The association of Alzheimer’s disease-related SNPs with mild cognitive impairment susceptibility in the Chinese population

Abstract

Previous Genome-wide association studies have identified several single nucleotide polymorphisms (SNPs) associated with Alzheimer’s disease (AD), whereas their associations with mild cognitive impairment (MCI) remain unclear. To evaluate the associations between 100 representative AD-associated SNPs and susceptibility to MCI in the Chinese population. We recruited 200 MCI patients and 200 cognitively-healthy controls from the community, matched for age and sex. Associations between SNPs and MCI risk were estimated using lasso regression, adjusted for APOE status, using different genetic models. Fifteen SNPs in nine genes (including CLU, SORL1, PICALM, BDNF, NOS3, MTHFR, TOMM40, BIN1, and PVRL2) were associated with MCI in single-SNP analysis. In the multi-SNP association test, rs1801133 and rs9331888 of CLU were consistently associated with MCI risk in the dominant model. TOMM40 rs2075650 (G) was associated with MCI risk in the dominant model by age and education (OR = 2.41, 95%CI = 1.27–4.59), but disappeared when further adjusted for APOEε4 status. PICALM rs561655 (G) (OR = 0.52, 95%CI = 0.30–0.92) and NOS3 rs1549758 (T) (OR = 0.53, 95%CI = 0.30–0.94) were identified as protective genetic factors of MCI for the first time in dominant model combined with the APOEε4 allele. Moreover, MTHFR rs1801133 (A) and CLU rs9331888 (G) showed more susceptibility to MCI in the additive model. SORL1 rs641120(G) showed a protective effect, whereas BIN1 rs5733839 consistently showed a risk effect for MCI in the overdominant model, regardless of APOEε4 status. This study suggests that some AD-associated SNPs are associated with cognitive decline and may have important implications for future studies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37309-0.

Introduction

Alzheimer’s disease (AD) is a neurodegenerative disease and the most prevalent age-related dementia, adversely affecting more than 55 million people worldwide¹. The prevalence of AD is still increasing dramatically, especially in developing countries, and is projected to nearly triple by 2050². The continuum of AD spans 15 to 25 years and includes various clinical stages from preclinical, mild cognitive impairment (MCI) to dementia stages³. As an intermediate stage on the continuum of AD, MCI is defined as an pre-dementia stage for AD patients, characterized by subtle cognitive deficits in one or more domains such as memory, attention, language, or executive function, which are greater than expected for an individual’s age and educational background but do not significantly impair daily activities and independence⁴. MCI patients have been reported to progress to AD at an annual rate of approximately 10%⁵. After a six-year observation period, the cumulative rate of conversion from MCI to AD could reach 80%⁶. WHO Guidelines to reduce the risk of cognitive impairment and dementia recommend multiple factors for intervention before symptoms manifest⁷. Future directions of AD treatment also emphasize the importance of multidisciplinary lifestyle intervention that combines pharmacological and nonpharmacological approaches at a relatively earlier stage compared with current studies (mostly at the dementia stage)⁸. Thus, early detection, diagnosis, and treatment of MCI patients may prevent AD progression.

To date, studies have tended to view AD as a disease with multiple causes, with onset associated with both environmental and genetic factors⁹. A twin study found a high heritability for AD, estimated at about 60% to 80%¹⁰. In addition to the strongest genetic risk factor for AD, the APOEε4 allele, several large genome-wide studies in different ethnic populations have identified more than 40 genes associated with AD through different pathways¹¹. There are several established AD-related genes, such as amyloid-β precursor protein (AβPP), sortilin-related receptor 1 (SORL1), clusterin or apolipoprotein J (CLU or APOJ), complement receptor 1(CR1), etc^12–14. Most of these genes are crucial for the production or degradation of amyloid-β (Aβ) metabolism or tau tangles, a landmark for diagnosing the presence of AD^15,16. In addition to Aβ and tau metabolism, genes that play important roles in immune response, lipid dysfunction, endocytosis, and vascular disease also play a role in the development of AD, modifying the hypothesis of a single amyloid cascade for the disease¹⁷. Identification of novel loci is critical to understanding the complicated etiology of a heterogeneous disease.

Compared with the number of studies about the association between genetic risk factors and AD or MCI to AD stage, rather fewer relevant studies paid particular attention to the appearance of MCI solely^18,19. Additionally, most current studies were carried out on individuals of European descent²⁰, and the repeating results from the Chinese population could be a good validation. The effects of variants of different populations enable us to better interpret the disease mechanism²¹. Thus, this study aimed to evaluate whether well-established AD-associated SNPs, originally identified in GWAS of clinically diagnosed AD patients, already exert measurable effects at the MCI stage in a Chinese community-based sample. Rather than seeking to identify MCI-specific loci across all etiological subtypes, our study focused on AD-related genetic architecture that may contribute to early cognitive decline.

