Contrast Sensitivity Predicts 30-month Functional Brain Network Integrity in Cognitively Unimpaired Older Adults: the Brain Networks and Mobility Study

Abstract

Background

Visual contrast sensitivity (CS) is critical to many functions in older adults and is associated with brain network community structure, but the direction of the relationship between CS and the brain remains unclear.

Methods

We evaluated whether baseline binocular CS predicts 30-month functional brain network organization in 172 community-dwelling older adults (mean age 76.4 ± 4.8 years, 56.4% female, 11.6% non-White/Hispanic) that underwent functional MRI at rest and during a motor imagery task. We constructed separate distance regression models for each of the 8 subnetworks covering the entire brain, while controlling for the baseline brain networks, sex, and the number of volumes removed during motion scrubbing from head motion in the scanner.

Results

Worse baseline CS predicted lower community structure integrity at 30 months in the visual network (β = 0.0115; p < .0001), dorsal attention network (β = 0.0075; p = .0089), and default mode network both at rest (β = 0.0173; p < .0001) and during the motor imagery task (default mode network, β = 0.0103; p = .0002). No other networks showed significant associations. The dorsal attention network did not have a relationship with CS at baseline but was significant at 30 months. Similar findings were observed in models that additionally controlled for baseline Montreal Cognitive Assessment and change in Montreal Cognitive Assessment score over 30 months.

Conclusions

Poor CS may identify a subset of older adults at risk of future decrements in brain circuits important for vision, cognitive, and mobility functions. Future studies should explore if improving CS increases functional brain health.

Contrast sensitivity (CS) is the ability to discern differences in light and dark and is crucial for spatial perception and pattern recognition (1,2). Impairment in CS has been linked to poor cognitive and mobility function in older adults (3,4). Moreover, we have suggested that CS impairment may be more integral to gait and balance issues compared to visual acuity discrimination of high contrast optotypes (3,5). We have also observed that even minor deficits in CS are associated with worse performance on the expanded Short Physical Performance Battery in cognitively unimpaired older adults with good visual acuity (3,5). Decreased CS may thus identify a subset of older adults at risk for downstream mobility limitations who are well-suited for interventions to improve visual function and decrease risk of mobility disability.

In a prior study, we used data from the Brain Networks and Mobility (B-NET) study to identify the functional neural correlates of CS (6). The B-NET study was a cohort study of community-dwelling older adults who were cognitively unimpaired at baseline. We observed that CS was cross-sectionally associated with the visual network (VN), default-mode network (DMN), and sensorimotor (SMN) brain networks at baseline. These networks are important for visual processing, self-referential mentation, and mobility function, respectively. While these results contributed to a greater understanding of the relationship between CS function and brain network organization, they do not imply a causal relationship.

Attempting to further elucidate this relationship, and to understand if low CS contributes to future brain network organization, we utilized the same B-NET cohort to test if baseline CS could predict the 30-month brain network organization at rest and during a motor imagery (MI) task. In this analysis, we hypothesized that CS drives changes in brain connectivity rather than the reverse because there is a biological connection between CS impairment and certain age-related eye diseases (such as cataract, macular degeneration, and glaucoma) which could affect visual input to the brain networks (7,8). From a neurobiological perspective, the loss of CS input into the brain networks could result in network decline due to disuse. However, we also know that many older adults exhibit CS deficits in the absence of specific ophthalmic or neurologic disease, which could suggest a generalized functional decline related to normal aging (4,9).

Based on our prior baseline findings and our understanding of functional networks, we hypothesized that baseline CS would specifically predict community structure from the 30-month fMRI in the DMN, VN, and SMN. The DMN is a large brain network that is primarily composed of prefrontal cortex, posterior cingulate cortex, precuneus, temporal lobe, and lateral parietal lobe (10) and is functionally implicated in self-referential thought (11,12). The DMN mediates context-dependent cognition (13) and has been classified as generating self-oriented predictive models of the world (11). The VN largely consists of the occipital lobes (14), which receive visual information from the retina via the thalamus (15) and is primarily responsible for processing visual information (16), visuospatial processing, and is instrumental to CS (17). The SMN consists of primary motor neurons that initiate movement and has a well-documented role in maintaining mobility (18,19). Our previous work suggests that CS is cross-sectionally related to sensory-motor processing in the SMN (6).

