Fecal microbiota transplant overcomes resistance to anti-PD-1 therapy in melanoma patients

Abstract

Anti-programmed cell death protein 1 (PD-1) therapy provides long-term clinical benefits to patients with advanced melanoma. The composition of the gut microbiota correlates with anti-PD-1 efficacy in preclinical models and cancer patients. To investigate whether resistance to anti-PD-1 can be overcome by changing the gut microbiota, this clinical trial evaluated the safety and efficacy of responder-derived fecal microbiota transplantation (FMT) together with anti-PD-1 in patients with PD-1-refractory melanoma. This combination was well tolerated, provided clinical benefit in 6 of 15 patients, and induced rapid and durable microbiota perturbation. Responders exhibited increased abundance of taxa that were previously shown to be associated with response to anti-PD-1, increased CD8⁺ T cell activation, and decreased frequency of interleukin-8-expressing myeloid cells. Responders had distinct proteomic and metabolomic signatures, and transkingdom network analyses confirmed that the gut microbiome regulated these changes. Collectively, our findings show that FMT and anti-PD-1 changed the gut microbiome and reprogrammed the tumor microenvironment to overcome resistance to anti-PD-1 in a subset of PD-1 advanced melanoma.

Trial registration: ClinicalTrials.gov NCT03341143.

Fig. 1.. Radiographic responses from a phase 2 study of anti–PD-1 responder–derived FMT and pembrolizumab in PD-1–refractory melanoma.

Melanoma patients who had primary refractory disease to anti–PD-1 therapy received FMT derived from individual melanoma patients with durable OR to anti–PD-1 therapy. FMT was administered colonoscopically on day 0 along with pembrolizumab (200 mg). Pembrolizumab was repeated every 3 weeks. Restaging scans were performed at weeks 9 to 12 and repeated every 9 to 12 weeks while in the study. Patients remained in the study until intolerable toxicity, RECIST v1.1–confirmed disease progression, or completion of 35 cycles of pembrolizumab. (A and B) Treatment exposure and response duration by RECIST v1.1 (investigator assessed; n = 15). (A) FMT donor and best response to prior line(s) of anti–PD-(L)1 therapy singly or in combination are shown for each FMT-recipient patient. The length of each bar corresponds to the duration of time that patients received treatment (in weeks). Response status is color coded (R, blue; NR, red). Response symbols represent status at first restaging scan (9 to 12 weeks) and at most recent review. Patients with ongoing response in the study are depicted with horizontal arrows. (B) Radiographic change of tumor burden from baseline (investigator assessed per RECIST v1.1; n = 15). One patient (PT-18–0018) had initial disease stability with subsequent progression after antibiotic therapy and was offered a retransplant with the same donor, with subsequent disease stabilization. (C) Representative CT scans from one responding patient. CT scans from patient PT-19–0024 at four separate time points depict initial tumor growth after FMT followed by eventual PR. L, left; LN, lymph node.

Fig. 2.. Microbiome analyses before and after FMT in melanoma patients.

(A) Dimensionality reduction using t-distributed UMAP (t-UMAP) plot of microbial taxa abundances by last known taxon of FMT recipients at different time points. Each color corresponds to a different FMT-treated patient. Pre-FMT stool samples are depicted as squares and post-FMT stool samples as triangles. Ellipses encapsulate each recipient’s pre- and post-FMT samples, and the size of the ellipse spans two standard deviations from the centroid. Rs and NRs are distinguished by solid and dashed lines, respectively. PT-19–0026 (PD) is not depicted because the patient had a single post-FMT sample. (B) Intrapatient variance of stool samples from donors and recipients after standardization and dimensionality reduction. Donors (n = 3) and recipients (n = 15) who contributed at least three fecal samples are depicted. Data were standardized, PCA was performed, PC loading was computed, and variances of patients for every PC loading were calculated as the squared standard deviation divided by the mean and were multiplied by the PC variance contribution. Resultant values were added together to produce a combined variance number, which was compared between donors (n = 3) and recipients (n = 15) by using the nonparametric t test. AU, arbitrary units. (C) Rate of taxonomic change of stool samples sequentially obtained from treated patients. The rate of taxonomic change for each sample sequentially obtained from each patient was calculated by using speed of traversion (Euclidean distances traversed per day), which was calculated by dividing total Euclidean distance traveled by days. Euclidean distance was calculated by using log-transformed normalized taxonomic data from shotgun sequencing between Rs (n = 6) and NRs (n = 9) by using the nonparametric t test. (D) Plot of Euclidean distance over time from patients’ gut microbiota to corresponding FMT donor’s microbiota. To assess the efficiency of FMT engraftment, Euclidean fitted curves were generated by using points on the graph in both NRs (red, above) and Rs (blue, below). A positive percentage of each curve indicates similarity to the corresponding donor, whereas a negative percentage indicates changes in the recipient microbiota even further from the donor microbiota. Graphs were normalized by truncating at 60 days in both Rs and NRs. The chi-square test was calculated by using an even distribution as null hypothesis. (E) Recipient IgG response against donor microbiota induced by FMT. Donor fecal bacteria were incubated with recipient sera at 1:200 dilutions, washed and stained with phycoerythrin-labeled antihuman IgG, and fixed and analyzed on a flow cytometer. Change in IgG positivity of donor fecal bacteria for each FMT recipient was calculated in relation to the first FMT recipient serum sample available compared with the subsequent two time points up to 50 days later. Percent IgG-positive donor fecal bacteria was assessed, and area under the curve (AUC) for percent IgG-positive donor bacteria was calculated, adjusting all recipient time points relative to the baseline time point set to zero. Difference in reactivity between sera from Rs and NRs was evaluated by Student’s t test. (F) Cladogram visualization of phylogenetic distribution of differentially abundant taxa before and after FMT in responding patients. Fisher’s method for meta-analyses was used to validate statistical significance and calculate effect size of the differential abundances of taxa in Rs (fig. S10). Differentially abundant taxa are color coded on the basis of relative abundance in post-compared with pretransplant samples (green, higher; yellow, unchanged; red, lower). The most significantly associated taxa are highlighted at the family level.

