text
string | predicted_class
string | confidence
float16 |
|---|---|---|
Tumors belonging to the MSI subtype displayed strikingly different associations between epigenetic features and mutation patterns. We observed little association between mutation rate and open chromatin marks or replication timing in MSI tumors. This is likely because mismatch repair (MMR) deficient MSI tumors have lost MMR-coupled enhanced repair efficiency at early replicating open-chromatin regions31. In addition, we found that MSI mutation profiles showed a strong positive association with heterochromatin (H3K9me3) and repressive domains (H3K27me3) (Fig. 1b). This is in contrast with a previous study by Supek et al. reporting that mutations generated after MMR inactivation are no longer enriched in heterochromatin regions, arguing that genome-wide regional mutation rate variation is mostly a result of MMR31. Instead, our data suggests that, in addition to MMR, other repair or mutational processes may further contribute to variation of the GC mutation landscape. Principal component analysis (PCA) on the correlation matrix between the mutation profiles of individual tumors and the epigenetic covariates also revealed MSI tumors as a distinct cluster (Fig. 1c). Accordingly, we removed the small number of MSI tumors (N = 19) from the discovery cohort to ensure all tumors had similar mutational biases.
|
study
| 100.0 |
To identify positive selection in cancer genomes, it is essential to build an accurate background mutation rate model that corrects for covariates (features) that impact regional mutation rate variation, such as local sequence context and chromatin profiles. We considered a range of genetic and epigenetic features that could be correlated with GC somatic mutation rates. The features included 33 general and 36 gastric-specific chromatin features, 133 transcription factor binding profiles, and DNA replication timing profiles (Supplementary Data 2; Methods). To model the effect of local sequence context on mutation rate, previous studies have considered the single or tri-nucleotide sequence context of each mutation5,6,8,9. However, as mutation rates may also be influenced by wider sequence contexts33, we thus used an expanded sequence context model that considers the effects of tri-nucleotide (1 bp flanks) and penta-nucleotide (2 bp flanks) contexts on the mutation probability of each base. Least absolute shrinkage and selection operator (LASSO) logistic regression was used to identify the most predictive epigenetic and sequence context features (Supplementary Fig. 2). We used these features to estimate sample-specific background mutation probabilities, and to identify individual focal regions (21 bp) exhibiting mutational recurrence across samples beyond chance expectation (Fig. 2a; Methods). Overlapping significantly mutated regions were merged to obtain a list of unique hotspots.Fig. 2Genome-wide analysis of non-coding indel recurrence. a Workflow of the method to detect recurrently mutated non-coding regions. b Genome-wide negative log P-values of indel recurrence of 21 bp regions with at least one indel. The horizontal line marks the Bonferroni adjusted P-value of 0.01. c Negative log P-value of indel recurrence in merged non-coding regions of each gene. The top three significantly mutated genes are highlighted. The horizontal line marks the Bonferroni adjusted P-value of 0.01. d–f Gene expression of LIPF (d), PGC (e), and MUC6 (f) in normal gastric samples, tumors wildtype for the gene of interest, and tumors with non-coding indels in the gene of interest
|
study
| 100.0 |
Genome-wide analysis of non-coding indel recurrence. a Workflow of the method to detect recurrently mutated non-coding regions. b Genome-wide negative log P-values of indel recurrence of 21 bp regions with at least one indel. The horizontal line marks the Bonferroni adjusted P-value of 0.01. c Negative log P-value of indel recurrence in merged non-coding regions of each gene. The top three significantly mutated genes are highlighted. The horizontal line marks the Bonferroni adjusted P-value of 0.01. d–f Gene expression of LIPF (d), PGC (e), and MUC6 (f) in normal gastric samples, tumors wildtype for the gene of interest, and tumors with non-coding indels in the gene of interest
|
study
| 100.0 |
We used this statistical framework to identify somatic mutation hotspots (both indels and point mutations) across the non-coding genome (Fig. 2b and Fig. 3a). The top indel hotspot was located ~100 kbp upstream of the AFDN gene, which is frequently translocated in leukemia and down-regulated in multiple cancer types34–36. The effect of hotspot mutations on AFDN expression could not be tested, as we lacked paired tumor expression data for the mutated samples. The second most significant indel hotspot was located in an intron of the PGC gene, which encodes the precursor of gastric proteinase pepsinogen (Supplementary Table 1). PGC is expressed at 11940 TPM in the stomach, 39 TPM in the lung, and ≤2 TPM in all other tissues in the Genotype-Tissue Expression (GTEx) project37,38. Interestingly, a recent study reported that LIPF, a lineage-specific gastric lipase, has broad enrichment of indels in GC39. Hypothesizing that other lineage-specific genes could show similar patterns of indel enrichment, we performed a gene-based recurrence analysis to identify all genes with broad enrichment of indels in their non-coding regions (combining promoter, untranslated, and intronic regions for each gene; Methods). Interestingly, the top three genes in this analysis were all lineage-specific genes highly expressed in stomach tissue: LIPF, PGC, and MUC6 (Fig. 2c; Supplementary Table 2). MUC6 encodes a mucin glycoprotein that is a major constituent of the gut mucosa, and is expressed at 133 TPM in stomach tissue, 38 TPM in the pancreas, and ≤2 TPM in all other tissues in GTEx37,38. However, consistent with the previous report39, non-coding indels in these three recurrently mutated lineage-specific genes were not associated with expression change (Fig. 2d–f).Fig. 3Genome-wide analysis of non-coding SNV hotspots. a The negative log P-values of SNV recurrence for all 21 bp regions genome-wide, only regions with at least three mutations are displayed. Significantly mutated hotspots overlapping CBSs are highlighted. The horizontal line marks the Bonferroni adjusted P-value of 0.01. b Log odds ratio of the enrichment of hotspot mutations and non-hotspot mutations in transcription factor binding regions and conserved regions. Error bars indicate the s.e.m of the log odds ratio. c Gastric cancer samples sorted by molecular subtype, with each row representing a significant mutation hotspot. Mutated samples are highlighted in black in the matrix. The mutation load of each sample is shown in the bottom panel. The right panel annotates the location of each hotspot with respect to annotated functional regions
|
study
| 100.0 |
Genome-wide analysis of non-coding SNV hotspots. a The negative log P-values of SNV recurrence for all 21 bp regions genome-wide, only regions with at least three mutations are displayed. Significantly mutated hotspots overlapping CBSs are highlighted. The horizontal line marks the Bonferroni adjusted P-value of 0.01. b Log odds ratio of the enrichment of hotspot mutations and non-hotspot mutations in transcription factor binding regions and conserved regions. Error bars indicate the s.e.m of the log odds ratio. c Gastric cancer samples sorted by molecular subtype, with each row representing a significant mutation hotspot. Mutated samples are highlighted in black in the matrix. The mutation load of each sample is shown in the bottom panel. The right panel annotates the location of each hotspot with respect to annotated functional regions
|
study
| 100.0 |
We then performed a genome-wide analysis of SNVs in non-coding regions and identified 34 significant mutation hotspots (Bonferroni adjusted P-value < 0.01; Fig. 3a; Supplementary Table 3). These hotspots were enriched in conserved sequences and TF binding regions, suggesting that many hotspot mutations may disrupt functional elements (Fig. 3b). Strikingly, of the 34 mutation hotspots, 11 were located in CBSs (Fig. 3a, c). The majority of mutations at CBS hotspots occurred in CIN tumors (71%, P = 0.012 by two-sided Fisher’s exact test), which is the most common GC subtype, accounting for ~50% of all GC cases (Fig. 3c). The remaining 23 non-CBS hotspots often overlapped gene regions, but never co-located with TF binding regions. Furthermore, we observed a depletion of somatic mutations at gastric-specific TFBSs among the non-hotspot mutations (Fig. 3b). Overall, gastric tissue TFBSs comprises about 1% of the genome, but only 0.58% of the non-hotspot mutations were located in these regions. A similar depletion of mutations was observed for constitutive TFBSs (Supplementary Fig. 3). This is striking, as two recent studies have found that somatic mutation rates are elevated at transcription factor binding sites (TFBSs), and that this higher overall mutation load at TFBSs may be explained by reduced accessibility to nucleotide-excision repair (NER) enzymes at these sites16,17. This phenomenon is primarily observed in melanoma and lung adenocarcinoma where NER plays an important role in repairing carcinogen induced DNA lesions16. In contrast, our finding demonstrates that NER and TF occupancy is not a cause of regional mutational bias in GC.
|
study
| 100.0 |
To test if the 21 bp window size was adequate to capture most mutation hotspots, we repeated the hotspot analysis using larger 41 bp windows. In general, the rankings of the hotspots remained stable (Supplementary Fig. 4). 17/34 hotspots remained significant and only two additional hotspots were identified (P < 0.01, Bonferroni correction).
