In this case–control study, the study population was drawn from the residents of Gongcheng County, Guangxi Zhuang Autonomous Region, who participated in a health survey in December 2018–December 2019. The following criteria were observed: (a) residents of the study area; (b) individuals who did not suffer from psychosis or other neurological disorders that may cause cognitive impairment, such as epilepsy, schizophrenia, alcoholism, and nervous system tumor; (c) individuals who did not suffer from severe visual or auditory impairment; (d) individuals who had not taken medication that could severely affect neurological function within the last 2 weeks. Then, 339 individuals with cognitive impairment according to the MMSE results were selected as the case group. The control population was matched by sex and age (± 5 years) at a 1:1 ratio. A total of 678 individuals aged 30–91 years were included. The research protocol was approved by the medical ethical committee of Guilin Medical University. All participants provided written informed consent.

Under the principle of informed consent, a standardized and uniform questionnaire was used. A face–to–face conversational questionnaire was administered to the study subjects by investigators who had received uniform and rigorous training. The main content of the questionnaire included information on demographics, past medical history, socioeconomic status, lifestyle, and cognitive functional status. Height (measured with a rangefinder with accuracy of up to 0.1 cm), weight (the participants were weighed with minimal clothing and with shoes off; accuracy was up to 0.1 kg), and waist circumference were measured as part of the physical examination. Body mass index (BMI) was calculated from the weight and height data, i.e., weight (kg)/height² (m²). Fasting venous blood was taken on the day of the physical examination and transported to the Laboratory Department of Gongcheng Yao Autonomous County People’s Hospital through.

cold chain transportation method in the morning of the same day. Fasting blood glucose (FPG) was tested by a blood routine analyzer (Sysmex CS–1600, Shanghai, China). The concentrations of glycosylated hemoglobin, uric acid (UA), serum total cholesterol (TC), high–density lipoprotein cholesterol (HDL–C), low-density lipoprotein cholesterol (LDL–C), and serum triglycerides (TG) were determined using a blood biochemical detector (Hitachi 7600–020, Kyoto, Japan).

Cognitive function assessment was assessed by face-to-face interviews using the Chinese version of the MMSE. The MMSE scale is the most commonly used tool to screen for cognitive impairment in clinical, research, and community settings (Folstein et al., 1975; Ismail et al., 2010; Arevalo-Rodriguez et al., 2015) and is widely used in large–scale epidemiological studies of dementia with high validity and reliability (Cao et al., 2012; Bickel et al., 2018). MMSE assesses five dimensions, namely, orientation (10 points), attention and calculation (5 points), memory (6 points), language skills (8 points), and visual–spatial (1 point) for a total score of 30 points. High MMSE scores indicate good cognitive function. Several studies have proved that education exerts an impact on cognitive function (Crum et al., 1993; Li H. et al., 2016; Lovden et al., 2020). Given the strong influence of education on cognitive function, we used the MMSE score cut–off points for different literacy levels to determine whether the study participants suffered from cognitive decline. The cognitive impairment group was divided on the basis of a cut-off point of 16/17 (MMSE score < 17) for the participants with no formal education, 19/20 (MMSE score < 20) for those with 1–6 years of education, and 23/24 (MMSE score < 24) for those with > 6 years of education (Li H. et al., 2016); the rest belonged to the cognitively normal group.

A two--part locus screening strategy involving functional single nucleotide polymorphisms (SNPs) and validated SNPs, was used to screen for the important candidate gene TFEB. The strategy was as follows: (1) Functional regional SNP screening: The TFEB gene was searched in the NCBI-SNP¹ website, and the TFEB gene promoter (upstream variant 2KB), 5′UTR, Exon (missense, synonymous), and functional SNP loci in the 3′UTR region (the relevant optimization parameter was minor allele frequency (MAF) in CHB > 0.05 according to HapMap or 1000Genomes database). The relevant literature was checked, and the results with disease susceptibility were annotated. The screened SNP loci were used for functional prediction using the NIH website². The SNP loci screened by³ were used to perform linkage disequilibrium (LD) analysis on the SNP loci screened in the first step and annotate the fully linked loci with R (Jia et al., 2020) = 1. (2) Validated and hot SNP screening: SNP loci with susceptibility were screened via a Google Scholar search, and then the screened SNP loci were verified before functional prediction at http://snpinfo.niehs.nih.gov/. Finally, the SNP loci of the appealing candidate genes was optimized by discarding the loci with R (Jia et al., 2020) = 1 and keeping the loci with R (Jia et al., 2020) > 0.8 in the promoter region. After the appeal screening, rs1015149, rs1062966, rs14063, rs2273068, rs11754668, and rs73733015 loci of the TFEB gene for typing were screened.

Blood samples were drawn from the participants by a professional nurse. DNA was extracted from 1 mL of blood samples by using the Blood DNA Kit (Tiangen, Beijing, China). Genotyping of SNPs was conducted, and primer design and synthesis were performed by Bio Miao Biological Technology Co., Ltd. (Beijing, China) by using the Sequenom MassARRAY matrix–assisted laser desorption ionization time–of–flight mass spectrometry platform (Sequenom Inc., San Diego, CA, United States). The primer sequence information is shown in Supplementary Table 1. PCR (ABI) was used to amplify the DNA. A 384–well SpectroCHIP bioarray was used as the microarray.

The data were statistically analyzed using SPSS22.0 statistical software. Normally distributed quantitative data were expressed in terms of mean ± SD, and the t-test of independence was used for comparison between groups. MMSE scores were not normally distributed and expressed as median and interquartile range, and comparison between groups was performed via the Wilcoxon or Kruskal–Wallis rank sum test. The qualitative data were presented in terms of percentage and were analyzed for the control and cognitive impairment groups by using a chi-square test or Fisher’s exact probability method. After adjusting for gender, age, education, and ethnicity as covariates, a multifactorial logistic regression model was used to calculate the odds ratio (OR) and 1–α confidence interval (CI) for assessing the association between the SNPs and cognitive function. A Bonferroni corrected p-value was applied to the multifactorial logistic regression p–values to account for the multiple testing of six different SNPs in the same samples (corrected α = 0.05/6 = 0.00833). LINK 1.90 software was used for the analysis of the Hardy--Weinberg equilibrium (HWE) and MAF statistics for the study population. Pair--wise LD and haplotype frequencies among the SNPs were analyzed using Haploview (Broad Institute of MIT and Harvard, United States, version 4.2). The analysis of the haplotypes comprising strongly linked SNP loci and risk of cognitive impairment was performed using SHEsis Main online software⁴. The criterion for significance was set at P < 0.05 for all tests.

The demographic parameters of the 678 study subjects are summarized in Table 1, where the median MMSE score was 21 and the interquartile range was 17–25. No statistically significant differences in mean age, sex ratio, BMI, ethnicity, smoking status, and alcohol consumption between the control and cognitively impaired groups (P > 0.05) were observed. In addition, a significant difference in the distribution of different education levels between the two groups (P < 0.05) was noted.

The SNP loci rs1015149, rs1062966, rs11754668, rs14063, rs2273068, and rs73733015 of the TFEB gene were successfully typed. Hence, all the above loci were located on chromosome 6. The MAFs of each locus were all greater than 0.05, which indicated non-low frequency variants. The genotype distribution of six SNPs in the cognitive impairment and control groups met the HWE (P_HWE > 0.05 for all). Table 2 shows that the allele and genotype frequencies of the TFEB rs1062966 and rs14063 SNPs in this study differed between the control and cognitive impairment groups (P < 0.05). The frequency of TFEB rs73733015C allele was higher in the cognitive impairment group than in the control group, and the difference was statistically significant between the two groups (P < 0.05).

