Introduction

Oral cancer (OC) comprises malignant lesions in the oral cavity and ranks among the most prevalent cancers globally. Although some progress has been made in understanding OC etiology over the years, the pathogenesis has yet to be elucidated.

Significant changes in the oral microbiota have been observed during OC, highlighting its important role in patients1,2,3. Multiple studies have found that Prevotella and Veillonella are associated with OC, but the evidence is inconsistent. For example, Nouri Z et al.4 found reduced Prevotella abundance in OC, while Herreros-Pomares A et al.5 reported the opposite. Previously, case-control studies were the most common; however, effectively controlling for confounding factors like lifestyle, dietary habits, and environment in observational studies of the oral microbiota and OC remains a challenge. For instance, smoking—a major oral cancer risk factor—affects oral microbiota abundance, creating spurious associations in unadjusted analyses. These conditions pose significant limitations on drawing causal inferences between OC and oral microbiota.

It is in this context that Mendelian randomization (MR) emerges as an innovative approach to investigate the causal relationship between OC and the oral microbiota. MR uses genetic variants to construct instrumental variables of exposure to estimate the causal association between exposure and disease outcome. Because the allocation of genotypes from parent to offspring is random, the association between genetic variants and outcome is not affected by common confounding factors, and a causal sequence is reasonable. Microbiota and various diseases, such as cancer and metabolic disorders, have been extensively studied using MR6. There have also been several MR studies analyzing the relationship between microbiota and OC, but all of these have focused on gut microbiome. Given that most studies on microbiota and OC have focused on the oral microbiota, and considering its physical proximity to OC sites, it is reasonable to posit that the oral microbiota may have a more significant impact on OC than the gut microbiota. Thus, this research employed genome-wide association study (GWAS) summary statistics from the ADDITION-PRO study (2009–2011) and the UK Biobank (UKB) consortium to perform a two-sample MR analysis, assessing the causal link between OC and oral microbiota.

Methods

Data sources

As the data in this research come from existing, published GWASs, ethical approval and informed consent have been secured by the original studies.

Genetic variants for oral microbiota were obtained from the latest GWAS for oral microbiota composition7. The study included 610 individuals of European ancestry from the Danish ADDITION-PRO cohort. To profile the microbial composition, the hypervariable region V4 of the 16S rRNA gene was targeted. The eHOMD database (version V15.22) and dada2 algorithm were used for taxonomic classification. An analysis of microbiota quantitative trait loci mapping identified genetic variants linked with bacterial taxa abundance in the oral microbiome. In the original study, the oral microbiota was classified into 43 taxa, comprising 2 phyla, 1 classes, 4 orders, 5 families, 12 genera, and 19 species, and all of these taxa were included in this study. GWAS summary statistics for OC were derived from the UKB8. A total of 407,464 European subjects were included in this summary statistics, with 643 cases and 406,821 controls. These GWAS sample populations are primarily of European descent and mostly independent. For detailed information on population recruitment criteria and genetic data quality control, refer to the original paper7,8.

Instrumental variable (IV)

To ensure causal inference accuracy and robustness, IVs must meet MR’s key assumptions (Fig. 1). The IVs were selected based on the following criteria: (1) a set of single nucleotide polymorphisms (SNPs) related to each taxon, which met the significance threshold (P < 1.0 × 10–5), were selected as potential IVs; (2) linkage disequilibrium calculated using 1000 Genomes project European samples, retaining SNPs with R2 < 0.001 and the lowest P-values within a 10,000 kb clumping window; (3) removal of SNPs with minor allele frequency of 0.01 or lower; (4) to reduce bias, SNPs with F-statistics greater than ten are selected, and (5) alignment of exposure and outcome datasets to remove SNPs with discordant allele pairings.

Fig. 1
Fig. 1
Full size image

Schematic representation of the two-sample bi-directional Mendelian randomization analysis. Mendelian randomization was used to evaluate the causal links between 43 microbial taxa and oral cancer. Three key assumptions of Mendelian randomization: (1) genetic variants must be associated with exposures, (2) genetic variants must not be associated with confounders, and (3) genetic variants must affect outcomes only through exposures, not through other pathways.

STROBE-MR checklist was followed in the conduct and interpretation of the MR study9,10 (Additional file 1: Table S1).

Statistical analysis

This study employed various methods—inverse variance weighted (IVW), weighted model, MR-Egger, maximum likelihood (ML), MR-PRESSO, and weighted median—to investigate the causal relationship between OC and oral microbiota. In the absence of heterogeneity or horizontal pleiotropy, the ML method is comparable to the IVW method. The results are more likely to be unbiased if these conditions are met, and the standard errors are typically lower than those obtained using IVW11. We utilized the MR-Egger intercept to assess pleiotropy. In scenarios where horizontal pleiotropy is present, the MR-Egger regression estimate tends to offer a more precise depiction of the causal relationship. When the intercept of this regression is close to zero, it indicates that the MR-Egger regression is in concordance with the IVW model. Conversely, a non-zero intercept may hint at the presence of horizontal pleiotropy among the IVs under consideration12. Furthermore, the weighted median method is known to provide consistent estimates of the causal effect, provided that less than 50% of the SNPs are deemed invalid13.

