The intratumor microbiome is associated with microsatellite instability

Abstract

Intratumoral microbes may have multifunctional roles in carcinogenesis. Microsatellite instability (MSI) is associated with higher tumor immunity and mutational burden. Using whole transcriptome and whole genome sequencing microbial abundance data, we investigated associations of intratumoral microbes with MSI, survival, and MSI-relevant tumor molecular characteristics across multiple cancer types including colorectal cancer (CRC), stomach adenocarcinoma, and endometrial carcinoma. Among 451 CRC patients, our key finding was strong associations of multiple CRC-associated genera, including Dialister and Casatella, with MSI. Dialister and Casatella abundance was associated with improved overall survival (hazard ratio_mortality = 0.56, 95% confidence interval = 0.34 to 0.92, and hazard ratio_mortality = 0.44, 95% confidence interval = 0.27 to 0.72), respectively, comparing higher relative to lower quantiles. Multiple intratumor microbes were associated with immune genes and tumor mutational burden. Diversity of oral cavity–originating microbes was also associated with MSI among CRC and stomach adenocarcinoma patients. Overall, our findings suggest the intratumor microbiota may differ by MSI status and play a role in influencing the tumor microenvironment.

Accumulating evidence supports that solid tumors contain a distinctive tissue-specific microbiome (1). Microsatellite instability (MSI), caused by the inactivation of mismatch repair genes, is a key cancer hallmark for colorectal, gastric, and endometrial cancers. Tumors with MSI-high (MSI-H) phenotype tend to exhibit dramatically higher mutation burden and display clinically significantly better response to anti–programmed cell death–1 therapy than MSI-low (MSI-L) tumors (2). It was previously found that intratumor bacteria, specifically Fusobacterium, might differ by MSI status in colorectal cancer (CRC) (3-5); however, few studies investigated differences in abundance of other bacteria and differences across cancer types. Further, converging evidence from preclinical and human studies suggests that the microbiome and immune system interact to play a significant role in influencing the overall tumor microenvironment (6). Additional evidence is needed to support associations of the intratumor microbiome with important clinically relevant characteristics such as immunity and tumor mutational burden (TMB). Therefore, we systematically compared intratumor microbiota between MSI-H and MSI-L tumors in CRC and then considered these intratumor microbes in context of survival, the tumor immune microenvironment, and the TMB.

Pan-cancer tumor microbial abundances in The Cancer Genome Atlas (TCGA) were generated using whole transcriptome sequencing (RNA-sequencing) and whole genome sequencing data, as described previously (1). We calculated alpha diversity overall and among oral genera [based on accumulating evidence that oral cavity–originating bacteria are associated with presence of CRC (7)] using Phyloseq. To account for differences in sequencing depth, the count data were rarefied.

To characterize MSI, immune genes, and TMB, we downloaded the standardized, normalized, batch-corrected and platform-corrected RNA-sequencing expression and clinical data matrices generated by the PanCancer Atlas consortium, described elsewhere (https://gdc.cancer.gov/about-data/publications/pancanatlas). We used 5 well-known immune genes (8-10) and an immune cytolytic activity score (11).

The most extensive research has been conducted on microbiome-CRC associations. Therefore, we first used a hypothesis-driven approach to focus on bacteria that are extensively supported to be CRC associated, as derived from multiple studies (see bacteria in Figure 1) (12). Wilcox rank sum tests and a zero-inflated quantile approach were used to calculate P values comparing alpha diversity (Shannon, InvSimpson, and Chao1) and normalized relative abundance between MSI-H and MSI-L patients.

Figure 1.

