Microbiota-Centered Interventions: The Next Breakthrough in Immuno-Oncology?

Revolutionary progress has been made over the last decade, as indicated by the approval of several immune checkpoint inhibitors (ICI) by the FDA to block the PD-1/PD-L1 pathway for treatment of ≥16 cancer types: (i) the first-ever tissue-agnostic approval of cancer drugs for the treatment of all cancers with DNA mismatch repair defects; (ii) real-world practice of “synthetic immunology” using adoptive T-cell therapy with two CD19-directed chimeric antigen receptor T-cell (CAR-T) products for relapsed or refractory B-cell malignancies; (iii) the 2018 Nobel Prize to Honjo and Allison “for their discovery of cancer medicine by inhibition of negative immune regulation” (www.nobelprize.org/prizes/medicine/2018); and (iv) the emerging concept of normalizing the mucosal microbiome to modulate systemic immunity (1–8). In fact, most cancers amenable to PD-1/PD-L1 blockade reside at portals of entry, where a rich microbial ecosystem contributes to the development of the local immune system. Unfavorable changes in gut microbial composition and function, which are referred to as “dysbiosis,” have recently been linked to alterations in the cancer immune set point and primary resistance to cancer immunotherapy (1).

Here, we review the impact of antibiotics on cancer immunotherapy, discuss how the balance between beneficial and harmful microbial species can predict the success of immunotherapy, and introduce the potential for microbiota-centered interventions (MCI) to overcome primary or secondary resistance to immunogenic chemotherapy (oxaliplatinum, cyclophosphamide) and ICIs.

Antibiotics not only kill pathogenic bacteria. They also markedly affect gut microbiome diversity and composition (1–5), with subsequent impact on the peripheral immune system (4). These preclinical findings have led to a deeper understanding of how the gut microbiome might influence cancer immunosurveillance and response to cancer immunotherapy (5–8). Since these results were published, a first retrospective study examined the impact of antibiotics in 249 patients treated with ICIs for advanced non–small cell lung cancer (NSCLC), renal cell carcinoma (RCC), or urothelial carcinoma (6). The study highlighted that antibiotic treatment before ICI initiation was associated with a significant decrease in both progression-free survival (PFS) and overall survival (OS), thus corroborating the murine experiments. Considering the potential impact of such findings on routine oncology, a burst of retrospective and, later, prospective studies examined the impact of antibiotics use on ICI efficacy across a wide range of tumors. A meta-analysis published in 2020, which included 2,889 patients from 18 studies, demonstrated that antibiotic use before ICI was more deleterious on OS than antibiotic use after ICI (9). These results were consolidated by an additional meta-analysis limited to patients with NSCLC (10). For this review, we performed a study-level meta-analysis to evaluate the consequence of antibiotics therapy on the clinical benefit of ICI. Hence, we conducted a new review of the literature on Medline and Embase as of December 23, 2020, including all abstracts from the American Society of Clinical Oncology (ASCO) and the European Society of Medical Oncology (ESMO; refer to Supplementary Fig. S1 for methodology). We identified 38 studies including 11,959 patients treated with anti-PD-1/PD-L1 antibodies, anti-CTLA4 antibodies, or both. All included studies were published between 2017 and 2020. Two of 38 studies were prospective in design (11, 12). A total of 2,804 patients (23.4%) were treated with antibiotics in the period before or after their first ICI treatment. Here again, antibiotic uptake was associated with increased mortality at a statistically significant HR of 1.81 [95% confidence interval (CI), 1.50–2.17], in the absence of stratification on timing of administration of antibiotics (Supplementary Fig. S2); HR was even superior when antibiotics were administered before ICI (HR, 2.26; 95% CI, 1.76–2.90; Fig. 1; Supplementary Table S1). This meta-analysis reinforces the notion of the deleterious effects of antibiotics use before ICI initiation, across many malignancies, in stage III and IV cancers, irrespective of the study design.

Beyond these epidemiologic observations, two recent studies characterized the effects of antibiotics on the stool taxonomic composition at diagnosis in patients amenable to ICI. In the first study, shotgun metagenomic profiling of feces was performed before anti–PD-1 in 69 patients with metastatic RCC (6). When considering only patients who did not take antibiotics, microbiota richness at baseline dictated the prognosis, a higher α-diversity being associated with better clinical outcome. Sixteen percent received antibiotics before ICI and had worse clinical outcomes. At baseline, patients on antibiotics had a lower α-diversity than antibiotic-free patients. At the taxa level, antibiotics use was associated with enrichment in Clostridium/Hungatella hathewayi and decreased relative abundance of the Eubacterium genus. In addition, two Clostridium hathewayi species were overrepresented in the group of patients who failed to respond to ICIs. In the second study, fecal samples of 70 patients with NSCLC treated with ICIs were sequenced using 16S rRNA (13). In this cohort, antibiotics use was associated with lower baseline microbiome diversity, as well as increased and decreased prevalence of Hungatella versus Ruminococcaceae UCG 013 and Agathobacter genus, respectively. In antibiotic-free patients, the prevalence of Ruminococcaceae UCG 013 and Agathobacter genus were both associated with improved overall response rate (ORR), PFS, and OS (13).

These data suggest a causal link between antibiotics use, microbiome dysbiosis, and poor ICI efficacy, although with some caveats. First, it is difficult to disentangle antibiotics uptake from poorer performance status and increased susceptibility to infection. Although deleterious for ICIs, antibiotics did not affect the tumoricidal activity of chemotherapy against bladder cancer (14). Second, the stratification of patients based solely on antibiotics use does not consider other important variables (such as antibiotics spectrum, duration, or route of administration) on clinical outcome. Third, the mechanistic link behind the harmful effects of antibiotics before the start of ICI remains sketchy. Long-term antibiotics decreased intestinal IL4 and TGFβ gene expression, colonic lamina propria (LP), and IL9-producing tumor-infiltrating lymphocytes (TIL), while increasing the susceptibility to melanoma development (15). Fecal microbial transplantation (FMT) or recombinant IL9 into tumor bearers could restore intestinal microbiota diversity and the TH9 pool in lung metastases, leading to efficient tumor control (15).

Like antibiotics, proton pump inhibitors (PPI) have a negative impact on the efficacy of immunotherapy (16). The consequences of their uptake in 1,012 patients with cancer amenable to ICI in 20 Italian cancer centers have been reviewed. Whereas baseline uptake of statins, aspirin, and β-blockers were independently related to an increased objective response rate, steroids, antibiotics, PPIs, anticoagulants, and opioids were significantly associated with a higher risk of disease progression (16). Three cohorts comprising a total of 1,815 users of PPIs concordantly exhibited a significant enrichment of “oral” microbes (17). These data were confirmed in another cohort of >2,700 people, unveiling the stool dominance of Streptococcaceae and Micrococcaceae family members (18). Similar species were isolated from the ileal lumen or stools from patients with colorectal cancer resistant to oxaliplatin-based chemotherapy or ICIs (6, 19).

Altogether, oncologists will need guidelines to cope with antibiotics and PPI administration before ICIs. Delaying administration of ICIs or repairing dysbiosis by FMT or compensatory probiotics (such as Clostridium butyricum; ref. 20) are the options discussed in the final section of this review.

