Этнорасовые диспропорции в атрофии серых веществ сдерживаются структурным разъединением в множестве склерозов.

Introduction

Multiple sclerosis (MS) is the most prevalent chronic neurologic disease in young adults in the United States (US),
1
and a leading cause of non‐traumatic disability.
2
Recent studies have shown how MS is more prevalent among minority populations than previously thought,
3
highlighting significant ethnoracial disparities. Non‐Hispanic Black (B‐MS) and Hispanic/Latinx (H‐MS) patients with MS (PwMS) often present earlier, with greater disease burden and faster disability accumulation compared to non‐Hispanic White (W‐MS) PwMS.
4
,
5
,
6
,
7
Imaging studies have shed light on some of the structural underpinnings of such disparities. B‐MS and H‐MS were previously reported to have higher cortical gray matter (GM) atrophy relative to W‐MS, most pronounced in temporal and parietal regions in B‐MS. H‐MS were also shown to have greater Deep GM (DGM) atrophy.
8
,
9
Moreover, B‐MS and W‐MS were shown to have greater inflammatory white matter (WM) lesion burden compared to W‐MS, especially periventricularly and infratentorially.
10
It is thought that a host of biological, genetic, and sociocultural factors contribute to the observed disparities.
11

MS pathology is characterized by concomitant neuroinflammation and neurodegeneration, manifesting as focal demyelination, axonal disruption, and brain atrophy.
12
GM atrophy, a chief correlate of physical disability and cognitive decline in PwMS, appears early and worsens with disease progression.
13
,
14
Various mechanisms are thought to contribute to GM neuroaxonal loss, including retrograde degeneration secondary to WM lesions, focal cortical demyelination,
15
,
16
,
17
,
18
and DGM degeneration, with thalamic atrophy serving as a significant marker of neurodegeneration due to its role as a central hub in brain networks.
19
Voxel‐based morphometry
20
(VBM) quantifies and analyzes regional GM atrophy using magnetic resonance imaging (MRI), which can be particularly useful in deriving group‐level insights.
21
,
22
It has shown regional GM loss in PwMS which correlates with disease duration and severity,
23
though its univariate approach analyzing voxels independently limits its ability to distinguish between different atrophy patterns, which can be clinically relevant in PwMS.
24
,
25
An alternative multivariate data‐driven technique, source‐based morphometry (SBM), pools signals across different voxels exhibiting covariance within brain regions using independent component analysis (ICA).
26

However, while these techniques provide valuable insights into GM atrophy, they do not address the relationship between white matter lesions and GM atrophy. MS‐related WM lesions disrupt brain networks, leading to structural disconnectivity that correlates with both GM pathology and clinical outcomes.
27
,
28
The Network Modification tool (NeMo) quantifies this disconnectivity by estimating WM pathway disruption and GM disconnection using the Human Connectome Project normative dataset,
29
demonstrating comparable utility in predicting clinical disability in PwMS relative to diffusion MRI measures.
28
To untangle the complex relationships between multiple outcome‐related variables, mediation analysis provides a statistical framework for understanding how variables may influence each other. This approach helps determine whether the effect of one variable on another occurs directly or indirectly through an intermediary variable (mediator), offering insights into the possibly contributing mechanisms.
30
It has been recently used to investigate the mediating role of paramagnetic rim lesions on the relationship between race and disability in PwMS.
31

We aimed to investigate ethnoracial disparities in gray matter (GM) atrophy in PwMS in a large single‐center cohort, testing the following hypotheses: (1) B‐MS, H‐MS, and W‐MS exhibit different extent and patterns of GM atrophy; (2) WM lesion burden and resultant structural disconnectivity mediate these ethnoracial differences in GM atrophy.

Methods

Subjects

Demographic, clinical, and imaging data were obtained under human research and subject protocol approved by the McGovern Medical School at UTHealth institutional review board. This retrospective study included patients with MS diagnosis, seen between May 2009 and August 2022 at the comprehensive MS care clinic at UT Physicians. MS diagnosis was initially identified by ICD‐9 (340) and/or ICD‐10 (G35) codes, subsequently individual clinical, radiological, and laboratory data were reviewed by two MS specialists (C.A.P. and J.A.L.) to confirm dissemination in time and space meeting the 2017 McDonald diagnostic criteria. Previous publications describe the studied cohort in detail.
6
,
7
,
9
Inclusion criteria included MRI scans obtained within one month of the clinical evaluation visit and the absence of brain trauma or other demyelinating central nervous system conditions upon record review. Exclusion criteria included patients with incomplete medical records, lacking MRI data, or patients with MRI data obtained within 30 days of a clinically documented relapse, corticosteroid administration, and/or disease‐modifying therapy (DMT) switching. Demographic and clinical data were extracted as previously described
9
from medical records including age, sex, disease duration, treatment delay (time interval between first clinical symptom onset and DMT initiation), disease phenotype, Expanded Disability Status Score (EDSS), DMT, body mass index (BMI), and other comorbidities as evidenced by diagnosis codes or corresponding pharmacotherapy. Subjects were divided into three groups according to their self‐reported race/ethnicity: B‐MS, H‐MS, and W‐MS. We initially selected equal numbers of patients, 100, across the three groups (1:1:1), of frequency matching was used to ensure comparable distributions of age and sex across the groups.

Image acquisition

MRI was performed on a single 3.0 T Philips Ingenia research scanner with a maximum gradient amplitude of 45 mT/m and a 15‐channel SENSE‐compatible head coil (Philips Medical Systems, Best, Netherlands). High‐resolution 3D T 1 ‐weighted (T 1 ‐w) magnetization prepared rapid gradient echo images (voxel size: 1 × 1 × 1 mm
3
, field‐of‐view (FOV): 256 × 256 mm 2 , repetition time/echo time: 8/3.7 msec), as well as T 2 ‐weighted fluid‐attenuated inversion recovery (T 2 ‐w FLAIR) images (voxel size: 1 × 1 × 1 mm 3 , FOV: 256 × 256 mm 2 , repetition time/echo time: 4800/300 msec, and inversion time: 1650 msec) were acquired.

Image analysis

An overview of the image processing workflow is illustrated in Figure  1 . Details on the image analysis methods can be found in the Supplementary Materials .

Image analysis workflow. (1) Image co‐registration and WM hyperintense lesions segmentation. (2) Inpainting of lesioned voxels on T 1 ‐w images. (3) Image normalization and tissue class segmentation. (4) Univariate analysis of GM group differences. (5) Multivariate analysis of GM group differences. (6) Mean group lesion distribution and their differences. (7) Quantification of interrupted white matter fibers attributed to lesion distribution. (8) Calculation of regional gray matter disconnection secondary to lesion‐related white matter fiber interruption.

