Utilising Polygenic Risk Score Analysis for AD to Determine the “Sphere of Influence” of the APOE Isoform SNPs
Connor Farrell, Keeley J Brookes*
Biosciences, Nottingham Trent University, Clifton Campus, Nottingham, NG8 11NS, UK
Abstract
The APOE gene and particularly the ε4 allele have been a long-established risk factor for Alzheimer’s disease (AD), demonstrating the largest genetic effect size in this complex disease. In light of the odds ratios observed for the risk allele, many studies disregard neighbouring association signals as merely “tagging” this effect. Polygenic risk score (PRS) analyses in this field regularly use low linkage disequilibrium parameters (r2≥0.1) when selecting SNPs for analysis across the genome and remove kilobases of data surrounding the APOE locus, preventing confounding factors influencing their results. This study investigated a 500kb region surrounding the APOE locus, utilising PRS analysis to explore whether additional SNPs in this region could be providing contributory effects to AD predictability. The data presented here suggest that the “sphere of influence” of the APOE isoform SNPs covers a region of around 92kb; SNPs in Linkage Disequilibrium (LD) at r2<0.4 with rs429358 potentially contribute independently to the PRS predictability for AD, and that there are additional independent SNPs in this region that have increased effects in an APOE ε4 negative sample. This study concludes that further consideration is required when selecting LD parameters for PRS analysis and that additional investigation into the region surrounding APOE may yield polymorphisms that may play a pivotal role in the development of AD.
Introduction
The Apolipoprotein E (APOE) gene is singularly the most replicated genetic association with Alzheimer’s disease (AD), located on chr19q13, its potential role has been highlighted since the early 1990’s through linkage studies1. The APOE ε4 allele/isoform is determined by the genotypes present at two coding SNPs, rs429358 and rs7412, and has been observed to have a multiplicative dosage effect on AD risk, with odds ratio estimated at 3 for heterozygous carriers and up to 12 for homozygotes2,3. Despite its large effect size, this genetic variant does not account for the entire estimated heritability observed for AD4, and therefore genetic variants have been sought and found in genome-wide studies over the past 30 years5. In recent years, association analyses have given way to polygenic risk score (PRS) studies, summarizing the effect sizes of multiple alleles into a single score with the aim to differentiate between cases and controls6. These studies have highlighted the involvement of multiple polymorphisms in the AD phenotype; however, these studies often exclude the APOE region. This exclusion is due to observations of multiple association signals in neighbouring SNPs to rs429358 and rs7412, believed to be tagging the effect of the isoform SNPs via linkage disequilibrium (LD), therefore large regions of the genome surrounding APOE are often removed prior to PRS analysis to prevent confounding effects of these variants. This excluded region ranges from 14kb to over 2Mb7–12, despite multiple studies suggesting that additional loci in this region may be having independent effects13–19. Therefore, the removal of this area could be missing key contributory variants from PRS models. This study aimed to investigate the extent of the “sphere of influence” of the APOE isoform SNPs by exploring the clumping parameters within a 500kb region, removed in our previous studies20, and identify potential SNPs within this region that may be independent contributors to the polygenic risk score for AD.
Methods
Datasets
The IGAP_stage 1 (IGAP_S1) summary statistics21 were used as the base dataset, and genotype data from the Brains for Dementia Research (BDR) project22 was used as the target dataset to generate the polygenic risk scores. The BDR dataset underwent standard quality control with PLINK v1.923. removing SNPs with a minor allele frequency of less than 1%; genotype calls of less than 95% and which deviated significantly from Hardy-Weinberg Equilibrium (P<0.0001) in the control samples. Samples that had less than 95% call rate were also removed. This resulted in an analysis dataset of 520 samples consisting of 356 pathological confirmed AD cases, and 164 controls.
Clumping
The IGAP_S1 summary statistics were clumped using the 1000Genomes dataset in PLINK v1.923, using the parameters –clump-p1 1 –clump-p2 1 –clump-kb 250 and –clump-r2 ranging from 0.1 to 0.9. Using R v4.0.324 the clumped output file for each r2 was changed from wide-to-long format for comparison with the .bim files from the target datasets, allowing each common SNP to be tagged with a clump number identifier.