Over the last decade, scientists have learned which microbes play a role in individual diseases, which harmful microbes are gained or helpful commensals lost in patients with particular diseases, if certain changes in gut microbes occur across many pathophysiologic disorders, and how to deal with confounding factors such as age (21). Knowledge is expanding on how distinct bacterial commensals may individually or collectively mediate innate and cognate immune responses in the LP, the mesenteric lymph node (mLN), and distant tissues (22, 23). Moreover, ileal microbiota may be more important than colonic microbiota for cancer immunosurveillance. This may be explained by the elective presence of specific immune cell types in the ileum, notably IL12-producing LP dendritic cells (DC), follicular Th cells (TFH), and pathogenic Th1 cells coproducing IFNγ and IL17; the ileal presence of bacterial species and strains (22, 24); the particular cell biology of the crypt (with increased susceptibility to apoptosis (19); and specificities related to lymphatic draining (25). The attempts of immuno-oncologists to understand the anticancer immunostimulatory properties of distinct microbial taxa present in the gut rely on multiple arguments (Fig. 2).

Stool genetic fingerprints associated with objective responses to ICIs have been described across many geographic sites (Europe, United States, Japan, China) and several solid cancers (summarized in Supplementary Table S2). Although most of these studies found an impact of the α or β diversity of the fecal taxonomic composition on the prognosis of patients, they did not always converge with respect to the ranking of clinically relevant bacteria, with variations at the levels of order, genera, or species. These apparent discrepancies could be attributed to cancer types and staging, clinical parameters (ORR vs. PFS vs. OS), extraction and sequencing technologies [16S rRNA sequencing vs. shotgun MetaGenomics (MG)], diet, clinical center, cohort size and geography, and, finally, confounding factors (such as socioeconomic differences, comedications, comorbidities, etc.). Regardless of these considerations, some bacteria with clinical and functional significance were shared among these intestinal blueprints. Thus, responses to PD-1/PD-L1 blockade are associated with an increased abundance of the Verrucomicrobiaceae family (mostly Akkermansia muciniphila), species belonging to the Ruminococcus (including Faecalibacterium prausnitzii), Collinsella (notably Collinsella aerofaciens), Eubacterium and Bifidobacterium (B. longum) genera, members from the Lachnospiraceae family (Coprococcus comes, Dorea formicigenerans), and species from the Bacteroidales order (Barnesiella intestinihominis, Prevotella copri) in several independent studies. In fact, there are arguments in favor of the convergence of these findings. Thus, species-interacting groups can be built using co-occurrence networks that cluster bacteria according to their abundance and interactive links within a large cohort of individuals (6). The residence of Akkermansia muciniphila in the intestines of patients with advanced cancer predicted the long-term benefit to PD-1 inhibition in patients with RCC and NSCLC, while it coincided with a highly diverse gut microbiota dominated by Ruminococcus, Alistipes, and Eubacterium genera and many members of the Lachnospiraceae family (6, 26). In addition, “guilds” represent miniaturized ecosystems containing bacteria overlapping in their niche requirements and sharing similar biological functions (27). For instance, a guild of 15 bacteria belonging to three taxa has been identified during fiber-enriched diet to mediate acetate and butyrate production and to regulate insulin sensitivity in patients with type 2 diabetes (27). Kasper and colleagues (28) demonstrated that selected Bacteroides spp. (such asB. fragilis, B. thetaiotaomicron) can modulate colonic RORγ⁺ regulatory T cells (Treg) through the bile acid (BA) receptor VDR (vitamin D3 receptor). This guild may have clinical significance in gastrointestinal (GI) and hepatic cancers, in which dysregulated intestinal BAs mediate immunosuppression (Supplementary Table S3). In oncoimmunology, A. muciniphila, Faecalibacterium, and Ruminococcus genera constitute an “immunologic” guild. Indeed, Xavier and colleagues (29) identified Faecalibacterium, Ruminococcus 2, and Akkermansia as dominant taxa highly associated with immune cell dynamics during white blood cell recovery after hematopoietic stem cell transplantation (HSCT). Mager and colleagues (30) described an adenosine A2A receptor agonist–related guild composed of Bifidobacterium pseudolongum, A. muciniphila, and Lactobacillus johnsonii that are capable of producing inosine or hypoxanthine, which both are potent ligands of the adenosine A2A receptor and elicit costimulatory signals on T cells to enhance the efficacy of CTLA4 blockade. Of note, inosine monophosphate and hypoxanthine were also highly elevated in the cecum and serum of mice colonized with a consortium of 11 immunogenic bacteria, improving ICI efficacy (31).

Aside from taxonomic and metabolomic classification, the functional profile of the microbiota can be inferred by studying the genes expressed by the microbiota (21, 32, 33). Importantly, patients who responded to immunotherapy tended to harbor high relative abundances of bacteria found in high-fiber diets and endowed with antidiabetic effects, known to produce short-chain fatty acids (SCFA; refs. 34–36) and secondary BAs (ref. 37; Fig. 2).

Successful cancer immunotherapy relies on the host capacity to generate tumor antigen–specific, IFNγ-secreting Th1 cells and cytotoxic effector T (Tc1) cells (38). The anticancer effects induced by CTLA4 and PD-1 co-blockade involve DCs that, in mice, depend on the transcription factor Batf3 (38) as well as IL12p70 (3, 4) and the Th1-associated Cxcr3 receptor (39). Hence, gut commensals capable of correcting defective Th1/Tc1 immune responses in antibiotics-treated tumor-bearing mice turned out to be good candidates for MCI (Supplementary Table S3).

A number of studies showed that Enterococcus hirae, B. fragilis, A. muciniphila, and Bacteroides rodentium in conjunction with cyclophosphamide, anti-CTLA4, anti–PD-1, or Rnf5 gene deficiency, respectively, elicited IFNγ-, DC-, and IL12–dependent Th1/Tc1 immune responses that mitigated tumor progression (3, 4, 40, 41) Bifidobacteria spp. and their active metabolite inosine were mandatory for the IL12Rβ2-dependent T-cell triggering and full-blown Th1 immune responses during immunotherapy of colon cancer with anti-CTLA4 and CpG (30). The cooperation between the tumor-derived chemokine CCL5 and IFNγ-inducible CXCR3 ligands secreted by myeloid cells was required for orchestrating T-cell infiltration of immunoreactive tumors (39). Similar chemokine-driven pathways orchestrated activation of antitumor T cells in the gut. In Rnf5^−/− mice, gut commensals dominated by B. rodentium stimulated CCL5 release from the intestinal barrier, leading to mobilization of antigen-presenting cells and Th1 priming and trafficking into melanoma lesions (41).

Anti–PD-1 and anti-CTLA4 antibodies mediated efficacy against poorly immunogenic breast tumors with high mutational burden by triggering FTH cells and B-cell responses, generating antibody-dependent cellular cytotoxicity–eliciting antitumor IgG2b antibodies (42). In microsatellite-stable (MSS) colorectal cancer, distinct ileal commensals (including B. fragilis, Alistipes ondordonkii, Erysipelatoclostridium ramosum) contributed to oxaliplatin-induced crypt cell apoptosis that elicited antitumor TFH immune responses and IgG2b antibodies. Such immunogenic bacteria could promote the mobilization of LP DCs producing IL1β and IL12p70 that were indispensable for the priming of TFH after oxaliplatin. Oxaliplatin together with PD-1 blockade became efficient against MSS colon cancers normally resistant to PD-1 blockade when immunogenic ileal commensals colonized ileal crypts (19). Similarly, A. muciniphila, one of the most immunogenic commensals, is capable of triggering various types of adaptive immune responses (TFH, Th1, Th17, IgA, and IgG1; ref. 43). Accordingly, patients with cancer capable of mounting Th1 or Tc1 memory T-cell responses against A. muciniphila, B. fragilis, or E. hirae exhibited longer PFS than patients who did not (3, 4, 40). Again, patients with melanoma resistant to PD-1 blockade who benefited from allogeneic FMT from responding donors mounted immune responses harboring the hallmarks of TFH (plasma IL21 and Cxcl13) associated with IgG responses against donor microbiota (37).

