Gut microbiome is associated with recurrence-free survival in patients with resected Stage IIIB-D or Stage IV melanoma treated with immune checkpoint inhibitors

Abstract

The gut microbiome (GMB) has been associated with outcomes of immune checkpoint blockade therapy in melanoma, but there is limited consensus on the specific taxa involved, particularly across different geographic regions. We analyzed pre-treatment stool samples from 674 melanoma patients participating in a phase-III trial of adjuvant nivolumab plus ipilimumab versus nivolumab, across three continents and five regions. Longitudinal analysis revealed that GMB was largely unchanged following treatment, offering promise for lasting GMB-based interventions. In region-specific and cross-region meta-analyses, we identified pre-treatment taxonomic markers associated with recurrence, including Eubacterium, Ruminococcus, Firmicutes, and Clostridium. Recurrence prediction by these markers was best achieved across regions by matching participants on GMB compositional similarity between the intra-regional discovery and external validation sets. AUCs for prediction ranged from 0.83-0.94 (depending on the initial discovery region) for patients closely matched on GMB composition (e.g., JSD ≤0.11). This evidence indicates that taxonomic markers for prediction of recurrence are generalizable across regions, for individuals of similar GMB composition.

Figure 1.. Beta Diversity, Regional Variation and Melanoma Recurrence

(A) illustrates a PERMANOVA analysis of essential clinical and demographic variables within our study, using JSD distance. Color indicates the analysis type: crude in blue and adjusted (adjustment for each other variable) in orange. The x-axis denotes R², reflecting the proportion of overall gut microbiota composition variance, with stars adjacent to the bars indicating significance (p-values: 0.05 *, 0.01 **, 0.001 ***, >0.05 NS). Panel (B) presents a map of the geographic regions, with paired PERMANOVA results for each geographic pair displayed on the plotted curves. All pairs exhibited significant differences. Donut charts plotted over each geographic region represent the top 10 genera across (based on abundance) all samples within that region. Panels (C-G) depict the principal coordinate analysis (PCoA) for each of the five geographic regions, considering recurrence as the outcome (R² and p-values are provided for each region).

Figure 2.. Region Stratified Analysis of Individual GMB Taxa and Melanoma Recurrence

(A) shows the region-stratified analysis as a forest plot with each point and associated confidence interval colored by the geographic region in which the identified gut microbiome markers prospectively associated with melanoma recurrence. All strains shown are significant after adjustment for multiple testing (FDR<0.05), with effects adjusted for participant age, sex, tumor stage, BRAF mutation and study arm. (B-K) show the meta-analysis of region-specific gut microbiome markers associated with recurrence in melanoma patients across geographic regions. Each panel shows analysis of a specific microbiome strain by region and, in meta-analyses, for the full cohort and for the full cohort minus the discovery region in (A) (N.A. stands for North America in the exclusion line). Meta analyses were performed using random-effect meta-analysis models (Schwarzer, 2007). Bacteria from (A) that are not significant in meta-analysis (Aeromonas salmonicida, Clostridiales bacterium 1–7-47 FAA and Peptostreptococcus anaerobius) are not shown in (B-K).

Figure 3.. Association of Functional Pathways with Recurrence Biomarkers.

(A) depicts the correlation between z-score normalized KEGG Level 3 bacterial pathways (presented in rows) of the recurrence-associated taxa (displayed in columns). The color of the column strip indicates the direction of association of the bacteria with recurrence (red for positive association and green for negative). The color of the row strip indicates the category (KEGG Level 2) of the functional pathway. The color of each cell represents the level of correlation. Functions are included only if they have a correlation FDR<0.0001 for a minimum of five recurrence-associated taxa. (B) exclusively displays Carbohydrate metabolism pathways derived from (A). (C) presents carbohydrate-associated pathways significantly correlated with recurrence in the North American region (no significant correlations in other regions), accounting for factors: age, sex, tumor stage, BRAF mutation and study arm. Y-axis of figure C represents the normalized z-score of the pathways.

Figure 4.. Recurrence Risk Prediction Models in Patients Using Independent Cross-regional Replicates Matched on GMB

Panel A depicts the patient matching method employed to generalize markers (i.e. using the region specific markers in other geographic areas for patients with the “same” GMB). JSD was used to match patients across region (testing patients are always from a different region from training patients), and subsequently, the predictive power of biomarkers was evaluated in the subsequent panels with adjustment for age, sex, tumor stage, BRAF mutation and study arm. Panel B shows the relationship between prediction measured using AUC vs. increasing JSD distance (spearman correlation = −0.85, p<0.001). For each point (200 total) a non-North American patient is matched to a North American subject at each JSD threshold and the final independently matched set is modeled using the North America markers to obtain AUC. Panel C, D and E show the same analysis, but using the ROW, Eastern Europe and Western Europe as discovery sets respectively. Panel F shows the comparison between the original JSD beta-diversity as well as Bray-Curtis dissimilarity, Jaccard index, and Aitchison dissimilarity with respect to AUC predictive power. Metrics were standardized by setting the lower limit to the median intra-sample distance and the upper limit to the median inter-sample distance, with 200 equal intervals for testing.

Figure 5.. GMB Stability Across Time

Figure shows the bacterial β-diversity measured using Jensen Shannon divergence between measured visits (intra-patient variation) as well as between all unpaired samples for reference (inter-patient variation). Overall GMB was largely unchanged across baseline, week 7 and week 29 measurements with global PERMANOVA R² = 0.867, p-val<0.001. Comparison between baseline and week 7 samples had an R² = 0.930, p-val<0.001; week 7 vs. week 29 R² = 0.900, p-val<0.001.