Abstract
Recent research suggests genetic variation in the Klotho locus may modify the association between APOE ɛ4 and cognitive impairment. We tested for associations and interactions between these genotypes versus risk of dementia, cognitive abilities, and brain structure in older UK Biobank participants. Klotho status was indexed with rs9536314 heterozygosity (versus not), in unrelated people with versus without APOE ɛ4 genotype, corrected for various confounders. APOE ɛ4 associated with increased risk of dementia, worse cognitive abilities, and brain structure. Klotho was associated with better reasoning. There were no interactions; potentially suggesting an age- and pathology-dependent Klotho effect.
INTRODUCTION
Preserving cognitive abilities such as memory is a common concern into older age, and in the absence of reliable treatments, the public health priority is prevention and delay of cognitive impairment [1], including understanding effect modifiers. APOE ɛ4 is a known risk factor for Alzheimer’s disease (AD) and cognitive decline [1]. Genetic variation in the KL locus has been associated with ageing-related phenotypes including insulin resistance and brain function [2]. A recent study of AD cohorts, longitudinal conversion, and amyloid-beta samples showed a statistically significant ‘modification’ effect where the deleterious effects conferred by APOE ɛ4 were offset by heterozygosity based on two KL polymorphisms in strong linkage disequilibrium: F352V (rs9536314) and C370S (rs9527025), possibly due to correlations with increased serum Klotho [3]. UK Biobank is a relatively large general population cohort [4] where we have previously shown deleterious effects of APOE ɛ4 on cognitive [5], structural brain imaging [6], and AD/dementia phenotypes [7]. This brief report tested the hypothesis that based on recent research, genetic KL variation would interact with APOE ɛ4 genotype in relevant cognitive, brain, and dementia phenotypes.
METHODS
Study design and participants
UK Biobank is a prospective cohort study including 502,628 participants who attended one of 22 baseline assessment centers from 2006 to 2010, aged 40–70 years [8]. In 2014, MRI scanning of a sub-group of 100,000 participants began, and this is ongoing. This project was completed using UK Biobank application #17689 except for the hospital admission and episode (HES) analyses which were conducted using project #7155.
Ethical approval and data availability
This analysis was conducted under generic approval from the NHS National Research Ethics Service (approval letter dated 17 June 2011, ref 11/NW/0382). Written informed consent was obtained from all participants in the study.
Dementia outcomes
Dementia/AD outcomes were generated in two ways: firstly, derived by UK Biobank using self-report, hospital admission, and death record data, with data utilizing International Classification of Diseases version 10 (ICD-10 codes). Individuals were designated as cases (“all-cause dementia” or “Alzheimer disease”) if they had indicated either in self-report or hospital/death records—derived by UK Biobank [7, 9]. Those coded as missing were designated as controls (i.e., did not self-report dementia, and diagnoses not present in hospital/death records). The UK Biobank-derived ascertained cases were the latest available as of October 2020. Secondly we supplemented this with record-based hospital admission/episode (HES) data using open-access methods described previously [10] where dementia was defined as ICD-10 codes F00 (dementia in Alzheimer disease), F01 (vascular dementia), F02 (dementia in other diseases), or F03 (unspecified dementia); hospital admission records were available until February 2018 for the full cohort, whereas linkage to primary care records was available for 45%of the UK Biobank cohort (approximately 230,000 participants) until May 2017, for Scotland, September 2017, for Wales, and August 2017 for England.
Imaging data
The release of brain MRI data as of July 2020 was used (i.e., approximately 40k). All imaging data used here was processed and quality checked by UK Biobank [11]. We selected imaging phenotypes a priori shown to be associated with worse cognitive ability and decline total white and grey volumes adjusted for skull size (WM/GM respectively); log WM hyperintensity volume (WMH); overall hippocampal volume; general factors of fractional anisotropy (FA) and mean diffusivity (MD), and frontal lobe GM (gFrontal) [12] based on principal components analysis (PCA). Total WM hyperintensity volumes were calculated based on T1 and T2 fluid-attenuated inversion recovery (FLAIR), derived by UK Biobank.
