Abstract
Introduction Alzheimer’s disease (AD) disproportionately affects females. We determined whether physiological biomarkers (neuroplasticity, immune, stress, epigenetic) explain why females are more susceptible to AD than males using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database.
Methods Using the complete ADNI cohort, we analysed the effect of sex and APOE genotype (number of ε4 alleles) and sex and diagnosis (cognitively normal (CN), mild cognitive impairment (MCI), AD) on (1) AD related endpoints: memory scores, executive function scores, hippocampal volume, cerebrospinal fluid (CSF) amyloid beta, tau and p-tau; (2) markers of the immune system (interleukins, C-reactive protein, and immunoglobulins), neuroplasticity (intercellular adhesion molecule, ICAM1), and stress (cortisol); and (3) epigenetic age.
Results Females had higher levels of tau and p-tau compared to males and increasing alleles of APOEε4 disproportionately increased tau and p-tau compared to males. Females had larger hippocampal volume (corrected with intracranial volume) and better memory scores (that include verbal memory) than males, regardless of APOE genotype and diagnosis. There were also sex differences in biomarkers with females having higher levels of plasma C-reactive protein and lower levels of CSF IL-8, IL-16, immunoglobulin A, and ICAM1. We did not observe an association between sex, diagnosis, or APOE genotype and blood epigenetic age acceleration or intrinsic epigenetic age acceleration.
Conclusion In females tau pathology was increased but memory scores were higher and corrected hippocampal volume were larger compared to males suggesting females have a reserve against brain damage that delays either the onset of cognitive decline or diagnosis. In this ADNI cohort more males than females were diagnosed with MCI but with no significant difference in AD diagnosis, although more females presented with AD, suggesting the progression from CN, MCI to AD may be sex-specific. We found sex differences in immune biomarkers indicating that the underlying physiology may participate in differential aging with and without a diagnosis of AD or MCI between the sexes.
Introduction
Alzheimer’s disease (AD) is a neurodegenerative disease characterized by severe cognitive decline (Alzheimer’s Association, 2017). Modifiable risk factors associated with AD include stress (Caruso et al., 2018), sociocultural or lifestyle factors (e.g., education, marital status, exercise), and conditions (diabetes, obesity, and cardiovascular disease; Baumgart et al., 2015; Nebel et al., 2018; Xu et al., 2015). Non-modifiable risk factors include age, biological sex, and APOE genotype (Riedel et al., 2016). Females are more likely to be diagnosed with AD in Europe and Asia, although this sex difference may depend in part on geographic location as the sex difference is not always observed in studies from the United States (reviewed by Ferretti et al., 2018; Mielke et al., 2014; Nebel et al., 2018). Nevertheless, regardless of prevalence, females show greater neuropathology (brain atrophy, neurofibrillary tangles) and cognitive decline with AD than males in both Europe and the United States (Ardekani et al., 2016; Barnes et al., 2005; Holland et al., 2013; Hua et al., 2010; Irvine et al., 2012; Koran and Hohman, 2017; Lin et al., 2015).
The hippocampus is one of the first brain areas to show atrophy with AD (Apostolova et al., 2006; Jack et al., 2000; Kidron et al., 1997) and hippocampal atrophy correlates with cognitive decline (Petersen et al., 2000) and AD pathology (neurofibrillary tangles; Jack et al., 2002). Previous studies using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) indicate that females have greater atrophy rates and cognitive decline than males with AD (Holland et al., 2013; Hua et al., 2010; Lin et al., 2015). However, there is limited research into the role of sex in the possible mechanisms underlying AD. In addition, few studies have examined the interaction of genetic polymorphisms and biological sex in AD. The ε4 allele of the APOE gene is a well-known genetic risk factor of AD (Corder et al., 1993) and is associated with accumulation of amyloid beta protein (Ossenkoppele et al., 2015). In females between 65 and 75 years, one allele of ε4 increases the risk of AD by 4-fold relative to males, indicating that the APOE genotype affects males and females differently (meta-analysis by Neu et al., 2017). Understanding why females are at a higher risk and have a higher burden of the disease is important for the development of tailored treatments based on sex and genetics.
Chronic inflammation is a hallmark of AD, as evidenced by increased expression of proinflammatory cytokines in the brains of AD patients which can exacerbate AD pathology (Heppner et al., 2015; Kinney et al., 2018; Swardfager et al., 2010). There are sex differences in immune responses (Klein and Flanagan, 2016) which can affect neuroplasticity (Dantzer, 2018; de Miranda et al., 2017) and interact with stress (Dantzer, 2018), but it is not known how these may be related to sex differences in AD. Biomarkers are highly sought after to predict disease onset and progression and to understand the possible underlying mechanisms of AD to develop better treatments. Therefore, the first objective of this study was to investigate potential physiological biomarkers (neuroplasticity, immune, stress) that may explain sex differences in AD and in people at risk for AD using the ADNI database.
