Serum levels of iCAF-derived osteoglycin predict favorable outcome in pancreatic cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is characterized by abundant stroma, the main cellular constituents of which are cancer-associated fibroblasts (CAFs). Stroma-targeting agents have been proposed to improve the poor outcome of current treatments. However, clinical trials using these agents showed disappointing results. Heterogeneity in the PDAC CAF population was recently delineated demonstrating that both tumor-promoting and tumor-suppressive activities co-exist in the stroma. Here, we aimed to identify biomarkers for the CAF population that contribute to a favorable outcome. RNA-sequencing reads from patient-derived xenografts (PDXs) were mapped to the human and mouse genome to allocate the expression of genes to the tumor or stroma. Survival meta-analysis for stromal genes was performed and applied to human protein atlas data to identify circulating biomarkers. The candidate protein was perturbed in co-cultures and assessed in existing and novel single-cell gene expression analysis from control, pancreatitis, pancreatitis-recovered and PDAC mouse models. Serum levels of the candidate biomarker were measured in two independent cohorts totaling 148 PDAC patients and related them to overall survival. Osteoglycin (OGN) was identified as a candidate serum prognostic marker. Single-cell analysis indicated that Ogn is derived from a subgroup of inflammatory CAFs. Ogn-expressing fibroblasts are distinct from resident healthy pancreatic stellate cells and arise during pancreatitis. Serum OGN levels were prognostic for favorable overall survival in two independent PDAC cohorts (HR = 0.47, P = .042 and HR = 0.53, P = .006). Altogether, we conclude that high circulating OGN levels inform on a previously unrecognized subgroup of CAFs and predict favorable outcomes in resectable PDAC.

Keywords: CAF subtypes; liquid biopsy center; pancreatic ductal adenocarcinoma; stroma.

FIGURE 1

Identification of stromal genes associated with favorable prognosis in PDAC. (A) Diagram indicating the analysis workflow to interrogate gene expression specific to the mouse host (stroma), and the grafted human tumor cells. (B) Survival meta‐analysis was performed on the top 1000 most stromal‐associated genes using a random‐effect model. The volcano plot shows the effect size on the x‐axis and P‐value on the y‐axis for each biomarker. The colors indicate the concentrations found in blood by the Human Protein Atlas. (C) Forest plot of hazard ratios (HRs) for the prognostic value of OGN across PDAC gene expression dataset. All seven data sets show a decreased hazard ratio with increasing OGN expression levels

FIGURE 2

OGN is required for fibroblast activation and marks a distinct iCAF population. (A) Quantification of mOgn expression by qPCR in mouse embryonic fibroblasts (C3H/10T1/2, NIH/3T3) after knockdown. Values were normalized to scrambled control (mean ± SD, n = 3 biological replicates per group). (B) Quantification of stromal activation markers in mOgn‐KD MEFs after coculture with human tumor cells using qPCR. Values were normalized to scrambled control (mean ± SD, n = 3 biological replicates per group). (C) As for experiment shown in panel B, showing expression of human Vimentin (hVIM). Data are mean ± SD, n = 3 biological replicates per group, normalized to scrambled control (shc002). (D) As for panels A to C, using PANC‐1 cells and NIH/3T3 fibroblasts. Cells were processed for flow cytometry using an antibody against human N‐cadherin. In addition, cancer cells were identified on scatter profile. Shown are gMFI, normalized to scrambled control (included in the panel). xData are technical triplicates. (E) UMAP embedding of CAFs from KPC mice, showing heterogenous expression of mOgn. (F,G) Violin plot of mOgn expression for each CAF subtype, and for each iCAF subcluster, respectively. (H) Kaplan‐Meier survival analysis of TCGA PAAD patients based on iCAF.1 signature as identified by stromal gene ranking analysis. The samples were dichotomized by the median. n.s. = not significant, *P < .05, **P < .01, ***P < .001 and ****P < .0001.

FIGURE 3

OGN⁺ fibroblasts pre‐exist in the nonmalignant pancreas. (A) UMAP embedding of 3691 cells from eight mice across three experimental conditions. Cell types identified based on marker expression are indicated. (B) Expression of selected stellate cell and fibroblast markers using the UMAP representation from in silico subset fibroblasts. (C) Proportion of Ogn⁺ fibroblasts from the total fibroblast population, grouped by experimental condition (median ± minimum/maximum). (D) Feature plot of expression distribution for Fabp4 (red) and Ogn (blue) across pseudotime. Fabp4 associates with the resident pancreatic stellate cell population. Ogn is part of an activated fibroblast population, which increases in pseudo‐time

FIGURE 4

Serum OGN levels correlate with favorable outcomes in resectable PDAC patients. (A) Kaplan‐Meier survival analysis of AMC PDAC patients (BioPAN, n = 75) using a cut‐off value of 40.1 ng/mL, determined by ROC. (B) Validation of the prognostic power of serum OGN in an independent cohort of PDAC patients (DPCG, n = 108). Upper panel shows the validation of the 40.1 ng/mL threshold. The lower panel shows the most optimal cut‐off (55.4 ng/mL) for this cohort determined by ROC in this cohort