Androgens among men are associated with a decrease in Alzheimer’s disease risk

Abstract

INTRODUCTION:

Increased sex hormones have been hypothesized to decrease Alzheimer’s disease (AD) risk. We assessed the association between sex steroid hormones with AD using a Mendelian randomization (MR) approach.

METHODS:

An inverse-variance weighting (IVW) MR analysis was performed using effect estimates from external GWAS summary statistics. We included independent variants (linkage disequilibrium R²<0.001) and a P-value threshold of 5×10⁻⁸.

RESULTS:

An increase in androgens was associated with a decreased AD risk among men: testosterone (OR:0.53; 95%CI:0.32–0.88; P-value:0.01; FDR P-value:0.03); dehydroepiandrosterone sulfate (DHEAS; OR:0.56; 95%CI:0.38–0.85; P-value:0.01; FDR P-value:0.03); and androsterone sulfate (OR:0.69; 95%CI:0.46–1.02; P-value:0.06; FDR P-value:0.10). There was no association between sex steroid hormones and AD among women, though analysis for estradiol had limited statistical power.

DISCUSSION:

A higher concentration of androgens was associated with a decreased risk of AD among men of European ancestry, suggesting androgens among men might be neuroprotective and could potentially prevent or delay an AD diagnosis.

INTRODUCTION

Alzheimer’s disease (AD) is more prevalent among females although it remains unclear whether females are at an increased risk after adjusting for their increased overall survival.[1] Several studies indicated a higher incidence among women of advanced age (>80 years) compared to similarly aged men,[2–6] while other studies did not replicate this finding.[7–9] Various hypotheses have been proposed for the potential sex difference in AD risk. These include differences in lifestyle and risk factors, such as educational attainment, smoking, or physical inactivity.[2,10–12] Another hypothesis is that differences or changes in sex hormone concentrations are associated with risk for AD.[13–16] Animal studies and cell-line studies have consistently shown that estrogen, progesterone, testosterone and other androgens possess neuroprotective effects.[17,18] Unfortunately, observational studies have been less consistent. Several studies have indicated that surgical menopause, bilateral salpingectomy, low or nulliparity, and shorter reproductive periods are associated with an increase in AD risk, although these findings are not always consistent (for an extensive review see[19]). In addition, clinical trials studying the effect of hormone replacement therapy (HRT) in postmenopausal women on AD have been inconsistent.[20] One hypothesis for this discrepancy is that the correct timing of HRT may be critical though further research is warranted.[21–23] A recent review indicated that men with androgen deprivation therapy (often prescribed due to prostate cancer) had an increased risk for cognitive decline and AD, while testosterone therapy was associated with a reduced incidence of AD or dementia.[17] One Mendelian randomization study, using genetic variants associated with age at menopause and menarche, did not find an association with Alzheimer’s disease.[24] However, a major limitation of the study was that the analysis was not sex stratified. As we have previously shown for age of menopause and Parkinson’s disease,[25] sex-stratified MR analysis are necessary when analyzing sex-specific exposures.

Here, we analyze the association between sex steroid hormones (androsterone sulfate, dehydroepiandrosterone sulfate (DHEAS), estradiol, sex hormone binding globulin (SHBG) and total testosterone) and AD status. Based on the hypothesis that sex hormones are neuroprotective, we assessed whether an expected higher concentration of various sex hormones are associated with a decreased risk of AD among men and women.

METHODS

We used a Mendelian randomization (MR) analysis approach. A MR analysis estimates the effect of exposure on outcome by using the effect estimates of genetic variants on the exposure of interest, and the effect estimates for the genetic variants on disease status, through an inverse weighting calculation and using the genetic variants as instrumental variables.²⁰ One benefit of MR analysis is limiting the risk for confounding bias.

