Abstract

Introduction

Patients with inflammatory bowel disease (IBD) are at a substantially increased risk for cancers of the colon/rectum or (for Crohn's Disease) the small bowel, and the risk increases the longer IBD is active¹. Colitis-associated cancers (CAC) are an especially feared complication of IBD, as they are frequently diagnosed at an advanced stage, with locally advanced or metastatic disease. Unlike more common sporadic colorectal cancers (CRC) that arise from polyps, it may be difficult to identify pre-cancerous dysplastic lesions or early CAC in patients with IBD. Their flat appearance makes the total extent of the pre-cancerous area difficult to determine and to remove endoscopically. CAC commonly present with multifocal tumors and signet ring cell histology, features associated with a worse prognosis^2-6.

The biologic changes that underlie the development of CAC are not well understood, but current data suggest that the genetic events leading to the initiation of sporadic CRC are different than those leading to CAC. Chronic inflammation is assumed to lead to genomic changes that increase the risk of CAC, and animal models suggest that both the initiation and the progression of colonic neoplasia can be exacerbated or expedited by an inflammatory insult^{7, 8}. TP53 mutations are an early event in CAC and are often detected in non-dysplastic areas or in dysplasia in patients with IBD^{9, 10}, while TP53 mutations are considered late events in sporadic CRC. In sporadic colorectal adenocarcinoma, WNT pathway activation, particularly through APC mutational inactivation, is often the initiating event^{11, 12}. It is not yet clear if subsequent genomic alterations in CAC are similar or different from those occurring in sporadic CRC. This is of importance as new therapies are frequently designed to affect specific alterations, and potentially targetable alterations in CAC have not been clearly defined.

After a genomic alterations analysis revealed an unexpected EML4-ALK fusion protein in a tumor specimen from a patient with CAC, we set out to more comprehensively characterize the mutational landscape of CAC. In this study, we use hybrid capture-based next-generation sequencing (NGS) to define recurrent or targetable genomic events in CAC from patients with both ulcerative colitis (UC) or Crohn's disease (CD) and compare the spectrum of genomic alterations (GA) in CAC with those occurring in sporadic CRC.

Methods

Results

Genomic profile of CAC

Overall, 6.2 GA per tumor were detected (range 1-19). We defined a recurrent alteration in CAC as one in which the gene was altered in at least 10% of cases (Figure 1). GA in TP53, the most commonly altered gene, occurred in 42 cases (89%). Most of the TP53 mutations occurred in the DNA-binding domain (Supplementary Figure 1) and largely consisted of point mutations. Of the latter, 29 cases harbored missense mutations. Supplementary Table 3 lists the TP53 alterations, with case order corresponding to the oncoprint in Figure 1. Sixteen cases had mutations reported to confer p53 oncogenic gain-of-function (GOF)²⁰ (C176F, V173L, R175H, R273H, R248Q in five samples, R248W in three samples, P278S, and R282W in three samples).

Open in a separate window

Figure 1

Oncoprint showing genes mutated in at least 10% of CAC cases. Each column denotes an individual tumor and each row represents a gene. Colors indicate type of genomic alteration as indicated in the legend below the oncoprint.

WNT pathway alterations in CAC were less frequent than in sporadic CRC; APC alterations were seen in 21% of cases, CTNNB1 alterations in two cases, and RNF43 mutations in three cases. KRAS mutations or amplification occurred in 40% of cases. SMAD4 alterations were seen in 17% of cases, and GNAS alterations in 13% of cases. IDH1 hotspot R132H/C/L mutations were detected in five cases of CAC, all associated with CD.

The most common copy number alteration was MYC amplification, in 26% of cases, affecting both CD- and UC-associated cases. Recurrent alterations in cell cycle genes, consisting primarily of copy number alterations of cyclin D (CCND1, CCND2, CCND3), cyclin dependent kinases 4, 6, and 8, and cyclin E, occurred in 16 cases (35%). Additionally, a segment of chromosome 20, containing SRC, BCL2L1, TOP1, and ZNF217, was recurrently amplified in four cases (data not shown).

Key GA associated with early onset IBD were not identified. Several cases had alterations in the Notch gene family. Sixteen cases (34%) had mutations involving NOTCH1, NOTCH2, NOTCH3, or NOTCH4, including a truncating mutation in one case. Of the alterations in the Notch gene family, seven cases had mutations previously described in at least one other tumor specimen in the Catalogue of Somatic Mutations in Cancer (COSMIC) (data not shown).

