The present study is a secondary analysis utilizing the baseline data from the Alzheimer's Prevention through Exercise (APEX, NCT02000583) study, a 52-week exercise intervention conducted at the University of Kansas Alzheimer's Disease Research Center examining change in AD-related neuroimaging biomarkers. Baseline medical history, neuropsychological testing, and MRI and amyloid-β imaging with PET (18F-AV-45) were assessed in all participants during baseline evaluations. The APEX study recruited a convenience sample of 121 cognitively unimpaired, sedentary older adult participants. Inclusion criteria for APEX required age of 65 years and older, a Clinical Dementia Rating (CDR) Scale of 0, a Mini Mental State Examination (MMSE) score of 27 or greater, normal cognitive test performance on the Uniform Data Set neuropsychiatric battery (74, 75) for age and years of education (<1.5 SD below the mean), stable 30-day medication regimen, sedentary or underactive as defined by the Telephone Assessment of Physical Activity (76), and amyloid-β positive measured by PET (18F-AV-45). Persons with insulin-dependence, significant hearing or vision problems, clinically evident stroke, cancer in the previous 5 years, or recent history (<2 years) of major cardiorespiratory, musculoskeletal, or neuropsychiatric impairment, were excluded.

APOE genotype status was determined from frozen whole blood samples (acid citrate dextrose anticoagulant) using an allelic discrimination assay to identify single nucleotide polymorphisms (ThermoFisher). We distinguished APOE4, APOE3, and APOE2 alleles using Taqman probes to APOE-defining polymorphisms: rs429358 (C_3084793_20) and rs7412 (C_904973_10). This was included in the model as the number of ε4 alleles (0, 1, 2).

The presence or absence of the following cardiovascular risk factors were included as predictive measures in the present analysis: (1) Optimal fasting glucose: A fasting glucose measurement was collected twice during the baseline visit and an average of these measurements was calculated. A fasting glucose level of above 100 mg/dl was considered non-optimal (77). (2) High mean arterial pressure: Mean arterial pressure was collected at the baseline visit and pressures that read above 100 was considered high (78). (3) Higher than optimal waist to hip ratio: Waist to hip ratio was calculated at the baseline visit and dichotomized as yes/no depending on whether it was above the sex-specific cutoff of greater than or equal to 0.85 in women and greater than or equal to 0.90 in men (79). A Dual-energy x-ray absorptiometry (DEXA) scan was completed during the baseline visit to measure total body, android (trunk and abdomen adipose tissue), and gynoid (hip and thigh adipose tissue) fat percentages. (4) Higher than optimal total body fat percentage: Total body fat percentage was calculated by dividing fat mass (g) by the sum of lean mass (g), bone mineral content (g), and fat mass (g). Total body fat percentage greater than 38% in women and greater than 25% in men was considered nonoptimal (80). (5) Total android fat percentage: Android fat percentages were calculated by dividing android fat (g) by total fat mass (g). (6) Total gynoid fat percentage: Gynoid fat percentages were calculated by dividing gynoid fat (g) by total fat mass (g).

For the proposed analysis, sex differences were examined among three neuroimaging techniques (18F-AV-45 PET imaging, T1 weighted MRI, and ASL-MRI, described in detail below).

Data from 121 participants were available for analysis. We excluded 9 participants (6 with incomplete genetic testing data and 3 with incomplete MRI data). The remaining 112 participants (ages 65–87 years) were examined, including 74 women and 38 men who were college educated on average and 95.5% of which identified as non-Hispanic, White. There were no quality control issues with the neuroimaging data from these 112 participants. Results from the Community Healthy Activities Model Program for Seniors (CHAMPS) self-report questionnaire suggest the sample did not engage in a substantial amount of physical activity and were primarily sedentary. Participants’ clinical, demographic, and anthropometric characteristics are given in Table 2. Notably, none of our participants in our sample carried 2 APOE4 alleles therefore APOE4 status in the subsequent models were reduced from three levels to two (0,1 alleles). Males and females did not differ by age, APOE4 carrier status, fasting plasma glucose, mean arterial pressure, or physical activity engagement. The female group was less educated (p = 0.049) and included a higher percentage of participants who identified as African American (p < 0.001), had a lower waist to hip ratio (p < 0.001), lower total body fat percentage (p < 0.001), higher android fat percentage (p < 0.001), and lower gynoid fat percentage (p < 0.001) than the male group.

