KRT17high/CXCL8+ Tumor Cells Display Both Classical and Basal Features and Regulate Myeloid Infiltration in the Pancreatic Cancer Microenvironment

Abstract

Purpose: Pancreatic ductal adenocarcinoma (PDAC) is generally divided in two subtypes, classical and basal. Recently, single-cell RNA sequencing has uncovered the coexistence of basal and classical cancer cells, as well as intermediary cancer cells, in individual tumors. The latter remains poorly understood; here, we sought to characterize them using a multimodal approach.

Experimental design: We performed subtyping on a single-cell RNA sequencing dataset containing 18 human PDAC samples to identify multiple intermediary subtypes. We generated patient-derived PDAC organoids for functional studies. We compared single-cell profiling of matched blood and tumor samples to measure changes in the local and systemic immune microenvironment. We then leveraged longitudinally patient-matched blood to follow individual patients over the course of chemotherapy.

Results: We identified a cluster of KRT17-high intermediary cancer cells that uniquely express high levels of CXCL8 and other cytokines. The proportion of KRT17high/CXCL8+ cells in patient tumors correlated with intratumoral myeloid abundance, and, interestingly, high protumor peripheral blood granulocytes, implicating local and systemic roles. Patient-derived organoids maintained KRT17high/CXCL8+ cells and induced myeloid cell migration in a CXCL8-dependent manner. In our longitudinal studies, plasma CXCL8 decreased following chemotherapy in responsive patients, while CXCL8 persistence portended worse prognosis.

Conclusions: Through single-cell analysis of PDAC samples, we identified KRT17high/CXCL8+ cancer cells as an intermediary subtype, marked by a unique cytokine profile and capable of influencing myeloid cells in the tumor microenvironment and systemically. The abundance of this cell population should be considered for patient stratification in precision immunotherapy. See related commentary by Faraoni and McAllister, p. 2297.

Figure 1.

Epithelial subtyping reveals subpopulations of basal, classical, and intermediary cancer cells. A, Uniform Manifold Approximation and Projection (UMAP) of scRNA-seq of 18 treatment-naïve PDAC. B, UMAP of tumor epithelial cells extracted from scRNA-seq of PDAC. Numbers represent unbiased clustering of populations. C, Dotplot showing average expression and percent cells of top expressing markers in each epithelial cluster. Cluster 5 markers outlined in green box. Genes highlighted yellow represent classical subtype markers. Genes outlined in magenta represent basal subtype markers. D, Feature plots showing expression of CLDN18, GATA6, KRT17, and CXCL8 in tumor epithelial cells. Cluster 5 epithelial cells denoted by dashed oval. E, Feature plots of gene set scoring of classical and basal signatures on tumor epithelial cells. F, Gene set scoring of each epithelial cluster for classical and basal gene sets. Numbers represent unbiased clustering of epithelial populations. Blue arrow denotes epithelial Cluster 5 (cluster with highest expression of KRT17, CLDN18, and CXCL8). G, Violin plots showing normalized expression of top expressing markers from epithelial Cluster 5 within the basal clusters (Clusters 3, 8, and 10), classical clusters (Clusters 0 and 7), and intermediary clusters (Clusters 5 and 6). H, Functional annotation showing top pathways expressed in epithelial Cluster 5 utilizing KEGG. The size of each dot represents gene count.

Figure 2.

CXCL8⁺ tumor prevalence is associated with worse OS. A, Surgically resected PDAC tissue stained with combined antibody immunostaining for KRT17 (white) and ISH utilizing a probe for CXCL8 (green) and CXCL1 (red). B, Surgically resected PDAC tissue stained with combined antibody immunostaining for KRT17 (green)/CLDN18 (red) and ISH utilizing a probe for CXCL8 (magenta). White arrow denotes cell that is positive for CXCL8, KRT17, and CLDN18. C, Histogram of single-cell epithelial cluster frequencies by patient. D, Multivariate survival analysis of Puleo cohort using significant covariates from univariate survival analysis of top expressing markers from epithelial Cluster 5 and clinical features as variables. Pink highlighted genes represent significant increase in hazard while blue highlighted genes represent significant decrease in hazard. E, Kaplan–Meier curve of Puleo cohort of patients with high versus low IL8 tumoral gene expression (determined by top and bottom quartile), n = 288.

