Tonsillar microbiome diversity in healthy individuals
As a first step for the assessment of microbial alterations in tonsil cancer, we have analyzed the tonsillar microbiome of a well-defined control cohort of patients that underwent tonsillectomy due to OSA. In total, 21 OSA patients have been enrolled between May 2017 and May 2020. 16S rRNA gene amplicon sequencing has been performed on the samples of the first 14 recruited patients (Table 1), while recruitment continued to reach the final cohort size (Supplementary Table 1). To assess whether the crypt microbiome represents the global tonsillar bacterial composition, we acquired two punch biopsies each from the surface epithelium, the lymphoid tissue parenchyma and the crypts (Fig. 1A). Data from the same patient and sampling site were regarded as technical replicates and were merged in downstream bioinformatics analyses. In total, 160 biopsies have been analyzed in the OSA cohort with 130 samples remaining after quality control. We found a uniform within-sample (α-)diversity across different collection sites with the mean number of observed species and Chao1 species richness estimators showing no significant differences between the three sampling sites (Fig. 1B). Likewise, both Shannon and Simpson indices, which denote species richness and evenness, were not significantly different (Fig. 1C). Principle coordinate analysis using Bray Curtis (Fig. 1D), UNIFRAC (Fig. 1E) and weighted UNIFRAC (Fig. 1F) distance matrices did not reveal significant differences in β-diversity between the tonsillar compartments. Bacteroidetes, Fusobacteria, and Firmicutes were the dominant phyla in the three different anatomical locations (Fig. 1G). The abundance of the various phyla did not vary significantly between the sampling sites suggesting that microbial colonization occurs in a homogenous fashion across the different tonsillar niches.
Table 1
Characteristics of patients with 16S rRNA gene amplicon-based analysis of the tonsillar microbiome in the tonsillar carcinoma and obstructive sleep apnea cohorts.
Cohort
|
Main diagnosisa
|
N
|
Age (mean ± SD)
|
Male/female ratio
|
HPV p16 positivity
|
HR-HPV DNA positivity
|
1
|
Obstructive sleep apnea
|
14
|
38.9 ± 11.5
|
13:1
|
0/14
|
0/14
|
2
|
SCC of the tonsil
|
18
|
66.9 ± 12.8
|
2.6:1
|
16/18
|
14/16b
|
a Main clinical diagnosis as documented in the patient information system; SCC, squamous cell carcinoma. |
b Data of two patients could not be obtained. |
To further assess to what extent microbial community composition in tonsillar niches of individual patients is conserved, we compared the microbiome in ipsi- and contralateral tonsils of OSA patients (Fig. 2A). Wilcoxon signed-rank testing did not indicate significant differences in species richness (Fig. 2B) or richness and evenness (Fig. 2C) in the microbiome of opposing tonsils. Analysis of β-diversity revealed a significant dissimilarity for paired Permanova testing based on the Bray Curtis distance (p = 0.0275) (Fig. 2D), but not for UNIFRAC (Fig. 2E) or weighted UNIFRAC distances (Fig. 2F). The assessment of individual patients in these analyses highlights the substantially higher variance between individual patients compared to the differences between ipsi- and contralateral tonsils (Fig. 2D-F). In sum, these data indicate that tonsils of OSA patients harbor diverse bacterial communities and that the microbial composition in tonsillar niches of individual patients is similar across the oropharynx.
Tonsillar crypt microbiome composition in tonsil cancer
Tissue biopsies of 18 patients suffering from squamous cell carcinoma of the tonsils have been analyzed by 16S rRNA gene amplicon sequencing with almost 90% of the patients being HR-HPV-positive (Table 1). In addition to the HPV-status, age, sex, TNM tumor staging, nicotine and alcohol exposure and secondary carcinomas were documented (Supplementary Table 2). Two punch biopsies from the contralateral crypt were collected during the surgery and were included in the analysis (Fig. 3A). The assessment of four α-diversity metrics did not reveal statistically significant differences between the crypt microbiome of tumor-affected and contralateral tonsils (Fig. 3B and C). Principle coordinate analysis using Bray Curtis, UNIFRAC and weighted UNIFRAC estimators did not indicate statistically significant dissimilarities in the microbial community composition of tumor-affected vs. contralateral tonsils (Fig. 3D-F). Extended analyses indicated that the severity of the disease, here assessed as pathological tumor stage, had a significant effect on the β-diversity of the microbial communities in the tumor-affected tonsils (Supplementary Fig. 1A-C). At phylum level, tumor-affected tonsils exhibited elevated (p = 0.056) relative abundance of Firmicutes (Fig. 3G). The genera Fusobacterim, Prevotella, Prevoella_7 and Veillonella displayed the highest relative abundance in tonsillar crypts of tumor patients (Fig. 3H). These data suggest that the growth of HR-HPV-associated tonsil cancer alters the microenvironment in tonsillar crypts leading to changes in microbial communities both on phylum and genus level.