We utilized the summary statistics data from several large external Genome-Wide Association Studies (GWAS).[26–29] These previously performed studies identified and quantified the effect estimates for genetic variants on the sex hormones of interest to this study. The summary statistics of genetic variants on AD status were derived from the sex-stratified Alzheimer’s Disease Genetics Consortium (ADGC) GWAS.[30]

GWAS of exposures

We used publicly available external summary statistics for five sex hormones (androsterone sulfate,[26] DHEAS,[27] estradiol (binary only),[28] SHBG,[29] and total testosterone).[29] When multiple GWAS summary statistics were available for an individual sex hormone, the GWAS with the largest sample size was chosen and the alternative GWAS summary statistics were used in a sensitivity analyses (see supplemental table 1). All GWAS summary statistics were at minimum adjusted for age and sex. Additionally, the GWAS for SHBG and estradiol were adjusted for body mass index (BMI) as BMI is related to SHBG and estradiol, and might confound the association with Alzheimer’s disease.[31–36] We restricted the sex hormone GWA studies to those of European ancestry, as the AD-related genetic data were restricted to this subpopulation.

For each of these sex hormone GWAS, we restricted the genetic variants to those with minor allele frequencies (MAF) of at least 1%, that were bi-allelic and on autosomal chromosomes. The androsterone sulfate summary statistics by Shin et al.,[26] consisting of 7,338 subjects and 2,544,104 SNPs. DHEAS summary statistics were derived from Prins et al.,[27] consisting of 5,414 female and 4,308 male subjects with 6,245,773 SNPs. The summary statistics for testosterone and SHBG were retrieved from Ruth et al.[29] This study only provided summary statistics for SNPs with a genome-wide statistically significant P-value of 5×10⁻⁸, i.e. 501 SNPs for SHBG and 183 SNPs for testosterone. The GWAS analysis on SHBG were based on 368,929 participants and was adjusted for age, sex, BMI, batch, and menopausal status. Participants with hormone therapy were excluded and SHBG concentrations below the measurement threshold were set to the lower limit.[29] The GWAS analysis for testosterone was based on 425,097 subjects.[29] For estradiol, we used data from Schmitz et al, which dichotomized the estradiol concentration to those above and below the threshold of 175 pmol/L. Data for 147,690 men and 163,985 females (38,194 pre-menopausal and 125,791 postmenopausal) were collected. The estradiol GWAS analyses among men and women were adjusted for age, BMI, array type, and 10 principal components (PCs) identified in the genetics data in order to adjust for any potential population stratification.[38] For women the analyses were also adjusted for hormone therapy, oral contraceptives, parity, menopausal status, and hysterectomy. Summary statistics for men and women were provided separately and a fixed-effects meta-analysis was performed using the software program METAL.[37] This provided summary statistics for 6,933,662 SNPs.

GWAS of Alzheimer’s disease

We received sex-stratified summary statistics for late-onset AD from the Alzheimer’s Disease Genetics Consortium (ADGC).[30] These data are available through the NIAGADS website. The ADGC GWAS results were derived from fifteen AD studies and consisted of two phases (18,844 and 5,342 subjects). Except for one individual, all participants were aged 60 years and above. There were 5,705 female and 4,010 male cases, and 7,067 female and 4,672 male controls. The analyses were adjusted for age at onset (or age at visit for controls), cohort indicators, and top 10 PCs. We excluded SNPs with a MAF<0.01 and multi-allelic variants leaving 7,713,203 SNPs available for male and female subjects. We performed a fixed-effect meta-analysis using METAL [37] to establish the effect estimates of the genetic variants on AD risk for the combined male and female population (from here on referred to as the total population).

Mendelian randomization analysis

For each sex hormone in each GWAS, we determined the independent SNPs by clumping them based on European ancestry reference data (1000 Genomes Project). In the process of clumping, the most statistically significant variant (the index variant) is selected and other variants in the region that are correlated with the index variant are removed from further analysis. This ensures all variants chosen in the final model are strongly associated with the exposure and are independent of one another. Utilizing the TwoSampleMR package in R, we set an R-squared threshold of 0.001 and a genomic region of 10,000kb.[39] We restricted the SNPs within the exposure GWAS (sex hormone) to those that were available in the AD GWAS before clumping. After clumping, we harmonized the data from the exposure GWAS with the AD GWAS to ensure correct orientation of the exposures and outcome effect alleles. For the main analysis, we restricted the genetic variants to those with a P-value threshold of 5×10⁻⁸, resulting in 3, 6, 8, 250, and 120 SNPs for androsterone sulfate, DHEAS, estradiol, SHBG, and testosterone, respectively.

