Genetic effects on longitudinal cognitive decline during the early stages of Alzheimer’s disease

Cognitive decline in early-stage Alzheimer’s disease (AD) may depend on genetic variability. In the Swedish BioFINDER study, we used polygenic scores (PGS) (for AD, intelligence, and educational attainment) to predict longitudinal cognitive change (measured by mini-mental state examination (MMSE) [primary outcome] and other cognitive tests) over a mean of 4.2 years. We included 260 β-amyloid (Aβ) negative cognitively unimpaired (CU) individuals, 121 Aβ-positive CU (preclinical AD), 50 Aβ-negative mild cognitive impairment (MCI) patients, and 127 Aβ-positive MCI patients (prodromal AD). Statistical significance was determined at Bonferroni corrected p value < 0.05. The PGS for intelligence (beta = 0.1, p = 2.9e−02) was protective against decline in MMSE in CU and MCI participants regardless of Aβ status. The polygenic risk score for AD (beta =  − 0.12, p = 9.4e−03) was correlated with the rate of change in MMSE and was partially mediated by Aβ-pathology (mediation effect 20%). There was no effect of education PGS on cognitive measures. Genetic variants associated with intelligence mitigate cognitive decline independent of Aβ-pathology, while effects of genetic variants associated with AD are partly mediated by Aβ-pathology.


Study participants
The study population consisted of 381 cognitively unimpaired (CU) elderly participants and 177 patients with mild cognitive impairment (MCI) from the prospective and longitudinal Swedish BioFINDER sample (clinical trial no. NCT01208675; www.biofinder.se), for which baseline cognitive tests, age, education, gender, Ab status and at least 3 follow up data for MMSE (including baseline) were available. Among consecutively included patients with mild cognitive symptoms, some were classified as MCI and some as subjective cognitive decline (SCD). Following research guidelines, control participants and patients with SCDs were combined to make CU individuals [1]. The participants were recruited between September 2010 and December 2014 at three different memory clinics as previously described [2,3]. Briefly, clinical assignment of MCI and SCD was performed after patient recruitment based on a neuropsychological battery as previously described [4]. The subjects were thoroughly assessed for their cognitive complaints by physicians with a special interest in dementia disorders. The inclusion criteria for patients with SCD and MCI were as follows: (i) cognitive complaint; (ii) not fulfilling the criteria for dementia; (iii) a Mini-Mental State Examination (MMSE) score of 24-30 points; (iv) age 60-80 years; and (v) fluent in Swedish.
The exclusion criteria were as follows: (i) cognitive impairment that without doubt could be explained by a condition other than prodromal dementias; (ii) severe somatic disease; and (iii) refusing lumbar puncture or neuropsychological investigation. Cognitively normal controls were eligible for inclusion if they (i) were aged 60 years old, (ii) scored 28-30 points on the Mini-Mental State Examination (MMSE) at the screening visit, (iii) did not have cognitive symptoms as evaluated by a physician, (iv) were fluent in Swedish, (v) did not fulfill the criteria of MCI or any dementia. The exclusion criteria were as follows: (i) presence of significant neurologic or psychiatric disease (e.g., stroke, Parkinson's disease, multiple sclerosis, major depression), (ii) significant systemic illness making it difficult to participate, (iii) refusing lumbar puncture and (iv) significant alcohol abuse. The Regional Ethics Committee in Lund, Sweden, approved the study. All subjects gave their written informed consent.

Genotyping and preparation of genetic data
Genotyping was conducted using the Illumina platform GSA-MDA v2. Before imputation, subject-level quality control (QC) included removing sexual incompatibility between chip-inferred sex and self-reported sex, low call rates (1% cut-off), and extreme heterozygosity. Relatedness among the samples was eliminated by removing one participant from each pair of close relatives (first or second degree) identified as, # ³ 0.0625. Using PLINK2 [5], multi-dimensional scaling was done to create principal components in genetic analyses to account for ancestry. Standard QC steps were performed for SNP-level filter to ensure conformity with the reference panel used for imputation (strand continuity, names of the alleles, position and assignments for Ref / Alt). To sum up, for imputation, 685494 high-quality variants (autosomal, non-monomorphic, biallelic variants with Hardy-Weinberg Equilibrium (HWE) P > 5 × 10 -8 and with a call rate of > 99%) were used. Imputation was carried out using the Sanger Imputation Server (https://imputation.sanger.ac.uk/) with SHAPEIT for phasing [6], Positional Burrows-Wheeler Transform (PWBT) [7] for imputation and the entire Haplotype Reference Consortium (release 1.1) reference panel [8].
Multi-allelic variants and SNPs with a data imputation score < 0.8 have been excluded as part of post-imputation QC and genotype calls with a posterior likelihood < 0.9 have been set to fail (i.e., hard-called). SNPs with a genotyping rate >0.9 were retained. SNPs with Minor Allele Frequency (MAF) ³ 5% were taken for the analysis. Further information on the imputation and QC process is detailed in https://rpubs.com/maffleur/452627.

Statistical Analyses (GWAS)
For the GWAS, we used generalized linear regression models using PLINK2 [5] to test for genetic association with longitudinal MMSE scores. The models were adjusted for age, gender, education, baseline MMSE (not for the intercept), APOE ε2 and ε4 count, and top 10 principal components (PC) from the principal component analysis (PCA) on the entire set of genotype data. All the statistical analysis was conducted in R programming (version 4.0.2) using standard R packages.

GWAS of the rate of cognitive decline
We next turned to exploratory analysis of individual genetic variants and rate of cognitive decline adjusting with and without APOE e4 burdens.
The genomic inflation factor for association analysis in both analyses was close to unity (l = 1), indicating a subtle population structure. No variant reached genome-wide significance, but 18 variants in APOE e4 burdens adjusted analysis and 22 variants in APOE e4 burdens not adjusted analysis were significant at the suggestive level significance of p £ 5 * 10 -5 (supplementary table S29 rs10492328 is located near a pseudogene GLULP5 and variant rs4747634 is located near an RNA gene C10orf16. rs10492328 was associated with cognitive decline (beta = -0.28), whereas rs4747634 showed a protective effect against cognitive decline (0.54). Recent studies have shown that although pseudogenes are not transcribed themselves, they may contribute to the regulation of gene expression [12], making it possible that the variants identified here modulate cognitive decline through regulation of other (unknown) genes

Education as a proxy for cognitive reserve
Though there was no significant association between PGS-Edu and the cognitive scores, we did find the expected association between education in itself and cognitive decline (supplementary table S7 and S8). To validate PGS-Edu, we performed this supplemental analysis by taking baseline education as a proxy for cognitive reserve. We found the expected results that all education PGSs (PGS-Edu 1-7) we strongly associated with education at baseline (p-value in range of 2.7e-12 to 9.8e-04, after Bonferroni correction).
A similar result was observed for all the intelligence PGSs that showed significant association with the baseline education (p-value in range of 1.4e-04 to 3.7e-02, after Bonferroni correction). However, we did not find any significant association between AD PRSs (PRS-Alz) and education at baseline (supplementary table S39 and supplementary figure 6). Table S1: PRS-Alz association with the slope of longitudinal MMSE score when not adjusting for Aβ-status Table S2: PRS-Alz association with the intercept of longitudinal MMSE score when not adjusting for Aβ-status