Immunotherapy harnesses the body’s immune system to target and eliminate cancer cells (1). The various forms of immunotherapy include immune checkpoint inhibitors (ICIs)—such as programmed cell death protein 1 (PD-1), programmed cell death ligand 1 (PD-L1), and cytotoxic T-lymphocyte–associated antigen 4 (CTLA-4) inhibitors—, adoptive cell therapies—such as chimeric antigen receptor T-cell therapy, tumor-infiltrating lymphocyte therapy, and natural killer (NK) cell therapy—, cancer vaccines, and cytokine therapies (2). These therapies work by either enhancing the immune system’s ability to detect and destroy cancer cells or infusing cancer-fighting immune cells that effectively target cancer cells (3).

Among various types of cancer immunotherapy, ICI therapy has emerged as a promising therapeutic approach for a variety of solid tumors, including for non-small cell lung cancer (NSCLC) and melanoma (4). ICIs are monoclonal antibodies that can target and block immune checkpoints, such as PD-1 and CTLA-4 on T cells or PD-L1 on tumor cells, thus inhibiting tumor immune escape and enhancing the antitumor function of T cells (5). ICIs offer clear advantages such as high accuracy, response durability, wide adaptability, and low toxicity (2). However, the effectiveness of ICI therapy can vary greatly between individuals. For instance, in patients with NSCLC with high PD-L1 expression treated with pembrolizumab (a PD-1 inhibitor), the five-year overall survival (OS) rate is approximately 25–30% (6). A recent study estimated that only 19.60% of US patients with cancer responded to ICI treatment in 2023. This high variability in ICI response has been linked to several factors, including tumor mutational burden, PD-L1 expression, immune infiltration, altered interferon (IFN)-γ signaling, and epigenetic modifications (7, 8). Nevertheless, the factors determining the efficacy of ICI therapy remain poorly understood, posing considerable challenges for improving response rates and developing effective, patient-specific ICI treatment strategies.

Recent studies have shown that the gut microbiota plays a significant role in shaping drug responses by either modifying drug structures or affecting host physiological functions (9–11). Moreover, the gut microbiota can modulate local and systemic immune responses via microbial metabolites or components (12). For example, gut microbiota can affect T-cell expression profiles (e.g., regulatory T cells (Tregs) and CD8+ T cells) and regulate the differentiation of monocytes into immunogenic dendritic cells (DCs) (13, 14). Furthermore, short-chain fatty acid (SCFA) butyrate produced by gut microbiota can enhance CD8+ T-cell effector function and promote the transition of activated CD8+ T cells to T memory cells (15, 16). These immunomodulatory capacities of the gut microbiota suggest its possible influence on the efficacy of ICI therapy. In 2015, Vétizou et al. showed that antibiotic-treated or germ-free mice with tumors exhibited poor responses to CTLA-4 blockade, and the supplementation of specific Bacteroides spp. restored treatment efficacy by activating T-helper (Th) 1 cells in tumor-draining lymph nodes and promoting the maturation of intratumoral DCs (17). Importantly, three studies in 2018 have revealed a strong link between the gut microbiota and outcomes of ICI therapy in human (18–20), positioning it as a potential biomarker for predicting immunotherapy response and a modifiable target for enhancing treatment efficacy.

This review focuses on the past and current state of research on role of the gut microbiota in the responses to ICI therapy in human patients with solid tumors ( Figure 1 ). We explore findings from studies investigating whether gut microbial dysbiosis due to antibiotic treatment is associated with ICI therapy outcomes in patients with solid tumors and determining which specific gut microbes are differentially abundant in responders and non-responders. We also examine research on outcome prediction using machine learning models based on gut microbiome data before starting ICI therapy, as well as clinical trials determining whether gut microbiota modulation can improve ICI treatment outcomes. In addition, we briefly outline proposed mechanisms by which the gut microbiota impacts the efficacy of ICI therapy via cell components or metabolites. As most human studies on this subject have been conducted specifically on gut bacteria, the term “gut microbiome” in this review will refer to the gut bacterial microbiome.

Patients with cancers are often prescribed antibiotics to treat infectious diseases such as dental, respiratory tract, and urinary tract infections. Antibiotic use is necessary to control infections, but it can disrupt gut microbial homeostasis by altering microbial composition and function (21). The gut dysbiosis induced by antibiotic use can affect the host immune system (22) and has been linked to increased risks of various immunological diseases such as allergy, asthma, and rheumatoid arthritis (23–25). These findings indicate that antibiotic use in patients with cancers may influence the efficacy of ICI therapy by impairing immune responses. Since the first study reporting worsened responses to ICI therapy in antibiotic-treated mice (17), many studies have examined the impact of antibiotic use on outcomes of ICI therapy in human cohorts ( Table 1 ).