For each SNP, we calculated the proportion of variance of the exposure explained by the SNP (R²), and the F statistic. We summed the R²s for our power calculation providing a range between 0.002 (estradiol) and 0.15 (SHBG).[40] Given the estimated R² and assuming a standardized and normalized (mean 0; standard deviation 1) dataset for the sex hormone GWAS summary statistics, we had sufficient power (>80%) to detect an odds ratio between 0.26 (males and estradiol) and 0.90 (total population and SHBG). These predictions were highly dependent on the subset of AD population (males, females, or total) and the hormone exposure of interest (see supplemental table 2); the power for estradiol was low compared to the others. As our primary analysis, we performed a random-effect inverse variance weighted (IVW) MR analysis and adjusted for multiple testing using a Benjamini-Hochberg False discovery rate (FDR).

Sensitivity analysis

We performed MR PRESSO to identify and adjust for, if necessary, potential outliers influencing our overall findings. Additional sensitivity analysis included removing palindromic SNP (A/T or G/C alleles with potential uncertainty on the effect or reference allele due to ambiguity about the forward versus reverse strand) with a MAF between 0.3 and 0.5, as well as all palindromic SNPs. These two sensitivity analyses ensure the findings are not biased by possible incorrect harmonization. Lastly, we performed an analysis restricting to SNPs with a MAF of more than 0.05 in the original exposure and disease GWAS to ensure the findings are not based on rare variants with potential type II errors.

To test for reverse causation, we performed a bidirectional MR analysis for each sex hormone. Here the genetic variants were those strongly associated with AD, the exposure variable was AD status, and the outcome was the sex hormone concentration. We identified independent SNPs associated with AD risk (P-value<5×10⁻⁸, clumping-R² of 0.001) and then used the effect estimates for each sex hormone of these specific genetic variants to perform the IVW MR analysis. As the dataset from Ruth et al.[29] did not provide genome-wide effect estimates for SHBG and testosterone, we used the UK-biobank derived dataset downloaded from the MRC Integrative Epidemiology Unit website instead (“ukb-d-30850_irnt”, “ukb-d-30830_irnt”).[41]

MR Phenome-Wide Association Study (PHEWAS) and multivariable MR analysis

For each of our sex hormone exposures, we performed an MR Phenome-Wide Association Study (PHEWAS) for the top 10 hormone-related SNPs that were identified after clumping using the “ieugwasr” R-package with a P-value threshold of 1×10⁻⁵. This analysis identified three potential pleiotropic pathways that previously have been hypothesized to be related with AD status and are associated with the genetic variants used in this MR analysis. These three pathways were: anthropometric measures (BMI), cholesterol (triglycerides), and cell type counts (lymphocytes and neutrophils count). We then performed a multivariable MR analysis for each of the sex hormones adjusting for these pathways by including their summary statistics (“ieu-b-40”,”ieu-b-111”,”ieu-b-34”,”ieu-b-32”) in the multivariable analysis. For this analysis, we used the UK-biobank derived datasets (“ukb-d-30850_irnt”, “ukb-d-30830_irnt”, “ukb-d-30800_irnt”).[41]

Data availability and informed consent

All study protocols regarding human subjects have been approved by their local Institutional Review Board and informed consent was given by all participants. All data are either publicly available or available through the NIAGADS website.