Cognitive data
Five tests were completed at baseline (2006–2010), of which we examine three here which have shown sufficient intra-participant reliabilities: Pairs-matching 6-pair (memory), verbal-numeric reasoning and log reaction time (processing speed) [13]. We also examined four cognitive tests administered from 2014 onwards. These were: Trail making test a + b (processing speed/executive function) and Digit symbol substitution (executive function) assessed via online follow-up, plus Matrix pattern completion (nonverbal reasoning) and Tower rearranging (executive function) at MRI [12].
Genetic data
UK Biobank genotyping was conducted by Affymetrix using a bespoke BiLEVE Axiom array for ∼50,000 participants and the remaining ∼450,000 on the Affymetrix UK Biobank Axiom array. All genetic data were quality controlled by UK Biobank as described by the protocol paper [4].). APOE ɛ4 ‘risk’ genotype presence (versus non-ɛ4) was genotyped based on rs7412 and rs429358. KL was indexed using rs9536314 where G/T is considered protective (versus G/G; T/T) and synonymous with KL-VS diplotype heterozygosity [14].
Covariates
Participants self-reported their smoking history: current, past, or never, medication use for dyslipidemia, hormone replacement therapy, blood pressure, oral contraceptive, or insulin. We excluded participants for whom these data were missing (< 5%). Townsend deprivation indices were derived from postcode of residence.
Statistical analysis
PLINK v1.90 was used for genetic quality controlling and Stata V.14 was used for statistical analyses. We removed participants who reported neurological conditions as described previously [13]. We statistically controlled for: age, sex, Townsend, ever-smoking, genotypic array, baseline/MRI assessment center, 8 principal components, array, and medication (concurrent to the phenotype under study; dementia outcomes used baseline values). We focused on participants aged≥60 years at baseline or imaging (respectively for those analyses) [3]. We excluded participants with non-white British ancestry, self-report versus genetic sex mismatch, putative sex chromosomal aneuploidy and excess heterozygosity. In terms of quality controlling, we accounted for relatedness between participants by removing one random participant in cases where two individuals were 2nd cousins or closer, and this was based on central UK Biobank-derived relatedness coefficients (https://biobank.ndph.ox.ac.uk/ukb/label.cgi?id=263). We included polymorphisms in Hardy Weinberg equilibrium (p > 1*10–6), polymorphisms with missingness rate less than 0.1 and minor allele frequency > 0.01 and imputation score > 0.8. We have previously reported power calculations indicative of > 95%confidence to find ‘true’ effect sizes at Cohen’s D = 0.1 (i.e., small) with regard to APOE genotype and outcomes in UK Biobank [6]. Multiple power calculations using G*Power 2, for a ‘true’ effect at Cohen’s D = 0.1 (where 0.2 is considered a small effect size) at p = 0.05, estimated the least power to find a ‘true’ effect was in the imaging sample, with 96%estimated power post-hoc based on current group sample sizes.
RESULTS
Descriptives
After exclusions there were baseline N = 169,374 (mean age 64.1, SD = 2.89) participants; imaging n = 26,903 (68.14 years; SD 5.02). Allele frequencies for baseline/imaging analyses are shown in Supplementary Table 1. There were n = 1,570 UKB-ascertained dementia cases (0.9%) of which n = 634 were AD (0.4%). There were n = 3,346 (1.9%) dementia cases based on HES of which n = 1,577 (0.9%) were AD.