SNPs in a 500kb region (hg19_chr19:45,160,844-45,660,844) surrounding the APOE isoform SNPs were extracted from both datasets, and common SNPs present in both datasets were carried forward into the analysis.
Polygenic Risk Score Generation
Polygenic risk scoring was carried out using the –score parameter in PLINK v1.923. Logistic regression was carried out in R v4.0.324, followed by calculating the Area Under the Curve (AUC) using the pROC package25. Using R v4.0.3 a ‘magic for loop’ 26 was set up to allow the inclusion of SNPs from the base dataset individually across the APOE region (R script available upon request).
Results
Sixty-two SNPs were present in both the IGAP_Stage 1 summary statistics and the BDR dataset across the 500kb APOE region under investigation (Supplemental 1). Individual SNP AUC values within the BDR dataset identified 26 SNPs of interest, inclusive of the APOE isoform SNPs and spanning 92kb (92,040bp). Out of the 26 SNPs, 14 achieve AUC’s of over 0.55, with 8 over 0.60; these mapped on to those SNPs which were found to be significantly associated with the AD phenotype in the BDR dataset (P<0.05). SNPs with AUC’s ≥0.60 are all found within a 39kb region (39,220bp), in addition to 3 SNPs with AUCs ≥0.55 and 2 SNPs with AUCs less than 0.55. Both rs429358 (AUC=0.7025) and rs7412 (AUC=0.5304) are within this region, with their combined isoform predictability providing an AUC of 0.7082 (P=6.89x10-15). Notably the region of significant SNPs and those that have higher AUCs are upstream to the APOE isoform SNPs with little indication for the involvement of SNPs downstream of the rs7412 SNP (Table 1).
Table 1: Summary table of data collected from the IGAP_Stage1 summary statistics, clumping of the IGAP_Stage1 summary statistics at various levels of r2, and association data and individual SNP AUC generated on the BDR dataset covering a 92kb region of interest. SNPs with high Area Under the Curve (AUC) measures align with association P values within the BDR dataset.
Visualisation of the clump assignment of these SNPs at various levels of r2, demonstrate two clear blocks of LD covering the TOMM40 and APOE genes (Table 1). It was hypothesised that SNPs that were tagging each other would not alter the AUC statistic of the polygenic risk scores generated, as the additional effects sizes of the SNPs would alter the scores for both cases and controls within the same margin, therefore not altering the difference between them in the overall risk score. To test this, SNPs assigned to “clump one” were added into the polygenic risk score model as the level of r2 decreased (Table 2). The SNPs incorporated into the PRS model up to an r2 ≥0.6 have similar effect sizes, leading to marginal changes in the AUC generated by these scores. As more SNPs with more variable effect sizes are added, the AUC also varies and increases, suggesting these SNPs may be independently contributing to the PRS.
Table 2: Summary table of AUC generated for the SNPs assigned as belonging to clump 1 in the IGAP_Stage 1 summary statistics at each level of r2. Both SNPs rs34404554 and rs34342646 are in high LD and remain in the same clump throughout the r2 parameters. Individual they provide similar AUC, which is not dissimilar when they are combined into the PRS. Addition of SNP rs71352238 throughout r2 parameters of 0.8-0.6 does not see a significant change in the polygenic risk score statistics. However additional SNPs added at lower levels of linkage disequilibrium appear to improve both the significance of the PRS and discriminability.
A secondary block of SNPs (rs8106922-rs1160985-rs405509) consistently identified as belonging to the same clump (although the clump number changes across the r2 parameters) was also observed and subjected to the same exploration to see if the same pattern of additional SNPs contributing to the model when lower levels of LD is utilised was seen. Again, this clump suggests that at lower parameters of r2, additional SNPs are contributing to the model rather than tagging other SNPs (Table 3).
Table 3: Secondary block of SNPs in strong LD across markers rs8106922-rs1160985-rs405509 displayed no improvement of AUC when SNPs were in strong LD, however at lower levels of r2 (<0.4) additional SNP contributed to the model.
