Modifiable dementia risk associated with smaller white matter volume and altered 1/f aperiodic brain activity: cross-sectional insights from the LEISURE study Authors: Thomas Pace, Jacob M Levenstein, Toomas E Anijärv, Alicia J Campbell, Ciara Treacy, Daniel F Hermens, Sophie C Andrews Published: 2024 ABSTRACT AbstractBackgroundThe rising prevalence of dementia necessitates identifying early neurobiological markers of dementia risk. Reduced cerebral white matter volume and flattening of the slope of the electrophysiological 1/f spectral power distribution provide neurobiological markers of brain ageing alongside cognitive decline. However, their association with modifiable dementia risk remains to be understood.MethodsA cross-sectional sample of 98 healthy older adults (79 females, mean age = 65.44) underwent structural magnetic resonance imaging (sMRI), resting-state electroencephalography (EEG), cognitive assessments and dementia risk scoring using the CogDrisk framework. Univariate and multivariate linear regression models were conducted to investigate the relationships between modifiable dementia risk and sMRI brain volumes, the exponent of EEG 1/f spectral power, and cognition, whilst controlling for non-modifiable factors.ResultsSmaller global white matter volume (F(1,87) = 6.884, R2 = 0.073, P = .010), and not grey (F(1,87) = 0.540, R2 = 0.006, P = .468) or ventricle volume (F(1,87) = 0.087, R2 = 0.001, P = .769), was associated with higher modifiable dementia risk. A lower exponent, reflecting a flatter 1/f spectral power distribution, was associated with higher dementia risk at frontal (F(1,92) = 4.096, R2 = 0.043, P = .046) but not temporal regions. No significant associations were found between cognitive performance and dementia risk. In multivariate analyses, both white matter volume and the exponent of the 1/f spectral power distribution independently associated with dementia risk.ConclusionsStructural and functional neurobiological markers of early brain ageing, but not cognitive function, are independently associated with modifiable dementia risk in healthy older adults. Key Points Modifiable dementia risk is associated with global white matter volume, even after adjustment for non-modifiable factors including demographics and intracranial volume. Modifiable dementia risk is associated with global white matter volume, even after adjustment for non-modifiable factors including demographics and intracranial volume. Higher modifiable dementia risk is associated with a flatter slope of 1/f spectral power distribution at frontal electrodes. Higher modifiable dementia risk is associated with a flatter slope of 1/f spectral power distribution at frontal electrodes. Modifiable dementia risk was not associated with cognitive function. Modifiable dementia risk was not associated with cognitive function. Introduction Identifying both modifiable and non-modifiable dementia risk factors that predict dementia is necessary for early interventions and for monitoring cognitive decline and healthy brain ageing. Modifiable dementia risk factors encompass potentially alterable aspects of an individual’s life, such as dietary habits, sleep patterns, and physical activity levels. In contrast, non-modifiable risk factors include inherent characteristics that cannot be changed, such as genetic predisposition and age. Consequently, specialised dementia risk tools have been created that estimate an individual’s lifetime dementia risk based on a combination of factors [1, 2]. Notably, 14 modifiable risk factors alone account for ~45% of dementia cases worldwide [3]. One such tool is the Cardiovascular Risk Factors, Aging and Incidence of Dementia (CAIDE) [2], which has shown to predict dementia, particularly Alzheimer’s disease, up to 20 years pre-diagnosis [2, 4]. Similarly, the Lifestyle For Brain Health (LIBRA) covers a range of health and lifestyle factors, recently expanded to include risks such as hearing impairment, life course inequalities, social contact, atrial fibrillation and sleep [5, 6], and has been validated across 21 cohorts [7]. Likewise, the Cognitive Health and Dementia Risk Assessment (CogDrisk) tool includes lifestyle factors such as sleep, diet, depression, physical activity and more, and also incorporates age- and sex-specific risk scoring [8]. CogDrisk integrates empirically substantiated risk and protective factors for dementia and has achieved external validation across four heterogeneous cohort studies [9]. Notably, comparative studies across multiple cohorts with follow-up periods spanning 7–22 years demonstrated CogDrisk’s comparable accuracy in predicting dementia onset compared to these established risk scores [10]. Overall, CogDrisk’s predictive capability may facilitate the identification of early neurobiological markers of brain ageing and has not yet been used to investigate the effects of dementia risk on brain health. One potential type of neurobiological marker of