- Open Access
Distinct microbes, metabolites, and ecologies define the microbiome in deficient and proficient mismatch repair colorectal cancers
- Vanessa L. Hale†1, 2, 3,
- Patricio Jeraldo†2, 3,
- Jun Chen2, 4,
- Michael Mundy5,
- Janet Yao2,
- Sambhawa Priya6,
- Gary Keeney7,
- Kelly Lyke3,
- Jason Ridlon8,
- Bryan A. White8,
- Amy J. French7,
- Stephen N. Thibodeau7,
- Christian Diener9,
- Osbaldo Resendis-Antonio9, 10,
- Jaime Gransee11,
- Tumpa Dutta11,
- Xuan-Mai Petterson11,
- Jaeyun Sung2, 3,
- Ran Blekhman6,
- Lisa Boardman12,
- David Larson13,
- Heidi Nelson2, 13 and
- Nicholas Chia2, 3Email authorView ORCID ID profile
© The Author(s). 2018
- Received: 13 June 2018
- Accepted: 8 October 2018
- Published: 31 October 2018
Links between colorectal cancer (CRC) and the gut microbiome have been established, but the specific microbial species and their role in carcinogenesis remain an active area of inquiry. Our understanding would be enhanced by better accounting for tumor subtype, microbial community interactions, metabolism, and ecology.
We collected paired colon tumor and normal-adjacent tissue and mucosa samples from 83 individuals who underwent partial or total colectomies for CRC. Mismatch repair (MMR) status was determined in each tumor sample and classified as either deficient MMR (dMMR) or proficient MMR (pMMR) tumor subtypes. Samples underwent 16S rRNA gene sequencing and a subset of samples from 50 individuals were submitted for targeted metabolomic analysis to quantify amino acids and short-chain fatty acids. A PERMANOVA was used to identify the biological variables that explained variance within the microbial communities. dMMR and pMMR microbial communities were then analyzed separately using a generalized linear mixed effects model that accounted for MMR status, sample location, intra-subject variability, and read depth. Genome-scale metabolic models were then used to generate microbial interaction networks for dMMR and pMMR microbial communities. We assessed global network properties as well as the metabolic influence of each microbe within the dMMR and pMMR networks.
We demonstrate distinct roles for microbes in dMMR and pMMR CRC. Bacteroides fragilis and sulfidogenic Fusobacterium nucleatum were significantly enriched in dMMR CRC, but not pMMR CRC. These findings were further supported by metabolic modeling and metabolomics indicating suppression of B. fragilis in pMMR CRC and increased production of amino acid proxies for hydrogen sulfide in dMMR CRC.
Integrating tumor biology and microbial ecology highlighted distinct microbial, metabolic, and ecological properties unique to dMMR and pMMR CRC. This approach could critically improve our ability to define, predict, prevent, and treat colorectal cancers.
The gut microbiota has been linked to colorectal cancer (CRC) in many studies [1–9] and serves as a very promising target for diagnostic, prophylactic, and therapeutic applications. Yet, despite intense study, only a few microbial species—like Fusobacterium species—are consistently observed across studies [10–14], while many microbial associations appear to be cohort-specific. Meta-analyses have attempted to overcome the limited statistical power of smaller studies  but are limited by the strong biases introduced through varying collection, sequencing, and data processing methodologies . Mechanistic studies in mouse models have identified strong causative links between specific microbes (e.g., Fusobacterium nucleatum, Bacteroides fragilis) and CRC development and progression [11, 17–24], but these models have limited applicability in genetically diverse human populations. Capturing some of this genetic diversity, on the other hand, may improve our ability to discriminate tumor and normal microbial communities and more clearly define pathways to CRC.
One genetic subtype of CRC is based on the presence or absence of mutations in the DNA mismatch repair system. This system involves multiple protein complexes that recognize, remove, and correct mismatched DNA base pairs. Mutations in these protein complexes can render the mismatch repair system defunct—allowing mutations to accumulate. This hypermutable subtype is known as deficient mismatch repair (dMMR) and occurs in approximately 15% of sporadic CRCs . CRCs that do not exhibit mutations in the mismatch repair system are known as proficient mismatch repair (pMMR) CRCs . In general, dMMR CRCs are microsatellite instable (MSI-H), hypermethylated, and associated with BRAF V600E mutations and low nuclear beta-catenin expression, whereas pMMR CRCs are more commonly microsatellite stable (MSS) and associated with KRAS mutations [27, 28]. Clinically, MMR status is associated with patient prognosis and age, as well as tumor location and stage: Specifically, dMMR CRCs have a better prognosis and occur more often on the right side of the colon in older patients with early-stage CRC . Finally, dMMR and pMMR CRC not only have different endpoints, but may also have different paths to tumorigenesis  as supported by emerging evidence that dMMR CRC arises from sessile serrated adenomas  as opposed to the more classic tubular adenoma associated with pMMR CRC .
The distinct phenotype of dMMR CRC suggests that host—and possibly also microbial—dynamics are greatly altered in association with deficient mismatch repair. Previous work has examined the role of other differentiating factors in the CRC microbiome including location , MSS/MSI status , and consensus molecular subtypes . However few CRC microbiome studies account for MMR status [32–34] or microbial dynamics , and no studies, to our knowledge, have assessed both MMR status and microbial community dynamics. In addition, our study examines demographic, genetic, and tumor features together in a relatively large prospectively collected cohort.
Here, we undertook a new approach in a study involving 83 patients who underwent partial or total colectomy for CRC. From each patient, we collected colon tissue and mucosal samples at tumor and normal-adjacent sites. MMR status was extracted from patient records or determined by testing formalin-fixed paraffin-embedded tumor tissue for the expression of four MMR proteins (MLH1, MSH2, MSH6, PMS2). From this, patient tumors were characterized as either deficient (dMMR) or proficient (pMMR) mismatch repair. Microbial composition was assessed via 16S rRNA gene sequencing. A subset of colon tissue samples additionally underwent targeted metabolomic analysis to quantify amino acids. A portion of these data was published previously  in a study that highlighted the value of integrating in silico genome-scale metabolic model predictions and in vivo experimental metabolomic data.