The discriminability of PRS consisting of the most significant SNP (by IGAP_Stage1 statistics) for each clump at various levels of linkage disequilibrium were compared to identify the most significant model, utilising only those SNPs whose individual AUC was greater than 0.55. This suggests that using an r2 of ≥0.7 provides the most discriminatory model (Table 4) incorporating 10 SNPs within this region, although negligible differences in AUC were observed between the r2 range of 0.5-0.7.
Table 4: Table showing the results of polygenic risk score models consisting of the most significant SNP within each clump in the region of interest at each r2 parameter. For this only SNPs with individual AUCs of greater than 0.55 were included. This suggests that an r2 parameters of ≥0.7 provides the most discriminatory model, though there is minimal different in AUC between r2 measures of 0.5-0.7.
The BDR target dataset was divided into two groups based on whether the individual was APOE ε4 allele positive or not. This resulted in a dataset of 247 cases and 53 controls in the APOE ε4 allele positive group (ε4 pos): and 108 cases and 111 controls in the APOE ε4 allele negative group (ε4 neg). The 10 SNP PRS model was applied to these sub-groups and as expected the AUC in the APOE ε4 allele positive was not dissimilar to that of the full dataset (AUC 0.7368 v 0.7477 in the full dataset) whilst the score for those with no APOE ε4 alleles display minimal discriminability (0.5284), confirming the presence of the APOE ε4 allele is a strong contributory factor in predicting AD (Table 5, header row).
To test if each SNP within this 10 SNP model was contributing to the 0.7477 AUC, a drop-out analysis was carried out, removing a single SNP from the model, and observing if a drop of greater than 0.005 was observed in the AUC. The removal of SNPs, rs34404554 and rs1457582 that reside within “clump 1” of the IGAP_S1 summary statistics across all r2 parameters does not result in a large drop in the AUC when analysed in the entire dataset or in the APOE ε4 allele positive/negative datasets. Likewise, SNPs, rs1160985, rs104022771, rs519825 and rs8104483 are not making strong contributions to the discriminatory value of the 10 SNP model. However large drops in AUC are observed when the APOE isoform SNP rs429358 is removed from the model, supporting its role as a key contributory factor. In addition, removal of rs157580, consistently reduced the predictability of the model across full dataset and in the APOE ε4 allele positive/negative sub-groups, suggesting this SNP may also have a contributory role independent of the rs429358 isoform SNP (Table 5). The roles of SNPs rs12721046 and rs6859 are less clear, although lower AUCs are observed when these SNPs are removed from the entire dataset and in the APOE ε4 allele positive it does not make the >0.005 cut-off, however in the APOE ε4 allele negative dataset, the removal of these SNPs from the model indicate key contributory roles, which may only be observed in the absence of the rs429358 APOE ε4 isoform SNP.
Table 5: Summary of “drop-out” analysis to determine the level of contribution each SNP is having to the polygenic risk score model. Grey shading indicates drops in AUC of ≥0.005 when SNP is removed from the model. This analysis that SNPs rs429358 and rs157580 are important contributory SNPs in the polygenic risk score model for AD. In the absence of the APOE ε4 allele, two further SNPs, rs12721046 and rs6859, which display a low level of linkage disequilibrium with the APOE isoform SNP rs429358 (r2<0.4), are also important contributing SNPs in the discriminatory model.
To confirm the findings of the drop-out analysis, a “drop-in” analysis was performed adding in the key contributory SNPs identified to observe the impact on the AUC (Table 6). The resultant 4 SNP model had a higher AUC than the original 10 SNP model, indicating that the removal of SNPs that do not contribute to the model, reduces noise, and improves the discriminatory accuracy. However, this was only observed in the entire dataset and the APOE ε4 allele negative dataset. In the APOE ε4 allele positive sub-group the original model fairs better.
Table 6: Drop-in analysis, adding in contributory SNPs identified singly to obtain a 4-SNP model that provides a greater discriminatory accuracy than the 10 SNP model in the entire dataset the APOE ε4 allele negative sub-group. However, no improvement was observed in the APOE ε4 allele positive sub-group.
For completeness, rs7412 was added into this 4-SNP model to ensure capture of the full APOE isoform effect. Inclusion of this SNP only marginally changed the discriminatory accuracy of the 4-SNP model with AUCs of 0.7525, 0.7134, and 0.5746 obtained for the full cohort and APOE ε4 allele positive and negative subgroups respectively.