In a large European cohort of patients with advanced NSCLC (n = 140), renal cell carcinoma (RCC) (n = 67), and urothelial carcinoma (UC) (n = 42) who received PD-1/PD-L1 inhibitors, progression-free survival (PFS) and/or OS were significantly shorter in the antibiotic-treated group within two months before or one month after the start of ICI treatment, compared to the untreated group (PFS: 3.5 months for antibiotic users vs. 4.1 months for non-users, OS: 11.5 vs. 20.6 months) (26). Similarly, detrimental effects of antibiotic use within 30 days of the first ICI administration on therapeutic efficacy were observed in patients with advanced RCC (n = 121) and NSCLC (n = 239) treated with anti-PD-L1 inhibitor (PFS: 1.9 vs. 7.4 months, OS: 17.3 vs. 30.6 months) (18). In a cohort of Korean patients with solid tumors (n = 234), including NSCLC and melanoma, antibiotic treatment within 60 days of starting ICI therapy was associated with significantly reduced PFS (2 vs. 4 months) and OS (5 vs. 17 months) (27). However, other studies have shown minimal or no significant effects of antibiotic use on outcomes of ICI therapy (28, 29). Despite these mixed findings, a meta-analysis of 23 studies on NSCLC patients reported that antibiotic use before or during ICI treatment reduced median PFS (n = 2208) by 1.2 months and median OS (n = 5560) by 6.7 months (30). Another meta-analysis of 37 studies across 13 cancer types also found that antibiotic use was significantly associated with poorer prognosis (OS and PFS) in cancer patients treated with ICIs (31).

These findings suggest that the gut microbiota plays a critical role in determining responses to ICI therapy and underscore the importance of considering a patient’s antibiotic treatment history when planning ICI treatment. More detailed studies are needed to explore the effects of different class of antibiotics, doses, treatment durations, and time of administration on the efficacy of ICI therapy in various cancer types and stages.

Identifying the specific microbes that play pivotal roles in influencing responses to ICI therapy is crucial for understanding their immune modulation mechanisms, predicting the efficacy of ICI therapy, and developing personalized microbiota modulation strategies to enhance outcomes of ICI therapy. In 2018, three landmark studies were published. Routy et al. found that Akkermansia muciniphila, Enterococcus faecium, and Alistipes indistinctus were significantly enriched in responders compared to non-responders among patients with NSCLC and RCC treated with PD-1 inhibitors (n = 100) (26). Matson et al. reported enrichment of Bifidobacterium longum, Collinsella aerofaciens, and E. faecium in responders (n = 42), while Gopalakrishnan et al. found enrichment of Faecalibacterium in responders (n = 43) in melanoma patients treated with PD-1/CTLA-4 inhibitors (19, 20).

Since these initial studies, many others have reported distinct gut microbial features associated with responses to ICI therapy ( Table 2 ). In patients with advanced NSCLC treated with anti-PD-1 blockade (n = 25), responders had a higher abundance of Alistipes putredinis, Prevotella copri, and B. longum, while non-responders had more Ruminococcus (32). A large cohort study of NSCLC patients receiving ICIs (n = 338) found that baseline A. muciniphila presence was associated with higher objective response rates (ORR) and OS (33). Another study of NSCLC patients (n = 65) undergoing ICI therapy reported significant enrichment of Ruminococcus, Akkermansia, and Faecalibacterium in responders (34). In addition to NSCLC, several studies have been conducted on melanoma. In melanoma patients receiving ICI therapy (n = 27), prolonged PFS was linked to higher abundances of Faecalibacterium prausnitzii, Coprococcus eutactus, and Prevotella stercorea, while shorter PFS was associated with higher levels of Bacteroides ovatus, B. dorei, B. massiliensis, Ruminococcus gnavus, and Blautia producta (35). Another study of advanced melanoma patients treated with combined CTLA-4 and PD-1 blockade (n = 40) found that responders had enriched levels of Bacteroides stercoris, Parabacteroides distasonis, and Fournierella massiliensis (36). In RCC (n = 31), Bifidobacterium adolescentis, Barnesiella intestinihominis, Odoribacter splanchnicus, and Bacteroides eggerthii were enriched in responders (37). Another RCC study (n = 58) showed that responders to PD-1 inhibitors had higher levels of A. muciniphila, Bacteroides salyersiae, and Eubacterium siraeum (38). In hepatocellular carcinoma (HCC), an increase in Faecalibacterium was observed in patients who responded to ICI therapy (39, 40).

