Harnessing the Microbiome in Cancer Immunotherapy: Regulation, Prediction, and Therapeutic Targeting

1. INTRODUCTION

Humans are metaorganisms, composed of both host cells and an extensive community of commensal microorganisms—collectively termed the microbiome—that reside primarily at epithelial barrier surfaces (1–3). These microbial communities, particularly abundant in the lower gastrointestinal tract, include bacteria, archaea, fungi, viruses, and protists. Together, they encode over 100 times more genes than the human genome, contributing to critical physiological functions such as digestion, nutrient absorption, bile acid metabolism, vitamin synthesis, and immune regulation (4, 5).

The gut microbiome co-evolves with the host and participates in bidirectional interactions via microbial metabolites, immune signaling molecules, and receptor-mediated pathways (6, 7). This crosstalk influences the development and homeostasis of multiple organ systems—including the nervous, endocrine, cardiovascular, and hematopoietic systems—highlighting its central role in maintaining organismal health (8). Of relevance is its impact on immune development: the microbiome helps mature both innate and adaptive immunity, modulates cytokine profiles, and influences the function of antigen-presenting cells, T cells, and B cells (Figure 1) (9, 10).

In this context, cancer emerges as a disease not only of the host genome but of the metaorganism. The microbiome can contribute to tumorigenesis by promoting genome instability, chronic inflammation, immune evasion, and alterations in the tumor microenvironment (TME) (11–13). Specific microbial taxa, such as Fusobacterium nucleatum in colorectal cancer, have been linked to tumor progression, immune suppression, and poor clinical outcomes (14, 15). Moreover, bacteria like Escherichia coli carrying the pks pathogenicity island produce genotoxins such as colibactin, which directly induce DNA mutations and have been associated with distinct mutational signatures in human colorectal tumors (16, 17).

Microbiome composition is established early in life and influenced by delivery mode and exposure to vaginal and skin maternal microbiome; in the first 2-3 year of age, it remains plastic and modified by exposure to the microbiome of family members and social contacts and to environmental microorganisms (18–20). While the adult microbiome is resilient, it remains somewhat plastic in response to lifestyle and medical conditions and interventions, with the possibility of lateral acquisition of new strains by social contacts (21). Disruptions—termed “dysbiosis”—have been implicated in a wide range of inflammatory conditions and cancers (8, 22). However, defining a “healthy” microbiome remains challenging due to inter-individual variability, geographic heterogeneity, and reliance on fecal samples, which may not fully capture mucosa-associated microbial communities where host-microbe interactions are most critical (23, 24).

The introduction of immune checkpoint blockade (ICB), such as antibodies targeting cytotoxic T-lymphocyte associated protein 4 (CTLA-4) and programmed cell death 1/ programmed death-ligand 1 (PD-1/PD-L1), has transformed cancer therapy. These agents can elicit durable responses in subsets of patients with melanoma, NSCLC, renal cell carcinoma, and other malignancies. However, response rates remain limited, and immune-related adverse events (irAEs) can lead to substantial morbidity (25). Understanding the factors that influence both therapeutic efficacy and toxicity is essential to improve patient outcomes.

A growing body of evidence implicates the gut microbiome as a key modulator of response to immunotherapy. Preclinical studies using germ-free or antibiotic-treated mice have shown that gut microbial composition can influence the effectiveness of ICB, chemotherapy, and adoptive cell therapies through mechanisms involving immune priming, cytokine production, and antigen presentation (26–30). These findings were corroborated by clinical studies linking specific gut bacterial taxa to favorable responses to ICB in patients with melanoma and other cancers (31–33). Notably, fecal microbiome transplantation (FMT) from therapy-responsive donors into PD-1-refractory patients has restored sensitivity to ICB in approximately 40% of cases, supporting a causal relationship between microbiome composition and treatment response (34, 35). Additional microbiome-modulating strategies, including diet, probiotics, selective antibiotics, and engineered bacterial consortia, are being explored to enhance immunotherapy outcomes and reduce toxicity (36–38).

This review will examine the emerging role of the gut microbiome as a regulator, predictor, and therapeutic target in cancer immunotherapy. We will explore the mechanistic underpinnings of microbiome-immune crosstalk, assess its role in treatment efficacy and irAEs, evaluate machine learning approaches for predicting patient outcomes, and highlight clinical strategies—both current and future—for microbiome-based interventions.

2. BACTERIA AS CARCINOGENS AND TUMOR PROMOTERS

Eleven infectious agents have been classified by the International Agency for Research on Cancer as proven human carcinogens (Group 1). These include seven viruses, three macroparasites, and a single bacterial species—Helicobacter pylori—which is causally linked to gastric cancer and is also associated with mucosa-associated lymphoid tissue lymphoma (11, 39–41). H. pylori colonizes deep within the gastric glands, where it affects gastric stem and progenitor cells. Through the action of virulence factors such as the CagA toxin, H. pylori disrupts DNA repair mechanisms, promotes genomic instability, and induces hypermutation by inhibiting RAD51 expression, activating the ACVR1/IRF3/POLD1 pathway, and impairing p53 tumor suppressor functions (42, 43). In the early stages of colonization, H. pylori typically dominates the gastric microbiome (41). However, during the progression of chronic gastritis and the development of oxyntic atrophy—which leads to increased gastric pH—H. pylori levels decline and may even be eliminated. In its place, a dysbiotic microbiome emerges, often dominated by potentially pathogenic bacteria (pathobionts) such as Lactobacillus, Streptococcus, Fusobacterium, Prevotella, and various Proteobacteria spp (41). These microbes may further contribute to carcinogenesis by promoting chronic inflammation, releasing reactive oxygen species (ROS), and generating N-nitroso compounds, all of which can exacerbate DNA damage and induce mutagenesis (41).

