Whole-exome Sequencing analyses of Inflammatory Bowel Disease-associated Colorectal Cancers

Whole-Exome Capture and Sequencing and Copy Number Analysis

DNA samples were purified from archived formalin-fixed paraffin-embedded (FFPE) blocks. Regions of these tumors that were enriched for neoplastic cells (>50%) were micro-dissected, avoiding foci of inflammation. Library construction was performed with a protocol developed specifically for DNA from archival samples that might contain damaged nucleotides⁵⁶. The resulting libraries were compatible for sequencing on Illumina instrumentation. Exonic regions were captured in solution using the Agilent SureSelect v.5 kit according to the manufacturer’s instructions (Agilent, Santa Clara, CA) and modifications as published elsewhere⁵⁷. Next-generation sequencing and bioinformatic analyses were performed at the Goldman Sequencing Center at Johns Hopkins or at Personal Genome Diagnostics (PGDx, Baltimore, MD). Known polymorphisms recorded in dbSNP were removed from the analysis. Potential somatic mutations were filtered and visually inspected as described previously⁵⁸. Selected somatic mutations were validated using SafeSeqS, a digital method developed for precisely quantifying mutations⁵⁹. Copy number alterations were identified by comparing normalized average per-base coverage for a particular gene in a tumor sample to the normalized average per-base coverage in the matched normal sample. Mutation prevalence in sporadic CRC was obtained from exome-sequencing reported by the Cancer Genome Atlas Research Network⁶⁰.

See the Supplementary Materials and Methods for detailed description of tissue specimens and sequencing, immunohistochemical staining, analysis of mutation patterns and pathway analysis.

Landscape of somatic mutations in IBD-Associated Colon Cancers

To generate a census of the genetic alterations that characterize IBD-associated colorectal tumors, we performed whole-exome sequencing on 32 formalin-fixed paraffin-embedded (FFPE) specimens from 31 patients with IBD. These included two patients with indeterminate colitis, 14 CD and 15 UC patients (Supplementary Table 1).

An average of 11.6 Gb were sequenced per sample. The average depth of quality coverage of the targeted region was 62-fold (range, 29 to 100-fold), with 76% of targeted bases yielding at least 10 distinct reads (range, 52% – 90%) in the tumors; and 64-fold (range, 17- to 123-fold) with at least 78% of targeted bases yielding at least 10 distinct reads in the matched non-tumor samples (Supplementary Table 2).

Somatic alterations were identified by comparing variants in paired tumor and non-tumor tissues. The number of somatic mutations showed great variability among tumors, with a range of 16 to 1,118 mutations per sample. There were two outlier samples, IBD128 and IBD003, harboring 1,118 and 916 somatic alterations each, with mutation rates of 22.20 and 18.19 mutations/Mb, respectively (Supplementary Figure 1). IBD128, from a CD patient, contained a missense as well as a splice site mutation of the POLE proofreading DNA polymerase and a splice site mutation of POLD1. IBD003, from a UC patient, contained single base pair deletions resulting in frameshift mutations of the DNA mismatch repair pathway genes MLH1 and MSH6. The mutation spectrum of these two hypermutated samples was dominated by small deletions, which constituted over 30% of the somatic sequence alterations (Figure 1A; Supplementary Table 3). This profile is consistent with that observed in hypermutated and Microsatellite Instability-High (MSI-H) colorectal⁶⁰ and gastric^{61, 62} tumors. After exclusion of these two outliers, the median mutation rate for the remaining colorectal tumors was 1.33 mutations/Mb (range, 0.32 to 5.10 mutations/Mb), with an average of 71 somatic mutations per sample. These mutation rates are similar to those observed in non-hypermutated and Microsatellite-Stable (MSS) samples from other cancers of the GI tract^{58, 60–64}, including those that have inflammatory etiologies, as well as sporadic tumors not readily associated with inflammation (Figure 1B; Supplementary Table 4). The pattern of base changes in non-hypermutated tumors showed a preponderance of C to T transitions (62%), a majority of which (48% of all mutations) occurred at 5′-CpG-3′ sites, a pattern commonly found in other cancers⁶⁵ and linked to spontaneous deamination of methylated cytosine⁶⁶ (Figure 2A; Supplementary Figure 2). An excess of A to C transversions at AA dinucleotides, particularly in a context of AAG trinucleotides, was also noted. This kind of enrichment has been described in esophageal^{67, 68} and gastric⁶² adenocarcinomas but has not been observed in sporadic CRC (Supplementary Figure 3). There were no apparent differences in the spectrum of substitutions between tumors arising in UC and CD patients (Figures 2B and 2C, respectively) with only a slightly elevated median number of mutations in tumors from UC patients vs. CD patients (77 vs. 53, respectively, excluding the two hypermutated samples).

