Sex differences in the genetic predictors of Alzheimer’s pathology

Despite known sex differences in Alzheimer’s disease risk, sex is rarely systematically evaluated in genetic studies. Hohman et al. examine sex-specific associations with Alzheimer’s disease neuropathology across the genome and identify one male-specific association with neurofibrillary tangles. This locus on chromosome 7 appears to confer male-specific protection against tau pathology.

Abstract

Autopsy measures of Alzheimer’s disease neuropathology have been leveraged as endophenotypes in previous genome-wide association studies (GWAS). However, despite evidence of sex differences in Alzheimer’s disease risk, sex-stratified models have not been incorporated into previous GWAS analyses. We looked for sex-specific genetic associations with Alzheimer’s disease endophenotypes from six brain bank data repositories. The pooled dataset included 2701 males and 3275 females, the majority of whom were diagnosed with Alzheimer’s disease at autopsy (70%). Sex-stratified GWAS were performed within each dataset and then meta-analysed. Loci that reached genome-wide significance (P < 5 × 10⁻⁸) in stratified models were further assessed for sex interactions. Additional analyses were performed in independent datasets leveraging cognitive, neuroimaging and CSF endophenotypes, along with age-at-onset data. Outside of the APOE region, one locus on chromosome 7 (rs34331204) showed a sex-specific association with neurofibrillary tangles among males (P = 2.5 × 10⁻⁸) but not females (P = 0.85, sex-interaction P = 2.9 × 10⁻⁴). In follow-up analyses, rs34331204 was also associated with hippocampal volume, executive function, and age-at-onset only among males. These results implicate a novel locus that confers male-specific protection from tau pathology and highlight the value of assessing genetic associations in a sex-specific manner.

Introduction

Two-thirds of Alzheimer’s disease cases are female (Mielke et al., 2014; Mazure and Swendsen, 2016) and emerging evidence has highlighted notable sex differences in Alzheimer’s disease risk (Altmann et al., 2014; Neu et al., 2017; Buckley et al., 2018), presentation (Barnes et al., 2005; Apostolova et al., 2006; Hua et al., 2010; Hohman et al., 2018), and progression (Barnes et al., 2005; Koran et al., 2017). Notably, the apolipoprotein E (APOE) gene, which is the strongest genetic risk factor for Alzheimer’s disease shows a stronger association among females compared to males, particularly between ages 65 and 75 years (Neu et al., 2017). Despite growing evidence of sex differences in the genetic drivers of Alzheimer’s disease (Deming et al., 2018), limited work has systematically explored sex-specific genetic associations with Alzheimer’s disease neuropathology across the genome.

Autopsy measures of neuropathology, including the Consortium to Establish a Registry for Alzheimer’s disease (CERAD) neuritic plaque staging and Braak neurofibrillary tangle staging, have been leveraged in previous genome-wide association studies (GWAS) to identify novel genetic loci for Alzheimer’s disease (Beecham et al., 2014). These endophenotypes provide an invaluable opportunity to better understand the underlying disease process by providing biological measures that are more proximal to gene function. Moreover, these metrics provide ideal outcomes for sex-specific analyses because identified associations will highlight points along the disease cascade where sex differences emerge.

This study leverages six autopsy cohorts to assess sex-specific genetic associations with Alzheimer’s disease neuropathology. First, we perform a sex-stratified GWAS in 5976 participants with autopsy measures of plaques and tangles. Second, we validate observed sex-specific associations leveraging complementary biomarker data from independent datasets. Our central hypothesis is that certain genetic factors act in a sex-specific manner to drive the neuropathological presentation of Alzheimer’s disease. The identification of sex-specific effects will advance our understanding of the genetic architecture of Alzheimer’s disease.

