Gray Matter Covariance Networks Associated With Parental Longevity—Results From the LonGenity Study

Abstract

Background

Older adults with exceptional longevity—and their offspring—are protected from age-related diseases and have a prolonged health span. Prior research suggests that offspring of parents with exceptional longevity have larger temporal and sensorimotor cortices in mid to late-adulthood—but the association between brain health and parental longevity is not well understood in older adults. This study aimed to identify a gray matter volume pattern (or network) associated with parental longevity in older adults—and to determine if individual expressions of this gray matter pattern vary with cognitive performance.

Methods

Participants consisted of 139 older adults of Ashkenazi Jewish descent from the LonGenity study (M age 79.4 ± 6.5 years; 56.11% women). Eighty-four (60.4%) were offspring of parents with exceptional longevity and 55 (39.6%) were offspring of parents with usual survival. A gray matter network associated with parental longevity was derived with multivariate covariance-based analyses that were adjusted for potential confounders. Participant-specific expressions of this network were then regressed against cognitive test performances.

Results

The derived gray matter covariance network associated with parental longevity was primarily composed of frontal, insular, and hippocampal regions. The extent to which older adults displayed this pattern was associated with overall cognition, Free Recall, Digit Symbol Substitution, Boston Naming, and Trail Making Test A performance.

Conclusions

Parental longevity is associated with a widespread network of brain regions in aging. This brain network also varies as a function of cognitive performance.

Physical and cognitive health problems are common in aging. As the number of older adults increases, so do the societal and economic consequences of aging (1,2). Previous magnetic resonance imaging (MRI) research has demonstrated that aging, even in the absence of cognitive impairment and dementia, is associated with global and region-specific brain atrophy—including prefrontal, hippocampal, thalamic, caudate, cingulate, and parietal atrophy (3–6). What makes individuals more or less resilient to age-related brain atrophy, however, is not well-understood.

Individuals with exceptional longevity prolong their independence and delay the onset of many age-related diseases—including cancer, hypertension, diabetes, and heart disease (7–10). Their offspring go on to live long and healthy lives as well (9,11–13) and can be used as a model for studying longevity and brain health (7,14). Parental longevity is associated with a slower rate of cognitive decline (11,15). Yet, common risk factors for cognitive decline such as smoking, physical inactivity, high body mass index, and poor diet are not different in those with parental longevity compared to those without parental longevity (16–18). These studies suggest that genetics and gene–environment interactions play a significant role in exceptional longevity.

The effect of parental longevity on brain health is not commonly studied—particularly in older adults. One study of adults aged 40–70 years (M age = 64.4 years) suggests that offspring of parents with longevity (at least 1 parent lived ≥85 years) have relatively preserved hippocampal, parahippocampal, and middle temporal brain volumes compared to those without parental longevity (19). This study also found that parental longevity was associated with lower mean diffusivity (less water content/atrophy) in the thalamic radiation and the uncinate fasciculus. Another study of adults aged 55 years and above (M age = 66 years) suggests that parental longevity (at least 1 parent lived ≥85 years) is associated with less white matter hyperintensities (a neuroimaging marker of cerebral white matter or small vessel disease), less increase in temporal horn volume (less water content/atrophy), and a slower rate of cognitive decline (20). These initial studies suggest that parental longevity may be associated with better brain and cognitive health, but additional studies of older adults (when atrophy is more pronounced) using different definitions of exceptional longevity (eg, ≥85; ≥95, centenarian) are needed to determine the reliability of these findings.

