https://orcid.org/0000-0002-2139-987X
Figures
Abstract
Understanding the genetic foundations of dementia is critical to unraveling its complex molecular basis. Given that a clinical diagnosis of Alzheimer’s disease (AD) dementia often results from interplay between multiple underlying neuropathologic co-morbidities, previous genome-wide association studies (GWAS) of clinically diagnosed AD are restricted in their ability to translate genetic associations to potential targeted therapeutics. The current study seeks to address these limitations by presenting the largest GWAS to date (n = 12,509) of neuropathologic hallmarks of AD and AD related dementias (ADRDs). We further performed a candidate-variant analysis using loci previously identified in GWAS of clinically diagnosed AD dementia and Parkinson’s disease (PD). Finally, we conducted heritability and genetic correlation analyses using linkage disequilibrium (LD) score regression. We found broad genome-wide significant associations with APOE across AD and ADRDs but not cerebrovascular disease and vascular brain injury. We further identified 12 significant loci across 10 neuropathologic phenotypes, including 5 loci previously implicated in GWAS of clinical AD and ADRDs (variants on BIN1, PICALM/ EED, TMEM106B, GRN, and SNCA/ SNCA-AS1) and 7 novel genome-wide associations (variants on EPHA5, PSMG1, LINC00276, VAPA, LINC00290, DOCK4 and SLAIN2/ SLC10A4). Our analysis of AD and PD clinical candidate variants demonstrated several that were associated with AD neuropathologic change and Lewy body disease, as well as substantial overlap with neuropathologic lesions other than the primary neuropathologic hallmarks of these diseases. Heritability analyses demonstrated heritability that was high for amyloid plaques (78%) relative to prior clinical AD heritability analyses, intermediate for TDP-43 inclusions (41%), and low for remaining AD and ADRD pathologic features. This study underscores the importance of investigating the underlying neuropathologic hallmarks of AD and ADRDs as a step toward refining the translation of genetic associations to biomarker interpretation and development of targeted therapeutics.
Author summary
Clinically diagnosed Alzheimer’s disease (AD) dementia is commonly associated with its hallmark pathologic changes plus neuropathologic features of prevalent co-morbid diseases such as cerebrovascular disease, Lewy body disease, and more recently discovered abnormalities in protein called TDP-43 (collectively, AD related dementias; ADRD). As a result, previous studies that associated clinical diagnosis of AD with specific genes may not tell us the whole story. For this study, we gathered autopsy and genetic data to identify relationships between genes and dementia-associated brain changes. We found some relationships between these diseases and genes that had been previously identified as contributing to clinical dementia, as well as some new relationships that had been previously unknown. We also found that some genes that had previously been identified in relation to AD were associated with different dementia-associated brain lesions. Finally, we found that the various brain lesions differ in the proportion that can be attributed to genetic vs. environmental differences. These results support that the pathway to a diagnosis of dementia can be caused by multiple factors and are an important step in beginning to identify individually based dementia treatments.
Citation: Cholerton B, Godrich D, Pasteris J, Rivero J, Martin ER, Kunkle BW, et al. (2026) Genome wide association study meta-analysis of neuropathologic lesions of Alzheimer’s disease and related dementias in a multi-site autopsy cohort. PLoS Genet 22(6): e1012170. https://doi.org/10.1371/journal.pgen.1012170
Editor: J. Nicholas Cochran, HudsonAlpha Institute for Biotechnology, UNITED STATES OF AMERICA
Received: July 24, 2025; Accepted: May 18, 2026; Published: June 29, 2026
Copyright: © 2026 Cholerton et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Genotype and phenotype data are available through the contributing cohorts. ROSMAP data can be requested at https://www.radc.rush.edu and https://www.synapse.org. ADGC data can be requested from NIAGADS at https://www.niagads.org/resources/related-projects/alzheimers-disease-genetics-consortium-adgc-collection. NACC neuropathology data can be requested at https://naccdata.org/. ACT data can be requested at https://actagingresearch.org/. Harmonized neuropathology data are available through NIAGADS at https://dss.niagads.org/datasets/ng00067/. IDIBAPS data can be requested at https://www.clinicbarcelona.org/en/idibaps. TGen data can be requested at https://www.tgen.org/. NIA-LOAD data can be requested at https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000168.v2.p2. Analysis-related scripts are available at https://github.com/beecham-lab/ADRD-Neuropathology-Scripts/.
Funding: This work was supported in part by the National Institutes for Health (NIH/NIA grant number R01 AG062695).GWB, TJM, BC, DG, JP, and JR were supported by NIH: AG062695. GWB, TJM, and TJH were supported by NIH: AG019085. ERM, BWK, CAN, KLH-N, HW, W-PL, MAPV, GS, and the ADGC, were supported by NIH: AG032984, AG036528, AG041689. WK was supported by NIH: AG016976. RPM was supported by NIH: AG026395, AG041797, AG008702, AG015473. DAB was supported by NIH: AG010161. MAPV, GWB, KLH-N, CAN, WS, and MC were supported by NIH: AG027944. MAPV, GWB, and WS were supported by NIH: AG010491, AG021547, AG019757. EBL, PKC, CDK, CSL, and SM, were supported by NIH: AG066567. MIK was supported by NIH: AG0066468, AG0064877, AG007562, AG023651. Additional data collection and coordination under the ADGC banner are listed in the acknowledgements.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Although rare early onset dementia is commonly caused by an aggressive form of a single disease, data from multiple cohorts throughout the world strongly support that much more common late onset dementia is a complex interplay among multiple diseases. Indeed, there is substantial evidence that some mix of multiple neuropathologies, rather than Alzheimer’s disease (AD) only, is the most prevalent pathologic scenario among people clinically diagnosed with AD dementia [1–7]. These commonly co-morbid AD related dementias (ADRDs) include cerebrovascular disease (CBVD) with resulting vascular brain injury (VBI), Lewy body disease (LBD), hippocampal sclerosis (HS), and limbic-predominant age-related TDP-43 encephalopathy (LATE). Importantly, dementia risk in older individuals, already a looming public health crisis, increases geometrically with increasing co-morbidity among AD and ADRDs [2,7,8].
Understanding the genetic architecture of dementia is a critical component to unraveling the molecular basis for dementia. Many of the previous genome-wide association studies (GWAS) for AD focused on genes that correlated with a clinical diagnosis of AD dementia [9–13]. Given the widely validated evidence for multiple neuropathologic processes underlying the clinical diagnosis of AD dementia [7,14,15], unaccounted co-morbidity may significantly undermine the translatability of clinical AD GWAS discoveries because it is unclear which of multiple co-morbidities are actually associated with the identified genetic locus [16]. Translatability is further confounded by AD itself being a complex, multifaceted disease. For example, genetic loci identified by GWAS of AD dementia could operate through amyloid plaque pathways, neurofibrillary tangle pathways, other AD-related neuropathologic changes (e.g., neuroinflammation or vascular contributors, et cetera), or a combination of factors. GWAS of AD CSF biomarkers and amyloid PET may provide additional insights into genes that associate with risk, presence, and progression of AD, including possible comorbid conditions. However, these are limited in scope by their focus on specific proteinopathies. These are all potentially important distinctions when considering translation of genetic associations to therapeutics and interpretation of biomarkers [17–20].
Given these potential limitations of clinical AD GWAS, the inclusion of neuropathologic hallmarks of AD and ADRDs may be potentially useful endophenotypes. However, only a few studies have attempted to include these pathologic endophenotypes in GWAS to disentangle genetic associations with specific diseases rather than their combined impact on clinical expression [21–25]. A particular challenge to focusing on pathologic endophenotypes is limited sample sizes sufficient to power GWAS. Further complicating the effort is historical variation in how neuropathologic data were collected and described.
To address these challenges, we amassed the largest sample to date (n = 12,509) of harmonized neuropathologic and genetic data gathered from the AD Genetics Consortium (ADGC). We hypothesized that using consensus methods for AD and ADRD hallmark lesions within a large sample would permit identification of genetic features that align with each clinic-pathologic entity rather than with the syndromic, functional endpoint of dementia that most commonly derives from multiple co-morbidities. Herein, we (1) present the largest GWAS meta-analysis conducted thus far of the commonly co-morbid neuropathologic lesions of AD and ADRDs assessed by current consensus protocols, (2) analyze previously identified clinical AD and PD GWAS candidates for their associations with AD neuropathologic change (ADNC) and ADRD neuropathologic lesions, and (3) present SNP heritability analysis of ADNC and ADRD lesions in the sample.
Results
Dataset characteristics and pathologic phenotype correlations
After exclusions and quality control, the final dataset consisted of 12,509 individuals. A description of contributing data sets and their demographics are included in Table 1. The dataset was mostly female (53.5%) with a mean age at death of 81.7 (±9.8). The prevalence of APOE ε4 allele was 37% ε4 heterozygotes and 9% ε4 homozygotes.
