The Gut Microbiome Associates with Immune Checkpoint Inhibition Outcomes in Patients with Advanced Non–Small Cell Lung Cancer

The therapeutic landscape of advanced non–small cell lung cancer (NSCLC) has been revolutionized by immune checkpoint inhibitors (ICI). Initial landmark trials compared single-agent anti–PD-1/PD-L1 mAbs to docetaxel in previously treated advanced NSCLC patients, which demonstrated improvements in overall survival (OS) with durable responses in the anti–PD-1/PD-L1 groups (1–4). As a result, the study of ICI in NSCLC rapidly expanded to first-line settings with or without platinum-doublet chemotherapy (5). Given the improved OS observed with first-line single-agent pembrolizumab in PD-L1–positive tumors (≥50%, using 22C3), this strategy is now the standard of care for this patient population (6).

Despite these unprecedented advances, disease progression with ICI therapy is often inevitable with primary resistance ranging from 35% to 44% (6–9). Predictive biomarkers of therapeutic success, based on PD-L1 expression and tumor mutational burden by genome profiling with next-generation sequencing, are inadequate given their limited sensitivity and specificity (10, 11). Finally, immune-related adverse events (irAE) remain an important therapeutic hurdle, leading to discontinuation of ICI in 7% of patients in clinical trials and up to 30% of patients in real-world settings (12).

Addressing these shortcomings in the immune-oncology landscape, the discovery that gut microbiome (GM) influences response to ICI, not only in NSCLC but also in melanoma and other tumors, illuminates the GM as a potential therapeutic target and biomarker of response. Indeed, in preclinical models, the efficacy of anti–CTLA-4 and PD-L1 antibodies require the presence of distinct Bacteroides species and Bifidobacterium, respectively (13, 14). Microbiome profiling in patients amenable to ICI identified that high bacterial GM diversity and specific commensals associate with CD8⁺ T- and CD4⁺ T-cell phenotypes and correlate with favorable response to ICI (15–18). Building on these findings, fecal microbial transplantation (FMT) from responder NSCLC patients into germ-free or antibiotic-treated specific pathogen–free mice restores the efficacy of ICI treatment in a CD4⁺ CXCR3⁺–dependent manner, whereas FMT from “nonresponder (NR)” patients abrogates ICI response (16).

Clinically, antibiotic use prior to the initiation of ICI associates with worse OS (19–21). The successful use of FMT to treat ICI-associated colitis reveals the potential role of the GM in abrogating irAE (22), demonstrating that the GM can be a therapeutic target to improve ICI and to dampen toxicity.

Notwithstanding these contributions to the understanding of local gut immunity on tumor immunosurveillance, the impact of antibiotics on the specific GM signature of patients with NSCLC remains to be fully defined. Whether GM profiling, at baseline or after antibiotic, could represent a biomarker of both response and irAE in advanced NSCLC during ICI therapy remains to be established. Here, we characterized the GM composition in a cohort of 70 patients with advanced NSCLC treated with single-agent anti–PD-1/PD-L1.

We prospectively collected fecal samples from 70 Japanese patients with advanced NSCLC receiving ICI between December 2017 and September 2019 who were treated at the Tokyo Metropolitan Cancer and Infectious Diseases Center Komagome Hospital (Bunkyo City, Tokyo, Japan). Patients who fulfilled the following criteria were eligible for this study: (i) histologically or cytologically confirmed unresectable advanced (stage III or IV) or recurrent NSCLC and (ii) treatment with ICI (nivolumab, pembrolizumab, or atezolizumab) monotherapy at recommended dose either as the first-line or secondary therapy. Written informed consent was obtained from all patients. The study protocol was approved by the Ethics Committee of the Tokyo Metropolitan Cancer and Infectious Diseases Center Komagome Hospital (Bunkyo City, Tokyo, Japan, approval number: #1744) and conducted in accordance with the tenets of the Declaration of Helsinki. The study was registered with the UMIN Clinical Trials Registry (ID: UMIN000021734).