From these data, we assessed the relative importance of MMR status compared to other biological factors reported to alter the microbiome . MMR status was the strongest predictor of microbial community variance in comparison to sample location (proximal/distal and on/off tumor), body mass index (BMI), age, and sex. Separate analyses of the dMMR and pMMR microbial communities revealed that many common CRC-associated microbial signatures —including Fusobacterium nucleatum, Fusobacterium periodonticum, and Bacteroides fragilis—were all enriched in dMMR but not pMMR tumors. Functional differences were examined using a combination of metabolomics and community metabolic modeling. Our results indicate greater hydrogen sulfide production (inferred through amino acid proxies) in dMMR CRC and greater metabolic suppression of B. fragilis in pMMR CRC. Our work demonstrates distinct microbial, metabolic, and ecological attributes of dMMR and pMMR microbial communities, serving to further emphasize the importance of considering tumor biology and microbial interactions in studies of the CRC microbiome.
Human subject enrollment
Demographic and tumor features of individuals identified as having dMMR or pMMR CRC
Sex, n (%)
Smoke ever? n (%)
Tumor location, n (%)
p < 0.0001 between proximal and distal
Stage, n (%)
0.0007 between early and late
MMR status determination
Mismatch repair (MMR) pathway and microsatellite instability (MSI) test results were extracted from patient records if available. For patients without MMR test results, banked formalin-fixed paraffin-embedded colon tumor tissue blocks were submitted to the Mayo Clinic Pathology Resource Core for sectioning into 10-μm-thick slices. Slices were then submitted to the Mayo Clinic Molecular Genetics Laboratory for immunohistochemistry staining of MMR proteins (MLH1, PMS2, MSH2, MSH6).
16S DNA extraction, sequencing, and sequence processing
DNA extraction  and library preparation on colon tissue (tumor and normal-adjacent) and mucosa were performed as described previously in the Mayo Clinic Microbiome Laboratory . Samples were submitted for 16S rRNA gene sequencing (V3–V5 region) at the Mayo Clinic Medical Genomics Facility (Illumina MiSeq, 2 × 300, 600 cycles, Illumina Inc.). Sequencing yielded a total of 41,400,384 reads with a median of 70,208 reads per sample. Reads were processed using DADA2 v1.6 to obtain error-corrected amplicon sequence variant representatives—analogous to operational taxonomic units with single-nucleotide resolution (sOTUs) . sOTUs were annotated with genus-level taxonomy using the RDP Naïve Bayesian Classifier  as implemented in DADA2 and, if possible, to species level using DADA2, both against the SILVA 16S database, v132 . sOTUs annotated as chloroplast and mitochondria were removed. Resulting sOTUs were filtered for possible non-specific amplification using SortMeRNA v2.0  and Infernal v1.1.2 . sOTUs with fewer than 10 reads across all samples were excluded. Multiple sequence alignment of the sOTUs was performed using Infernal v1.1.2 , and an approximate Maximum Likelihood phylogeny was calculated using FastTree v2.1.9 .
Statistical analyses of 16S rRNA microbial community data
UniFrac distance matrices  based on the microbial communities in all samples were generated using the phyloseq  package v1.22.3. A permutational multivariate analysis of variance (PERMANOVA) was then performed on the distance matrix to assess the effects of MMR status and sample location (proximal/distal and on/off tumor) on variance between microbial communities. The PERMANOVA additionally accounted for subject age, sex, BMI, and sample type (mucosa versus colon tissue) and was performed based on the adonis function in the vegan  package v2.5-1, with 999 permutations. Different permutation schemes were used to maintain the original correlation structure when testing the significance of relevant variables.
A generalized linear mixed model (GLMM)  was calculated for each sOTU to estimate its abundance (read counts) in relation to predictors that included MMR status and sample location (proximal/distal and on/off tumor). Models were corrected for subject intervariability, specimen type (mucosal vs tissue biopsy), and sequencing read depth, allowing for interactions. We used the package glmmTMB  v0.1.4 to estimate the abundance of each microbe under a zero-inflated Poisson distribution. For each predictor, sOTUs were excluded where the method did not converge or the Akaike Information Criterion (AIC) for model quality was not defined. Multiple hypothesis correction was calculated using the Benjamini–Hochberg procedure.
Validation of differentially abundant microbes using an independent cohort
To validate the differentially abundant microbes associated with dMMR status, we investigated data from a recent study that included microbiome profiling in tumor and matched normal tissue samples in 44 CRC patients . Individuals with microsatellite instable (MSI-H) tumors or downregulation of any of the 4 MMR genes (MLH1, MSH2, MSH6 and PMS2)—as assessed using RNA-Seq—were categorized as dMMR. A cutoff of log2(normal/tumor) ≥ 1 was used to call a gene as downregulated in tumor. Individuals with microsatellite stable (MSS) tumors were categorized as pMMR. Altogether, we identified 9/44 patients as dMMR and the remaining 35/44 as pMMR. Using the 16S rRNA gene to characterize these samples (as described in detail in ), we identified sOTUs associated with dMMR tumor/normal and pMMR tumor/normal conditions. We first filtered rare sOTUs, only preserving sOTUs found in at least 50% of our samples, and then performed differential abundance analysis using phyloseq  (which uses DESeq2 to build negative binomial generalized linear models). We used the Benjamini–Hochberg method to control for the false discovery rate (FDR).