Cochran’s IVW Q statistics quantified IV heterogeneity. A ‘leave-one-out’ approach was utilized to detect potential heterogeneity SNPs by sequentially excluding each IV from the analysis. We utilized MR-PRESSO tests to address horizontal pleiotropy by eliminating IV outliers. The MR-Egger regression was employed to evaluate pleiotropy, providing a robust causal estimate adjusted for pleiotropy under the assumption of no measurement error and independence between instrument strength and direct effects. In order to determine the causal impact of OC on oral microbiota, we conducted reverse MR analysis on microbes that were previously identified as causally linked to OC in the forward MR analysis. The methodology and parameters used were aligned with those applied in the forward MR analysis.

Instrument strength was quantified by determining the F-statistic, calculated as \(\:\text{F}=\frac{{R}^{2}(N-2)}{(1-{R}^{2})}\), with R2 denoting the proportion of variance in the exposure explained by the genetic variants and N signifying the sample size14,15. R2 was derived using the formula:\(\:\:{R}^{2}=\frac{2\times\:EAF\times\:(1-EAF)\times\:{beta}^{2}}{2\times\:EAF\times\:\left(1-EAF\right)\times\:{beta}^{2}+2\times\:EAF\times\:(1-EAF)\times\:N\times\:{SE\left(beta\right)}^{2}}\), where EAF denotes the effect allele frequency, beta represents the genetic impact on oral microbiota, and SE (beta) is the standard error of this genetic effect. The absence of significant weak instrument bias was indicated when the F-statistic was great than 1016.

In this research, we decided against implementing multiple testing corrections due to the intricate interplay and relationships among microbial features. While standard in many analyses, we contend that such corrections could be excessively stringent for our data, risking the obscuration of genuine causal links. This approach is justified by the intricate nature of microbial ecosystems, where inter-microbe influences are prevalent. Considering the existing interactions, the use of multiple testing corrections for individual microbial features might be excessively stringent, potentially masking signals indicative of true causality17.

All statistical analyses were performed using R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria), with the R packages “TwoSampleMR” (version 0.6.6) and MR-PRESSO (version 1.0)18.

Results

In line with the IV selection criteria, a total of 477 SNPs were designated as IVs for 43 distinct bacterial taxa. Further details regarding these chosen IVs can be found in Additional File 2: Table S2.

Causal effect of oral microbiota on OC

As shown in Table 1, Additional file2: Table S3, and Fig. 2, two bacterial genera, specifically, Prevotella and Veillonella, were identified as being associated with OC across at least one MR analytical approach. IVW estimate indicates that Prevotella exerted a protective influence on OC (OR = 0.89, 95% CI: 0.81–0.99, P = 0.03), while Veillonella had a deleterious effect on OC (OR = 1.14, 95% CI: 1.01–1.31, P = 0.04).

Table 1 MR estimates for the association between oral microbiota and OC.
Full size table
Fig. 2
Fig. 2
Full size image

Scatter plots for the causal association between oral microbiota and oral cancer.

Cochran’s Q test for IVW revealed no significant heterogeneity among the IVs (Additional file 2: Table S4). Additionally, the MR-Egger intercept analysis also indicated no significant directional pleiotropy (Additional file 2: Table S5).

Visual inspection of scatter (Fig. 2) and leave-one-out (Fig. 3) plots noted potential outlier IVs for Prevotella. Nevertheless, subsequent MR-PRESSO analysis revealed no significant outliers (global test P > 0.05, Additional file 2: Table S6). Thus, there was inadequate evidence for horizontal pleiotropy in the Prevotella-OC association.

Fig. 3
Fig. 3
Full size image

leave-one-out plots for the causal association between oral microbiota and oral cancer.

Causal effect of OC on oral microbiota

Reverse MR analysis revealed no causal impact of OC on these bacteria (Additional file 2: Tables S7 and S8). Cochran’s Q test for IVW indicated no significant heterogeneity among OC IVs (Additional file 2: Table S9). Neither the MR-Egger regression intercept analysis (Additional file 2: Table S10) nor the MR-PRESSO analysis (Additional file 2: Table S11) found significant horizontal pleiotropy.

Discussion

In this study, using summary statistics of oral microbiota from the latest GWAS analysis and the summary statistics of OC from the UKB, we first conducted a two-sample MR analysis to assess the causal link between oral microbiota and OC. Result indicated that Prevotella exerted a protective influence on OC, while Veillonella had a deleterious effect on OC.