Materials and methods

Participants

The population-based cross-sectional community study, which was carried out in conjunction with the elderly medical examination program, recruited 2287 adults (including males and females, aged 60 years and older) from Dapuqiao Street, Huangpu District, Shanghai, China, from March 2019 to November 2019. For the purpose of this study, 200 MCI patients and 200 cognitively-healthy age/sex-matched controls, who had been living in the community for at least 12 months preceding data collection according to the Family Doctor Contract Service Information System, were selected. The sample size is enough to achieve 80% statistical power in genetic models under assumptions of 15.6% disease prevalence (approximately the prevalence of MCI in China)²², 5% minor allele frequency (MAF) (exclusion criteria for SNP), complete LD, 1:1 case-to-control ratio, and 5% type I error rate (α) according to a review of sample size and statistical power calculation in previous genetic association studies²³. All participants were required to sign the informed consent form that contained information such as the purpose and content of the study, the possible risks and benefits, privacy protection, and the collection of blood samples for scientific research. Data collection and informed consent templates for this study were approved by the Public Health and Nursing Research Ethics Review Committee, Shanghai Jiao Tong University School of Medicine (SJUPN-202103-X1).

Demographic information and cognition assessment

The general demographic information includes age, sex, education, smoking status, drinking status, exercise, activities of daily living (ADL), self-assessments of health status, hearing loss, hypertension, diabetic mellitus, and other indicators of physical examination. The Mini-Mental State Examination (MMSE) was used to assess global cognitive function, which includes orientation, immediate and delayed word recall, attention and calculation, language, and visual space. The diagnosis of MCI was made based on medical history, physical examination, and MMSE (total score < 24) by one or more experts in dementia²⁴.

DNA extraction, SNP selection, and genotyping

DNA was extracted from white cells in the blood sample using QIAsymphony SP (QIAGEN GmbH, QIAGEN Strasse1,40724 Hilden, Germany), which were stored in a vacutainer with ethylene diamine tetra acetic acid (2 ml) under − 80℃ until genotyping. Single-nucleotide polymorphism (SNP) genotyping was performed using DNA for rs429358 (T, C) and rs7412 (C, T), which determine the APOE status (TT: ε2; TC: ε3; CC: ε4). Genotyping was performed on the Real-Time PCR 7500 System (Applied Biosystems, Thermo Fisher Scientific, Carlsbad, United States, RRID: SCR_018051). In addition to that, other SNPs’ selection was based on a review of the literature. Specifically, we selected 100 SNPs that had been repeatedly identified in large-scale AD-related GWAS or meta-GWAS studies^11,20,25, prioritizing variants with biological plausibility, cross-population replication, and functional annotation. These SNPs were selected to balance relevance and statistical feasibility within our sample size. All selected loci were detected using the Hi-SNP high-throughput genotyping method with technical support from the Shanghai Biowing Applied Biotechnology CO.LTD (www.biowing.com.cn). Three rounds of multiplex PCR with next-generation sequencing were used for genotyping²⁶. Amplification of primer sequences was conducted by Primer3 online software, Version 0.4.0 (http://frodo.wi.mit.edu/). All selected SNPs and their primers were shown in Supplemental Table 1. After multiplex PCR, 5ul products were mixed in a centrifuge tube, then the tube was sealed with a parafilm. After mixing overnight, such mixture was purified by a TIANgel Midi Purification Kit (Tiangen Biotech, China) and then sequenced on Illumina X-10, the high-throughput genotyping platform.

One hundred representative AD-associated SNPs were selected, which were categorized in several etiological mechanisms as followed: (1) synaptic function; (2) Cholesterol metabolism; (3) Neuroinflammation; (4) Aβ production; (5) Tau; (6) cell adhesion; (7) Energy and biomacromolecule metabolism; (8) endocytosis; (9) epigenetic and others^27–32. These categories are intended as a heuristic framework to facilitate interpretation and discussion, rather than as rigid or mutually exclusive classifications, acknowledging that many of the biological processes involved in AD pathogenesis are highly interconnected.

Statistical analysis

Allele and genotype frequencies were calculated and Hardy-Weinberg equilibrium (HWE) was tested using Pearson’s Chi-squared test. The quantitative variables are presented as mean (SD) and qualitative data are described as frequency and percentage. Chi-squared test and t-test were used to further tease out the observed association between cognitive status and categorical and continuous variables, respectively. The risk of each SNP was estimated using logistic regression analysis by adjusting for age and education (Model 1); further adjusting for age, education, and APOEε4 status (Model 2), and four different genetic models were applied. The following definitions were used assuming A represents the major allele and a represents the minor allele: (1) dominant, AA vs. Aa + aa, (2) Recessive, AA + Aa vs. aa, (3) Additive, AA vs. Aa vs. aa, (4) overdominant, Aa vs. AA + aa. The additive, dominant, recessive, and overdominant genetic models were applied to investigate potential heterogeneity in genetic effects across inheritance modes. The positive status of APOEε4 (APOEε4 carriers) was defined as carrying one or more ɛ4 alleles. Three haplotypes (ɛ2(388 T-526 T), ɛ3(388 T-526 C), ɛ4(388–526 C)) were judged by a combination of rs7412 and rs429358 genotype³³. The data were stratified by exercise, gender, education, and diabetes mellitus for subgroup analysis. The Lasso regression model was used to select the potential susceptibility SNPs of MCI by adjusting for age and education (Model 1); further adjusting for age, education, and APOEε4 status (Model 2). False discovery rates (FDR) correction was applied for multivariable analyses and adjusted 2-sided P < 0.05 was used to determine statistical significance. We also performed subgroup analyses to explore potential effect modification of genetic risk by key covariates, including physical activity, diabetes mellitus status, sex, and educational attainment, which represent well-established risk factors for cognitive decline in older adults³⁴. Finally, to strengthen the robustness of our findings, we performed two sensitivity analyses: (1) additional adjustment for sex, because the population is disproportionately female; (2) setting the threshold at an MMSE score of 23 for MCI classification. All analyses were conducted with SPSS software package version 22.0 (SPSS Inc., Chicago, IL, United States) and the statistical program R 4.1.0 (www.r-project.org).

