Age-related effects on dynamic postural stability and prefrontal cortex activation during precision fitting tasks

Participants

We recruited right-handed healthy young, middle-aged, and older adults in the current study. The inclusion criteria were as follows: (1) age 60 or older in the older group, age from 45 to 55 in the middle-aged group, age from 18 to 22 in the young group; (2) able to stand and walk at least 2 min without any assistance; (3) no injury or surgery at lower extremity in the past 6 months; (4) no neurological diseases, such as mild cognitive impairment, permanent memory loss, stroke, Parkinson’s disease, and brain tumors; (5) Mini-Mental State Examination (MMSE) score ≥ 24 for all three groups, except for participants with 0 to 6 years of schooling (MMSE score ≥ 22) (Cui et al., 2011); and (6) no history of drug and alcohol abuse. The authors have permission to use MMSE from the copyright holders. This study was approved by the Ethics Committee of the Affiliated Hospital of Yangzhou University (2020-YKL12-23-02). Each participant signed the informed consent form before participation.

Instrumentation setup

A custom-built instrument was utilized in this study, featuring a large whiteboard with two openings: a larger one measuring 130 × 130 mm and a smaller one measuring 100 × 100 mm. These openings were positioned in the upper middle section of the whiteboard, spaced 150 mm apart. Additionally, a fitting block measuring 90 × 90 mm, equipped with a cylindrical handle (20 mm in length and 10 mm in diameter), was placed on a custom-built base situated on a small table. Both the fitting board and the small table were designed with adjustable heights and positions. Finally, four pairs of optical sensors were employed, which are synchronized with Vicon Nexus system (Version 2.5, Vicon, Inc., Oxford, UK). Of these, two pairs were affixed to the top middle edge of each opening and on the whiteboard’s backside, while the remaining pairs were attached to each side of the custom-built base.

The CoP data was collected using an embedded force plate (Kistler 9285BA, Kistler Corporation, Winterthur, Switzerland) at a sampling rate of 2,000 Hz. Data collection from the force plate was conducted using Vicon Nexus system (Version 2.5, Vicon, Inc., Oxford, UK). Changes of the cortical activation in the PFC were measured using an fNIRS device (Brite 24, Artinis medical systems, Einsteinweg, Netherlands) at a sampling rate of 50 Hz, utilizing two wavelengths of near-infrared light (760 and 850 nm). The setup included 10 sources and eight detectors, constituting 24 channels in total, positioned on the head’s surface via a standard NIRS cap (10–10 international system) covering the PFC. The differential pathlength factor (DPF) was calculated was age-dependent: for participants younger than 55 years, DPF was determined using the formular: 4.99 + 0.067 × Age^0.814 (Duncan et al., 1996). For participants older than 55, the DPF was fixed set to 6.61 (Claassen, Levine & Zhang, 2009). To identify the subregions of PFC, five anatomical landmarks (nasion, inion, Cz, left and right preauricular points) were digitized using a Polhemus digitizer. Oxysoft was used for the collection prefrontal cortex activation. All devices were synchronized.

Study protocol

This study was a cross-sectional design. Participants performed the precision fitting task in this study. The precision fitting task entails both the execution of typical goal-directed task and the need to maintain dynamic postural stability (Pan et al., 2020; Pan et al., 2016). Moreover, task constraints could be simply manipulated by decreasing the opening size (enhancing the fitting precision) and increasing the opening distance (enhancing the postural constraint) (Coats et al., 2016; Huang & Brown, 2013; Potocanac & Duysens, 2017; Sarlegna, 2006).

