Harnessing the microbiome for cancer therapy

Abstract

The microbiome is increasingly recognized as a key player in cancer pathogenesis and treatment response, acting through both local and systemic mechanisms. Microbial communities and their metabolites can directly influence drug metabolism, shape the immune landscape, and alter transcriptional and epigenetic programmes in the gut, systemically and in the tumour microenvironment. Emerging data support the potential of microbiome-targeted interventions (such as faecal microbiota transplantation, diet, prebiotics and probiotics) as adjuncts to conventional cancer therapies, with the goal of enhancing efficacy and reducing toxicity. This Review highlights the promise of the microbiome as a prognostic and predictive biomarker, a modifiable factor in cancer care and prevention, and a therapeutic target. We also discuss major knowledge gaps, limitations in current study designs, and the need for mechanism-guided, personalized strategies to advance clinical translation.

Introduction

The gastrointestinal tract and other tissues in the human body are colonized by diverse communities of microorganisms from all kingdoms of life¹. These microorganisms and their products, enzymatic activities and metabolites can affect processes related to human health in a myriad of ways, for example, via regulation of host physiology including transit and barrier function, modulation of the immune system, and metabolism of drugs², dietary nutrients and host components. These processes are also relevant for human cancer.

The field of cancer microbiome research began with historical observations that infections could cause tumour regression. Another breakthrough was achieved in the late 1800s with the successful treatment of tumours using inactivated bacteria as an early immunotherapy (Coley’s toxins). Ultimately, viruses and bacteria (such as human papillomavirus, hepatitis B virus and Helicobacter pylori) were also recognized as contributors to cancer³. These milestones formed the foundation for modern studies on the complex relationship between microorganisms and tumour biology³. Advances in culture-independent next-generation sequencing have revolutionized the field by broadening it beyond single candidate pathogens to also encompass complex microbial communities and their collective metabolic activities. The functional relevance of the microbiome in cancer is highlighted by the observation that broad-spectrum antibiotics, which disrupt the microbiome, reduce the efficacy of multiple cancer therapies across tumour types^4–8. Recent human multi-omics studies and mechanistic preclinical models suggest that the microbiome exerts its effects through multiple processes in the gut or tumour microenvironment (TME), including via microbial metabolism of anticancer drugs, by affecting circulating levels of metabolites, and through the modulation of antitumour immunity, such as promoting the numbers and capacity of cytotoxic CD8⁺ T cells that can destroy cancer cells^9–11.

Many different strategies are used in the treatment of cancer. Chemotherapy involves combinations of drugs focused broadly on inhibiting the growth of fast-growing cells, whereas targeted therapies interfere with specific molecules, proteins or signalling pathways that some cancers depend on for growth and survival. Other approaches include surgery to remove or radiation therapy to kill tumour tissue. In addition, immunotherapies have been developed that use recombinant antibodies to block inhibitory checkpoints of the immune system and, thereby, stimulate its capacity to recognize and destroy cancer cells. All of these therapies can lead to adverse events of varying severity in subsets of patients.

Therapeutic strategies have also been explored to harness the interaction between the microbiome and cancer. These strategies can be grouped based on their shared proposed mechanisms of action. For example, genotoxic microbial metabolites can be inactivated, their production prevented, or the resulting damage repaired. Specific microbial modifications of cancer drugs could be inhibited at either the enzyme or taxa level. Finally, the immune system could be modulated via the microbiome through a wide range of approaches that include diet, faecal microbiota transplantation (FMT), prebiotics, (engineered) probiotics, synbiotics (mixtures of prebiotics and probiotics), and postbiotics (preparations of inanimate microorganisms and/or their components that confers a health benefit on the host¹²).

In this Review, we explore how microbially mediated processes in the gut and the TME^13–16 contribute to interindividual variability in cancer risk and treatment outcomes. We begin by discussing mechanisms through which the microbiome drives cancer development and progression, and then examine associations between the microbiome and treatment outcomes. Next, we review novel and proposed therapeutic strategies, followed by emerging diagnostic and artificial intelligence approaches. Finally, we outline the limitations of current approaches and discuss future prospects for integrating microbiome science into oncology.

Cancer development and progression

Historically, microorganisms have been implicated in cancer using frameworks such as Koch’s postulates, which attribute disease to a single causative agent with a defined incubation time. However, microorganisms are now recognized to function as complex communities that can exert collective effects on the host rather than just as isolated pathogens. These effects include modulating inflammation, immunity, metabolism and epigenetic regulation, in ways that could profoundly influence carcinogenesis. These mechanisms have been comprehensively described in prior reviews^17,18, so, in this section, we briefly discuss some key examples of how microbial products and metabolites might affect host processes that underlie cancer (Fig. 1). Notably, most of these studies are on colorectal cancer (CRC), whereas fewer studies address the role of the microbiome in carcinogenesis of cancers distant from the gastrointestinal tract. Here, we focus on bacteria as although increasing attention has been given to non-bacterial communities (particularly fungi), the influence of the mycobiome on cancer is still nascent, with mechanistic insights only beginning to emerge^19,20.

Fig. 1 |. Mechanisms underlying microbiome-driven cancer pathogenesis.

The gut microbiome can be involved in cancer initiation and/or progression via inducing genomic damage (part a), epigenetic modulation (part b) or disruption of the gut barrier function (part c). a, Microbiome-induced genotoxicity is mainly driven by pathobionts secreting DNA-damaging toxins, including indolimines by the inflammatory bowel disease-associated microbiome, tilimycin and tilivalline by Klebsiella oxytoca, colibactin mainly by Escherichia coli strains, and Bacteroides fragilis toxin (BFT) by B. fragilis. The BFT-secreting and mucolytic B. fragilis enables colibactin to reach the epithelial layer, wherein it induces DNA alkylation and double-strand breaks, an effect that is further potentiated by microbial spermidine synthase in the gut. Microbial transformation of primary biliary acids into secondary biliary acids and of luminal ethanol into acetaldehyde can also result in genomic damage. Microbial β-glucuronidases can also be involved in genotoxicity by transforming the nitrosamines N-butyl-N-(4-hydroxybutyl)nitrosamine (BBN) into N-butyl-N-(3-carboxypropyl)nitrosamine (BCPN), as well as promoting circulating oestrogen levels, which could have distant effects on bladder and oestrogen-sensitive cancers. b, Butyrate is a histone deacetylase (HDAC) inhibitor, which makes microbially produced butyrate an important factor in epigenetic modulation of the epithelium, resulting in chromatin modulation, suppression of proto-oncogenes and activation of tumour suppressor genes. However, inflammation and other metabolites (for example, nitrosamines or folate) can also induce DNA acetylation and methylation, modulating the expression of cancer-associated genes. c, Inflammation can disrupt gut barrier integrity by degrading the mucus layer and weakening tight junctions between epithelial cells. A weaker gut barrier can result in translocation of bacterial components into the circulation, and in the activation of the Toll-like receptor–myeloid differentiation primary response 88 (TLR–MyD88) inflammatory cascade in the colonic mucosa. In addition, chronic inflammation and excessive bacterial translocation may lead to a systemic immunosuppressive phenotype that decreases immunosurveillance and promotes extraintestinal cancer development. 1° BA, primary bile acid; 2° BA, secondary bile acid; ETBF, enterotoxigenic B. fragilis; IL, interleukin; LCN, lipocalin; LPS, lipopolysaccharides; MMP, matrix metalloproteinase; MUC2, mucin 2; NF-κB, nuclear factor-κB; ROS, reactive oxygen species; PGN, peptidoglycan; TNF, tumour necrosis factor^{9,22,28,45,46,49–52,54–56,59,60,213–218}.