risk is MRI-derived structural brain volume. The total volume of cortical grey and cerebral white matter, and lateral ventricles, are well-documented indicators of age-related brain changes [11–13]. Whilst gross-brain volumetric changes across the human lifespan are consistent with normal ageing, brain volumes also relate to cognitive decline and risk of dementia [11, 14, 15]. For example, individuals with elevated midlife CAIDE scores showed greater white matter hyperintensities 2–3 decades later [16], whilst high CAIDE scores have been associated with smaller grey matter and hippocampal volume [17, 18]. Similar cross-sectional findings have been found using LIBRA, with higher risk scores associated with larger volumes of white matter hyperintensities and, specifically in men, lower grey matter and higher ventricular volume [19]. Importantly, prior research has shown that brain structures are impacted by individual modifiable factors including physical activity [20, 21], diet [22, 23] and sleep [24], making tools like CogDrisk particularly suited for investigating associations between modifiable risk and early markers of brain ageing. In addition to structural markers, the ‘exponent’ derived from the aperiodic activity of the electroencephalography (EEG) power spectrum has emerged as a promising functional indicator of brain ageing. Whilst periodic EEG components represent oscillatory neural activity at specific frequencies (e.g. alpha, gamma), the aperiodic component reflects background neural activity across a broad frequency range [25]. This non-oscillatory broadband activity follows a 1/f power-law distribution [26, 27], where power systematically declines as frequency increases. This pattern is believed to represent the brain’s excitation–inhibition (E/I) balance [28, 29], providing insights into the functional state of neuronal circuits. The ‘exponent’ quantifies the slope of this distribution, with a lower value representing a flatter slope (i.e. slower decline) and implying a shift towards asynchronous excitation disrupting the E/I balance in the brain [28, 30, 31], although this interpretation remains contested [32, 33]. Significantly, the exponent decreases with age in older adults [34], with lower values correlating with cognitive decline [35], whilst higher values associate with better performance across cognitive domains [36–39]. Consequently, the exponent may serve as a promising EEG biomarker for early brain ageing and dementia risk. This study thus explored the link between dementia risk scores, structural brain volumes, functional EEG activity, and cognitive function in healthy older adults. We hypothesised that higher modifiable risk would be associated with structural and functional brain characteristics indicative of advanced neurobiological ageing, specifically smaller white and grey matter volumes, larger ventricle cerebrospinal fluid volumes and lower EEG exponents. These hypotheses were grounded in the premise that dementia risk factors may drive accelerated neurobiological ageing processes, affecting both brain structure and function. Methods Study population The cross-sectional sample for the current analysis was drawn from individuals who successfully completed the baseline evaluation of the Lifestyle Intervention Study for Dementia Risk Reduction (LEISURE) [40]. In brief, the LEISURE study protocol included MRI, EEG, cognitive test batteries, self-reported psychological measures, actigraphy monitoring and physical fitness evaluations (see the full protocol for additional details [40]). For this investigation, a total of 98 independent community-dwelling healthy older adults aged 50–84 (females = 79, mean age = 65.44, SD ± 8.44) from the Sunshine Coast (Queensland, Australia) were included. Exclusion criteria included diagnosis of mild cognitive impairment or dementia, traumatic brain injury, stroke, cardiovascular disease, metabolic disorders, diagnosis of serious psychiatric or neurological disorder, and epilepsy. Dementia risk assessment Dementia risk scores were calculated post hoc using baseline data from participants, following a methodology adapted to the CogDrisk framework [8]. Risk scores were derived from a range of data sources including demographics (age, sex, education), medical history (diabetes, stroke, traumatic brain injury), health assessments (body mass index, hypertension, cholesterol levels, atrial fibrillation, insomnia), psychological factors (depression), lifestyle factors (smoking history, physical activity, cognitive engagement, social engagement) and dietary patterns (fish consumption). Individual risk and protective factors were scored according to CogDrisk protocol and aggregated into an overall risk score, with higher scores indicating greater dementia risk (see Appendix 1—Supplementary Table 1). Table 1Values are means (SD). RVPA, rapid visual processing accuracy; WM, white matter; GM, grey matter; EEG, electroencephalography. *The ‘offset’ reflects the broadband shift of power across all frequencies in log(μV2) units and is linked to neuronal population spiking. The ‘exponent’ is typically unitless and describes the rate (i.e. slope) at which this power decays with increasing frequency in log–log space [25]. A higher exponent indicates a steeper decline, which is thought to reflect greater inhibition relative to excitation in neural activity, whilst a lower exponent suggests a flatter decline, potentially indicating more asynchronous, noisy neural activity.CharacteristicNumber of ParticipantsMean (standard deviation)Age9865.44 (8.44)Sex9879 females19 malesEducation9814.89 (3.06)CogDrisk982.04 (5.82)Speed98391.45 (41.84)Executive Composite9853.67 (1.62)Spatial Working Memory988.19 (3.68)Episodic9819.68 (14.69)Attention RVPA980.90 (0.04)Total WM volume89488091.50 (5521.96)Total GM volume89529729.27 (4605.25)Total ventricle volume8940241.68 (1633.86)EEG Frontal Exponent*940.95 (0.22)EEG Temporal Exponent*940.90 (0.03)EEG Frontal Offset*940.37 (0.27)EEG Temporal Offset*940.62 (0.03) Values are means (SD). RVPA, rapid visual processing accuracy; WM, white matter; GM, grey matter; EEG, electroencephalography. *The ‘offset’ reflects the broadband shift of power across all frequencies in log(μV2) units and is linked to neuronal population spiking. The ‘exponent’ is typically unitless and describes the rate (i.e. slope) at which this power decays with increasing frequency in log–log space [25]. A higher exponent indicates a steeper decline, which is thought to reflect greater inhibition relative to excitation in neural activity, whilst a lower exponent suggests a flatter decline, potentially indicating more asynchronous, noisy neural activity. We categorised dementia risk factors as non-modifiable (age, sex, education, history of diabetes, stroke and traumatic brain injury) or modifiable (high cholesterol, hypertension, atrial fibrillation, insomnia, depression, physical inactivity, cognitive engagement, social engagement, diet and smoking). To isolate the effects of modifiable factors, we excluded participants with non-modifiable risk factors (traumatic brain injury, diabetes and stroke) and adjusted for age, sex and education in our linear regression models. This approach enabled a focused examination of the impact of modifiable risk factors of dementia on pre-symptomatic healthy older adults. Neurobiological measures For full details, see Appendix 1—Neurobiological Measures. Briefly, all eligible participants underwent a structural MRI brain scan using a 3-T Siemens Skyra (Erlangen, Germany) with a 64-channel head and neck coil. Structural MRI images were pre-processed and analysed using in-house scripts, FSL [41], and a modified version of SIENAX [42] to obtain brain volumes. Resting-state EEG was recorded using a 32-channel BioSemi system (BioSemi Active-Two; V.O.F., Amsterdam, Netherlands), at 1024 Hz, with aperiodic parameters extracted using custom scripts [43] and the ‘specparam’ algorithm [25]. Cognitive measures Cognitive functions were assessed using the CANTAB [44], a comprehensive, computerised battery which assessed processing speed, attention, episodic memory and spatial working memory. Executive function was assessed through a composite score derived from traditional neuropsychological assessments of Verbal Fluency (DKEFS-FAS) [45], Semantic Fluency (DKEFS-Category) [45] and Response Inhibition (Colour-Word Interference (Stroop) task) [46]. Statistical analysis Multivariate linear regression models were first used to regress out demographic factors (age, sex and education) in relation to MRI volumes, EEG measures, cognitive function metrics and dementia risk scores. Then, univariate linear regression models were constructed using the standardised residuals from these initial models, with dementia risk score residuals as predictors and neuroimaging/cognitive measure residuals as dependent variables (see Appendix 1—Statistical Analysis). Results Sample characteristics All participants completed the self-report questionnaires and assessments that accounted for the CogDrisk measure, the cognitive assessment and the CANTAB cognitive tests battery (n = 98). Of the participants who completed all assessments needed for generating a dementia risk score, 89 also completed MRI brain scans (mean age = 65.54, female = 82%), 94 completed the EEG resting-state task (mean age = 65.20, female = 81.9%), and 86 completed both the MRI and EEG (mean age = 65.2, female = 83.7%). Associations of dementia risk with MRI structural volumes After adjusting for age, sex and education in relation to both risk score and MRI volumes, and additionally adjusting for intracranial volume only for MRI volumes, univariate linear regression models showed a significant negative association between dementia risk and white matter volume (Figure 1A, overall model: F(1,87) = 6.884, R2 = 0.073, P = .010; predictor: dementia risk, β = −0.271, t(87) = −2.624, P = .010). However, no significant associations were found between dementia risk and grey matter volume (overall model: F(1,87) = 0.540, R2 = 0.006, P = .468; predictor: dementia risk, β = −078, t(87) = −0.729, P = .468) or ventricle cerebrospinal fluid volumes (overall model: F(1,87) = 0.087, R2 = 0.001, P = .769; predictor: dementia risk, β = −0.032, t(87) = −0.295, P = .769). Associations between dementia risk and biomarkers of brain ageing. (A) Scatter plot of dementia risk against standardised white matter volume residuals, adjusting for modifiable factors. The scatter plot illustrates individual data points with a superimposed linear regression line (β = −0.271, P = .010) and a shaded 95% confidence interval, indicating a significant inverse association between dementia risk and white matter volume. (B) Scatter plot of standardised residuals for dementia risk against EEG frontal electrode exponents, adjusted for modifiable factors. The scatter plot illustrates individual data points with a superimposed linear regression line (β = −206, P = .046) and a shaded 95% confidence interval, indicating a significant inverse association between dementia risk and the exponent. Inset: topographical map of the EEG electrode configuration highlighting the frontal regions of interest in orange, denoting the electrode sites used to calculate the EEG frontal exponent. Associations between dementia risk and resting-state EEG 1/f aperiodic slope Figure 1B reflects the results of the regression model assessing the relationship between dementia risk (CogDrisk) and the 1/f aperiodic exponent from resting-state EEG, averaged over frontal electrodes, whilst adjusting for age, sex, years of education and aperiodic offset similar to prior research [47, 48]. The analysis indicated a significant negative association between dementia risk and the frontal EEG exponent (Figure 1B, overall model: F(1,92) = 4.096, R2 = 0.043, P = .046; predictor: dementia risk, β = −0.206, t(87) = −2.024, P = .046). Conversely, dementia risk did not demonstrate a significant association with the exponent derived from temporal electrodes (overall model: F(1,92) = 2.608, R2 = 0.028, P = .110; predictor: dementia risk, β = −1.66, t(87) = −1.615, P = .110), pointing to a possible EEG region-specific relationship with dementia risk. To test regional specificity, a paired samples t-test compared frontal and temporal exponents, revealing a significantly larger frontal exponent (M = 0.050, SD = 0.187) than temporal (t(93) = 2.617, P = .010, 95% confidence interval [0.012, 0.089]). Associations between dementia risk and cognition We next sought to investigate the relationship between dementia risk and cognitive performance [49]. Table 2 reflects the results of associations between CogDrisk and cognitive tests, controlling for demographics. We observed no significant associations between dementia risk and several key cognitive domains, including response speed (P = 0.870), executive function (P = 0.278), spatial working memory (P = 0.300), episodic memory (P = 0.877) and attention (P = 0.681). Table 2Relationships of dementia risk with cognitive functionsVariable n Standardised coefficients B t P-valueResponse speed98−.017−0.164.870Executive function98−.111−1.091.278Spatial working memory98.1061.041.300Episodic memory98−.016−0.156.877Attention RVP98−.042−0.412.681 Relationships of dementia risk with cognitive functions Multimodal analysis of the associations of brain structure and function with dementia risk We used multivariate regression to assess the independence of associations between significant brain ageing markers (white matter, frontal exponent), controlling for demographics and cognition (see Appendix 1—Statistical Analysis). The overall model was significant (Figure 2, R2 = 0.171, adjusted R2 = 0.140, F(3, 82) = 5.627, P = .001), with both white matter volume (Figure 2, blue line, β = −0.361, t(82) = −3.489, P < .001) and the frontal exponent (Figure 2, yellow line, β = −0.279, t(82) = −2.678, P = .009) significant negative independent predictors of dementia risk. The global cognition composite (β = −0.027, t(82) = −0.267, P = .790) did not significantly predict dementia risk. Partial residual plots for structural and functional predictors of dementia risk. This plot visualises the independent associations of brain structure (white matter volume) and function (frontal 1/f exponent) to dementia risk, controlling for age, sex and education, and holding each predictor constant. Blue dots represent white matter volume, and orange dots represent frontal 1/f exponent in relation to dementia risk. Each circle represents an individual participant (n = 86). Lines show the best linear fit for each predictor. The x-axis displays standardised predictor values (white matter volumes and EEG exponents), whilst the y-axis shows the residual values for dementia risk. Discussion We examined associations between dementia risk and markers of neurobiological