Fecal samples at baseline (pre-ICI) were collected using a commercial sampling kit containing guanidine solution following the manufacturer's instructions (catalog no.: FS-0006, TechnoSuruga Laboratory Co., Ltd.). Fecal samples were immediately stored at 4°C and frozen at −80°C within 24 hours. Patient baseline characteristics including the use of antibiotic within 1 month before the first ICI injection were recorded.

Genomic DNA was extracted from fecal samples using NucleoSpin DNA Fecal Kit following the manufacturer's instructions (catalog no.: 740472.10, Macherey-Nagel GmbH & Co. KG) and immediately stored at −80°C. Isolated DNA was sent to TechnoSuruga Laboratory Co., Ltd. and analyzed using 16S rRNA gene sequencing to investigate the microbial composition in fecal samples. The V3–V4 hypervariable regions of the 16S rDNA were amplified using prokaryotic universal PCR primer pair Pro341F and Pro805R for the simultaneous analysis of bacteria and archaea (TechnoSuruga Laboratory Co., Ltd.). In addition to the V3–V4-specific priming regions, these primers were complementary to standard Illumina forward and reverse primers. To reduce the formation of spurious byproducts during the amplification process, the touchdown PCR methods for thermal cycling were used with a Rotor-Gene Q Quantitative Thermal Cycler (Qiagen). Sequencing was conducted using a paired-end, 2 × 250-bp cycle run on an Illumina MiSeq sequencing system and MiSeq Reagent Nano Kit version 2 (500 cycles) chemistry (23).

Gene sequence processing and analysis were performed using R v3.6.1 and GraphPad Prism v8.3.1. DADA2 (24) R package (25) v1.14.0 was used to generate exact amplicon sequence variants (ASV) of each sample from raw amplicon sequences. Sequences were corrected for Illumina amplicon sequence errors, dereplicated, chimera removed, and merged of paired‐end reads with 260‐bases for forward reads and 190-bases for reverse reads. The taxonomy assignment was performed against the SILVA reference database (v132; ref. 26). Archea and Eukaryota residual sequences were removed. Alpha diversity, defined as the number of distinguishable taxa, was analyzed at the genus level and computed with phyloseq R package (27) v1.30.0. The alpha diversity was estimated with different metrics: observed ASV, Shannon index, Inverse Simpson index, as well as weighted and unweighted Faith Phylogenetic Diversity (Faith pd, a phylogenetic index, and an analogue of taxon richness, and is expressed as the number of tree units which are found in a sample). Bray–Curtis distance (28) and weighted UniFrac distance (29) were used as beta diversity metrics (which show the difference in taxonomic abundance profiles from different samples) and visualized through NMDS method (30). Sequencing data were deposited at the SRA NCBI and are available at https://www.ncbi.nlm.nih.gov/bioproject/PRJNA606061/.

Statistical analysis was performed using R software v3.6.1. The Mann–Whitney U test was used to determine significant differences among the different groups using alpha diversity, which shows the diversity in each individual sample. Linear discriminant analysis (LDA) effect size (LEfSe; ref. 31) and DESeq2 (32) were used to perform differential abundances analysis at the genus level. The P values were corrected with the Benjamin–Hochberg procedure for the DEseq2 differential abundance analyses only. Both methods statistically enrich plausible bacterial strain to explain differences consistency and effect relevance by means of features using bioinformatics procedure. Multivariate analysis was performed by Cox proportional hazards model with a relative abundance of bacteria transformed with an arcsine-square root transform.

Clinical endpoints were OS (defined by the time of the first injection of ICI to the date of death from any cause), progression-free survival [PFS, defined by the time of the first injection of ICI to the first event (tumor progression or death from any cause)], objective response rate (ORR) based on RECIST v1.1 (33) criteria, and incidence and grade of irAE [graded by Common Terminology Criteria for Adverse Events (CTCAE) v4.0; ref. 34]. Patients with no events were censored at the date of the last follow-up. Survival curves were estimated through the Kaplan–Meier method and compared with the log-rank test (35, 36). Patients were defined as responders (favorable ORR) if they achieved complete response, partial response, or continuous stable disease for more than 6 months. Otherwise patients were defined as NRs (unfavorable ORR). For irAE, the highest grade toxicities during each therapy were recorded. To determine the effect of antibiotic on these clinical endpoints, the same analyses were performed in eubiotic patients who did not receive antibiotic. All tests performed were two-tailed. A P < 0.05 was considered statistically significant