|
study
| 100.0 |
Despite the general depletion of somatic mutations at TFBSs in gastrointestinal tumors, several studies have reported an increased mutation rate specifically at CBSs in gastrointestinal tumors15,18,40. Indeed, when we examined all CBS across the genome, we found a 3-fold increased mutation rate at CBSs (11 mutations/Mb) compared to their 1 Kb flanking regions (3.6 mutations/Mb). In addition, the mutation frequencies at CBSs were very different among tumors of different molecular subtypes. The somatic mutation rate was 7.1 and 4.7-fold higher at CBSs compared to flanking regions in CIN and GS tumors, respectively (Fig. 4a–e). There was no enrichment of somatic mutations at CBSs in MSI tumors, likely due to impaired DNA MMR. Surprisingly, EBV tumors, which are not MMR-deficient, only had a modest 1.7-fold increase in mutation load at CBSs. The enrichment of somatic mutations at CBSs is therefore unlikely the result of differential DNA repair alone.Fig. 4Analysis of CBS mutations in different gastric cancer subtypes. a–d Mutation count per tumor around CBSs in the four gastric cancer subtypes. e Elevated mutation rates at CBSs compared to flanking regions. f Somatic substitution patterns within CTCF motifs for hotspot mutations and all mutations, respectively. g The negative log P-values of mutation recurrence of all CBSs evaluated with a CBS-specific background model. CBS hotspots identified in Fig. 3a are highlighted in magenta. The horizontal line marks the Bonferroni adjusted P-value of 0.01
|
study
| 100.0 |
Analysis of CBS mutations in different gastric cancer subtypes. a–d Mutation count per tumor around CBSs in the four gastric cancer subtypes. e Elevated mutation rates at CBSs compared to flanking regions. f Somatic substitution patterns within CTCF motifs for hotspot mutations and all mutations, respectively. g The negative log P-values of mutation recurrence of all CBSs evaluated with a CBS-specific background model. CBS hotspots identified in Fig. 3a are highlighted in magenta. The horizontal line marks the Bonferroni adjusted P-value of 0.01
|
study
| 100.0 |
Consistent with findings by Katainen et al. in colorectal cancer15, we found that somatic mutations at CTCF motifs, including the CBS hotspot mutations, were predominately A.T>C.G and A.T>G.C substitutions (Fig. 4f), suggesting that hotspot mutations are generated by the same mutational process as other CBS mutations. The mutation pattern at CBS hotspots was overall similar to that of all CBSs. However, while a conserved base at position 9 of the 19 bp CTCF binding motif was the most commonly mutated position at CBSs in general, the CBS hotspot mutations had the highest enrichment in the 4 bp sequence flanking the 5′ end of the CTCF motif. Furthermore, C>T changes, which are relatively common among all CBS mutations are much rarer among the CBS hotspot mutations (Fisher’s exact test P-value = 4.4 × 10−7). These differences could indicate a functional difference between CBS hotspot and non-hotspot mutations.
|
study
| 100.0 |
To explicitly test if the CBS hotspots could be explained by the genome-wide elevated mutation rate at CBSs, we constructed a CBS-specific background mutation model. Since CBS mutation rates varied across tumor subtypes, this model further included the tumor subtype as a covariate. Also, since CBSs located at chromatin loop boundaries have higher somatic mutation burden than non-boundary CBSs15,18, the CBS-specific background model differentiated between CBSs inside and outside chromatin loop boundaries. CTCF loop domains have not been profiled in gastric tissue but tend to be cell-type invariant41,42. We therefore used a constitutive set of CTCF domains shared across three cell lines (CM12878, Jurkat, and K562) to define CTCF loop boundaries13,43. In addition, since the mutation spectrum at CBSs is distinct from the overall genomic mutation spectrum, we performed LASSO logistic regression to identify sequence context features correlated with the somatic mutation rate at CBSs. To identify other mutational processes that might be associated with the occurrence of CBS mutations, we calculated the correlation between the proportion of CBS mutations in each tumor and the percentage contribution of each COSMIC mutation signature to each tumor44. While CBS mutations are known to be positively associated with signature 1715, we found that CBS mutations were also strongly negatively associated with COSMIC mutation signature 1, an age related signature (Pearson correlation = −0.41; Supplementary Fig. 5). Therefore, we added the percentage contributions of mutation signatures 1 and 17 in each individual as covariates. Finally, this model also corrected for replication timing and local mutation rate. With this model, 9/11 CBS hotspots remained significant at the Bonferroni corrected significance threshold of 0.01 and the other two were borderline with adjusted P-values of 0.025 and 0.086 (Fig. 4g). Furthermore, seven additional CBSs became significant with the restricted hypothesis testing (Supplementary Table 4; Supplementary Figs. 6, 7). Mutations at these specific sites can therefore not be explained by a genome-wide elevated mutation rate at CBS, indicating that mutations at these focal sites are may be positively selected in gastric tumors.
|
study
| 100.0 |
We next examined the possibility that the CBS hotspots were associated with changes in expression of nearby genes. We restricted the analysis to the four CBS hotspots that had at least three mutated samples with gene expression data in the TCGA cohort (N = 35 samples). We validated the results using expression data from the SG cohort (N = 14 samples). Since the chromatin structure is generally cell-type invariant41,42 and there is no published Hi–C data from gastric tissue, we used the Hi–C data from IMR90 cells published by Dixon et al.41 to examine the 3D chromatin structure around each hotspot (Supplementary Figs. 8–10). We identified the flanking TAD boundary nearest to each hotspot, and tested the association between the mutation status of each hotspot and the expression of genes within the two adjacent TADs. We found genes with nominally altered expression for three of the four hotspots (Fig. 5), two of them remain significant after correcting for multiple testing in each region.Fig. 5Association of CBS hotspot mutations and cis-gene expression. a, d, g Association between mutation status of the CBS hotspot and expression levels of neighboring genes (two-sided Wilcoxon rank-sum test). Upregulated genes are shown above the x-axis, and downregulated genes are shown below the x-axis. Non-expressed genes are shown with empty circles on the x-axis (normalized count < 10 in all samples). b, e, h The reference sequence and mutated alleles at the three CBS hotspots. The mutations in tumors with expression data are underlined. c, f, i The gene expression of CENPQ (c), KCNQ5 (f), and SPG20 (i) in normal gastric tissue, and tumors with and without mutations at the corresponding CBS hotspot. P-values were adjusted using the Benjamini–Hochberg method
|
study
| 100.0 |
Association of CBS hotspot mutations and cis-gene expression. a, d, g Association between mutation status of the CBS hotspot and expression levels of neighboring genes (two-sided Wilcoxon rank-sum test). Upregulated genes are shown above the x-axis, and downregulated genes are shown below the x-axis. Non-expressed genes are shown with empty circles on the x-axis (normalized count < 10 in all samples). b, e, h The reference sequence and mutated alleles at the three CBS hotspots. The mutations in tumors with expression data are underlined. c, f, i The gene expression of CENPQ (c), KCNQ5 (f), and SPG20 (i) in normal gastric tissue, and tumors with and without mutations at the corresponding CBS hotspot. P-values were adjusted using the Benjamini–Hochberg method
|
study
| 100.0 |
The first hotspot we identified is located in a CBS on chromosome 6 and has mutations in 12 samples (Fig. 5a–c). The expression of two neighboring genes, CENPQ and MUT, ~1 Mb upstream of this hotspot was significantly elevated in the mutated samples (P = 0.007 and 0.0021 respectively, Benjamini–Hochberg adjusted P = 0.026 and 0.042 respectively, two-sided Wilcoxon rank-sum test; Fig. 5a–c). A similar trend of CENPQ expression was observed using the expression data from the SG cohort (Supplementary Fig. 11a). CENPQ is a subunit of a centromeric complex, and is involved in mitotic progression and chromosomal segregation45. Interestingly, the tumor with the highest expression of CENPQ was mutated at the highly conserved position 9 of the CTCF motif, while the other two tumors were mutated at position 2 of the CTCF motif. This indicates that different mutations in the same hotspot may have different disruptive potentials. However, a formal evaluation of such effects requires a larger set of tumor samples with both CBS mutations and RNA-seq data available.
|
study
| 100.0 |
The next hotspot we tested is located on chromosome 6 with 9 mutated samples. Tumors with mutations at this hotspot had significantly lower expression of the KCNQ5 gene (Wilcoxon P = 0.0059, adjusted P = 0.047), located ~200 kb downstream of the hotspot (Fig. 5d–f). A similar trend in KCNQ5 expression was observed using the expression data from the SG cohort (Supplementary Fig. 11b). A recent study by Umer et al. found the same mutation hotspot by analyzing motif-breaking mutations40. Using an electrophoretic mobility shift assay, Umer et al. confirmed that the chr6:73,122,103 A>G mutation disrupts CTCF binding. In addition, it has been reported that CTCF is involved in the spatial organization of the KCNQ5 locus, and knockdown of CTCF downregulates KCNQ5 expression46.
|
study
| 100.0 |
At the third hotspot located on chromosome 13, mutated tumors had on average a 3-fold decrease in SPG20 expression (Wilcoxon P = 0.045, adjusted P = 0.65; Fig. 5g–i). However, only three tumors with expression data were mutated at this hotspot, and the expression change was not significant after correcting for multiple testing. A larger sample size is needed to evaluate if this is a spurious or true correlation. A similar trend in SPG20 expression is observed using the expression data from the SG cohort (Supplementary Fig. 11c). SPG20 is involved in epidermal growth factor receptor trafficking47 and was previously found to be significantly mutated in the exome of esopheagal cancer48.
|
study
| 100.0 |
In all three cases, we confirmed that the expression changes of these genes were significant after correcting for variation in DNA copy numbers and tumor purity between samples (Supplementary Fig. 12). We would expect mutations at these hotspots to be associated with expression change of the altered allele. However, allele-specific expression analysis is challenging to evaluate across distant sites. Indeed, for all three candidate CBS hotspots, we could not unambiguously resolve the two alleles because the maximum inter-SNP distance between hotspot and candidate target gene were >5 kb. As CBSs are essential in maintaining the chromosomal architecture, it is likely that these CBS hotspot mutations cause altered expression of nearby cancer driver genes by disrupting the local chromosomal organization. Indeed, using the set of constitutive CTCF-CTCF loops, we observed chromatin contacts between the KCNQ5 and SPG20 loci and their corresponding CBS hotspots (Supplementary Figs. 9, 10). Interestingly, the three genes were also differentially expressed in GC tumors compared to normal gastric tissue. CENPQ expression was upregulated in tumors (Wilcoxon P = 0.0028; Fig. 5c), while both KCNQ5 and SPG20 expression was downregulated in tumors compared to normal gastric samples (Wilcoxon P = 3.2 × 10−7 and 0.00082 respectively; Fig. 5f, i). Therefore, it is plausible that the expression of these three genes could be altered in GC through additional mechanisms. Indeed, KCNQ5 and SPG20 were found to be downregulated in colorectal cancer compared to the normal mucosa due to promoter hypermethylation49–51. These results further support the contributions of these genes to GC tumorigenesis.