GA occurring in CAC differ significantly from GA in sporadic CRC

The spectrum of GA we identified suggests a different biology for CAC than that of sporadic CRC. We compared the incidence of recurrent GA found in CAC with their frequency in The Cancer Genome Atlas CRC (TCGA-CRC)¹⁷ and in a large Foundation Medicine dataset of 3117 patients with CRC¹⁹ (Figure 2). As our CAC cohort consisted predominantly of operable primary tumors (91%) (Table 1), we chose to use the TCGA-CRC data as one of the comparison populations; TCGA-CRC consists of surgical specimens from newly diagnosed colon and rectal tumors with predominantly resectable disease. We also compared the spectrum of CAC GA with that of the Foundation Medicine database (FM-CRC); we recognize that sporadic CRC tumor samples studied using FoundationOne assay are largely from patients with more advanced disease, frequently metastatic.

Open in a separate window

Figure 2

Comparative analysis of the frequency of alterations in recurrently altered genes in Colitis-Associated Cancers (CAC) and in sporadic colorectal cancer. Top: Table showing the frequency of alterations in the indicated genes in the CAC cases overall (UC-associated plus CD-associated) versus the frequency of alterations in the same genes as found in the TCGA-CRC, and in the Foundation Medicine database (FM-CRC), and associated p-values based on Fisher's exact test. Bottom: Bar graph showing the relative frequency of genomic alterations in the indicated genes in CAC associated with UC (UC-CAC), CAC associated with CD (CD-CAC), TCGA-CRC, and FM-CRC. Differences in the frequency of alterations that were statistically significant, based on Fisher's exact test, are indicated with a star.

APC alterations were much less common in our CAC cases than in the TCGA-CRC or FM-CRC (p<0.001). Genomic alterations in TP53 were nearly universal in CAC compared to about half of the cases in TCGA-CRC (p<0.001) and about three-quarters of Foundation Medicine cases of advanced CRC (p=0.02). MYC amplification occurred in 26% of CAC compared to 4% of TCGA-CRC and 9% of advanced FM-CRC (p<0.01).

IDH1 mutations seen in 11% of CAC (and only in CD CAC) are extremely rare in sporadic CRC, with 1% mutation frequency in TCGA-CRC and only 0.6% in FM-CRC¹⁹. The IDH1 mutations identified in CAC were at the R132 hotspot, producing a mutant protein that could potentially be targeted by selective IDH1 inhibitors that are currently in clinical trials.

An understanding of genomic differences between CAC and sporadic CRC may help in the clinical assessment of bowel cancers developing in patients with IBD. In an additional patient with a long prior history of CD that had been quiescent for 10 years, who underwent ileocolectomy for a rectal tumor, review of his pathologic specimen showed no evidence of active colitis (Figure 3). Genomic analysis of the resected rectal tumor (not included in our 47 patients with clear CAC) showed no TP53 mutations, but identified APC and KRAS mutations, a genomic profile more consistent with sporadic CRC. This patient's case suggests that genomic analysis of CAC may help distinguish bowel cancer developing from underlying colitis-inflammation versus sporadic tumorigenesis.

Open in a separate window

Figure 3

Photomicrograph of rectal mucosa adjacent to tumor that shows no active colitis in a patient with history of quiescent CD and a tumor genomic alterations analysis mutation profile consistent with sporadic CRC. The horizontal sizing bar indicates 500um.

Differences in the GA signature for CD-associated and UC-associated CAC

The GA identified in the CAC cases suggest that different genes are recurrently altered in CD-associated CAC when compared with UC-associated CAC (Figures 1 and ​and2).2). APC alterations were significantly more common in the CD-associated CAC (CD 39% versus UC 10%, p=0.02). Activating alterations in KRAS appeared more common in the UC-associated CAC, but the difference did not reach statistical significance (CD 33% versus UC 45%, p=0.5). TP53 alterations predicted to cause p53 loss-of-function (deletions, frameshifts, and stop codons, expected to truncate p53) were more common in UC-associated CAC compared to CD-associated CAC, but again did not reach statistical significance (UC 32% versus CD 12%, p=0.16) (Supplementary Table 3).