The identification of H. pylori as a bacterial carcinogen has spurred investigation into the potential roles of other bacteria in human cancer. Several studies have analyzed the gut microbiomes of large cohorts of colorectal cancer (CRC) patients, identifying bacterial, archaeal, and fungal species enriched in individuals with adenomas and/or carcinomas. These findings have been used to develop interpretable machine learning (ML) models—based on either taxonomic or functional gene profiles—that allow for relatively accurate CRC diagnosis (44–48). Among the most consistently implicated microbes are oral pathobionts such as Fusobacterium nucleatum, Parvimonas micra, Peptostreptococcus stomatis, Bacteroides fragilis, and Proteobacteria, including Escherichia coli. These species have been found within CRC lesions, particularly those in the proximal colon, which frequently exhibit bacteria-containing biofilms (49–51). F. nucleatum has been the most extensively studied and is known to promote tumor progression, immune evasion, and poor clinical outcomes in CRC (14, 15). It contributes to carcinogenesis by modulating host cell proliferation, metabolism, and both innate and adaptive immune responses (15). Importantly, F. nucleatum is a heterogeneous species consisting of four subspecies: animalis, nucleatum, polymorphum and vincentii. Notably, F. nucleatum subsp animalis is further divided into two distinct clades, Fna C1 and C2, with only the latter (Fna C2) being associated with CRC (52). Moreover, F. nucleatum is predominantly detected in CRC lesions that exhibit biofilm formation, where it is found in large, structured communities, while it is sparse in biofilms on normal colonic mucosa (50). This supports the hypothesis that F. nucleatum, rather than initiating CRC, may migrate from the oral cavity to the colon—alongside other oral microbiome—where it preferentially adheres to the exposed neoplastic epithelium. In this environment, it forms biofilms, thrives, and promotes tumor progression through immune modulation and local inflammation (50, 53).

Until recently, there was limited evidence that bacteria could directly contribute to cancer initiation by inducing oncogenic mutations. However, three classes of bacterial genotoxins have now been identified that can cause DNA damage: cytolethal distending toxins (CDTs), typhoid toxin (from Salmonella enterica serovar Typhi), and colibactins. The first two are multi-protein virulence factors produced by over 30 Gram-negative bacterial species, while colibactins are polyketide–peptide genotoxins synthesized by Escherichia coli strains carrying the pks pathogenicity island (54). These genotoxins trigger a DNA damage response in host cells, which may lead to cell cycle arrest, senescence, or apoptosis. However, if cells survive this damage, they may accumulate permanent genetic mutations (54). The host response can also provoke chronic inflammation, promoting tumorigenesis through the generation of ROS that further damage DNA (55). Indeed, the gut microbiome has been shown to prime inflammatory cells in the TME for ROS production, enhancing DNA double-strand breaks in response to platinum-based chemotherapy (26). Salmonella Typhi has been epidemiologically linked to the initiation of gallbladder cancer, and CDTs from various bacteria have been proposed to contribute to tumorigenesis through their genotoxic effects (56, 57). For example, CDT produced by Campylobacter jejuni has been shown to facilitate metastasis formation in CRC (58). Among the most compelling evidence for a direct bacterial mutagenic effect comes from pks+ E. coli strains. These strains have been shown to induce distinct mutational signatures in human cells in vitro, which were subsequently identified in a subset of human cancers—particularly CRC (16). Colibactin-induced mutations were more frequent in patients from countries with high CRC incidence, and notably in cases of early-onset, left-sided CRC with favorable prognosis and APC driver mutations (17, 59), These findings suggest that early-life exposure to colibactin-producing bacteria may initiate tumorigenesis by accelerating age-associated somatic mutation accumulation.

DNA is constantly subjected to damage from both endogenous and exogenous mutagens, including those derived from commensal and pathogenic microorganisms. As a result, each cell acquires thousands of somatic mutations with age, including alterations in oncogenic driver genes (60). While somatic mutations are found across all tissues, they occur more frequently—and with distinct mutational signatures—in tissues exposed to specific environmental mutagens, such as ultraviolet light in skin, tobacco smoke in the lungs and chemotherapy agents in hematopoietic cells (60). In hematopoietic tissues, clonal hematopoiesis can emerge, wherein a small number of mutated clones expand and dominate the population. In solid tissues, however, clonal expansion is often limited, possibly due to intrinsic cellular competition or surveillance by the innate and adaptive immune system (61, 62). Despite these constraints, some mutated cells escape control and ultimately progress to cancer. The gut microbiome may play a role in amplifying mutational burden, as demonstrated by the involvement of E. coli and F. nucleatum in CRC. Notably, age-associated changes in microbiome composition may influence mutagenesis. For example, gram-negative bacteria can produce ADP-heptose, a metabolite that impairs DNA repair mechanisms and binds to the ALPK1 receptor, promoting NF-kB–driven inflammation and enhancing the expansion of pre-leukemic clones (63). In murine models, variations in microbiome composition across animal colonies influence lymphoma development and survival in ATM-deficient mice by modulating inflammation, oxidative stress, and genotoxicity (64). Similarly, lung tumor formation in Kras-mutant, Tp53-deficient mice is suppressed under germ-free conditions, and colitis-associated colon carcinogenesis is attenuated in laboratory mice colonized with microbiome from wild mice, suggesting that the microbiome modulates cancer risk in multiple tissues (65, 66). Commensal microbiome can also shape systemic immunity in ways that promote tumor progression. For instance, activation of TLR5 by gut microbiome induces immunosuppressive myeloid cells, facilitating tumor growth at distant sites (67). In the liver, microbiome-derived secondary bile acids inhibit the recruitment of tumor-suppressive natural killer T cells, while lipopolysaccharide (LPS) from Proteobacteria spp drives the accumulation of tumor-promoting myeloid cells (68, 69).

3. HUMAN MICROBIOME DIVERSITY AND PLASTICITY

In healthy pregnancies, initial exposure to bacteria and other microorganisms occurs at birth (19, 70) (Figure 2). The infant microbiome is gradually shaped through contact with the maternal microbiome across various tissues, as well as through exposure to the microbiomes of family members, social contacts, pets, and the surrounding environment (19, 20, 70, 71). Due to limited alpha diversity in early life, the microbiome remains highly plastic during the first few years, but it subsequently stabilizes and becomes relatively resilient (19). Nevertheless, microbial strain transmission from close contacts remains possible throughout life, and changes in microbiome composition in later years can influence health, morbidity, and longevity (19, 72, 73). Microbiomes are more similar among individuals who are closely related or socially connected, and diverge progressively with increasing social distance, geographic separation, or limited interaction (19, 74). Lifestyle factors—such as diet, medication use (including antibiotics, probiotics, and acid suppressants), infections, and disease—can temporarily or permanently alter microbiome composition, disrupting the gut ecosystem and influencing health outcomes (19).

The high taxonomic heterogeneity of the gut microbiome presents a major challenge in identifying disease-associated microbial signatures and often hinders the development of reliable diagnostic or predictive biomarkers (75). Geographical variation is also a critical contributor to microbiome diversity, with studies showing greater variability between cohorts from different regions than within individual cohorts (20, 76, 77). Despite migration, microbial profiles can persist across ethnic groups for many years, indicating that lifestyle, dietary practices, vertical transmission of commensals, and other poorly understood factors contribute to long-term and transgenerational microbiome resilience (78–81). Therefore, to successfully identify microbiome-based biomarkers for treatment response and to develop personalized therapeutic strategies, it is essential to understand and account for the many variables that shape individual microbiome composition.