The baseline characteristics of the 70 patients with NSCLC on ICI included in the study are presented in Supplementary Table S1. The median age in this cohort was 70 and most patients had an Eastern Cooperative Oncology Group (ECOG) performance status (PS) of ≤1. All patients received anti–PD-1/PD-L1 monotherapy with 50% of patients treated in the first-line setting. Median follow-up for this cohort was 9.7 months. Sixteen patients (23%) received antibiotic 1 month prior to ICI initiation.

First, we measured the alpha diversity for all patients in the cohort and found no difference in diversity between patients with OS >12 months and those with OS ≤12 months (Supplementary Fig. S1A). We then compared GM composition between groups according to clinical outcomes. In patients with PFS >6 months, certain bacteria were enriched, including Ruminococcaceae UCG 13 and Agathobacter (Supplementary Fig. S1B). In patients with OS >12 months, Lachnospiraceae UCG 001, a member of the Clostridiales order, was also overrepresented (Supplementary Fig. S1C).

We examined the impact of antibiotic on GM composition. Antibiotic use associated with significantly decreased alpha diversity by both Shannon (Fig. 1A) and Inverse Simpson methods (Supplementary Fig. S2). Two objective clusters were also found when analyzing beta diversity for antibiotic compared with antibiotic-free groups (Fig. 1B). Patients who did not receive antibiotic had feces that were enriched with Clostridia, specifically Ruminococcaceae UCG 13, Clostridiales, and Agathobacter (Fig. 1C), whereas feces from patients who received antibiotic were enriched in Hungatella (Fig. 1D).

Because antibiotic use significantly altered GM composition, we characterized GM composition in eubiotic patients by excluding patients who received antibiotic. In the antibiotic-free group, decreased alpha diversity associated with shorter OS (Fig. 2A). Of note, no objective clusters in the beta diversity analysis were found with regard to different outcomes. Ruminococcaceae UCG 13 and Agathobacter were overrepresented in those individuals with favorable ORR (Fig. 2B) and PFS >6 months (Fig. 2C). Ruminococcaceae UCG 13 was enriched in patients with OS >12 months (Fig. 2D). In antibiotic-free patients, Clostridiales order was also enriched in patients with OS >12 months (Fig. 2D). In addition, using DESeq2 analysis, dominance in Ruminococcaceae UCG 13 was also observed in patients with favorable ORR and PFS >6 months (Supplementary Fig. S3). Because Ruminococcaceae UCG 13 and Agathobacter were consistently enriched, we decided to focus on these particular members for subsequent analysis.

Hence, the Kaplan–Meier survival curves for patients with NSCLC whose GM contained or lacked Ruminococcaceae UCG 13 and Agathobacter significantly diverged. In antibiotic-free patients, the presence of Ruminococcaceae UCG 13 associated with longer OS [not reached vs. 4.4 months; HR, 0.24, 95% confidence interval (CI), 0.08–0.78; P < 0.001; Fig. 3A] and PFS (P = 0.04; Supplementary Fig. S4A). This was also true for Agathobacter in antibiotic-free patients for OS (not reached vs. 6.8 months; HR, 0.08; 95% CI, 0.02–0.3; P < 0.0001; Fig. 3C) and PFS (P = 0.0001; Supplementary Fig. S4B). Similarly, when analyzing all patients, the presence of Ruminococcaceae UCG 13 was associated with longer OS (not reached vs. 6.81 months; HR, 0.40; 95% CI, 0.16–0.96; P = 0.03; Fig. 3B). This association was also seen when analyzing Agathobacter for all patients for both OS (not reached vs. 5.5 months; HR, 0.13; 95% CI, 0.05–0.31; P < 0.0001) and PFS (P = 0.0002; Fig. 3D; Supplementary Fig. S4C). We then performed a multivariate analysis of the effect of Ruminococcaceae UCG 13 on OS taking into account standard prognostic factors relevant for advanced NSCLC including ECOG PS, histology type, lines of treatment, stage, and PD-L1 expression. The multivariate analysis further supported the presence of Ruminococcaceae UCG 13 associated with improved OS (P = 0.004; Supplementary Fig. S5). We performed multivariate analysis for Agathobacter as well, however, there was no statistically significant difference, P = 0.59 (Supplementary Fig. S6).

