Associations of Midlife Lifestyle and Health Factors with Long-Term Changes in Blood-Based Biomarkers of Alzheimer’s Disease and Neurodegeneration

Study population

The Beaver Dam Offspring Study (BOSS) is a longitudinal cohort study of sensory and cognitive aging. Baseline BOSS examinations took place in 2005–2008 with follow-up examinations at 5 (2010–2013) and 10 (2015–2017) years [25]. Examinations included a blood draw, measures of vascular health, and questionnaires on demographics and behavioral and medical history. The study was approved by the Health Sciences Institutional Review Board of the University of Wisconsin and written informed consent was obtained from all participants prior to each examination. Participants who provided consent for future testing of blood samples with samples available from all three BOSS examination phases were included in this study. More study details have been published [26] and this study sample was similar in their baseline characteristics to the complete baseline BOSS cohort [26].

Measurements

Statistical analyses

All statistical analyses were conducted with SAS 9.4 (SAS Institute, Inc., Cary, NC USA) and visualizations with R Studio version 4.1.2 with packages tidyverse and ggplot2 [33, 34]. All biomarkers were assayed in duplicate, and the average of the duplicates was used as a measure of biomarker concentration. Samples with a biomarker concentration below the limit of detection were assigned a value halfway between zero and the limit of detection for that biomarker (Aβ₄₂ set to 0.09 pg/ml (N = 9); TTau set to 0.04 pg/ml (N = 109); no samples below limit of detection for NfL and Aβ₄₀). The Aβ₄₂/Aβ₄₀ ratio was calculated by dividing the concentration of Aβ₄₂ by the concentration of Aβ₄₀ to normalize Aβ₄₂ for the total amount of Aβ peptides that are present in the specimen [35]. The Aβ₄₂/Aβ₄₀ ratio has previously shown better performance for AD diagnostics compared to Aβ₄₂ alone [35]. A higher ratio represents less pathology. We present results of the ratio multiplied by 1000 for display purposes. We natural log transformed the highly skewed data of TTau and NfL.

To determine the effect of baseline lifestyle and health-related factors of vascular, inflammation, and metabolic syndrome on change in biomarker levels (Aβ₄₂/Aβ₄₀, TTau, and NfL), we used linear mixed-effect models with the baseline risk factors as determinants and the repeated measures of biomarkers as the outcome (Supplementary Equation). The interaction of baseline risk factor with time indicates differences in rate of biomarker change by baseline risk factor. We tested for sex interactions in our models. In our study, men had significantly lower baseline Aβ₄₂/Aβ₄₀ levels as compared to women and women had a more rapid decline over time than men, which we published earlier [26]. We thus stratified models for Aβ₄₂/Aβ₄₀ by sex. We have also previously investigated age effects adjusting for sex. There were no age differences in 10-year change of Aβ₄₂/Aβ₄₀ or TTau. Older age groups had greater increases in NfL [26]. Previous studies have shown that blood-based neurodegenerative biomarker levels might be lower in individuals with a higher BMI and with larger blood volume [36–40] and suggested to adjust for BMI in analyses [39, 40]. We thus adjusted all models for age, sex, random intercept, random slope, and BMI. When analysis resulted in a non-positive definite G matrix of the random effects, indicating poor model fit, we reran models with random participant-specific intercepts only (in models of Aβ₄₂/Aβ₄₀ ratio in men). Results from linear mixed-effects models of Aβ₄₂/Aβ₄₀ were presented in Aβ₄₂/Aβ₄₀ ratio × 1000, where a higher value reflects less pathology. Results from models of log transformed TTau and NfL were exponentiated and are presented as % change, where a higher value reflects more neurodegeneration.

Sensitivity analyses

Models were repeated excluding participants with biomarker levels reported above the upper limit of quantification to ensure they were not overly influencing findings (removed N = 2 for analysis involving Aβ₄₂/Aβ₄₀ and N = 1 for TTau). We also repeated models excluding participants below the age of 30 (N = 18).

