Lifespan cardiometabolic exposures are associated with midlife white matter microstructure: The Bogalusa Heart Study

Abstract

Introduction:

Compartment model analysis of diffusion MRI data provides unique information on the microstructural properties of white matter. However, studies relating compartment model microstructural measures to longitudinal cardiometabolic health data are rare.

Methods:

130 cognitively healthy participants in the Bogalusa Heart Study completed diffusion MRI scans. Compartment model analysis was performed, and summary metrics were measured in organized and diffuse white matter. Multiple linear regression models were used to relate the white matter microstructure metrics to demographics and cardiometabolic risk factors.

Results:

In both organized and diffuse white matter, age was associated with worse diffusion metrics, women had better diffusion metrics than men, and African American participants had worse diffusion metrics compared to White participants. Greater blood pressure in pre-adulthood was associated with worse diffusion metrics in midlife.

Discussion:

Summary metrics from compartment model analyses of diffusion MRI data were associated cardiometabolic risk factors from youth to midlife as well as demographic factors.

1. Background

The structural integrity of white matter tracts is critical to interregional brain connectivity, and the degeneration of this connectivity is believed to contribute to phenotypes in cognitive aging and aging-related neurodegenerative diseases (Bennett and Madden, 2014). Diffusion weighted magnetic resonance imaging (dMRI) based proxy measures of white matter integrity have been associated with differences in cognitive aging in older adults (O’Sullivan et al., 2001), as well as white matter differences in populations with Alzheimer’s disease and other neurological conditions (Damoiseaux et al., 2009). Midlife cardiometabolic risk factors have been associated with poorer late-life white matter integrity (Burgmans et al., 2010; Kullmann et al., 2015), leading to the hypothesis that the well documented effects of midlife cardiometabolic risks on late life cognition (Qiu and Fratiglioni, 2015) may be mediated in part by cardiometabolic effects on white matter integrity.

However, to our knowledge almost no work to date has assessed cardiometabolic risk factors experienced in pre adulthood and early adulthood as determinants of white matter integrity later in life. The lack of knowledge about relationships between early-life cardiometabolic factors and midlife or late life white matter integrity is important because the risk factors develop insidiously starting in youth (Flynn et al., 2017; Stierman et al., 2021) and their prevalence is both high and rising over time (Anderson et al., 2019; Din-Dzietham et al., 2007). Understanding how early life risk factors associate with white matter integrity later in life could help to clarify the mechanisms by which cardiometabolic factors impact cognitive aging, as well as the benefits of early-life cardiometabolic intervention.

The few studies to date that have related early life factors to midlife or late life dMRI metrics have lacked modeling of white matter cellular composition. Instead, prior studies have utilized white matter summary metrics derived from mathematical models of local water diffusion directions, especially the diffusion tensor model. Limitations of the diffusion tensor model are well studied, including its inability to resolve multiple white matter fiber directions within a single location, and the lack of direct connection between diffusion tensor summary metrics and biological properties such as myelination, fiber density, and fiber dispersion (Jones et al., 2013). Compartment models (Alexander et al., 2019) such as Neurite Orientation Dispersion and Density Imaging (NODDI) (Zhang et al., 2012), white matter compartmental model, is an alternative to the diffusion tensor that has been validated against histological measures of optical myelin staining intensity and stereological estimates of axonal volume fraction (Jespersen et al., 2010). In healthy populations, NODDI-based metrics of neurite density, fiber orientation dispersion (a measure of the directional distribution of neurons), and free water fraction have been sensitive to aging, sex differences (Beck et al., 2021; Lawrence et al., 2021), and cardiometabolic risks in midlife and late life (Lockhart et al., 2021; Coffin et al., 2022). However, NODDI studies to date have been cross-sectional, leaving relationships between early-life risk factors and midlife or late life white matter compartmental properties unexplored.

