|Home | About | Journals | Submit | Contact Us | Français|
Approximately 15% of colorectal carcinomas (CRC) exhibit a hypermutated genotype accompanied by high levels of microsatellite instability (MSI-H) and defects in DNA mismatch repair. These tumors, unlike the majority of colorectal carcinomas, are often diploid, exhibit frequent epigenetic silencing of the MLH1 DNA mismatch repair gene, and have a better clinical prognosis. As an adjunct study to The Cancer Genome Atlas consortium that recently analyzed 224 colorectal cancers by whole exome sequencing, we compared the 35 CRC (15.6%) with a hypermutated genotype to those with a non-hypermutated genotype. We found that 22 (63%) of hypermutated CRC exhibited transcriptional silencing of the MLH1 gene, a high frequency of BRAF V600E gene mutations and infrequent APC and KRAS mutations, a mutational pattern significantly different from their non-hypermutated counterparts. However, the remaining 13 (37%) hypermutated CRC lacked MLH1 silencing, contained tumors with the highest mutation rates (“ultramutated” CRC), and exhibited higher incidences of APC and KRAS mutations, but infrequent BRAF mutations. These patterns were confirmed in an independent validation set of 250 exome-sequenced CRC. Analysis of mRNA and microRNA expression signatures revealed that hypermutated CRC with MLH1 silencing had greatly reduced levels of WNT signaling and increased BRAF signaling relative non-hypermutated CRC. Our findings suggest that hypermutated CRC include one subgroup with fundamentally different pathways to malignancy than the majority of CRC. Examination of MLH1 expression status and frequencies of APC, KRAS, and BRAF mutation in CRC may provide a useful diagnostic tool that could supplement the standard microsatellite instability assays and influence therapeutic decisions.
Colon and rectal cancers (CRCs) are the third most frequently arising cancer in both men and women in the United States . While progress has been made in CRC treatment, mortality rates remain high . Some inherited and somatically arising CRC exhibit genetic instability in simple tandem repeat sequences called microsatellite DNA, often associated with defective DNA mismatch repair [3,4]. About 15% of CRC exhibit microsatellite instability (MSI) . Pathologically, these tumors tend to arise in the proximal colon, display lymphocytic infiltration, are often mucinous, and respond to chemotherapeutics differently than microsatellite stable tumors . About 75% of CRCs with microsatellite instability display hypermethylation and transcriptional silencing of the MLH1 gene [6,7]. A frequent genetic alteration in CRC with MSI is the oncogenic BRAF V600E mutation. BRAF mutations in sporadic CRC have been associated with increased mortality .
Truncating mutations in one or both adenomatous polyposis coli (APC) alleles occur early during CRC evolution, occurring at roughly 80% frequency . APC truncating mutations result in constitutive activation of WNT-mediated growth signaling . Later genetic events in CRC include oncogenic KRAS mutations and inactivating p53 tumor suppressor mutations . KRAS mutation status has been a useful marker in CRC therapeutic interventions with epidermal growth factor receptor-specific antibodies. CRC with wildtype KRAS were more likely to show therapeutic benefit when treated with these therapeutic antibodies . Sporadic CRC with and without MSI display distinctive patterns of mutations in cancer-associated genes. Unlike CRC with microsatellite stability, CRC with MSI exhibit reduced frequencies of APC, p53 and KRAS mutations, as well as increases in TGFBR2 mutations and the aforementioned BRAF mutations [12–15].
The Velculescu, Kinzler, and Vogelstein laboratories have pioneered genomic surveys of colorectal cancers, identifying novel altered genes and chromosomal aberrations [16–18]. Recently, The Cancer Genome Atlas (TCGA) research consortium analyzed 224 tumor/normal CRC pairs by integrating data from whole exome sequencing, DNA copy number variation, promoter methylation, and global mRNA and microRNA expression . In the TCGA study CRC were subdivided into 15% hypermutated (>12 non-silent exonic mutations/megabase) and 85% non-hypermutated (<12 non-silent exonic mutations/megabase) populations. Hypermutated CRC were distinguished by high rates of MLH1 gene silencing and high MSI. Here, in an adjunct paper to the TCGA study, we analyze key features of the hypermutated CRC. We show that the hypermutated CRC can be usefully subcategorized into MLH1-silenced and non-MLH1-silenced groups. The MLH1-silenced hypermutated CRC exhibit low numbers of APC mutations and high rates of BRAF V600E mutations. Analysis of mRNA and miRNA expression patterns indicates that this group of tumors may not rely on activated WNT signaling for tumor initiation and progression. Thus, we propose that classification of CRC by MLH1 silencing status in addition to MSI status may provide an alternative and useful way to predict CRC pathobiology and therapy responses.