|
study
| 100.0 |
Many of the hotspot mutations were located in the 5′ flanks of the consensus CTCF motif (Fig. 4f). Previous studies have found increased conservation in the flanking sequences of weak CTCF and REST binding sites, suggesting that the sequence context is important for TF binding at these sites52,53. We examined the evolutionary conservation54 at the CTCF binding motifs and their flanking sequences. In general, the 5′ flanks of the CTCF motifs are not conserved (Supplementary Fig. 13a). However, in the hotspot upstream of CENPQ, the mutation cluster in the 5′ flank co-occurred with conserved bases (Supplementary Fig. 13b). In addition, we found another CBS hotspot with nine 5′-flank mutations that coincided with a highly conserved base (Supplementary Fig. 13c). Such hotspot mutations, affecting conserved 5′ flanks of CTCF motifs, could disrupt context-specific binding of CTCF.
|
study
| 100.0 |
We also examined the possibility that mutations in the flanking regions of CTCF motifs create or disrupt binding motifs of other TFs. We used DeepBind55 to predict the binding scores of wildtype and mutated sequences for 472 transcription factors. However, we only found mutations at three CBS sites with predicted change in TF binding (Supplementary Table 5). Lastly, it is also possible that some mutations at CBS flanks are passenger mutations arising due to the overall elevated mutation rates at CBSs. While our model identifies individual CBS regions with overall mutation enrichment, it does not allow us to distinguish between passenger and driver mutations within such regions.
|
study
| 100.0 |
Taken collectively, 25% of all gastric tumors are mutated in at least one of the 11 CBS hotspots, representing the second most mutated functional region in GC after TP53 (50% of gastric tumors). To study if these hotspots could also play a role in other cancer types, we examined the recurrence of these 11 hotspots in 826 non-hypermutated tumors of 18 other cancer types5 (Fig. 6 and Supplementary Fig. 15). Strikingly, we found that 19% of colorectal cancer tumors were mutated at one or more of the CBS hotspots (Fig. 6a). Since colorectal cancer have pathological and molecular similarities to GC56, the CBS hotspot mutations may drive cancer progression in colorectal cancer through similar mechanisms as in GC. The CBS hotspots were mutated at lower frequencies in breast cancer, liver cancer, lung cancer, pancreas cancer, and lymphoma. Interestingly, while melanoma and bladder carcinoma also have high genome-wide mutation rates at CBS, none of the CBS hotspots were mutated in these two cancer types. Similarly, we found that mutations at all CBS hotspots had previously been reported in COSMIC57 or other genome-wide studies of gastrointestinal tumors15,40 (Supplementary Table 6). This suggests that the CBS hotspot mutations are generated and act in a cancer-specific manner.Fig. 6Pan-cancer analysis of mutation recurrence at the 11 CBS mutation hotspots. a Fraction of samples with mutation in at least one of the CBS hotspots in different cancer types. b Mutation rate of CBSs in different cancer types. c Mutation recurrence of individual CBS hotspots in different cancer types
|
study
| 99.94 |
Pan-cancer analysis of mutation recurrence at the 11 CBS mutation hotspots. a Fraction of samples with mutation in at least one of the CBS hotspots in different cancer types. b Mutation rate of CBSs in different cancer types. c Mutation recurrence of individual CBS hotspots in different cancer types
|
study
| 99.9 |
Enrichment of CBS mutations was highest in CIN tumors, which are characterized by increased chromosomal aneuploidy. This prompted us to examine if mutations at CBSs in CIN tumors were correlated with somatic copy number alteration (SCNA) breakpoints. Strikingly, the distance between a CBS hotspot and its nearest SCNA breakpoint was significantly shorter in mutated than non-mutated tumors (P = 0.0018, two-sided Wilcoxon rank-sum test; Fig. 7a). In contrast, non-CBS mutation hotspots showed no such association (P = 0.53). The median distance between CBS hotspot mutations and its nearest SCNA breakpoint in the same sample was ~1 Mbp, notably shorter than the ~2 Mbp distance for non-CBS hotspots (Fig. 7a). To study whether this correlation between CBS mutations and SCNA breakpoints was specific to the CBS hotspots, we extended the analysis to all CBSs. Interestingly, we found that CBS mutations were correlated with occurrence of nearby SCNA breakpoints in the same samples, especially for mutations affecting CBSs at loop boundaries (Wilcoxon P = 5.7 × 10−16; Fig. 7b). Conversely, when we grouped 1 Mb windows of the genome according to SCNA breakpoint density, we found that the normalized CBS mutation rate was positively associated with SCNA breakpoint density (Fig. 7c–d). Overall, these results highlight a link between regional chromosomal instability and mutations at both CBS hotspots and boundary CBSs in general.Fig. 7Association between CBS mutations and genomic instability. a Distance to the nearest CNV breakpoint from CBS hotspots and other non-CBS mutation hotspots. b Distance to the nearest CNV breakpoint from CBSs at loop boundary and non-boundary CBSs. Wilcoxon rank-sum test P-values are shown. c Correlation of mutation rates with SCNA breakpoint density. d Correlation of normalized mutation rates with SCNA breakpoint density, correcting for the background mutation rate in each bin. Error bars represent the s.e.m. e The violin plots show the VAF distributions of somatic mutations in diploid regions of individual tumors. VAFs of the mutations at CBS hotspots are marked by red vertical lines. f Comparison between VAFs of the CBS hotspot mutations and VAFs of non-silent coding mutation on GC driver genes. The red points represent the median VAFs in each group. The dashed lines match mutations from the same samples. P-value is calculated by paired Wilcoxon rank-sum test
|
study
| 100.0 |
Association between CBS mutations and genomic instability. a Distance to the nearest CNV breakpoint from CBS hotspots and other non-CBS mutation hotspots. b Distance to the nearest CNV breakpoint from CBSs at loop boundary and non-boundary CBSs. Wilcoxon rank-sum test P-values are shown. c Correlation of mutation rates with SCNA breakpoint density. d Correlation of normalized mutation rates with SCNA breakpoint density, correcting for the background mutation rate in each bin. Error bars represent the s.e.m. e The violin plots show the VAF distributions of somatic mutations in diploid regions of individual tumors. VAFs of the mutations at CBS hotspots are marked by red vertical lines. f Comparison between VAFs of the CBS hotspot mutations and VAFs of non-silent coding mutation on GC driver genes. The red points represent the median VAFs in each group. The dashed lines match mutations from the same samples. P-value is calculated by paired Wilcoxon rank-sum test
|
study
| 100.0 |
As the CBS mutation rate was also elevated in GS tumors (Fig. 4b), we investigated if there was a similar association between CBS mutations and SCNA in GS tumors. Although we found that mutated CBSs also tended to be closer to SCNA breakpoints compared to the non-mutated CBSs in GS tumors, the difference was not statistically significant (Supplementary Fig. 14), and the relative difference was greater in CIN (2.17-fold difference in distance to nearest breakpoint) compared to GS (1.58-fold difference) tumors. This may indicate that the coupling of CBS mutations and nearby chromosomal instability is a process that is specific to, or exacerbated in, the CIN tumors.
|
study
| 100.0 |
In this study, we performed a comprehensive and unbiased analysis of non-coding SNVs and indels in 212 GC genomes, the largest studied cohort thus far. In addition to a previously identified indel enrichment at LIPF39, our analysis identified two other gastric lineage-specific genes with broad enrichment of non-coding indels (PGC and MUC6). Our results show that the accumulation of indels occur in multiple lineage specific genes in GC. Yet, indels at these three genes were not associated with change in gene expression. The functional consequences of these indels are therefore still unclear. Strikingly, genome-wide analysis of somatic SNVs revealed 34 significant mutation hotspots (Bonferroni adjusted P-value < 0.01) that were disproportionately enriched in CBSs. Katainen et al. previously reported an increased mutation load at CBSs in colorectal cancer15, and a subsequent study by Kaiser et al. confirmed the general hypermutation at CBSs in 11 cancer types18. Both studies generally discounted CBS mutations as passengers, yet they did not rigorously explore the hypothesis that a subset of these mutated CBSs may be undergoing positive selection within individual cancer types. Indeed, a recent study on motif-breaking mutations identified a recurrent CBS mutation that disrupts CTCF binding40, confirming the motif-breaking potential of CBS mutations. Here, we used a large cohort of GC genomes in combination with rigorous statistics, to show that mutation rates at 11 specific CBSs are unexpectedly high and cannot alone be explained by a genome-wide elevated mutation burden at CBS, indicating positive selection at these sites. Out of the four CBS hotspots we examined, three of them were associated with nominally significant expression changes of neighboring genes (CENPQ, KCNQ5, and SPG20), and these associations were validated in an independent tumor cohort. Furthermore, it is possible that mutations at these CBS hotspots also have long-range or spatio-temporal regulatory effects on gene expression that are not captured by bulk tumor transcriptome profiling. Overall, our analyses nominate these CBS hotspots as potential drivers in GC, and support the hypothesis that driver mutations may arise as a by-product of the increased mutation load at CBSs followed by positive selection at specific CBSs. This is comparable to a model of genomic rearrangement hotspots in breast cancer, where rearrangements initially arise from defective homologous-recombination-repair and those affecting cancer risk loci are subsequently positively selected, forming rearrangement hotspots58.
|
study
| 100.0 |
We found that gastric tumors of the genomic instable subtype (CIN) exhibited the highest mutation rate at CBSs compared to tumors of the other GC subtypes. Furthermore, CBS mutations were associated with the occurrence of nearby chromosomal breakpoints, suggesting a general link between CBS mutations and genomic instability. A previous study has suggested a model where genome higher-order interactions are directly poised for chromosomal breaks59. One important open question is whether these processes are coupled, and if so, what is the temporal order of CBS mutations and chromosomal breaks. Interestingly, somatic variant allele frequencies (VAFs) of the CBS hotspot mutations supported that these were generally clonal and likely early events in tumor evolution (Fig. 7e). Furthermore, we found that the VAFs of CBS hotspot mutations were comparable to non-silent coding mutations of known GC driver genes from the same sample (paired Wilcoxon P-value = 0.49; Fig. 7f).
