The Cancer Microbiome: Recent Highlights and Knowledge Gaps

The community of bacteria, archaea, fungi, and viruses on and in various sites of the body constitutes the human microbiome. A flourishing literature now links microbiome composition to an assortment of diseases from inflammatory bowel disease to schizophrenia (1, 2). Bacteria are an abundant component of the microbiome, particularly in the gut, where bacteria are estimated to number ∼3.8 × 10¹³ in total (3). Recent advances in our understanding of the mechanisms by which bacterial members of the microbiome may initiate or promote tumorigenesis throughout not just the gastrointestinal tract but also other organs are the focus of this review. Experimental models suggest that bacteria can foster the induction and development of tumor formation through multiple mechanisms: (i) direct DNA-damaging effects of bacterial toxins, (ii) bacterial metabolites (such as products of Western diet metabolism), (iii) inflammation driven by direct physical interaction with host cells, chronic infections, and invasive biofilm formation, and (iv) inhibition of antitumoral immune responses (Fig. 1). Multiple theories have arisen regarding bacterial involvement in tumorigenesis, such as the driver–passenger model, including extension to the keystone hypothesis; the hit-and-run model; chronic dysbiosis; and alterations in the spatial distribution of the microbiome or barrier integrity in epithelial tissues (Fig. 2). For the first, an individual or collection of driver procarcinogenic bacteria collaborate with microbiome passengers to promote tumorigenesis. In the keystone hypothesis, the carcinogenic driver is sufficient to drive carcinogenesis. For the hit-and-run model, transient colonization and damage by a procarcinogenic bacterium is necessary and sufficient to drive tumorigenesis (4). An important note is, with the exception of Helicobacter pylori, no additional bacterial species to date has been identified as the causative agent of human tumorigenesis in the absence of a prior host cell genetic mutation. The common link between each model is the disruption of a healthy microbiome, termed dysbiosis. However, a well-defined healthy microbiome membership remains elusive due to significant person-to-person variation. Methods to analyze the composition of the microbiome continue to evolve rapidly with uneven application and best practices uncertain. Culturing bacteria from stool samples and other patient tissues is valuable but also inconsistent among laboratories, with a subset of bacteria remaining unculturable to this day. An early breakthrough in this field was recognition of the utility of sequencing the 16S rRNA gene to study human microbiota. Next-generation sequencing now allows for increased sequence depth, which, when combined with improved computational analyses, allows for in-depth phylogenetic and taxonomic analyses of the microbiome, even to the species level. Sequence detection of all microbial genomes (metagenomics) or transcripts (metatranscriptomics) within a sample have further advanced microbiome analyses. Together, identification of what microbes are present, what genes they contain, and their transcriptional profiles is now feasible in some instances, allowing characterization of both microbiome membership and function. Nonetheless, in tissues or samples with a low abundance of microbes, both accurate taxonomic and functional analyses are challenging. Recent large-scale studies encompassing multiple tumor types have sought (i) to improve database and computational approaches (particularly to limit sample contaminant reads), thus facilitating microbiome analyses across cancer cohorts; (ii) to develop a deeper understanding of host variables confounding microbiome analyses; and (iii) to struggle with the nuances and definition of comparator “healthy host” microbiomes (refs. 5–9; Table 1). Emerging metabolomics and proteomics methods may provide additional insights into the human microbiome. Whereas each approach provides its own unique set of benefits and limitations, a more complete picture of microbiota complex communities and their impact on tumorigenesis will be realized by use of combined methods and incorporation of host cell responses (e.g., RNA sequencing).

Thus, the microbiome—microbes and their genomes—presents a tantalizing research frontier for understanding cancer pathogenesis and devising previously unimagined approaches to cancer prevention, diagnosis, and therapy. Herein, we provide a perspective on recent investigations (a collection of literature highlights is found in Table 2) of putative microbiome, primarily bacterial, contributions to the pathogenesis of select gastrointestinal (GI) and non-GI cancers.

Colorectal cancer is the second most frequently diagnosed cancer and ranks second in cancer-related deaths worldwide (10). The induction of oncogenes and suppression of tumor-suppressor genes in the large bowel results from a series of mutations and epigenetic changes over time, leading to the onset of tumor formation (11, 12). Inherited genetic predisposition syndromes, such as familial adenomatous polyposis (FAP), Lynch syndrome, and Peutz–Jeghers syndrome, constitute a minority of colorectal cancer cases (11), with the heritability of colorectal cancer estimated to be between 12% and 35% (13). Therefore, sporadic colorectal cancer developing from environmental stimuli constitutes the majority of cases (11). Lifestyle factors including obesity, diabetes, a Western diet, alcohol consumption, and smoking are recognized as colorectal cancer risk factors (11). Each of those colorectal cancer risk factors alters the gut microbiome. These data, combined with the knowledge that individual strains of bacteria harbor the ability to contribute to tumorigenesis, provided strong support for investigations into the role of the gut microbiome in the initiation and progression of colorectal carcinogenesis (11, 13).

Multiple models of bacteria-induced tumorigenesis have been theorized (see Introduction and Figs. 1 and 2), yet a defined sequence of bacteria-driven events in colorectal cancer and the fraction of colorectal cancer most clearly linked to bacterial species remains poorly defined (14). This lack of clarity suggests that multiple microbial and nonmicrobial mechanisms may contribute to colorectal cancer development. Despite this, specific bacterial strains have been linked to human colorectal cancer using a combination of human epidemiologic studies that demonstrate plausible associations with tumorigenesis and animal models that demonstrate potential mechanisms (15, 16).

