Periodontal pathogens are a risk factor of oral cavity squamous cell carcinoma, independent of tobacco and alcohol and human papillomavirus

Abstract

Over the past decade, there has been a change in the epidemiology of oral cavity squamous cell cancer (OC-SCC). Many new cases of OC-SCC lack the recognized risk factors of smoking, alcohol and human papilloma virus. The aim of this study was to determine if the oral microbiome may be associated with OC-SCC in nonsmoking HPV negative patients. We compared the oral microbiome of HPV-negative nonsmoker OC-SCC(n = 18), premalignant lesions(PML) (n = 8) and normal control patients (n = 12). Their oral microbiome was sampled by oral wash and defined by 16S rRNA gene sequencing. We report that the periodontal pathogens Fusobacterium, Prevotella, Alloprevotella were enriched while commensal Streptococcus depleted in OC-SCC. Based on the four genera plus a marker genus Veillonella for PML, we classified the oral microbiome into two types. Gene/pathway analysis revealed a progressive increase of genes encoding HSP90 and ligands for TLRs 1, 2 and 4 along the controls→PML → OC-SCC progression sequence. Our findings suggest an association between periodontal pathogens and OC-SCC in non smoking HPV negative patients.

Keywords: Fusobacterium; streptococcus; leukoplakia; microbiome; nonsmoking; oral cavity squamous cell carcinoma; periodontal pathogens; risk factor.

Figure 1.. Selection of bacterial genera as markers to differentiate among and between negative controls, oral PML and oral cavity squamous cell carcinoma (OC-SCC).

Initial analysis was done with all genera by principal coordinate analysis for the global differences of microbiome with weighted UniFrac distances matrices. Comparisons were made among negative control, oral PML, and OC-SCC, between negative controls and OC-SCC (A). P values were calculated by Adonis test. Selective analyses were based on relative abundance of five genera (Prevotella, Veillonella, Fusobacterium, Alloprevotella, and Streptococcus) selected by Random Forest coupled with Support Vector Machine and statistical tests with Kruskal Wallis test for difference among three groups and Mann-Whitney test between two groups (B), Jonckheere trend test for trend (C) and principal coordinate analysis with weighted UniFrac distances matrices among the five genera (D). P value is marked with * if <0.1, ** if <0.05, or *** if <0.01 after FDR-adjustment.

Figure 2.. Classification of bacterial communities into microbiome types and their correlation with diseases.

The bacterial communities sampled were classified into two microbiome types based on the relative abundance of five genera informative of disease status as determined by statistical analyses, including Prevotella, Veillonella, Fusobacterium, Alloprevotella, and Streptococcus. A Euclidean metrics between all possible pairs of samples were calculated and used to cluster the samples by Ward’s linkage algorithm (A). Major clusters generated were termed periodontal pathogen-low (PPL) and periodontal pathogen-high (PPH) types of microbiome, respectively. Separation between the two types of microbiome was further analyzed by principal coordinate analysis with weighted UniFrac distances matrices with p values calculated by Adonis (B). Correlation of the microbiome types with diseases was evaluated using taxonomic analysis by Mann-Whitney test (C) and two tailed Fisher exact test (D). P value is marked with * if <0.1, ** if <0.05, or *** if <0.01 after FDR-adjustment.

Figure 3.. Co-occurrence/avoidance network among major genera in the oral microbiome and their correlations with disease states.

The four large nodes are genera that directly correlated with the disease states and the termini show genera that indirectly correlated with the disease states through the four genera (A). Connecting lines represent strong (r>0.3) and significant (p<0.05) correlations. Red lines indicate a positive (collaborative) correlation while blue lines indicate a negative (inhibitive) correlation. The thickness of a connection line is in proportion to the r value in Spearman’s correlation coefficient. Known periodontal pathogens are bold-faced. Difference among negative controls, PML, and OC-SCC in the combined relative abundance of all periodontal pathogens were analyzed by with Kruskal Wallis test for difference among the three case control groups and Mann-Whitney test between two groups (B). Trend of changes in the combined relative abundance of all 14 periodontal pathogens found in the samples along the negative controls(NC)→PML→OC-SCC sequence was analyzed by Jonckheere trend test.

Figure 4.. Progressive enrichment of proinflammatory genes/pathways along the sequence of negative controls→PML→OC-SCC and enrichment of Capnocytophaga in patients with OC-SCC.

Microbial gene contents in the samples were deduced from 16S rRNA gene profiles using PICRUSt (A). A gene shown is specified by name and KEGG ID. Relative abundance of each gene was stratified to source genera by color codes. Difference between patients with and without recurrence of OC-SCC after tumor resection was shown in relation to the relative abundance of Capnocytophaga (B). P value is marked with * if <0.1 or ** if <0.05 after FDR-adjustment.