Theses Doctoral

CDX2 as a Predictive Biomarker of Drug Response in Colon Cancer

Raab, William

Colon cancer is one of the most common cancers in both the United States (US) and throughout the world. Over the last 30 years, despite the development of multiple classes of effective anti-tumor agents, colon cancer has consistently remained the second leading cause of mortality amongst all cancers and is today responsible for over 50,000 deaths a year in the US alone. Among the greatest challenges to the successful treatment of colon cancer is its heterogeneity in terms of drug-sensitivity, whereby it is often difficult to identify which patients will benefit from a specific class of anti-tumor agents before treatment has begun. It is therefore imperative to identify predictive biomarkers that can be leveraged to distinguish which colon tumors are most likely to respond to individual anti-cancer drugs. This will help develop new therapeutic algorithms that can maximize patient survival by rapidly matching individual patients with the specific treatment combinations that are most likely to benefit them as well as sparing them the toxicities from drugs that would be ineffective. Previous studies have reported that human colon carcinomas lacking expression of the caudal-type homeobox 2 (CDX2) transcription factor can be leveraged as a predictor of benefit from adjuvant chemotherapy containing 5-fluorouracil (5-FU). Lack of CDX2 expression associates with microsatellite instability (MSI), as well as several histopathological and molecular features that associate with exceptionally poor prognosis such as poor differentiation, lympho-vascular invasion, and BRAF mutation. However, the molecular mechanisms linking lack of CDX2 expression with increased drug sensitivity are currently unknown.

In the first section of this study, we conducted a high throughput screen (HTS) aimed at identifying clinically approved anti-tumor drugs that display selective activity against colon carcinomas lacking CDX2 expression (CDX2-negative). The results of our screening, which compared an isogenic pair of CDX2+/+ and CDX2-/- cell lines generated by genetic inactivation of CDX2 using CRISPR/Cas9 constructs, revealed that CDX2-negative colon cancer cells display increased sensitivity to anti-tumor drugs that are substrates of the ATP binding cassette sub-family B member 1 (ABCB1) transporter. ABCB1 is a drug-efflux protein known for its capacity to extrude multiple classes of anti-tumor agents from the cytoplasm, therefore contributing to drug-resistance in cancer cells. Importantly, analysis of CDX2 and ABCB1 expression in two independent gene-expression databases (NCBI-GEO: n=2115; TCGA: n=478) revealed that a lack of CDX2 expression is invariably associated with lack of ABCB1 expression in human primary colon carcinomas. Furthermore, our molecular studies revealed that forced expression of CDX2 in human CDX2-negative colon cancer cells was capable of inducing expression of ABCB1, while genetic inactivation of CDX2 in human CDX2-positive cancer cells using CRISPR/Cas9 constructs resulted in loss of ABCB1 expression, thus establishing CDX2 as a direct mechanistic regulator of ABCB1 expression. Amongst all of the anti-tumor drugs identified as being ABCB1 substrates with preferential activity against CDX2-negative colon cancer cells, we observed that paclitaxel was the FDA-approved drug with the greatest degree of selectivity with a 10-fold difference in IC50. When tested in vivo against a collection of human patient derived xenograft (PDX) lines representative of both CDX2-negative and CDX2-positive colon carcinomas, paclitaxel displayed selective activity against CDX2-negative models, often inducing volumetric regression of established lesions. Our study, therefore, identified paclitaxel as a clinically approved anti-tumor agent that should be investigated for use in the treatment of CDX2-negative colon carcinomas.

In the second portion of our study, we sought to conduct a preliminary evaluation of the possibility of using immune checkpoint inhibitors (ICIs) for the treatment of CDX2-negative colon carcinomas. ICIs have been shown to display substantial anti-tumor activity against colon carcinomas with microsatellite instability (MSI) and against epithelial malignancies over-expressing the immune-suppressive molecule PD-L1/CD274. Because CDX2-negative tumors are enriched for MSI and high levels of PD-L1/CD274, they are predicted to include a subgroup that is responsive to ICIs. However, not all MSI tumors respond to ICIs and, contrary to the majority of MSI tumors, the subgroup of MSI tumors characterized by a CDX2-negative phenotype is often associated with poor prognosis. Because the clinical activity of ICIs is dependent upon expression of class-I HLA molecules by tumor cells, we decided to evaluate whether CDX2-negative tumors were associated with inactivating mutations in class-I HLA genes. Our attention focused on a highly conserved poly-cytosine repeat region in the coding sequence of HLA-A (c.621_627) and HLA-B (c.621_626) genes. Because this sequence fulfilled the molecular definition of microsatellite, we predicted it to be highly susceptible to frameshift mutations (insertions or deletions) in MSI colon tumors. Indeed, a search across three independent genetic databases (TCGA, COSMIC, EBI) confirmed that this highly conserved poly-cytosine repeat region was targeted by recurrent and deleterious mutations in at least one HLA-A or HLA-B allele of at least 13% (n=21/156) of human MSI colon tumors, as compared to 0.3% (n=2/770) of human colon tumors with a microsatellite stable (MSS) phenotype (p<0.0001). Among tumors assessable for CDX2 expression, this specific type of class-I HLA mutations was more frequent among CDX2-negative (12%; n=6/49) as compared to CDX2-positive (1.5%; n=5/340) colon tumors (p<0.001), but was similar within MSI CDX2-negative (21%; n=6/28) and MSI CDX2-positive (17%; n=5/30) subgroups.

In summary, this work achieved two main results: 1) it identified paclitaxel, a clinically approved anti-tumor drug, as a new treatment option for patients with CDX2-negative colon cancers, which represents an extremely aggressive subgroup of colorectal malignancies; 2) it revealed that, in human MSI colon tumors, class-I HLA genes are prone to recurrent frameshift mutations in a genomic hotspot, mutations that are likely to associate with tumor resistance to ICIs and that they are therefore likely to represent a new class of actionable predictive biomarkers for both MSI and CDX2-negative colon carcinomas. These findings will help advance our understanding of colon cancer biology, and hopefully improve treatment algorithms for the clinical management of colon cancer patients.

Geographic Areas


  • thumnail for Raab_columbia_0054D_16776.pdf Raab_columbia_0054D_16776.pdf application/pdf 4.05 MB Download File

More About This Work

Academic Units
Pathobiology and Molecular Medicine
Thesis Advisors
Dalerba, Piero D.
Ph.D., Columbia University
Published Here
August 25, 2021