Bacteroides fragilis (B. fragilis) is a commensal member of the human microbiota composing 0.1% to 0.5% of total gut bacteria (17). The majority of B. fragilis strains remain benign; however, a subset of strains produces the zinc-dependent metalloprotease toxin, B. fragilis toxin (BFT; ref. 17). Toxin-producing strains [enterotoxigenic B. fragilis (ETBF)] induce colonic inflammation, which has been associated with diarrhea, inflammatory bowel disease, and colorectal cancer (17). ETBF stimulates carcinogenesis by activating host colonic epithelial cell (CEC) NFκB and STAT3 pathways while additionally recruiting procarcinogenic myeloid inflammation and inducing mucosal IL17 production (17). BFT binds to CECs through an unknown receptor triggering E-cadherin cleavage, resulting in increased barrier permeability, activation of CEC Wnt signaling, induction of c-Myc expression, and amplified CEC proliferation (17). ETBF modulates host gene expression through BFT, increasing chromatin accessibility. Effects on genes by BFT include upregulation of CEACAM6, which functions as a receptor for adherent-invasive Escherichia coli (E. coli), and downregulation of MUC2, the primary glycoprotein of colonic mucus (18). Combined with its cytoskeletal impact and ability of all B. fragilis to digest mucin, ETBF modifies colonic barrier function and the CEC apical membrane. Additional alterations to the host genome stem from ETBF-induced genomic hypo- and hypermethylation of human colorectal cancer cell line genomes and tumors (18–20). Although some results support an association between methylation changes and, for example, reduced expression of genes with known tumor-suppressive functions, direct, consistent linkage of methylation changes to transcriptional changes remains uncertain and varies based on tumor type (18, 20, 21). Furthermore, BFT has been identified more frequently in patients with colorectal cancer than controls in both the colon mucosa and the stool (15). Together, these data lend support for ETBF to be a potent contributor to colorectal tumorigenesis.

E. coli is another common gut commensal bacterium that is mainly considered benign. Harmful strains produce the genotoxin colibactin through a 50-kb hybrid polyketide–nonribosomal peptide synthase operon (pks⁺ E. coli; ref. 14). Detailed studies have demonstrated that colibactin causes DNA interstrand cross-links, DNA double-strand breaks, chromosomal aberrances, and cell-cycle arrest in human cells in vitro (12). Despite the unstable nature of colibactin, specific DNA adducts are formed, leading to cytotoxicity and mutations (22). The damage to DNA in combination with intestinal inflammation fosters tumorigenesis in mice (12). In clonal organoids injected with pks⁺ E. coli, a mutational signal was identified that was absent in organoids injected with an isogenic pks⁻ E. coli (14). These data provide evidence that colibactin can modify CEC DNA, which is required for colorectal cancer development (14). Despite this mutational signature being present in up to 16% of human cancer genomes and being associated with APC mutation, predominantly in colorectal cancer, whether certain pks⁺ E. coli strains initiate colorectal cancer remains undefined (14). Ultimately, these results suggest that pks⁺ E. coli modifies host genetics to favor the initiation and propagation of tumorigenesis in the colon.

The Gram-negative anaerobic oral commensal Fusobacterium nucleatum (F. nucleatum) has been identified as a potential colorectal cancer biomarker in stool and is predominantly found in the tumor microenvironment (TME; ref. 23). However, as with the previously described tumorigenic bacteria, the causality of F. nucleatum in colorectal cancer is uncertain. Surprisingly, F. nucleatum encodes limited canonical virulence factors and no currently described toxins (23). F. nucleatum includes four major subspecies: F. nucleatum animalis, F. nucleatum nucleatum, F. nucleatum polymorphum, and F. nucleatum vincentii. To date, using repetitive inoculation, only select strains of F. nucleatum have been shown to affect colon carcinogenesis in SPF Apc^Min/+ mice (e.g., EAVG_002; 7_1; ref. 24). No strain has been found to induce colonic tumors in germ-free mice (25). F. nucleatum uses the virulence factor FadA to bind to the extracellular domain of E-cadherin, which induces colon cancer cell proliferation through activation of host CEC Wnt/β-catenin signaling in addition to TLR4-activated signaling to NFκB (23). In support of this mechanism, FadA gene expression in human colorectal cancer tissue is significantly upregulated compared with healthy tissue controls (26). F. nucleatum further induces a proinflammatory tumor-promoting microenvironment by expanding myeloid-derived immune cells. Interestingly, F. nucleatum also inhibits antitumor responses by Fap2 binding of the human TIGIT receptor, subsequently inhibiting the cytotoxic function of natural killer (NK) cells and other tumor-infiltrating lymphocytes (TIL; ref. 23). Importantly, this interaction occurs only in human cells, as F. nucleatum does not bind to mouse TIGIT (27). It remains unclear whether this lack of murine TIGIT binding may partly explain the disconnect between the strong association studies with colorectal cancer in human cross-sectional cohorts versus the relatively weak and variable tumorigenesis seen in mouse models. Finally, F. nucleatum was found to persist with its associated microbiome in distant liver metastases of colorectal cancer tumors (28), suggesting that F. nucleatum and its associated metastatic microbiota affect antitumoral responses and could play a role in metastatic lesions (29). Taken together, these data show that F. nucleatum appears to colonize and expand in the TME to promote colorectal cancer through impacts on the host immune response; well-defined F. nucleatum procarcinogenic virulence factors deserve more study.

Collectively, the list of putative tumorigenic bacteria and their connection to colorectal cancer continues to expand. Like F. nucleatum, Streptococcus gallolyticus (S. gallolyticus) has been shown to modify the TME. S. gallolyticus induces a proinflammatory state marked by high NFκB and IL8 mRNA tissue expression while simultaneously recruiting TILs and myeloid cells that cause an immune-suppressive microenvironment promoting neoplasia (30). Peptostreptococcus stomatis (P. stomatis) and P. anaerobius are other candidate bacteria that have been shown to modulate the TME. P. stomatis contributes to acidity and hypoxia, whereas P. anaerobius leads to reactive oxygen species (ROS) accumulation, promoting bacterial colonization and cellular proliferation, respectively (31, 32).