Discussion
The presence of the APOE ε4 allele is one of the strongest risk factors associated with the onset of AD1,2, due to this many PRS investigations on AD, remove a substantial region of genotype data surrounding this locus due to the assumption that additional association signals in this region are due to the SNPs being in LD with the isoform SNPs rs429358 and rs7412. This study has investigated this region of chr19, utilising PRS analysis and clumping algorithms to determine if by removing this region of the genome predictive SNPs for AD are being lost. The data presented here suggest that the “sphere of influence” of the APOE isoform SNPs covers a region of around 92kb; SNPs in LD at r2<0.4 with rs429358 potentially contribute to the PRS predictability for AD, and that there are additional independent SNPs in this region that demonstrate independent contributory effects in an APOE ε4 negative sample.
AUC statistics obtained for the APOE isoform SNPs in the BDR sample present here provide an overall model accuracy of 0.7082, with the increase in score significantly correlated to disease outcome (P=6.89x10-15). This is comparable to PRS outcomes observed in other studies8,27–29, confirming the BDR cohort is showing the same genetic architecture of larger cohorts.
LD is where there is a non-random association of alleles between loci, suggesting they are co-inherited at a frequency that is higher than chance. The r2 parameter indicates the level at which 2 alleles are correlated, and therefore when one is known provides an approximate prediction of the allele at the second loci, with r2=1 indicating the two alleles are in perfect correlation, with a certainty of allelic prediction, and 0 suggesting no correlation, and therefore random chance of allele prediction. Consequently, the higher the r2 between two SNPs the more correlated they are and the more accurate the allele prediction will be. Using this assumption many PRS studies opt to clump the SNP dataset being used so that only a single (most significant) SNP is used from a haplotype where the correlation (or r2) between them is ≥0.1, although the genetic distance in which the SNP lie in proximity to each other to assess varies from 250kb (default PLINK parameter) to 1000kb windows7,8,10,11,20,30,31. Although this almost certainly ensures independence of the SNPs being entered into the PRS model, it could also lead to many key independent SNPs being omitted from the analysis. Traditionally “tag” SNPs were genetic variants that were genotyped as proxies for additional SNPs in high LD around them, reducing redundancy in genotyping efforts for GWAS studies, investigations into the selection of these “tag” SNPs to capture the maximum variation of the genome suggest using SNPs with r2 ≥0.532,33.
In this analysis we clumped the base dataset and labelled each SNP with the clump identifier it belongs to at each r2 parameter. This demonstrated that the SNPs within the 500kb surrounding the APOE are not in high LD, with only 2 SNPs sharing at single clump at the highest r2 parameter of ≥0.9. As the r2 metric decreases, more SNPs are assigned to the same clump and by r2≥0.1, the majority of the 26 SNPs are tagged by 3 haplotype blocks. When exploring the 2 main haplotype blocks with the immediate region upstream to the isoform SNPs, including SNPs with higher levels of LD (>0.5) into the PRS models, does not greatly alter the AUCs observed (Tables 2 and 3). It is only when SNPs with lower levels of r2 are added into the model is there an increase in predictability. This suggests that a) when SNPs in high LD are included together in PRS models it does not artificially increase the AUC and b) the current practise of clumping SNPs at an r2 ≥0.1 may be forcibly removing SNPs that could be contributing to the phenotype. Interesting, the LD level at which additional SNPs are informative to these models is around an r2 of 0.5; which is the same level suggested by those involved in the HapMap and Tag SNP selection algorithms suggest is required to sufficiently capture the variation in the surrounding genome32,33.
This study has identified 3 potentially contributory SNPs to the PRS (in addition to rs429358) that are likely to be omitted from analyses. The SNP rs157580, becomes part of the secondary haplotype block (Table 3) when the clumping of SNP occurs with r2 less than 0.5, and therefore was included in as an independent SNP in the best observed model including the most significant SNPs in each clump at an r2 of ≥0.7. Drop-out analysis consistently suggested that the SNP was making contributory effects to the PRS model in the whole dataset and in APOE ε4 sub-groups. The minor G-allele of this SNP is less frequent in cases than in controls and therefore has a beta effect size of -0.378 in the IGAP_S1 summary statistics (P=1.21 x 10-101). The SNP resides within the TOMM40 gene which has been hypothesized as having an association with AD via an intron 6 poly-T variant17,34–36, although not consistently observed across studies37,38, and has been attributed to the level of LD observed between this variant and the APOE isoform39,40. Rs157580 is located further upstream within intron 1 and has been observed to be association with AD in multiple studies14,15, potentially altering intron excision rates16.