Given the possible association between the GM and irAE, we analyzed the differences in GM composition between patients who experienced clinically relevant (≥grade 2) irAE compared with those with nonsevere irAE (grade 1 or absent). In antibiotic-free patients and all patients, there was no difference in the bacterial diversity between these groups (Supplementary Fig. S7A and S7B). However, Lactobacillaceae on LEfSe analysis and Raoultella on DESeq2 analysis were enriched in feces of patients who did not experience severe irAE profiles (Fig. 4A and B). Although Akkermansia species was not associated with improved clinical outcomes in our cohort, it was associated with less severe irAE profile (Fig. 4B). Lactobacillaceae and Raoultella also associated with a less severe irAE profile when examining all patients by LEfSe and DESeq2, respectively. Despite its association with favorable outcome, Agathobacter was associated with more severe irAE profile (Supplementary Fig. S8A and S8B).

Altogether, we demonstrated the negative influence of antibiotic on GM composition and identified a differential bacteria repertoire in patients experiencing favorable clinical outcomes (specifically, Ruminococcaceae UCG 13) or low grade irAE.

In this prospective analysis of 70 patients with advanced NSCLC treated with monotherapy ICI, 16S rRNA sequencing revealed that patients who received antibiotic had lower alpha diversity at baseline and underrepresentation of Clostridiales, Ruminococcaceae UCG 13, and Agathobacter. When analyzing antibiotic-free patients, lower alpha diversity was observed in patients with lower OS. In addition, Ruminococcaceae UCG 13 and Agathobacter were overrepresented in patients with favorable ORR and PFS >6 months. Ruminococcaceae UCG 13 was enriched in those with OS >12 months. Clostridiales order was also enriched in patients with OS >12 months. Compositional GM differences were also observed between the patients who experienced clinically significant (≥grade 2) irAE; Lactobacillaceae and Raoultella were enriched in patients who had less severe irAE profile, whereas Agathobacter associated with more severe irAE profile.

Our findings corroborate the key role of the GM diversity, defined by the abundance distribution of microorganisms colonizing the gut, in predicting beneficial responses to ICI. Indeed, several investigators profiling the microbiome across tumor histology, ICI type, and geographic distribution demonstrate the association between higher baseline diversity and favorable clinical outcomes (15–18, 37). Beyond diversity metrics, specific commensals associate with improved outcomes to ICI such as objective response and survival (38). Increases in both Clostridiales and Ruminococcus in 43 patients with advanced melanoma correlate with favorable response to anti–PD-1. When analyzing systemic immune responses in patients enriched with these commensals, there were higher frequencies of effector CD4⁺ and CD8⁺ T cells in the circulation and CD8⁺ T-cell infiltration in the tumor. Similarly, Ruminococcus associates with ICI benefit in melanoma, NSCLC, and renal cell carcinoma (16, 39). For example, a mixture of 11 bacteria including Ruminococcaceae bacterium cv2 in preclinical models associated with high colon IFNγ production from CD8⁺ T cells and correlated with improved anti–PD-1 efficacy (39). The positive prognostic significance of Ruminococcaceae was also seen in patients with melanoma who underwent neoadjuvant immune checkpoint blockade (40). Our Japanese cohort of patients with advanced NSCLC validated this association between Ruminococcaceae and clinical responses to anti–PD-1/PD-L1. We illustrated the potentially suppressive role of specific bacteria, such as Lactobacillaceae, Raoultella, and Agathobacter in the development of irAE. Preclinical data investigating the anti-inflammatory properties of some Lactobacillaceae species (41) may explain its potential modulation of off-target immune responses.

