Walking performance is a robust predictor of adverse outcomes in aging (Studenski et al., 2011) including dementia (Quan et al., 2017). Furthermore, gait is dependent on cognitive control of cortical resources (Paraskevoudi et al., 2018; Yogev-Seligmann et al., Reference Yogev-Seligmann, Hausdorff and Giladi2008); notably, gait speed has been associated with attention and executive functions (Atkinson et al., Reference Atkinson, Rosano, Simonsick, Williamson, Davis, Ambrosius, Rapp, Cesari, Newman, Harris, Rubin, Yaffe, Satterfield and Kritchevsky2007).
Dual-task walk designs, combining walking with a cognitive interference task, have been used to experimentally manipulate resources of executive control of walking by imposing competing task demands (Holtzer et al., 2012, 2014). Poor dual-task walking, in particular, predicts progression to dementia (Montero-Odasso et al., 2017), frailty, disability, and mortality (Verghese et al., 2012) in aging. Further, dual-task designs provide greater ecological validity than single-task walking as multisensory interference is present in natural environments. The prefrontal cortex (PFC) is implicated in cortical control of attention and executive functions (Koechlin et al., 2003), and has been shown to increase in activation from single-tasks to dual-task walking (Holtzer et al., 2015). Consistent with the key role of the PFC in executive functions, functional near-infrared spectroscopy (fNIRS) has been used extensively to quantify changes in oxygenated hemoglobin (HbO2) in the PFC during walking. Specifically, a recent consensus guide (Menant et al., 2020) and meta-analysis (Bishnoi et al., Reference Bishnoi, Holtzer and Hernandez2021) demonstrated reliable increases in fNIRS-derived HbO2 in the PFC in dual compared to single-task walking conditions.
Central tendency measures of change in HbO2 levels from single-tasks to dual-task walking have been examined most often, however, variability in brain activation may provide additional information (Holtzer et al., 2020). Intraindividual variability (IIV) is psychometrically distinct from central tendency measures (Nesselroade & Salthouse, 2004) and has been associated with poor clinical outcomes in aging (Costa et al., Reference Costa, Dogan, Schulz and Reetz2019). Much of the literature on IIV has focused on variability in cognitive performance, showing that cognitive IIV increases in aging and is predictive of cognitive decline (Haynes et al., 2017). However, IIV in gait performance is also increased in aging (Smith et al., 2017) and associated with negative outcomes (Moon et al., 2016).
Research on neural IIV is limited and less conclusive (Dinstein et al., Reference Dinstein, Heeger and Behrmann2015; Dubois & Adolphs, Reference Dubois and Adolphs2016; MacDonald et al., 2006; Uddin, 2020). Traditional functional neuroimaging measurements of the hemodynamic response utilize an average signal which does not consider variability of activity within individuals (Garrett et al., Reference Garrett, Kovacevic, McIntosh and Grady2010). While some studies show increased IIV associated with better performance (Garrett et al., Reference Garrett, Kovacevic, McIntosh and Grady2013), other studies show negative or differential associations between neural IIV and behavior (Boylan et al., Reference Boylan, Foster, Pongpipat, Webb, Rodrigue and Kennedy2021; Guitart-Masip et al., 2016). The mixed findings may underscore distinctive relationships between neural IIV and cognitive domain, task difficulty, age, and disease (Armbruster-Genç et al., Reference Armbruster-Genç, Ueltzhöffer and Fiebach2016; Garrett et al., Reference Garrett, Kovacevic, McIntosh and Grady2013; Grady & Garrett, 2018; Guitart-Masip et al., 2016). For instance, increased neural IIV may be adaptive in learning, but excessive IIV may have negative associations (Boylan et al., Reference Boylan, Foster, Pongpipat, Webb, Rodrigue and Kennedy2021; Dinstein et al., Reference Dinstein, Heeger and Behrmann2015; Steinberg et al., 2022). We have previously found that increased neural IIV in the PFC from single-tasks to dual-task walking was greater in men and people with cognitive impairment (Holtzer et al., 2020).