Further to this two SNPs that also clumped with one of the two major haplotypes within this region also appeared to make independent contributions to the PRS. SNPs rs6859 and rs12721046 reside on opposite ends of the region of interest spanning the APOE isoform SNPs, with both being tagged by the rs429358 SNP at low levels of LD (r2 of 0.1 and 0.3 respectively). The rs12721046 resides in the downstream gene to APOE, Apolipoprotein C1 (APOC1), and has also been shown to be associated with the AD phenotype19,41,42. A recent study43 suggests that a haplotype across the APOE locus consisting of rs2075650 (TOMM40) - APOE ε4 – rs12721046 (APOC1), has a stronger association with AD than the ε4 allele alone. This would support the suggestion from this investigation that TOMM40 contributes to the AD phenotype, however the SNP Kulminski and colleagues43 identified, although is not in the BDR analysis and so therefore is absent from our analysis, on exploration of the clump tagged IGAP_S1 dataset, it was found that rs2075650 resided in the linkage disequilibrium block “clump1” at all measures of r2 (0.1-0.9) and suggests it may be part of the association block of rs429358.
The SNP rs6859 which resides in the NECTIN2/PVRL2 gene upstream to the TOMM40-APOE-APOC1 LD block, has also been identified in several AD investigations15,18,19,44–48. The data present here would suggest that both rs6859 and rs12721046 are only contributing to the predictability of AD in the absence of the ε4 allele. The increase in AUC in the whole datasets and the ε4 negative sub-group, but negligible change in the ε4 positive subgroup when they are removed from the PRS 10 model demonstrates this and is supported by the improved predictability observed in the 4 SNP haplotype model in those same datasets, possibly suggesting that the large effect size of the ε4-allele overshadows their effects. This is support by the observations by Zhou and colleagues exploration of the APOE region, observing both the rs6859 and rs12721046 were associated in samples homozygous for ε3 allele19. This along with the data presented here suggests that some key variants in the APOE locus may be independently contributing to the AD phenotype and consist of a disease risk haplotype when in combination with the ε4 allele. It is perhaps because of the high frequency of the ε4 allele found in AD cohorts that these variants have been overlooked and be obscuring association and PRS analyses, as on an ε4 positive background it would seem these SNPs have little effect. Additional studies on this region, and a deeper look into the haplotypes is warranted, especially in AD cases that do not carry the ε4 allele, and across ethnic groups.
Acknowledgements
We would like to gratefully acknowledge all donors and their families for the samples provided for the BDR cohort and additional datasets who genetic data was also utilised here.
Tissue samples from the BDR cohort were obtained from the Southwest Dementia Brain Bank, London Neurodegenerative Diseases Brain Bank, Manchester Brain Bank, Newcastle Brain Tissue Resource and Oxford Brain Bank, and we thank our colleagues of the BDR Network, in particular the neuropathologists at each centre and BDR Brain Bank staff for the collection and classification of the samples. The BDR is jointly funded by Alzheimer's Research UK and the Alzheimer's Society in association with the Medical Research Council. Brains for Dementia Research has ethics approval from London – City and East NRES committee 08/H0704/128+5 and has deemed all approved requests for tissue to have been approved by the committee.
The genotyping of the Brains for Dementia cohort is supported by funding provided by an ARUK project grant, entitled ‘Enabling high-throughput genomic approaches in Alzheimer’s disease’ and an ARUK extension grant entitled ‘NeuroChip analysis of the entire Brains for Dementia Research (BDR) resource of 2000 samples’, awarded to KJB.