Results

Characteristics of participants

The characteristics of 200 MCI patients and 200 age and sex-matched healthy controls were shown in Table 1. The mean age was 78.42 years (SD = 7.02) and 70.50% were women in the MCI group. The mean age was 77.58 years (SD = 6.90) and 70.00% were women in the control group. MCI patients gained lower MMSE scores (20.29 ± 3.55) than controls (27.48 ± 1.96) with P less than 0.001. MCI was significantly associated with lower education level (P < 0.001), fewer exercise (P = 0.017), and APOEε4 carriers (P = 0.009). There were no statistically significant differences between MCI patients and controls in age, sex, smoking status, and drinking status, ADL, self-assessments of health status, hearing loss, hypertension, and diabetic mellitus. The distribution of MMSE scores across groups is provided in Supplementary Fig. 1.

Table 1.

Demographics and characteristics of the study populations. MMSE, Mini-Mental state Examination; APOEε4 carriers, people with more than one copy of ε4 allele.

	MCI (N = 200)	Control (N = 200)	P value
Age, mean ± SD, y	78.42 ± 7.02	77.58 ± 6.80	0.228
Gender, No. (%)			0.887
Female	141 (70.50)	140 (70.00)
Male	59 (29.50)	60 (30.00)
Education, No. (%)			< 0.001
Illiteracy	42 (21.99)	2 (1.03)
Primary school	41 (21.47)	22 (11.34)
Junior high school	55 (28.80)	60 (30.93)
Senior high school	33 (17.28)	41 (21.13)
University or above	20 (10.47)	69 (35.57)
Smoking, No. (%)			0.853
Never	186 (93.00)	186 (93.47)
Ever and current	14 (7.00)	13 (6.53)
Drinking, No. (%)			0.594
No	183 (91.50)	179 (89.95)
Yes	17 (8.50)	20 (10.05)
Exercise, No. (%)			0.194
No	94 (53.50)	107 (46.76)
Yes	106 (46.50)	93 (53.23)
Self-care ability, No. (%)			0.337
Independence	193 (96.50)	197 (98.50)
Partial dependence	7 (3.50)	3 (1.50)
Self-satisfaction, No. (%)			0.293
Not clear	8 (4.08)	5 (2.50)
Dissatisfied	6 (3.06)	12 (6.00)
Satisfied	182 (92.86)	183 (91.50)
Hearing, No. (%)			0.173
Hearing loss or disorder	23 (11.50)	15 (7.50)
Normal hearing	177 (88.50)	185 (92.50)
Hypertension, No. (%)			0.613
No	86 (43.00)	81 (40.50)
Yes	114 (57.00)	119 (59.50)
Diabetic Mellitus, No. (%)			0.231
No	150 (75.00)	160 (80.00)
Yes	50 (25.00)	40 (20.00)
MMSE score, mean ± SD	20.29 ± 3.55	27.48±1.96	< 0.001
APOEε4 carriers, No. (%)			0.009
Non-carriers	137 (78.74)	179 (89.50)
Carriers	37 (21.26)	21 (10.50)

Genotype distribution

The frequency distributions of alleles in MCI patients and controls were detailed in Table 2. The MAF of SNPs were all above 10%, except rs2075650 (MCI: 0.11, control: 0.07). We found significant differences in the allele frequencies of rs1801133 (OR = 1.51, 95% CI = 1.13–2.01, P = 0.005), rs2030324 (OR = 1.41, 95% CI = 1.07–1.87, P = 0.016) and rs2075650 (OR = 1.68, 95% CI = 1.02–2.79, P = 0.042) between MCI patients and controls. In addition, the allele distributions of CLU (rs11136000, rs867230, rs9331888 and rs9331896, except rs1532278) were found to be significantly associated with MCI. No deviations from Hardy-Weinberg equilibrium (HWE) for the potential polymorphisms were observed (Supplemental Table 2).

Table 2.

Allele distribution of SNPs in MCI and control group.

Gene		SNPs	Chr.	MA	P allele	OR (95%CI)	MAF
							Control	MCI
Neuroinflammation	NOS3	rs1549758	7	T	0.627	0.91 (0.62, 1.34)	0.18	0.16
Synaptic function	BDNF	rs2030324	11	G	0.016	1.41 (1.07, 1.87)	0.41	0.49
Amyloid-β	SORL1	rs2070045	11	T	0.180	1.21 (0.92, 1.60)	0.42	0.46
	SORL1	rs641120	11	G	0.179	1.21 (0.92, 1.58)	0.46	0.50
	SORL1	rs11218343	11	C	0.645	1.07 (0.79, 1.45)	0.29	0.31
	PICALM	rs561655	11	G	0.056	0.76 (0.58, 1.01)	0.53	0.46
	CLU	rs11136000	8	T	0.039	0.70 (0.50, 0.98)	0.25	0.19
	CLU	rs1532278	8	T	0.052	0.71 (0.51, 1.00)	0.24	0.19
	CLU	rs867230	8	C	0.030	0.69 (0.49, 0.97)	0.25	0.19
	CLU	rs9331888	8	G	< 0.001	1.70 (1.28, 2.59)	0.38	0.51
	CLU	rs9331896	8	C	0.005	0.62 (0.44, 0.87)	0.26	0.18
Metabolism	MTHFR	rs1801133	1	A	0.005	1.51 (1.13, 2.01)	0.35	0.45
	TOMM40	rs2075650	19	G	0.042	1.68 (1.02, 2.79)	0.07	0.11
Endocytosis	BIN1	rs6733839	2	T	0.295	1.61 (0.88, 1.54)	0.43	0.46
	PVRL2	rs6859	19	A	0.634	1.08 (0.80, 1.45)	0.32	0.34

