models is decreased since the predecessor machine learning methodology has provided a partial solution. Last but not least, new learning methodologies such as convolution neural network regression, and deep geometric learning can be investigated, as well as performing the prediction of breast shape directly on the 3D surface data.
|
study
| 90.25 |
Rheumatoid arthritis (RA) is the prototype of polyarticular inflammatory disease, affecting ~1% of the world population. Other forms of arthritis specifically in children affect a single or very few joints. Pigmented villonodular synovitis (PVNS) is a tumour that occurs inside the synovial membrane, with a high tendency of recurrence despite surgery. The use of biological drugs has been a major advance for the treatment of RA. However, ~30% of RA patients do not respond to these drugs, which are expensive and can cause severe side-effects1, 2. Intra-articular treatment with radio-isotopes for instance, has been effective in RA and PVNS but has major restrictions related to the use of radio-active material. There is therefore a necessity for improvement or alternatives in the local treatment of arthritis.
|
review
| 99.9 |
In the inflamed joint, the uncontrolled proliferation and accumulation of fibroblast-like synoviocytes (FLS) are the main cause of chronic inflammation and its progression to joint damage3, 4. This results in part from acquired molecular changes in FLS leading to reduced sensitivity to cell death signals. Apoptosis-inducing strategies targeting FLS have been considered as treatment of arthritis5–7. In vitro experiments using a plasmid vector to express the proapoptotic gene PUMA (p53 upregulated modulator of apoptosis) in FLS, showed the efficacy of PUMA in inducing cell apoptosis8, 9, a phenomenon which was independent of the p53 status of the synovium9. These preliminary data suggested that the strategy of PUMA-induced apoptosis in FLS could block the hyperplasia of the synovial intimal lining.
|
study
| 99.94 |
A variety of non-viral and viral vectors have been tested for the local and systemic treatment of rheumatic diseases by gene therapy10. The human adenovirus type 5 (HAdV5)-based vectors gave the best results, despite low efficiency in transduction of rheumatoid synovium in RA animal models11. HAdV5 infection is initiated by the attachment of the viral vector to its high-affinity receptor, the Coxsackie-adenovirus receptor (CAR), on the surface of cells12. However, human FLS do not express CAR on their surface and are thus poorly transduced by HAdV5 vectors13.
|
study
| 70.6 |
To overcome this problem of vector inefficiency, we design a novel gene delivery strategy, in which HAdV5-PUMA was ‘piggybacked’ on a baculovirus vector carrying CAR on its envelope14, resulting in the efficient cell entry of the vector BVCARHAdV5-PUMA into the FLS. We demonstrate in this study that PUMA gene transfer into FLS by BVCARHAdV5-PUMA results in rapid and extensive cell death by PUMA-induced apoptosis. The pro-apoptotic effect is not substantially reduced in the presence of proinflammatory cytokines, which mimic the environment of inflamed joints. Using the adjuvant-induced arthritis (AIA) rat model, we find that a single intra-articular injection of BVCARHAdV5-PUMA significantly decreases joint inflammation, and improves joint function with reduced joint damage and bone loss. The results of this study show that the intra-articular administration of a PUMA-expressing vector has therapeutic potential as a treatment for various forms of arthritis in which FLS proliferation is implicated.
|
study
| 99.94 |
The use of HAdV5 as a gene transfer vector for FLS has been limited due to their non-permissiveness to HAdV5 as they do not express CAR, the cellular receptor of HAdV5 on their surface. To overcome this hurdle, we design an efficient gene delivery strategy using a dual vector system by complexing HAdV5 to a baculovirus vector carrying CAR molecules on its envelope (BVCAR; Fig. 1a,b)14. The BVCARHAdV5 complex binds to the cell surface via the baculoviral envelope glycoprotein Gp64 (Fig. 1a,b), and is internalized by endocytosis, even in cells lacking CAR, such as FLS14. BVCAR serves as a transporter to bring the HAd5V vector into the target cells. Transduction of human and rat FLS with HAdV5-GFP alone, at a vector dose of 20 vector particles per cell (vp/cell), gave <2% of GFP-positive cells (Fig. 1c, d; grey bars), whereas more than 50% GFP-positive cells were obtained with the BVCARHAdV5-GFP complex at the same vector dose (Fig. 1c,d; grey bars). To mimic the inflammatory conditions in arthritic joints, FLSs were pre-treated with proinflammatory cytokines (TNFα, IL-17) before transduction with BVCARHAdV5-GFP. This resulted in almost 70% GFP-positive cells (Fig. 1c,d; black bars).Fig. 1Vector duo and transduction of human and rat synoviocytes. a The BVCARHAdV5 binary complex can be visualize by electron microscopy. The HAdV5 vector with a icosahedral shape, is bound to BVCAR, a rod-shaped baculovirus which has incorporated the CAR glycoproteins in its envelope. Scale bar, 100 nm. b The schematic representation shows the BVCARHAdV5 binary complex, with HAdV5 bound to BVCAR via the CAR glycoprotein at the baculoviral membrane. The baculoviral envelope glycoproteins Gp64 are localized at the head of the baculoviral particle. c Human and d rat fibroblast-like synoviocytes (FLS) non-treated (grey bars) or pre-treated (black bars) with the proinflammatory cytokines were transduced with HAdV5-GFP alone, or with the BVCARHAdV5-GFP complex. The figures represent the percentage of GFP-positive FLS cells analysed by flow cytometry. The average of three independent experiments, each in triplicates are shown. Statistical analysis using one-way ANOVA test, ***P < 0.001, **P < 0.01, *P < 0.05
|
study
| 100.0 |
Vector duo and transduction of human and rat synoviocytes. a The BVCARHAdV5 binary complex can be visualize by electron microscopy. The HAdV5 vector with a icosahedral shape, is bound to BVCAR, a rod-shaped baculovirus which has incorporated the CAR glycoproteins in its envelope. Scale bar, 100 nm. b The schematic representation shows the BVCARHAdV5 binary complex, with HAdV5 bound to BVCAR via the CAR glycoprotein at the baculoviral membrane. The baculoviral envelope glycoproteins Gp64 are localized at the head of the baculoviral particle. c Human and d rat fibroblast-like synoviocytes (FLS) non-treated (grey bars) or pre-treated (black bars) with the proinflammatory cytokines were transduced with HAdV5-GFP alone, or with the BVCARHAdV5-GFP complex. The figures represent the percentage of GFP-positive FLS cells analysed by flow cytometry. The average of three independent experiments, each in triplicates are shown. Statistical analysis using one-way ANOVA test, ***P < 0.001, **P < 0.01, *P < 0.05
|
study
| 100.0 |
An earlier study shows that transfection of FLSs with a plasmid expressing PUMA resulted in rapid apoptosis of the transfected cells in vitro 8, 9. In the present study, a HAdV5-based vector carrying the PUMA gene under the control of the cytomegalovirus (CMV) immediate-early promoter (HAdV5-PUMA) was tested in vitro in human and rat FLSs and evaluated for the cytological effects due to PUMA expression and a HAdV5-GFP vector is used as control.