Real-time PCR for the Bacteroides fragilis toxin gene
Real-time PCR was performed as described previously  to test colon tissue and mucosal samples for the presence of the Bacteroides fragilis toxin (BFT) genes in the 22 dMMR individuals and 53 pMMR individuals. Primers included: BFT-F (5′-GGATAAGCGTACTAAAATACAGCTGGAT-3′), BFT-R (5′-CTGCGAACTCATCTCCCAGTATAAA-3′), and the probe (5′-FAM-CAGACGGACATTCTC-NFQ-MGB-3′) .
Modeling microbial hydrogen sulfide production
We predicted hydrogen sulfide production within dMMR and pMMR tumor and normal-associated microbial communities as described previously . Briefly, we aligned 16S rRNA gene sequences for dMMR tumor and normal samples (colon tissue and mucosa) and pMMR tumor and normal samples against complete genomes in PATRIC and then generated genome-scale metabolic models of each microbe (Additional file 1: Table S1). Genome-scale metabolic models use gene annotations from a microbial genome to predict the metabolic inputs and outputs of that microbe. To predict how a microbe might interact within a community, we used MICOM, an open-source platform to assess microbial community metabolism (https://github.com/resendislab/micom). Specifically, we used flux balance analysis with MICOM’s community growth objective and constraint formulation in order to evaluate hydrogen sulfide flux as a measure of hydrogen sulfide production within each microbial community.
Microbial influence network
To select sOTUs for the Microbial Influence Networks (MINs), we used GLMM results to choose tumor and normal-associated microbes in dMMR and pMMR samples with a linear effect size greater than 0.25, regardless of statistical significance. Effect size captures biological impact potential while significance measures certainty. In this case, we wanted to assess the metabolic influence (i.e., biological impact) of microbes in relation to their respective microbial communities; as such, it was more appropriate to filter by effect size. For each sOTU, the 16S rRNA gene consensus sequence was aligned against complete genome in the PATRIC system using VSEARCH v2.7.1, with a minimum nucleotide identity of 90%. When this procedure generated multiple top hits, we selected a genome, in order, to the most complete genome (fewer contigs), a type strain, a strain with a binomial name, and the closest match to the 16S taxonomy (when possible). For each genome, we then reconstructed and downloaded its corresponding genome-scale metabolic model using the PATRIC service. When sOTUs mapped to the same model, we used that model only once, effectively merging those sOTUs in further analysis, with an exception for when two sOTUs were associated with opposite conditions (i.e., tumor and normal-adjacent samples), in which case, we discarded that model from further consideration. The decision to discard was also based on the observation that low identity hits or sOTUs with taxonomy not sufficiently resolved were typically involved in these few cases.
This scoring closely follows the spirit of the scoring from the global metabolic interaction modeling in Sung et al. ; to derive interactions, we used growth rates that were computationally inferred from comprehensive metabolic models in contrast to using experimentally verified transport reactions from a limited number of microbes and metabolites. Metabolic modeling based on flux balance analysis, as described here, provides a means to calculate a rate of steady-state growth, as normalized per unit mass, allowing us to take a simple sum in order to calculate influence under anaerobic conditions.
The percentage of negative interactions was calculated by counting the number of negative interactions over the number of total interactions in each microbial influence network (MIN). Statistical significance was based on the probability of getting equivalent results in dMMR and pMMR networks using the measured distributions of negative and positive interactions in each network and a scheme of random selection with replacement.
Finally, the resulting MIN  was visualized using Cytoscape v3.6.1  with node size and edge weights set according to influence score and influence, respectively. The entire list of microbial influences in dMMR and pMMR subjects (Additional file 1: Tables S8 and S9) are too dense for direct visualization, and therefore, only a part of them are presented. More specifically, interactions below an influence of 10 in the case of both dMMR and pMMR were excluded. Unconnected nodes that had no influence were not included in the visualization.
Estimating the effect of whole-community metabolic interactions on growth suppression
i.e., the abundance-weighted sum of the metabolic influences on microbe m. When this sum is negative (as would be generally true in eubiosis), this yields a suppression score that reflects the magnitude of the negative interactions affecting microbe m. For the purpose of this calculation, we calculate this in every sample, we use anaerobic conditions and only consider microbial species that make up greater than 5% of the relative abundance of the community in at least one sample, ensuring we do not miss any microbes that may have a significant effect on the suppression score.
dMMR tumors associated with older-age and early-stage, proximal tumors
A total of 25 individuals with dMMR CRC and 58 individuals with pMMR CRC were involved in this study. Individuals with dMMR CRC were significantly older than individuals with pMMR CRC and significantly more likely to have an early-stage, proximal tumor (Table 1)—in alignment with other studies on dMMR CRC . Thus, to address potential confounding effects due to age and sample location (proximal/distal), we adjusted these variables in subsequent analyses.
Tumor MMR status explains the largest variance between microbial communities
Factors contributing to variance between microbial communities
Sample location—on/off tumor
Distinct microbial communities associated with pMMR and dMMR tumors
To validate these results, we used publicly available data from tumor and matched normal samples from 44 CRC patients . Our validation analysis showed several overlapping associations of microbial genomes with respect to dMMR and pMMR in tumor and matched normal samples (Additional file 1: Tables S7, S8). dMMR tumors were found enriched for B. fragilis (p = 0.02, FDR p = 0.37) and Fusobacterium (p = 0.03, FDR p = 0.37) while dMMR normal samples were enriched for Dorea (p = 0.03, FDR p = 0.37) and an Erysipelotrichaceae bacterium (p = 0.007, FDR p = 0.31) (Additional file 1: Figure S4). Even though these associations were not statistically significant after correcting for FDR, their trend of association overlaps with the results from the present study. Differentially abundant sOTUs between pMMR tumors versus normal included Ruminococcaceae, Faecalibacterium prausnitzii, and Bacteroides caccae, which were also differentially abundant in the present study.