The primary aim of this exploratory study assessed relationships between lifespan cardiometabolic risk factors measured across three epochs—pre-adulthood (less than 20 years old), early-adulthood (21 to 40 years old), and midlife (41 to 60 years old)— and midlife dMRI-based white matter tissue composition metrics in the Bogalusa Heart Study, a cohort that was middle-aged at the time of neuroimaging and has provided cardiometabolic measurements since 1973. A secondary aim was to assess differentiation in diffusion MRI metrics of white matter health between White and African American participants given well known disparities in cardiovascular exposures between White and African American individuals (Carnethon et al., 2017). We used a neural network method to synthesize high b-value dMRI data from low b-value data, thus enabling estimation of white matter composition.

2. Methods

2.1. Study participants

The Bogalusa Heart Study began in 1973 as a longitudinal prospective cohort study of atherosclerosis and cardiovascular risk factors in a White and African American rural community in southeastern Louisiana (Berenson, 2001). Race disparities in lifespan cardiovascular health outcomes have been a major theme of BHS since its inception (Berenson et al., 1992; Srinivasan et al., 1991). Details on the participant selection and methodology of BHS can be found in existing published work (CROFT et al., 1984). This study utilized a subset of participants from the Bogalusa Heart Study who had completed brain scans between 2020 and 2023. To date, 140 participants have completed brain MRI scans at Pennington Biomedical Research Center, FDG and Amyloid PET scans at Mary Bird Perkins Cancer Center, and cardiometabolic assessments at the Bogalusa Heart Study clinic (Table 1). Of the 140 participants, 10 were excluded from this work due to incomplete dMRI acquisitions or artifacts that prevented analysis. Study participants provided written informed consent. The study was overseen by the Institutional Review Boards of Pennington Biomedical Research Center and Tulane University. All Bogalusa Heart Study data may be made available following an approval process through the Bogalusa Heart Study Steering Committee.

Table 1.

Participant characteristics. Some measures were obtained multiple times throughout the lifespan. Cells containing NA indicates no measurements during a given life epoch. For smoking and drinking, overall column indicates the total percentage with some history.

	Overall	Pre-adulthood (<20)	Early-adulthood (21–40)	Mid-life (40–60)
N	130	130	127	65
Sex (% male)	40	40	41	43
Race (% African American)	21	21	20	17
Age at MRI (years)	55.41 ± 4.54	–	–	–
Education (% high school, % vocational, % college, % post graduate)	20, 22, 36, 11	–	–	20, 22, 36, 11
Smoking (% smoker)	48	30	25	14
Drinking (% drinker)	27	1	12	6
BMI (kg/m3)	25.9 ± 6.8	20.1 ± 3.9	26.8 ± 5.4	30.8 ± 6.1
SBP (mmHg)	109 ± 11	103.7 ± 10.2	110.6 ± 8.8	115.7 ± 12.9
DBP (mmHg)	71 ± 9	65.0 ± 6.6	72.6 ± 5.9	79.5 ± 8.6
Hemoglobin A1c (%)	5.7 ± 0.9	–	–	5.7 ± 0.9
Fasting glucose level (mg/dL)	85.5 ± 16.2	84.2 ± 5.6	82.3 ± 10.6	95.9 ± 22.9
HOMA-IR	2.7 ± 1.8	2.6 ± 1.4	2.4 ± 1.6	3.4 ± 2.5
Fasting insulin level (mIU/mL)	12.2 ± 6.9	12.6 ± 6.7	11.3 ± 5.9	13.4 ± 8.6
APOE-ε4 (% ε4 carriers)	28	–	–	–
Gray Matter Volume (% of TCV)	42.8 ± 3.0	–	–	42.8 ± 3.0
White Matter Volume (% of TCV)	34.1 ± 3.7	–	–	34.1 ± 3.7
WMH Volume (% of TCV)	0.04 ± 0.05	–	–	0.04 ± 0.05

2.2. Clinical evaluation

Validated questionaries were used to obtain demographic and lifestyle information including age, race, and sex. Race was self-reported by participants with no categories pre-defined by the investigators. Details of clinical evaluation methods can be found in the Supplementary Materials. Measurements of BMI, SBP, DBP, fasting glucose, fasting insulin, and HOMA-IR were collected at multiple ages from childhood to midlife and averaged within three epochs of life: pre adulthood (less than 20 years old), early adulthood (21–40 years old), and mid-life (more than 40 years old).