Consensus scores for ADNC and ADRD lesions were correlated as previously described [26], validating the pathologic assessments in our study. ADNC, including amyloid plaques (APs; assessed by Thal phase or its derivative National Institute on Aging and Alzheimer’s Association [NIA-AA] A score), neurofibrillary tangles (NFTs; assessed by Braak stage or its derivative NIA-AA B score), neuritic plaques (NPs; assessed by Consortium to Establish a Registry for AD [CERAD] NP score or its derivative NIA-AA C score), and the combined NIA-AA ADNC ABC score [27] were strongly positively correlated with each other (r2 = 0.70 to 0.87) and had moderate correlations with APOE ε4, cerebral amyloid angiopathy (CAA), and LBD (r2 = 0.14 to 0.51, S1 Fig). Two measures of CBVD correlated moderately (r2 = 0.31 for atherosclerosis and arteriolosclerosis) while CAA was weakly correlated with these measures of CBVD (r2 = 0.05 to 0.07 for CAA with atherosclerosis or arteriolosclerosis). Two types of VBI (infarcts/lacunes and microinfarcts) also correlated positively but moderately with each other (r2 = 0.28). TDP-43 proteinopathy score and HS were correlated moderately (r2 = 0.32). In terms of frequency, the AD pathologies and CBVD measures were the most frequent (S1 Table). Approximately 85% of the sample set was positive for amyloid at some level, and 81% showed moderate to high neurofibrillary tangles (Braak); when assessed, 90% of the samples were positive for one or more CBVD measure (atherosclerosis, arteriolosclerosis, or CAA).
Genome-wide significant APOE associations with neuropathologic lesions
We observed genome-wide significant (GWS) APOE associations for NFTs, NPs, APs, CAA, LBD, HS, and TDP-43 proteinopathy but not CBVD or VBI, after controlling for age at death and sex (S2 Table).
GWS associations with ADNC
All GWAS were well-controlled for type-1 error (S2 Fig). As expected, we found strong GWS associations in the APOE locus for the dual proteinopathies of AD (Fig 1, S2 Table). Other (non-APOE) genome-wide significant variants are found in Table 2. GWAS of extent of APs (A score; S3A Fig), extent of NFTs (B score, S3B Fig), and NPs (C score S3C Fig) identified GWS signal in the BIN1 locus (top SNP: rs6733839; minimum PC = 1.6 × 10−13 and minimum PB = 1.4 × 10−16; S4 Fig). Effect sizes of top variants on BIN1 were similar for NPs and NFTs (Tables 2, S3). We also found a GWS signal with NPs (C score) on the PICALM/EED locus (top SNP: rs3851179; minimum P = 4.0 × 10−8), another previously identified AD region (S5 Fig). Additional ADNC variables include presence/absence of any amyloid plaques (NPs and/or APs), Thal phase [28], AD Braak stage, and NIA-AA ADNC ABC score. For the presence of APs, we identified 2 novel GWS associations: one near the Ephrin Receptor A5 (EPHA5) gene/ the long intergenic non-protein coding RNA 2232 (LINC02232) and (rs144823952; P = 2.5 × 10−8) and another around the Proteasome Assembly Chaperone 1 gene (PSMG1, top SNP: rs2836880; minimum P = 2.4 × 10−8) which is also known as Down syndrome critical region 2 (S6 Fig). As expected, Thal phase and Braak stage showed similar results to their derived NIA-AA A and B scores (S7 Fig). Excepting the EPHA5 locus, there was no evidence of effect-size heterogeneity at these loci; the EPHA5 locus showed nominal evidence of heterogeneity (Cochran’s Q p-value = 0.0448; Table 2).
Color denotes -log10(pval), capped at 25; size of points denotes absolute value of the effect size from the logistic regression models (e.g., “betas”).
GWS associations with ADRD lesions
CBVD.
We identified three GWS signals among CBVD traits (S8 Fig): two for atherosclerosis and one for presence of CBVD (any atherosclerosis, arteriolosclerosis, or CAA). The atherosclerosis variants include a locus on chromosome 2 (LINC00276, top SNP = rs41446051, P = 4.2 × 10−9) and one variant near the Vesicle-Associated Membrane Protein (VAMP) Associated Protein A (VAPA, rs206499, P = 1.5 × 10−8) (S9 Fig, Table 2). The CBVD locus resides in an intergenic region of chromosome 4 (rs112992465, P = 1.4 × 10−8). There were no genome-wide significant associations with CAA, apart from APOE (S10A Fig). There was, however, one nominally associated variant on chromosome 17 (rs17669645, P = 9.3 × 10−7) near the Cytochrome C Oxidase Assembly Factor Heme A:Farnesyltransferase (COX10) gene (S10B Fig; S3 Table).
VBI.
GWAS of presence of infarcts/lacunes (S11 Fig) showed a GWS signal on the Dedicator of Cytokinesis 4 (DOCK4) gene (top SNP = rs6976029, P = 1.1 × 10−8) (S11 Fig, Table 2). Additional variables related to VBI, such as microinfarcts and to some extent white matter rarefaction, are presented in supplemental material (S12 Fig, S3 Table).
LBD.
All primary models for each LBD categorization revealed GWS association with the APOE region with the same top SNP (rs429358; LBD 5-group, P = 9.2 × 10−17; LBD 3-group, P = 5.7 × 10−18; LBD 2-group, P = 1.9 × 10−18) (S2 Table; S13, S14 Figs). None showed further GWS associations in the primary model. However, in the secondary model in which APOE ε4 count was a covariate, all three LBD groupings showed GWS associations with the alpha-synuclein gene and its antisense transcript (SNCA/ SNCA-AS1) locus with the same top SNP (rs3806789; LBD 5-group, P = 4.3 × 10−8; LBD 3-group, P = 4.1 × 10−8; LBD 2-group, P = 2.8 × 10−8) (S13 Fig). In primary models, top signals in the region reached suggestive significance (rs3806789; LBD 5-group, P = 7.3 × 10−8; LBD 3-group, P = 7.3 × 10−8; LBD 2-group, P = 5.3 × 10−8) (S14 Fig).
TDP-43 proteinopathy and HS.
TDP-43 proteinopathy and HS are correlated (S1 Fig) and are commonly observed in FTLD-TDP, LATE-NC, and ADNC. Three loci achieved GWS with HS: APOE (lead SNP = rs429358, minimum P = 4.1 × 10 − 9), transmembrane protein 106B (TMEM106B; lead SNP = rs4721058, minimum P = 1.9 × 10 − 15), and granulin precursor gene (GRN; lead SNP = rs5848, minimum P = 1.3 × 10 − 10) (S15, S16 Figs). There was nominal evidence of effect-size heterogeneity for the TMEM106B locus with the HS phenotype (Cochran’s Q p-value = 0.0471). Primary results for these phenotypes have been previously reported elsewhere and are more fully discussed there [29].
Candidate variant associations AD-candidate genetic variant associations with ADNC and ADRD lesions.
Given the phenotypic correlations (S1 Fig) and cooccurrence of neuropathological lesions, we assessed candidate variants from clinic-based AD and PD association studies for pleiotropic effects across the spectrum of neuropathological lesions. Of the 78 variants in our data that overlap with the Bellenguez et al [30]. clinical AD GWAS, 20 were nominally associated (P < 0.05) with A score for ordinal ranking of APs, 30 with B score for ordinal ranking of NFTs, and 21 with C score for ordinal ranking of NPs; 12 of the loci were associated at a more stringent Bonferroni correction (p-value < 0.05/78) for one or more of the tests. Variants with nominal significant in A, B, or C score are shown in Table 3. Of these, 11 associations are shared across all three scores; 9 were shared by ranking for NFTs and APs (B Score plus A and/or C Scores); eight variants were associated with ranking for APs alone (A and/or C Scores); and ten variants were associated with ranking NFTs alone (B Score) (Table 3). Of the 78 variants, 40 were not associated with any of the three ADNC scores.
We extended this candidate variant analysis to ADRD lesions and again observed shared genetic risk with ADNC, although not as strong as APOE. Of note, the majority of candidate variants associated with the neuropathologic changes assessed were not associated with ADNC alone; indeed, 24 variants associated with ADNC were also associated with one or more ADRDs, and 24 variants were associated with ADRDs but not ADNC (Fig 2). Eight variants were associated with LBD presence or severity (SORL1, SC1MP, IGH gene cluster, ADAM17, BIN1, SPI1, TMEM106B, and USP6NL), 3 with CAA severity (TPCN1, MINDY2, and EED) and an additional 7 with presence of CAA (SORL1, COX7C, IGH gene cluster, CLU, MAF, SLC2A4RG, and ACE). Additional nominal association signals were observed with HS, TDP-43, CBVD, and VBI (S4, S5 Tables).
Plain text = AD candidate genes from Bellenguez et al. (2022) GWAS. Bold text = PD candidate genes from Nalls et al. (2019) GWAS. Other ADRD = HS, CAA, CBVD, VBI, or TDP-43 proteinopathy. *Genes with multiple variants represented within the diagram. Abbreviations: AD, Alzheimer’s disease; ADRD, AD related dementias; CAA, cerebral amyloid angiopathy; CBVD, cerebrovascular disease; PD, Parkinson’s disease; VBI, vascular brain injury.
PD-candidate genetic variant associations with ADNC and ADRD lesions.
Mounting clinical, pathological, and biochemical data support some degree of overlap between AD and PD. For this reason, we examined the 90 previously identified PD clinical variants by Nalls et al. [31]. for association with the pathologic endophenotypes for AD and ADRD. Of these, 11 were associated with LBD (INPP5F, LINC00693, SEMA4A, TMEM163, SNCA, RIT2, SCARB2, TMEM175, KCNS3, SATB1, and KRTCAP2) in our analyses; most of these signals were robust against changes in how LBD was categorized (e.g., 5 ordinal ranks, 3 ordinal ranks, or presence/absence) (S6 Table). Several PD-candidate variants also were associated with ADNC (Fig 2, S7 Table), including 8 associated with A score, 10 associated with B score, 11 associated with C score, and 4 associated with ADNC ABC score. The PD candidate variants also were associated with other ADRD pathologic features, including CAA severity (5 variants), CAA presence (9 variants), HS (10 variants), TDP-43 proteinopathy presence (4 variants), and TDP-43 severity (9 variants) (S8 Table).
