Cognition and brain network connectivity in timed up & go performance

Abstract

Gait performance depends on numerous aspects of brain functioning that are also relevant to key cognitive processes throughout the lifespan. The Timed Up and Go (TUG) test has been shown to be a reliable tool for assessing age-related mobility changes and risk of falls in older adults. This study aimed to predict TUG performance using motor-cognitive inter-network connectivity, cognitive performance and socio-demographics. 189 participants without cognitive impairment were included. Mobility was measured with the TUG through a wireless inertial sensor device. Functional connectivity between the somatomotor hand, cerebellar, and cognitive-related networks was assessed using pairwise correlations among 264 predefined regions assigned to large-scale functional networks. Connectivity between networks was summarized by averaging Fisher z-transformed values across all ROI pairs linking each network pair. Three linear regression models were constructed including the dependent variables TUG total time, turning time and turning peak-angular velocity (PAV) and the independent variables age, sex, years of education, memory and executive performance, and inter-network connectivity patterns. Somatomotor hand – ventral attentional and cerebellar – DMN inter-network connectivity were significant predictors of TUG total time. Memory performance, somatomotor hand – salience and –cingulo opercular inter-network connectivity were significant predictors of turning time. Age, years of education, somatomotor hand – dorsal attentional and cerebellar – cingulo opercular network activation were significant predictors of turning PAV. Findings indicate different contributions of attentional systems to gait performance in aging.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-025-02016-7.

Introduction

Gait and cognition are interrelated throughout the lifespan, with gait performance depending on numerous aspects of brain functioning that are also relevant to key cognitive processes. The co-occurrence of decline in gait speed and cognitive performance suggests common neural pathways [1].

Gait assessment provides an opportunity to understand the interrelationships between cognitive performance, cognitive impairment and fall risk in older adults [2]. The Timed Up and Go (TUG) test has been shown to be a reliable, valid tool for assessing age-related mobility changes and risk of falls in older adults [3–5]. The time taken to complete the TUG test has been found to be significantly related to ageing, including acceleration and angular velocity measurements [3, 6].

Imaging studies have explored the structural and functional links between brain status, cognition and motor performance. Maidan et al., using a whole-brain approach with principal component regression to identify specific brain areas associated with mobility, detected associations only between cortical thickness and performance in a simple, usual walking task [7]. However, more sensitive functional neuroimaging techniques such as functional MRI (fMRI) and resting state (rs-) fMRI could be used to identify more complex, active compensatory mechanisms [1, 8, 9]. Recent studies have highlighted that gait performance in aging does not rely exclusively on motor systems but also involves the flexible recruitment of cognitive networks. As aging progresses, reconfiguration of large-scale brain networks occurs, with decreased within-network connectivity and increased inter-network connectivity between motor and associative systems [8, 10]. The shift from segregation to integration is thought to reflect a compensatory adaptation aimed at maintaining function despite declining efficiency.

Beyond cognitive systems, two motor networks are central to mobility. The somatomotor hand (SMH) network indexes core sensorimotor function and has been related to gait speed and mobility in aging [1, 9] The cerebellar (CB) network supports on-line motor coordination, timing, and error-based updating,its long-range coupling with associative systems shows inter-individual variability in older adults and relates to cognitive performance [11, 12]. Guided by this evidence, we examine inter-network connectivity between SMH/CB and cognitive networks in relation to TUG performance. In support of this, Geerligs et al. observed increased connectivity between cognitive control and sensorimotor networks in older adults, which was associated with preserved cognitive performance [13].

Different attentional systems contribute differently to gait regulation in ageing. Greater integration with stimulus-driven attentional networks, such as the salience (SAL) and ventral attention networks (VAN), has been linked to more efficient motor adjustments under increased cognitive demands [8, 9]. However, increased coupling with systems supporting sustained attention and task-set maintenance, including the cingulo-opercular network (CON) [14] and dorsal attention network (DAN) [15], has been shown to have more variable associations. This includes reduced gait regularity and also poorer anticipatory postural adjustments during motor tasks, suggesting potential interference with automatic control mechanisms [8, 16]. Altogether, these findings indicate the importance of determining how the balance between functional segregation and integration across motor and cognitive systems shapes gait performance in aging.