In addition to the networks we observed at baseline, we hypothesized that CS would also be linked to future dorsal attention network (DAN) integrity. Anatomically, the DAN consists of the frontal eye fields and inferior parietal sulcus, and as a circuit, it is sensitive to stimulus distinctiveness for a range of visual features (20). The DAN is crucial for detecting novel environmental elements and is involved in visuospatial attention, working memory for spatial tasks, and integrating stimuli from one’s environment for balance (20,21). It supports visually guided actions by providing orientation information about objects and environments relevant to goal-oriented movements (22). For instance, before standing up and walking across the room, the DAN integrates visual and spatial cues necessary for initiating movement. Together, these networks provide reasonable structural and functional neurological targets where we expected CS to be related.

Thus, given our previously observed association of CS with mobility (3,5), and a cross-sectional relationship with brain network connectivity in regions important for not just vision but also cognition and mobility (6), we expected that poor CS would predict lower community structure integrity in networks integral to not only vision (VN) but also cognitive (DMN), SMN, and spatial attention (DAN) functions.

Methods

The B-NET Study

Brain Networks and Mobility (NCT03430427) is a longitudinal, observational study of community-dwelling older adults aged 70 and older recruited from Forsyth County, North Carolina, and the surrounding areas. Prior publications from the B-NET study have detailed methods that are briefly described here with references to supplemental material and prior publications for detailed descriptions (6,23). This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of the Wake Forest School of Medicine (IRB protocol #IRB00046460; approval date: 08/27/2020). All B-NET participants provided written informed consent for the B-NET study.

Brain Networks and Mobility study inclusion and exclusion criteria were designed to yield a participant population at baseline without major cognitive disability or physical function disabilities, meaning that participants were able to walk without assistive devices. Potential participants were excluded if they self-reported major uncorrectable hearing or vision impairment on a screening telephone questionnaire (“Do you have any major uncorrectable hearing or vision impairments?”) as this would preclude them from being able to complete the task in the fMRI scanner. They were also excluded if they reported single or double amputations, severe musculoskeletal disease, dependence on a walker or another person to walk, recent surgery or hospitalization within the previous 6 months, serious or uncontrolled chronic diseases, hazardous alcohol use (> 21 drinks per week), clinical manifestation of a neurological disease that affects mobility, history of traumatic brain injury with residual deficits, history of brain tumors, seizures within the last year, unwillingness or inability to undergo an magnetic resonance imaging (MRI) brain scan, plans to relocate within the next 2 years, or participation in a behavioral interventional trial. Cognitive status was assessed using the Montreal Cognitive Assessment (MoCA), excluding those with scores less than 21 (24,25). For participants with MoCA scores between 21 and 25, the study neuropsychologist excluded participants with mild cognitive impairment based on a review of a full battery of cognitive tests—MoCA, semantic fluency, verbal fluency, craft story, digit symbol coding, auditory verbal learning, trail making A and B, and flanker (26).

Visual Function Testing

Each B-NET participant underwent both binocular visual acuity and CS testing at the baseline visit, which took place 30 months before the follow-up MRI. The Early Treatment Diabetic Retinopathy Study (ETDRS) eye chart was utilized for participant binocular visual acuity testing. Visual acuity was recorded as the logarithm of the minimum angle of resolution (logMAR), and Snellen visual acuity was categorized as worse than 20/40 or 20/40 or better. Participants were tested while wearing corrective lenses, if applicable. Contrast sensitivity testing was performed utilizing the Pelli–Robson eye exam chart (27) at a distance of 5 feet away while wearing corrective lenses, if applicable. The total number of letters correctly read was recorded and converted to log CS for analysis, as demonstrated in eqn (1,27).

$${\displaystyle \begin{array}{l}{\displaystyle LCS=(0.05\ast x)-0.15}\end{array}}$$ (1)

Lower log CS indicates more dysfunction. Binocular log CS was selected as the independent variable of interest for the current study. In the remainder of this document, we refer to binocular log CS as CS for simplicity.