Commensals can also regulate the innate arm of the immune system, locally or at distant sites. The capacity of distinct intestinal bacteria to control the type I IFN fingerprint of DCs located in secondary lymphoid organs or lungs was crucial for protecting mice against melanoma progression or influenza viruses, respectively. The indigenous microbiota controls constitutive production of type I IFNs by plasmacytoid DCs (44, 45). Indeed, a Clostridium orbiscindens–associated metabolite, desaminotyrosine, could protect mice from influenza virus through augmented type I IFN signaling (46). In melanoma-bearing mice, colonic Bifidobacterium genus controls the type I IFN production by DCs and modulates the effectiveness of PD-L1 blockade (47).

Some taxa might have a therapeutic benefit by decreasing drug-induced intestinal toxicity and shifting the gut ecosystem toward a healthier status. PD-1 and CTLA4 co-blockade is effective in transplantable cancer models and patients but exacerbates autoimmunity. IL1β and TNF were increased in the LP of patients with cancer experiencing ICI-induced colitis and a dominance of stool Bacteroides intestinalis that coincided with increased colonic IL1β and ICI-induced side effects (48, 49). Hence, treating mice with the TNF or IL1R1 inhibitors improved the preclinical efficacy of ICI (48), perhaps by attenuating the dysbiosis. In fact, TNFα inhibition tended to correct gut microbial dysbiosis by increasing richness and SCFA-producing bacteria while decreasing pathogenic commensals in Crohn disease (50, 51).

In many chronic inflammatory processes, intestinal barrier fitness is compromised, leading to increased intestinal permeability, systemic inflammation, and eventually immunosuppression (52, 53). Whether and how progressive carcinogenesis could mediate this syndrome remains an open conundrum.

Select bacteria (such as Bifidobacterium breve) are capable of mitigating intestinal inflammation, as shown for dextran sulfate sodium–induced colitis and CTLA4 blockade–induced colon toxicity. They do so by shifting the colonic microbial ecosystem toward Lactobacilli spp. (in particular L. rhamosum) and by increasing the mitochondrial fitness of IL10-producing Tregs (54) or boosting intestinal regulatory B cells. Toll-like receptor-2 (TLR2) and TLR9 ligands and enteric bacterial lysates preferentially induce IL10 production by, and regulatory capacity of, intestinal B cells (55). The anti-inflammatory role of specific probiotics on intestinal barrier fitness had marked consequences on the efficacy of cisplatin-based chemotherapy against a murine hepatocarcinoma. Probiotics modulated the gut microbiome composition toward beneficial bacteria (B. fragilis, Parabacteroides distasonis, Prevotella, and Oscillibacter), which produced anti-inflammatory metabolites and favored the local differentiation of anti-inflammatory IL10-producing Tregs at the expense of the proangiogenic Th17 population. Therefore, the combination of probiotics with chemotherapy yielded synergistic antitumor effects against hepatocarcinoma (56).

In contrast, other bacteria need to translocate from the small intestine to secondary lymphoid organs to stimulate the cognate immune system during immunogenic cell death–mediating chemotherapy. E. hirae regulated many components of the ileal epithelial barrier (goblet and Paneth cells, intraepithelial γδT cells) to contribute to the efficacy of cyclophosphamide (56). Translocation of the aerobic E. hirae from the gut to the spleen was inhibited by epithelial NOD2 molecules (56). A villin-driven defective expression of Atg4b or Atg5 components of the autophagy machinery abrogated the antitumor effects of E. hirae and cyclophosphamide against a sarcoma model (57). However, more comprehensive analyses will be required to better define the intrinsic effect of the carcinogenesis process on the dysfunctions of the epithelial barrier and its commensalism to best appreciate the anti-inflammatory and/or immunostimulatory bioactivity of cancer therapies.

Cancer therapies may affect the intestinal microbiota by direct mechanisms (through their mode of action on enterocytes or crypt stem cells) or indirectly (through the gut immune system or the metabolism of xenobiotics). Hence, in patients with RCC, PD-1 blockade for 6 months increased gut microbiota richness (6). Tyrosine kinase inhibitors (such as cabozantinib, axitinib, sunitinib) induced a significant and prototypic shift in microbiota composition characterized by a relative overabundance of immunostimulatory commensals (such as A. muciniphila, Alistipes senegalensis; ref. 31) in mice and patients (6). Systemic androgen deprivation and oral abiraterone acetate affected the intestinal microbiota of patients with prostate cancer. Depletion of androgens reduced the abundance of androgen-using and proinflammatory Corynebacterium spp. As a consequence, oral abiraterone acetate selectively reshaped the gut microbiota by favoring the overrepresentation of the health-associated and anti-inflammatory gut commensal A. muciniphila (58).

Altogether, beneficial gut bacteria play a central role in maintaining intestinal barrier integrity and metabolic homeostasis to avoid systemic inflammation as well as boost cancer immunosurveillance during immunotherapy (Fig. 2).

Besides their anti-inflammatory and proimmunostimulatory functions, gut microbes can promote tumor growth and dissemination by (i) favoring the epithelial-to-mesenchymal transition and favoring cancer cell proliferation, (ii) steering the immune system toward Treg and Th17 responses, and (iii) subverting healthy metabolism, as discussed in several reviews (59).

Age and co-medications influence the microbiota composition toward unfavorable patterns that have also been found in ICI-resistant patients. Younger people gradually acquire disease-associated gut microbes, whereas older people tend to lose health-associated gut microbes. Moreover, distinct microbes can be gained in many diseases and across age groups. Distinct species that were enriched across diseases regardless of age included a group of Clostridia (C. bolteae, C. symbiosum, C. hathewayi, C. citronae, C. asparagiforme; ref. 60; Fig. 2). Most if not all of these taxa were found to be overrepresented in nonresponders to ICI or in poor-prognosis cancers [refs. 6, 19, 61; Derosa and colleagues, ASCO Merit Award 2021 (https://ascopubs.org/doi/abs/10.1200/JCO.2021.39.15_suppl.9019); Supplementary Table S1]. Co-medications interfering with the success of ICI may be informative about immunosuppressive pathobionts. As mentioned above, antibiotics administered 1 to 2 months before immunotherapy curtail its clinical benefit, presumably by interfering in the balance between beneficial and harmful bacteria. Derosa and colleagues (6) and Hakozaki and colleagues (13) both identified the antibiotics-induced overrepresentation of Clostridia spp. (such as Hungatella/C. hathewayi) or Erysipelotrichaceae family coinciding with reduction of the relative abundance of beneficial Eubacterium rectale or Lachnospiraceae family (Dorea, Roseburia, Coprococcus spp.), and Ruminococcus (Faecalibacterium spp.) genus. Similarly, PPIs, which have a negative impact on the efficacy of immunotherapy (16), promote significant enrichment of oral microbes consisting of Enterococcus, Staphylococcus, Streptococcus, and Rothia genera in the feces (17). In fact, these bacterial species were recovered from stools in patients with colorectal cancer resistant to chemoimmunotherapy (6, 19).

