Age-Related Changes in Global Motion Coherence: Conflicting Haemodynamic and Perceptual Responses

Our aim was to use both behavioural and neuroimaging data to identify indicators of perceptual decline in motion processing. We employed a global motion coherence task and functional Near Infrared Spectroscopy (fNIRS). Healthy adults (n = 72, 18–85) were recruited into the following groups: young (n = 28, mean age = 28), middle-aged (n = 22, mean age = 50), and older adults (n = 23, mean age = 70). Participants were assessed on their motion coherence thresholds at 3 different speeds using a psychophysical design. As expected, we report age group differences in motion processing as demonstrated by higher motion coherence thresholds in older adults. Crucially, we add correlational data showing that global motion perception declines linearly as a function of age. The associated fNIRS recordings provide a clear physiological correlate of global motion perception. The crux of this study lies in the robust linear correlation between age and haemodynamic response for both measures of oxygenation. We hypothesise that there is an increase in neural recruitment, necessitating an increase in metabolic need and blood flow, which presents as a higher oxygenated haemoglobin response. We report age-related changes in motion perception with poorer behavioural performance (high motion coherence thresholds) associated with an increased haemodynamic response.


Results
Data Analysis. fNIRS data pre-processing was completed with a custom-written MATLAB script, which included the following: normalisation, smoothing and average response calculation. Using a simple subtraction method, all data were normalised with respect to the pre-stimulus baseline (the last 20 seconds of a 60 seconds baseline). This ensured that the normalisation procedure was carried out according to the most stable average response to the control image (baseline fixation cross). A conservative 3-point moving average filter was computed on the normalised data. This low-pass filtered data were then de-trended. The mean of 10 to 20 seconds post-stimulus onset was used, representing the most reliable response to the stimulus [75][76][77] . This timing allowed the HDR to peak immediately after the stimulus was presented, ensuring the recording of a stable perceptual response. The mean of this global motion response was compared with the baseline (grey screen with white cross) visual stimulation. The robust Median Absolute Deviation (MAD) method of outlier analysis was used. As there were no statistically significant differences between the hemispheres these were averaged to create an overall V5 response.
Group Descriptives. Group descriptives can be seen in Table 1. There were n = 28 younger adults (mean age of 28 ± 7.3), n = 22 middle-aged adults (mean age of 50 ± 6.9), and n = 23 older adults (mean age of 70 ± 6.9). A multivariate ANOVA showed the expected age-related differences (systolic and diastolic BP, p < 0.05). Pairwise comparisons revealed that both younger and middle-aged adults had lower systolic BP (F 2, 67 = 8.08, p < 0.01, η 2 = 0. 19) and diastolic BP (F 2, 67 = 3.64, p < 0.05, η 2 = 0.09), in contrast to older adults. These group statistics were supported by Pearson's skipped correlations with 95% CI using MATLAB's Robust Correlations toolbox 78 . Skipped correlations is a robust method of Pearson's r using the linear association but ignoring outliers detected by taking into account the overall structure of the data via bootstrapping techniques 78 . Age was significantly correlated with both systolic and diastolic BP (r = 0.41, CI = [0.21 0.61], r = 0.25, CI = [0.04 0.41], respectively). When entering sex as a between-subject factor for each of these outcome measures, there were no statistically significant results, indicating that there were no sex differences in our sample (p > 0.05), unlike a handful or previous studies 18,25,26 . None of these descriptive outcome measures (BP, HR, eCRF) correlated significantly with the other data (behavioural or neuroimaging).