Our study has several limitations. First, we did not correlate the GM composition with a dietary history. The OS in the antibiotic-free group was numerically longer compared with those who received antibiotic (16.1 vs. 12.1 months), with no statistical significance, and this may be due to a low sample size. 16S rRNA gene sequencing may be underpowered to illustrate the whole GM signature. The use of deeper GM profiling techniques such as metagenomics sequencing is now available, and despite the current financial burden of this tool, this will likely become more affordable in the future due to increasing demand. To relay the results of GM studies to the bedside, the accessibility of this technique, including its cost and efficiency, is required.

In sum, we demonstrated the negative influence of antibiotics on GM composition and identified a differential bacteria repertoire in patients experiencing favorable clinical outcomes or low-grade irAE. Our data reinforced the importance of developing diagnostic tools aimed at identifying gut dysbiosis to predict resistance or irAE in patients with advanced NSCLC treated with ICI. Many questions remain unanswered, including whether GM composition could influence cancer incidence and severity, neoplasia histology, tumor genetics, and tumor microenvironment, and whether lines of therapy, comorbidities, comedications, and geo-distribution alter GM. Further efforts to prospectively sequence the fecal samples of patients with NSCLC are ongoing to identify specific and minimalist GM signatures associated with favorable or unfavorable clinical outcomes for ICI therapy.

A. Elkrief reports grants from AstraZeneca outside the submitted work. Y. Hosomi reports personal fees from AstraZeneca, Eli Lilly Japan, Taiho Pharmaceutical, Chugai Pharmaceutical, Ono Pharmaceutical, Bristol-Myers Squibb, Kyowa Kirin, and CSL Behring outside the submitted work. L. Derosa reports other from Philantropia (PhD fellowship) and grants from Fondation Dassault outside the submitted work. L. Zitvogel is the main founder of EverImmune and reports grants and personal fees from Transgene (board of directors), and grants from Kaleido and 9 meters during the conduct of the study; personal fees from Lytix Biopharma (scientific advisory board) outside the submitted work; and patents for EP 18306282.7 licensed to EverImmune and EP 19306246.0 pending. B. Routy reports personal fees from Vedanta Company (advisory board) outside the submitted work. Y. Okuma reports grants from Grant-in-Aid for scientific research (KAKENHI) during the conduct of the study, as well as personal fees from AstraZeneca K.K, Boehringer Ingelheim Japan, Chugai Pharmaceutical Co., Ltd., Ono Pharmaceutical Co., Ltd., and Bristol-Myers Squibb outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

T. Hakozaki: Resources, data curation, investigation, writing–original draft, writing–review and editing. C. Richard: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. A. Elkrief: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. Y. Hosomi: Resources, supervision, writing–review and editing. M. Benlaïfaoui: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. I. Mimpen: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. S. Terrisse: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. L. Derosa: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. L. Zitvogel: Investigation, methodology, project administration, writing–review and editing. B. Routy: Data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, project administration, writing–review and editing. Y. Okuma: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported by JSPS KAKENHI (grant no.: JP19K16820). B. Routy was supported by H2020 ONCOBIOME. L. Zitvogel was supported by the Ligue contre le Cancer (équipe labelisée); Agence Nationale de la Recherche (ANR) francogermanique ANR-19-CE15-0029 and Association pour la recherche sur le cancer; Cancéropôle Ile-de-France; Fondation pour la Recherche Médicale; a donation by Elior; Fondation Carrefour; Institut National du Cancer; Inserm (HTE); the LabEx Immuno-Oncology; the RHU Torino Lumière (ANR-16-RHUS-0008); H2020 ONCOBIOME and the Seerave Foundation; the SIRIC Stratified Oncology Cell DNA Repair and Tumor Immune Elimination (SOCRATE); FHU CARE, Dassault, and Badinter philantropia; and the Paris Alliance of Cancer Research Institutes. L. Derosa's salary was supported by the RHU Torino Lumière (ANR-16-RHUS-0008) and H2020 ONCOBIOME. The authors thank TechnoSuruga Laboratory Co., Ltd. (https://www.tecsrg.co.jp/) for sequencing of DNA from fecal samples. The authors also thank Dr. Kageaki Watanabe, Dr. Kana Hashimoto, Dr. Shoko Kawai, Dr. Makiko Yomota, and other residents for patients' care and sample collection. In addition, B. Routy would like to thank the Prefontaine Family for their generous support of this project.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.