Greater neural activation may reflect proper engagement due to increased task difficulty. However, when task-related increases are greater than expected and without accompanying increases in behavioral performance, this may reflect neural inefficiency (Haier et al., 1988; Neubauer & Fink, 2009). As healthy aging is accompanied by atrophy of gray matter in the frontal cortex and throughout the brain (Giorgio et al., 2010; Salat et al., 2004), reduced brain integrity may be related to functional changes in executive control of gait (Burzynska et al., Reference Burzynska, Nagel, Preuschhof, Gluth, Bäckman, Li, Lindenberger and Heekeren2012). Consistent with the neural inefficiency hypothesis, over-activation during dual-task walking may be a compensatory function to meet task demands with limited neural resources (Daselaar et al., Reference Daselaar, Iyengar, Davis, Eklund, Hayes and Cabeza2015). Previous studies have shown that increased PFC activation in dual-task walking was related to reduced gray matter volume (Wagshul et al., 2019), cortical thickness (Ross et al., 2021), and white matter integrity (Lucas et al., 2018) in older adults.
The current study was designed to address an important gap in the literature concerning the interaction between structural brain integrity and IIV in cortical activation of gait in older adults. Specifically, we examined the moderating effect of cortical thickness in the PFC and other regions on IIV in PFC activation from single-tasks to dual-task walking, while adjusting for behavioral performance to capture neural efficiency. We focused on cortical thickness as it is related to executive functioning, gait, and brain atrophy in aging (Burzynska et al., Reference Burzynska, Nagel, Preuschhof, Gluth, Bäckman, Li, Lindenberger and Heekeren2012; Maidan et al., 2021; Salat et al., 2004). Cortical thickness is genetically and phenotypically distinct from gray matter volume measurement and is suggested to be more sensitive to age-related change (Hutton et al., 2009; Winkler et al., 2010). Our primary aim was to examine cortical thickness in the PFC as a moderator of IIV from single-tasks to dual-task walking. Our secondary aim was to examine the moderating effect of cortical thickness in other cortical regions. We hypothesized that increased neural IIV in the PFC from single to dual-task walking would be associated with thinner cortex in the PFC and in other brain regions implicated in cortical control of walking. While previous findings have implicated task-related changes to integrity of cortical regions involved in sensory, motor, and cognitive processing (Ross et al., 2021), we predicted that the examination of IIV rather than central tendency measures would provide novel additional information regarding the complex interaction of structural and functional brain correlates of gait.
Method
Participants
A subset of participants from “Central Control of Mobility in Aging” (CCMA) who completed an MRI protocol were included. Procedures for the CCMA study were described previously (Holtzer et al., 2014). Briefly, older adults were identified from population lists in Westchester country, New York, USA. After being mailed a letter, potential participants were contacted by telephone during which verbal consent and initial eligibility were obtained via structured interview. Screening included assessment of cognitive, medical, psychological, and physical status. Eligible participants (age ≥ 65 years) completed in-person visits comprised of comprehensive psychological, mobility, functional, and neuropsychological assessments. Written informed consent was obtained the first study visit. The dual-task walking protocol was completed within one session, while the MRI protocol was completed on a separate visit.
Exclusion criteria were: current or history of severe neurological or psychiatric disorder, significant impairment in vision or hearing, dementia, inability to speak English, inability to ambulate independently, and current or anticipated medical procedures that would hinder ambulation. Additional MRI exclusion criteria were left-handedness and MRI contraindication (e.g., presence of metal in the body and tolerance of the MRI procedure). A total of 73 right-handed older adults completed the MRI protocol. All study procedures were in accordance with the ethical standards of the 1964 Helsinki declaration and its later amendments and approved by the institutional review board of Albert Einstein College of Medicine.
Measures
Dual-task walking protocol
The walking protocol included three task conditions: single-task walk (STW), single-task alpha (STA), and dual-task walk (DTW). All tasks were completed on a 4 × 20 foot electronic walkway. For the STW condition, participants were instructed to walk at their normal pace, walking three continuous counterclockwise loops. For the STA condition, participants were instructed to stand in place while reciting alternate letters of the alphabet out loud (A, C, E, …) for 30 seconds. For the DTW condition, participants were instructed to perform the two tasks simultaneously and pay equal attention to both tasks to minimize task prioritization. Task order was counterbalanced across participants using a Latin square design. This walking protocol has been well-validated (Holtzer et al., 2014).