Oncogenic effects exerted by the microbiome

Genotoxicity.

Pathogenic or opportunistic bacterial taxa colonizing epithelial surfaces can exert oncogenic effects through direct host cell signalling, inter-microbial communication, or via the production of secreted toxins. Direct bacterial contributions to epithelial transformation have been identified, including in the gastrointestinal tract and more distant organ systems^21–26. Bacterial toxins include tilimycin, indolimines, tilivalline, colibactin and Bacteroides fragilis toxin (BFT), as well as suspected ones that remain to be described^22,27–29. Among these, the mechanism of colibactin is the most extensively characterized. Colibactin is encoded by the polyketide synthase (pks) genomic island found in Proteobacteria, such as specific strains of Escherichia coli^30,31. This genotoxin induces DNA double-strand breaks and alkylation, triggering genomic instability and promoting carcinogenesis^31–34 (Fig. 1).

Beyond de facto genotoxins, microbial metabolites can also have DNA damaging effects. For example, secondary bile acids produced through microbiota-mediated conversion of primary bile acids can promote colorectal tumorigenesis by inducing DNA damage, modulating inflammatory pathways and suppressing antitumour immune responses, including inhibition of cytotoxic CD8⁺ T cells that are crucial in cancer immunosurveillance⁹.

Inflammation and gut barrier dysfunction.

Chronic microbial infections, such as Salmonella typhi in gallbladder cancer and H. pylori in gastric cancer, cause cancer through the induction of sustained inflammation, which promotes cumulative cellular damage and dysregulated epithelial proliferation^35–37. Accordingly, substantive evidence supports H. pylori eradication to prevent cancer development^38,39. However, microbiome imbalances can also promote carcinogenesis by modulating local inflammatory responses in the absence of overt infection (Fig. 1). For example, Fusobacterium nucleatum is normally found in the oral microbiota but can spread to the gut under certain conditions, wherein it can induce inflammation of the TME in CRC tumours^40,41. Furthermore, chronic intestinal inflammation occurring owing to inflammatory bowel disease and propagated through the gut microbiome can result in the development of CRC⁴². Microbiome-induced inflammation is mediated in part through pattern-recognition receptor signalling, including Toll-like receptors, and inflammasome activation. Consequently, cytokines and reactive oxygen species are produced that stimulate epithelial proliferation, immune evasion, and cancer cell dissemination and implantation at distant sites^43–47.

Intestinal inflammation is tightly linked to gut barrier dysfunction, defined as the inability of the mucus layer and epithelium to provide sufficient space between the luminal community and the host epithelium, thus increasing susceptibility to damage by procarcinogenic microorganisms and secreted genotoxins^48,49. Indeed, an impaired gut barrier that fosters chronic subclinical mucosal inflammation might be a pivotal step in colorectal carcinogenesis⁵⁰. Compromised barrier integrity could also facilitate translocation of microbial products and components, such as lipopolysaccharides (LPS), and metabolites such as bile acids into systemic circulation^45,51. These factors could drive systemic inflammation or immunosuppression in distal tissues, potentially contributing to the development of distant epithelial and non-epithelial cancers and metastases^45,52,53. By contrast, some microbial metabolites (such as short-chain fatty acids (SCFAs) produced via fermentation of dietary fibre in the colon) exert protective effects on the gut barrier by promoting tight junction formation and stimulating mucus production^45,54.

Epigenetic modulation.

The gut microbiome is a potent epigenetic regulator, capable of shaping gene expression and tumour behaviour through a combination of metabolic, inflammatory and diet–microorganism interactions (Fig. 1). The best described microbial-driven mechanism of epigenetic modulation is inhibition of histone deacetylases (HDAC) by the SCFA butyrate. This process could be relevant in CRC, as in healthy epithelial cells, butyrate supplies an energy source that supports proliferation and differentiation. However, as metabolism shifts towards aerobic glycolysis in neoplastic cells (the Warburg effect), butyrate is no longer consumed efficiently. The resulting accumulation leads to excessive HDAC inhibition, altering chromatin conformation and promoting the transcriptional activation of tumour-suppressor genes that can suppress carcinogenesis⁵⁴. Other microbial or dietary components can also influence the epigenome. For example, folate, nitrosamines and obesogenic diets affect DNA methylation and histone acetylation, contributing to intestinal tumorigenesis in mouse models^55,56. Interestingly, enterotoxigenic B. fragilis colonization in mouse models promotes colorectal tumorigenesis through disruption of DNA repair mechanisms and induces widespread CpG island hypermethylation, a hallmark of epigenetic change in tumour tissue^57,58. Finally, the host epithelium also responds to microbially induced inflammation through DNA methylation and host chromatin remodelling^59,60.

Effects of microbiome disruption across lifespan

The classical multi-hit model of cancer posits that sequential genetic and epigenetic alterations, driven by environmental exposures, ageing and DNA replication stress, lead to malignant transformation. Emerging evidence suggests that microbiome disruption could represent an additional interconnected factor within this multi-hit model. Microbiome development begins in early life, shaped by mode of delivery, feeding practices and dietary transitions^61,62. Disruption during this critical window can result in lasting reductions in microbial diversity and absence of keystone taxa essential for immune and metabolic homeostasis^63,64. This microbial imbalance potentially leads to a pro-tumorigenic state analogous to that seen in some germ-free or antibiotic-treated preclinical models⁶⁵. Across the lifespan, cumulative insults (including antibiotic overuse, chronic stress, and low-fibre, ultra-processed diets) can further erode microbial diversity and function^66–70. This progressive loss of microbial resilience might compromise epithelial integrity, increase inflammation, impair metabolic signalling and weaken immune surveillance. We propose that this longitudinal degradation of the microbiome could lower the threshold for cancer initiation and progression, paralleling the stepwise genetic events in classical oncogenesis (Fig. 2).

Fig. 2 |. Progressive degradation of microbiome health might contribute to cancer development across the human lifespan.

Two contrasting trajectories of gut microbiome health and resiliency over time are shown. The desirable trajectory (blue line) represents optimal microbiome maintenance through beneficial factors (blue boxes), and the deteriorating trajectory (red line) shows the cumulative effects of microbiome-disrupting factors (red boxes). Each acute disturbance is followed by incomplete recovery, leading to progressive degradation of microbiome health and resiliency. The grey-shaded area represents a degraded baseline state associated with increased cancer risk. Persistent stressors could push individuals along the deteriorating trajectory towards a threshold wherein cancer development becomes more probable, as illustrated by the lung cancer and melanoma representation at the end point of the trajectory. C-section, caesarean section.