|
study
| 100.0 |
Both HAdV5 vectors were used at the same vector dose of 30 vp/cell and in complex with BVCAR. No morphological changes were observed in both the mock-transduced and BVCARHAdV5-GFP-transduced FLS at 24 h post transduction (Fig. 2a,b,g,h). However, both human and rat FLS transduced with BVCARHAdV5-PUMA show massive cell death (Fig. 2c,i). To mimic the inflammatory conditions in arthritic joints, human and rat FLSs were pre-treated with proinflammatory cytokines TNFα and IL-17 before vector transduction (Fig. 2d,e,f,j,k,l). Similarly, massive cell death was observed in both types of FLS transduced with BVCARHAdV5-PUMA (Fig. 2f,l). This implied that cell death of both human and rat FLS could be induced by an HAdV5 vector expressing a proapoptotic gene such as PUMA, and achieved with the cooperation of a BVCAR vector. Since rat FLS are sensitive to PUMA-induced apoptosis, a rat in vivo model of arthritis could be evaluated with the BVCARHAdV5-PUMA vector.Fig. 2Cell death of human and rat FLS induced by BVCARHAdV5-PUMA. Cultures of human and rat FLS were non-treated a–c,g–i or treated d–f,j–l with proinflammatory cytokines, TNFα and IL-17. The FLS samples were mock-infected a,d,g,j, infected with BVCARHAdV5-GFP b,e,h,k or infected with BVCARHAdV5-PUMA c,f,i,l. Note the rounding-up of FLS after transduction with BVCARHAdV5-GFP in c,f,i,l. Scale bars, 20 μm
|
study
| 100.0 |
Cell death of human and rat FLS induced by BVCARHAdV5-PUMA. Cultures of human and rat FLS were non-treated a–c,g–i or treated d–f,j–l with proinflammatory cytokines, TNFα and IL-17. The FLS samples were mock-infected a,d,g,j, infected with BVCARHAdV5-GFP b,e,h,k or infected with BVCARHAdV5-PUMA c,f,i,l. Note the rounding-up of FLS after transduction with BVCARHAdV5-GFP in c,f,i,l. Scale bars, 20 μm
|
study
| 99.94 |
For clinical relevance, we evaluated FLS derived from synovium explants of three RA patients, for their sensitivity to PUMA-induced cell death. BVCARHAdV5-PUMA transduction was carried out at a constant vector dose of 50 vp/cell, with or without pretreatment with proinflammatory cytokines, and the cell survival measured by the MTT assay over a period of 40 h post infection. The cell survival curves from the three clinical samples show similar profiles, with a rapid cell death at 16 h post infection, followed by a more gradual effect until 40 h (Fig. 3a–c). The percentage of surviving cells at 40 h post transduction was 20% or less in all three samples, and minor differences in the cell survival curves observed in the presence or absence of cytokines (Fig. 3a–c). The results show that BVCARHAdV5-PUMA transduction of the FLS derived from the three clinical samples induced rapid cell death in all samples, with or without cytokines.Fig. 3BVCARHAdV5-PUMA-induced apoptosis of FLS derived from clinical samples. FLS derived from synovium explants from three patients with RA a–c, nontreated or treated with cytokines TNFα and IL-17, were infected with BVCARHAdV5-PUMA at a constant adenoviral vector dose of 50 vp/cell. The cell survival was monitored over 40 h, using the MTT assay. The number of surviving cells was expressed as the percentage of control cells taken at time 0 of infection, which was attributed the 100 per cent value. The average from three independent experiments, each performed in triplicates is presented. The error bars indicate standard error of the mean. Statistical analysis using one-way ANOVA test, ***P < 0.001, **P < 0.01, *P < 0.05. d The expression of PUMA protein was analysed in FLS transduced or not with BVCARHAdV5-PUMA at 50 vp/cell. Cells were harvested at 40 h post transduction, lysed and whole-cell lysates probed for PUMA protein using SDS–polyacrylamide gel electrophoresis and western blot analysis. (bottom) Luminogram of blot reacted with anti-PUMA RabMAb, peroxidase-conjugated anti-rabbit IgG antibody and enhanced chemiluminescence Detection Kit. (Top) Loading control (stained blot). Blot was reacted with anti-β-actin MoMAb, peroxidase-conjugated anti-mouse IgG antibody, and staining reaction performed with H202 and 3,3’-Diaminobenzidine. e To quantitatively measure cell apoptosis, FLS untreated or treated with individual cytokines TNFα and IL-17, or a mixture of TNFα and IL-17, were incubated with increasing doses of BVCARHAdV5-PUMA, and harvested at 24 h post infection. DNA fragmentation and nucleosome release into the cytoplasm were determined using an immunological detection of histone-complexed DNA fragments. Lysates from mock-infected cells served as negative controls to evaluate the background level, i.e., the physiological nucleosome content of the cytoplasm. The results are expressed as the fold change over the background level, which was attributed the value of 1. The control consisted of FLS infected by BVCARHAdV5-GFP at the highest dose of 50 vp/cell. Results presented are from three independent experiments, each in triplicates, using the one-way ANOVA test, ***P < 0.001, **P < 0.01, *P < 0.05
|
study
| 100.0 |
BVCARHAdV5-PUMA-induced apoptosis of FLS derived from clinical samples. FLS derived from synovium explants from three patients with RA a–c, nontreated or treated with cytokines TNFα and IL-17, were infected with BVCARHAdV5-PUMA at a constant adenoviral vector dose of 50 vp/cell. The cell survival was monitored over 40 h, using the MTT assay. The number of surviving cells was expressed as the percentage of control cells taken at time 0 of infection, which was attributed the 100 per cent value. The average from three independent experiments, each performed in triplicates is presented. The error bars indicate standard error of the mean. Statistical analysis using one-way ANOVA test, ***P < 0.001, **P < 0.01, *P < 0.05. d The expression of PUMA protein was analysed in FLS transduced or not with BVCARHAdV5-PUMA at 50 vp/cell. Cells were harvested at 40 h post transduction, lysed and whole-cell lysates probed for PUMA protein using SDS–polyacrylamide gel electrophoresis and western blot analysis. (bottom) Luminogram of blot reacted with anti-PUMA RabMAb, peroxidase-conjugated anti-rabbit IgG antibody and enhanced chemiluminescence Detection Kit. (Top) Loading control (stained blot). Blot was reacted with anti-β-actin MoMAb, peroxidase-conjugated anti-mouse IgG antibody, and staining reaction performed with H202 and 3,3’-Diaminobenzidine. e To quantitatively measure cell apoptosis, FLS untreated or treated with individual cytokines TNFα and IL-17, or a mixture of TNFα and IL-17, were incubated with increasing doses of BVCARHAdV5-PUMA, and harvested at 24 h post infection. DNA fragmentation and nucleosome release into the cytoplasm were determined using an immunological detection of histone-complexed DNA fragments. Lysates from mock-infected cells served as negative controls to evaluate the background level, i.e., the physiological nucleosome content of the cytoplasm. The results are expressed as the fold change over the background level, which was attributed the value of 1. The control consisted of FLS infected by BVCARHAdV5-GFP at the highest dose of 50 vp/cell. Results presented are from three independent experiments, each in triplicates, using the one-way ANOVA test, ***P < 0.001, **P < 0.01, *P < 0.05
|
study
| 100.0 |
Protein analysis of the three clinical samples taken before BVCARHAdV5-PUMA transduction show low basal levels of PUMA protein (Fig. 3d). After transduction of the FLS by BVCARHAdV5-PUMA, high expression of PUMA protein was detected in all three samples (Fig. 3d and Supplementary Fig. 1). The results strongly implied the role of PUMA protein expression in the induction of FLS cell death.