MA minor allele, MAF minor allele frequency, Chr. Chromosome, P allele p.value of allele.

Single SNPs association with MCI

The associations between AD-associated SNPs and MCI risk using four genetic models were illustrated in Table 3, and did not change materially after FDR correction (Supplemental Table 3). Notably, these associations remained robust in sensitivity analyses accounting for sex and stricter MCI classification (Supplemental Tables 4 and Supplemental Table 5).

Table 3.

Association of target SNPs and adjusted odds ratio to MCI risk.

Gene	SNPs	Model 1, OR (95%CI)*				Model 2, OR (95%CI)*
		Dominant	Recessive	Additive	Overdominant	Dominant	Recessive	Additive	Overdominant
NOS3	rs1549758	0.78 (0.47,1.28)	2.39 (0.41,13.93)	0.87 (0.56, 1.36)	0.71 (0.42, 1.18)	0.64 (0.38,1.10)	2.91 (0.48,17.65)	0.76 (0.47, 1.23)	0.56 (0.32, 0.98)
BDNF	rs2030324	1.46 (0.88, 2.40)	2.20 (1.22, 3.96)	1.54 (1.10, 2.16)	0.85 (0.54, 1.32)	1.43 (0.84, 2.42)	2.05 (1.11, 3.78)	1.49 (1.04, 2.12)	0.87 (0.55, 1.39)
SORL1	rs11218343	0.87 (0.56, 1.35)	2.61 (1.02, 6.71)	1.07 (0.75, 1.52)	0.68 (0.44, 1.06)	0.87 (0.55, 1.38)	2.58 (0.97, 6.86)	1.06 (0.73, 1.55)	0.68 (0.43, 1.09)
SORL1	rs2070045	1.06 (0.65, 1.71)	1.72 (0.97, 3.05)	1.22 (0.89, 1.68)	0.75 (0.48, 1.17)	1.04 (0.62, 1.72)	1.81 (0.99, 3.30)	1.23 (0.88, 1.73)	0.72 (0.45, 1.15)
SORL1	rs641120	0.85 (0.51, 1.40)	1.82 (1.07. 3.10)	1.16 (0.85, 1.59)	0.58 (0.37, 0.91)	0.77(0.45, 1.31)	2.03 (1.17, 3.54)	1.17 (0.84, 1.63)	0.49 (0.30, 0.79)
PICALM	rs561655	0.58 (0.35, 0.95)	0.77 (0.47, 1.27)	0.74 (0.55, 1.01)	0.80 (0.51, 1.24)	0.57 (0.34, 0.96)	0.85 (0.50, 1.43)	0.76 (0.56, 1.05)	0.73 (0.46, 1.16)
CLU	rs11136000	0.62 (0.39, 0.97)	0.71 (0.24, 2.17)	0.67 (0.46, 0.99)	0.64 (0.40, 1.01)	0.60 (0.37, 0.97)	0.45 (0.13, 1.58)	0.62 (0.41, 0.94)	0.67 (0.41, 1.08)
CLU	rs1532278	0.62 (0.40, 0.98)	0.84 (0.27, 2.66)	0.69 (0.47, 1.02)	0.63 (0.40, 0.99)	0.60 (0.37, 0.97)	0.50 (0.14, 1.83)	0.63 (0.41, 0.96)	0.65 (0.40, 1.06)
CLU	rs867230	0.59 (0.38, 0.93)	0.87 (0.29, 2.58)	0.67 (0.46, 0.99)	0.59 (0.37, 0.94)	0.58 (0.36, 0.93)	0.55 (0.16, 1.87)	0.62 (0.41, 0.94)	0.62 (0.38, 1.01)
CLU	rs9331888	2.20 (1.33, 3.63)	2.58 (1.42, 4.69)	1.93 (1.38, 2.71)	1.09 (0.70, 1.71)	2.97 (1.70, 5.18)	2.32 (1.25, 4.30)	2.13 (1.48, 3.06)	1.41 (0.88, 2.27)
CLU	rs9331896	0.55 (0.34, 0.87)	0.58 (0.17, 2.00)	0.59 (0.40, 0.89)	0.58 (0.36, 0.93)	0.53 (0.33, 0.87)	0.27 (0.06, 1.24)	0.54 (0.35, 0.83)	0.62 (0.38, 1.00)
MTHFR	rs1801133	1.77 (1.11, 2.84)	1.73 (0.91, 3.27)	1.55 (1.11, 2.17)	1.28 (0.82, 2.00)	1.93 (1.17, 3.18)	1.91 (0.98, 3.73)	1.67 (1.17, 2.38)	1.30 (0.82, 2.07)
TOMM40	rs2075650	2.08 (1.14, 3.81)	0.86 (0.03,28.24)	1.97 (1.10, 3.52)	2.12 (1.15, 3.91)	1.20 (0.44, 3.23)	0.46 (0.01,16.33)	1.12 (0.43, 2.89)	1.26 (0.48, 3.30)
BIN1	rs6733839	1.42 (0.87, 2.32)	0.72 (0.40, 1.31)	1.06 (0.76, 1.48)	1.61 (1.03, 2.51)	1.37 (0.82, 2.29)	0.71 (0.37, 1.33)	1.04 (0.74, 1.47)	1.58 (0.99, 2.53)
PVRL2	rs6859	1.40 (0.90, 2.19)	0.51 (0.25, 1.04)	1.04 (0.75, 1.44)	1.85 (1.18, 2.91)	1.10 (0.67, 1.81)	0.43 (0.20, 0.92)	0.86 (0.60, 1.23)	1.58 (0.97, 2.56)