|
study
| 100.0 |
Previous studies found kilobase sized regions of hypermutation, termed “kataegis”, that tend to co-occur with genomic rearrangements in cancer44,60. Importantly, our data suggest that the mutational mechanism underlying the association between CBS mutations and DNA breakpoints is distinct from that of kataegis. While kataegis is characterized by C>T and C>G substitutions, CBS mutations are mostly T>G and T>C substitutions. In addition, kataegis is defined by mutation clusters with inter-mutation distance <1 kb. CBS hotspots are confined focal regions of <30 bps including the CTCF motif and its 5′ flanking sequence.
|
study
| 100.0 |
Only a subset of tumor samples in our cohort had paired gene expression data (49/187 samples). This limited our ability to test for functional consequences of CBS hotspot mutations. Additional focused experiments involving transcriptome, copy number, and chromatin structure data should further clarify the regulatory and functional effects of the CBS mutations. We did not uncover a shared theme for the 23 significant non-CBS hotspots. Among the non-CBS hotspots, seven of them are intronic, one is downstream of a gene and the rest are intergenic. None of the genes associated with the hotspots are known cancer drivers. We did not observe any mutation hotspot near TERT, confirming that the reactivation of TERT is very rare in GC3. For the non-CBS hotspots that overlapped gene regions, focused functional validation experiments could be performed on a case-by-case basis.
|
study
| 100.0 |
The statistics of cancer driver identification is still limited by our knowledge of the somatic mutation and repair processes. Although our background model corrected for many covariates of the somatic mutation rate, such as epigenetic and sequence context features, false positives and false negatives could still arise from the current model not considering as yet unknown mutational biases.
|
study
| 99.94 |
Taken collectively, 25% of GC tumors and 19% of colorectal cancer tumors are mutated in at least one of the 11 CBS hotspots. Overall, our analyses nominate these CBS hotspots as potentially common drivers of gastrointestinal cancers. Furthermore, the data supports a general link between CBS mutations and chromosomal instability. This suggests that non-coding regulatory mutations could potentially drive tumor evolution through interfacing with cancer genome and epigenome plasticity.
|
study
| 100.0 |
We performed whole-genome sequencing of 40 gastric GC tumors and matched normal samples from patients from Singapore (study protocol approved by National University of Singapore Institutional Review Board). Informed consent was obtained from all participating patients. Genomic DNA of tumors and matched normal gastric tissues was extracted (QIAGEN). Libraries were constructed with 300–400 bp insert length, and 101 bp or 151 bp paired-end sequencing was performed on Illumina Hiseq instruments. The tumors were classified into four molecular subtypes as described previously by TCGA19.
|
study
| 100.0 |
We obtained WGS data of 40 GC tumors from TCGA (https://gdc.cancer.gov), 32 tumors from ICGC (https://ega-archive.org/datasets/EGAD00001003132), and 100 tumors from Wang et al. (HK)20. The molecular subtypes of tumors from the TCGA cohort were defined by TCGA. For the HK cohort, only EBV and MSI subtype status was available. The molecular subtypes of tumors from the ICGC cohort were unavailable, but we identified one MSI sample from the ICGC cohort using MSIseq61.
|
study
| 100.0 |
Raw sequencing data was uniformly processed using the bcbio-nextgen pipeline (v0.9.3). Briefly, sequencing reads were aligned to the human reference genome (hg19) using BWA62. Duplicated reads marked by Picard were removed. Indel regions were realigned using GATK63. Somatic mutations were called by four independent mutation callers: VarScan64, MuTect65, VarDict66, and FreeBayes67 using default parameters of the bcbio-nextgen pipeline. As the nature of our analyses requires high specificity in somatic mutation calling, we developed a random forest predictor, SMuRF68, trained on manually curated true somatic mutations to identify high confidence somatic mutation calls from the output of the four mutation callers. For each GC WGS sample, a set of high confidence consensus calls were obtained by running the random forest prediction algorithm.
|
study
| 100.0 |
False positive somatic calls could arise from sequencing and mapping errors. More false positives tend to be called in the non-coding regions of the genome because these regions are enriched for repeats and low sequence complexity regions. As the downstream mutation recurrence analysis is extremely sensitive to recurrent artefacts in somatic mutation calling, we applied additional post-processing filters to eliminate potential false positive calls. We removed candidate somatic mutation calls that:Are found at >1% allele frequency in the 1000 Genomes Project69 (potential germline mutations)Are found in more than 10% of the matched normal samples (potential systematic sequencing errors)Are found in more than 1% of the matched normal samples and are within 20 bp to a common indel in the 1000 Genomes Project (potential errors arising from mapping errors near indels).
|
study
| 100.0 |
We performed RNA-sequencing on 19 matched tumor-normal pairs. Total RNA was extracted using the Qiagen RNeasy Mini kit. RNA-seq libraries were constructed according to manufacturer’s instructions using Illumina Stranded Total RNA Sample Prep Kit v2 (Illumina, San Diego, CA), Ribo-Zero Gold option (Epicenter, Madison, WI), and 1 μg total RNA. We validated the completed libraries with Agilent Bioanalyzer (Agilent Technologies, Palo Alto, CA), and applied the libraries to an Illumina flow cell via the Illumina Cluster Station. RNA-seq reads (2 × 101 bp) were aligned to the human genome (hg19) using TopHat2-2.0.12 (default parameter and—library-type fr-firststrand). Transcript abundances at the gene level were estimated by Cufflinks70. The normalized counts of RNA sequencing data of 35 tumors from the TCGA cohort were obtained from the Genomic Data Commons Portal.
|
study
| 100.0 |
The somatic mutation rate is correlated with epigenetic features such as histone modification and chromatin accessibility30, especially those derived from the cell type of origin of the cancer32. We compiled 36 gastric specific and 24 general chromatin features that potentially affect mutation rate in GC. These 66 histone modification profiles and chromatin accessibility profiles were obtained from Roadmap Epigenomics29 and in-house data. We obtained P-value signal tracks of 853 DNaseI and histone modification profiles of 111 primary tissues and cell types from the Roadmap Epigenomics project. Among them, 27 epigenetic profiles were derived from gastric related tissues. For the 24 histone marks that were not assayed in gastric-related tissues, we created meta histone modifications profiles by taking the median profile of each mark across all tissues and cell-types assayed. In addition, we included histone modifications profiles of H3K4Me1, H3K4me3, and H3K27Ac of 19 GC tumor/normal samples and 13 GC cell lines (FU97, KATO3, MKN7, NCC24, NCC59, OCUM1, RERF-GC-1B, SNU16, SNU1750, YCC3, YCC7, YCC21, and YCC22)24,25. We used the median signal of each histone mark over all tumor samples, all normal samples, and all cell lines, respectively.
|
study
| 100.0 |
Replication timing profiles were not available for gastric tissue. We therefore used the mean replication timing profile of 13 cell lines (Bj, Nhek, K562, Mcf7, Gm06990, Gm12812, Imr90, Hepg2, Helas3, Gm12801, Huvec, Gm12878, and Gm12813) generated by ENCODE71.
|
study
| 100.0 |
Binding profiles of 132 transcription factors and a meta-profile of all TFBSs were obtained from the Ensembl Regulatory Build72. We used generic TF binding profiles as there is no comprehensive TF binding assay done in gastric tissue. In total, we considered 194 candidate epigenetic covariates potentially informative of somatic mutation rates in GC (Supplementary Data 2).
|
study
| 100.0 |
To identify sequence context features affecting somatic mutation accumulation in GC, we considered 1-mer, 3-mer, and 5-mer nucleotide motifs centered at the mutated site, as well as 1-bp and 2-bp left/right flank motifs of the site. All nucleotide context features were grouped into reverse compliment pairs. As indels tend to occur in poly-monomer sequences, especially poly-A and poly-T sequences, we used the presence of poly-A, poly-T, poly-G, and poly-C sequences at the indel sites as features in the indel background mutation model.
|
study
| 100.0 |
Lastly, we included local mutation rate as a covariate to account for other unknown factors affecting mutation rate. The local mutation rate was calculated for 100 kb non-overlapping bins across the genome after masking CDS regions, immunoglobulin loci, and poorly mappable regions (mappability score < 1 in the ENCODE 75mers Alignability track).
|
study
| 100.0 |
The genome was divided into 1 Mb non-overlapping windows. CDS regions, immunoglobulin loci, and poorly mappable regions were masked from the genomic windows. Windows smaller than 250 kb after masking were removed. We calculated the mean signal of each epigenetic feature (in Fig. 1b) and the mutation rate of each tumor in each window. The Pearson correlations between the epigenetic features and mutation rates of the each tumor were calculated. To identify the contributions of epigenetic features to the variance in the mutation rate of individual tumors, we performed PCA on the correlation matrix between the mutation rates of individual tumors and epigenetic features using the “prcomp function in R. The contribution of each feature to a principal component is calculated as the feature’s loading (rotation) divided by the sum of loadings of all features for that principal component.
|
study
| 100.0 |
The LASSO is a regularized regression approach commonly used for automated feature selection. LASSO penalizes the sum of the absolute size of the regression coefficients, forcing some of the regression coefficients to shrink to zero, thereby selecting a simpler and more interpretable model. The LASSO objective function can be written as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop {{\min }}\limits_{\beta _0,\beta } \frac{1}{N}\mathop {\sum }\limits_{i = 1}^N l(y_i,\beta _0 + \beta ^Tx_i) + \lambda \parallel \beta \parallel _1$$\end{document}minβ0,β1N∑i=1Nl(yi,β0+βTxi)+λ∥β∥1Where l is the negative log-likelihood function and λ is the regularization parameter.
|
other
| 98.75 |
We used LASSO logistic regression to identify the most informative features for modeling the somatic mutation rate in GC. As it is computationally expensive to run a logistic regression on all positions in the non-coding genome with a large number of predictor variables, we used all mutated sites and an equal number of randomly sampled non-mutated sites as the input for feature selection in the LASSO logistic regression model. We regressed the binary mutation status of each site against the mean signal of each feature over an 11 bp region centered at the site. The regularization parameter λ was chosen by 10-fold cross-validation such that the error of the selected model was within one standard deviation from the minimum error. LASSO regression and cross validation were performed using the “glmnet” package in R.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{glmnet}(y\sim \beta X,\mathrm{family} = \mathrm{logistic})$$\end{document}glmnet(y~βX,family=logistic)
|
study
| 100.0 |
We bootstrapped 100 samples with 50% of the data at each bootstrap, and performed LASSO regression using the bootstrap samples. Assuming that the most informative features would be robustly selected, we used features selected in more than 95% of the bootstrap samples for the final regression model.