Quantitative gait assessment
Gait was assessed using a 4 × 20 foot Zeno electronic walkway (Zenometrics, LLC, Peekskill, NY) in conjunction with ProtoKinetics Movement Analysis Software (PKMAS). Entry and exit points of footfalls were determined algorithmically under single and dual-task walk conditions, allowing for extraction of stride velocity and walk time (England et al., Reference England, Verghese, Mahoney, Trantzas and Holtzer2015). Split-half intra-class correlations of quantitative gait measurements in both walking conditions were greater than 0.95, revealing excellent internal consistency (Holtzer et al., 2015).
Functional near-infrared spectroscopy
As in previous studies, the fNIRS Imager 1100 (fNIR Devices, LLC, Potomac, MD) measured oxygenated (HbO2) and deoxygenated (Hb) hemoglobin levels in the PFC during all DTW protocol tasks (Bunce et al., Reference Bunce, Izzetoglu, Izzetoglu, Onaral and Pourrezaei2006; Holtzer et al., 2015). The fNIRS sensor contains four light-emitting diode light sources and 10 photoreceptors 2.5 cm apart, allowing for 16 channels of data collection placed using the standard international 10–20 system (Ayaz et al., Reference Ayaz, Izzetoglu, Platek, Bunce, Izzetoglu, Pourrezaei and Onaral2006). Light sensors emitted peak wavelengths at 730, 805, and 850 nm and data were collected at a 2-Hz sampling rate.
fNIRS preprocessing and hemodynamic signal extraction methods have been previously described (Izzetoglu & Holtzer, 2020). Briefly, this included identification and elimination of saturation, dark current conditions, or extreme noise artifacts through visual inspection. Daubechies 5 (db5) wavelet for spiky noise suppression was applied to remove motion artifacts from 730 and 850 nm wavelengths (Molavi & Dumont, 2012). The modified Beer–Lambert law was used to calculate changes from artifact removed measurements, and account for age and wavelength adjusted differential pathlength factor (DPF) and wavelength and chromophore dependent molar extinction coefficients (ϵ) (Izzetoglu & Holtzer, 2020; Kim & Liu, 2007; Scholkmann & Wolf, 2013). Finally, baseline and physiological artifacts were removed by applying spline filtering (Scholkmann et al., 2010) followed by a finite impulse response low-pass filter with cutoff frequency at 0.08 Hz.
Data points were extracted separately for each task at all 16 channels. E-Prime 2.0 software (Psychology Software Tools, Inc.) synchronized fNIRS signal extraction with PKMAS data. Participants were instructed to stand still, counting silently, with a fixed gaze to compare task-related changes to a baseline measure of HbO2 (Holtzer et al., 2015). For the current study, we used HbO2 as the measure of neural activation in the PFC instead of Hb due to its better reliability and sensitivity to locomotion-related cerebral activity (Miyai et al., 2001).
Measurement of Intraindividual Variability
IIV of PFC activation was operationalized by calculating standard deviation (SD) of fNIRS-derived HbO2 during the first 30 seconds of each task. This time period ensured comparability across task conditions, as walking time under STW and DTW depended on participants’ gait speed, and STA was fixed at 30 seconds. This resulted in 61 data points for each channel, as data was collected every 0.5 seconds (2-Hz sampling rate). IIV could not be calculated using relative measures of dispersion (e.g., coefficient of variation) because several participants’ mean HbO2 approached zero (i.e., no increase from the individual baseline to the experimental condition), most often under STW which imposes significantly less cognitive demands. Analyses adjusted for mean HbO2 to address this potential limitation.
Magnetic resonance imaging
Magnetic resonance imaging was performed at Albert Einstein College of Medicine’s Gruss Magnetic Resonance Research Center (Bronx, NY) in a 3T Philips scanner (Achieva TX; Philips Medical Systems, Best, the Netherlands) equipped with a 32-channel head coil. Cortical thickness was extracted from T1-weighted images (MPRAGE – TE/TR/TI = 4.6/9/8/900 ms, voxel size 1 mm isotropic, SENSE acceleration factor 2.6).