Brain ROIs showing biomarker differences between sex groups were first identified among amyloid-β deposition (18F-AV-45 PET), T1 weighted volumetrics (MRI), and cerebral blood flow (ASL-MRI) among the whole sample.

LASSO regressions were used to identify which risk factor variables were most strongly associated with the Aim 1 identified brain ROIs showing neuroimaging biomarker differences between sex groups. LASSO regressions selected the most informative predictors for MRI volumetric measurements of the amygdala and entorhinal cortex as well as ASL-MRI cerebral blood flow measurements of the amygdala, temporal lobe, anterior cingular cortex, precuneus, superior parietal lobe, parahippocampal gyrus, and hippocampus regions. Ten AD risk factor variables were included as potential predictors including age, education, sex, APOE4 carrier status, non-optimal fasting plasma glucose, non-optimal mean arterial pressure, non-optimal waist-to-hip ratio, non-optimal total body fat percentage, android fat percentage, and gynoid fat percentage. To evaluate sex differences, interaction terms with each predictor multiplied by sex were included in the LASSO models (females = 1, males = 0).

We did not find any associations between sex and amyloid-β burden among the anterior cingulate gyrus, lateral temporal lobe, precuneus, superior parietal lobule or global amyloid-β burden after adjusting for age, education, and APOE4 status. It is important to note that we did not expect a large amount of variance among participants given the inclusion criteria of the primary APEX study required sub-threshold or elevated amyloid-β, which may have contributed to the lack of sex differences we observed. These results are consistent with literature using cross-sectional methodology among cognitively unimpaired older adults (100) and across the lifespan (49). Some research suggests sex differences in amyloid-β burden may exist but occur earlier in the aging process, such as during the menopause transition (47, 71). According to the estrogen hypothesis, estrogen may reduce aggregation of amyloid-β. This suggests that the decrease of estrogen during menopause may result in greater amyloid-β burden. Future studies may benefit from exploration of the potential effect of reproductive history, including years since menopause, and its relationship with amyloid-β burden among postmenopausal women. An alternative possibility for the lack of observed sex differences in amyloid-β may relate to a lack of control for AD family history in our study, for which prior studies for have demonstrated sex differences (101). Future studies should consider family history when examining the interaction between sex and asymptomatic cerebral β-amyloidosis.

Our volumetric findings are consistent with the UK Biobank, the largest single-sample study of structural sex differences in the human brain ages 40–69 (102). The study found males generally had larger total brain volume and among similar brain regions of interest to the present study to include the hippocampus, nucleus accumbens, amygdala, caudate nucleus, dorsal pallidum, putamen, and thalamus, accounting for total intracranial volume. Interestingly, they also found women had thicker cortices compared to men globally and among these regions, except notably the entorhinal cortex which was found to be thicker among men. It is important to note that greater brain volume does not necessarily indicate better brain health and can be region specific (103). For example, greater amygdala volumes are associated with depression, while greater hippocampal volumes are associated with higher levels of education (104, 105). Future studies should independently examine sex differences among regional cortical thicknesses, which have been shown to be phenotypically independent from structural volume measurements (106). In a sample slightly younger than the current study, Rahman et al. and Mosconi et al. also demonstrated lower MRI gray and white matter volume among women in the amygdala compared to men. Contrary to our results, they also found women had significantly lower volumes in the hippocampus, parahippocampal gyrus, insula, and caudate in addition to the superior, middle, and orbital frontal gyrus, anterior cingulate (ACC), and putamen among 46–65 year old females compared to age-matched males (47, 71). Future studies with larger sample sizes and across age groups during post-menopause should explore the potential underlying mechanisms of sex hormones in volumetric sex differences accounting for reproductive system health history.