|
study
| 99.94 |
The patient specific background mutation probabilities were estimated by fitting a logistic regression model on all genomic sites after masking CDS regions, immunoglobin loci, and poorly mappable regions. Replication timing was discretized into eight equally sized bins, the local mutation rate was discretized into ten equally sized bins, and the chromatin features and TF-binding profiles were binarized. P-value signal tracks of the histone modification profiles from the Roadmap Epigenomics were binarized using a cutoff of 10−4. ENCODE TF-binding profiles were binarized according to the presence of a peak in any cell line assayed. We performed logistic regression using the frequency table of the counts of mutated and non-mutated sites for each combination of the covariates. Separate logistic regression models were fit to estimate the background mutation probabilities of SNVs and indels. This is to account for the different mutational processes from which SNVs and indels arise, as well as the different uncertainties associated with SNV and indel calls.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{glm}\left( {y\sim \mathrm{rep} + \mathrm{epi} + \mathrm{sequence} + \mathrm{pid},\,\mathrm{family} = \mathrm{logit}} \right)$$\end{document}glmy~rep+epi+sequence+pid,family=logit
|
study
| 100.0 |
For a specific region of interest, the probability, pi, of mutation in tumor i is a function of the length of that region and the expected mutation rates of individual nucleotides in that region under the null hypothesis. Assuming qi, j is the mutation probability of nucleotide j in tumor i, and l is the length of the region of interest:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_i = 1 - \mathop {\prod }\limits_{j = 0}^l (1 - q_{i,j})$$\end{document}pi=1-∏j=0l(1-qi,j)
|
study
| 99.94 |
Mutation recurrence is then modeled using the Poisson binomial distribution, which accounts for variation in mutation rate across tumors. For a specific region of interest, the probability of having mutations in k or more individuals is given by:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Pr \left( {K \ge k} \right) = \mathop {\sum }\limits_{m = k}^n \mathop {\sum }\limits_{A \in F_m} \mathop {\prod }\limits_{i \in A} p_i\mathop {\prod }\limits_{j \in A^c} (1 - p_j)$$\end{document}PrK≥k= ∑m=kn ∑A∈Fm ∏i∈Api ∏j∈Ac(1-pj)
|
study
| 99.94 |
Here n is the total number of tumors sequenced, k is the number of tumors with mutations in the region of interest, Fm is the set of all subsets of k integers selected from {1,2,…,n}, A is a subset of Fm, Ac is the complement of set A, pi is the probability of mutation in tumor i, and pj is the probability of mutation in tumor j. The Poisson binomial probability is calculated using an efficient and accurate normal approximation in the “poibin” R package.
|
study
| 96.1 |
The hotspot analysis aims to identify small focal regions with high mutation rates. We first considered all mutated 21 bp regions by taking 10 bp flanks on each side of each mutation. Then we calculated the mutation recurrence scores for all 21 bp regions with three or more mutated samples (two or more for indels). The P value of mutation recurrence of each hotspot was calculated using the Poisson binomial model described in the previous section. The total number of hypothesis tested is equal to the number of bases in the masked non-coding genome. We used the Bonferroni correction to adjust for multiple testing of 2,533,374,732 hypotheses, to maintain the overall α at 0.01.
|
study
| 100.0 |
We scanned for non-coding regions of genes with recurrence of indels. Gene regions were defined by Ensembl v75 annotations. We considered the merged non-coding regions of each gene by masking all coding regions of each gene, and extending the gene boundaries by 1 kb to take into account its promoter region. We calculated the mutation recurrence scores for all protein-coding genes, and their individual merged non-coding regions, using the Poisson binomial model described in the previous section. The Bonferroni correction was used to maintain the overall α at 0.01.
|
study
| 100.0 |
We calculated the log odds ratio of the enrichment of hotspot mutations in TF binding regions and conserved DNA elements. Gastric-specific TFBSs were defined as a ChIP-seq peak of a TF in any of the ENCODE cell lines that overlaps a gastric tissue DNaseI hypersensitivity site (data from Roadmap Epigenomics). Constitutive TFBSs are defined as TFBSs with Ptfbs > 0.75, where Ptfbs is the probability that the TFBS is bound by a TF for any given ENCODE cell line. Ptfbs for all TFBSs were obtained from the ENSEMBL regulatory build. Conserved elements generated by GERP73 from the alignment of hg19 to 36 mammals were downloaded from the UCSC genome browser.
|
study
| 100.0 |
The expected fraction of hotspot (or non-hotspot) mutations in the functional region type (p2) is the fraction of the genome that constitutes the functional region. The observed fraction of hotspot (or non-hotspot) mutations in the functional region is calculated by adding all mutations in the functional region type and dividing by the total number of mutations genome-wide (p1). The log odds ratio of the enrichment of hotspot (or non-hotspot) mutations in a functional region type is given by,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{LOD} = {\mathrm{ln}}(\frac{{p_1/(1 - p_1)}}{{p_2/(1 - p_2)}})$$\end{document}LOD=ln(p1∕(1-p1)p2∕(1-p2))The standard error of the LOD is calculated as,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SE}}_{\mathrm{LOD}} = \sqrt {\frac{{{{\mathrm{SE}}_{p1}}^2}}{{{p_1}^2 - \left( {1 - p_1} \right)^2}} + \frac{{{{\mathrm{SE}}_{p2}}^2}}{{{p_2}^2 - \left( {1 - p_2} \right)^2}}}$$\end{document}SELOD=SEp12p12-1-p12+SEp22p22-1-p22The statistical significance of the enrichment was evaluated by the Z-test.
|
study
| 99.94 |
The position weight matrix of the CTCF binding motif was obtained from JASPAR74. Genomic locations of CTCF binding motifs were identified using the FIMO75 function of the MEME tool suite76 with a P-value threshold of 0.01. Gastric-specific CBSs were defined as CBS motifs overlapping both a CTCF ChIP-seq peak in at least one ENCODE cell line and a DNaseI hypersensitivity site in gastric tissue from Roadmap epigenomics. We used the set of constitutive CTCF–CTCF loops shared across three cell lines (GM12878, Jurkat, and K562) obtained from the supplementary information of Hnisz et al.13 CBSs that overlap the boundaries of these constitutive CTCF loops were defined as boundary CBSs.
|
study
| 100.0 |
For the CBS-specific background model, we limited the model and search space to CBS regions and their 5 bp flanking DNA.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{glm}}\left( {{y}_{{\mathrm{CBS}}}\sim {\mathrm{rep}} + {\mathrm{subtype}} + {\mathrm{boundary}} + {\mathrm{sequence}} + {\mathrm{pid}} }\right.\\ \left.{+ {\mathrm{mutsig}}1 + {\mathrm{mutsig}}17,{\mathrm{family}} = {\mathrm{logit}}} \right)$$\end{document}glmyCBS~rep+subtype+boundary+sequence+pid+mutsig1+mutsig17,family=logit
|
study
| 100.0 |
Here subtype is the tumor subtype, boundary indicates if the CBS is located at a CTCF loop boundary, and mutsig1 and mutsig17 represent the percentage contributions of signature 1 and signature 17 of the tumor. We used deconstructSigs77 to quantify the prevalence of each of the 30 COSMIC consensus mutation signatures in each tumor.
|
study
| 99.94 |
We extracted the ±40 bp sequence context around each mutation, and used DeepBind to predict the binding scores of 472 TFs for the reference (ref score) and mutated sequences (alt score) of each mutation. Since the binding scores output by DeepBind are on an arbitrary scale and vary between different TF models, we estimated the background distributions of the binding scores of each TF by applying DeepBind to 10,000 randomly sampled non-hotspot mutations. For a particular TF, a mutation is predicted to be motif-disrupting if its reference sequence scores higher than 99.9% of the random mutations, and the score difference between its alternate and reference sequences (alt score – ref score) is smaller than 99.9% of the random mutations for that TF. A mutation is predicted to create a motif for a specific TF if its alternate sequence scores higher than 99.9% of the random mutations, and the score difference between its alternate and reference sequences (alt score – ref score) is greater than 99.9% of the random mutations for that TF.
|
study
| 100.0 |
Somatic mutations of 858 tumors from 22 cancer types were downloaded from the supplementary information of Weinhold et al.5 Hypermutated tumors with more than 200,000 mutations were excluded from the analysis. Cancer types with less than ten samples were excluded from the analysis. For CBS mutation rate calculation in Fig. 6b, CBSs were defined as CTCF motifs overlapping a CTCF ChIP-seq peak in at least one ENCODE cell line. We further defined tissue-specific CBSs for 14/19 cancer types for which DNaseI profiles in the matched tissue types are available in Roadmap Epigenomics. Tissue-specific CBSs were defined as generic CBSs that fall under DNaseI peaks in the respective tissue. Supplementary Figure 15 shows the mutation rates at tissue-specific CBSs.
|
study
| 100.0 |
Copy number segmentations were generated by CNVkit78 using default settings (bcbio-nextgen v0.9.3). SCNA breakpoints were defined as the ends of non-diploid segments. Assuming tumor purity of 50%, the estimated mean purity of these tumors, non-diploid segments were defined as segments with log2(tumor coverage/normal coverage) < log2(1.5/2) or log2(tumor coverage/normal coverage) > log2(2.5/2).
|
study
| 99.94 |
The list of known GC driver genes was collated from the Cancer Gene Census79 and the driver genes identified by TCGA19 and Wang et al.20 We excluded TP53 from the analysis as TP53 frequently undergo deletions and loss of heterozyosity. We identified non-synonymous and truncating mutations on known GC driver genes, and compared their VAFs to the VAFs of CBS hotspot mutations from the same samples using a matched Wilcoxon rank-sum test. Only mutations in diploid regions in each sample were included in the analysis.