Recently, a meta-analysis of colorectal cancer 16S rRNA amplicon sequence data revealed that a limited consortium of bacteria—B. fragilis, F. nucleatum, P. stomatis, Parvimonas micra, and Gemella morbillorum—was reliably associated with human colorectal cancer (15). All are members of the oral microbiome and were detected in both the oral cavity and tumors of patients with colorectal cancer, presenting a strong trend between colorectal cancer pathogenesis and the oral microbiome (15, 33, 34). Notably, periodontitis was found to generate oral microbiome–reactive Th17 cells imprinted with gut tropism, which subsequently migrated to the inflamed gut. These translocated (mouth to gut) Th17 cells led to the development of colitis (35). The precise mechanism of translocation of oral bacteria to cancerous lesions, whether by descent through the digestive tract or by hematogenous routes as a result of chewing, dental hygiene, and dental procedures, remains undefined. Furthermore, recent evidence suggests that at least some gut strains of F. nucleatum are identical to those found in the oral cavity in patients, and that hematogenous (tail-vein injection) administration of F. nucleatum led to better gut colonization in mice than oral gavage (36). Whether the route of administration alters tumorigenesis in mouse models—let alone in patients—remains to be understood but is another important area of consideration for future preventive measures. Continued exploration into the gut microbiome of patients with colorectal cancer followed by experimental analyses is imperative to divulge key bacteria and their interactions that function to initiate or promote colorectal cancer.

The collective microbiome community likely demonstrates, at a minimum, an equal and critical contribution versus individual bacteria in influencing colorectal cancer. Immune activation through TLR and NOD-like receptor (NLR) signaling during dysbiosis results in low-level colonic inflammation, a known common factor of all colorectal cancer (2). Furthermore, the diverse metabolites produced by the microbiome in response to host diet directly interface with CECs (Fig. 1). Consumption of a Western diet, rich in red meats and processed foods and low in dietary fiber, is associated with an increased risk of colorectal cancer, potentiated by lower levels of beneficial short-chain fatty acids (SCFA) and increases in potentially deleterious metabolites such as secondary bile acids and hydrogen sulfide. Fermenting bacteria produce SCFAs, such as butyrate, after the consumption of nondigestible carbohydrates (13). Butyrate elicits anti-inflammatory properties by downregulating proinflammatory cytokines, modulating colonic regulatory T (Treg) cells, regulating gene expression through inhibition of histone deacetylases, and inducing apoptosis in colorectal cancer cell lines (2). In contrast to these anticarcinogenic properties, butyrate induced proliferation of harvested MSH2-deficient CECs in vitro (37). These data suggest that butyrate may require tight regulation to benefit the host, potentially depending on host cell genotype. The conversion of primary bile acids to secondary structures is also dependent on the gut microbiome (13). A high abundance of secondary bile acids has the potential to induce oxidative DNA damage and stimulate colonic tumor formation (13), despite these molecules providing numerous benefits. Although the data describing the influence and subtle balance of the metabolites produced by the gut microbiome are far from comprehensive, they lend to the significance of the community composition in colorectal cancer.

Colonic biofilms, highly concentrated communities of bacteria that invade the inner, dense, typically sterile mucus layer of the colon, are assemblies that may provide continual microbial interaction with the host CECs (3). Mucus-invasive biofilms were present on more than 50% of sporadic colorectal cancers in both a U.S. and a Malaysian cohort compared with <15% in healthy colonoscopy controls (15, 38). Interestingly, colonic biofilms on histologically normal colonic tissues displayed increased STAT3 activation and a loss of E-cadherin in CECs (38). Three major microbiologically divergent colonic biofilm types were described in sporadic colorectal cancer patient samples: polymicrobial (Bacteroidetes, Lachnospiraceae), polymicrobial with Fusobacteria blooms, and Proteobacteria-dominant (15). Human colonic biofilms from both patients with colorectal cancer and otherwise healthy controls were found to be directly tumorigenic in murine models (25), whereas biofilm-negative normal mucosal tissues were not tumorigenic. biofilm-positive colonic tissues may enable procarcinogenic bacteria adherent to the colonic epithelium to deliver virulence factors to CECs. Patients with FAP also harbored nearly ubiquitous mucus-invasive biofilms, although they were primarily composed of ETBF and pks⁺ E. coli (39). The compounding effects of a dual-pathogen ecosystem were tested in mouse and in vitro models, in which ETBF promoted mucin degradation that facilitated mucosal colonization of pks⁺ E. coli and led to a coordinated increase in tumorigenesis compared with either bacteria alone (39). In subsequent work, human colonic biofilms were found to be directly tumorigenic in murine models (25). Together, these data suggest that the structure and physical position of the bacterial community in the colon have a critical role in colorectal cancer tumorigenesis.

The recognition of H. pylori as causal to gastric cancer provides some of the most compelling evidence for the microbiome as a cancer promoter and a road map for demonstrating microbial causality in other malignancies (Fig. 3; ref. 40). Gastric cancers of the intestinal type follow a proinflammatory trajectory (termed the Correa cascade), advancing from inflamed mucosa/gastritis to gastric atrophy, intestinal metaplasia, intraepithelial neoplasia, and, finally, gastric cancer. As reviewed by Engstrand and Graham (41), the earliest studies (late 1800s) that examined gastric cancer microflora identified Lactobacilli overgrowth initially presumed to contribute to cancer initiation itself; however, later studies demonstrated that the Lactobacilli and fungal overgrowth was, in fact, likely a bystander effect due to increase in pH (hypo- or achlorhydria), making a more hospitable environment for these organisms to thrive (41). In the 1980s, however, H. pylori emerged as a potential cause of gastric cancer and is now recognized as a type I human carcinogen, with approximately 90% of gastric cancer cases worldwide attributable to H. pylori infection (42) and approximately 10% to Epstein–Barr virus (EBV) infection. Additional microbial players are now being investigated subsequent to the advent of next-generation sequencing (see below).