RESULTS

Main analysis

In our main IVW-MR analysis, among men, we identified an inverse association between three androgens and AD status, specifically: DHEAS (OR:0.56; 95%CI:0.38–0.85, P-value:0.01, FDR P-value:0.03), testosterone (OR:0.53; 95%CI:0.32–0.88, P-value:0.01, FDR P-value:0.03), and possibly androsterone sulfate (OR:0.69; 95%CI:0.69–1.02, P-value:0.06, FDR P-value:0.10). Though the effect estimates for sex hormones on AD status were in the same inverse direction for females, none were statistically significant. The effect estimates for the total population, men and women combined, are mainly driven by the effects among men. There was no indication for an association between estradiol or SHBG and AD. A visual representation for the MR analysis is provided in figure 1 (forest plots and leave-one-out plots are available in supplemental figure 1 and 2). The leave-one-out analysis indicated a consistent effect for DHEAS, testosterone and androsterone sulfate (see supplemental table 3). Although the global test indicated potential outliers for testosterone among females and in the total population, no outliers were detected in the MR-PRESSO analysis. Both the leave-one-out analysis and the MR-PRESSO analysis indicate that the results are not due to a few influential SNPs.

Figure 1.

Visual representation of MR of sex hormones (androsterone sulfate, DHEAS, estradiol, SHBG, and testosterone) on Alzheimer’s disease status. Effect estimates on the X-axis are for SNP on exposure, Y-axis indicates effect estimates (log odds) on AD. Line represents the estimated effect of exposure on AD using the inverse variance weighted MR method.

Sensitivity analysis – different SNP sets

Results are consistent when we used different P-value thresholds (see supplemental table 4). Removing ambiguous alleles with a MAF between 0.3 and 0.5, or removing all ambiguous SNPs independent of MAF, did not change the overall findings (supplemental table 5a and 5b). When limiting to SNPs that had a MAF of at least 5%, results were very similar (see supplemental table 5c).

Alternative GWAS-based hormone exposures

As a second sensitivity analysis, we performed the IVW-MR analysis analyzing the association between sex hormones and AD status using summary statistics from different GWA studies.[26–29,41–44] As these were less well-powered, they were mainly utilized to further examine associations between each of the sex hormones and AD. Except for one androgen (4-androsten-3beta-17-betadioldisulfate2), all androgens showed a negative association with AD although not all were statistically significant. Similarly, we identified an inverse association between DHEAS and AD among men when using the two other relatively large GWAS that were based on a total (men and women) population.[27,42] For the estrogens, the results are not consistent across various GWAS, with some indicating a decrease and some an increased risk of developing AD. The effect estimates for estrone was derived from one GWAS study, indicating a potential protective effect of estrone on AD among females (OR:0.46; 0.27–0.79; P-value:0.005). Using data from two of the largest GWAS of total testosterone indicated a potential protective effect of total testosterone on AD risk among males. This suggests that there may be a small protective effect of total testosterone on AD risk among males, while no or only a small effect among females was detected. There was no indication for associations between SHBG, bioavailable testosterone, 17-hydroxy-progesterone, progesterone with AD status using these alternative GWA studies.

Bidirectional MR (AD as exposure, sex hormones as outcome)

We performed a reverse MR analyzing associations of AD status on each sex hormone concentration (see table 2). AD status is associated with an increase in SHBG concentration among all sexes (OR for men:1.64; 95%CI:1.39–1.92, P-value:2×10⁻⁹, FDR P-value:9×10⁻⁹; OR for women:1.48; 95%CI:1.10–2.20, P-value:0.01, FDR P-value:0.05; OR for total population:1.48; 95%CI:1.18–1.85, P-value:7×10⁻⁴, FDR P-value:3×10⁻³). In addition, there is a small negative association between AD status and DHEAS concentration among females and in the total population (OR for women: 0.98; 95%CI:0.96–1.00, P-value:0.03, FDR P-value:0.08; OR for total population: 0.98; 95%CI:0.96–1.00, P-value:0.03, FDR P-value:0.07). There were no associations between AD status and the sex hormone concentrations after removing genetic variants within the APOE gene from the subset of variants used in the MR analysis.

Table 2a.