ageing and cognitive function, controlling for non-modifiable health and demographic factors, to elucidate the relationship between modifiable risk factors and early brain ageing using baseline data from the LEISURE study [40]. Univariate regression analyses revealed inverse associations between modifiable dementia risk and both structural (white matter volumes) and functional (frontal EEG 1/f exponents) brain health measures. A unified multivariate regression model incorporating these measures along with global cognitive performance demonstrated that white matter volume (β = −0.361, P < .001) and frontal exponents (β = −0.279, P = .009) independently predict modifiable dementia risk. Despite a supplementary analysis showing a small mediation effect of white matter on dementia risk and frontal exponents (see Appendix 2), the strong independent relationships in the multivariate regression model suggest that each brain ageing marker’s relationship to dementia risk remained largely independent. This multidimensional approach reflects the complementary roles of brain structure and function in cognitive ageing, suggesting distinct pathways of vulnerability in early, pre-symptomatic stages of brain ageing among healthy older adults. The association between reduced white matter volume and ageing has been consistently demonstrated in previous literature [11–15, 50], including in our sample (see Appendix 2). Although this relationship is primarily linear [51], some evidence indicates an acceleration in white matter volume loss during advanced ageing [52]. This decline may arise from the loss of small-diameter myelinated fibres [53], particularly evident in those >59 years old [54], leading to cognitive decline [55]. Furthermore, recent research has suggested that larger white matter volume might mitigate the effect of age-related accumulation of white matter lesions on cognitive function [56], whilst a negative association has been identified between white matter volume and cognitive decline [57, 58], and dementia [59]. Recently, a study employing deep learning models showed that white matter volume was a stronger predictor of brain ageing than either grey matter or ventricle volumes [60]. Similarly, our study demonstrated a relationship only between dementia risk and white matter volume. Previous research has shown that white matter volume loss precedes grey matter atrophy [61], can occur in the absence of grey matter decline [54] and declines relatively more sharply than grey matter in advanced age (~90 years old) [62]. Therefore, white matter may be inherently more vulnerable to early structural brain changes associated with brain ageing, potentially worsening due to lifestyles with increased modifiable dementia risk factors. Previous studies have demonstrated that elevated CAIDE scores are primarily linked to reductions in grey matter volume, especially among populations with high risk [17, 18]. One study did report a relationship between CAIDE scores and global white matter hyperintensities; however, this association became statistically nonsignificant after age adjustments [63]. In contrast, our LEISURE study examined a low-risk cohort and found an age-independent negative association between dementia risk (using CogDrisk) and white matter volume, but not grey matter volume. This deviation suggests that white matter volume may emerge as a more effective biomarker for assessing dementia risk particularly in low-risk cohorts. Importantly, previous research has shown that white matter volume is significantly affected by lifestyle behaviours, making it a great target for dementia risk interventions. Studies have shown that behaviours such as alcohol use [64], sedentariness [20] and shorter sleep time [24] can lead to decreased white matter volumes. Alternatively, learning new skills [65], physical activity [21, 66], cognitive engagement [67, 68] and adherence to the Mediterranean diet [22] have been shown to act as protective factors for white matter decline. Improving these modifiable lifestyle factors could delay the ageing of brain structures by 2.0 to 5.8 years [69], with prior research indicating that optimal risk reduction should occur before age 65 [70]. Notably, the CogDrisk includes multiple of these factors in its scoring, and our results show a significant negative link between modifiable risk and white matter volume, highlighting the impact of lifestyle on brain ageing and the need for early intervention. One such intervention trial was the Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER), which pioneered a multimodal approach to reduce dementia risk [71]. Whilst the intervention improved cognition in at-risk older adults (CAIDE ≥6) [72] and reduced hippocampal atrophy in those with lowered dementia risk scores [73], it did not affect regional brain volumes or white matter lesions [74]. Interestingly, the efficacy of the FINGER intervention appeared to be moderated by baseline