Preprocessing and lesion segmentation

In FSL (version 6.0.6, FMRIB Software Library; http://www.fmrib.ox.ac.uk/fsl ), T 1 ‐w and T 2 ‐FLAIR images were bias‐field corrected and co‐registered. WM hyperintense lesions were segmented on the T1‐w and T2‐w FLAIR images estimating total lesion distribution and volume, as well as labeled regional volumes (periventricular, subcortical, juxtacortical, and infratentorial) using the Lesion Segmentation Toolbox.
32
Produced lesion masks were used to fill T 1 ‐w images
33
to mitigate the effect of lesioned voxels on image normalization.
34
The produced lesion masks and the quality of the filled T 1 ‐w images were visually assessed.

GM assessment

We utilized complementary techniques to comprehensively evaluate GM differences across the studied groups. VBM was utilized as a univariate method to assess overall differences in global GM atrophy, and SBM was used to provide insights on topographically co‐occurring GM loss shedding light on the atrophy patterns.

VBM

The lesion‐filled T 1 ‐w images were processed using the standard Computational Anatomy Toolbox 12 (CAT12; https://neuro‐jena.github.io/cat ) pipeline, which has been validated in cross‐sectional MS studies and shown to provide robust segmentations.
34
,
35
The images were registered to standard space and segmented, estimating total intracranial (TIV), GM, WM volumes, and the different brain tissue maps including the modulated normalized GM maps. The T 1 ‐w images and produced GM maps were assessed for image quality and sample inhomogeneity, outliers (>±2 standard deviations) were excluded from the analysis.
36
Groups were compared for regional GM differences including age, sex, and TIV as covariates. After probabilistic threshold‐free cluster enhancement,
37
results were produced controlling for false discovery rate (FDR).

SBM

To investigate GM regions with covarying atrophy, the GIFT toolbox ( https://github.com/trendscenter/gift ) was used. We estimated the number of components
26
and performed independent component analysis using the Infomax algorithm
38
on the modulated normalized GM maps. The reliability of the identified GM components (GMCs) was ensured using 100‐fold cross‐validation in ICASSO ( https://research.ics.aalto.fi/ica/icasso/ ). Subjects' loading coefficients for the GMCs were obtained and converted to Z ‐scores.

Structural disconnectometry

NeMo
29
was used to calculate the change in connectivity (ChaCo) scores to the different GM regions. The standardized lesion masks were used to define the subsequent pattern of structural disconnectivity as the percentage of tractography streamlines connected to the region intersecting with the lesion mask. The target GM regions used were whole‐brain GM atlas parcellations as well as the identified gray matter patterns, GMCs, obtained using SBM.

Statistical analysis

Statistical analyses were conducted in jamovi (Version 2.3, https://www.jamovi.org/ ), p ‐values < 0.05 were considered significant. P ‐values were adjusted for FDR using the Benjamini–Hochberg procedure when appropriate. Continuous demographic, clinical, and imaging variables were assessed for normality of distribution using Q‐Q plots and equal variances using Levene's test. Demographic and clinical variables were compared using analysis of variance (ANOVA), Kruskal–Wallis, and chi‐square tests whenever appropriate. GM fraction (GMF) was calculated as total GM volume divided by TIV multiplied by 100, and lesion load (LL) was calculated as total lesion volume divided by TIV multiplied by 1000. Analysis of covariance (ANCOVA) was used to compare the identified GMCs among the ethnoracial groups, controlling for age, sex, and TIV. Partial eta squared ( η p 2 ) was calculated to estimate the variance explained by ethnoracial group membership, and post‐hoc comparisons were conducted using the Games‐Howell test. As the structural disconnectivity, ChaCo, scores did not follow normal distribution, the Kruskal–Wallis test was used to calculate group differences. The Dwass–Steel–Critchlow–Fligner test was used to estimate pair‐wise differences. Voxel‐wise analysis of the WM lesions was conducted estimating group means and differences from 10,000 non‐parametric permutations performed using randomize in FSL on the standardized lesion masks.

Finally, to examine the relationship between the ethnoracial group, WM lesions/structural disconnectivity, and GM atrophy differences observed among the groups, classical non‐parametric mediation analyses were conducted. We examined how ethnoracial group membership affected global GM atrophy via total WM lesion volume, LL, as well as how it affected GMC differences via its corresponding structural disconnectivity secondary to WM lesions. The predictor, ethnoracial group, was used as a categorical independent variable with three contrasts: (1) combined B‐MS + H‐MS groups vs. W‐MS, (2) B‐MS vs. W‐MS, and (3) H‐MS vs. W‐MS, with the W‐MS group being the reference level. We estimated total, direct and indirect effects ( η p 2 ), including age, sex, disease duration, clinical phenotype, DMT, and TIV (when appropriate), as covariates, using a maximum likelihood estimator with confidence intervals obtained from 10,000 percentile‐corrected bootstraps in a structural equation modeling framework using lavaan.
39
For the first contrast (B‐MS + H‐MS vs. W‐MS), which had the largest sample size and thus greater statistical power, we calculated the coefficient of determination ( R
2 ) for the direct ( R D 2 ) and total effects ( R T 2 ), quantifying the proportion of variance in GM atrophy explained by each. The percent mediation (Pm%) was estimated by dividing indirect effect η p 2 by total effect η p 2 multiplied by 100.

Results

Patient characteristics

A total of 316 records were assessed for inclusion, out of which 19 outliers were excluded (2 from B‐MS, 10 from W‐MS, and 7 from H‐MS). The analysis included 297 PwMS, divided into three groups: B‐MS ( n  = 82, mean age = 39.3 ± 10.5 years, 79.3% female), W‐MS ( n  = 117, mean age = 40.8 ± 12.5 years, 71.8% female), and H‐MS ( n  = 98, mean age = 37.4 ± 11.6 years, 76.5% female). Table  1 summarizes demographic, clinical and imaging data for included subjects. The three groups did not show significant differences in MS disease duration (B‐MS: mean disease duration = 6.9 ± 6.2 years, W‐MS: mean disease duration = 6.9 ± 7.9, and H‐MS: mean disease duration = 6.3 ± 7.3 years, p  = 0.4), treatment delay, EDSS, disease phenotype, DMT, or other comorbidities. Total gray matter fraction was significantly lower in B‐MS and H‐MS compared to W‐MS (B‐MS: 43.7 ± 3.6%, W‐MS: 45.2 ± 3.1%, and H‐MS: 44.8 ± 3.5%, p  = 0.02), and lesion load was greater (B‐MS: 13.0 ± 15.9 mL, W‐MS: 5.6 ± 8.3 mL, and H‐MS: 10.5 ± 13.3 mL, p  < 0.001). Figure  2 depicts the lesion probability maps for each group and their differences. Compared to W‐MS, B‐MS and H‐MS had significantly higher callosal and periventricular lesion volumes, specifically interrupting commissural pathways, forceps minor anteriorly, and forceps major posteriorly, connecting corresponding cortical regions bilaterally as well as association pathways connecting various cortical regions ipsilaterally. More details on the distribution of lesion differences can be found in Table  S1 .