Functional MRI Protocol

At both the baseline and the 30-month visit, brain images were acquired on a Siemens 3T Skyra MRI Scanner, which has a 32-channel head coil. Each MRI scan session lasted for about 1 h. A T1-weighted 3D volumetric MPRAGE sequence was employed to acquire an anatomical image, with parameters as follows: TR = 2 300 ms; TE = 2.98 ms; number of slices = 192; slice thickness = 1.0 mm; voxel dimensions = 1.0 × 1.0 × 1.0 mm; FOV = 256 mm. Functional MRI (fMRI) data was collected utilizing blood oxygenation level-dependent (BOLD) imaging (28) during resting-state, a MI task, and an attention task (TR = 2 000 ms; TE = 25 ms; number of slices = 35; slice thickness = 5.0 mm; voxel dimensions = 4.0 × 4.0 × 5.0 mm; FOV = 256 mm). Only resting-state and MI data are presented here. In all participants, the resting-state scan first took place and was followed by the MI task. The resting-state scan contained 217 volumes. The MI task scan contained 127. The BOLD scans were conducted parallel to the anterior commissure–posterior commissure using multislice gradient-echo planar imaging (29).

During the scans, all participants were positioned supine in the MRI scanner with a view of an MR-compatible monitor positioned at the head-end of the scanner, which they observed through a mirror. During the resting-state scan, a fixation cross was displayed on the monitor, whereas for the visual MI tasks, continuous feed videos were played on the monitor. The adapted videos were from the Mobility Assessment Tool—short form (MAT-sf) (30,31) and were played on a computer via a standard media player. The video featured an avatar performing mobility tasks classified as “easy” based on previous research involving older adults (30). Before entering the MRI scanner, participants practiced engaging in the MI task while watching a shortened preview of the MAT-sf videos so that they knew what to reasonably expect. They were instructed to envision themselves as the avatar shown ambulating through space. This task is thought to engage spatial attention and includes visual features that require CS processing.

Brain Networks and Mobility participants were asked 3 verification questions after completing the MI task. We utilized a visual analog scale (VAS) of how well participants were able to perform the visualization task based on their self-report. The VAS tool in RedCap was a slider on a line with no numbers or values except the labeled ends which indicated how well the participant felt they performed. Once the MI task finished, participants were asked the question and then were asked to indicate where to move the slider over the MRI speaker. The position of the slider was verified with the participant another time before moving on to the next question or task. This slider position was then logged as a score that ranged from 0 to 100. The mean participant VAS scores at baseline are reported in Table 1. Participants were asked these questions following the MI task and indicated where to move the slider on the VAS:

Table 1.

Participant Demographic Characteristics and Descriptive Statistics at 30 Months

	Older Adults (N = 172)
Age—mean (SD)	76.4 (4.76)
Race/ethnicity—n (%)
Non-White/Hispanic	20 (11.6)
White/Non-Hispanic	152 (88.4)
Sex—n (%)
Female	97 (56.40)
Male	75 (43.60)
Binocular log CS—mean (SD)	1.71 (0.14)
Binocular logMAR Visual Acuity—mean (SD)	0.09 (0.11)
MoCA—mean (SD)	25.6 (2.22)
VAS—mean (SD)
1.How well were you able to imagine yourself doing the actual task?	90.96 (16.11)
2.Were you able to stay with the task for the entire 4 min?	95.98 (10.12)
3.The activities you just visualized have a similar level of difficulty. In general, could you perform tasks of that difficulty?	96.39 (11.94)

Notes: The table outlines the baseline characteristics of the B-NET participants that completed a second fMRI at 30 months of study enrollment. Binocular visual acuity and Montreal Cognitive Assessment (MoCA) were restricted based on inclusion criteria. 2 B-NET participants failed to complete the VAS during their baseline brain imaging. logMAR = logarithm of the minimum angle of resolution; VAS = visual analog scale.