4. THE GUT MICROBIOME MODULATES CANCER IMMUNOTHERAPY OUTCOMES

About a decade ago, a series of murine studies clearly demonstrated that the gut microbiome profoundly influences the efficacy of cancer chemotherapy and immunotherapy. These effects occur through priming of inflammatory cells to produce cytokines and ROS, or through direct or indirect licensing of adaptive immune lymphocytes to mediate antitumor immunity (26–29). In 2018, three seminal clinical studies (31–33), followed by many others since (12), described that specific gut bacterial taxa were associated with either favorable or unfavorable clinical responses to anti-PD-1 therapy in patients with melanoma and epithelial cancers. However, the specific taxa identified varied significantly across studies, with little taxonomic or functional overlap, which introduced uncertainty about which microbial features are truly predictive of response (76). Nonetheless, confidence in the conclusion that the gut microbiome modulates immunotherapy outcomes in humans was greatly enhanced by functional studies using germ-free or antibiotic-treated mice. These experiments showed that FMT from responder or non-responder patients into microbiome-depleted mice recapitulated the donor’s clinical response, confirming a causal role for the microbiome (31–33). Further, adding back specific bacterial taxa associated with favorable responses into non-responder-derived microbiomes restored responsiveness to ICB in mice, reinforcing the evidence for therapeutic potential of microbiome manipulation (32, 33). Meta-analyses of multiple patient cohorts have begun to reveal consistent patterns across studies investigating the relationship between the gut microbiome and cancer immunotherapy outcomes, not only at the level of specific bacterial species but also at higher taxonomic ranks and in functional pathways inferred from microbial gene content (76, 82–84). When data from different studies of melanoma patients treated with anti-PD-1 were aggregated, taxa associated with favorable responses to therapy included multiple species from the phylum Actinobacteria and from the families Oscillospiraceae and Lachnospiraceae within the phylum Firmicutes (76). In contrast, non-responders were enriched in taxa belonging to the Bacteroidetes and Proteobacteria phyla (76).

One plausible explanation for the initially discordant results among individual studies is that while specific bacterial species were inconsistently represented across cohorts, taxa identified at higher taxonomic levels may share convergent functional traits relevant to immunotherapy response. Indeed, beta-diversity analysis revealed that inter-cohort variation in gut microbiome composition was greater than the variation observed between responders and non-responders within individual cohorts. Additionally, both patient and healthy donor microbiomes clustered into distinct enteric microbiotypes, characterized by differing taxonomic dominance and unequal geographic distribution (76). Importantly, the probability of response to ICB was not uniformly distributed across these microbiotypes, suggesting that geographic and ecological factors contribute significantly to therapeutic outcomes (76). These findings help explain why taxa associated with response may vary by region, and they underscore the importance of accounting for batch effects and geographic variability when conducting multi-cohort analyses. Supporting this, a large phase III adjuvant trial comparing combination anti-PD-1/CTLA-4 to anti-PD-1 monotherapy, conducted across three continents and five regions, demonstrated that different pre-treatment taxonomic markers were associated with recurrence in different regions. (85). Notably, recurrence prediction remained robust when intra-regional training and external validation datasets were matched based on microbiome compositional similarity, highlighting the potential for regionally informed, microbiome-based predictive models (85).

Differences in DNA purification methods, sequencing strategies, and bioinformatic pipelines are likely to contribute to the variability in bacterial taxa reported to be associated with favorable or unfavorable clinical responses to immunotherapy. In addition, variations in clinical immunotherapy protocols—which rely on distinct immunological mechanisms—are also likely influenced by microbial pathways associated with different bacterial species (12, 86). Furthermore, association of bacterial taxa with a clinical response does not prove causation and precise identification of the species involved and of their mechanism of action will be very important for a rational planning of therapeutic protocols targeting the microbiome to improve clinical response to cancer immunotherapy (87, 88). Although many seminal studies have uncovered specific mechanisms by which certain bacterial strains modulate cancer immunity, most of this work has been conducted in experimental animals. These models have limitations, as they often fail to fully replicate the clinical and immunological context of cancer patients, and important differences exist between human and murine immune systems. Thus, these results may not always be translatable to the clinic as corroborated by the difficulty in clinical trials to reproduce promising mouse results by treatment with single taxa or consortia of few taxa (Figure 1) (87, 88).

While most microbiome–immunotherapy studies have focused on bacteria, growing evidence points to important roles for viruses and fungi. The gut virome—primarily composed of bacteriophages and enteric viruses—remains poorly characterized. A recent metagenomic study identified nearly 200,000 viral sequences in gut samples, of which only ~5% matched known phage genomes, underscoring the virome’s unexplored diversity (89). Early-life virome composition can shape immune development and influence childhood asthma as strongly as the bacteriome (90). In mice, murine norovirus (MNV) mimics microbiota-driven immune stimulation via type I IFNs and IL-22 in germ-free conditions (91–93), but can exacerbate colitis in microbiota-sufficient mice through NOD2 activation (91, 94). In cancer, virome elements influence immunity: HCMV infection in lung cancer promotes senescent CD8⁺ T cells linked to anti–PD-1 resistance (95), and endogenous retroviruses (ERVs) reactivated in tumors can act as antigens recognized by T cells (96, 97). ERVs also mediate microbiota-induced skin inflammation (98), and the microbiome can enhance MuLV carcinogenicity via NOD1/2 signaling (99). Fungi are likewise implicated in cancer and mucosal immunity. Loss-of-function mutations in AIRE or gain-of-function mutations in STAT1—linked to chronic mucocutaneous candidiasis—are associated with increased risk of oral and esophageal cancers, a link confirmed in Ikka-deficient mouse models (100–102). Specific intratumoral fungi have been associated with tumor progression and immune modulation, although low microbial biomass complicates detection (103). In radiotherapy, antibiotic-induced bacterial depletion impairs anti-tumor immunity and drives fungal overgrowth that activates Dectin-1 in the TME (104). Importantly, including mycobiome data improves machine learning prediction of immunotherapy response, and fungal enterotypes help identify FMT donors capable of transferring anti–PD-1 sensitivity in mice (105, 106).

