Introduction

Colorectal cancer (CRC) is a leading cause of cancer morbidity and mortality, with 1.8 million new cases and >800,000 deaths annually1. Microbial dysbiosis is increasingly recognized as a factor in colon tumor initiation and progression. This dysbiosis can manifest as enrichment or depletion of specific bacterial taxa or in alterations of bacterial communities, such as the development of mucus-invasive biofilms2. Fusobacterium nucleatum, a Gram-negative anaerobe and common member of the human oral microbiota, is known to be associated with CRC3,4,5,6. Clinical studies from North American, European, and Asian cohorts have demonstrated enrichment of F. nucleatum in a subset of CRC compared to healthy colon tissues3,4,5,6. Data from animal models and translational studies have suggested potential roles for this organism in colon tumor development, progression, metastasis and/or treatment response7. In the mouth, F. nucleatum plays an important role as a bridging species linking early and late colonizers of polymicrobial biofilms8. Biofilms have also been recognized as common features of colon tumors9,10, where it is posited that F. nucleatum and other organisms invade into the mucosa to contact host cells and promote inflammation and tumorigenesis. F. nucleatum is a heterogenous species consisting of four subspecies anticipated to become distinct species: animalis, nucleatum, polymorphum, and vincentii11. Recently, subspecies animalis has been further classified into two clades, with identification of clade Fna C2 as prominently featured in the colon tumor microenvironment12. In this study, we sought to characterize the tumor microbiome of a large cohort of individuals with CRC through a combination of sequencing and spatial analysis, and to identify features associated with enrichment of F. nucleatum and its subspecies and clades.

Results

Colon tumors frequently have polymicrobial biofilms containing Fusobacteria

The study population consisted of 116 individuals diagnosed with colorectal cancer in Malaysia, from whom paired tumor and distant normal colon tissues were collected at the time of CRC tumor resection surgery (Fig. 1a). Four of these individuals underwent surgery for resection of masses later identified as adenomas (polyps). This 116 person cohort is inclusive of two previously published smaller cohorts (MAL1 and MAL2)10, to which we added an additional 72 individuals with CRC, yielding an assessment of 143 additional tumor and/or paired normal samples (MAL3). Demographic and clinical data collected included age, sex, ethnicity, tumor location, and tumor stage (Fig. 1b). The proximal colon through the hepatic flexure was defined as right colon, and distal to the hepatic flexure as left colon, as previously described10. To assess for the presence of biofilms, we performed fluorescence in situ hybridization (FISH) on methacarn-fixed tissues of colon tumor and paired normal samples (Fig. 2a). We found that biofilms were prevalent on colon tumors, visualized in 68.8% (77/111) (Fig. 2b). Biofilms were more frequently observed on right-sided tumors (33/36; 91.7%) compared to left (44/75; 58.7%; p = 0.0003) (Fig. 2c). Of 70 biofilm-positive tumors available for multi-probe FISH, we identified broad patterns of microbial biofilm community membership, consisting of either polymicrobial biofilms (94.3%; 66/70) with (Fig. 2d) or without (Fig. 2e) Fusobacterium (genus), or, rarely, biofilms predominated by proteobacteria (5.7%; 4/70) (Fig. 2f). Among the 66 polymicrobial biofilms, we visualized Fusobacterium spp. in 68.2% (45/66) of tumors, ranging from sparse populations of individual rods to dense blooms where Fusobacterium was the predominant organism. Although biofilms were not associated with tumor stage (Fig. 2g), the presence of Fusobacterium in tumor biofilms was associated with later tumor stage compared to polymicrobial biofilms where Fusobacterium was absent (p = 0.0212) (Fig. 2h). There was high concordance in biofilm status between tumors and paired normal tissue; in individuals who had biofilms observed on their tumors, 79.0% (60/76) also had biofilms on distant normal tissue (Fig. S1a). In those with no biofilms visualized on their tumors (n = 34), there were no biofilms observed in paired normal tissue. Similar to tumor biofilms, biofilm positive normal tissues screened by multiprobe FISH were predominantly polymicrobial (89.3%; 25/28) (Fig. S1b). In contrast to tumors, polymicrobial biofilms on normal tissue typically had sparse Fusobacterium, with no consistent observations of dense blooms or epithelial invasion of Fusobacterium in normal tissue (Fig. S1c, d).

Fig. 1: The CRC cohort.
figure 1

A Flow diagram of study participants and downstream analysis of samples. B Demographic characteristics of the cohort (n = 116), including age, sex, ethnicity, stage, and tumor location. EO-CRC early-onset colorectal cancer, defined as age under 50. LO-CRC late-onset colorectal cancer, defined as age 50 or above.

Full size image
Fig. 2: Biofilm analysis of the CRC cohort.
figure 2

A The location and biofilm status of CRC tumors overlaid on a diagram of a colon. One tumor marked as unknown had no fixed tissue suitable for screening. B Biofilm status of CRC tumors classified as either positive or negative by FISH. C Percent of left and right-sided tumors positive for biofilms. D–F Representative 40X images of multi-probe FISH of each biofilm subtype stained for DAPI (blue), Bacteroidetes (green), Fusobacterium (yellow), Lachnospiraceae (red), and Proteobacteria (pink). G Percent of tumors positive for biofilms stratified by tumor location and stage. C, G analyzed by Fisher’s exact test, with p < 0.05 considered significant. H Stacked bar graph depicting percent of tumor biofilms of each subtype screened by multi-probe FISH (n = 70) and stratified by tumor stage. Analyzed by nonparametric Kruskal-Wallis one-way ANOVA, with p < 0.05 considered significant.

