MOLECULAR FEATURES AND MOUSE MODELS OF COLORECTAL CANCER

Abstract

Colorectal cancers (CRCs) harbor accumulated defects in key signaling pathways that regulate cell phenotypes, including proliferation, survival, metabolism, and differentiation. To study the functional contributions of the accumulated molecular defects in CRC, we have developed approaches to inactivate selected tumor suppressor and/or activate oncogenes in mouse colon epithelium. Conditional inactivation of the CDX2 tumor suppressor protein in conjunction with oncogenic activation of the BRAF protein promotes development of serrated glandular benign and malignant tumors in the mouse colon. The mouse tumors share significant morphological and molecular relationships with the 8% to 10% of human CRCs that manifest serrated morphology at diagnosis. The gene and protein expression patterns in the mouse tumors have informed understanding of the relationships between benign and malignant human serrated colon tumors. Our findings are consistent with prior work suggesting that perhaps upwards of one-third of human CRCs may arise from a precursor lesion with serrated morphology rather than a conventional adenoma.

My efforts in the cancer research field have been greatly influenced by the work of Peter Nowell. Dr. Nowell had a remarkable and highly impactful career in cancer research. Among his many important contributions included the discovery, along with David Hungerford, of the Philadelphia chromosome as a recurrent chromosomal alteration present in the neoplastic cells of patients with chronic myelogenous leukemia (1). Other pioneering contributions from Dr. Nowell were concepts summarized in his 1976 Science manuscript on the clonal evolution of tumor cell populations (2). In that manuscript, he presented a unifying view that acquired genetic lability and the biological selection of variant subpopulations are key driving factors in cancer progression. Although published nearly 4 decades ago, the paper still provides many highly relevant points for the cancer genetics field, including the notion that clonal selection acts on variant cell populations with particular constellations of molecular alterations, resulting in the outgrowth of neoplastic cells with the most robust proliferative and survival context in a given patient and in a given tissue and biological context. The manuscript also highlighted the critical role of intratumoral heterogeneity in advanced and metastatic cancers, presciently predicting a fundamental role for intratumoral heterogeneity in therapy resistance, including as we now know not only for classical chemotherapeutic approaches but also for so-called “molecularly targeted” therapies, such as the inhibition of the epidermal growth factor receptor (3).

The work of Dr. Nowell was an inspiration to me as an MD-PhD student training in Dr. Bert Vogelstein’s laboratory at Johns Hopkins Medical School during 1983–1990, where among various projects there, I contributed to work that defined somatic chromosomal and molecular defects associated with different phases of colorectal tumor development (4,5). In short, we found that certain defects, such as chromosome 5q and adenomatous polyposis coli (APC) alterations were roughly equally prevalent in small colorectal adenomas as in advanced colorectal cancers (CRCs), whereas other alterations were more often seen in advanced lesions compared to their prevalence in early lesions (4,5). For instance, activating or oncogenic point mutations in the KRAS gene were seen in roughly 40% to 50% of colorectal adenomas greater than 1 cm in size and roughly similar frequencies of CRCs (4,5). Defects in the chromosome 17p region where the TP53 gene resides and in the TP53 gene itself were frequent in most CRCs, but were rarely found in adenomas (5). Taken together, the findings indicated that selected, recurrent somatic molecular defects affecting key signaling and regulatory pathways are associated with clonal expansions at preferential times in the natural history of colon tumors. The studies also offered strong molecular support for the clinical and pathological presumptions at that time that many CRCs arose from pre-existing adenomatous lesions.

The apparent preferences for the clonal expansion of tumor cell populations with certain molecular defects in selected phases of the natural history of CRC development was not due to the fact that the somatic defects arose at particular times in tumor development (5). Rather, as Nowell had suggested in his 1976 Science manuscript, the findings indicated that there was potent biological selection for certain variant cell populations at preferred phases in the natural history of a given colorectal tumor (5). While the order of somatic genetic events was not invariant when large numbers of colorectal tumors were studied, the findings indicated that there might be preferred molecular pathways in the development of a given type of cancer — i.e., CRC. Finally, the findings suggested that the accumulation of somatic defects in multiple independent pathways — perhaps four or more different pathways — were likely to be important in the genesis of advanced and metastatic CRCs (5).