Recurrently mutated genes in IBD-Associated Colon Cancers

Overall, we identified 4,158 somatic sequence alterations in 3,254 genes, comprising 3,297 nucleotide substitutions, of which 2,997 were missense mutations (Supplementary Table 5). When hypermutated tumors were excluded, we identified 2,124 somatic mutations, 1,953 substitutions and 171 indels in 1,806 genes. Among these, 28 genes showed mutation rates > 10 mutations/Mb and were recurrently mutated in three or more non-hypermutated samples (Table 1). One of the goals of this study was to determine the genomic landscape of colorectal tumors associated with IBD and compare it to that of sporadic CRC, the first tumor type in which genomic landscapes were defined^{63, 69}. To this end, we used the sporadic colorectal tumor landscape recently reported by the Cancer Genome Atlas Research Network as a fiducial⁶⁰. Although there were similarities between the most commonly mutated genes in both IBD-associated and sporadic CRC, there were also some striking differences (Figure 3; Table 2). APC, the most commonly mutated gene in sporadic colorectal tumors⁶⁰, was mutated in only four IBD-CRC tumors (13%), a prevalence much lower than expected (Benjamini-Hochberg adjusted Fisher’s exact test P = .0017, Supplementary Table 6). This finding is in agreement with prior single-gene based sequencing efforts^15–17. Three of the four IBD-CRC tumors with APC mutation were from CD patients, while the fourth was from a patient with indeterminate colitis. Mutations in KRAS were also less prevalent in IBD-CRC compared to sporadic CRC, again concordant with previous single-gene studies^{18–21, 60} (adjusted P =.019). No NRAS mutations were found in IBD-CRC tumors, although this gene is often altered in sporadic CRC⁶⁰. Instead, several genes were identified that were more frequently or uniquely mutated in IBD-CRC, including IL16, RADIL, RRBP1, SRRM4, DOCK3, TRRAP, CACNA1D, NRG1, EP300, and RIMS2 (Figure 3; Supplementary Table 6).

We identified four tumors with IL16 mutations. Interleukin-16 is a chemoattractant cytokine that can stimulate the secretion of tumor-associated inflammatory cytokines by monocytes⁷⁰, thus supporting a sustained inflammatory response. NRG1, encoding a ligand for the ERBB3 and ERBB4 tyrosine kinase receptors⁷¹ was also recurrently mutated, as were genes coding for RAC1 guanine nucleotide exchange factors (GEFs), including two members of the dedicator of cytokinesis (Dock) family, DOCK2 and DOCK3, as well as PREX2. Recent exome sequencing studies have revealed recurrent DOCK2 mutations in esophageal^{67, 72} and colon⁷³ cancers, and recurrent PREX2 mutations in melanoma⁷⁴. We also found novel somatic mutations in RADIL, which, together with its paralog RASIP1, controls Rho signaling associated with cell spreading⁷⁵. Additionally, we observed recurrent mutations in EP300 and TRRAP, which function in chromatin remodeling through histone acetylation and which facilitate p53-mediated transcription⁷⁶. TP53 was the most commonly mutated gene (19/30 cases, 63%), with mutation prevalence similar to that in sporadic CRC (Table 2; Figure 3). Other genes similarly mutated in IBD-CRC and sporadic CRC included SMAD2/4, PIK3CA, SOX9, FBXW7, and TCF7L2 (Table 2). All three PIK3CA mutations occurred at hotspots previously associated with CRC and other cancers. Likely oncogenic driver gene mutations in BRAF, CTNNB1, and CREBBP were found in one IBD-CRC case each (CHASM scores < 0.2).

The mutation spectrum of TP53 is of particular interest in light of its prevalence and documented association with environmental exposures⁷⁷. As expected, the majority of TP53 mutations (14/19) observed in IBD-CRC tumors were missense (Supplementary Table 7), and located predominantly in the DNA binding domain of the protein (aa.100–300, Figure 4). However, there were some notable differences in the spectrum of TP53 single substitutions in IBD-associated neoplasia and sporadic CRC. Specifically, no mutations were observed at the hotspot R273, and only one mutation was found at each of the hotspots R248, G245 and R175, in IBD-associated colorectal tumors. Instead, three of the 19 tumors harbored the less common R282W mutation, and two tumors each had mutations in R158, H179, or R342, which are rarely found in sporadic CRC (Supplementary Table 7). TP53 alterations were found in eight UC- and nine CD-associated tumors. In IBD-associated colorectal tumors, the predominant substitution was a C:G>T:A transition at CpG dinucleotides (n=10, 52.6%). This type of substitution in TP53 has previously been positively correlated with expression of the enzyme inducible NO synthase (iNOS)⁷⁸ in colonic tumors, where it has been hypothesized to be readout of inflammation-associated DNA damage. Indeed, colonic iNOS expression is known to be increased in IBD^{79, 80}.

Copy Number Alterations in IBD-Associated CRCs

We identified 15 regions with at least five-fold amplification many of which were recurrent or had clinically-relevant genes within them (Supplementary Table 8). These included 8q24.21, containing the MYC oncogene, and 17q12, involving ERBB2, as previously reported in sporadic CRC^{60, 81}. CDX2, a homeobox transcription factor involved in intestinal development⁸², was the most commonly amplified gene. Recurrently amplified regions containing IRS2, involved in activation of the PI3K pathway and recently identified as a candidate driver of CRC⁸³, or containing Suppressor of Cytokine Signaling 1 (SOCS1), were also identified. The only recurrently deleted region spanned the HLA-DQB1 locus in 6p21, a region previously associated with UC risk⁸⁴. Interestingly, this alteration was found in 6 patients with UC, 1 patient with CD and 1 patient with indeterminate colitis.