Bidirectional Mendelian randomization analysis for AD status and its estimated effects on sex hormones (androsterone sulfate, DHEAS, estradiol, SHBG, and total testosterone). The genetic variants for the MR analysis were derived from the ADGC GWAS, using the sex-stratified summary statistics.

	Males						Females						Total
Exp	N SNP	OR	LL	UL	P	FDR P	N SNP	OR	LL	UL	P	FDR P	N SNP	OR	LL	UL	P	FDR P
AD--> Andro	1	1.00	0.98	1.02	0.82	0.82	2	1.01	0.99	1.02	0.56	0.59	5	1.01	0.99	1.03	0.44	0.72
AD--> DHEAS	1	0.99	0.97	1.01	0.32	0.79	2	0.98	0.96	1.00	0.03	0.08	5	0.98	0.96	1.00	0.03	0.07
AD--> E2	1	1.00	0.98	1.03	0.76	0.82	2	0.99	0.97	1.01	0.54	0.59	5	0.99	0.98	1.01	0.61	0.72
AD--> SHBG	1	1.64	1.39	1.92	2×10−9	9×10−9	2	1.48	1.10	2.00	0.01	0.05	5	1.48	1.18	1.85	7×10−4	3×10−3
AD--> Tot T	1	1.00	0.99	1.02	0.61	0.82	2	1.00	0.99	1.02	0.59	0.59	5	1.00	0.99	1.02	0.72	0.72

MR Phenome-Wide Association Study (PHEWAS)

We performed an MR PHEWAS for the top 10 SNPs that were independently associated with each sex hormone (see supplemental table 6). Each of these genetic variants were highly correlated with various other sex hormone concentrations. Other phenotypic characteristics in the MR PHEWAS that were related to the top genetic variants included body composition, immune cell composition, cholesterol, and to a lesser extent vitamin D concentration, hypertension, cardiovascular disease, and gallstones.

Multivariable MR analysis

After adjusting for BMI, triglycerides, neutrophil and lymphocyte counts, the number of SNPs associated with the sex hormones were diminished to only 1, 0, 0, 40, and 9 SNPs for androsterone sulfate, DHEAS, estradiol, SHBG, and testosterone, respectively. Thus, no SNPs were remaining for DHEAS and estradiol after correction for these exposures. Even with a loss of SNPs after adjusting for the other exposures, the effect estimates of the MR analysis on AD among men for androsterone sulfate, SHBG, and testosterone remained very similar (androsterone sulfate: 0.72; 95%CI:0.52–0.99, P-value:0.04, FDR P-value:0.13; and testosterone: 0.43; 95%CI:0.15–1.26, P-value:0.12, FDR P-value:0.18; see supplemental table 7).

DISCUSSION

This MR study suggests that a higher concentration of DHEAS, testosterone, and, possibly, androsterone sulfate are associated with a lower risk of developing Alzheimer’s disease in men. In women, the estimated effect estimates for androgens were too small to be statistically identifiable in this study. The results were robust, even when changing P-value thresholds or using alternative sex hormone GWAS summary statistics. This is the first study that utilizes MR to analyze the association between sex hormones and AD. The strengths of this study are that it is based on summary statistics from large GWAS and that we examined sex-specific associations.

Some previous studies indicated that DHEAS supplementation may be beneficial, though results were inconsistent and potential benefits limited. Interestingly, certain studies indicated a positive effect among males only.[45–48] However, most of these studies were cross-sectional making it difficult to assess the direction of the association, as discussed in various reviews.[49,50] In addition, some studies identified a potential association between low testosterone concentration and stronger cognitive decline among men. However, two recent studies did not find such an association: one a randomized clinical trial with testosterone supplementation; the other a multicenter study.[51,52] Previous Mendelian randomization or genetic studies have not identified an association between testosterone and cognitive decline or all-cause dementia.[53–55] The lack of findings could be due to limited statistical power (i.e., low prevalence of outcome or restricted genetics) or the possibility of different causal pathways between all-cause dementia/cognitive decline and AD. Our study focused specifically on AD. It is possible that our findings are specific to Alzheimer’s disease and androgens may have no or limited effects on all-cause dementia.