SNP heritability of ADNC and ADRD pathologic features
Heritability and genetic correlations.
We performed SNP heritability analyses using linkage disequilibrium (LD) score regression (LDSC) on the summary statistics from the above GWAS analyses. AP severity by Thal phasing showed the highest SNP heritability (h2: 0.78; Table 4), neuritic plaque severity by CERAD score had intermediate heritability (h2: 0.24), and NFT severity by Braak staging had lowest heritability (h2: 0.19) among ADNC. TDP-43 proteinopathy measures (h2:0.21-0.41) and presence of CAA (h2 = 0.26) also had intermediate heritability. Other ADRD pathologic features (LBD, CBVD, and VBI) tended to have lower heritability (h2 < 0.20). While most of the heritability estimates were significantly larger than zero, many had large standard errors and wide confidence intervals (Table 4; full results in S9 Table). Phenotypic correlations (S1 Fig) motivated the assessment of genetic correlations among traits with significant heritability. These analyses suggested shared genetic architecture of CAA and presence of AP (P = 0.04 for presence of CAA and presence of AP; P = 0.044 for severity of CAA and presence of AP). However, overall, the genetic correlation estimates tended to be unstable (i.e., high variance on correlation estimates, leading to wide confidence intervals), suggesting the need for analysis with larger sample sizes (S10 Table). Additional heritability analyses are described in the supplemental material (Supplemental Text; S17, S18 Figs; S11–S14 Tables); there, we describe the assessment of cell and tissue-specific polygenic effects for neuropathologic changes [32]. That is, we assess whether the SNP heritability estimates noted above are concentrated or enriched in specific cell or tissue types. Referent cell types were generated by a genome-wide study of tissue-specific gene expression in humans (GTEx), and a study of immune cell types in mice (ImmGen) [32,33].
Pathway and set-based tests
Pathway and set-based analyses were performed using MAGMA enrichment analysis (S15–S17 Tables). The pathway analyses (GO, Biocarta, etc) yielded results overlapping with known AD pathways (S15 Table). Top pathways for AD-related neuropathology phenotypes were dominated by immunity-related pathways (e.g., association between the Reactome pathway “Activation of RAS in B cells” and Thal Phase; p-value = 6.1 x 10-7). Other known AD pathways (cell signaling, amyloid, lipid metabolism) were also present, though with more nominal associations. Additional tissue-specific analyses are also included, based on MAGMA rather than LDSC (S16, S17 Tables).
Discussion
We present results from the largest GWAS of AD and ADRD neuropathologic lesions conducted to date, using data from several cohorts that collected research quality clinical, neuropathologic, and genetic data. We confirmed significant association of APOE with both hallmark lesions of AD as well as with all other ADRDs excepting cerebrovascular arteriolosclerosis, cerebrovascular atherosclerosis, and VBI. Furthermore, we mapped ten significant loci to ten disease-specific hallmark lesions, including five loci previously associated with AD dementia (BIN1, PICALM/EED, TMEM106B, GRN, and SNCA/SNCA-AS1) and five novel loci (EPHA5, PSMG1, LINC00276, VAPA, and DOCK4). While the largest study of its type, we recognize that the size of our cohort is still modest for GWAS. For this reason, we also analyzed clinical AD and PD GWAS candidates and found substantial overlap of candidate variants with multiple ADNC and ADRD lesions. Finally, we found variable heritability among ADNC and ADRD lesions.
GWS associations with neuropathologic hallmarks
The APOE region was significantly associated with all five proteinopathies – AP, CAA, NFT, LBD, and TDP-43 inclusions - validating a broad role for APOE variants in AD and commonly comorbid ADRD [25,34–36]. Indeed, associations between APOE and clinical AD GWAS may be especially strong because its association with pathologic features of both ADNC and ADRD. Many proposals, spanning from co-seeding of protein aggregates to shared influences of the biology of aging, have been offered to explain why LBD and TDP-43 inclusions co-occur with ADNC more commonly than expected by chance alone [37–39]. Our data support that the increased likelihood of co-occurrence among ADNC, LBD, and TDP-43 inclusions is at least in part because these distinct pathologic features share genetic risk at the APOE locus. Although functional validation prior to drawing any conclusions as to the clinical implications of these findings will be vital, these results suggest that explorations into potential treatments targeting ApoE variants may be appropriate not only in AD but also multiple ADRDs. Interestingly, other than CAA, our results show only nominal association between the APOE locus and CBVD, and no association with forms of VBI. This could suggest a difference in the influence of ApoE isoforms in central vs. peripheral circulations [40].
Consistent with previous smaller autopsy GWAS, we observed strong GWS association between BIN1 variants and NFT measures but weak GWS associations between BIN1 and PICALM/ EED variants with NP measures [21,25,41]. We also identified two novel GWS variants associated with presence of APs on PSMG1 and EPHA5. Variants on PSMG1 (also known as Down Syndrome Critical Region Gene 2) have been implicated in several cardiometabolic phenotypes [42–48]. Such associations, together with appropriate functional genetic follow-up, may provide a biological context for exploring the connection between PSMG1 and AD/ADRD, and may help to identify future potential vascular therapeutic targets. Variants in EPHA5 have been associated with general cognitive ability and are in the same ephrin receptor family as EPHA1 [49,50], a known clinical AD-associated genetic risk factor [51,52]. Given the role of ephrin receptors in cell and axon guidance and in synaptic development and plasticity [53,54], EPHA5 represents a biologically plausible target through which modulation of ephrin receptor pathways could support synaptic resilience in AD/ADRD. It is important to note that these associations require replication, and functional validation is needed to confirm mechanistic relevance and inform potential future therapeutic development.
Our data suggest that VAPA and two long noncoding RNA (LINC00276, LINC00290) may be associated with CBVD. Variants in LINC00276 are associated with multiple cardiometabolic, hematological, and immune measures, including platelet, monocyte, and lymphocyte counts, waist-to-hip ratio [55]. LINC00290, similarly, has been associated with immune phenotypes, as well as age of onset for AD [56]. For macroinfarcts, our GWAS identified a significant association with variants at DOCK4, which participates in the transport of low-density lipoproteins [57]. DOCK4 has also been associated with multiple immune and cardiovascular phenotypes [50,55,57–62].
Clinical candidate variants
Our candidate-variant approach based on clinical AD GWAS [30] revealed 38 of 78 candidate variants associated with one or multiple ADNC features with nominal significance (12 significant after Bonferroni correction). Once validated, these may represent targets for translating these genetic risk associations into therapeutics. For example, CLU and FERMT2 were both associated with APs but not NFTs, and their encoded proteins alter Aβ aggregation and clearance [63] as well as APP metabolism and Aβ peptide production [64]. Of note, 11 of the clinical AD GWAS candidate variants did not associate with any ADNC but instead were associated with one or more ADRDs, and 19 clinical AD GWAS variants that were associated with one or more ADNCs also were associated with one or more ADRD pathologic endophenotypes. These results provide compelling support for the heterogeneous nature of clinical AD, such that multiple pathways may lead to similar clinical outcomes. Interestingly, 40 candidate variants were not associated with any of the ADNC lesions. This last group is especially important because they may provide novel insights into largely unexplored hallmark-independent potential therapeutic approaches as well as resilience factors that suppress clinical expression of disease without altering pathologic hallmarks.
We do note some overlap between the current study and prior CSF and imaging GWAS. For example, BIN1 has been reportedly associated with both CSF Aβ42 and tau, but not with amyloid PET [18,19]. This suggests that BIN1 may act early in the disease process to influence both Aβ42 and tau, and over time contribute to the accrual of plaques and tangles, while mid-stage markers may not capture the phase at which BIN1 exerts its strongest effects. Alternatively, we report associations with amyloid and FERMT, a finding that is supported by amyloid PET, but not CSF, studies, suggesting that the variant on FERMT may influence more moderate to late stages of amyloid accumulation. Additional studies to determine stage-dependent genetic architecture of AD and genetic overlap between biomarker modalities will ultimately be needed to further understand these relationships and facilitate identification of treatment targets.
Similarly, a candidate variant approach for clinical PD [31] found significant candidate associations with LBD, but also with ADNC, CAA, HS, and TDP-43. Of the 103 clinical PD candidate variants assessed 11 were nominally associated with LBD. However, 23 were associated with ADNC and 29 with ADRD lesions (CAA, HS, TDP-43 proteinopathy). For example, KRTCAP2 was associated with LBD, but also nominally associated with measures of AP, NFT, and TDP-43 proteinopathy. These results further underscore how ADNC and ADRD lesions share partial genetic overlap with clinically diagnosed AD and PD.
Heritability of neuropathology
Our heritability analyses suggest a strong polygenic effect for AP severity with high SNP heritability (up to 78%) relative to GWAS of clinical AD dementia (e.g., 9–30%) [65,66]. Heritability estimates for presence of TDP-43 inclusions was intermediate (41%) while NFT severity and CBVD measures showed relatively low heritability (11–26%) with presence of CAA, another form of amyloidosis, being the highest. This is consistent with moderate heritability of cardiovascular traits, and the higher heritability of amyloid measures.