Therapeutic efficacy and adverse events

Variability in cancer treatment response is influenced by multiple host-related and tumour-related factors, including tumour and host genetic heterogeneity, the extent of immune surveillance, treatment-associated toxicity and drug metabolism. The gut microbiome is an important contributor to this variability in drug metabolism, which spurred the field of pharmacomicrobiomics². In this section, we focus on two broad mechanisms through which the microbiome can modulate cancer treatment outcomes. First, microbial biotransformation of therapeutic agents and the reciprocal effects of drugs on the microbiome. Second, microbiome-mediated modulation of the host immune response (Fig. 3).

Fig. 3 |. Microorganism-mediated mechanisms that affect cancer treatment outcomes.

a, Chemotherapeutic agents including 5-fluorouracil (5-FU), capecitabine and gemcitabine can be metabolized or inactivated by bacterial taxa, via enzymes including bacterial dihydropyrimidine dehydrogenase (DPD, encoded by the preTA operon) and cytidine deaminase (CDA). Fusobacterium nucleatum in the gut limits the bioavailability of 5-FU and leads to a microRNA (miRNA)-dependant cascade culminating in autophagy in the tumour microenvironment. b, Several commensal bacterial taxa were shown to modulate adaptive immunity and the function of CD4⁺ and CD8⁺ T cells, leading to elevated higher immune surveillance, tumour immune infiltration and tumour clearance by immune checkpoint inhibitors (ICIs), both locally and systemically. This process is further enhanced with innate immunity contributing to cancer cell recognition and clearance. Bacterial metabolites, including butyrate, microbial tryptophan and its catabolites, as well as inosine and pentanoate, can modulate the balance between immunosurveillance and immunosuppression, specifically between CD8⁺ T cells and myeloid-derived suppressor cells (MDSCs) and regulatory T cells, which ultimately influence tumour immune recognition and the antitumour effects of ICIs. These processes might occur both locally in epithelial surfaces and systemically owing to translocation of bacteria, their metabolites and conditioned immune cells. A. muciniphila, Akkermansia muciniphila; B. longum, Bifidobacterium longum; B. pseudolongum, Bifidobacterium pseudolongum; C. aerofaciens, Collinsella aerofaciens; CRC, colorectal cancer; CTL, cytotoxic T lymphocyte; DC, dendritic cell; E. faecium, Enterococcus faecium; FasL, Fas ligand; IFN, interferon; IL, interleukin; IPA, indole-3-propionic acid; LPS, lipopolysaccharide; M. hyorhinis, Mycoplasma hyorhinis; MyD88, myeloid differentiation primary response 88; NK, natural killer cell; NSCLC, non-small-cell lung cancer; P. distasonis, Parabacteroides distasonis; PD1, programmed cell death protein 1; PDL1, programmed cell death 1 ligand 1; T_H1, T helper 1 cell; TLR, Toll-like receptor^{10,43–45,73–77,85–87,89–95,102,103,135}.

Microbial biotransformation and effects of drugs

Like the host genome, the gut microbiome encodes a vast repertoire of genes capable of metabolizing both oncologic and non-oncologic drugs^71,72, with implications for drug bioavailability, efficacy and toxicity. A well-characterized example involves the chemotherapeutic antimetabolite 5-fluorouracil (5-FU) and its oral prodrug capecitabine. Certain species of Escherichia and Anaerostipes express bacterial dihydropyrimidine dehydrogenase (DPD), which inactivates 5-FU by converting it to dihydrofluorouracil, potentially reducing its therapeutic efficacy^73,74. In addition, E. coli and Parabacteroides distasonis can deglycosylate capecitabine, prematurely activating it before systemic absorption⁷⁵. A similar mechanism is seen with gemcitabine (another antimetabolite), which is deaminated by the bacterial cytidine deaminase (CDA) expressed by certain tumour-associated Gammaproteobacteria, leading to loss of drug efficacy⁷⁶. Although drug inactivation by microbial enzymes can compromise efficacy, the same mechanism has also been shown to reduce toxicity in preclinical models⁷⁷.

By contrast, microbiome-encoded genes can also activate a drug. This effect could be beneficial, but only when it takes place at the desired location. For example, the topoisomerase I inhibitor irinotecan (commonly used in CRC treatment⁷⁸) is hydrolysed in the liver to its active metabolite SN-38 and subsequently inactivated by hepatic glucuronidation to SN-38G before being excreted into the gut via bile. Gut bacterial β-glucuronidases (GUS) can in turn reactivate SN-38G to SN-38, which causes epithelial injury, mucositis and diarrhoea⁷⁸. Although intratumoural reactivation might theoretically boost efficacy, gut reactivation contributes to dose-limiting toxicity, thereby highlighting the importance of the site of microbial activity.

Beyond the effect of the gut microbiota on drug metabolism, cancer drugs can themselves alter gut microbiome composition and function, via direct antimicrobial activity, or secondarily via effects on the host. For instance, 5-FU has antimicrobial activity, which could deplete beneficial taxa and increase susceptibility to opportunistic infections⁷⁹. Platinum-based agents (such as oxaliplatin) accumulate in the enteric nervous system, inducing oxidative stress and neuronal loss that can secondarily alter gut microbial communities and contribute to chemotherapy-induced mucositis and gut–brain axis-related symptoms, such as fatigue and cognitive decline⁸⁰. Cancer therapies can also impair epithelial barrier integrity, either directly or through microbiome-mediated pathways. For example, PPARγ signalling is central to the barrier-protective effects of butyrate and reduced bacterial activation of this pathway has been associated with impaired barrier integrity^45,54. Loss of barrier integrity facilitates translocation of bacteria and microbial products, exacerbating mucosal inflammation and increasing the risk of mucositis, surgical site infections and even cancer recurrence^43–45.

Collectively, these studies reveal a dynamic system in which therapeutic agents, the microbiome and host physiology are tightly interlinked. Effects can originate from the drug, the host or the microbiome, but are often propagated through self-reinforcing feedback or feedforward loops. This complexity underscores the holobiont concept, the combination of the host and all its associated microorganisms, which highlights the importance of considering the host–microbiome system as a unified therapeutic target.

The immune system and therapy outcomes

The gut microbiome can influence cancer therapy outcomes by modulating the balance of immune activation and suppression in both the gut and TME, potentially affecting the levels and activity of cytotoxic CD8⁺ T cells, dendritic cells, macrophages, neutrophils, regulatory T (T_reg) cells and myeloid-derived suppressor cells (MDSCs)⁸¹. In addition, the microbiome affects (mucosal) immune responses through alterations in local and systemic signalling, disruption of gut barrier integrity, and regulation of metabolic and nutritional pathways within the TME. In this section, we highlight key microbiome–immune interactions relevant to cancer treatment outcomes.

Immunotherapy response: efficacy.