Full size image

Fusobacteria are enriched in the colon tumor microbiome

We performed 16S rRNA amplicon sequencing on tumor biopsies and paired normal tissue to identify bacterial taxa enriched or depleted in the tumor microenvironment. Two subsets of this cohort were previously analyzed by 16S rRNA amplicon sequencing and published10: MAL1 (consisting of 22 tumors and 1 adenoma with 20 paired normal samples) and MAL2 (consisting of 24 tumors with 22 paired normal samples, 3 of which were replicate pairs from MAL1). We performed 16S rRNA amplicon sequencing on an additional 79 tumors, 3 adenomas, and 71 paired normal samples; this subset, termed MAL3, included 10 replicate tumors previously sequenced in the MAL1 and/or 2 cohorts. Although all three cohorts were sequenced independently, each was sequenced using the V3–V4 hypervariable region as the amplification target, and data were analyzed using the same computational pipeline to assign species-level taxonomic identification (Resphera Insight, Baltimore, MD). Analysis of these ten replicates demonstrated no significant variation by sequencing cohort (Fig. S1a), as measured by Bray–Curtis dissimilarity, with significance estimated by Permutational multivariate analysis of variance (PERMANOVA). Linear discriminant Effect Size (LEfSe) analysis identified few taxa that were significantly enriched in one cohort relative to replicates in other cohorts (Fig. S1b), and notably did not include Fusobacterium or other taxa specifically investigated herein. Each set of replicate samples had consistent taxonomic assignments across independent sequencing runs for replicate samples at the phylum through species levels (Table S1a–g; Figs. S2–4). This included F. nucleatum and other Fusobacterium species, which are the particular focus of this study (Fig. S5). Given the consistent findings of concordant taxonomic assignments between replicates by multiple analytical approaches, we pooled MAL1, MAL2, and MAL3 data for all subsequent analyses (see Table S2 for all taxonomic assignments). Altogether, the combined MAL cohort presented herein consists of 108 unique tumor and normal tissue pairs, 4 adenoma and normal tissue pairs, and an additional 4 tumors for which paired normal sequencing data is unavailable.

Consistent with prior studies13,14, tumors had lower alpha diversity than normal tissue as measured by the number of observed species (p = 0.0042). Biofilm-negative tumors had lower alpha diversity than biofilm-negative normal tissue, as measured by number of observed species (p = 0.019) or Chao Index (p = 0.0346) (Fig. S6a). Although biofilm-positive tumors were more diverse than biofilm-negative tumors by Chao Index (p = 0.0096), there was no significant difference in alpha diversity between biofilm-positive tumor and normal samples, by Chao Index, Shannon Index, Simpson Index, or number of observed species. Neither sample type (tumor vs normal) nor biofilm status (positive vs negative) was associated with differences in beta diversity, as measured by Bray–Curtis dissimilarity, with significance estimated by PERMANOVA (Fig. S6b).

F. nucleatum was present in 86.6% of CRC tumors in this cohort (97/112), with relative abundance ranging from 0.01% to 40% of total microbial reads (Table S2). Of 112 tumors in the study, 108 had paired normal tissue available for 16S rRNA sequencing. Analysis of the 108 tumor/normal pairs revealed significant enrichment of F. nucleatum in the tumor microbiome compared to paired normal samples (p < 0.0001) (Fig. 3). Resphera Insight attempts to achieve species-level resolution; however, when a confident single species assignment is not feasible, the method minimizes false positives by providing ambiguous assignments that may include multiple closely related species. This significant enrichment in the tumor microenvironment was observed whether F. nucleatum abundance was calculated using confident (Fig. 3) or confident + ambiguous (Fig. S7a) assignments. Other Fusobacterium species enriched in tumors included F. peridonticum (p = 0.0005), F. simiae (p = 0.0293), and F. varium (p = 0.0094) (Fig. 3). The latter has previously been found to be enriched only in Southern Chinese populations15. In our cohort, F. varium was enriched in tumors vs paired normal tissue only for individuals of Malay descent (Fig. S7b). Also of note, Bacteroides fragilis was significantly enriched in the tumor microbiome relative to paired normal tissue (p = 0.0002), whereas other taxa previously linked to the CRC microbiome were not enriched in our cohort, including Akkermansia muciniphila, Enterococcus faecalis, Escherichia coli, and Streptococcus gallolyticus (Fig. S8). We also observed relative depletion of Alistipes onderdonkii (p = 0.0261), Bifidobacterium longum (p = 0.0029), Parabacteroides distasonis (p = 0.0149), and Ruminococcus bromii (p < 0.0001), with a trend toward depletion of Alistipes senegalensis (p = 0.0620) in tumor samples.

Fig. 3: Abundance of Fusobacterium species in tumor/normal pairs.
figure 3

Relative abundance on a log scale by 16S rRNA amplicon sequencing of each depicted Fusobacterium species in pairs of tumor (blue circle) and normal (clear circle) tissues. Bars indicate median. Of 112 individuals with CRC in the study, 108 tumor and normal pairs were available for 16S rRNA sequencing analysis. Numbers below each graph depict the number of samples out of the total in which each species was not detected by sequencing (i.e. 0% abundance). Analyzed by two-tailed Wilcoxon matched-pairs signed rank test, with p < 0.05 considered significant. See Fig. S7 for abundance of F. nucleatum as a sum of confident and ambiguous species assignments, and for F. varium abundance stratified by ethnicity.