The work that Dr. Vogelstein and I summarized in our 1990 Cell review article emphasized the view that many CRCs arose from adenomatous precursor lesions (5), but the manuscript did not address the possibility that there might be other potential types of precursor lesions for CRCs besides adenomatous polyps. Dr. Jeremy Jass was among those who presented important concepts and data regarding the potential role of a collection of serrated neoplastic lesions in the colon and rectum as potentially significant precursor lesions for CRCs. In a 2007 review article, Jass described the use of two molecular features — the so-called microsatellite instability high-frequency (MSI-H) phenotype versus the microsatellite instability low-frequency (MSI-L) and microsatellite stable (MSS) phenotypes as well as the so-called CpG island methylator phenotype-high (CIMP-H) versus CIMP-low (CIMP-L) and CIMP-negative states — to define subgroups of CRC (6). Based on his analysis of how these two molecular features — MSI and CIMP status — assorted with selected clinical and pathological features as well as somatic mutations in the APC, KRAS, TP53, and BRAF genes, Jass proposed there might be five distinct groups of CRC (6). Of note, Jass proposed that certain benign colorectal lesions where the epithelial glands had a serrated (sawtooth) morphology, especially sessile serrated adeonomas, were potentially key precursors in the natural history of perhaps 30% of all CRCs, including many of the CRCs that harbored BRAF mutations and lacked mutations in KRAS and where mutations in the APC and TP53 tumor suppressor genes were also uncommon.

One potential challenge to the proposal of Jass that perhaps 30% of CRCs may arise from benign lesions with serrated epithelial glandular morphological features is that only approximately 8% to 10% of CRCs manifest clear-cut serrated morphological features at diagnosis. Assuming two things — first, that Jass was on point in proposing that approximately 30% of CRCs arise from serrated precursor lesions, and second, that the 8% to 10% of CRCs that manifest serrated features at diagnosis arose from benign serrated precursor lesions — one must then infer that a very significant fraction of the CRCs that arose from a serrated precursor lesion must have lost recognizable serrated morphological features at some point before detection and removal. The absence of molecular markers to define those CRCs that arose from a serrated precursor lesion and that then lost serrated morphology represents one of the challenges for the serrated precursor-CRC progression model.

To generate some new insights into the pathogenesis of serrated morphology CRCs and the possible relationships of serrated benign lesions to serrated CRCs, we first characterized selected somatic molecular alterations and protein expression in a group of 36 serrated human CRCs. We found that approximately 45% to 50% of the serrated human CRCs had concurrent oncogenic BRAF^V600E missense mutations and loss or markedly reduced expression of the CDX2 homeobox protein (7), a presumptive tumor suppressor gene in CRCs. The CDX2-null and BRAF^V600E-mutant human serrated morphology CRCs also frequently expressed gastric epithelial markers, such as mucin 5AC and annexin A10 (ANXA10) (7).

To address functionally the potential cooperative roles of somatic loss of CDX2 expression and BRAF^V600E mutations as driving factors in serrated tumorigenesis individually and cooperatively, we developed a new mouse model of serrated colon tumor formation. We used a tamoxifen-regulated Cre recombinase transgenic mouse line where Cre recombinase was expressed in epithelium of very distal ileum, cecum, and proximal colon (i.e., CDX2-CreER^T2 mice) (8). The Cre recombinase was used to conditionally modify LoxP-containing alleles in intestinal epithelium to inactivate mouse Cdx2 alleles, to activate an oncogenic BRAF^V600E allele, or to concurrently inactivate Cdx2 alleles and activate the BRAF^V600E allele (7). Inactivation of Cdx2 on its own or activation of BRAF^V600E on its own had only modest effects on cell phenotypes and/or proliferation in the targeted epithelium and no significant adverse effects on mouse survival (7). In contrast, targeting of both Cdx2 and BRAF^V600E led to multiple epithelial tumors in the distal small intestine, cecum, and proximal colon of each mouse and the mice had a dramatically shortened lifespan (7). A subset of the tumors arising in each mouse were carcinomas. The Cdx2-null BRAF^V600E-mutant epithelium in the tumors showed serrated morphological glandular features similar to that seen in serrated morphology CRCs (7).

