We hypothesize that unique neural substrates and EEG activation patterns underlie exceptional gait phenotypes in older adults. To test this, we will compare structural and functional brain measures between Super Movers and usual movers aged ≥80 years using MRI, predicting that Super Movers will exhibit superior indices of brain health (Aim 1.1). We will further examine mobile EEG activation patterns during complex walking tasks, predicting that Super Movers will display distinct cortical signatures relative to usual movers of the same age (Aim 1.2). Finally, we will extend this analysis to all participants aged ≥65 years by comparing MRI and mobile EEG markers between fast movers and usual or slow movers (Aim 1.3).

We hypothesize that the accumulation of dementia-related pathology produces measurable functional and neurophysiological brain changes that, over time, manifest as clinically observable cognitive and gait dysfunction. To test this, differential gait-related EEG responses will be quantified using both categorical (absence of differential cortical activation) and continuous (increased variability in frontal premotor activity) metrics. We will then determine whether these differential gait-related mobile EEG responses at baseline predict longitudinal cognitive decline in adults aged ≥65 years (Aim 2.1). We will further assess whether poor baseline MSI similarly predicts cognitive decline over time (Aim 2.2). Exploratory analyses, stratified by gait status (slow, usual, and fast movers), will be conducted to identify differential neurophysiological patterns that may reflect underlying protective mechanisms. Dual cognitive-motor decline, operationalized as incident MCR syndrome, will serve as a secondary outcome.

We hypothesize that dementia-related brain pathology and accelerated biological aging drive the early clinical emergence of both cognitive and motor dysfunction. This aim addresses an existing gap by examining biomarkers of neurodegeneration and biological aging in relation to gait performance and cognitive trajectory. We will examine whether, among adults aged ≥65 years, slow movers exhibit greater burden of neurodegenerative pathology, characterized by elevated phosphorylated-tau (p-tau) 217, amyloid beta 40 and 42 (Aβ-40 and 42), neurofilament light chain (NfL), as well as accelerated biological aging as indexed by epigenetic clock measures [44] relative to usual movers, with fast movers hypothesized to show the inverse pattern (Aim 3.1). We will further determine whether higher baseline neurodegenerative pathology predicts longitudinal cognitive decline, and whether slow-mover status moderates the strength of this association (Aim 3.2).

Gait speed is a well-validated biomarker of aging [12] and cognitive health and is strongly associated with mobility limitations, fall risk, disability, and dementia [82]. In this study, we will implement both simple and complex dual-task walking paradigms to quantify differential neural activation, defined as complex task performance minus simple task performance. As gait depends on the integration of motor and cognitive processes, increasing task demands may enable early identification of neural vulnerability associated with dementia risk. The WWT test increases cognitive demand during walking and is sensitive to early motor and cognitive decline. Performance on this task has been shown to predict falls, mortality, and dementia [12,101,102]. We will develop quantitative gait-related EEG protocols and select gait speed as the outcome given that it is standardized, reproducible, reliable, and well-tolerated across participants [10,72,103,104].

Gait speed will be measured using two 26-foot straight instrumented walkways with embedded pressure sensors (Zenometrics, LLC) that form a track—where turning area purposefully excludes pressure sensors in order to accurately capture quantitative steady-state gait metrics. Participants will walk around the track at their usual pace during baseline and annual visits. Prior studies demonstrate excellent reliability across repeated assessments and strong predictive validity for mobility limitations, falls, and dementia [71,81,82]. The Zenometrics system captures additional spatiotemporal gait parameters that will be available for in-depth secondary analyses [10].

Gait Classification and Normative Framework

Super Movers will be defined as fast movers aged ≥80 years with a gait speed at least 1.5 SD above sex-adjusted normative cut scores [33]; this threshold corresponds to the typical gait speed of adults approximately 30 years younger. Cut scores will be derived from gait speed measured at a normal, self-selected pace across 2 straight walking segments, providing 26 feet of steady-state gait per segment. Slow movers will be defined as individuals with a gait speed at least 1.5 SD below the calculated mean based on established normative data [104,105].