|
study
| 100.0 |
To further confirm that the cell death of human FLS transduced with BVCARHAdV5-PUMA was due to apoptosis, an enzyme-linked immunosorbent assay (ELISA) was performed to measure the degree of chromatin fragmentation and nucleosome released in the cell cytoplasm. Non-transduced FLSs served as the control of background cytoplasmic nucleosomes, whereas FLSs transduced with BVCARHAdV5-GFP served as control for cytotoxic effects of HAdV5 vector transduction. For FLS samples transduced with the lowest vector dose of 10 vp/cell BVCARHAdV5-PUMA, the amount of cytoplasmic nucleosomes was significantly higher compared to control samples (Fig. 3e), suggesting induction of FLS apoptosis. The level of cytoplasmic nucleosomes increased at higher vector doses, reaching a plateau at the relatively low dose of 20 vp/cell (Fig. 3e). A similar profile was obtained with FLS pre-treated with the cytokine TNFα (Fig. 3e). For FLS pre-treated with IL-17, individually or in combination with TNFα, the overall levels of cytoplasmic nucleosomes were lower, compared to those of non-treated or TNFα-treated FLS (Fig. 3e).
|
study
| 100.0 |
We next explored the effect of BVCARHAdV5-PUMA injection into the synovial cavity of rats with adjuvant-induced arthritis (AIA). This is an animal model of oligo-articular arthritis in which the ankles are particularly affected, with the development of arthritis 8–10 days after adjuvant injection. Rats were given different vector preparations for a single intra-articular injection at day 14 post induction, i.e., at the active stage of the disease, as would be the case in a clinical setting. Thirty rats were divided into six groups (three control and three therapeutic) of five rats each. Rats of control groups were injected with (i) BVCAR alone (105 plaque forming unit (PFU) per joint), (ii) BVCARHAdV5-null (105 PFU BVCAR + 109 PFU HAdV5-null per joint) and (iii) HAdV5-PUMA alone (109 PFU per joint). Rats of the three therapeutic groups were administered BVCAR (105 PFU per joint) complexed with HAdV5-PUMA at three different doses (107, 108 and 109 PFU per joint, respectively) and monitored for 4 days.
|
study
| 100.0 |
Rats which received the control vectors BVCAR or HAdV5-PUMA alone, or the BVCARHAdV5-null, showed no therapeutic effect on the expected course of arthritis (Fig. 4). However, rats treated with BVCARHAdV5-PUMA showed significant reduction of the ankle circumference in a vector dose-dependent manner (Fig. 4a). Likewise, there was a decrease in the ankle articular index score, although this was only significant for the group treated with the highest HAdV5-PUMA vector dose (Fig. 4b). The body weight of rats which received BVCARHAdV5-PUMA was superior to the values of control animals for all the vector doses tested (Fig. 4c). Collectively, the intra-articular administration of BVCARHAdV5-PUMA after arthritis onset showed beneficial effect as early as 4 days after injection, with significant reduction of the ankle circumference indicating decreased joint inflammation.Fig. 4BVCARHAdV5-PUMA-mediated reduction of joint inflammation. Rats develop arthritis 8–10 days after adjuvant injection. After the onset of arthritis, at day 14 post injection, vectors (10 μl) were delivered intra-articularly in each ankle. Thirty rats were enrolled and divided into six groups (five rats per group). Three control groups consisted of BVCAR alone (105 PFU per joint), BVCAR (105 PFU per joint) complexed with an empty adenoviral vector (HAdV5-null; 109 PFU per joint), and HAdV5-PUMA alone (109 PFU per joint). The three therapeutic groups consisted of BVCAR (105 PFU per joint) complexed with HAdV5-PUMA at increasing concentrations (107, 108 and 109 PFU per joint). Values represented in the bar graphs are the means ± s.e.m. Differences (Delta) in the biological parameters are defined by the values obtained on the day of follow-up minus the values obtained on the day of intra-articular injection. a Differences in ankle circumferences. Values of ankle circumferences were obtained from perpendicular caliper measurements of ankle diameter, using a geometric formula. b Differences in ankle articular index scores. c Differences in rat body weights. D day after intraarticular injection of vector; *P < 0.05 vs BVCAR alone; $ P < 0.05 vs BVCARHAdV5-null; # P < 0.05 vs HAdV5-PUMA alone; § P < 0.05 vs BVCARHAdV5-PUMA (109 PFU per joint); ♮ P < 0.05 for ANOVA one-way test
|
study
| 100.0 |
BVCARHAdV5-PUMA-mediated reduction of joint inflammation. Rats develop arthritis 8–10 days after adjuvant injection. After the onset of arthritis, at day 14 post injection, vectors (10 μl) were delivered intra-articularly in each ankle. Thirty rats were enrolled and divided into six groups (five rats per group). Three control groups consisted of BVCAR alone (105 PFU per joint), BVCAR (105 PFU per joint) complexed with an empty adenoviral vector (HAdV5-null; 109 PFU per joint), and HAdV5-PUMA alone (109 PFU per joint). The three therapeutic groups consisted of BVCAR (105 PFU per joint) complexed with HAdV5-PUMA at increasing concentrations (107, 108 and 109 PFU per joint). Values represented in the bar graphs are the means ± s.e.m. Differences (Delta) in the biological parameters are defined by the values obtained on the day of follow-up minus the values obtained on the day of intra-articular injection. a Differences in ankle circumferences. Values of ankle circumferences were obtained from perpendicular caliper measurements of ankle diameter, using a geometric formula. b Differences in ankle articular index scores. c Differences in rat body weights. D day after intraarticular injection of vector; *P < 0.05 vs BVCAR alone; $ P < 0.05 vs BVCARHAdV5-null; # P < 0.05 vs HAdV5-PUMA alone; § P < 0.05 vs BVCARHAdV5-PUMA (109 PFU per joint); ♮ P < 0.05 for ANOVA one-way test