Immune checkpoint inhibitors (ICIs), including anti-programmed cell death protein 1 (anti-PD1), anti-programmed cell death 1 ligand 1 (anti-PDL1) and anti-cytotoxic T lymphocyte protein 4 (anti-CTLA4) antibodies block inhibitory signals on T cells, thereby enabling cytotoxic CD8⁺ T cells expressing IFNγ and secreting perforin and granzymes to mount effective antitumour responses^82,83. However, this immune activation is counterbalanced by T_reg cells and MDSCs, which facilitate immune evasion and contribute to therapeutic resistance⁸⁴.

A key mechanism by which the microbiome can influence ICI efficacy is by modulating cytotoxic CD8⁺ T cell activity in the gut, systemic circulation or TME. Multiple early studies have independently demonstrated that gut microbial composition correlates with response to anti-PD1 and anti-PDL1 therapy^6,85–87. However, the specific taxa linked to response varied, with some studies highlighting Akkermansia, whereas other studies have found associations with Bifidobacterium or Ruminococcus^87,88. Despite this taxonomic variability, the immune-modulatory mechanisms frequently appear to converge on increased numbers or activity of CD8⁺ T cells, potentially through modulation of distinct pathways. For example, in preclinical models, oral administration of Bifidobacterium enhanced dendritic cell activation and CD8⁺ T cell priming, which enhanced the response to PDL1 blockade⁸⁹. Similarly, Akkermansia muciniphila enhanced the response to PD1 inhibition by promoting CD8⁺ T cell recruitment to the tumour⁶. These findings suggest that ICI response-associated taxa might share the ability to induce dendritic cell-dependent CD8⁺ T cell priming in mesenteric lymph nodes or intestinal mucosa, followed by (chemokine-mediated) infiltration into the TME. Notably, although favourable taxa have been associated with ICI efficacy across cancers in pan-cancer cohorts⁸⁸, cancer type should still be considered when identifying beneficial taxa and underlying immune mechanisms.

Distinct microbial communities could yield similar metabolic or immunologic outputs and this functional redundancy might be the key determinant of therapeutic response⁹⁰. For example, the purine derivative inosine, which can be produced by multiple gut microorganisms, can enhance IFNγ production by CD8⁺ T cells through activation of the adenosine A2A receptor⁹¹. Likewise, any bacterial peptidoglycan-derived muropeptides that stimulate NOD2 are expected to expand cytotoxic CD8⁺ T cell populations⁹². Similarly, SCFAs, including butyrate and pentanoate, can be produced by different bacteria and promote CD8⁺ T cell function, and enhance responses to chemotherapy and immunotherapy in both intestinal and extraintestinal cancers in preclinical models^93–95. In addition, the immunologic effects of microbial metabolites are context dependent. For example, butyrate can enhance cytotoxic CD8⁺ T cell function and support antitumour immunity, yet butyrate might also promote immunosuppressive T_reg cell expansion under other conditions^96–98.

The TME has a key role in determining treatment response. Moderate inflammation supports CD8⁺ T cell infiltration and activity, enhancing antitumour immunity⁹⁹, whereas chronic inflammation can promote the expansion of T_reg cells and MDSCs, leading to immune exhaustion and ICI resistance¹⁰⁰. Tryptophan metabolism exemplifies how immune tone is shaped by host–microorganism interactions. Host indoleamine 2,3-dioxygenase (IDO) activity converts tryptophan into immunosuppressive kynurenine, dampening T cell responses^101,102. By contrast, microbial metabolism of tryptophan in the TME can generate indole derivatives such as indole-3-acetic acid and indole-3-propionic acid, which enhance CD8⁺ T cell function and improve responses to chemotherapy and immunotherapy across multiple cancers through metabolic or epigenetic effects^10,103.

Although most of the literature focuses on the interplay between the gut microbiota and ICIs, emerging evidence highlights the role of the microbiota in other immune-modulating therapies, including dendritic cell vaccines¹⁰⁴, adoptive T cell approaches (such as chimeric antigen receptor (CAR) T cell therapy)^105–107, haematopoietic stem cell transplantation¹⁰⁸ and radiation therapy¹⁰⁹, potentially through analogous underlying mechanisms.

In summary, tissue context, local cytokine milieu and microbial niche all have a critical influence in determining the immune-modulatory effects of microbial metabolism^110–112 (Fig. 3).

Immunotherapy response: adverse events.

Immune activation by ICIs is a double-edged sword, as it can also lead to off-target effects in the form of immune-related adverse events (irAEs). These irAEs seem to be driven by autoreactive CD4⁺ and CD8⁺ T cells and resemble auto-immune or inflammatory disorders^113,114. Like studies investigating ICI efficacy, several reports have linked baseline gut microbiome composition to the risk of developing irAEs^114–117. However, these studies have failed to identify consistent pre-treatment microbial signatures predictive of irAE risk. Although the lack of detectable microbial signatures might partly reflect functional redundancy among microorganisms (as seen in efficacy studies), additional limitations are particularly relevant in the context of irAEs. These include heterogeneity in patient populations, small cohort sizes and inconsistencies in the definition and classification of irAEs across studies^{115,117–121}. Moreover, many studies aggregate irAEs into a single outcome category, overlooking the likelihood that the mechanisms driving irAEs might vary across organ systems.

Recent preclinical studies using ICI-colitis mouse models have begun to provide mechanistic insights into the role of the gut microbiome in irAEs. In one model, colitis was triggered by colonization of wild-type BALB/c and C57BL/6 mice with a pro-inflammatory gut microbiome from donor TRUC mice (which develop spontaneous colitis)¹²². By contrast, in another study, a gut microbiome from wild-caught mice was sufficient to induce colitis in C57BL/6 wild-type mice following ICI treatment¹²³. The dextran sodium sulfate colitis model has also been used in combination with ICI injections to model ICI colitis^124,125. These findings support a causal role for an inflammatory gut microbiome in driving ICI-induced intestinal toxicity and underscore the potential for identifying specific pre-treatment microbial compositions, functional pathways or metabolites that predispose to ICI colitis or other irAEs¹¹⁴.

Chemotherapy response.

Similar to immunotherapy, several studies have shown that the baseline gut microbiota composition associates with chemotherapy response. For example, specific microbial profiles were associated with the response to platinum-based chemotherapy in patients with epithelial ovarian cancer¹²⁶. Similarly, in patients with oesophageal or gastroesophageal junction cancer receiving folinic acid, 5-FU and oxaliplatin, faeces from responders had higher relative abundances of Bifidobacterium adolescentis, Ruminococcus gnavus and A. muciniphila, taxa also linked to improved ICI efficacy and CD8⁺ T cell priming¹²⁷. This consistency in observed taxa suggests that overlapping immune-related microbial mechanisms might modulate responses to both chemotherapy and immunotherapy.

More broadly, chemotherapeutic agents often damage the gut epithelium, compromising barrier integrity and facilitating microbial translocation, an effect that, although potentially harmful, can also stimulate systemic antitumour immunity. For example, cyclophosphamide promotes bacterial translocation to secondary lymphoid organs, inducing T helper 17 (T_H17) cell responses that contribute to its therapeutic effect¹²⁸. Similarly, translocated microbial products (such as LPS) activate monocytes and dendritic cells, providing immunostimulatory cues during treatment^129–132. Bacteria in the small intestine can also influence this immune stimulation. For example, B. fragilis in the ileal mucosa enhanced oxaliplatin efficacy through immune activation triggered by epithelial apoptosis¹³³.