1. How well were you able to imagine yourself doing the actual task? (Not at all to Extremely clear)

2. Were you able to stay with the task for the entire 4 min? (Not at all to Engaged for the full 4 min)

3. The activities you just visualized have a similar level of difficulty. In general, could you perform tasks of the difficulty? (Not at all to Could perform all the tasks)

Neuroimaging Analyses

High-resolution anatomical image

Structural image segmentation was performed by using Statistical Parametric Mapping v.12 (SPM12, available at http://www.fil.ion.ucl.ac.uk/spm). Segmented gray and white matter brain images were combined and all voxels possessing a probability greater than 0.5 were kept, which generated a mask including the brain parenchyma while excluding nonbrain tissue and cerebral spinal fluid (CSF). This mask was applied to the structural image. To confirm accuracy and ensure comprehensive coverage of the entire brain, the masked structural images were visually inspected and manually cleaned to remove any remaining extraparenchymal tissues. This cleaning process was performed using MRIcron software (accessible at https://www.nitrc.org/projects/mricron). The masked and cleaned T1-weighted images were then spatially normalized to the Montreal Neurological Institute (MNI) 152 template (https://www.bic.mni.mcgill.ca/ServicesAtlases/ICBM152NLin2009) using Advanced Normalization Tools (ANTS).

Functional image analyses

Image distortion correction was executed using FMRIB’s “Topup” Software Library (FSL, www.fmrib.ox.ac.uk/fsl). To enable signal normalization, the first 10 volumes of the BOLD images were excluded. SPM12 was utilized for slice time correction and realignment of functional images. Blood oxygenation level-dependent images were aligned with anatomical images and spatially normalized to MNI space by applying the previously defined warp using ANTS. A motion scrubbing procedure (32) identified and removed volumes with excessive movement (> 0.5 mm FD, framewise displacement) and excessive signal change (> 0.5 DV_GM, average change in mean gray matter signal) (32). One hundred and seventeen participants required additional scrubbing for resting-state images and 112 participants for the MI task out of 172 older adult participants. An average of 7.9 ± 14.2 (SD) volumes were removed for resting-state and 8.8 ± 12.1 volumes for the MI task. Finally, the data underwent band-pass filtering (0.009–0.08 Hz) to correct for low-frequency drift and physiological noise. Global signals from white matter, gray matter, CSF, and realignment parameters were regressed from the filtered data. White matter and CSF segments were thresholded to 0.9 and 0.999, respectively.

Brain Network Generation

A voxel-wise correlation was conducted to generate a cross-correlation matrix with Pearson’s correlation coefficient in each cell to represent connectivity strength between each pair of nodes. The correlation matrices were thresholded to generate sparse networks with the density matched across all participants by using the following formula: S = log(N)/log(K), where N is the number of nodes and K is the average number of connections per node. S was empirically determined (33) to be 2.5. All values above the threshold were assigned a value of 1, indicating the presence of a connection, while all other values were set to 0, indicating no connection.

Network Community Structure

A network community is a group of nodes exhibiting stronger connectivity among themselves compared to other network nodes (34). Each participant’s binary network was divided into distinct communities where each voxel (node) was assigned to a single community. A dynamic Markov process was used to optimize modularity, or Q, to identify the network community partition at the maximum value of Q (34,35). The modularity algorithm was performed 100 times and the partition with the highest Q was selected for each participant.

To compare the spatial alignment of the communities across participants, Scaled Inclusivity (SI) was computed (36) using predefined a priori networks covering the entire brain. These networks were generated from a previous study with healthy adult resting-state brain network data (37) and included these 8 intrinsic subnetworks typically identify across studies: basal ganglia network, central executive network (CEN), DMN, DAN, frontotemporal network (FTN), SMN, salience network (SN), and VN (38). The SI values range from 0 to 1, where a value of 1 indicates complete spatial alignment between the subnetwork and the community structure in the participant. Lower SI value indicates that the network communities do not align well with the spatial pattern of the predefined subnetworks. The SI maps for each individual participant were used as the dependent variable in the regression analyses described below.