Figure 3.

CXCL8 is expressed in a unique subpopulation of tumor epithelial cells and in tumor-infiltrating granulocytes. A, Surgically resected PDAC stained with combined antibody immunostaining for epithelial cells (E-cadherin, white), immune cells (CD45, red), and ISH utilizing a probe for CXCL8 (green). B, Three surgically resected treatment-naïve PDAC (left) and three surgically resected treated PDAC (right) stained with combined antibody immunostaining for epithelial cells (E-cadherin, white), immune cells (CD45, red), and ISH utilizing a probe for CXCL8 (green).

Figure 4.

Functional analysis ex vivo reveals reciprocal interactions between cancer cells and granulocytes in the pancreatic cancer microenvironment mediated by CXCL8. A, Light microscopy (left) and H&E staining (right) of organoid lines isolated from tissue of 5 patients with PDAC. B, Organoids stained with antibody immunostaining for KRT17 (white) and ISH utilizing a probe for CXCL8 (green) and CXCL1 (red). C, Scheme of coculture to assay chemotaxis of PDAC patient myeloid cells to PDAC organoid conditioned media. D, Migration assay showing chemotaxis index (ratio of cells migrated through transwell in treatment condition compared with cells migrated in control media) in organoid lines with and without CXCL8-blocking antibody. Recombinant CXCL8 (rhCXCL8) was used as a positive control. Comparisons were made using one-way ANOVA. E, Scheme of assay to profile CXCL8 levels in healthy donor CD11b⁺ myeloid cells in response to PDAC organoid conditioned media. F, Relative fold change in CXCL8 mRNA levels in myeloid cells isolated from healthy donor blood treated with conditioned media in CXCL8^high versus CXCL8^low tumor organoid lines (1225 and 1253) with and without anti CXCL8-blocking antibody.

Figure 5.

Tumor-derived CXCL8 correlates with specific changes in the local and systemic immune system. A, Scheme of workflow to compare scRNA-seq of matched tumor-infiltrating granulocytes from patients with tumors high in CXCL8⁺ epithelial cells versus tumors low in CXCL8⁺ epithelial cells. B, Unbiased differential expression between tumor-infiltrating granulocytes from tumors high in CXCL8⁺ epithelial cells versus tumors low in CXCL8⁺ epithelial cells. Significantly upregulated and downregulated genes are plotted as the average expression. Purple arrows denote genes associated with protumor phenotype, upregulated in epithelial CXCL8^high group. Blue arrows denote upregulation of CXCL8 receptors CXCR1 and CXCR2 in epithelial CXCL8^low group. C, UMAP visualization of myeloid cells extracted from PDAC scRNA-seq. D, Average expression of CXCL8 in the epithelial and myeloid subsets by patient. E, Scatterplot of percent positive CXCL8 myeloid cells versus percent positive CXCL8 epithelial cells. Each dot represents one patient. F, Feature plot of CXCL8 in tumor-infiltrating myeloid cells (left). Extracted granulocyte cells, colored by cluster (right). G, Top expressed genes within granulocyte Clusters G1 through G5. H, Violin plots showing normalized expression of select markers in granulocyte Clusters G1 through G5. I, Scheme of workflow to compare scRNA-seq of matched peripheral blood granulocytes from patients with tumors high in CXCL8⁺ epithelial cells versus tumors low in CXCL8⁺ epithelial cells. J, Unbiased differential expression between peripheral blood granulocytes from patients with tumors high in CXCL8⁺ epithelial cells versus tumors low in CXCL8⁺ epithelial cells. Significantly upregulated and downregulated genes are plotted as the average expression. Purple arrows denote genes associated with protumor phenotype, upregulated in epithelial CXCL8^high group. Blue arrows denote upregulation of genes associated with antitumor phenotype, upregulated in epithelial CXCL8^low group. K, UMAP visualization of all captured cells from PDAC blood of our single-cell cohort. Populations defined by color. L, Top: UMAP visualization of extracted granulocytes from PDAC blood single-cell sequencing. Numbers represent unbiased clustering of populations. Bottom: CXCL8 feature plot of extracted granulocytes. M, Dotplot showing average expression and percent cells expressed of CXCL8, CXCR1, CXCR2, CXCR4, SPP1, CCL4, SELL, and VEGFA in each peripheral blood granulocyte cluster. Green box outlines Cluster 0, the cluster with highest expression of CXCL8.

