Immunological modifications following chemotherapy are associated with delayed recurrence of ovarian cancer

Abstract

Introduction: Ovarian cancer recurs in most High Grade Serous Ovarian Cancer (HGSOC) patients, including initial responders, after standard of care. To improve patient survival, we need to identify and understand the factors contributing to early or late recurrence and therapeutically target these mechanisms. We hypothesized that in HGSOC, the response to chemotherapy is associated with a specific gene expression signature determined by the tumor microenvironment. In this study, we sought to determine the differences in gene expression and the tumor immune microenvironment between patients who show early recurrence (within 6 months) compared to those who show late recurrence following chemotherapy.

Methods: Paired tumor samples were obtained before and after Carboplatin and Taxol chemotherapy from 24 patients with HGSOC. Bioinformatic transcriptomic analysis was performed on the tumor samples to determine the gene expression signature associated with differences in recurrence pattern. Gene Ontology and Pathway analysis was performed using AdvaitaBio's iPathwayGuide software. Tumor immune cell fractions were imputed using CIBERSORTx. Results were compared between late recurrence and early recurrence patients, and between paired pre-chemotherapy and post-chemotherapy samples.

Results: There was no statistically significant difference between early recurrence or late recurrence ovarian tumors pre-chemotherapy. However, chemotherapy induced significant immunological changes in tumors from late recurrence patients but had no impact on tumors from early recurrence patients. The key immunological change induced by chemotherapy in late recurrence patients was the reversal of pro-tumor immune signature.

Discussion: We report for the first time, the association between immunological modifications in response to chemotherapy and the time of recurrence. Our findings provide novel opportunities to ultimately improve ovarian cancer patient survival.

Keywords: chemoresistance; cold tumors; hot tumors; immune response; ovarian cancer.

Figure 1

Schematic illustration of the experimental workflow followed in this study. Enrolled HGSOC patients had pre-chemo biopsy taken, after which they received 6 cycles of Carboplatin and Taxol (C/T). Following this, interval debulking surgery was performed and post-chemo samples obtained. Both pre-C/T and post-C/T samples were prepared and examined by a Board-Certified Pathologist, after which RNA extraction, cDNA library preparation and sequencing were performed on high quality samples. RNA-seq data was then analyzed in multiple steps for differential gene expression detection, pathway impact, and gene ontology enrichment detection as well as immune cell fraction estimation.

Figure 2

Chemotherapy induced gene expression changes in high grade serous ovarian cancer patients. (A) A volcano plot showing significantly upregulated (red) or downregulated (blue) genes (FDR < 0.05 and |log2-foldchange| > 0.6) post-C/T. Genes with FDR < 0.01 or |log2-foldchange| > 2 are labeled with HGNC symbols. (B) A heatmap showing the top 30 upregulated (red) and top 30 downregulated (blue) genes post-C/T (log2-foldchange was used as the ranking metric). For each gene, log2-foldchange in transcripts per kilobase million (TPM) values was computed per patient, comparing pre-C/T versus post-C/T expression values (pre-C/T as reference). Each column corresponds to a distinct patient.

Figure 3

Chemotherapy impacted pathways related to cancer, cell cycle, and immune response in HGSOC patients. (A) A dot-plot showing the significantly enriched (adjusted p-value < 0.05) gene ontology (GO) biological processes identified with the iPathwayGuide software. pv_weight = adjusted p-values obtained using the Smallest Common Denominator pruning method for correcting for multiple comparisons. Enrichment Factor (%) = 100*(number of genes differentially expressed in a GO term)/(number of all genes on the GO term). Highest % enrichment is colored red with lowest enrichment colored green according to the color scale. DE_GeneNumber = number of differentially expressed genes in the GO term. (B) Differentially expressed genes in the Oxygen Transport GO term. Red color indicates gene upregulation. (C) A dot-plot showing the significantly impacted KEGG pathways identified with the iPathwayGuide software. pORA_fdr = FDR-adjusted p-values obtained from pathway overrepresentation analysis. pv_fdr = FDR-adjusted global p-values obtained from combining pORA and pathway perturbation p-values. The most significant global pv_fdr is colored red with the least significant global pv_fdr colored green according to the color scale. DE_GeneNumber = number of differentially expressed genes in the KEGG pathway term. (D) Principal Component Analysis using top 30 upregulated and top 30 downregulated genes shows partial separation of pre-C/T samples from post-C/T samples.