Aging biomarkers also include epigenetic alterations, and these have been associated with a variety of pathologies and adverse health conditions, including normal cognitive aging and neurodegenerative phenotypes such as AD (Hannum et al., 2013; Horvath, 2013; Levine et al., 2015; Yokoyama et al., 2017). Recently, molecular biomarkers of aging known as “epigenetic clocks” have been developed based on DNA methylation signatures (Hannum et al., 2013; Horvath, 2013). Epigenetic age or “DNAmAge” is a measure of the biological age of a sample (cell or tissue), and can be calculated across a range of tissues and time points, providing an accurate estimation of a sample’s chronological age based on the presence or absence of methylation at the 5’ carbon of informative CpG dinucleotides throughout the human genome (Horvath, 2013). Positive deviations of epigenetic age from chronological age (positive epigenetic age acceleration) reflect more rapid biological aging and have been associated with numerous factors including smoking, obesity, Parkinson’s disease, Trisomy 21, and cancer (Gale et al., 2018; Horvath, 2013; Horvath et al., 2015; Horvath and Ritz, 2015), while negative deviations of epigenetic age from chronological age (negative epigenetic age acceleration) have been associated with high life-expectancy populations and memory retention (Degerman et al., 2017; McEwen et al., 2017). In AD, epigenetic age acceleration of the frontal cortex was associated with amyloid load, neuritic plates, and cognitive decline (Levine et al., 2015). Intra-individual DNA methylation profiles in peripheral tissue are correlated with the epigenetic signature in the brain, likely due both to identical genetic background affecting DNAme, and common signatures of epigenetic aging (Braun et al., 2019), thus it is reasonable to hypothesize that epigenetic age acceleration may also be detectable in peripheral tissues such as blood in AD participants. In healthy individuals, aging males exhibit more positive epigenetic age acceleration than females in blood and buccal tissue, and multiple brain regions (Horvath et al., 2016); in AD and other diseases with a sex difference, it is possible that the underlying sex-specific pathological mechanisms may be reflected in epigenetic age acceleration measures – for example, in AD females could potentially have more positive epigenetic age acceleration than males.
Our aims were to first examine sex differences in cognitive ability, volume of the hippocampus, neuropathological markers of AD and the potential underlying physiological mechanisms (neuroplasticity, immune, stress) and how these may be affected by APOE genotype (number of ε4 alleles), and secondly by dementia status (cognitively healthy (CN), mild cognitive impairment (MCI), AD). Our third objective was to investigate epigenetic age in peripheral tissue of CN, MCI and AD participants, and to study the relationship between sex, APOE genotype, dementia status, and epigenetic age acceleration.
Methods
ADNI database
Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer’s disease (AD). For up-to-date information, see www.adni-info.org. Data used in this article were downloaded on or before Jan 16, 2019.
Statistical Methods: Sex and APOE genotype and sex and diagnosis
We included all participants that had a baseline diagnosis in the ADNI database (total n = 1,460, n= 630 females, n=830 males). Data included in our analyses were: demographics (age, years of education, and ethnicity), baseline diagnosis (cognitively normal, CN; early MCI, EMCI; late MCI, LMCI; or AD), number of APOE ε4 alleles (0, 1 or 2), ADNI executive function Z-scores, ADNI memory Z-scores (using data from the ADNI neuropsychological battery and validated in Crane et al., 2012; Gibbons et al., 2012), hippocampal volume (mm3), cerebrospinal fluid (CSF) amyloid beta (pg/ml), CSF tau (pg/ml), and CSF p-tau (pg/ml). The executive function score included WAIS-R Digit Symbol Substitution, Digit Span Backwards, Trails A and B, Category Fluency, and Clock Drawing (Gibbons et al., 2012). The composite memory score included Rey Auditory Verbal Learning Test, AD Assessment Schedule - Cognition, Mini-Mental State Examination, and Logical Memory data (Crane et al., 2012). A small subset of participants also had inflammatory markers measured in CSF (N = 279), and plasma (N = 527) listed in Table 2A. Hippocampal volume was divided by intracranial volume to correct for differences in brain size, as sex differences in hippocampal volume are influence by intracranial volume (Lotze et al., 2019; Tan et al., 2016) and is presented as a ratio.
We compared all available data for each study variable between the sexes using the Wilcoxon rank sum test for continuous variables and Fisher’s exact test for categorical variables. We used general linear models to determine the relationships between (1) sex and APOE genotype or (2) sex and dementia diagnosis and cognitive ability, corrected hippocampal volume, and biomarkers. All models included age as a covariate. To test the main question, all models initially included an interaction between sex and APOE genotype or sex and dementia diagnosis; if this interaction was not significant, it was removed from the model to estimate the main effects of sex and APOE genotype or diagnosis. Significance was based on the likelihood ratio test, and all p-values for comparisons of sex and either APOE or diagnosis for all outcomes combined were corrected for multiple testing using the Benjamini-Hochberg false discovery rate method (Benjamini and Hochberg, 1995). All regression analyses were carried out in R v3.5.1 (R Core Team 2018).