Limitations
There are several potential limitations to our GWAS of ADNC and ADRD lesions. First, while this study represents the largest brain autopsy AD GWAS to date, it is still limited in size compared to the clinical AD GWAS that surpasses one million individuals. Because of this, we cannot resolve whether the lack of association of some candidate variants to neuropathologic hallmarks is due to low statistical power or to the variants associated with hallmark-independent processes. Second, here we utilized a fixed-effects meta-analysis to maximize power, so effect-size heterogeneity may be of some concern for this study, due to sample sets coming from multiple sources and genotyping arrays. Genome-wide heterogeneity statistics, however, do not support this: genome-wide heterogeneity statistics closely followed the null expectation, both within and across neuropathological phenotypes (see S18 Table), and there was minimal evidence of test-statistic inflation (GIF; see S2 Fig). However, an additional random-effects meta-analysis may provide additional insight into this question. Third, this study lacks racial and ethnic diversity among participants. Genetic variants may have different frequencies and/or relevance in other populations, and thus risk prediction based on these studies may not translate to non-European populations. The lack of generalizability to wider populations highlights the need for improved outreach and diversity in brain autopsy programs. Fourth, our explorations into AD and PD clinical candidate variants revealed several nominal associations, raising questions about multiple testing correction, and appropriate modeling. The correlated nature of some of the neuropathological phenotypes (S1 Fig) may allow for improved efficiency and power with more complex multivariate modeling (e.g., MANCOVA), at least for some ensemble hypotheses; this will be the subject of future studies. Similarly, the multiple-testing correction of the candidate variants is worthy of consideration. Typically, multiple-testing corrections are relaxed when there is strong a priori evidence for the hypothesis. In this case, we have strong evidence for those loci from the clinical AD and PD GWAS, reducing the need for correction, for this subset of analyses. It is important to note here that these analyses are by nature exploratory and require independent replication due to the relatively small sample size in the context of GWAS. The wide confidence intervals noted in many of the heritability and genetic correlation estimates are likewise likely related to relatively small sample size, again highlighting the need for additional confirmation studies. Fifth, it is of note that white matter phenotypes (rarefaction), while commonly reported, do not yet have consensus protocols and reporting procedures. As such, we largely relegated analyses related of these phenotypes to supplemental material (S1, S2, S3, S5 Table). Finally, while additional replication of our GWS associations is needed, we note that the meta-analysis methodology used here serves as a form of replication.
Conclusion
Clinical AD GWAS are well-powered to identify candidate genes yet leave unaccounted the impact of common co-morbidities that potentially limit the translatability of findings. By offering greater pathologic specificity, GWAS-identified loci of neuropathologic endophenotypes, once validated, may provide important insights into potential disease-relevant pathways and enable the nomination of genetic and molecular biomarkers that may serve as targets for disease detection and intervention. Our study underscores a spectrum of genetic risk that is partially shared, most notably for APOE and TMEM106B variants, and partially distinct across pathologically verified AD and ADRD, providing a potential explanation for how these classically distinct clinical entities share neuropathologic features that co-occur more commonly than by chance alone. Although autopsy studies are limited by sample size, and thus genes not associated with any pathologic endophenotype might be underpowered for discovery in our cohort, the associations revealed here provide an important basis for further discovery. Alternatively, genes not identified here may be associated with neuropathologic hallmark-independent or yet to be discovered disease mechanisms, an important caveat for ongoing investigations into the genetic architecture of dementia.
Methods
The methodology of the study is summarized and outlined in Fig 3.
Sample acquisition and selection
Participant selection.
Participants were selected from the ADGC (G.D. Schellenberg, PI) and affiliated studies. Contributing studies include the National Institute on Aging (NIA) funded AD Research Centers (ADRCs), the Adult Changes in Thought Study (ACT; L.K. McEvoy, P. Crane, A.Z. LaCroix, PIs), the Religious Orders Study/Memory and Aging Project (ROSMAP; David Bennett, PI), NIA Late-Onset AD Family Study (NIA-LOAD; Richard Mayeux, PI), Translational Genomics Research Institute (TGen; Eric Reiman, MD), the University of Pittsburgh ADRC (Ilyas Kamboh, PI), the Neurological Tissue Bank at the Biobank-Hospital Clínic, Instituto de Investigaciones Biomédicas August Pi i Sunyer, Barcelona (IDIBAPS; Laura Molina Porcel, PI), the Mayo Clinic ADRC (R.C. Petersen, PI), and the University of Miami Hussman Institute for Human Genomics (AD, Margaret Pericak-Vance, PI; Parkinson’s Disease, Jeffrey Vance, PI). The samples were predominantly from clinical dementia patients (Table 1), mostly with AD as primary suspected etiology. This reflects both the focus of contributing studies (largely AD dementia and cognitive aging) as well as bias in participation in autopsy programs. IDIBAPS is an autopsy-based study with no clinical data on participating samples (beyond sex and age at death). ADRC phenotypic data were obtained through the National Alzheimer’s Coordinating Centers (NACC; Walter Kukull, PI). The study here reported is a secondary data analysis study conducted with approval of the Wake Forest University School of Medicine IRB. No samples or primary data were collected as part of this study, and all data analyzed are deidentified, and from deceased individuals. As such, in the study falls under exempt human subjects research. For contributing studies all participants (or representatives) provided written informed consent; all protocols and assessments were performed with approval by the institutional internal review boards of the contributing studies.
Inclusion criteria.
ADGC participants were included if age at death was greater than 50 years and both neuropathologic data and genetic array data were available. Participants with dementia whose primary dementia etiology was determined to be non-AD or non-ADRD (e.g., traumatic brain injury, chronic drug/alcohol use, etc.) were excluded.
Genotyping and quality control
Genotyping of the ADC sets was performed at the Children’s Hospital of Pennsylvania, but genotyping chips differed across ADCs. For smaller sample sets we combined like chips into batches after initial QC and imputation (Table 1). Genotyping of ADGC-collaborating sets was performed on a variety of genotyping platforms and is described in elsewhere [11,52]. IDIBAPS participant samples underwent genotyping with the Illumina NeuroBooster array at the University of Miami. Imputation quality (r2) for genome-wide significant loci are noted in the supplementary material (S19 Table).
The standardized ADGC quality control pipeline was performed on the sample and variant level, detailed elsewhere [11,52]. Briefly, samples or variants with low call rates (sample missingness > 2%; variant missingness > 5%), sex discordance, or deviations from Hardy-Weinberg Equilibrium (PHWE < 10-6 among controls) were dropped. Relatedness checks were performed with the KING algorithm from the SNPRelate package [67,68]. The analysis revealed identical pairs (kinship ≥ 0.480) that were subsequently dropped. For pairs with lower kinship (0.177 ≤ kinship < 0.480), only one individual from each related pair was kept, whichever belonged to the larger dataset. If that was accounted for, the sample with the most complete (non-missing) neuropathology phenotype information was kept. The samples were imputed with the Trans-Omics for Precision Medicine program server (r2) [69]. Genetic variants with minor allele frequency (MAF) ≥ 0.01 and imputation quality score R2 ≥ 0.40 were used for analysis. When like sets were combined, MAF and R2 scores were weighted by comprising sets. After imputation, principal components analysis was conducted using PC-AiR to assess and account for population substructure [70]. Outliers for genetic ancestry (>6 standard deviations from mean within any of the first ten PCs) were dropped.
Phenotype harmonization
To the extent possible, we used consensus methods for neuropathologic lesion assessment and scoring [27,71,72], as we have done previously (S20 Table) [21]. When scoring methods were incompatible with consensus methods, we first restricted categorization to achieve compatibility or, when that was not achievable, we created less granular categorizations to accommodate the incompatible coding or less distinct codings [21]. For example, LBD was categorized in three different ways: grouped into five categories (0 = none, 1 = olfactory or unspecific region, 2 = brainstem predominant, 3 = limbic, and 4 = neocortical), grouped into 3 categories (0 = none, 1 = olfactory, unspecific region, or brainstem predominant, and 2 = limbic/neocortical), and binary to insure compatibility across datasets while retaining the largest possible sample set. Meta-analyses were performed on seven AD phenotypes and 19 ADRD phenotypes, with some phenotypes representing derivations of others (primarily ordinal phenotypes reduced to an any/none binary phenotype) (S19 Table).
Statistical analyses
Unless noted, statistical analyses were conducted in R version 4.2.2 [73]. Spearman correlation and the corrgram R package were used to assess correlations [74].
Single variant GWAS meta-analysis.
GWAS was conducted with either logistic or ordinal logistic regression model testing for effects of dosages of imputed genotypes and adjusting for age at death, sex, and the first three principal components within dataset/batch. Regression for GWAS was performed using RVtests for binary endpoints or the “ordinal” package in R for ordinal endpoints, on each of the available datasets [75,76]. Batch-specific results were then combined across datasets in a fixed-effect meta-analysis with an inverse-variance weighted approach, as implemented in METAL, excluding genetic variants appearing in less than 30% of datasets [77]. Fixed-effects meta-analysis was applied to maximize statistical power for genetic discovery. Genomic control was not applied at the individual cohort level; heterogeneity (Q statistic) was calculated for each test. Proportion of heterogeneity statistics reaching nominal significance (p-value < 0.05) are noted in the supplemental material (S18 Table). QQ plots were generated along with genomic inflation factors. As a sensitivity analysis, we also conducted GWAS adjusting for APOE ε4 count. These results were very similar to the primary analyses, besides the lack of genome-wide signal from variants in APOE.
QQ plots and genomic inflation factor (λ) were generated in R using the “qqman” package to assess possible inflation from false positives and excluded the APOE region (chr19:44-46Mb) [78]. Manhattan plots to visualize the GWAS meta-analyses were created using the “ggplot2” package in R [79]. Figures for regional association signals were created with LocusZoom [80].
Functional analysis.
