Integrative multi-omics analysis uncovers tumor-immune-gut axis influencing immunotherapy outcomes in ovarian cancer

Abstract

Recurrent ovarian cancer patients, especially those resistant to platinum, lack effective curative treatments. To address this, we conducted a phase 2 clinical trial (NCT02853318) combining pembrolizumab with bevacizumab, to increase T cell infiltration into the tumor, and oral cyclophosphamide, to reduce the number of regulatory T cells. The trial accrued 40 heavily pretreated recurrent ovarian cancer patients. The primary endpoint, progression free survival, was extended to a median of 10.2 months. The secondary endpoints demonstrated an objective response rate of 47.5%, and disease control in 30% of patients for over a year while maintaining a good quality of life. We performed comprehensive molecular, immune, microbiome, and metabolic profiling on samples of trial patients. Here, we show increased T and B cell clusters and distinct microbial patterns with amino acid and lipid metabolism are linked to exceptional clinical responses. This study suggests the immune milieu and host-microbiome can be leveraged to improve antitumor response in future immunotherapy trials.

Fig. 1. Clinical cohort overview.

Recurrent ovarian cancer patients were previously enrolled in a single-arm phase 2 clinical trial combining pembrolizumab, bevacizumab, and oral metronomic cyclophosphamide therapy. Baseline (upon enrollment) and On-Treatment (Cycle 4, Day 1; C4D1) assessments, including tumor biopsies, blood samples, and stool collections, were obtained. The study diagram provides a thorough outline of the primary, secondary, and exploratory endpoints for the trial.

Fig. 2. Analysis of tumor transcriptomes indicates patients with durable clinical benfit exhibit greater immune activation compared to patients with limited clinical benefit.

a Through Gene Set Enrichment Analysis (GSEA) of tumor transcriptome data, immune population changes are found to be enriched in durable clinical benefit (DCB) groups, with this enrichment further amplified during treatment (On.TX) compared to limited clinical benefit (LCB) groups. Additionally, MYC-associated and neuronal signatures are more prevalent in LCB groups compared to DCB groups, irrespective of the timepoint (Baseline or On.TX). b Quantification of immune-specific signatures based on GSEA, demonstrates increases in immune-associated transcripts in DCB, as compared to LCB populations, including those associated with T cells, B cells, CD40 signaling, and tertiary lymphoid structures (TLS). Further, this corresponds with (c) enrichment of MCP Counter immune deconvolution signatures in DCB patient populations compared to LCB patient populations at baseline, and further exacerbated by treatment. d This is further exemplified in comparing MCP counter cytotoxicity, T cell, B cell, and myeloid dendritic cell (DC) signatures. *p < 0.05, **p < 0.01, two-sided Mann–Whitney U-Test. Exact p-values can be found in Data Table 1.

Fig. 3. Analysis of digital spatial profiling within tumor and stromal regions indicates patients with durable clinical benefit exhibit enhanced immune trafficking and activation compared to patients with limited clinical benefit.

a Representative images from two DCB patients showing large CD3+ aggregates forming while On.TX. b Heatmaps representing all markers detected in the DSP assay in both tumor and stromal compartments. Tumoral immune marker abundances are increased with treatment in DCB (On.TX_DCB, red), as compared to both baseline DCB (Baseline_DCB, orange) and LCB baseline and on treatment (Baseline_LCB, light blue and On.TX_LCB, dark blue; respectively) patient populations. c Volcano plots demonstrate differential expression of these markers between our Baseline and On.TX DCB and LCB patient populations, highlighting interesting biology like the appearance of Ki67 enrichment in Baseline_DCB, as compared to On.TX_DCB. d Quantifying immune and proliferative markers in the DSP data demonstrated significant differences in stromal and tumor fractions between important markers like PD-L1 and CD20 between On.TX_DCB and On.TX_LCB, suggesting altered immune phenotypes. *p < 0.05, **p < 0.01, Mann–Whitney U-Test. e TLS scores are significantly elevated in DCB patients On.TX relative to baseline DCB and On.TX LCB. *p < 0.05, **p < 0.01, ***p < 0.001; derived from linear regression of log-normalized abundances, by clinical group, correcting for the patient in paired analyses when appropriate, and multiple test corrections.