Figure 6.

Longitudinal matched blood immunoprofiling shows that myeloid-derived CXCL8 is modulated by chemotherapy. A, Scheme of longitudinal matched blood collection from patients with PDAC for scRNA-seq analysis with relevant clinical information (OS = overall survival, PFS = progression-free survival). B, UMAP visualization of all captured peripheral blood immune cells before and on chemotherapy. Populations defined by color. C, Scatterplot of differentially expressed genes comparing scRNA-seq of treatment-naïve blood immune cells to matched on-treatment blood immune cells. Myeloid genes are labeled in red; natural killer (NK) and T cell genes are labeled in green. Each dot represents a significantly upregulated or downregulated gene. D, UMAP of extracted myeloid cells from longitudinal blood single-cell sequencing. Left: Myeloid cells from patient blood before treatment. Right: Myeloid cells from matched patient blood on treatment. E, Dotplot showing markers defining different myeloid populations. F, Histogram of single-cell myeloid cluster frequencies comparing treatment-naïve (T1) with on chemotherapy (T2) timepoints. G, Average expression of differentially expressed genes in granulocytes, alternatively activated monocytes, and classical monocytes comparing treatment-naïve (teal) with treated (coral) matched patients. Blue markers denote CXCL8 significantly decreased in all myeloid populations with chemotherapy. Pink marker denotes CXCL1 is significantly decreased in granulocyte population with chemotherapy.

Figure 7.

CXCL8 levels in patient plasma correlate with abundance of KRT17⁺ epithelial cells and with prevalence of myeloid cells in matched patient tumors and portend poor prognosis. A, Surgically resected PDAC tissue costained with antibodies for myeloid cells, marked by CD11b (green) and KRT17 (red). Matched plasma from patient tissues on the left were low for CXCL8; matched plasma of patient tissues on the right were high for CXCL8. Top row represents treatment-naïve tumors; bottom row represents tumors from patients who were treated with chemotherapy before surgical resection. s/p, status post. B, Surgically resected treatment-naïve PDAC tissue costained with antibody immunostaining for KRT17 (white) and ISH utilizing a probe for CXCL8 (green). C, Fold change in plasma CXCL8 in matched longitudinally sampled patient blood, plotted against days on chemotherapy. Each patient represents one line, where nonresponders to chemotherapy are denoted by a red line and responders to chemotherapy are denoted by a black line. D, Scatterplot showing correlation of plasma CA19-9 with plasma CXCL8 in the Michigan cohort. E, Barplot comparing plasma CA19-9 in patients with low and high plasma CXCL8 (defined by lowest and highest quartiles, respectively) in the Michigan cohort.

Figure 8.

Graphical abstract of working model. KRT17^high;CXCL8⁺ cancer cells express both classical and basal subtype features and attract tumor-promoting neutrophils in a CXCL8⁺-dependent manner; these neutrophils, in turn, express more CXCL8 in a feed-forward mechanism.