Together, these findings underscore the pivotal role of microbial regulation of immune tone in shaping both the efficacy and toxicity of cancer therapies, including immunotherapy and chemotherapy (Box 1).

Box 1 |. Lessons learnt from microbiome outcome studies.

1. The holobiont concept: host–microorganism–drug as an integrated unit

The host and microbiome form a dynamic, co-regulated system wherein therapy outcomes emerge from their interaction rather than isolated effects. Drugs modify the microbiome, which in turn alters drug metabolism, immune tone and barrier function, often in feedback loops. This interdependence makes it difficult to disentangle cause from correlation and could partly explain why many microbial signatures fail to replicate across studies.

2. Causal inference is challenged by redundancy and plasticity

Functional redundancy (for example, different microorganisms producing the same immune-modulatory metabolites) and plasticity (for example, microbiome shifts in response to therapy or disease) further obscure direct cause–effect relationships. What seems to be a predictive microbial signature could, in fact, be a surrogate for broader ecosystem dynamics or host factors.

3. Microbiome effects are highly context-dependent

The effect of the microbiome on therapy is not uniform. Microbial metabolites can either enhance cytotoxic immune responses or promote immune suppression, depending on tissue type, local cytokines and host immune tone. Similarly, the same bacterial species might improve efficacy in one setting or cancer type and drive resistance or toxicity in another.

For example:

• Butyrate can enhance cytotoxic CD8⁺ T cell function and support antitumour immunity in some settings, while stimulating regulatory T cells and leading to an immunosuppressive environment in other settings^96–98.

• Intratumoural formate produced locally by Fusobacterium nucleatum promotes tumour progression and metastasis²¹⁹, whereas elevated systemic formate supports immune checkpoint inhibitor-induced antitumour immunity by enhancing CD8⁺ T cell activation¹⁸⁴.

• F. nucleatum in the tumour promotes chemoresistance in colorectal cancer (CRC)²²⁰, yet paradoxically, in microsatellite-stable CRC (a subtype typically resistant to immunotherapy), it enhances PD1/PDL1 blockade responses²²¹.

• β-Glucuronidase activity can reactivate irinotecan and worsen gut toxicity, but similar reactivation in tumours might enhance efficacy⁷⁸.

• Moderate levels of Akkermansia are associated with a favourable response to PD1 blockage in non-small-cell lung cancer, but high levels are associated with resistance to treatment⁸⁷.

These examples illustrate that the same taxa or functions can be either detrimental or beneficial depending on the treatment, abundance and/or site of action.

Microbiome therapies in cancer treatment

With growing evidence linking the gut microbiome to both the efficacy and toxicity of cancer therapies, a logical next step is to explore whether manipulation of the microbiome can enhance treatment outcomes. A range of strategies is under investigation, including general or targeted dietary interventions, administration of individual strains or defined microbial consortia (either native or engineered) driving specific microbial functions^134,135, FMT and postbiotics^12,136,137 (Table 1). Additional microbiome-targeted approaches, such as vaccines¹³⁸, narrow-spectrum antibiotics¹³⁹ and bacteriophage therapies¹⁴⁰, are also being explored. Below, we highlight how these interventions align with the previously described mechanisms by which the microbiome affects the results of cancer treatment.

Table 1 |.

Microbiome-based therapeutics strategies in cancer

Target area	Strategies	Maturity rankinga	Examples
Genotoxin	Inactivation	i	Hydrolytic inactivation of colibactin by ClbS Blockade of colibactin maturation Expression of MfsX by Klebsiella, an efflux transporter for tilimycin141–143
	Mitigation of effects and detoxification	i	Repair of tilimycin-induced damage by UvrX Upregulation of PXR and host detoxifying enzymes by microbially produced indole Stimulation of DNA repair141,145,146,212
	Prevention of expansion	i	Inhibition of bacterial growth or decolonization (for example, supplementation with putrescine, bacterial consortia and dietary carbohydrates)144,150,152
Drug metabolism	Inhibitors of GUS enzymes	i	Inhibition of the reactivation of therapeutic metabolites and prevention of toxicity (for example, probiotics and FMT)78,154,157
	Bacterial inactivation of chemotherapy to prevent toxicity	ii	Conversion of 5-FU into an inactive form by a bacterial homologue of DPD79
Immune tone	FMT	iv	Improved response to ICIs in patients with melanoma158–160
	Probiotics	iii	Improved response to ICIs with bacterial supplementation (for example, Lactobacillus reuteri and Lactobacillus gallinarum)11,176
	Prebiotics and dietary interventions	iii	Improved response to immunotherapy with a high-fibre diet, ketogenic diet and oral supplements (for example, castalagin and vitamin D)168,170,171

5-FU, 5-fluorouracil; DPD, dihydropyrimidine dehydrogenase; FMT, faecal microbiota transplantation; GUS, β-glucuronidase; ICIs, immune checkpoint inhibitors; Mfs, major facilitator superfamily; PXR, pregnane X receptor.

^a‘iv’ indicates the closest therapeutic to broad clinical adoption whereas ‘i’ shows the farthest.

Genotoxin targeted therapies

Given the role of microbial genotoxins (such as colibactin and tilimycin) in cancer pathogenesis, several therapeutic strategies are being explored to mitigate their effects. These include disrupting genotoxin biosynthesis, leveraging bacterial self-immunity mechanisms, modulating host detoxification pathways, using small-molecule inhibitors, dietary interventions and microbial preparations¹⁴¹. Such strategies can be particularly relevant in early life, given that that the mutational signature of colibactin could represent an early mutation occurring in the development of in colorectal cancer³⁰.

Bacteria have evolved intrinsic mechanisms to protect themselves from their own genotoxins, which could be repurposed therapeutically. For example, E. coli produces ClbS, which chemically inactivates colibactin. Moreover, some Klebsiella species express MfsX, an efflux transporter for tilimycin, and UvrX, a DNA repair enzyme specific to tilimycin-induced damage^27,141–143. Supplementation with putrescine, a microbiota-derived polyamine, can also suppress expansion of pks⁺ E. coli in C57BL/6 wild-type mice¹⁴⁴.

An alternative strategy is to enhance host detoxification. For instance, microbially produced indole non-enzymatically reacts with tilimycin to produce tilivalline, which in turn activates the pregnane X receptor (PXR), a nuclear receptor that upregulates host detoxifying enzymes conceivably to mitigate broad damage from toxin exposure¹⁴⁵. A promising small-molecule approach includes the development of a boronic acid-based inhibitor targeting ClbP, the peptidase required for colibactin maturation. This inhibitor effectively blocked genotoxic effects of colibactin in human cells without broadly disrupting the gut microbiota¹⁴⁶.