Unfortunately, our study was still underpowered to detect an association between estradiol and AD. Hence, additional larger studies are needed to re-examine this hypothesis. Such studies will need to take age or time since menopause into account to be able to decipher a possible association between estradiol and AD.

The three hormones associated with AD risk in our study (androsterone sulfate, DHEAS, and testosterone) are all androgens. As they are metabolites within the same hormonal pathways, they are strongly correlated with each other. Androsterone is an endogenous steroid hormone that is a weak androgen compared to testosterone. Both androsterone and DHEAS are precursors for (dihydro)testosterone. Thus, it is not surprising that we noted strong correlations in our MR PHEWAS. It is therefore not possible to identify whether one or all of these hormones are causally associated with AD risk.

The GWAS for our exposure instruments used various adjustments for additional covariates. The GWAS for estradiol and SHBG but not for other sex hormones adjusted for BMI. Since BMI itself was not associated with Alzheimer’s disease in our study, it is less of a concern that some of the GWAS were not BMI adjusted.

One assumption of MR analysis is that the genetic variants only have an effect through the exposure of interest. Our analysis suggests pleiotropy as the MR PHEWAS the genetic variants are also related to body composition and cholesterol concentration, especially for those genetic variants associated with testosterone and SHBG. However, as our results do not change in the multivariable MR analysis when adjusting for BMI, triglyceride, and blood cell type composition, the effects in our analysis can be attributed to sex hormones and not due body composition, cholesterol, or blood cell composition pathways. However, we cannot completely exclude the possibility of remaining pleiotropic effects that influence our overall results, though we expect this to be limited.

One concern of MR analysis is the risk of reverse causation. However, the bidirectional MR analysis did not show an association between the genetic risk for AD and sex hormone concentration, except for SHBG which was not associated with AD. There was some indication that the genetic risk for AD was associated with a slightly lower DHEAS concentration among females, though there was no effect of AD status on DHEAS among males. Hence, there is no evidence for reverse causation. Another potential reason why reverse causation is unlikely is that the GWAS for each of the sex hormones were performed on large study populations that were middle-aged. At this life stage, sex hormone levels often have stabilized. Most women will be postmenopausal and, unlike menopausal women, the timing of the blood draw in relation to hormonal cycles does not need to be considered. Therefore, we believe that the genetic variants we identified in the GWAS likely represent a lifetime of higher or lower sex hormone concentration and thereby indicating a higher or lower sex hormone concentration prior to the AD onset and diagnosis. Overall, these results indicate that the androgen levels would likely influence AD risk and not vice versa.

Lastly, our study only analyzed individuals of European descent as both the ADGC as well as the large GWAS studies for sex hormones were conducted amongst an exclusive European ancestry population. Therefore, further studies need to evaluate how well these genetic variants perform as genetic instruments in other ethnicities.

CONCLUSION

This study provides evidence that higher concentrations of androgens are associated with a decreased risk of AD among men of European ancestry. Estimated effect for androgens were small or null among women. No effects for SHBG or estradiol were found, though the statistical power to detect an association with estradiol was limited.

Though our study indicates the strongest effect for DHEAS, due to high correlations between androgens and genetic pleiotropy, it was impossible to decipher which androgen might be causally associated with AD. Overall, our study indicates androgens might be neuroprotective in men and could potentially help prevent or delay an AD diagnosis.

Table 1.

Mendelian randomization for sex hormones (androsterone sulfate, DHEAS, estradiol, SHBG and total testosterone) and their estimated effect on Alzheimer’s disease status, stratified by sex. We adjusted for multiple testing using the Benjamini-Hochberg FDR P-value.