Associations of intratumor microbes with microsatellite instability (MSI), survival, and tumor immune gene expression among colorectal cancer patients in The Cancer Genome Atlas. A) Differences in relative abundance of a priori–selected genera by microsatellite instability status with Wilcox rank sum (WRS) test and zero-inflated quantile approach (ZINQ) P values. B) Differences in genera of family Clostridiales by microsatellite instability status with WRS test and ZINQ P values. C) Among MSI-low colorectal cancer patients, log-rank tests to compare survival between higher and lower quantiles (based on the median) of Dialister and Casatella relative abundance. D) Association of Parvimonas abundance with immune gene expressions (CD3E and CD274) using WRS tests as an example. E) Mean differences of genus abundance between high and low immune gene expression groups with Wilcox rank sum test calculated P values. MSI-high = microsatellite instability high; MSI-low = microsatellite instability low; PD-L1 = programmed death-ligand 1.

We estimated associations of microbial abundances (dichotomized as higher relative to lower quantiles based on the median) with survival (overall and stratified by stage and MSI) using log-rank tests and Cox proportional hazards models. We used Wilcox rank sum tests to compare expressions of microbial relative abundance (continuous) across immune gene abundance (dichotomized to low vs high). We compared microbial relative abundance across MSI status using Wilcox rank sum tests among CRC, uterine corpus endometrial carcinoma, and stomach adenocarcinoma cancer types (cancers with a high proportion of MSI-H subtypes) using an exploratory approach of all microbes, adjusting for multiple testing using the false discovery rate method (alpha < 0.05 considered statistically significant). Across all cancers included in TCGA, microbial relative abundances were compared between first and fourth quartiles of TMB using Wilcox rank sum tests. In all multivariable regression models, we adjusted for stage, race, sex, and age.

We found strong differences in multiple genera across MSI-H and MSI-L CRC patients (Figure 1, A and B; patient characteristics in Supplementary Table 1, available online), even after adjustment for potential confounders (Supplementary Table 2, available online). Notably, multiple bacteria that primarily reside in the oral cavity and that were consistently associated with CRC, such as Dialister and Prevotella, were more abundant, on average, among MSI-H CRC tumors. Dialister and Casatella were also statistically significantly associated with improved CRC survival. For example, among all CRC patients with higher relative to lower abundance of Dialister and Casatella, the hazard ratios (HRs) were 0.56 (95% confidence interval [CI] = 0.34 to 0.92) and 0.44 (95% CI = 0.27 to 0.72), respectively. Associations among MSI-L CRC patients (who have poorer prognosis than MSI-H) were stronger: the hazard ratios were 0.49 (95% CI = 0.28 to 0.84) and 0.36 (95% CI = 0.21 to 0.64), respectively (Figure 1, C; Supplementary Table 3, available online). Fusobacterium, which has been consistently associated with CRC aggressiveness (13), was not abundant in our data; however, multiple genera within family Fusobacteriaceae were more abundant among MSI-H CRCs (Supplementary Figure 1, available online). Though many prior studies focused on Fusobacterium nucleatum only (3-5), our findings agree with the plausibility of microbe-MSI-survival associations.

We found that MSI-H abundant microbes were also associated with higher abundance of immune genes. Dialister was positively associated with CD3E (a marker of overall tumor-infiltrating lymphocytes) and CD8E (a marker of cytotoxic T cells), both which have favorable prognostic value (Figure 1, E; Supplementary Table 4, available online). Previously, Fusobacterium nucleatum was also found to be associated with immune markers but conversely in an immunosuppressive manner (eg, inversely associated with CD3) (14). Experimental animal studies and human epidemiologic studies support the idea that gut microbes may promote or inhibit tumor growth and influence cancer therapy response via direct and indirect modulation of innate and adaptive immune responses (14-16). We also found that the intratumor microbiome was associated with TMB, which has been associated with better response among patients treated with immune checkpoint inhibitors (Supplementary Figure 2, available online) (17).

In an exploratory analysis of microbial abundances among CRC (n = 451), endometrial carcinoma (n = 528), and stomach adenocarcinoma (n = 440) (Figure 2, A), we found 11 common microbes that statistically significantly and commonly differed across MSI status in all 3 cancer types. Oral genera diversity based on Shannon diversity was higher among MSI-H CRC patients (P = .012; Figure 2, B and C). Among those with stomach cancer, oral genera alpha diversity was higher for the Shannon and inverse Simpson index (P < .0001). The findings were generally similar after adjusting for potential confounders (Supplementary Table 5, available online).