Outcomes
Supplementary Table 2 shows associations between APOE ɛ4 genotype and multiple worse outcomes: UK Biobank-ascertained dementia (odds ratio [OR] = 3.27 for ɛ4 versus not) and AD (OR = 5.06), HES-based dementia (OR = 3.27) and AD (OR = 5.06), cognitive scores on Matrix Completion (–0.04 SDs for ɛ4 versus not), frontal lobe GM (–0.05), hippocampal volume (–0.06), log TMT total time (0.037) and Digit symbol (–0.075; significant p-value range < 0.001 to 0.038). KL heterozygosity (versus not) was associated with better reasoning only (0.021 SDs; p = 0.042). There were no statistically significant APOE/KL interactions.
As sensitivity analyses all models were re-run: unadjusted then covariates added incrementally; excluding people with concurrent neurological conditions [13]; testing a dose effect of 0/1/2 G allele copies (i.e., including an extra n = 4,489 with GG) rather than the primary T/T versus G/T test throughout, and using the full sample (aged < 60). These made no difference to the results.
DISCUSSION
A recent study by Belloy et al. [3] reported a protective modifying effect of KL heterozygosity on APOE ɛ4 genotype’s conferred risk on cognitive impairment and dementia, in a collation of longitudinal, AD and amyloid-beta cohorts totaling N = 24,743. Using unrelated UK Biobank data, we tested whether a similar effect could be seen in multiple outcomes: AD/all-cause dementia versus not, and non-demented cognitive and structural brain MRI phenotypes known to underlie cognitive decline. Participants were≥60 years as per Belloy et al. We identified individual APOE and KL genotype/outcome associations but no interactions, against our hypothesis. There could be an underestimation of true effect due to cognitive test imprecision [13] or generally preserved participant health [12]. No interaction here versus Belloy et al. could reflect the use of different phenotypes: the original study investigated AD case versus control status, conversion to impairment and amyloid-beta while this study investigated AD status and non-demented cognitive/brain structure values. UK Biobank derived dementia status largely from ICD codes whereas Belloy et al. used clinical and/or pathological ascertainment. This could suggest that the age-dependent changes in KL expression, and interaction with APOE status, manifest only beyond at least moderate AD-related neuropathology (e.g., amyloid or tau) [15]. The null AD case/control interaction could reflect that UK Biobank is relatively healthy and well-educated. It is possible the UK Biobank participants were not sufficiently old; analysis of longitudinal cognitive and brain imaging data (in independent data) is indicative of more pronounced KL heterozygosity effects particularly into later life [16]. By contrast our findings are more in-line with de Vries et al. [16] who reported protective rather than deleterious effects of KL heterozygosity on longitudinal cognitive decline.
Limitations
This study did not explore an exhaustive list of structural imaging phenotypes. There is some degree of healthy volunteer bias in UK Biobank participants, and probably more so in participants who returned for imaging. Some imaging participants would have been unable to complete scanning due to contraindications related to poorer health, e.g., pacemakers/stents [12]. Dementia ascertainment was based on a mixture of self-report, HES, and death data; participants did not have regular cognitive assessment, e.g., with Mini-Mental State Exam, and hence there may be some degree of underestimation of dementia in the current data.
Summary
A recent relatively large-scale study including cohort, case/control longitudinal and amyloid-beta data showed a significant interaction whereby KL genotype modified the well-known APOE ɛ4 and dementia association. Using independent cognitive, structural brain and dementia data, we did not support these prior findings; this could reflect some degree of bias or imprecision in UK Biobank participants or phenotypes, or that the interaction while ‘true’ is contingent on AD-related neuropathology: future studies should investigate this further in deeply-phenotyped cohorts.
Footnotes
ACKNOWLEDGMENTS
This research has been conducted using the UK Biobank resource; we are grateful to UK Biobank participants. Thanks to Dr. Breda Cullen for devising exclusion criteria, Dr. Michael A. Belloy and Dr. Luke Pilling for helpful comments on an earlier draft.
RT is funded by a Baillie Gifford doctoral student fellowship. UK Biobank was established by the Wellcome Trust medical charity, Medical Research Council, Department of Health, Scottish Government and the Northwest Regional Development Agency. It has also had funding from the Welsh Assembly Government and the British Heart Foundation.