The analyses described here were from 474 tumor/normal CRC on which whole exome sequencing was performed under the auspices of The Cancer Genome Atlas (TCGA) Consortium. Details of sample acquisition, DNA sequencing, DNA copy number studies, RNA expression, miRNA expression, microsatellite instability analyses, and epigenomic analyses were described in the parent TCGA CRC publication  (which featured 224 of the 474 tumors). The 224 samples in the discovery cohort were sequenced by the genome centers at Washington University School of Medicine and Baylor College of Medicine . The 250 samples in the validation cohort were sequenced at the Baylor College of Medicine Human Genome Sequencing Center. All data collected by the TCGA Consortium provided for full patient privacy protection and were in accordance with established ethical standards.
The 224 CRC discovery set (featured in the initial publication ) that were whole exome sequenced were divided into two categories based on mutation rates: hypermutated (>12 non-silent mutations per 106 exonic nucleotides) and non-hypermutated (<12 non-silent mutations per 106 exonic nucleotides). The 35 hypermutated CRC were then subdivided into 22 MLH1-silenced CRC and 13 non-MLH1-silenced CRC, and were analyzed separately from non-hypermutated CRC. While all 22 MLH1-silenced hypermutated CRC appeared to have normal germline mismatch repair genes (MSH2, MSH6, MLH1, and PMS2), 2 of the 13 non-MLH1-silenced hyprmutated tumors may have been Lynch syndrome CRC, as patient AA-A00A exhibited a germline frameshift mutation in PMS2 and patient AA-A01R displayed a frameshift mutation in MSH6. Of the 189 non-hypermutated CRC, 163 had complete methylation data including methylation status of the MLH1 gene. Validated mutations in APC, KRAS, NRAS, BRAF, TP53, SMAD2, SMAD3, SMAD4, ACVR2A, ACVR1B, and TGFBR2 were determined for all tumors and mutation frequencies of individual genes or gene groups (e.g. TGF-β pathway members) in each of the three CRC categories were compared by Fisher’s exact test or chi-square test as indicated.
A second validation set of 250 CRC was analyzed by whole exome sequencing and global methylation status (including MLH1). Of this second set, 29 (11.6%) were categorized as MLH1-methylated (top 20% of methylation beta values, based on Illuminal array probe cg02279071 and the initial analysis of the discovery set) and hypermutated (>12 mutations per exome Mb), 16 (6.4%) were hypermutated and did not exhibit MLH1 methylation, while the remaining 205 (82%) CRC were classified as non-hypermutated (<12 mutations/Mb).
To compare mRNA expression patterns, we acquired the RNAseq data for those tumors among the 224 discovery CRC with this data available. The tumors were subdivided into the three categories, MLH1-silenced hypermutated (N = 20), non-MLH1-silenced hypermutated (N = 9) and non-hypermutated (N = 140). Mean expression levels for each of 20,533 genes in the three CRC categories were calculated and ratios of mean expression between MLH1-silenced CRC and non-hypermutated CRC was determined. Student’s t test was used to identify those genes whose expression were significantly different between these two categories. Those genes with the lowest P values by t test were considered to be most significantly different in their expression.