Pathways Altered in IBD-CRC

The set of mutated genes was enriched for gene ontologies associated with cell communication, cell-cell signaling, and cell adhesion (Supplementary Table 9). These findings are consistent with dysregulated cytokines and inflammatory mediators playing a role in IBD-associated CRC.

We grouped mutated genes into pathways and functions to better understand the overall consequences of mutations in IBD-CRC and how the pathways and functions altered in IBD-CRC compare to those altered in sporadic CRC. We found recurrent alterations in the p53, RTK, RAS, PI3K, WNT, and TGF-ß pathways, as has been previously described for sporadic CRC⁶⁰. In addition to these established CRC-associated pathways, genes involved in the Rho GTPase network, cell communication/adhesion, cytokine signaling, and chromatin remodeling were recurrently altered in IBD-CRC.

WNT signaling pathway alterations involving inactivation of APC or activation of CTNNB1 were found in only 17% (5/30) non-hypermutated IBD-CRC tumors, however, recurrent mutations were found on EP300 and SOX9, also thought to be involved in WNT signaling. Several other genes in this pathway, including FZD8, AXIN1, TCF7L2, NKD2, MAP3K7, NLK, ARID1A, and CREBBP harbored private mutations (Supplementary Figure 4). Even considering these, only 53% of the IBD-CRCs had mutations in the WNT signaling pathway, a much lower proportion than has been reported in sporadic CRC⁶⁰. To directly evaluate the status of WNT signaling in IBD-CRC, we performed beta-catenin immunohistochemical staining in 17 tumors with available material (Supplementary Figure 5; Supplementary Table 10). Not surprisingly, two tumors with APC inactivating and one with CTNNB1 activating mutations showed nuclear accumulation of beta-catenin. TCF7L2 and SOX9 are also considered part of the WNT canonical pathway and two tumors with inactivating mutations in these genes also showed beta-catenin nuclear accumulation. In addition, two tumors without obvious mutations in genes that are part of the canonical WNT pathway showed beta-catenin nuclear staining, although we can’t exclude large deletions of APC that would be missed by exome sequencing. Thus, despite the lower rate of inactivating APC or activating CTNNB1 mutations, the WNT pathway is still dysregulated in IBD-CRC through activating mutations in other components of the pathway. However, the rate of 41% nuclear accumulation of beta-catenin is still much lower than that in sporadic CRCs. The overall rate of mutation in TGF-ß signaling pathway was similar to that of sporadic CRC⁶⁰. We found SMAD2/4 mutations in 7 tumors (23%), more commonly than in a previous single gene study⁸⁵. Infrequent mutations of other genes in this pathway, including solitary mutations of TGFBR2, ACVR1, ACVR2B, and SMAD3, were identified in IBD-CRCs (Supplementary Figure 4). Recurrent alterations in genes involved in receptor-mediated signaling pathways were identified in 60% of IBD-CRCs (Supplementary Figure 4). These included mutually exclusive amplification of IRS2 and ERBB2, and mutually exclusive mutations of PIK3CA and KRAS, as well as NRG1 mutations.

Four recurrently mutated genes, DOCK2, DOCK3, PREX2, and RADIL, belonging to the Rho and Rac GTPase network, were mutated in 30% of IBD-CRCs (9/30). Perusing the list of genes with at least one non-silent somatic mutation, we found mutations in several other activating RhoGEFs, along with inactivating RhoGTPase-activating proteins (RhoGAPs)⁸⁶. Overall, the Rho and Rac pathway was altered in 50% of cases (Supplementary Figure 4). Moreover, a striking 50% of cases (15/30) presented mutations in chromatin remodeling genes, including recurrent mutations in TRRAP and EP300, and low-frequency mutations in CREBBP, ARID1A, and others (Supplementary Figure 4). Dysregulation of cytokine signaling is characteristic of IBD⁸⁷. We found recurrent mutually exclusive mutations in IL-16 and amplification of SOCS1 in 7 (24%) of cases (Supplementary Figure 4). Genetic alterations functionally associated with cell-cell communication or adhesion were found in 60% of IBD-CRCs, owing to recurrent alterations in CSMD2/3, DOCK2/3, CTNND2, DSCAML1, PTPRF and RADIL (Supplementary Figure 4).

Comparison of genes and pathways altered in UC and CD-associated CRC

Although both UC and CD patients are at elevated risk of developing CRC compared to the general population, the incidence is higher in patients with extensive colitis⁸⁸. This may be related to more sustained and pervasive inflammation of the colonic mucosa in UC patients. We looked at the overall involvement of each major pathway identified per tumor and compared how these pathways were differentially altered in UC vs. CD (Figure 5). We found that the Rho and Rac pathway was affected in 10 CRCs from confirmed UC patients but in only 3 CRCs from confirmed CD patients (Fisher’s exact test P = .025). With the caveat of our relatively small sample size, it appears that the Rho and Rac pathway may be preferentially activated in UC patients.