No genetic causal association between dental caries and Alzheimer’s disease: a bidirectional two-sample Mendelian randomization analysis

GWAS summary statistics of dental caries

The summarized GWAS data of dental caries were obtained from a publicly acquired GWAS meta-analysis containing 487,823 participants (Shungin et al., 2019), which incorporated evidence from the Gene-Lifestyle Interactions in Dental Endpoints (GLIDE) consortium (Shungin et al., 2015) and UK Biobank (UKB) (Bycroft et al., 2018). All individuals were of European descent. The GLIDE consortium is a cohort-based project with detailed clinical measures to explore individual susceptibility or resistance to dental caries, while the UKB is characterized by self-reported measures of dental condition through questionnaires. In the GWAS meta-analysis, forty-seven single nucleotide polymorphisms (SNPs) with genome-wide significance were obtained for further analysis.

GWAS summary statistics of Alzheimer’s disease

We only chose GWAS in the European population to remain in accordance with the population of dental caries. Therefore, genetic association estimates for AD were obtained from the FinnGen database (Kurki et al., 2022). The FinnGen consortium, launched in Finland, is a large public project that involves genome and health data from 500,000 Finnish biobank participants. We chose the following phenotypes from the FinnGen database: AD (AD, 3,899 cases and 214,893 controls), early-onset AD (ADEO, 587 cases and 214,885 controls) and late-onset AD (ADLO, 2,670 cases and 214,871 controls) (Kurki et al., 2022). The patients were divided into an early-onset group and a late-onset group according to whether they were older than 65 years based on the International Statistical Classification of Diseases (version 2016) (https://icd.who.int/browse10/2016/en). Furthermore, we also validated our MR results by using AD GWAS data from different databases, including data from the International Genomics of Alzheimer’s Project consortium (IGAP) and the UKB (Bycroft et al., 2018; Kunkle et al., 2019). The IGAP consortium, an authoritative consortium of AD, analyzed the genetic data of 63,926 European participants (21,982 AD cases and 41,944 controls) and reported more than 10 million SNPs. However, we finally decided to abandon the AD GWAS, which mostly harbored UKB participants (Larsson, Woolf & Gill, 2022), considering the influence of population overlaps between the exposure and the outcomes.

MR design

In our study, bidirectional two-sample MR was performed to investigate the causality between dental caries and AD (Fig. 1). Genetic variants were used as instrumental variables in the MR analysis. For the dental caries-AD association, SNPs associated with dental caries were used as IVs to explore the effect of dental caries on AD. SNPs associated with dental caries were obtained from a recent meta-analysis of large GWASs, and AD GWASs were obtained from FinnGen. Regarding the AD-dental caries association, SNPs associated with AD were used as IVs and were extracted from the GWAS from the IGAP consortium, while the dental caries GWAS was sourced from the UKB dataset. All the mentioned GWASs were conducted in European individuals.

Selection of genetic instrumental variables

To determine whether the association between the risk factor (exposure) and the outcome is causal, the genetic variants associated with the modifiable factors were chosen as IVs in the MR, which are similar to the random assignment in a randomized trial (Emdin, Khera & Kathiresan, 2017). Additionally, the variants used as IVs should satisfy three basic requirements. First, the selected IVs are associated with the exposure (the relevance assumption). Second, there are no associations between the IVs and the confounders (the independence assumption). Third, the IVs could only exert an impact on the outcomes through exposure rather than other pathways (the exclusion restriction).

In the causal analysis of dental caries on AD risk, SNPs from GWASs that are significantly relevant with dental caries at the genome-wide level are used as IVs. To satisfy the first assumption, forty-seven SNPs are finally identified to be significantly associated with dental caries in a GWAS meta-analysis containing 487,823 participants and included in our MR analysis. Additionally, SNPs that were associated with factors that might influence the exposure (dental caries) or the outcome (AD) were removed to guarantee the second and third assumptions. Based on a literature review, smoking, obesity, hypertension and diabetes seem to be modifiable factors for AD (Durazzo, Mattsson & Weiner, 2014; Yu et al., 2020; Zhang et al., 2021). Therefore, these factors seem to be confounding factors that intervene in the causal relationship between dental caries and AD. Thus, we used the PhenoScanner database (http://www.phenoscanner.medschl.cam.ac.uk/) to manually screen and remove SNPs related to confounding factors and AD outcomes. Those SNPs that were directly associated with AD were manually removed as well. Consequently, nine SNPs (rs3865314, rs11676272, rs62106258, rs9366651, rs898797, rs10851907, rs8054556, rs28822480, rs7429279) related to those confounding factors were abandoned to minimize horizontal pleiotropy.