|
study
| 100.0 |
Peri-articular bone damage was assessed using micro-computed tomography (µ-CT). The three-dimensional images allow an overall evaluation of bone loss in the ankles, while the parasagittal slices would provide more refined details. When compared to the healthy ankle (Fig. 5a), substantial bone damage was observed in the joints injected with BVCAR alone (Fig. 5b), and extensive bone loss was also visible in the joints treated with HAdV5-PUMA alone (Fig. 5c). In contrast, only minor morphological changes were observed in joints administered with BVCARHAdV5-PUMA (Fig. 5d,e), as compared to ankle joints from healthy rats (Fig. 5c). The only bone loss marker detectable in BVCARHAdV5-PUMA-treated joints was a modification in porosity, characteristic of the early phase of arthritis before the administration of any treatment. Further microarchitecture alterations consecutive to extended inflammation, was not observed in the BVCARHAdV5-PUMA-treated joints.Fig. 5Analysis of rat ankle tissues. Ankle tissue imaging by micro-computed tomography (µ-CT) was performed on rat ankle joints from a a healthy rat, rats injected with b BVCAR alone, c HAdV5-PUMA alone and animals injected with BVCAR HAdV5-PUMA at vector doses of d 107 PFU per joint, and e 109 PFU per joint respectively. (left) Parasagittal three-dimensional views of the whole rat ankles. (right) The enlarged images of the µ-CT parasagittal slices showing, from left to right, the intermediate cuneiform, navicular and distal talus. The white arrows point at local bone erosions. Scale bars, 1 mm. f Histological analysis and quantification of infiltrates was also performed. The bar graph shows the number of infiltrating cells, expressed as the ratio to their number in BVCAR control samples (mean ± s.e.m.; data from 30 slides per group). ANOVA: analysis of variance; §§§: ANOVA, P < 0.001; ***P < 0.0001 by Newman–Keuls post-hoc tests. Note that (i) BVCARHAdV5-PUMA strongly reduced the density of inflammatory cells in the joints, compared to control BVCAR or HAdV5-PUMA; (ii) BVCARHAdV5-PUMA was more effective at 109 PFU, compared to 107 PFU, demonstrating a dose-dependent effect
|
study
| 100.0 |
Analysis of rat ankle tissues. Ankle tissue imaging by micro-computed tomography (µ-CT) was performed on rat ankle joints from a a healthy rat, rats injected with b BVCAR alone, c HAdV5-PUMA alone and animals injected with BVCAR HAdV5-PUMA at vector doses of d 107 PFU per joint, and e 109 PFU per joint respectively. (left) Parasagittal three-dimensional views of the whole rat ankles. (right) The enlarged images of the µ-CT parasagittal slices showing, from left to right, the intermediate cuneiform, navicular and distal talus. The white arrows point at local bone erosions. Scale bars, 1 mm. f Histological analysis and quantification of infiltrates was also performed. The bar graph shows the number of infiltrating cells, expressed as the ratio to their number in BVCAR control samples (mean ± s.e.m.; data from 30 slides per group). ANOVA: analysis of variance; §§§: ANOVA, P < 0.001; ***P < 0.0001 by Newman–Keuls post-hoc tests. Note that (i) BVCARHAdV5-PUMA strongly reduced the density of inflammatory cells in the joints, compared to control BVCAR or HAdV5-PUMA; (ii) BVCARHAdV5-PUMA was more effective at 109 PFU, compared to 107 PFU, demonstrating a dose-dependent effect
|
study
| 100.0 |
Synovium changes were evaluated by light microscopy of fresh-frozen sections of the joints of euthanized animals, and conventional staining with haematoxylin and eosin (H&E). In comparison to the control joints treated with BVCAR or HAdV5-PUMA alone, the infiltrate was significantly reduced in the BVCARHAdV5-PUMA-treated joints (Fig. 5f). This effect was vector dose-dependent, with a lower number of infiltrating cells in ankle joints treated with 109 PFU compared to 107 PFU of BVCARHAdV5-PUMA (Figs 5f and 6).Fig. 6Histological analysis of ankle joints. Fresh-frozen sections (10 µm) of joints treated respectively with: a, BVCAR alone; b, HAdV5-PUMA alone (109 PFU); c, BVCARHAdV5-PUMA (107 PFU); d, BVCARHAdV5-PUMA (109 PFU); were stained with haematoxylin and eosin staining, and compared for the presence of synovial tissue infiltrates. The representative photograph from each group is provided at various magnifications (×200, left; ×400, middle; and ×800, right). The righmost panels are enlargements of the rectangular area delineated in the middle panels. Scale bar, 200 µm
|
study
| 100.0 |
Histological analysis of ankle joints. Fresh-frozen sections (10 µm) of joints treated respectively with: a, BVCAR alone; b, HAdV5-PUMA alone (109 PFU); c, BVCARHAdV5-PUMA (107 PFU); d, BVCARHAdV5-PUMA (109 PFU); were stained with haematoxylin and eosin staining, and compared for the presence of synovial tissue infiltrates. The representative photograph from each group is provided at various magnifications (×200, left; ×400, middle; and ×800, right). The righmost panels are enlargements of the rectangular area delineated in the middle panels. Scale bar, 200 µm
|
study
| 99.94 |
To assess the long-term therapeutic effect, an animal experiment was performed with rats treated with BVCARHAdV5-PUMA and HAdV5-PUMA (109 PFU per joint), and the study period extended up to 21 days post injection. Following restrictions by the animal ethic committee, the negative control groups (BVCAR and BVCARHAdV5-null) were not included in this long-term study due to the high level of arthritis severity.