Cortical thickness measures and cortical segmentation was extracted from all participants using the FreeSurfer image analysis suite (https://surfer.nmr.mgh.harvard.edu) (Fischl, Reference Fischl2012). Preprocessing included brain extraction, Talairach transformation, subcortical segmentation, identification of gray-white matter boundaries, and atlas registration. These methods were detailed previously (Dale et al., Reference Dale, Fischl and Sereno1999; Fischl et al., Reference Fischl, Salat, Busa, Albert, Dieterich, Haselgrove, van der Kouwe, Killiany, Kennedy, Klaveness, Montillo, Makris, Rosen and Dale2002; Fischl et al., Reference Fischl, van der Kouwe, Destrieux, Halgren, Ségonne, Salat, Busa, Seidman, Goldstein, Kennedy, Caviness, Makris, Rosen and Dale2004). The 68 regions identified by FreeSurfer’s cortical parcellation tools (Desikan et al., Reference Desikan, Ségonne, Fischl, Quinn, Dickerson, Blacker, Buckner, Dale, Maguire, Hyman, Albert and Killiany2006) were visually inspected in FSLeyes by overlaying the segmentation on each subject’s T1 image (Smith et al., 2004). Surface-based smoothing was applied at FWHM = 5 mm prior to extraction, and cortical thickness values at each region were mean centered prior to statistical analysis.
Covariates
We adjusted for factors that may impact DTW performance, cognition, and brain integrity, and for behavioral performance to evaluate inefficiency. Covariates were age, sex, global cognitive functioning, global health, and correct letter generation performance and walking performance under the DTW condition. Global cognitive functioning was measured using the total score from the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) (Randolph et al., 1998). Global Health Score (GHS) is a summary score ranging from zero to 10, indicating presence or absence of 10 health conditions (diabetes, chronic heart failure, arthritis, hypertension, depression, stroke, Parkinson’s disease, chronic obstructive lung disease, angina, and myocardial infarction) taken via self-reported interview (Holtzer et al., 2008). Cognitive performance under DTW was assessed by dividing the correct letters generated by time walked, and walking performance was measured as stride velocity extracted from the PKMAS system. Statistical models were further adjusted for mean DTW HbO2.
Statistical analysis
Analyses were performed utilizing SPSS statistical software package (version 26; SPSS, Inc., Chicago, IL). Linear mixed models (LMEMs) were used to examine the main effects of task and cortical thickness on HbO2 IIV, as well as the moderating effect of cortical thickness. These models included data from all 16 channels and accounted for correlations across repeated measures within the DTW paradigm. The outcome was HbO2 IIV, with task entered as fixed effects, channel entered as random effects, and task by channel entered as repeated effects. A compound symmetry covariance structure was selected to account for the correlation between HbO2 measured among channels within task within each person.
Sixty-eight LMEMs were run, one for each cortical region, to examine the moderating effect of cortical thickness across the cortex on the change in HbO2 IIV from STW and STA to DTW. The moderating effect of cortical thickness, entered as a covariate, on the change in fNIRS-derived HbO2 IIV across task conditions was assessed via three-level 2-way interactions of regional cortical thickness by task. For all models, DTW HbO2 IIV was the reference group. Analyses adjusted for all covariates described above. False discovery rate (FDR) was used to adjust for multiple comparisons (Benjamini & Hochberg, Reference Benjamini and Hochberg1995).
Results
A total of 73 participants completed the MRI protocol. Participants were excluded for the following reasons: more than one year between MRI and fNIRS (n = 8), poor quality fNIRS data (n = 7), and data exploration outliers (e.g., unusually high variance in gait velocity or HbO2; n = 3). Thus, 55 participants (mean age = 74.84 ± 4.97 years; % female = 49.1) had complete data available for both the DTW and MRI protocols and were included in analyses. Participants were relatively healthy (GHS mean = 1.36 ± 1.08) with average global cognitive functioning (RBANS total score mean = 92.71 ± 11.28). Complete descriptive statistics of the sample are included in Table 1. Descriptive statistics of regional cortical thickness values extracted from FreeSurfer’s cortical parcellation tools are included in Table 2.
Demographic characteristics of the study sample
Note. N = 55. RBANS = repeatable battery for the assessment of neuropsychological status; STW = single-task-walk; STA = single-task-alpha; DTW = dual-task-walk; HbO2 = oxygenated hemoglobin.
a N = 53 for stride velocity measures.