Other studies have found no sex differences among structural volumetric measurements of the hippocampus and amygdala among samples of healthy older adults (51, 107). Contrary to these studies, our sample included sedentary older adults with asymptomatic cerebral β-amyloidosis. Further, findings from the Buckley et al. review suggest a greater amyloid-β burden sensitivity in females relative to males such that females with asymptomatic cerebral β-amyloidosis had steeper cognitive decline and neurodegeneration compared to males (100). Taken in context of our volumetric findings, this may suggest faster neurodegeneration among women with preclinical AD risk factors compared to men. However, the cross-sectional design of the present study limits this interpretation, and longitudinal studies are needed to further investigate this potential sex difference in AD trajectory.

Our study found that older adult women with asymptomatic cerebral β-amyloidosis had higher blood flow in regions associated with AD pathology and estrogen receptor network ROIs compared to men. The mechanism explaining the interaction between sex and cerebral blood flow in our results remains unclear. It is important to note that greater blood flow does not necessarily indicate better brain health. Prominent theories in the aging neuroscience literature regarding compensation, maintenance, reserve, and age-related brain restructuring or reorganizing are examples of when higher blood flow in a certain region is not a marker of health (108, 109). Mosconi et al. demonstrated the role of estrogen on cerebral blood flow during the menopause stages. The study found higher cerebral blood flow in postmenopausal women compared to perimenopausal women in the supramarginal gyrus, middle and superior temporal gyrus, superior and inferior front gyrus of both hemispheres (71). In their discussion of these findings, they proposed a compensatory reaction to glucose hypometabolism, and increased ketone metabolism after menopause as an underlying reason for the increased cerebral blood flow in these women (71). These results could also be interpreted as suggesting that blood flow is reduced during the fluctuations of perimenopause. During post menopause, blood flow may be returning to normal or possibly higher functioning as part of the compensatory reaction. Given our postmenopausal sample was older than the Mosconi et al. study, in addition to our volumetric sex difference findings, there could be a similar compensatory process occurring for women later in the aging process.

Only one other study to our knowledge has evaluated associations of demographic and cardiometabolic risk factors with MRI T1 weighted volumetric sex differences in cognitively unimpaired older adults (110) and none have examined predictors of ASL-MRI cerebral flood flow. In contrast to our methodological approach, Armstrong et al. followed its sample for 20 years and measured volumetric loss using cardiovascular predictors defined differently than in our study (110). They found that for males, hypertension and higher HDL cholesterol were protective against volume loss in the hippocampus and parahippocampal gyrus. In females, obesity (as measured by body mass index ≥30 kg/m² vs. <30 kg/m²) was found to protect against volume loss in the temporal gray matter but hypertension was associated with steeper volumetric decline in gray matter compared to men (110). Although we did not replicate their hypertension findings in the hippocampus or parahippocampal gyrus, we also found sex differences to include high mean arterial pressure predicted higher volume in the entorhinal cortex for males and lower volume for females.

Contrary to their study, our study measured obesity using more precise methods including waist to hip ratio, DEXA measured total fat percentage, and type of fat distribution (android and gynoid fat percentages). Our study found that for men, more android fat but less gynoid fat and total fat was associated with smaller entorhinal cortex volumes. In contrast, we found for females, waist to hip ratios greater than or equal to 0.85 was associated with smaller amygdala and entorhinal cortex volumes and more android fat predicted higher cerebral blood flow. It is worth noting that measures of body composition have shown non-linear age-related associations with cognitive performance and dementia such that body fat or higher body weight may be protective at some life stages and deleterious at other stages (111). Thus, timing may be a critical element to understanding sex differences in the influence of cardiometabolic risk factors on the brain.