H. pylori is a common gastric mucosal inhabitant, colonizing approximately 50% of individuals worldwide, and is typically acquired in childhood then persists for life unless treated. H. pylori colonization always induces chronic inflammation, and although most individuals are asymptomatic, this inflammation can cause numerous sequelae, including peptic ulcers (10% of infected individuals), gastric adenocarcinomas (1%–3%), and mucosa-associated lymphoma (<0.1%; ref. 43). In seminal studies in the early 1990s, prospective case–control studies revealed an association between anti–H. pylori antibodies in banked serum samples and increased risk of gastric cancer several years later (44, 45). A recent meta-analysis established that clearance of H. pylori with antibiotics reduced the risk of incident gastric cancer from 3% to 1.6% in healthy, asymptomatic H. pylori–positive individuals over a range of 4 to 22 years of follow-up (46). In a separate study, antibiotic clearance of H. pylori similarly reduced the risk of metachronous gastric cancer nearly in half, from 13.4% to 7.2% in H. pylori–positive patients with prior endoscopic resection of early gastric cancer or high-grade adenoma (47). As gastric tumors emerge, the gastric environment and TME are less hospitable to ongoing H. pylori colonization. Consistent with these observations, a recent study did not observe an enrichment of H. pylori in gastric tumors compared with adjacent normal tissue, providing support for the hit-and-run model of bacteria-initiated tumorigenesis, albeit after likely protracted, chronic colonization as occurs inH. pylori gastritis in humans (8). Thus, absence of a bacterium at the time of tumor diagnosis does not exclude that a bacterium contributed to tumor pathogenesis.

Mechanistically, inflammation is hypothesized to drive H. pylori–associated cancer, as H. pylori does not contain any directly genotoxic virulence factors. Rather, H. pylori induces the recruitment of neutrophils and macrophages, which in turn produce ROS and reactive nitrogen species (RNS). As recently reviewed by Kidane (48), inflammation-mediated ROS/RNS can directly trigger single-strand DNA breaks (SSB) and induce the NFκB proinflammatory pathway that can trigger double-strand DNA breaks (DSB).H. pylori virulence factors (CagA, VacA) may influence the degree of stomach inflammation; thus, strain differences as well as polymorphisms in host inflammatory cascades influence the risk of H. pylori–associated gastric cancer (43). For example, H. pylori peptidoglycans delivered to host cells via the cag pathogenicity island type IV secretion system are recognized by host Nod1, activating NFκB signaling; mice deficient in Nod1 are more susceptible to H. pylori colonization than wild-type mice (43). Coinfection with EBV may also exacerbate H. pylori–associated gastric cancer in approximately 10% of cases (41). Finally, the proinflammatory responses induced by H. pylori, including IL1β and TNFα, inhibit stomach parietal cell acid secretion; the resulting hypochlorhydria may promote secondary bacterial overgrowth that may modify chronic gastric inflammation, consistent with the earliest associations between benign Lactobacilli overgrowth and gastric cancer (43).

Although H. pylori is the strongest risk factor for gastric cancer, recent next-generation sequencing studies identified oral microbes [similar to those seen in colorectal cancer and head and neck squamous cell carcinomas (HNSCC)] with gastric cancer (49). Support for a role of oral microbes in gastric cancer also comes from a meta-analysis demonstrating a potential link between tooth loss (a marker for periodontal disease) and risk of gastric cancer (50), although this linkage remains controversial and needs validation. Other studies highlight additional, but inconsistent, microbes putatively contributory to gastric cancer; thus, a meta-analysis of the taxa and functional predictions from published 16S rRNA sequencing data may be of value to assess associations with gastric cancer.

In summary, gastric cancer and H. pylori currently represent the strongest link between a single bacterium and cancer causality (Fig. 3; ref. 40), with prospective data demonstrating that H. pylori precedes tumorigenesis and clearance of the organism with antibiotics reduces rates of gastric cancer. Although other microbes may influence this tumorigenesis process, the dozens of other microbes associated with gastric cancer from 16S rRNA gene-sequencing studies (including the enrichment of oral microbes) are inconsistent and lack clear mechanisms and prospective studies to demonstrate causality. Indeed, the lessons from the early observations of enrichment of Lactobacillus in gastric cancer represent an important cautionary tale regarding tumorigenesis associations versus causality.

Pancreatic ductal adenocarcinoma (PDAC) is lethal, with few patients surviving 5 years after diagnosis, in part because diagnosis most often occurs at an advanced stage of disease. Thus, novel approaches to early diagnosis and treatment are being sought, including exploration of gut and tumor microbiota. The normal human pancreas tissue may harbor microbiota, both bacterial and possibly fungal, as well as produce antimicrobial peptides (10% of the proteins in exocrine pancreatic fluid), suggesting that the pancreas modulates both its intrinsic microbiota and that in the duodenal lumen and gut (51). The mouth, duodenum, and gut microbiota all likely seed the pancreatic microbiota. Exactly how the human pancreatic microbiota modulate pancreatic function, immunologic homeostasis, and susceptibility to pancreatic disease remains unknown.

As with colorectal cancer, oral microbiota have been associated with PDAC in 16S rRNA microarray and sequencing studies, as well as by detection of plasma antibodies to oral bacteria (51–53). These results are consistent with the epidemiologic association of PDAC with periodontitis. Furthermore, in intraductal papillary mucinous neoplasms (IPMN), increased intracystic bacterial copy numbers enriched in oral bacteria taxa and inflammatory signals (IL1β) associate with IPMNs with high-grade dysplasia or cancer compared with non-IPMN pancreatic cystic neoplasms, suggesting that oral bacteria, likely through chronic inflammation, contribute to PDAC pathogenesis. Putative contributing oral microbiota members include bacteria with positive (e.g., Porphyromonas gingivalis), negative (e.g., Neisseria elongata, Streptococcus mitis), and variable (e.g., F. nucleatum) associations with PDAC, as well as associations with more complex consortia of oral bacterial species (51). Studies of fecal microbiota in patients with PDAC display lower α-diversity (within-sample diversity) and phyla shifts versus healthy individuals, a result consistent with a wide range of microbiota comparisons in disease versus health (51, 54). Nonetheless, interindividual variability of the fecal microbiome among patients with PDAC, as in healthy individuals, is high, and preliminary observations found little concordance between PDAC-associated microbial signals and the microbiota in precancerous lesions, suggesting limited utility for use in early detection of pancreatic neoplasia (54).