*OR of model 1: odds ratio adjusted by age and education. *OR of model 2: odds ratio adjusted by age, education and APOEε4 carriers. Only SNPs found to be significant in at least one single genetic model or utilization in multi-genetic models were shown. Bold indicates statistically significant.

In the dominant model, CLU rs11136000 (T), rs1532278 (T), rs867230 (C), and rs9331896 (C) and PICALM rs561655 (G) were associated with decreased risk of MCI. Meanwhile, CLU rs9331888 (G), MTHFR rs1801133 (A), and TOMM40 rs2075650 (G) were associated with increased risk of MCI after adjusting for age and education (Model 1). After further adjusting for APOEε4 carriers, the association between TOMM40 rs2075650 (G) and MCI was disappeared (Model 2).

In the recessive model, BDNF rs2030324 (G), SORL1 rs641120 (G), SORL1 rs11218343 (C) and CLU rs9331888 (G) were associated with increased risk of MCI (Model 1). After further adjusting for APOEε4 carriers, SORL1 rs11218343 (C) was no longer associated with MCI, while PVRL2 rs6859 (A) showed a significant association with decreased risk of MCI (Model 2).

In the additive model, CLU rs11136000 (T), rs867230 (C), and rs9331896 (C) were associated with decreased risk of MCI. Meanwhile, BDNF rs2030324 (G), CLU rs9331888 (G), MTHFR rs1801133 (A), and TOMM40 rs2075650 (G) were associated with increased risk of MCI after adjusting for age and education (Model 1). After further adjusting for APOEε4 carriers, the association between TOMM40 rs2075650 (G) and MCI disappeared, while CLU rs1532278 (T) showed a significant association with MCI (Model 2).

In the overdominant model, SORL1 rs641120, CLU rs1532278 (T), rs867230 (C), and rs9331896 (C) were significantly associated with decreased risk of MCI, while TOMM40 rs2075650 (G), BIN1 rs6733839 (T), and PVRL2 rs6859 (A) were significantly associated with increased risk of MCI (Model 1). After further adjusting for APOEε4 carriers, only the association between SORL1 rs641120 and MCI was persistent (Model 2). NOS3 rs1549758 (T) gained a significant protective effect for MCI (Model).

Multiple-SNP analysis

In the dominant model, MTHFR rs1801133 (A) and CLU rs9331888 (G) were consistently found to be associated with MCI risk (Models 1 and 2). PICALM rs561655 (G) showed no significant association with MCI in Model 1 but showed a significant protective effect after further adjusting for APOEε4 carriers (Fig. 1).

Fig. 1.

Multiple SNPs analysis using lasso regression by four genetic models (dominant, recessive, additive and overdominant model). Only significant set of SNPs in models were shown. a model 1, adjusted by age and education; b model 2, adjusted by age, education and APOEε4 carriers; OR, odds ratio.

In the recessive model, BDNF rs2030324 (G), BDNF rs2070045 (T) and CLU rs9331888 (G) showed significant associations with risk of MCI, while PVRL2 rs6859 (A) were significantly associated with decreased risk of MCI (Model 1). MTHFR rs1801133 (A) and SORL1 rs641120 (G) showed significant associations with the risk of MCI after adjusting for APOEε4 carriers (Model 2).

In the additive model, MTHFR rs1801133 (A) and CLU rs9331888 (G) were associated with MCI risk with or without adjustment for APOEε4 carriers.

In overdominant model, SORL1 rs641120(G) showed a protective effect, while BIN1 rs6733839 (T) showed a risk effect for MCI, independent of APOEε4 status. However, the influence of CLU rs9331896 (C) was eliminated after adjustment for APOEε4 status.

Subgroup analysis of single SNPs

In the dominant model (Fig. 2), rs1801133 was associated with increased risk of MCI in no exercise, advanced education, and no diabetes groups, while rs561655 was associated with decreased disease risk in exercise and advanced education groups. The protective effects of rs9331896 were showed in no exercise and diabetes population. The sex-specific associations were observed in two SNPs (rs2030324 and rs9331888).

Fig. 2.

Subgroup SNPs of single SNPs analysis using logistic regression by four genetic models (dominant, recessive, additive and overdominant model). Only significant set of SNPs in models were shown. Model was adjusted by age, education and APOEε4 carriers. OR odds ratio, 95%CI 95% confidence interval.