Weight distribution during aging differs between males and females and may further explain some of our findings. For males, adipose tissue slowly but consistently accrues in the trunk and abdomen during aging over time (android fat distribution). For females, prior to menopause, adipose tissues accrue in the hips and thighs (gynoid fat distribution). After menopause, gynoid fat mass stabilizes while android fat distribution steeply increases into older adulthood (112, 113). Android fat distribution is associated with visceral fat accumulation which is directly related to cardiovascular disease and metabolic disease compared to other types of fat, both of which have been recognized as risk factors for Alzheimer's disease. Females in our sample had significantly more android fat percentage on average compared to males, suggesting postmenopausal females in our sample have more visceral fat and susceptibility for increased cardiovascular risk factors.

Our sample was sedentary by design. Research has found sedentary behavior is associated with additional risk factors for Alzheimer's disease including impaired glucose and lipid metabolism (114). Sedentary behavior over time may lead to abdominal obesity and insulin resistance, the most prominent underlying risk factors for cardiovascular disease and metabolic syndrome (115–119). In addition, according to the estrogen hypothesis, postmenopausal women are also more susceptible to these cardiovascular and metabolic risk factors due to the loss of estrogen and its key protective effect for glucose transport regulation and aerobic glycolysis. This area of aging research may provide a potential understanding of the underlying cause for the association finding between waist to hip ratio with brain volume among AD pathology and estrogen receptor network region ROIs in our sample of sedentary older adult females. More research is warranted to further explain these findings.

We did not find any impactful risk factors associated with the ASL-MRI cerebral blood flow of these regions. The ASL-MRI technique is helpful with providing information on vascular physiology and neurodegeneration related to AD pathology (120). The present study was limited by the amount of risk factors included in the model directly related to vascular physiology. Future studies should consider including more risk factors, as acknowledged in our limitations section. Research has shown neurofibrillary tangle tau biomarkers and cerebral blood flow are correlated, independent of other well-known AD-related risk factors (i.e., APOE, amyloid, small vessel disease markers) (121–123). These studies suggest significant cerebral blood flow changes may be associated with more downstream AD pathology. Given our study sample consisted of cognitively unimpaired individuals that may be at high risk for developing AD, it may be too early to associate ASL-MRI measurements with any risk factors. More studies examining ASL-MRI in AD at-risk populations with consideration of tau are needed. Methodological limitations and inconsistencies of the ASL-MRI technique may have also contributed to our findings. The accuracy and reproducibility of results in studies using the ASL-MRI technique has been critiqued in the literature given its sensitivity to motion and reliance on the scanner's magnetic field strength (124). Studies also discussed concerns with lack of standardization in acquisition across the literature (125, 126).

The current analysis was unique in that it included neuroimaging techniques that considered two of the three categories as part of the A/T/N biomarker classification scheme for AD pathology (43). Uniquely, it adds ASL-MRI imaging, a unique imaging technique for highlighting sex and age differences in cerebral blood flow. However, the current analysis is missing consideration of neurofibrillary tangle tau biomarkers, which is commonly measured by elevated CSF phosphorylated tau and elevated NFT-tau ligand uptake on PET imaging analyses. Future studies should consider examining sex differences among all three categories of the A/T/N biomarker AD pathology scheme in preclinical AD individuals.

While the LASSO regression statistical approach is designed for predication accuracy, literature also suggests concerns with variable selection consistency (127, 128). Certain techniques were used in the current study to accommodate for conducting a LASSO regression with a smaller sample size (i.e., k2 and leave-one-out cross validation) that we acknowledge may have oversimplified the model threatening the reliability of our Aim 2 findings (129). Results should be interpreted with caution and future replication of these findings are needed. Further, LASSO regressions are designed to estimate the relationships between variables and make predictions however do not estimate causal effects. Our study's purpose was to identify the most influential risk factor associated with sex-specific AD biomarkers to inform future hypotheses and studies. Future studies are needed to study the underlying mechanisms among these predicted associations.