Participants wore uniform socks provided by the laboratory. Each participant stood on the center of the force plate with feet forming a 30° angle, heels being apart at 8% of the height, arms relax on each side, and align their middle line with the opening’s center (Pan et al., 2020; Pan et al., 2016). Then, the height of opening was adjusted to the participant’s shoulder height and aligned with the midline of body, and the distance of the board was adjusted to either their arm’s length or 1.3 times arm’s length, and the small table’s height to the participant’s waist height. Participants were instructed to fit the block into either a large or small opening on the custom-made board under upright stance, positioned either an arm’s length or at 1.3 times an arm’s length from the board. Thus, four different conditions were performed by participants, including close-large, close-small, far-large, and far-small conditions. Participants began with the close condition, and the size condition was randomly selected. The order was counterbalanced across participants. Moreover, the purpose of performing close conditions first was to avoid participants starting with the most difficult far-small condition. It could minimize the learning effects that affects the outcomes in the close conditions. If participants either moved their feet or if the block touched the edge of the opening during the fitting task, the trial was categorized as “failed”. For each condition, participants were required to complete five consecutive successful trails under the supervision of our experimental operator. Once a “failed” trial occurred, the participant was instructed to redo the five trials. Meanwhile, one experimental operator was required to stand in the back of the board, take the block, and put it back on the base as soon as possible. A 10-second baseline of quiet standing was recorded before the fitting task. Participants were given at least a 2 min break between conditions.

Data analysis

The Vicon Nexus system was used to preprocess the CoP data, which was filtered with a fourth-order Butterworth low-pass filter with a cutoff frequency of 10 Hz (Rocchi, Chiari & Horak, 2002) and then exported in the format of CSV file for further analysis. Then, CoP data was divided into five trials based on the event signals from the optical sensors, where the onset of each trial was defined as the moment of rising the block and ending once the block completely passed through the opening. The standard deviation (SD) of CoP (CoP variability, SD_AP & SD_ML), average CoP velocity (V_AP & V_ML) at anterior-posterior (AP) and medial-lateral (ML) directions, and the 95% ellipse of sway area (SA) were calculated using the MATLAB (2021b, MathWorks, Natick, MA, USA). All these dependent variables were averaged over five trials under each condition. The higher value of these dependent variables represents worse dynamic postural stability.

To analyze the fNIRS data, the trail was defined as the first raising the block to the fifth passing the block through the opening under each condition based on the event signals from the optical sensors. The average duration of each condition was 19.74 ± 2.35 s, 24.88 ± 3.51 s, 22.45 ± 3.11 s and 30.47 ± 4.03 s in the young group, 19.77 ± 2.10 s, 28.39 ± 3.16 s, 24.24 ± 3.91 s and 33.15 ± 6.35 s in the middle-aged group, and 21.48 ± 3.47 s, 30.24 ± 7.94 s, 26.97 ± 4.47 s and 35.41 ± 8.50 s in the older group during the close-large, close-small, far-large, and far-small conditions, respectively. In the current study, only oxygenated hemoglobin (HbO₂) data were used for further analysis. It is because that a superior signal-to-noise ratio was observed for HbO₂than deoxygenated hemoglobin (HHb) signals (Strangman et al., 2002) and HbO₂ and HHb were negatively associated (Cui, Bray & Reiss, 2010). The coefficients of variation (CV) for each channel of every participant were computed. Channels presenting a CV greater than 15% were excluded from subsequent data analyses, as they may include physical artifacts (e.g., motion-induced instabilities of the coupling efficiency at the tissue-optical interfaces) and physiological artifacts (e.g., blood-pressure-induced hemodynamics) (Pinti et al., 2019).

The preprocessing of the PFC activation used the Homer3 toolbox within MATLAB (The MathWorks, Inc. Natick, MA, USA), following distinct steps outlined in previous studies (Koren, Parmet & Bar-Haim, 2019; Öztürk et al., 2021): (1) converting raw data to optical density data; (2) removal of motion artifacts using the principal component analysis (tMotion = 1.0, tMask = 1.0, StdevThresh = 50 and AmpThresh = 0.5); (3) correction of motion artifacts using the spline interpolation (p = 0.99); (4) correction of motion artifacts using the wavelet based filter (iqr = 0.1); (5) corrected of physiological artifacts using the band-pass filter at a cutoff frequency of 0.01–0.14 Hz; (6) baseline-corrected by subtracting each trial individually from the 5 s of quiet standing; and (7) converting optical density data to concentrations. The HbO₂ concentration signals were exported to the TXT file to further calculate. According to the channels’ positions, the average concentration of HbO₂ on six regions of interest (Fig. 1), including right dorsolateral PFC (R_PFC_DL), left PFC_DL (L_PFC_DL), right dorsomedial PFC (R_PFC_DM), left PFC_DM (L_PFC_DM), right ventrolateral PFC (R_PFC_VL) and left PFC_VL (L_PFC_VL), were calculated.