Figure 4

Ovarian cancer patients with delayed recurrence, but not those with early recurrence, demonstrated significant gene expression changes post-C/T. (A) A volcano plot showing significantly upregulated (red) or downregulated (blue) genes (FDR < 0.05 and |log2-foldchange| > 0.6) post-C/T in Late Recurrence patients. Genes with FDR < 0.01 or |log2-foldchange| > 3 are labeled with HGNC symbols. (B) A volcano plot showing that there was no significantly upregulated or downregulated gene (FDR < 0.05 and |log2-foldchange| > 0.6) post-C/T in Early Recurrence patients.

Figure 5

Ovarian cancer patients with delayed recurrence showed uniquely upregulated and uniquely downregulated genes post-C/T. (A) A heatmap showing the top 30 upregulated (red) and top 30 downregulated (blue) genes post-C/T in Late Recurrence patients (log2-foldchange was used as the ranking metric). For each gene, log2-foldchange in transcripts per kilobase million (TPM) values was computed per patient, comparing pre-C/T versus post-C/T expression values (pre-C/T as reference). Each column corresponds to a distinct patient. (B) A Venn diagram comparing upregulated genes from the initial Global differential gene expression analysis with upregulated genes from the Late Recurrence differential gene expression analysis. 93 genes (in green square) were uniquely upregulated in Late Recurrence patients post-C/T. (C) A Venn diagram comparing downregulated genes from the initial Global differential gene expression analysis with downregulated genes from the Late Recurrence differential gene expression analysis. 81 genes (in green square) were uniquely downregulated in Late Recurrence patients post-C/T.

Figure 6

Genes uniquely upregulated post-C/T in Late Recurrence patients were associated with immunological response while uniquely downregulated genes involved cell cycle, DNA repair and angiogenesis. (A) Boxplots of all 93 uniquely upregulated genes (left facet) and top 10 uniquely upregulated genes (right facet). Log2-foldchanges post-C/T are shown separately for Late Recurrence (blue) and Early Recurrence (red) patients. For left facet (i.e., all 93 genes), the median log2-foldchange per gene was plotted as an individual dot. For right facet (i.e., top 10 genes), the individual log2-foldchange per patient was plotted as an individual dot. The p-value shown was obtained from Wilcoxon Rank Sum Test. (B) Boxplots of all 81 uniquely downregulated genes (left facet) and top 10 uniquely downregulated genes (right facet). Log2-foldchanges post-C/T are shown separately for Late Recurrence (blue) and Early Recurrence (red) patients. For left facet (i.e., all 81 genes), the median log2-foldchange per gene was plotted as an individual dot. For right facet (i.e., top 10 genes), the individual log2-foldchange per patient was plotted as an individual dot. The p-value shown was obtained from Wilcoxon Rank Sum Test. ****P < 0.0001.

Figure 7

C/T induced immunological modifications in late recurrence ovarian cancer patients. (A) Circos plot of significantly impacted KEGG pathways in Late Recurrence patients identified using the iPathwayGuide software. Significantly impacted pathways were those with FDR-adjusted global p-values (i.e., combined pORA and pPerturbation) < 0.05. Pathways are color-coded and shown in the right half of the plot (legend shown beneath the plot). All differentially expressed genes within the impacted pathways are shown in the left half of the plot along with their corresponding log2-foldchanges shown in adjacent boxes, color-coded according to the logFC color scale beneath the plot. Negative logFC (blue) indicates downregulation and positive logFC (red) indicates upregulation. (B) Differentially expressed genes in the Complement and Coagulation Cascade Pathway. Red color indicates gene upregulation.

Figure 8

Ovarian cancer patients with delayed recurrence showed enrichment of positive regulation of T cell differentiation post-C/T. (A) A dot-plot showing the significantly enriched (adjusted p-value < 0.05) gene ontology (GO) biological processes identified with the iPathwayGuide software. pv_weight = adjusted p-values obtained using the Smallest Common Denominator pruning method for correcting for multiple comparisons. Enrichment Factor (%) = 100*(number of genes differentially expressed in a GO term)/(number of all genes on the GO term). Highest % enrichment is colored red with lowest % enrichment colored green according to the color scale shown. DE_GeneNumber = number of differentially expressed genes in the GO term. (B) Differentially expressed genes in the Positive Regulation of T cell Differentiation GO term. Red color indicates gene upregulation and blue color indicates gene downregulation. (C) STRING network diagram of IL1B as an upstream regulator predicted as activated due to the shown IL1B targets being upregulated (red fill). Red arrows indicate activation, blue lines with perpendicular blocked ends indicate expression, and grey lines with perpendicular blocked ends indicate inhibition. (D) STRING network diagram of STAT3 as an upstream regulator predicted as activated due to the shown STAT3 targets being upregulated (red fill). Red arrows indicate activation, blue lines with perpendicular blocked ends indicate expression, and grey lines with perpendicular blocked ends indicate inhibition.