Growing evidence suggests that interactions between motor systems and the default mode network (DMN) may also play a critical role in shaping gait performance in later life. The DMN is typically deactivated during externally oriented, goal-directed behaviour, and its functional segregation from task-positive systems is considered essential for efficient cognitive and motor execution [17, 18]. However, in aging, this segregation may be compromised, and increased coactivation between the DMN and motor-related networks has been interpreted as a sign of dedifferentiation [19, 20], potentially interfering with automatic motor processing by sustaining internally oriented activity during movement. Such loss of functional specialization could hinder the rapid, efficient engagement of motor circuits required for stable and adaptive gait [11, 21].

Motor-cognitive internetwork connectivity and TUG performance have been related to cognitive processing [7, 22–24]. In addition, age and educational levels have a potential role in the use of motor-cognitive connectivity resources in TUG performance [8]. The present study aimed to investigate the relationship between motor-cognitive inter-network connectivity and TUG performance in middle-age and older adults and according to their educative levels. In light of the above findings, it also explored the complementary role of executive and memory scores in TUG performance.

Method

Participants

Participants were selected from the 2nd and 3rd cohorts of the CompAS project, an ongoing longitudinal study involving assessment and follow-up of middle-aged and older adults with self-reported cognitive complaints recruited from primary care centres in the Santiago de Compostela health area [22, 23]. Of the 189 participants with complete cognitive, physical and rs-fMRI data, 142 who were not diagnosed with MCI were selected for this study (mean age = 64.78, d.s. = 7.79,107 female, 75.4%). None of the participants had objective cognitive impairment based on their scores on the MMSE and CAMCOG-R according to the normative reference scores for their age and educational level [25, 26]. None of the participants had previous diagnosis of dementia, psychiatric or neurological disorders, severe illness, deafness or blindness. In addition, none of the participants were receiving chemotherapy and none had a history of alcohol or other substance abuse.

Cognitive and gait assessment

Verbal learning memory was measured with The Spanish version of the California Verbal Learning Test (CVLT) [27]. In this test, a list of 16 words is presented orally to participants, who are required to recall the words immediately and after short and long delays. This study includes total immediate recall after each of the 5 presentations of the word list, a short delay free recall (SDFR) score, and a long delay free recall (LDFR) score.

Executive functioning was measured with the Trail Making Test form A (TMT-A), the TMT form B (TMT-B) and a phonological fluency test. In the TMT-A, participants were asked to draw lines to connect the numbers 1 to 25 distributed on a sheet of paper, while attentional processes, motor speed and visuomotor and conceptual tracking were measured. In the TMT-B participants were asked to draw lines to connect the numbers 1 to 13 alternating with the letters A to L, while working memory and attentional shifting were measured [28]. The time taken to complete the test was recorded, as the test score. In the phonological fluency test, applied in accordance with the procedures established by the Spanish Consortium for Normative Data on Ageing (SCAND) [29], each participant was asked to produce as many common words as possible beginning with the letter P in 60 s, excluding proper nouns and repetitions of the same word with different endings.

Mobility was measured with the Timed-Up and Go (TUG) test of dynamic balance. In this test, the person is instructed to get up from a chair from an initial sedentary position, walk a distance of 3 m, turn round bypassing a cone, walk the 3 m back to the chair, and sit down again [5, 22, 23]. A wireless inertial sensor device with Bluetooth connection (Wiva®, Loran Engineering, Bologna, Italy) was used to record and transmit data in the instrumented TUG performance. The Wiva® device includes an accelerometer, a triaxial magnetometer and a gyroscope. Reliable information on the angular velocities reached throughout the test are obtained with the device attached to the body at the level of the lumbar region [30]. The device records the split time of the different TUG phases, including sit-to-stand, gait-to-go, turn around the cone, walk back and stand to sit. During turning, the subject rotates the body 180º to return to the chair. This study includes the total time taken to complete the task, the time taken to turn around and the peak angular velocity (PAV) during turning as dependent variables.

Acquisition parameters

Magnetic resonance imaging (MRI) data were acquired using a 3 T Philips Achieva scanner (Philips Medical Systems). High-resolution structural images were obtained through a sagittal T1-weighted 3D Magnetization-Prepared Rapid Acquisition Gradient Echo (MPRAGE) sequence. Functional images were collected using a gradient echo-planar imaging (EPI) sequence optimized for blood oxygenation level-dependent (BOLD) contrast during resting-state conditions with the eyes open and a centrally presented fixation cross. To reduce head motion, the participant’s head was stabilized using a vacuum cushion. Prior to image acquisition, four dummy volumes were excluded to ensure steady-state magnetization.

For the MPRAGE sequence, the parameters were repetition time (TR) = 7.45 ms, echo time (TE) = 3.40 ms, flip angle = 8º, voxel dimensions = 1 × 1 × 1 mm, field of view (FOV) = 240 × 240 mm² and image matrix = 240 × 240. The EPI sequence was defined by a TR of 2000 ms, TE of 30 ms, flip angle = 87º, 37 interleaved axial slices, voxel size = 3 × 3 × 3.5 mm, FOV = 240 × 240 mm², and a matrix size of 80 × 80 mm.

Data pre-processing and denoising

Structural and functional MRI datasets were pre-processed and analysed using the CONN toolbox (v19c) [31], in combination with SPM12 (Welcome Centre for Human Neuroimaging, London, UK; http://www.fil.ion.ucl.ac.uk/spm/), both implemented in Matlab R2019a.

Preprocessing followed the default pipeline recommended by the CONN toolbox. Functional images were realigned and unwarped, before being spatially normalized to central coordinates (0, 0, 0). Slice-timing correction was applied, and outlier scans were detected. Functional volumes were then directly segmented and normalized to standard Montreal Neurological Institute (MNI) space. Structural images were similarly centred, segmented and normalized. A spatial smoothing step using an 8 mm full width at half-maximum (FWHM) Gaussian kernel was then applied to the functional data.

To account for physiological artefacts, the anatomical CompCor approach (aCompCor) was used [32]. This approach removes noise derived from white matter and cerebrospinal fluid regions, thereby minimizing the impact of physiological fluctuations such as heartbeat and respiration. This denoising strategy has been shown to reduce artefactual anticorrelations while preserving meaningful functional connectivity patterns [33].

Motion-related artefacts were identified using the Artifact Detection Tools (ART) software package [31, 34], with thresholds set at z = ± 5 for global signal changes and 0.9 mm for framewise displacement. Denoising included the regression of nuisance variables, including the principal components derived from white matter and cerebrospinal fluid masks (as estimated by aCompCor), 12 motion parameters, identified outlier scans (scrubbing), and resting-state effects. This step was carried out using Ordinary Least Squares regression. Finally, a temporal band-pass filter (0.008–0.09 Hz) was used to isolate the BOLD signal within the frequency range typically associated with relevant resting-state fluctuations.

Statistical analysis

Inter-network functional connectivity between the somatomotor hand and cerebellar networks and the rest of the other functional networks was calculated using the Power et al. atlas [35], a functional brain parcellation comprising by 264 regions of interest defined based on resting-state connectivity patterns. ROIs were masked to grey matter to ensure the analyses were restricted to this tissue. Standard bivariate correlation measures were used, and correlations of the BOLD signal across the time series were calculated for each ROI pair involving these networks. Fisher's z-transformation was applied to the resulting correlation coefficients, yielding ROI-to-ROI connectivity matrices. The matrices containing Fisher's z-transformed values were then used to quantify inter-network connectivity between the somato-motor hand and cerebellar networks and the remaining networks, following the approach described by Varangis et al. [36]. The spatial distribution of the nodes belonging to each network used to compute inter-network connectivity values is illustrated in Fig. 1.

Fig. 1.

Spatial distribution of the regions of interest (ROIs) used to compute inter-network functional connectivity. Each panel depicts a pairwise combination of the somatomotor hand (SMN, blue) or cerebellar (CB, cyan) networks and one of the following large-scale networks: default mode network (DMN, red), dorsal attention (DAN, magenta), ventral attention (VAN, yellow), frontoparietal control (FPCN, orange), salience (SAL, black), and cingulo-opercular (CON, green). Nodes are displayed on lateral/medial views. These ROI combinations were used to compute z-transformed correlation values representing inter-network connectivity in subsequent analyses

Cognitive tests results were transformed into Z-scores, after which a principal component analysis with Varimax rotation was conducted. The Kaiser–Meyer–Olkin measure (KMO = 0.796) and Bartlett’s test of sphericity (< 0.001) indicated sampling adequacy. Two factors were obtained: Factor 1, Memory Performance, which included CVLT immediate recall, SDFR and LDFR scores, explained 47.41% of the variance; Factor 2, Executive Performance, including TMT-A, TMT-B and fluency performance, explained 32.46% of the variance.

Three linear regression models were constructed including the dependent variables TUG total time, turning time and turning PAV and the independent variables age, sex, years of education, memory and executive performance, and inter-network connectivity of somatomotor hand – DMN, somatomotor hand – SAL, somatomotor hand – cingulo opercular, somatomotor hand – frontoparietal, somatomotor hand – dorsal attentional, somatomotor hand – ventral attentional, cerebellar – DMN, cerebellar – SAL, cerebellar – cingulo opercular, cerebellar – frontopartietal, cerebellar – dorsal attentional, and cerebellar – ventral attentional inter-networks. The R lm function was used for the regressions. To obtain the standardized parameters, the ‘parameters’ package and its function ‘standardize_parameters’ were used with the default refit method, a method based on a complete refit of the model with a standardized version of the data. This method is especially recommended for complex models that include interactions or transformations [37]. Descriptive analyses were conducted prior to regression analysis. Memory and executive performance in CVLT, TMT and phonological fluency were transformed to factorial scores, and collinearity between independent variables were tested using IBM SPSS 29.

To limit multiplicity, we fitted one prespecified multivariable linear model per outcome (TUG total time, turning time, and turning peak angular velocity) including a priori predictors. We report standardized coefficients and 95% confidence intervals (CIs) and interpret effects whose CIs exclude zero. No post-hoc multiple correction was applied,rather, inference emphasizes effect sizes with CIs and converging patterns across outcomes, and robustness was examined in a sensitivity analysis excluding participants with low cognitive performance (performance under expected in at least two cognitive tests according to normative scores [29, 38, 39]). All tests were two-tailed (α = 0.05).

Results

Descriptive statistics are summarized in Table 1, and bivariate relationships among continuous variables are displayed in the correlation‐matrix (Fig. 2). Standardized results of the linear regression models are shown in Table 2 and Fig. 3. Somatomotor hand – ventral attentional (SMN-VAN) and cerebellar – DMN (CB-DMN) inter-network connectivity were significant predictors of the model including TUG total time as the dependent variable. Memory performance and somatomotor hand –salience (SMN-SAL) and –cingulo opercular network (SMN-CON) activation were significant predictors of the turning time model. Age, years of education, somatomotor hand – dorsal attentional (SMN-DAN) and cerebellar – cingulo opercular network (SMN-CON) activation were significant predictors of the model including turning PAV.

Table 1.

Descriptive statistics for demographic, cognitive and resting state functional connectivity Z scores

	Minimum	Maximum	Mean	Standard deviation
Age	51	83	64.78	7.79
Years of education	4	28	13.74	5.27
TUG total time (secs.)	5.56	25.56	10.72	2.95
TUG turning time (secs.)	0.4	3.1	1.38	0.55
TUG turning PAV (º/secs.)	81.2	311.7	164.28	44.91
CVLT total immediate recall (items correctly recalled)	31	76	55.74	10.08
CVLT SDFR (items correctly recalled)	2	16	11.80	3.33
CVLT LDFR (items correctly recalled)	2	16	12.45	3.02
TMT-A (secs.)	19	180	42.85	19.64
TMT-B (secs)	39	260	98.49	48.71
Phonological fluency (correct words)	1	29	15.85	4.94
Memory factor score	−2.20	2.66	0.33	0.80
Executive factor score	−1.39	3.62	−0.10	0.90
Somatomotor hand – DMN	−0.13	0.11	−0.02	0.04
Somatomotor hand – SAL	−0.12	0.11	0.02	0.04
Somatomotor hand – cingulo opercular	−0.04	0.29	0.10	0.05
Somatomotor hand – frontoparietal	−0.15	0.09	−0.01	0.04
Somatomotor hand – dorsal attentional	−0.09	0.15	0.02	0.04
Somatomotor hand – ventral attentional	−0.09	0.15	0.02	0.04
Cerebellar – DMN	−0.32	0.07	−0.05	0.06
Cerebellar – SAL	0.011	0.25	0.04	0.05
Cerebellar – cingulo opercular	−0.09	0.22	0.06	0.06
Cerebellar – frontoparietal	−0.15	0.11	−0.03	0.05
Cerebellar – dorsal attentional	−0.12	0.21	0.04	0.06
Cerebellar – ventral attentional	−0.13	0.14	0.01	0.05

TUG timed-up and go, PAV peak angular velocity, CVLT california verbal learning test, SDFR short delay free recall, LDFR long delay free recall, TMT trail making test, DMN default mode network, SAL salience

Fig. 2.

Bivariate correlation matrix among gait performance measures, demographic factors, cognitive scores and resting-state functional connectivity. Each circle represents one pairwise correlation (r): circle area is proportional to |r|, and colour hue and intensity indicate both direction and strength of the association (blue = negative, orange = positive). All values are rounded to two decimal places. Variables include motor outcomes (TUG total time in seconds; TUG turning time in seconds; TUG turning peak angular velocity [PAV] in degrees/seconds), demographics (Age; Years of education), cognitive performance (Memory performance; Executive performance), and inter-network connectivity for the somatomotor hand network and cerebellar network with the default mode network (DMN), salience network (SAL), cingulo-opercular network, frontoparietal network, dorsal attention network and ventral attention network

Table 2.

Standard coefficients and 95% confidence intervals (between brackets) of linear regression models with TUG total time, turning time and turning PAV as dependent variables (columns) and sociodemographic, cognitive and inter-network connectivity as independent variables (rows)

	TUG total time	TUG turning time	TUG turning PAV
(Intercept)	3.67e-16 (−0.15, 0.15)*	−1.15e-16 (−0.16, 0.16)*	−3.93e-16 (−0.15, 0.15)***
Age	0.20 (−0.04, 0.43)	−0.10 (−0.35, 0.14)	−0.35 (−0.57, −0.12)**
Sex	0.03 (−0.15, 0.22)	−0.04 (−0.23, 0.15)	−0.13 (−0.31, 0.04)
Years of education	−0.13 (−0.32, 0.06)	0.12 (−0.07, 0.32)	0.20 (0.02, 0.38)*
Memory performance	3.63e-03 (−0.22, 0.23)	−0.26 (−0.49, −0.02)*	−0.20 (−0.42, 0.02)
Executive performance	0.20 (−0.01, 0.41)	0.07 (−0.15, 0.29)	−0.06 (−0.27, 0.15)
Somatomotor hand – DMN	0.13 (−0.06, 0.33)	−0.01 (−0.22, 0.19)	−0.07 (−0.26, 0.12)
Somatomotor hand – SAL	0.10 (−0.13, 0.33)	−0.30 (−0.54, −0.06)*	−0.12 (−0.35, 0.11)
Somatomotor hand – cingulo opercular	0.11 (−0.12, 0.34)	0.27 (0.03, 0.51)*	6.21e-03 (−0.22, 0.23)
Somatomotor hand – frontoparietal	−0.10 (−0.29, 0.10)	0.05 (−0.15, 0.25)	0.01 (−0.18, 0.20)
Somatomotor hand – dorsal attentional	0.14 (−0.04, 0.32)	0.18 (−0.01, 0.37)	−0.23 (−0.40, −0.05)*
Somatomotor hand – ventral attentional	−0.24 (−0.45, −0.04)*	−0.03 (−0.24, 0.19)	0.14 (−0.06, 0.34)
Cerebellar – DMN	0.26 (0.05, 0.47)*	0.13 (−0.09, 0.35)	−0.17 (−0.37, 0.04)
Cerebellar – SAL	−0.10 (−0.33, 0.13)	0.10 (−0.14, 0.34)	0.18 (−0.05, 0.41)
Cerebellar – cingulo opercular	0.18 (−0.08, 0.44)	0.09 (−0.36, 0.17)	−0.28 (−0.53, −0.02)*
Cerebellar – frontopartietal	−6.20e-04 (−0.19, 0.19)	−0.12 (−0.33, 0.08)	0.09 (−0.10, 0.28)
Cerebellar – dorsal attentional	−0.08 (−0.27, 0.10)	0.10 (−0.10, 0.29)	0.07 (−0.11, 0.25)
Cerebellar – ventral attentional	0.11 (−0.07, 0.29)	−0.06 (−0.25, 0.12)	−0.12 (−0.30, 0.05)

TUG timed-up and go, PAV peak angular velocity, DMN default model network, SAL salience. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001

Fig. 3.

Standardized regression coefficients (Std) and 95% confidence intervals (CI) for the three linear models predicting components of gait performance: total TUG time (left panel), turning time (middle panel), and turning peak angular velocity (PAV; right panel). Predictors include inter-network resting-state functional connectivity measures, memory performance (MEM), age (AGE) and years of education (YrsEd). Functional connectivity variables are indicated using the following abbreviations: SMN = somatomotor hand network; CB = cerebellar network; DMN = default mode network; SAL = salience network; CON = cingulo-opercular network; DAN = dorsal attention network; VAN = ventral attention network

In a sensitivity analysis excluding 5 participants with low cognitive performance, the overall pattern of findings remained essentially unchanged (see Supplementary Table 1). For TUG total time, the same inter-network predictors were retained (SMN-VAN negatively associated; CB – DMN positively associated). For turning time, the effects of SMN – SAL (negative) and SMN – CON (positive) persisted, and one additional association emerged: SMN – DAN was positively related to turning time, consistent with the PAV model where the somatomotor hand – dorsal attentional inter-network predicted lower PAV. Finally, in the PAV model effect directions were stable: age (negative) and years of education(positive) remained significant, while the CB – CON effect attenuated to a nonsignificant trend. The memory effect on turning time shifted from significant to marginal (same direction). Overall, the sensitivity analysis supports the robustness of the main results with only minor changes in statistical significance.

Discussion

We evaluated the role of memory and executive performance, motor and cognitive rs-fMRI connectivity, aging, sex and years of education in TUG performance. Regression models for TUG total time, turning time and turning PAV were all significant. Different patterns of motor-cognitive activation were significant predictors: SMN-VAN and CB–DMN connectivity for TUG total time; SMN-SAL and SMN-CON connectivity for turning time; and SMN-DAN and CB-CON connectivity for turning PAV.

Among these, greater SMN–VAN connectivity was associated with better TUG performance. This result is consistent with previous findings suggesting that integration between sensorimotor systems and bottom-up attentional networks may support gait performance in aging, potentially by facilitating the allocation of attentional resources during movement. Geerligs et al. observed increased connectivity between cognitive-control and somatomotor networks in older adults, which was associated with preserved cognitive task performance [13]. Similarly, Droby et al. reported that usual walking performance was related to connectivity between motor and sustained attention networks, while dual-task walking relied more strongly on divided attention networks such as the VAN [8]. Jockwitz and Caspers emphasized that age-related increases in inter-network communication may reflect compensatory mechanisms aimed at maintaining function despite reduced segregation [40]. In line with these findings, our results may reflect a compensatory adaptation, whereby increased SMN–VAN connectivity supports the optimization of attentional resource allocation during dynamic tasks assessing parameters such as gait. This interpretation is consistent with the idea that preserved or enhanced integration involving the VAN is functionally relevant for mobility in older adults.

Beyond its traditional role in motor coordination, the cerebellum has emerged as a key structure involved in a wide range of cognitive processes, including memory, attention and executive function. In healthy older adults, the cerebellum exhibits high inter-individual variability in its functional connectivity with cortical brain networks. Notably, many of the long-range connections -especially those linking the cerebellum with associative networks such as the frontoparietal and default mode networks- have been associated with cognitive performance across multiple domains [11, 12]. This suggests that cerebellar integration with large-scale networks is functionally meaningful and may reflect individual differences in adaptive compensation during the aging process.

At rest, CB-DMN connectivity typically exhibits an anti-correlated relationship, which reflects the functional segregation between internally and externally oriented processes [11]. The DMN is typically deactivated during externally oriented, goal-directed behaviour, and its functional segregation from task-positive systems is considered essential for efficient cognitive and motor execution. However, in aging, this segregation may be compromised: increased coactivation between the DMN and motor-related networks has been interpreted as a sign of dedifferentiation [19, 20], potentially interfering with automatic motor processing by sustaining internally oriented activity during movement. Such loss of functional specialization could hinder the rapid and efficient engagement of motor circuits required for stable and adaptive gait [8, 11, 21]. Our observation that increased CB-DMN connectivity predicted slower TUG performance, together with the findings of previous studies, suggests that CB–DMN connectivity may be an important marker of suboptimal neural organization in later life.

Memory performance was a significant predictor of turning time, highlighting the involvement of cognitive resources in the control of complex motor subcomponents such as turning. The association between gait and memory is supported by a shared neural substrate including the right hippocampus; e.g., studies of larger hippocampal volumes in London taxi drivers support the link between hippocampus and spatial learning and spatial memory processes [41, 42]. Rosso et al. found that right hippocampus was the only region related to both gait slowing and cognitive impairment in a cohort of old adults followed over 14 years [43]. Nevertheless, although verbal list learning is often linked to left medial temporal structures, prior longitudinal and cross-sectional work indicates that hippocampal integrity relates to gait without a strict laterality rule. Longitudinally, right hippocampal volume has been shown to predict subsequent gait slowing [43], and evidence from navigation research supports bilateral hippocampal contributions to spatial mapping and locomotor control [41, 42, 44]. Accordingly, our memory–gait association should not be over-interpreted as strictly left-lateralized.

Additionally, two functional connectivity features emerged as significant predictors of turning time: increased connectivity between the SMN–SAL was associated with shorter turning time, while increased connectivity between the SMN–CON was associated with longer turning time. Previous studies have reported age-related increases in connectivity between large-scale networks, including SAL and SMN, which may reflect compensatory integration processes in the aging brain [8, 40, 45]. In the present study, greater SMN–SAL connectivity was associated with faster turning performance, suggesting a potentially adaptive role of this interaction in supporting motor adjustments. This is consistent with prior findings linking the integrity of both SMN and SAL to better performance in motor tasks requiring attentional control [46, 47]. These outcomes were also related to attention control capacity, particularly distractor resistance, highlighting the role of salience-based mechanisms in facilitating motor performance under attentional demands [47]. Additionally, SAL has been implicated in stimulus detection and bottom-up attentional shifts during dynamic gait phases, including dual-task walking [8]. The observed increased SMN–SAL connectivity may facilitate timely attentional reorientation and enhance motor control under increased cognitive demands, reflecting flexible and efficient integration between sensorimotor and attentional systems in healthy aging.

Increased SMN–CON connectivity was associated with slower turning performance. This is consistent with previous findings suggesting that increased coupling between motor and control-related networks may interfere with automaticity of gait in older adults. For example, Droby et al. reported that increased SMN–CON connectivity was linked to reduced stride regularity during routine walking, a pattern interpreted as reflecting compensatory (potentially maladaptive) recruitment of control systems to support basic motor functions [8]. Evidence from developmental research also supports this interpretation. Cignetti et al. associated stronger connectivity between the dACC/msFC (a hub of CON) and bilateral motor areas (M1, SMA) with poorer anticipatory postural adjustments in adults, suggesting that increased SMN–CON coupling may reflect over-involvement of cognitive control in motor planning [16]. In this context, our finding may indicate that, in older adults, stronger CON engagement during turning could reduce execution efficiency by overriding more automatic, sensorimotor-driven strategies.

Age and years of education are significant predictors of turning PAV, a parameter previously associated with fall risk in older adults [4, 6]. Our observation that greater SMN–DAN connectivity was related to lower angular velocity suggests that increased integration between motor and top-down attentional systems may not always support more efficient performance. This is consistent with prior observation of a negative association between SMN–DAN connectivity and gait regularity under dual-task conditions in aging populations [8]. Moreover, in the sensitivity analysis excluding low cognitive performers, an additional association emerged: SMN–DAN connectivity was also positively related to turning time, consistent with its inverse relation with PAV. This convergence across outcomes strengthens the interpretation that excessive engagement of top-down attentional systems may compromise the automaticity of gait. It is possible that when cognitive reserve is limited, the participants tend to recruit attentional control mechanisms, disrupting motor automaticity. Turning tasks, which demand anticipatory adjustments and postural reorientation, may be particularly vulnerable to such imbalances. In line with this view, we also observed that a higher number of years of education -often considered a proxy for cognitive reserve- was positively associated with turning performance, suggesting that individuals with greater reserve may rely less on compensatory control and preserve more efficient, automated motor strategies. This interpretation is consistent with findings of Droby et al., who reported a positive association between education and gait regularity [8].

Our observation of increased connectivity between the CB and CON networks was associated with reduced angular velocity during turning may reflect a compensatory mechanism that, while supporting higher-order cognitive functions, interferes with efficient motor execution. This interpretation is consistent with evidence showing that in healthy aging, functional connectivity between these two networks tends to decrease [48]. Moreover, prior research has linked stronger CB–CON connectivity to better fluid intelligence in adults, reinforcing the idea that this integration may reflect cognitive compensation [49]. Clark et al. reported that connectivity between CB and CON does not follow a linear pattern with age, as some connections become stronger while others become weaker in older individuals [50]. Such variability may reflect diverse compensatory processes depending on individual aging trajectories and/or the specific demands of the task. In the context of a motor task like turning, greater engagement of the cingulo-opercular system -a network involved in cognitive control and salience detection- may interfere with the automatic motor coordination typically supported by the cerebellum, ultimately leading to slower and less efficient performance. However, in the sensitivity analyses excluding participants with low cognitive performance, the CB–CON association attenuated to trend level (p = 0.083) but remained directionally consistent (see Supplementary Table 1), thus we must treat this finding with caution.

Our findings indicate different contributions of attentional systems to gait performance in aging. Specifically, increased functional connectivity between motor networks and those implicated in divided attention (SAL, VAN) was associated with better motor performance, such as faster turning and reduced total TUG timer. By contrast, greater connectivity between motor networks and systems involved in sustained attention, task-set maintenance, and internally directed processes (CON, DAN, DMN) was linked to poorer motor outcomes, including slower turning and reduced angular velocity [8, 14, 15, 51]. These opposing patterns suggest that while integration with reactive, stimulus-driven attentional networks may facilitate adaptive responses during dynamic gait phases, coupling with sustained control systems may reflect over-recruitment or inefficiency, potentially disrupting automatic motor execution. Altogether, our results highlight the dual nature of neurofunctional adaptations in aging. While some forms of network integration support performance, others may emerge as maladaptive responses in certain contexts.

Although the executive-function composite did not reach significance, we observed a small trend in the model predicting total TUG time (standardized β ≈ 0.20; p = 0.065). This pattern persisted in sensitivity analyses excluding low cognitive performers but remained non-significant. The heterogeneous nature of the composite (phonemic fluency, processing speed, and set-shifting) may have diluted component-specific variance, limiting power to detect effects. Notably, at the neural level, inter-network coupling with control-related systems (e.g., CON, DAN) showed clear associations with gait dynamics in the same direction of the executive-function composite score (slower turning and reduced peak angular velocity), suggesting that control-network engagement may influence motor regulation.

This study has several limitations that should be acknowledged. First, the cross-sectional design precludes conclusions about causality or changes over time. Future research should adopt longitudinal approaches to better characterize how patterns of network integration and segregation evolve with aging and how they relate to trajectories of motor and cognitive performance. It would also be useful to examine the extent to which these functional connectivity patterns interact with proxies of cognitive reserve, such as education and lifestyle factors, to modulate age-related changes in gait. The exclusion of major neurological and psychiatric disorders, intended in the study design to ensure the required homogeneity of the data, implies limitations on the generalisability of the data to the entire population over the age of 50. Somatic symptoms disorders were not systematically assessed in this sample, and performance validity were not assessed, which could be a major concern in the cognitive and neuropsychological evaluation of these patients. Finally, the memory performance score was based on multiple scores from a single measure, whereas the executive performance score was based on different measures, including processing speed, set-shifting and phonological fluency. Furthermore, the reliability of the memory measure has been extensively studied within the CompAS [52, 53], whereas executive performance has been less studied in this cohort study. Accordingly, future works should evaluate its reliability and dimensionality, also considering component-specific executive performance composites and larger samples to detect small effects.

In conclusion, improving the comprehension of those central processes involved in the performance of a TUG test, including total speed, turning speed and angular velocity when turning, contribute to increase our knowledge about the involvement of cognitive and motor-cognitive processes in mobility risks such as risk of falling, and about the role of processing resources in the interplay between cognitive and motor tasks in daily life throughout the lifespan. Findings regarding the positive effect of greater functional connectivity between motor networks and those involved in divided attention, and the negative effect of greater connectivity between motor networks and sustained attention networks, along with the DMN, support the ongoing debate about changes in attentional processes, shifting from segregation to integration, throughout the ageing process. Although the sample size is relatively large for a neuroimaging study, replication in larger cohorts is warranted, particularly given the number of predictors included in the models and the inter-individual variability inherent to aging populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Conceptualization: BVL, SGA, CH, YS, DF; Data Curation: BVL, AF; Formal Analysis: BVL, SGA, DF; Investigation: BVL, SGA, AF, DF; Methodology: BVL, SGA, CH, DF; Supervision: SGA, YS, DF; Visualization; BVL; Writing – original draft: BVL, DF; Writing – review & editing: SGA, AF, CH, YS.

Funding

The CompAS study is supported though FEDER funds provided by the Spanish National Research Agency (Spanish Ministry of Science, Innovation and Universities : Refs. PID2020-114521RB-C21 and PID2023-151659OB-C21) and by the Galician Government (GI-1807-USC-2021-PG011: Ref. ED431C 2021/04).

Data Availability

The datasets analyzed during the current study are available from the first author upon reasonable request.

Declarations

Competing interests

There are no competing interests.