Systematic analyses uncover robust salivary microbial signatures and host-microbiome perturbations in oral squamous cell carcinoma

Abstract

Oral squamous cell carcinoma (OSCC) is a prevalent malignancy in the oral-maxillofacial region with a poor prognosis. Oral microbiomes play a potential role in the pathogenesis of this disease. However, findings from individual studies have been inconsistent, and a comprehensive understanding of OSCC-associated microbiome dysbiosis remains elusive. Here, we conducted a large-scale meta-analysis by integrating 11 publicly available data sets comprising salivary microbiome profiles of OSCC patients and healthy controls. After correcting for batch effects, we observed significantly elevated alpha diversity and distinct beta-diversity patterns in the OSCC salivary microbiome compared to healthy controls. Leveraging random effects models, we identified robust microbial signatures associated with OSCC across data sets, including enrichment of taxa such as Streptococcus, Lactobacillus, Prevotella, Bulleidia moorei, and Haemophilus in OSCC samples. The machine learning models constructed from these microbial markers accurately predicted OSCC status, highlighting their potential as non-invasive diagnostic biomarkers. Intriguingly, our analyses revealed that the age- and gender-associated signatures in normal saliva microbiome were disrupted in the OSCC, suggesting perturbations in the intricate host-microbe interactions. Collectively, our findings uncovered complex alterations in the oral microbiome in OSCC, providing novel insights into disease etiology and paving the way for microbiome-based diagnostic and therapeutic strategies. Given that the salivary microbiome can reflect the overall health status of the host and that saliva sampling is a safe, non-invasive approach, it may be worthwhile to conduct broader screening of the salivary microbiome in high-risk OSCC populations as implications for early detection.

Importance: The oral cavity hosts a diverse microbial community that plays a crucial role in systemic and oral health. Accumulated research has investigated significant differences in the saliva microbiota associated with oral cancer, suggesting that microbiome dysbiosis may contribute to the pathogenesis of oral squamous cell carcinoma (OSCC). However, the specific microbial alterations linked to OSCC remain controversial. This meta-analysis reveals robust salivary microbiome alterations. Machine learning models using differential operational taxonomic units accurately predicted OSCC status, highlighting the potential of the salivary microbiome as a non-invasive diagnostic biomarker. Interestingly, age- and gender-associated signatures in the normal salivary microbiome were disrupted in OSCC, suggesting perturbations in host-microbe interactions.

Keywords: age; gender; meta-analysis; oral squamous cell carcinoma; salivary microbiome.

Fig 1

Study collection (A) and analysis workflow (B).

Fig 2

Study heterogeneity and batch effect correction. (A) PCoA plot showed significant differences between studies. (B) Different studies were aligned after ConQuR adjustment. (C) Adonis test evaluated the effects of various factors on the oral microbiome before and after ConQuR adjustment.

Fig 3

Alterations in the oral microbiome associated with OSCC across studies. (A) Forest plots depicting alpha diversity differences between healthy and OSCC samples. Dots on the left of the gray dashed line represent higher diversity in healthy groups; dots on the right indicate higher diversity in OSCC samples. Horizontal lines represent the 95% confidence intervals. (B) Differences between healthy and OSCC microbiome communities were calculated using the ADONIS analysis based on Bray-Curtis, Jaccard, unweighted UniFrac, and weighted UniFrac distance metrics. The combined R² was computed by setting the study as the stratifying factor in the ADONIS test. (C) Top 20 differential OTUs with the largest effect sizes between healthy controls and OSCC samples.

Fig 4

Performance of discriminating OSCC samples from healthy controls using 29 differential features. The heatmap depicts AUC scores from RF classifiers. The diagonal elements represent the cross-validation performance within each individual data set. The off-diagonal elements denote study-to-study predictions, with the training data sets along the rows and the test data sets along the columns. The “average”’ row shows the mean values for each column, with the last entry being the overall average of the diagonal values. The “leave-one-out” row displays the AUCs obtained by training the classifier with all data sets except the data set in the corresponding column and predicting the disease classes for the left-out data set. The “average” column provides the mean AUC values across each row.

Fig 5

Age-related oral microbiome shifts in OSCC. (A–C) The RF model was trained on the healthy oral microbiome to predict host microbial age. The average MAEs were 1.76, 5.14, and 5.82 for the healthy training set, healthy cross-validation set, and OSCC set, respectively. (D) The delta between predicted and chronological ages in healthy training, healthy cross-validation, and OSCC samples.

Fig 6

Gender-related oral microbiome shifts in OSCC. (A–C) The RF model was trained on healthy oral microbiome to classify host gender. The average AUCs were 0.97, 0.71, and 0.5 for the healthy training set, healthy cross-validation set, and OSCC set, respectively.