Gene transcription signature of pathways were defined as follows: p53, canonical bound and up-regulated p53 gene targets, as catalogued in the p53 IARC database (http://www-p53.iarc.fr/TargetGenes.html); KRAS, genes differentially expressed between KRAS-dependent and KRAS-independent cell lines, as previously defined by Singh et al. ; BRAF, genes induced in melanocytes after activation of BRAF (P<0.05, t-test, fold>1.4, using GSE13827 dataset); TGF-β, genes differentially expressed in TGF-beta-treated epithelial cell lines, as previously defined by Padua et al. ; Using the RNA-seq expression profiles, the TCGA colorectal tumors were scored for relative activity of a given pathway signature; for signatures consisting of up-regulated genes the signature score within each tumor was defined as the Spearman’s rank sum statistic of the RPKM values of the genes; for signature consisting of both up-regulated and down-regulated genes, the signature score was defined as the Spearman’s rank sum statistic of the up-regulated genes, minus the rank sum statistic of the down-regulated genes. Java Treeview  represented mutation/LOH/gene signature patterns as heat maps.
In the recently reported TCGA study, 224 CRC normal/tumor sample pairs were analyzed by whole exome sequencing, global DNA copy number, promoter methylation, mRNA and miRNA expression . Of 224 CRC, 35 (15.6%) exhibited “hypermutation” with at least 12.0 non-silent mutations per exome megabase. All but one of the remaining tumors had a mutation rate below 7.0 non-silent mutations per megabase and were classified “non-hypermutated” . For our discovery studies we compared the 35 hypermutated and 189 non-hypermutated tumors for which there was complete exome sequencing data. As a validation step we analyzed an additional 250 CRC (not represented in the original TCGA study) by exome sequencing and assessed their global methylation status. Among the validation set, 45 of 250 CRC were categorized as hypermutated and the remaining 205 CRC were non-hypermutated (Figure 1). Four CRC had extremely high mutation rates and were classified “ultramutated”.
In the TCGA study the majority of hypermutated tumors exhibited high microsatellite instability (MSI-H) and silenced MLH1 expression . Overall, hypermutated CRC had fewer APC, KRAS, and TP53 mutations compared to the non-hypermutated CRC . In contrast, mutations in TGF-β signaling genes and BRAF were dramatically elevated in the hypermutated tumors. To determine whether these mutational patterns were more specific to a subset of the hypermutated CRC, we stratified the 224 CRC discovery set by MLH1 silencing or MSI status. When the hypermutated tumors were subcategorized according to MLH1 silencing status (MLH1 promoter methylation and low MLH1 expression), such tumors had a very low APC mutation rate (Figure 2A). In the 224 tumor discovery set, only 3 of 22 MLH1-silenced hypermutated CRC displayed a truncating APC mutation, while 12 of 13 non-MLH1-silenced hypermutated CRC exhibited at least one truncating APC mutation (P = 3 X 10−12) (Figure 2A,B). Consistent with previous reports, 152 of 189 (80.4%) non-hypermutated CRC contained at least one truncating APC mutation. In addition, hypermutated CRC with MLH1 silencing had the highest frequency of MSI-H (Figure 2A).
Stratification of the 45 hypermutated CRC in the 250 tumor validation set by MLH1 promoter methylation status indicated 29 MLH1 methylated CRC and 16 non-MLH1 methylated CRC. Only 6 of 29 (20.7%) MLH1 methylated CRC contained APC truncating mutations, whereas 8 of 16 (50%) non-MLH1 methylated CRC contained truncating APC mutations (Figure 2C; Table S1). In contrast, 176 of 205 (85.9%) non-hypermutated CRC exhibited at least one truncating APC mutation (Figure 2C, Figure S1). Thus, the significantly reduced APC mutation rate in the MLH1 methylated hypermutated CRC of the validation set corroborated the low APC mutation numbers observed in the original discovery set.
For the discovery set, the non-MLH1-silenced hypermutated CRC exhibited more variation in MSI status. While 21 of 22 MLH1-silenced hypermutated tumors displayed MSI-H (high microsatellite instability), only 6 of 13 non-MLH1-silenced CRC were in this category (and non-hypermutated CRC rarely had MSI-H). Despite having lower frequencies of MSI-H, the non-MLH1-silenced hypermutated CRC had higher mean mutation rates (100 non-silent mutations per megabase) than their MLH1-silenced counterparts (27.8 non-silent mutations per megabase) (P = 0.032). The hypermutated CRC with the seven highest mutation rates were considered “ultramutated” and were all non-MLH1-silenced tumors (Figure 2A).