Taken together, a wide range of individual gut microbes have been associated with outcomes of ICI therapy, with Akkermansia and Faecalibacterium frequently linked to favorable responses across different cancers. However, there is no clear consensus, making it difficult to use these microbes as universally applicable biomarkers. Recently, Derosa et al. proposed a biomarker based on the ecological topology of the gut microbiome that predicts responses to ICI therapy in various cancers (41). Using co-abundance networks, they identified two species-interacting groups (SIGs) related to PD-1 blockade response in NSCLC patients. SIG1 comprised bacteria associated with reduced OS, such as the Enterocloster genus and Streptococcaceae, Veillonellaceae, and Lactobacillaceae families, while SIG2 consisted of longer OS-related bacteria, such as F. prausnitzii and Ruminococcus bicirculans. The topological score (TOPOSCORE), which is a combination of the SIG1/SIG2 ratio and A. muciniphila abundance, was significantly associated with immunotherapy outcome in the NSCLC discovery cohort and could predict patients’ responses to ICI therapy with an area under the receiver operating characteristic curve (AUC) of 0.66. The association of the TOPOSCORE with responses to ICI was also validated in patients with NSCLC and those with genitourinary cancer, suggesting the potential of TOPOSCORE as a robust biomarker for predicting immunotherapy response (41). These findings highlight the importance of considering not only individual bacteria but also bacterial interactions when predicting ICI responses based on the gut microbiome.

The identification of significant associations between the gut microbiome and ICI response have led researchers to investigate machine learning models to stratify patients into responders and non-responders to immunotherapy using the gut microbial profiles before starting treatment ( Table 3 ).

Limeta et al. (42) developed a predictive model using three previously published gut microbiome datasets from 103 patients with melanoma undergoing ICI therapy (19, 20, 43). They used 49 differentially abundant species and functional pathways between responders and non-responders, including unknown Ruminococcaceae species, unknown Faecalibacterium species, Bacteroides thetaiotaomicron, and aerobic respiration, as input for training a random forest (RF) classifier. The AUC of the model on the training data was 0.604. They also evaluated the classification performance of the model on the independent validation dataset (n = 27) (35), achieving an AUC of 0.624 (42). McCullochet et al. (44) also developed machine learning models to predict the PD-1 response of patients with melanoma using 155 metagenomic datasets from a new cohort and four published independent cohorts (19, 20, 35, 43). In this study, batch-effect-corrected taxon count data were used to reduce the inter-study heterogeneity of metagenomic data, and a modified leave-one-out cross-validation method, where models were trained on the four cohorts and tested on the remaining cohort, was applied to overcome the limitation of small sample size in each cohort. Three machine learning algorithms, including RF, a generalized linear model, and a polynomial support vector machine, were used, and the resulting AUC values were between 0.54 and 1.00 (44). This study showed that machine learning models trained on gut microbiome data can consistently predict the efficacy of PD-1 therapy in patients with melanoma prior to ICI treatment.

A machine learning model for gastrointestinal (GI) cancer was developed using microbial species data from 18 patients with gastric cancer (GC) and 59 patients with colorectal cancer (CRC) undergoing anti-PD-1/PD-L1 treatment (45). The light gradient-boosting machine model showed high predictive performance (AUC of 0.90 in the training dataset), and the performance of this model was confirmed in the independent validation set (n = 31) with an AUC of 0.79. In this model, Bacteroides caccae, Veillonella parvula, V. atypica, and Clostridiales bacteria significantly contributed to distinguishing responders from non-responders (45). Recently, it was reported that the outcome of anti-PD-1/PD-L1 immunotherapy can be also predicted in biliary tract cancer (BTC) (46). This study used microbiome data from 77 patients with BTC and built three different RF models based on bacteria, metabolites, and a combination of bacteria and metabolites. The model trained only with six bacteria, including Streptococcus anginosus, Olsenella profusa, and A. putredinis, showed high predictive performance with an AUC of 0.897 for the training dataset and of 0.722 for the test dataset (46).

For NSCLC, the aforementioned study introducing the TOPOSCORE approach (41) also developed a machine learning model based on the relative abundance of all microbial species to compare the predictive performance of TOPOSCORE across 245 patients with NSCLC. The TOPOSCORE distinguished the responders and non-responders with an AUC of 0.66, while the RF model using microbiome data achieved an AUC of 0.651, indicating that these two approaches have similar performance in predicting the outcome of ICI therapy in patients with NSCLC (41). Another study including small number of patients with NSCLC (n = 37) also demonstrated RF models based on differentially abundant species and/or pathways between patients with long and short PFS. The model based on species could discriminate between long- and short-PFS patients with an AUC of 0.74 (47).

