Spatiotemporal architecture of immune cells and cancer-associated fibroblasts in high-grade serous ovarian carcinoma

Abstract

High-grade serous ovarian carcinoma (HGSOC), the deadliest form of ovarian cancer, is typically diagnosed after it has metastasized and often relapses after standard-of-care platinum-based chemotherapy, likely due to advanced tumor stage, heterogeneity, and immune evasion and tumor-promoting signaling from the tumor microenvironment. To understand how spatial heterogeneity contributes to HGSOC progression and early relapse, we profiled an HGSOC tissue microarray of patient-matched longitudinal samples from 42 patients. We found spatial patterns associated with early relapse, including changes in T cell localization, malformed tertiary lymphoid structure (TLS)-like aggregates, and increased podoplanin-positive cancer-associated fibroblasts (CAFs). Using spatial features to compartmentalize the tissue, we found that plasma cells distribute in two different compartments associated with TLS-like aggregates and CAFs, and these distinct microenvironments may account for the conflicting reports about the role of plasma cells in HGSOC prognosis.

Fig. 1.. IMC experimental methods and phenotyping.

(A) Patient-matched primary, synchronous metastasis, and recurrent metastasis HGSOC samples from 42 patients were assembled into a TMA. The Kaplan-Meier survival curve below shows the time to relapse, with patients classified as early or late relapse (<15 or >15 months to relapse, respectively). Scale bars, 4 mm. (B) Imaging mass cytometry was performed by staining the tissue with metal ion–tagged antibodies, ablating the tissue, and performing image data analysis. (C) Three major cell types and their protein expression patterns are shown in sample IMC images and a heatmap: immune cells, fibroblasts, and epithelial (cancer) cells. Scale bar, 25 μm. (D) Cell composition of (regions of interest) ROIs was approximately 50 to 55% epithelial cancer cells, 25 to 30% fibroblasts, and 15 to 20% immune cells in primary, synchronous metastasis, and recurrence tumor types. (E) Lymph node metastases contained elevated levels of immune cells, P = 1.1 × 10⁻⁵, analysis of variance (ANOVA). (F) Early and late relapse conditions were associated with similar proportions of immune cells, fibroblasts, and epithelial cells. Stacked bar plots are shown to illustrate the summation to 1, error bars denote SD, and the error bar is single-sided to reduce visual clutter. Cell proportion differences are not statistically significant by ANOVA unless noted with *.

Fig. 2.. Fine-grained phenotyping.

(A) Major immune cell types are plotted after applying Uniform Manifold Approximation and Projection (UMAP). Nine distinct immune subtypes were categorized. (B) The relative proportion and average expression pattern of immune subtypes are shown, identified by positive and negative selection. (C) Fibroblast subtypes categorized by the relative expression level of up to three markers—FAP, α–smooth muscle actin (α-SMA), and PDPN—are shown, along with their proportions. (D and E) Relative proportions of immune subtypes (D) and fibroblast subtypes (E) are shown for each tumor type. CD8 T cells as a fraction of immune cells were observed at significantly higher proportions in metastasis samples than primary samples; P = 0.028, Tukey’s test. Stacked bar plots are shown to illustrate the summation to 1, error bars denote SD, and the error bar is single-sided to avoid visual clutter. Cell proportion differences were not statistically significant by ANOVA unless noted with *.

Fig. 3.. IMC spatial analysis.

(A) For a sample image colored by cell type, Gcross analysis generated a cumulative distribution function of minimum cell-cell distances, and ΔGcross calculated the mean difference between the observed and expected cell-cell distances. High fibroblast-epithelial separation and low epithelial-epithelial separation are highlighted in blue and green, respectively. (B) A heatmap of clusters generated by ΔGcross comparisons shows manually annotated spatial patterns. Examples of ROIs in cluster 3 (immune-excluded tumors) and cluster 2 (well-mixed cells) are shown. (C) The spatial relationship of cells to other cell types was calculated by measuring the distance to five nearest neighbors, truncating the distance at 100 μm, and measuring the average of the five distances. (D) Spatial analysis of a sample IMC image shows immune cells, subtypes, and a lymphoid aggregate. (E) Single-cell spatial biases for T cells versus macrophages were calculated for each cell. Epithelial cells in synchronous metastatic tumors showed an increase in spatial T cell bias relative to other tumor types (P < 1 × 10⁻¹⁵, Tukey’s test), and recurrence tumors showed a smaller spatial T cell bias relative to primary tumors (P = 8.7 × 10⁻⁵, Tukey’s test). All scale bars, 100 μm.