5. MICROBIOME-DERIVED LIGANDS AND METABOLITES ENHANCE ANTI-TUMOR IMMUNITY VIA DIVERSE IMMUNOMODULATORY PATHWAYS

Many bacteria have been shown to produce ligands for innate immune receptors that enhance the efficacy of ICB in tumor therapy. For example, muropeptides derived from Enterococcus spp. can activate NOD2, promoting antitumor immune responses (107, 108). Similarly, certain strains of Bifidobacterium bifidum produce peptidoglycans that stimulate cancer immunity via TLR2 signaling (109). Constitutive expression of type I interferons (IFNs), which supports antitumor immunity (110–112), has been observed to depend on the gut microbiome. In germ-free mice, or those treated with antibiotics to deplete the microbiome, this basal type I IFN signaling is markedly reduced, impairing immune responsiveness (26, 37). Bacterial products induce IFN production in a variety of cell types through innate receptors such as TLR3, TLR4, TLR7, TLR9, Dectin-1, stimulator of interferon genes (STING), DDX41 and RIG-I (37, 113–117). Notably, commensals such as Akkermansia muciniphila, Bifidobacterium spp., and Lactobacillus rhamnosus have been shown in mice to activate the STING–IFN signaling axis, which correlates with favorable responses to ICB in melanoma patients (37, 116, 117). However, it is essential to note that type I IFN signaling can exhibit dual roles, as it has been associated with resistance to ICB therapy in certain experimental contexts (118–121). Clinical trials of STING agonists have so far shown modest efficacy (122).

The earliest experimental evidence linking the microbiome to immunotherapy came from murine studies demonstrating that the potentiating effect of total body irradiation in adoptive T cell therapy (ACT) was due to bacterial translocation of LPS-producing microbes across the gut epithelium, enhancing immune activation (30). Similarly, vancomycin-induced expansion of Proteobacteria in the gut of mice was shown to improve ACT efficacy by stimulating IL-12 production from type 1 conventional dendritic cells (123). In clinical settings, leukemia patients treated with engineered T cells expressing chimeric antigen receptors (CAR-T) often receive prophylactic antibiotics to prevent infection. However, broad-spectrum antibiotic use that reduces microbial alpha diversity or depletes anaerobes has been associated with lower remission rates and decreased survival following CAR-T treatment (124–127). In contrast, enrichment of short-chain fatty acid (SCFA)–producing bacteria improves outcomes of CAR-T therapy. SCFAs, through their ability to inhibit histone deacetylases (HDACs), enhance IL-12 production, thereby augmenting anti-tumor immune responses (127–129).

Microbial metabolites are emerging as key modulators of anti-tumor immunity and the efficacy of ICB. Certain bacteria, including Bifidobacterium pseudolongum, Lactobacillus johnsonii, and Akkermansia muciniphila, produce inosine and its downstream metabolite hypoxanthine, which bind to the adenosine A2a receptor and enhance anti-CTLA-4 therapy in preclinical melanoma models (130). Similarly, a defined consortium of 11 bacterial species that promotes T cell–mediated anti-tumor activity also induces hypoxanthine accumulation in colonized mice (131). Beyond receptor binding, inosine also supports cancer immunity metabolically, acting as an alternative carbon source for CD8⁺ T cells under nutrient-limited conditions (132). Another class of microbiome-derived metabolites, SCFAs—including butyrate, propionate, and acetate—are generated in the colon through the fermentation of dietary fiber. SCFAs serve as an energy source for colonocytes, but they are also absorbed systemically, where they exert anti-inflammatory and mucosal protective effects (8). While SCFA-producing bacteria have been repeatedly associated with favorable responses to ICB, the precise immunomodulatory functions of individual SCFAs in the context of cancer immunity remain to be fully elucidated (128, 133–136). Additional microbiome-derived metabolites have recently been implicated in modulating immunotherapy response. For instance, trimethylamine N-oxide (TMAO), a product of microbial choline metabolism, was found to be elevated in immunotherapy-responsive patients with pancreatic and triple-negative breast cancer, and shown in mouse models to act by activating macrophages via type I IFN signaling or enhancing CD8⁺ T cell function by inducing tumor cell pyroptosis (137, 138). Tryptophan metabolites represent another class of microbiome-influenced immunomodulators. The metabolite indole-3-propionic acid (IPA), produced by Lactobacillus johnsonii in cooperation with Clostridium sporogenes, enhances immunotherapy-induced anti-cancer immunity in mouse models by promoting Tcf7 expression and maintaining the stemness of progenitor exhausted CD8⁺ T cells (139). Other tryptophan catabolites such as indole-3-acetic acid (3-IAA) and indole-3-aldehyde (I3A) have been found in higher concentrations in patients responding to chemo-immunotherapy (139–141). Notably, 3-IAA enhances chemotherapy by inhibiting tumor anti-oxidative defenses (139) whereas I3A, partially produced by Lactobacillus reuteri, has dual roles: stimulating CD8⁺ T cells while suppressing pro-inflammatory macrophage phenotypes via aryl hydrocarbon receptor signaling (140, 141).

Exercise has been shown to slow tumor growth and enhance the efficacy of immune checkpoint blockade (ICB) by boosting T cell–mediated immunity in both mice and humans (142, 143). While exercise alters gut microbiome composition, its role in mediating anti-tumor effects via microbial pathways was unclear (144). A recent study demonstrated that exercise reshapes the microbiome to stimulate microbial one-carbon metabolism, leading to increased formate production, which enhances CD8⁺ T cell cytotoxicity and improves response to anti–PD-1 therapy (145). This effect is mediated through Nrf2, not the aryl hydrocarbon receptor. Notably, in melanoma patients treated with anti-PD-1, clinical response correlated with higher abundance of the microbial gene encoding pyruvate formate-lyase (pfl), the enzyme responsible for producing formate from pyruvate (145).

In another mechanism of microbiome-immune interaction, commensal bacteria such as Coprobacillus cateniformis have been shown to enhance anti–PD-(L)1 responses by downregulating PD-L2 and its binding partner repulsive guidance molecule b (RGMb) (146). These findings suggest that targeting both PD-1/PD-L1/2 and PD-L2/RGMb axes may provide a therapeutic strategy to overcome microbiome-associated resistance to ICB. However, such approaches may not fully counteract the broader immunosuppressive effects of a dysbiotic microbiome, which may operate through multiple redundant mechanisms.