Materials and methods

Participants

Data were drawn from a previous GWAS (Beecham et al., 2014), which provided detailed descriptions of the following six well-characterized autopsy cohorts: the National Institute on Aging Late-Onset Alzheimer’s Disease Family Study (LOAD), Mayo Clinic (Mayo), the Adult Changes in Thought (ACT) study, the National Alzheimer’s Coordinating Center (NACC), the Religious Orders Study and Rush Memory and Aging Project (ROS/MAP), and the Translational Genomics Research Institute (TGEN). All participants agreed to brain donation and were evaluated at each site. Alzheimer’s disease diagnoses were made through consensus criteria excluding participants with a documented history of stroke, substantial cerebrovascular disease, if they met criteria for another dementia/aphasia, or with an active neurological disease or medication/medical co-morbidity that may impact cognition (McKhann et al., 1984, 2011). All neuropathological data were reviewed and harmonized by a single neuropathologist.

Quantification of neuropathology outcomes

Autopsy measures of neurofibrillary tangles (Braak staging) and neuritic plaques (CERAD score) were collected and harmonized previously (Beecham et al., 2014). Thal stage was not collected or included in our staging definitions. Both measures were analysed as binary outcomes. Binary neuritic plaque status was defined based on established neuropathological criteria for Alzheimer’s disease (Hyman et al., 2012) whereby a CERAD score of ‘none’ or ‘sparse’ was considered ‘neuritic plaque negative’, and ‘moderate’ or ‘frequent’ was considered ‘neuritic plaque positive’. Similarly, the binary neurofibrillary tangles (NFT) status was defined whereby Braak stages 0/I/II were considered ‘NFT negative’ and stages III/IV/V/VI were considered ‘NFT positive’ (Hyman et al., 2012).

Genotyping and quality control

Genome-wide genotyping was carried out by each study on a variety of platforms. All participants were of European descent, and all genotype data were processed using the same imputation and standard quality control protocols (Supplementary material).

Statistical analyses

Descriptive statistics were quantified in R v3.3.1 (https://www.r-project.org/). Sex-stratified analyses of NFT and neuritic plaques were performed using logistic regression in PLINK (https://www.cog-genomics.org/plink/1.9). Analyses were carried out within each cohort, used additive coding, and co-varied for age at death. Fixed-effects meta-analysis was performed using GWAMA (http://www.geenivaramu.ee/en/tools/gwama) (Mägi and Morris, 2010). Meta-analysis results were limited to SNPs (single nucleotide polymorphisms) that were genotyped or imputed, passed quality control, and were polymorphic in at least three of the six cohorts. Statistical significance was set at the standard GWAS level (α = 5 × 10⁻⁸). All significant sex-stratified effects were assessed for sex × SNP interactions. Miami plots were generated using EasyStrata v16.0 (Winkler et al., 2014). Genomic inflation factors for the GWAS analyses ranged from λ = 0.99–1.04 (Supplementary Fig. 1). SNP annotation was performed using ANNOVAR (v2018Apr16). Forest plots were generated using the R package Metafor.

Six haplotype tagging SNPs were used to test for associations between the MAPT locus and NFT positivity. Tags included the H1 haplotype (rs8070723) and H1 sub-haplotypes (rs1467967, rs242557, rs3785883, rs2471738, and rs7521) (Pittman et al., 2005; Höglinger et al., 2011).

Analysis of Alzheimer’s disease endophenotypes

SNPs with sex-specific associations were further assessed for correlation with relevant Alzheimer’s disease endophenotypes using linear regression in R, co-varying for baseline age. Hippocampal volumes were normalized by intracranial volume using established procedures (Voevodskaya et al., 2014). Finally, putative SNPs were also evaluated in sex-specific associations with age-at-onset using data from a previously published survival analysis of Alzheimer’s disease (Huang et al., 2017). See Supplementary material for additional details.

Expression quantitative trait analysis

SNPs that showed a sex-specific association were further assessed for expression quantitative trait locus (eQTL) associations using data from Braineac (http://caprica.genetics.kcl.ac.uk/BRAINEAC/). Correction for multiple comparisons was completed using the false discovery rate (FDR) procedure.

Significant eQTL genes were further assessed for sex-specific associations with Alzheimer’s disease neuropathology leveraging prefrontal cortex gene expression data from ROS/MAP participants at autopsy (Supplementary material).