Participants will complete an established simple reaction time (RT) task consisting of visual, somatosensory, and combined visual-somatosensory (VS) conditions, along with a catch condition in which no stimulus is presented and no response is expected [106]. Stimuli will be presented in random order with equal frequency. Each block includes 15 trials per each of the 4 conditions, and participants will complete 3 blocks for a total of 180 trials. Interstimulus intervals will vary randomly between 1‐3 seconds. Blocks will be separated by 20-second rest periods to reduce fatigue and maintain attention. This task will be conducted while sitting and once while walking on the above-noted quantitative gait mat.

Visual and somatosensory stimuli will be delivered bilaterally for a duration of 100 ms using a custom-built stimulus generator developed by Zenometrics, LLC. Stimulus triggering will be controlled using Matlab code (MATLAB R2025a, MathWorks) on a PC connected to the stimulus generator. Participants will wear headphones delivering continuous white noise to ensure somatosensory stimuli are inaudible. Participants will be instructed to respond to all stimuli as quickly as possible by pressing a pressure sensor located under their fingers.

RT from omitted trials will be set to infinity [107]. RTs outside the range of mean±2.5 SD for each stimulus type will also be set to infinity. Accuracy will be calculated as the number of correct responses divided by 45 trials per condition. Participants with task accuracy below 60% will be excluded from analyses, consistent with prior work [28].

MSI magnitude will be quantified using established cumulative probability models [106]. Robust probability models will compare the combined cumulative distribution functions (CDF) of the unisensory visual and somatosensory RTs (the predicted CDF) capped at min [P(RT_V≤ t)+ P(RTS ≤ t), 1], with the CDF of multisensory VS RTs [P(RT_VS≤ t)] [28]. A violation of the race-model inequality, indicated when the observed multisensory CDF exceeds the predicted CDF, reflects MSI. The predicted CDF will be subtracted from the observed multisensory CDF to generate a CDF difference curve, and the area under the significantly violated portion of this curve will serve as the continuous measure of MSI magnitude.

Continuous EEG data will be recorded using a 64-channel mobile SAGA system (TMSi) at a sampling rate of 512 Hz, with a passband of 0.05 to 100 Hz and a roll-off of 24 dB per octave. Data will be high-pass filtered at 1 Hz, and notch-filtered (60 Hz) to remove line noise. Noisy channels will be identified using automated detection procedures [115] and confirmed by visual inspection prior to interpolation. Gait data will be precisely time-synchronized with EEG via transistor-transistor logic (TTL) to enable alignment of biomechanical events, neural activity, and visual attention.

EEG data will be processed using a validated analysis pipeline following current guidelines for the removal of movement-induced EEG artifacts [116]. Independent component (IC) analysis will be applied to decompose signals into ICs. Resultant ICs will be localized using inverse source modeling, and only ICs with dipoles located within the brain volume and whose dipole model residual variance was below 15% will be retained. Cortical ICs will be clustered across participants [108,117].

Three analyses will be run on the retained ICs. First, log spectral power will be computed for each IC cluster and averaged across the gait cycle, providing the condition-level measure that underlies the primary Aim 1.2 outcome. Second, event-related spectral perturbations (ERSPs) will be computed from single-trial spectrograms time-locked to right heel strike, using sliding-window time–frequency decomposition normalized to the mean spectrum, to characterize gait-phase-resolved modulation in the theta, alpha, and beta bands. Third, event-related potentials (ERPs) will be extracted time-locked to the 4 VSI task conditions (unisensory visual, unisensory somatosensory, multisensory VS, and catch stimuli) to assess MSI during walking. Established guidelines for artifact rejection, component classification, and automated processing will be followed to ensure reproducibility and analytic rigor [118,119].

Two different resilience mechanisms will be examined: cognitive reserve and brain maintenance. Cognitive reserve is defined as preserved cognitive and physical functions despite evidence of brain changes or pathology, whereas brain maintenance refers to preserved function associated with minimal or no brain changes or detectable pathology [120]. The more general term resilience—defined as the ability to maintain cognition despite poorer brain health, function, or disease—will be used in cross-sectional examinations. Because no gold standard exists for defining protective brain morphology in aging, multimodal MRI metrics will be used to determine the presence of cognitive reserve and/or brain maintenance. These complementary frameworks will guide image selection and analytic strategy.

All images will be acquired on a 3T MRI system. Image preprocessing and analyses will be conducted on a centralized Linux-based high-performance computing cluster. The following image sequences will be acquired: (1) T1-weighted structural images (3D-magnetization-prepared rapid gradient-echo [MPRAGE]), (2) diffusion-weighted images (in 66 diffusion directions), (3) 3D-fluid attenuated inversion recovery [FLAIR] imaging, and (4) T2-weighted (blood-oxygen-level-dependent [BOLD]) images during rest.

Blood samples will be collected by a certified nurse practitioner within the Stony Brook University Clinical Trial Unit in Commack, New York, and processed according to standardized biospecimen handling protocols by trained study personnel. Samples will be processed and distributed promptly for laboratory analyses to minimize preanalytical variability and preserve sample integrity. Plasma and serum aliquots will be cryopreserved and banked for future investigations. Aliquots of plasma, serum, and whole blood will be transferred to Stony Brook University Freezer Farm, a centralized biorepository where samples are tracked and archived using the Freezer Works database, ensuring secure long-term storage at −80 °C for future research.

Stored serum and plasma will support a broad spectrum of biomarker analyses, including proteomic, metabolomic, and biochemical profiling. Genomic DNA will be isolated from whole blood using QIAamp DNA Blood Mini Kits from Qiagen, followed by quality assessment and quantification using the Agilent 2200 TapeStation or the Invitrogen Qubit fluorometer. Extracted DNA will be stored at -80 °C until further analysis.

Fasting will be recommended but not required. Fasting status will be recorded at the time of blood collection. Blood samples will be collected using ethylenediaminetetraacetic acid (EDTA) tubes and serum separation tubes. EDTA tubes will be used for plasma isolation and genomic DNA extraction, enabling downstream methylation array analyses for epigenetic aging clocks.

Blood-based biomarkers including Aβ40, Aβ42, phosphorylated (p)-tau217, and apolipoprotein E will be evaluated as indicators of neurodegeneration and amyloid pathology [43]. Selection of these biomarkers is based on prior validation studies demonstrating their relevance to AD and Alzheimer disease and related dementias (ADRD) [128,129]. NfL has emerged as a robust marker of neurodegeneration [36]. The plasma Aβ42/Aβ40 ratio, quantified using high-precision assays, has demonstrated sensitivity for detecting abnormal amyloid status in cognitively impaired and cognitively normal adults prior to amyloid positron emission tomography (PET) positivity [128,129]. Elevated baseline levels of blood p-tau, particularly p-tau217, have been shown to predict subsequent cognitive decline [130-133]. The ApoE ε4 allele is the strongest known genetic risk factor for AD, whereas the ε2 allele is a known protective factor [134].

To expand biomarker coverage within the current study, additional blood-based markers will be assessed. Serum, plasma, and whole blood samples will be analyzed for a metabolic panel, including creatinine, alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, bilirubin, hemoglobin A_1c, a lipid panel with glycoprotein acetylation (GlycA), C-reactive protein, and complete blood count measures such as hematocrit, mean cell volume, and red cell distribution width. Samples will be shipped to Sunrise Medical Laboratories (Hicksville) for analysis. Plasma collected during the same study visit will be analyzed using the Alamar Biosciences NULISAseq Neuro 220 Panel, a multiplex proteomic assay measuring 220 biomarkers associated with neurodegenerative disease, including markers of amyloid and tau pathology, neurodegeneration, neuroinflammation, synaptic function, and vascular dysfunction [135,136].

Epigenetic aging will be assessed using DNA methylation markers. DNA methylation profiling will be performed using the Illumina Infinium HumanMethylationEPIC v.2.0 BeadChip array at the Roswell Park Comprehensive Cancer Center Genomics Shared Resource (Buffalo, NY, United States), which measures methylation at over 930,000 CpG sites across the genome.

Proteomic profiling will be conducted using the SomaScan assay from SomaLogic to explore biological pathways associated with biological aging and neurodegeneration, and to inform future mechanistic studies focused on gait phenotypes. Proteomic assessments will follow established protocols [137]. Plasma samples will be stored at -80 °C, and 130 microliter aliquots of plasma will be used to perform the SomaScan assay.

Primary hypothesis testing will be limited to prespecified primary outcomes within each aim to reduce concerns related to multiplicity. Secondary and exploratory analyses, including region-wise neuroimaging measures, EEG cluster analyses, proteomic profiling, and methylation analyses, will control the false discovery rate (FDR) within modality-specific outcome families. Analyses not designated as primary will be considered hypothesis-generating.

Two sources of missing data will be distinguished and handled separately. Modality-specific unavailability at baseline (eg, MRI contraindication, refusal, or noncompletion of a procedure) will be documented with its reason and treated as missing by design; affected participants will be excluded from analyses of that modality but retained for all aims for which their data are complete. Longitudinal missingness due to attrition or missed follow-up visits will be accommodated within the mixed-effects framework under a missing-at-random assumption. Missing data patterns will be summarized separately by source and by gait phenotype and baseline risk factors, and baseline characteristics will be compared between completers and noncompleters of each modality to assess selective unavailability.

Descriptive statistics and graphical methods will be used to summarize data distributions and identify outliers. Transformations will be applied when needed to address skewness.

For Aim 1.1, the primary neuroimaging outcome will be brain age. Linear regression models will compare this measure between Super Movers and comparison groups, adjusting for age, sex, education, total intracranial volume, and comorbidities. Secondary neuroimaging analyses will examine additional structural and functional MRI measures using multivariate covariance-based approaches to reduce dimensionality and account for collinearity and participant variability.

For Aim 1.2, the primary EEG outcome is gait-related cortical activation, defined as the spectral power difference between complex and simple walking, averaged across the gait cycle within the a priori frontomedial theta (4‐7 Hz) independent-component cluster. The complex condition is the multisensory VS walking task, in which participants respond to unisensory and multisensory stimuli while walking. This condition is used for the primary contrast in preference to the WWT task because overt speech introduces orofacial and glossokinetic artifact that is present in the complex condition only and would therefore bias the complex-minus-simple difference. The primary comparison contrasts Super Movers with usual movers aged ≥80 years. Because Super Movers are an uncommon phenotype and group sizes may be small, the primary analysis treats this signature as a continuous measure: a linear regression of frontomedial theta activation on gait phenotype, adjusting for age, sex, education, and comorbidities (the Aim 1.1 covariate set). Effects will be reported as standardized coefficients with 95% CIs.

Gait speed: Because cortical oscillatory activity is modulated by walking speed [112,113] and gait phenotypes are defined by self-selected gait speed, speed and phenotype are partially entangled by design. The primary outcome is a within-participant difference score, with both conditions performed at the individual’s own pace, so additive person-level effects of habitual walking speed are subtracted out. Beyond this, gait speed derived from the instrumented walkway will be modeled explicitly, separating within-participant from between-participant variation, and unadjusted and speed-adjusted estimates will be reported side by side. Because gait speed constitutes the operational definition of the phenotype, complete conditioning on speed would adjust away the exposure of interest; the unadjusted estimate is therefore interpreted as the total phenotype effect and the speed-adjusted estimate as the residual component. Phenotype-by-speed interactions and dual-task-related change in walking speed will be examined, and sensorimotor alpha and beta, the bands with the most consistently documented speed dependence, will be analyzed alongside frontomedial theta as a specificity check.

Secondary analyses will (1) repeat the simple-versus-complex contrast using the WWT condition as a sensitivity analysis, with stringent independent-component removal of speech and movement artifact; (2) model presence versus absence of the differential activation signature with logistic regression, summarizing diagnostic performance using sensitivity, specificity, and area under the receiver operating characteristic curve; (3) characterize the full time-frequency response using event-related spectral perturbation across cortical IC clusters in the theta, alpha, and beta bands, with cluster-based permutation testing; (4) treat frontomedial theta variability as a continuous marker of neural stability, linking to Aim 2; and (5) examine the VS ERP measure of MSI recorded during walking. The single prespecified primary outcome limits multiplicity for confirmatory testing; all secondary EEG analyses will control the FDR within the EEG outcome family.

Longitudinal MRI data from repeat imaging in 100 participants will be analyzed using linear mixed-effects models to compare rates of brain aging over time, focusing on regions differentiating groups at baseline. Multiple comparison correction will be applied using family-wise error correction or FDR procedures, as appropriate.

For Aim 2, the primary cognitive outcome will be longitudinal change in Hopkins Verbal Learning Test (HVLT) or Hopkins Verbal Learning Test Revised (HVLT-R) performance. Associations of baseline gait-related EEG activation patterns and MSI with cognitive decline during follow-up will be examined using linear mixed-effects models. These models account for within-participant correlation, accommodate unequally spaced observations, and handle missing outcome data under a missing-at-random assumption.

