Background: Immune activation by immune checkpoint inhibitors (ICI) has shown promise in the treatment of diverse cancer types at the cost of different immune-related adverse events (irAEs). While specific microbial signatures have been associated with both ICI efficacy and irAEs, in this study we aimed to investigate whether common pre-treatment microbiome-driven pathways contribute to both ICI efficacy and irAEs.

Methods: Stool and plasma samples were prospectively collected from patients with varied cancers prior to starting ICI therapy (including both CTLA4 and PD1/PDL1 blockers) and submitted for metagenomics, metabolomics and immune protein analysis as outlined in figure legends. Prospectively collected clinical data was used to assess outcomes (RECIST) and adverse events.

Results: Among the 183 patients, 57 patients had complete response. The most common indication was melanoma, while the most prevalent irAE’s were colitis (n=19) and hepatitis (n=24). There were no significant differences in the clinical/demographic factors among patients with or without favorable outcomes, colitis or hepatitis (Figure 1a-c).

A significant difference in microbial community structure (GUniFrac, p<0.05) was seen when comparing our cancer cohort (n=183) with a pre-defined cohort of healthy controls (n=200), however, this difference was not associated with specific treatment outcomes or irAEs. The development of an irAE was not associated to outcomes of ICI therapy (equality of proportions, p=0.7). There was also no overlap among patients with different irAEs like colitis and hepatitis. This was also reflected in the microbiome where distinct microbial taxa were associated with the irAE’s and outcomes (Figure 1d) suggesting different microbes might drive immune processes underlying these phenomena. Disregarding irAEs, Collinsella aerofaciens is associated with a favorable treatment outcome, as also previously reported. Meanwhile, Lachnospira rogosae A and UBA5416 sp900539175 were linked to a positive outcome in the absence of irAEs.

Next, we integrated additional omics layers and found 6 immune markers (e.g. IL8, TNF), eight species, and two metabolites (e.g. succinate) were positively associated, while 1 immune marker and 31 metabolites (e.g. orotic and indolelactic acid) were negatively associated with ICI-colitis (Figure 2) with orotic acid exhibiting the strongest effect size. Integration of omics data using a partial least squares regression framework identified potential metabolites and immune markers driven by specific microbes (e.g., negative correlation of Blautia spp and TNF; Figure 2).

Conclusion: Our findings identify distinct microbial pathways pre-treatment that underlie ICI outcomes and irAEs which can be leveraged to stratify available treatments as well as develop microbiome-based therapies to optimize treatment outcomes and reduce irAE’s.

<b>Figure 1 Distinct microbial markers of ICI therapy and adverse events a-c</b> Clinical and demographic characteristics of patients with (RECIST 4) and without (RECIST 1,2,3) favorable outcome or with and without colitis or hepatitis. <b>d.</b> Heatmap showing microbial taxa significantly associated with ICI outcomes or irAE (colitis). Stool samples were sequenced using shotgun metagenomics (Illumina, 2 x 150 bp). Taxonomy was mapped using Kraken2 with the GTDB R202 database (median 5.9M reads annotated).

Figure 1 Distinct microbial markers of ICI therapy and adverse events a-c Clinical and demographic characteristics of patients with (RECIST 4) and without (RECIST 1,2,3) favorable outcome or with and without colitis or hepatitis. d. Heatmap showing microbial taxa significantly associated with ICI outcomes or irAE (colitis). Stool samples were sequenced using shotgun metagenomics (Illumina, 2 x 150 bp). Taxonomy was mapped using Kraken2 with the GTDB R202 database (median 5.9M reads annotated).

<b>Figure 2. Interactions among Omics features associated with ICI outcomes and colitis: </b>Select omics features are shown chosen based on prevalence and/or association with ICI colitis or outcomes. Interactions between omics features are from a multiblock(s) PLS model (mixOmics DIABLO) and reveal possible biologically relevant interactions, e.g negative association of TNF plasma cytokine with <i>Blautia</i> spp. and branched chain amino acids valine and isoleucine in pre-treatment stool samples. Quantitative metabolomics was performed (multiple LC-MS based methods, TMIC MEGA, The Metabolomics Innovation Center) on 172 stool samples; 618 metabolites were quantified, of which 202 metabolites were detected in more than 25% of samples. A panel of 96 inflammatory markers was assessed in plasma for 151 samples (Olink inflammation panel).

Figure 2. Interactions among Omics features associated with ICI outcomes and colitis: Select omics features are shown chosen based on prevalence and/or association with ICI colitis or outcomes. Interactions between omics features are from a multiblock(s) PLS model (mixOmics DIABLO) and reveal possible biologically relevant interactions, e.g negative association of TNF plasma cytokine with Blautia spp. and branched chain amino acids valine and isoleucine in pre-treatment stool samples. Quantitative metabolomics was performed (multiple LC-MS based methods, TMIC MEGA, The Metabolomics Innovation Center) on 172 stool samples; 618 metabolites were quantified, of which 202 metabolites were detected in more than 25% of samples. A panel of 96 inflammatory markers was assessed in plasma for 151 samples (Olink inflammation panel).

Guilherme Piovezani Ramos

Mayo Clinic

Appearances