|
study
| 100.0 |
The results show that the articular index decreased from day 2 to day 21 post-intra-articular injection (Fig. 7a). A similar pattern was observed for the mobility index (Fig. 7b) and ankle circumference (Fig. 7c). In comparison to the rats which received HAdV5-PUMA, the mobility of the BVCARHAdV5-PUMA-treated animals improved progressively from day 8 to day 21 and their body weight improved rapidly from day 3, remaining significantly higher until the end of the study (Fig. 7d). All the parameters measured confirmed the therapeutic benefits of BVCARHAdV5-PUMA treatment with improvement of the joint function, and absence or minimal joint and bone damage, as observed in histopathology and micro-computed tomography. Control animals administered with HAdV5-PUMA alone showed no detectable beneficial effect.Fig. 7Long term in vivo and in vitro analysis of PUMA gene transfer. Rat ankle joints were injected with HAdV5-PUMA alone, or BVCARHAdV5-PUMA at a vector dose of 109 HAdV5-PUMA per joint, after the onset of arthritis at day 14. The parameters monitored over a period of 21 days were the a ankle articular index, b loss of function index, c ankle circumferences and d body weights. Histology was performed on undecalcified samples with various staining. Tartrate-resistant acid phosphatase (TRAP) staining provided red colour in multinucleated osteoclasts on the bone counter-stained with Anilin blue e. Compared to HAdV5-PUMA group, TRAP+ osteoclast staining with the TRAP was reduced at several sites of the ankle area, including distal tibia, talus, calcaneus and navicular bones in the HAdV5-PUMA group. Arrows showed important TRAP+ osteoclast concentration in the bones. In the same sites, the bone matrix was stained with Goldner trichrome staining f. Mineralized matrix was stained in green, while non-mineralized matrix was stained in red and the marrow in yellow. Arrows showed loss of bone mineral. More mineralized trabeculae and less non-mineralized areas were observed in the bones of the BVCARHAdV5-PUMA group in comparison with the HAdV5-PUMA group f. Then, Safranin O–Fast green staining was performed with bone tissue staining in turquoise and cartilage staining in red g. Compared to HAdV5-PUMA, cartilage thickness and its regularity were preserved in the BVCARHAdV5-PUMA group. Arrows indicated the loss of mineral in the bones. Inflammatory infiltrate (indicated by the ‘i’) was importantly diminished in the HAdV5-PUMA group and the bone cavity conserved its histological integrity. Scale bar, 500 µm
|
study
| 100.0 |
Long term in vivo and in vitro analysis of PUMA gene transfer. Rat ankle joints were injected with HAdV5-PUMA alone, or BVCARHAdV5-PUMA at a vector dose of 109 HAdV5-PUMA per joint, after the onset of arthritis at day 14. The parameters monitored over a period of 21 days were the a ankle articular index, b loss of function index, c ankle circumferences and d body weights. Histology was performed on undecalcified samples with various staining. Tartrate-resistant acid phosphatase (TRAP) staining provided red colour in multinucleated osteoclasts on the bone counter-stained with Anilin blue e. Compared to HAdV5-PUMA group, TRAP+ osteoclast staining with the TRAP was reduced at several sites of the ankle area, including distal tibia, talus, calcaneus and navicular bones in the HAdV5-PUMA group. Arrows showed important TRAP+ osteoclast concentration in the bones. In the same sites, the bone matrix was stained with Goldner trichrome staining f. Mineralized matrix was stained in green, while non-mineralized matrix was stained in red and the marrow in yellow. Arrows showed loss of bone mineral. More mineralized trabeculae and less non-mineralized areas were observed in the bones of the BVCARHAdV5-PUMA group in comparison with the HAdV5-PUMA group f. Then, Safranin O–Fast green staining was performed with bone tissue staining in turquoise and cartilage staining in red g. Compared to HAdV5-PUMA, cartilage thickness and its regularity were preserved in the BVCARHAdV5-PUMA group. Arrows indicated the loss of mineral in the bones. Inflammatory infiltrate (indicated by the ‘i’) was importantly diminished in the HAdV5-PUMA group and the bone cavity conserved its histological integrity. Scale bar, 500 µm
|
study
| 100.0 |
Histo-morphometry analysis was performed on the treated joints, to study the protective effects of BVCARHAdV5-PUMA treatment on the bone and cartilage. The tartrate-resistant acid phosphatase+ (TRAP+) osteoclast staining was reduced at several sites of the ankle area, including distal tibia, talus, calcaneus and navicular bones (Fig. 7e). Two areas remained TRAP+ in distal tibia and calcaneus, corresponding to the remodeling of growth plates, which are active in rats at this age. Elsewhere, pathologic osteoclastogenesis was abolished in the BVCARHAdV5-PUMA group. Mineralized bone was also protected, in the same sites as described with less TRAP+ osteoclasts (Fig. 7f). More mineralized trabeculae and less non-mineralized areas were observed in the bones of the BVCARHAdV5-PUMA group in comparison with the HAdV5-PUMA group. Finally, the regularity and the thickness of the cartilage were preserved in the BVCARHAdV5-PUMA group, especially at the talo-tibial joint site (Fig. 7g).
|
study
| 100.0 |
To assess if the HAdV5-PUMA and BVCARHAdV5-PUMA vectors remain localized within or escaped from the injected joints, we performed PCR assays on total DNA extracted from liver samples and sera from rats killed at day 4 and day 21 after treatment. No amplification of either adenovirus or baculovirus genome were obtained from the liver and serum samples. These results implied that there was no detectable dissemination of either vectors from the injected joints.
|
study
| 100.0 |
To address the persistence of the vectors in the injected joints, synovial tissues from animals injected with HAdV5-PUMA and BVCARHAdV5-PUMA and sacrificed at day 21 were analysed. Synovial tissues from four individual rats from each group were chosen at random and the total DNA extracted. PCR assays, performed using primers recognizing the hexon gene of HAdV5, amplified the expected hexon band of 465 bp in all the samples (Fig. 8a; arrowhead). However, no specific baculovirus band of 1.1 kbp, corresponding to a sequence within the baculovirus CAR gene, was detected in any of the synovium samples (Fig. 8b). These results implied the persistence of the HAdV5-PUMA genome in the synovial tissues at least until day 21 post treatment, but not that of the BVCAR genome.Fig. 8Analysis of vector persistence and immune response to vectors. a The HAdV5-PUMA and BVCAR DNA genomes in synovial tissues were extracted from synovial tissues samples of rats killed at day 21 after intra-articular administration of HAdV5-PUMA alone (B2, B4, B5 and B11) or BVCARHAdV5-PUMA (B10, B12, B13 and B14), and used for PCR analysis to detect the viral genomes. a The 465-bp amplified fragment corresponding to the HAdV5 hexon gene (a, arrowhead) was found in all the samples and in the positive control. Note the occurrence of a non-specific amplified fragment of 1500 bp (black dots). b, For the BVCAR genome, no specific band of 1.1 kbp, corresponding to the baculovirus CAR gene was detected in any of the samples, but was present in the positive control. b To evaluate the immune response of the treated animals to BVCAR and HAdV5-PUMA vectors, the sera of the animals treated for 4 days and 21 days with HAdV5-PUMA alone or with BVCAR HAdV5-PUMA were analysed by ELISA for the presence of antibodies against adenovirus and baculovirus. A single serum dilution (1:10) was used for all samples. No significant adenovirus or baculovirus antibodies were detected in the sera of the animals treated for 4 days (grey bars, c,d). In the animal group treated for 21-days, adenovirus antibodies were detected in the sera of 5/8 animals (black bars, c) and baculovirus antibodies detected in the sera of 2/8 animals (black bars, d). Similar results were obtained in an independent experiment