Top signals from GWAS meta-analyses were followed up for functional assessment using scoring and eQTL databases via in Functional Mapping and Annotation (FUMA; v1.5.2) [81] combined annotation-dependent depletion [82], and RegulomeDB [83]. MAGMA, also implemented in FUMA, was utilized for set-based analysis of biochemical pathways and ontology sets, as well as GTEx-based tissue-based sets.
AD-candidate variants analysis.
We leveraged summary statistics from a recent clinical AD GWAS meta-analysis by Bellenguez et al. totaling 111,326 clinically diagnosed or proxy AD cases and 677,663 controls that identified 83 independent GWS lead variants (excluding APOE) [30]. We located these variants within our dataset and found 78 variants, dropping 5 variants with either MAF < 0.01, not imputed on our reference panel, or imputed with R2 < 0.40. The AD-candidate variants dropped from analysis were on TREM2, SORT1, TREML2, NCK2, and PLCG2. Additionally, we did not consider any variants on APOE for this analysis. Using these 78 AD-candidate variant dosages in our dataset, we tested for association with both AD (A score, B score, C score, and ADNC) and comorbid pathologies (LBD, CAA, HS, and LATE-NC) using either logistic or ordinal logistic regression. A priori and a posteriori power calculations for AD-candidate variants are reported in S21 Table.
PD-candidate variants analysis.
A recent meta-GWAS of Parkinson’s disease (PD) by Nalls et al. including 37.7K cases, 18.6K UK Biobank proxy-cases (having a first degree relative with PD), and 1.4M controls revealed 107 lead variants with independent GWS associations with PD, of which 90 passed quality control [31]. We were able to identify 100 of these 107 variants in our dataset, again excluding rare variants (MAF < 0.01) that were not imputed on our reference panel or variants that did not impute well (R2 < 0.40), including variants on GXYLT1, FGD4, GBA, LRRK2, PMVK, SEMA4A, and DPM3. For each of these 100 variants, we tested for association with three categorizations of LBD. We also tested AD lesion scores (A score, B score, C score, and ADNC) and other common comorbid diseases (CAA, HS, and LATE-NC) for association with PD-candidate variants.
Heritability and genetic correlation analyses.
We performed heritability and genetic correlation analyses using LD Score Regression (LDSR) [84] using the LDSC software (https://github.com/bulik/ldsc/). To calculate heritability, we used the APOE adjusted summary statistics, with the APOE region removed (+/- 500kb of APOE coding region). Reference LD scores were computed using the 1000 Genomes European subset (obtained via [href: https://github.com/bulik/ldsc/wiki/] https://github.com/bulik/ldsc/wiki/, January 2025). The H2 intercept was constrained to 1 (--intercept-h2). Prevalences were included (--samp_prev, --pop_prev), with sample prevalence based on our dataset. Since population prevalences for pathology are not widely available, we estimated the pathology prevalences within dementia cases and cognitively intact subsets (using NACC, ROSMAP, and ACT subsets). These dementia/intact-specific prevalences were then weighted by the population-prevalences of dementia (0.13 for this age-range) to estimate the population prevalence of neuropathological lesions. Genetic correlations were also estimated using LDSR (--rg) with the above noted references and prevalences, but without constraining H2; the intercept for rg was not constrained to account for sample overlap. Cell-specific enrichment of heritability methodology is reported in supplemental material (Supplemental S1 Text), together with the results (S11, S12, S13, S14 Tables; S17, S18 Fig).
Supporting information
S1 Text. The Supplemental Text includes details on additional SNP heritability analyses, including cell and tissue-specific enrichment analyses.
Text includes results, discussion, and methodology specific to SNP heritability analyses.
https://doi.org/10.1371/journal.pgen.1012170.s001
(DOCX)
S1 Fig. Correlation plot of neuropathology and related variables.
Spearman correlation of neuropathology phenotypes and relevant covariates. Acronyms and abbreviations: APOE4, count of e4 allele; AAD, age at death; Thal, Thal Phase; Braak, NFT Braak stage; CERAD, CERAD NP score; ADNC, ADNC score (ABC score); ATH, cerebrovascular atherosclerosis; ART, cerebrovascular arteriolosclerosis; CAA, cerebral amyloid angiopathy; LBD, Lewy body disease; INFA, macroinfarcts/lacunes; MICR, microinfarcts; WMR, white matter rarefaction; VBI, vascular brain injury; HS, hippocampal sclerosis; TDP43, TDP-43 proteinopathy.
https://doi.org/10.1371/journal.pgen.1012170.s002
(DOCX)
S2 Fig. QQ plots for genome-wide association analyses.
Quantile-quantile plots for each genome-wide association study. λ denotes the genomic inflation factor for the study.
https://doi.org/10.1371/journal.pgen.1012170.s003
(DOCX)
S3 Fig. P-value by genomic position for A score, B score, and C score.
Genome-wide association results for AD hallmark pathologies. P-values reported on the -log(10) scale. Variants at APOE with -log10(p-value) greater than 17 were censored to improve readability.
https://doi.org/10.1371/journal.pgen.1012170.s004
(DOCX)
S4 Fig. Regional association plot for the BIN1 locus, for A Score (NP/Thal), B Score (NFT Braak), and C Score (CERAD).
Regional association plots for the BIN1 locus for AD hallmark pathologies. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s005
(DOCX)
S5 Fig. Regional association plot for the PICALM/EED locus, for A Score (NP/Thal), B Score (NFT Braak), and C Score (CERAD).
Regional association plots for the PICALM locus for AD hallmark pathologies. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s006
(DOCX)
S6 Fig. Regional association plots and forest plots for genome-wide significant variants from the amyloid plaque (presence/absence) analysis.
Regional association and forest plots for the EPHA5 and PSMG1 loci for presence/absence of amyloid plaques. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s007
(DOCX)
S7 Fig. P-value by genomic position for association with Thal phase, NFT Braak, and ADNC (ABC) score.
Genome-wide association results for Thal Phase, Braak (NFT), and ADNC composite score. P-value was capped at 1e-15. The minimum p-values for APOE were: p-value(THAL) = 6.379e-63, p-value(NFT BRAAK) = 8.06e-147, and p-value(ADNC) = 1.916e-55.
https://doi.org/10.1371/journal.pgen.1012170.s008
(DOCX)
S8 Fig. P-value by genomic position for cerebral atherosclerosis (any/none) and cerebrovascular disease (any/none).
Genome-wide association results for atherosclerosis (any/none) pathology, and cerebrovascular disease (any/none).
https://doi.org/10.1371/journal.pgen.1012170.s009
(DOCX)
S9 Fig. Regional association plots and forest plots for genome-wide significant variants from cerebral atherosclerosis (any/none) and cerebrovascular disease (any/none) analyses.
Regional association and forest plots for the significant chromosome 2 locus (atherosclerosis any/none), VAPA (atherosclerosis any/none), and the significant chromosome 4 locus (CBVD any/none).
https://doi.org/10.1371/journal.pgen.1012170.s010
(DOCX)
S10 Fig. P-value by genomic position for association with cerebral amyloid angiopathy (CAA) analysis, and regional association analysis of the COX10 region.
Genome-wide association results for cerebral amyloid angiopathy, and regional association plot for the COX10 region. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s011
(DOCX)
S11 Fig. P-value by genomic position for association with infarcts/lacunes analyses, and regional association analysis of the DOCK4 region, and forest plot for the index variant.
Genome-wide association results for infarcts/lacunes, and regional association plot for the DOCK4 region. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s012
(DOCX)
S12 Fig. P-value by genomic position for association with microinfarcts, hemorrhages, and whole brain vascular disease (WBVD).
Genome-wide association results for microinfarcts, hemorrhages, and whole brain vascular disease pathologies. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s013
(DOCX)
S13 Fig. P-value by genomic position for association with lewy body (PD Braak) analyses, and regional association analysis of the SNCA/MMRN1 region.
Genome-wide association results for Lewy body pathology (any/none), and regional association plot for the SNCA/MMRN1 region. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s014
(DOCX)
S14 Fig. P-value by genomic position for association with lewy bodies.
Model 1 includes age, sex and principal components in the model; model 2 includes APOE e4 allele count as well as age, sex, and principal components. Three different parameterizations of lewy bodies were included: PD Braak is the standard 5 category Braak staging for severity of LBD; the “3 cat(egory)” parameterization collapses these into three groups; “Any/None” parameterizes LBD to presence/absence.
https://doi.org/10.1371/journal.pgen.1012170.s015
(DOCX)
S15 Fig. P-value by genomic position for association with TDP-43 proteinopathy and hippocampal sclerosis.
Genome-wide association results for TDP-43 proteinopathy (any/none) and hippocampal sclerosis. P-values reported on the -log(10) scale.
https://doi.org/10.1371/journal.pgen.1012170.s016
(DOCX)
S16 Fig. Regional association plot for the TMEM106B locus, for TDP-43 proteinopathy and hippocampal sclerosis, and the GRN locus for hippocampal sclerosis.
Regional association plot for the TMEM106B locus, for TDP-43 proteinopathy (presence/absence) and hippocampal sclerosis, and the GRN locus for hippocampal sclerosis.
https://doi.org/10.1371/journal.pgen.1012170.s017
(DOCX)
S17 Fig. Cell-specific heritability enrichment results (multi-tissue GTEx).
Results from cell-specific heritability enrichment analyses. Y axis denotes -log10(p-value) of the enrichment score. The dashed line denotes nominal association. Study-wide significance by false discovery rate (FDR) would be at -log10(p-value) = 2.75. Variant sets are derived from human gene expression data from the GTEx study, as in Finucane et al. See Supplemental Text for more details.
https://doi.org/10.1371/journal.pgen.1012170.s018
(DOCX)
S18 Fig. Cell-specific heritability enrichment results (ImmGen).
