Effects of Cognitive- and Sleep- Related Single Nucleotide Polymorphisms on Cognitive Functions in the Han Chinese Community-dwelling Elderly

Objective: The genetic biomarkers on Alzheimer’s disease (AD) have been widely studied in different groups. Since the lack of ecacy therapeutic methods for AD, early recognition in preclinical stage becomes increasingly important. Evidence of AD and cognitive-related single nucleotide polymorphisms (SNPs) in high risk population is insuciency. Our aim was to assess whether these SNPs within cognitive- and sleep-associated genes are correlated with cognitive impairment independently or through gene-gene interactions in community-dwelling elderly in Beijing. Methods: Eight single-nucleotide polymorphisms (SNPs) were genotyped in 2133 Northen Han Chinese elderly from ten communites in Chaoyang District, Beijing. The short version of the Montreal Cognitive Assessment-Beijing (MoCA-s), Ascertain Dementia 8 (AD8), Digit Span Backwards (DSB), Digital Symbol Substitution Test (DSST), and Paired Associative Learning Test (PALT) were used to detect different cognitive domains. Results: Logistic regression analyses showed a signicant correlation between APOE (rs429358), ABCC9 (rs11046205) and cognitive impairment, and an interaction between ABCC9 (rs11046205)/APOE promoter (rs405509) and APOE ε4 status. Additionally, we found a signicant negative association between the additive model of APOE (rs429358) and MMSE score. Moreover, our analysis revealed that the gene-gene interaction between ABCC9 (rs11046205) and APOE (429358) may contribute to the etiology of congnitive impairment. Conclusions: This study conrmed the independent contribution of AD, memory and sleep-related genes in the cognitive impairment of the community-dwelling elderly in China, as well as through gene-gene interactions. reveal more valuble information into the multiple complex pathogeneses of cognitive impairment. AD: Alzheimer’s MCI: mild cognitive impairment; SNP: single-nucleotide polymorphisms; MMSE: Mini-Mental State Examination; MoCA-s: the short version of the Montreal Cognitive Assessment-Beijing; AD8: Ascertain Dementia DSB: Digit Span Backwards; DSST: Digital Symbol Substitution Test; PALT: Paired Associative Learning Test; APOE: apolipoprotein E; CESD: Center for Epidemiological Studies-Depression; HWE: Hardy-Weinberg equilibrium; GMDR: generalized multifactor dimensionality reduction; CI: condence interval; OR: odds ratio; MAF: minor allele frequency; CV: cross-validation consistency;

elderly [11,12]. For aspects of cognitive function, the de cit of memory is a typical characteristic of people with cognitive impairment, in the transitional stage between normal aging and dementia, especially episodic memory [13,14]. Two related genes, DTNBP1 (rs1047631) and CLSTN2 (rs6439886), have been found to be involved in memory phenotype. The DTNBP1 variation, which could in uence glutamate signaling and used to be associated with schizophrenia, has been proven to be associated with general cognitive ability (especially working memory) and cognitive impairment [15,16]. As the most typical factor in cognitive impairment, episodic memory has been paid more attention in the research of the preclinical stage of dementia. As one of the well-known genes associated with episodic memory, CLSTN2 (rs6439886) was also included in our research [17,18].
Numerous cross-section and cohort studies have con rmed the relationship between sleep problem and the risk of developing cognitive impairment [19,20]. About 60% of people with cognitive impairment have trouble sleeping [21]. Due to the high incidence of sleep disorder in this population, we selected the ABCC9 (rs11046205) gene in our research, a SNP which has been proved to be associated with sleep duration [22].
In this study, we intended to verify the correlation between AD-related risk SNPs and cognitive impairment susceptibility in the community-dwelling elderly in Beijing, and attempted to discover potential pathways the cognitive-and sleep-related SNPs may serve in the early recognition.

Study population
We recruited 2084 participants (and excluded three participants who didn't nish the MMSE) consecutively from ten communities in Chaoyang District, Beijing, China, between November 2018 and May 2019. The inclusion criteria were: 1) have permanent-resident status in Beijing; 2) community residence; 3) age 50-85; 4) no history of severe brain injury; 5) normal daily living ability; 6) ethnic Han; 7) and no neuropsychiatric diseases (such as Alzheimer's disease, depression, Parkinson's disease, etc.). We excluded those participants who could not guarantee the entire study would be completed and each signed the approved informed consent form. The trail protocol was approved by the research ethics committees of the Institute of Psychology, Chinese Academy of Science (H11036) (ChiCTR1900025487).