Statistical Methods: Epigenetic Age
We used DNAme data quantified with the Illumina Infinium HumanMethylationEPIC BeadChip array (“EPIC” array) for 1905 blood samples from 640 unique ADNI participants (n=284 females, n= 356 males; Vasanthakumar et al., 2017) with CN, MCI and AD diagnosis. DNAme IDAT files were read into R v3.5.1 (R Core Team, 2018) using the ‘minfi’ package, and annotated with the most recent version of the EPIC manifest, the Infinium MethylationEPIC v1.0 B4 Manifest File, (available from https://support.illumina.com/downloads.html) (Aryee et al., 2014; Fortin et al., 2017). We excluded 11 low quality samples from 9 unique participants from further analyses on the basis of having a median methylated or unmethylated probe intensity <10.5 (Aryee et al., 2014; Fortin et al., 2017), the remaining samples were background normalized and dye-bias adjusted with normal exponential out-of-band (“noob”) normalization (Triche et al., 2013). DNAme data were converted to beta values and biological sex for all samples was confirmed by clustering samples on all DNAme probes mapping to the X and Y chromosomes. Beta values were calibrated to Horvath’s 21,368-probe training dataset, and epigenetic age was calculated using R code modified for compatibility with the EPIC array using the 334/353 epigenetic clock probes present on the array from https://horvath.genetics.ucla.edu/ (Horvath, 2013; Teschendorff et al., 2013). The missing DNAme values at these CpG sites can also be imputed based on the k-nearest neighbors method. We observed a very high correlation between epigenetic age values calculated with the missing probes removed versus imputed with k=10 (R=0.99, p<2.2e-16), in agreement with previous reports; we therefore chose to remove missing probes (Fiorito et al., 2017; McEwen et al., 2018).
Prior to statistical analyses we removed all technical replicates. Epigenetic age acceleration was calculated as the residual of epigenetic age regressed on chronological age and technical/batch covariates, including the laboratory collection site at which blood samples were drawn, and EPIC microarray chip and row. Intrinsic epigenetic age acceleration, a measure designed to be independent of age-related changes in whole blood cell-type proportions, was calculated as described in Chen et al. (Chen et al., 2016) as the residual of epigenetic age regressed on chronological age, technical covariates of collection site, row, and chip, and the proportions of six blood cell types (CD8T, CD4T, NK, B cells, monocytes, and granulocytes) estimated from noob-normalized methylation data with the Houseman algorithm (Houseman et al., 2012). For participants who contributed more than one blood DNAme sample within the 2-year collection period, we determined that longitudinal data collected within the median 3.6-year error of the epigenetic clock could not be meaningfully evaluated, and therefore calculated mean epigenetic age acceleration measures per participant from all available time points and performed all statistical analyses on these mean values.
Statistical analyses of epigenetic age acceleration were conducted using data from the remaining 640 participants (see Table 2B). To determine if epigenetic age acceleration or intrinsic epigenetic age acceleration differed by sex, dementia diagnosis, or APOE genotype, we used unbalanced two-way ANOVA designs. With CSF biomarker (amyloid beta, tau and p-tau) data available from the ADNI repository for a smaller subset of participants with matched EPIC DNAme data, (n=533, see Table 2C) we used linear regression to test whether APOEε4 genotype, amyloid beta, tau, p-tau, dementia diagnosis, or sex were significantly associated with epigenetic age acceleration.
Results
Demographic and biomarker information
Table 1 gives a summary of the variables for the overall data set (N=1460). Overall, females were significantly younger and had fewer years of education than males (P<0.0001 for both). There were more white males than white females in our sample and there were more non-white females compared to non-white males (P<0.05). In terms of APOE genotype, there were no sex differences in distribution of APOE genotype with 11% females and 12 % of males possessing two alleles of APOEε4. In the overall data set, the proportion of participants in each of the diagnosis categories was significantly different for females and males (P<0.05). There were more females with a baseline diagnosis of AD compared to males (23.7% compared to 21.7%, unadjusted P = 0.41), although not significantly, and more females were cognitively normal than males (26.7% compared to 20.8%, unadjusted P = 0.01). However, there were more males with a diagnosis of late MCI (39.5% versus 32.5%, unadjusted P=0.007) and early MCI (18.0% versus 17.1%, unadjusted P=0.74) compared to females, although not significantly.