Gene expression studies in the CDX2-null BRAF^V600E-mutant epithelium showed dramatic cooperative interactions of CDX2 loss and BRAF^V600E in activating the expression of hundreds of different genes, with the ANXA10 gene among the most strongly activated (7). The gene expression patterns of the mouse tumors resemble those of serrated CRCs. Among the other genes that we found to be strongly activated in the mouse Cdx2-null BRAF^V600E–mutant tumors was that for the PDX1 homeobox protein (7). PDX1 expression is normally restricted to proximal gastrointestinal tract epithelium in adult tissues and is not detected in normal adult colon. However, ectopic PDX1 expression was found in more than 90% of human serrated CRCs. Ectopic PDX1 expression was also found in all sessile serrated adenomas studied, as well as all other benign serrated lesions studied, including traditional serrated adenomas and microvessicular and goblet cell hyperplastic polys (7). The ectopic PDX1 expression in all benign serrated morphology lesions was accompanied by reduction or less of expression of the CDX2 protein. When PDX1 expression was studied in a collection of nearly 400 unselected CRCs, PDX1 expression was seen in roughly 33% of CRCs (7), which intriguingly is not far from the roughly 30% fraction of CRCs that Jass had suggested might arise from serrated precursor lesions (6).

In summary, our work has yielded new insights into the pathogenesis of the 8% to 10% of CRCs that manifest a serrated morphology at diagnosis. Concurrent loss of CDX2 expression and BRAF^V600E mutations are seen in approximately 45% to 50% of serrated morphology CRCs. We developed a new mouse model for serrated CRCs, where conditional concurrent CDX2 inactivation and BRAF^V600E oncogene activation in mouse colon epithelium leads to much shortened mouse survival due to tumor formation, including the development of carcinomas. The mouse and human serrated CRCs that are CDX2-null and BRAF^V600E-mutant express gastric epithelial markers, suggesting aberrant differentiation and reprogramming of the colon epithelial cells to a cell fate characteristic of more proximal regions in the gastrointestinal tract underlies the pathogenesis of serrated colon tumors. With regard to the potential relationships of benign serrated colorectal precursor lesions to serrated morphology CRCs, we found that all sessile serrated adenomas that we studied had lost CDX2 expression and ectopically expressed PDX1. Other benign serrated colorectal lesions had markedly reduced or absent CDX2 expression and ectopic PDX1 expression, highlighting the possibility that some serrated CRCs might arise from benign lesions currently thought to have very limited or no potential to progress to malignancy, such as hyperplastic polyps carrying BRAF mutations. Finally, the expression of PDX1 in one-third of CRCs at diagnosis is consistent with but does not establish that a serrated precursor-CRC pathway may be a critical factor in the pathogenesis of a major subset of CRCs. Further work is needed on that topic, as better understanding of the origins and biology of benign and malignant serrated colorectal tumors may well have significant relevance for cancer prevention, early diagnosis, and treatment, much has been the case for advances in understanding of the adenoma-CRC progression pathway.

DISCUSSION

Schuster, New York: Beautiful work. Can you rescue the BRAF mice with clinical BRAF inhibitors?