Proxies for hydrogen sulfide production enriched in the dMMR CRC tumors
dMMR and pMMR tumor and normal-adjacent microbial community predicted to be highly influenced by differing Bacteroides species
pMMR microbial community predicted to enhance suppression of Bacteroides fragilis
Given this finding and the well-established links between toxigenic B. fragilis and colorectal cancer [19, 21, 24], we next looked for the presence of the B. fragilis toxin (BFT) gene in dMMR and pMMR tissue and mucosa samples. Of the 22 individuals with dMMR CRC, only one was BFT positive (5%); of 53 individuals with pMMR CRC, only five were BFT positive (9.4%). There was no significant difference in BFT presence between individuals with dMMR or pMMR CRC (Chi-squared, p = 0.477).
This study integrates tumor biology and microbial ecology in a novel and powerful approach to understanding colorectal cancer. Our results indicate that MMR status is one of the strongest predictors of microbial community variance; however, few studies [32–34], to date, include MMR status in microbial community analysis of colorectal cancer. Interestingly, we also identified several differentially abundant microbes associated with dMMR but not pMMR tumor samples including F. nucleatum, F. periodonticum, and B. fragilis. We further validated these findings in an independent cohort , which underscores the importance of including MMR status in future CRC microbiome studies. We additionally characterized the predicted and actual metabolic profiles of dMMR and pMMR individuals in relation to hydrogen sulfide production, and we generated a network of predicted interactions within the dMMR and pMMR microbial communities.
Hydrogen sulfide has been reported to both promote and inhibit colorectal cancer [54–57]. To assess the role of hydrogen sulfide within our study, we looked for sulfidogenic bacteria, predicted hydrogen sulfide production using community metabolic models, and indirectly measured hydrogen sulfide concentrations through targeted metabolomics for amino acid proxies. We found two significantly enriched hydrogen sulfide-producing Fusobacterium species and significantly increased proxies for hydrogen sulfide in dMMR tumor samples. In the microbial influence network, both Fusobacterium species exhibited zero predicted interactions—positive or negative—with other microbes in the network. Together, this suggests that these Fusobacterium species may grow unchecked by other microbes and have the potential to produce large quantities of hydrogen sulfide.
These intriguing results lead us to speculate on the relationship between Fusobacterium species, hydrogen sulfide production, and dMMR CRC. Notably, Fusobacterium species have previously been associated with hypermethylation of MLH1, MSI, BRAF mutations, and poorly differentiated tumors [12, 22]—all of which are characteristics of dMMR CRC . Hydrogen sulfide—a cytotoxic, genotoxic gas—has also been associated with CRC [54, 55], although there have been conflicting reports on its role [56, 57]. A recent report indicates that colon cancer cells may respond to hydrogen sulfide in a bell-shaped dose-dependent manner: at high concentrations, hydrogen sulfide inhibits the proliferation of cancer cells, while at lower concentrations, hydrogen sulfide can stimulate the proliferation of cancer cells [57, 58]. In dMMR, if high levels of hydrogen sulfide (and hydrogen sulfide producers) inhibit cancer proliferation, then we would expect individuals with dMMR to present with earlier-stage cancer—which is indeed the case in our cohort and other reported cohorts .
Epidemiologically, it is worth noting that dMMR CRC has also been associated with lower recurrence rates and a better prognosis . In seeming opposition to these findings are studies showing that F. nucleatum can potentiate tumorigenesis and that F. nucleatum-associated CRCs have a worse prognosis [11, 12]. However, these findings are not contradictory with our data. A more detailed examination of the effects of location (Additional file 1, Tables S5 and S6) shows that Fusobacterium is associated with the proximal colon in both dMMR and pMMR patients. This raises a subtle, but important, point. Fusobacterium-associated pMMR tumors are very likely to be found in the proximal colon alongside normal-adjacent tissue that is also enriched for Fusobacterium. Stated another way, while pMMR tumors are not especially associated with Fusobacterium, the proximal colon is. (In contrast, dMMR tumors show enrichment for Fusobacterium that goes beyond the effect of location in the colon.) When put into context with other epidemiological findings that identify right-sided (proximal) colon cancer to have lower overall survival , certain inferences come to light. Where right-sided dMMR CRCs have a relatively better prognosis, right-sided pMMR CRCs have a worse one. This would then allow us to make sense of both the overall lower survival in right-sided CRC  and the results indicating F. nucleatum-associated CRCs have a worse prognosis [11, 12]. In sum, the prognosis of F. nucleatum-associated CRCs is likely be dependent upon both location and tumor MMR status, and our study highlights the importance of evaluating these covariates simultaneously when determining tumor prognosis.
Besides Fusobacterium, B. fragilis was also found to be significantly enriched in dMMR tumor samples. Toxigenic B. fragilis has well-established and causative links to inflammation and CRC [19, 21, 24], and inflammation has been linked to hypermethylation . Our own metabolic modeling reflects a metabolic basis for higher ratios of B. fragilis in dMMR communities, and greater metabolic suppression in pMMR. We tested dMMR and pMMR tissue and mucosa samples for the presence of the B. fragilis toxin (BFT) gene but did not find a significant difference in the presence of the BFT gene between dMMR and pMMR individuals. Given these results, it is unclear what the significance of toxigenic B. fragilis is in the dMMR tumor samples.
Overall, our study demonstrates the importance and value in considering tumor biology (MMR status) and ecological interactions when evaluating microbial community data. Our work is primarily descriptive and incorporates host clinical features, microbiome, metabolome, and modeling data. While we make speculations based on these data, future prospective and mechanistic studies are needed to test these ideas. We also recognize that selecting sequenced genomes available in the database to represent 16S rRNA sOTUs cannot fully replace metagenomic sequencing given well-known strain-to-strain variation in gene content. However, these variations between strains are often largely in secondary metabolite pathways, rather than core metabolic function, which is the main target of our modeling analysis. Differences in secondary metabolite pathways (i.e., non-core genome within a species) are commonly associated with functional adaptations to various environmental niches .