Conflict of Interest Statement
The author declares there is no conflict of interest.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
References:
- Pericak-Vance MA, et al. Linkage studies in familial Alzheimer disease: evidence for chromosome 19 linkage. Am J Hum Genet 48, 1034–1050 (1991).
- Corder EH, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science (80-. ). 261, 921–923 (1993).
- Kukull WA, et al. Apolipoprotein E in Alzheimer’s disease risk and case detection: A case-control study. J. Clin. Epidemiol. 49, 1143–1148 (1996).
- Farrer LA. et al. Effects of age, sex, and ethnicity on the association between apolipoprotein E genotype and Alzheimer disease. A meta-analysis. APOE and Alzheimer Disease Meta Analysis Consortium. JAMA 278, 1349–1356 (1997).
- Brookes KJ & Morgan K. Genetics of Alzheimer’s Disease. eLS. (2017) doi: 10.1002/9780470015902.a0020228.pub2.
- Baker E & Escott-Price V. Polygenic Risk Scores in Alzheimer’s Disease: Current Applications and Future Directions. Front. Digit. Heal. 2, 14 (2020).
- Fulton-Howard B, et al. Greater effect of polygenic risk score for Alzheimer’s disease among younger cases who are apolipoprotein E-ε4 carriers. Neurobiol. Aging 99, 101. e1-101. e9 (2021).
- Leonenko G, et al. Identifying individuals with high risk of Alzheimer’s disease using polygenic risk scores. Nat. Commun. 12, 4506 (2021).
- Escott-Price V, Shoai M, Pither R, et al. Polygenic score prediction captures nearly all common genetic risk for Alzheimer’s disease. Neurobiol Aging 49, 214 e7-214 e11 (2017).
- Najar J, et al. Polygenic risk scores for Alzheimer’s disease are related to dementia risk in APOE É4 negatives. 13, e12142 (2021).
- Stocker H, et al. Prediction of clinical diagnosis of Alzheimer’s disease, vascular, mixed, and all-cause dementia by a polygenic risk score and APOE status in a community-based cohort prospectively followed over 17 years. Mol. Psychiatry 2020 2610 26, 5812–5822 (2020).
- Logue MW, et al. Use of an Alzheimer’s disease polygenic risk score to identify mild cognitive impairment in adults in their 50s. Mol Psychiatry (2018) doi: 10.1038/s41380-018-0030-8.
- Blom ES, et al. Does APOE explain the linkage of Alzheimer’s disease to chromosome 19q13? Am J Med Genet B Neuropsychiatr Genet 147B, 778–783 (2008).
- Potkin SG, et al. Hippocampal atrophy as a quantitative trait in a genome-wide association study identifying novel susceptibility genes for Alzheimer’s disease. PLoS One 4, e6501 (2009).
- Jia L, et al. Prediction of Alzheimer’s disease using multi-variants from a Chinese genome-wide association study. Brain 144, 924–937 (2021).
- Belloy ME, Eger SJ, Guen Y Le, et al. Two APOE splice sQTLs reduce Alzheimer’s disease risk in APOE 4/4 carriers. Alzheimer’s Dement. 16, e043539 (2020).
- Roses AD, et al. A TOMM40 variable-length polymorphism predicts the age of late-onset Alzheimer’s disease. Pharmacogenomics J. 10, 375 (2010).
- Logue MW, et al. A comprehensive genetic association study of Alzheimer disease in African Americans. Arch. Neurol. 68, 1569–1579 (2011).
- Zhou X, et al. Non-coding variability at the APOE locus contributes to the Alzheimer’s risk. Nat. Commun. 10, 1–16 (2019).
- Chaudhury S, et al. Alzheimer’s disease polygenic risk score as a predictor of conversion from mild-cognitive impairment. Transl. Psychiatry 9, 154 (2019).
- Lambert J-C, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease. Nat. Genet. 45, 1452–1458 (2013).
- Young J, et al. Genome-wide association findings from the brains for dementia research cohort. 107, 159–167 (2021).
- Purcell S, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 81, 559–575 (2007).
- R Core team. R: A language and environment for statistical computing. (2021).
- Robin X, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 12, 77 (2011).
- Makiyama K. magicfor: Magic Functions to Obtain Results from for Loops. (2016).