Table 1 Patient demographic, clinical, and imaging characteristics. B‐MS ( n  = 82) H‐MS ( n  = 98) W‐MS ( n  = 117)
p ‐value Age (years) 39.3 ± 10.5 37.4 ± 11.6 40.8 ± 12.5 0.1 Sex (% female) 79.3 76.5 71.8 0.5 Disease duration (years) 6.9 ± 6.2 6.3 ± 7.3 6.9 ± 7.9 0.4 Treatment delay
a
(years) 0.3, 1.3 0.6, 1.9 0.6, 3.7 0.06 EDSS
a

2.0, 2.0 2.0, 2.0 1.5, 2.5 0.3 MS phenotype (%) 0.2 PPMS 11.0 3.1 7.7 RRMS 81.7 83.7 77.8 SPMS 7.3 13.3 14.5 DMT (%) 0.4 Injectable
b

69.5 57.1 60.7 Oral
c

12.2 17.3 16.2 Infusion
d

13.4 13.3 16.2 None 4.9 12.2 6.8 BMI (kg/m 2 ) 28.4 ± 7.2 28.5 ± 7.2 27.5 ± 7.1 0.5 Smoking (%) Current 11.0 6.1 6.8 0.5 Never 79.3 79.6 76.1 Previous 9.8 14.3 17.1 Comorbidities (%) Hypertension 19.5 32.7 27.4 0.1 Hyperlipidemia 8.5 9.2 9.4 1.0 Diabetes mellitus 7.3 6.1 4.3 0.6 GMF (%) 43.7 ± 3.6 44.8 ± 3.5 45.2 ± 3.1
0.02
WMF (%) 36.1 ± 2.9 35.7 ± 2.8 36.1 ± 2.5 0.5 Lesion load (mL) 13.0 ± 15.9 10.5 ± 13.3 5.6 ± 8.3
<0.001
Periventricular 14.9 ± 19.3 12.0 ± 17.0 6.3 ± 10.5
<0.001
Subcortical 0.3 ± 0.3 0.3 ± 0.4 0.3 ± 0.3 0.8 Juxtacortical 1.3 ± 1.5 1.3 ± 1.8 0.9 ± 1.4 0.1 Infratentorial 0.2 ± 0.2 0.2 ± 0.4 0.1 ± 0.3 0.5 Listed values are the mean group values and standard deviation where applicable. Bold values indicates p ‐values < 0.05. BMI, body mass index; B‐MS, non‐Hispanic Black PwMS group; DMT, disease‐modifying therapy; EDSS, expanded disability status scale; GMF, gray matter fraction; H‐MS, Hispanic PwMS group; MS, multiple sclerosis; PPMS, primary progressive multiple sclerosis; RRMS, relapsing‐remitting multiple sclerosis; SPMS, secondary progressive multiple sclerosis; WMF, white matter fraction.; W‐MS, non‐Hispanic White PwMS group.
a
Median and interquartile range are represented.
b
Injectable DMTs: glatiramer acetate, interferon beta‐1a, interferon beta‐1b, and peginterferon beta‐1a.
c
Oral DMTs: diroximel fumarate, dimethyl fumarate, teriflunomide, fingolimod, and siponimod.
d
Infusion DMTs: ocrelizumab, natalizumab, alemtuzumab, rituximab, and intravenous immunoglobulin (IVIG).

Patient demographic, clinical, and imaging characteristics.

Listed values are the mean group values and standard deviation where applicable. Bold values indicates p ‐values < 0.05.

BMI, body mass index; B‐MS, non‐Hispanic Black PwMS group; DMT, disease‐modifying therapy; EDSS, expanded disability status scale; GMF, gray matter fraction; H‐MS, Hispanic PwMS group; MS, multiple sclerosis; PPMS, primary progressive multiple sclerosis; RRMS, relapsing‐remitting multiple sclerosis; SPMS, secondary progressive multiple sclerosis; WMF, white matter fraction.; W‐MS, non‐Hispanic White PwMS group.

Median and interquartile range are represented.

Injectable DMTs: glatiramer acetate, interferon beta‐1a, interferon beta‐1b, and peginterferon beta‐1a.

Oral DMTs: diroximel fumarate, dimethyl fumarate, teriflunomide, fingolimod, and siponimod.

Infusion DMTs: ocrelizumab, natalizumab, alemtuzumab, rituximab, and intravenous immunoglobulin (IVIG).

Lesion probability maps and their group differences. (A–C) represent lesion probability maps for each group estimated using non‐parametric voxel‐wise statistics in B‐MS, W‐MS, and H‐MS respectively. (D and E) represent group differences in lesion distribution between B‐MS versus W‐MS (D), and H‐MS versus W‐MS (E). Higher t ‐values denote higher probabilities.

Univariate GM differences

Figure  3 shows the pairwise regional GM differences among the groups. In general, both B‐MS and H‐MS groups showed greater GM atrophy relative to W‐MS, with more evident atrophy in the B‐MS group. B‐MS had notable GM decreases in the bilateral lingual, temporal, parahippocampal, precuneus, and cuneus regions when compared to W‐MS, meanwhile, H‐MS showed the largest decreases in the bilateral fusiform, temporal, lingual, frontal, and parahippocampal regions. Compared to H‐MS, B‐MS also showed decreases in the bilateral cuneus, orbital, precuneus, and occipital regions. Whereas no areas showed significant decreases in W‐MS compared to B‐MS and H‐MS, and no areas showed significant decreases in H‐MS compared to B‐MS. Detailed results showing pair‐wise regional GM differences statistics found using VBM can be found in Table  S2 .

Univariate gray matter group differences. Pair‐wise differences in gray matter estimated using VBM depicting areas with more pronounced gray matter atrophy in B‐MS relative to W‐MS (A), H‐MS relative to W‐MS (B), and B‐MS relative to H‐MS (C). Higher t ‐values denote larger differences.

