Characterization of cell states in biliary tract cancers identifies mechanisms of therapeutic resistance in a phase II trial of DKN-01/nivolumab

Abstract

Biliary tract cancers demonstrate profound therapeutic resistance, and broadly effective therapies for refractory disease are lacking. We conducted a single-arm, second-line phase II trial combining DKN-01, a humanized monoclonal antibody targeting Dickkopf-1 (DKK-1), and nivolumab to treat patients with advanced biliary tract cancer (NCT04057365). No objective responses were seen. To identify mechanisms of treatment failure, we analyzed paired pre-treatment and on-treatment biopsies using scRNA-seq and constructed a detailed molecular classification of malignant and immune cells. We annotated five biliary tract cancer malignant cell states: classical, basal, mesenchymal, neural-like, and endothelial-like. Neural-like and endothelial-like states, which drive therapeutic resistance in other cancers, have not previously been described in BTC. Malignant cell states co-varied with distinct immune cell states, revealing diverse mechanisms of myeloid and T-cell mediated immune suppression, including M2 myeloid and terminally exhausted T cell programs that were induced by DKN-01/nivolumab. Here, we provide the first systematic classification of functionally annotated cell states in biliary tract cancer and provide new insight into resistance mechanisms to an immunotherapy combination that can inform the next generation of trials.

Figure 1.

Single-cell profiling of paired patient biopsies in a clinical trial of biliary tract cancers patients treated with DKN-01 and nivolumab. (A) Schematic of trial treatments, timing of biopsies and analysis performed in this study. (B) UMAP embedding of single-cell RNA-seq (scRNAseq) profiles of all cells obtained across biopsy samples labeled by major cell types. (C) Dot plot showing expression level and proportion of cells expressing marker genes for all major cell types from scRNAseq data. Differential expression analysis was performed using a Wilcoxon rank-sum test. (D) Cell type proportions of all major cell types in baseline and on-treatment samples analyzed using all samples (left; n=11) and only-paired samples (right, n=6).

Figure 2.

DKN-01 and nivolumab treatment leads to a redistribution of immune cell types in the tumor microenvironment. (A) UMAP embedding of all T cells profiled by scRNAseq labeled by granular subsets. Heatmap showing marker genes for (B) CD4+ and (C) CD8+ T cell subsets. (D) UMAP embedding of myeloid cells profiled by scRNAseq labeled by granular subsets. (E) Heatmap showing marker genes for myeloid cell subsets. Proportion of granular (F) CD4+ T cell, (G) CD8+ T cell and (H) myeloid cell subsets relative to their parent group in pre- and on-treatment biopsies.

Figure 3.

Defining novel tumor cell states in biliary tract cancers. (A) UMAP embedding of all tumor cells profiled by scRNAseq labeled by tumor cell states. (B) Dot plot showing marker genes for each tumor cell state. Differential expression analysis was performed using a Wilcoxon rank-sum test. (C) Proportion of each tumor cell state relative to all tumor cells in pre- and on-treatment biopsies. (D)-(E) Pathway scoring of (D) known cell cycle and signaling and (E) EMT gene programs in tumor cells belonging to each cell state. (F) Prediction of regulons (using SCENIC+) with highest specificity in each cell state. (G) UMAP embedding of all tumor cells showing activity of select tumor cell state-specific regulons.

Figure 4.

Heatmap showing Pearson correlation in abundance of all tumor cell states, T cell subsets and myeloid cell subsets across all samples.