Full size image

Fusobacterium nucleatum and other oral organisms are associated with colon tumor biofilms

We next investigated individual and tumor characteristics associated with F. nucleatum abundance in the tumor microbiome. Enrichment of F. nucleatum in this cohort was independent of sex (Fig. 4a). Enrichment was seen among individuals with late-onset (LO) CRC, but not those with early-onset (EO) CRC (Fig. 4b), although the latter constituted a small proportion of the study population (n = 14). The three major ethnic groups in Malaysia are Malay, Chinese, and Indian, with data suggesting disparities in CRC incidence and mortality between these groups16,17. F. nucleatum was enriched in tumors of individuals of Chinese and Malay descent, but not statistically significant for the participants of Indian heritage (n = 14) (Fig. 4c). F. nucleatum was enriched in both left and right-sided tumors (Fig. 4). Notably, F. nucleatum enrichment was observed only in later cancer stages (p = 0.0041, 0.1037, and 0.0046, respectively, for stages II–IV) (Fig. 4e) and in biofilm-positive tumors (p < 0.0001) (Fig. 4f; Fig. S7c). This enrichment in later cancer stages is, in fact, driven by higher abundance of F. nucleatum in tumors that have biofilms (p = 0.0012, 0.0084, and p < 0.0001, respectively, for biofilm-positive tumors at stages II-IV, whereas biofilm-negative tumors had no significant enrichment at any cancer stage (Fig. 4g). This finding supports our observations from multiprobe FISH (Fig. 2h), highlighting that although F. nucleatum is present in a majority of tumors in this cohort, it is specifically enriched in association with biofilms in advanced disease.

Fig. 4: Fusobacterium nucleatum abundance by individual and tumor characteristics.
figure 4

Relative abundance by 16S rRNA amplicon sequencing of Fusobacterium nucleatum in pairs of tumor (blue circle) and normal (clear circle), stratified by A sex, B age, C ethnicity, D tumor location, E tumor stage, F biofilm (BF) status, and G both biofilm status and tumor stage. Bars indicated median. Of 112 individuals with CRC in the study, 108 tumor and normal pairs were available for 16S rRNA sequencing analysis. Numbers below each graph depict the number of samples out of the total in which F. nucleatum was not detected by sequencing. Analyzed by two-tailed Wilcoxon matched-pairs signed rank test, with p < 0.05 considered significant. EO-CRC early-onset colorectal cancer. LO-CRC late-onset colorectal cancer. See Fig. S7 for abundance of F. nucleatum as a sum of confident and ambiguous species assignments, stratified by biofilm status. H Matrix of F. nucleatum and other oral biofilm organism abundances in tumor samples, analyzed by nonparametric Spearman correlation coefficient. See Fig. S8 for relative abundance of each organism in the matrix in tumor/normal pairs.

Full size image

Because of the association of F. nucleatum with oral biofilms, we assessed for enrichment of other oral microbiome community members, including taxa known to associate with Fusobacterium in dental plaque biofilms18,19. 16S rRNA amplicon sequencing analysis revealed enrichment of Actinomyces odontolyticus (p = 0.0127), Capnocytophaga sputigena (p = 0.0004), Gemella morbillorum (p < 0.0001), Haemophilus influenzae (p = 0.0164), Lachnoanaerobaculum orale (p < 0.0001), Leptotrichia hofstadii (p = 0.0019), Leptotrichia trevisanii (p < 0.0001), Parvimonas micra (p = 0.0255), and Porphyromonas gingivalis (p = 0.0057) in the tumor microbiome (Fig. S9). Of these organisms, the relative abundance of C. sputigena (p = 0.015), G. morbillorum (p = 0.005), L. orale (p = 0.006), L. trevisanii (p = 0.002), and P. micra (p < 0.0001) were positively associated with abundance of F. nucleatum in the tumor microbiome (Fig. 4h). Relative abundances of two additional oral organisms, Leptotrichia buccalis (p = 0.019) and Peptostreptococcus stomatis (p < 0.0001) were associated with that of F. nucleatum, despite not being enriched in tumors relative to paired normal tissues (Fig. S9).

F. nucleatum subspecies animalis clade 2 is the predominant Fusobacterium enriched in colon tumors with biofilms

We next assessed the full cohort for the presence of the recently delineated clades within subspecies animalis using clade-specific amplicon sequence variants, as previously described12. We found that both subspecies animalis clade 1 (Fna C1) and clade 2 (Fna C2) were significantly enriched in the tumor microenvironment compared to normal tissues (p = 0.0046 and p < 0.0001, respectively) (Fig. 5a). However, Fna C2 was strikingly enriched in both tumors and normal tissues compared to Fna C1 (p < 0.0001). Similar to our findings at the species level (Fig. 4f), when compared to paired normal tissues, Fna C1 and C2 were only enriched in tumors that have biofilms (p = 0.0027 and p < 0.0001, respectively) (Fig. 5b). Abundance of Fna C2 was modestly enriched in biofilm-positive tumors (median 0.78%) relative to biofilm-negative tumors (median 0.19%) (p = 0.0140). When biofilm-positive tumors were stratified by stage, Fna C1 only reached statistically significant enrichment in stage 3 tumors (Fig. 5c), whereas Fna C2 was significantly enriched at all tumor stages (Fig. 5d). In our small cohort of four individuals with surgical adenomatous polyps, two had biofilms and two did not (Fig. S10A). F. nucleatum, including Fna C1 and Fna C2, were detected in a majority of these adenomas (Fig. S10B), although the small sample size precluded comparative analysis. Altogether, we find striking enrichment of Fna C2 in tumors that span the adenoma to advanced carcinoma sequence, and particularly in tumors displaying biofilms.