Statistical analysis

The two sets of dependent variables are dynamic postural stability, which includes SD_AP, SD_ML, V_AP, V_ML and SA; and PFC activation, which includes HbO₂concentration in the R-PFC_DL, L-PFC_DL, R-PFC_DM, L-PFC_DM, R-PFC_VL, and L-PFC_VL. The errors in fNIRS data are not independent across measurement channels (Huppert, 2016). Therefore, the Shapiro–Wilk’s test was only used to assess the normality of each dependent variable of dynamic postural stability (α = 0.05). Additionally, the multivariate normality and outliers of two sets of dependent variables were examined using Mahalanobis’ distance.

Then, two two-way repeated measure MANOVAs (between-subject factor: group & within-subject factor: condition) were performed to investigate the effects of group and condition on dynamic postural stability and PFC activation, respectively. When a repeated measure MANOVA was significant, follow-up repeat measure ANOVAs and pairwise comparisons with Bonferroni correction were employed to detect significant main effects (group and condition) and interaction effect (group × condition) for each dependent variable. Partial Eta Squared (η²p) was calculated for overall effects and interactions for MANOVAs and ANONAs, where η²p = .010 is considered a small effect size, .060 medium effect size, and .14 or higher large effect size (Cohen, 2013). Cohen’s d effect size was calculated to interpret the magnitude of specific post hoc comparisons, with d = .20 considered a small effect size, .50 medium effect size, and .80 or higher a large effect size (Cohen, 2013). The significant level was set at .05. All statistical analyses were performed using SPSS (28.0, IBM Inc., Armonk, NY, USA).

Dynamic postural stability

Some dependent variables of dynamic postural stability presented non-normal distribution based on the Shapiro–Wilk’s test. These non-normally distributed dependent variables were log10 transformed before statistical analysis. Additionally, our dependent variables of the dynamic postural stability were multivariate normally distributed (MD < 20.52). Table 2 showed the Mean ± standard deviation values of dynamic postural stability among three groups under different conditions.

There were significant group (Wilk’s lambda = .455, F(10, 76) = 3.672, p < .001, η²p = .326) and condition (Wilk’s lambda = .048, F(15, 28) = 36.892, p < .001, η²p = .952) effects, and group × condition interaction (Wilk’s lambda = .217, F(30, 56) = 2.139, p = .007, η²p = .534) effect on the association of dependent variables. Based on follow-up ANOVA with repeated measure tests, the significant effects of group, condition, and group × condition interaction were observed in the SD_AP (group effect: F(2, 42) = 8.926, p = .001, η²p = .298; condition effect: F(3, 126) = 42.311, p < .001, η²p = .502; & interaction effect: F(6, 126) = 2.819, p = .020, η²p = .118) and V_AP (group effect: F(2, 42) = 4.129, p = .023, η²p = .164; condition effect: F(3, 126) = 136.575, p < .001, η²p = .765; & interaction effect: F(6, 126) = 3.960, p = .002, η²p = .159) (Table 2). Additionally, there was significant condition effect (F(3, 126) = 91.357, p < .001, η²p = .952) and group × condition (F(6, 126) = 2.680, p = .021, η²p = .113) interaction in the V_ML (Table 2). Post hoc analysis indicated that the older group presented greater SD_AP than the young group in the close-large (p = .002 & Cohen’s d = 1.31), far-large (p = .002 & Cohen’s d = 1.21), and far-small (p < .001 & Cohen’s d = 1.37) conditions (Fig. 2). Also, the older group showed greater V_AP and V_ML than the young group in the close-large (V_AP: p = .029 & Cohen’s d = 1.73 & V_ML: p = .034 & Cohen’s d = 0.97) and far-small (V_AP: p = .001 & Cohen’s d = 1.27& V_ML: p = .021 & Cohen’s d = 0.83) conditions. Additionally, the middle-aged group showed grater SD_AP (p = .005 & Cohen’s d = .77) and V_AP (p = .029 & Cohen’s d = .77) compared to the young group in the close-large condition (Fig. 2).