Results

A total of 2701 males and 3275 females across six independent autopsy datasets were analysed. In general, females were older than males (males: 79 ± 9 years, females: 81 ± 9 years, P < 0.001) and were more frequently Alzheimer’s disease cases (males: 68%, females: 71%, P = 0.02), APOE ɛ4 carriers (males: 49%, females: 46%, P = 0.03), and neuritic plaque- and NFT-positive individuals (males: 72% and 77%, females: 76% and 85%, P-values < 0.01) than males. Participant characteristics by cohort are presented in Supplementary Table 1.

In the sex-stratified GWAS analysis of NFT, one intergenic SNP on chr7p21.1 (rs34331204; Fig. 1 and Supplementary Fig. 2) outside of the APOE locus reached genome-wide significance in males (β = −0.720; P = 2.48 × 10⁻⁸) but not in females (β = −0.027; P = 0.85). Furthermore, rs34331204 showed an interaction with sex on NFT (β = 0.71; P = 2.93 × 10⁻⁴), whereby the minor allele C (MAF = 0.07) was associated with a lower risk of NFT positivity in males (Fig. 2). It is notable that this association did not meet genome-wide significance in the previously published GWAS that did not incorporate sex-stratified models (β = −0.39; P = 2.63 × 10⁻⁵). Additional models using different Braak staging cut-off points are presented in Supplementary Figs 3–6.

Figure 1.

Sex-stratified genome-wide association results for tangle positivity. (A) Miami plot illustrating neurofibrillary tangle positivity genome-wide association results stratified by males and females. Male findings are plotted in blue and grey on the top and female results are plotted in pink and grey at the bottom. The red lines represent the genome-wide threshold for statistical significance (P < 5 × 10⁻⁸). Regional association plots for the rs34331204 association with neurofibrillary tangle positivity within (B) males and (C) females.

Figure 2.

Male-specific SNP (rs34331204) associated with protection from neurofibrillary tangles also relates to hippocampal volume and executive function. Sex-specific association of rs34331204 with (A) NFT, (B) hippocampal volume, and (C) executive function. Within each panel, males are presented on the left and females on the right. Outcomes are presented on the y-axes. Bar colours represent rs34331204 genotype. Homozygous carriers of the A allele are presented in dark green on the left, heterozygotes in light green in the middle, and homozygous carriers of the C allele in the lightest green on the right. A neuroprotective effect of the rs34331204 C allele is observed among males, but not females.

No associations reached genome-wide significance in the sex-stratified GWAS analysis of neuritic plaque (Supplementary Fig. 7). The top meta-analysis results from sex-stratified GWAS of neuritic plaque and NFT are presented in Supplementary Tables 2–5.

As expected, a strong signal was seen at the APOE locus in all four sex-stratified GWAS (P-values < 4.96 × 10⁻²⁰) (Fig. 1 and Supplementary Fig. 7). As the sex-specific effect of APOE on Alzheimer’s disease biomarkers has been previously reported in detail (Hohman et al., 2018), our focus was on associations outside of the APOE locus. However, as it is possible that some significant signal(s) in the APOE locus could be independent of APOE haplotype, we co-varied for both APOE ɛ4 and ɛ2 allele carrier status. All signals in the APOE locus were strongly attenuated after co-varying for APOE haplotype (P-values > 1.46 × 10⁻³) (Supplementary Figs 8 and 9).

Another candidate locus we evaluated was MAPT, which encodes the tau protein. Of the six MAPT locus haplotype tagging SNPs tested, rs242557 was nominally associated with NFT positivity in males (P = 0.0043) but not in females (P = 0.30; sex-interaction P = 0.015) (Supplementary Table 6).

The putative sex-specific GWAS locus (rs34331204) was then assessed for associations with cognition, amyloidosis, neurodegeneration, and age-at-onset using publicly available data sources. Results are presented in Table 1. SNP rs34331204 showed a comparable male-specific association with executive function performance and hippocampal volume (Fig. 2), with mixed evidence of a sex difference in the age-at-onset analysis. No sex-specific associations with CSF tau or p-tau were observed.

Table 1.