Informative translational studies of the microbiota in the PDAC tumor bed combined with preclinical studies in mouse models have emerged. Overall, in short-term survivors of PDAC (the vast majority of patients), the microbiome assessment of tumor resection samples supports dominance of the phyla Proteobacteria, Firmicutes, and Bacteroidetes. Among the Proteobacteria, Enterobacteriaceae (family), Pseudomonas (genus), and Elizabethkingia (genus) appear notable and, in part, likely represent translocation from the gut into the pancreas where Proteobacteria may flourish (55–57). In one murine study, a protumorigenic role for the species Bifidobacterium pseudolongum (B. pseudolongum; phylum Actinobacteria) is highlighted (56), contrasting with work in nonpancreas tumors, suggesting that B. pseudolongum or B. longum promote antitumorigenic mechanisms (58, 59). Using preclinical mouse models, both germ-free mice and antibiotic-treated mice display limited PDAC growth, strongly supporting that the intratumoral or fecal microbiota can promote PDAC progression. Mechanistically, murine models support that the intratumoral microbiota promote PDAC progression through innate and adaptive immunosuppression mechanisms, whereas microbial ablation with antibiotics fosters T-cell proliferation and immune activation, including tumor responsiveness to checkpoint inhibitor therapy (56). The intratumoral bacteria (likely Proteobacteria) may foster PDAC therapeutic resistance by metabolizing gemcitabine, a common PDAC chemotherapeutic, to an inactive form (55). Consistent with these observations, retrospective clinical data suggest that coincident antibiotic therapy may improve outcomes in patients with PDAC treated with gemcitabine (55–57, 60–62). In marked contrast, in a study focused on examining the infrequent long-term PDAC survivors (LTS; survival >5 years), Riquelme and colleagues (57) presented compelling data, using multiple approaches, that distinct intratumoral microbiome features increased α-diversity, and that a consortium of the genera Saccharopolyspora (phylum Actinobacteria), Pseudoxanthomona (phylum Proteobacteria), Streptomyces (phylum Actinobacteria), and species Bacillus clausii (phylum Firmicutes) are highly predictive of LTS and, again, likely act through modulation of the tumor microenvironment. Further, PDAC LTS has been linked to identical circulating and intratumoral T-cell clones reactive to both high-quality tumor neoantigens and infectious disease–derived sequences consistent with neoantigen molecular mimicry, suggesting the hypothesis that an individual's exposure to particular microbes over time may serve, in part, to prime an effective neoantigen-specific immune response as PDAC outgrowth emerges (63).

Collectively, the assessment of the PDAC microbiome at the time of PDAC therapy initiation may offer insight into prognosis and help design experimental studies to improve patient survival through modulation of the pancreatic and/or gut microbiome. Longitudinal and family studies of microbiota characteristics and onset of PDAC might further enhance the understanding of microbiota contributions to PDAC derived, to date, only from cross-sectional studies.

The lung, with the largest mucosal surface area in the body, displays a complex microbiota molded by both intrinsic (e.g., upper vs. lower lobe) and extrinsic (e.g., airborne microbes, smoking, oral microbiome) environmental factors. Lung cancer is a highly heterogeneous tumor, and largely cross-sectional studies provide evidence of lung cancer–associated microbiome dysbiosis but with highly variable results (64, 65). One study using both primary lung tissue samples and a validation cohort from The Cancer Genome Atlas (TCGA) suggested that Proteobacteria overall increase in the lung cancer microbiome, whereas increased Acidovorax (phylum Proteobacteria) abundance was specifically found in squamous cell carcinoma with TP53 mutations in smokers, suggesting microbiome–gene and microbiome–exposure interactions (66). Two recent studies provide pivotal data on the local lung microbial and immune mechanisms contributing to lung cancer, whereas data supporting gut–lung axis mechanisms are lacking. Jin and colleagues (67) used the KP murine model of lung adenocarcinoma (LUAD), driven by an activating point mutation of Kras and loss of Trp53, to clearly demonstrate that an increased lung bacterial load, and even a limited bacterial consortium or bacterial molecules, accelerate LUAD by activating a myeloid cell/lung–resident γδ-T-cell amplification loop that drives procarcinogenic inflammation and tumor cell proliferation through IL17 and polymorphonuclear cell–mediated mechanisms. In a prospective study, Tsay and colleagues (68) tackled the human lung microbiota of stages I–IIIA versus IIIB–IV lung cancer, providing evidence that, independent of disease stage, a lower airway microbiota enriched for oral commensals (e.g., Streptococcus, Prevotella, Veillonella, termed SPT, supraglottic predominant taxa, pneumotype) associated with a worse prognosis as well as upregulation of inflammatory cancer–related pathways (e.g., ERK/MAPK, PI3K/AKT). As proof of concept, using the KP mouse model, addition of Veillonella parvula alone to the lung microbiota accelerated LUAD progression. It is hoped that these results will provide translational approaches to improve the lethality of lung cancer.

In 2014, demonstration of a breast microbiome, potentially sourced from the skin, mouth, or gut, emerged. Highly variable studies suggest that, whereas patients with breast cancer display dysbiosis at the genus and species level, the phyla Proteobacteria, Firmicutes, and Bacteroidetes are prominent in both healthy breast and breast cancer tissues; however, Lactobacilli (phylum Firmicutes) abundance may be lower in breast cancer tissues. To date, limited and inconsistent taxa differences appear to distinguish tumor and adjacent normal tissue microbiota and benign versus malignant tumors, although breast cancer subtypes may possess distinct microbial signatures (69). Bacterial regulation of estrogen bioavailability or induction of DNA damage in the breast is proposed to mediate carcinogenesis. In contrast, increased F. nucleatum genomic DNA in breast cancer tissues has been associated with breast tissue Gal-GalNAc levels that increase with breast cancer progression; results are consistent with the known binding of Fap2, a F. nucleatum adherence protein, to Gal-GalNAc. In a murine model, Fap2-sufficient, but not FAP2-deficient, F. nucleatum colonized orthotopic breast cancer and suppressed tumor-infiltrating T cells to promote tumor growth and metastasis (70). Whether the gut microbiome affects breast cancer biology is unknown. However, Parida and colleagues (71) bioinformatically identified increased B. fragilis, a common colon anaerobe, in both benign and malignant breast tumor tissue. Using murine models, both breast intraductal and colon colonization with toxin-producing (ETBF; see colorectal cancer section), but not non–toxin-producing, B. fragilis promoted breast tumor growth and metastatic progression involving β-catenin and Notch1 pathways. Notably, breast tissue cells exposed to the B. fragilis toxin retained procarcinogenic memory, potentially increasing disease risk. The breast or gut microbiome in patients with breast cancer may help stratify breast cancer risk and response to therapy.