In the MR analysis of AD on dental caries risk, to satisfy the relevance assumption, we only included SNPs from large GWASs that were significantly associated with AD as IVs to investigate the causal relationship between AD and dental caries risk. To satisfy the independence assumption and the exclusion restriction, SNPs were dismissed if they were associated with confounders that might influence the risk of developing dental caries. Diabetes, sugar intake and education level were reported to be associated with the occurrence of dental caries (Martignon et al., 2021). Therefore, we manually removed SNPs that were associated with those factors based on the PhenoScanner database. Additionally, SNPs holding direct association with dental caries were also omitted.

We set genome-wide significance with a p value of 5 × 10⁻⁸ as a threshold to identify significant SNPs associated with dental caries used as IVs. A PLINK clumping method was used to prune SNPs in linkage disequilibrium (Purcell et al., 2007), and the clumping window of r² was set as 0.001, while the distance was set as 10,000 kb. SNPs were sorted according to the p value, and the SNP with the lowest p value ranked first and was retained. Within a 10,000 kb distance, the SNP was dumped when its correlation with the most significant SNP was greater than 0.001. The genetic SNPs associated with the outcomes were also removed by searching the variant-trait associations on the public database PhenoScanner V2 (http://www.phenoscanner.medschl.cam.ac.uk/). In addition, the proxy would be taken as a substitute for SNPs that could not be found in the outcome GWAS according to LDlink (https://ldlink.nci.nih.gov/). To minimize the impact of confounding factors on the outcomes, we excluded SNPs associated with confounders based on PhenoScanner (Kamat et al., 2019). We extracted the remaining SNPs from the GWAS of outcomes with a minor allele frequency cutoff set as 0.01.

Statistical analysis

Various MR analytic methods could be employed to estimate the causal effect using summarized statistics (Burgess, Butterworth & Thompson, 2013; Richmond & Davey Smith, 2022). Among them, the inverse variance weighted (IVW) method was the most commonly used method in MR analysis. Based on a meta-analysis approach, the IVW method was often used to obtain a pooled estimation of the ratio of the causal effect, especially when there were multiple IVs (Dudbridge, 2021; Richmond & Davey Smith, 2022). The MR‒Egger method was able to provide a test for a causal effect when the genetic variants were weak IVs, as well as to detect whether there was directional pleiotropy. When the MR‒Egger intercept value was not markedly different from zero, the MR‒Egger estimation would be in parallel with the IVW estimation (Burgess & Thompson, 2017). The median-based method was adopted as a supplement to the IVW method to reduce the bias produced by abnormal SNPs, and the maximum likelihood method was used if there was an overlap in the investigated population between the exposure and the outcome.

After MR analysis, heterogeneity and pleiotropy tests were performed in the sensitivity analysis. Cochran’s Q statistic was widely used to measure heterogeneity, and heterogeneity was present when the p value of Q was less than 0.05. A large Q value indicates that the SNPs associated with dental caries could exert various effects on outcomes even with large differences and that the findings should be cautiously interpreted. The MR‒Egger regression test was adopted to explore the directional pleiotropy of the dental caries-associated SNPs by comparing the intercept and the origin, and horizontal pleiotropy was absent if the p value was larger than 0.05. To detect any possible horizontal pleiotropy and identify any outliers, Mendelian randomization pleiotropy RESidual Sum and Outlier (MR-PRESSO) analysis was also employed (Verbanck et al., 2018). MR-PRESSO detected horizontal pleiotropy via the global test, identified pleiotropic variants via the outlier test, and then performed IVW to estimate the causal effect by removing the pleiotropic variants (Zhu, 2021). Leave-one-out analysis was also utilized to determine the potential SNPs that might impact dental caries-outcome associations.

We conducted our two-sample MR on R software (version 4.1.1) for Windows. Three packages were mainly employed for statistical analysis, data output and visualization: (1) ‘TwoSampleMR’ (https://github.com/MRCIEU/TwoSampleMR), (2) ‘MRPRESSO’ (https://github.com/rondolab/MR-PRESSO), and (3) ‘forestplot’ (https://cran.r-project.org/web/packages/forestplot/index.html). A p value was indicative of statistical significance based on a defined cutoff of 0.05.