|
study
| 100.0 |
Analysis of vector persistence and immune response to vectors. a The HAdV5-PUMA and BVCAR DNA genomes in synovial tissues were extracted from synovial tissues samples of rats killed at day 21 after intra-articular administration of HAdV5-PUMA alone (B2, B4, B5 and B11) or BVCARHAdV5-PUMA (B10, B12, B13 and B14), and used for PCR analysis to detect the viral genomes. a The 465-bp amplified fragment corresponding to the HAdV5 hexon gene (a, arrowhead) was found in all the samples and in the positive control. Note the occurrence of a non-specific amplified fragment of 1500 bp (black dots). b, For the BVCAR genome, no specific band of 1.1 kbp, corresponding to the baculovirus CAR gene was detected in any of the samples, but was present in the positive control. b To evaluate the immune response of the treated animals to BVCAR and HAdV5-PUMA vectors, the sera of the animals treated for 4 days and 21 days with HAdV5-PUMA alone or with BVCAR HAdV5-PUMA were analysed by ELISA for the presence of antibodies against adenovirus and baculovirus. A single serum dilution (1:10) was used for all samples. No significant adenovirus or baculovirus antibodies were detected in the sera of the animals treated for 4 days (grey bars, c,d). In the animal group treated for 21-days, adenovirus antibodies were detected in the sera of 5/8 animals (black bars, c) and baculovirus antibodies detected in the sera of 2/8 animals (black bars, d). Similar results were obtained in an independent experiment
|
study
| 100.0 |
To evaluate if there was immune response to the viral vectors, sera from rats treated for 4 days and 21 days with BVCAR, HAdV5-PUMA or BVCARHAdV5-PUMA, were analysed by ELISA for the presence of antibodies to HAdV5 and BVCAR. The results show that antibodies against both HAdV5 and BVCAR were either absent or not detectable from the day-4 sera (Fig. 8c,d). However, from the day-21 sera, anti-HAdV5 antibodies were detected in 2/4 animals of the HAdV5-PUMA group, and in 3/4 animals of the BVCARHAdV5-PUMA group (Fig. 8c). Low levels of anti-BV antibodies were detected in the sera of 2/4 animals of the BVCARHAdV5-PUMA group (Fig. 8d). Although no viral genomes were detectable in the sera and tissues, the presence of antibodies against the viral vectors in certain individual animals suggested that the immune response observed could be due some vector leakage from the treated joints or induction of Ig by activated B cells inside the synovium15.
|
study
| 100.0 |
The aim of this study is to show the feasibility of controlling arthritis severity by an apoptosis-inducing gene therapy strategy targeting the FLSs in arthritic joints. HAdV5 vectors have been widely used as gene transfer vectors for cancer therapy. However, the use of HAdV5 vectors for arthritis treatment are limited as FLSs are poorly transduced and attempts to improve the gene transfer efficacy by genetic modification of the vector were not satisfactory13, 16. The innovation in the present study is the development of a strategy for efficient HAdV5 gene transfer to FLS coupled to a potent apoptosis-inducing gene for the treatment of chronic synovitis.
|
study
| 99.94 |
Our strategy consists of an HAdV5 vector carrying a proapoptotic gene PUMA, complexed on recombinant baculovirus (BV) displaying CAR on the baculoviral envelope (BVCAR)14. The BVCARHAdV5-PUMA complex exploits the capacity of the BVs to enter FLSs, resulting in efficient HAdV5 transduction of FLS in vitro and in vivo. Both the HAdV5 and BVs are non-integrative vectors and are considered biologically safe. BVs have a biosafety profile due to their incapacity to replicate in mammalian cells17, 18.
|
study
| 100.0 |
The functionality and efficiency of our strategy for PUMA gene delivery by an HAdV5-PUMA vector was demonstrated in vitro as well as in vivo in a rat AIA model. Gene transfer of PUMA into FLSs by BVCARHAdV5-PUMA, results in a rapid and massive cell death of FLS in vitro. In the context of chronic joint inflammation, the persistence of the disease results in part from the reduced sensitivity of FLS to death signals19. We also show that the FLS death cell induced by the HAdV5-PUMA vector is not inhibited in the presence of inflammatory cytokines. Using the rat AIA model, a single intra-articular injection of BVCARHAdV5-PUMA decreases significantly joint inflammation and improves joint function, resulting in minimal joint damage and bone loss. The two videos (Supplementary Movies 1 and 2) show that the treatment not only restores the joint functions and ambulation but also the mannerisms of the treated animal. The HAdV5-PUMA vector remains detectable in the injected joints 21 days post treatment, suggesting that the PUMA therapeutic effect can prolong in the joints beyond the observation period. The induction of antibodies against the adenovirus vector and its effect on the efficacy of intra-articulation treatment in humans is unknown. However, the possible adverse effects should be limited if a single therapeutic administration is proposed. Alternatively, chimeric adenovirus vectors with capsid proteins from other serotypes could replace the conventional HAdV5 vector to escape the circulating and local adenovirus neutralizing antibodies.
|
study
| 100.0 |
The intra-articular treatment options today are limited to intra-articular steroids with only transient anti-inflammatory effects. Other options such as radiochemical synovectomy with radio-active isotopes or osmic acid are not easily or not at all available. Our results show that local intra-articular treatment with the administration of a PUMA-expressing vector represents an efficient strategy for the control of abnormal synoviocyte proliferation. In addition to the local protective effect, the control of local inflammation appears to have a systemic effect. Nevertheless, this apoptosis-based therapeutic strategy has potential as an alternative to surgical or radiochemical synovectomy in a long list of arthritis conditions including RA, mono- or oligo-articular arthritis which includes juvenile oligo articular arthritis, inflammatory osteoarthritis, or the orphan disease pigmented-villo-nodular synovitis. A first proof of concept trial to treat this cancer-like disease by inducing targeted cell death may help in preventing the high rate of recurrences.
|
study
| 92.25 |
The use of human patient biological samples (FLS) in this study was approved by the Institutional Review Board of the Lyon University Hospital and the French Ministry of Education and Research (Authorization #AC-2010-1164). Informed consent was obtained from the patients. Animal experiments were performed in full compliance with the Ethical Committee Guidelines and Regulations for Animal Protection of the Loire in accordance with the legislation of the European Community, and received approval from the Ethical Committee for Animal Experiments of the University Jean Monnet in Saint-Etienne (Authorization #CU-14N1402).
|
other
| 99.94 |
Human embryonic kidney cells HEK293 cells were obtained from the American Type Culture Collection (ATCC, Manassas, VA), and maintained in Dulbecco’s modified Eagle’s medium (DMEM; Life Technologies) supplemented with 10% fetal bovine serum (FBS; Life Technologies), penicillin (100 U ml−1), and streptomycin (100 mg ml−1) at 37 °C, 5% CO2. Spodoptera frugiperda cells (purchased from Life Technologies) were maintained at 28 °C in Grace insect medium supplemented with 10% FBS and penicillin and streptomycin (Life Technologies). FLS were obtained from synovial tissue of RA patients undergoing joint surgery who fulfilled the ACR/EULAR criteria for RA20. Human and rat FLS were isolated by enzyme digestion and cultured in DMEM supplemented with 10% FBS and antibiotics and used between passages 4 and 9, as previously described13. All cell lines were routinely checked for mycoplasma every 6 months.