In the recessive model, rs2070745 (T) and rs641120 (G)was associated with increased risk of MCI in no exercise and advanced education individuals. For rs2030324, exercise, education, and diabetes were found to be modifying factors influencing MCI risk. rs9331888 in females and rs1801133 in the no-diabetes group significantly increased MCI risk, while rs6733839 in the no-exercise group decreased the risk.

In the additive model, apart from SNPs mentioned in dominant and recessive models, rs9331896 was associated with decreased MCI risk for individuals with advanced education. In diabetes patients, rs1532278 and rs867230 were found to be protective factors.

In overdominant model, the heterozygote of rs641120 showed a protective effect of MCI in no exercise, female, and advanced education participants, while heterozygote of rs6733839 was a MCI risk factor in no exercise, less education, and no diabetes participants. The significant associations between rs1549758 and MCI were restricted to females. The diabetes status also modified the effect of rs1136000 and rs867230.

Subgroup analysis of multiple SNPs

The multiple models stratified by exercise, gender, education, and diabetes mellitus were shown in Fig. 3. We merely present the included SNPs after the selection of SNPs by lasso regression. In dominant model, the modifying factor of rs1801133 compared with the single SNP model remained unchanged. rs9331888 (G) was found to be associated with MCI risk in females, less education, and no diabetes patients.

Fig. 3.

Subgroup SNPs analysis of multiple genes using lasso regression by four genetic models (dominant, recessive, additive and overdominant model). Only SNPs selected in models by lasso regression were shown. Model was adjusted by age, education and APOEε4 carriers. OR odds ratio, 95%CI 95% confidence interval.

In recessive model, rs2070045, 2,030,324, and rs1801133 were associated with higher MCI risk in no exercise individuals. In females, rs1801133 and rs9331888 were associated with MCI risk. Education also modified the effect of rs2030324.

In additive model, the modifying effect was mostly observed for rs1801133. In overdominant model, The significant associations between rs6733839 heterozygote with increased MCI risk were merely observed in no exercise, females, less education, and no diabetes mellitus individuals.

Discussion

In this cross-sectional study, fifteen SNPs in nine genes were found to be associated with MCI risk in the southern Han Chinese population. The functional annotations of these genes are related with Aβ production or clearance (SORL1, PICALM, CLU), Synapse regulation (BDNF), Neuroinflammation (NOS3), metabolism (MTHFR, TOMM40), and Endocytosis (BIN1, PVRL2). The diverse functions and pathways related to development of MCI indicates the multi-causal nature of the disease. The associations between MTHFR rs1801133 (A), CLU rs9331888 (G) and risk of MCI were evaluated consistently in most genetic models with or without adjusting for APOEε4 status. PICALM rs561655 (G) and NOS3 rs1549758 (T) were firstly identified to be associated with MCI risk in dominant and overdominant models. The association between TOMM40 rs2075650 (G) and risk of MCI was dependent with APOEε4 carriers. Although genome-wide association studies offer a more comprehensive strategy for novel locus discovery, our candidate SNP approach enables targeted evaluation of well-established AD risk loci in an underrepresented population, providing insights into whether these variants contribute to early cognitive decline at the MCI stage.

CLU PICALM SORL1

In our study, several potentially functional SNPs were identified at the CLU locus. Of these, rs11136000, rs1532278, rs867230, and rs9331896 of CLU were associated with decreased risk of MCI, whereas rs9331888 was associated with increased risk of MCI. A meta-analysis of GWAS studies found a significantly decreased risk for AD in carriers of the T allele at rs11136000 and rs1532278, and of the C allele at rs867230 and rs9331896, which is consistent with our study^14,35. In addition, carriers of the risk allele rs11136000 in a Caucasian population showed a faster rate of memory loss compared with non-carriers in the pre-dementia stages of the AD continuum³⁶. A large-scale GWAS identified rs9331888 in the CLU gene as significantly associated with AD in Caucasian ancestry. However, a previous study reported no association between the rs9331888 polymorphism and AD in the East Asian population³⁷. The strong statistical evidence for the involvement of SNPs in CLU in the development of MCI is supported by biological plausibility. CLU, also known as APOJ, is a multifunctional glycoprotein³⁸. It binds with the soluble form of Aβ, forming new complexes and facilitating Aβ to cross the blood-brain barrier³⁹. Similarly, CLU suppresses Aβ deposition in cooperation with APOE and alters Aβ clearance⁴⁰. Increased CLU levels have been found in brain injury or chronic inflammation⁴¹. Moreover, PICALM and CLU interact with hippocampal degeneration in elder populations⁴². After APOE and BIN1, PICALM is one of the most important genetic factors for AD²⁵. In prior studies, the corresponding protein was found to be involved in Aβ pathology through the processing of AβPP processing and Aβ transcytosis, as well as in tau pathology and to act as a sensitizing factor for the progression of tau pathology⁴³. A previous GWAS meta-analysis identified rs561655 in PICALM as a significant protective factor for the development of AD⁴⁴. Our study revealed that rs561655 (G) of PICALM was a protective genetic allele for MCI. Moreover, three SNPs in SORL1, including rs11218343, rs2070045, and rs641120, were associated with a higher likelihood of MCI in the recessive model and overdominant models. Two of these SNPs, rs2070045 and rs641120, were associated with the risk of MCI in the northern Han population⁴⁵. Minor allele C within rs11218343 was found to be protective in both Han and Caucasian populations. However, in our study, rs11218343 (C) was found not to be associated with MCI in all models except the recessive model. SORL1 was identified as having an important role in influencing AD pathology via altering Aβ production^46,47. The related protein is located in neuronal multivesicular bodies and functions as a sorting factor for AβPP, directing intracellular movement and procession of AβPP⁴⁸.