Cortical thickness of the study sample
a Cortical Parcellation based on the Desikan-Killiany atlas (Desikan et al., Reference Desikan, Ségonne, Fischl, Quinn, Dickerson, Blacker, Buckner, Dale, Maguire, Hyman, Albert and Killiany2006).
HbO2 IIV increased from STW to DTW (estimate = −0.07 µM, p < .001, 95% CI [−0.09, −0.05]) and from STA to DTW (estimate = −0.03 µM, p = .005, 95% CI [−0.04, −0.01]). Split-half intra-class correlations revealed excellent internal consistency of HbO2 IIV in each task (STA = 0.82; STW = 0.84; DTW = 0.87).
After confirming the main effect of task on HbO2 IIV, full models including the main and moderating effects of task and cortical thickness were run for each region. The primary analyses with the 12 prefrontal regions, the caudal middle frontal, rostral middle frontal, frontal pole, superior frontal, lateral orbitofrontal, and medial orbitofrontal regions (left and right), are displayed in Table 3. All PFC regions moderated the task effect of HbO2 IIV except for the bilateral orbitofrontal cortex, both lateral and medial regions.
Adjusted main and interaction effects of PFC thickness and task HbO2 IIV
Note. DTW is the reference group. For all results, including covariates, see supplemental materials. HbO2 = oxygenated hemoglobin; IIV = intraindividual variability; STW = single-task-walk; STA = single-task-alpha; DTW = dual-task-walk.
Secondary analyses examined the 54 remaining regions of the cortex. All models were FDR-corrected (p < .02). Models in which the moderating effects of cortical thickness outside the PFC were significant are included in Table 4. In summary, 16 regions of the left hemisphere and 21 regions of the right hemisphere significantly moderated task effects of HbO2 IIV. Full models including covariates and nonsignificant effects can be found in Supplementary Table 1. In all models, the effect of sex was significant (estimate ∼ 0.09, p < .05), showing that males had higher HbO2 IIV across tasks.
Significant adjusted interaction effects of posterior cortical thickness and task HbO2 IIV
Note. DTW is the reference group. Only models with significant FDR-corrected interaction effects are displayed. For all results, see supplemental materials. HbO2 = oxygenated hemoglobin; IIV = intraindividual variability; STW = single-task-walk; STA = single-task-alpha; DTW = dual-task-walk.
To display the distinct and overlapping regions, pictorial representations of the results are shown in figures. Regions where cortical thickness significantly moderated the task effect of HbO2 IIV from STW to DTW are shown in Figure 1, and from STA to DTW in Figure 2. Significant regions spanned the cortex: PFC: bilateral caudal middle frontal, bilateral rostral middle frontal, bilateral superior frontal, bilateral frontal pole; non-PFC frontal lobe: left caudal anterior cingulate, left pars orbitalis, bilateral pars triangularis, right pars opercularis, right precentral; parietal lobe: bilateral precuneus, bilateral superior parietal, right inferior parietal, right isthmus; temporal lobe: bilateral banks of the superior temporal sulcus, bilateral inferior temporal, bilateral temporal pole, right fusiform, right middle temporal, bilateral superior temporal; occipital lobe: left cuneus, left lingual, bilateral pericalcarine, right lateral occipital, right lingual; and right insula.
Regions in which cortical thickness significantly moderated the DTW vs. STW task effect on change in HbO2 IIV. Note. FDR-corrected significant regions are highlighted in color. Yellow-red colors indicate positive interaction effects, while blue color indicates negative interaction effects. STW = single-task-walk; DTW = dual-task-walk; HbO2 = oxygenated hemoglobin; IIV = intraindividual variability. Drawings generated using BrainPainter (Marinescu et al., 2019).
Regions in which cortical thickness significantly moderated the DTW vs. STA task effect on change in HbO2 IIV. Note. FDR-corrected significant regions are highlighted in color. Yellow-red colors indicate positive interaction effects, while blue color indicates negative interaction effects. STA = single-task-alpha; DTW = dual-task-walk; HbO2 = oxygenated hemoglobin; IIV = intraindividual variability. Drawings generated using BrainPainter (Marinescu et al., 2019).