The current study of older adults above 65 years of age employed multivariate covariance-based statistics to derive structural brain networks (or patterns) that differ as a function of parental longevity—defined as having at least 1 parent that lived ≥95 years. Multivariate covariance-based statistics were used to address this issue because they are particularly resistant to between-subject variability and collinearity (21–23) and eliminate the need to correct for multiple comparisons. Currently, it is unclear if the protective effects of parental longevity extend into more advanced old age when brain atrophy is accelerated, and if the effects are comparable across different definitions of parental longevity and different populations (5). Prior research has focused on regional brain volumes and functional resiliency but not whole brain data in people with parental longevity. We further sought to determine if parental longevity brain networks also vary with cognitive performance. We examined this issue in a genetically homogenous sample of adults above 65 years of age (all 4 grandparents were of Ashkenazi Jewish descent) with educational and socioeconomic homogeneity—further increasing our power for discovery. We hypothesized that we could derive a gray matter network associated with parental longevity after adjusting for potential confounders. We further hypothesized that the extent to which older adults displayed this network would be associated with cognitive performance—particularly on measures of episodic memory that rely on the hippocampus and neighboring temporal brain regions and attention that rely on frontal, parietal, and thalamic regions. Altogether, this would demonstrate that brain networks associated with parental longevity overlap with those that experience age-related atrophy and presumably disrupt cognitive performance. The identification of preserved and nonpreserved brain regions associated with parental longevity could be important for developing interventions for maintaining or restoring these regions while promoting resiliency to neurocognitive aging.

Method

Study Participants

A subset of LonGenity study participants that underwent MRI before August 4, 2023, were included in this analysis. LonGenity study participants are older (≥65 years) adults of Ashkenazi Jewish descent, free of dementia at baseline. Approximately half are offspring of those enriched for exceptional longevity (OPEL; at least 1 parent lived 95 years or more) while the other half are offspring of those of usual survival (OPUS; both parents lived less than 95 years). A detailed description of the study as well as the inclusion and exclusion criteria can be found in previous literature (24). The exclusion criteria for MRI acquisition included: history of neurodegenerative diseases at baseline, head trauma, stroke, and contraindication to MRI (pacemaker or any permanent magnetic metal implants). An MRI sub-sample of 141 participants was initially included in this analysis. Two participants with corrupted T1-weighted images files were excluded prior to analysis; however, leaving a total of 139 participants for final analysis. All study participants provided written informed consent and Einstein’s institutional review board approved the study protocol.

Magnetic Resonance Imaging Acquisition

Image acquisition was performed at the Gruss Magnetic Resonance Research Center in Bronx (NY), using a Philips 3T Elition multinuclear MRI/MRS system. The T1-weighted images included in this analysis were acquired using axial 3D-MP-RAGE parameters over a 240 mm FOV; 1.0 mm isotropic resolution, echo time = 4.6 ms, repetition time = 9.9 ms, α = 8°, and SENSE factor 2.5.

Magnetic Resonance Imaging Preprocessing

T1-weighted images were preprocessed with a streamlined voxel-based morphometry pipeline using SPM12 (Wellcome Centre for Human Neuroimaging)—which was implemented in MATLAB R2020b (Mathworks, Natick, MA). Voxel-based morphometry is a common method for identifying systematic volumetric brain differences, and voxel values reflect the volume of a specific brain region relative to neighboring regions (25). Each T1-weighted image was segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using the unified segmentation procedure (26), which combines image registration, tissue classification, and bias correction in the same generative model. This procedure includes Diffeomorphic Anatomical Registration Through Exponentiated Linear algebra (DARTEL (27))—which provides precise intersubject alignment by utilizing 3 parameters for every voxel to morph the shape of the brain while concurrently aligning both GM and WM to generate a study-specific template. The GM-, WM-, and CSF-segmented images are then aligned to this study-specific template, normalized to Montreal Neurological Institute (MNI) space, and spatially smoothed using an isotropic Gaussian kernel (full width half max = 8 mm). Only the smoothed GM images were used as input for statistical analysis.

