Iron and risk of dementia: Mendelian randomisation analysis in UK Biobank

Background Brain iron deposition is common in dementia, but whether serum iron is a causal risk factor is unknown. We aimed to determine whether genetic predisposition to higher serum iron status biomarkers increased risk of dementia and atrophy of grey matter. Methods We analysed UK Biobank participants clustered into European (N=451284), African (N=7477) and South Asian (N=9570) groups by genetic similarity to the 1000 genomes project. Using Mendelian randomisation methods, we estimated the association between genetically predicted serum iron (transferrin saturation [TSAT] and ferritin), grey matter volume and genetic liability to clinically defined dementia (including Alzheimer’s disease [AD], non-AD dementia, and vascular dementia) from hospital and primary care records. We also performed time-to-event (competing risks) analysis of the TSAT polygenic score on risk of clinically defined non-AD dementia. Results In Europeans, higher genetically predicted TSAT increased genetic liability to dementia (Odds Ratio [OR]: 1.15, 95% Confidence Intervals [CI] 1.04 to 1.26, p=0.0051), non-AD dementia (OR: 1.27, 95% CI 1.12 to 1.45, p=0.00018) and vascular dementia (OR: 1.37, 95% CI 1.12 to 1.69, p=0.0023), but not AD (OR: 1.00, 95% CI 0.86 to 1.15, p=0.97). Higher TSAT was also associated with increased risk of non-AD dementia in participants of African, but not South Asian groups. In survival analysis using a TSAT polygenic score, the effect was independent of apolipoprotein-E ε4 genotype (with adjustment subdistribution Hazard Ratio: 1.74, 95% CI 1.33 to 2.28, p=0.00006). Genetically predicted TSAT was associated with lower grey matter volume in caudate, putamen and thalamus, and not in other areas of interest. Discussion Genetic evidence supports a causal relationship between higher TSAT and risk of clinically defined non-AD and vascular dementia, in European and African groups. This association appears to be independent of apolipoprotein-E ε4.


Ascertainment of dementia diagnosis in UK Biobank
Specific diagnostic codes for dementia were identified from the UK NHS National Institute for Health and Care Excellence (NICE) Quality and Outcomes Framework (QOF) Business Rules (https://qof.digital.nhs.uk),version 37.0 (version date 09/06/2017).Read v2 and CTV3 codes were converted to ICD-9 and ICD-10 codes using UK Biobank Resource 592 (Clinical coding classification systems and maps).A detailed list of ICD-9, ICD-10, Read2 and CTV3 codes used can be found in Supplementary Table 2.
We also performed secondary exploratory analysis of vascular dementia.We were unable to analyze "dementia with Lewy bodies" or frontotemporal dementia due to very low number of cases ascertained.UK Biobank only provided the first 5 digits of the diagnostic codes, rendering us unable to differentiate some subtypes (for example, ICD-10 code G31.83 "Dementia with Lewy bodies" is 6 digits but the HES diagnosis data only includes G31.8 i.e., the first 5 digits).Patient with missing data were excluded from analysis.

Derivation of grey matter volume phenotypes
Grey matter volume regions were created as the sum of the left and right hemisphere using data from UK Biobank, except when multiple data fields were aggregated into one phenotype (details below)..For subcortical regions the amygdala was created as the sum of data fields 25888 and 25889, caudate 25880 and 25881, hippocampus 25886 and 25887, pallidum 25884 and 25885, putamen 25882 and 25883 and thalamus 25878 and 25879.For frontal regions the inferior frontal gyrus was created as the sum of data fields 25790 to 25793, middle frontal gyrus 25788 and 25789, precentral gyrus 25794 and 25795, superior frontal gyrus 25786 and 25787and supplementary motor cortex 25832 and 25833.For parietal regions precuneus was created as the sum of data fields 25842 and 25843, postcentral gyrus 25814 and 25815, superior parietal lobe 25816 and 25817.For temporal regions interior temporal gyrus was created as the sum of data fields 25808 to 25813, middle temporal gyrus 25802 to 25807 and parahippocampal gyrus 25848 to 25851.All resulting phenotypes were adjusted for brain volume (data field 25009)

Mendelian randomization (MR) IVW estimate and standard error
MR relies on several assumptions: the exposure SNPs are robustly associated with the relevant measured exposure.This is quantified by the F-statistic, which can be approximated by the ratio of the SNPexposure association estimate, β ând its standard error, SE(β ), squared (equation 1) 3 .
2 the exposure SNPs are not associated with confounding factors that bias conventional epidemiological associations.the exposure SNPs are only associated with the outcome through the risk factor.
For IVW the causal estimate β is: BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) and the approximate standard error (se) of the estimate is: where Xn is the estimate of the genetic association with the risk factor with standard error σXn and Yn is the estimate of the genetic association with the outcome with standard error σYn, for each of the N genetic variants (n=1,…,N).

Polygenic score derivation
We derived a polygenic score for transferrin saturation in the UKB participants by summing the number of TSAT-increasing alleles carried, multiplied by the known effect of the allele on TSAT levels 4 .In mathematical notation the polygenic score  ̂ is derived: where   is the number of TSAT-increasing alleles weighted by their published effect  ̂, for each of the  genetic variants ( = 1 … ).Patient with missing data were excluded from analysis.

Defining Groups in UK Biobank using Principal Component Analysis
To determine genetic groups of individuals in the UK Biobank, we performed principal components (PC) analysis using samples from the 1000 Genomes Project (1GP) 5 .Individuals from the UK Biobank were subsequently projected into the 1GP PC space using the SNP loadings derived from the initial PC analysis to minimise confounding of PC values due to varying degrees of relatedness within UK Biobank.Next, we calculated the means of the first 4 PCs stratified by individuals known to be from each of the 5 main genetic ancestral superpopulations in the 1GP data: European (EUR), South Asian (SAS), East Asian (EAS), African (AFR) and Admixed American (AMR).Using the means derived from the 1GP reference dataset, we subsequently performed K-means clustering analyses to determine which individuals from UK Biobank could be classified as EUR-like, SAS-like or AFR-like.Using 1GP ancestry specific mean values for PCs 1-4, we classified 451,454 individuals in UK Biobank to be similar to EUR genetic ancestry.Next, using the 1GP mean values described above for principal components 2-4, we classified 10,207 individuals to be similar to SAS genetic ancestry.Finally, using 1GP means for PCs 1-4, we identified 7,968 individuals estimated to be similar to AFR genetic ancestry.In the manuscript we refer to these groups defined by similarity to 1GP as EUR-like, SAS-like and AFR-like groups.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)
..... 14 Analysis in UK Biobank participants, adjusted for the competing risk of mortality.X-axis restricted to 82 years, as very few participants attained age >82 during follow-up.AFR=African-like group, EUR=European-like, SAS=South Asian-like.Other ancestry groups were too small to study.BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)SFig 4: Scatter plot of individual transferrin saturation-associated genetic variant effects on all-cause dementia diagnosis in EUR-like grouprs79220007 BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)SFig 5: Scatter plot of individual transferrin saturation-associated genetic variant effects on non-Alzheimer dementia diagnosis in EUR-like males.BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)SFig 7: Scatter plot of individual transferrin saturation-associated genetic variant effects on vascular dementia diagnosis in EUR-like group (female participants).BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) rs79220007