The majority of head and neck cancers arise from mucosal membrane squamous epithelial cells (HNSCC). HNSCC encompasses a broad number of malignancies, including cancers of the oral cavity, pharynx (including nasopharyngeal, oropharyngeal, and hypopharyngeal), larynx, paranasal sinuses and nasal cavity, and salivary glands. Multiple general and tumor type–specific risk factors for HNSCC exist, including human papillomavirus (HPV) type 16 (primarily in oropharyngeal cancers), EBV (nasopharyngeal and salivary gland cancers), preserved or salted foods (nasopharyngeal cancers), radiation (salivary gland cancers), poor oral hygiene (i.e., periodontitis in oral cavity cancers) along with tobacco and alcohol products (linked to >75% of all HNSCC cases), occupational exposures (e.g., asbestos, pesticides, and industrial solvents in multiple HNSCC types), and host genetics (https://www.cancer.gov). Bacteria may play a direct (e.g., periodontal disease) or indirect (e.g., alcohol metabolism) role in the tumorigenesis pathways of many of these risk factors. Approximately 15% of HNSCCs lack known risk factors, spurring increased interest in the role for novel bacteria or groups of bacteria in the etiology of HNSCC.

The oral microbiome is a diverse community of >750 species (Human Oral Microbiome Database, homd.org), many of them anaerobes, that form complex multispecies biofilms on both tooth surfaces and oral cavity mucosal membranes. Within a healthy individual, the oral microbiome is largely stable. Left unchecked through poor dental hygiene, however, periodontal biofilms can trigger disease, resulting in tooth decay, tooth loss, and potentially oral cavity cancers. Distinct oral microbial niches (>25) and sampling tools exist, from buccal swabs to oral rinses to tumor biopsies, which markedly complicates microbiome-based analyses.

Most studies analyzing the microbiome and HNSCC have focused on oral cavity squamous cell carcinomas (OSCC). Using both sequencing- and culture-based methods, OSCC biopsies appear to harbor dozens of both intratumoral and surface biofilm–associated bacterial species not found in healthy oral biopsy samples from the same patients (see review by Minarovits; ref. 72) and include several species also found at higher abundances in colorectal cancer tissues compared with normal colonic tissues: F. nucleatum, Fusobacterium periodonticium, G. morbillorum, and P. stomatis (72). Bacterial biofilms found on OSCC also harbored a higher overall abundance of total anaerobic and aerobic bacteria by colony-forming units (CFU), which parallels data from the colon, where colorectal cancer–associated mucus-invasive biofilms are thicker and harbor more bacteria than paired normal tissues (38). However, inconsistencies between studies exist, and each studied cohort was relatively small. In a larger study of oral rinse samples from 197 patients with OSCC compared with 52 healthy individuals, the Fusobacteria phylum again strongly associated with OSCC, increasing from a relative abundance (RA) of 2.98% in healthy controls to 7.92% RA in stage 4 OSCC alongside decreases in Bacteroidetes and Actinobacteria phyla. At the species level, this trend was largely driven by F. periodonticium (∼1% RA in healthy tissues to 2% RA, stage 1; 3% RA, stage 2 and 3; and 5% RA, stage 4 OSCC; ref. 73). Four other bacteria also increased alongside tumor stage in the oral rinse samples: P. micra (also increased in colorectal cancer–associated tissues), Streptococcus constellatus, Haemophilus influenzae, and Filifactor alocis (73). A meta-analysis of Fusobacterium in swab or tissue samples from 13 HNSCC studies found the Fusobacterium genus to be consistently enriched in tumor sites compared with nontumor controls, with F. nucleatum the most abundant Fusobacterium species detected, followed by F. naviforme, F. periodonticum, and others (74).

Only a single large prospective study exists to date for HNSCC, which did not validate the above organisms, although all HNSCCs were included (not solely OSCC; ref. 75). In the nested case–control study of >100,000 patients within the American Cancer Society Prevention Study II Nutrition Cohort (CPS-II) and the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO), 129 new cases of HNSCC were identified over an average of 3.9 years of follow-up. Prediagnostic, baseline oral mouthwash samples were examined by 454 pyrosequencing of the 16S rRNA V3–V4 region. HNSCC cases were associated with more tobacco usage, alcohol consumption, and HPV-16 carriage compared with controls, consistent with prior data. The oral microbiome β-diversity was not different between the total HNSCC cases or specific cancer types compared with controls (75). Surprisingly, the strongest trends observed were for protective bacteria against HNSCC: the genera Corynebacterium (phylum Actinobacteria) and Kingella (phylum Proteobacteria) were independently associated with reduced HNSCC risk. The protective effects of Corynebacterium and Kingella remained even after excluding current smokers or HPV-16–positive patients. Results remained similar by age, sex, and alcohol usage. Divided by cancer subtype, the genera Corynebacterium, Kingella, Neisseria, Abiotrophia, Capnocytophaga, and species Kingella dentificans and Streptococcus sanguinins were associated with reduced risk of larynx cancer; Actinomyces oris and Veillonella denticariosi with reduced risk of pharynx cancer; and P. micra and Neisseria siccas with reduced risk of oral cavity cancer. The only bacteria positively associated with HNSCC were phyla Actinobacteria (HNSCCs overall) and Actinomyces (oral cavity cancer). The microbial associations in the prospective study were strongest in patients with a history of tobacco usage and those who developed larynx cancer. These associations are biologically plausible because larynx cancer is the HNSCC most strongly associated with tobacco usage, and Corynebacterium and Kingella may neutralize several toxicants found in cigarette smoke. Thus, among tobacco users, having Corynebacterium or Kingella spp. may be particularly protective against larynx cancer. However, the average time from sample collection to HNSCC detection in this cohort (3.9 years) may be insufficient to truly demonstrate potential microbial modulators of cancer. Replication studies are needed.