Most striking, all the IDH1 mutations identified were in CD-associated CAC. The IDH1 mutations that we identified in CAC occurred exclusively in the CD cases and were seen in 5/18 CD cases and zero in 29 UC cases (CD 28% versus UC 0%, p<0.01). Notably, in one case where a patient carried a long-standing diagnosis of UC, but was found at surgery to have histopathologic features more consistent with CD, including relative sparing of the distal segment of colon and the presence of well-formed submucosal granulomas and transmural lymphoid aggregates, an IDH1 mutation was identified (Figure 4). This suggests that the genomic profile of CAC may help clarify cases developing in the setting of UC from those developing in the setting of CD, as separating UC from CD can be difficult in a subset of IBD.

Open in a separate window

Figure 4

Photomicrographs of colon mucosa showing (A) transmural chronic colitis and (B) granuloma formation (arrow), histologic changes characteristic of Crohn's colitis, in a patient with clinical history of UC, but found to have tumor IDH1 R132 mutation. The horizontal sizing bars indicate 500um.

Potentially actionable alterations in CAC

Excluding TP53 alterations, potentially actionable GA were identified in 17/47 (36%) of CAC (Figure 5). Activating alterations were commonly seen in receptor tyrosine kinase (RTK) signaling, including the fibroblast growth factor receptor (FGFR) pathway that was altered in 8 patients, ERBB2 amplification in 2 patients, ERBB2 S310F, EML4-ALK fusion, and PDGFRA T134M. One patient had an FGFR2-TACC2 fusion and FGFR2 amplification. One patient had a BRAF V600E mutation and TSC2 truncation, two alterations in mitogenic signaling pathways that are potentially targetable with RAF inhibitors and rapalogs, respectively. As noted above, potentially targetable IDH1 hotspot R132H/C/L mutations were detected in five cases of CAC. IDH1/2 inhibitors are now in clinical trials, as are agents targeting the fibroblast growth factor (FGF) pathway. Drugs targeting HER2, EML4-ALK, and V600E BRAF are approved for other indications. Figure 5 also shows the overlap in these GA with RAS activation and indicates that the GA in receptor tyrosine kinase signaling were largely mutually exclusive of KRAS activation. KRAS and IDH1 mutations were also not co-incident.

Open in a separate window

Figure 5

Oncoprint showing alterations in genes that are potentially actionable, defined as genes whose altered product can be targeted either by agents that are already FDA-approved for other indications or are currently in clinical trials. Each column denotes an individual tumor and each row represents a gene. Colors and symbols indicate type of genomic alteration as indication in the legend below the oncoprint.

Activating alterations were also seen in insulin growth factor receptor signaling, including IRS2 amplification in 3 cases and IGF1R amplification in 1 case (data not shown). Activating mutations in the gene encoding the catalytic subunit of phosphatidyl inositol 3-kinase (PI3K), PIK3CA, occurred in three cases. These GA may be targeted, alone or in combination regimens as antibodies against the insulin growth factor 1 receptor and small molecule inhibitors of PI3K are in clinical development.

Figure 6 summarizes pathways altered in UC- and CD-associated CAC, integrating data on gene mutations and copy number alterations. Alterations in the p53 pathway were seen in 100% of CD-associated CAC and 83% of UC-associated CAC. Activation of RTK/RAS signaling was common, occurring in 72% of CD-associated CAC and 57% of UC-associated cases. FGFR2 amplification and translocations and EGFR activation were more common in CD-associated cases. Alterations in PI3K/mTOR signaling occurred in 22% of CD-associated CAC and 7% of UC-associated cases. Cell cycle alterations were seen in 17% of CD-associated CAC and 24% of UC-associated cases. Alterations in WNT signaling, TGFβ, and MYC were seen in about half of the CAC, with inactivation of APC more common in CD-associated CAC and MYC amplification affecting both CD and UC-associated cases, as noted above.

Open in a separate window

Figure 6

Pathway alterations diagram integrating gene mutations and copy number alterations to identify pathways altered in UC- and CD-associated CAC. Alteration frequencies are expressed as a percentage of CD-associated cases (left side of box) and of UC-associated cases (right side of box). Red denotes activated genes and blue denotes inactivated genes, with the brightness of these colors corresponding to the percentage of cases altered. Amplifications and missense mutations in recurrent positions in oncogenes were considered activating, and deletions, truncating mutations as well as recurrent missense mutations in tumor suppressor genes were considered inactivating.