Nonresponders to ICIs harbor gut microbiome signatures that are dominated by Proteobacteria (E. coli, Shigella, Klebsiella, often diagnosed after antibiotics use), Fusobacteria genera, Porphyromonadaceae family members, and Actinobacteria (Eggerthella lenta, Atopobium parvulum), but their deleterious role in mediating immunosuppression has not been formally demonstrated. It remains unclear whether lack of response to ICI results from the absence of beneficial commensals or the presence of harmful commensals (Fig. 2; Supplementary Table S2). As an example of harmful commensals, Fusobacteria and Prevotella spp., which dominate the ileal commensals in patients with proximal colon cancers, elicited Th17 responses instead of TFH and B-cell responses, compromising the tumor growth–reducing activity of oxaliplatin alone or in combination with PD-1 blockade (19).

More controversial is the clinical significance of commensals that are found within tumors (62) as well as the unique microbial signature in the blood of patients with cancer (63). The intricate roles of microbes in cancer have been extensively discussed in a recent review article (64). On one hand, distinct sets of bacteria colonize tumors within tumor nests or outside by infecting myeloid cells. Intracellular bacteria can be processed and peptides presented onto MHC class I and II molecules to elicit a recall response from TILs, at least in melanoma (65). On the other hand, accumulating evidence points to a proinflammatory and negative prognostic impact of this local microbiota. In advanced NSCLC, microaspirations and swallowing disorders may account for the identification of supraglottic pathobionts (Veillonella, Fusobacteria, Staphylococcus, Streptococcus genera) in the lower bronchoalveolar airways, associated with dismal prognosis (61). Bacterial translocation experiments suggested interactions between the intestinal and pancreatic compartments, presumably via the pancreatic duct. Proteobacteria and Actinobacteria were isolated and incriminated in the transformation of pancreatic inflammatory lesions (66). These bacteria have been cultivated, and causative links between their residency or their TLR ligands (TLR2/5) and γδT17, Th17, and myeloid cells have been established. Bacterial ablation by antibiotics triggered immunogenic reprogramming of the pancreatic ductal adenocarcinoma microenvironment, with M1 macrophage and Th1 differentiation boosting cytotoxic T-cell effector functions (66).

Hence, the dual role of the local and distant (gut) microbiota may add complexity to the effects of antibiotics on patients with cancer. Depending on the precise location of the tumor and its stage, specific antibiotics may be used with the purpose of eliminating pathogenic microbes.

The gut microbiota regulates a variety of different functions in the meta-organism within the epithelial, immune, or hematopoietic components of the intestinal barrier, which connect with advanced cancers through a gut–bone marrow–thymus–secondary lymphoid organ axis. The functional effects of the gut microbiota on local and distant tissues vary according to the α and β diversity, the abundance and geodistribution of the live biomass, the dynamics of growth of selected commensals, the presence of regulatory strains, their absolute or relative representation, and species-interactive groups. Interestingly, distinct species may have pleiotropic modes of action (57).

Functional links between the epithelial component of the intestinal mucosa and the tumor microenvironment (TME) were recently unveiled. Knockout mice and bone marrow chimeras allowed the identification of the role of RnF5, a membrane-anchored E3 ubiquitin ligase implicated in endoplasmic reticulum (ER)–associated protein degradation and the recruitment of DCs and T cells into aggressive melanomas. RNF5 controls clearance of misfolded proteins and the stability of ATG4B and STING. Mice lacking RNF5 exhibited activation of the ER stress response, which coincided with increased expression and signaling through TLR4/Myd88, and the NLRP3/NLRP6 inflammasome platforms in the gut. Decreased secretion of antimicrobial peptides and increased cell death in the ileal crypts caused intestinal dysbiosis in Rnf5^−/− mice. This bowel injury allowed the exacerbated recruitment, activation, and mobilization of CCR7-expressing DCs to Peyer patches, mLNs, and melanoma draining lymph nodes, culminating in the massive infiltration of these poorly immunogenic melanomas by plasmacytoid and conventional DCs and IFNγ-producing CD4⁺ and CD8⁺ T lymphocytes that kept tumor progression in check (41). Cohousing of Rnf5^−/− with wild-type (WT) mice or administration of antibiotics to Rnf5^−/− mice restored tumor aggressiveness, whereas oral gavage with a cocktail of 11 species (enriched in Bacteroides and Parabacteroides spp. including Bacteroides rodentium) overrepresented in Rnf5^−/− animals recapitulated tumor immunosurveillance mechanisms. Coculture of Rnf5^−/− duodenal epithelial cells with B. rodentium elicited the TLR signaling pathway and ER stress, as well as inflammatory chemokine/cytokine release triggering bone marrow (BM)–DC activation and antigen presentation to T-cell receptor transgenic T cells in vitro. Hence, reduced ER stress in tumor cells and ileum coincided with deviated microbiota and exacerbated antitumor immunity (Fig. 3).

Ileal crypt apoptosis was a phenotypic feature of Rnf5^−/− mice that could be recapitulated by chemotherapy in WT animals (19). Indeed, oxaliplatin-induced, caspase-3/7–dependent apoptosis of enterocytes located in ileal crypts favored the expansion of immunogenic bacteria in the ileal mucosa. These commensals (including B. fragilis and E. ramosum) governed epithelial apoptosis and the subsequent priming of type 1 TFH in mLNs, culminating in B-cell activation, IgG2b production, and infiltration of colorectal cancer by TILs in mice and patients (19).

A hypovitaminosis concerning all-trans-retinoic acid (atRA) is an important factor for colorectal cancer development. atRA concentrations in the colon result from the balance between atRA-synthesizing retinaldehyde dehydrogenase (ALDH1A1) and atRA-catabolizing cytochrome P450 (CYP26A1). Depletion of the intestinal microbiota prevented the colitis-induced imbalance of atRA enzymes, leading to atRA depletion and colorectal cancer progression (67). MHC class I genes are transactivated by atRA-elicited transcription factors in normal enterocytes and colon cancer cells, explaining why atRA boosts the cytotoxic activity of tumor-infiltrating CD8⁺ T cells. Along the same lines, Ziegler and colleagues (68) showed that Stat3 gene deficiency in enterocytes leads to the mitophagy-dependent induction of MHC class I molecules, which can be transferred (“cross-dressed”) to DCs that then prime T cells against colorectal cancer.

These examples illustrate that barrier injury is accompanied by a deviation of the local microbiome, which participates to enterocyte stress or death, which in turn mobilizes the recruitment and migration of DCs to and outside the gut-associated lymphoid tissue (GALT), contributing to the infiltration of tumor beds by helper or cytotoxic T cells.

The gut microbiota has broad effects contributing to innate and adaptive immunity at steady state and in pathophysiologic conditions including carcinogenesis (7). Anticancer therapies highlighted the links between distinct commensals and protective antitumor T-cell responses. Cyclophosphamide enabled E. hirae to translocate and stimulate pathogenic Th17 responses and IFNγ-producing CD8⁻ T-cell effectors that controlled tumor growth (1, 40). CTLA4 blockade favored ileal dominance of B. fragilis and B. thetaiotaomicron that promoted MHC class II–restricted TH1 memory responses that were beneficial against tumors (4). PD-L1 or CD47 blockade leading to T-cell priming against melanoma were more effective in hosts harboring species from the Bifidobacteria genus in their colons. Adoptive T-cell transfer after total body irradiation depended on the taxonomic composition of the intestinal microbiota, gut permeability, and the translocation of TLR4 ligands (69, 70). Oxaliplatin-induced cell death of ileal enterocytes inversely governed the immunogenic Erysipelotrichaceae and tolerogenic Fusobacteriaceae proportions in the ileum, dictating the balance between antitumor TFH and deleterious Th17 responses in colon cancer (19). The immunogenicity of these commensals relied on both antigenicity and adjuvanticity (Fig. 3; Supplementary Table S3).