Fig. 4. Gut microbiome diversity and metabolic profile associated with durable responses to immunotherapy.

a Top 20 differentially abundant microbes (adj. p. value < 0.05, |LogFC|>1.5) between Baseline_DCB and Baseline_LCB displayed in a heatmap. Similar to tumor transcriptome, heatmaps demonstrate that across all microbes, patients with durable clinical benefit express higher microbe levels, and this expression is increased with treatment in DCB (On.TX_DCB, red), as compared to baseline DCB (Baseline_DCB, orange), baseline LCB (Baseline_LCB, light blue), and on treatment LCB (On.TX_LCB, dark blue) patient populations. b Violin plots quantify OTU abundance in all four groups for select bacterial species, all of which demonstrate higher levels of the bacterial species at baseline in DCB (Baseline_DCB, orange) as compared to LCB (Baseline_LCB, light blue), and more so with treatment in the DCB group (On.TX_DCB, dark red). *p < 0.05, **p < 0.01; derived from linear regression of log-normalized abundances, by clinical group, with multiple test corrections. c Volcano plots demonstrate variety of significant (|logFC|>1.5, adjusted p < 0.1) differentially abundant metabolites between Baseline LCB (Baseline_LCB, light blue), Baseline DCB (Baseline_DCB, orange), On-Treatment LCB (On.TX_LCB, dark blue), and On-Treatment DCB (On.TX_DCB, red) patient populations. d Expression levels of the 20 most highly differentially expressed metabolites separate Baseline_DCB and Baseline_LCB fecal samples to a high degree based on Euclidean distances, across all 4 groups. Some metabolites were decreased with treatment in DCB and LCB (indole), some were solely increased in DCB with treatment (C3-OH, TMAO, Glu, and Ser), and some lipids were increased more so in LCB, regardless of timepoint (TGs and DGs). *p < 0.05, derived from linear regression of log-normalized abundances, by clinical group. Exact p-values can be found in Data Tables 2 and 3.

Fig. 5. Integration of bacterial species, metabolites, and immune populations identify points of convergence.

a Tumor transcriptomic data was utilized for metabolic pipeline assessment to understand which metabolic pathways were transcriptionally dysregulated significantly between the baseline DCB (orange) and LCB (light blue). Pathways like glycerophospholipid and tryptophan metabolism were among the most highly dysregulated pathways at both baseline (top) and on-treatment (bottom). b Significant differentially abundant metabolites from the same baseline comparison were used for Metabolite Set Enrichment Analysis (MSEA), highlighting altered glycerophospholipid metabolism (top left). Further, the differentially abundant OTUs from the baseline DCB vs. LCB analysis were functionally enriched to reveal alterations to similar pathways, like tryptophan/amino acid metabolism, glycerophospholipids, and fatty acids (top right). c Significantly dysregulated pathways identified between the tumor RNA-sequencing, stool metabolomics, and stool microbiome study overlapped, revealing 3 altered pathways common to all analyses. d Correlation network limited to metabolites (green rectangle), microbes (orange ellipses), and immune scores (purple diamond) nodes highly correlated with CD8 + T Cells, B cells, or Macrophages/Monocytes (R² > 0.39 (red), R² < −0.39 (blue), width of connectors increase with increased correlation value). Patient phenotype is designated by node borders with orange indicating the immune signature/microbe/metabolite was more highly expressed in DCB or blue for LCB. e Model demonstrating microbial (microbiome and metabolomic) and immune changes that underlie response to ICB in both DCB with DCB (top, orange) and LCB (bottom, blue) with no clinical response.