Overcoming intra-tumoral heterogeneity for biomarker discovery in the high-grade serous ovarian cancer proteome

Abstract

Improved biomarkers of treatment response are needed for patients with high-grade serous ovarian cancer (HGSC). A challenge is substantial anatomical site-to-site variation in expression. We completed data-independent acquisition-mass spectrometry (DIA-MS) analysis of 404 fresh frozen and 78 formalin-fixed, paraffin-embedded HGSC tissue samples from the ovary (adnexal) and a common secondary site (omentum) in 11 patients. This was compared with mutation testing, gene expression, and whole-genome copy number profiling. Proteins with relatively stable intra- and variable inter-individual expression (n = 1651), included a 52-protein module reflecting interferon-mediated tissue inflammation, indicative of a cGAS-STING pathway cytosolic double-stranded (ds) DNA response. The dsDNA sensing/inflammation score was higher in the omentum compared with the ovary. Ovarian HGSC samples showed marked inter-individual differences in inflammatory and immune responses to DNA damage. Stable discriminative features of the HGSC proteome, a prerequisite for clinical predictive biomarkers, are detectable in ovary (adnexal) tissue samples.

Fig. 1. Multi-omic analysis of HGSC in ovary and omentum from 11 individuals.

a The cohort included 11 individuals with HGSC. Tumor tissue from both the ovary/adnexal site and omentum was extensively sampled and analyzed. The final dataset included DIA-MS proteomic analysis of multiple biopsy-sized samples of FF tissue (n = 404, 6–49 per tissue, 11–80 per individual), and FFPE tissue sections (n = 78, 2–4 per tissue, 5–8 per individual). Additional data were multigene targeted next generation sequencing (NGS) mutation analysis (FF ovary), whole-genome single nucleotide polymorphism (SNP) microarray copy number variation (CNV) analysis (FF ovary and omentum), whole-genome gene expression (RNASeq) (FF ovary and omentum) and gene expression molecular subtype (NanoString PrOTYPE) analysis (FFPE ovary and omentum). b Summary of phenotypic and genotypic features. TAI telomeric allelic imbalance (TAI), LST large-scale transitions, HRD-LOH loss of heterozygosity, IHC immunohistochemistry. c The global proteomic landscape of 404 individual FF samples was influenced by inter-individual differences (tSNE) (proteins n = 7232). d The same plot colored by relative stromal score showed similarity between cases with high stromal content from disparate individuals. e Distribution of stromal scores in individual FF samples taken from ovary and omentum. The range of scores from a single piece of tissue was variable reflecting heterogeneity in tissue content. Stromal scores tended to be higher in omentum samples, and the difference was significant in eight out of ten cases (n = 404) (t test, p < 0.05).

Fig. 2. Inflammatory response is characteristic in the HGSC proteome.

a Filtering steps to identify stable intra-individual and variable inter-individual expression. b Biological pathways represented by proteins in the final protein matrix. Dot plot showing enriched hallmark pathways (adj p < 0.05, Fisher exact) using gene set enrichment. The size of circle corresponds to number of genes. The combined score is natural log of the p value multiplied by the z-score, where the z-score is the deviation from the expected rank. c WGCNA identified six protein co-expression modules from the 1648 mapped stable discriminative proteins in 319 FF tissue samples. Different modules gave representation to DNA repair (module 1, aqua), oxidative phosphorylation (module 3, brown), fatty acid metabolism (module 4, yellow), interferon α (IFNα) and IFNɣ signaling (module 5, green) and Epithelial Mesenchymal Transition (EMT) (module 6, red). Red-blue spectrum represents relative protein abundance (z score normalized protein expression). d The range of ssGSEA scores derived from expression of distinct pathway related proteins in modules 1 (aqua), 3 (brown) and 5 (green) respectively, across multiple FF samples from a single piece of ovary or omental HGSC tissue (samples n = 404, protein matrix n = 4715).

Fig. 3. Expression of inflammatory proteins and immune cell infiltrates are different in primary and omental tissues.

a STRING network of 23/52 core connected proteins DSI proteins. Nodes colored by Reactome pathways: Interferon signaling (red) [Reactome stable identifier R-HSA-913531], Innate immune system (green) [R-HAS-168249], adaptive immune system (purple) [R-HAS-1280218], ISG15 antiviral mechanism (yellow) [R-HSA-1169408]. b The range of DSI scores across multiple FF tissue samples from primary and omental tissues (samples n = 404, 6–49 samples per tissue), normal fallopian tube tissue (two samples), a pooled normal ovarian tissue sample, two HGSC cell lines (three samples per cell line) (protein matrix n = 4715). Scores were significantly higher in omental samples in seven individuals (t test independent *: 1.00e-02 < p ≤ 5.00e-02, **: 1.00e-03 < p ≤ 1.00e-02, ***: 1.00e-04 < p ≤ 1.00e-03, ****: p ≤ 1.00e-04). c The range of DSI scores across multiple FFPE HGSC tissue sections from ovary (n = 37) and omental tissues (n = 41) (2–4 per tissue), fallopian tube tissue (n = 5), normal ovarian tissue (n = 6) (protein matrix n = 4715). d DSI and ESTIMATE immune scores were positively correlated (n = 404, R² = 0.72, Pearson correlation). e There was no significant correlation between the DSI and ESTIMATE stromal scores (n = 404, R² = 0.16, Pearson correlation). f CD8+ T cell infiltration scores derived from CIBERSORTx analysis of RNA-Seq data. Only two ovary samples showed appreciable CD8+ T-cell infiltration. g The DSI score was significantly higher in HGSC samples taken from omentum (n = 38) compared with samples from ovary (n = 40) in an independent published proteomic study (*: p < 0.05, T test).

Fig. 4. Module 5 proteins are elevated in HR-deficient HGSC.

a There were 158 differentially expressed proteins between FF samples (≥20% tumor content) from HR-intact (n = 220 samples, 6 cases) and HR-deficient (n = 90 samples, 4 cases) HGSC (p < 0.05, Limma t test). b Biological processes enriched in proteins that were differentially abundant in HR-intact samples (p value < 0.05, Fisher’s exact test). c Scores for n = 53 HR-intact proteins in HR-intact (n = 65) and HR-deficient (n = 118) HGSC cases from independent published data^, (protein matrix n = 9600) (t test, p < =0.02). d Biological processes enriched in proteins that were differentially abundant in HR-deficient samples (p value < 0.05, Fisher’s exact test). e DSI scores in individual FF samples taken from HGSC that were HR-intact (n = 264 samples) and HR-deficient (n = 124 samples) (independent t test, p < 0.05). f DSI scores for HR-intact (n = 65) and HR-deficient (n = 118) samples from independent published data^, (available DSI proteins n = 50).

Fig. 5. Inflammatory markers are variable in ovarian HGSC tissue, and different from omentum.

a Correlation between CIBERSORTx CD8+ T cells scores and CCL5 mRNA expression in RNA-Seq data (n = 22, Pearson correlation, R² = 0.65). b Correlation between CCL5 and CD274 (PDL1) mRNA expression in RNA-Seq data (n = 22, Pearson correlation R² = 0.92). c, d HR-deficiency, DSI scores, CCL5 mRNA, CD8+ T cell CIBERSORTx scores and CD274 (PDL1) mRNA in ovarian and omental tumor tissues. Color scale indicates position in range.