For visualization purposes, group community structure maps were generated by averaging voxel-wise SI values for each analysis (eg, DMN during the MI task) across participants for the upper and lower tertiles of baseline CS. These brain images were used to demonstrate the directionality of the significant associations but were not used in the performance of the statistical analyses. All images were generated with MRIcron.

Statistical Analyses

Separate distance regression models were constructed to examine if baseline CS was significantly associated with each of the 8 intrinsic functional brain networks covering the entire brain at 30 months. The regression model operates in MATLAB (v. R2021a) and was specifically developed to examine the relationships between the spatial pattern of brain network organization and continuous and/or categorical phenotypes (39). This model allowed us to control for sex, the number of volumes removed during motion scrubbing from head motion in the scanner (covariate name “Volumes removed”), and baseline brain network community structure organization. Sex was included because we know there are sex differences in brain networks. In a supplemental analysis, we also included baseline MoCA score and change in MoCA score (from baseline to 30 months) in the model. Separate regression models were built for resting-state and the MI task with 30-month network community structure as the dependent variable and CS as the independent variable of interest. A Jaccardized Czekanowski similarity index (40) was calculated to compare the SI maps across participant pairs. The Jaccard distance (1-Jaccard index) was computed and used in the regression analyses. An absolute distance between participants was utilized for each of the independent variables. The power of this regression method is that it allows one to identify associations between the spatial pattern of the network communities and other variables of interest, such as CS. However, the method was not developed to assess longitudinal change within a given subject’s brain network community structure. Such a longitudinal analysis would require computing the distance between distance (change) matrices and the methodology has not been validated on such data structures. However, we included the baseline brain networks as a covariate in our model to ensure the relationship of baseline CS with 30-month brain networks is not due to baseline networks.

Statistical significance was set at p < .05 for all analyses. To correct for multiple comparison across the multiple tested models, an adapted false discovery rate (FDR) analysis was used (41).

Results

Demographic data for B-NET participants at 30 months is listed in Table 1. Over 98.5% of participants saw 20/40 or better with corrective lenses if applicable. In total, 172 of the 192 original B-NET participants were retained at 30 months (mean age of 76.4 ± 4.8 years; 56.4% female; 11.6% non-White/Hispanic). The mean participant VAS scores at baseline are sufficient as confirmation that the MI task was successfully completed. Out of 100 points, participants averaged 90.96, 95.98, and 96.39 for questions 1–3, respectively (Table 1), allowing us to be confident that B-NET participants completed the MI task well.

Table 2 describes the associations of baseline CS with the 30-month brain networks at rest and during the MI task while controlling for baseline brain network community structure. Volumes removed during head motion correction and sex were also included as covariates in each model (Table 2). Contrast sensitivity was significantly associated with DMN (p < .0001) community structure at rest and during the MI task (p = .0002). Contrast sensitivity was also significantly associated with the DAN (p = .0089) and VN (p < .0001) during the MI task, but not at rest (DAN, p = .9640; VN, p = .5914). Sex was not significantly associated with any of the brain networks (Table 2). None of the other brain networks exhibited associations with CS at rest or during the MI task at 30 months (see Supplementary Table 1). The associations of CS with DMN, DAN, and VN remained significant after adjusting for FDR.

Table 2.

Associations Between Baseline Contrast Sensitivity and 30-month Brain Networks in 172 Participants