|
study
| 100.0 |
All R code used to generate the figures and statistics of the paper is included in Supplementary Data 5. Source code for the ensemble somatic mutation caller, SMuRF68, can be found at https://github.com/skandlab/SMuRF. Source code for estimating background mutation rate from genomic covariates and identification of non-coding mutation hotspots is available at: https://github.com/skandlab/MutSpot.
|
other
| 99.94 |
There has been an intense renewal of interest in primaquine (PQ) after over 60 years since its introduction into clinical medicine [1, 2]. This is understandably so because, despite spirited research efforts, no suitable alternative has yet been approved to rival the unique therapeutic profile of PQ. As the prototype 8-aminoquinoline (8-AQ), PQ is the only licensed drug capable of treating the relapsing liver stages (hypnozoites) of Plasmodium vivax . It is gametocytocidal against Plasmodium falciparum and capable of blocking transmission. Additionally, PQ has been used as a prophylactic option against all forms of malaria . It has found use in the treatment of artemisinin-resistant P. falciparum strains . The major limitation in the clinical utility of PQ is its haemolytic toxicity in individuals with genetic deficiency in glucose-6-phosphate dehydrogenase (G6PD). Both the therapeutic and haemotoxic effects of PQ have been strongly linked to its metabolite(s) [6–8]. While a number of human metabolites of PQ have been described , the identity of the specific PQ metabolites responsible for efficacy and toxicity are not well understood .
|
review
| 99.9 |
Interactions between various anti-malarial drugs have been reported to improve efficacy and mitigate toxicity of PQ. In particular, there have been a number of clinical evidences presented over the years showing that quinine (QN) and chloroquine (CQ) can favourably affect the efficacy of 8-AQs; further, there are intriguing suggestions that the haemolytic toxicity of the 8-AQ class can be reduced. For instance, the addition of CQ increased the effectiveness of PQ and pamaquine while reducing the associated toxicity in Korean patients . In a study to assess the toxicity of pentaquine, using doses two- to three-fold higher than therapeutic doses, ten patients were experimentally infected with malaria. Each patient received 120 mg pentaquine with or without QN . Five subjects receiving pentaquine only experienced marked drops in haemoglobin (average 2.6 g/dl), but the five concomitantly receiving QN showed no change in haemoglobin. Moreover, no patient in the pentaquine/QN group had a relapse in 12 months compared to two in the pentaquine-only group, who had relapses in 12 and 26 days, respectively. Associated side effects of syncope and postural hypotension were not reported in the pentaquine/QN group compared to pentaquine-only group . Thus, this study not only demonstrated the attenuation of the toxic effects of pentaquine by QN, it showed that QN improved the therapeutic efficacy of pentaquine. A similar observation, including an increase in plasma pentaquine concentration in the presence of QN, was reported by Alving and co-workers . Being an 8-AQ like PQ, whose mechanism of toxicity might be similar, this observation suggests similar beneficial drug interactions with PQ. Since the pharmacological activity of the 8-AQs are dependent on their metabolites, the knowledge of the roles of each metabolites may be deductible from observed pharmacodynamic changes due to inhibition or facilitation of the formation of known metabolites by other agents.
|
study
| 73.0 |
The potential mechanisms underlying these interactions are unknown at present, but could be presumptively metabolism-linked. Until recently, the focus on PQ metabolites in clinical studies has been towards carboxy-primaquine (CPQ), which has been reported to be the major circulating metabolite . In one study conducted to investigate interactions among anti-malarial drugs, QN caused a decrease in the area under the curve (AUC0–24h) of CPQ by about 50 % . More recently, Pukrittayakamee et al. conducted an open-label, cross-over, pharmacokinetic interaction study in 16 health human subjects. Volunteers received single doses of 30 mg PQ, 600 mg CQ, or the combination thereof. In contrast to the QN results, CQ caused a significant increase in the plasma levels of both PQ and its major metabolite, CPQ. Thus, these two drugs, while sharing a beneficial effect on PQ efficacy and toxicity, apparently do so by affecting mechanisms other than the monoamine oxidase-mediated metabolism of PQ to CPQ. Other pathways of PQ metabolism, now known to be prominent in human liver—but so far difficult to pinpoint their ultimate disposition—have not been explored from the perspective of CQ and QN effects.
|
review
| 99.7 |
More recent studies with PQ have developed compelling evidence that the metabolism to its active metabolites is dependent on the activity of cytochrome P450 2D6 (CYP2D6) [7, 17]. This evidence is based on the failure of PQ therapy in CYP2D6-null metabolizers and the ablation of the anti-malarial effect of PQ in CYP2D knockout mice. Thus, it would naturally be expected that inhibition of CYP2D6 would impair the efficacy of PQ. However, both QN and CQ have been shown to be moderate inhibitors of CYP2D6 with significant effects at clinically used doses [18, 19]. It is puzzling then, that they do not effectively antagonize the anti-malarial effects of PQ, and rather are potentiators. This suggests that either the putative dependence of PQ efficacy on CYP2D6 is incorrect, or that CQ and QN may be acting by other mechanisms.
|
study
| 67.5 |
Recent studies of PQ metabolism have revealed interesting new information about the metabolic pathways, suggesting that in humans the two enantiomers of PQ (which is used as a racemic mixture) are differentially metabolized to CPQ . While the R(–) form of PQ is readily converted to CPQ, the S(+) form is much less so, and in fact >95 % of the CPQ circulating is attributable to R(–) PQ. This raises the question as to the primary metabolic route for the S(+) form. Furthermore, other studies revealed several other major metabolites of PQ generated in human hepatocytes and some of these are formed predominantly or exclusively from one enantiomer of PQ [9, 21].
|
study
| 99.94 |
Racemic, (S)-(+)- and (R)-(−)- enantiomers of 13C(6)-labelled PQ (MW 259.1317 g/mol); PQ alcohol, 2-, 3-, and 4-hydroxyprimaquine and CPQ were synthesized as previously reported [9, 22–24]. The identity of the compounds synthesized was confirmed by spectral infra-red (IR), nuclear magnetic resonance (NMR), high-resolution mass spectrometry and physical data in comparison with published values.
|
study
| 100.0 |
Co-factors for enzyme activity, including glucose-6-phosphate and its dehydrogenase, reduced nicotinamide adenine dinucleotide phosphate and magnesium chloride were purchased from Sigma-Aldrich (St Louis, MO, USA). Analytical grade solvents, including acetonitrile and methanol were purchased from Fisher Scientific (Fair Lawn, NJ, USA). Milli-Q system (Millipore, Bedford, MA, USA) was used to purify water for analytical utility.
|
other
| 99.9 |
Recombinant cytochrome-P450 supersomes containing CYP2D6 (1 nmol CYP/mL), and recombinant monoamine oxidase-A (MAO) were purchased from BD Biosciences (Billerica, MA, USA) and stored at −80 °C until used. Cryopreserved, metabolism-qualified hepatocytes (mixed-gender pooled) were obtained from Corning Life Sciences (Woburn, MA, USA) and stored in liquid nitrogen tank until use. Just before use, the metabolic systems were thawed according to supplier instructions. Typically, the vials containing the cryopreserved metabolic systems were dipped (without submerging) into water bath set at 37 °C for not more than 2 min in order to defrost the contents. The vials are then wiped with 70 % alcohol before the contents are poured into the recovery medium or incubation buffers.
|
study
| 99.9 |
Typical assay with rCYP2D6 and MAO involved the dilution of the thawed enzymes with 0.1 mM phosphate buffered saline (pH 7.4), aliquot dispensing in clear 96-well plates, and the subsequent addition of the substrate (PQ) and the co-factor solution to initiate metabolic reactions. The final concentrations of additives in the incubation mixture of 500 µL are as follows: 20 µM PQ, 0.5 mg/mL of enzyme (CYP2D6 or MAO-A), 1.0 unit/ml of G6PD, 1.3 mM reduced nicotinamide adenine dinucleotide phosphate and 3.3 mM each of glucose-6-phosphate and magnesium chloride. Stock solutions of CQ/QN was prepared in water and added to the wells to a final concentration of 5 µM; an equivalent volume of buffer was included in control wells. Metabolic activity of the enzymes was halted at pre-determined times by the addition of one-part ice-cold methanol containing 0.5 µg/mL D3-primaquine as internal standard. The samples were kept on ice for an hour and then centrifuged (14,000 rpm; 4 °C; 10 min). Clear supernatants were collected for ultra-performance liquid chromatography-mass spectrometry (UHPLC-MS) analysis. All assays were performed in duplicate to observe intra-assay variations and repeated at separate days for inter-day comparisons. Controls included substrate incubation without enzyme or start solutions and incubations without the substrate.
|
study
| 100.0 |
Suspensions of pooled human hepatocytes were thawed from cryopreserved vials and recovered with the supplier-provided recovery media (containing buffered saline with 2 % fetal bovine serum and 300 mM glucose). This was followed by centrifugation (300g, 5 min) to pellet the cells, which were then re-suspended in the plating media provided by the suppliers. Cell viability was determined based on cell counts using a Bio-Rad automated cell counter (Hercules, CA, USA). The cell density was adjusted to approximately 1 million viable cells/mL in the plating media and incubated with 20 µM PQ at 37 °C in a humidified atmosphere of 95 % air and 5 % CO2 in an Eppendorf incubator (Hauppauge, NY, USA) with an attached shaker set at 100 rpm. In addition, incubation plates contained 5 µM CQ/QN or equal volume of the plating media. In parallel, substrate-free and cell-free incubations were performed as controls. At each time point, aliquots were taken for cell viability measurement. Reactions were quenched using two volumes of ice-cold methanol containing 0.5 µg/mL 6-D3-primaquine as internal standard at pre-determined time points (between 0 and 120 min).
|
study
| 100.0 |
Once metabolic reactions were quenched, the incubation mixtures were vortexed and kept at −20 °C for at least 4 h before further analysis. Clear supernatants of centrifuged samples (10,000g, 4 °C, 10 min) were dried in a Speedvac® and reconstituted in methanol for UHPLC-MS analysis.