		Males					Females					Total
Exp	N SNP	OR	LL	UL	P	FDR P	OR	LL	UL	P	FDR P	OR	LL	UL	P	FDR P
Andro	3	0.69	0.46	1.02	0.06	0.10	0.84	0.58	1.23	0.38	0.47	0.78	0.59	1.03	0.08	0.13
DHEAS	6	0.56	0.38	0.85	0.01	0.03	0.82	0.59	1.13	0.22	0.47	0.71	0.55	0.92	0.01	0.05
E2	8	0.91	0.59	1.40	0.66	0.82	0.84	0.64	1.11	0.23	0.47	0.87	0.70	1.09	0.22	0.27
SHBG	250	0.98	0.67	1.42	0.91	0.91	0.97	0.72	1.30	0.85	0.85	0.97	0.77	1.23	0.83	0.83
Tot T	120	0.53	0.32	0.88	0.01	0.03	0.82	0.52	1.28	0.38	0.47	0.69	0.48	0.99	0.04	0.10

Abbreviations: N: Number of SNPs used in the MR analysis; OR: Odds ratio; LL: 95%CI Lower Limit; UL: 95%CI Upper Limit; P: P-value; FDR P: False Discovery Rate P-value; Andro: Androsterone sulfate; DHEAS: dehydroepiandrosterone sulfate; E2: Estradiol; SHBG: Sex Hormone Binding Globulin; Tot T: Total Testosterone.

Table 2b.

Bidirectional Mendelian randomization analysis for AD status and its estimated effects on sex hormones (androsterone sulfate, DHEAS, estradiol, SHBG, and total testosterone). The genetic variants for the MR analysis were derived from the ADGC GWAS, using the sex-stratified summary statistics excluding SNPs within the APOE gene.

	Males						Females						Total
Exp	N SNP	OR	LL	UL	P	FDR P	N SNP	OR	LL	UL	P	FDR P	N SNP	OR	LL	UL	P	FDR P
AD--> Andro	0						1	1.02	0.95	1.10	0.54	0.68	4	1.02	0.98	1.06	0.39	0.88
AD--> DHEAS	0						1	0.94	0.84	1.06	0.32	0.68	4	0.97	0.91	1.03	0.28	0.88
AD--> E2	0						1	0.96	0.86	1.07	0.45	0.68	4	1.00	0.94	1.06	0.88	0.88
AD--> SHBG	0						1	0.65	0.30	1.39	0.26	0.68	4	0.95	0.50	1.81	0.88	0.88
AD--> Tot T	0						1	1.01	0.93	1.10	0.79	0.79	4	0.99	0.94	1.05	0.81	0.88

Abbreviations: N: Number of SNPs used in the MR analysis; OR: Odds ratio; LL: 95%CI Lower Limit; UL: 95%CI Upper Limit; P: P-value; FDR P: False Discovery Rate P-value; Andro: Androsterone sulfate; DHEAS: dehydroepiandrosterone sulfate; E2: Estradiol; SHBG: Sex Hormone Binding Globulin; Tot T: Total Testosterone.

HIGHLIGHTS.

• Sex hormones are hypothesized to play a role in developing Alzheimer’s disease (AD).

• The effect of sex hormones on Alzheimer’s disease was assessed using MR analysis.

• Among females, genetically determined effects of sex hormones were limited or null.

• Among men, higher concentration for androgens decreased AD risk.

• This study suggests a causal relationship between androgens and AD among men.

Research in Context.

Systematic Review:

We reviewed the literature using online databases. Previous animal and cell-line studies indicate neuroprotective effects for sex hormones. Unfortunately, observational human studies have shown inconsistent results.

Interpretation:

We used a two-sample Mendelian randomization (MR) approach to evaluate the relationship between five sex hormones (androsterone sulfate, DHEAS, estradiol, testosterone an SHBG) and AD status. This study suggests that an increase in androgens (androsterone sulfate, DHEAS, and testosterone) among men is associated with a decrease in AD risk.

Future Directions:

Though our study estimated the strongest effect for DHEAS, this study is not able to define which androgen is causally associated with AD due to pleiotropy. Future studies should determine which androgen or other correlated hormone may be reducing AD risk.