Network/Condition	Variable	Estimate	SE	T score	p Value	FDR
DMN/Rest	Binocular CS	0.0173	0.0032	5.3681	<.0001	<0.0001
	Volumes removed	0.0003	<0.0001	7.9756	<.0001
	Sex	0.0007	0.0006	1.0957	.2732
	Baseline BN	0.2255	0.0087	25.8679	<.0001
DMN/Task	Binocular CS	0.0103	0.0028	3.6793	.0002	0.0009
	Volumes removed	0.0006	<0.0001	15.5745	<.0001
	Sex	<0.0001	0.0006	0.0331	.9736
	Baseline BN	0.3538	0.0083	42.7900	<.0001
DAN/Rest	Binocular CS	0.0001	0.0028	0.0451	.9640	0.9640
	Volumes removed	0.0002	<0.0001	6.8234	<0.0001
	Sex	0.0004	0.0006	0.8002	.4236
	Baseline BN	0.0614	0.0084	7.2908	<.0001
DAN/Task	Binocular CS	0.0075	0.0029	2.6164	.0089	0.0237
	Volumes removed	0.0004	<0.0001	11.1696	<.0001
	Sex	<0.0001	0.0006	0.0089	.9929
	Baseline BN	0.1787	0.0099	18.0757	<.0001
VN/Rest	Binocular CS	0.0020	0.0036	0.5369	.5914	0.9640
	Volumes removed	0.0002	<0.0001	4.3453	<.0001
	Sex	-0.0005	0.0007	-0.6826	.4948
	Baseline BN	0.1013	0.0093	10.9152	<.0001
VN/Task	Binocular CS	0.0115	0.0027	4.3471	<.0001	0.0001
	Volumes removed	0.0005	<0.0001	14.2105	<.0001
	Sex	-0.0001	0.0005	-0.2618	.7935
	Baseline BN	0.1795	0.0089	20.2594	<.0001

Notes: The table includes the 30-month community structure statistical model results in the 172 older adults that completed 30-month fMRI in B-NET while controlling for baseline brain networks (BN). All models included binocular contrast sensitivity (CS), brain volumes removed (motion correction), sex, and baseline brain networks as the independent variables. The underlined networks/conditions p values were significant (p < .05). The bolded text indicates a variable that reached significance. SE = standard error; the motion correction variable reflects the number of brain volumes removed during image correction. The last column denotes the p value after a false discovery rate (FDR) procedure was applied for multiple comparisons. The underlined networks/conditions of the default mode network (DMN), dorsal attention network (DAN), and visual network (VN) significance with binocular CS did not change post FDR-correction. Supplementary Table 1 contains the results for the remaining 5 subnetworks we analyzed.

The community structure maps shown in Figure 1 illustrate the directionality of the significant association between CS and brain network community structure at 30 months. All images exhibit similar spatial patterns of community structure between the upper and lower tertile groups based on CS. The low CS tertile group exhibited significantly lower connectivity, seen through predominantly cool colors, compared to the upper CS group. The figure also shows that there was overall greater consistency in the DMN during rest than during the MI task (note the difference in color bar scale).

Figure 1.

30-month community structure brain maps for the groups in upper and lower tertiles of baseline contrast sensitivity (CS). (A) Maps for the default mode network (DMN) at Rest. (B) Maps for the DMN, dorsal attention network (DAN), and visual network (VN) during the motor imagery (MI) task. Brain regions with hotter colors on the community structure scale were more frequently part of the community across participants, indicating greater spatial consistency. Note that all of the MI task brain maps are on the same community structure scale, and the scale for the DMN at Rest is higher than the scale for the MI task. Each image collage contains a sagittal slice (x: DMN = −5, DAN = −41, VN = 11), an axial slice (z: DMN = 28, DAN = 3, VN = 4), and a coronal slice (y: DMN = −55, DAN = −66, VN = −84).

Supplementary Table 2 incorporates baseline cognitive scores (MoCA) in the 3 networks we found to be significant, the DMN, DAN, and VN. This was done to ensure that any changes in brain connectivity were related to CS and not differences in cognition in our cohort. We found that CS was significantly associated with the DMN (p < .0001) and VN (p = .0336) community structure at rest, and during the MI task (DMN, p = .0001; VN, p < .0001). Contrast sensitivity was also significantly associated with the DAN (p = .0109) during the MI task, but not at rest (p = .1896). Supplementary Table 3 included the change in MoCA scores, calculated by subtracting baseline MoCA from the 30-month MoCA score, in our model. The findings were similar for the models with baseline MoCA (Supplementary Table 2) and change in MoCA (Supplementary Table 3). The VN at rest is the only network/condition in Supplementary Tables 2 and 3 that did not remain significant when controlling for multiple comparisons.