These models demonstrated moderate prediction performances, indicating the potential for gut microbial profiles to serve as markers for treatment response. Notably, the models for cancers of the GI system (GC, CRC, and BTC) exhibited higher predictive performances than those for other cancers, suggesting that ICI therapy for GI cancers may be more directly influenced by the gut microbiota. Given the complexity and variability of the gut microbiome, it is evident that further validation is necessary to enhance the clinical applicability of these predictive models.

Applications using the gut microbiota to enhance ICI therapy outcomes are of considerable clinical interest. However, a comprehensive understanding of the molecular mechanisms underlying this effect is crucial for deepening our knowledge and developing more effective strategies to harness the gut microbiota for ICI therapy. In this context, we will briefly explore the molecular mechanisms by which the gut microbiota may impact the efficacy of ICI therapy, with a particular focus on the bacterial cell components and metabolites.

The reviewed studies make it evident that the gut microbiota plays a significant role in influencing patients’ responses to ICI therapy. While the potential for utilizing the gut microbiome to predict or enhance outcomes of ICI therapy is promising, several challenges must be addressed before it can be widely implemented in clinical practice.

A recurring issue is the lack of consensus regarding gut microbial features associated with responses to ICI therapy. Many studies have identified specific microbes which were enriched in responders and non-responders to ICI treatment in various types of solid cancer, but an individual’s gut microbial compositions are influenced by numerous intrinsic and extrinsic factors such as age, diet, geography, medication, and co-morbidities (79). These factors may results in differences in ICI response-related microbes among the study populations. Therefore, large-scale research projects including diverse populations are needed to identify robust biomarkers applicable to diverse populations. It is also important to collect comprehensive patient metadata such as antibiotic use, co-morbidities, co-medications, and dietary records. These metadata can be used to account for confounding factors in statistical analyses and to identify ICI response-associated features specific to certain subgroups.

Studies investigating gut microbiome-based machine learning models to predict patient responses to ICI therapy have shown promising results. To determine whether a developed model can be applied to other populations, the predictive performance of the model should be evaluated in independent validation cohorts (80). However, it may be inappropriate to use microbiome data produced using different protocols for sample collection, storage, processing, and/or sequencing for validation, as these differences can introduce systemic bias in sequence-based microbiome data (81). Therefore, standardized protocols for microbiome analysis are required in this research area. McCulloch et al. used an empirical Bayesian framework to reduce between-cohort heterogeneity when integrating and analyzing microbiome data from five independent cohorts (44). This method may represent a solution that enables data produced by various protocols to be utilized in the creation and validation of predictive models. Additionally, most predictive models reported to date have been generated using only gut microbiome information. Research is also needed to determine whether adding other known biomarkers for responses to ICI therapy—such as tumor mutational burden and PD-L1 expression—or other omics data can improve the predictive performance of machine learning models.

The gut microbiota holds potential as a modifiable target to enhance the efficacy of ICI therapy. Gut microbes enriched in responders could serve as candidates for LBPs to improve treatment outcomes. Given that strains from the same species can have different functional capabilities (82), high-resolution microbiome studies are necessary to identify specific strains with the greatest potential for boosting the therapeutic efficacy. Additionally, verifying the functional characteristics of candidate microbes through in vitro and in vivo experiments is critical.

One challenge in gut microbiota interventions is the pre-existing microbial environment. As demonstrated by the SER-401 trial, where the effectiveness of ICI therapy was influenced by antibiotic preconditioning (61), the gut microbial dysbiosis induced by antibiotic treatment before intervention may impact the intervention outcome. Among the four FMT clinical trials mentioned in the previous section, those by Baruch et al. and Kim et al. included the administration of oral antibiotics before FMT treatment for 3 and 5 days, respectively, to remove the native gut microbiota (55, 58). However, the studies by Davar et al. and Routy et al. did not include any antibiotic treatment prior to FMT (56, 57). As these four FMT studies have shown the clinical benefits of FMT combined with ICI therapy regardless of antibiotic preconditioning, the procedure for native microbiota depletion may be omitted in combined treatment using ICIs and gut microbiota manipulation. Further research is needed to determine preconditioning regimens for preparing the gut microbial environment to maximize the effect of the combination therapy. In addition, understanding the relationship between the baseline gut microbial composition and intervention outcomes could help develop microbiome-based stratification strategies for patient treatment.

In addition to FMT and LBPs, ongoing clinical studies are investigating the role of diet and prebiotics in modulating the gut microbiota to enhance the efficacy of ICI therapy. However, our understanding of the mechanisms through which the gut microbiota influences ICI therapy outcomes is still limited. Therefore, large-scale observational and intervention studies, combined with mechanistic investigations, are essential for advancing gut microbiome-based personalized therapeutic strategies applicable in real-world clinical settings.