6. INTRATUMORAL BACTERIA AFFECT TUMOR PROGRESSION AND RESPONSE TO THERAPY

The presence of intratumoral bacteria, mostly present intracellularly in tumor and infiltrating cells, and fungi has been demonstrated in many human tumors including breast, bone, pancreatic, ovarian and lung cancers, melanoma, glioblastoma and primary and metastatic colorectal carcinoma (147–152). Except for tumors originating from epithelial barriers such as CRC, the number of bacteria present in the tumor is, however, extremely low and estimated to be in most patients one bacterium every 1000 to 100 tumor cells even in pancreatic cancer patients who appear to harbor the largest number of intratumoral bacteria (153). Thus, the expected amount of bacterial DNA vs human DNA in a tumor is expected to be 1 vs 10⁵/10⁶. Because of this extremely low bacterial biomass and the unavoidable presence of environmental, laboratory and reagent contaminants, the technical and bioinformatics identification of intratumoral bacteria remains difficult and has been at the origine of numerous artifacts and inaccuracies (154, 155). Thus, while the concept of the presence of an ecologically balanced microbiome within tumors is not any longer widely accepted, individual bacterial species or small consortia with mutualistic interactions are likely present within tumors where they may play a role in oncogenesis, tumor progression, metastasis formation and response to therapy (156, 157). Bacterial cytidine deaminase in pancreatic cancer associated Gammaproteobacteria inactivates the cytotoxic drug gemcitabine and in a mouse model the antibiotic ciprofloxacin re-establishes sensitivity to the drug (148). In some cases, intratumoral bacteria have been shown to affect the response to immunotherapy. The modulatory effects on immunotherapy by the TMAO and I3A metabolites produced by tumor associated bacteria have been discussed above (137, 140). In mouse models, the presence of bacteria in tumors attracted immunosuppressive neutrophils: in lung tumors bacteria via IL-1b and IL-23 production stimulated IL-17-producing T cells whereas in cholangiocarcinoma LPS-producing Proteobacteria released CXCR2-binding chemokines (65, 69). Similarly, in human CRC, CXCL8 induction by intratumoral F. nucleatum attracts immunosuppressive neutrophils (158). Microbial peptides from human tumor associated bacteria have been shown to activate tumor infiltrating lymphocytes upon presentation by HLA class II in glioblastoma and both class I and II in melanoma, suggesting that bacteria may be involved in the activation and/or adjuvancy of cancer immunity and responses to immunotherapy (159, 160). However, because of the possibly dominant role of tumor-associated bacteria in carcinogenesis, metastasis formation and immunosuppression, elimination of tumor associated bacteria may be expected to have an overall beneficial role in slowing tumor progression or boosting immunotherapy response. In mice, vancomycin and neomycin aerosolization reduced immunosuppression and prevented tumor implantation while aerosolization of a vancomycin/neomycin resistant probiotic, such as Lactobacillus rhamnosus GG, boosted anti-tumor resistance (161). Clinical trials of aerosolized antibiotics are ongoing in patients with lung cancer (NCT05777603). However, intracellular bacteria that escape RAB11-mediated phagocytosis may localize in stable vacuoles where they are resistant to antibiotics (162). Therefore, using locally cell membrane penetrating antibiotics such as doxycycline or combined treatment with cell-penetrating peptides and nucleic acids may maximize the efficacy of antibiotics against tumor-associated bacteria while avoiding the negative effect on immunotherapy of systemic antibiotic treatment (163–165).

7. THE GUT MICROBIOME OF CANCER PATIENTS IS OFTEN ASSOCIATED WITH INFLAMMATION AND IMMUNE DYSREGULATION

An eubiotic gut microbiome—a balanced and diverse community of microorganisms—plays a vital role in supporting digestion, regulating immune responses, and defending against pathogens. In cancer patients, this balance is frequently disrupted, resulting in dysbiosis—a state that promotes chronic inflammation, facilitates tumor progression, and impairs the efficacy of immunotherapies such as immune checkpoint blockade (ICB) (46, 166, 167). Consequently, restoring or modulating the gut microbiome has emerged as a promising strategy to improve cancer treatment outcomes. Dysbiosis can arise not only from the malignancy itself but also as a consequence of cancer-related interventions, including chemotherapy, radiation, antibiotic exposure, and lifestyle changes such as dietary alterations and psychological stress (46, 167). This imbalance is particularly pronounced in patients undergoing hematopoietic stem cell transplantation, where intensive conditioning regimens and prolonged immunosuppression often lead to a profound loss of microbial diversity (168). Such depletion increases susceptibility to infections, promotes graft-versus-host disease, and is associated with poorer treatment outcomes (168).

Microbiome alterations in cancer patients often mirror those observed in other inflammatory or immune-mediated conditions, such as inflammatory bowel disease or infections (46, 80). This overlap in microbial signatures—including loss of diversity, expansion of pro-inflammatory taxa, and depletion of beneficial commensals—makes it challenging to identify disease-specific microbial markers that can reliably distinguish cancer-related dysbiosis from other forms of inflammation (46, 80). To overcome the limitations of non-specific microbial signatures, researchers have begun characterizing bacterial “guilds”—functionally defined groups of microbes that exert similar effects on host-microbiome interactions (169). Health-associated guilds are typically enriched in genes responsible for producing SCFAs, such as butyrate, which help maintain gut barrier integrity and regulate immune responses (169). In contrast, disease-associated guilds often exhibit an overrepresentation of virulence factors, pro-inflammatory pathways, and antibiotic resistance genes, indicating a shift toward dysbiosis and pathogenicity (76, 169, 170).

Dysbiosis can promote chronic inflammation and impair responses to ICB by recruiting regulatory T cells and immunosuppressive myeloid cells (171, 172). Markers of systemic chronic inflammation such as high neutrophil-to-lymphocyte ratio (NLR), interleukin-8 (IL-8), serum amyloid A, and C-reactive protein (CRP) correlate with poor responses to ICB, whereas interferon-induced markers like IP-10 are associated with better outcomes (173–176). In melanoma, increased abundance of Bacteroidetes correlates with a higher NLR and reduced efficacy of ICB (76). High NLR has been consistently associated with poor response to immunotherapy, according to a meta-analysis of over 100 studies involving nearly 20,000 cancer patients (177). High levels of CRP and low serum albumin are associated with poor response to ICB. These two markers are included in the Glasgow Prognostic Score, a robust predictor of resistance to ICB (178, 179).

