Molecular classification and therapeutic targets in extrahepatic cholangiocarcinoma

Abstract

Background & aims: Cholangiocarcinoma (CCA), a deadly malignancy of the bile ducts, can be classified based on its anatomical location into either intrahepatic (iCCA) or extrahepatic (eCCA), each with different pathogenesis and clinical management. There is limited understanding of the molecular landscape of eCCA and no targeted therapy with clinical efficacy has been approved. We aimed to provide a molecular classification of eCCA and identify potential targets for molecular therapies.

Methods: An integrative genomic analysis of an international multicenter cohort of 189 eCCA cases was conducted. Genomic analysis included whole-genome expression, targeted DNA-sequencing and immunohistochemistry. Molecular findings were validated in an external set of 181 biliary tract tumors from the ICGC.

Results: KRAS (36.7%), TP53 (34.7%), ARID1A (14%) and SMAD4 (10.7%) were the most prevalent mutations, with ∼25% of tumors having a putative actionable genomic alteration according to OncoKB. Transcriptome-based unsupervised clustering helped us define 4 molecular classes of eCCA. Tumors classified within the Metabolic class (19%) showed a hepatocyte-like phenotype with activation of the transcription factor HNF4A and enrichment in gene signatures related to bile acid metabolism. The Proliferation class (23%), more common in patients with distal CCA, was characterized by enrichment of MYC targets, ERBB2 mutations/amplifications and activation of mTOR signaling. The Mesenchymal class (47%) was defined by signatures of epithelial-mesenchymal transition, aberrant TGFβ signaling and poor overall survival. Finally, tumors in the Immune class (11%) had a higher lymphocyte infiltration, overexpression of PD-1/PD-L1 and molecular features associated with a better response to immune checkpoint inhibitors.

Conclusion: An integrative molecular characterization identified distinct subclasses of eCCA. Genomic traits of each class provide the rationale for exploring patient stratification and novel therapeutic approaches.

Lay summary: Targeted therapies have not been approved for the treatment of extrahepatic cholangiocarcinoma. We performed a multi-platform molecular characterization of this tumor in a cohort of 189 patients. These analyses revealed 4 novel transcriptome-based molecular classes of extrahepatic cholangiocarcinoma and identified ∼25% of tumors with actionable genomic alterations, which has potential prognostic and therapeutic implications.

Keywords: Biomarkers; Extrahepatic cholangiocarcinoma; Immunotherapy; Liver cancer; Molecular classification; Targeted therapies.

Fig. 1.. Landscape of structural genomic alterations in eCCA.

(A) Frequency of recurrent mutations and copy number alterations in eCCA ranked by their prevalence. Color of bars represents the type of genomic alteration: orange: missense; blue: nonsense; green: frameshift indel; purple: splice site; red: amplification; and yellow: TERT promoter mutations. (B) Pathway diagrams showing the percentage of samples from each eCCA anatomical location with structural genomic alterations in genes from RTK-RAS-PI3K, TP53-RB, histone modification and TGFβ pathways. Red and blue mean alterations leading to activation or inactivation of the gene, respectively.

Fig. 2.. Molecular classes of eCCA and their biological features defining the tumor and its microenvironment.

(A) Solutions for transcriptome-based unsupervised classification of eCCA using non-negative matrix factorization consensus are shown for k = 2 to k = 5 classes; being four the number of classes with the highest cophenetic coefficient. Heatmaps of: (B) hallmark gene sets from MSigDB collections and (C) immune subpopulations inferred by gene expression of immune metagenes described in The Cancer Immunome Atlas significantly enriched in any of the four molecular classes of eCCA. Single-sample Gene Set Enrichment Analysis (ssGSEA) was used to obtain the enrichment score, representing the degree of which the genes in a particular gene set are coordinately up- or down-regulated. Samples from the same molecular class were represented with a normalized enrichment score. P values between a specific molecular class and the rest were calculated using T-Test. Box plots representing the estimation of: (D) stromal; and (E) immune compartment in each eCCA molecular class according to virtual microdissection of tumor-microenvironment using gene expression data (ESTIMATE package). P values were calculated using a two-sided T-test. (F) Relative RNA expression of cell of origin markers (stem cell, hepatocyte and biliary) in the four molecular eCCA classes in comparison to normal bile duct. P values were calculated using a two-sided T-test. Error bars represent 95% confidence intervals.