Discussion

We demonstrated that baseline CS predicted 30-month brain network community structure in networks important for cognition, physical function, and vision, including the DMN, DAN, and VN, respectively. Moreover, we observed that the strength of association between baseline CS and these networks measured at 30 months was qualitatively stronger than the associations with networks measured at baseline. Contrast sensitivity was measured 30 months prior to the brain imaging collection and this suggests that poor CS may predict future degradations in brain networks. Given the disruptions in community structure integrity associated with low CS, cognitive, physical function, and visual deficits may emerge in the DMN, DAN, and VN, respectively. Future studies should investigate whether interventions that improve CS could be used to improve future brain health within these 3 networks.

The DMN was the only network that was significantly associated with baseline CS during both states, likely corresponding to the diversified functional responsibilities the DMN possesses. The association at rest is most likely because the DMN is known to be a key network engaged at rest (11,42). Moreover, the CS-DMN relationship observed during the MI task may be consistent with low CS impairing the self-centered forecast that internally evaluates an older adult’s physical function capabilities and mobility. If an individual struggles to detect their environment due to CS deficits, it makes sense that their internal processes regarding their own mobility reflects the mobility level in a real setting (43).

The DAN’s role in processing and integrating sensory cues, particularly those important for moving through the environment, is relevant to our findings. We have previously shown that DAN community structure is associated with physical function most notably when performing the MI task that includes considerable contrast cues (44). We propose that individuals with low CS may have difficulties with orientation, movement, and integration of sensory cues. Typically, the DAN prepares responses to external stimuli by transmitting crucial visuospatial information downstream of motor planning regions. As demonstrated in older adults with poor CS, this impaired visuospatial sensory integration process is likely underpinned by disrupted DAN. Consequently, these individuals may struggle with visualizing movement during the MI task and maintaining balance due to DAN degradation. For individuals with poor CS, garnering the necessary attention for a task may be more difficult to achieve, which may be observed as DAN dysfunction. This is supported by evidence that associates processing speed and attentional deficits with low CS in older adults (45).

Notably, CS was associated with the SMN at rest in the baseline analysis and this was not observed at 30 months. This may indicate that the mobility-related effects associated with poor CS are more complicated and cannot be precisely detected by brain network analyses. Another possibility is that, at baseline, low CS leads to early degradation of SMN community structure and no further degradation over the 30 months. One may also hypothesize that the SN would be involved with CS as the SN is important for determining if attention is directed toward internal or external processing. In essence, the SN helps switch between the DMN and CEN processing (46). We did not see any SN or CEN findings in our original baseline work (6). It is possible that CS is not critical to the selection of attention toward external stimuli and is not related to SN or CEN but is essential to spatial processing once attention is directed toward the external environment.

It makes intuitive sense that older adults with poor CS demonstrate a degraded VN when visualizing the MI task because the input to the visual cortex carries deficient contrast details. Thus, low CS may be leading to some disuse of the visual cortex. Cortical CS is continuously adjusted by the visual cortex so that the brain captures all the most crucial contrasts, which differ in bright versus dark scenes (47). At the brain level, changes in CS are facilitated by the visual cortex as it must process both contrast and luminance changes (48). These 2 cortical signals continuously interact to sample the visual environment. Potentially, degraded VN community structure in older adults with low CS may be a representation of deficient cortical CS. However, we cannot determine whether the observed relationship between poor CS and VN dysfunction may instead be a result of the visual cortex not receiving enough contrast information.

While the current study suggests a relationship between baseline CS and future brain networks due to its longitudinal design, it does not provide conclusive causal evidence that baseline CS results in future declines in brain network community structure. Such a claim would require that we directly examine brain network change scores. Since we were regressing continuous measures against spatial patterns, a distance regression was used. Unfortunately, using distances of distances (ie, change scores) in the regression model is not currently validated. In addition, simplifying the data by reducing the community structure into a single numeric value cannot be done while retaining the complex spatial patterns that exist in the brain. Although we were not able to examine the community structure change directly, we included the baseline brain networks as covariates to demonstrate that the relationship of CS with future brain networks was not dependent on the baseline brain function.