Another limitation of this study is our inability to attribute a source to metabolomic data. Hydrogen sulfide and its amino acid proxies can be produced by both humans and bacteria. Thus, the enriched hydrogen sulfide we detect in dMMR tumor samples could potentially be attributed to increased hydrogen sulfide production within tumor tissue, and indeed, this has been reported . If this was the solely case here however, we might expect to see similar increases in hydrogen sulfide in pMMR tumors—most of which are later in stage than dMMR tumors. We did not see this, suggesting that it is feasible that the increased hydrogen sulfide production in dMMR tumors is coming from an exogenous (microbial) source. Notably, microbially produced hydrogen sulfide can be generated from multiple pathways including the respiration of dietary taurine and sulfate as well as the degradation of sulfomucins. The amino acid proxies we use to assess hydrogen sulfide production only capture some, but not all of these potential pathways, so we may have underestimated hydrogen sulfide production.
Finally, the field of genome-scale metabolic modeling has only recently encompassed tools for community metabolic analyses , and many of the tools [51, 52, 63] are sensitive to the underlying quality of the metabolic models [64, 65]. Models vary greatly depending on the presence and accuracy of genome annotations which will generally improve over time. Future work aimed at understanding and verifying microbial dynamics in relation to MMR status or other CRC subtypes could dramatically improve our ability to define, predict, prevent, and treat colorectal cancers.
Host–microbe interactions: Tumor MMR status strongly predicted microbial community variance and was associated with distinct microbial, metabolic, and interaction profiles. Our approach incorporating tumor MMR status, microbiome, metabolome, and modeling data allowed us unique insights into the role of hydrogen sulfide and hydrogen sulfide producers within the dMMR microbial community. Tumor biology (e.g., MMR status) and microbial ecology are inextricably linked, and it is critical that future studies account for both in order to understand and more precisely classify the many pathways to CRC.
Microbe–microbe interactions: Microbial influence networks provided in silico predictions of microbial interactions that aligned with in vivo metabolomics data: Enrichment of sulfidogenic F. nucleatum and significantly higher hydrogen sulfide production in dMMR CRC, and depletion of B. fragilis and significantly higher suppression in pMMR CRC. The validation of in vivo findings and in silico modeling provides support for a future of precision medicine tools that can accurately predict disease and the potential effects of prophylactic or therapeutic interventions on the microbiome. Microbes act within communities, and understanding and predicting these interactions will be key to developing targeted mechanisms to help prevent or treat colorectal cancer.
We would first like to thank the patients who volunteered for this study. We also thank the many other individuals who made this work possible including members of the Mayo Clinic Microbiome Laboratory, study coordinators, students, colorectal surgeons, program directors, and pathology assistants. We also specially acknowledge Donna Felmlee Devine and Caitlin Foss-Baumgard for their assistance with patient records in relation to this study.
We gratefully acknowledge the following funding sources: NIH (R01CA179243; N.C. and V.L.H. and R01CA170357; L.B.), the Mayo Clinic Center for Cell Signaling in Gastroenterology (NIDDK P30DK084567), the Mayo Clinic Metabolomics Resource Core Pilot and Feasibility Award (U24DK100469), the Fred C. Andersen Foundation (H.N. and N.C.), the Mayo Clinic Center for Individualized Medicine, The Randy Shaver Cancer Research and Community Fund (R.B.), the Minnesota Partnership for Biotechnology and Medical Genomics (R.B.), and the Alfred P. Sloan Foundation (R.B.). O.R.A thanks the financial support coming from the National Institute of Genomic Medicine (INMEGEN) to develop the computational tool used for the microbiome analysis (MICOM).
Availability of data and materials
The 16S datasets generated and/or analyzed during the current study are available in the NCBI Short Read Archive, under BioProject PRJNA445346: https://www.ncbi.nlm.nih.gov/bioproject/445346, PRJNA284355: https://www.ncbi.nlm.nih.gov/sra/PRJNA284355, and Additional files 2, 3, and 4. Metabolomics data is available here: https://figshare.com/articles/AminoAcid_Metabolomics_ForPublicSharing_xlsx/7065053
NC, HN, BAW, VLH, and PJ conceived of various aspects of this project. VLH, JY, GK, KL, LAB, DL, and HN were involved in the sample collection and processing for microbial sequencing. PJ, VLH, JC, MM, JS, CD, and OR-A were involved in the analysis of 16S and community metabolic modeling data and figure generation. X-MP, TD, JG, and VLH were involved in metabolomic sample prep and analysis. JR provided critical insights on microbial metabolite production pathways. AJF and SNT were involved in MMR status identification via immunohistochemistry. SP and RB provided analysis of the independent cohort. All authors read and approved the final manuscript.
Ethics approval and consent to participate
This study was performed with the approval of the Mayo Clinic Institutional Review Board (IRB# 14-007237 and IRB# 622-00) in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all individuals in the study.
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Flemer B, Lynch DB, Brown JMR, Jeffery IB, Ryan FJ, Claesson MJ, et al. Tumour-associated and non-tumour-associated microbiota in colorectal cancer. Gut. 2017;66:633–43. https://doi.org/10.1136/gutjnl-2015-309595.View ArticleGoogle Scholar
- Chen W, Liu F, Ling Z, Tong X, Xiang C. Human intestinal lumen and mucosa-associated microbiota in patients with colorectal cancer. PLoS One. 2012;7:e39743.View ArticleGoogle Scholar
- Zackular JP, Baxter NT, Chen GY, Schloss PD. Manipulation of the Gut Microbiota Reveals Role in Colon Tumorigenesis mSphere 2016;1:e00001–e00015. doi:https://doi.org/10.1128/mSphere.00001-15.