The associations of the six SNPs with cognitive impairment are shown in Table 3. Using multifactorial logistic regression, the associations of SNP locus genotypes and four SNP genetic models with cognitive impairment were analyzed. After Bonferroni correction and adjustment for age, gender, ethnicity, and education level as covariates, the association of TFEB rs1062966 and rs14063 SNPs with cognitive impairment was significant in the dominant model in the total population (P < 0.00833). Subjects carrying the TFEB rs1062966T allele (CT + TT) showed a lower risk of cognitive impairment than the subjects with the CC genotype (dominant model: adjusted OR = 0.636, Bonferroni correction confidence interval = 0.405–0.998, P = 0.008). The risk of developing cognitive impairment was 1.547 times higher in those carrying the TFEB rs14063A allele (AG + AA) than in those with the GG genotype (dominant model: adjusted OR = 1.547, Bonferroni correction confidence interval = 1.021–2.345, P = 0.006). In the co-dominant model, subjects carrying the TFEB rs14063AG genotype had an increased risk of cognitive impairment relative to those with the GG genotype, and the risk of cognitive impairment was 1.553 times higher in the AG carriers than in the GG carriers (co-dominant model: adjusted OR = 1.553, Bonferroni correction confidence interval = 1.007–2.397, P = 0.007). The association between genotype and cognitive functional status for the remaining SNP loci was not statistically significant.

Given the effect of education level on cognitive function, the study population was further stratified into three groups for analysis in accordance with the number of years of formal education received. As shown in Supplementary Table 2, after Bonferroni correction and adjustment for gender, age and ethnicity, results showed a significant association between TFEB rs1062966 and cognitive function in the group with no formal education. In the co-dominant model, the presence of the TFEB rs1062966 CT genotype was more strongly associated with a reduced risk of cognitive impairment than the presence of the CC genotype (co-dominant model: adjusted OR = 0.510, Bonferroni correction confidence interval = 0.275–0.948, P = 0.004). In the dominant model, subjects carrying the TFEB rs1062966T allele (CT + TT) had a lower risk of cognitive impairment than those carrying the CC genotype (dominant model: adjusted OR = 0.510, Bonferroni correction confidence interval = 0.281–0.928, P = 0.003). In the overdominant model, the presence of the TFEB rs1062966CT genotype was more closely associated with a reduced risk of cognitive dysfunction than the presence of the CC + TT genotype (overdominant model: adjusted OR = 0.527, Bonferroni correction confidence interval = 0.285–0.974, P = 0.006). The remaining SNP loci were not statistically associated with cognitive function in any of the four genetic models (P > 0.0833).

Ethnic differences in genetic background might exist in different racial/ethnic groups. Therefore, the study population was further stratified by ethnicity for analysis. As shown in Supplementary Table 3, after Bonferroni correction and adjustment for sex, age, and education level, results showed a significant association between TFEB rs1062966 and cognitive function in the Yao minority group. In the co-dominant model, subjects carrying the CT genotype had a lower risk of cognitive impairment than those carrying the CC genotype (co-dominant model: adjusted OR = 0.563, Bonferroni correction confidence interval = 0.317–0.999, P = 0.008). In the dominant model, subjects carrying the TFEB rs1062966T allele (CT + TT) showed a lower risk of cognitive impairment than those carrying the CC genotype (dominant model: adjusted OR = 0.543, Bonferroni correction confidence interval = 0.311–0.946, P = 0.004). In other ethnic groups, SNP loci were not statistically different from cognitive function in any of the four genetic models (P > 0.0833).

Figure 1 shows the LD analysis of the SNPs of TFEB. Multilocus LD analyses indicated that the tested sites in the study population were not statistically independent and that the control and cognitive impairment groups showed strong LD (D’ = 0.89–0.99). As shown in Table 4, the dominant haplotypes were rs14063G–rs1062966C–rs2278068C–rs1015149C (>40% of the samples) and rs14063A–rs1062966C–rs2278068C–rs1015149T (>20% of the samples). The rs14063A–rs1062966C–rs2278068C–rs1015149T and rs14063G–rs1062966T–rs2278068C–rs1015149C haplotype frequencies were statistically different between the two groups (P < 0.05 for all). The rs14063A–rs1062966C–rs2278068C–rs1015149T haplotype was associated with an increased risk of cognitive impairment (OR = 1.370, 95% CI = 1.044–1.797, P = 0.022797), whereas the rs14063G–rs1062966T–rs2278068C–rs1015149C haplotype.

Was protective against cognitive function (OR = 0.690, 95% CI = 0.517–0.922, P = 0.011908).