Results from cell-specific heritability enrichment analyses. Y axis denotes -log10(p-value) of the enrichment score. The dashed line denotes nominal association. Study-wide significance by false discovery rate (FDR) would be at -log10(p-value) = 3.03. Variant sets are derived from mouse derived immune gene expression data from the GTEx study, as in Finucane et al. See Supplemental Text for more details.
https://doi.org/10.1371/journal.pgen.1012170.s019
(DOCX)
S19 Fig. QQ Plots and lambdas of individual cohort analyses.
https://doi.org/10.1371/journal.pgen.1012170.s020
(DOCX)
S1 Table. Frequency and proportion of neuropathological lesions.
Abbreviations: AP, amyloid plaques; CAA, cerebral amyloid angiopathy; CBVD, cerebrovascular disease; PD Parkinson disease (e.g., Lewy bodies); HS, hippocampal sclerosis; VBI, vascular brain injury; WBVD, whole brain vascular disease; WMR, white matter rarefaction.
https://doi.org/10.1371/journal.pgen.1012170.s021
(XLSX)
S2 Table. Association of APOE e4, across neuropathology phenotypes (rs429358; chr19:44,908,684).
Abbreviations: AP, amyloid plaques; CAA, cerebral amyloid angiopathy; CBVD, cerebrovascular disease; RD, (AD) related dementias; PD Parkinson disease (e.g., Lewy bodies); HS, hippocampal sclerosis; VBI, vascular brain injury; WBVD, whole brain vascular disease; SE, standard error WMR, white matter rarefaction Het Q, Het df and Hetp-value note the heterogeneity test statistic, degrees of freedom, and p-value, respectively.
https://doi.org/10.1371/journal.pgen.1012170.s022
(XLSX)
S3 Table. All genome-wide association results with p < 0.0001.
https://doi.org/10.1371/journal.pgen.1012170.s023
(XLSX)
S4 Table. Association of Bellenguez et al AD risk loci with AD related dementia pathologies.
Odds ratios (OR) refer to the risk allele (RA) from Bellenguez et al. GWAS. Highlighted and bolded ORs indicate significance (P < 0.05). Only variants showing significant association with at least one lesion is included in the table. Abbreviations: RA, risk allele; OA, other allele; OR, odds ratio.
https://doi.org/10.1371/journal.pgen.1012170.s024
(XLSX)
S5 Table. Association of Bellenguez et al AD risk loci with cerebrovascular and vascular brain injury pathologies.
https://doi.org/10.1371/journal.pgen.1012170.s025
(XLSX)
S6 Table. Association of Nalls et al PD risk loci with lewy body pathology.
https://doi.org/10.1371/journal.pgen.1012170.s026
(XLSX)
S7 Table. Association of Nalls et al PD risk loci with AD hallmark pathologies.
https://doi.org/10.1371/journal.pgen.1012170.s027
(XLSX)
S8 Table. Association of Nalls et al PD risk loci with related dementia pathologies.
https://doi.org/10.1371/journal.pgen.1012170.s028
(XLSX)
S9 Table. Estimated SNP heritability across all AD/RD pathologies.
Analyses used Model 2 throughout, excluding APOE region; liability scale; intercept constrained to 1. Shaded p-values indicate significance (p < 0.05). Abbreviations: AD, Alzheimer’s disease; ADNC, AD neuropathologic change; CAA, cerebral amyloid angiopathy; CI, confidence interval; CERAD, Consortium to Establish a Registry for AD neuritic plaque score; CBVD, cerebrovascular disease; HS, hippocampal sclerosis; LBD, Lewy body disease; SE, standard error; TDP-43, TAR DNA-binding protein 43; VBI, vascular brain injury; WMR, white matter rarefaction.
https://doi.org/10.1371/journal.pgen.1012170.s029
(XLSX)
S10 Table. Estimated SNP-based genetic correlation (shared heritability) across between neuropathological phenotypes.
Genetic correlations only calculated when individual SNP heritabilities were significantly greater than zero (i.e., p < 0.05 in S8 Table). Some comparisons not included due to poor model convergence.
https://doi.org/10.1371/journal.pgen.1012170.s030
(XLSX)
S11 Table. Cell/Tissue-specific heritability enrichment p-values (GTEx, multi tissue).
See Supplemental Text for methodology.
https://doi.org/10.1371/journal.pgen.1012170.s031
(XLSX)
S12 Table. Cell/Tissue-specific heritability enrichment coefficients and p-values (GTEx, multi tissue).
This table contains the same pvalues as S10 Table, but with coefficients and in “long” format. See Supplemental Text for methodology.
https://doi.org/10.1371/journal.pgen.1012170.s032
(XLSX)
S13 Table. Cell/Tissue-specific heritability enrichment p-values (ImmGen).
See Supplemental Text for methodology.
https://doi.org/10.1371/journal.pgen.1012170.s033
(XLSX)
S14 Table. Cell/Tissue-specific heritability enrichment coefficients and p-values (ImmGen).
This table contains the same pvalues as S12 Table, but with coefficients and in “long” format. See Supplemental Text for methodology.
https://doi.org/10.1371/journal.pgen.1012170.s034
(XLSX)
S16 Table. MAGMA tissue-specific associations.
https://doi.org/10.1371/journal.pgen.1012170.s036
(XLSX)
S17 Table. MAGMA tissue-specific associations, general tissues.
https://doi.org/10.1371/journal.pgen.1012170.s037
(XLSX)
S18 Table. Heterogeneity statistics across study.
*het p-value determined in METAL, using the Q statistic. “Prop” indicates proportion of total tests with Het(Q) p-value less than 0.05.
https://doi.org/10.1371/journal.pgen.1012170.s038
(XLSX)
S19 Table. Genotype Imputation Quality (R2) by batch, for genome-wide associated variants.
https://doi.org/10.1371/journal.pgen.1012170.s039
(XLSX)
S21 Table. A priori and a posteriori power calculations for clinical-AD risk variants.
Abbreviations: chr, chromosome; pos, position; min, minor allele; maj, major allele; MAF, minor allele frequency; OR, odd’s ratio. Yellow highlights denote moderate power (0.5-0.8) and green highlights denote strong power (0.8-1). A priori power denotes the power of observing an association with the designated variant, if the current study has the same additive effect size as those reported in the Bellenguez et al GWAS. Power (A) denotes the power if that effect size is observed with ‘A Score’ and its sample size and ordinal trait distribution, under the proportional odds assumption. Similarly, Power (B) is with B Score (e.g., AD Braak), and Power (C) is with CERAD. The a posteriori power section denotes power to observe association when the true effect sizes are those observed in our study. “A Score” uses the effect sizes we estimated for A Score, “B Score” with B score, etc. Power (Bonf.) denotes power when using a Bonferroni-correct alpha threshold (p < 0.05/78) and Power (Nom.) denotes power when using a nominal alpha threshold (p < 0.05). All calculations were performed using simulations. Briefly, a population of 10,000,000 was created for each SNP, with the appropriate effect size (under proportional odds), ordinal trait distribution, and minor allele frequency (assuming Hardy-Weinberg Equilibrium). To calculate power, 1000 samples were drawn from this population, with sample size equivalent to the “N” for the A Score, B Score, and C Score phenotypes. Each iteration was then tested for association between variant and ordinal trait using ordinal logistic regression under the proportinal odds assumption. Power is then the portion of iterations with effect-size p-value less than the required alpha threshold (Bonferroni or nominal).
https://doi.org/10.1371/journal.pgen.1012170.s041
(XLSX)
References
- 1. Brenowitz WD, Hubbard RA, Keene CD, Hawes SE, Longstreth WTJ, Woltjer RL. Mixed neuropathologies and estimated rates of clinical progression in a large autopsy sample. Alzheimers Dement. 2017;13(6):654–62.
- 2. Kapasi A, DeCarli C, Schneider JA. Impact of multiple pathologies on the threshold for clinically overt dementia. Acta Neuropathol. 2017;134(2):171–86.
- 3. Kawas CH, Kim RC, Sonnen JA, Bullain SS, Trieu T, Corrada MM. Multiple pathologies are common and related to dementia in the oldest-old: The 90+ Study. Neurology. 2015;85(6):535–42. pmid:26180144
- 4. Montine TJ, Sonnen JA, Montine KS, Crane PK, Larson EB. Adult Changes in Thought study: dementia is an individually varying convergent syndrome with prevalent clinically silent diseases that may be modified by some commonly used therapeutics. Curr Alzheimer Res. 2012;9(6):718–23. pmid:22471861
- 5. Rahimi J, Kovacs GG. Prevalence of mixed pathologies in the aging brain. Alzheimers Res Ther. 2014;6(9):82. pmid:25419243
- 6. Schneider JA, Aggarwal NT, Barnes L, Boyle P, Bennett DA. The neuropathology of older persons with and without dementia from community versus clinic cohorts. J Alzheimers Dis. 2009;18(3):691–701. pmid:19749406
- 7. White LR, Edland SD, Hemmy LS, Montine KS, Zarow C, Sonnen JA, et al. Neuropathologic comorbidity and cognitive impairment in the Nun and Honolulu-Asia Aging Studies. Neurology. 2016;86(11):1000–8. pmid:26888993
- 8. Bennett DA, Buchman AS, Boyle PA. Religious Orders Study and Rush Memory and Aging Project. J Alzheimers Dis. 2018;64(s1):S161–89.