- Escott-Price V, et al. Common polygenic variation enhances risk prediction for Alzheimer’s disease. Brain 138, 3673–3684 (2015).
- Chaudhury S, et al. Polygenic risk score in postmortem diagnosed sporadic early-onset Alzheimer’s disease. Neurobiol. Aging 62, 244.e1-244.e8 (2018).
- Daunt P, et al. Polygenic Risk Scoring is an Effective Approach to Predict Those Individuals Most Likely to Decline Cognitively Due to Alzheimer’s Disease. J. Prev. Alzheimer’s Dis. 8, 78–83 (2021).
- Logue MW, et al. Use of an Alzheimer’s disease polygenic risk score to identify mild cognitive impairment in adults in their 50s. Mol. Psychiatry 2018 243 24, 421–430 (2018).
- Escott-Price V, et al. Genetic analysis suggests high misassignment rates in clinical Alzheimer’s cases and controls. Neurobiol. Aging 77, 178–182 (2019).
- De Bakker PIW, et al. Transferability of tag SNPs in genetic association studies in multiple populations. Nat. Genet. 38, 1298–1303 (2006).
- Carlson CS, et al. Selecting a Maximally Informative Set of Single-Nucleotide Polymorphisms for Association Analyses Using Linkage Disequilibrium. Am. J. Hum. Genet. 74, (2004).
- Johnson SC, et al. The Effect of TOMM40 Poly-T length on Gray Matter Volume and Cognition in Middle-Aged Persons with APOE ε3/ε3 Genotype. Alzheimers. Dement. 7, 456 (2011).
- Roses AD, An inherited variable poly-T repeat genotype in TOMM40 in Alzheimer disease. Arch. Neurol. 67, 536–541 (2010).
- Maruszak A, et al. TOMM40 rs10524523 polymorphism’s role in late-onset Alzheimer’s disease and in longevity. J. Alzheimers. Dis. 28, 309–322 (2012).
- Cruchaga C, et al. Association and expression analyses with single-nucleotide polymorphisms in TOMM40 in Alzheimer disease. Arch. Neurol. 68, 1013–1019 (2011).
- Jun G, et al. Comprehensive search for Alzheimer disease susceptibility loci in the APOE region. Arch. Neurol. 69, 1270–1279 (2012).
- Linnertz C, et al. Characterization of the Poly-T Variant in the TOMM40 Gene in Diverse Populations. PLoS One 7, e30994 (2012).
- Guerreiro RJ & Hardy J. TOMM40 association with Alzheimer disease: tales of APOE and linkage disequilibrium. Arch. Neurol. 69, 1243–1244 (2012).
- Gao L, Cui Z, Shen L, et al. Shared Genetic Etiology between Type 2 Diabetes and Alzheimer’s Disease Identified by Bioinformatics Analysis. J. Alzheimer’s Dis. 50, 13–17 (2016).
- Zhou X, et al. Identification of genetic risk factors in the Chinese population implicates a role of immune system in Alzheimer’s disease pathogenesis. Proc. Natl. Acad. Sci. U. S. A. 115, 1697–1706 (2018).
- Kulminski AM, Philipp I, Shu L, et al . Definitive roles of TOMM40-APOE-APOC1 variants in the Alzheimer’s risk. Neurobiol. Aging 110, 122–131 (2022).
- Xiao Q, et al. The Relationship Between Low-Density Lipoprotein Cholesterol and Progression of Mild Cognitive Impairment: The Influence of rs6859 in PVRL2. Front. Genet. 13, 194 (2022).
- Xiao Q, et al. Risk prediction for sporadic Alzheimer’s disease using genetic risk score in the Han Chinese population. Oncotarget 6, 36955–36964 (2015).
- Zhou X, et al. Genetic and polygenic risk score analysis for Alzheimer’s disease in the Chinese population. 12, e12074 (2020).
- Rao S, et al. An APOE-independent cis-eSNP on chromosome 19q13.32 influences tau levels and late-onset Alzheimer’s disease risk. Neurobiol. Aging 66, 178. e1-178.e8 (2018).
- Jun G, et al. Comprehensive Search for Alzheimer Disease Susceptibility Loci in the APOE Region. Arch. Neurol. 69, (2012).