Figure 9

Global results obtained from immune deconvolution of RNA-seq data from HGSOC patients. Stacked bar-plots showing estimated fractions of 22 immune cell types determined using CIBERSORTx’s in-built LM22 signature matrix. Early Recurrence patients are labelled in red while Late Recurrence patients are labelled in black.

Figure 10

There were no significant differences pre-C/T in immune infiltration of tumors from late recurrence versus early recurrence patients. (A) Boxplots comparing pre-C/T estimated cell fractions of T cell and NK cell subtypes between Late Recurrence patients (n=13) and Early Recurrence patients (n=11). P-values shown on plots were calculated from Wilcoxon Rank Sum Test and corrected for multiple comparison using the Benjamini-Hochberg method. (B) Boxplots comparing pre-C/T estimated cell fractions of Macrophage subtypes between Late Recurrence patients (n=13) and Early Recurrence patients (n=11). P-values shown on plots were calculated from Wilcoxon Rank Sum Test and corrected for multiple comparison using the Benjamini-Hochberg method. ^ns P > 0.05; ns, not significant.

Figure 11

Delayed recurrence of ovarian cancer was associated with significant decrease in T regulatory cells in the tumor microenvironment post-C/T. (A) Boxplots comparing estimated cell fractions of T cell and NK cell subtypes between pre-C/T (n=13) versus post-C/T (n=13) samples from Late Recurrence patients. P-values shown on plots are calculated from Wilcoxon Signed Rank Test and corrected for multiple comparison using the Benjamini-Hochberg method. (B) Boxplots comparing estimated cell fractions of T cell and NK cell subtypes between pre-C/T (n=11) versus post-C/T (n=11) samples from Early Recurrence patients. P-values shown on plots are calculated from Wilcoxon Signed Rank Test and corrected for multiple comparison using the Benjamini-Hochberg method. ^* P < 0.05, ^ns P > 0.05; ns, not significant.

Figure 12

C/T did not induce statistically significant shift in macrophage subtype in Late Recurrence or Early Recurrence patients. (A) Boxplots comparing estimated cell fractions of Macrophage subtypes between pre-C/T (n=13) versus post-C/T (n=13) samples from Late Recurrence patients. P-values shown on plots are calculated from Wilcoxon Signed Rank Test and corrected for multiple comparison using the Benjamini-Hochberg method. (B) Boxplots comparing estimated cell fractions of Macrophage subtypes between pre-C/T (n=11) versus post-C/T (n=11) samples from Early Recurrence patients. P-values shown on plots are calculated from Wilcoxon Signed Rank Test and corrected for multiple comparison using the Benjamini-Hochberg method. ^ns P > 0.05; ns, not significant.

Figure 13

Validation of immune deconvolution results in an independent patient cohort confirms post-C/T decrease of Treg fraction in Late Recurrence patients. (A) Boxplots comparing pre-C/T estimated cell fractions of T cell and Macrophage subtypes between Late Recurrence patients (n=10) and Early Recurrence patients (n=8). P-values shown on plots are calculated from Wilcoxon Rank Sum Test and corrected for multiple comparison using the Benjamini-Hochberg method. (B) Boxplots comparing estimated cell fractions of T cell and Macrophage subtypes between pre-C/T (n=10) versus post-C/T (n=10) samples from Late Recurrence patients. P-values shown on plots are calculated from Wilcoxon Signed Rank Test and corrected for multiple comparison using the Benjamini-Hochberg method. (C) Boxplots comparing estimated cell fractions of T cell and Macrophage subtypes between pre-C/T (n=8) versus post-C/T (n=8) samples from Early Recurrence patients. P-values shown on plots are calculated from Wilcoxon Signed Rank Test and corrected for multiple comparison using the Benjamini-Hochberg method. ^* P < 0.05, ^ns P > 0.05.

Figure 14

Proposed model of the differential response to Carboplatin plus Taxol in HGSOC patients and its association with different recurrence profiles. In Late Recurrence patients, treatment with Carboplatin and Taxol induces significant gene expression changes and immunological modifications, turning the cold ovarian tumor hot. In Early Recurrence patients, treatment with Carboplatin and Taxol does not induce significant gene expression changes or immunological modifications; hence the tumor remains immunologically cold.