2.3. Structural MRI acquisition and processing

MRI scans were performed on a GE Discovery MR750w 3.0T scanner at Pennington Biomedical Research Center. T1-weighted structural images were acquired with a sagittal MPRAGE sequence as well as T2-FLAIR images. These images were analyzed using in-house software described elsewhere (Yoshita et al., 2006; Decarli et al., 2008; Carmichael et al., 2012; Carmichael et al., 2019). The main processing steps for FLAIR images include manual segmentation of the total cranial volume by tracing the dura mater within the cranial vault; non-uniformity correction of the segmented image (DeCarli et al., 1996); thresholding the total cranial volume into brain matter and cerebrospinal fluid (CSF) voxels (DeCarli et al., 1992); fitting the intensity distribution of the brain tissue to a Gaussian distribution and labeling voxels with an intensity greater than 3.5 standard deviations from the mean as white matter hyperintensities (WMH) (DeCarli et al., 2005). The main processing steps for structural images include non-uniformity correction and segmentation of gray matter (GM), white matter (WM), and CSF using a Bayesian maximum-likelihood expectation-maximization algorithm (Fletcher et al., 2012a, b). Volume measurements of GM, WM, and WMH expressed as percentages of the total cranial volume were used for further statistical analysis.

2.4. PET acquisition and processing

¹⁸F-FDG and ¹⁸F-florbetapir PET scans were performed on GE Discovery 610 PET/CT and Discovery 600 PET/CT systems at Mary Bird Perkins Cancer Center. The acquisition and reconstruction parameters specific to the scanner were set according to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) protocol for the aforementioned scanner models (Jagust et al., Jul. 2015). Briefly, attenuation corrected PET images were acquired, and the preprocessing of the images, including co-registration, averaging, resampling to standard space with 1.5 mm³ voxel size, and spatial smoothing, followed the ADNI PET procedure to produce uniform results from both scanners (Jagust et al., Jul. 2015).

Fully preprocessed FDG images were spatially normalized to an MNI standard stereotactic brain atlas. For FDG PET semi quantification the Berkeley FDG MetaROI was used (Caroli et al., 2012). For florbetapir PET semi-quantification Avid florbetapir MetaROI was used (Joshi et al., 2015). The standard uptake value ratios (SUVRs) were calculated for the individual ROIs as a ratio of mean uptake in the ROI with respect to the mean uptake in the reference region. The global SUVR was calculated as the average of the ROI SUVRs.

2.5. Diffusion MRI acquisition and processing

Diffusion weighted images were acquired on the same scanner as structural MRI images. Four b-0 vol were acquired with 40 diffusion weighted volumes at b = 1000 s/mm². Preprocessing steps included skull stripping, susceptibility distortion correction, eddy current distortion and motion correction, and diffusion tensor fitting. The SynB0-DISCO tool was used to estimate the undistorted b-0 vol using the structural T1-weighted scan as an undistorted reference image (Schilling et al., 2019). Geometric distortion correction was performed using the acquired b-0 vol and the estimated distortion-free volume to estimate and correct the off-resonance field using the topup tool from FSL (Andersson et al., 2003; Smith et al., 2004). Eddy current-induced distortions and subject motion between volumes were corrected using the eddy tool from FSL (Andersson and Sotiropoulos, 2016). A neural network approach was then used to generate diffusion data at b = 2000 and b = 3000 s/mm² from the acquired b = 1000 s/mm² data (Dugan and Carmichael, 2023).

The resulting multi-shell diffusion data was then fit to a NODDI-Watson model using the Diffusion Microstructure Imaging in Python toolbox (Fick et al., 2019). The fitted model was used to generate parameter maps of neurite density index (NDI), orientation dispersion index (ODI), and free water fraction (FWF). Parameter maps for FA, MD, NDI, ODI, and FWF were nonlinearly registered to 2mm³ MNI space using structural T1-weighted scans as an intermediary via the FSL linear and nonlinear registration tools FLIRT and FNIRT (Jenkinson and Smith, 2001).