Because not all data were available for each subject we created a summary table for the participants: with CSF biomarkers (Table 2A; N=279), with whole blood EPIC DNAme data (Table 2B; N=640) and with matched EPIC data and measured CSF biomarkers (Table 2C; N=533). Among those with measured CSF biomarkers, demographics were very similar as per results from overall data set in Table 1(see legend of Table 2). For the data applicable to the participants with available EPIC DNAme data (Table 2B) and participants with EPIC DNAme data and CSF biomarkers (Table 2C), most of the demographics were similar to the entire data set except the proportion of participants in each of the diagnosis categories was not significantly different between females and males.
In the overall data set, females had a smaller uncorrected hippocampal volume but larger corrected hippocampal volume, greater CSF amyloid beta, tau and p-tau, and higher memory function z-scores than males (Table 1). Biomarkers in the CSF were measured in a subset of participants (Table 2A). In this smaller cohort, females and males had similar levels of CSF CRP, CD 40 antigen and IL-6 receptor. However, females had lower CSF cortisol, interleukin-3, interleukin 8, interleukin-16, immunoglobulin A, and intercellular adhesion molecule compared to males (Table 2A).
Sex and APOE genotype are associated with changes in memory, hippocampus volume, AD and CSF inflammatory markers
Our first aim was to investigate whether sex and APOE genotype interact to influence cognitive ability, volume of the hippocampus, and biomarkers of AD and inflammation. There were significant interactions between sex and APOEε4 genotype for CSF tau, p-tau, and IL-16 (Table 3). Tau and p-tau levels were significantly higher in females with one or two alleles of APOEε4 compared to males (Fig 1 A and B). Although CSF p-tau and tau levels also increase in males with APOEε4 genotype, they do not rise to the same extent as in females. IL-16 levels were significantly lower in females with no APOEε4 alleles compared to males, whereas levels were similar between the sexes with one or two APOEε4 alleles (Fig 1 C and D).
Both sex and APOE genotype were independently (main effects of sex or APOE genotype) associated with memory z-scores and corrected hippocampal volume (Table 3). Females had higher memory z-scores and larger corrected hippocampal volume across all APOE genotypes (Fig 1 E and F). Lower memory z-scores were associated with increasing number of APOEε4 alleles in both sexes. Similarly, corrected hippocampus volume was significantly lower with increasing number of APOEε4 alleles in both sexes. Increasing APOEε4 alleles was also associated with lower executive function z-scores, lower amyloid beta, and lower C-reactive protein (Table 3; Fig 1 G-I), however there was no additional association of these variables with sex. Finally, results were similar for biomarkers in plasma (Supplementary Table S3).
Sex and diagnosis are associated with changes in memory, hippocampus volume, AD and CSF inflammatory markers
We next tested whether sex and dementia status (CN, MCI, and AD) influenced cognitive ability, corrected hippocampal volume, and CSF biomarkers of AD and inflammation. There were no significant interactions between sex and diagnosis for any of the tested variables (memory, executive function, corrected hippocampal volume, CSF tau, p-tau, amyloid beta, and CSF and plasma inflammatory markers). However, overall both sex and diagnosis were independently associated with memory z-scores, corrected hippocampal volume and CSF tau and p-tau (Table 4). Females had higher memory scores, larger corrected hippocampus volume, and higher tau and p-tau compared to males, irrespective of diagnosis. As expected, increasing severity of diagnosis was associated with lower memory and executive function scores, smaller corrected hippocampus volume, and higher CSF tau and p-tau irrespective of sex (Fig 2 A-D).
We found that although females had higher CSF levels of interleukin 16 (IL-16), and lower levels of interleukin 8 (IL-8), immunoglobulin A (IgA), and intercellular adhesion molecule 1 (ICAM1), controlling for age, compared to males, there was no association between these variables and diagnosis (Fig 2 E-H). Finally, there were associations between diagnosis and executive function z-scores, and amyloid beta, controlling for age, but not between these variables and sex (Fig 2 I and J).
The results for biomarkers and inflammatory markers in plasma were similar (Supplementary Table S4), with the exception of a significant relationship between plasma C-reactive protein (CRP) and sex (adjusted p=0.03), and also between plasma cortisol and baseline diagnosis (adjusted P=0.01; Fig 2 K and L). Males have lower levels of CRP compared to females and we observed a trend between diagnosis and CRP levels in plasma with lower CRP levels in late MCI and AD (adjusted P=0.08). Plasma cortisol was lower in late MCI compared to CN but higher in AD compared to CN. In summary, although we detected associations between sex and diagnosis and various parameters, we did not find evidence for a clear sex and diagnosis interaction.
Epigenetic age, sex, dementia diagnosis, and AD biomarkers
We investigated the hypothesis that sex and dementia diagnosis affect epigenetic age acceleration in blood samples of ADNI participants (see Table 5).