Fig. 3.. Associations between eCCA molecular classes and clinical, pathological and molecular traits.

(A) Heatmap representing the clinical-pathological characteristics and previously described CCA-HCC-PDAC molecular classes for each single sample and grouped according to the proposed eCCA molecular classes. Tumor, stromal and immune cellularity was analyzed by pathological evaluation. Molecular classes described in other tumors were inferred using Nearest Template Prediction (NTP). P values were calculated using a two-sided Fisher’s exact test for categorical variables and T-test for categorical and continuous data. The color of p value identifies the molecular class (red: Metabolic; green: proliferation; blue: mesenchymal; and orange: immune) that presents significant differences when compared with the rest. Total number of eCCA samples per molecular class with a positive expression of (B) ERBB2 (circumferential membrane staining that is complete, intense, and within > 10% of tumor cells), (C) PD-L1 (membranous staining of tumor cells or stromal cells in >1% over the total number of cells) and (D) PD-1 (cytoplasmic staining of >5% lymphocytes over the total number of intra-tumoral lymphocytes) assessed by IHC. P values were calculated using a two-sided Fisher’s exact test between a specific molecular class and the rest. (E) Kaplan-Meier curves comparing OS in the four molecular eCCA classes. Subclass mapping comparing the transcriptome of the four molecular eCCA classes with external cohorts of: (F) solid tumors including melanoma, non-small cell lung carcinoma and head and neck squamous cell carcinoma treated with anti PD-1 monoclonal antibodies[54]; and (G) iCCA defining proliferation and inflammation classes[12]. P values were obtained after 100 random permutations and Bonferroni correction. CCA: Cholangiocarcinoma; iCCA: Intrahepatic cholangiocarcinoma; HCC: Hepatocellular carcinoma; PDAC: Pancreatic ductal adenocarcinoma; BilIN: Biliary intraepithelial neoplasia; IPNB: intraductal papillary neoplasm of the bile duct.

Fig. 4.. External validation of eCCA molecular classifier.

(A) Heatmap showing molecular eCCA class prediction in the ICGC validation cohort using a predefined list of 174 marker genes (eCCA classifier) analyzed by Nearest Template Prediction (NTP). As measure of significance of the prediction for each sample, false discovery rate (FDR) <0.05 was selected. (B) Prevalence of eCCA molecular classes for pCCA and dCCA in both discovery and validation (ICGC) cohorts. (C) Distribution of eCCA molecular classes according to anatomical location in the ICGC validation cohort containing 181 BTC. iCCA: Intrahepatic cholangiocarcinoma; pCCA: Perihilar cholangiocarcinoma; dCCA: Distal cholangiocarcinoma; GBC: Gallbladder cancer.

Fig. 5.. Summary of characteristics of eCCA molecular classes.

Schematic representation of the main tumor/microenvironment molecular features and clinical characteristics describing Metabolic, Proliferation, Mesenchymal and Immune classes. In the bottom of the figure, candidate targeted therapies with a potential benefit in each molecular class. *Recommendation of treatment based on molecular data at transcriptomic and protein level. Phase 2/3 clinical trials to investigate these findings should be conducted to confirm the efficacy of the proposed targeted therapies. **Suggested targeted therapies supported by molecular associations at transcriptomic level. Further functional preclinical studies should be conducted prior testing them in the setting of clinical trials. EMT: epithelial-to-mesenchymal transition; IPNB: intraductal papillary neoplasm of the bile duct.