Multivariate GM differences

We identified 12 distinct covarying GMCs using SBM, generally corresponding to structurally or functionally connected regions. The GMCs were labeled using the significantly contributing brain regions, details on their distribution can be found in Table  S3 . Figure  4 illustrates the identified GMCs and group differences. When comparing the three groups for the identified GMCs, significant differences were found in 9 out of the 12 components. The largest differences were observed in the temporal GMC ( F  = 23.1, η p 2  = 0.14, p

FDR
 < 0.001), followed by the precuneus ( F  = 21.6, η p 2  = 0.13, p

FDR
 < 0.001), the posterior cingulate ( F  = 19.6, η p 2  = 0.12, p

FDR
 < 0.001), then the temporooccipital component ( F  = 15.6, η p 2  = 0.10, p

FDR
 < 0.001). Post‐hoc comparisons revealed that B‐MS had the most prominent decreases in the posterior cingulate ( t  = −8.5, p  < 0.001), the precuneus ( t  = −7.8, p  < 0.001), and the temporal ( t  = −7.6, p  < 0.001) components compared to W‐MS, meanwhile H‐MS showed the greatest decreases in the temporooccipital ( t  = −5.9, p  < 0.001), posterior cingulate ( t  = −3.8, p  < 0.001), and frontal ( t  = −3.6, p  = 0.001) components compared to W‐MS. Results are detailed in Table  2 .

Multivariate gray matter assessment and group differences. (A) Topographic distribution of GMCs identified using SBM. (B) Group differences in the identified GMCs. GMCs with significant group differences are denoted by an asterisk. CERBa, anterior cerebellar component; CERBp, posterior cerebellar component; Cinga, anterior cingulate component; Cingp, posterior cingulate component; DGM, deep gray matter component; Fr, frontal component; Ins, insular component; pCun, precuneus component; PHC, parahippocampal component; Temp, temporal component; TempOcc, temporooccipital component; TempPar, temporoparietal component.

Table 2 Group differences in gray matter atrophy patterns and their lesion‐driven structural disconnectivity. Differences in GM atrophy in the identified components (GMCs)
a

GMC
F‐ statistic
η p 2

p

FDR

B‐MS < W‐MS B‐MS < H‐MS H‐MS < W‐MS
t‐ statistic
p ‐value
t ‐statistic
p ‐value
t ‐statistic
p ‐value Cinga 7.3 0.05 0.002 −5 <0.001 −1.9 0.1 −3 0.01 Cingp 19.6 0.12 <0.001 −8.5 <0.001 −4.3 <0.001 −3.8 <0.001 DGM 3.8 0.03 0.03 −3.9 <0.001 −1.1 0.5 −2.9 0.01 Fr 5.2 0.03 0.009 −3.5 0.002 −0.1 1 −3.6 0.001 Ins 6.4 0.04 0.003 −4.9 <0.001 −2.5 0.03 −2.3 0.07 pCun 21.6 0.13 <0.001 −7.8 <0.001 −5.7 <0.001 −1.9 0.1 PHC 7.3 0.05 0.002 −6.5 <0.001 −3 0.008 −3 0.007 Temp 23.1 0.14 <0.001 −7.6 <0.001 −5 <0.001 −2.9 0.01 TempOcc 15.6 0.1 <0.001 −6 <0.001 −0.3 0.9 −5.9 <0.001 TempPar 1.9 0.01 0.2 – – – – – – CERBa 2.3 0.02 0.1 – – – – – – CERBp 0.8 0.01 0.4 – – – – – – Differences in structural connectivity to GMCs
b

GMC
χ
2

p

FDR

B‐MS > W‐MS B‐MS > H‐MS H‐MS > W‐MS
W‐ statistic
p‐ value
W‐ statistic
p‐ value
W‐ statistic
p‐ value Cinga 11.1 0.02 −4.2 0.008 −0.5 0.9 −3.7 0.02 Cingp 11.2 0.05 −4.3 0.007 −0.5 0.9 −3.7 0.03 DGM 10.6 0.02 −4.2 0.009 −0.6 0.9 −3.6 0.03 Fr 7.5 0.03 −3.6 0.03 −0.9 0.8 −2.9 0.1 Ins 8.2 0.02 −3.7 0.03 −0.7 0.9 −3.1 0.07 pCun 9.5 0.01 −4.1 0.01 −0.7 0.9 −3.2 0.06 PHC 9.7 0.02 −4 0.01 −0.5 0.9 −3.4 0.04 Temp 4.7 0.1 – – – – – – TempOcc 9.1 0.02 −4.1 0.01 −1.1 0.7 −3 0.09 TempPar 10 0.02 −4.2 0.008 −1 0.8 −3.2 0.06 CERBa 10.4 0.02 −4.2 0.008 −0.8 0.8 −3.4 0.04 CERBp 1.2 0.6 – – – – – – B‐MS, non‐Hispanic Black PwMS group; CERBa, anterior cerebellar component; CERBp, posterior cerebellar component; Cinga, anterior cingulate component; Cingp, posterior cingulate component; DGP, deep gray matter component; Fr, frontal component; GMC, gray matter component; H‐MS, Hispanic PwMS group; Ins, insular component; pCun, precuneus component; p

FDR
, false‐discovery rate adjusted p ‐value; PHC, parahippocampal component; Temp, temporal component; TempOcc, temporooccipital component; TempPar, temporoparietal component; W‐MS, non‐Hispanic White PwMS group; W ‐statistic, Wilcoxon rank‐sum test statistic; η p 2 , partial eta squared; χ
2 , Chi‐squared statistic.
a
Calculated using ANCOVA controlling for age, gender and TIV. Post‐hoc comparisons were conducted using Games‐Howell test.
b
Calculated using Kruskal–Wallis test. Post‐hoc comparisons were conducted using Dwass–Steel–Critchlow–Fligner (DSCF) test.

Group differences in gray matter atrophy patterns and their lesion‐driven structural disconnectivity.

B‐MS, non‐Hispanic Black PwMS group; CERBa, anterior cerebellar component; CERBp, posterior cerebellar component; Cinga, anterior cingulate component; Cingp, posterior cingulate component; DGP, deep gray matter component; Fr, frontal component; GMC, gray matter component; H‐MS, Hispanic PwMS group; Ins, insular component; pCun, precuneus component; p

FDR
, false‐discovery rate adjusted p ‐value; PHC, parahippocampal component; Temp, temporal component; TempOcc, temporooccipital component; TempPar, temporoparietal component; W‐MS, non‐Hispanic White PwMS group; W ‐statistic, Wilcoxon rank‐sum test statistic; η p 2 , partial eta squared; χ
2 , Chi‐squared statistic.