- 9. Escott-Price V, Hardy J. Genome-wide association studies for Alzheimer’s disease: bigger is not always better. Brain Commun. 2022;4(3):fcac125. pmid:35663382
- 10. Jansen IE, Savage JE, Watanabe K, Bryois J, Williams DM, Steinberg S, et al. Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk. Nat Genet. 2019;51(3):404–13. pmid:30617256
- 11. Kunkle BW, Grenier-Boley B, Sims R, Bis JC, Damotte V, Naj AC, et al. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Aβ, tau, immunity and lipid processing. Nat Genet. 2019;51(3):414–30. pmid:30820047
- 12. Wightman DP, Jansen IE, Savage JE, Shadrin AA, Bahrami S, Holland D, et al. A genome-wide association study with 1,126,563 individuals identifies new risk loci for Alzheimer’s disease. Nat Genet. 2021;53(9):1276–82. pmid:34493870
- 13. Willett JDS, Waqas M, Choi Y, Ngai T, Mullin K, Tanzi RE, et al. Identification of 16 novel Alzheimer’s disease loci using multi-ancestry meta-analyses. Alzheimers Dement. 2025;21(2):e14592. pmid:39998322
- 14. Nelson PT, Alafuzoff I, Bigio EH, Bouras C, Braak H, Cairns NJ, et al. Correlation of Alzheimer disease neuropathologic changes with cognitive status: a review of the literature. J Neuropathol Exp Neurol. 2012;71(5):362–81. pmid:22487856
- 15. White LR, Corrada MM, Kawas CH, Cholerton BA, Edland SE, Flanagan ME, et al. Neuropathologic Changes of Alzheimer’s Disease and Related Dementias: Relevance to Future Prevention. J Alzheimers Dis. 2023;95(1):307–16. pmid:37522210
- 16. Xavier C, Pinto N. Navigating the blurred boundary: Neuropathologic changes versus clinical symptoms in Alzheimer’s disease, and its consequences for research in genetics. J Alzheimers Dis. 2025;104(3):611–26. pmid:39956949
- 17. Jansen IE, van der Lee SJ, Gomez-Fonseca D, de Rojas I, Dalmasso MC, Grenier-Boley B, et al. Genome-wide meta-analysis for Alzheimer’s disease cerebrospinal fluid biomarkers. Acta Neuropathol. 2022;144(5):821–42. pmid:36066633
- 18.
Cruchaga C, Timsina J, Jiang C. GWAS meta-analysis of CSF Alzheimer’s disease biomarkers 18,948 individuals reveal novel loci and genes regulating lipid metabolism, brain volume and autophagy. 2025.
- 19. Amanullah AA, Mirbod M, Pandey A, Singh SB, Gandhi OH, Ayubcha C. The Genetics of Amyloid Deposition: A Systematic Review of Genome-Wide Association Studies Using Amyloid PET Imaging in Alzheimer’s Disease. J Imaging. 2025;11(8):280. pmid:40863490
- 20. Raghavan NS, Dumitrescu L, Mormino E, Mahoney ER, Lee AJ, Gao Y, et al. Association Between Common Variants in RBFOX1, an RNA-Binding Protein, and Brain Amyloidosis in Early and Preclinical Alzheimer Disease. JAMA Neurol. 2020;77(10):1288–98. pmid:32568366
- 21. Beecham GW, Hamilton K, Naj AC, Martin ER, Huentelman M, Myers AJ, et al. Genome-wide association meta-analysis of neuropathologic features of Alzheimer’s disease and related dementias. PLoS Genet. 2014;10(9):e1004606. pmid:25188341
- 22. Chibnik LB, White CC, Mukherjee S, Raj T, Yu L, Larson EB, et al. Susceptibility to neurofibrillary tangles: role of the PTPRD locus and limited pleiotropy with other neuropathologies. Mol Psychiatry. 2018;23(6):1521–9. pmid:28322283
- 23. Guerreiro R, Ross OA, Kun-Rodrigues C, Hernandez DG, Orme T, Eicher JD, et al. Investigating the genetic architecture of dementia with Lewy bodies: a two-stage genome-wide association study. Lancet Neurol. 2018;17(1):64–74. pmid:29263008
- 24. Nelson PT, Estus S, Abner EL, Parikh I, Malik M, Neltner JH, et al. ABCC9 gene polymorphism is associated with hippocampal sclerosis of aging pathology. Acta Neuropathol. 2014;127(6):825–43. pmid:24770881
- 25. Shade LMP, Katsumata Y, Abner EL, Aung KZ, Claas SA, Qiao Q, et al. GWAS of multiple neuropathology endophenotypes identifies new risk loci and provides insights into the genetic risk of dementia. Nat Genet. 2024;56(11):2407–21. pmid:39379761
- 26. Godrich D, Martin ER, Schellenberg G, Pericak-Vance MA, Cuccaro M, Scott WK. Neuropathological lesions and their contribution to dementia and cognitive impairment in a heterogeneous clinical population. Alzheimers Dement. 2022;18(12):2403–12.
- 27. Hyman BT, Phelps CH, Beach TG, Bigio EH, Cairns NJ, Carrillo MC, et al. National Institute on Aging-Alzheimer’s Association guidelines for the neuropathologic assessment of Alzheimer’s disease. Alzheimers Dement. 2012;8(1):1–13. pmid:22265587
- 28. Thal DR, Rüb U, Orantes M, Braak H. Phases of A beta-deposition in the human brain and its relevance for the development of AD. Neurology. 2002;58(12):1791–800. pmid:12084879
- 29. Godrich D, Pasteris J, Martin ER, Kunkle B, Naj AC, Hamilton K, et al. Genome-wide association studies of TDP-43 proteinopathy and hippocampal sclerosis reveal shared genetic associations with APOE and TMEM106B. Alzheimers Dement. 2025;21(11):e70760. pmid:41208712
- 30. Bellenguez C, Küçükali F, Jansen IE, Kleineidam L, Moreno-Grau S, Amin N, et al. New insights into the genetic etiology of Alzheimer’s disease and related dementias. Nat Genet. 2022;54(4):412–36. pmid:35379992
- 31. Nalls MA, Blauwendraat C, Vallerga CL, Heilbron K, Bandres-Ciga S, Chang D, et al. Identification of novel risk loci, causal insights, and heritable risk for Parkinson’s disease: a meta-analysis of genome-wide association studies. Lancet Neurol. 2019;18(12):1091–102. pmid:31701892
- 32. Finucane HK, Reshef YA, Anttila V, Slowikowski K, Gusev A, Byrnes A, et al. Heritability enrichment of specifically expressed genes identifies disease-relevant tissues and cell types. Nat Genet. 2018;50(4):621–9. pmid:29632380
- 33. Heng TSP, Painter MW, Immunological Genome Project Consortium. The Immunological Genome Project: networks of gene expression in immune cells. Nat Immunol. 2008;9(10):1091–4. pmid:18800157
- 34. Jin Y, Li F, Sonoustoun B, Kondru NC, Martens YA, Qiao W, et al. APOE4 exacerbates α-synuclein seeding activity and contributes to neurotoxicity in Alzheimer’s disease with Lewy body pathology. Acta Neuropathol. 2022;143(6):641–62. pmid:35471463
- 35. Tsuang D, Leverenz JB, Lopez OL, Hamilton RL, Bennett DA, Schneider JA, et al. APOE ε4 increases risk for dementia in pure synucleinopathies. JAMA Neurol. 2013;70(2):223–8. pmid:23407718
- 36. Yang H-S, Yu L, White CC, Chibnik LB, Chhatwal JP, Sperling RA, et al. Evaluation of TDP-43 proteinopathy and hippocampal sclerosis in relation to APOE ε4 haplotype status: a community-based cohort study. Lancet Neurol. 2018;17(9):773–81. pmid:30093249
- 37. Ferencz B, Gerritsen L. Genetics and underlying pathology of dementia. Neuropsychol Rev. 2015;25(1):113–24. pmid:25567624
- 38. Spires-Jones TL, Attems J, Thal DR. Interactions of pathological proteins in neurodegenerative diseases. Acta Neuropathol. 2017;134(2):187–205. pmid:28401333
- 39. Tomé SO, Tsaka G, Ronisz A, Ospitalieri S, Gawor K, Gomes LA, et al. TDP-43 pathology is associated with increased tau burdens and seeding. Mol Neurodegener. 2023;18(1):71. pmid:37777806
- 40. Rasmussen KL, Luo J, Nordestgaard BG, Tybjærg-Hansen A, Frikke-Schmidt R. APOE and vascular disease: Sequencing and genotyping in general population cohorts. Atherosclerosis. 2023;385:117218. pmid:37586954
- 41. Katsumata Y, Shade LM, Hohman TJ, Schneider JA, Bennett DA, Farfel JM, et al. Multiple gene variants linked to Alzheimer’s-type clinical dementia via GWAS are also associated with non-Alzheimer’s neuropathologic entities. Neurobiol Dis. 2022;174:105880. pmid:36191742
- 42. Ahola-Olli AV, Würtz P, Havulinna AS, Aalto K, Pitkänen N, Lehtimäki T, et al. Genome-wide Association Study Identifies 27 Loci Influencing Concentrations of Circulating Cytokines and Growth Factors. Am J Hum Genet. 2017;100(1):40–50. pmid:27989323
- 43. Borges MC, Haycock PC, Zheng J, Hemani G, Holmes MV, Davey Smith G, et al. Role of circulating polyunsaturated fatty acids on cardiovascular diseases risk: analysis using Mendelian randomization and fatty acid genetic association data from over 114,000 UK Biobank participants. BMC Med. 2022;20(1):210. pmid:35692035
- 44. Chen VL, Du X, Chen Y, Kuppa A, Handelman SK, Vohnoutka RB, et al. Genome-wide association study of serum liver enzymes implicates diverse metabolic and liver pathology. Nat Commun. 2021;12(1):816. pmid:33547301
- 45. Graham SE, Clarke SL, Wu K-HH, Kanoni S, Zajac GJM, Ramdas S, et al. The power of genetic diversity in genome-wide association studies of lipids. Nature. 2021;600(7890):675–9. pmid:34887591
- 46. Richardson TG, Leyden GM, Wang Q, Bell JA, Elsworth B, Davey Smith G, et al. Characterising metabolomic signatures of lipid-modifying therapies through drug target mendelian randomisation. PLoS Biol. 2022;20(2):e3001547. pmid:35213538
- 47. Richardson TG, Sanderson E, Palmer TM, Ala-Korpela M, Ference BA, Davey Smith G, et al. Evaluating the relationship between circulating lipoprotein lipids and apolipoproteins with risk of coronary heart disease: A multivariable Mendelian randomisation analysis. PLoS Med. 2020;17(3):e1003062. pmid:32203549
- 48. Schoeler T, Speed D, Porcu E, Pirastu N, Pingault J-B, Kutalik Z. Participation bias in the UK Biobank distorts genetic associations and downstream analyses. Nat Hum Behav. 2023;7(7):1216–27. pmid:37106081
- 49. Davies G, Lam M, Harris SE, Trampush JW, Luciano M, Hill WD, et al. Study of 300,486 individuals identifies 148 independent genetic loci influencing general cognitive function. Nat Commun. 2018;9(1):2098. pmid:29844566
- 50. Lee JJ, Wedow R, Okbay A, Kong E, Maghzian O, Zacher M, et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nat Genet. 2018;50(8):1112–21. pmid:30038396
- 51. Hollingworth P, Harold D, Sims R. Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, CD33 and CD2AP are associated with Alzheimer’s disease. Nature Genetics. 2011;43(5):429–35.