Gene expression profiles from inflammatory and epithelial cells shed in the small intestine can be evaluated using shotgun metatranscriptomic analysis of human fecal samples (76). Analysis of human transcripts in fecal samples from ICB-resistant patients revealed an NF-kB-driven inflammatory signature, in part triggered by LPS (76). In mice, the translocation of gram-negative Proteobacteria to the liver recruits suppressive neutrophils (69). Similar immunosuppressive effects have been observed with Fusobacterium nucleatum in colorectal cancer (180). Early in tumor development, transient inflammation disrupts the intestinal barrier, leading to lasting dysbiosis dominated by Clostridia spp, which further promote tumor growth (167). Antibiotic use prior to ICB can worsen outcomes by expanding Enterocloster spp, which inhibit the mucosal adhesion molecule MAdCAM-1 and facilitate the migration of Treg17 cells into tumors (165, 181). Immunotherapy itself can alter the microbiome. Data from melanoma models suggest that anti-PD-1/CTLA-4 therapy promotes bacterial translocation, which activates dendritic cells and enhances T-cell priming, whereas antibiotics suppress this effect (182). Within tumor-associated bacteria, some, like Lactobacillus reuteri, support anti-tumor immunity by producing I3A, while others, like Fusobacterium nucleatum, promote immunosuppression and metastasis (140, 158). These findings underscore the microbiome’s context-dependent dual role in cancer and immunotherapy, highlighting its potential as a therapeutic target.

8. MACHINE LEARNING PREDICTION OF RESPONSE TO THERAPY

Given the critical role of the gut microbiome in modulating responses to cancer therapy, ML models have been employed to predict outcomes in patients receiving ICB. Early models demonstrate only modest predictive performance, with area under the receiver operating characteristic curve (AUROC) values rarely exceeding 0.7 (82, 83). However, prediction accuracy improves significantly when microbiome functional features—rather than taxonomic composition—are used, highlighting the importance of microbial activity and metabolic potential in shaping therapeutic response. (82, 83, 183). Two studies that achieved higher predictive performance (AUROC > 0.8) using refined taxonomic indices and batch-corrected, pooled microbiome datasets, suggested that microbiome signals linked to poor outcomes are more consistent than those linked to favorable responses (76, 84). Significantly improved predictive performance was observed when machine learning models incorporated not only bacterial taxonomic or genomic data but also complementary information from gut fungi (mycobiome) and bacterial metabolites, highlighting the value of multi-kingdom and multi-omic integration in modeling ICB therapy response (105, 106, 184, 185). Notably, models trained exclusively on microbiome data have demonstrated greater accuracy in predicting clinical responses to immunotherapy than conventional ICB biomarkers, such as the expression levels of immune checkpoint targets, tumor mutational burden, or markers of the TME. In addition, several promising predictive models based on easily obtainable clinical parameters or routine blood tests have emerged (186, 187). These models in part utilize biomarkers such as NLR and blood albumin levels, which have been shown to correlate with gut microbiome composition (76, 188), further underscoring the systemic influence of the microbiome on host immunity and treatment outcomes.

Overall, the findings from ML models and biomarkers provide strong evidence for a robust association between specific microbiome biomarkers and responses to ICB. However, establishing a definitive causal relationship between individual microbial taxa or functions and treatment outcomes remains an ongoing challenge. Emerging approaches such as Mendelian randomization, which leverages host genetic variants associated with gut microbiome composition, and causal-inference frameworks like transkingdom network analysis applied to human datasets, are beginning to shed light on the potential mechanistic roles of specific microbes (37, 189–191). These computational insights, when coupled with experimental validation in germ-free or gnotobiotic mouse models, help to build a stronger case for the causal influence of particular microbial species or their metabolites on immunotherapy efficacy (37, 189–191). Due to the intraspecies heterogeneity in the presence, function, and expression of metabolic genes, accurately identifying bacterial functions that influence immunotherapy response will require a focus on genomic differences at the strain level. This includes examining structural variations in gene products as well as environmental factors that regulate gene expression (76, 192–194).

In NSCLC patients treated with immunotherapy, ML models trained on metagenomic data have demonstrated limited predictive power, largely due to the low taxonomic overlap observed across different patient cohorts. (195). However, species-interaction networks revealed two distinct microbial groups: one associated with longer survival and another linked to poor outcomes. These findings formed the basis of the TOPOSCORE, a composite biomarker that integrates the presence or absence of species from these two clusters along with a tripartite scoring system for Akkermansia muciniphila abundance. The TOPOSCORE demonstrated moderate predictive power for overall survival (OS) in patients with NSCLC and genitourinary cancers but showed limited predictive ability in melanoma (195). Microbial biomarker scores, such as the TOPOSCORE, which are developed by identifying defined groups of taxa using methods like polymerase chain reaction (PCR), may represent a feasible clinical test for predicting response to immunotherapy (195).

Geographic variability in microbiome composition and enterotypes represents a major confounding factor affecting the generalizability of prediction models; however, this can be addressed through appropriate correction methods within the models (76, 184). Indeed, a large multicenter trial in stage III melanoma demonstrated that, without batch correction, ML models were predictive only within individual centers or among patients with similar microbiome profiles (105, 196). Cancer- and therapy-specific microbiome signatures may underlie the observed differences in microbiome associations across tumor types such as melanoma, NSCLC, and genitourinary cancers (32, 33, 76, 195, 197). Therapy-specific microbiome interactions have also been identified. For example, in melanoma, TLR9 agonists depend on gram-negative bacteria to transform immunosuppressive myeloid environments into immunostimulatory ones, while the same bacteria are associated with an unfavorable response to anti-PD-1 (198). Similarly, response to anti-CTLA-4 therapy (with or without PD-1 blockade) has been associated with Bacteroidetes and Faecalibacterium spp, a microbiome profile not optimal for anti-PD-1 response (29, 199–201). Recent data also demonstrate that ML models trained on microbiome data matched to specific ICB regimens but not those trained on data from a different regimen can predict treatment response across diverse geographic cohorts (199).

Importantly, some microbiome signatures are predictive of overall survival or disease progression rather than direct therapy response. For example, microbial guilds associated with age-related frailty or chronic diseases may serve as more prognostic markers than predictive biomarkers of treatment efficacy (80, 169, 195). Distinguishing between models that predict treatment response and those that forecast disease trajectory is crucial, particularly when these models are used for patient stratification or donor selection in microbiome-based interventions. The slow progress in clinical translation may be attributed to several factors: fundamental differences between human and mouse innate and adaptive immunity, especially when relying on specific pathogen-free or microbiome-depleted mouse models (202); poor colonization in humans by the bacterial species used in preclinical studies; and the effects of antibiotic preconditioning in patients (203). Based on our advancing understanding of gut microbiome ecology, consortia of symbiotic bacterial species can be rationally designed to create therapeutic formulations that promote efficient colonization and long-term persistence of the transferred microbes (204).