Instrumental variable selection

The SNPs that conformed with all three requirements were chosen as IVs to estimate the causal effect size between dental caries and AD. As demonstrated in Tables 1 and S1, for dental caries-AD association analysis, forty-seven genome-wide significant and independent SNPs that were associated with dental caries identified by the latest GWAS were selected. After the deletion of SNPs that did not satisfy the three consumptions and SNPs absent from the outcome GWASs from the FinnGen database, 30 SNPs remained in the analyses on AD, 30 SNPs in the analyses on ADEO and 30 SNPs on ADLO. Meanwhile, only 26 SNPs were incorporated into MR analyses based on the AD GWAS from the IGAP consortium.

For the AD-dental caries association analysis, after the process of clumping and filtering SNPs, 11 SNPs independently associated with AD taken from the IGAP consortium were ultimately retained for further analysis, and no SNP was retained from the FinnGen datasets. Thus, we only investigated the causal effect of AD on dental caries with GWAS data from the IGAP and UKB.

Effect of dental caries on Alzheimer’s disease

The estimation of the effect of dental caries on AD according to the FinnGen from different MR methods is displayed in Fig. 2A. For the FinnGen dataset, as shown in Fig. 3, the causal estimation from the IVW method revealed that there was no significant association between dental caries and AD (odds ratio (OR): 1.041, 95% CI [0.630–1.722], p = 0.874). Moreover, as shown in Table 1, no obvious significance was identified from either the MR‒Egger or median method (MR–Egger OR: 0.873, 95% CI [0.177–4.298], p = 0.869; simple median OR: 1.244, 95% CI [0.607–2.549], p = 0.550; weighted median OR: 0.993, 95% CI [0.456–2.163], p = 0.986). In the sensitivity analysis, displayed in Table 2, there was no evidence of heterogeneity by Cochran’s Q statistic (Q = 29.530, p = 0.386). The MR‒Egger method also identified no significant direction pleiotropy by comparing the intercept value and zero (intercept value = 0.004, p = 0.821). Meanwhile, no robust horizontal pleiotropy was revealed by MR-PRESSO analysis (global test p = 0.467), and no outliers were detected.

MR analysis was also performed with the GWAS from the IGAP dataset to confirm the results. As shown in Figs. 2B and 3, no causal association between dental caries and AD risk was suggested (IVW OR: 1.162, 95% CI [0.813–1.661], p = 0.409). Consistent with the IVW results, evidence obtained from other MR methods also supported that the AD risk did not increase with increasing dental caries (MR‒Egger OR: 1.537, 95% CI [0.495–4.775], p = 0.464; simple median OR: 1.122, 95% CI [0.70510–1.772], p = 0.662; weighted median OR: 1.006, 95% CI [0.630–1.607], p = 0.979). No significant findings were identified from either Cochran’s Q heterogeneity test or the pleiotropy test with the MR‒Egger and MR–PRESSO methods (Table 2).

Effect of dental caries on Alzheimer’s disease onset ages

The effect of dental caries on the onset age of AD was also investigated, and the scatter plot (Figs. 2C and 2D) demonstrated the relationship between dental caries and AD onset age. We found no causality between dental caries and the onset age of AD. Dental caries could not alter the onset age of AD. For the ADEO subtype, no apparent statistical evidence for the causal effect of dental caries was observed (IVW OR: 0.515, 95% CI [0.146–1.817], p = 0.302), as displayed in Fig. 3. In the meantime, the evidence for increased dental caries associated with a higher risk of ADLO was absent from the IVW approach (OR: 1.329, 95% CI [0.735–2.403], p = 0.347). Similar results were obtained by the MR‒Egger and weighted median methods either for the effect of dental caries on ADEO or on ADLO (Table 1). Next, we performed a sensitivity test to detect pleiotropy, and all the MR‒Egger intercepts were near zero (p = 0.637 for ADEO, p = 0.959 for ADLO), suggesting that there was no directional pleiotropy. The MR–PRESSO global test also detected no horizontal pleiotropy in ADEO and ADLO (p = 0.255 and 0.738, respectively).

Effect of Alzheimer’s disease on dental caries

We detected no causal effect of AD on dental caries, and having AD did not alter the risk of developing dental caries (IVW OR: 1.000, 95% CI [0.999–1.001], p = 0.717) (Fig. 4). Consistent with the IVW results (Table 1), evidence obtained from other MR methods also supported that increased AD risk would not increase the chance of developing dental caries (MR‒Egger OR: 1.000, 95% CI [0.997–1.004], p = 0.859; simple median OR: 1.000, 95% CI [0.997–1.004], p = 0.723; weighted median OR: 1.000, 95% CI [0.999–1.001], p = 0.723). No significant findings were identified from either Cochran’s Q heterogeneity test (p = 0.379) or the pleiotropy test with the MR‒Egger (p = 0.916) and MR–PRESSO methods (p = 0.415).