To assess whether associations between predictor variables and dMRI outcome metrics were primarily driven by highly organized white matter tracts as opposed to diffuse white matter, statistical analyses were performed on mean values of the dMRI metrics within two ROIs: a composite mask of all the white matter tracts in the JHU ICBM-81-DTI atlas and a global white matter mask including all voxels with white matter probability greater than 50% but excluding the voxels of the composite JHU atlas mask. The JHU ICBM-81-DTI white matter labels atlas is a hand-segmented atlas of white matter labels from a standard space average of 81 scans provided by the ICBM DTI group (Mori et al., 2005). Additional details on all imaging acquisition and processing can be found in the Supplementary Materials.

2.6. Statistical analysis

Statistical analysis began with tests of the validity of our dMRI measurements. To assess the validity of our parcellation of white matter into organized and diffuse regions, we checked for statistical differences between the dMRI metrics in the two ROIs with a t-test. To confirm the relevance of our dMRI metrics as measures of brain health, we assessed the degree to which they associated with well-established markers of structural brain health, including gray matter volume and amyloid PET burden, both of which have been shown to associate with cognitive functioning and neuropathology burden in older adults (Fjell and Walhovd, 2010; Rice and Bisdas, 2017). This was done with models that used one of the following imaging variables as a predictor for one of the dMRI metrics: FDG PET SUVR, Amyloid PET SUVR, GM volume, WM volume, and WMH volume. In these models, age, sex, and race were included as covariates. These results are reported in the Supplementary Materials.

After these verifications, we assessed demographic and lifespan cardiometabolic measurements as predictors of midlife dMRI summary metrics. Separate multivariable linear regression models were run, each with a separate dMRI metric (FA, MD, NDI, ODI, and FWF) within a specified white matter compartment (organized tracts or diffuse) as the outcome variable. In the context of white matter, high values of FA and NDI, and low values of MD, ODI, and FWF are considered to indicate healthier white matter tissue.

An initial set of models specifically tested for associations between demographic variables and mid-life dMRI metrics. These models included age, sex, and race as simultaneous predictors for one of the dMRI metrics. A second set of models added one of the following cardiometabolic predictors, measured in one of the three lifespan epochs: SBP, DBP, BMI, fasting glucose level, fasting insulin level, HOMA-IR. Hemoglobin A1c was measured once per participant in middle age (minimum/mean/maximum ages of 33.6/48.1/58 years). These models tested for associations between the cardiometabolic variables and mid-life dMRI metrics while controlling for age, sex, and race. For all models, predictor variables with a significant p-value at the p < 0.05 level are reported in the results section.

This analysis is exploratory, and steps were not taken to control for type 1 error. For each model, normality of residuals was confirmed with a Kolmogorov–Smirnov test to ensure validity of the linear regression model. Outliers in the dataset were defined as values above the 75th or below the 25th percentile by a factor of 3 times the interquartile range. They were identified and removed before statistical analysis. Outlier details can be found in the Supplementary Materials.

3. Results

3.1. Associations between white matter and demographic measures

Organized white matter showed significantly greater FA, MD, NDI, and FWF, and significantly lower ODI, than the diffuse white matter (based on t-test with p-values < 0.05) (Table 2). In organized white matter, greater age was associated with greater FWF (p = 7.8e-06 β = 0.000950) and lesser NDI (p = 0.019 β = −0.000960), women had lower FWF than men (p = 0.002 β = −0.00592), African American participants had greater ODI than White participants (p = 0.001 β = 0.00615), and number of years of education was not associated with any of NDI, ODI, or FWF (Table 3). Education level was not associated with any diffusion metrics in the organized white matter region. Associations between MD and age mirrored those of FWF, race differences in FA mirrored those of ODI, there were no sex differences in either FA or MD, and African American participants had greater MD than White participants (Supplementary Materials Table 1).