Fig. 3.. Single-cell analyses of circulating and tumor-infiltrating immune cells.

(A) Unsupervised multiparameter flow cytometry analysis of circulating immune cells. UMAP visualization of 100,000 live single cells from Rs and NRs at three time points—pretreatment (D0), day 21 (D21), and D42—from 30-parameter flow panel analysis (n = 14) after merging clusters on the basis of expression of CD3, CD4, CD8, CD19, CD14, CD56, Tγδ1, and Tγδ2. Myeloid cells were identified as lineage-negative cell clusters on the basis of presence or absence of CD14⁺ cells. (B) Frequency of CD56⁺CD8⁺ clustered T cells in PBMCs of patients. Whisker boxes show frequencies of CD56⁺CD8⁺ clustered T cells in PBMCs between Rs and NRs at D0, D21, and D42. We observed a significant increase of CD56⁺CD8⁺ T cells in Rs at D42 using the unpaired t test. *p < 0.05. (C and D) Phenotypic analysis of circulating CD8⁺ T cells. Whisker boxes show markers that are significantly differentially expressed (normalized mean fluorescence intensity) in CD8⁺ T cells (C) and CCR7⁺CD45RA⁺-naïve, TEMRA, CCR7⁺CD45RA⁻ effector memory (EM), and CCR7⁻CD45RA⁻ central memory (CM) cells (D) between Rs (n = 5 to 6) and NRs (n = 5 to 7) at the three time points. Analysis was performed on live single CD3⁺ and TCRgd⁻ T cells. In Rs, we observed up-regulation of TIGIT, Lag-3, and T-bet after treatment and down-regulation of CD27 in CD8⁺ T cells using the unpaired t test. *p < 0.05. (E) Phenotypic analysis of circulating MAIT cells. Whisker boxes comparing MAIT cells between Rs (n = 5 to 6) and NRs (n = 5 to 7) at the three time points. Analysis was performed on live single CD3⁺ and TCRgd⁻ T cells. In Rs, MAIT cells up-regulated granzyme B expression and down-regulated CD27 after treatment using the unpaired t test. *p < 0.05. (F) scRNA-seq analysis of tumor-infiltrating immune cells. UMAP projection of 64,000 CD45⁺ cells that were clustered and manually identified on the basis of their expression profile. NK, natural killer; pDC, plasmacytoid dendritic cell; T FH, T follicular helper; T FOS, T cells expressing FOS; Tregs, T regulatory cells. (G) Whisker boxes showing the abundance of myeloid cells and CD4⁺ Tregs in CD45⁺ tumor-infiltrating cells. We observed decreased abundance of myeloid cells and CD4⁺ Tregs in Rs compared with NRs using the unpaired t test. *p < 0.05. (H) Cell-associated expression of two markers (CXCL8 and SPP1) in UMAP projection. These markers are predominantly expressed in suppressive myeloid cells. (I) Volcano plots showing the differences in phenotype of CD8⁺ T cells and myeloid cells between Rs and NRs after FMT. Rs show a CD8⁺ T phenotype with increased activation markers (GZMK, class II HLA genes, CD74), whereas NRs show a myeloid phenotype with an increased suppressive signature (CXCL8 and SPP1) at day 56 after treatment. Adjusted p values were obtained by Wilcoxon rank sum test.

Fig. 4.. Serum proteomics, metabolomics, and lipidomics signatures before and after FMT.

(A) PCA and heatmap of serum cytokines of Rs and NRs before and after FMT. Data show that Rs after treatment (orange) separate from Rs before treatment (green), along with NRs before (red) and after (blue) treatment, as assessed by two-way analysis of variance (ANOVA; *p < 0.05). (B) PCA and heatmap of serum metabolites of Rs and NRs before and after FMT. Data show that Rs after treatment (orange) separate from Rs before treatment (green), along with NRs before (red) and after treatment (blue), as assessed by using two-way ANOVA (q < 0.05). (C) PCA and heatmap of serum lipidomics of Rs and NRs before and after FMT. Serum lipidomic analyses show that Rs after treatment (orange) distinctly clustered separately from Rs before treatment (green), along with NRs before (red) and after (blue) treatment, as assessed by using two-way ANOVA (q < 0.05). (D) Transkingdom network analysis of multi-omic data. Data for microbial (octagons), metabolites (squares), cytokines (triangles), and multiparameter flow cytometry (hearts) were analyzed to identify highly differentially abundant elements between Rs and NRs to FMT and pembrolizumab. To identify nodes (i.e., any of these four types of elements) and their groups with potential contribution to a regulatory activity, a “transkingdom” network integrating -omics data was constructed by using their correlations within the different groups (Rs or NRs, before or after FMT and pembrolizumab). Network interrogation revealed that “microbiome” and “metabolite” as well as “microbiome” and “cytokine” subnetworks were the most interconnected. We identified a dense subnetwork (module) containing the highest number of nodes from different -omics data (nodes highlighted in yellow, positively correlated edges in red, and negatively correlated edges in blue). (E) Subnetwork identified in (D). Network analyses established that CXCL8 (IL-8), IL-10, and CCL3 (MIP-1α) were positively correlated with organisms enriched in NRs before treatment (Bacteroides uniformis, Bacteroides nordii, Phascolarctobacterium faecium, etc.) and negatively correlated with organisms enriched in Rs after treatment (e.g., Ruminococcus flavefaciens and F. prausnitzii).