|
study
| 99.9 |
Methods for simultaneous analysis of PQ and its metabolites using the UHPLC-MS were reported earlier . This involved the use of a BEH Shield RP18 column (100 mm × 2.1 mm ID, 1.7 mm) equipped with an LC-18 guard column (Vanguard 2.1 × 5 mm, Waters Corp, Milford, MA, USA) on an ACQUITY UHPLC™ system (Waters) to which a conditioned auto-sampler (at 20 °C) was attached. Metabolites earlier identified, based on retention time, high resolution mass and MS–MS fragmentation pattern, were individually searched for and profiled against time in the presence and absence of CQ/QN.
|
study
| 100.0 |
PQ is metabolized rapidly by both CYP2D6 and MAO. CYP2D6 is known to mediate ring oxidation and demethylation steps, while MAO catalyzes oxidative deamination of the side chain. The effect of 5 µM each of CQ and QN on the generation of known CYP2D6- and MAO-dependent metabolites of 20 µM PQ or its individual enantiomers was assessed using recombinant human enzymes. The influence of the two anti-malarial drugs on the generation of secondary metabolites in human hepatocytes was also investigated. In general, the oxidation of the quinoline ring proved very sensitive to the presence of CQ/QN, whereas the formation of the deaminated products was generally not markedly inhibited. The structures of the major metabolites are presented in Fig. 1.Fig. 1A putative pathway of PQ metabolism in human hepatocytes. Uridine-diphosphate-glucuronic acid which is co-factor necessary for glucuronidation catalyzed by UDPGT (Uridine-diphosphate glucuronosyltransferase); CYP cytochrome P450; ALDH aldehyde dehydrogenase) (First published by Fasinu et al. )
|
study
| 100.0 |
A putative pathway of PQ metabolism in human hepatocytes. Uridine-diphosphate-glucuronic acid which is co-factor necessary for glucuronidation catalyzed by UDPGT (Uridine-diphosphate glucuronosyltransferase); CYP cytochrome P450; ALDH aldehyde dehydrogenase) (First published by Fasinu et al. )
|
other
| 96.5 |
PQ was steadily depleted to varying degrees in the incubation mixtures with recombinant CYP2D6 such that by 2 h, 65, 50 and 72 % of the racemic, (+)- and (−)-PQ remained, respectively. Both CQ and QN suppressed this depletion with all three substrates such that more than 90 % of the substrates remained after 2 h (Fig. 2a–c).Fig. 2The influence of chloroquine and quinine on the CYP2D6-mediated depletion of racemic PQ (a) and its enantiomers (b, c); and formation of 2-hydroxyprimaquine from racemic PQ (d) and e, f over 2 h of incubation. Each point represents values: mean ± S.D. (n = 4)
|
study
| 100.0 |
The influence of chloroquine and quinine on the CYP2D6-mediated depletion of racemic PQ (a) and its enantiomers (b, c); and formation of 2-hydroxyprimaquine from racemic PQ (d) and e, f over 2 h of incubation. Each point represents values: mean ± S.D. (n = 4)
|
study
| 99.94 |
The major metabolites from CYP-2D6-mediated pathway were 2-, 3-, and 4-hydroxyprimaquine, 5,6-orthoquinone product from 5-hydroxyprimaquine and PQ alcohol (as earlier reported) . The formation of 2-hydroxyprimaquine and the 5,6-orthoquinone predominated with (+)-PQ, while the 3-hydroxy- and 4-hydroxyprimaquine were formed more from the (−)-PQ. Quantitatively, 2-hydroxyprimaquine was four times more abundant from (+)-PQ than from (−)-PQ. Although the concentration of 2-hydroxyprimaquine peaked at 20 min with all substrates, 50 % of the peak concentration was already achieved by 5 min. CQ and QN completely suppressed the initial metabolite surge, with some metabolite appearing after an initial lag time (Fig. 2).
|
study
| 100.0 |
Unlike 2-hydroxyprimaquine, 3- and 4-hydroxyprimaquine were formed more readily from (−)-PQ than in (+)-PQ (2:1 and 5:1, respectively) (Fig. 3). Both 3-hydroxy- and 4-hydroxyprimaquine have sharp early rises, which were again completely suppressed by CQ and QN at the early stages, then appearing in limited amounts after 5 min. The inhibitory effect of QN on the formation of the monohydroxyprimaquines appeared to be slightly more than the equimolar CQ, but the two were generally comparable. Both CQ and QN completely suppressed the formation of the 5,6-orthoquinone (Fig. 4), presumptively derived from 5-hydroxyprimaquine. Both CQ and QN show only moderate inhibition on CYP2D6-catalyzed formation of the PQ terminal alcohol. Interestingly, in contrast to the effect on ring-hydroxylated metabolites, CQ caused greater inhibition (about 70 %) of CYP2D6-mediated formation of PQ alcohol from (+)-PQ while QN caused only about 45 % inhibition, respectively (Fig. 5).Fig. 3The influence of chloroquine and quinine on the CYP2D6-mediated formation and kinetics of 3-hydroxyprimaquine from racemic PQ (a) and its individual enantiomers (b, c); and on the formation of 4-hydroxyprimaquine from racemic PQ (d) and its enantiomers (e, f). Each point represents values mean ± S.D. (n = 4)Fig. 4The inhibitory effect of chloroquine and quinine on the formation and time-course of the PQ 5,6-orthoquinone generated from racemic PQ (a) and its individual enantiomers (b, c) by CYP2D6; and on the formation of primaquine alcohol from CYP2D6-catalyzed metabolism of racemic PQ (d) and its enantiomers (e, f). Each point represents values mean ± S.D. (n = 4)Fig. 5Comparative depletion of racemic PQ (a) and its enantiomers (b, c) on incubation with recombinant human monoamine oxidase; and the formation of PQ alcohol (d–f) from racemic PQ (d) and its enantiomers (e, f) by the same enzyme. Each point represents values mean ± S.D. (n = 4)
|
study
| 100.0 |
The influence of chloroquine and quinine on the CYP2D6-mediated formation and kinetics of 3-hydroxyprimaquine from racemic PQ (a) and its individual enantiomers (b, c); and on the formation of 4-hydroxyprimaquine from racemic PQ (d) and its enantiomers (e, f). Each point represents values mean ± S.D. (n = 4)
|
study
| 100.0 |
The inhibitory effect of chloroquine and quinine on the formation and time-course of the PQ 5,6-orthoquinone generated from racemic PQ (a) and its individual enantiomers (b, c) by CYP2D6; and on the formation of primaquine alcohol from CYP2D6-catalyzed metabolism of racemic PQ (d) and its enantiomers (e, f). Each point represents values mean ± S.D. (n = 4)
|
study
| 100.0 |
Comparative depletion of racemic PQ (a) and its enantiomers (b, c) on incubation with recombinant human monoamine oxidase; and the formation of PQ alcohol (d–f) from racemic PQ (d) and its enantiomers (e, f) by the same enzyme. Each point represents values mean ± S.D. (n = 4)
|
study
| 99.94 |
Other minor metabolites generated by CYP2D6-mediated metabolism of PQ include dihydroxyprimaquine (m/z 292) and two that appear to be quinone-imine derivatives of dihydroxyprimaquine (m/z 290). All of these minor metabolites were completely suppressed by CQ and QN.
|
study
| 99.9 |
The metabolism of PQ by MAO-A showed a marked enantioselectivity, with (−)-PQ being depleted four times faster than (+)-PQ (Fig. 2b). This depletion was not significantly affected by CQ or QN with any of the substrates until 1 h, after which a very modest inhibition was observed (Fig. 5a–c). The only major metabolite from MAO-dependent metabolism was the alcohol formed by oxidative deamination of the terminal amine. It is formed twice as much from (−)-PQ compared to (+)-PQ. By 2 h, CQ and QN caused about 60 % inhibition of the formation of PQ alcohol with both PQ enantiomers (Fig. 5d–f).
|
study
| 100.0 |
PQ depletion and metabolite production from hepatocyte incubation monitored via UHPLC-MS/MS demonstrated a faster metabolic rate for (−)-PQ as compared to (+)-PQ. The rate of metabolism was profiled against time, yielding an initial half-life of 3.65 and 1.08 h for (+)- and (−)-PQ, and elimination rate constants (ʎ) of 0.19 and 0.64 h−1, respectively. The calculated in vitro intrinsic clearance (Clint) was 2.55 and 8.49 (µL/min)/million cells for (+) and (−)-PQ, respectively. Metabolism was observed via oxidative deamination of the terminal amine, hydroxylation of the quinoline ring, and what appears to be a third path, carbamoylation and glucuronidation of the terminal amine (m/z 480). The major deaminated metabolites were CPQ, PQ terminal alcohol, a cyclized side chain derivative from the aldehyde (m/z 241), cyclized carboxylic acid derivative (m/z 257), a quinone-imine product of hydroxylated CPQ (m/z 289), CPQ glucuronide (m/z 451), and the glucuronide of PQ alcohol (m/z 437). All were preferentially generated from the (-)-PQ. The major quinoline oxidation product (m/z 274) and 4-hydroxyprimaquine were preferentially formed from (+)-PQ. A prominent conjugate (m/z 422) (seemingly a glycosylated PQ, (but perhaps reflecting a fragment or degradation product of another metabolite) was also preferentially (+)-PQ-derived. The PQ carbamoyl glucuronide metabolite (m/z 480) accumulated linearly over the incubation period, and was exclusively generated from (+)-PQ.