Associations between rs34331204 and relevant Alzheimer’s disease endophenotypes

Outcome	n	Males		Females		Sex interaction
		Beta	P	Beta	P	Beta	P
CSF tau	2926	0.006	0.69	−0.009	0.58	−0.013	0.55
CSF p-tau	2759	−0.002	0.89	−0.011	0.47	−0.010	0.63
Episodic memory	1182	0.104	0.14	0.038	0.73	−0.063	0.62
Executive function	1182	0.266	0.001	−0.016	0.88	−0.283	0.039
Hippocampal volume	1086	252.17	0.014	−33.18	0.80	−284.70	0.09
Age of onset in ADGC	17 603	−0.091	0.052	0.043	0.27	0.136	0.022
Age of onset in CERAD	3552	0.076	0.47	0.077	0.35	0.015	0.91

Bold signifies P < 0.05.

Lastly, we used eQTL mapping in Braineac to identify candidate genes within the rs34331204 locus and analysed expression of these genes in brain tissue. Significant eQTL associations (FDR-corrected aveALL P-value < 0.05 in Braineac) were seen for eight genes (BZW2, TSPAN13, AGR3, ANKMY2, LRRC72, AGR2, ISPD, and AHR; Supplementary Table 7). Two of these genes were not highly expressed in ROS/MAP PFC (AGR3 and LRRC72), so we assessed six genes for sex-specific associations with tau load (Table 2). Surprisingly, BZW2 and ANKMY2 showed evidence of female-specific associations with tau load (P-values < 0.002), but no male-specific associations or sex-interactions were observed.

Table 2.

Associations between tau load and rs34331204 cis gene expression in brain tissue

Gene	Males		Females		Sex interaction
	Beta	P	Beta	P	Beta	P
AGR2	−2.069	0.59	−0.824	0.60	1.263	0.76
AHR	−0.233	0.42	0.189	0.29	0.420	0.21
ANKMY2	−0.037	0.22	−0.078	9.93×10−4	−0.041	0.28
BZW2	−0.027	0.48	−0.093	2.02×10−3	−0.064	0.18
ISPD	−0.395	0.18	−0.537	0.018	−0.145	0.70
TSPAN13	−0.001	0.94	−0.010	0.049	−0.010	0.30

Gene expression data were collected from prefrontal cortex tissue of participants from the Religious Orders Study/Memory and Aging Project (males: n = 213, females: n = 380).

Discussion

The present study evaluated sex-specific genetic associations with Alzheimer’s disease neuropathology measured at autopsy. Results implicate one novel genetic locus, rs34331204 on chromosome 7 proximal to BZW2, that is associated with neurofibrillary tangles only among males. Additional evidence of a male-specific neuroprotective effect was observed in follow-up analyses in which the minor allele of rs34331204 was also associated with larger hippocampal volume, better executive function, and a later age-at-onset among males. It is important to note that the association between rs34331204 and NFT fell below the threshold of genome-wide significance when males and females were combined, and sex was simply included as a covariate in a post hoc analysis, highlighting the utility of sex-stratified analyses in uncovering novel potential disease loci.

There are a number of potential candidate genes within the associated locus, and rs34331204 was a strong eQTL for eight of them, complicating the picture. Among the implicated genes, ANKMY2 and BZW2 showed some weak evidence of association with tangle burden. It should be noted that the gene expression effects of these two genes were observed among females rather than males, counter to the male-specific SNP effects. The female-specific gene expression association may suggest that there is a male-specific eQTL effect (which we could not test from available data as Braineac does not offer results stratified by sex), or that these two genes are not the functional genes driving the male-specific association. However, given the sex × gene expression interaction was not significant, and there are more females than males in the ROS/MAP expression sample, it is probably safest to assume the gene expression effect is not sex-specific while the SNP effect is male-specific.