General Information Collection
Trained personnel visited all the participants at the local hospital for data collection, including sociodemographic characteristics, lifestyle habits and patient history by using a designed questionnaire. Smoking status was divided into current smoking and non-smoking (including never smoking and smoking cessation); alcohol drinking frequency was divided into high and low frequency -depending on whether 3 or more times per week. Physical activity was divided into active and inactive groups depending on the amount of time people spent and de ned as: 0 = no activity, 1 = less than 1 day per month, 2 = 1-3 days per month, 3 = 1-2 days per week, 4 = 3-4 days per week, 5 = 5-6 days per week and 6 = every day, and we combined the 0-2 as 0, 3-6 as 1. Past medical history including hypertension, diabetes, heart disease and cerebral infarction were all collected and de ned as: 0 = none and unknown, 1 = did have. The missing values of medical history were all coded as 0, which accounts for 11.4% for the total.

Cognitive Assessment
The global cognitive assessment was measured by the Mini-Mental State Examination (MMSE), which includes orientation, registration, attention and calculation, recall and language domains. Since the education year of the participants in the study varied, we divided the participants by using both education year and MMSE scores. Based on a previous study of a Chinese residential population aged 55 years and over in the urban and rural areas of Beijing [23], the cutoff for determining the presence of cognitive impairment is 19 for illiterate, 22 for elementary and 26 for more than high school. The short version of the Montreal Cognitive Assessment-Beijing (MoCA-s) was used to assess the cognitive status of the participants. The MoCA-s was revised by our research group, which contains 15 items. The total score is 17, higher scores indicate better cognitive status. The discrimination of MoCA-s was con rmed equal to the full version. The Ascertain Dementia 8 (AD8) questionnaire with 8 questions was developed by Washington University. It is a screening tool for differentiating non-demented from demented individuals (even at the very early stage) and used in our project as a self-rated test. The higher score of AD8 indicate higher risk of dementia. A series of tests were used to detect different cognitive domains, including (1) working memory (Digit Span Backwards, DSB) [24]; (2) attention (Digital Symbol Substitution Test, DSST) [25]; (3) episodic memory (Paired Associative Learning We used the Center for Epidemiological Studies-Depression (CESD) scale [27] to assess the emotional state of the elderly. Higher scores indicate higher depression level.

Statistical Analysis
Differences in the demographic characteristics and cognitive tests of each group were evaluated by the t or X 2 test. The Fisher's exact test was also used to assess whether the genotypic frequencies conformed to Hardy-Weinberg equilibrium (HWE). And the Pearson's Chi-square test was used to compare the genotypic and allelic frequencies between the cognitive impairment and control group. Differences in groups and cognitive function between the participants carrying the different genotypes were calculated by the Mann-Whitney U test. Logistic regression analyses, adjusting for gender, age, education year, smoking status, alcohol drinking frequency, physical activity, hypertension, diabetes, heart disease, cerebral infarction, CESD and APOE status (except for rs429358 and rs7412), were used to estimate 95% con dence interval (CI) and ORs under additive genetic models which was de ned as 0 (AA) vs 1 (Aa) vs 2 (aa). The signi cance of SNP and APOE interaction was also tested for each additive model of the SNP by logistic regression. Multiple linear regression analyses, adjusting for gender, age, education year, smoking status, alcohol drinking frequency, physical activity, hypertension, diabetes, heart disease, cerebral infarction, CESD and APOE status (except for rs429358), were used to estimate the association between additive model and different cognitive domains. All the analyses were conducted by using SPSS 22.0 software, and two-tailed p < 0.05 was considered statistically signi cant.
To further investigate the interactions betweem gene-gene and gene-lifestyle, we employed the generalized multifactor dimensionality reduction (GMDR) method [29], which permits adjustment for covariates and provides a uni ed framework for coherently handling both dichotomous and quantitative phenotypes. We used 10-fold cross-validation to test two to four-way interactions. Here are some output parameters from GMDR, such as testing accuracy, sign test and cross-validation consistency. The age, gender, and education year were used as covariates for gene-gene and gene-lifestyle interaction analyses (including smoking status, alcohol drinking frequency and physical activity), and we also used both logistic and linear regression models ( using MMSE as the dependent variable) in the interaction analyses. Statistical signi cance was set at twotailed p < 0.05.

Characteristics of the study group
The baseline characteristics of all participants in the study are summarized in Table 1. We studied 1924 ethnic Han community-dwelling elderly including a total of 411 cognitively impaired participants (21.4%). There were signi cant differences between groups in terms of age (p < 0.001), gender (p = 0.002), education year (p = 0.001), and all the cognitive tests (p < 0.001) and CESD score (p = 0.003). The frequency of the APOE ε4 allele was higher in the cognitive impairment group than in the normal group (χ 2 = 12.76, p < 0.001), which indicates that APOE ε4 increases the risk of cognitive impairment in the Han community-dwelling elderly.

Allele And Genotype Distribution Of The Snps
All these SNPs were consistent with the Hardy-Weinberg equilibrium in both normal and cognitive impairment groups (p > 0.05) (see supplementary material Table S1). Here we observed signi cant differences in the minor allele frequency of rs2075650 (p = 0.047), rs429358 (p = 0.002) and rs6439886 (p = 0.040) between normal and cognitive impairment groups. We detected the ORs of 1.309, 1.557 and 0.717 for rs2075650, rs429358 and rs6439886, respectively.
According to APOE ε4 status, we strati ed the participants into APOE ε4 allele carrier and non-carrier subgroups (see supplementary material Table S2). The allele frequency and genotype distribution of SNPs in each subgroup are supplied. For rs1047631, the distribution of genotype showed signi cantly difference between normal and cognitive impairment groups in APOE ε4 carriers (p = 0.020). For rs11046205, the frequency of genotype and allele showed signi cant difference between normal and cognitive impairment groups in APOE ε4 carriers, with the minor allele (A allele) signi cantly decreasing the cognitive impairment risk (genotype: p = 0.003, allele: OR = 0.387, 95%CI = 0.191-0.784, p = 0.008). For rs2070045, the genotype distribution showed a signi cant difference between normal and cognitive impairment groups in both the whole group (p = 0.010) and the APOE ε4 non-carriers group (p = 0.018). For rs2075650, the genotype distribution differed signi cantly between normal and cognitive impairment groups (p = 0.034). For rs405509, allele frequency and genotype distribution showed signi cantly difference between normal and cognitive impairment groups in APOE ε4 carriers, with the minor allele (G allele) signi cantly increasing the cognitive impairment risk (genotype: p = 0.001, allele: OR = 2.242, 95%CI = 1.397-3.600, p = 0.001). For rs429358 (one of the APOE SNPs), the allele frequency and genotype distribution showed signi cant difference between normal and cognitive impairment groups, with the minor allele (C allele) signi cantly increasing the cognitive impairment risk (genotype: p = 0.001, allele: OR = 1.611, 95%CI = 1.217-2.132, p = 0.001). For rs6439886 and rs7412, both the genotype distribution and minor allele frequency showed signi cant differences between the two groups even after strati cated by APOE ε4 status (p > 0.05). We further analyzed the impact of rs11046205/rs405509 on the APOE ε4 risk for cognitive impairment (table   3). After adjusting for rs11046205, the OR of APOE ε4 decreased from 1.759 to 1.752 (P < 0.001). When adjusted for rs405509, the OR of APOE ε4 increased from 1.759 to 1.816 (P < 0.001).

Effect Of Snp Genotype On Cognitive Test Scores
Selected by logistic regression, ABCC9 (rs11046205) and APOE (rs429358) were included in these analyses.
Since there is only 1 participant that had the C/C genotype in rs429358, we regrouped the two SNPs into AA or a-carrier (A: major, a: minor), and compared the different cognitive scores in the two groups (see supplementary material Table S3). The scores of the MMSE, MoCA-s, AD8, DSB, DSST and PALT were strati ed according to genotype. For rs429358, participants carrying the C allele had signi cantly higher AD8 score than non-carriers in the cognitive impairment group (p = 0.002, Fig. 1). There is no signi cant difference in cognitive scores between the genotypes of rs11046205.

Snp And Phenotype Association Analysis
The association analyses were performed using an additive model adjusted by age, gender, educational year, smoking status, alcohol drinking frequency, hypertension, diabetes, heart disease, cerebral infarction, physical status, CESD and APOE status (except for rs429358). For rs429358, there was a signi cant negative association between the additive model and MMSE score (p = 0.003). None of the other SNPs showed an association with cognitive scores (p > 0.05) (table 4).

Gene-gene And Gene-lifestyle Interaciton
Then, we selected two key SNPs into the subsequent analyses with evidence of cognitive impairment association, including ABCC9 rs11046205 (p = 0.039) and APOE rs429358 (p < 0.001). We used GMDR analysis to assess the gene-gene and gene-lifestyle interaction between the two SNPs and smoking, alcohol drinking frequency and physical activity -including gender, age and education year as covariates. Cognitive impairment and MMSE score were employed as outcome for both gene-gene and gene-lifestyle interaction analysis. Table 5 summarizes the results obtained from the GMDR analysis for gene-gene interaction models after adjusting covariates. In the logistic regression analyses, there were signi cant interaction models involving rs11046205 and rs429358 (p = 0.0107, CV = 10/10). This result indicate potential gene-gene interactions between ABCC9 and APOE on in uencing cognitive impairment. In the linear regression analyses, there were signi cant interaction models involving rs429358 and physical activity (p = 0.0107, CV = 6/10), indicating potential gene-lifestyle interactions between APOE (rs429358) and lifestyles on in uencing general cognitive function (MMSE). (Table 6) Discussion Due to the increased prevalence of AD in the elderly, relevant risk genes for AD are widely utilized in cognitive aging research, such as BDNF, TOMM40, COMT, DTNBP1 and SORL1 [6, 30]. These related risk genes have been identi ed by genomewide association studies (GWASs) in many populations [31]. Evidence con rms that the most well replicated gene is APOE, which accounts for 50% or less in the risk of developing AD [32]. These results remind us of the polygenic trait character of the disease and the importance of prevention -not only identi cation; we should not only focus on the genes directly associated with AD, but the ones related with general cognitive function and simultaneous phenomena. In this study, we veri ed the association between AD risk genes and cognitive impairment in our Han community-dwelling elderly. The allele or genotype frequency of APOE (rs429358), APOE promoter (rs405509), TOMM40 (rs2075650) and SORL1 (rs2070045) showed a signi cant difference between normal and cognitive impairment groups and APOE ε4 strati ed subgroups. Logistic regression analyses showed a signi cant correlation between APOE (rs429358), ABCC9 (rs11046205) and cognitive impairment and an interaction between ABCC9 (rs11046205)/APOE promoter (rs405509) and APOE ε4 status. A conditional analysis test was used to nd whether APOE ε4 has an association independent of the ABCC9 (rs11046205)/APOE promoter (rs405509) SNPs. After adjusting the covariates and ABCC9 (rs11046205)/APOE promoter (rs405509) SNPs, there is still signi cant risk for APOE ε4 in cognitive impairment. The ABCC9 (rs11046205) could reduce the risk of APOE ε4, and APOE promoter (rs405509) could increase the risk of APOE ε4 in cognitive impairment. We further substantiated the linkage between these risk genes and cognitive phenotypes. For the additive model of APOE (rs429358) showed a signi cant negative association with MMSE scores. Additionally, our data revealed that gene-gene interaction between ABCC9 and APOE may contribute to the etiology of cognitive impairment.
As the 2 SNPs relevant to classifying the APOE genotype, the rs429358 (not rs7412) signi cantly increased the risk of cognitive impairment in the Han community-dwelling elderly in our data, with an OR of 1.639 (1.223-2.196), and the interaction between the APOE ε4 status and APOE promoter (rs405509) could increase the OR of APOE ε4 from 1.759 to 1.816. TOMM40 (rs2075650) in our sample could also increase the risk of cognitive impairment with an OR of 1. 309(1.007-1.702). The genes we conducted above were all in the APOE/TOMM40 region, which has recently been proven to be associated with cognitive aging [8,33]. There is no clear boundary between cognitive aging and cognitive impairment, the genetic contribution to 'normal' and 'pathological' cognitive variation in old age overlaps [8]. Here we con rmed these results in cognitively impairment elderly in Han community-dwelling in Beijing.
The DTNBP1 (rs1047631) and CLSTN2 (rs6439886) were two genes which have been proven to be associated with general cognitive function (working memory and episodic memory) [34]. In our data, the genotype of rs1047631 and minor allele of rs6439886 polymorphisms showed protective effect in the development of cognitive impairment in the APOE ε4 carrier and whole sample, respectively. Up to now, the evidence for DTNBP1 mainly focused on Schizophrenia [35,36], which could in uence the risk of the disease. However, recently the variation of DTNBP1 has also been linked with human cognitive functioning, although the evidence is inconsistent [37]. The association studies differed across study population (vary in sample size, ethnic composition, age and educational background), which could add complexity to the genotype-phenotype association, but were generally observed for memory traits in the two older cohorts [38]. Our results are consistent with the situation above and another research in Japanese older adults [39]. The CLSTN2 (rs6439886) gene has con rmed the association with episodic memory by GWAS studies in young adults (age range: 18-48 years) [40]. The protective effect of CLSTN2 (rs6439886) on cognitive function has been proven in both young adults and the elderly, which was consistent with our results [41,42].
We found the additive model of the ABCC9 (rs11046205) gene had a protective effect for the development of cognitive impairment (OR = 0.775, 95%CI = 0.607-0.900), and the interaction between ABCC9 and the APOEε4 status could reduce the OR of APOE ε4 from 1.759 to 1.752. As an ATP-sensitive potassium channel, the ABCC9 gene has been proven to be related with sleep duration in samples from two large European consortiums (n = 4251 and n = 5949) [43,44]. The minor allele (A-allele) distribution showed protective effect in the development of cognitive impairment in the APOE ε4 carrier, which is the same genotype that has been previously associated with longer sleep duration [44]. The U-shape association between cognitive impairment and night sleep duration has been con rmed in a study of Northern Chinese elderly, which means the cognitive impairment was linked to getting either too little or too much sleep [19]. We cannot assess the sleep duration of the A-allele represented, but it should not exceed the normal time horizon since it showed a protective effect for people with depression, which is one of the risk factors for AD [22]. Given the high incidence of disturbed sleep in cognitive impairment (60% in dementia and MCI) [21], it is reasonably believe that the same SNP which increases sleep duration having a protective effect on cognitive impairment. To our knowledge, this is the rst association of cognitive impairment-related traits with an ATP-sensitive potassium channel gene, which may suggest that this group of potassium channels serves as an additional pathway to affect the development of cognitive impairment.
In the GMDR analysis of gene-gene interactions, we found the interaction between ABCC9 (rs11046205) and APOE (rs429358) with cognitive impairment (p < 0.05). ABCC9 is a K(ATP) channel gene which affects sleep duration [44], and we have proven the protective effect of rs11046205 in cognitive impairment above. The interaction between APOE (rs429358) and ABCC9 (rs11046205) here further proofed the underlying effect of potassium channels in cognitive impairment. Here we also tracked down the interplay between MMSE scores (general cognitive function) and APOE (rs429358) with a kind of lifestyles, which is physical activity. The result were consistent with others' [45]. Evidence showed that APOE-lifestyle interaction could increase the risk of late-life cognitive decline by 11 times, in which sedentary lifestyles is included [46]. It was well known that the multifactorial characteristic of cognitive impairment including genetic, lifestyle and environmental factors [47] suggesting the essential role of lifestyle and gene-lifestyle interaction in the future studies.
Several limitations of this study should be acknowledged. First, we didn't make a distinction between MCI and cognitive impairment in our study participants. Second, we didn't examine the interaction between genetic variants and measurable physiology, such as blood lipids and glucose, which may add more information to the pathological character of cognitive impairment [48]. Further studies may add more physiological data into account to reveal potential interactions.
In conclusion, the current study support the association of APOE (rs429358), APOE promoter (rs405509) and

Acknowledgments
We are grateful to all the participants who agreed to participate in this study and the fully support by the staff of local hospital.

Funding
This study was supported by grants from the Beijing Municipal Science & Technology Commission (Z171100000117006). The funding supported the collection of the participants' data.

Availability of data and materials
The datasets used and analyzed in the current study are available from the corresponding author on reasonable request.  Adjusted for age, gender, education year, APOE status (except for rs429358), smoking status, alcohol drinking frequency, physical activity, hypertension, diabetes, heart disease, cerebral infarction and CESD. Table 3 The association between APOE ε4 and cognitive impairment. Models Total sample P a OR(95%CI) P b OR(95%CI) P c OR(95%CI) APOE ε4 <0.001 1.759(1.296-2.386) <0.001 1.752(1.290-2.378) <0.001 1.816(1.333-2.474) a Adjusted for age, gender, education year, smoking status, alcohol drinking frequency, physical activity, hypertension, diabetes, heart disease, cerebral infarction and CESD. b Adjusted for age, gender, education year, smoking status, alcohol drinking frequency, physical activity, hypertension, diabetes, heart disease, cerebral infarction, CESD and rs11046205 additive model. c Adjusted for age, gender, education year, smoking status, alcohol drinking frequency, physical activity, hypertension, diabetes, heart disease, cerebral infarction, CESD and rs405509 additive model.  Table 5 Gene-gene interaction models identified by the GMDR method (logistic regression).