Calculated using ANCOVA controlling for age, gender and TIV. Post‐hoc comparisons were conducted using Games‐Howell test.

Calculated using Kruskal–Wallis test. Post‐hoc comparisons were conducted using Dwass–Steel–Critchlow–Fligner (DSCF) test.

Lesion‐driven structural disconnectivity

Figure  5 shows the pairwise differences in structural disconnectivity to regional GM. Overall, the largest differences were observed in the hippocampus, cingulate, and precuneus as well as deep gray matter connectivity, detailed results can be found in Table  S4 . Furthermore, we computed the change in connectivity to the GMCs identified using SBM shown in Table  2 . The largest differences in structural disconnectivity secondary to lesions were observed in the anterior cingulate ( χ
2  = 11.1, p

FDR
 = 0.02), posterior cingulate ( χ
2  = 11.2, p

FDR
 < 0.05), and DGM ( χ
2  = 10.6, p

FDR
 = 0.02) components.

Gray matter disconnectivity group differences. Depicting pair‐wise differences in regional gray matter structural disconnectivity secondary to WM hyperintense lesions in B‐MS versus W‐MS (A), and H‐MS versus W‐MS (B). Higher W‐values denote larger differences.

Mediation effects of structural disconnectivity on GM atrophy differences

The analyses examined how lesion load and structural connectivity possibly influenced ethnoracial differences in GM atrophy. For the combined minority groups comparison (B‐MS + H‐MS vs. W‐MS, Table  3 ), total lesion load significantly mediated differences in global GM fraction, accounting for 52.4% of the total effect (total effect η p 2 = −1.43, p  < 0.001; direct effect η p 2 = −0.68, p  < 0.05; indirect effect η p 2 = −0.75, p  < 0.001). Meanwhile, structural disconnectivity had varying impact across the GMCs, with the strongest mediation observed in DGM (Pm = 82.9%), followed by insular (Pm = 29.6%) and anterior cingulate (Pm = 24.3%), and temporal components (Pm = 9.7%).

Table 3 Mediation effects of MS‐related lesions and resulting structural disconnectivity on differences in GM atrophy between W‐MS and combined B‐MS and H‐MS groups. Mediation effects of lesion load on differences in total gray matter fraction Indirect effect Direct effect
R D 2

p

FDR

Total effect
R T 2

p

FDR

Pm% GMF
−0.75 ***

(−1.1, −0.42)

−0.68 *

(−1.22, −0.17)
0.45 9 × 10 −33

−1.43 ***

(−2.02, −0.85)
0.59 4 × 10 −49
52.42 Mediation effects of structural disconnectivity on group differences in gray matter atrophy patterns GMC Indirect effect Direct effect
R D 2

p

FDR

Total effect
R T 2

p

FDR

Pm% Cing‐a
−0.08 **

(−0.14, −0.03)

−0.25 **

(−0.4, −0.09)
0.59 1 × 10 −48

−0.33 ***

(−0.49, −0.16)
0.62 6 × 10 −53
24.30 Cing‐p
−0.08 **

(−0.15, −0.02)

−0.39 ***

(−0.6, −0.2)
0.40 2 × 10 −26

−0.47 ***

(−0.68, −0.27)
0.44 5 × 10 −30
16.68 DGM
−0.27 ***

(−0.42, −0.12)

−0.06
(−0.2, 0.08)
0.34 7 × 10 −21

−0.33 **

(−0.53, −0.12)
0.70 2 × 10 −66
82.87 Fr
−0.04 *

(−0.09, −0.003)

−0.17 *

(−0.33, −0.02)
0.62 7 × 10 −53

−0.21 **

(−0.37, −0.05)
0.63 4 × 10 −54
16.88 Ins
−0.09 **

(−0.16, −0.03)

−0.22 *

(−0.39, −0.04)
0.47 6 × 10 −34

−0.31 ***

(−0.49, −0.13)
0.53 6 × 10 −40
29.61 pCun
−0.06 *

(−0.11, −0.02)

−0.21 *

(−0.38, −0.04)
0.55 2 × 10 −43

−0.27 **

(−0.45, −0.1)
0.58 6 × 10 −47
23.16 PHC
0.03
(−0.01, 0.08)

−0.33 ***

(−0.53, −0.12)
0.47 1 × 10 −33

−0.3 **

(−0.48, −0.1)
0.47 2 × 10 −33
– Temp
−0.04 *

(−0.1, −0.002)

−0.39 ***

(−0.58, −0.19)
0.46 4 × 10 −33

−0.43 ***

(−0.62, −0.24)
0.48 4 × 10 −34
9.72 TempOcc
−0.02
(−0.05, 0.01)

−0.47 ***

(−0.67, −0.28)
0.52 2 × 10 −39

−0.49 ***

(−0.67, −0.31)
0.52 3 × 10 −39
– Values represent point estimates and 95% confidence intervals. B‐MS, non‐Hispanic Black PwMS group; Cing‐a, anterior cingulate component; Cing‐p, posterior cingulate component; DGM, deep gray matter component; Fr, frontal component; GMC, gray matter component; GMF, gray matter fraction; H‐MS, Hispanic PwMS group; Ins, insular component; pCun, precuneus component; PHC, parahippocampal component; Pm%, percent mediated; Temp, temporal component; TempOcc, temporooccipital component.; W‐MS, non‐Hispanic White PwMS group. *
p  < 0.05. **
p  < 0.01. ***
p  < 0.001.

Mediation effects of MS‐related lesions and resulting structural disconnectivity on differences in GM atrophy between W‐MS and combined B‐MS and H‐MS groups.