Truncating mutations in APC in colorectal cancer result in stabilization and nuclear localization of beta-catenin and transcriptional activation of WNT signaling targets . About 80% of CRC have APC truncating mutations that activate WNT signaling, but MLH1-silenced hypermutated CRC have low frequencies of APC truncating mutations (Figure 2). This suggests that most MLH1-silenced CRC may not have activated WNT signaling. To test this, we compared RNA expression profiles as determined by RNAseq. Using the 224 tumor discovery set, we compared RNA expression levels of 20,533 genes in 20 MLH1-silenced hypermutated CRC and 9 non-MLH1-silenced hypermutated CRC to that in 140 non-hypermutated CRC for which complete RNA expression data was available. The genes were then sorted by those most differentially regulated between each of the two hypermutated CRC categories and the non-hypermutated CRC. As expected, when comparing the MLH1-silenced CRC to the non-hypermutated CRC, the most differentially regulated gene was MLH1, with 4-fold higher expression in the non-hypermutated CRC compared to the hypermutated MLH1-silenced CRC (P = 2.29 X 10−41) (Figure 3A). The second most differentially regulated gene was AXIN2, a WNT signaling regulator and target, which was increased 7-fold in the non-hypermutated CRC compared to the MLH1-silenced hypermutated CRC (P = 5.61 X 10−27) (Figure 3A).
The reduced AXIN2 in the hypermutated MLH1-silenced CRC indicated that WNT signaling is downregulated in these tumors. We compared RNA expression datasets for over-represented gene ontogeny categories using the software provided by the Database for Annotation, Visualization and Integrated Discovery (DAVID) and the Broad Institute Gene Set Enrichment Analysis (GSEA) Websites. Both software packages showed that WNT signaling pathway genes were significantly over-represented among the most differentially regulated genes between MLH1-silenced CRC and non-hypermutated CRC. Invariably the WNT signaling genes were expressed at higher levels in the non-hypermutated CRC compared to the MLH1-silenced hypermutated CRC (Figure 3A). We also compared gene expression of the non-MLH1-silenced hypermutated tumors to the non-hypermutated CRC and these tumors generally displayed higher expression of WNT genes than their MLH1-silenced counterparts but less expression than the non-hypermutated CRC (Figure 3A). Heat maps were generated from the expression data for 41 WNT antagonists, 41 WNT agonists, and 17 WNT targets and graphically show differences in expression among the three categories of CRC (Figure 3B). Non-hypermutated tumors show the highest levels of WNT gene expression, while the MLH1-silenced hypermutated CRC show the lowest overall levels of WNT gene expression (Figure 3B). High expression of WNT antagonist genes in the non-hypermutated CRC may seem inconsistent with elevated WNT signaling, but deregulated WNT signaling resulting from APC loss triggers an intense but ineffective negative feedback loop accompanied by high expression of WNT antagonists [23,24]. Thus, based on expression data and APC mutation patterns, there appears to be a marked reduction in WNT signaling in the MLH1-silenced hypermutated CRC compared to the non-hypermutated CRC.
Global microRNA expression patterns were consistent with reduced WNT signaling in MLH1-silenced CRC. In the 224 tumor discovery set, we compared the three categories of CRC for differences in expression of each miRNA species and found several dozen that were highly up- or downregulated in the MLH1-silenced CRC compared to the non-hypermutated CRC (Figure 3C,D). For the 17 most significantly downregulated miRNAs in the MLH1-silenced CRC, we found that 9 of these miRNAs were reported in the literature to be upregulated by WNT and/or TGF-β signaling (Figure 3C). Thus, the non-hypermutated CRC showed evidence of increased WNT and TGF-β signaling based on miRNA signatures. The non-MLH1-silenced hypermutated tumors showed intermediate levels of downregulation for these 17 miRNAs. Thus, microRNA expression data support the mRNA expression analyses indicating that the MLH1-silenced CRC display decreased WNT signaling activity.