- Ahn J, Sinha R, Pei Z, Dominianni C, Wu J, Shi J, et al. Human gut microbiome and risk for colorectal cancer. J Natl Cancer Inst. 2013;105:1907–11.View ArticleGoogle Scholar
- Arthur JC, Gharaibeh RZ, Muhlbauer M, Perez-Chanona E, Uronis JM, McCafferty J, et al. Microbial genomic analysis reveals the essential role of inflammation in bacteria-induced colorectal cancer. Nat Commun. 2014;5:4724. https://doi.org/10.1038/ncomms5724.View ArticlePubMedPubMed CentralGoogle Scholar
- Brennan CA, Garrett WS. Gut microbiota, inflammation, and colorectal cancer. Annu Rev Microbiol. 2016;70:395–411. https://doi.org/10.1146/annurev-micro-102215-095513.View ArticlePubMedPubMed CentralGoogle Scholar
- Marchesi JR, Dutilh BE, Hall N, Peters WH, Roelofs R, Boleij A, et al. Towards the human colorectal cancer microbiome. PLoS One. 2011;6:e20447. https://doi.org/10.1371/journal.pone.0020447.View ArticlePubMedPubMed CentralGoogle Scholar
- Sobhani I, Tap J, Roudot-Thoraval F, Roperch JP, Letulle S, Langella P, et al. Microbial dysbiosis in colorectal cancer (CRC) patients. PLoS One. 2011;6:e16393.View ArticleGoogle Scholar
- Louis P, Hold GL, Flint HJ. The gut microbiota, bacterial metabolites and colorectal cancer. Nat Rev Microbiol. 2014;12:661–72.View ArticleGoogle Scholar
- Castellarin M, Warren RL, Freeman JD, Dreolini L, Krzywinski M, Strauss J, et al. Fusobacterium nucleatum infection is prevalent in human colorectal carcinoma. Genome Res. 2012;22:299–306.View ArticleGoogle Scholar
- Kostic AD, Chun E, Robertson L, Glickman JN, Gallini CA, Michaud M, et al. Fusobacterium nucleatum potentiates intestinal tumorigenesis and modulates the tumor-immune microenvironment. Cell Host Microbe. 2013;14:207–15.View ArticleGoogle Scholar
- Mima K, Nishihara R, Qian ZR, Cao Y, Sukawa Y, Nowak JA, et al. Fusobacterium nucleatum in colorectal carcinoma tissue and patient prognosis. Gut. 2016;65:1973–80.View ArticleGoogle Scholar
- Flanagan L, Schmid J, Ebert M, Soucek P, Kunicka T, Liska V, et al. Fusobacterium nucleatum associates with stages of colorectal neoplasia development, colorectal cancer and disease outcome. Eur J Clin Microbiol Infect Dis. 2014;33:1381–90.View ArticleGoogle Scholar
- Kostic AD, Gevers D, Pedamallu CS, Michaud M, Duke F, Earl AM, et al. Genomic analysis identifies association of Fusobacterium with colorectal carcinoma. Genome Res. 2012;22:292–8.View ArticleGoogle Scholar
- Drewes JL, White JR, Dejea CM, Fathi P, Iyadorai T, Vadivelu J, et al. High-resolution bacterial 16S rRNA gene profile meta-analysis and biofilm status reveal common colorectal cancer consortia. NPJ Biofilms Microbiomes. 2017;3:34.View ArticleGoogle Scholar
- Sinha R, Abu-Ali G, Vogtmann E, Fodor AA, Ren B, Amir A, et al. Assessment of variation in microbial community amplicon sequencing by the Microbiome Quality Control (MBQC) project consortium. Nat Biotechnol. 2017;35:1077–86. https://doi.org/10.1038/nbt.3981.View ArticlePubMedPubMed CentralGoogle Scholar
- Rubinstein MR, Wang X, Liu W, Hao Y, Cai G, Han YW. Fusobacterium nucleatum promotes colorectal carcinogenesis by modulating E-cadherin/β-catenin signaling via its FadA adhesin. Cell Host Microbe. 2013;14:195–206.View ArticleGoogle Scholar
- Yang Y, Weng W, Peng J, Hong L, Yang L, Toiyama Y, et al. Fusobacterium nucleatum increases proliferation of colorectal cancer cells and tumor development in mice by activating TLR4 signaling to NFκB, upregulating expression of microRNA-21. Gastroenterology. 2016;152:851–866.e24. https://doi.org/10.1053/j.gastro.2016.11.018.View ArticlePubMedPubMed CentralGoogle Scholar
- Housseau F, Sears CL. Enterotoxigenic Bacteroides fragilis (ETBF)-mediated colitis in min (Apc+/−) mice: a human commensal-based murine model of colon carcinogenesis. Cell Cycle. 2010;9:3–5.View ArticleGoogle Scholar
- Purcell RV, Pearson J, Aitchison A, Dixon L, Frizelle FA, Keenan JI. Colonization with enterotoxigenic Bacteroides fragilis is associated with early-stage colorectal neoplasia. PLoS One. 2017;12:e0171602.View ArticleGoogle Scholar
- Chung L, Thiele Orberg E, Geis AL, Chan JL, Fu K, DeStefano Shields CE, et al. Bacteroides fragilis toxin coordinates a pro-carcinogenic inflammatory cascade via targeting of colonic epithelial cells. Cell Host Microbe. 2018;23:203–214.e5. https://doi.org/10.1016/j.chom.2018.01.007.View ArticlePubMedPubMed CentralGoogle Scholar
- Koi M, Okita Y, Carethers JM. Fusobacterium nucleatum infection in colorectal cancer: linking inflammation, DNA mismatch repair and genetic and epigenetic alterations. J Anus, Rectum Colon. 2018;2:37–46. https://doi.org/10.23922/jarc.2017-055.View ArticleGoogle Scholar
- Abed J, Emgård JEM, Zamir G, Faroja M, Almogy G, Grenov A, et al. Fap2 mediates fusobacterium nucleatum colorectal adenocarcinoma enrichment by binding to tumor-expressed Gal-GalNAc. Cell Host Microbe. 2016;20:215–25.View ArticleGoogle Scholar