Both implicated genes, ANKMY2 and BZW2, are interesting candidates. BZW2 is a basic leucine zipper protein with a known role in cell proliferation through the Akt/mTOR pathway, particularly in cancer (Cheng et al., 2017). Associations between dual leucine zipper proteins and neurodegenerative disease have been reported in the literature recently (Le Pichon et al., 2017). While no functional association between BZW2 and neurodegeneration has been reported to date, it is notable that a SNP within BZW2 (rs58370486) previously showed an association with cognitive decline in Alzheimer’s disease (P = 6 × 10⁻¹¹) (Sherva et al., 2014). The protein product of ANKMY2 has been shown to interact with FKBP38 in the mouse brain, regulating the Sonic hedgehog signalling pathway (Saita et al., 2014), but FKBP38 also acts as a BCL2 chaperone that has been implicated in an apoptotic pathway downstream of amyloidosis (Kudo et al., 2012). However, ANKMY2 has not been directly implicated in Alzheimer’s disease previously. The present results suggest future functional and fine-mapping work in the rs34331204 region should focus on potential sex-specific effects.

In addition to the GWAS associations, we also observed weak evidence of a male-specific association between an SNP (rs242557) that partially tags the H1c sub-haplotype and tangle burden. This sub-haplotype has been associated with increased risk of Alzheimer’s disease (Myers et al., 2005) and increased expression of tau (Myers et al., 2007) in previous work, and our findings suggest a possible sex difference may contribute to this effect.

It is important to note that our results provide evidence of a sex-specific effect of rs34331204 on tangle load at autopsy, but not on CSF tau or CSF p-tau (Table 1). Furthermore, our results do not appear to overlap with those of a recent large sex-specific GWAS of CSF Alzheimer’s disease biomarkers (Deming et al., 2018). Deming and colleagues identified three female-specific associations with CSF biomarker levels, but none of these loci showed evidence of sex-specific associations with the relevant neuropathology in this study (Supplementary Table 8). Interestingly, in a previous study, we observed a similar lack of consistency across autopsy and CSF datasets for sex differences in the APOE association, whereby sex-specific effects of APOE ɛ4 on CSF tau, but not on autopsy measures of neurofibrillary tangles, were seen (Hohman et al., 2018). It is also notable that the loci identified in GWAS for CSF biomarkers and autopsy measures of neuropathology (across sexes) similarly did not show overlap with one another, highlighting that the discrepant results are characteristic of the endophenotypes rather than the sex-specific analytical pipelines. While the differing results between CSF tau and autopsy measures of tangles seem counterintuitive, there are multiple contributing factors. First, autopsy staging of tangles reflects a process that is distinct from CSF biomarkers of tau. Autopsy measures (and PET tau measures) are best characterized as ‘stage’ markers that signify how far the disease process has progressed, while CSF biomarkers are ‘state’ biomarkers that appear to measure the intensity of the disease process (Blennow and Hampel, 2003; Mattsson et al., 2017). In addition to the different biological processes that are tagged by autopsy and CSF metrics, there are also notable cohort differences between the CSF datasets and autopsy datasets that could contribute to the discrepancy. The autopsy datasets included here were older on average and include a higher proportion of individuals with clinical Alzheimer’s disease compared to the younger cohorts evaluated in the previous CSF studies. It is certainly possible that the genetic architecture of Alzheimer’s disease neuropathology is different at older ages than at younger ages, and different during the preclinical stages of disease compared to end stage disease. Ultimately, larger CSF and autopsy datasets will be needed to disentangle the complex contributors to discrepant signals in CSF and autopsy-based endophenotype analyses.

This study had multiple strengths including the large sample size gathered across multiple cohort studies, the comprehensive follow-up analyses in independent studies using complementary endophenotypes of amyloidosis, tau, neurodegeneration, and cognition, and the functional assessment of gene expression in prefrontal cortex tissue providing evidence of sex-specific associations at the gene level. However, there were also important limitations, including the noted age difference between males and females, the high percentage of Alzheimer’s disease cases, and the high percentage of APOE ɛ4 carriers within the leveraged datasets. These limitations leave open the strong possibility that additional sex-specific genetic loci for amyloid and tau pathology, particularly in the preclinical stages of disease, were probably undetected in our analyses. Further, the present analyses were restricted to individuals of European ancestry, leaving open the possibility that findings may not extend to other racial or ancestral backgrounds. Future work extending to datasets with more cognitively normal individuals and a more representative sample will be important to better understand sex-specific associations across the spectrum of normal ageing and dementia. Nevertheless, our results highlight a novel sex-specific candidate locus for Alzheimer’s disease and demonstrate the utility of incorporating sex considerations into genetic models of disease.