Aβ₄₂/Aβ₄₀

Baseline effects and effects on rates of change in Aβ₄₂/Aβ₄₀ by lifestyle and health factors are displayed separately for women and men in Table 2 and Supplementary Figures 1–3.

In men, higher education was associated with higher Aβ₄₂/Aβ₄₀ at baseline. Those with 16 or more years of education had a 6.07 (95% confidence interval (CI) 2.75,9.39) higher Aβ₄₂/Aβ₄₀ ratio and those with 13–15 years had a 4.54 (95% CI 1.19,7.89) higher Aβ₄₂/Aβ₄₀ ratio compared to those with 12 years or less. Those with more education also tended to have a faster decrease in Aβ₄₂/Aβ₄₀ ratio over time. Individuals with 16 years of education or more decreased –0.49 faster per year (95% CI –0.91,–0.07). Similar patterns for baseline effects and effects of education on the rate of change in Aβ₄₂/Aβ₄₀ were observed in women. There were no considerable differences by alcohol consumption, exercise, or BMI. Current smokers tended to have lower Aβ₄₂/Aβ₄₀ (–4.21, 95% CI –8.83,0.40 in women; –3.36, 95% CI –7.04,0.32 in men), although these differences were not statistically significant.

At baseline, the mean Aβ₄₂/Aβ₄₀ was lower (–4.45, 95% CI –8.11,–0.79) in women with hypertension as compared to women without hypertension. However, women with hypertension had a slower Aβ₄₂/Aβ₄₀ decline per year as compared to normotensive women (0.47 per year, 95% CI 0.09,0.85). Men with carotid plaque had lower mean baseline Aβ₄₂/Aβ₄₀ (–3.24, 95% CI –6.36,–0.13). Women with higher baseline inflammatory marker levels (IL-6, sICAM-1, sVCAM-1) had lower mean baseline Aβ₄₂/Aβ₄₀ and a less steep rate of decline in Aβ₄₂/Aβ₄₀ over time. In men, these patterns were only significant for baseline sICAM-1.

TTau

At baseline, having 16 or more years of education was associated with mean baseline TTau levels that were 9.08% lower (95% CI –17.04,–0.35) than TTau in those with 12 years or less education. There were no associations of smoking, alcohol consumption, or exercise with TTau but mean TTau levels increased 1.32% faster per year (95% CI 0.14,2.52) in obese participants compared to individuals with a normal weight (Table 3 and Supplementary Figure 4). Participants with diabetes had baseline TTau levels 14.96% higher (95% CI 3.41,36.82) than participants without diabetes. (Table 3 and Supplementary Figure 5).

Individuals with higher non-HDL levels had lower TTau –2.00 (95% CI –3.86,–0.10). Higher levels of hsCRP were associated with lower baseline TTau levels (–1.43%, 95% CI –2.23,–0.62), whereas higher sICAM-1 and sVCAM-1 were associated with slightly higher baseline TTau levels (0.06%, 95% CI –0.001,0.11; 0.03%, 95% CI 0.01,0.05, respectively) (Table 3 and Supplementary Figure 6).

NfL

More years of education were associated with a slower increase in NfL over time. NfL levels increased 0.47% (95% CI –0.95,0.01) more slowly per year in people with 13–15 years of education and 0.64% (95% CI –1.12,–0.16) more slowly per year in people with≥16 years of education as compared to those who had up to 12 years of education (Table 3 and Supplementary Figure 7). NfL levels increased 1% faster (95% CI 0.35,1.75) per year for those who did not consume alcohol (versus light consumers) and were lower at baseline in individuals with heavy alcohol consumption (>140 g, –6.81, 95% CI –12.23,–1.04). There was also a faster increase in NfL over time in individuals who did not exercise (0.51% per year, 95% CI 0.90,0.12) as compared to those participants who reported exercising regularly, and in obese individuals as compared to normal weight (0.87% faster per year, 95% CI 0.36,1.38). Obese and overweight individuals had lower baseline NfL levels.