- Dejea CM, Fathi P, Craig JM, Boleij A, Taddese R, Geis AL, et al. Patients with familial adenomatous polyposis harbor colonic biofilms containing tumorigenic bacteria. Science (80- ). 2018;359:592–7.View ArticleGoogle Scholar
- Richman S. Deficient mismatch repair: read all about it (review). Int J Oncol. 2015;47:1189–202.View ArticleGoogle Scholar
- French AJ, Sargent DJ, Burgart LJ, Foster NR, Kabat BF, Goldberg R, et al. Prognostic significance of defective mismatch repair and BRAF V600E in patients with colon cancer. Clin Cancer Res. 2008;14:3408–15. https://doi.org/10.1158/1078-0432.CCR-07-1489.View ArticlePubMedPubMed CentralGoogle Scholar
- Guinney J, Dienstmann R, Wang X, De Reyniès A, Schlicker A, Soneson C, et al. The consensus molecular subtypes of colorectal cancer. Nat Med. 2015;21:1350.View ArticleGoogle Scholar
- Mårtensson A, Oberg A, Jung A, Cederquist K, Stenling R, Palmqvist R. Beta-catenin expression in relation to genetic instability and prognosis in colorectal cancer. Oncol Rep. 2007;17:447–52.PubMedGoogle Scholar
- Morkel M, Riemer P, Bläker H, Sers C. Similar but different: distinct roles for KRAS and BRAF oncogenes in colorectal cancer development and therapy resistance. Oncotarget. 2015;6:20785–800. https://doi.org/10.18632/oncotarget.4750.View ArticlePubMedPubMed CentralGoogle Scholar
- Sweetser S, Jones A, Smyrk TC, Sinicrope FA. Sessile serrated polyps are precursors of colon carcinomas with deficient DNA mismatch repair. Clin Gastroenterol Hepatol. 2016;14:1056–9.View ArticleGoogle Scholar
- Dejea CM, Wick EC, Hechenbleikner EM, White JR, Mark Welch JL, Rossetti BJ, et al. Microbiota organization is a distinct feature of proximal colorectal cancers. Proc Natl Acad Sci. 2014;111:18321–6. https://doi.org/10.1073/pnas.1406199111.View ArticlePubMedGoogle Scholar
- Purcell RV, Visnovska M, Biggs PJ, Schmeier S, Frizelle FA. Distinct gut microbiome patterns associate with consensus molecular subtypes of colorectal cancer. Sci Rep. 2017;7:11590.View ArticleGoogle Scholar
- Belcheva A, Irrazabal T, Robertson SJ, Streutker C, Maughan H, Rubino S, et al. Gut microbial metabolism drives transformation of msh2-deficient colon epithelial cells. Cell. 2014;158:288–99.View ArticleGoogle Scholar
- Lennard KS, Goosen RW, Blackburn JM. Bacterially-associated transcriptional remodelling in a distinct genomic subtype of colorectal cancer provides a plausible molecular basis for disease development. PLoS One. 2016;11:e0166282. https://doi.org/10.1371/journal.pone.0166282.View ArticlePubMedPubMed CentralGoogle Scholar
- Hale VL, Jeraldo P, Mundy M, Yao J, Keeney G, Scott N, et al. Synthesis of multi-omic data and community metabolic models reveals insights into the role of hydrogen sulfide in colon cancer. Methods. 2018. https://doi.org/10.1016/j.ymeth.2018.04.024.View ArticleGoogle Scholar
- Chen J, Ryu E, Hathcock M, Ballman K, Chia N, Olson JE, et al. Impact of demographics on human gut microbial diversity in a US Midwest population. PeerJ. 2016;4:e1514. https://doi.org/10.7717/peerj.1514.View ArticlePubMedPubMed CentralGoogle Scholar
- Hale VL, Chen J, Johnson S, Harrington SC, Yab TC, Smyrk TC, et al. Shifts in the fecal microbiota associated with adenomatous polyps. Cancer Epidemiol Prev Biomarkers. 2017;26:85–94.View ArticleGoogle Scholar
- Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJA, Holmes SP. DADA2: high-resolution sample inference from Illumina amplicon data. Nat Methods. 2016;13:581–3.View ArticleGoogle Scholar
- Wang Q, Garrity GM, Tiedje JM, Cole JR. Naïve Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Appl Environ Microbiol. 2007;73:5261–7.View ArticleGoogle Scholar
- Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P, et al. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res. 2013;41:D590–6.View ArticleGoogle Scholar
- Kopylova E, Noé L, Touzet H. SortMeRNA: fast and accurate filtering of ribosomal RNAs in metatranscriptomic data. Bioinformatics. 2012;28:3211–7.View ArticleGoogle Scholar
- Nawrocki EP, Eddy SR. Infernal 1.1: 100-fold faster RNA homology searches. Bioinformatics. 2013;29:2933–5.View ArticleGoogle Scholar
- Price MN, Dehal PS, Arkin AP. FastTree 2 - approximately maximum-likelihood trees for large alignments. PLoS One. 2010;5:e9490.View ArticleGoogle Scholar
- Lozupone C, Knight R. UniFrac: a new phylogenetic method for comparing microbial communities. Appl Environ Microbiol. 2005;71:8228–35. https://doi.org/10.1128/AEM.71.12.8228-8235.2005.View ArticlePubMedPubMed CentralGoogle Scholar
- McMurdie PJ, Holmes S. Phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One. 2013;8:e61217. https://doi.org/10.1371/journal.pone.0061217.View ArticlePubMedPubMed CentralGoogle Scholar
- Oksanen J, Kindt R, Legendre P, O’Hara B, Simpson GL, Solymos PM, et al. The vegan package. Community Ecol Packag. 2008:190. https://doi.org/10.4135/9781412971874.n145.