Cognitive Assessments

A subset of 119 participants also completed a comprehensive neuropsychological test battery. Global cognition was generated by calculating a standardized composite score from 10 neuropsychological tasks based upon 7 different cognitive domains (executive function, working memory, processing speed, language, visual memory, visuospatial function, and verbal episodic memory) as described in Jayakody et al. 2022 (28,29). The individual scores examined were: (a) executive function (Trail Making Test B-A [TMT interference]) (30), (b) verbal episodic memory (Free Recall on Free and Cued Selective Reminding Test) (31), (c) visuomotor speed (Trail Making Test-A), (d) language (Boston Naming) (32), (e) working memory (Digit Span test of the Wechsler Adult Intelligence Scale-III [WAIS-III]) (33), and (f) attention (Digit Symbol Substitution test of WAIS-III). Individual neuropsychological tests were examined as a raw score but z-standardized for inclusion in the global cognition score. These assessments were completed within 4 months of the MRI, most on the same day of acquisition.

Statistical Analysis

Baseline characteristics of OPEL and OPUS participants were compared using bivariate analyses. Data were examined both graphically and with the appropriate statistical test measures for model assumptions. Visual examinations were conducted regarding normality for continuous variables; t tests were applied to variables deemed normally distributed. Mann–Whitney tests were applied to continuous variables that were not normally distributed. Chi-square tests were applied to dichotomous variables.

Multivariate covariance-based analyses (www.nitrc.org/projects/gcva_pca) were used to identify gray matter networks or patterns associated with history of parental longevity—after adjusting for age, sex, years of education, estimated total intracranial volume, and white matter hypointensities (a measure of overall WM disease burden automatically generated from FreeSurfer 7.2; http://surfer.nmr.mgh.harvard.edu/ (23,34)). The key analytical steps can be summarized as follows. First, GM images were masked using a standardized template provided by SPM12 so that only voxels with >30% probability of being GM would be considered in subsequent analysis. Each image was then z-transformed by subtracting the overall participant means from each voxel. A principal component analysis (PCA) was then performed on these z-transformed images. A PCA is a common statistical technique used to reduce complex data sets into more manageable components. Derived components (that maximally covaried together) were then regressed against parental longevity (OPEL vs OPUS) while adjusting for covariates. The best linear combination of principal components that predicted parental longevity was selected using the Akaike Information Criteria (35). A bootstrap resampling procedure (1 000 iterations, n − 1 with replacement) was then performed to determine the stability of each voxel in the generated component/s. The resulting spatial map contained a specific z-value for each voxel. Statistical significance was set at p < .05, z ± 1.96. Significant voxels are considered nodes in the brain pattern associated with parental longevity status (36). Positively weighted voxels are associated with more volume in OPEL than OPUS. Negatively weighted voxels are associated with less volume in OPEL than OPUS. Finally, linear regression was used to identify associations between participant-specific expression scores and attention, episodic memory, executive function, language, visuomotor speed, working memory, and overall cognition. Participants-specific expression scores reflect the extent to which individuals display the derived brain network.

Results

The demographic characteristics of the study sample are summarized in Table 1. A total of 139 participants were analyzed. The 84 OPEL were significantly younger (median 77.71 years) than the 55 OPUS participants (median 80.18 years) with a statistically significant difference of p = .009. About 60% of the participants in each group were female. There was no statistically significant difference in years of education between OPEL and OPUS. All cognitive measures showed no statistical difference between groups. Both groups of older adults were relatively healthy with no statistical significance difference in their global health score: a composite of common diseases observed in older adults.

Table 1.

Baseline and Clinical Characteristics of Offspring of Parents With Exceptional Longevity Versus Offspring of Parents With Usual Survival (N = 139)

	OPEL (n = 84)	OPUS (n = 55)	p Value
Demographics Age* (years)	77.7 (72.8–83.0)	80.1 (78.4–83.8)	0.009
Sex, n (%)   Female	49 (58.33)	29 (52.73)	0.52
Years of education*	18 (16.5–20)	18 (16–20)	0.22
eTIV* (mm3)	1 437 992 (1 266 470–1 541 138)	1 396 298 (1 269 035–1 525 339)	0.49
White matter hypointensities* (mm3)	2 262.85 (1 514.75–3 904.8)	2 279.6 (1 553.4–3 387.9)	0.81
Global health score*,†,‡ (range 1–8)   Arthritis, n (%)   Chronic heart failure, n (%)   COPD, n (%)   Depression, n (%)   Diabetes, n (%)   Hypertension, n (%)   Myocardial infarction, n (%)   Stroke, n (%)	1 (0–1) 32 (38.55) 0 (0) 1 (1.20) 7 (8.33) 6 (7.14) 28 (35.00) 6 (7.14) 0 (0)	1 (0–2) 10 (18.51) 1 (1.81) 0 (0) 3 (5.56) 2 (3.64) 23 (41.82) 3 (5.45) 0 (0)	0.45 0.02 0.39 1.00 0.74 0.48 0.51 1.00 N/A
Cognitive tests * , § Global Cognition Digit Symbol Substitution Trail Making Test (B-A) Free Recall Boston Naming Trail Making Test—A Digit Span	2.54 (−0.84 to 7.82) 64 (53–75) 36 (26–57) 36 (32–39) 56 (52–60) 35 (28–49) 18 (15–20)	2.11 (0.29–5.95) 62 (53–71) 36.5 (21–72) 35 (31–38) 56 (56–60) 38.5 (32–45) 17 (14–21)	0.97 0.60 0.90 0.46 0.83 0.40 0.55

Notes: eTIV = estimated total intracranial volume; N/A = not applicable; OPEL = offspring of parents with exceptional longevity (at least 1 parent reached the age of 95 or beyond); OPUS = offspring of parents with usual survival (neither parent reached the age of 95 or beyond).

^*Median (IQR) reported.

^†Global Health Score is obtained from the dichotomous ratings of the presence or absence of diabetes, chronic heart failure, arthritis, hypertension, depression, stroke, chronic obstructive pulmonary disease, angina and myocardial infarction.

^‡Total participants with global health score values, N = 132.

^§Total Participants with cognitive assessments, N = 119 except for global cognition where N = 101 and language where N = 117.

The GM covariance network—or the best linear fit of sequential principal component/s—associated with parental longevity was 1 principal component and explained 11% of the variance (R² = 0.11). Table 2 summarizes the peak structures that were positively and negatively associated with the outcome of parental longevity status after adjusting for the same set of covariates mentioned previously. The largest peak structures, seen in red, that were associated with greater volume in OPEL included rectus, insula, rolandic operculum, middle frontal gyrus, middle temporal gyrus, and inferior frontal gyrus, triangular (see Figure 1). The largest peak structures, seen in blue, that were associated with lower volume in OPEL included cerebellar, supplementary motor, and precuneus regions (see Figure 1).

Table 2.

Gray Matter Covariance Network Associated With Parental Longevity Status After Adjusting for Age, Sex, Years of Education, Estimated Total Intracranial Volume, and White Matter Hypointensities