Epigenetic age acceleration was not associated with sex, dementia diagnosis (CN, EMCI, LMCI, and AD), or the interaction of sex and diagnosis after multiple test correction (Figure 3). Intrinsic epigenetic age acceleration was also not significantly associated with participant sex, diagnosis, or their interaction term.
To assess the effect of sex and more broadly defined dementia-associated cognitive impairment on epigenetic age acceleration, we compared epigenetic age acceleration between participants with any form of clinically ascertained cognitive impairment (AD + LMCI + EMCI, n=423, proportion female 41%) and those without (CN, n=217, proportion female 50%). By two-way unbalanced ANOVA neither sex, dementia status, nor their interaction were significantly associated with epigenetic age acceleration after correction for multiple comparisons.
Matched biochemical data including APOEε4 genotype and CSF concentrations of amyloid beta, tau, and phosphorylated tau was available for a subset of participants with EPIC DNAme data (n=533). Based on the hypothesis that epigenetic age acceleration may be more strongly associated with concentrations of pathologically relevant compounds than with diagnosis, we assessed the impact of sex, APOEε4 genotype, amyloid beta concentration, tau and p-tau concentration on epigenetic age acceleration and intrinsic epigenetic age acceleration with linear regression. None of these variables was significantly associated with epigenetic age acceleration (Table 6, results for intrinsic epigenetic age acceleration not shown).
In addition to dementia diagnosis for all participants, we also had access to two composite scores designed by ADNI collaborators to reflect executive function and memory; these scores have been demonstrated to be independently predictive of the transition from mild cognitive impairment to a formal diagnosis of Alzheimer’s disease (Gibbons et al. 2012, Gale et al. 2013). By a two-way unbalanced ANOVA models investigating the effect of sex and memory score on epigenetic age acceleration, neither sex (p=0.248), memory score (p=0.486), nor their interaction (p=0.227) were associated with epigenetic age acceleration. In a similar model, neither sex (p=0.260), executive function (p=0.105), or the interaction term of sex and executive function (p=0.153) were associated with epigenetic age acceleration.
Discussion
In the present study, we found that tau related pathology in the CSF was disproportionately elevated by APOEε4 genotype in females compared to males. However, diagnosis and APOE genotype were independently associated with reduced memory scores, hippocampal volume (corrected by intracranial volume) and reduced CSF amyloid beta which was similar in males and females. Furthermore, there were main effects of sex as females had lower CSF cytokines (IL-8, IL-16, IL-18) and CSF and plasma immunoglobulins (IgA, IgE, respectively) but higher plasma CRP and tau related pathology compared to males, regardless of diagnosis and APOE genotype. Interestingly, females had larger corrected hippocampal volume and better memory scores which may contribute to their delayed diagnosis (Sundermann et al., 2017). Finally, we found no differences in epigenetic age acceleration by dementia diagnosis or sex in this cohort of samples with available whole blood EPIC DNAme data. In this ADNI cohort, slightly more females presented with a diagnosis of AD compared to males, whereas significantly more males presented with a diagnosis of MCI supporting the prevalence observed in bigger populations (Winblad et al., 2016; Mielke et al., 2014). Previous work has demonstrated sex differences in rates of AD and symptoms of AD (reviewed in Ferretti et al., 2018; Mielke et al., 2014; Nebel et al., 2018), and the current study also suggests that biomarkers of AD may be different between males and females between genotypes, and this should be considered in future studies and researchers should be cautioned to use sex as a biological variable in all analyses.
Females show greater tau neuropathology disproportionately affected by APOE genotype
In the present study, we found that females have significantly higher baseline tau and p-tau levels in CSF than males and these are indicative of the formation of neurofibrillary tangles and AD pathology (Blennow et al., 2015; Henriques et al., 2018). This is in agreement with a recent ADNI study (Sundermann et al., 2018; but see an earlier ADNI study Holland et al., 2013) and with animal models (Lewis et al., 2001). Intriguingly, we also found that levels of tau and p-tau were disproportionately elevated with APOEε4 allele expression in females compared to males. Previous studies indicate that females with the APOEε4 allele are at a greater risk for developing AD than are males with this allele (Altmann et al., 2014), and sex differences in tau and p-tau may be one underlying mechanism by which this occurs. In females (65-75 years of age) one allele of ε4 increases the risk of AD by 4-fold relative to males, indicating that genotype may affect females differently (Neu et al., 2017). Levels of CSF tau are hypothesized to increase after CSF amyloid beta declines and amyloid beta aggregates and deposits in the brain (Blennow et al., 2015). However, in this study although we found sex differences in CSF tau and p-tau levels, no significant differences were seen in CSF amyloid beta after controlling for age (see below) indicating that the pathway may be different in females compared to males or that the timeline of tau and amyloid beta deposition may not be consistent.