Fig. 4.. Spatial analysis of early and late relapse patients.

(A) A spatial strategy was used to obtain “digital cell biopsies” that excise specific zones of each tumor. For example, two types of biopsies were generated for tumor cells to separate cells with no tumor contact (left), from cells with any tumor contact (middle), and cells in tumor-enriched areas (right). (B) In the image shown (IHC, top left; IMC phenotypes, top right), digital biopsies isolated different spatial regions based on tumor density with an inset region shown, highlighting the tumor cells (bottom left), cells selected and removed based on tumor contact (bottom middle), and tumor enrichment (bottom right). Scale bars, 100 μm. (C) Of all proportions measured, three terms were significantly (Benjamini-Hochberg procedure, q < 0.05) associated with early relapse after considering the tumor type. (D) Among cell proportions in digital-biopsied spatial areas, four fibroblast-related terms were significantly associated with early relapse (Benjamini-Hochberg procedure, q < 0.05). (E) After measuring changes between tissues taken from the same patient, digital biopsied terms were significantly associated with early relapse (P < 0.05). These terms were primarily immune and fibroblast cells found in immune-, fibroblast-, or epithelial-enriched areas. (F) For the percentage of B cells found in fibroblast-enriched areas between patient primary and recurrence samples, the percentage increased in late relapse patients but decreased in early relapse patients.

Fig. 5.. PDPN and plasma cell spatial analysis.

(A) Spatial transcriptomic selection of ROIs for digital spatial profiling was aided by multiplex immunofluorescence (mIF) staining with antibodies for PDPN, CD19, CD20, and CD38 (n = 14). Most PDPN-positive and PDPN-negative ROIs were in different cores; one core with both types of ROIs is shown for illustration. Scale bars, 200 μm. (B) Volcano plot of differential gene expression of ovarian cancer The Cancer Genome Atlas samples stratified by PDPN and TLS gene signature expression (ligands, red; receptors, blue; Benjamini-Hochberg adjustment). (C) QuPath AI-aided annotation of plasma cells and B cells located in PDPN-positive and PDPN-negative areas of the tumor. Scale bars, 100 μm. (D) Boxplots with scatter plots of the fold change difference of each sample between plasma cells within PDPN-positive areas versus PDPN-negative areas, and plasma cells versus B cells (q = 0.0071 and q = 0.026, respectively, Wilcoxon rank sum test, Benjamini-Hochberg adjustment).

Fig. 6.. IMC and pathology comparisons.

(A) IMC was compared to proximal and distal H&E sections analyzed by cell morphology classification. (B to D) All IMC cell counts were significantly associated with H&E analysis (P < 1 × 10⁻¹⁶, linear regression). Fibroblast comparisons had the most similar coefficient, while epithelial cell comparisons were the most correlated (R²). (E) For six samples available, three cores were individually analyzed by IMC and proximal H&E slices and compared to the whole image slide (equivalent distance to distal), showing concordance (pairwise t test, Benjamini-Hochberg correction).

Fig. 7.. Proposed spatial importance of B cells, plasma cells, TLS, and CAFs.

(A) TLS is an organized aggregate of multiple cell types, including immune cells and FRCs, which mediate the interaction between B and T cells to promote differentiation of B cells into plasma cells with antitumor implications. In contrast, plasma cells found within CAFs do not have direct contact with B and T cells to initiate antitumor functions, which may contribute to the negative impact of plasma cells on prognosis in these cases. (B) TLS FRCs and CAFs exhibit similar expression profiles; however, FRCs secrete a range of chemokines that attract B cells, whereas CAFs predominantly express CXCL12, which specifically attracts plasma cells. (C) Preclinical experiments have shown that depleting plasma cells with bortezomib can reverse the mesenchymal characteristics of ovarian cancer and inhibit tumor growth. Targeting plasma cells with bortezomib has the potential to disrupt the arrangement of CAFs, which could enhance the accessibility of chemotherapy agents to the tumor site.