Fig. 5: Fusobacterium nucleatum subspecies animalis abundance by clade and correlated with FISH.
figure 5

A Relative abundance by 16S rRNA amplicon sequencing of Fusobacterium nucleatum subspecies animalis clade 1 (Fna C1) and clade 2 (Fna C2) in pairs of tumor (blue circle) and normal (clear circle). Tumor/normal pairs analyzed by two-tailed Wilcoxon matched-pairs signed rank test, and tumor/tumor or normal/normal comparisons analyzed by two-tailed Mann Whitney test, with p < 0.05 considered significant. B Relative abundance of Fna C1 and C2 stratified by biofilm (BF) status and C, D by both biofilm status and tumor stage. Bars indicated median. Of 112 individuals with CRC in the study, 108 tumor and normal pairs were available for 16S rRNA sequencing analysis. Numbers below each graph depict the number of samples out of the total in which each clade was not detected by sequencing. E Heatmap depicting relative abundance by 16S rRNA amplicon sequencing for the Fusobacterium genus and each labeled Fusobacterium species or clade for n = 29 tumors where dense blooms of Fusobacteria were visualized by FISH.

Full size image

We assessed whether Fna C2 abundance by 16S rRNA amplicon sequencing correlated with the density of Fusobacterium visualized by FISH (Fig. 5e). Of note, the Fusobacterium FISH probe is genus-specific. In tumor biofilms with dense blooms of Fusobacteria (n = 29), a small number of tumors had higher abundance of non-nucleatum Fusobacteria by sequencing analysis. However, in general, the abundance of Fna C2 was similar to the total estimated abundance of F. nucleatum (confident + ambiguous assignments) (Table S2), consistent with Fna C2 being dominant in the tumor microenvironment.

To identify whether other F. nucleatum subspecies were detectable in CRC tissues and to cross compare methodologies, we complemented our 16S rRNA amplicon sequencing analysis with a PCR-based method for distinguishing the F. nucleatum subspecies20. Of 34 tumors tested, F. nucleatum was detectable by PCR in 25, and among these, subspecies animalis was most frequently detected (Fig. S11a), consistent with previous reports20,21. When stratified by biofilm status, subspecies animalis was most frequently detected in biofilm positive tumors, whereas subspecies vincentii was more frequently detected in biofilm negative tumors (Fig. S11b). Notably, multiple subspecies were detected in 20% of tumor biopsies tested (5/25).

In parallel to sequencing and PCR analysis of Fusobacterium in the tumor microbiome of our cohort, we also used selective culture media to isolate unique Fusobacterium strains. Of the 13 Fusobacterium strains isolated (from 111 tumors), 6 were F. nucleatum, 50% of which were subspecies animalis (Table S3).

Biofilms and Fna C2 abundance correlate with predicted metabolic shifts

To begin to investigate functional consequences of Fna C2 enrichment and biofilm formation in the tumor microenvironment, we used the PICRUSt2 computational analysis pipeline to predict differences in metabolic pathways inferred from our 16S rRNA amplicon data (Fig. 6; Figs. S12, S13, S14; Table S2I). We also identified taxa that contributed to enrichment of these pathways (Fig. 6b; Table S4). Analysis was performed on 112 tumor samples and 112 normal samples (inclusive of 4 normal samples from patients with surgical polyps). We identified 67 PICRUSt2 pathways that predicted differential metabolic function between biofilm-positive tumor vs. biofilm-positive normal samples, and only 9 that predicted differences in metagenome content between biofilm-negative tumor and biofilm-negative normal samples (Fig. 6a). These data suggest that the presence of biofilms strongly impacts metabolic shifts in the microenvironment during tumor development. For example, in biofilm-positive tumors, there was a significant increase in gene content related to CMP-pseudaminate biosynthesis. Pseudaminic acids are sialic-acid-like sugars on the surface of many gram-negative bacteria, including Bacteroides and Campylobacter22,23, which were the top taxa contributing to enrichment of this pathway in tumors compared to normal tissues (Fig. 6b). Data suggests that microbial-derived sialic-acid like sugars may facilitate immune evasion and biofilm formation24.

Fig. 6: PICRUSt2 pathways associated with tumors, biofilms, and Fna C2 abundance.
figure 6

A Heatmap of all statistically significant PICRUSt2 pathways with differences in gene content between tumor (n = 112) and normal (n = 112) samples for either biofilm-positive tissues (left) or biofilm-negative tissues (right), as analyzed by Mann Whitney test, with p < 0.05 considered significant. Data depicted as Log2 Fold Change between tumor and normal samples (>0 indicates higher in tumor). B For each indicated comparison, the listed pathways are the 4 PICRUSt2 pathways with the greatest Log2 Fold Change between groups. For each pathway, the genera listed are the top 3 taxa contributing to enrichment of that pathway. See Figs. S12, S13, and S14A for the full display of significantly enriched or depleted pathways for each comparison. See Table S4 for all taxa significantly contributing to each enriched pathway.