Neuroimaging HDR to Motion. The (Fig. 2h). This is a good effect size with robust confidence intervals demonstrating a clear linear age-related effect to the motion HDR. These individual level results were also found at group level with the ANOVA indicating a small 3-way interaction between stimulation, speed, and age group (F 4, 128 = 2.98, p < 0.05, η 2 = 0.08). To determine exactly where group differences occurred, independent T-tests were performed post-hoc with adjusted CIs (99.99%). During both 1.5 and 3°/s motion stimuli, age groups were equivalent in their [HbO] in their HDR. However, during the faster 9°/s motion stimuli there were age-related effects, thus supporting the robust individual correlational results. Older adults showed significantly increased  (Fig. 2i). Once more, group results supported individual level findings with a 3-way interaction between stimulation, speed, and group (F 4, 122 = 5.57, p < 0.001, η 2 = 0.15). Adjusted post-hoc T-tests showed group differences at both 3 and 9°/s stimuli, but not at the slower 1.5°/s. There were statistically significant differences between younger and middle-aged adults (mean difference = 0.07, t 47 = 2.76, p < 0.01, d s = 0.79), and younger and older adults (mean difference = 0.08, t 49 = 4.74, p < 0.01, d s = 1.33).
There were mixed findings for the two slower speeds of stimulus (1.5°/s, 3°/s). However, results for the fastest stimuli (9°/s) showed robust and distinct age-related effects wherein [HbO] and [HbR] mirrored each other. This HDR to motion presented as a characteristic fNIRS response with an increase in [HbO] and decrease in [HbR] during stimulus presentation compared to the baseline fixation cross. Subsequent age-related changes in motion perception were strongest during the fastest stimulus speed with a dot presentation of 9°/s. Note that these HDR results were in response to a stimulus with high global motion coherence, because trials always began with 100% coherence and the algorithm subsequently adjusted stimulus presentation depending on individuals' performance (and response speed). There is fMRI evidence of a direct linear relationship between coherence of globally moving stimuli and V5 activation 53,[79][80][81] . Therefore, as this HDR represents such a motion response to stimuli with high coherence, rather than being specific to individuals' thresholds, it represents a general perceptual HDR to global motion. Overall results can be seen in Table 2 with a summary of the behavioural and neuroimaging results for each speed and age group comparison.

Discussion
This is the first fNIRS study to investigate global motion perception in a healthy ageing population. We used a global motion task to study the perceptual thresholds and associated physiological changes of the HDR in adults aged 18-85 years old. The expected behavioural results were present with a reduced global motion coherence perception in older adults demonstrated by higher motion coherence thresholds. In contrast to these behavioural findings, the HDR was greater in older adults. When using robust statistical methods, we report greater [HbO] and [HbR] HDR as a function of older age, i.e. there was a linear correlation between age and the motion HDR.
In accordance with previous literature, the psychophysical results show an age-related decline in global motion perception [5][6][7][8][9][10] . In terms of motion coherence thresholds, i.e. a measure of individually determined motion perception, this decline was much greater than previously reported 30 . The mean difference between young and older adults was 7%, and middle-aged and older adults 6%. Regardless of the speed of the RDK stimulus used, older adults had the highest thresholds (poorest performance). Interestingly, these results are the opposite of Pilz et al., (2017) who reported older adults to perform similar to younger adults detecting vertical motion 12 . However, there are a number of methodological differences that may underpin this: the older adult group was slightly younger than the current group, individual MCTs were calculated using the classical method of constant stimuli rather than the current adaptive staircase procedure, and the stimuli contained just under double the number of dots than ours did. The current differences in age groups support previous evidence on decade group thresholds and propose the greatest difference to be with the 'oldest of old' , namely over the age of 70 years old 7,28 . Similarly, the greatest behavioural differences we report were with the older adults (aged 60-85), compared to younger and middle-aged adults (see Fig. 1). Crucially, what we add here is the correlational data, as our sample included adults aged 18-85 years old. Robust statistics revealed the 3 and 9°/s stimulus speeds were associated with poorer motion perception with increasing age. As Fig. 2 shows, our rigorous methodology ensured that the age-related decline in motion perception could not be attributed to differences in acuity or speed of processing. Particularly noteworthy is the data distribution for the fast stimulus speed (9°/s, Fig. 2g); there is less variability in data distribution compared to the other speeds, which suggests that the results at 9°/s were reliable. Real world implications for this task have been shown by Conlon et al., (2015) who reported thresholds at the same speed and showed older adults' motion coherence thresholds were related to self-reported driving difficulties 82 . The data provide a clear physiological correlate of motion perception as reflected by the HDR. This is the first study to use FDMD-fNIRS during a motion task in healthy adults providing absolute values of cerebral oxygenation measures. Regardless of speed, all participants showed a distinctive HDR with an increase of [HbO] and decrease of [HbR] during stimulation. This HDR to motion processing and associated statistical effect sizes, show the magnitude of the changes between the baseline fixation cross and subsequent motion stimulation. These results support two previous fNIRS studies using a passively recorded HDR to different motion stimuli (motion illusion 83 and radial motion 66 ), both of which report an increase in [HbO] and decrease in [HbR] during stimulation. The crux of the current findings lies in the differences in responses recorded across the lifespan: both poor motion perception (high motion coherence thresholds) and an increased HDR were modulated by age. We hypothesize that older adults required the recruitment of additional neurons to complete the task successfully. The speculated increased neural recruitment necessitated an increased metabolic need and subsequent increase in blood flow, presenting as higher [HbO] responses to global motion stimulation in older adults. Whilst we acknowledge the spatial limitations of fNIRS, our data show unequivocally, a HDR that is a physiological reflection of global motion perception recorded over the parieto-occipital cortex approximating V5.