Meanwhile, higher expression of SORL1 in cells reduced the Aβ in the mouse model⁴⁹. The advanced mechanism study also showed that mutations in SORL1 impaired Aβ uptake in microglia-like cell lines and reduced Aβ clearance in the mouse brain⁵⁰. Thus, variants in SORL1 could be predictors of function loss, but their molecular effect should be investigated by experimental studies⁵¹. Taken together, our findings support the hypothesis that allele dosage may influence CLU and SORL1 expression, thereby modulating Aβ clearance efficiency. For example, individuals carrying two risk alleles at rs9331888 or rs2070045 may exhibit a more pronounced impairment in Aβ trafficking or receptor sorting, accelerating synaptic dysfunction and early cognitive deficits. This gene-dose relationship may operate through mechanisms where the presence of specific risk alleles alters transcriptional regulation or mRNA stability, potentially reducing the abundance of functional CLU or SORL1 proteins available for Aβ binding and clearance. This aligns with previous observations showing altered CLU or SORL1 mRNA and protein levels in AD brains, which provide direct evidence for gene expression dysregulation in AD pathogenesis, as well as with functional studies demonstrating variant-specific differences in Aβ binding and endocytosis^49,52. Future studies are warranted to quantify these effects longitudinally and assess how allele combinations shape the trajectory from normal aging to MCI.

MTHFR TOMM40

High plasma homocysteine and folate levels were linked to cognitive performance, even in dementia-free seniors⁵³. A dose-response meta-analysis indicates that an increase in blood homocysteine levels of 5umol/L each is linearly associated with a 15% increased risk of AD⁵⁴. MTHFR is one of the key enzymes regulating folic acid metabolism and affecting DNA methylation and nucleic acid synthesis⁵⁵. The SNP rs1801133 in MTHFR is associated with an increased risk for AD⁵⁶. However, the association between rs1801133 and MCI remains controversial. The minor allele of rs1801133 was a genetic risk factor for MCI in our study but does not appear to influence cognition loss with age in several previous studies⁵⁷. This inconsistency might be due to gene-dose effects. Homozygous carriers (TT) have significantly reduced enzyme activity. This leads to higher homocysteine levels and weaker methylation ability, which could make neurons more vulnerable early on⁵⁸. TOMM40 mediates the formation of a complex with BAP31, an endoplasmic reticulum protein, and the movement of NADH dehydrogenase (complex I) from the cytosol into the mitochondria, which ultimately assembles the mitochondrial membrane respiratory chain⁵⁹. rs2075650 in TOMM40 was identified in strong linkage disequilibrium with APOE⁶⁰. Our findings indicated a significant association between rs2075650 and MCI risk, but no association after adjusting for APOEε4 status, consistent with previous studies⁶¹. Although its statistical signal may be confounded by APOE, rs2075650 has also been implicated in mitochondrial dysfunction, which may compromise neuronal energy metabolism and contribute to neurodegeneration⁶².

NOS3 BDNF

Early and substantial involvement of inflammation plays a crucial role in the development of AD continuum or MCI³⁰. We found that rs1549758(T), located at exon 6 of NOS3, a key gene related to immune response, was associated with a significantly decreased risk of MCI in a dominant model with adjustment for APOE4 status. Inducible NOS in microglia could be stimulated by cytokines, leading to the production of high concentrations of nitric oxide, which may be possibly toxic to neurons⁶³. Apart from direct neurotoxicity, overexpression of NOS could induce posttranslational modification of Aβ, such as nitration and S-nitrosylation, which could eventually lead to functional and structural damage in the brain⁶⁴. More importantly, nitrated Aβ could activate plague formation and be found in the core of amyloid plague⁶⁵. The above biological evidence suggests the important role of nitric oxide in the early stages of AD, which may be consistent with the significant association between SNP in NOS3 and MCI. NOS3 may be another promising biomarker for the early detection of MCI.

In our study, rs2030324 (G) in BDNF was found to be associated with a higher risk of MCI in recessive and additive models independent of APOEε4 status, confirming the effect of rs2030324 on MCI in previous reports⁶⁶. Although large-scale meta-analyses have not consistently found a strong link between rs2030324 and AD, mounting evidence suggests this SNP may influence earlier neurodegenerative stages. Rs2030324 lies in the 5’ untranslated region (UTR) of the BDNF gene and could affect gene transcription or mRNA stability. Carriers of the risk allele show reduced BDNF expression. This reduction is linked to impaired hippocampal function, as BDNF is crucial for synaptic plasticity, long-term potentiation, and memory formation. Notably, the functionally related BDNF Val66Met polymorphism disrupts activity-dependent BDNF secretion. This leads to reduced hippocampal activation during memory tasks and poorer memory performance^67,68. These findings support the idea that BDNF dysfunction, potentially modulated by regulatory variants like rs2030324, may increase susceptibility to MCI. This likely occurs through diminished neuroplasticity and impaired memory consolidation.