BRAF mutations are increased and KRAS mutations are reduced in MSI-H CRC . When we stratified the discovery set hypermutated CRC into MLH1-silenced and non-MLH1-silenced groups, only 1 of 22 (4%) MLH1-silenced CRC had a canonical oncogenic KRAS mutation (mutation of codons 12, 13, or 61), while 5 of 13 (38%) MLH1-non-silenced hypermutated CRC had oncogenic KRAS or NRAS mutations (Figure 4A,B) (P = 0.02). The 38% KRAS mutation incidence in the non-MLH1-silenced hypermutated CRC was not significantly different from the 48% RAS mutation frequency observed in the non-hypermutated tumors (Figure 4B). In contrast, 15 of 22 (68%) MLH1-silenced CRC had the oncogenic BRAF V600E mutation, whereas none of the 13 MLH1-non-silenced CRC and only 5 of 189 (2.6%) non-hypermutated CRC exhibited this mutation (Figure 4A,C). Analysis of BRAF AND RAS gene expression signatures (defined previously in cell line studies) showed that the MLH1-silenced CRC displayed evidence of enhanced BRAF signaling and reduced RAS signaling compared to their non-MLH1-silenced hypermutated and non-hypermutated counterparts (Figure 4F). The presence of oncogenic KRAS mutations correlated with higher levels of a “KRAS dependency” gene expression signature  (P<0.01, t-test), while the presence of BRAF V600E mutations correlated with a higher BRAF-activation signature (P<0.0001, t-test).
Analysis of the validation set for KRAS, NRAS, and BRAF mutations showed similar patterns to those observed in the discovery set. Only 10% of MLH1 methylated and hypermutated CRC exhibited canonical KRAS or NRAS mutations, while 25% of non-MLH1 methylated and hypermuted CRC and 47% of non-hypermutated CRC contained RAS mutations (Figure 4D; Table S1). Moreover, BRAF mutations were very frequent in the MLH1 methylated hypermutated CRC relative to the other CRC categories. Over 59% of MLH1 methylated and hypermutated CRC exhibited BRAF V600E mutations, while 13% of non-MLH1 methylated hypermutated CRC and 3% of non-hypermutated CRC contained BRAF mutations (Figure 4E; Table S1).
We examined mutation frequencies in tumor suppressors, including TP53 and TGF-β pathway genes, as these genes are frequently altered in CRC [17,25,26]. In the 224 tumor discovery set, TP53 mutations were significantly less frequent in the hypermutated CRC compared to the non-hypermutated CRC. Almost 60% of non-hypermutated CRC displayed TP53 mutations while only 18% of MLH1-silenced and 23% of non-MLH1-silenced hypermutated CRC exhibited TP53 mutations (Figure 5A,B). Analysis of TP53 status in the 250 tumor validation set showed a similar pattern as the discovery set, with 24%, 37%, and 67% p53 mutations in the MLH1-silenced, non-MLH1-silenced, and non-hypermutated CRC, respectively (Figure 5C). Analysis of a p53 expression signature in the discovery set confirmed that p53 target genes were more upregulated (P<0.0001, t-test) in the hypermutated CRC compared to the non-hypermutated CRC (Figure 5G), and that p53 signature correlated with both p53 somatic mutation and copy loss (P < 10−10 for each).
The CRC hypermutated tumors in the discovery and validation sets also exhibited high rates of mutations in the TGF-β pathway member genes such as SMAD2/3/4, ACVR1B, ACVR2A, and TGFBR2, while such mutations in the non-hypermutated CRC were much lower in frequency (Figure 5A,D,E). There were no significant differences in TGF-β pathway member mutations between the MLH1-silenced and non-MLH1-silenced hypermutated CRC. In many cases, the TGF-β pathway alterations were the result of frameshift mutations observed in long mononucleotide repeats that are present in the coding sequences of ACVR1B, ACVR2A, and TGFBR2. When the TGF-β RNA expression signatures were compared among the tumors, TGF-β signaling was decreased in a large fraction of the MLH1-silenced CRC relative to the majority of non-HM CRC (Figure 5F).