Distinct tonsillar crypt microbiome composition in tonsil cancer
Next, we compared the crypt microbiome of ipsi- and contralateral tonsils of tumor patient with the crypt microbiome of OSA patients. α-diversity measures revealed equal richness (Fig. 4A) and richness and evenness (Fig. 4B) in the microbial communities. In contrast, principle coordinate analysis revealed a significant shift of the microbiome composition in tumor patients compared to OSA patients for all distance metrics (Fig. 4C-E and Supplementary Fig. 2A-C). In particular, we found a substantial dissimilarity in the microbial composition on the phylum level with significantly elevated abundance of Firmicutes and Actinobacteria and significantly reduced abundance of Spirochaetes, Synergistetes and Fusobacteria in tumor patients (Fig. 4F and Supplementary Fig. 2D). These results were further corroborated using projection of taxa abundance onto the phylogenetic tree. As shown in Supplementary Fig. 3, consistent patterns emerged along tree branches with significant differences between tumor and OSA patients and substantially higher abundance of the phylum Firmicutes and its genera Veillonella, Streptococcus and Megasphaera in tumor patients. The microbial composition analysis on the genus level revealed differential abundance of several genera in tonsillar crypts of tumor vs. OSA patients (Fig. 4G and Supplementary Fig. 4). Fusobacterium was the most abundant in OSA tonsils, whereas Veillonella and Prevotella_7 were highly abundant in OSA tonsils (Fig. 4G). The crypt microbiome of contralateral tonsils of tumor patients consistently showed an intermediate relative abundance on the genus level (Fig. 4G and Supplementary Fig. 4A). The observed shift in relative genera abundance was coherent across patients in both cohorts (Supplementary Fig. 4B). Together, these data unveil a distinct tonsillar crypt microbiome composition in patients suffering from tonsil cancer.
Identification of tonsil cancer patients based on predictive microbiome pattern
To further elaborate a potential relationship between the microbiota and tonsil cancer, we compared the crypt microbiome of tumor-affected and OSA tonsils at the species level using highly resolved taxonomic classification according to the human oral microbiome database [24]. Since bacterial operational taxonomic units typically do not follow a Gaussian distribution, differences on the species level were assessed using the DESeq2 method [25], which permits the analysis of count data based on the negative binomial distribution. The heatmap shown in Fig. 5A lists the top 45 differentially abundant bacterial species in a pattern that clearly separates tumor from OSA patients. To assess whether the abundance of certain bacterial species in tumor tonsils could be used as a predictor for the presence of tonsil tumor, we trained a machine-learning algorithm using the available data from OSA and tonsil cancer patients. The applied random forest classification method was shown previously to outperform other supervised classifiers on microbiome data [26]. Based on the complete 16S rRNA gene amplicon datasets from tumor and OSA patients, the trained model was able to detect a tumor-affected patient with high accuracy (0.89), sensitivity (0.89) and specificity (0.88) (Fig. 5B). The top ten tumor-predictive species included Treponema denticola, Fusobacterium periodonticum, Filifactor alocis, Fusobacterium nucleatum subsp.vincentii, Megasphaera micronuciformis, Prevotella melaninogenica and Veillonella atypica (Fig. 5B), which showed all significantly different abundances in tumor vs. OSA patients (Fig. 5C). To validate these results, we performed quantitative PCR analysis using biopsy DNA and published primers that allow for the reliable detection of bacterial species. For the ten potentially predictive bacterial species, the published 16S rRNA gene amplicon PCR primers for F. alocis [27] and P. melaninogenica [28] produced the most robust abundance measures. Using crypt DNA samples from the extended cohorts (21 OSA patients and 28 tonsil tumor patients, Supplementary Tables 1 and 2), we found a significantly higher relative abundance of F. alocis in crypt biopsies of OSA compared to tumor patients (Fig. 5D). As expected from the random forest analysis, P. melaninogenica could be detected by PCR in all samples with significantly higher relative abundance in tumor-affected tonsils compared to OSA tonsils (Fig. 5E). Collectively, these data reveal that assessment of the composition of microbial communities in tonsillar crypts can serve as diagnostic means for the detection of tonsil cancer.