Genotyping of the Alzheimer’s Disease Genome-Wide Association Study Index Single Nucleotide Polymorphisms in the Brains for Dementia Research Cohort

INTRODUCTION

The knowledge on the genetic etiology of late-onset Alzheimer’s disease (LOAD) has been vastly enhanced over the last decade. Whole genome association studies (GWAS), and next-generation sequencing investigations have identified numerous genetic risk variants for the disease in large collaborative samples [1]. The largest GWAS study to date combined the data from four previous GWAS datasets (ADGC, CHARGE, EADI, and GERAD) to create the IGAP discovery sample of 17,008 cases and 37,154 controls and imputed over 11 million SNP genotypes for analysis [2]. The addition of data from a replication dataset increased the sample size to 25,580 cases and 48,466 and yielded the now accepted 19 risk loci (excluding the APOE locus) for LOAD from GWAS.

The Brains for Dementia Research (BDR, http://brainsfordementiaresearch.org.uk/) project is a recently established longitudinal cohort (2008) which aims to provide brain tissue, alongside serially collected clinical assessments, for research purposes from neuropathologically defined samples [3]. The in-life assessments include measures of cognition, mood, behavior, general health, and lifestyle and occur every 1–5 years depending on age and cognitive status. All such data is available through the Dementias Platform UK (https://portal.dementiasplatform.uk). An extensive neuropathological examination of each brain at postmortem provides diagnosis and full details of muti-morbidities according to standard criteria. Data for deceased participants is available through the Medical Research Council (https://mrc.ukri.org/research/facilities-and-resources-for-researchers/brain-banks/). The original cohort included 3,276 participants of whom over 2,300 remain alive. These include more than 75% who have no reported memory problems. Brain donation has been achieved for 610 participants so far with 70–100 participants dying each year. Blood samples have been collected from 400 participants, almost all from those without memory problems. To date, DNA has been extracted from 600 postmortem brains for the purpose of genetic analysis. Previously we have published the results of our initial exome sequencing project on a sub-sample of the dataset [4], here we present the findings from association analysis on the full sample set to-date for the established GWAS index SNPs [2], in order to determine if the BDR sample is genetically representative of LOAD.

METHODS

RESULTS

The BDR sample is a growing cohort, with DNA available for over 600 brain samples for scientific use. The genetic analysis presented here consists of the current neuropathology-confirmed diagnosed sample, of 315 AD samples and 149 control samples. The demographics are presented in Table 1. A number of samples were excluded from the analysis due to the dementia being other than AD, including Lewy body dementia, vascular dementia, frontal temporal lobe dementia, mixed dementias, and early-onset Alzheimer’s disease dementias. Other diagnoses where dementia was present but were comorbid with other pathologies included three cases of Parkinson’s disease, three cases of cerebrovascular disease, two cases of corticobasal syndrome, and single cases of argyrophilic grain disease and Pick’s disease. There was also an individual who presented with dementia but for which there was no underlying neuropathology present. There were also a number of control samples where although no dementia symptoms were present, they have been excluded due to other disorders being present, including seven cases of cerebrovascular disease, six cases of Parkinson’s disease, and single cases of Huntington’s disease, corticobasal syndrome, progressive supranuclear palsy, motor neuron disease, and cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) syndrome.

Allele frequencies did not deviate from Hardy-Weinberg equilibrium and minor allele frequencies were similar to those observed in the Lambert et al. discovery dataset [2]. One discrepancy arose from a single SNP (rs35349669) associated with the INPP5D gene, even though the minor allele frequency was similar to that observed in the Lambert et al. study and in the population data generated from gnomAD, the minor allele was the opposite to what was expected suggesting that the C-allele is the minor allele in our dataset, whereas in gnomAD and the Lambert datasets the minor allele in the T-allele (Table 2).