There was a faster increase in NfL levels over time for individuals with hypertension (1.43% per year, 95% CI 1.03,1.83) and for those with carotid plaques (0.72% per year, 95% CI 0.25,1.19). Individuals with diabetes had markedly elevated levels of NfL at baseline (11.20%, 95% CI 2.23,20.97) and a 2.71% faster (95% CI 1.79,3.64) increase in NfL over time (Table 3 and Supplementary Figure 8).

Individuals with higher non-HDL levels had lower NfL –1.09 (95% CI –1.99,–0.17). Higher levels of inflammatory markers, hsCRP, IL-6, sICAM-1, and sVCAM were also associated with faster increases in NfL levels over time (Table 3 and Supplementary Figure 9).

Sensitivity analyses

Excluding participants with biomarker levels reported above the upper limit of quantification and those below the age of 30 did not change any of the results.

Aβ₄₂/Aβ₄₀

Since the accumulation of proteinopathies of amyloid is one of the hallmark pathological changes in AD [3], blood-based levels of Aβ₄₂/Aβ₄₀ have been widely studied and been found to be associated with changes in cognitive function and the development of AD and dementia [45–47] (for meta-analysis see [48]) and with brain amyloidosis [6, 50]. While their research questions often had a different focus, some existing studies reported the associations of education, health-related behavioral factors, health history, and blood health measures with Aβ₄₂/Aβ₄₀ cross-sectionally [40, 51–55]. The long-term effect of midlife lifestyle and health factors on changes in Aβ₄₂/Aβ₄₀ has not yet been characterized. Particularly, the study of associations of midlife inflammation, measured with blood-based markers, with change in Aβ₄₂/Aβ₄₀ is novel.

Education is known for its protective effect on cognition and brain health [13–15]. Correspondingly, we found that, in men, higher education at baseline was associated with less Aβ pathology in blood. However, those with more education had a faster decrease in Aβ₄₂/Aβ₄₀, which would reflect a faster accumulation of Aβ. The pattern was comparable in women, yet, not reaching significance. Few previous cross-sectional studies reported on Aβ₄₂/Aβ₄₀ level differences with regard to levels of education in their cohorts. Two larger cohort studies indicated no difference [51, 55] and one other study found no difference in the entire sample, but a higher ratio with more years of education in the cognitively unimpaired participants only [52]. Longitudinal studies in AD research have shown similar patterns, i.e., faster cognitive decline in people with higher education after AD diagnosis [56] and sharper decline in those with higher education in the 4 years prior to their diagnosis of AD [57]. The exact mechanisms for how education could serve as a protective factor in cognitive aging are not yet understood and likely complex. Scale characteristics in longitudinal studies need to be taken into account as well: People with a higher level of Aβ₄₂/Aβ₄₀ to start with (i.e., those with higher education) will, from a measurement point of view, have a greater possibility to show larger declines.

Current smokers had higher levels of Aβ pathology at baseline compared to never or past smokers. These group differences were comparably large in effect size but only borderline significant. Midlife smoking has previously been identified as a risk factor for dementia and cognitive decline [11, 21]. One possible mechanism for smoking to act negatively on the brain is through its harmful effects on the cardiovascular system [58].

Moreover, other risk factors of vascular health and metabolic syndrome have been identified as risk factors of AD, dementia, and cognitive decline [11, 21–23]. Cross-sectionally, two previous studies found no difference in Aβ₄₂/Aβ₄₀ levels in their primarily middle-aged cohorts with regards to hypertension [53, 55]; one study found higher Aβ₄₂/Aβ₄₀ levels in older cognitively unimpaired individuals with hypertension as compared to normotensive individuals [51]. In our longitudinal study, the strengths of associations of different cardiovascular factors with blood-based Aβ pathology varied between men and women. Among women, having hypertension at baseline was associated with a lower mean baseline Aβ₄₂/Aβ₄₀, which reflects more pathology. We also found that individuals with hypertension at baseline had less accumulation of blood-based AD pathology. This unexpected direction of effect of change in Aβ₄₂/Aβ₄₀ might be due to a floor effect, i.e., women with worse health status had a lower starting point, which means less room for decline over time. It could also be that women with more inflammation or hypertension may have ‘aged faster’ and their decline in Aβ₄₂/Aβ₄₀ had plateaued. This finding could also be due to chance. In men, having any carotid plaque was associated with lower Aβ₄₂/Aβ₄₀ at baseline and thus more pathology.

