Microbiomic subprofiles and MDR1 promoter methylation in head and neck squamous cell carcinoma

Abstract

Clinical observations and epidemiologic studies suggest that the incidence of head and neck squamous cell carcinoma (HNSCC) correlates with dental hygiene, implying a role for bacteria-induced inflammation in its pathogenesis. Here we begin to explore the pilot hypothesis that specific microbial populations may contribute to HNSCC pathogenesis via epigenetic modifications in inflammatory- and HNSCC-associated genes. Microbiomic profiling by 16S rRNA sequencing of matched tumor and adjacent normal tissue specimens in 42 individuals with HNSCC demonstrate a significant association of specific bacterial subpopulations with HNSCC over normal tissue (P < 0.01). Furthermore, microbial populations can separate tumors by tobacco status (P < 0.008), but not by alcohol status (P = 0.41). If our subhypothesis regarding a mechanistic link from microorganism to carcinogenesis via inflammation and consequent aberrant DNA methylation is correct, then we should see hypermethylation of relevant genes associate with specific microbiomic profiles. Methylation analysis in four genes (MDR1, IL8, RARB, TGFBR2) previously linked to HNSCC or inflammation shows significantly increased methylation in tumor samples compared with normal oral mucosa. Of these, MDR1 promoter methylation associates with specific microbiomic profiles in tumor over normal mucosa. Additionally, we report that MDR1 methylation correlates with regional nodal metastases in the context of two specific bacterial subpopulations, Enterobacteriaceae and Tenericutes (P < 0.001 for each). These associations may lead to a different, and potentially more comprehensive, perspective on the pathogenesis of HNSCC, and support further exploration of mechanistic linkage and, if so, novel therapeutic strategies such as demethylating agents and probiotic adjuncts, particularly for patients with advanced or refractory disease.

Figure 1.

Scatter plot of methylation status of IL8, TGFBR2, MDR1 and RARB gene promoters. Each dot represents the presence of promoter hypermethylation of that gene in that particular tumor sample compared with its corresponding normal oral mucosal tissue. There are 42 paired tumor and normal mucosa samples analyzed for these four genes.

Figure 2.

Microbial subpopulational content at the phylum level of HNSCC and matched normal mucosa. Each bar represents a sample, and the bars are split showing the microbial content quantitatively of that sample at the phylum level. There are 42 matched HNSCC (top) and normal oral mucosal (bottom) samples where a total of 7 microbial phyla identified.

Figure 3.

Microbiomic content of the paired normal and tumor samples. (A) Biological classifications of the bacteria are shown. The baseline is the kingdom bacteria, and each level represents the next taxonomic rank (phylum, class, order, family, genus). (B) rRNA-based count of each microbial population in normal and tumor samples are represented by boxplots. Separate circles represent the outliers. (C) Site-specific microbial classifications are shown. Distribution of bacteria in HNSCC sites is indicated by columns of triangles to the right of the taxonomy tree, where the left triangle represents tumor samples and the right represents the corresponding normal mucosa samples.

Figure 4.

MANCOVA of bacterial subpopulations, DNA methylation status and clinical and exposure information in HNSCC and matched normal tissue. The heat map depicts –log(P-values), i.e. orange and darker are significant at least at the P < 0.05 level. The bacterial populations detected are listed on the x-axis, and the models that are considered in the analysis are listed on the y-axis. Models with variables MDR1 methylation, RARB methylation, sex, age (dichotomized at cut-off 62.5 years), clinical tumor stage (cT, dichotomized at stages 2 or more) and clinical nodal involvement (cN, dichotomized at 0 versus 1 or more). The models that were not significant (e.g. IL8) are not shown.

Figure 5.

MANOVA correlates tobacco use with microbiome but not alcohol use. The plots depict MANOVA results. The two axes are the first two canonical variables from the analysis where these variables are calculated based on linear combinations of the mean-centered original values of the microbiome in the samples analyzed. (A) The scatter plot of tobacco use canonical variables shows more separation between groups (P < 0.0036), whereas (B) alcohol use scatter plot shows two clouds that have a larger overlap (P = 0.41).