Despite the absence of a clear “smoking gun” microbial trigger for HNSCC, at a mechanistic level, oral bacteria may facilitate tumorigenesis via both specific metabolic activities and broad inflammatory properties. For example, oral bacteria facilitate the metabolism of alcohol into its carcinogenic by-product, acetaldehyde, at levels that can induce DNA damage and enhance epithelial cell proliferation (76). The genus Neisseria is particularly adept at alcohol dehydrogenase activity, whereas Lactobacillus metabolizes acetaldehyde to relatively nontoxic forms. Interestingly, alcohol consumption was associated with increased Neisseria RA and decreased Lactobacillus RA in oral wash samples from a large >1,000-person cohort (76). However, Neisseria have not been elevated in OSCC patient samples, either in comparison to paired normal control biopsies from the same patient or in oral wash samples from cancer-free individuals (72). This disconnect between the alcohol-associated microbiome and the OSCC-associated microbiome suggests that early microbial drivers or potential contributors to OSCC (i.e., mediators of alcohol-induced damage) may be lost by the time the tumor has overtly formed (hit-and-run model).

Alternatively, oral bacteria may promote tumorigenesis via inflammation induction. Using the functional prediction tool PICRUSt, Al-Hebshi and colleagues (77) proposed that OSCC tissues harbor an “inflammatory bacteriome” characterized by enrichment of genes for bacterial mobility, flagellar assembly, bacterial chemotaxis, and lipopolysaccharide synthesis, whereas control samples were enriched for more homeostatic genes such as amino acid biosynthesis, DNA repair, purine metabolism, ribosome biogenesis, and glycolysis/gluconeogenesis. Similarly, the link between periodontal disease and HNSCC revolves around proinflammatory mechanisms. Although links between periodontal disease and OSCC are inconsistent, in studies using clinical measurements of periodontal disease, a 4–10-fold higher risk of HNSCC was associated with severe periodontitis (78). Periodontal disease has also variably been linked to total cancer risk, as well as specifically lung cancer, PDAC, and colorectal cancer (78). Mechanistically, the link between periodontal disease and cancer has primarily focused on the proinflammatory bacteria F. nucleatum and P. gingivalis. P. gingivalis is a keystone periodontal disease pathogen that may inhibit apoptosis and enhance proliferation pathways via increased JAK/STAT signaling, suppression of Bcl2, alteration of p53 pathway cyclins, or alteration of dendritic cell–specific intercellular adhesion molecule-3–grabbing nonintegrin (DC-SIGN; ref. 78).

In summary, HNSCCs are associated with numerous oral microbiome changes that vary from precancerous to cancer stages. Risk factors (e.g., alcohol consumption and periodontal disease) induce early microbial changes, although data are lacking on whether these bacteria, a mixture of oral microbes led by the Fusobacterium genus, directly impact human cancer initiation. However, these associations may be merely bystander associations. The only prospective cohort study yielded no species identified as HNSCC risk factors; rather, Corynebacterium and Kingella appeared protective. Prospective cohorts in other study populations would be invaluable. A meta-analysis of all 16S rRNA or metagenomic studies in HNSCC examining all potential bacteria (not solely Fusobacterium) would be beneficial.

Cancers of the genitourinary (GU) tract (i.e., adrenal, bladder, kidney, penile, prostate, and testicular cancer) have limited, but growing, data suggesting a role of the microbiome in disease etiology. Urine, previously assumed to be a sterile fluid, is now known to contain a limited but variable microbiome that may affect GU cancers. The urinary microbiome is proposed as a source of proinflammatory bacteria that reflux to, for example, the prostate or kidney. The urinary microbiome varies by sex, potentially contributing to the higher rates of several GU cancers in men (bladder, renal, and prostate) compared with women; men harbor more Actinobacteria including Corynebacterium and Propionibacterium in healthy urine samples, whereas women tend to harbor more Lactobacillales (79).

A microbial role in bladder cancer has been evident for decades, as infection with the parasitic flatworm Schistosoma haematobium is associated with very high rates of bladder cancer in endemic areas, including the Middle East and Africa, particularly before effective treatments for schistosomiasis were developed (80). Inflammation triggered by the embedding of parasitic eggs into the bladder wall may be the primary disease mechanism. The resident bacterial microbiome may still contribute, however, as even in Schistosoma-positive cases, multiple genera including Fusobacterium, Bacteroides, Veillonella, Aerococcus, and Facklamia were enriched in patients with bladder cancer versus those with only Schistosoma infection or no infection (81). In addition, in the U.S. population where Schistosoma is not endemic, a history of three or more urinary tract infections (UTI) is an established risk factor for bladder cancer (82). The vast majority (>70%) of UTIs are E. coli, with preliminary data suggesting that approximately 20% of patients with UTI harbor DNA-damaging pks+ E. coli (see colorectal cancer section). Although diverse genera are reported as enriched in bladder cancers, taxa overlap between studies is limited, suggesting that further study is needed (reviewed in ref. 80).