−0.75 ***

(−1.1, −0.42)

−0.68 *

(−1.22, −0.17)

−1.43 ***

(−2.02, −0.85)

−0.08 **

(−0.14, −0.03)

−0.25 **

(−0.4, −0.09)

−0.33 ***

(−0.49, −0.16)

−0.08 **

(−0.15, −0.02)

−0.39 ***

(−0.6, −0.2)

−0.47 ***

(−0.68, −0.27)

−0.27 ***

(−0.42, −0.12)

−0.06

(−0.2, 0.08)

−0.33 **

(−0.53, −0.12)

−0.04 *

(−0.09, −0.003)

−0.17 *

(−0.33, −0.02)

−0.21 **

(−0.37, −0.05)

−0.09 **

(−0.16, −0.03)

−0.22 *

(−0.39, −0.04)

−0.31 ***

(−0.49, −0.13)

−0.06 *

(−0.11, −0.02)

−0.21 *

(−0.38, −0.04)

−0.27 **

(−0.45, −0.1)

0.03

(−0.01, 0.08)

−0.33 ***

(−0.53, −0.12)

−0.3 **

(−0.48, −0.1)

−0.04 *

(−0.1, −0.002)

−0.39 ***

(−0.58, −0.19)

−0.43 ***

(−0.62, −0.24)

−0.02

(−0.05, 0.01)

−0.47 ***

(−0.67, −0.28)

−0.49 ***

(−0.67, −0.31)

Values represent point estimates and 95% confidence intervals.

B‐MS, non‐Hispanic Black PwMS group; Cing‐a, anterior cingulate component; Cing‐p, posterior cingulate component; DGM, deep gray matter component; Fr, frontal component; GMC, gray matter component; GMF, gray matter fraction; H‐MS, Hispanic PwMS group; Ins, insular component; pCun, precuneus component; PHC, parahippocampal component; Pm%, percent mediated; Temp, temporal component; TempOcc, temporooccipital component.; W‐MS, non‐Hispanic White PwMS group.

p  < 0.05.

p  < 0.01.

p  < 0.001.

In the pairwise comparisons (Table  4 ), lesion load significantly mediated GMF differences in both B‐MS versus W‐MS (total effect η p 2 = −1.81, p  < 0.001; direct effect η p 2  = −0.93, p  < 0.01; indirect effect η p 2 = −0.88, p  < 0.001; Pm = 48.6%), and H‐MS vs W‐MS (total effect η p 2 = −1.09, p  < 0.01; direct effect η p 2 = −0.47, p  > 0.05; indirect effect η p 2 = −0.62, p  < 0.001; Pm = 56.9%). Both pairwise comparisons demonstrated significant structural disconnectivity effects in DGM, insular, and anterior cingulate components, while the H‐MS versus W‐MS comparison additionally showed mediation effects in posterior cingulate, precuneus, and temporal components.

Table 4 Mediation effects of MS‐related lesions and resulting structural disconnectivity on differences in GM atrophy in B‐MS and H‐MS compared to W‐MS. Mediation effects of lesion load on pair‐wise differences in GMF B‐MS versus W‐MS H‐MS versus W‐MS Indirect effect Direct effect Total effect Pm% Indirect effect Direct effect Total effect Pm% GMF
−0.88 ***

(−1.32, −0.48)

−0.93 **

(−1.59, −0.29)

−1.81 ***

(−2.56, −1.07)
48.62
−0.62 ***

(−1.03, −0.26)

−0.47
(−1.11, 0.14)

−1.09 **

(−1.76, −0.42)
56.88 Mediation effects of structural disconnectivity on pair‐wise differences in GMCs GMC B‐MS versus W‐MS H‐MS versus W‐MS Indirect effect Direct effect Total effect Pm% Indirect effect Direct effect Total effect Pm% Cing‐a
−0.08 *

(−0.15, −0.02)

−0.3 **

(−0.49, −0.1)

−0.37 ***

(−0.57, −0.17)
21.62
−0.08 **

(−0.15, −0.03)

−0.21 *

(−0.39, −0.04)

−0.29 **

(−0.48, −0.11)
27.59 Cing‐p
−0.07
(−0.14, −0.004)

−0.68 ***

(−0.93, −0.45)

−0.75 ***

(−0.99, −0.51)
–
−0.1 **

(−0.18, −0.03)

−0.17
(−0.38, 0.05)

−0.26 *

(−0.47, −0.04)
38.46 DGM
−0.26 **

(−0.43, −0.08)

−0.12
(−0.29, 0.04)

−0.38 **

(−0.61, −0.14)
68.42
−0.28 **

(−0.47, −0.1)

0
(−0.17, 0.15)

−0.29 *

(−0.53, −0.06)
96.55 Fr
−0.04
(−0.08, −0.001)

−0.04
(−0.22, 0.14)

−0.08
(−0.26, 0.11)
–
−0.04
(−0.09, −0.001)

−0.28 **

(−0.46, −0.11)

−0.31 ***

(−0.49, −0.13)
– Ins
−0.08 *

(−0.16, −0.01) *


−0.33 **

(−0.53, −0.12)

−0.41 ***

(−0.63, −0.19)
19.51
−0.1 **

(−0.18, −0.02)

−0.13
(−0.33, 0.07)

−0.23 *

(−0.44, −0.02)
43.48 pCun
−0.06
(−0.12, −0.004)

−0.52 ***

(−0.73, −0.32)

−0.58 ***

(−0.79, −0.37)
–
−0.07 *

(−0.13, −0.02)

0.04
(−0.14, 0.22)

−0.03
(−0.21, 0.15)
– PHC
0.02
(−0.01, 0.06)

−0.52 ***

(−0.73, −0.28)

−0.5 ***

(−0.71, −0.27)
–
0.03
(−0.01, 0.09)

−0.18
(−0.39, 0.05)

−0.15
(−0.36, 0.07)
– Temp
−0.03
(−0.09, 0.003)

−0.69 ***

(−0.92, −0.45)

−0.72 ***

(−0.94, −0.49)
–
−0.06 *

(−0.14, −0.004)

−0.15
(−0.37, 0.04)

−0.21 *

(−0.42, −0.001)
28.57 TempOcc
−0.02
(−0.05, 0.01)

−0.39 ***

(−0.62, −0.17)

−0.41 ***

(−0.62, −0.19)
–
−0.02
(−0.06, 0.01)

−0.54 ***

(−0.75, −0.33)

−0.56 ***

(−0.76, −0.36)
– Values represent point estimates and 95% confidence intervals. B‐MS, non‐Hispanic Black PwMS group; Cing‐a, anterior cingulate component; Cing‐p, posterior cingulate component; DGM, deep gray matter component; Fr, frontal component; GMC, gray matter component; GMF, gray matter fraction; H‐MS, Hispanic PwMS group; Ins, insular component; pCun, precuneus component; PHC, parahippocampal component; Pm%, percent mediated; Temp, temporal component; TempOcc, temporooccipital component.; W‐MS, non‐Hispanic White PwMS group. *
p  < 0.05. **
p  < 0.01. ***
p  < 0.001.

Mediation effects of MS‐related lesions and resulting structural disconnectivity on differences in GM atrophy in B‐MS and H‐MS compared to W‐MS.