Discussion

In this study, we find support for the hypothesis that there is a substantive genomic difference between CAC and sporadic CRC: genes commonly altered in CRC, such as APC, were much less commonly altered in CAC. Genes rarely altered in sporadic CRC, such as IDH1, were recurrently altered in CAC. The genomic differences we identified between CAC and sporadic CRC suggest that these tumors develop differently. As summarized by Ullman and Itzkowitz, the development of carcinoma in CAC appears to progress through a sequence consisting of indefinite dysplasia–low-grade dysplasia–high-grade dysplasia–carcinoma, in contrast to the classical progression of a discrete focus of neoplasia from a polypoid adenoma to an invasive carcinoma in sporadic CRC²¹.

Novel findings in our NGS of a large series of CAC include recurrent IDH1 R132 mutations, which we find only in the CD-associated cases, alterations of the FGFR pathway, and recurrent amplification of MYC. We note that the GA identified included those that may be targeted by agents already approved for other indications or in clinical trials. These include IDH1, ERBB2, EML4-ALK, and FGF pathway alterations. We also find that the spectrum of GA in UC CAC may differ from that of CD CAC: for example, besides IDH1 mutations, APC alterations are significantly more common in CD-associated CAC than UC-associated cases. A limitation of our analysis is the use of targeted gene sequencing with NGS panels analyzing about 300 genes.

Recently, Robles et al²² reported the results of an analysis, using whole exome sequencing, of 31 cases of CAC (in one patient, two distinct primary tumors were studied). Fifteen patients had UC, 14 CD, and 2 indeterminate colitis. There were both similarities and differences between our study and the findings in Robles et al (shown also in Supplementary Table 6). As had Robles et al, we also find that the spectrum of GA in CAC is different from that found in sporadic CRC (in both studies, the TCGA-CRC dataset was used as a comparator for sporadic CRC). Both Robles et al and we note that there are also differences in recurrent GA between CAC arising in the setting of UC versus CAC arising in CD. There are also different findings in the two studies. Although in both studies TP53 mutations were the most frequent alteration, the incidence of TP53 alterations in Robles et al was similar to that of TCGA-CRC (63% versus 60%), whereas we found a much higher frequency of TP53 mutations (89%). In agreement with Robles et al, we also find a lower frequency of TP53 mutations in the hotspot residues 175 (1 mutation), 273 (1 mutation), and 245 (0 mutations). However, at variance with Robles et al, mutations in another hotspot, Arg248 (to either Trp or Gln), were the most abundant in our cohort, comprising 26% (8/31) of the total missense mutations found in these patients. The reasons for this difference between the two studies remain to be elucidated. We noted MYC amplification in 26% of patients higher than seen by Robles et al. (3%) or in TCGA-CRC (4%). Last, as discussed in more detail below, we found IDH1 R132 mutations in 5/18 CD CAC patients. The differences in the incidence of GA seen between the two studies may be due to the relatively small number of patients studied; the higher coverage (500×) in the more focused NGS platform may have allowed us to identify less frequent variants.

As noted above, the TP53 gene in our study was altered in 42 of 47 CAC cases, a much higher percentage than in sporadic CRC. Previous studies have identified TP53 abnormalities as early events in the dysplasia-carcinoma sequence in CAC⁹. Our study supports this finding. In addition we note many alterations with potential GOF. The TP53 alterations identified consisted largely of coding mutations, rather than truncating mutations, and included sixteen patients with neomorphic mutations that have been described as causing GOF activities, including enhancement of invasive properties, attenuation of apoptosis, and increased genomic instability²⁰. It is possible that due to limited experimental data, some of the other missense mutations might also possess GOF capacity. This remains to be formally demonstrated. A recent study by Cooks et al suggests that the early presence of mutant p53 in the inflamed colon of IBD patients may actually be a driver of the subsequent progression to carcinoma by invigorating inflammation in the immediate microenvironment of the cells harboring mutant p53⁷. In this study, Cooks et al found that mice with mutant TP53 exposed to dextran sodium sulfate, which elicits a condition resembling human IBD, not only developed more frequent inflammation-associated colon cancer, but also developed carcinoma much earlier than mice with knockout of one TP53 allele, suggesting that mutant p53 may not only make the mice more susceptible to chronic inflammation, but also accelerate the development of carcinoma on an inflammatory background. Both of the TP53 mutations (TP53 R273H and TP53 R175H) implicated in their study as actively contributing to the development of CAC were found in the tumors sequenced in our patients with CAC.