|
study
| 100.0 |
Mouse monoclonal antibody (MAb) against CAR (clone E1.1)21 obtained from Silvio Hemmi (University of Zürich, Switzerland) was used at dilution 1:50. MAb against the baculoviral glycoprotein Gp64 (clone AcV1) used at dilution 1:50, were purchased from Santa Cruz Biotechnology. Rabbit MAb against PUMA (clone EP512Y) used at dilution 1:1000 (Epitomics Inc.) was purchased from Abcam (Cambridge, UK). MAb against β-actin-peroxidase (clone AC-15) from Sigma-Aldrich was used at dilution 1: 10,000. TNFα was used at 1 ng ml−1 (Biosource, Camarillo, CA), IL-17A at 50 ng ml−1 (R&D Systems, Minneapolis, MN).
|
other
| 99.9 |
The replication-defective (E1-deleted) HAdV5-based vectors expressing the green fluorescent protein (HAdV5-GFP) and the human PUMA−β protein (HAdV5-PUMA) under the control of the CMV immediate-early promoter were propagated in HEK293 cells. The HAdV5-null vector, harboring no transgene in the E1-deleted region of its genome, was purchased from GeneCust (Dudelange, Luxembourg). The vector stocks were prepared and purified by CsCl gradient ultracentrifugation according to conventional methods14, 22, 23. The recombinant baculovirus (BV) vector expressing CAR (BVCAR) was derived from the Autographa californica Multiple Nucleopolyhedrosis virus (AcMNPV) by inserting the human CAR-encoding sequence into the Nhe I and Kpn I cloning sites of pBlueBac (Life Technologies) under the control of the polyhedrin promoter14. The BV vector expressing GFP under the CMV promoter (BV-GFP) was a kind gift of Norman J. Maitland (University of York, UK). BV vectors were propagated by infection of Sf9 cells at a multiplicity of infection (MOI) of 1–514. The infected cell culture media were harvested at 48 h post-infection and clarified by centrifugation for 10 min at 2400 r.p.m. Concentrated stocks of BV vectors were prepared by ultracentrifugation of the infected cell supernatant through a 20% sucrose cushion at 30,000 r.p.m. for 1 h at 4 °C. The viral pellet was then resuspended in sterile phosphate-buffered saline (PBS) with gentle shaking overnight at 4 °C, and used for transduction assays.
|
study
| 100.0 |
The infectious titre of a virus is defined as the number of infectious virions, determined by the plaque assay method and expressed as (PFU) per ml. In our transduction assays, cells were prepared in 24-well plates containing 105 cells per well. Complexes of adenovirus and baculovirus vectors in the ratio 5:100 PFU respectively, were prepared by preincubation of each virus in a total volume of 100 μl of cell culture media for 1 h at 37 °C. The complexes were then added to the cells and incubated for another hour at 37 °C, after which 200 μl prewarmed media was added to each well and incubated at 37 °C.
|
study
| 100.0 |
Twenty-four hours post transduction, transduced cells were observed directly from the tissue culture plate using a Zeiss Axiovert inverted microscope (Zeiss). Images were taken using an Axiovert digital camera and analysed using an Axio Vision program (Zeiss).
|
other
| 99.25 |
Vector samples were diluted in 0.14 M NaCl, 0.05 M Tris-HCl buffer pH 8.2 (Tris-buffered saline, TBS), and adsorbed onto carbon-coated Formvar membranes on grids, and incubated with primary antibody (monoclonal anti-CAR or anti-gp64 antibody) at a dilution of 1:50 in TBS for 1 h at room temperature (RT). After rinsing with TBS, the grids were postincubated with 20-nm colloidal gold-tagged goat anti-mouse immunoglobulin G antibody (British Biocell International Ltd., Cardiff, UK; diluted to 1:50 in TBS) for 30 min at RT. After rinsing with TBS, the specimens were negatively stained with 1% uranyl acetate in H2O for 1 min at RT, rinsed again with TBS, and examined under a JEM1400 JEOL electron microscope equipped with an Orius-Gatan digitalized camera (Gatan, Grandchamp, France).
|
study
| 99.94 |
Cells grown in 96-well flat-bottom plates were transduced with our different vectors and analysed at 36 h post-transduction. The cell culture media was then removed, and 30 μl of MTT solution (7.5 mg ml−1 thiazolyl blue tetrazolium bromide (Sigma-Aldrich) in phosphate buffered saline) was added per well, followed by incubation at 37 °C for 4 h. The MTT solution was then removed and 100 μl of DMSO (dimethylsulfoxide, Sigma-Aldrich) was added to each well. The optical density of the supernatants in the 96-well plates was read at 570 nm.
|
study
| 99.94 |
The cell apoptosis assay was carried out using a Cell Death Detection ELISA kit (Roche Diagnostics). Briefly, aliquots of FLS (105 cells) were harvested at different times post-infection and analysed by immunodetection to measure the degree of release of histone-complexed DNA fragments from each sample, according to the manufacturer’s recommendations.
|
study
| 99.94 |
Total nucleic acids were extracted from rat sera or synovium, using the NucleoSpin Tissue kit from Macherey Nagel, following the manufacturer’s protocol. The presence of adenovirus and baculovirus DNA genomes in the rat sera and synovium were then evaluated by PCR using primers specific for each virus. For the HAdV5, the forward primer 5′-GCTGTATTTGCCCGAC-3′ and reverse primer 5′-CATGGCCTCAAGCGTG-3′, amplified a fragment of 465 nt of the hexon gene. For the baculovirus, the forward primer 5′-GCTGTATTTGCCCGAC-3′ and reverse primer 5′-CATGGCCTCAAGCGTG-3′, amplified a fragment of 1095 nt of the baculovirus-CAR gene.
|
study
| 100.0 |
Purified HAdV5-PUMA in PBS were dissociated by incubation with 0.05% deoxycholate at 56 °C for 2 mins, while BVCAR virions in PBS were dissociated by a treatment of four cycles of freeze-thawing. Microtitre plates were coated with 100 μl of the viral proteins (1 μg ml−1) diluted in PBS and left overnight at 4 °C in a moisture chamber. The coated wells were washed four times with PBS and incubated with 200 μl blocking solution (2% bovine serum albumin in TBS) for 1 h at RT. After washing, 100 μl of each sera was added to the wells and incubated for 1 h at RT. Plates were then washed and incubated with anti-rat-horseradish peroxidase diluted to 1: 15,000 in blocking solution for 1 h at RT. After washing, 100 μl of OPD (o-phenylenediamine, Sigma-Aldrich) substrate was added, and the reaction was blocked by addition of 1 N HCl. OD was measured at 450 nm using an ELISA plate reader (Wallac 1420 Multilabel Counter, PerkinElmer).