TGF-β pathway member genes such as ACVR1B, ACVR2A, and TGFBR2 show increased frameshift mutations within monucleotide repeats within their coding sequences [5,19]. We expanded this frameshift analysis to 20 genes that had mononucleotide repeats of six nucleotides or longer. In the 224 tumor discovery set, the hypermutated tumors displayed dramatically higher frameshift mutation rates in those 20 genes compared to the non-hypermutated tumors (Figure 6A,B). The MLH1-silenced CRC showed significantly higher overall mononucleotide frameshift rates compared to their MLH1-non-silenced hypermutated counterparts (Figure 6B). For the 20 examined genes, the average rates were 5.14 mononucleotide repeat frameshift mutations per MLH1-silenced CRC, 2.08 per non-MLH1-silenced CRC, and 0.05 per non-hypermutated CRC. The difference between the MLH1-silenced and non-silenced hypermutated CRC was significant (P = 0.0044). This pattern was repeated in the 250 tumor validation set (Figure 6D). When a set of genes that included the 20 genes with mononucleotide repeats and 6 genes important for DNA mismatch repair were analyzed in the discovery set tumors, missense and nonsense mutation rates were higher in the non-MLH1-silenced CRC (8.08 mutations/tumor) than the MLH1-silenced CRC (1.86 mutations/tumor) and the non-hypermutated CRC (0.21 mutations/tumor) (Figure 6C). The non-MLH1-silenced CRC had significantly more missense and nonsense mutations in the various mismatch repair genes and homopolymer containing genes than did the MLH1-silenced CRC (P = 0.028) (Figure 6C). However, in the validation tumor set this difference in the two hypermutated tumor types was less pronounced (Figure 6E).
Sporadic colorectal cancers (CRC) are usually subclassified according to whether they display microsatellite instability (MSI) or not . The 15% of CRC with MSI are frequently associated with distinctive molecular features, including defects in mismatch repair, high genome mutation rates, and epigenetic silencing of the MLH1 mismatch repair gene. Another useful way to subcategorize CRC is by mutation rate. The TCGA study showed that over 15% of all CRC had distinctly elevated gene mutation rates (>12 non-silent mutations/106 exonic nucleotides) compared to the remainder of the CRC. The hypermutated CRC had a high fraction of tumors with MSI-H (27/35) and MLH1 silencing (22/35). The TCGA CRC study also revealed that the hypermutated CRC had fewer APC, KRAS, and TP53 mutations, while these tumors also displayed higher numbers of BRAF and TGF-β pathway-related mutations. These mutation patterns were observed in both the discovery set of 224 CRC and the validation set of 250 CRC (Table 1, Table S1).
The gene mutation differences between hypermutated and non-hypermutated CRC suggested that these tumors might undergo distinct pathways to tumorigenesis. To examine this possibility further, we subdivided the 35 hypermutated CRC by MSI and MLH1 silencing status and found that MLH1 silencing best defined a subcategory in which APC and KRAS mutations were infrequent and BRAF V600E mutations were frequent. The MLH1-silenced hypermutated CRC were almost invariably MSI-H, while hypermutated CRC without MLH1 silencing were highly variable in MSI status. This latter subcategory of hypermutated CRC also displayed high APC and KRAS and low BRAF mutation rates similar to that of the non-hypermutated tumors. However, both categories of hypermutated CRC do exhibit high rates of TGF-β pathway mutations and low frequencies of p53 mutations relative to non-hypermutated CRC. These studies suggest that MLH1 silencing defines a subset of about 10% of sporadic CRC that are distinct from both non-hypermutated CRC and other hypermutated CRC that do not exhibit MLH1 silencing.
Truncating mutations in the APC tumor suppressor gene are the signature mutations of colorectal cancer, occurring in roughly 80% of all tumors. The low APC mutation rates in MLH1-silenced hypermutated CRC were not accompanied by increased mutation rates in other WNT pathway regulators, suggesting that these tumors do not initiate or progress by augmented WNT signaling. This was supported by analysis of CRC mRNA expression levels showing that expression of WNT regulatory genes and target genes were significantly reduced in the MLH1-silenced CRC relative to non-hypermutated CRC. In addition, the CRC microRNA analyses showed that upregulated miRNAs associated with augmented WNT signaling are overrepresented in the non-hypermutated CRC compared to the MLH1-silenced CRC. The combined mutational and expression data argue that enhanced WNT signaling does not play a major role in initiation or progression of MLH1-silenced hypermutated CRC, in contrast to the non-hypermutated CRC. The non-MLH1-silenced hypermutated CRC generally displayed intermediate levels of WNT signaling (Figure 3, Table 1).