We identified MYC amplification in 26% of CAC cases, a substantially higher frequency than seen in sporadic CRC. The higher incidence of MYC amplification in our CAC cases may be due to the relative low frequency of activation of the canonical WNT pathway in these cases. Mutational activation of the canonical WNT pathway is a very early event in sporadic colorectal carcinogenesis¹¹. This is expected to lead to early transcriptional upregulation of Myc expression²³, greatly obviating the selective pressure for MYC gene amplification. Indeed, MYC gene amplification is relatively infrequent in sporadic colorectal cancer¹⁷. In contrast, in CAC, WNT pathway activation is less common; in those cases where it does occur, this is believed to be a late event²¹. This is likely to provide a selective drive for MYC gene amplification, providing a plausible explanation for the markedly higher frequency of such events in CAC. We note that in our study only 2/11 MYC amplified cases were coincident with APC mutations.

In the pathway analysis we performed, we found that several of the same pathways were altered in sporadic CRC and in CAC, but there were some significant differences. While p53 signaling was altered in 60% of CRC cases, p53 alteration appeared to be a near universal feature of the CAC. RTK/RAS signaling was activated in the majority of both sporadic CRC and CAC, with a similar frequency of RAS mutations, but a slightly higher frequency in CAC of activation of receptor tyrosine kinases, such as EGFR, ERBB2, and FGFR1/2; PI3K activation occurred in a large subset of both sporadic CRC and CAC. WNT/MYC signaling, which is altered in nearly all sporadic CRC, was altered in two-thirds of CD-associated CAC and one-third of UC-associated cases. We also note some differences in the pathways altered between CD and UC-associated CAC: In addition to the more frequent alterations in WNT/MYC signaling in CD-CAC, RTK and PI3K pathway alterations appeared more common in CD-CAC.

We found that targetable genomic alterations were frequent in CAC (36% of cases); agents targeting these alterations are already FDA-approved for other indications or are currently in clinical trials. Many of the potentially targetable alterations identified would not have been predicted based on previous genomic studies in sporadic CRC. For example, IDH1 mutations, which are extremely rare in sporadic CRC, were significantly enriched in the CAC. This finding is consistent with a recent report from Hartman et al where they found IDH1 mutations in 3/23 (13%) of CAC cases analyzed, 2 with CD and 1 with UC²⁴. They used PCR-based sequencing to genotype IDH1 at codon 132 and IDH2 at codon 172 in CAC from 12 patients with CD and 11 with UC. In their more limited study, they also evaluated KRAS and BRAF mutation status in a subgroup of patients (a broader NGS platform was not used). Interestingly, recurrent IDH1 mutations were first identified in glioblastomas, where they occur as an early event in the development of secondary glioblastomas that often harbor TP53 mutations²⁵.

The presence of IDH1 activating mutations only in the CD cases analyzed in our series suggests a distinct pathogenesis for CAC in CD compared to UC and also has potential therapeutic implications. KRAS/BRAF mutations were more common in the UC cases, raising the possibility that activation of this mitogenic pathway may play a role in the development of CAC in the setting of UC. As RAF and MEK inhibitors have recently been developed and are now entering the clinic, these findings if validated in larger studies may offer additional therapeutic targets.

In summary, our data support the hypothesis that CAC is a biologically different malignancy from sporadic CRC. A better understanding of the genomic events underlying the development of CAC could have implications for early detection of CAC in patients with IBD and for new therapeutic options for more advanced CAC. Different TP53 mutations may confer a different risk profile for the development of frank carcinoma in dysplastic lesions. The identification of “second genomic hits” in dysplastic lesions, the majority of which already have TP53 mutation, may also serve as biomarkers guiding management. For example, the identification of either an activating TP53 mutation or a second genomic event in a dysplastic lesion may guide the extent of regional intervention, e.g. surgical resection instead of endoscopic mucosal resection. Further, our data suggest that there may be a spectrum of differences in genomic alterations between UC and CD cases and raises the possibility that IDH1 mutation may be both a marker for CD-associated CAC and a promising therapeutic target.

Supplementary Material

Acknowledgments

Footnotes

References