Full size image

To further investigate the impact of biofilms on predicted tumor microbiome metabolism, we compared biofilm-positive and biofilm-negative tumors, identifying 106 PICRUSt2 pathways that significantly differed (Fig. S12a). Among the PICRUSt2 pathways with the highest fold change between biofilm-positive and negative tumors (Fig. 6b) were pathways related to glucose metabolism and nucleotide biosynthesis, both of which are known to be dysregulated in cancer25,26. Interestingly, enrichment of these pathways in biofilm-positive tissues relative to biofilm-negative tissues was driven both by oral bacteria such as Streptococcus, as well as several commensal taxa typically thought of as beneficial to the host (e.g., Clostridium XIVa, Blautia, and Faecalibacterium).

Comparison of tumor samples with low Fna C2 abundance (<1%) and high abundance (≥1%) yielded 83 differentially predicted PICRUSt2 pathways. 78.3% (65/83) of the pathways associated with Fna C2 abundance in tumors were unique from those associated with biofilm status (Fig. S14a). Many of these pathways are involved in amino acid and cofactor biosynthesis, carbohydrate degradation, energy metabolism, or represent metabolic superpathways. Among the PICRUSt2 pathways with the highest fold-change in high Fna C2 abundance relative to low-abundance tumors (Fig. 6b) were protein N-glycosylation, which is known to be aberrant in tumors and is potentially a critical factor in tumor immune evasion27. Also associated with Fna C2 abundance was L-1,2-propanediol degradation, with Fusobacterium itself as the top contributing taxon, altogether consistent with the previous observation that Fna C2 has conservation of genetic content associated with 1,2-propanediol (1,2-PD) metabolism12. Several of the genera associated with these Fna C2-associated PICRUSt2 pathways (Fig. 6b) contain oral microbiome members (e.g., Campylobacter, Leptotrichia, and Selenomonas), further linking the oral microbiota with putative functional changes in the tumor microenvironment.

Lastly, we sought to investigate whether Fna C1 abundance was associated with a metabolic signature similar to Fna C2. When Fna C1 is present in the tumor microenvironment, it is typically at low abundance (median 0.02%) and absent in nearly 40% of tumors (Fig. 5a). We therefore sought to identify PICRUSt2 pathways significantly associated with presence or absence of Fna C1. We identified 34 pathways differentially represented in tumors with or without Fna C1, with many pathways associated with aromatic compound degradation and carbohydrate metabolism (Fig. S14b). Of these 34 pathways, only 5 pathways were significantly associated with both Fna C1 and Fna C2 abundance: L-glutamate degradation V, L-1,2-propanediol degradation, protein N-glycosylation, tRNA processing, and teichoic acid biosynthesis.

Discussion

Fusobacterium nucleatum, which typically resides in the oropharynx, has been consistently linked to the colon cancer microbiome in numerous studies across diverse cohorts3,4,28,29,30,31. F. nucleatum enrichment has been associated with specific pathological and clinical features such as high levels of microsatellite instability (MSI-high), the CpG island methylator phenotype (CIMP), and poor prognosis32,33,34,35. Controversy remains as to the precise role(s) F. nucleatum plays in tumor initiation, progression, and immune evasion, although a large body of data argues against it merely existing as a passive colonizer of colon tumors. One key barrier to uncovering critical host-bacterial interactions in the tumor microenvironment is the diversity of F. nucleatum strains employed to date in mechanistic studies that may not accurately reflect the genomic content or adhesion factor repertoire of CRC-associated strains. Therefore, a key question has been which subspecies of this diverse organism are enriched in colon tumors. Inconsistent phenotypes in mouse models have also raised the question as to the critical community partners with which F. nucleatum may collaborate to colonize the lower gastrointestinal tract and promote carcinogenesis36,37,38.

In this large cohort of individuals with CRC, we report that F. nucleatum is significantly enriched in the colon tumor microbiome when compared to distant normal tissues. A combination of sequencing, culture, and spatial evaluation has, in particular, allowed us to link F. nucleatum abundance in the tumor microenvironment with the presence of tumor biofilms as well as with enrichment of other oral microbiome members, particularly in advanced cancer stages. These data build on previous analyses linking F. nucleatum to CRC biofilms10, and confirm the finding that biofilms on colon tumors frequently contain Fusobacteria. Although F. nucleatum has previously been linked to CRC biofilms, we report the surprising finding that F. nucleatum is only enriched in tumor tissue relative to normal tissue in individuals who have biofilms, a phenomenon largely driven by subspecies animalis clade 2 (Fna C2).

We find that subspecies animalis is most frequently detected by PCR and isolated by culture, consistent with other reports20,21. One limitation of our study was inability to perform strain-level comparisons of Fusobacterium isolates because of low culture yield, generating a library of only 13 Fusobacteria strains from the cohort of 112 individuals. Culture yield was likely impacted by long-term storage of tumor biopsies for 5–10 years prior to attempts at isolating Fusobacterium strains. Recent work has identified two clades (Fna C1 and C2) within subspecies animalis that were frequently detected in the mouths of healthy individuals, whereas only clade Fna C2 was enriched in colon tumors12. We confirm this finding in our cohort, with Fna C2 representing the most abundant Fusobacterium in the CRC microbiome, although our data show that other Fusobacterium species are also significantly enriched in the tumor microenvironment. Although we consider Fna C2 a high priority group for future investigation, fusobacterial communities in CRCs are complex, strongly suggesting that investigation of community features beyond Fna C2 is required to understand the mechanistic determinants of biofilms, Fusobacterium, and CRC.