9. TARGETING THE GUT MICROBIOME TO ENHANCE OR RESTORE IMMUNOTHERAPY RESPONSE

Robust evidence from clinical studies indicates that the composition of the gut microbiome is linked to immunotherapy response and toxicity. Moreover, transferring individual bacterial taxa or larger consortia linked to favorable responses into tumor-bearing mice can enhance their ability to respond to immunotherapy (32, 33, 131). These findings have encouraged the design of clinical trials targeting the microbiome to enhance anti-tumor efficacy and reduce treatment-related toxicity (Figure 3). Initial clinical trials have shown promising results, indicating that the administration of Clostridium butyricum MIYAIRI 588, a probiotic strain commonly used in East Asia for digestive tract disorders, enhances the response to combined ICB in lung and kidney cancers (205). C. butyricum is believed not only to colonize the gut but also to reach the TME, where it produces butyrate. This metabolite inhibits indoleamine 2,3-dioxygenase, thereby contributing to CD8⁺ T cell activation by preventing tryptophan degradation and the accumulation of kynurenine (206). However, numerous clinical trials in recent years using individual bacterial strains or consortia, including over-the-counter probiotics that demonstrated activity in experimental animal models (32), have not reported beneficial clinical outcomes in immunotherapy (36).

The most compelling proof-of-concept for targeting the gut microbiome to enhance immunotherapy efficacy comes from two seminal single-arm clinical trials, in which FMT was administered via colonoscopy from PD-1-responsive melanoma donors to PD-1-refractory patients (Table 1) (34, 35). Despite differences in trial design, including the inclusion of only primary PD-1-refractory patients or not, the use of antibiotic preconditioning, and the administration of oral FMT capsules post-treatment, both studies reported clinical benefit in approximately 40% of treated patients. These benefits included RECIST 1.1-defined radiographic responses and prolonged stable disease (34, 35). Importantly, both trials demonstrated durable shifts in gut microbiome composition following FMT. In patients who responded, there was a notable enrichment of bacterial taxa previously linked to favorable anti-PD-1 responses, accompanied by improvements in immunological markers indicative of enhanced anti-tumor immunity (34, 35). Similar clinical results were reported in a later similar trial in patients with gastrointestinal tract cancers (207).

In multicenter, single-arm phase I trials, FMT from individual healthy donors was administered as a single oral dose, without antibiotic preconditioning, one week before first-line treatment of advanced melanoma patients with anti-PD-1 (n=20) or combined anti-PD-1/CTLA-4 (n=20) therapy and treatment of advanced NSCLC patients with PD-L1 positivity of >50% (n=20) with anti-PD-1 (208, 209). The response rate in the three groups were 65%, 75% and 80%, respectively, much higher than expected from historical controls (208, 209). As in other FMT trials in cancer immunotherapy (35), favorable clinical responses were associated with an enrichment of beneficial taxa such as Ruminococcus, Eubacterium, and Faecalibacterium spp, alongside a decrease in unfavorable taxa like Enterocloster and Catabacter (208). In a phase two randomized trial, 50 patients with metastatic renal cell carcinoma receiving anti-PD-1 therapy combined with a tyrosine kinase inhibitor were randomized to receive either FMT—consisting of a single donor fecal infusion via colonoscopy followed by oral capsules—or placebo (210). Progression-free survival (PFS) at 12 months (n=44) was 66.7% in the FMT group compared to 35% in the placebo group (p = 0.036) (210).

Although all FMT studies to date have reported that responder patients exhibit a shift in gut microbiome toward a stable taxonomic composition, enterotype classification, and metabolomic profile supportive of anti-tumor immunity (184), the overall proportion of donor strain engraftment did not differ significantly between responders and non-responders (211). This suggests that recipient-derived strains often persist post-transplant, adapting their ecological abundance, while additional strains may be acquired from the surrounding environment (211). Collectively, these trials support the potential clinical utility of FMT either as a concurrent first-line treatment with ICB or as a salvage strategy in cases of ICB resistance. However, a key concern in the first-line setting is the possibility that patients who would have otherwise responded to ICB might fail to benefit— or even experience detrimental effects—if paired with an incompatible or mismatched microbiome donor. As discussed, ML approaches hold promise for guiding personalized immunotherapy by identifying the most effective ICB strategy for individual patients and assessing whether FMT could enhance their likelihood of response. ML models have already been developed to predict post-FMT microbiome composition based on donor and recipient microbiome features (211), offering a potential framework for rational donor selection. Such models could be integrated into clinical workflows to match patients with the most compatible donors, thereby optimizing FMT efficacy and minimizing the risk of unfavorable outcomes.

Although FMT is FDA-approved for preventing recurrent Clostridioides difficile infection and is being explored in various other clinical settings, its application in cancer patients remains challenging due to standardization issues and potential safety concerns. A more controlled and potentially safer approach involves the use of single bacterial strains or well-defined microbial consortia. However, the clinical translation of these strategies has been sluggish. This may stem from fundamental differences in innate and adaptive immune responses between humans and preclinical models, especially specific pathogen-free or germ-free mice (66), as well as from the possibly limited colonization efficiency of administered strains in the human gut. Additionally, antibiotic preconditioning, often used to facilitate engraftment, may inadvertently alter the gut environment in ways that hinder the establishment of a stable microbiome (203). Building on our growing understanding of gut microbiome ecology, it is now feasible to design consortia of symbiotic bacterial species as therapeutic formulations that promote efficient colonization and long-term persistence of introduced microbes (204). However, it is essential to acknowledge that bacterial strains and mechanisms identified as beneficial in mouse models may not demonstrate the same efficacy in human patients. Their therapeutic potential may be limited by an underlying immunosuppressive dysbiosis present in some individuals, which could dampen microbiome-driven immune activation. In this context, FMT has been proposed as a strategy to reprogram the gut microbiome in patients refractory to ICB, thereby restoring a more balanced and immunostimulatory ecosystem. This reconfiguration may help counteract chronic inflammation–driven dysbiosis and re-sensitize tumors to immunotherapy (35, 76, 88).

The identification of gut microbiome alterations that favor or impede the effectiveness of CAR-T cell therapy in hematopoietic malignancies suggest the possibility of targeting the microbiome composition to improve therapy (126), FMT trials in CAR-T cell therapy have been registered (NCT06218602, NCT07042438) but results have not yet been reported.