If enhanced WNT signaling is not important in the MLH1-silenced CRC, what is the positive growth stimulus that drives these particular tumors? The frequent presence of oncogenic BRAF V600E mutations the MLH1-silenced CRC suggest that this event be a key driver of tumor cell proliferation. Since BRAF is a major downstream target of RAS signaling, it is not surprising that KRAS and NRAS oncogenic mutations are rare in the MLH1-silenced CRC while occurring much more frequently in the non-MLH1-silenced CRC that do not have BRAF mutations. We hypothesize that positive growth signaling in the MLH1-silenced CRC is driven in part by BRAF activation, whereas proliferation in the non-silenced CRC is driven largely by enhanced WNT and RAS signaling.
The MLH1-silenced CRC exhibited significantly higher rates of frameshifts in mononucleotide repeats, including key genes of the TGF-β signaling pathway. In particular, the MLH1-silenced CRC exhibited frequent frameshifts in mononucleotide repeats present in mismatch repair genes MSH3 and MSH6. Because MLH1, MSH3, and MSH6 play an important role in resolving small insertions/deletions in newly replicated DNA that arise from DNA polymerase slippage on long mononucleotide repeats , absence of their function in CRC would be expected to greatly increase the frequency of frameshifts in genes containing such repeats. Another interesting observation was that the non-MLH1-silenced hypermutated CRC displayed higher mutation rates than their MLH1-silenced counterparts. The mechanisms for this extreme mutability in the non-silenced cohort remain unclear, but this group of CRC had frequent mutations in the exonuclease domain of the POLE DNA repair polymerase .
Finally, does the reclassification of CRC by MLH1 silencing status instead of by MSI status provide any clinical benefit? We found that APC, BRAF, and KRAS mutation status was more definitive in MLH1-silenced CRC compared to MSI-H CRC. The 68% frequency of BRAF mutations in the MLH1-silenced CRC make this subgroup particularly attractive candidates for treatment with BRAF AND MEK inhibitors [28,29] and unattractive candidates for treatment with anti-EGFR molecules such as cetuximab . Thus, we propose that a simple test to identify MLH1 expression levels in CRC could potentially be as useful as the microsatellite instability test for clinical diagnoses and therapeutic decisions.
This study was supported by a grant to the Baylor College of Medicine Human Genome Sequencing Center U24 CA143843.
Conflict of Interest Statement: None of the authors have any conflicts of interest as defined by the guidelines for Journal of Pathology.
Statement of Author Contributions:L.A. Donehower initiated and carried out the majority of the analyses of the paper and did much of the writing of the paper.
C. Creighton did a significant amount of bioinformatic analyses, generated all of the figures for this paper, contributed to the writing, and performed virtually all of the statistics.
N. Schultz provided the initial data and analyses of the hypermutated CRC that formed the foundation for this paper.
E. Shinbrot generated the final CRC mutation data used for the CRC validation set.
K. Chang generated all of the homopolymer repeat mutation data used for Figure 6.
P. Gunaratne performed some of the analyses for the microRNA data presented in Figure 3.
D. Muzny supervised the generation of much of the sequencing data used in this paper. C. Sander supervised the initial data analyses of the TCGA CRC data that was the foundation for this paper.
S. Hamilton is a physician/scientist with expertise in clinical aspects of CRC who assisted in the writing and editing of this paper.
R. Gibbs is the director of the Baylor College of Medicine Human Genome Sequencing Center who provided the resources and overall guidance for this study.
D. Wheeler is the director of this research and supervised its progress at all steps and performed writing and editing of the paper.
List of Online Supporting Information:
Supplementary Table 1: Mutations in Key Genes of the 250 CRC Validation Set