brain health. Participants with greater cortical thickness, larger hippocampal volumes and lower white matter lesion burden at baseline exhibited more pronounced cognitive benefits [74], whilst higher baseline LIBRA scores were related to less cognitive improvement over time [75]. These findings suggest that lifestyle interventions may be most effective when implemented before the onset of lifestyle risk–related brain changes. Additionally, we observed a significant negative association between dementia risk and our functional brain marker, the frontal EEG aperiodic exponent, aligning with previous research linking decreased exponents to cognitive decline [76], dementia [77] and Alzheimer’s disease [35]. Derived from the slope of the 1/f-like power spectrum, the exponent has largely been interpreted as reflecting neural excitation–inhibition balance [28–30, 34]. However, recent modelling has revealed that the exponent may also result from complex interactions among biophysical parameters beyond E/I ratio including neural leak conductance and synaptic properties [32], whilst studies in mice showed pharmacological induced E/I balance alterations that produced exponent changes which both supported and contradicted theoretical predictions [33]. Conversely, spectroscopy findings indicate that an increase in exponent correlates significantly with higher glutamate–GABA asymmetry, primarily driven by glutamate levels, supporting its role in reflecting E/I dynamics especially related to excitatory processes [78]. Despite these theoretical challenges, the exponent’s clinical relevance remains compelling. Longitudinal studies have shown that changes in the exponent predict cognitive decline [34], particularly in executive functioning [37], and linked to performance in various cognitive tasks, including visuospatial memory [36], processing speed [39] and response strategies in attention tasks [38]. Our result suggest that dementia risk has an inverse relationship with the frontal exponent, possibly reflecting an increasing excitation–inhibition imbalance influenced by modifiable dementia risk factors [34, 35]. The lack of association between dementia risk and cognition was surprising, considering the established relationship between dementia, white matter volume and cognitive decline. Cognitive markers, such as performance on neuropsychological tests, can predict early cognitive decline [79], with subtle changes emerging years before dementia diagnosis [49], and correlating with preclinical dementia biomarkers [80]. In this study, we investigated cognition using both CANTAB and traditional neuropsychological assessments to assess cognitive functions. Importantly, CANTAB has proven effective in detecting cognitive decline across various neurological conditions, including Alzheimer’s and mild cognitive impairment [81], is sensitive to age-related cognitive decline [82–85] and demonstrated predictive power for dementia onset [86–88]. However, despite using CANTAB alongside traditional testing, we found no association between dementia risk and cognitive function. This may be due to our cross-sectional design, which limits detection of subtle cognitive changes, or our relatively young and healthy cohort (mean age = 65), shaped by strict exclusion criteria [40]. Strengths and limitations A strength of this study was its multimodal neurophysiological approach, which utilised both MRI and EEG to explore structural and functional indicators of brain ageing and their relationship to modifiable dementia risk. It also benefited from its focus on a low-risk, healthy population, which makes the findings especially relevant for preventive strategies. However, this focus could also be seen as a limitation as our results may not generalise to populations with higher risk or those with existing health conditions, whilst the cross-sectional design constrains our ability to establish causality. Additionally, the CogDrisk scores used in our analyses were constructed from our existing dataset and not obtained through direct administration of the CogDrisk tool, a methodological compromise that may have introduced variance potentially affecting the interpretability from our post hoc CogDrisk metric and other studies using the full CogDrisk assessment tool. Lastly, our study did not investigate region-specific volumetric relationships, which may be necessary for understanding the relationship between the observed frontal electrode EEG results and dementia risk. Conclusions Overall, our study has demonstrated significant independent relationships between modifiable dementia risk with structural and functional markers of early brain ageing. This was despite the use of strict exclusion criteria and controlling for non-modifiable demographic and health factors, highlighting the critical importance of early intervention, both in younger adults and in populations who have not yet shown signs of cognitive decline. Supplementary Material