Complementary to these strategies, dietary antigenotoxic compounds (including flavonoids, polyphenols, isothiocyanates and dietary fibre) have demonstrated potential to reduce DNA damage through direct detoxification, free radical scavenging and stimulating DNA repair mechanisms¹⁴⁷. These mechanisms are conceptually aligned with how dietary phytochemicals modulate epigenetic pathways to prevent cancer¹⁴⁸. Many genotoxins are encoded by Enterobacteriaceae, such as E. coli and Klebsiella; therefore, strategies to reduce these species could be beneficial. Relevant approaches include increasing dietary fibre diversity^149–151, administering defined bacterial consortia designed to displace pathogenic strains¹⁵², or using human milk oligosaccharide (HMO)-degrading Bifidobacterium strains in combination with HMOs as synbiotics to crowd out Enterobactericeae¹⁵³.

Targeting microbial metabolism of drugs

As outlined earlier, microbial metabolism of chemotherapeutic agents can notably influence both efficacy and toxicity. Although mechanistic understanding is growing, targeted interventions in humans remain largely conceptual. For example, microbial GUS enzymes can reactivate metabolites of tyrosine kinase inhibitors (TKIs) and irinotecan in the gut, thereby contributing to gastrointestinal toxicity^78,154. Thus, GUS inhibitors or depletion of GUS-producing taxa could mitigate such adverse effects. However, although bacterial GUS inhibitors have shown promise in mitigating irinotecan and TKI-induced toxicity in preclinical studies^78,154, clinical translation has been limited by the heterogeneity of bacterial β-glucuronidases and lack of human studies¹⁵⁵. In a small study, FMT resolved TKI-induced diarrhoea in seven of ten patients¹⁵⁶, although whether this effect was owing to reduced GUS activity in donor microbiota remains unclear. Given the proposed similar GUS-dependent mechanism, FMT or probiotics might also hold promise for irinotecan-induced toxicity¹⁵⁷. Similarly, delivery of native or engineered bacteria expressing DPD that degrade 5-FU locally in the gut could help counter 5-FU-induced gastrointestinal toxicity without compromising systemic efficacy.

Boosting the systemic or TME immune system

Microbiome-based interventions have primarily aimed at enhancing cancer immunosurveillance by boosting CD8⁺ cytotoxic T cell activity, either systemically or within the TME. Although this mechanism is not the only relevant one, it is central to the success of strategies such as FMT, dietary modulation, probiotics, engineered microorganisms and microbial metabolite supplementation. Two seminal FMT studies have demonstrated that microbiota transfer from patients with melanoma who responded to ICI could convert a subset of ICI refractory patients into responders, accompanied by increased CD8⁺ T cell activation¹⁵⁸ or favourable changes in immune cell infiltrates and gene expression profiles in both the gut lamina propria and the TME¹⁵⁹. More recently in 2023, a phase I FMT study on first-line anti-PD1 has demonstrated that faecal samples from patients with advanced melanoma who had received healthy donor FMT and were treatment responders restored anti-PD1 efficacy in antibiotic-treated or germ-free tumour-bearing mice. This was accompanied by increased memory CD8⁺ T cells and TIM3⁺ CD8⁺ T cells in the TME. By contrast, pre-FMT faecal samples from these patients did not affect anti-PD1 efficacy in the mice. These findings raise the possibility that FMT could be incorporated into first-line melanoma treatment¹⁶⁰. Many clinical trials are ongoing that combine FMT with ICI regimens in different cancers. Early results seem promising; for example, preliminary results of the TACITO trial (NCT04758507) show promise for FMT in increasing the efficacy of PD1-based therapies in patients with metastatic renal cell cancer¹⁶¹. Whether the mechanisms identified in early FMT studies can be verified by these larger clinical trials remains to be seen.

Although enhanced CD8⁺ T cell responses and reduced CD4⁺ FOXP3⁺ T_reg cells can potentiate ICI efficacy, this same immune skewing might contribute to irAEs¹¹⁴. In a preliminary study of two patients with ICI-induced colitis, FMT led to symptom resolution and was associated with a reduction in CD8⁺ T cells and a concomitant increase in CD4⁺ FOXP3⁺ T cells¹⁶². Whether FMT can selectively enhance efficacy while also mitigating toxicity remains an open and important question¹⁶³. Related to this question, selecting faecal material from donors who have shown a favourable response to the administered ICI without developing irAEs seems prudent.

Dietary modifications have been a major focus of microbiome-targeted interventions. In an observational study, high fibre intake (without probiotic use) was associated with improved ICI outcomes. This finding was supported by preclinical studies, which showed that mice on a low-fibre diet (or receiving probiotics) during ICI treatment exhibited reduced frequencies of IFNγ⁺ CD8⁺ cytotoxic T cells within the TME¹⁶⁴. This study, along with other favourable associations with fibre intake, has led to several ongoing trials investigating the therapeutic impact of fibre modulation. The BE GONE Trial (NCT02843425) randomized patients to their usual diet with or without a daily cup of study beans. The consumption of beans increased Faecalibacterium, Eubacterium and Bifidobacterium, accompanied by favourable shifts in systemic nutrients, microbiome-derived metabolites and proteomic markers of inflammation¹⁶⁵. Long-term cancer outcomes, however, remain to be determined. In addition, the DIET (NCT04645680) double-blinded study has recently been completed, which recruited patients with melanoma receiving immunotherapy who were randomized 2:1 to receive either 20 or 50 g fibre per day from whole foods¹⁶⁶. However, adherence was limited owing to challenges of a fully controlled feeding protocol during active cancer treatment¹⁶⁷.

Alternatively, the ketogenic diet (a high-fat, low-carbohydrate diet in which ketone bodies are the primary energy source) has shown promise in enhancing CD8⁺ T cell responses during ICI therapy. In preclinical models, a ketogenic diet improved immunotherapy efficacy, with the primary ketone body (β-hydroxybutyrate) promoting the expansion of CXCR3⁺ activated T cells via the GPR109A receptor¹⁶⁸. Notably, supplementing a standard mouse diet with β-hydroxybutyrate alone recapitulated this benefit, highlighting its potential to be tested in humans. Other targeted dietary approaches have focused on polyphenols and vitamin D. Polyphenols are naturally found in plant foods, such as berries and tea, and are interesting targets for dietary interventions as they are inaccessible by the host and pass to the colon, wherein they feed specific members of the microbiome¹⁶⁹. In mice, oral administration of polyphenol extract was found to increase cytotoxic CD8⁺ T cells, which was recapitulated by a single polyphenol, castalagin¹⁷⁰. Castalagin feeding increased abundance of Ruminococcaceae and enhanced the CD8⁺:FOXP3⁺CD4⁺ T cell ratio in the TME, favouring cytotoxic over T_reg cell responses¹⁷⁰. Finally, vitamin D supplementation in preclinical models modulated the gut microbiome and enhanced ICI efficacy, presumably through enhancement of antitumour immunity¹⁷¹, although the efficacy of vitamin D in the clinical setting is yet to be better characterized.