PVRL2 BIN1

BIN1 is an important regulator of endocytosis, membrane recycling, and apoptosis⁶⁹. There is no previous evidence on rs6733839 and MCI risk, but rs744373 in BIN1 is related to development of MCI in the Chinese population⁷⁰. Minor allele A of rs6859 in PVRL2 was associated with risk of MCI in a logistic regression model but not with MCI when adjusted for APOEε4 status. In a recent study, rs6859 was found to be associated with MCI in elderly APOEε4 carriers⁷¹. The associations between rs6733839 in BIN1 and rs6859 in PVRL2 and MCI remain controversial and further studies are needed to investigate them clearly.

Mechanistically, the BIN1 locus contributes to tau pathology. Risk alleles enhance tau propagation and neuronal toxicity in AD models⁷². The rs6733839 variant may alter BIN1 expression in brain regions like the hippocampus and temporal cortex. This could disrupt endocytic processes essential for synaptic function. Similarly, PVRL2 encodes a brain-expressed cell adhesion protein. The rs6859 variant may regulate its expression via epigenetic mechanisms, particularly in APOEε4 carriers, as demonstrated by eQTL analyses⁷³. These findings suggest potential dosage-dependent and context-specific effects. They underscore the need for analyses stratified by APOE genotype and tissue-specific expression.

Modification of lifestyle and gene effect

The subgroup analysis found the modification effect of personal traits and lifestyle on gene. Our findings showed the modification of gender and exercise to effect of BDNF rs2030324 polymorphism in BDNF on MCI risk. Previous studies revealed the different role of BDNF as an inflammatory mediator between male and female, resulting from the similar pathways between estrogen and BDNF, especially in hippocampus⁷⁴. Meanwhile, emerging data showed that participants might benefit from regular exercise, but such effect vary by BDNF genotype and gender⁷⁵. Another meta-analysis indicated the positive impact of regular exercise on BDNF level in blood and this impact could be moderated in female group⁷⁶. Above findings indicate a potential link between gender, exercise, and BDNF.

CLU rs9331888 was another SNP with wide disparity in different subgroup analysis. A plasma proteomic analysis showed that physical exercise could slow cognitive decline and neurodegeneration via increasing CLU level⁷⁷. The direct evidence between CLU polymorphism of participants with different educational background was rare, while level of CLU protein in plasma increased in Chinese MCI population with no education⁷⁸. However, the complex mechanisms behind these modifications remain unclear.

Diabetes mellitus status modified effect of four SNPs in CLU, including rs1532278, rs867230, rs11136000, and rs9331888. Reported evidence showed the association between CLU polymorphisms with diabetes, via increasing insulin resistance and decreasing insulin secretion⁷⁹. In Chinese population, diabetes is a common risk factor for both dementia and MCI, similar to many chronic diseases²². These findings indicated glycemic control could be beneficial for cognitive function, suggesting the importance of multi-intervention of diverse chronic diseases⁸⁰. However, the evidence with modifying effects to other SNPs were less established, more researches are still need to investigate the modifying effect of lifestyle on MCI related genes.

Strengths and limitations

The strengths of the paper include general and comprehensive literature review to cover most of the genetic factors of AD in previous studies. Moreover, we adjusted diverse risk factors associated with AD such as age, education, and APOEε4 status in different models, indicating the stability of our models. Most importantly, in our study, we recruited relatively healthy elderly people from communities to find participants with risk genetics alleles, which may be helpful for prevention of AD at an early stage. Finally, we conducted subgroup analysis to elucidate modification effect of personal traits and lifestyle on SNPs.

This study has several limitations. First, the original data included only cohort baseline data, with no follow-up data. In the future, we would follow the cohort participants and observe change of cognitive function, to figure out which SNPs are related to the conversion of MCI to AD in the Chinese population. Second, MMSE was used to assess global cognitive function, which have relatively lower sensitivity when compared with the MoCA and might lead to missed diagnosis of MCI³¹. However, there were previous studies merely adopting MMSE to evaluate cognition, because MMSE is very convenient for community survey⁸¹. Third, 100 SNPs were selected for statistical analysis, which might require a large sample size. The sample size limited our ability to perform GWAS, which generally requires larger cohorts to achieve genome-wide significance for polygenic traits. Fourth, the study was cross-sectional and observational. Causal inferences cannot be made and reverse causality bias cannot be ruled out.

Conclusion

Our study demonstrates the associations between several AD-associated SNPs with MCI risk, and lifestyle-gene modification effect. Findings of this study provide novel data that will identify high-risk subgroups for early precision prevention of MCI or preclinical-AD.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

ZX, WT and NF was responsible for SNPs test. ZX and WT wrote the main manuscript text. NF participated in providing ideas for the article and reviewing the manuscript. XZ, XY, LH, and XZ participated in recruiting. ZX, WT and NF participated in revising the manuscript. All authors contributed to the article and approved the submitted version.

Funding

This work was supported by grants from the National Natural Science Foundation of China (to N.F., Grant NO.81602929) and Startup Fund for Youngman Research at SJTU (to N.F., 17 × 100040016).

Data availability

All data supporting the findings of this study are available within the paper and its Supplementary Information. Selected SNPs and primers sequences are provided in Supplementary Table 1. They are available from the corresponding author (nnfeng@shsmu.edu.cn) for researchers who meet the criteria for access to confidential data.

Declarations

Competing interests

The authors declare no competing interests.