The current study is cross-sectional in design, and therefore causal direction of associations cannot be determined. Longitudinal studies are needed to determine if the identified sex differences are predictive of dementia or suggest healthy brain aging differences over time, especially considering changes that result from the menopause transition. The use of regression analysis to examine a potential relationship between neuroimaging biomarkers and cardiovascular risk factors in a cross-sectional study is also problematic. Neurodegeneration can only be measured using longitudinal study designs. It also cannot be determined whether blood flow is increasing or decreasing over time for either men or women. Lastly, cardiovascular risk factors and health are cumulative over time therefore analyses examining health outcomes should be controlling for health and related behaviors over a lifetime to examine potential effects in later life.

Our sample consisted of a preponderance of non-Hispanic White, educated, and high socioeconomic status older adults, diminishing the external validity of the results of the current study. However, our sample consisted of significantly more women who identified as African American (n = 5) compared to men (n = 0) (Table 2). African American populations are 2 times more likely than non-Hispanic white populations to be clinically diagnosed with Alzheimer's disease or dementia related disorders, with African American women shown to be at an even higher risk (130, 131). A recent review also suggested African American women may demonstrate more AD-related neurodegeneration and cerebral small vessel disease when compared to non-Hispanic white populations (132). Therefore, the sex differences identified for both volumetric MRI and ASL-MRI in our study may have been strengthened with the inclusion of significantly more African American women in our study sample. As for the Royse et al. meta-analysis findings suggested, more studies with larger, more diverse samples examining AD-related biomarkers and risk factors taking both sex and ethnoracial factors into consideration are needed. The imbalance between sample size of men and women and relatively small sample size of men may have limited the sex-associated difference analyses. However, the imbalanced sample size represents the ratio of Alzheimer's disease prevalence between men and women with two thirds of patients with AD being women. Our results are only generalizable to persons who identify as female or male and may not apply to trans-women or trans-men. Sex and gender exist on spectrums. Research on sex-differences must consider the health and aging experiences of trans-people or for persons who identify elsewhere on the sex/gender spectrum and examine factors unique to this population such as hormone therapy.

Another limitation of the current study was the binomial (yes/no) categorization of AD risk factors as predictive measures in the analysis including impaired fasting glucose, high mean arterial pressure, higher than optimal total body fat percentage, and higher than optimal waist to hip ratio. Although dichotomizing the risk factors aligned with typical approaches, we acknowledge it simplifies what is an underlying continuous process. Measurement imprecision inevitably results in some classification errors, particularly for values close to cutoff points. Future studies should consider the use of continuous biomarker variables to provide a more accurate representation of the severity of the risk factors.

The current study is limited by its lack of control for other well-established risk factors for AD with known sex and gender differences including other health conditions (e.g., dyslipidemia, hypercholesterolemia), health behaviors (e.g., diet, sleep, smoking), socioeconomic considerations (e.g., occupation, household income), and psychological factors (e.g., depression) (133–135). Another major limitation of the current study is its lack of reproductive history consideration, specifically menopause type (e.g., natural, medically or surgically induced) and treatment (e.g., hormone therapy) as potential contributors to AD risk. Natural menopause has been associated with AD-risk-specific neuropathology including amyloid-β deposition and decreased hippocampal volume (45, 46). Further, research has identified a greater risk for AD associated with early medically or surgically induced menopause that occurs prior to natural menopause (136, 137). A recent review and meta-analysis of research on hormone replacement therapy and its relationship to cognition and AD risk indicate a lack of benefits and potential harmful effects, except in women who undergo early surgically induced menopause. Current research suggests the formulation, route of administration, and timing of hormone replacement therapy initiation produce different effects and are critical to understanding the efficacy of hormone therapy for AD prevention (15, 138–149). Future studies should consider the combination of these risk factors, especially reproductive history, among one individual and the potential sex differences among these risk factors in relation to preclinical AD biomarkers.