Although some probiotics can mitigate treatment-related toxicities such as diarrhoea¹⁷², oral mucositis¹⁷³ and infections¹⁷⁴, mechanistic clarity remains limited¹⁷⁵. One example of a mechanism was the finding that the probiotic Lactobacillus reuteri translocated into melanoma tumours and produced indole-3-aldehyde, a tryptophan-derived AhR agonist that promoted IFNγ⁺ CD8⁺ T cell responses and improved ICI efficacy in mice¹¹. Notably, serum levels of indole-3-aldehyde correlated with ICI responses in patients with advanced melanoma¹¹. By contrast, in absence of bacterial translocation, luminal production of the tryptophan metabolite indole-3-carboxylic acid by the probiotic Lactobacillus gallinarum improved anti-PD1 efficacy in preclinical models through reduction of intratumoural infiltration of T_reg cells, hence enhancing the activity of CD8⁺ T cells¹⁷⁶. The next generation of probiotics offer promising strategies to enhance antitumour immunity. One approach exploits the tumour-localizing capacity of certain bacteria to deliver immunostimulatory payloads. For example, E. coli Nissle 1917 engineered to convert ammonia into L-arginine (a metabolite known to fuel T cell metabolism and augment ICI responses) enhanced the efficacy of PDL1 blockade in preclinical models^177–180. In some cases, microbial metabolites might not require production within the TME but can be supplemented orally. Inosine was shown to activate antitumour T cells systemically⁹¹, and oral inosine enhanced ICI efficacy in a phase II trial for advanced solid tumours¹⁸¹.

A complementary approach involves leveraging molecular mimicry. Gut microbial peptides that resemble tumour-associated antigens can be used in cancer vaccines to elicit cross-reactive CD8⁺ T cells capable of targeting tumours. This approach has shown preclinical efficacy¹⁸² and is being explored in multiple clinical trials (ROSALIE (NCT04116658) and Spencer (NCT04187404) trials). Finally, the protective effect of exercise on cancer survival¹⁸³ might be mediated by the microbiome, as exercise in preclinical models increased circulating levels of the microbial fermentation product formate, which in turn boosted antitumour CD8⁺ T cell function¹⁸⁴.

Despite mounting evidence linking the microbiome to cancer therapy outcomes, microbiome-based interventions face several critical challenges (Box 2). To overcome these limitations, future strategies should assign priority to mechanism-based interventions, especially in large pragmatic intervention trials across diverse cancer types and stages. Incorporating the microbiome as a variable in cancer drug pharmacokinetic and pharmacodynamic studies will enable us to develop better therapeutic strategies. Moving beyond FMT, scalable and targeted modalities (such as defined microbial consortia, engineered strains and postbiotics) could offer more practical and control-lable solutions. Crucially, clinical trials must account for the unique constraints faced by patients with cancer, including reduced mobility, appetite and treatment burden, and focus on functional and clinically meaningful end points within real-world settings.

Box 2 |. Challenges in strategies harnessing the microbiome for cancer treatment.

1. Mechanistic uncertainty of generalized approaches

Many current interventions (for example, high-fibre or ketogenic diets, and traditional probiotics) — are explored without mechanistic understanding of how specific microbial taxa or functions modulate therapy efficacy or toxicity. This lack of understanding has led to generalized strategies targeting broad pathways, often without validating whether the intended microbial or immune targets are engaged in individual patients. The gut epithelial barrier is frequently targeted, yet its clinical measurement is inconsistent across studies, ranging from histology and in vitro assays to surrogate markers. Moreover, the functional role of barrier modulation remains paradoxical: although barrier protection is thought to reduce toxicity, some anticancer effects may require microbial translocation to stimulate immune pathways.

2. Context dependence and interindividual variability

The effect of the microbiome is shaped by host genetics, tumour type, treatment regimen, baseline microbiota composition and modifiable environmental factors (such as diet, alcohol consumption, tobacco use and physical activity). This strong context dependence makes generalization challenging and highlights the importance of precision-guided approaches.

3. Lack of individualized biomarkers

Linked to the points above is a lack of validated biomarkers to guide patient stratification or mechanistically align interventions with clinical needs, for example, whether to target microbial drug metabolism, barrier function or immune tone (for example, enhancing CD8⁺ T cells versus expanding regulatory T cells). The lack of biomarkers limits personalization and could contribute to conflicting outcomes across studies and heterogeneity in patient outcomes.

4. Limited focus of diet intervention studies

In addition, most diet intervention studies incorporating the gut microbiome have been focused on cancer immunotherapy outcomes. However, diet is important in cancer efficacy and cancer-specific adverse events. For example, a need exists for dietary interventions that are agreeable for patients and ameliorate cachexia²²². One study has found a beneficial effect of the Mediterranean diet in mitigating cachexia²²³, which is also considered beneficial for the gut microbiome.

5. Translational and practical barriers

Many promising strategies, such as engineered microorganisms, bacterial vaccines or microbial metabolite supplementation, remain preclinical. Translating these into scalable, safe and regulatory-compliant therapies is non-trivial. Dosing, pharmacokinetics and context-dependent host responses remain poorly defined.

6. Limitations of current clinical trial design

Microbiome trials in cancer patients often face feasibility challenges. Many patients undergoing chemotherapy or immunotherapy are immunocompromised, cachectic or experiencing gastrointestinal toxicity, making adherence to demanding dietary regimens or oral therapies difficult. For example, high dropout rates in fully controlled feeding trials underscore the impracticality of some interventions in real-world oncology settings. In addition, the prevalent use of antibiotics in this patient population might also complicate microbiome composition assessment and interfere with the colonization of beneficial bacterial communities or strains. Finally, end points in trials often focus on microbiome composition or measurement of immune markers rather than validated functional or clinical readouts.

Outlook on microbiome diagnostics in cancer

The microbiome offers considerable potential across the cancer care continuum, from enhancing treatment efficacy and minimizing toxicity to enabling early detection, prevention and personalized treatment strategies. However, realizing this promise requires addressing key scientific, technical and clinical challenges.

Early detection and risk stratification

The gut microbiome has been explored as a potential tool for early cancer detection, particularly for CRC. Most studies to date have used cross-sectional designs, comparing the microbiome profiles of patients with cancer and healthy control individuals to identify compositional differences interpreted as diagnostic signatures^185,186. However, several limitations undermine the clinical applicability of these findings. First, most microbiome studies rely on relative rather than absolute abundance data, which introduces compositional and technical biases¹⁸⁷. In addition, tissue microbiome studies that aim to detect intratumoural or tumour-associated microorganisms are complicated by the challenges inherent to low-biomass analyses, which affect both amplicon sequencing and whole-genome sequencing approaches^3,188–192. Second, patients with cancer often differ from healthy individuals in multiple ways unrelated to cancer itself (such as comorbidities, medication use or systemic inflammation) that can independently alter microbiome composition^193–195. These confounders limit the specificity of identified microbial signatures.