In this ADNI cohort, more females presented with a diagnosis of AD compared to males. Although the ADNI cohort is relatively small, this result supports the prevalence observed in bigger populations (Winblad et al., 2016). Together with the disproportionate effect of APOE genotype on tau-related pathology it supports the idea that females have a higher burden of the disease. On the other hand, more males presented with a diagnosis of MCI and this is in line with the research that males are more likely to be diagnosed with MCI compared to females (Mielke et al., 2014). Females progress faster from MCI to AD (Lin et al., 2015) and sex differences in tau related pathology found in the current study may be the underlying mechanism for this accelerated transition.
Sex differences in hippocampal volume depend on correction for intracranial volume. Females have better memory scores than males that may have been driven by verbal memory
In the present study, we found that increasing APOEε4 alleles and AD diagnosis was associated with reduced corrected hippocampal volume, memory and executive function scores consistent with past literature (Apostolova et al., 2006; Buckner, 2004; Ewers et al., 2012; Jack et al., 2000; Li et al., 2016; Mungas et al., 2010; Petersen et al., 2000; Pievani et al., 2011; Shi et al., 2014). Surprisingly, although females have higher levels of tau and p-tau, they presented with larger corrected hippocampal volume and better memory and executive function scores than males, regardless of diagnosis and APOE genotype. Previous studies have suggested that there are sex differences in hippocampal volume, favoring males, but the sex differences depend on whether hippocampal volume is corrected for by intracranial volume (Tan et al., 2016), a finding that is supported by the current study. In a number of studies, including the present study, males have a larger hippocampus without correcting for intracranial volume (Cavedo et al., 2018; Jack et al., 2015; Murphy et al., 1996; Ritchie et al., 2018; Sohn et al., 2018; Sundermann et al., 2018; Tan et al., 2016). However after correcting for intracranial volume, either the sex difference disappears (Cavedo et al., 2018; Ritchie et al., 2018; Tan et al., 2016) or females have larger corrected hippocampal volume (this study; Jack et al., 2015; Murphy et al., 1996; Sohn et al., 2018; Sundermann et al., 2018). Regardless of hippocampal volume, volume loss is greater in aging females (Ardekani et al., 2016; Koran et al., 2017; Murphy et al., 1996) and in females with one or two APOEε4 alleles (Fleisher et al., 2005). Although in the present study we did not examine longitudinal data, we found that increasing APOEε4 alleles reduced corrected hippocampal volume similarly in males and females. In contrast, when CN, MCI and AD individuals were analysed separately in the ADNI database, APOEε4 was associated with a smaller corrected hippocampal volume in CN males only, controlling for age and education (Sundermann et al., 2018). In addition, also using the ADNI database, Koran et al. (2017) found that females with low CSF amyloid beta had more hippocampal atrophy and faster decline in memory and executive function than males and this sex difference was more pronounced in APOEε4 carriers. Therefore, sex and APOE genotype can interact to affect corrected hippocampal volume reduction with age in certain subgroups and across time (e.g., in CN or individuals with low CSF amyloid beta). Differences in results between studies are likely due to differences in statistical analyses (e.g., analysing diagnosis groups separately, partitioning the data based on amyloid beta levels, and differences in covariates included) and/or whether longitudinal data analyses are included.
We found that in addition to larger corrected hippocampal volume, females also had better composite memory scores (but not executive function scores) than males, regardless of diagnosis and APOE genotype. Previous studies have found that females have better verbal memory in cognitively normal individuals (Jack et al., 2015), and in MCI and AD ADNI cohorts compared to males (Sundermann et al., 2018, 2016). Here we used the ADNI memory score developed by Crane et al. (2012) to detect abnormal memory including language, attention, and logical memory so it is possible that verbal memory may be driving the sex difference favouring females in the present study. In contrast, Buckley et al.(2018) found no sex differences using a composite cognitive score that includes memory and executive function (Preclinical Alzheimer’s Cognitive Composite score with semantic processing, PACC5) using ADNI and two other cohorts. In this study using the current ADNI cohort, males were slightly more educated than females, and although we did not use education as a covariate, one would expect education levels would have positive effects on memory, suggesting that education is not a factor for the observed sex difference in memory. Altogether, we found that in females tau pathology was increased but memory scores, which included verbal memory, were higher and corrected hippocampal volume were larger compared to males suggesting females have a reserve against brain damage that delays either the onset of cognitive decline (Stern, 2002) or diagnosis (Sundermann et al., 2017). However, once cognitive decline begins, females show higher rates of declines compared to males (this was observed by Buckley et al., 2018; Holland et al., 2013; Hua et al., 2010 using the ADNI database) perhaps because the underlying pathology is elevated in females.