By combining microbial sequencing and spatial analysis, we find that F. nucleatum, dominated by Fna C2, is frequently present in CRC biofilms and can bloom into dense populations, the latter rarely identified in biofilms on normal tissue10. This observation supports the theory that F. nucleatum is itself not a tumor initiator, but blooms in the tumor environment under specific conditions that allow it to flourish and subsequently promote tumor progression. A recent study in a screening colonoscopy cohort found that biofilms co-evolve with adenomas, increasing in density as polyps increase in size39. Likewise, our data suggest that biofilms likely precede or at least coincide with tumor initiation, as they are consistently concordant between tumor and distant normal tissue. However, while biofilms are observed in normal tissue in our cohort, Fusobacterium appears to colonize and bloom within polymicrobial biofilms only as tumors progress.

Notably, among the subset of tumors with Fusobacterium blooms, the overall abundance of F. nucleatum and other Fusobacterium species ranged in the tumor microenvironment from 0.06 to 40.8% of total microbial reads, suggesting that there are concentrations of Fusobacteria in specific niches in the tumor microenvironment that are underappreciated by nonspatial sequencing techniques. One factor to consider in this analysis is that tumor samples processed for 16S rRNA sequencing and FISH were from different biopsy regions. Although microbiome alpha-diversity is greater between individuals than between multiple biopsies from within a single tumor40, there is a great deal of spatial heterogeneity of tumor microbiome composition41. Future investigations into the role of F. nucleatum in CRC pathogenesis will need to harness emerging spatial omics analyses to better understand the specific interactions of F. nucleatum and biofilm community members with host cells.

Although we saw only a modest impact of biofilms on compositional diversity, PICRUSt2 analysis predicted a marked impact on metabolic diversity, perhaps owing in part to the spatial organization of the microbial community. One key open question is whether F. nucleatum serves a similar role in colonic biofilms as in oral biofilms—that of a bridging species orchestrating the composition, structure, and potentially the function of polymicrobial biofilm communities. PICRUSt2 analysis suggests that both biofilms and Fna C2 abundance in the tumor microenvironment may be associated with altered, but differential, metabolic functions, a key hypothesis for future study. Whether the combined metabolic impact of biofilms and Fna C2 accelerates tumor development and/or metastatic disease is an open question. Consistent with previous studies, we also find enrichment of other oral organisms in colon tumors and contributions of oral organisms to enriched PICRUSt2 pathways, suggesting that F. nucleatum expansion occurs in the context of an oral microbiome community10,42,43.

These analyses build on existing data demonstrating an association of F. nucleatum with the colon tumor microenvironment, and add critical resolution that highlights the abundance of F. nucleatum, and specifically subspecies animalis clade 2, in mucus-invasive biofilms observed particularly in late-stage disease. Our findings suggest that F. nucleatum:bacterial interactions within mucus-invasive biofilms are critical, yet understudied, features of CRC carcinogenesis.

Methods

Study participants

This study was approved by the Johns Hopkins Institutional Review Board and the University of Malaya Medical Centre (UMMC, Kuala Lumpur, Malaysia) Medical Ethics Committee. Written informed consents (provided in Malay and English) were obtained from all participants. All samples were obtained in accordance with the Health Insurance Portability and Accountability Act. Individuals who had received pre-operative radiation, chemotherapy or had a prior history of CRC were excluded. All individuals in the study underwent a standard mechanical bowel preparation. Standard pre-operative intravenous, but not oral, antibiotics were administered in all surgical cases. We collected data on the following variables: age, sex, ethnicity, BMI, smoking history, alcohol consumption, tumor location, tumor pathology, tumor grade, and tumor stage. Individuals under age 50 were defined as having early-onset CRC (EO-CRC), whereas individuals aged 50 and over were defined as having late-onset CRC (LO-CRC). Study data were collected and managed using REDCap electronic data capture tools hosted at Johns Hopkins University.

Sample collection

Excess colon tumor and paired normal tissues were collected for analysis from individuals undergoing surgery at UMMC. Surgical tissues were fixed in methacarn or flash frozen for later analysis.

Fluorescence in situ hybridization (FISH) analysis of biofilms

FISH was performed on methacarn-fixed, paraffin-embedded tissue sections as previously described10. Briefly, sequential sections were stained with the Eub338 universal bacterial probe for the presence of biofilms. Serial sections of samples confirmed to have biofilms were stained with specific bacterial probes for Fusobacterium, Bacteroidetes, Betaproteobacteria, Gammaproteobacteria, and Lachnospiraceae (see Oligonucleotide Table). Quantification of bacteria, lambda scanning, and linear unmixing were performed on a Zeiss 780 laser scanning confocal microscope as previously described10. A biofilm was defined as a dense aggregation of bacteria, with at least 20 bacteria within 1 μm of the epithelium (equivalent to 2 × 109 bacteria/ml)9, covering an expanse of at least 200 μm of the CEC layer, in at least one of three screened areas.

16S rRNA gene Illumina library generation and sequencing

16S rRNA amplicon sequencing of cohorts MAL1, MAL2, and MAL3 were performed at different facilities with different methodologies for the three sets of samples. For each cohort, the V3-V4 hypervariable region of the 16S rRNA gene was amplified and sequenced.