In all models the main effect of task was negative, indicating that HbO2 IIV increased from single-tasks to DTW. For the majority of cortical regions, the main effect of thickness was negative and the interaction effects of task by thickness were positive, indicating that increased HbO2 IIV in the PFC was associated with thinner cortex and that HbO2 IIV increased more from single-tasks to DTW when individuals had thinner cortex in these regions. However, a number of regions showed a positive main effect of thickness on HbO2 IIV and a negative interaction effect of task by thickness, indicating that higher HbO2 IIV and an attenuated increase from STW to DTW were associated with greater thickness in these regions (left banks of the superior temporal sulcus, left superior temporal gyrus, bilateral temporal poles, right isthmus, right middle temporal gyrus, and right insula). Finally, a few regions demonstrated positive thickness main effect and interaction effects, indicating that higher IIV and a greater increase from single to dual task were associated with greater thickness in these regions (STA vs. DTW: left cuneus, right banks of the superior temporal sulcus, right middle temporal, and right pars opercularis models; STW vs. DTW: left pericalcarine).
Discussion
We examined the moderating effects of regional cortical thickness on changes in PFC activation IIV from single-tasks to DTW. Findings revealed that thinner cortex in the PFC, and in specific regions in all cerebral lobes, was related to greater increases in fNIRS-derived HbO2 IIV in DTW compared to STA and STW. This supported our hypothesis that lower cortical thickness would be related to increased IIV and poor dual-task PFC efficiency. As we controlled for gait and cognitive performance, the moderation of HbO2 IIV by cortical thickness suggests inefficient PFC response.
While previous literature on IIV in aging suggests that greater variability is indicative of poor outcomes (Costa et al., Reference Costa, Dogan, Schulz and Reetz2019), the literature on neural IIV is less conclusive. It is important to note that neural IIV can be measured as hemodynamic (e.g., fNIRS or fMRI) or electrophysiological (e.g., EEG) and different findings arise depending on the method used, likely reflecting different physiological processes (Kumral et al., 2020). Greater IIV in the hemodynamic response has been associated with greater task difficulty, greater variability in movement, and cognitive impairments (Haar et al., 2017; Holtzer et al., 2020). Neural IIV has been suggested to be task and region-dependent (Armbruster-Genç et al., Reference Armbruster-Genç, Ueltzhöffer and Fiebach2016; Boylan et al., Reference Boylan, Foster, Pongpipat, Webb, Rodrigue and Kennedy2021; Guitart-Masip et al., 2016). It has been postulated that neural IIV may follow a u-shaped curve, with some variability necessary for learning but too much variability being disadvantageous and associated with clinical disorders (Dinstein et al., Reference Dinstein, Heeger and Behrmann2015). Indeed, one study showed that increased neural IIV during DTW was associated with cognitive impairments and greater behavioral IIV (Holtzer et al., 2020). A recent study, however, found that people with Parkinson’s disease demonstrated greater neural IIV during normal walking and that this IIV did not increase from single- to DTW. Over repeated learning trials, DTW neural IIV increased in Parkinson’s disease and decreased in healthy controls (Maidan et al., 2022). In sum, recent research supports the notion that a degree of neural IIV is adaptive, but that excessive IIV may be indicative of inefficiency.
Most models in the current study followed the pattern of lower cortical thickness associated with greater DTW PFC IIV. However, in five models of the frontal, temporal, and occipital lobes, greater thickness in these regions was associated with greater increases in DTW IIV. This is in line with previous studies showing regional differences in neural variability (Armbruster-Genç et al., Reference Armbruster-Genç, Ueltzhöffer and Fiebach2016; Boylan et al., Reference Boylan, Foster, Pongpipat, Webb, Rodrigue and Kennedy2021; Guitart-Masip et al., 2016). Further, it is important to note that the statistical analyses examined regional thickness measurements as distinct, when in fact these regions are not independent from one another (Habeck & Stern, 2007). Given the suspected differential cost-benefit implications of neural IIV and the complex networks involved in task-related activation in DTW, it is not surprising that all regions did not show the same patterns.