- Bolker BM, Brooks ME, Clark CJ, Geange SW, Poulsen JR, Stevens MHH, et al. Generalized linear mixed models: a practical guide for ecology and evolution. Trends Ecol Evol. 2009;24:127–35. https://doi.org/10.1016/j.tree.2008.10.008.View ArticlePubMedGoogle Scholar
- Brooks ME, Kristensen K, van Benthem KJ, Magnusson A, Berg CW, Nielsen A, et al. glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. R J 2017;9:378–400.Google Scholar
- Burns MB, Montassier E, Abrahante J, Priya S, Niccum DE, Khoruts A, et al. Colorectal cancer mutational profiles correlate with defined microbial communities in the tumor microenvironment. PLoS Genet. 2018;14:090795. https://doi.org/10.1371/journal.pgen.1007376.View ArticleGoogle Scholar
- Orth JD, Thiele I, Palsson BØ. What is flux balance analysis? Nat Biotechnol. 2010;28:245–8. https://doi.org/10.1038/nbt.1614.View ArticlePubMedPubMed CentralGoogle Scholar
- Mendes-Soares H, Mundy M, Soares LM, Chia N. MMinte: an application for predicting metabolic interactions among the microbial species in a community. BMC Bioinformatics. 2016;17:343. https://doi.org/10.1186/s12859-016-1230-3.View ArticlePubMedPubMed CentralGoogle Scholar
- Sung J, Kim S, Cabatbat JJT, Jang S, Jin Y-S, Jung GY, et al. Global metabolic interaction network of the human gut microbiota for context-specific community-scale analysis. Nat Commun. 2017;8:15393. https://doi.org/10.1038/ncomms15393.View ArticlePubMedPubMed CentralGoogle Scholar
- Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13:2498–504.View ArticleGoogle Scholar
- Attene-Ramos MS, Nava GM, Muellner MG, Wagner ED, Plewa MJ, Gaskins HR. DNA damage and toxicogenomic analyses of hydrogen sulfide in human intestinal epithelial FHs 74 int cells. Environ Mol Mutagen. 2010;51:304–14.PubMedGoogle Scholar
- Wolf PG, Parthasarathy G, Chen J, O’Connor HM, Chia N, Bharucha AE, et al. Assessing the colonic microbiome, hydrogenogenic and hydrogenotrophic genes, transit and breath methane in constipation. Neurogastroenterol Motil. 2017;29:1–9.View ArticleGoogle Scholar
- Lee ZW, Zhou J, Chen CS, Zhao Y, Tan CH, Li L, et al. The slow-releasing hydrogen sulfide donor, GYY4137, exhibits novel anti-cancer effects in vitro and in vivo. PLoS One. 2011;6:e21077.View ArticleGoogle Scholar
- Hellmich MR, Coletta C, Chao C, Szabo C. The therapeutic potential of cystathionine β-synthetase/hydrogen sulfide inhibition in cancer. Antioxid Redox Signal. 2015;22:424–48. https://doi.org/10.1089/ars.2014.5933.View ArticlePubMedPubMed CentralGoogle Scholar
- Cai W, Wang M, Ju L, Wang C, Zhu Y. Hydrogen sulfide induces human colon cancer cell proliferation: role of Akt, ERK and p21. Cell Biol Int. 2010;34:565–72. https://doi.org/10.1042/CBI20090368.View ArticlePubMedGoogle Scholar
- Lim DR, Kuk JK, Kim T, Shin EJ. Comparison of oncological outcomes of right-sided colon cancer versus left-sided colon cancer after curative resection. Med (United States). 2017;96:e8241.Google Scholar
- Maiuri AR, Peng M, Sriramkumar S, Kamplain CM, DeStefano Shields CE, Sears CL, et al. Mismatch repair proteins initiate epigenetic alterations during inflammation-driven tumorigenesis. Cancer Res. 2017;77:3467–78.View ArticleGoogle Scholar
- Lind AL, Wisecaver JH, Lameiras C, Wiemann P, Palmer JM, Keller NP, et al. Drivers of genetic diversity in secondary metabolic gene clusters within a fungal species. PLoS Biol. 2017;15:e2003583.View ArticleGoogle Scholar
- Magnúsdóttir S, Thiele I. Modeling metabolism of the human gut microbiome. Curr Opin Biotechnol. 2018;51:90–6.View ArticleGoogle Scholar
- Benedict MN, Mundy MB, Henry CS, Chia N, Price ND. Likelihood-based gene annotations for gap filling and quality assessment in genome-scale metabolic models. PLoS Comput Biol. 2014;10:e1003882.View ArticleGoogle Scholar
- Magnúsdóttir S, Heinken A, Kutt L, Ravcheev DA, Bauer E, Noronha A, et al. Generation of genome-scale metabolic reconstructions for 773 members of the human gut microbiota. Nat Biotechnol. 2017;35:81–9.View ArticleGoogle Scholar
- Wattam AR, Abraham D, Dalay O, Disz TL, Driscoll T, Gabbard JL, et al. PATRIC, the bacterial bioinformatics database and analysis resource. Nucleic Acids Res. 2014;42:D581–91.View ArticleGoogle Scholar