Funding

This research was supported in part by K01 AG049164, R21 AG05994, K12 HD043483, K24 AG046373, HHSN311201600276P, S10 OD023680, R01 AG034962, R01 HL111516, R01 NS100980, R01 AG056534, P30 AG10161, RF1 AG15819, R01 AG17917, R01 AG30146, R01 AG019085, R01 AG15819, R01 AG30146, R01 AG027161, R01 AG021155, R01 AG037639, U01 AG46152, U01 AG006781, U01 AG032984, U01 HG004610, U01 HG006375, U24 AG021886, U24 AG041689, R01 AG044546, P01 AG003991, RF1 AG053303, R01 AG035083, R01 NS085419, UL1 TR000445, KL2 TR000446, TL1 TR000447, S10 OD023680, and the Alzheimer’s Association (NIRG-11–200110), further supported in part by the Intramural Research Program, NIA, NIH and the Vanderbilt Memory & Alzheimer’s Center. Y.D. was supported by an NIMH training grant (T32MH014877). Support for P.D.J. was provided by R01 AG048015. H.Z. is a Wallenberg Academy Fellow and is further supported but the Swedish and European Research Councils and the UK Dementia Research Institute. KB holds the Torsten Söderberg professorship at the Royal Swedish Academy of Sciences.

The NACC database is funded by NIA/NIH Grant U01 AG016976. NACC data are contributed by the NIA-funded ADCs: P30 AG019610 (PI Eric Reiman, MD), P30 AG013846 (PI Neil Kowall, MD), P50 AG008702 (PI Scott Small, MD), P50 AG025688 (PI Allan Levey, MD, PhD), P30 AG010133 (PI Andrew Saykin, PsyD), P50 AG005146 (PI Marilyn Albert, PhD), P50 AG005134 (PI Bradley Hyman, MD, PhD), P50 AG016574 (PI Ronald Petersen, MD, PhD), P50 AG005138 (PI Mary Sano, PhD), P30 AG008051 (PI Steven Ferris, PhD), P30 AG013854 (PI M. Marsel Mesulam, MD), P30 AG008017 (PI Jeffrey Kaye, MD), P30 AG010161 (PI David Bennett, MD), P30 AG010129 (PI Charles DeCarli, MD), P50 AG016573 (PI Frank LaFerla, PhD), P50 AG016570 (PI David Teplow, PhD), P50 AG005131 (PI Douglas Galasko, MD), P50 AG023501 (PI Bruce Miller, MD), P30 AG035982 (PI Russell Swerdlow, MD), P30 AG028383 (PI Linda Van Eldik, PhD), P30 AG010124 (PI John Trojanowski, MD, PhD), P50 AG005133 (PI Oscar Lopez, MD), P50 AG005142 (PI Helena Chui, MD), P30 AG012300 (PI Roger Rosenberg, MD), P50 AG005136 (PI Thomas Grabowski, MD, PhD), P50 AG033514 (PI Sanjay Asthana, MD, FRCP), and P50 AG005681 (PI John Morris, MD).

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests

E.B.L. reports royalties from UpToDate. J.A.S. reports personal fees from Avid Radiopharmaceuticals, personal fees from Navidea Biopharmaceuticals, outside the submitted work. H.Z. has served at advisory boards of Eli Lilly, Roche Diagnostics and Pharmasum Therapeutics and is one of the founders of Brain Biomarker Solutions in Gothenburg AB, a GU Ventures-based platform company at the University of Gothenburg. K.B. has served at advisory boards of Alzheon, BioArctic, Eli Lilly, IBL International, Fujirebio, Merck, and Roche Diagnostics and is one of the founders of Brain Biomarker Solutions in Gothenburg AB, a GU Ventures-based platform company at the University of Gothenburg.

Glossary

Abbreviations

eQTL

expression quantitative trait locus

GWAS

genome-wide association studies

NFT

neurofibrillary tangles