Likewise, increased levels of blood-based inflammatory markers have been shown as risk factors of cognitive decline and dementia [24, 60]. Levels of cell adhesion molecules ICAM and VCAM are elevated in response to pro-inflammatory cytokines [61] and have been associated with atherosclerosis and coronary artery disease [61, 62], which are known risk factors for dementia and AD [22, 64]. However, associations of inflammatory markers with neurogenerative and AD markers in blood have not been extensively studied. In line with the hypothesis of inflammation contributing to cognitive decline and dementia [24, 60], we saw that among women, having higher inflammatory marker levels (IL-6, sICAM-1, sVCAM-1) was associated with more pathology (lower baseline Aβ₄₂/Aβ₄₀). We also found that individuals with higher levels of inflammation (sICAM-1, sVCAM-1) had slower increases of the blood-based AD pathology measured over ten years. Again, this could be due to a floor effect, those with more inflammation could have ‘aged faster’ and reached a Aβ₄₂/Aβ₄₀ plateau, or due to chance. In men, higher levels of ICAM were also associated with more Aβ₄₂/Aβ₄₀ pathology.

Sex differences in dementia and cognition have been widely acknowledged [2, 65–67]. Women have a higher likelihood for developing dementia than men, particularly at older ages [2]. Underlying reasons for these differences are still under debate: Selective survival among men, and/or biological differences related to hormonal differences, menopausal changes, and vascular and metabolic or brain structure differences [65, 69]. Such sex-specific pathological mechanisms or trajectories of AD and neurodegeneration might also be reflected in the differences between women and men that we observed in Aβ₄₂/Aβ₄₀ biomarker levels and the effect of risk factors on Aβ₄₂/Aβ₄₀.

TTau

TTau is another marker that has been studied in association with neurodegenerative diseases and has been associated with cognitive decline [70]. Results on its association with AD pathology are inconclusive [47, 70–74]. It has thus been suggested that TTau may be a prognostic marker of neuronal injury and level of neurodegeneration not specific to AD [10, 70].

Some previous studies with other research foci in smaller samples reported on potential cross-sectional differences in TTau levels based on education, but did not identify significant differences [51, 52]. In our study of > 1,500 participants, we found that more years of education at baseline were associated with lower baseline levels of tau protein in the blood, which is in line with the cognitive reserve hypothesis [13–15].

We also found associations of lifestyle and health factors of metabolic syndrome and inflammation with TTau. Longitudinally, those who were obese had a faster tau level increase over time. There was a strong association of diabetes with TTau at baseline: diabetic individuals had 15% higher levels of TTau. Two other studies also report on the detrimental effects of diabetes on levels of TTau [51, 53]. This is consistent with extensive literature showing associations of diabetes with cognitive decline, dementia, and AD [11, 21] and with reduced cortical thickness [75, 76]. Diabetes may exert its impact on neurodegeneration and increased risk for brain aging and cognitive disorders as a direct result of hyperglycemia or through neuroinflammation [77]. Also, other associated comorbidities of hypertension, dyslipidemia, and hyperinsulinemia may play a role [77].

Regarding blood cholesterol levels, we saw that individuals with higher non-HDL levels had lower TTau. Another study also found inconclusive results; they showed that both higher HDL and higher total cholesterol were associated with lower levels of plasma TTau [53]. This area has been understudied and whether there are biological explanations needs to be determined.