Figure 2.

Associations of intratumor microbes and their diversity with microsatellite instability in The Cancer Genome Atlas among microsatellite instability (MSI)–enriched cancers. The cutoff for MSI-high (MSI-H) was based on MSI Mantis Score of 0.35 (cervical, breast, and bladder were below threshold). A) A comparison of common genera abundance by MSI status using Wilcox rank sum tests among colorectal, endometrial, and stomach cancers. P values are adjusted using false discovery rate method. Eleven genera were statistically significantly different between MSI across 3 cancers. Three of these are viruses; others are bacteria. B) Comparison of alpha diversity (Shannon, Inverse Simpson, and Chao1) calculated based on all bacterial genera across MSI status using Wilcox rank sum tests. C) Comparison of alpha diversity (Shannon, Inverse Simpson, and Chao1) calculated based on oral-originating bacterial genera using Wilcox rank sum tests across MSI. The count data were rarefied to 1 458 979 reads (the minimum) for overall alpha diversity analyses and 66 566 for oral alpha diversity for colon and rectal patients: 870 535 reads (the minimum) for overall bacteria alpha diversity analyses and 72677 for the oral alpha diversity analyses for uterine corpus endometrial carcinoma; 1 819 037 reads (the minimum) for overall bacteria alpha diversity analyses and 131 404 for the oral alpha diversity analyses in stomach adenocarcinoma; 1 905 000 reads (the minimum) for overall bacteria alpha diversity and 83 312 for the oral alpha diversity analyses in cervical squamous cell carcinoma and endocervical adenocarcinoma; 1 551 086 reads (the minimum) for overall bacteria alpha diversity and 91 127 for oral alpha diversity in breast invasive carcinoma; 1 644 614 reads (the minimum) for overall bacteria alpha diversity and 87 663 for the oral diversity analyses in bladder urothelial carcinoma.

There are many limitations of the microbiome data generated in TCGA that have mostly previously been described by Poore et al. (1) and Hermida et al. (18) such as the poly(A) tail capture. Diverse sets of molecular and genetic signatures may have been harbored within a single tumor as found by Liu et al. (19), who studied multiple biopsy sections of the same CRC tumor finding intraneoplasia heterogeneity in microbes. Finally, contamination is of particular concern for tissue samples and potentially influenced our detection of Fusobacterium in our study; however, we expect that contamination is nondifferential in nature. Poore et al. (1) made huge efforts to decontaminate and validate the data described herein. As suggested by others, detailed protocols need to be developed for future studies of the intratumor microbiome.

Taken together, we found the intratumor microbiome may be associated with MSI and its associated features, such as immunity, survival, and mutational burden. These associations remain incompletely understood and require mechanistic studies among diverse CRC subtypes to further elucidate microbial roles in carcinogenesis. A broader understanding of the currently unknown mechanisms underlying associations of tumor microbiomes, host immune responses, and tumor progression may facilitate clinical decision making.

Data availability

The data presented herein are publicly available at: https://gdc.cancer.gov/about-data/publications/pancanatlas and https://www.cbioportal.org/datasets.

Author contributions

Doratha A Byrd, PhD, MPH (Conceptualization; Formal analysis; Investigation; Methodology; Writing–original draft; Writing–review & editing), Wenyi Fan, Phd (Formal analysis; Writing–original draft; Writing–review & editing), K. Leigh Greathouse, PhD MPH (Methodology; Writing–original draft; Writing–review & editing), Michael C. Wu, PhD (Methodology; Writing–original draft; Writing–review & editing), Hao Xie, MD (Methodology; Writing–original draft; Writing–review & editing), Xuefeng Wang, PhD (Conceptualization; Formal analysis; Investigation; Methodology; Supervision; Writing–original draft; Writing–review & editing).

Funding

National Cancer Institute and National Human Genome Research Institute.

Conflicts of interest

The authors declare no potential conflicts of interest.