Many observational studies have identified the link between oral microbiota and OC. Previous research indicates a potential link between Prevotella and OC, yet the evidence is still inconsistent; some studies have found reduced Prevotella abundance in OC4, while others have observed an increase5. Here, we demonstrated Prevotella had protective effects on OC in European population. Currently, the mechanism of Prevotella and OC is unknown. However, as a proteolytic/amino acid-degrading bacterium, Prevotella can cleave proteins and peptides into amino acids and further metabolize them through specific pathways to generate short-chain fatty acids (SCFAs). Nouri et al.4 demonstrated that alterations in oral microbiota composition reduce SCFA levels and FFAR2 expression, which may trigger inflammation by increasing TNFAIP8 and activating the IL-6/STAT3 pathway, potentially elevating the OC risk. In addition, as the second prevalent genus in human oral cavity, Prevotella comprises over 50 identified species, which could have diverse physiological functions19. This is possibly one of the reasons of inconsistent results of previous studies. Thus, future studies should focus on conducting species-level GWAS of the oral microbiota using metagenomic data, followed by MR analyses to further investigate the casual association between the oral microbiota and OC. Furthermore, multiple studies have consistently indicated a lower prevalence of Prevotella in Westernized populations compared to non-Westernized ones20,21,22,23,24,25, which exhibit a higher prevalence of several distinct Prevotella species19. However, this study was conducted in two European populations, caution is needed when extrapolating our results to different populations, particularly to non-Westernized populations.

In alignment with prior research26,27, our findings also suggest that Veillonella have a deleterious effect on OC. Although the precise mechanism of Veillonella effect on OC is remain unclear, Tsay JJ et al.28 found that Veillonella from oral cavity could promotes lung cancer through PI3 K (phosphoinositide 3-kinase) pathway. And Hirschfeld J et al.29 noted that Veillonella could promote the release of reactive oxygen species, which are associated with the development of OC. Furthermore, many studies have found that Veillonella provides adhesion sites and nutrients for the pathogen P. gingivalis30,31,32,33, which promotes the progression of OC through various mechanisms3,34,35,36,37,38. More data will be needed in order to better determine the mechanism of Veillonella effect on OC in the future.

Albeit several MR studies have analyzed the link between microbiota and OC, they have all focused on the gut microbiome. Given that the majority of research on the microbiota’s association with OC has concentrated on the oral microbiota, and considering its closer proximity to OC sites, it is plausible to posit that its impact on OC may exceed that of the gut microbiota. Thus, we conducted the world’s first MR analysis to investigate the causal link between OC and oral microbiota. This study is of significant importance for deepening our comprehension of the interplay between the oral microbiota and the development of OC. In the last two decades, numerous studies have shown that probiotic supplementation can reestablish equilibrium and functionality in the oral microbiota of oncology patients, following disruptions caused by cancer therapy39. Additionally, oral probiotics can effectively manage oral dysbiosis, sans any adverse side effects40. Our study provides further evidence supporting the integration of microbiome-based adjunct therapies into the treatment regimen for OC. Consequently, future research should prioritize species-level GWAS for oral microbiota, with a particular emphasis on probiotic species. Further MR analyses, informed by these GWAS results, could potentially provide additional evidence to substantiate the application of probiotics as an adjunct therapy in OC therapy.

However, this research comes with several limitations to consider when interpreting the findings. To enhance sensitivity analysis and detect horizontal pleiotropy, a more comprehensive set of genetic variants should be considered as IVs. Consequently, the SNPs utilized in this analysis didn’t reach the conventional GWAS significance threshold (P < 5 × 10–8). Given the limited sample size of the oral microbiota GWAS, the reproducibility and reliability of the GWAS results remain uncertain. And given that microbiome taxonomic profiling based on the V4 region of the 16S rRNA gene cannot achieve species-level resolution, future studies should prioritize large-scale GWAS of oral microbiota utilizing shotgun metagenomic sequencing to enable precise species- and strain-level characterization. Furthermore, the limited sample size of the oral microbiota may have introduced weak instrument bias into the reverse MR analysis, meaning a reverse causal association cannot be definitively ruled out. Moreover, we opted not to apply multiple testing corrections due to the intricate interplay and relationships among microbial features. While standard in many analyses, we contend that such corrections could be excessively stringent for our data, risking the obscuration of genuine causal links. We acknowledge this limitation and suggest that future studies further investigate these interactions, refine statistical methodologies, and conduct larger-scale studies for confirmation.

Conclusion

In summary, this two-sample MR study found that Prevotella had a protective effect on OC, while Veillonella had a deleterious effect on OC. Further studies are necessary to elucidate their underlying mechanism. Furthermore, while the reverse MR estimates did not support a causal effect of OC on oral microbiota, we cannot exclude the possibility that OC might influence oral microbiota. This aspect warrants further investigation in future studies.