For the MAL1 cohort, samples were lysed by suspending in 700 μL PBS and incubating with lysozyme (5 μL of 10 mg/mL stock), mutanolysin (15 μL of 1 mg/mL stock), and lysostaphin (5 μL of 1 mg/mL stock) in lysing matrix tubes at 37 °C. After 30 min, 10 μL Proteinase K and 50 μL 10% SDS was added, and samples were vortexed and then incubated at 55 °C for 45 min. Mechanical lysis was performed in a FastPrep-24 5G instrument at 6.0 m/s for 40 sec, then centrifuged at 10,000 g for 3 min. The ZR Fecal DNA MiniPrep kit (Zymo Research) was then used for DNA extraction. The V3-V4 region was amplified using primers 319 F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) containing a linker sequence required for Illumina MiSeq 300 bp paired-end sequencing and a 12-bp heterogeneity-spacer index sequence. Sequencing was performed on the Illumina MiSeq (Illumina, San Diego, CA) according to the manufacturer’s protocol.

For the MAL2 cohort, samples were lysed and DNA extracted using the MasterPure DNA Purification Kit (Epicentre/Illumina) according to the manufacturer’s instruction. Primers used for amplification were S-D-Bact-0341-b-S-17 forward (5′-NNNNCCTACGGGNGGCWGCAG-3′) and S-D-Bact-0785-a-A-21 reverse (5’-GACTACHVGGGTATCTAATCC-3’), designed to include Illumina-compatible adaptors. Sequencing was performed on the MiSeq System (Illumina, San Diego, CA, USA) at the Monash University Malaysia Genomics Facility using the MiSeq 500-cycle reagent kit V2 on standard flow cell.

For the MAL3 cohort, DNA was extracted using the ZR Quick-DNA Fecal/Soil Microbe 96 Kit or Miniprep Kit (Zymo Research) with the following modification: 1.0 mm Zirconia-Silicate beads (BioSpec) were added to the 0.1/0.5 mm beads provided in the kit for improved tissue homogenization. Samples were homogenized and lysed using the Mini-beadbeater 96 (BioSpec) at maximum speed for 60 s x 3. Lysate was then processed according to manufacturer instructions. Sequencing was performed at Novogene with DNA amplified with primers 341 F (5’-CCTAYGGGRBGCASCAG-3’) and 806R (5’-GGACTACNNGGGTATCTAAT-3’), and sequencing performed on the Illumina NovaSeq 6000 platform.

Analysis of all 16S rRNA amplicon sequence data sets

Paired-end 16S rRNA amplicon sequences were trimmed for quality and length using Trimmomatic, merged using FLASH (Fast Length Adjustment of Short reads)44, and quality screened using QIIME45. Spurious hits to the PhiX control genome were identified using BLASTN (Basic Local Alignment Search Tool) and removed. Passing sequences were trimmed of 16S rRNA primers, evaluated for PCR chimeras and filtered for host-associated contaminant using Bowtie246. Chloroplast contaminants were identified and filtered using the RDP (Ribosomal Database Project) classifier47 with a confidence threshold of 50%. High-quality 16S rRNA amplicon sequences were assigned to a high-resolution taxonomic lineage using Resphera Insight, which uses a hybrid global-local alignment strategy to a manually curated 16S rRNA database with 11,000 unique species10,48,49. This approach attempts to achieve species-level resolution when possible; however, when a confident single species assignment is not feasible, the method minimizes false positives by providing “ambiguous assignments” i.e., a list of candidate species reflecting the ambiguity. Sequences that cannot be confidently assigned to any known species are clustered de novo (at 97% identity) and resulting clusters are given an OTU # identifier and annotated with a closest relative (or set of relatives). Taxonomic assignments are presented as unambiguous assignments by Resphera Insight unless otherwise indicated. Normalization of observations were performed by rarefaction followed by alpha and beta diversity characterization.

Taxonomic abundances were converted from counts to percentages. LEfSe was run to identify strong effect differentiating taxa among the three replicate cohorts of interest, with visualization of LDA scores by barplot and taxonomic cladogram. The method was run requiring a nonparametric P-value threshold of 0.01, and LDA score threshold of 2.5. Taxonomic profiles for samples reflecting replicates were compared pairwise for quantitative concordance of detected genera and species and using Pearson and Spearman and correlation. Statistical associations with beta-diversity utilized PERMANOVA. Functional inference of microbial gene content was evaluated using PICRUSt250. Predicted PICRUSt2 metagenomic contributions per taxon to each pathway were calculated per sample and converted to relative abundance values, followed by comparisons between groups of interest using the Mann-Whitney U Test with P-value threshold of 0.05. Classification of Fusobacterium nucleatum C1 or C2 subgroups was performed for individual 16S rRNA amplicon sequences, using BLASTN (ungapped alignment; word size 124; minimum 99% identity) against a reference database of C1 and C2 amplicon sequence variant representatives12. Candidate sequences with alignments exclusively to C1 or C2 were labeled accordingly.

Subspecies PCRs

DNA was extracted from 3 mm tumor punch biopsies using the ZR Quick-DNA Fecal/Soil Miniprep Kit (Zymo Research) with the following modification: 1.0 mm Zirconia-Silicate beads (BioSpec) were added to the 0.1/0.5 mm beads provided in the kit for improved tissue homogenization. Samples were homogenized and lysed using the Mini-beadbeater 96 (BioSpec) at maximum speed for 60 s × 3. Lysate was then processed according to manufacturer instructions. DNA concentration was measured on a nanodrop, and ranged from 30.4 to 260.5 ng/μl. DNA was screened for F. nucleatum using primers specific for nusG (forward, 5’-CAACCATTACTTTAACTCTACCATGTTCA-3’; reverse, 5’- GTTGACTTTACAGAAGGAGATTATGTAAAAATC-3’)3 at a primer concentration of 0.9 μM using a PCR program of 98 °C × 60 s; 35 cycles of 98 °C × 30 s, 58 °C × 60 s, 72 °C × 20 s; and 72 °C × 7 min. DNA was then screened using primers specific for each F. nucleatum subspecies (See Oligonucleotide Table)20 at a primer concentration of 0.5 μM, using a PCR program of 98 °C × 60 s; 30 cycles of 98 °C × 30 s, 56 °C × 15 s, 72 °C × 20 s; and 72 °C × 7 min.