−0.88 ***

(−1.32, −0.48)

−0.93 **

(−1.59, −0.29)

−1.81 ***

(−2.56, −1.07)

−0.62 ***

(−1.03, −0.26)

−0.47

(−1.11, 0.14)

−1.09 **

(−1.76, −0.42)

−0.08 *

(−0.15, −0.02)

−0.3 **

(−0.49, −0.1)

−0.37 ***

(−0.57, −0.17)

−0.08 **

(−0.15, −0.03)

−0.21 *

(−0.39, −0.04)

−0.29 **

(−0.48, −0.11)

−0.07

(−0.14, −0.004)

−0.68 ***

(−0.93, −0.45)

−0.75 ***

(−0.99, −0.51)

−0.1 **

(−0.18, −0.03)

−0.17

(−0.38, 0.05)

−0.26 *

(−0.47, −0.04)

−0.26 **

(−0.43, −0.08)

−0.12

(−0.29, 0.04)

−0.38 **

(−0.61, −0.14)

−0.28 **

(−0.47, −0.1)

0

(−0.17, 0.15)

−0.29 *

(−0.53, −0.06)

−0.04

(−0.08, −0.001)

−0.04

(−0.22, 0.14)

−0.08

(−0.26, 0.11)

−0.04

(−0.09, −0.001)

−0.28 **

(−0.46, −0.11)

−0.31 ***

(−0.49, −0.13)

−0.08 *

(−0.16, −0.01) *

−0.33 **

(−0.53, −0.12)

−0.41 ***

(−0.63, −0.19)

−0.1 **

(−0.18, −0.02)

−0.13

(−0.33, 0.07)

−0.23 *

(−0.44, −0.02)

−0.06

(−0.12, −0.004)

−0.52 ***

(−0.73, −0.32)

−0.58 ***

(−0.79, −0.37)

−0.07 *

(−0.13, −0.02)

0.04

(−0.14, 0.22)

−0.03

(−0.21, 0.15)

0.02

(−0.01, 0.06)

−0.52 ***

(−0.73, −0.28)

−0.5 ***

(−0.71, −0.27)

0.03

(−0.01, 0.09)

−0.18

(−0.39, 0.05)

−0.15

(−0.36, 0.07)

−0.03

(−0.09, 0.003)

−0.69 ***

(−0.92, −0.45)

−0.72 ***

(−0.94, −0.49)

−0.06 *

(−0.14, −0.004)

−0.15

(−0.37, 0.04)

−0.21 *

(−0.42, −0.001)

−0.02

(−0.05, 0.01)

−0.39 ***

(−0.62, −0.17)

−0.41 ***

(−0.62, −0.19)

−0.02

(−0.06, 0.01)

−0.54 ***

(−0.75, −0.33)

−0.56 ***

(−0.76, −0.36)

Values represent point estimates and 95% confidence intervals.

B‐MS, non‐Hispanic Black PwMS group; Cing‐a, anterior cingulate component; Cing‐p, posterior cingulate component; DGM, deep gray matter component; Fr, frontal component; GMC, gray matter component; GMF, gray matter fraction; H‐MS, Hispanic PwMS group; Ins, insular component; pCun, precuneus component; PHC, parahippocampal component; Pm%, percent mediated; Temp, temporal component; TempOcc, temporooccipital component.; W‐MS, non‐Hispanic White PwMS group.

p  < 0.05.

p  < 0.01.

p  < 0.001.

Discussion

This study uniquely combines various neuroimaging analysis techniques aimed at providing a comprehensive assessment of the ethnoracial disparities in MS focusing on GM atrophy. Our results underscore significant ethnoracial disparities in GM atrophy in PwMS, with some of these differences being mediated by WM lesions and consequent structural disconnectivity. This analysis sheds light on disparities in degenerative MS outcomes and the potential underlying mechanisms, underscoring the importance of considering ethnoracial factors in MS research and clinical practice.

Our results show that B‐MS and H‐MS exhibit lower GMF, and greater lesion burden compared to W‐MS. The pattern of GM atrophy topographically aligns with previous studies in MS,
40
and suggests that ethnoracial factors may influence the extent and pattern of neurodegeneration. Notably, B‐MS patients showed the most pronounced atrophy in the posterior cingulate, precuneus, and temporal regions, while H‐MS patients exhibited greater decreases in the temporooccipital, posterior cingulate, and frontal regions. These patterns suggest that specific GM regions are differentially affected across ethnoracial groups, potentially contributing to the variability in functional clinical outcomes observed in MS.
4
The largest differences in lesion‐related structural disconnectivity were observed in the hippocampus, cingulate, precuneus, and deep gray matter, regions known to be critical hubs in the brain's network and heavily impacted by MS pathology.
41
,
42
,
43
,
44
Structural disconnectivity, driven by lesion burden and distribution, emerged as a significant mediator of some of the observed GM atrophy differences, echoing the hypothesis that MS‐related lesions disrupt the connectivity of neural networks through white matter pathways, leading to secondary neurodegeneration in connected GM regions.
15
,
16
,
17
The mediation effects were particularly notable in the deep gray matter, insular and anterior cingulate GM regions.

Although the groups showed no significant differences in disability as quantified by EDSS, they exhibited significant differences in lesion load, global GM atrophy, and its patterns. This can be explained by the nature of lesion topography, which showed notable differences in periventricular lesion volume. Periventricular lesions can be particularly clinically ineloquent, possibly due to the extent of cerebral reserve and functional compensation, especially when contrasted to infratentorial lesions.
45
,
46
,
47
Moreover, EDSS can fail to capture sub‐threshold neurologic deficits, compared to other more sensitive measurement tools.
48
In context, the exhibited differences might be compensated for by functional cerebral reserve or contributing to subclinical disability not fully expressed in terms of EDSS.

The lack of substantial mediation effects in some of the identified GM components can be explained by several factors. First, we did not assess demyelinating cortical pathology, which can disproportionately affect temporal and parahippocampal regions
49
and contribute to focal GM atrophy. Additionally, we did not systematically include demographic factors from the Social Determinants of Health framework in our analyses. Social adversity has been shown to contribute to and mediate GM atrophy.
50
Future studies incorporating these factors could provide a more comprehensive understanding of the complex interplay between lesion distribution, structural disconnectivity, and GM atrophy in diverse MS populations.