The current results fit the dedifferentiation theory of healthy ageing. Visual dedifferentiation has been reported in older adults (compared to younger adults) in both early and late stages of visual processing 38,84 . Both univariate and multivariate studies show an age-related reduction in neural distinctiveness in visual regions in older adults 13,[36][37][38][39][40][41] . For example, Biehl et al. 13 , recently showed older adults presented with differential fMRI activation patterns compared to younger adults. Age-related differences in processing radial motion were present with older adults recruiting larger areas of the middle/superior temporal gyrus and additional brain regions compared to younger adults 13 . Similarly, electrophysiological work has shown older adults to present with a greater topographical voltage distribution in response to translational motion compared to younger adults, despite equivalent behavioural performance on the associated task 43 . Our fNIRS data can provide absolute quantification of the HDR in terms of [HbO] and [HbR]. As can be seen in Fig. 3, there is tight physiological coupling between these two measures of cerebral oxygenation, regardless of age. This is in line with previous literature reporting no age-related effects of coupling 85,86 . Here we report an increased HDR as measured by fNIRS in ageing adults in response to seeing a global motion stimulus. We propose that older adults required an additional recruitment of neurons for processing and perceiving global motion.
Event-related processing was not possible in this experimental design. This is a limitation that may easily be addressed in future work where triggering could provide information about the shape of the HDR to specific coherence levels. The HDR recorded was in response to the initial presentation of the motion stimuli, i.e. when the image had high coherence. As individual performance varied, the HDR could not be attributed to a specific coherence level. However, previous neuroimaging evidence indicates a linear correlation between high coherence of moving stimuli and the V5 response 45,53,[79][80][81] . Therefore, we propose our HDR to be a sound reflection of global motion processing.
To conclude, we report an age-related decline in global motion perception thresholds (across all speeds of stimuli used). The fastest speed (9°/s) generated the most robust results for both behavioural and neuroimaging data. Regardless of age, all participants showed a distinctive increase of [HbO] and decrease of [HbR] during visual stimulation. Our novel results show that fNIRS can reliably capture the HDR associated with global motion perception. Moreover, strong vascular coupling between [HbO] and [HbR] was shown regardless of age or experimental manipulation. The significance of this work lies in the age-related differences in the HDR, with an increased HDR from younger, to middle-aged, to older adults. The significant results and large effect sizes reported here relate each of the behavioural and neuroimaging data to be affected as a function of age, in line with the dedifferentiation theory of ageing. We hypothesise that there is an increase in neural recruitment, and therefore a consequent increase in cerebral blood flow, in response to an age-related decline in motion processing.