Studies in men with prostate cancer similarly report differentially abundant microbes encompassing broad genera, with little overlap between studies, with the exception of Bacteroides and Streptococcus (reviewed in Table 3 in Nicolaro and colleagues; ref. 80). However, those studies were performed on diverse samples (urine, rectal swabs, or stool samples). In one of the largest studies to date, urine obtained from men with and without prostate cancer did not reveal broad clustering of cancer versus noncancer patients, although six largely uropathogenic bacteria (capable of inducing inflammation) were enriched in a subset of prostate cancer cases (Streptococcus anginosus, Anaerococcus lactolyticus, Anaerococcus obesiensis, Actinobaculum schaalii, Varibaculum cambriense, and Propionimicrobium lymphophilum; ref. 83). In mouse models, uropathogens (e.g., E. coli) lead to long-lasting inflammation even once the pathogen subsides to lower levels, suggesting a plausible mechanism for infection-associated cancers (79). Epithelium damage in bacteria-induced prostatitis may lead to impaired antimicrobial defenses, potentiating a feed-forward cycle of recurrent bacterial infections and epithelial damage resulting in chronic inflammation (84).

The microbiome of kidney (renal cell) cancer (RCC) is the least well studied of the GU cancers. However, a history of UTIs of the bladder or kidney in a U.S. population was associated with increased risk for RCC, particularly in men who smoked. Complex interactions between bacteria and other epidemiologic risk factors in RCC may exist (85).

The growing associations between the microbiome and various cancers offer an opportunity to develop screening modalities that may target cancer prevention and treatment. Current cancer diagnosis usually requires invasive techniques (e.g., colonoscopy for colorectal cancer, biopsies of potential tumorigenic tissue; ref. 11). Other less invasive tests, including blood tests, urine tests, imaging, fecal immunochemical tests, and multitarget stool DNA tests, provide alternative but less accurate diagnosis (4, 11). Early cancer detection is key to patient outcomes. Exploiting defined microbial signatures specific to individual cancer types may enhance accurate early and less invasive methods of diagnosis. Currently, no microbial screening tools exist outside of H. pylori for gastric cancer, but ever-increasing investigations provide promise for future development.

The concept of targeting microbes in cancer originates from the removal of H. pylori as a treatment for gastric cancer (Fig. 3; ref. 13). Therefore, targeting the microbiome for other cancers may influence therapeutic outcomes as well as provide an unparalleled opportunity to develop microbial-targeted treatments. Microbiota manipulation is an intense research area (86). Dietary intervention, prebiotics, synbiotics, and probiotics that enrich or provide beneficial bacteria in the gut are currently under investigation to prevent or improve therapy (2). Naturally occurring or engineered probiotic bacteria may outcompete detrimental species through colonization displacement and niche exclusion or by producing therapeutic molecules in situ (86). The complete replacement or restoration of a patient's microbiome through fecal microbiota transfer (FMT) is another area of focused research. Nonetheless, human FMT has resulted in patient infections with enteric and multiple drug-resistant bacteria, leading to patient death from lack of quality control or use of ill-defined microbial products (87). Removal of procarcinogenic members of the microbiota through narrow-spectrum antibiotics, monoclonal antibody therapy, or species-specific bacteriophages provides putative direct therapeutic approaches to microbial modulation (13). To date, antibiotics have been primarily used as a tool to discern the contribution of the microbiota to tumorigenesis in various murine disease models (88). In human studies, only associations, not causal links, between antibiotic exposure and onset of cancers have been reported. Further, the potential impact of antibiotic exposure on cancer pathogenesis likely varies by route of administration, the timing of exposure, and antibiotic class, with even cancer-specific impacts (89). Vaccination against the microbial virulence factors involved in tumorigenesis, such as toxins or adhesion factors, or the specific bacteria themselves has the potential to elicit an immunomodulatory benefit. Finally, designing therapeutic molecules to scavenge or negate tumorigenic molecules produced by the microbiome (e.g., preventing DNA modifications) may emerge. Despite numerous theoretical approaches for microbiota therapy, the application of these ideas remains in its infancy. However, clinical trials are ongoing for numerous types of cancers and diseases, suggesting that live microbial therapeutics might become available as a treatment option (86).

Survivability of every type of cancer described in this review can be attributed to the magnitude of disease progression when first detected. Although many factors propel cancer progression, the state of the microbiome is now entertained as a harbinger, cause, or promoter of disease. Microbial signals in cancer are likely highly confounded by the heterogeneity of cancer, combined with observations that single bacterial species–oncogene interactions modify cancer biology (20, 66), making well-defined cancer microbiome “signatures” difficult to classify or detect. Herein, we focused on how specific bacterial microbiome members and the microbiome community structure affect cancer progression. We delineated mechanisms, when described, by which the production of harmful microbial molecules, microbe-driven host immune responses, and microbe-triggered changes in host cell function or genome may contribute to cancer biology. Beyond viruses, H. pylori, and schistosomiasis, microbiota causality in cancer is not yet established, in part due to the absence of longitudinal microbiome studies antecedent to human cancer. Further investigations to understand microbiome transitions in cancer emergence and oncogenic mechanisms may direct development of targeted screening modalities and microbial-based treatments with the vital objective of enhancing patient care and outcomes. The ever-increasing importance of the microbiome in a multitude of human cancers heralds boundless research opportunities to inform and direct new standards of cancer clinical practice.

J.L. Drewes reports grants from NIH/NCI during the conduct of this work. C.L. Sears reports grants from Janssen, Cancer Research UK Grand Challenge, Bloomberg Philanthropies, and NCI during the conduct of this work; and grants from Bristol Myers Squibb outside the submitted work; in addition, C.L. Sears has a patent for US 10,203,329 issued and a patent for US 16/237,112 pending. No disclosures were reported by the other author.

The authors regret the inability to cite additional authors and their primary research within the constraints of this review. The authors thank the many members of the Sears laboratory for the inspiration they provided over time in considering the role of the microbiome in cancer. Figures 1, 2, and 3 are unpublished original works created with a licensed copy of BioRender.com. This work was supported by Cancer Research UK CRUK (#C10674/A27140, to C.L. Sears), Bloomberg Philanthropies (to C.L. Sears), NCI R01CA196845 (to C.L. Sears), NCI R00CA230192 (to J.L. Drewes), and the Johns Hopkins School of Medicine.