AD affects amyloid beta similarly in both sexes
We found that AD diagnosis was associated with lower CSF amyloid beta, as expected, and this was irrespective of sex, which indicates greater amyloid deposition with AD (Henriques et al., 2018). These findings are consistent with data from studies in AD patients (Buckley et al., 2018) and in cognitively normal individuals (Jack et al., 2015). Other studies have found using PET that males have higher amyloid beta levels or lower amyloid beta burden compared to females dependent on APOE genotype (Sundermann et al., 2018) or in cognitively normal adults in the anterior cingulate (Cavedo et al., 2018). In this study, we used CSF amyloid beta data which detects abnormal amyloid deposition earlier than amyloid beta by PET (reviewed in Blennow et al., 2015). Thus, taken together, sex differences in amyloid beta may be detected in specific brain regions and later in the disease, although more research is needed investigating sex differences in AD biomarkers.
Females have higher CRP levels but lower cytokine and immunoglobulin levels compared to males
In this study, we investigated whether sex interacted with APOE genotype or dementia diagnosis to influence inflammatory, neurotrophic and neuroplasticity markers. We found that plasma CRP, a widely used inflammatory and cardiovascular marker (Koenig et al., 1999; Ridker et al., 1998), was affected by sex and APOE genotype. Females, regardless of diagnosis or APOE genotype, had significantly higher plasma CRP relative to males, consistent with findings in healthy individuals (Khera et al., 2005). Higher levels of peripheral CRP may suggest higher inflammation in females, which is associated with an increased risk in all-cause dementia (Koyama et al., 2013). In contrast, APOEε4 genotype decreased circulating CRP levels, consistent with previous research in large population studies (Hubacek et al., 2010; Yun et al., 2015). Recent meta-analyses, without regard to sex, did not find differences in peripheral levels of CRP in AD compared to control patients (Gong et al., 2016; Ng et al., 2018). However, in patients with mild and moderate dementia only, CRP levels were lower compared to the healthy control group (Gong et al., 2016). To our knowledge, no other study has examined sex differences in CRP in relation to AD.
We also found that CSF IL-16 was affected by sex and APOE genotype. CSF IL-16 levels were lower in females with no APOEε4 alleles compared to males, but with increasing number of ε4 alleles, no sex differences were detected. IL-16 has been implicated in AD (Rosa et al., 2006) and IL-16 levels decrease with disease severity (analysis without regard to sex; Motta et al., 2007). In this ADNI cohort, IL-16 levels were not affected by diagnosis but our results suggest that APOE genotype can modulate levels in a sex-dependent way. We also found biomarkers that were affected by sex but not diagnosis or APOE genotype for example, females had lower CSF levels of ICAM1 compared to males, but there was no influence of APOE genotype or diagnosis. Consistent with our findings, ICAM1 serum levels were lower in healthy females compared to males (Ponthieux et al., 2003). ICAM1 is a type of adhesion molecule associated with microvascular endothelial activation (Zenaro et al., 2017) and plasma ICAM1 levels (but not CSF levels; Nielsen et al 2007) were higher in patients with AD (Huang et al 2015; Nielsen et al 2007; Rentzos et al 2004). However, it is intriguing that females have lower CSF levels of cytokines (IL-8, IL-16, IL-18), and immunoglobulins (IgE and IgA) but higher tau pathology compared to males. Neuroinflammation is associated with AD but it can have both beneficial and detrimental roles (Walters et al., 2016). Increased expression of pro-inflammatory cytokines contributes to neuronal loss, while anti-inflammatory effects contribute to amyloid beta clearance (Heneka et al., 2015). In AD mouse models, some pro-inflammatory mechanisms reduced plaque pathology, while anti-inflammatory cytokines increased amyloid beta deposition (Chakrabarty et al., 2012, 2011, 2010a, 2010b; Ghosh et al., 2013; Shaftel et al., 2007). It has been suggested that there are beneficial pro-inflammatory mechanisms and detrimental anti-inflammatory mechanisms in AD (Heneka et al., 2015). It is possible that males and females have varying levels of beneficial vs detrimental immune responses which can affect how the disease progresses in each of the sexes but it is also important to remember that CSF levels may not match levels in different regions of the brain.
Sex, AD and biochemical markers do not affect blood epigenetic age acceleration
We did not observe an association between either sex or diagnosis and epigenetic age acceleration or intrinsic epigenetic age acceleration. To our knowledge, no other study has similarly probed epigenetic age acceleration in peripheral tissue in the presence of AD, or whether epigenetic age acceleration in AD is associated with sex.
