Genome-wide association study of hippocampal atrophy rate in non-demented elders

Hippocampal atrophy rate has been correlated with cognitive decline and its genetic modifiers are still unclear. Here we firstly performed a genome-wide association study (GWAS) to identify genetic loci that regulate hippocampal atrophy rate. Six hundred and two non-Hispanic Caucasian elders without dementia were included from the Alzheimer’s Disease Neuroimaging Initiative cohort. Three single nucleotide polymorphisms (SNPs) (rs4420638, rs56131196, rs157582) in the TOMM40-APOC1 region were associated with hippocampal atrophy rate at genome-wide significance and 3 additional SNPs (in TOMM40 and near MIR302F gene) reached a suggestive level of significance. Strong linkage disequilibrium between rs4420638 and rs56131196 was found. The minor allele of rs4420638 (G) and the minor allele of rs157582 (T) showed associations with lower Mini-mental State Examination score, higher Alzheimer Disease Assessment Scale-cognitive subscale 11 score and smaller entorhinal volume using both baseline and longitudinal measurements, as well as with accelerated cognitive decline. Moreover, rs56131196 (P = 1.96 × 10-454) and rs157582 (P = 9.70 × 10-434) were risk loci for Alzheimer’s disease. Collectively, rs4420638, rs56131196 and rs157582 were found to be associated with hippocampal atrophy rate. Besides, they were also identified as genetic loci for cognitive decline.


INTRODUCTION
The hippocampus is a vulnerable and plastic structure buried deep in the medial temporal lobe of human body [1].The atrophic hippocampus is often accompanied by poor memory performance, and changes in the hippocampus provide a neural substrate for cognitive impairment that may be associated with normal aging, post-traumatic stress disorder, recurrent depression, and Cushing's syndrome [2].Hippocampal atrophy rate has been demonstrated to be closely correlated with cognitive disorders, including the most commonly reported Alzheimer's disease (AD) and some non-AD disorders, such as frontotemporal dementia (FTD) and impaired memory [3][4][5][6][7][8].As for mechanism, hippocampal atrophy rate may relate to the deposition of amyloid-beta (Aβ) and tau [9,10].Clinically, the atrophy rate was greater in subjects with normal cognition (NC) who converted to mild cognitive impairment (MCI) or AD than in those who remained AGING stable; it was greater in MCI subjects who converted to AD than in those who remained stable; and it was greater in fast AD progressors than slow ones [11,12].Consequently, the reduction in hippocampal volumes over time may be promising in predicting individuals at high risk of developing cognitive decline, monitoring disease trajectories at early stage, and assessing treatment efficacy in clinical practice or drug trials.
Previous genome-wide screening identified novel susceptibility genes for AD using baseline hippocampal volumes as quantitative traits [13].However, the genetic predictors of longitudinal changes in hippocampal volumes remain poorly understood.Use of quantitative traits in genome-wide association studies (GWAS) provides novel and important insights into broader trends in correlations between genes and their associated pathways [14].Furthermore, magnetic resonance imaging (MRI) has great advantages in visualizing structural and functional brain changes, such as sufficient sensitivity, non-invasiveness, ease of availability, and good tolerance [15].And hippocampal volumes can be reliably measured in vivo.Therefore, we conducted a GWAS with longitudinal MRI measures of hippocampal volumes to identify genetic risk factors influencing hippocampal atrophy rate in non-demented elders.These genetic contributors may be involved in cognition-related pathophysiological process.

Characteristics of included subjects
This study included 226 NC (111 women, 74.7±5.3 years) and 376 MCI (152 women, 72.3±7.2 years) subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort of non-Hispanic Caucasian ancestry.The summarized characteristics of included subjects were shown in Table 1.MCI group (47.5%) had a higher frequency of ε4 allele within APOE gene than NC group (25.7%).MCI group also had bigger baseline whole brain volume (P = 0.020) and smaller baseline hippocampal volume (P < 0.001) compared to NC group (P < 0.001).

Impact of rs4420638 and rs157582 on cognitive scores and brain structures
Associations of two top SNPs with cognitive scores and brain structures were analyzed.In the crosssectional analyses, both the minor allele of rs4420638 (G) and the minor allele of rs157582 (T) were associated with lower Mini-mental State Examination (MMSE) score, higher Alzheimer Disease Assessment Scale-cognitive subscale 11 (ADAS-cog 11) score and smaller entorhinal volume (Supplementary Figure 5).In the longitudinal study, subjects with a minor allele of rs4420638 (G) showed faster cognitive decline in MMSE score (P < 0.0001) and ADAS-cog 11 score (P < 0.0001), as well as greater rates of entorhinal atrophy (P = 0.0001) and ventricular enlargement (P < 0.0001).Besides, subjects with a minor allele of rs157582 (T) also showed faster cognitive decline in a The MCI group had lower baseline hippocampal volumes compared to the NC group (P < 0.001).MMSE score (P < 0.0001) and ADAS-cog 11 score (P < 0.0001), as well as greater rates of entorhinal atrophy (P = 0.0049) and ventricular enlargement (P < 0.0001) (Figure 3).

Effect of SNPs on risk of cognitive decline
After the GWAS, we conducted survival analysis to further explore the influence of two top SNPs on cognitive decline (Figure 4 and Supplementary Table 3).The COX regression analysis was performed on minor allele homozygotes, heterozygotes and major allele homozygotes by including age and gender as covariates.Both the minor allele of rs4420638 (G) (P = 2.04 × 10 -12 ) and the minor allele of rs157582 (T) (P = 8.23 × 10 -8 ) appeared to accelerate cognitive decline, conferring increased risk to homozygous and heterozygous carriers of the minor allele and confirming the positive direction of effects detected in our GWAS.

Bioinformatics analyses
The International Genomics of Alzheimer's Disease Project (IGAP) is by far the largest genetic epidemiological survey of AD risk, which was performed in two stages comprising a discovery step (stage 1) and a replication step (stage 2).Based on several grand-scale meta-analyses, the IGAP has identified many susceptibility loci for AD [16,17].After checking the top loci linked to hippocampal atrophy rate in the stage 1 meta-analysis from IGAP database, we identified rs56131196 (P = 1.96 × 10 -454 ) and rs157582 (P = 9.70 × 10 -434 ) as risk loci for AD.

DISCUSSION
This study was the first to conduct a GWAS of hippocampal atrophy rate in non-demented elders.We identified genome-wide significant associations of 3 SNPs (rs4420638, rs56131196, rs157582) in the TOMM40-APOC1 region with hippocampal atrophy rate and 3 additional suggestive association loci (in TOMM40 gene and near MIR302F gene).The minor allele of rs4420638 (G) and the minor allele of rs157582 (T) showed associations with lower MMSE score, higher ADAS-cog 11 score and smaller entorhinal volume using both baseline and longitudinal measurements, as well as with accelerated cognitive decline.Our findings provide evidence that TOMM40 and APOC1 as candidate genes may promote the application of hippocampal atrophy rate as an early biomarker for predicting cognitive progression and detecting disease trajectories.Both APOC1 and TOMM40 are genes adjacent to APOE, which showed no significant associations with hippocampal atrophy rate after adjusting for APOE ε4 dosage.Although the most concise explanation may be that APOE ε4 was driving the observed associations, the extensive LD structure around APOE and various associations reported in APOE region made it difficult to further explain the result.As in any SNP-trait association, only statistical association was not enough to infer causality [19].Besides, it's impossible for us to analyze all potentially causal variants in the region.A previous study also supported that the correlation between APOE region and cognitive function was not mediated solely by the APOE ε4 allele [20].And the polymorphic poly-T variant in the TOMM40 gene among different populations provided more possibilities for the interpretation of our results [21].Compared with those having shorter poly-T repeats, APOE ε3 carriers having a long poly-T repeat in the TOMM40 gene not only have an earlier age of onset of late-onset AD, but also have decreased memory abilities and grey matter volumes [20].It seems that the polymorphic poly-T variant in the TOMM40 gene provides greatly improved accuracy in the estimation of cognitive disorders for APOE ε3 carriers.Furthermore, a large-scale longitudinal study reported that both APOE ε4 and poly-T repeats in the TOMM40 gene were associated with cognitive decline, but there was no interaction between the two genes [22].Thus, it's difficult to say whether the correlations between TOMM40-APOE-APOC1 region and cognitive function were mediated by the APOE ε4 allele or by poly-T repeats, or were mediated by more complex mechanisms.More research is warranted to explore the related pathogenesis.
Rs56131196 and rs157582 have been reported in GWAS studies on AD and aging-related verbal memory respectively [23,24], thus lending validation and confidence to our analytic procedure and results.Previous studies suggested that the elderly carrying APOC1 gene tended to perform worse in cognitive scores and showed more severe hippocampal abnormalities [25,26].Besides, the APOC1-knock-out mice might have worse memory performance than those carrying APOC1 [27].As for mechanism, ApoC1 (apolipoprotein C1) encoded by APOC1 is a member of the apolipoprotein family, which may be involved in multiple biological processes comprising cholesterol metabolism and neuronal apoptosis [28].But the specific mechanisms by which APOC1 gene modulates the risk of cognitive impairment remain controversial, although some research has been done in this field [26,29,30].The protein that TOMM40 encodes, TOM40 (translocase of the outer mitochondrial membrane 40), constitutes an external mitochondrial membrane channel that promotes the transport of many aggregating proteins to mitochondria [31,32].TOM40 protein may also act as a molecular chaperone that could accelerate the movement of ribosomal preproteins through the channel and assemble them in the mitochondria after translation [31,32].By mediating the dynamic functions of mitochondria, TOMM40 may contribute to changes in cognitive status.
Both TOMM40 and APOC1 are in strong LD with APOE on chromosome 19.Several SNPs in TOMM40-APOE-APOC1 region have been detected to be associated with cognitive impairment [19,33].Each of these 3 adjacent genes could encode proteins with biological values that may affect cognitive function.However, controversy still exists as to whether the associations between these 3 adjacent genes and cognition are independent of APOE allele or are driven by LD with APOE.The biochemical interaction between APOC1 and APOE may be associated with cognitive impairment, since the binding of triglyceride lipoproteins to the very low density lipoprotein receptor, mediated by APOE allele, could be interfered by overexpressed ApoC1 [26].Furthermore, APOC1 could increase the risk of cognitive impairment by modulating lipid metabolism.Additionally, apolipoprotein E (ApoE), amyloid, and synuclein proteins are interactive with Tom40 [34].The Tom40 protein forms the channel through which amyloid beta protein precursor (APP) and Aβ travel and aggregate to cause mitochondrial abnormalities [32].There was also evidence for the effects of APOE receptors on APP trafficking and Aβ production, and the effects of APOE on Aβ aggregation and clearance [35].Thus, it has been postulated that APOE and TOMM40 genes might share similar mechanisms in mediating disease risk [32].
A functional analysis targeting TOMM40-APOE-APOC1 region reported that various APOE locus cisregulatory elements affect both APOE and TOMM40 promoter activity [36].This indicates that gene expression patterns in this region may be modulated by a complicated transcriptional regulatory structure.Evidence also supported the role of epigenetic mechanisms such as deoxyribonucleic acid (DNA) methylation in the regulation of gene expression [19].The increase of DNA methylation often leads to downregulated gene expression by either blocking access of transcriptional factors or enrolling methyl-cytosineguanine-binding proteins [37].The repressed gene expression within the region was demonstrated to be correlated with cognitive dysfunction both in blood samples and brain tissues [38][39][40].Both the methylation-gene expression and gene expressioncognition associations in the TOMM40-APOE-APOC1 region are worth investigating in the future.AGING There was a suggestive finding in the MIR302F gene, which was a member of microRNA (micro ribonucleic acid) family that participated in post-transcriptional regulation of gene expression in multicellular organisms via influencing both the stability and translation of messenger ribonucleic acids [41].Studies have reported the associations of breast cancer [42], gastric cancer [42,43] and acute heart failure [44] with MIR302F gene, whereas little was known about the mechanism by which MIR302F gene correlated with cognitive deterioration.Further investigations are especially warranted to explore how MIR302F gene influences the progression of cognitive disorders.

Limitations
Some limitations must also be acknowledged.Firstly, the sample size was relatively small, which may not be representative of the general population.Secondly, our participants were restricted to non-Hispanic Caucasians and we didn't explore the diversity among different populations.Thirdly, we applied Bonferroni correction for multiple comparisons and set the MAF threshold at > 0.10, which could enhance statistical power to avoid false-positive results but may miss less common SNPs.

ADNI dataset
The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD, VA Medical Center and University of California-San Francisco.The primary goal of ADNI has been to investigate the effectiveness of integrating neuroimaging, genetic/biological markers, as well as clinical and neuropsychological assessments in measuring the progression of MCI and early AD.All ADNI individuals were recruited from over 50 sites across the United States and Canada, and most people were non-Hispanic Caucasians.

Participants
In this study, 602 non-Hispanic Caucasian nondemented individuals (NC = 226, MCI = 376) were enrolled from the ADNI cohort after applying quality control (QC) procedures.All participants received baseline and periodic clinical and neuropsychological assessments as well as serial MRI.Data used in the preparation for this article were derived from the ADNI database (http://adni.loni.usc.edu/).A total of 698 samples before QC were available with both GWAS data and hippocampus data.To reduce confounding effects by genetic ancestry that could lead to population stratification, the analysis data was restricted to non-Hispanic Caucasian participants (n = 648).To avoid the impact of AD pathology on results, all participants were restricted to cognitively normal individuals or those with mild cognitive impairment (n = 610).To determine cryptically related individuals and/or sample mix-ups, identify-by-descent estimates and MDS components were conducted using PLINK [45].This step excluded 5 participants who showed cryptically associated and clustering separately from the other subjects (Supplementary Figure 1), remaining 605 valid samples.Finally, all samples presented tight clustering with the population of European descent using the HapMap cohort.
Individuals were followed up to detect progressive cognitive decline defined as (1) losing > 3 points between the first and last MMSE measurements, (2) developing from NC to MCI or from MCI to dementia, or (3) having a score < 24 at last MMSE [46].

Hippocampal atrophy rate and QC
Longitudinal hippocampal volume measurements by MRI could be downloaded from the ADNI database, which was conducted by N. Schuff and his colleagues at UCSF via FreeSurfer version 4.3 [47].The hippocampal atrophy rates were obtained from a mixed-effects model using "arm, lme4 and lmerTest" packages in R software, after controlling for age of entry, gender, number of APOE ε4 allele, years of education, baseline diagnosis and total ICV.Individualized rate was then used as a quantitative outcome phenotype for the GWAS.QC was conducted to mitigate the impact of extreme values on statistical results.The mean and standard deviation (SD) of the hippocampal atrophy rate were calculated by experienced operators blinded to clinical data, and the figures greater or smaller than 4fold SD from the mean value were considered as extreme outliers and removed from this analysis.After eliminating 3 outliers, there were 602 valid subjects left.

Genotyping and QC
Genotyping for all samples was analyzed by the Illumina Human Hap610-Quad BeadChips featuring 2,379,855 SNPs.QC procedures were implemented with the PLINK software following the stringent criteria: call rates for SNPs > 98%, call rates for individuals > 95%, MAF > 0.10 and Hardy-Weinberg equilibrium test P > 0.001.We restricted the MAF value > 0.

Figure 1 .
Figure 1.Manhattan and regional plots for associations with hippocampal atrophy rate.(A) Genome-wide signal intensity (Manhattan) plots showing the -log10 (p value) for individual single nucleotide polymorphisms.(B) Regional association results for the 45.0 Mb to 45.8 Mb region of chromosome 19.(C) Association results for the 45.0 Mb to 45.8 Mb region of chromosome 19 controlling for rs4420638.(D) Association results for the 45.0 Mb to 45.8 Mb region of chromosome 19 controlling for rs157582.

Figure 2 .
Figure 2. Hippocampal atrophy rates of different genotypes.The y-axis showed the hippocampal atrophy rate and the x-axis corresponded to different genotypes.The effect of genotypes on hippocampal atrophy rate was examined with a multiple linear regression model using age, gender and diagnosis as covariates.(A) The minor allele of rs4420638 (G) showed association with hippocampal atrophy rate in a dose-dependent manner (P = 3.20 × 10 -7 ).(B) The minor allele of rs157582 (T) showed association with hippocampal atrophy rate in a dosedependent manner (P = 1.23 × 10 -8 ).

Figure 3 .
Figure 3. Impact of rs4420638 and rs157582 on longitudinal measurements in cognitive scores and brain structures.Associations of rs4420638 with longitudinal measurements in Mini-mental State Examination (MMSE) score (A), Alzheimer Disease Assessment Scale-cognitive subscale 11 (ADAS-cog 11) score (B), entorhinal volume (C) and ventricular volume (D) over time.Associations of rs157582 with longitudinal measurements in MMSE score (E), ADAS-cog 11 score (F), entorhinal volume (G) and ventricular volume (H) over time.

Figure 4 .
Figure 4. Kaplan-Meier survival curves for probability of cognitive progression.Numbers of individuals at risk at different followup time points were presented.Survival time was calculated according to the interval from the initial baseline evaluation to cognitive progression.(A) Subjects with the minor allele of rs4420638 (G) showed a greater rate of cognitive decline.(B) Subjects with the minor allele of rs157582 (T) showed a greater rate of cognitive decline.

Table 2 . Top SNPs related to hippocampal atrophy rates with APOE ε4 fitted as a covariate.
10 for SNPs to avoid potentially false-positive results and improve statistical power.Finally, all 602 subjects remained in the analysis and 695,203 SNPs passed QC protocols.The overall genotyping rate in remaining individuals was 99.7%.and Technology Major Project (No.2018SHZDZX01) and ZHANGJIANG LAB, Tianqiao and Chrissy Chen Institute, and the State Key Laboratory of Neurobiology and Frontiers Center for Brain Science of Ministry of Education, Fudan University.Data collection and sharing for this project was funded by the Alzheimer's Disease Abbreviations: BP, base pair (variant position); CHR, chromosome; MAF, minor allele frequency; SNP, single nucleotide polymorphism.