In our study, associations between inflammation markers and blood-based TTau varied between the different inflammatory markers. Individuals with higher levels of ICAM and VCAM had higher TTau at baseline, which is consistent with previous literature suggesting that inflammation may play an important role for neurodegeneration and dementia. There was also an association of higher hsCRP with a faster increase in TTau over time. Unexpectedly, higher hsCRP was associated with lower TTau levels at baseline. As compared to other markers of inflammation, CRP has not shown a consistent pattern of increase with increasing age [78]. Furthermore, longitudinal associations of CRP and dementia and cognitive impairment have been inconsistent in the literature [24, 80]. As compared to ICAM and VCAM, CRP can be considered an indicator of an acute infection and is an important first-line host defense molecule [81]. It may thus not be as reflective of long-term inflammation processes, which are thought to be involved in AD and neurodegeneration [82] as compared to ICAM and VCAM.

NfL

In neurological aging research, NfL has been studied increasingly. NfL is an axonal protein that is released into the brain interstitial fluid after neuronal or axonal injury [83]. Importantly, NFL levels are also elevated in the blood of individuals with mild cognitive impairment, dementia, AD, and other neurodegenerative diseases [9, 84], have been associated with brain pathology [84–86], and with cognitive decline in clinical and healthy cohorts [46, 86].

Recently, researchers started to investigate associations of lifestyle and health-related risk factors and NfL blood levels but most research is cross-sectional [9, 87]. Longitudinal research is sparse [88].

Consistent with the idea of cognitive reserve, in our study, we found that education was associated with a less steep increase in blood-based NfL over time, which is a novel finding. Not many studies have been reporting results on associations of NfL with education and cross-sectionally there has been no evidence for a difference with level of education [9, 89].

Regarding other lifestyle factors, we found associations of NfL with alcohol consumption and exercise. The association of neurodegeneration with alcohol consumption in human participants is likely complicated and not linear. While heavy drinking and alcohol use disorders are considered risk factors for dementia [90–92] and might influence an earlier development through neurotoxicity, thiamine deficiency, other conditions, and vascular pathways [91], there are also studies reporting neuroprotective effects of smaller doses of alcohol intake [93], and show an increased risk in people who are abstinent [94]. A longitudinal study of drinking and cognitive performance showed a nonlinear association between drinking patterns and later life cognitive performance, where nondrinkers and heavy drinkers had worst cognitive results [95]. In our study, NfL levels increased faster in those who did not consume alcohol (versus light consumers). Moreover, levels were lower at baseline in individuals with heavy alcohol consumption. Previous cross-sectional smaller cohorts reported no differences of NfL with regard to alcohol usage [36, 37]. Our negative results in individuals who are not drinkers might be because such individuals might have comorbidities and other underlying diseases or be taking medications, that may not allow them to drink. We found an accelerated neurodegeneration, i.e., faster increase in NfL over time, in individuals who did not exercise as compared to those participants who reported exercising regularly. This is consistent with existing literature suggesting protective effects of exercise for brain health and cognition [11, 20], for which pathways of improvements through cardiovascular and metabolic system, neurogenesis, and reduction of stress and inflammation have been discussed [96].

In obese individuals as compared to normal weight, there was a faster increase in NfL over time, while obese and overweight individuals had lower baseline NfL levels. A similar pattern of longitudinal associations has been shown in a previous study [88]. More studies have been conducted cross-sectionally and found similar results, with higher BMI being associated with lower levels of NfL [36, 51]. It has been suggested that BMI and blood volume might influence the measurement of NfL in blood [36–40]. Thus, we adjusted for BMI in all our models [39].