GALT or tumor-draining lymph node–sessile DCs sense immunogenic commensals, engaging in antigen processing and maturation. DC activation leads to type I IFN expression after exposure to Bifidobacteria spp. (47, 71) or IL12 release after encountering B. fragilis or Bifidobacteria spp. (3, 19), A. muciniphila (4), B. rodentium (41), or Bacteroidales S24-7 enriched flora (70). Thus, oral administration of a cocktail of Bifidobacteria failed to facilitate the antitumor effect of CD47 blockade in Ifnar^f/fCD11c^Cre mice (71). Antibiotics can provide microbe-associated molecular patterns that stimulate immune responses either locally or at a distance (72). Depletion of the intestinal flora compromises the clearance of systemic infections after BM transplantation and sensitizes mice to a semilethal dose of radiation. Indeed, microbiota-derived compounds can protect against irradiation-induced hematopoietic injury (73), by the release of microbial-associated molecular patterns known to maintain BM-derived myeloid cells and neutrophil function (74, 75). This effect may be explained, at least in part, by the delivery of endogenous ligands for RIG-I (such as 3pRNA and RNA derived from viruses, phages, or bacteria) that induce protective IFN1 signaling in enterocytes and hence favor intestinal barrier repair (ref. 72; Fig. 3).

To mediate their full adjuvanticity, select immunogenic commensals may or may not need to be alive. Pasteurized A. muciniphila or its specific outer membrane proteins containing TLR2 ligands effectively prevented colitis-associated colorectal cancer in a T cell–dependent manner (76). Conversely, only live, not dead, Bifidobacteria boosted the antigen cross-presenting capacity of DCs through the induction of a STING-dependent type 1 IFN response (71). Whereas the translocation of live intestinal Enteroccus hirae was regulated by enterocyte-specific NOD2, activation of DCs by live E. hirae depended on splenic NOD2 and TLR2 to enhance cyclophosphamide-mediated Th1 cell–dependent antitumor effects (40). All the above examples of bacterial immunogenicity during cancer therapies (B. fragilis, B. rodentium, E. ramosum, Bacteroidales S24-7) depended on TLR4 or TLR9 signaling pathways in the gut.

Bacteria can help DCs not only to prime anticancer T cells but also to participate in gut barrier fitness. The Bifidobacterium genus senses luminal iron and regulates the capacity of LP DCs in the colon to secrete the hormone hepcidin, which is critical to repair the intestinal barrier (77). The microbiota also affects the myeloid cells of the TME. Antibiotics blunted the TLR4-dependent capacity of oxaliplatin to mediate early tumor genotoxicity by inducing reactive oxygen species by the NADPH oxidase NOS2 in tumor-associated myeloid cells (2). These examples illustrate the manifold actions of specific bacteria on the function of myeloid cells.

Apart from providing adjuvants, the gut microbiota represents a paramount source of antigens that can elicit commensal-specific T-cell responses (refs. 43, 78; Fig. 3).

In silico predictions revealed that pancreatic tumors from long-term survivors have a high neoantigen burden associated with strong CD8⁺ CTL infiltration. These neoantigens exhibited homology to infectious bacteria-derived peptides, suggesting molecular mimicry between neoantigens and microbial epitopes (79). Fluckiger and colleagues demonstrated the immunogenicity of an enterococcal lytic phage belonging to the Siphoviridae family, containing a MHC class I–restricted CD8⁺ T-cell epitope within the tape measure protein (TMP; ref. 22). This phage-encoded antigen shared 78% structural homology with an H-2K^b-restricted epitope of PSMB4, an oncogenic driver. Mutations within the epitopes encoded by either the phage or PSMB4 reduced the capacity of the Enterococcus strain to synergize with cyclophosphamide or PD-1 blockade (22). Some bacteriophages have the potential to transfer immunogenic sequences to other strains within the host ecosystem. The lytic phage contained within distinct species of E. hirae or E. faecalis in children and adults infects and lysogenizes Enterococcus gallinarum in the murine gut, thus transmitting on an H-2K^b–restricted epitope that potentiates anticancer immune responses (22). Similarly, T cells targeting an octapeptidic epitope (SVYRYYGL), expressed by the commensal bacterium Bifidobacterium breve, cross-reacted with a model neoantigen (SIYRYYGL) expressed by mouse melanoma B16-SIY cells (80). Moreover, a fraction of human T cells specific for naturally processed melanoma epitopes recognize microbial peptides (22), supporting the clinical significance of these findings. Indeed, patients with RCC and NSCLC harboring the TMP phage sequence in their stools exhibited a better response to PD-1 blockade than patients with cancer who lacked this fecal phage (22). Moreover, melanoma was found to contain intracellular bacteria from which peptides could be presented onto MHC class I and II complexes to stimulate cytotoxic lymphocytes in primary and metastatic lesions (65).

By their interaction with DCs or epithelial cells, commensals trigger pathogen recognition receptors, leading to the release of chemokines that favor the infiltration of tumors by polarized effector T cells. Thus, in colorectal cancer, commensals and their TLR ligands act on cancer cells to modulate the production of chemokines that govern the intratumor trafficking of Th1, Th17, or TFH cells, hence affecting the immunoscore and patient prognosis (81). The natural killer (NK) T cell–dependent immunosurveillance of hepatocarcinoma is controlled by intestinal microbes transforming BA into regulators of CXCL16 release by liver sinusoidal endothelial cells. Outer membrane vesicles from Gram-negative bacteria as well as extracellular vesicles from Gram-positive commensals (Staphylococcus aureus or Lactobacillus acidophilus) may stimulate a pro-Th1 chemokine secretory pattern (Cxcl10, IFNγ) in necrotic areas of the TME that then favors the activation of CXCR3-expressing NK and T cells (ref. 82; Fig. 3).

Bacteria can trigger innate and cognate immune responses through metabolites that they produce themselves or induce in the host. During immunotherapy with anti-CTLA4 (combined with either CpG or anti–PD-L1 antibody), inosine produced by gut-resident microbial species (such as B. pseudolongum and A. muciniphila) leads to adenosine A2A receptor–dependent costimulatory effects on T cells and Th1/TC1 differentiation and accumulation in tumors. Indeed, A2A receptor signaling in T lymphocytes was mandatory for the ICI-promoting effects of B. pseudolongum and allowed for IL12Rβ2 and IFNγ transcription, whereas IL12 was delivered by cDC1 in contact with B. pseudolongum, likely in secondary lymphoid organs (ref. 30; Fig. 3).

Radiation at curative doses or for conditioning during HSCT induces hematopoietic and GI dysfunction. The guts from survivors (mice or patients with leukemia) to total-body irradiation are overcolonized by members of Lachnospiraceae and Enterococcaceae families. Such bacteria contribute to postradiation restoration of hematopoiesis and GI repair through the production of propionate and tryptophan metabolites (83). Similarly, certain gut-derived metabolites dampen immune responses to conventional radiotherapy. Irradiation of tumor lesions was more effective in the presence of vancomycin that eliminated Clostridiales-producing immunosuppressive metabolites (butyrate and propionate). These SCFAs impaired antigen presentation by DCs in vitro and abrogated the vancomycin-enhanced antitumor activity of radiotherapy in vivo (84). SCFAs were also associated with resistance to CTLA4 blockade in mice and patients with melanoma. High blood butyrate and propionate levels were associated with increased Treg proportions, reduced DC and effector T-cell activation, and lower responses to IL2 (85). These few examples illustrate the capacity of bacterial metabolites to affect antitumor immune responses, suggesting that many additional metabolites affecting the cancer–immune dialogue will be deciphered in the future.