The specific regions found to moderate the change in HbO2 IIV from STW to DTW are involved in sensory, motor, and cognitive functioning. Involvement of PFC (caudal middle, rostral middle, superior middle, and frontal pole regions) thickness suggests that neural IIV expresses direct inefficiency of the PFC, and is consistent with a previous study linking prefrontal gray matter volume with task-related change in mean PFC signal (Wagshul et al., 2019). The parietal findings are likely related to spatial processing and integration (Passarelli et al., 2021) as well as attention and arousal (Leech & Sharp, 2014). Temporal lobe regions implicated are involved in auditory integration and language (Bhaya-Grossman & Chang, Reference Bhaya-Grossman and Chang2022; Herlin et al., 2021; van Kemenade et al., 2019), and highlight the processes needed with the addition of the verbalized cognitive alphabet-based task. Occipital regions involved are necessary for primary visual processing (Wandell et al., 2007) and object processing (Mechelli et al., Reference Mechelli, Humphreys, Mayall, Olson and Price2000), basic visual field processing needed when walking. Finally, the insula is implicated in speech, attention, and sensorimotor processing (Uddin et al., 2017).
In addition to PFC regions implicated in STW to DTW change, the inferior frontal gyrus (pars triangularis, pars orbitalis, and pars opercularis) was implicated in the STA to DTW change. This contains Broca’s area and is essential to speech production (Amunts & Zilles, Reference Amunts and Zilles2012), potentially implicating competing demands of speech and walking in DTW. Similarly, the addition of the walking component in DTW implicated further areas involved in sensorimotor functioning. Significant regions found to moderate the change in HbO2 IIV from STA to DTW included the precentral gyrus which contains the motor cortex (Lim et al., 1994), the inferior parietal lobule which is involved in sensory integration and movement (Haaland et al., 2000), the precuneus which is involved in spatially guided movement (Cavanna & Trimble, Reference Cavanna and Trimble2006), and the fusiform and inferior temporal gyri which are involved in visual (Cohen et al., Reference Cohen, Dehaene, Naccache, Lehéricy, Dehaene-Lambertz, Hénaff and Michel2000; Ptak & Valenza, 2005) and object (Cant & Goodale, Reference Cant and Goodale2007; Grill-Spector et al., 2001) processing.
Aging is accompanied by changes in brain integrity and functioning, and there are well-documented sex differences in cognitive (Levine et al., 2021) and neurobiological aging (Kakimoto et al., 2016). All models in the current study showed a main effect of sex, specifically that males had higher HbO2 IIV across tasks. Prior studies have found that men showed increased activation during DTW when under increased stress (Holtzer et al., 2017) and demonstrated greater increases in HbO2 IIV from single- to dual-task-walk conditions (Holtzer et al., 2020). These findings may be related to increased PFC atrophy seen in men compared to women during aging (Curiati et al., Reference Curiati, Tamashiro, Squarzoni, Duran, Santos, Wajngarten, Leite, Vallada, Menezes, Scazufca, Busatto and Alves2009). While the moderating effect of sex was not examined in the current study, this warrants further investigation.
Previous studies reported the influence of gray and white matter integrity on PFC activation during DTW; this was the first study to examine brain integrity in relation to IIV in HbO2 activity. We found that the regions associated with increased HbO2 IIV were overlapping but also distinct compared to the regions associated with increased mean HbO2 (Ross et al., 2021). This suggests that HbO2 IIV in the PFC provides incremental information not available through central tendency measurements that may be uniquely sensitive to the effect of aging and disease on brain efficiency vis-à-vis walking, notably under attention-demanding conditions.
Strengths, limitations, and future directions
The participants were dementia-free community-dwelling older adults. Future studies should examine these findings in a larger cohort sample, as the current study, common to neuroimaging studies, included only a moderate sample size. As the current study was cross sectional, whether thinner cortex is indicative of pathological brain atrophy versus normal age-related variability cannot be unequivocally ruled out. However, the current sample underwent a consensus case conference to diagnose participants as free of dementia, and there is no evidence of neurological disease. We therefore assumed that the range of cortical thickness observed in the current study is representative of the normal aging population. Future longitudinal work may shed further light on the relationship between changes in cortical thickness, and PFC HbO2 IIV, assessed over repeated measurements. Whether or not the current results translate into longitudinal, age-related decreases in cortical thickness is of clinical importance.