This study was partially undertaken to investigate whether epigenetic age acceleration that has been associated with the AD brain is reflected in peripheral tissues. Levine et al. have previously demonstrated increased epigenetic age acceleration in AD, however Levine’s study was conducted on post-mortem prefrontal cortex tissue, and did not explicitly investigate the role of sex in epigenetic age acceleration (Levine et al., 2015). While brain-blood methylation profiles are reasonably correlated (r=0.86) (Braun et al., 2019), DNA methylation profiles of peripheral tissues are imperfect representatives of the brain, and do not recapitulate all epigenetic alterations with high fidelity. Thus, our findings do not contradict the finding of increased epigenetic age acceleration in the presence of AD in the prefrontal cortex, but suggest that accelerated epigenetic aging in AD is not a pan-tissue phenomenon. Our finding of a lack of significant association between AD, biological sex, and epigenetic age acceleration in whole blood DNA methylation profiles could suggest a tissue-specific dysregulation of an epigenetic maintenance system, in which the brain epigenome is most strongly affected by AD (Levine et al., 2015). The phenotype of patients affected by AD and global gene expression patterns of the APOE protein, with high expression in brain, and low expression in whole blood (GTEx Project, 2018) further support this hypothesis.
Intriguingly, epigenetic age was observed to be lower on average than chronological age (see Table 5). Horvath’s epigenetic clock was trained on DNAme data from older versions of the Illumina DNAme arrays with more limited genomic coverage; 19 of the CpG probes required to calculate epigenetic age via this method do not exist on the EPIC array. Two previous studies investigated the application of Horvath’s epigenetic clock to EPIC data with conflicting results (Dhingra et al., 2019; McEwen et al., 2018), the largest issue being chronic underestimation of epigenetic age due to the positive linear regression coefficients associated with the missing probes(Dhingra et al., 2019). Both imputing and removing the missing probes from the array resulted in a chronic underprediction of epigenetic age with Horvath’s clock, suggesting that this is likely an artefact of the array platform and probe-set rather than the method chosen to deal with missing values, although it is possible that an adjustment factor could be devised to more accurately apply Horvath’s clock to EPIC data. In future explorations of epigenetic age with EPIC DNAme array data this should be considered, as there are other epigenetic age predictors available that have been trained on EPIC data such as the PhenoAge and GrimAge clocks, although these tools have limitations as well; for example, both PhenoAge and GrimAge were trained only on blood DNAme data, as compared to the original pan-tissue epigenetic clock, and therefore may have limited applicability and relevance in other tissues (Levine et al., 2018; Lu et al., 2019).
Limitations
The ADNI cohort is not ethnically or socioeconomically diverse, being mostly composed of white (only 12 individuals were not-white) and highly educated individuals (average 15.69 years of education). As incidence, prevalence, and age of onset of AD varies by ethnicity (Hispanics, Fitten et al., 2014; Mayeda et al., 2016; African-Americans, Steenland et al., 2016) and education (Sharp and Gatz, 2011), our conclusions may not apply to more ethnically and socially diverse populations. In addition to sex, it is possible the underlying mechanisms of AD are different depending on ethnicity. Finally, the ADNI biomarker data set has a low sample size (279 total), especially when taking into account diagnosis, sex and APOE genotype. Small sample size is also a limitation of the epigenetic analyses presented. Even in the larger 640-participant cohort, only 37 participants (5.78%) had an AD diagnosis, so statistical analyses were underpowered to detect subtle differences by diagnosis group. Additionally other pathologies in these participants, such as cancer, cardiovascular disease, smoking status, or obesity may have influenced AD neuropathology, biomarkers and epigenomes and limited our interpretations.
Conclusion
As expected, more females presented with a diagnosis of AD whereas more males presented with MCI diagnosis compared to the opposite sex. AD biomarkers (CSF tau and p-tau but not amyloid beta) were disproportionately affected by APOE genotype in females compared to males supporting the idea that females share a higher burden of the disease. Interestingly, although females in this cohort had elevated AD biomarkers, they also had larger corrected hippocampal volume and higher memory function scores compared to males, regardless of APOE genotype and dementia diagnosis. Therefore, it is possible that females may have a reserve that protects the brain from damage to delay cognitive decline or delay diagnosis. Finally, we found that females had lower cytokine and immunoglobulin levels but higher CRP levels compared to males. Together our work suggests that that the underlying physiology of aging and AD may be sex-specific.
Acknowledgments
Data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California. Funding for this study was provided by a Canadian Institutes of Health Research (CHIR) grant to LAMG (PJT-148662) and a CIHR sex catalyst grant to WPR (Reference No. 158613). WPR receives salary support through an investigatorship award from the BC Children’s Hospital Research Institute. PDG is funded by the Alzheimer’s Association of the USA and Brain Canada with the financial support of Health Canada through the Brain Canada Research Fund (AARF-17-529705). The views expressed herein do not necessarily represent the views of the Minister of Health or the Government of Canada. AMI is funded by a University of British Columbia Four Year Doctoral Fellowship and the CIHR Frederick Banting and Charles Best Masters Research Award.
Footnotes
↵* Considered co-first authors.
↵δ Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf