The association between amyloid-beta deposition on dual-task gait performance is partially moderated by cognitive functions in healthy older adults Authors: Amgad Droby, Mihir Kurkure, Inbal Maidan, Christian G Habeck, Jeffrey M Hausdorff, Yaakov Stern, Anat Mirelman Published: 2025 ABSTRACT AbstractAmyloid beta (Aβ) deposition has been implicated in the disruption of neuronal activity and large-scale network connectivity in older adults, contributing to impairments in cognitive function and mobility. Nevertheless, the extent to which regional patterns of Aβ accumulation differentially influence gait and cognition, and whether cognitive function modulates the association between Aβ burden and gait performance in healthy aging, remain insufficiently understood. This study examined the complex relationships among regional Aβ deposition, gait performance, and cognitive function in healthy older adults. The study included 33 healthy older adults (mean age = 65.72 ± 2.77 years; 12 females, 21 males). Participants underwent PET scans, cognitive assessments, and gait evaluations in usual-walk (UW) and dual-task (DT) conditions. Pittsburg Compound B (PiB-PET) imaging quantified Aβ burden across 68 brain regions. A three-step statistical analysis was conducted. Hierarchical regression models explored connections between Aβ, cognitive function (reasoning, vocabulary, memory, processing speed), and gait performance. Path analysis examined inter-relationships. All models were adjusted for age, sex, and years of education. A statistical threshold of p ≤ 0.05, adjusted for multiple comparisons, was applied. Elevated Aβ levels in the pallidal (β=-0.05, p < 0.001), and entorhinal cortex (β=-0.35, p = 0.021), were significantly associated with poorer step regularity, and hippocampal Aβ levels (β=-1.3, p = 0.037) were significantly associated with poorer step symmetry under UW conditions. During DT walking, higher Aβ in the cerebellar was associated with reduced stride regularity (β=-1.38, p = 0.034), while higher cerebellar and pericalcarine Aβ was linked to lower step symmetry performance (β=-1.1, p < 0.001). In contrast, elevated Aβ in the cingulate (β = 0.75, p < 0.001) and cuneus (β = 0.49, p = 0.024) was positively associated with step symmetry. Path analyses indicated that cognitive scores in vocabulary and reasoning moderated the relationship between regional Aβ deposition and DT stride regularity and step symmetry. The obtained results indicate that regional Aβ deposition is differentially associated with specific gait metrics depending on brain region and walking condition. Cognitive performance in vocabulary and reasoning moderated these associations, indicating a potential protective role. These findings highlight the relevance of region-specific pathology and cognition in early detection of age-related motor and cognitive decline.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-28077-4.Keywords: Amyloid-beta, Older adults, Gait, CognitionSubject terms: Neuroscience, Cognitive ageing, Cognitive neuroscience Introduction Aging is a complex biological process characterized by a decline in physiological functions and by structural and functional brain alterations that contribute to cognitive impairment1,2. Among the principal biological markers linked with age-related cognitive decline is the amyloid beta (Aβ) burden. Aβ is a peptide produced through the proteolytic cleavage of the amyloid precursor protein, and its accumulation and aggregation within brain tissue represent a hallmark of Alzheimer’s disease (AD) pathology. The extent of cerebral Aβ deposition—referred to as “Aβ burden”—serves as a marker for the early and preclinical identification of AD, as this peptide plays a pivotal role in the pathogenesis of the disorder3,4. Recent population studies reported that Aβ deposition is detectable in approximately 6% of adults aged between 55 and 60 years and up to 30% of adults aged 80 years or above5,6. Elevated Aβ levels initiate a neurotoxic cascade, including synaptic dysfunction, neuroinflammation, and impaired neuronal signaling, that underlies progressive cognitive decline and is often accompanied by changes in gait, underscoring Aβ’s critical role in the pathophysiology of AD and related neurodegenerative disorders7,8. Advances in positron emission tomography (PET) with Aβ-targeted radiotracers enable direct in vivo visualization of Aβ activity in the brain, providing a robust tool for investigating the relationships between biological neural changes and functional outcomes across the lifespan and disease states5. As people age, cognitive performance naturally declines, with subtle but measurable deficits arising in domains such as memory, processing speed, visuospatial ability, and reasoning7–14. Importantly, findings from large-scale longitudinal studies—including the Mayo Clinic Study of Aging—demonstrate that even among cognitively healthy individuals, elevated baseline Aβ burden detected by PET reliably predicts accelerated cognitive decline, highlighting the significance of Aβ accumulation as an early marker of neural dysfunction in aging15. Deficits in executive functions and memory are associated with poor gait and increased fall risk, particularly during dual-task (DT) conditions, in which a cognitive task is performed simultaneously with walking16–20. DT paradigms are widely used to uncover the interplay between cognition and gait, revealing decrements in motor and cognitive performance when both are taxed concurrently21, such as reduced gait velocity and increased gait variability or poorer verbal fluency and working memory22–24. Historically, cognition and mobility were viewed as largely independent; however, extensive research now demonstrates that gait is a complex behavior integrating motor, sensory, and cognitive processes25,26. Previous studies show that Aβ in the frontal, parietal, and subcortical regions, key areas for motor planning, executive control, and sensorimotor integration, is associated with both automatic and cognitively controlled gait mechanisms, especially under DT conditions27. This region-specific impact may impair the integration of cognitive and motor processes, leading to less stable and more variable gait, as demonstrated in both clinical and imaging studies14,18,27. Thus, the interplay between Aβ, cognitive function, and gait reflects disruptions in brain networks underlying executive and motor control, substantiating the rationale for examining region-specific Aβ effects in dynamic gait tasks. Recent work has specifically examined the relationship between cerebral Aβ burden and gait performance. For example, Del Campo et al. (2016) found that elevated Aβ in brain regions including the putamen, occipital cortex, precuneus, and anterior cingulate was linked to slower gait speed and diminished physical performance, even before the onset of cognitive impairment27. Similarly, Nadkarni et al. (2017) observed that individuals with higher Aβ deposition in the anterior cingulate, frontal, temporal, and parietal brain regions experienced greater gait decline during both normal and DT walking, with Aβ burden correlating with increased DT cost on gait speed14. However, these studies mainly examined global cognitive status or usual gait speed and did not systematically investigate whether specific cognitive domains moderate the associations between Aβ burden and gait metrics, nor did they explore how DT–related gait changes correspond to regional Aβ deposition. The primary objective of this study was to elucidate the complex relationships between regional brain Aβ burden, cognitive domain performance, and quantitative gait measures in healthy older adults. By employing a multi-layered analytical approach, we aimed to identify both direct and indirect associations among these factors, with particular focus on the differential effects of Aβ in higher cortical regions and gait performance under cognitively demanding conditions (i.e., DT walking), and whether cognitive abilities mediate the influence of regional Aβ on gait performance. We hypothesize that greater Aβ deposition in prefrontal, inferior parietal, and motor regions will be associated with poorer gait performance, with stronger effects under cognitively demanding DT conditions. Specifically, Aβ burden in regions supporting attention and sensorimotor integration is expected to show the strongest associations with DT gait impairments. We further hypothesize that cognitive performance will moderate the association between Aβ burden and gait outcomes, such that individuals with lower cognitive scores will show stronger Aβ–gait relationships than those with higher cognitive function. Methods Study participants This work is a sub-analysis of data obtained from the Reserve and Reference Ability Neural Network (RANN) studies28,29. Thirty-three (Age [mean ± SD]: 65.72 ± 2.77; Range: 61–71 years, N = 12 females) healthy community-dwelling older adults were recruited from the New York City metropolitan area through a randomized market mailing procedure. Participants were initially screened over the phone to determine whether they met the basic inclusion criteria. Participants were enrolled in the sub-study if they met the inclusion criteria which included: (i) right-handed, (ii) English-speaking, (iii) no outstanding orthopedic, psychiatric and/or neurological disorders that can affect either gait and/or cognition, (iv) normal or corrected-to-normal vision, (v) no CNS-targeting medications. Exclusion criteria included: (i) standard MRI contraindications, (ii) significant medical or neurological conditions that could affect cognitive or motor function, such as brain disorders (e.g., stroke, tumor, epilepsy), major systemic diseases (e.g., uncontrolled hypertension, active hepatic or renal disease), (iii) psychiatric diagnosis including a history of psychosis, major depressive disorder, bipolar disorder, or anxiety disorder, (iv) A history of non-skin neoplastic disease (excluding prostatic carcinoma), HIV infection, or recent (within one month) use of CNS-targeting medication. Eligible individuals were further screened in person, and had a Mattis Dementia Rating Scale30 score of at least 130 indicating cognitively preserved function31. The study received ethical approval from the Institutional Review Board of the College of Physicians and Surgeons of Columbia University and the Institutional Review Board of Tel Aviv Sourasky Medical Center (TLVMC- approval number: 0060-18-TLV). Participant recruitment, study procedures, and data collection were conducted at Columbia University Medical Center (CUMC). Data processing and analysis were carried out at both CUMC and TLVMC. Informed written consent was obtained from all participants prior to enrollment, and all study procedures adhered to the relevant guidelines and regulations. See Fig. 1 for schematic illustration of the study design. Schematic illustration of the study design. Demographics & cognitive assessments Demographic data, including age, sex, education level, were collected through structured interviews and standardized assessments. Education level was recorded as the total years of formal schooling completed, and the Wechsler Adult Intelligence Scale (WAIS-III)32. These variables were included to characterize the study sample and control for potential confounding factors in the analyses. Cognitive performance was evaluated with a set of tasks targeting four core domains: vocabulary, processing speed, fluid reasoning, and memory. Cognitive testing was performed during fMRI scanning. To adapt these tasks for administration during fMRI scanning and to optimize statistical power for image analysis, certain modifications in task structure and presentation were necessary. Paper-and-pencil versions of the cognitive tasks used in this study have demonstrated high internal consistency and test–retest reliability, with coefficients typically exceeding 0.7, as well as moderate to large loadings on their respective reference abilities [28,29]. Prior to each scanning session, participants received computerized training on the subset of six tasks to be administered in that session. This training ensured participant familiarity and minimized learning effects during scanning, thereby enhancing data quality and reliability of the cognitive performance measures [See 29 for more details]. The Vocabulary domain included the Synonyms and Antonyms tasks33, where participants selected a synonym or antonym of a probe word from four on-screen options across 15 trials per task. Participants pressed one of four buttons corresponding to their choice, with each item presented for 13.5 s. The Processing Speed domain comprised: the Digit Symbol substitution task, requiring verification of digit-symbol pairs over 90 trials; and Letter and Pattern Comparison tasks34, where participants judged whether simultaneously presented pairs of letter strings or line patterns matched, with 60 trials each. All speed tasks emphasized rapid, accurate responses. The Fluid Reasoning domain included paper folding35, matrix reasoning36, and letter sets tasks; these involved selecting figures or sets that conformed (or violated) spatial or pattern rules. Trials ranged from 7 to 18 self-paced attempts per task, with response windows calibrated to participant speed. The Episodic Memory domain consisted of Logical Memory, Word Order, and Paired Associates tasks. Logical Memory presented stories in segments, followed by multiple-choice questions; Word Order required identification of sequentially presented words; and Paired Associates tested memory of word pairs, with probe trials following study phases. Accuracy measured performance in Vocabulary, Fluid Reasoning, and Memory tasks, while reaction time was the primary metric for Processing Speed tasks. To enable cross-domain comparison, all scores were z-transformed relative to the sample mean, with higher z-scores indicating better accuracy or faster reaction times as appropriate (See29 for further details). Gait assessments Gait was assessed using a lightweight, wearable sensor including a three-axis accelerometer and a three-axis gyroscope (DynaPort; McRoberts, The Hague, Netherlands). The device was placed on the lower back of the participants to assess gait during usual and dual-task walking conditions. The gait protocol comprised a one-minute trial along a 20-meter corridor under two conditions: (i) usual walking (UW) at the participant’s preferred speed, and (ii) dual-task walking (DT), in which participants walked while simultaneously generating words beginning with the letter “A”, with no restrictions on word length or syllable count. The participants received no instructions for task prioritization, and were not made aware of any errors or received any feedback on their performance. The spatiotemporal gait measures were obtained from the mean values of a minimum of 30 steps included. Signals were filtered using a low-pass Butterworth filter with a frequency cutoff of 3.5 Hz with a band pass of less than 0.5 dB. Only straight line walking was analyzed37. Gait measures across three domains: pace (velocity, stride length), variability (step and stride time variability, and step regularity), and symmetry (step asymmetry) were analyzed under both walking conditions37–40. To minimize redundancy, parameters showing strong correlations with gait speed, such as step length, were excluded from further analyses. Stride time variability was calculated as the coefficient of variation (standard deviation divided by mean, multiplied by 100) of stride intervals and reflects fluctuations in gait timing as increased variability is associated with impaired mobility in various populations. Step and stride regularity measures, ranging from 0 to 1, were derived from unbiased autocorrelation of acceleration signals; lower scores indicate greater inconsistency and have been linked to asymmetric gait patterns in conditions such as Parkinson’s disease, stroke, and osteoarthritis41–44. Step asymmetry was quantified by the normalized logarithmic ratio of shorter to longer mean swing durations of each step between feet, with higher values indicating greater asymmetry. Imaging All participants underwent MR imaging on a 3.0T Philips Achieva scanner. High-resolution 3D T1-weighted images were acquired for each subject with a Magnetization Prepared Rapid Gradient Echo (MPRAGE) sequence [TE/TR: 3/6.5 ms; Field of view: 256 mm; Flip angle: 8°; In-plane resolution: 256 × 256 voxels; Slice thickness/gap: 1/0 mm; Slices: 180]. In addition, participants underwent 18F-Florbetaben PET scans to assess amyloid-β burden. The PET scans were acquired on the same MCT PET/CT scanner (Siemens) in dynamic, 3D imaging mode beginning 50 min after injection. Participant preparation consists of intravenous catheterization followed by the bolus injection (over 10–20 s) of the tracer. Brain images were acquired in 4 × 5-minute frames over a period of 20 min. The images were immediately assessed for technical validity. If considered inadequate, the participant had an additional 20 min of continuous imaging. Transmission scans were done prior to the scan. If there was a repeat scan, transmission was done after the scan. Imaging data processing Aβ PET image processing was performed as described previously45 (See: Tahmi et al., 2019). Briefly, structural T1-weighted MRI scans were processed using the FreeSurfer pipeline (http://surfer.nmr.mgh.harvard.edu) to segment subcortical structures and parcellate cortical regions of interest (ROIs). The processing included bias field correction, intensity normalization, and skull stripping with a watershed algorithm, followed by segmentation based on white matter boundaries, which allowed for accurate definition of gray matter, white matter, and pial surfaces, as well as topology correction. For PET image analysis, four dynamic frames were first aligned using rigid-body registration (six degrees of freedom) to account for any head movement, then averaged voxel-wise to create a single static PET image. The corresponding CT scan, acquired for attenuation correction, was coregistered to the PET image using rigid-body registration with normalized mutual information, yielding a merged PET-CT image in native space. The participant’s FreeSurfer-processed T1 MRI image was then registered to the merged PET-CT image, enabling the transfer of all ROI masks into PET native space. Finally, standardized uptake value ratio (SUVR) maps were computed by normalizing each voxel’s uptake value to the mean uptake in the cerebellum grey matter (GM) reference region in 68 brain regions obtained using FreeSurfer cortical parcellations. Statistical analysis SPSS® v29 (IBM Statistics) was used for all performed statistical analyses. A repeated-measures t-test was used to assess DT effects on gait parameters by comparing measures performance scores between the UW and DT walking conditions. To investigate the relationships between regional Aβ deposition, cognitive performance, and gait, a three-step hierarchical regression procedure was employed. First, stepwise hierarchical regression models assessed the association between regional Aβ levels (measured SUVR in 68 brain regions from FreeSurfer parcellations) as independent variables and cognitive performance across four distinct domains as dependent variables. Second, similar hierarchical models examined the relationship between regional Aβ and gait performance during both UW and DT conditions. In each tested regression model, the first block controlled for demographic covariates including age, sex, education, and white matter hyperintensity burden, while the second block included the measured regional Aβ predictors of interest. Cognitive variables were entered as individual domain scores to preserve domain-specific effects. Due to the complexity of the hypothesized relationships, cognition and gait were initially modeled separately to elucidate their direct associations with Aβ. The third stage of the statistical analysis was performed based on the results of the step-two regression models described above. Here, moderation analyses were conducted to assess the interaction effects of regional Aβ and cognitive domain scores on gait measures. These analyses were performed using the PROCESS46, with cognitive performance scores specified as moderators in the models (W). This approach allowed us to explore how cognitive function modulates the relationship between Aβ burden and gait, complementing the main effect analyses and enhancing model interpretability. To address the issue of multiple comparisons, we implemented a structured correction approach tailored to the number and grouping of tests performed at each analysis stage. For repeated measures t-tests, we applied a Bonferroni correction by dividing the conventional alpha level (0.05) by the number of comparisons (5), resulting in an adjusted significance threshold of p ≤ 0.01. For hierarchical regression models examining associations between regional Aβ levels and cognitive performance across four cognitive domains, the threshold was set at p ≤ 0.0125 (0.05/4). For gait performance analyses, which considered three gait domains (pace, variability, symmetry) each represented by two measures, we adjusted the threshold to p ≤ 0.025 (0.05/2) per domain. In path analyses, p-value thresholds were further adjusted based on the number of independent brain regions tested for each gait outcome, as specified in Table 2. This tiered correction strategy aimed to maintain an appropriate family-wise error rate while preserving statistical power by accounting for the logical structure and dependencies among tests. Table 2Results of the step-wise regression analysis examining the association between measured regional Aβ and gait performance.ModelIndependent predictors R 2 Adjusted R2FdfP- valueΒt-valueP-valueUsual walking (UW) UW step regularity0.570.5215.791–21< 0.001 Left pallidum− 0.5− 4.26< 0.001 Right entorhinal cortex− 0.352.70.021 UW step symmetry0.190.154.941–210.037 Right hippocampus− 1.29− 2.220.037Dual-task walking (DT) DT stride length0.250.216.571–210.034 Sex− 0.14− 2.560.019 DT stride regularity0.20.165.121–210.025 Left cerebellar cortex− 1.38− 2.260.034 DT step symmetry0.610.565.771–21< 0.001 Left cerebellar cortex− 1.62− 3.87< 0.001 Left cuneus0.753.85< 0.001 Right isthms cingulate0.492.40.024 Right pericalcrine gyrus− 1.1− 4.19< 0.001 Results of the step-wise regression analysis examining the association between measured regional Aβ and gait performance. Results Demographics, gait performance, and cognitive scores of the study participants Demographic, cognitive scores, and gait performance in both walking conditions of the studied participants are displayed in Table 1; Fig. 2. Table 1Demographic, gait, and neuropsychological characteristics of the study group.VariableMean ± SDRangeN = 33DemographicsAge65.72 ± 2.7761–71GenderFemales, n = 12 (36.4%)Males, n = 21 (63.6%)IQ120.08 ± 797–128Years of education15.82 ± 2.1412–20Neuropsychological performance (domain z-scores)Processing speed-0.08 ± 0.59-1.37–0.81Vocabulary0.47 ± 0.61-1.47–1.214Episodic memory0.06 ± 0.88-2.07–1.32Fluid reasoning-0.12 ± 0.55-1.19–0.98GaitUW velocity [m/sec]0.89 ± 0.130.71–1.33UW stride length [m]1.13 ± 0.190.85–1.69UW stride regularity [a.u]0.75 ± 0.100.41–0.9UW step symmetry [a.u]0.79 ± 0.190.43–1.39UW step regularity [a.u]0.58 ± 0.120.31–0.83DT velocity [m/sec]*0.81 ± 0.100.61–1.17DT stride length [m]1.11 ± 0.140.85–1.47DT stride regularity [a.u]*0.59 ± 0.160.16–0.93DT step symmetry [a.u]*0.50 ± 0.150.28–0.87DT step regularity [a.u]#0.9 ± 0.310.46–2.03IQ = intelligence quotient, UW = usual walking, DT = dual-task, a.u = arbitrary unit.*Gait measures significantly reduced in DT walking condition (Repeated-measures t-test; p < 0.001 in all cases).#Gait measures significantly increased in DT walking condition (Repeated-measures t-test; p < 0.001). Demographic, gait, and neuropsychological characteristics of the study group. IQ = intelligence quotient, UW = usual walking, DT = dual-task, a.u = arbitrary unit. *Gait measures significantly reduced in DT walking condition (Repeated-measures t-test; p < 0.001 in all cases). #Gait measures significantly increased in DT walking condition (Repeated-measures t-test; p < 0.001). (A) Schematic illustration of the calculated gait parameters: gait velocity (X [m]/time [s]), stride length, stride regularity, step symmetry, and step regularity. (B) Bar plots showing gait parameters measured during usual walking (UW) and dual-task (DT) conditions. Repeated measures t-test revealed significant effects of DT walking conditions on gait parameters. Significant reductions of walking velocity (t (32) = 5.93, p < 0.001 Cohen’s d CI-95% [0.60, 1.45]), stride regularity (t(32) = 6.38, p < 0.001 Cohen’s d CI-95% [0.67, 1.54]), and step symmetry (t(32) = 9.39, p < 0.001 Cohen’s d CI-95% [1.10, 2.15]) were observed in DT walking compared to UW condition. Reduced stride length was also observed in DT vs. UW conditions; however, this did not reach statistical significance. On the other hand, step regularity was observed to be increased in DT vs. UW condition (t(32) = -6.96, p < 0.001 Cohen’s d CI-95% [-1.62, 0.74]). Association between regional Aβ and cognitive performance Hierarchical regression models examining the association between measured regional brain Aβ levels and cognitive performance showed the following. Reasoning performance scores were inversely associated with Aβ levels in the right temporal pole (β=-2.8, CI-95% [-4.92, -1.45]. The overall model fit was R2 = 0.28, F(30) = 11.93, p = 0.002, reflecting the proportion of variance in Reasoning performance scores explained by all predictors included in the model. In the memory cognitive domain, female gender was positively associated with memory performance, while higher Aβ levels in the left para-hippocampal cortex were negatively associated with memory scores (β = -3.87, 95% CI [-6.81, -0.93]); where the overall model fit was R² = 0.36, F(30) = 8.36, p = 0.001. Finally, measured Aβ levels in the right lingual gyrus were found to be associated negatively with processing speed performance scores (β = -2.174, CI-95% [-4.72, -0.72]) however this did not survive correction for multiple comparisons (overall model fit: R2 = 0.13, F(30) = 4.46, p = 0.043). See Supplementary Table 1 summarizing these results. No regional Aβ measures were included as predictor variables in the regression model assessing associations with vocabulary performance scores. Association between regional Aβ and gait performance Results of hierarchical regression models examining the association between measured regional brain Aβ levels and gait performance are summarized in Table 2. UW condition: Higher values of Aβ in the left pallidum (β = -0.5, CI-95% [-0.74, -0.25]) and right entorhinal cortex (β = -0.35, CI-95% [-0.63, -0.06]) were associated with lower values of step regularity. Additionally, right hippocampal Aβ levels were associated negatively with step symmetry (β = -1.29, CI-95% [-2.49, -0.08]). No regional Aβ measures were included as predictors in the regression models assessing associations with UW velocity, stride length, or stride regularity, indicating that no significant relationships were detected between these variables. DT walking condition: In DT walking, Aβ accumulation in the left cerebellar cortex was negatively associated with stride regularity (β = -1.38, 95% CI [-2.65, -0.11]). Significant negative associations were also observed between left cerebellar cortex Aβ (β = -1.32, 95% CI [-2.48, -0.76]) and right peri-calcarine Aβ (β = -1.10, 95% CI [-1.64, -0.56]) with step symmetry, whereas positive associations were identified for left cuneus Aβ (β = 0.75, 95% CI [0.35, 1.16]) and right isthmus cingulate Aβ (β = 0.49, 95% CI [0.071, 0.91]). In addition, female sex was associated with shorter stride length (β = -1.36, 95% CI [-0.25, -0.025]). No regional Aβ measures showed significant associations with DT walking velocity, step regularity, or stride regularity. Moderation effects between regional Aβ and gait performance In the UW condition, path analysis revealed no significant moderation effects of cognitive performance on the relationship between Aβ levels and gait performance. In contrast, during the DT walking condition, both direct and indirect effects were observed between Aβ levels in the left cerebellar cortex and gait parameters. For DT stride regularity, path analysis demonstrated a significant direct negative effect of left cerebellar Aβ (β = -2.06, SE = 0.65, p = 0.0037). In addition, approximately 28% of the total effect was moderated by an interaction between vocabulary cognitive performance and left cerebellar Aβ levels (β = 3.13, SE = 1.17, p = 0.012). Vocabulary performance exerted a significant conditional effect on the association between left cerebellar Aβ and stride regularity at values ≤ -1.5 standard deviations below the group mean (β = -2.54, p = 0.003, 95% CI [-4.41, -0.94]), whereas higher vocabulary scores did not significantly influence this association (Fig. 3A). Moderation analysis of associations between regional amyloid-beta (Aβ), cognitive performance, and gait measures. Analyses were restricted to regions showing significant Aβ-gait associations in hierarchical regression (corrected for multiple comparisons). (A) Anatomical illustration showing the location of the left cerebellar cortex (B) Vocabulary performance (W) significantly moderated the relationship between left cerebellar cortex Aβ (X) and DT stride regularity (Y) (β = 3.13, SE = 1.17, p = 0.012), explaining 28% of the total effect. (C) Reasoning performance (W) significantly moderated the relationship between left cerebellar cortex Aβ (X) and DT step symmetry (Y) (β = 2.14, SE = 0.99, p = 0.039), explaining ~ 31% of the total effect. X, independent variable; W, moderator; Y, dependent variable; DT, dual-task. Similarly, significant direct and indirect effects were found between left cerebellar Aβ levels and DT step symmetry. Path analysis showed a significant direct negative effect of left cerebellar cortex Aβ on DT step symmetry (β = -1.20, SE = 0.44, p = 0.011), with 31% of the total effect moderated by an interaction between reasoning cognitive performance and Aβ levels (β = 2.14, SE = 0.99, p = 0.039). Reasoning performance had a significant conditional effect on the relationship between left cerebellar Aβ and DT step symmetry within the range of ≤ -1.5 to 0 standard deviations of the group mean (β = -1.44, p = 0.004, 95% CI [-2.38, -0.5]), beyond which no significant moderation was observed (Fig. 3B). No significant mediation effects were observed among the tested variables during the third step of the statistical analysis in either walking condition. Discussion We assessed the interrelations between regional brain Aβ deposition, gait performance, and cognitive functioning among cognitively healthy, community-dwelling older adults. Consistent with previous studies, elevated Aβ levels in brain areas involved in higher cognitive functions, such as the para-hippocampal gyrus, temporal pole, and lingual gyrus, were associated with poorer cognitive performance across multiple domains, including reasoning, memory, and processing speed4,47,48. Additionally, better performance during UW was linked to lower Aβ levels in striatal regions of the brain. In contrast, DT walking performance showed a negative association with Aβ deposition in cerebellar regions and a positive association with Aβ in temporal and posterior cingulate areas. Path analysis further revealed that, in DT walking conditions, the relationship between Aβ accumulation in the left cerebellar cortex and gait performance was partially influenced by vocabulary and reasoning cognitive scores, indicating that cognitive abilities may moderate the impact of Aβ on motor function. These findings highlight region-specific Aβ effects on gait, in which striatal Aβ is associated with UW performance, while cerebellar, temporal, and posterior cingulate Aβ relate to DT walking measures including step regularity and symmetry. Preserved vocabulary and reasoning abilities appear to partially offset the impact of cerebellar Aβ on DT gait performance. The association between Aβ deposition and cognitive-motor function decline is well-documented in AD and aging14,27,47–50. For instance, studies in cognitively impaired individuals have demonstrated that global Aβ burden is associated with slower gait speed and lower scores on measures such as the Short Physical Performance Battery (SPPB), but not with performance on the Timed Up-and-Go Test (TUG-T), indicating that Aβ may not be consistently linked to all aspects of functional mobility50. Regional analyses further identify Aβ accumulation in motor-planning regions, such as the primary motor cortex, putamen, and caudate nucleus, as a key contributor to gait variability, including fluctuations in stride time and cadence51. These findings are consistent with our hierarchical regression analyses, which demonstrated associations between Aβ burden in motor-related areas like the pallidum and hippocampus and altered step regularity and symmetry during UW. Together, these results underscore the role of Aβ in disrupting neural circuits essential for gait control, even at early stages of pathological aging. During DT walking, elevated Aβ levels in the left cerebellar cortex were associated with poorer gait performance, particularly reduced stride regularity. Notably, DT step symmetry demonstrated positive associations with Aβ deposition in the left cuneus and right isthmus cingulate, but negative associations with Aβ in the peri-calcarine gyrus and left cerebellar cortex, regions critical for visuo-motor integration and spatial orientation52–55. The co-occurrence of both positive and negative regional associations between Aβ deposition and dual-task gait symmetry may reflect the interplay of localized neural damage and adaptive mechanisms that support task performance during cognitively demanding tasks. Alternatively, these contrasting associations may indicate a non-linear relationship between Aβ accumulation and functional outcomes27. Imaging studies consistently highlight the importance of fronto-parietal and cingulate regions in gait control among older adults; neural circuits that are particularly vulnerable to disruption by Aβ pathology in neurodegenerative conditions14,56–59. Such compromise impairs cortical signaling, memory, executive function, and DT gait performance14,60–64. These results indicate that Aβ accumulation across multiple brain regions interferes with the neural integration needed for complex gait, reinforcing the central roles of cognitive control (prefrontal cortex) and motor coordination (motor cortex, cerebellum) in maintaining gait stability under cognitive challenge19,65–68.​ Furthermore, the cerebellum’s established connections with cortical regions support its integral role in attention, language, working memory, and executive processes67–70. This architecture enables the cerebellum not only to facilitate cognitive functions, but also to compensate for deficits arising when other brain areas are affected by disease or aging26,67–72. The cerebellum’s adaptive capacity may help redistribute cognitive processing, mitigating the impact of damage or decline on gait performance. Although regional Aβ burden was not directly linked to vocabulary scores, path analysis showed that vocabulary performance moderated the relationship between left cerebellar cortex Aβ and DT stride regularity. Similarly, reasoning scores moderated the association between left cerebellar cortex Aβ and DT step symmetry, with these effects most pronounced among individuals with lower cognitive scores. These results suggest that older adults with reduced cognitive abilities may need to draw more on their remaining cognitive resources to maintain gait during DT walking, underscoring the intricate relationship between cognition and motor control in aging56,57. Vocabulary scores generally increase until around age 65 and then remain stable, making them a common indicator of cognitive reserve in older adults73,74. Our findings on fluid reasoning are consistent with previous research showing its importance for loco-motor learning and adaptation, suggesting that strong reasoning skills help older adults adjust to novel walking challenges like DT walking75. These results highlight a complex relationship between cognitive reserve, reasoning, and gait, challenging the traditional capacity-sharing theory, which assumes cognitive and motor tasks compete for limited resources76. Instead, our data suggest that cognitive and motor functions rely on distinct but interconnected neural networks, allowing individuals with preserved cognitive abilities to better maintain gait performance under cognitively demanding walking conditions77–79. Importantly, our path analysis indicates that enhancing reasoning and vocabulary through targeted cognitive training could help preserve mobility and reduce cognitive decline risk in older adults, especially those with elevated Aβ. This approach may support independence and quality of life in aging populations. Limitations While our path analysis provides insights into the interplay between regional Aβ burden, cognition, and gait, these findings should be interpreted with caution. Participants were not instructed to prioritize either the cognitive or walking task during dual-task assessments, allowing them to perform in their natural, habitual dual-task manner. While this approach may have introduced some variability in individual task prioritization strategies, it avoided the bias that explicit prioritization instructions could impose and better reflects real-world dual-task performance. The cross-sectional design of this study, combined with a small sample size and above-average participant IQ, limits generalizability and constrains the statistical modeling approaches that can be employed. An additional possible limitation is that we did not control for other widespread age-related brain changes, such as reductions in brain volume and the presence of lacunar infarcts, which may also influence the studied associations. Ultimately, a deeper understanding of the relationship among these variables could help address age-related gait disturbances and their consequences, such as falls and reduced mobility. Future research should focus on prospective studies with larger, more diverse populations, and consider genetic risk factors for amyloid accumulation. Such studies would enhance the generalizability of the findings and help clarify potential causal relationships within this important yet still poorly understood triad. Additionally, future studies with larger sample sizes are warranted to re-examine potential mediation pathways among cognitive, brain, and gait measures that may have remained undetected in the present study. Conclusions Our results suggest that in older adults, the interplay between regional Aβ burden and cognitive performance partly accounts for gait outcomes, especially under complex conditions like DT walking. These findings may have direct implications for intervention designs; targeting both cognitive function and amyloid pathology concurrently may prove more effective for fall prevention than motor training alone. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors would like to express their gratitude to all research participants for their committed involvement in this study. We also extend our thanks to Dr. Tal Kozlovski for the statistical review and guidance. Author contributions AD, JH, IM, and AM conceptualized the study design, analyzed the data, and outlined the manuscript. MK, CH, CGH, and YS collected, analyzed the data, and revised the manuscript. All authors contributed to the article and approved the submitted version. Data availability The data supporting the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants. Anonymized data can be made available from the corresponding author, A.D (email: [amgadd@tlvmc.gov.il] ) upon reasonable request. Declarations Competing interests The authors declare no competing interests. Ethics declaration This study was approved by the Internal Board of the College of Physicians and Surgeons of Columbia University and the Institutional Review Board at TLVMC. Informed written consent was obtained from all study participants after the nature and risks of the study were clearly explained. All procedures were conducted in accordance with the relevant guidelines and regulations. Footnotes Additional Information The original online version of this Article was revised: The original version of this Article contained an error in the author list. Caleb R. Haynes was incorrectly listed as an author of the original Article and has subsequently been removed. Additionally, the original version of this Article contained an error in the spelling of the author Jeffrey M. Hausdorff which was incorrectly given as Jefferey M. Hausdorff. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Change history 5/19/2026 A Correction to this paper has been published: 10.1038/s41598-026-46024-9 References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data supporting the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants. Anonymized data can be made available from the corresponding author, A.D (email: [amgadd@tlvmc.gov.il] ) upon reasonable request.