PFC IIV was measured using HbO2, rather than deoxygenated hemoglobin (Hb). A previous study demonstrated similar task-related outcomes when examining HbO2 and Hb (Izzetoglu & Holtzer, 2020). We used only HbO2 in the current study to limit type two errors. The device used in the current study did not have short channels, which limited our ability to remove artifact arising from extracerebral sources. However, we applied stringent processing and filtering of the fNIRS signal to limit motion artifacts and to account for the impact of age on absorption coefficients and DPF (Scholkmann & Wolf, 2013). To further protect the validity of the current study outcomes, we adjusted for possible confounding effects of mean HbO2 signal in each model. Additionally, we measured only PFC IIV during the first 30 seconds of each trial. This allowed us to make direct one-to-one comparisons between tasks eliminating the potential confounding effect of time on IIV estimates. Moreover, clinically validated gait assessments are often limited to short 25-foot (Motl et al., 2017) protocols, which are often completed within 30 seconds. As previous literature implicated neural variability in learning, it would be important to examine whether PFC IIV is reduced over repeated learning trials in a pattern similar to central tendency measures (Holtzer et al., 2019). Whether PFC IIV follows the pattern of traditional practice effects, and its relationship to structural integrity of the brain, would provide valuable information of potential clinical utility.
We only measured HbO2 IIV in the PFC. Additional work is warranted to examine neural activation and IIV in more posterior regions as well, as the current findings provide a partial representation of brain control of locomotion. We elected to examine whole-brain segmented regions to compare the current findings to previous work examining cortical thickness in relation to task-related changes in the mean HbO2 signal (Ross et al., 2021). Future studies may consider network approaches as cortical regions are not independent.
Relative measures of dispersion (e.g., coefficient of variation) could not be calculated as several participants’ mean HbO2 approached zero. Analyses using SD were adjusted for mean HbO2 to address this potential limitation. Previous studies have used and validated SD to operationalize IIV (Costa et al., Reference Costa, Dogan, Schulz and Reetz2019; Garrett et al., Reference Garrett, Kovacevic, McIntosh and Grady2011). Split-half intra-class correlations of HbO2 IIV across the 30 seconds of each task were examined to ensure that findings were not simply reflective of increased noise during the course of tasks and revealed excellent internal consistency of HbO2 IIV. An additional benefit of examining IIV as a task-related biomarker is that it eliminates the effects of relative measurements to a baseline condition. These effects are present in continuous-wave fNIRS measurements while using mean values. Thus, examining neural IIV in task-related research demonstrates both a clinical and methodological strength. Nonlinear measures of data complexity such as multi-scale entropy can provide additional information about the variability of data (Angsuwatanakul et al., Reference Angsuwatanakul, O'Reilly, Ounjai, Kaewkamnerdpong and Iramina2020) and may be used in future studies. There is no current established entropy calculation method in fNIRS analysis, and many require large amounts of datapoints or other parameters not available in the current study.
Finally, there remains a lack of consensus regarding optimal neural IIV (Dinstein et al., Reference Dinstein, Heeger and Behrmann2015). There may be an inflection point at which neural IIV differentiates between being adaptive to disadvantageous, or the utility of IIV may be dependent on population (Maidan et al., 2022). Multimodal neuroimaging approaches provide an advantage for understanding brain mechanisms underlying healthy and disease-related aging (Sui et al., 2014). Future studies should leverage multimodal neuroimaging approaches in healthy and clinical populations to elucidate how the structural integrity of the brain and neural IIV interact to influence cognitive and motoric task performance.
Conclusion
Consistent with the neural inefficiency hypothesis, we found that thinner cortex in the PFC and specific regions throughout the cortex was associated with greater increases in neural IIV in the PFC from single-tasks to dual-task walking without behavioral gains. Cortical thickness in multiple brain regions influenced the change in PFC HbO2 IIV across task conditions, highlighting the role of complex interactions between brain structure and function in supporting gait performance among older adults. Reduced cortical thickness and greater IIV of fNIRS-derived HbO2 during dual-task walking should be further investigated as possible markers of increased risk of developing mobility impairments in aging and disease populations.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355617723000371
Funding statement
This research was supported by the National Institutes of Health (grant numbers R01AG036921, R01AG044007, R01NS109023).
Competing interests
Dr. Izzetoglu has a very minor share in the company that manufactures the fNIRS device used in this study. All other authors declare no conflicts of interest.