Consistent with the literature on associations between cardiovascular and metabolic function and health with neurodegeneration, dementia, and cognition [11, 23], in our study, there were faster increases in NfL levels over time for individuals with hypertension, carotid plaque, and with diabetes. Moreover, individuals with diabetes had a high estimated baseline risk for elevated NfL. While two previous cross-sectional studies in older adults failed to find a difference with diabetes [36, 51], others also showed that participants with a diagnosis of diabetes had significantly higher NfL levels compared to those without diabetes [9, 37]. Furthermore, our results are in line with one longitudinal study that found an accelerated increase in NfL over time in people with higher hemoglobin A1 C levels [88] and in those with higher blood pressures and higher resting heart rates in similar models [88].

Unexpectedly, individuals with higher non-HDL levels had lower NfL. Cross-sectional results from other studies are similarly inconclusive: One study did not find an association of NfL with total cholesterol or HDL [53]. Others identified higher HDL and higher NfL [37, 53], while also seeing higher NfL in individuals with hypercholesterolemia [37]. Two studies show higher NfL in people with dyslipidemia [9, 51]. In similar analyses from one longitudinal study, higher total cholesterol was associated with NfL increase over time, while higher HDL was associated with higher baseline NfL and not with increase over time [88].

In our study, we found associations of higher levels of multiple inflammatory markers (hsCRP, IL-6, sICAM-1, and sVCAM) with faster increase in NfL levels over time. This is in line with one longitudinal study (average follow-up 7.7 years), which showed that higher hsCRP was associated with NfL increase over time [88]. We expand upon this work with the investigation of additional markers of inflammation. These changes were significant yet rather small in effect size. While short-term inflammation is an immune response which benefits health, chronically elevated levels of circulating pro-inflammatory markers observed with aging are associated with symptoms of chronic disease. Such age-related symptoms develop progressively and over a longer time frame [82] and longer follow-up may be needed to identify stronger effects.

Limitations and strengths

In our study, we only had stored serum from all three time points. Serum-based measures have been used less frequently in research and concentrations of Aβ₄₂, Aβ₄₀, and TTau have been reported to be lower in serum than plasma when measured by SIMOA [27, 28]. However, in our sample, we have shown the correlation of biomarker concentrations in serum and plasma and evaluated the ability of these serum biomarkers to differentiate between people with and without cerebrospinal fluid and/or brain neuropathology [26]. Kidney function has been associated with blood-based biomarkers as well [51] Unfortunately, we did not have any measures of kidney function available in this cohort. Recent reports suggest ethnoracial differences in biomarkers levels [9, 97]. Our study cohort is predominantly non-Hispanic White, which may thus limit our ability to generalize the findings to other populations. while we aimed to assess determinants of early changes in neurodegenerative biomarkers in midlife, the biomarker level changes we observed over time were rather small. Particularly, levels of TTau were low in this younger population and showed little change over time which may have limited our power to detect relevant predictors [26]. Future studies with longer follow-up are needed to determine how these emerging biomarkers change together with lifestyle and health-related risk factors from midlife to older age, how they predict cognitive changes and the onset of (a clinical diagnosis of) dementia in later life and what clinically relevant biomarker cut-offs are.

The strengths of our study are the large, well-characterized general population cohort. We were utilizing three repeated measures covering 10 years of follow-up and starting in midlife, when future intervention and treatment methods should be more efficacious. We were able to investigate a variety of inflammation, vascular, and metabolic health and lifestyle factors using objective physical examination data and standardized lab procedures. Preanalytical handling and storage of blood samples across all phases was in accordance with currently recommended protocols for measuring Aβ₄₀, Aβ₄₂, TTau, and NfL in blood [27, 28].

Conclusion

In conclusion, various risk factors of lifestyle and health, including education and inflammation, were associated with longitudinal Aβ₄₂/Aβ₄₀, TTau, and NfL levels in midlife. If confirmed, these findings could have important implications for developing early intervention strategies and might inform policy decisions about education and health care services. Particularly a healthy lifestyle (e.g., decreasing smoking, increasing physical activity) and medical care of cardiovascular, metabolic, or inflammatory conditions, if applied early, might have potential to slow processes of neurodegeneration or delay the onset of AD.