Table 2.

Mean and standard deviation of diffusion metrics for all participants. Mean values in organized and diffuse white matter regions were statistically different for each measure (based on t-test with p-value < 0.05).

	NDI	ODI	FWF	FA	MD
Mean in organized white matter region	0.393±0.0213	0.220±0.00860	0.0688±0.0117	0.513±0.0257	0.869±0.0447
Mean in diffuse white matter region	0.360±0.0176	0.422±0.0197	0.0388±0.0009	0.309±0.0222	0.826±0.0375

Table 3.

Statistically significant linear relationships between age, sex, race, and NODDI diffusion measures in organized white matter (OWM) and diffuse white matter (DWM). The 95% confidence interval of the model coefficients are reported in square brackets.

	Neurite Density Index	Orientation Dispersion Index	Free Water Fraction
Age at MRI	OWM (p = 0.0188 β = −0.000960 [−0.001776, −0.000163])		OWM (p = 7.8e-06 β = 0.000950 [0.000548, 0.001355]) DWM (p = 0.00136 β = 0.000550 [0.000219, 0.000889])
Sex			OWM (p = 0.00240 β = −0.00592 [−0.00971, −0.00214]) DWM (p = 0.0231 β = −0.00363 [−0.00683, −0.00051])
Race	DWM (p = 0.00353 β = −0.0112 [−0.0187, −0.0038])	OWM (p = 0.00109 β = 0.00615 [0.00251, 0.00980]) DWM (p = 0.0102 β = 0.0112 [0.0027 0.0197])

Within diffuse white matter, greater age was associated with greater FWF (p = 0.001 β = 0.000550), women had lesser FWF than men (p = 0.023 β = −0.00363), and African American participants had lesser NDI (p = 0.004 β = −0.0112) and greater ODI (p = 0.010 β = 0.0112) than White participants (Table 3). Race differences in FA mirrored those of NDI, and race differences in MD mirrored those of FWF (Supplementary Materials Table 3).

3.2. Association between white matter and cardiometabolic measures

In organized white matter, worse cardiometabolic measures were generally associated with lesser neurite density. Greater pre adulthood DBP (p = 0.031 β = −0.000651), early adulthood SBP (p = 0.019 β = −0.000539) and DBP (p = 0.041 β = −0.000674), and mid-life SBP (p = 0.007 β = −0.000623) and DBP (p = 0.031 β = −0.000721) were all associated with lesser NDI (Table 4). Greater mid-life fasting glucose level was associated with greater FWF (p = 0.019 β = 0.000194). Greater mid-life HOMA IR was associated with lesser NDI (p = 0.038 β = −0.00240). Greater hemoglobin A1c was associated with greater ODI (p = 0.007 β = 0.00651). Associations between cardiometabolic factors and FA mirrored those of NDI for all associations except early adulthood DBP where there was no significant relationship. Associations between cardiometabolic measures and MD mirrored those of FWF except for mid-life fasting glucose. (Supplementary Materials Table 2) Greater pre adulthood SBP and DBP, early adulthood SBP, mid-life SBP, and mid-life HOMA IR were associated with greater MD while there was no significant relationship with FWF and these measures.

Table 4.

Statistically significant linear relationships between demographic, cardiometabolic, and cognitive measures and NODDI diffusion measures in organized white matter (OWM) and diffuse white matter (DWM). The 95% confidence interval of the model coefficients are reported in square brackets.

	Neurite Density Index	Orientation Dispersion Index	Free Water Fraction
SBP Pre-Adulthood
DBP Pre-Adulthood	OWM (p = 0.0308 β = −0.000651 [−0.001242, −0.000061])
SBP Early-Adulthood	OWM (p = 0.0187 β = −0.000539 [−0.000987, −0.000091])
DBP Early-Adulthood	OWM (p = 0.0405 β = −0.000674 [−0.001318, −0.000029])
SBP Mid-Life	OWM (p = 0.00710 β = −0.000623 [−0.001070, −0.000175])
DBP Mid-Life	OWM (p = 0.0309 β = −0.000721 [−0.001374, −0.000068])
Fasting Glucose Level Early Adulthood			DWM (p = 0.0134 β = 0.000341 [0.000072, 0.000611])
Fasting Glucose Level Mid-Life			OWM (p = 0.0192 β = 0.000194 [0.000032, 0.000357])
HOMA IR Mid-Life	OWM (p = 0.0384 β = −0.00240 [−0.00467, 0.00013]) DWM (p = 0.0254 β = −0.00211 [−0.00394, 0.00026])		DWM (p = 0.0246 β = 0.000920 [0.000122, 0.001718])
Hemoglobin A1c		OWM (p = 0.00690 β = 0.00651 [0.00181, 0.01121])

In diffuse white matter, worse cardiometabolic measures and brain health were generally associated with greater FWF (Table 4). Greater early adulthood fasting glucose level was associated with greater FWF (p = 0.013 β = 0.000341). Greater mid-life HOMA IR was associated with greater FWF (p = 0.025 β = 0.000920) and lesser NDI (p = 0.025 β = −0.00211). In the corresponding DTI results (Supplementary Materials Table 4) associations between cardiometabolic measures and MD mirrored those of FWF except for early adulthood fasting glucose. Greater mid-life SBP was associated with greater MD while there was no significant relationship with FWF. There were no significant associations for FA. These results are also reported in the form of a heat map in Figure S1 of the Supplementary Materials.

4. Discussion

We believe this may be the first report of associations between blood pressure throughout the lifespan from pre-adulthood to midlife, and dMRI based metrics of white matter tissue composition in midlife. Greater blood pressure in every epoch (pre adulthood, early adulthood, midlife) was associated with poorer metrics of neurite density within highly organized white matter tracts in midlife; pre adulthood blood pressure continued to be a significant predictor even after controlling for midlife blood pressure, suggesting possible legacy effects of pre adulthood exposures. In addition, the same white matter tissue characteristics were associated with demographic factors, contemporaneous metabolic risks, and complementary measures of brain health. These findings reinforce the importance of cardiometabolic risk prevention in childhood, due to the possibility of long-term adverse neurological consequences associated with exposures in youth. The findings also emphasize the urgency of understanding mechanisms underlying legacy effects, with a goal of blocking any progression from childhood vascular damage to later-life brain effects.

The associations between greater midlife blood pressure and poorer white matter tissue composition are consistent with previous cross-sectional investigations of more traditional dMRI metrics in midlife (De Groot et al., 2015), as well as one dMRI-based tissue composition study (Suzuki et al., 2017), although we note that unlike Suzuki et al., we did not find significant associations between blood pressure and free water fraction. Our findings also largely align with the limited literature on elevated early adulthood blood pressure and poorer midlife indicators of brain health (including white matter hyperintensities and estimated brain age) (Dintica et al., 2023; Suvila et al., 2021), although we note that one study found no midlife brain health differences between early-onset (onset at < 35 years age) and late onset hypertension (Suvila et al., 2021). We extend this body of work by suggesting that associations between blood pressure and midlife brain health extend all the way back into childhood, and that these associations may represent possible effects at different points throughout the lifespan. Given the consistent finding that midlife blood pressure is associated with dementia risk late in life (Whitmer et al., 2005) our finding raises the question of whether waiting until midlife, or even young adulthood, to prevent elevated blood pressure is adequate to prevent such an elevation of dementia risk later on.

Race is a complex, multi-factorial social construct that is associated with biological consequences, rather than an innate biological trait that in and of itself drives differences in brain or cognitive phenotypes. Within that context, we believe we may be providing among the first findings of race differences in diffusion MRI metrics of white matter health. In both organized and diffuse white matter regions, we found that African American participants had suggestions of poorer white matter health compared to White participants. Related studies of race differences in other MRI metrics have been mixed to date, with two large cohorts identifying no such differences (Knopman et al., 2011; Lockhart et al., 2022) and another reporting poorer brain health among African Americans in alignment with our findings (Launer et al., 2015). Better characterizing race differences in intermediate phenotypes such as brain MRI metrics could play an important role in clarifying how stark race disparities in cardiovascular exposures lead to stark race disparities in dementia risk (Gibbons, 2004).

Our findings align with existing literature on age-related white matter changes and contribute to a mixed literature on sex differences. In both organized and diffuse white matter, we found that older age was associated with greater free water fraction, while in highly organized white matter tracts, older age was associated with lesser neurite density, suggesting that in midlife, advancing age may be associated with poorer white matter health. These results are broadly consistent with the general pattern of findings from cross-sectional studies that included a broader age range and reported either more traditional dMRI metrics (Bennett and Madden, 2014) or tissue composition characteristics (Beck et al., 2021; Lawrence et al., 2021). We found that free water fraction was lower in women compared to men in both organized and diffuse white matter regions, suggestive of poorer white matter health among middle aged men. Some prior studies have reported similar findings of better dMRI-based metrics of white matter health among women (Salminen et al., 2020) while others have shown precisely the opposite (Lawrence et al., 2021). The age and sex findings reinforce the need to better understand biological mechanisms underlying possible age-related and sex-related white matter differences, with possible implications for age- and sex-related differences in risks for late-life cognitive impairment.

We analyzed dMRI-based tissue compartmental characteristics because they are more specific to anatomical properties than more traditional diffusion tensor summary metrics are. For example, fractional anisotropy, the most commonly reported diffusion tensor metric, is influenced by both neurite density and fiber orientation dispersion, both of which are reported separately in our analysis, and both of which exhibited differential associations with cardiometabolic risk factors (Zhang et al., 2012). In addition, the tissue compartment parameters identified a larger number of associations with the predictors (13) than the diffusion tensor summary metrics did (9), suggesting a greater sensitivity to risk factor effects. Our free water fraction estimates associated with risk factors similarly to diffusion-tensor-based mean diffusivity, but free water fraction was the only dMRI metric to identify sex differences, and the only dMRI metric that identified age differences in the diffuse white matter. The greater anatomical specificity and sensitivity of the tissue composition measures suggests that the large existing literature featuring diffusion tensor metrics could be enriched by adding compartmental characteristics to the analysis.

The main strength of this study is its use of an established population-based cohort with nearly 50 years of longitudinal health data. Another strength is the use of white matter tissue compartment modeling for assessment of white matter structure. A potential limitation of our study is the sample size of 130, which compares favorably to many other dMRI studies but is smaller than many recent structural MRI studies. An additional limitation is that the angular resolution of our diffusion MRI acquisition is lower than the number of diffusion gradients used in some other NODDI studies. Greater angular resolution has been shown to impact the ability of the model to accurately predict high values of orientation dispersion (Zhang et al., 2012).

This work should be considered exploratory in nature. No corrections for multiple comparisons were applied during analysis, which increases the risk of Type I errors. Therefore, these findings should be interpreted with caution. Further validation through hypothesis-driven studies with more rigorous statistical adjustments is needed confirm the robustness of the observed associations. Readers are encouraged to view these results as hypothesis-generating rather than definitive conclusions.

In a racially diverse, cognitively healthy, middle-aged community cohort sample, summary metrics from diffusion tensor and compartment model analyses of multi-shell diffusion MRI data were associated cardiometabolic risk factors from youth to midlife as well as demographic factors.

Abbreviations and Acronyms:

NODDI

neurite orientation dispersion and density imaging

NDI

neurite density index

ODI

orientation dispersion index

FWF

free water fraction

SBP

systolic blood pressure

DBP

diastolic blood pressure

HOMA-IR

homeostatic model assessment of insulin resistance

NEX

number of excitations

Data and code availability

All Bogalusa Heart Study data may be made available following an approval process through the Bogalusa Heart Study Steering Committee. Code used in this work can be made available upon request to the corresponding author.

Data availability

Data will be made available on request.