- 52. Naj AC, Jun G, Beecham GW, Wang L-S, Vardarajan BN, Buros J, et al. Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are associated with late-onset Alzheimer’s disease. Nat Genet. 2011;43(5):436–41. pmid:21460841
- 53. Martínez A, Soriano E. Functions of ephrin/Eph interactions in the development of the nervous system: emphasis on the hippocampal system. Brain Res Brain Res Rev. 2005;49(2):211–26. pmid:16111551
- 54. Lai K-O, Ip NY. Synapse development and plasticity: roles of ephrin/Eph receptor signaling. Curr Opin Neurobiol. 2009;19(3):275–83. pmid:19497733
- 55. Cerezo M, Sollis E, Ji Y, Lewis E, Abid A, Bircan KO, et al. The NHGRI-EBI GWAS Catalog: standards for reusability, sustainability and diversity. Nucleic Acids Res. 2025;53(D1):D998–1005. pmid:39530240
- 56. Li Y-J, Nuytemans K, La JO, Jiang R, Slifer SH, Sun S, et al. Identification of novel genes for age-at-onset of Alzheimer’s disease by combining quantitative and survival trait analyses. Alzheimers Dement. 2023;19(7):3148–57. pmid:36738287
- 57. Huang L, Chambliss KL, Gao X, Yuhanna IS, Behling-Kelly E, Bergaya S, et al. SR-B1 drives endothelial cell LDL transcytosis via DOCK4 to promote atherosclerosis. Nature. 2019;569(7757):565–9. pmid:31019307
- 58. Bone WP, Siewert KM, Jha A, Klarin D, Damrauer SM, VA Million Veteran Program, et al. Multi-trait association studies discover pleiotropic loci between Alzheimer’s disease and cardiometabolic traits. Alzheimers Res Ther. 2021;13(1):34. pmid:33541420
- 59. Xie X, Lu Y, Wang B, Yin X, Chen J. DOCK4 is a Novel Prognostic Biomarker and Correlated with Immune Infiltrates in Colon Adenocarcinoma. Comb Chem High Throughput Screen. 2024;27(8):1119–30. pmid:37702239
- 60. Huang J, Huffman JE, Huang Y, Do Valle Í, Assimes TL, Raghavan S, et al. Genomics and phenomics of body mass index reveals a complex disease network. Nat Commun. 2022;13(1):7973. pmid:36581621
- 61. Kong M, Lee C. Genetic associations with C-reactive protein level and white blood cell count in the KARE study. Int J Immunogenet. 2013;40(2):120–5. pmid:22788528
- 62. Surendran P, Feofanova EV, Lahrouchi N, Ntalla I, Karthikeyan S, Cook J, et al. Discovery of rare variants associated with blood pressure regulation through meta-analysis of 1.3 million individuals. Nat Genet. 2020;52(12):1314–32. pmid:33230300
- 63. Foster EM, Dangla-Valls A, Lovestone S, Ribe EM, Buckley NJ. Clusterin in Alzheimer’s Disease: Mechanisms, Genetics, and Lessons From Other Pathologies. Front Neurosci. 2019;13:164. pmid:30872998
- 64. Chapuis J, Flaig A, Grenier-Boley B, Eysert F, Pottiez V, Deloison G, et al. Genome-wide, high-content siRNA screening identifies the Alzheimer’s genetic risk factor FERMT2 as a major modulator of APP metabolism. Acta Neuropathol. 2017;133(6):955–66. pmid:27933404
- 65. Wang H, Lo M-T, Rosenthal SB, Makowski C, Andreassen OA, Salem RM, et al. Similar Genetic Architecture of Alzheimer’s Disease and Differential APOE Effect Between Sexes. Front Aging Neurosci. 2021;13:674318. pmid:34122051
- 66. Zhang Q, Sidorenko J, Couvy-Duchesne B, Marioni RE, Wright MJ, Goate AM, et al. Risk prediction of late-onset Alzheimer’s disease implies an oligogenic architecture. Nat Commun. 2020;11(1):4799. pmid:32968074
- 67. Manichaikul A, Mychaleckyj JC, Rich SS, Daly K, Sale M, Chen W-M. Robust relationship inference in genome-wide association studies. Bioinformatics. 2010;26(22):2867–73. pmid:20926424
- 68. Zheng X, Levine D, Shen J, Gogarten SM, Laurie C, Weir BS. A high-performance computing toolset for relatedness and principal component analysis of SNP data. Bioinformatics. 2012;28(24):3326–8. pmid:23060615
- 69. Taliun D, Harris DN, Kessler MD, al et. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature. 2021;590(7845):290–9.
- 70. Conomos MP, Miller MB, Thornton TA. Robust inference of population structure for ancestry prediction and correction of stratification in the presence of relatedness. Genet Epidemiol. 2015;39(4):276–93. pmid:25810074
- 71. Montine TJ, Phelps CH, Beach TG, Bigio EH, Cairns NJ, Dickson DW, et al. National Institute on Aging-Alzheimer’s Association guidelines for the neuropathologic assessment of Alzheimer’s disease: a practical approach. Acta Neuropathol. 2012;123(1):1–11. pmid:22101365
- 72. Vonsattel JP, Myers RH, Hedley-Whyte ET, Ropper AH, Bird ED, Richardson EP Jr. Cerebral amyloid angiopathy without and with cerebral hemorrhages: a comparative histological study. Ann Neurol. 1991;30(5):637–49. pmid:1763890
- 73.
R Core Team. R: A Language and Environment for Statistical Computing. 2025. https://www.R-project.org
- 74. Friendly M. Corrgrams. The American Statistician. 2002;56(4):316–24.
- 75. Zhan X, Hu Y, Li B, Abecasis GR, Liu DJ. RVTESTS: an efficient and comprehensive tool for rare variant association analysis using sequence data. Bioinformatics. 2016;32(9):1423–6. pmid:27153000
- 76. Christensen RH. Cumulative link models for ordinal regression with the R package ordinal. 2023.
- 77. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics. 2010;26(17):2190–1. pmid:20616382
- 78. D. Turner S. qqman: an R package for visualizing GWAS results using Q-Q and manhattan plots. JOSS. 2018;3(25):731.
- 79.
Wickham H. Ggplot2: Elegant Graphics for Data. Springer. 2009.
- 80. Pruim RJ, Welch RP, Sanna S, Teslovich TM, Chines PS, Gliedt TP, et al. LocusZoom: regional visualization of genome-wide association scan results. Bioinformatics. 2010;26(18):2336–7. pmid:20634204
- 81. Watanabe K, Taskesen E, van Bochoven A, Posthuma D. Functional mapping and annotation of genetic associations with FUMA. Nat Commun. 2017;8(1):1826. pmid:29184056
- 82. Kircher M, Witten DM, Jain P, O’Roak BJ, Cooper GM, Shendure J. A general framework for estimating the relative pathogenicity of human genetic variants. Nat Genet. 2014;46(3):310–5. pmid:24487276
- 83. Boyle AP, Hong EL, Hariharan M, Cheng Y, Schaub MA, Kasowski M, et al. Annotation of functional variation in personal genomes using RegulomeDB. Genome Res. 2012;22(9):1790–7. pmid:22955989
- 84. Bulik-Sullivan BK, Loh P-R, Finucane HK, Ripke S, Yang J, Schizophrenia Working Group of the Psychiatric Genomics Consortium, et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat Genet. 2015;47(3):291–5. pmid:25642630