Other promising strategies to modulate the gut microbiome in cancer patients include the use of selective antibiotics and bacteriophages, which allow targeted depletion of specific microbial populations (68, 212). Another innovative approach involves engineering bacterial strains to produce chemokines or metabolites that promote T cell recruitment and function within the TME, offering a novel route to enhance anti-tumor immunity (206, 213). Simpler and more cost-effective alternatives include dietary interventions, prebiotics, and postbiotics. For example, high-fiber diets have been shown to improve ICB responses in both patients and mouse models by reshaping gut microbiome composition (36) and surpass FMT in reversing antibiotic-induced dysbiosis in murine studies (214). Specific prebiotics, such as inulin and pectin, further support anti-tumor immunity and enhance the efficacy of immune checkpoint blockade (36, 37). Castalagin, a polyphenol in the prebiotic berry camucamu, enhances tumor immunity by modifying the microbiome and is being clinically investigated (38). Moreover, microbial metabolites and other postbiotics (such as inactivated bacteria, extracellular vesicles, and bacterial cell wall components) with immunomodulatory properties could be developed as standalone therapeutic agents (215). Ultimately, the signaling pathways through which the microbiome influences immunotherapy outcomes may provide additional targets for pharmacological intervention using pathway-specific agonists or antagonists.

10. GUT MICROBIOME AND irAEs

Although ICB has achieved significant clinical success in cancers such as melanoma and NSCLC, its use is frequently accompanied by irAEs. These toxicities, particularly severe with combination regimens like anti-CTLA-4 plus anti-PD-1, can cause substantial morbidity and, in some cases, require treatment discontinuation (216). Interestingly, several studies report a positive correlation between irAEs and improved PFS and OS, suggesting that overlapping immune mechanisms may drive both tumor clearance and autoimmunity (217–219). However, this association is not universal, and other studies indicate that irAEs can occur independently of therapeutic benefit (76, 220).

Emerging evidence implicates the gut microbiome in shaping both the risk and severity of irAEs (76, 201). In melanoma patients treated with combined anti-PD-1/CTLA-4 therapy, specific taxa were associated with colitis risk. Notably, Bacteroides intestinalis was enriched in patients who developed colitis and induced IL-1β–mediated ileal inflammation in ICB-treated mice (201). Other studies found that patients who developed irAEs had significantly different microbiome beta-diversity compared to those who did not (76). Two distinct microbial clusters were identified: one enriched for Lachnospiraceae (associated with favorable responses), and another enriched for Streptococcus spp., which correlated with higher irAE incidence and poorer outcomes. Patients with high cumulative levels of seven Streptococcus spp. showed consistently shorter PFS and a distinct irAE profile, including increased arthritis (76). These patterns suggest that the gut microbiome influences not only treatment efficacy but also the nature and severity of irAEs. Notably, high Streptococcus abundance and poor ICB response were linked to proton pump inhibitor (PPI) use—medications known to induce “oralization” of the gut microbiome, leading to increased colonization by upper-GI taxa such as Streptococcus and Veillonella spp. (76, 221, 222).

Using batch-corrected microbiome data from over 400 patients across nine cohorts (primarily melanoma and lung cancer), a machine learning model was developed to predict irAE occurrence based on fecal microbial composition. The model achieved strong performance (AUROC = 0.88), with the top 14 random forest classifiers driving predictive accuracy (223). Taxa associated with protection from irAEs were enriched for genes involved in vitamin K2 biosynthesis, suggesting anti-inflammatory effects potentially mediated via NF-κB inhibition (223).

Finally, a Mendelian randomization study in >1,700 ICB-treated patients supported a causal link between Lachnospiraceae abundance and increased irAE risk, while members of the Verrucomicrobiaceae family were associated with irAE resistance (224).

The gut microbiome of long-term responders to ICB is notably consistent between individuals with and without severe irAEs, yet clearly diverges from that of therapy-resistant patients (76, 224). In both responder groups, Firmicutes taxa predominate; however, patients who remain irAE-free are enriched in Ruminococcaceae species, which produce SCFAs such as propionate and butyrate—metabolites with well-established anti-inflammatory and mucosa-protective effects (225). In preclinical models, ICB has been shown to provoke commensal-specific Th17 responses, driving cutaneous inflammation that mimics irAEs observed in patients (226).

These findings suggest that the microbiome of irAE-free, long-term responders may exhibit a uniquely beneficial ecological balance. Such communities appear to support both immune activation—potentially via mechanisms like molecular mimicry of tumor and self-antigens—and immune regulation through the generation of SCFAs, vitamin K2, and secondary bile acids (223, 224, 227). Understanding and leveraging these protective features could inform the rational development of microbiome-based therapies, such as personalized FMT, to enhance immunotherapy efficacy while minimizing toxicity (224).

However, key limitations persist. The mechanism-based toxicity of microbial therapies may not be fully preventable, and in some cases, could be worsened by FMT. While FMT trials in anti–PD–1–treated patients have not reported treatment-limiting toxicities (34, 35, 208), early-onset severe irAEs have been observed in melanoma patients receiving FMT in combination with anti-PD-1/CTLA-4 therapy (209). Conversely, FMT has demonstrated efficacy in treating ICB-induced colitis that is refractory to corticosteroids and biologics, highlighting its therapeutic potential in select dysbiosis-driven inflammatory settings (Table 1) (228–230).

11. FUTURE PERSPECTIVE

In conclusion, the growing body of evidence from both preclinical models and early-phase clinical studies underscores the critical role of the gut microbiome in shaping cancer immunotherapy outcomes. While challenges remain—such as patient-to-patient microbiome variability, lack of standardization across trials, and difficulties translating findings from mice to humans—FMT has emerged as the most effective microbiome-targeting intervention to date, particularly in overcoming resistance to immune checkpoint blockade. ML tools show promise for predicting treatment response and guiding donor selection but require further refinement and validation across diverse patient populations and cancer types. Future directions should include larger randomized clinical trials to evaluate microbiome-based interventions, development of standardized donor selection criteria, and exploration of host-microbiome-immune interactions through multi-omics approaches. Additionally, defined bacterial consortia, postbiotic compounds, dietary modulation, and host-targeted therapies may offer safer, scalable, and more targeted alternatives to FMT. Ultimately, integrating microbiome profiling into routine clinical care could help personalize immunotherapy, improve efficacy, and reduce toxicity across a broad spectrum of cancers.