Logistic regression analyses found that the ɛ4 allele was highly associated with the AD phenotype (p = 3.99×10^–12, OR = 3.76 (95% CI 2.59–5.46)) as would be expected. Association analysis for the GWAS index SNPs Bonferroni-corrected for multiple testing (n = 19 tests, p < 0.0026) yielded no significant results. Nominal association was observed for 4 SNPs (Table 3). SNP rs28834970 (PTK2B) displayed nominal significance with a p value of 0.044, while stronger associations for rs10792832 (PICALM), rs35349669 (INPP5D), and rs1476679 (ZWCPW1) were observed with nominal significance with p values of 0.023, 0.014, and 0.015 respectively. The PICALM and ZWCPW1 associated SNPs both indicated a protective effect with odd ratio of 0.7 and 0.65, respectively, in agreement with that observed in the Lambert analysis. Furthermore, the PTK2B and INPP5D SNPs both increased risk for the development of AD with odds ratios of 1.39 and 1.47, respectively. However, as previously noted, the association for the INPP5D SNP was with the minor allele which is the opposing allele to that observed by Lambert et al. [2]. In total 14 out of the 19 SNPs were in concordance with the Lambert et al. study with respect to the allele associated and direction of effect size (73.7%).

DISCUSSION

This study presents data for the association of the established 19 SNP loci associated with LOAD in the newly formed BDR cohort. Although still in its infancy the cohort has collected over 600 neuropathology-confirmed brain samples, which have been genetically analyzed, with further samples (up to 3,000) expected in the next few years. The aim of this study was to investigate if the BDR cohort was representative of other much larger cohorts of LOAD and controls. Using the SNPs identified in the meta-analysis GWAS study by Lambert et al. [2], we genotyped the BDR sample with the NeuroChip [5] to obtain the GWAS index SNP data and supplemented it with KASP assays for SNPs that were not present on the array. Minor allele frequencies of the 19 SNPs explored were similar to that produced by the discovery sample in the Lambert et al. study, indicating that, genetically speaking, the BDR cohort is representative of other LOAD datasets. The single exception was for rs35349669 (INPP5D), where opposing alleles were found in the minor frequencies (Table 2). The minor allele for this SNP is the T-allele with a frequency of 49.9% in the European (non-Finnish) population dataset from gnomAD. This is similar to that observed in the Lambert et al. dataset, where the T-allele frequency is reported as 48.8% in their discovery sample. The BDR cohort also has a high minor allele frequency for the rs35349669 SNP (45.6% when cases and controls are combined) however this frequency is for the C-allele. Where samples have high frequencies of the minor allele it is not uncommon to observe ‘allele flipping’ and may be indicative of subtle variation between the BDR cohort and the Lambert discovery dataset or the difference in sample size [8]. Furthermore, this SNP also indicated nominal significant association with the LOAD phenotype in the BDR sample with the C-allele (as opposed to the T-allele in the Lambert study).

Three further SNPs within the BDR cohort were indicative of significance for association rs28834970 (PTK2B), rs10792832 (PICALM), and rs1476679 (ZWCPW1). It is interesting to note that previous analysis of this cohort with whole exome sequencing also indicated association to the ZWCPW1 gene region with Burden analysis indicating association of the PILRA gene which has been shown to be in weak LD (r² = 0.5) with the ZWCPW1 GWAS index SNP rs1476679 [4]. Although no SNP displayed significant association after correction for multiple testing, 14/19 SNPs (73.7%) were concordant with the Lambert meta-analysis dataset for allele and direction of effect. Those that were non-concordant were all with the same allele (except rs35349669, INPP5D), with effect sizes around 1. Given the small effect sizes of GWAS and the sample size of the current BDR cohort fluctuation around an OR of 1 is expected.

Currently the BDR cohort is underpowered to significantly detect the effect sizes of the established GWAS hits for LOAD. However, as the cohort grows, it is envisaged that the data will become increasingly concordant with such studies as the Lambert meta-analysis, given the preliminary data presented here. Genetic data generated from the BDR cohort is publicly available upon a data request to BDR and therefore can serve the interests of the research community at large for small-scale projects wanting to investigate the effects of the GWAS hits. The analysis presented here utilizes only the AD phenotype; additional investigations of the BDR cohort allows far more complex analysis utilizing the extensive neuropathological data, cognitive and lifestyle data available as more genetic analysis is available.