During inflammation-induced carcinogenesis, most specifically IL17-associated inflammation, gut microbes can have a prophylactic or therapeutic anti-inflammatory effect, interfering with cancer progression (Fig. 3).

Prior experimental evidence in oncogene-driven autochthonous lung cancer models unveiled that local commensals may be perturbed by carcinogenesis, triggering an inflammatory cross-talk between alveolar macrophages and IL17-producing lung resident γδT cells, contributing to tumor progression (86). Tsay and colleagues (61) demonstrated that the lower airway microbiota may harbor supraglottic commensals (such as Veillonella parvula) that turn on IL17-mediated inflammatory pathways and reprogram host transcription to exacerbate NSCLC progression (86). Lower airway microbiota profiles containing predominantly supraglottic taxa were associated with advanced stages at NSCLC diagnosis, as well as with poor therapeutic responses. Pancreatic cancers—which are notoriously invaded by microbes (87)—may result from a vicious cycle between IL17-mediated neutrophil recruitment into tumor beds, as well as CD8⁺ T-cell exclusion (88). Blockade of IL17RA/IL17 or antibiotics could restore sensitivity to PD-1 blockade in preclinical models of pancreatic cancer (66, 89).

Likewise, IL10-producing Tregs might prevent inflammation-induced tumorigenesis. For example, GPR109a was found to be indispensable for butyrate- and niacin-mediated induction of IL18 in colonic epithelium and suppression of colitis-induced colorectal cancer. GPR109a signaling also promoted anti-inflammatory responses of colonic macrophages and DCs and enabled them to induce the differentiation of IL10-producing Tregs (90). Another strategy consisted in supplementing the gut with probiotics capable of shifting the gut microbial community toward species (Prevotella, Oscillibacter, P. distasonis) known to produce anti-inflammatory metabolites (such as the long-chain fatty acids palmitoleate and docosahexanoate). Those species maintained intestinal IL10-producing Tregs while decreasing the gut exodus of proangiogenic and protumorigenic Th17 cells, thereby increasing the anticancer effects of platinum salts in murine hepatocarcinoma (ref. 56; Supplementary Table S3).

With the increasing recognition of the gut microbiome as the second human genome, the concept of pharmacomicrobiomics has been introduced as a natural expansion of pharmacogenomics. Several examples of microbial enzymes detoxifying cytotoxic antineoplastic agents are outside the scope of this review: Proteobacteria destroying gemcitabine (91), intestinal bacterial β-glucuronidases destroying camptothecin (92, 93), and metabolic drug interconversion of fluoropyrimidines involving bacterial vitamin B6, vitamin B9, and ribonucleotide metabolism.

Animal studies have suggested a causal relationship between the taxonomic composition of the intestinal ecosystem and the host–tumor interplay for the clinical outcomes (Supplementary Table S3). Thus, distinct facets of the immune or metabolic tones can be transferred by fecal transplantation or gavage with specific bacterial strains or consortia. As a result, several MCI maneuvers are currently being evaluated in clinical trials dealing with patients with cancer. Host conditioning before MCI (such as pretreatment with oral vancomycin or polyethylene glycol–based osmotic solutions) and lifestyle changes (such as diet and exercise) may be important to improve the success of colonization and the bioactivity of the MCI.

This term refers to the transfer of the fecal microbial content by endoscopy or oral capsules from a healthy individual (or a cured patient) into the intestine of a patient. The basic principles of FMT preparation have been detailed elsewhere (94). Depending on the country, FMT is regulated as a drug, tissue, or a combined product comprising human cells and nonhuman components (microbial DNA and metabolites). Stool banks are recommended to operate under the designated authority of each country (94), specifically in the context of COVID-19. The use of FMT to treat refractory or recurrent (>3 episodes) C. difficile–related colitis is favored by professional guidelines but is still considered an investigational treatment by the FDA, at the time of this writing under a safety alert (https://www.fda.gov/safety/medical-product-safety-information/fecal-microbiota-transplantation-safety-alert-risk-serious-adverse-events-likely-due-transmission). Moreover, FMT is currently in experimental stage for the treatment of inflammatory bowel disease, irritable bowel syndrome, hepatic encephalopathy, stroke, heart failure, obesity (95), and cancer (ref. 96; Supplementary Table S4).

FMT could have various indications in patients with cancer. By reprogramming stool composition for increased α diversity and proper delivery of health-related bacteria (21), FMT is viewed as a potential cure for dysbiosis induced by broad-spectrum antibiotics. Thus, FMT might reverse the antibiotic-induced dominance of members of the Clostridia group XIVa and Proteobacteria class that dampen the efficacy of ICI (97). Second, enriching the host microbiota with species endowed with intrinsic immunogenicity may circumvent primary resistance and restore responsiveness to ICI (Supplementary Tables S3 and S4). Hence, the direct recycling of stools from cured patients into ICI-resistant individuals could solve this issue (96).

In a proof-of-concept phase I trial, Baruch and colleagues (NCT03353402; ref. 96) treated 10 patients with metastatic melanoma refractory to PD-1 blockade with allogeneic FMT derived from 2 patients who had fully responded to anti–PD-1 antibodies. PD-1 blockade was reintroduced during the FMT process. Three of 5 patients colonized with the same donor-derived FMT exhibited partial (n = 2) or complete (n = 1) responses. None of the patients with transplants from the other donor responded. Of note, there were taxonomic differences in microbial composition of these 2 donor samples at baseline. After treatment, all patients exhibited changes in their gut microbiota, and patients who received the more immunostimulatory feces had greater relative abundance of Ruminococci spp. (R. gnavus, R. callidus) and Bifidobacterium adolescentis, which are considered potentially favorable for immunotherapy, as well as Lactobacillus ruminis (an anti-inflammatory bacterium), whereas those who received the other donor's sample manifested an overrepresentation of Coprococcus spp. and Clostridiaceae family members (C. clostridioforme, Erysipelatotrichaceae) already described in antibiotics-treated individuals (6). These gut compositional shifts were associated with reprogramming of the TME in the 5 patients who received the more effective feces, with an increase in tumor-infiltrating CD8⁺ T cells, as well as IFN and effector T-cell differentiation gene patterns (PRF1, GrZMB, CD69, CD27, CD28, IDO1, PD-L1, and PD-L2).

A parallel study conducted at the University of Pittsburgh reported similar promising results. In the same patient population composed of 15 patients with refractory melanoma receiving only 1 FMT from responders through colonoscopy, together with the reintroduction of pembrolizumab (37), restoration of clinical benefit was observed in 6 patients (3 partial responses and 3 stable disease >12 months). Interestingly, 1 responding patient developed a soft-tissue infection requiring broad-spectrum antibiotics. However, antibiotics changed the microbiome in this patient, with a loss of beneficial Ruminococcaceae, Alistipes, and F. prausnitzii, and translated to clinical progression (37). Therefore, the investigators performed a second FMT with the same donor 1 year after the original one, which colonized the recipient and triggered a clinical response. Metagenomics profiling and donor bacteria–specific IgG responses further revealed that the responding recipient microbiota exhibited a significant shift toward the donor composition compared with the nonresponders (37). Of note, successful FMTs were enriched in Ruminococcaceae, Bifidobacteriaceae, Lachnospiraceae, and Erysiplotrichaceae, whereas Tannerellaceae, Sutterellaceae, and Bacteroidaeceae were dominant in ineffective FMTs (37). Finally, the blood immunomics and TME contexture were evaluated after FMT. Responder patients (R) displayed a higher frequency of blood CD8⁺CD56⁺ and activated mucosal-associated invariant T cells, both subsets overexpressing killer molecules (Granzyme, perforin, NK-activating receptors) coinciding with upregulation of class II molecules in TILs. In contrast, nonresponders (NR) accumulated osteopontin and CXCL8-producing myeloid cells and Tregs in tumors. Serum analytes also segregated responders from nonresponders after FMT, with decreased concentrations of inflammatory markers (such as CXCL8, IL18, CCL2) in responders, concomitantly with increased levels of Flt3L, CX3CL1, type 2, and TFH cyto/chemokines (37). Altogether, these two proof-of-concept studies confirmed that FMT can shift the microbiome composition of a patient with cancer upon successful engraftment and has the potential to reprogram the peripheral immune and inflammatory tone as well as the TME, culminating in a clinical benefit to ICIs. Overall, the favorable safety profile and early signs of efficacy support further investigations of allogenic FMT for enhancing immunotherapy effects.