Methodology
Participants. A total of 72 healthy adults between the ages of 18-85 were recruited, and were broadly divided into the following age groups: young adults (n = 28, mean age 28 ± 7.3 range 18-39, 22 females), middle-aged adults (n = 22, mean age 50 ± 6.9, range 40-59, 12 females), and older adults (n = 23, mean age 70 ± 6.9, range 60-85, 12 females). All participants had a Visual Acuity (VA) of at least 6/9 (6/6 equating to 'normal' vision) with optical correction where required and had no history of neurological or psychiatric disorders. A short medical history was taken and any current medication noted with none being reported that would influence the cerebral blood flow (e.g. those participants on hypertensive medication would not bias results 87 ). Measurements of height, weight, Blood Pressure (BP), and Heart Rate (HR), were taken at rest prior to beginning the main experiment. Estimated Cardiorespiratory Fitness (eCRF) was calculated using an algorithm including sex, body mass index, resting HR, and self-reported physical activity. The eCRF has previously been used with success and reliability providing an accurate estimate of cardiorespiratory fitness (compared to a NASA exercise test) regarding cerebral blood flow 88,89 . Lastly, all participants had a comparable number of years of education within each subgroup and scored >27 on the Mini-Mental State Examination (MMSE) for cognitive screening. Glasgow Caledonian University's Ethics Committee approved the research protocol, and informed written consent was obtained from Visual Stimuli. Participant's Motion Coherence Threshold (MCT) was assessing using a standard RDK paradigm 90 . The 80 white moving dots (luminance 100 cd/m 2 ) were presented within a central square display, on a homogenous grey background (luminance 50 cd/m 2 ). The motion array can be seen in Fig. 4b. The direction of signal dots (moving either up or down) had to be detected over random noise dots. Each dot was displayed for 160, 80, or 26.67 ms. Varying the dot duration created apparent motion at 3 stimulus speeds: 1.5, 3, and 9°/second. Dot density, size, brightness, displacement, radius, background brightness, and spatial offset were kept constant. Participants viewed the stimulus at 57 cm, so that 1 screen pixel subtended 0.04°. Before beginning the motion task, participants were presented with a white central fixation cross (Fig. 4a) during which a baseline imaging response of 60 seconds was recorded. A button press started the task, and a motion array did not appear until a response had been recorded to the previous array. Figure 4 depicts the experimental paradigm including the baseline (a) and motion array (b), and the timing of a trial (c).

Psychophysical Staircase Parameters.
Participants completed 3 trials of each speed, in a randomised counterbalanced order, and the mean MCT to each speed was calculated. The MCT is the minimum percentage of signal dots required to successfully determine the global motion direction. A 2AFC adaptive staircase procedure was used that tracked 79.4% correct threshold value. On each staircase reversal, the step size was halved (starting at 16). When this reached 1 dot, the mean of the 6 last reversals was used to calculate the threshold. The final MCT for each speed was calculated as a mean of the 3 trials completed for an accurate estimation of the individual's threshold. This methodology and stimulus has previously been used to assess the sensitivity of global motion detection in amblyopic adults 90,91 . HDR Recording. A Frequency-Domain Multi-Distance (FDMD) fNIRS system was used with two-channels (OxiplexTS TM ). This system is frequency modulated at 110 MHz with 1 Hz sampling. Two wavelengths of light (690 nm and 830 nm) are used to calculate changes of [HbO] and [HbR] in absolute concentration (μM/L). This instrumentation has been described in detail elsewhere 76,92,93 . Near infrared light detectors were at fixed distances from the emitters (ranging from 1.9-3 cm) in each of the two silicone sensors. These precise distances were entered into the OxiplexTS software to subsequently calculate cerebral oxygenation levels, using absorption, scattering and phase data and the modified Beer-Lambert Law 56,93 . HDRs to the grey screen (see Fig. 4) were recorded prior to stimulus onset. As the stimuli consisted of moving dots, we recorded over the parieto-occipital cortex, namely V5. During completion of the motion task, fNIRS recordings were taken from over the area thought to coincide with V5's location (Fig. 5). This approximation was based on an average of previously reported fMRI Talariach coordinates and subsequently converted to the EEG 10-20 International System of Electrode Data Availability. The datasets generated during the current study are available from the corresponding author on reasonable request.