Secondary analyses will evaluate association with additional cognitive measures and MCR status using generalized mixed-effects models, and time to incident MCR will be analyzed using Cox proportional hazards models. Proportional hazards assumptions will be evaluated graphically and using scaled Schoenfeld residuals. Potential moderation by gait phenotype will be assessed by including interaction terms and by conducting stratified analyses.

Primary analyses for Aim 3 will focus on baseline Aβ42/Aβ40 ratio, ptau217, and NfL blood-based biomarker levels that will be examined using linear regression models adjusting for relevant covariates.

Longitudinal associations of biomarkers, gait phenotype, and their interactions with cognitive decline and incident MCR will be evaluated using linear mixed-effects and Cox proportional hazards models, respectively. Similar analytic approaches will be applied to biological aging outcomes.

Secondary analyses will examine additional Alzheimer-related biomarkers. Exploratory proteomic, methylation, and mediation analyses will also be conducted to examine biological pathways linking gait phenotype, biomarkers, and cognitive decline. Mediation analyses will be exploratory and interpreted as decompositions of associations rather than definitive causal mediation unless assumptions are justified. FDR correction will be applied within biomarker modality-specific outcome families for secondary and exploratory analyses.

As extreme gait phenotypes are uncommon in the general older adult population, recruitment will be enriched for Fast and Super Movers rather than population-proportional, with target enrollment of approximately 50 Super Movers and 50 Fast Movers; slow and usual movers will be combined for the primary analyses. For Aims 2 and 3, accounting for anticipated refusal or noncompletion of aim-specific procedures, the analyzable sample is approximately 300, comprising approximately 75 Fast and Super Movers and 225 usual and slow movers. Power calculations assume a 2-sided alpha of .05 and 80% power; for aims involving two coprimary hypotheses, a conservative 2-sided alpha of .025 was used to account for multiplicity.

For Aim 1, conducted in participants aged 80 years and older, with 50 Super Movers and 100 usual movers completing MRI, we can detect a minimum between-group difference of 0.55 SD (medium effect size) in neural signature measures, adjusting for covariates. Given the overall MRI sample of 150 in this age group and an observed 67% absence of the EEG neural signature among usual movers, we can detect an odds ratio of 0.32 for absence of the neural signature in Super Movers versus usual movers.

For Aim 2, power calculations account for unbalanced numbers of repeated measures in the longitudinal data. With 300 participants completing EEG and MSI studies, cognitive outcomes collected every 6 months over 5 years, an expected attrition rate of 10% in the first year and 5% per year thereafter, and an intraclass correlation coefficient of at least 0.5 among repeated cognitive outcomes, we can detect a difference of 0.05 SD per year in the rate of cognitive decline per 1 SD difference in baseline gait-related EEG activation pattern or MSI. With Fast and Super Movers constituting approximately 25% (75/300) of this sample, we can detect a minimum difference of 0.25 SD per year (eg, 0.5 points on the HVLT) between Fast and Super Movers and usual and slow movers in the rate of cognitive change per 1 SD difference in baseline gait-related EEG activation pattern or MSI. Based on an incident MCR rate of 51 per 1000 person-years in our local cohorts, approximately 44 incident MCR cases are expected over 5 years, providing 80% power to detect a hazard ratio of 1.62 per 1 SD increase in baseline gait-related EEG activation pattern or MSI.

For Aim 3, with 300 participants and Fast and Super Movers constituting approximately 25% of the sample by design, we can detect a between-phenotype difference of 0.43 SD (moderate effect size; two-sided alpha=.025) in blood-based biomarker levels. Under the same longitudinal assumptions as Aim 2, we can detect a difference of 0.05 SD per year in the rate of cognitive decline per 1 SD difference in baseline biomarkers, and a difference of 0.25 SD per year between Fast and Super Movers and usual and slow movers in the rate of cognitive change per 1 SD difference in baseline AD biomarkers.

Longitudinal analyses will use linear mixed-effects models, which improve statistical efficiency by incorporating repeated observations and accommodating incomplete follow-up under a missing-at-random assumption. Missing data patterns will be summarized by gait phenotype and baseline risk factors, and sensitivity analyses using multiple imputation and/or inverse probability weighting will evaluate robustness to informative missingness. Because of cross-enrollment, recruitment source will be included as a covariate or stratification factor, and baseline characteristics will be compared between newly recruited and cross-enrolled participants.