|
study
| 99.94 |
Polyacrylamide gel electrophoresis of SDS-denatured protein samples, and immunoblotting analysis have been described in detail in previous studies14, 22–25. Briefly, proteins were electrophoresed in SDS-denaturing, 10%-polyacrylamide gel, along with prestained protein markers (PageRulerTM prestained protein ladder; Fermentas Inc., Hanover, MD), and electrically transferred to nitrocellulose membrane (HybondTM-C-extra; GE Healthcare Bio-Sciences). Blots were blocked in 5% skimmed milk in TBS containing 0.05% Tween-20 (TBS-T), rinsed in TBS-T, then successively incubated with primary antibody anti-PUMA protein (RabMAb; working dilution 1:1000) and peroxidase-conjugated secondary antibody (goat anti-rabbit IgG antibody; Sigma, St Louis, MO; working dilution 1:10,000), followed by enhanced chemiluminescence detection (Pierce Chemicals, Thermo Fisher Scientific Inc. IL).
|
study
| 99.94 |
Female Lewis rats weighing ~100 g (Janvier Laboratories, Saint-Germain-sur-l’Arbresle, France), were injected subcutaneously at the base of the tail with 300 µl (5 mg ml−1) of lyophilized Mycobacterium butyricum (Difco Laboratories, Detroit, MI), as previously described26. In this model, the first signs of joint inflammation and pain appear after day-8 of induction and become maximal during days 14–1826.
|
study
| 100.0 |
The various vectors were injected (intra-articular, 50 µl per ankle) 14 days after arthritis induction, defined as day 0. The first animal experiment was designed to assess short-term efficacy and to determine the best dose of vector. Thirty rats were divided into six groups (n = 5 per group). Three control groups were recombinant baculovirus alone (BVCAR; 105 PFU per joint), adenovirus vector harbouring the PUMA gene alone (HAdV5-PUMA; 109 PFU per joint), and BVCAR (105 PFU per joint) in combination with an empty adenoviral vector (HAdV5-null; 109 PFU per joint). The three therapeutic groups associated BVCAR (105 PFU per joint) with HAdV5-PUMA at three concentrations (107, 108 and 109 PFU per joint, respectively). In this experiment, the animals were euthanized on day 4 post-intra-articular injection.
|
study
| 100.0 |
The second animal experiment assessed the long-term efficacy of BVCARHAdV5-PUMA during 21 days compared to HAdV5-PUMA alone (both 109 PFU per joint; n = 7 per group). The clinical parameters monitored during these experiments, including articular index, mobility index, ankle circumference and body weight. Articular index scores were recorded for each hind joint by a consistent observer blinded to the treatment regimen and then averaged for each animal. Scoring was performed on a 0–4 scale where 0 = no swelling or erythema, 1 = slight swelling and/or erythema, 2 = low-to-moderate oedema, 3 = pronounced oedema with limited joint usage and 4 = excess oedema with joint rigidity. Mobility index were also recorded in the same way and provided limb function with a scale from 0 to 4 where 0 = regular mobility, 1 = gait irregularities, intermittent lameness, 2 = continuous lameness, with use of the forefoot only, 3 = continuous lameness, use of the top of the paw by bending the metatarsals, and 4 = loss of function, no use of the paw. Ankle circumferences were evaluated as the perimeter calculated after antero-posterior and latero-lateral ankle measurement with digital caliper.
|
study
| 100.0 |
Four days after intraarticular injection, rats were killed, and deskinned right ankles were collected, embedded in Neg50 (Thermo Scientific, Waltham, MA) and frozen in liquid nitrogen. Frozen right ankles were scanned by µ-CT (viva-CT40, Scanco, Brütisellen, Switzerland). Three-dimensional reconstructions were segmented using the following parameters: sigma, 2.8; support, 2; threshold, 289. After µ-CT acquisition, fresh-frozen sections of the joints (10 µm in thickness) from the different groups were stained with H&E, and examined for the presence of infiltrates26.
|
study
| 100.0 |
After µ-CT assessment, dehydrated samples were embedded in methylmetacrylate resin and 9 µm slices were obtained. Analysis included TRAP staining for osteoclasts as multinucleated TRAP+cells on the bone counter-stained with Anilin blue. Goldner trichrome stained non-mineralized, osteoid tissue in red with Fuchsine, and mineralized bone in green with Light green. Safranin.O.–Fast green staining was used to analyse cartilage integrity.
|
study
| 99.94 |
For in vitro and ex vivo cellular analyses, data were presented as the mean of triplicate experiments (m ± s.e.m.), and were representative of results obtained from three independent experiments that produced similar results. Statistical analyses were performed using the Mann–Whitney test. For in vivo experiments in rats, average of ankle joints per animal was treated independently for statistical purposes. Differences between groups were calculated using analysis of variance (ANOVA) two-way tests with Bonferonni post-hoc tests. Comparison inside each group was performed using ANOVA one-way tests. P < 0.05 was considered as significant.
|
study
| 100.0 |
Services that offer reliable and scalable determination and disambiguation of research identities are essential services that data repositories and digital libraries need to provide. Such services enable distributed grouping, linking, aggregation, and retrieval of scholarship; evaluation of the research productivity and impact of individuals, groups, and institutions; and the identification of expertise [1–3]. Publishers, libraries, universities, search engines, and content aggregators use many different research identity management systems, often referred to as research information management systems (RIMSs) or current research information systems (CRIS), with different data models, levels of coverage, and levels of quality (e.g., Florida ExpertNet, Google Scholar, ORCID, REACH NC, ResearchGate). These databases use different approaches to and mechanisms for curating research identity information: manual curation by information professionals or users, including the subjects of identity data; automated data mining and curation scripts (aka bots); or some combination thereof. With universities engaging in the curation of digital scholarship produced by their faculty, staff, and students through institutional repositories (IR), some of these universities and IRs try to manage the research identity profiles of their contributors locally (e.g., Stanford Profiles). Some large academic libraries use the VIVO ontology to make their data, including researcher identity information, discoverable and linkable for cross-institutional retrieval, processing, and analysis by both human and computational agents. The use of ontologies and Semantic Web technologies can make data machine-processable and “understandable,” and hence may reduce the cost of data aggregation and analysis. Ultimately, however, the completeness and accuracy of the data are what make RIMSs reliable. Although knowledge curation by professionals usually produces the highest quality results, it is costly and may not be scalable . Libraries and IRs may not have sufficient resources and expertise to control the quality of large-scale uncontrolled metadata, often batch harvested and ingested from faculty-authored websites and journal databases . The effective aggregation of data may require knowing community, discipline-based, and cultural differences in data and metadata quality requirements, rules, norms, and reference sources. They may need the participation of subject specialists, librarians, and especially the researchers themselves in data curation activities to ensure the quality and reliability of research identity data [3, 6–8].
|
review
| 99.56 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.