Another indication for FMT could be the mitigation of ICI-induced immune-related adverse effects such as colitis. Indeed, a low diversity index of stool taxonomic content was associated not only with resistance to ICI but also an increased risk of immune-related adverse effects upon neoadjuvant PD-1 and CTLA4 blockade (98). In a pioneering clinical study involving 2 patients, FMT successfully inhibited ICI-induced colitis. The resolution of severe colitis could be assigned to the expansion of Clostridia and Bifidobacterium taxa in 1 of these patients.

A further indication for FMT is the reduction of transplant-related mortality in the context of allogeneic HSCT, in which patients are preconditioned with broad-spectrum antibiotics, favoring the emergence of antibiotic-resistant facultative pathogens (E. faecium, E. faecalis; ref. 99). In allo-HSCT, FMT facilitates hematopoiesis and GI recovery (83), as it prevents acute graft-versus-host disease. FMT has the potential to reintroduce health-promoting microorganisms such as Blautia or Lachnospiraceae (E. limosum) and Actinomycetaceae that can contribute to reduce the risk of transplant-related mortality (refs. 100–102; Supplementary Table S4), as well as taxa (such as Faecalibacterium, Ruminococcus 2, and Akkermansia) associated with immune cell dynamics after HSCT (29). These beneficial effects could be achieved by conventional autologous or allogeneic FMT or by FMT enriched with specific health-promoting microorganisms (Supplementary Table S4). A randomized, controlled clinical trial of autologous FMT versus no intervention allowed analysis of the intestinal microbiota profiles of 25 patients with allo-HSCT. The auto-FMT intervention enhanced microbial diversity and reestablished the intestinal microbiota composition that the patient had before conditioning (Supplementary Table S4).

However, FMT harbors a significant safety risk (103), limiting its implementation in prophylactic usage. FMT should be tested in prospective randomized trials to confirm its efficacy in primary resistance to ICI and decipher the enterofecal compatibility rules for long-term colonization of the graft and its clinical significance for prolonged survival.

Fecal transplants present challenges in social acceptance, safety, manufacturability, and consistency. Live biotherapeutics (lyophilized encapsulated bacteria) can potentially resolve these issues. Probiotic strains differ greatly in their mechanisms of action, survival potential during GI transit, metabolic and immunologic local modulations, and competition with microbiota niches and pathogens. A number of randomized controlled trials have been conducted using commercially available “probiotics” for the treatment of chronic inflammatory or autoimmune diseases (Supplementary Table S4), with modest results. Many of the probiotic species used in these trials were oxygen-tolerant lactic acid producers associated with food production (Lactobacillus spp., Bifidobacteria spp., Bacillus coagulans, Streptococcus thermophilus, and Saccharomyces boulardii).

Probiotics have been used in patients with cancer to address some side effects of anticancer treatments, especially radiotherapy. However, most of the reported trials are small studies that vary in the types and dosing of probiotic strains. Despite limited data, it seems that such live biotherapeutics could be safely administered in the setting of chemotherapy- or radiotherapy-induced neutropenia (Supplementary Table S4). However, this notion is still debated, with case reports of infections, allergy, metabolic disturbances, and depression being reported after usage of specific strains administered to vulnerable patients. One retrospective study on 122 patients with operable colorectal cancer indicated that probiotic use was associated with increased CD8⁺ TIL density and better survival (Supplementary Table S4).

The effects of regular consumption of probiotics were analyzed in a pioneering study of 116 patients with advanced melanoma treated with ICI. Up to 40% of patients took food supplements during immunotherapy that appeared to be associated with a lower diversity of the microbiome repertoire and reduced response rates compared with patients who did not take them (104). Therefore, off-the-shelf probiotics may be contraindicated in immunosuppressed patients with cancer. Many strains of Lactobacilli induce tolerance (Supplementary Table S3). Not surprisingly, oral gavage with Lactobacilli failed to improve the control of mouse tumors (ref. 57; Supplementary Table S3). However, interesting results from a Japanese cohort of patients with advanced NSCLC treated with anti–PD-1/PD-L1 antibodies unveiled the potential capacity of the probiotic C. butyricum MIYAIRI 588 strain (which reportedly augments the Bifidobacteria genus and decreases Th17 cell differentiation) to restore the clinical benefit of ICI in those patients who took antibiotics within 60 days before the initiation of immunotherapy (20).

MCI may profit from the isolation of bacterial strains cultivated from healthy individuals or “elite” patients, who were cured by ICI, coupled with the precise elucidation of mechanisms of action. For example, A. muciniphila has been associated with leanness, fitness, and general health as well as successful aging and directly reduced obesity and metabolic disorders in animal models. Food supplementation withA. muciniphila reduced insulin resistance and dyslipidemia in prediabetic patients (Supplementary Table S2). Reinforcing this finding, a double-blind, randomized, placebo-controlled study demonstrated that a novel 5-strain probiotic formulation (WBF-011, which contained A. muciniphila, B. infantis, E. hallii, and 2 Clostridia spp.) significantly improved glycemia control in metformin-treated type 2 diabetes (Supplementary Table S2). A. muciniphila has been associated with clinical responses to ICI and immunostimulatory properties in tumor bearers (4, 6). One A. muciniphila strain isolated from a centenarian will be soon evaluated in patients with advanced NSCLC (26). Based on a similar rationale, several phase I/II clinical trials using bacteria endowed with intrinsic immunogenicity (such as Enterococci or Bifidobacteria strains, Clostridia spp., an 11 strain–based cocktail; ref. 31) are introduced into clinical trials enrolling patients with ICI-naïve or anti–PD-1–refractory cancer (Supplementary Table S4). Encouraging results have been reported for MRx0518, an oral E. gallinarum strain endowed with TLR5 agonistic potential, in 12 patients diagnosed with advanced kidney or lung cancers refractory to PD-1 blockade. Five of 12 patients were still on treatment after the phase I evaluation at 12 weeks, with 2 partial responses and 3 cases of stable disease (https://www.4dpharmaplc.com/en/newsroom/press-releases/comprehensive-clinical-benefit-data-part-a-mrx0518-and-keytruda).

However, here again, the strain- and tissue-associated enterofecal compatibility rules between the probiotics and the recipient microbiota remain an open conundrum. The patient's native microbiome may reject the engraftment of probiotic strains (105), and conversely, probiotics may disrupt the activity of the recipient microbiome (106), potentially causing a harmful disbalance of the microbiota ecosystem. These limitations to the development of MCI in humans will be circumvented by ancillary studies scrutinizing the serial evolution of bacterial engraftment and its pharmacodynamic effects on the holosystem (immunity, metabolism, epithelium, microbiota).