The observed ethnoracial disparities in GM atrophy and structural disconnectivity have important implications for management approaches in MS. These findings might not directly translate to clinical disability measured by cross‐sectional EDSS, though they might underlie longitudinal decline in related clinical function. Given the more pronounced GM atrophy and higher lesion loads in B‐MS and H‐MS patients, clinicians might consider earlier and more aggressive treatment initiation for these groups. This could involve prioritizing high‐efficacy DMTs aimed at curbing lesion formation and neurodegeneration,
51
potentially controlling silent progression.
52
The specific patterns of GM atrophy and disconnectivity observed in different ethnoracial groups could inform tailored cognitive rehabilitation strategies.
53
,
54
B‐MS patients exhibiting atrophy in temporal, posterior cingulate, and precuneus regions might benefit from interventions targeting attention and memory functions associated with these areas. Similarly, the pronounced frontal GM involvement in H‐MS patients suggests a potential need for executive function‐focused cognitive interventions.
53
,
54
,
55
Furthermore, the structural disconnectivity findings highlight the importance of preserving white matter integrity, underscoring the value of neuroprotective strategies and remyelinating therapies, which might be particularly crucial for B‐MS and H‐MS patients.
56
Regular monitoring of GM atrophy through more nuanced MRI techniques could be incorporated into routine clinical care, especially for patients from these higher risk groups, to guide treatment decisions and assess therapy effectiveness.
57
Additionally, these findings emphasize the need for increased representation of diverse ethnoracial groups in clinical trials for MS therapies,
58
ensuring that treatment efficacy and safety are adequately evaluated across different populations. In general, our results support the growing emphasis on personalized medicine approaches in MS management.
59
Lastly, our findings emphasize that health and brain structure disparities should not be solely attributed to biological or genetic differences, as sociocultural risk factors can be more actionable and modifiable.
11
Recent works have highlighted the fundamental impact of such sociocultural factors on health outcomes.
50
,
60
,
61

Further research is needed to elucidate the biological, genetic, and environmental factors contributing to the observed ethnoracial disparities in MS. Longitudinal studies examining the progression of GM atrophy and structural disconnectivity over time in diverse populations will be crucial for understanding the dynamic nature of these differences. Additionally, exploring the socioeconomic and healthcare access factors that may underlie these disparities will provide a more comprehensive understanding of the disease process, and possibly inform more equitable healthcare practices and effective management of MS.

Strengths

In this study, we strived to create a balanced cohort of PwMS, ensuring equal representation across diverse ethnoracial groups. Additionally, included participants did not exhibit significant differences in key confounding variables such as age, sex, disease duration, and other relevant clinical parameters minimizing bias and enhancing the validity of our findings. Our cohort underwent both clinical and radiological examinations at the same center, minimizing bias associated with variations in care providers, different MRI systems, and acquisition protocols. Furthermore, we employed both univariate and multivariate techniques to ensure the robustness of our findings.

Limitations

A principal limitation of this study is the absence of a normative healthy control group, which restricts our ability to contextualize the observed gray matter atrophy against a baseline of healthy brain anatomy. Nevertheless, the primary objective was to elucidate ethnoracial disparities in gray matter atrophy in B‐MS and H‐MS, using W‐MS as a referent group. Determination of ethnoracial group was limited to self‐reporting, thus it might not fully encompass biologic factors and genetic background. Social confounders including education, health care access, community, economic stability, and lifestyle were not explored in this study. This retrospective analysis may be constrained by its inability to encompass all pertinent variables and to prospectively control potential unrecognized confounders, and by potentially not accounting for relevant undocumented comorbidities. The use of cross‐sectional data limits our ability to infer causality or track the temporal progression of gray matter atrophy. Longitudinal studies are essential to clarify the dynamics and causative factors of these changes more definitively. Finally, the lack of validated standardized brain templates for minority populations might affect the interpretation of the results, albeit it has been previously shown that the choice of template does not introduce significant differences when using CAT12 for image normalization,
62
and the Human Connectome Project dataset used for the disconnectometry analysis strived to include a representative sample reflecting the ethnoracial diversity in the United States.
63

Conclusions

This study highlights significant ethnoracial disparities in GM atrophy among PwMS, with lesion‐driven structural disconnectivity serving as a significant mediator. The findings demonstrate more evident gray matter atrophy in Hispanic/Latinx and non‐Hispanic Black PwMS when compared to non‐Hispanic White PwMS. These disparities emphasize the critical need for tailored diagnostic and therapeutic approaches that consider ethnoracial factors. Moving forward, comprehensive longitudinal studies are necessary to better understand the progression and causative factors behind these disparities. exploring the influence of genetic, environmental, and socioeconomic variables will be vital in developing more effective, personalized MS management strategies.

Conflict of Interest

J.A.L. is supported by Bartels Family Endowment, Iglesias Family Foundation, and Brass Family Foundation. J.S.W. holds royalties for monoclonal antibodies out‐licensed through UTHealth Houston to Millipore. J.S.W has received compensation for consulting, scientific advisory boards, or other activities with the Cleveland Clinic Foundation, EMD Serono, Inmagene, Novartis, Roche/Genentech, Sandoz, and Zenas. J.S.W. has received compensation as a member of the data safety monitoring or advisory boards from the Cleveland Clinic Foundation, EMD Serono, and Novartis. C.A.P. has received postgraduate fellowship support from Genentech and has participated in advisory boards for Sanofi. C.M.O. has received postgraduate fellowship support from the National MS Society (Sylvia Lawry Grant). C.M.O has received presenting honorarium from the American Academy of Neurology. A.B., J.A.T., B.R.A., J.J., and K.M.H have nothing to disclose.

Author Contributions

A.B. and J.A.L. contributed to the conception and design of the study. A.B., J.A.T., J.J., B.R.A, C.M.O., C.A.P., and J.A.L. contributed to acquisition and analysis of data. A.B., K.M.H., J.S.W., and J.A.L. contributed to drafting a significant portion of the manuscript and preparing figures.

Supporting information

Data S1.

Acknowledgements

We acknowledge the generous contributions of the Bartels Family Endowment, Iglesias Family Foundation, and Brass Family Foundation to this work.

Data Availability Statement

Raw data was generated at UTHealth. Anonymized derived data may be shared upon reasonable request to the corresponding or senior authors with researchers who provide a methodologically sound non‐commercial proposal, subject to restrictions according to participant consent and data protection legislation.

References

Associated Data

Supplementary Materials

Data S1.

Data Availability Statement

Raw data was generated at UTHealth. Anonymized derived data may be shared upon reasonable request to the corresponding or senior authors with researchers who provide a methodologically sound non‐commercial proposal, subject to restrictions according to participant consent and data protection legislation.