The ANOVA with repeated measure tests further observed group and condition effects in the SD_ML (group effect: F(2, 42) = 10.919, p < .001, η²p = .342; & condition effect: F(3, 126) = 27.039, p < .001, η²p = .392) and SA (group effect: F(2, 42) = 13.014, p < .001, η²p = .383; & condition effect: F(3, 126) = 58.553, p < .001, η²p = .582). Post hoc test reported that the older group showed greater SD_ML (young group: p < 0.001 & Cohen’s d = .96 & middle-age group: p = 0.028 & Cohen’s d = .66) and SA (young group: p < 0.001 & Cohen’s d = .94 & middle-age group: p = 0.030 & Cohen’s d = .58) than the young and middle-aged groups across the conditions (Fig. 2). We did not report the condition effect since our interests were the effects of group and interaction between group and condition.

Prefrontal cortex activation

Our dependent variables of the PFC activation were multivariate normally distributed within each group of the independent variables (MD < 22.46). Table 3 showed the mean ± standard deviation values of HbO₂ concentration in the PFC among three groups under different conditions.

There were significant group (Wilk’s lambda = .467, F(12, 74) = 3.455, p = .003, η²p = .317) and condition (Wilk’s lambda = .295, F(18, 25) = 3.327, p = .003, η²p = .75) effects on the association of dependent variables. No significant group × condition interaction effect (Wilk’s lambda = .253, F(36, 50) = 1.371, p = .150, η²p = .497) was observed in the MANOVA model. Follow-up ANOVA with repeated measure tests reported group effect in the R_PFC_DL (F(2, 42) = 7.693, p = .001, η²p = .268), L_PFC_DL (F(2, 42) = 11.076, p < .001, η²p = .345), R_PFC_DM (F(2, 42) = 7.701, p = .001, η²p = .268), L_PFC_DM (F(2, 42) = 15.637, p < .001, η²p = .427), R_PFC_VL (F(2, 42) = 6.920, p = .003, η²p = .248), and L_PFC_VL (F(2, 42) = 12.325, p < .001, η²p = .370) and condition effect in the R_PFC_DL (F(3, 126) = 5.933, p = .002, η²p = .124), R_PFC_DM (F(3, 126) = 6.390, p < .001, η²p = .132), L_PFC_DM (F(3, 126) = 5.675, p = .001, η²p = .119), R_PFC_VL (F(3, 126) = 6.935, p = .001, η²p = .142), and L_PFC_VL (F(3, 126) = 10.674, p < .001, η²p = .203) (Table 3). Post hoc test indicated that the older group presented greater HbO₂ activation in the R_PFC_DL (p = .001 & Cohen’s d = .85), L_PFC_DL (p < .001 & Cohen’s d = 1.08), R_PFC_DM (p = .001 & Cohen’s d = .84), L_PFC_DM (p < .001 & Cohen’s d = 1.18), R_PFC_VL (p = .002 & Cohen’s d = .79), and L_PFC_VL (p < .001 & Cohen’s d = 1.04) than the young group, while older group also showed greater HbO₂ concentration in the L_PFC_DL (p = .007 & Cohen’s d = .38) and L_PFC_VL (p = .041 & Cohen’s d = .54) than the middled-aged group across the conditions (Fig. 3). Also, the middle-aged group presented greater HbO₂ concentration in the L_PFC_DM (p = .008 & Cohen’s d = .36) than the young group (Fig. 3). We did not report the condition effect since our interests were the group and interaction effects among group and between group and condition.