Fearon, Ann Arbor: In the case of the BRAF experiment, we haven’t yet done the specific experiment you suggest. But the mouse models are informative in some ways about what we might expect. We have a mouse model of tumorigenesis with APC, KRAS, and TP53 mutations. Many of the tumors that arise become invasive and some are metastatic to the liver and/or the lung. We also have another mouse model with APC, KRAS, and PIK3CA mutations. So, we thought that if we use KRAS pathway inhibitors, we could test if the single inhibitors affect tumor growth and how combinations of inhibitors affect tumor growth. The single KRAS pathway inhibitor has modest effects in the mouse models, perhaps similar to the situation in human patients. With inhibitor combinations, we see somewhat greater effects, but similar to the situation in humans, the inhibitor combinations make the mice quite ill, including significant weight loss. Nonetheless, we are interested in doing this in a general way with some of the newer models that we’re making. So, I don’t have an answer on a BRAF inhibitor yet but we think it would likely show at least some evidence of tumor reductive effects.

Marrazzo, Birmingham: Thanks for your talk. Just referring to your final slides about effects that reflect not just the biology of tumor cells — Cindy Sears and others at Hopkins have done some great work looking at biofilms and the role of the microbiome in colonic tumors, even depending on the specific sideness in the colon. I wondered, can you work any of that into these models or is it too complex? And what does that look like?

Fearon, Ann Arbor: No. It’s a great question. I’m convinced, based on some of the data from humans as well the mouse data not only from Dr. Sears but from others — where if you treat with broad spectrum antibiotics you have a dramatic effect on tumorigenesis in the mouse models. Obviously, this is not a strategy for prevention in humans, but I think there are important relationships between the microbial communities in the gut in the general sense as well as those specifically that may be adherent to the tumor, as you note. I think that it’s actually likely quite complicated in that tumors themselves are genetically and functionally heterogeneous and I suspect the microbial communities around different portions of the tumor themselves are heterogeneous. There is probably coselection for microbial communities and tumor cells, based on the natural products they make. I think studies of microbial communities and cancer offer a great opportunity not only for prevention strategies to understand how we reduce tumor incidence, burden and size, but as well for how the findings may relate to treatment approaches. I think it is perhaps one of the next hot areas in general in medicine and certainly in colorectal cancer.

Licht, Gainesville: Could you elaborate a bit on why there are differences between the organoid and in vivo models? Could you clarify whether those organoids derive from the genetically engineered mouse model, and are the ex vivo conditions hyperoxia? What kinds of things do you think could make the organoid a more perfect model?

Fearon, Ann Arbor: Yes — those points are excellent. We can model the organoids in different ways, but the ones that I described were taken from epithelium of mice that have been treated with tamoxifen. The colon epithelial cells in question are genetically targeted and just a few days after targeting we can derive organoids or we can derive the organoids from established tumors. We can also do it entirely in vitro where we would treat one with 4-hydroxytamoxifen, which is the active metabolite of tamoxifen. We don’t have a keen sense of when the drift in phenotype from the in vivo setting begins, but it’s a pretty potent effect on patterns of gene expression. A colleague from Ramesh Shivdasani’s Lab at the Dana Farber Cancer Institute who just visited to give a job talk and we were chatting about this and how epigenetic patterning may contribute to the drift. I think there are many things different about organoids versus the in vivo setting. You’re adding these growth factors or conversely taking them away, based on certain dependencies that the cancer cells lose when they become transformed. You’re growing them in Matrigel, which is obviously not exactly the same microenvironment and the stromal cells are sort of selected against by organoid conditions and important immune cell are also lost in organoid culture. In addition, you’re obviously losing the benefit of all the systemic factors that circulate and also the paracrine factors. So, I think everything is sort of a model if you’re not studying human patients. So, I’m still a fan of in vivo models in spite of the challenges associated with the models, such as whether the tumors are actually arising in the appropriate organ and tissue site, so that the model might have potentially considerable relevance to the situation in humans. We haven’t spent so much time thinking about how to make organoid models better, and have largely been emphasizing our current mouse models.