|
study
| 100.0 |
CQ and QN had no effect on the primary amine oxidation products (CPQ or PQ alcohol or the cyclized aldehyde) (Fig. 6). However, surprisingly they did inhibit substantially the appearance of the glucuronides of CPQ and PQ alcohol. They also inhibited the appearance of the m/z 257 metabolite with a similar pattern, suggesting that it may be derived from the CPQ conjugate. The formation of 4-OH-PQ and the quinone-imine (m/z 274) were essentially completely suppressed, and the m/z 422 metabolite was very strongly suppressed (90–95 %) by both CQ and QN. The apparent quinone-imine of CPQ (m/z 289) was partially suppressed, but more prominently by CQ than QN. The carbamoyl glucuronide of PQ (m/z 480) was not affected.Fig. 6The inhibitory effects of chloroquine and quinine on the formation of metabolites from primaquine and its (+)- and (−)-enantiomers after one-hour incubation in human hepatocytes. Each point represents mean of 4 determinations. The major metabolites were carboxyprimaquine [predominantly from (−)-PQ], carbamoyl primaquine glucuronide [exclusively from (+)-PQ] and glycosylated primaquine (formed from both enantiomers). The inhibitory effect of chloroquine and quinine on the formation of carboxyprimaquine and carbamoyl primaquine glucuronide was mild
|
study
| 100.0 |
The inhibitory effects of chloroquine and quinine on the formation of metabolites from primaquine and its (+)- and (−)-enantiomers after one-hour incubation in human hepatocytes. Each point represents mean of 4 determinations. The major metabolites were carboxyprimaquine [predominantly from (−)-PQ], carbamoyl primaquine glucuronide [exclusively from (+)-PQ] and glycosylated primaquine (formed from both enantiomers). The inhibitory effect of chloroquine and quinine on the formation of carboxyprimaquine and carbamoyl primaquine glucuronide was mild
|
study
| 100.0 |
The unique therapeutic indications for PQ and the absence of alternatives have made it indispensable in the treatment of P. vivax malaria, despite its known haemotoxicity in individuals with genetic deficiency in G6PD. The efficacy and the haemotoxicity of PQ, as well as other 8-AQs, appears to be related to active metabolites, although the specific pathways and metabolites responsible, as well as their sites of formation and mechanisms of action, remain unclear.
|
review
| 99.44 |
Following earlier substantive clinical reports [13, 15, 16] that CQ and QN are capable of altering the therapeutic and toxicological responses to 8-AQs, including PQ, and from recent findings of new human metabolites of PQ [9, 21], it is important to understand how CQ and QN might alter human PQ metabolism. The present study aimed to address these issues.
|
study
| 99.94 |
In an overview, PQ is metabolized via two primary phase I pathways, namely: hydroxylation products derived from CYP-mediated oxidation, especially CYP2D6; and oxidative deamination of the side chain affording the terminal alcohol or carboxylic acid, primarily mediated by MAO-A (though CYP2D6 can also produce the PQ alcohol). In addition, a number of glucuronide conjugates of these phase I metabolites are formed, and an unusual (but not unknown, [26, 27]) carbamoylation/glucuronidation of the parent is observed . CYP2D6-mediated metabolism of PQ is known to produce mono- and dihydroxylated primaquines and their quinone-imine products . In human hepatocytes, a more limited suite of metabolites was observed, with the ring oxidation products quinone-imine m/z 274 and the 4-OH-PQ predominant from (+)-PQ, while the oxidation of the terminal amine was the major metabolic route with (−)-PQ . In addition, the PQ-carbamoyl-glucuronide was exclusively formed from (+)-PQ.
|
review
| 99.9 |
Another metabolite (m/z 422, apparent PQ-glucoside) seems to be predominantly derived from (+)-PQ, perhaps indirectly. It is of interest in that although it can be formed non-enzymatically, it also appears to arise in hepatocytes by an apparent enzymatic mechanism. Further studies are underway on this metabolite to determine its origin.
|
study
| 99.56 |
The effects of CQ and QN observed in the present studies were most striking for CYP2D6 inhibition, with very modest effects on the oxidative deamination products of MAO. In hepatocytes, the inhibitory effects were most striking on the CYP2D6 related products. The presence of CQ and QN did not appear to significantly inhibit the MAO-dependent pathway in hepatocytes. The inhibitory effects on CYP2D6 appeared to enhance the MAO-mediated metabolism as demonstrated in a mild increase in formation of CPQ, and the metabolites corresponding to m/z 241. A surprising inhibition was observed, however, on the formation of the glucuronides of CPQ and PQ alcohol in the human hepatocytes. This would be expected to facilitate the accumulation of the primary oxidative deamination metabolites. This finding is consistent with the report that CQ increased circulating CPQ concentrations when co-administered with PQ in humans . The implications of this are not known, but such metabolites could theoretically exert more pronounced pharmacodynamic effects and may contribute to the observed alteration in PQ response in the presence of CQ and QN.
|
study
| 100.0 |
The profound effects of CQ and QN on the CYP2D6-generated metabolites was not particularly surprising result, since both of these drugs have demonstrated potential for inhibition of CYP2D6 in human studies [18, 28]. The metabolism results from the incubation with human hepatocytes might be presumed to provide a closer picture to in vivo realities. Although CQ/QN suppressed the extent of the formation of a number of metabolites in the human hepatocytes, their inability to alter the clearance and other pharmacokinetic parameters of PQ suggests that the facilitation of one metabolic pathway counters the inhibitory effect on the other. Any observed pharmacodynamic changes therefore could result from the changes in the metabolite profiles.
|
study
| 100.0 |
For moderate CYP2D6 inhibitors like CQ and QN, substrate-specific or orientation/conformation-specific inhibition may be expected. This may explain varying degrees of inhibition on the individual CYP2D6 metabolites of PQ or the individual enantiomer substrates. However, such dramatic inhibition of the formation of the mono-hydroxylated primaquines as was observed at the early time points may have significant pharmacodynamic implications. This might favour alternative metabolic pathways leading to significant changes in therapeutic response, and may play major roles in the earlier observation of mitigated toxicity when PQ was combined with CQ and QN.
|
study
| 100.0 |
However, it seems difficult to harmonize these findings with the well-demonstrated enhancement of PQ anti-malarial efficacy by QN and CQ and the recent observations suggesting that CYP2D6 activity is essential to the anti-malarial efficacy of PQ [17, 29]. Overall, it raises the question of other mechanisms of action for these drugs, and perhaps important roles of other PQ metabolites that are not CYP2D6-dependent. For example, CQ and QN may affect transport processes for cellular uptake or extrusion of PQ or metabolites in a manner that could affect the overall efficacy or toxicity of PQ. Or the pathway to hydroxylated PQ products from other CYPs could be favoured, yielding active metabolites not seen in the absence of CQ or QN. In addition, the abundance of deaminated products in the presence of CQ/QN might be shunted into pathways that yield active metabolites with reduced toxicity. All of these possibilities must await further study, and especially the study in the clinic, where the interaction of PQ with QN and CQ can be assessed in terms of new knowledge of PQ metabolites and new methods for quantitating these.
|
review
| 73.8 |
This study has been able to assess the influence of CQ/QN on the multiple pathways of PQ metabolism. The identified PQ metabolites showed products of CYP-dependent ring oxidation, MAO-catalyzed oxidation of the terminal amine and phase II conjugation products. Both CQ and QN strongly inhibited the ring oxidation pathway. The suppression of CYP-generated metabolites in favor of the oxidative deamination of the side chain and the abundance of the resultant deaminated metabolites may raise the question of the roles of these metabolites in PQ activity and toxicity. The phase II reactions were also moderately inhibited by CQ/QN. The oxidation of the terminal amines and associated metabolites were generally not influenced by the presence of CQ and QN.
|
study
| 100.0 |
Ionizing radiation is the main cause of death for cancer cells in radiation therapy, but many cancers for example melanoma are not sensitive to radiation therapy, resulting in poor clinic effects [1, 2]. Radiotherapy is one of the most important methods of cancer treatment along with surgery and chemotherapy. Radiotherapy was administered to shrink tumor in advanced melanoma, or prevent tumor relapse after surgical treatment [3, 4].
|
review
| 99.8 |
In addition to killing cancer cells, radiation therapy also leads to damage of normal cells and tissues. It is therefore an urgent medical concern to protect the normal cells in addition to killing cancer cells as many as possible. To this end, studies on radiosensitization are receiving more and more attention in radiobiology [5, 6]. As a comprehensively used reagent, a radiosensitizer can facilitate the sensitivity of cancer cells in response to radiation, which accordingly promotes the effects of therapy by increasing radiation-mediated cancer cell death [7, 8]. Initially, radiosensitizers were used in radiation-resistant anaerobic cells in solid tumors, but this application has now extended to other cell types in cancers . Though a radiosensitizer is effective in increasing the effectiveness of radiation therapy for cancer, most radiosensitizers are chemotherapy drugs, with unavoidable toxicity for normal cells. Therefore, the search for low-toxic and highly effective radiosensitization is an urgent need in tumor radiation therapy.
|
review
| 99.9 |
The nuclear magnetic resonance (NMR) spectrum–based cellular metabolome analysis is a novel method to assist radiosensitization study in tumor radiation therapy. 1H NMR analysis is an ideal high throughput method to detect small-metabolites in biological samples [10, 11]. With pattern recognition, 1H-NMR-based metabolome analysis screens differentially expressed metabolites between different samples . Followed by bioinformatics analysis on the basis of a public database, the changed metabolic pathways can be further identified.
|
study
| 99.94 |
In this study, using radiation-tolerant mouse melanoma cell line B16 as research model , cells were first radiated with sublethal doses of X-ray radiation; then cellular metabolites were collected, which were further analyzed by 600 MHz 1H NMR to screen differentially expressed metabolites. Following bioinformatic analysis for those quantified metabolites, potential targets of disturbed metabolic pathway can be used for radiosensitization study to discover high efficacy drugs for sensitization in radiotherapy. The underlying mechanism of metabolome changes and radiation resistance in B16 cells was discussed.
|
study
| 100.0 |
Mouse melanoma cell line B16 (obtained from China Center for Type Culture Collection, Wuhan University, Wuhan, China) was maintained in a complete RPMI-1640 medium supplemented with 10% fetal bovine serum and 1% Penicillin-Streptomycin. After a sublethal dose of X-ray radiation (18 Gy) using RS2000 X-ray irradiator (Rad Source Technologies Inc., GA, U.S.), 1×106 cells were subcultured in 25 cm2 plates in 5% CO2 at 37°C. Untreated cells cultured in the same conditions were used as control cells. Cells were harvested after 48 h culture for further experiments. The experiment was repeated 3 times.
|
study
| 99.94 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.