Selective culture

Three mm punches were taken from flash-frozen tumor biopsy tissues and homogenized with a sterile pestle in 50 μl brain heart infusion (BHI) media supplemented with hemin (10 μg/mL) and vitamin K (5 μg/mL). Total volume was brought up to 500 μl of BHI and vortexed. 10 μl of the homogenate was spread onto Fusobacterium selective agar (FSA; Anaerobe Systems) and 10 μl were subcultured in 3 mL of BHI. Each was incubated for 48–72 h under anaerobic conditions (75% N2, 5% H2, 20% CO2) in an anaerobic chamber. From the homogenate culture in BHI media, 10 μl was then streaked onto FSA plates and incubated as above. From these two FSA plates (1 directly plated, 1 subcultured first in BHI), unique colonies were picked and spread onto new FSA plates for isolation (minimum 6 colonies/tumor) and incubated. After 48–72 h, colonies were scooped up and swirled into 50 μl sterile water, which was boiled to release bacterial DNA. This crude DNA prep was used as a template for screening by quantitative real-time PCR (qRT-PCR) with Fusobacterium genus-specific 16S primers (forward, GGATTTATTGGGCGTAAAG; reverse, GGCATTCCTACAAATATCTACGAA)51 and probe (5’VIC- TGC AGG GCT CAA CTC TGT ATT GCG - 3’BHQ1), at a concentration of 0.2 μM, and with the following program: 50 °C × 2 min; 95 °C × 10 min; 40 cycles of: 95 °C × 15 s, 58 °C × 60 s. Colonies were considered positive if the CT value was <30. Each positive colony was subcultured into BHI media × 48 h, and genomic DNA was extracted from pelleted bacteria using the ZR Quick-DNA Fecal/Soil Miniprep Kit (Zymo Research). Species and subspecies were confirmed by Sanger sequencing of the 16S rRNA gene (Azenta Life Sciences).

Quantification and statistical analysis

To compare categorical data between defined groups (e.g., biofilm status by tumor location), data were analyzed using a two-tailed Mann Whitney nonparametric t-test. Comparison of biofilm subtype across tumor stages was analyzed by a nonparametric Kruskal-Wallis one-way ANOVA. Prior to downstream statistical comparisons, 16S rRNA amplicon profiles within each dataset were subsampled to an even level of coverage (9500 reads). For comparisons of relative taxonomic abundances between matched tumor-normal pairs, data were analyzed by a two-tailed non-parametric Wilcoxon signed-rank sum test. To identify associations of oral taxa abundances with F. nucleatum abundance, we performed a non-parametric Spearman correlation. Differences in PICRUSt2 pathway content was analyzed by Mann Whitney Test or by Kruskal-Wallis test with Dunn’s multiple comparisons, as indicated in the figure legend. For all analyses, differences with a P value of <0.05 were considered significant.

Oligonucleotide Table

Oligonucleotide

Source

Eub338 universal bacterial FISH probe (5′ - Cy3 -GCTGCCTCCCGTAGGAGT - 3′)

MBD0033

Sigma-Aldrich, USA

Fusobacterium FISH probe (Fus714; 5′ - Cy3 – GGCTTCCCCATCGGCATT - 3′)

Valm et al.52

Bacteroidetes FISH probe (CFB286; 5′-Alexafluor488-TCCTCTCAGAACCCCTAC - 3′)

Weller et al.53

Betaproteobacteria FISH probe (Bet42a; 5′ - Cy5 - GCCTTCCCACTTCGTTT - 3′)

Manz et al.54

Gammaproteobacteria FISH probe (Gam42a; 5′ - Cy5 - GCCTTCCCACATCGTTT - 3′)

Manz et al.54

Lachnospiraceae FISH probe (Lac435; 5′ - TXR - TCTTCCCTGCTGATAGA - 3′)

Kong et al.55

Animalis primers:

SR-a2-F, ACTCAAATTATTATGAATGTGATGAAAGA

SR-a2-R, GCTACTGAAGGATGAAATGCTGG

Bi et al.20

Nucleatum primers:

SR-n3-F, CAAGCAACTGAAAATGCTTTAAAAG

SR-n3-R, TCCAGGTAAGGAAATTACACCTACTG

Bi et al.20

Polymorphum primers:

SR-p1-F, TTAGGAAATCTTTTAGAAGCAAAAACA

SR-p1-R, TCTACTGTAATAGTTACAAACTCTGCWCC

SR-p3-F, CYTATGGYTTTGATTTTGACTTATTTG

SR-p3-R, CCAAAGTAATTAAAGCCTCTTGAGC

Bi et al.20

Vincentii primers:

SR-v1-F, GAGGCTATTGCAAATTAAACTGTTAAA

SR-v1-R, CTTTACCACTATTATAAACTAAATAAAT

Bi et al.20