A second limitation involves the visual stimuli presented to the B-NET participants in the fMRI both at rest and during the MI task. If participants with low CS perceived lower visual stimuli in the fMRI, their brain networks may not have been stimulated to the same degree as an individual with high CS. Thus, the finding may be more directly related to retinal and optic nerve signals in response to the stimuli rather than intrinsic brain connectivity. This issue could be accommodated for in future studies by testing participant CS prior to an fMRI and adjusting the contrast on the visual stimuli to offset their CS functional deficiency.

Third, the B-NET study population is limiting as this cohort was fairly homogeneous and was not representative of the general older adult population. B-NET participants were relatively healthy, cognitively unimpaired, and had good visual acuity at baseline. On the other hand, this allowed us to study the effects of CS independent of high contrast visual acuity, and it means that this cohort could have strong brain network associations with areas important for mobility.

Even though the MoCA score data from B-NET did not display much spread since participants were cognitively unimpaired at baseline (Table 2), we included supplemental models with baseline MoCA (Supplementary Table 2) and change in MoCA scores (Supplementary Table 3) to ensure that 30-month brain network connectivity was driven by CS independent of the MoCA score. We found that all the main findings for the DMN and DAN remained significant, but that the VN at rest (p = .0336, Supplementary Table 2; p = .0328, Supplementary Table 3) was no longer significant after FDR correction (p = .0503, Supplementary Table 2; p = .0492, Supplementary Table 3). Potentially, this is due to only the DMN, DAN, and VN being included in the FDR analyses for Supplementary Tables 2 and 3. We did this to limit the multiple comparisons considering that the other 5 networks showed no relationship with CS (Supplementary Table 1). Although the VN at rest was no longer significant after FDR correction (p was = 0.05), we believe it is more plausible that differences in brain network connectivity were related to CS and not cognitive differences in our cohort as the cognitive range was very limited due to the original intentional design of B-NET which excluded those with low MoCA scores.

Future studies should investigate whether change in CS is associated with change in brain networks, and if improving CS increases functional brain network connectivity, improves mobility in older adults, and strengthens self-reported vision measures. CS testing is not currently used in geriatric medicine. However, if future research demonstrates its significant relevance to brain health, CS testing may be warranted for integration into general medical care of older adults. The established predictive relationship between baseline CS and brain network connectivity at 30 months provides the basis to examine if mobility is associated with the same networks we identified to be associated with CS. However, further mobility investigations will require mediation analysis. Given that CS is an input, it is possible that poor CS leads to further degeneration possibly due to low signal, which causes CS to get weaker over time. This work has the potential to generate a more accurate understanding of the mechanism underlying the interrelatedness of vision, mobility, and cognitive changes observed during normal aging.

Conclusion

Poor CS predicted future brain network integrity in circuits essential for cognition, mobility, and vision—the DMN, DAN, and VN—and this strength of association demonstrated qualitative increases from baseline to 30 months. In all cases, these results demonstrated a decrease in the spatial consistency of the community structure with lower CS. These findings suggest that CS dysfunction could be an indicator of poor brain health in cognitively unimpaired older adults. Whether intervening on CS can improve brain health and cognitive or mobility function should be considered in future studies.

Funding

This work was supported by R01-AG052419, the Wake Forest University Claude D. Pepper Older Americans Independence Center (P30-AG021332) and the Wake Forest Clinical and Translational Science Institute (UL1-TR001420). Dr. Thompson is supported by K23EY030897 from the National Eye Institute. The funding sources had no role in the conduct of this study.

Conflict of Interest

None.

Author Contributions

Study design: P.J.L., S.B.K., A.C.T., A.D.T. Funding acquisition: P.J.L., S.B.K. Analyses: A.D.T., R.G.L., H.C., A.C.T. Statistical interpretation: M.E.M., P.J.L., H.C., R.G.L., A.D.T., A.C.T. Manuscript preparation: A.D.T., A.C.T., P.J.L. Manuscript revision and final approval: A.D.T., C.E.H., J.D.W., M.E.M., P.J.L., S.B.K., A.C.T., R.G.L., H.C.