Several short-term dietary studies with gut microbiome endpoints have shown the power of diet to induce taxonomic compositional shifts in the gut microbiome (27). However, randomized controlled trials testing specific foods or nutrients have shown modest effects on the microbiota and the immune functions (107). Indeed, diet-induced modulations of the microbiome were transient due to notorious difficulties in durably adopting salutary dietary habits and lifestyle changes. Notwithstanding these shortcomings, recent lines of evidence point to the biological and clinical significance of the food–gut–immune axis for the outcome of cancer immunotherapies.

Dietary fibers are nondigestible carbohydrate polymers that are recommended at a dose of >30 g/day, preferably coming from >30 different plant species weekly to shape a healthy and diverse microbiome (108). Such diets increase the abundance of potentially beneficial species that ferment dietary fibers to produce SCFAs. Preliminary results of the first clinical studies investigating the impact of the food–gut–tumor axis on the clinical benefit to anti–PD-1 antibodies indicate that food consumption may indeed modulate the intestinal microbiome, enriching for Ruminococcacae family members and eventually influencing objective response rates. Two recent studies in patients with melanoma (35, 107) or NSCLC (109) described metagenomics and metabolomics fecal changes associated with high-fiber diets coinciding with increased response rates to ICI and TME reprogramming. Prospective clinical trials aimed at enriching diet with nondigestible fibers are currently underway (Supplementary Table S2).

Ketosis is a metabolic state in which the body retrieves energy from the catabolism of ketone bodies (KB), as opposed to glucose, and to a lesser extent, of fatty acids and amino acids. KBs include acetoacetate, 3-hydroxybutyrate (3HB), and volatile acetone. Ketogenic diets (KD) are composed of high-fat, moderate-protein, and low-carbohydrate (usually <40 g/d) diets, inducing a surge in 3HB. This increase in 3HB also occurs after fasting, favoring mitochondrial respiration rather than glycolysis for energy metabolism (110). Long-term calorie restriction in nonhuman primates reduces cancer incidence and mortality (111). In mice, short-term starvation and alternate-day fasting reduces tumor progression (112). KD differs from high-fat diet (HFD) with respect to fecal metagenomics and metabolomics. Whereas HFD increased and decreased the relative abundance of the Firmicutes phylum and Bacteroidetes, respectively, KD reversed this trend and decreased the relative abundance of Actinobacteria, in particular Bifidobacteria and Lactobacilli species, in obese male patients, as well as in mice. Interestingly, this effect could be recapitulated with oral 3HB supplementation that directly affected Bifidobacterium growth. KD and 3HB supplementation to HFD or oral gavage with B. adolescentis decreased Th17 (but not Th1 nor Treg) cells in the small intestine and in adipose tissue (113). This is in line with previous findings showing the protective effects of KD on γδT cells in the visceral adipose tissue, as well as in lungs during infection with influenza A virus (114) or a murine coronavirus (115). 3HB also has immunostimulatory effects against aggressive cancers (116). In standard diet conditions in which anti–PD-1, alone or in combination with anti-CTLA4, failed to reduce tumor growth, intermittent KD or oral 3HB supplementation restored therapeutic responses. Supplementation of KD with sucrose (which breaks ketogenesis) or with a pharmacologic antagonist of the βHB receptor GPR109A abolished the antitumor effects (116). Mechanistically, 3HB prevented the IFNγ-mediated upregulation of PD-L1 on myeloid cells while favoring the expansion of CXCR3⁺ T cells (116). KD also induced compositional changes of the gut microbiota with distinct species such as Eisenbergiella massiliensis, commonly emerging in mice and humans subjected to low-carbohydrate diets, correlating with a concomitant loss of immunosuppressive Lactobacilli (116).

It is noteworthy that the anti-inflammatory effects of fasting or KD may be linked to 3HB-mediated inhibition of the NLRP3 inflammasome (117). Indeed, tumor cell–autonomous NLRP3 was identified as a novel molecular cue driving adaptive immune evasion from PD-1 blockade (Supplementary Table S3; ref. 118). All these results point to potent immunostimulatory effects of KD or KBs during ICI, prompting the design of new clinical trials (NCT03950635).

Food supplements that can modulate the composition of the gut microbiota may be able to influence intestinal barrier fitness, the local immune system, and eventually tumor immunosurveillance. Reportedly, inulin and mucin share a common ability to promote substantial alterations in the composition of fecal microbiota, leading to increases in the relative abundance of multiple bacterial phylotypes endowed with immunogenicity (such as Bifidobacteria, Bacteroides, Barnesiella, and Parabacteroides; refs. 3, 31, 40, 41, 47, 119). Neither mucin nor inulin boosted the effect of anti–PD-1 antibodies, but inulin operated through a different mode of action than mucin, favoring the dominance of Bifidobacteria spp. in the intestine, enhancing splenic CTL functions, and overcoming melanoma resistance to MEK inhibitors (Supplementary Table S3). Likewise, distinct classes of prebiotics may antagonize the effects of immunostimulatory therapeutics. Indeed, SCFAs could antagonize the immunogenicity of local radiotherapy of tumors (84) or CTLA blockade (85). More work is needed to identify precise clinical indications for prebiotics.

There is little doubt that MCI will complement immunotherapy in the future (Fig. 4). Although most of the literature focuses on the possibility of using MCI in combination with immunogenic chemotherapies and ICI, it appears plausible that MCI will also be used in conjunction with CAR T cells, antibody–drug conjugates, and yet-to-be-developed immunotherapies. Despite tremendous progress over the last decade, there is a need for diagnostic tools for defining the “healthy microbiome,” for instance, for identifying optimal FMT donors or for stratifying patient populations. Novel diagnostic tools are also needed for the definition of immunogenic or suppressive bacteria, species-interacting groups, and the geodistribution of the intestinal ecosystem. A novel set of pharmacokinetics tools will be needed to characterize MCI, including the colonization of the gut by allogeneic bacteria and shifts in the endogenous microbiota (Fig. 3). Molecular strategies must be developed for investigating the pharmacodynamics of MCI, including their effects on intestinal barrier function and distant target tissues, such as circulating metabolites, systemic humoral and cellular immune responses against commensal or tumor antigens, and effects on the (epi)genetic and immune landscape of tumors (Fig. 4). Engineering technologies must improve ingestible MCI formulations compatible with serial delivery (Supplementary Tables S2–S4), as well as devices that sample the luminal content or perform serial biopsies in the GI tract to obtain spatially resolved information on the microbiota and the status of the gut wall. Such information may guide optimal therapeutic interventions. Supporting this concept, colonic ¹⁸F-fluorodeoxyglucose uptake before ICI initiation may predict complete responses to CTLA4 blockade in patients with metastatic melanoma (Fig. 4; ref. 120). It is our hope that such technologies will help refine the tentative recommendations (Table 1) for the development of MCI in immuno-oncology.

B. Routy has provided scientific consulting to the following companies: Vedanta, Davoltera, and Kaleido. R. Daillère is an employee and scientific cofounder of EverImmune, a biotech company. G. Kroemer reports grants from Agence Nationale de Recherche, Institut Nationale du Cancer, Ligue National contre le Cancer, Cancéropôle Ile-de-France, Fondation Association pour la Recherche sur le Cancer, and Horizon 2020 European Union outside the submitted work; in addition, G. Kroemer reports he was a cofounder of EverImmune during the conduct of the study. No disclosures were reported by the other authors.

L. Zitvogel, G. Kroemer, and R. Daillère are founders of everImmune. L. Zitvogel received research contracts from Kaleido, Innovate Pharma, and Pilege. L. Derosa was supported by Fondation Philanthropia for her PhD at Gustave Roussy.