Moreover, microbiome-based classifiers rarely achieve the high sensitivity and specificity thresholds required for clinical diagnostics¹⁹⁶. To overcome these limitations, larger and more diverse cohorts, including individuals with various cancer types, could help isolate cancer-specific microbiome signals. Additionally, integrating microbiome data with complementary biomarkers can improve predictive performance. For example, in pancreatic ductal adenocarcinoma, combining a panel of 27 microbial species with serum carbohydrate antigen 19–9 levels substantially enhanced diagnostic accuracy¹⁹⁶. Besides improved cross-sectional studies, large longitudinal natural history studies of cancer prediction are needed, in which many healthy individuals are followed over time with microbiome and other omics measurements and subsequently assessed for disease onset. Such studies also help account for the considerable inter-individual variation, as longitudinal deviations from a personal baseline could be more informative than deviations from the population average.

Companion diagnostics

Most research has focused on using the microbiome as a diagnostic for early detection or prognosis; however, it might be best suited as a companion diagnostic to guide treatment decisions, stratify patients by probable response or risk, and mitigate adverse events. Measuring the pretreatment gut or tumour microbiome could help identify individuals that have higher tendency to respond to a given therapy, those at higher risk of irAEs, or those who might benefit from a microbiome-targeted intervention. Ideally, such diagnostics should be rooted in mechanistic understanding rather than purely associative data.

Emerging examples include stratifying patients for ICI therapy based on the abundance of Ruminococcus species, which has been associated with favourable ICI responses^164,197,198. Another clinical application could be selecting between anti-PD1 or anti-PDL1 monotherapy versus combination treatment with anti-CTLA4. As combination regimens carry a higher risk of colitis¹⁹⁹, patients predicted to be at high risk for ICI-colitis might be better suited for monotherapy.

Beyond prognostication, companion diagnostics should ideally guide therapeutic decision-making. For instance, a microbiome-based diagnostic could determine whether an invasive intervention such as FMT is warranted as a first-line adjunct, reserving it for patients predicted to respond poorly to immunotherapy alone. Similarly, high-fibre dietary interventions might only be effective if the microbiome of a patient retains the capacity to degrade fibre. If this capacity is severely impaired, a defined microbial community or FMT might need to precede dietary modulation. Another promising avenue is the development of personalized probiotic interventions tailored to the microbiome composition and functional profile of an individual.

Leveraging artificial intelligence

Artificial intelligence (AI) is poised to revolutionize microbiome research by enabling advanced data integration, pattern recognition and predictive modelling (Fig. 4). However, several key challenges must be addressed for AI to reach its full potential in this field. One major hurdle is the ‘curse of dimensionality’, wherein the number of measured microbial or multi-omics features far exceeds the number of available samples, which heightens the risk of overfitting and spurious associations. Most cancer microbiome studies to date include relatively small cohorts (typically fewer than 500 individuals), limiting statistical power and generalizability. To address this limitation, researchers have used strategies such as grouping microbial features into co-abundance groups or combining different types of data in multiple steps to simplify and reduce the number of variables they need to analyse^200–202. Although this approach can improve model stability, it may also obscure biologically important, strain-specific signals, as the strain is the ultimate functional unit in the microbiome²⁰³. Increasing the number of samples through data augmentation via synthetic sampling or generative models offers another potential solution, although these methods must be used cautiously to avoid introducing bias. Transfer learning presents a promising strategy in which models are pretrained on large, publicly available datasets and then fine-tuned on smaller, context-specific cohorts. However, these approaches depend on standardized metadata and harmonized protocols to not introduce bias²⁰⁴. Additionally, striking the right balance between model complexity and interpretability is critical. Highly complex models might yield strong performance but often lack transparency, which hinders their clinical utility. The biological relevance of AI-derived outputs must also be rigorously evaluated to ensure meaningful insights rather than mere statistical artefacts.

Fig. 4 |. Conceptual framework for AI-driven precision oncology using multiomics integration.

Patients with cancer exhibit substantial heterogeneity across multiple biological layers including microbiome composition (metagenome and metabolome), host features (clinical and demographic variables and comorbidities) and exposome, and host immune profiles (cytokines and serum metabolome). Artificial intelligence (AI) approaches including dimensionality reduction, synthetic data augmentation, transfer learning, complexity–interpretability balancing, multi-omics integration, neural networks and digital twin models are applied to integrate these diverse data types. AI analysis identifies pathogenic mechanisms that lead to disease and predicts clinical outcomes, including treatment efficacy and potential adverse effects. This integrated approach enables the development of targeted therapies that can be tested in preclinical models, ultimately leading to novel diagnostic tools and personalized treatment strategies.

Emerging innovations in this space include using variational Bayesian neural networks to integrate paired microbiome and metabolome data by modelling the uncertainty that is inherent in high-dimensional multi-omics datasets²⁰⁵, or using neural networks to predict enzyme function, as recently applied in studies of bile acid metabolism²⁰⁶. In addition, multi-omics datasets have been converted into image-like formats that can be analysed using convolutional neural networks, leveraging tools from the more mature field of AI-based image analysis²⁰⁷. Separately, neural net or large language models are being trained to parse raw sequencing data directly, potentially circumventing the limitations of current taxonomic classification systems^208,209. Together, these AI-driven approaches have the potential to substantially advance both discovery and implementation in microbiome science, but thoughtful application and rigorous validation remain essential²⁰⁴.

The microbiome can also be incorporated into the concept of a digital twin, which is a dynamic, virtual replica of a patient that integrates static data such as genetic information and real-time data such as clinical records, physiological measurements and omics data, to simulate, monitor and predict health outcomes²¹⁰. Multiscale AI models and Boolean network approaches are emerging to simulate such complex interactions, supporting hypothesis-driven interventions and personalized therapy adjustments²¹⁰. Developing digital twins tailored to different life stages (for example, one for infancy²¹¹ and another for adulthood) is a promising direction for the microbiome field. Such models account for each baseline microbiome of an individual, which reflects personal lifestyle factors, and can also provide a valuable readout for microbiome-targeted interventions across the lifespan.

Conclusions

Growing evidence positions the gut microbiome as a key modulator of cancer therapy outcomes, acting through diverse mechanisms including immune regulation, drug metabolism and epithelial barrier integrity. Although this insight has fuelled the development of novel diagnostic and therapeutic strategies, clinical translation remains constrained by mechanistic uncertainty, a lack of personalized biomarkers, and implementation challenges in cancer populations. Microbiome outcome studies reveal that the host and microbiota operate as an integrated system in which therapeutic outcomes emerge from dynamic host–microorganism–drug interactions. Functional redundancy, microbial plasticity and strong context dependence make causal inference difficult and might explain the inconsistent replication of microbial signatures across studies. Moreover, many current interventions are based on generalized strategies rather than mechanistic understanding, and interindividual variability further complicates personalization. Translational progress is limited by the scarcity of validated biomarkers, the narrow focus of dietary and probiotic studies, and the practical barriers to conducting large, well-controlled clinical trials. Advancing this field might require a transition towards AI-enabled, precision-guided interventions that are anchored in mechanistic understanding, supported by robust companion biomarkers, and validated through large, longitudinal pragmatic clinical trials. Achieving clinical impact will require microbiome therapeutics designed to meet both the biological demands and real-world constraints of cancer care. As microbiome science becomes increasingly integrated into oncology, it promises to redefine cancer care by coupling mechanistic precision with clinical relevance to achieve more effective and tolerable therapies.