Peak structure	X	Y	Z	K
Negative (less volume in OPEL)
Supplementary motor area, R	1.7	16.1	67.6	13 259
Precuneus, R	1.7	−71.7	58.0	2 637
Postcentral gyrus, R	29.7	−46.6	71.3	444
Cerebellum (VIII), R	35.6	−37.8	−41.5	54 444
Middle frontal gyrus, R	29.7	19.8	60.2	561
Middle frontal gyrus, L	−38.9	52.2	23.4	249
Precentral gyrus, R	23.8	−23.7	75.0	581
Inferior parietal gyrus, L	−50.7	−42.2	55.1	324
Inferior frontal gyrus, opercular, L	−53.6	22.0	33.0	222
Middle occipital lobe, L	−40.4	−83.5	34.4	495
Postcentral gyrus, R	44.4	−38.5	62.5	332
Middle frontal gyrus, orbital, L	−46.3	53.7	−3.2	162
Precentral gyrus, L	−35.9	2.1	63.9	513
Middle frontal gyrus, R	50.3	43.4	20.4	949
Superior frontal gyrus, R	22.3	69.9	7.1	405
Middle frontal gyrus, orbital, R	47.4	53.7	−3.2	181
Pallidum, L	−18.2	2.1	6.4	126
Postcentral gyrus, L	−50.0	−9.0	55.1	111
Cuneus, L	−12.3	−84.2	18.2	123
Inferior occipital lobe, R	34.1	−96.0	−8.3	65
Middle frontal gyrus, L	−31.5	63.3	8.6	158
Inferior frontal gyrus, triangular, L	−53.6	36.0	14.5	51
Superior frontal gyrus, L	−21.9	64.0	21.9	58
Superior frontal gyrus, L	−11.6	13.1	52.9	469
Postcentral gyrus, R	64.4	−16.4	41.8	94
Postcentral gyrus, L	−31.5	−40.7	69.8	528
Superior frontal gyrus, R	17.9	67.0	21.2	77
Precuneus, L	−11.6	−49.6	51.4	322
Middle frontal gyrus, L	−29.3	32.3	52.1	54
Middle frontal gyrus, R	31.9	16.8	38.1	54
Fusiform gyrus, R	35.6	−54.0	−7.6	181
Inferior temporal gyrus, R	51.1	−16.4	−20.9	91
Putamen, R	21.6	2.1	11.6	97
Precuneus, R	14.9	−56.2	43.3	76
Cingulate gyrus, posterior, R	12.0	−40.7	30.0	115
Middle frontal gyrus, L	−30.0	15.3	62.5	55
Fusiform gyrus, L	−33.0	−62.8	−6.1	74
Positive (more volume in OPEL)
Rectus, R	10.5	20.5	−20.9	61 323
Insula, L	−31.5	24.2	2.7	20 286
Middle frontal gyrus, R	26.7	54.4	2.7	2 305
Rolandic operculum, R	40.8	−20.8	18.9	22 949
Inferior frontal gyrus, triangular, L	−40.4	43.4	8.6	5 000
Inferior frontal gyrus, triangular, R	42.2	31.6	18.2	1 582
Middle frontal gyrus, orbital, L	−30.0	38.9	−12.0	501
Precuneus, R	20.8	−59.9	22.6	2 730
Middle temporal gyrus, L	−52.9	−43.7	6.4	6 562
Cingulate gyrus, posterior, L	0.2	−34.8	31.5	839
Lingual gyrus, L	−8.7	−65.8	6.4	3 121
Fusiform gyrus, L	−46.3	−52.5	−17.9	258
Inferior temporal gyrus, R	46.7	−46.6	−16.5	154
Inferior frontal gyrus, orbital, R	26.7	36.0	−16.5	429
Lingual gyrus, R	18.6	−68.7	−9.1	466
Fusiform gyrus, R	39.3	−67.3	−19.4	602
Cerebellum (IV-V)	22.3	−39.2	−17.9	233
Postcentral gyrus, R	55.5	−3.1	27.1	318
Lingual gyrus, R	7.6	−88.6	−9.1	102
Temporal pole, superior, R	44.4	18.3	−22.4	123
Cerebellum (Crus I), L	−22.7	−87.2	−22.4	517
Fusiform gyrus, L	−24.1	−68.0	−12.0	95
Lingual gyrus, L	−9.4	−74.6	−9.1	87
Calcarine, L	−2.8	−91.6	−10.6	223

Notes: Peak structure is the brain region where most of the cluster is located; XYZ coordinates are peak structure coordinates; L = left; R = right; Threshold z = ± 1.96, k >50.

Figure 1.

Gray matter network patterns associated with parental longevity. Areas in red are positively (greater volume) associated with parental longevity status. Areas in blue are negatively (lower volume) associated with longevity status. Data was adjusted for age, sex, years of education, estimated total intracranial volume, and white matter hypointensities. The threshold level was set to z ± 1.96, p < .05.

The associations between the participant-specific expression scores (the extent to which the GM covariance network associated with parental longevity was expressed at the individual level) and cognitive test performance are summarized in Table 3. The global cognitive measure was significantly associated with parental longevity status (β = −0.56, p < .001) as well as the individual measures for attention (β = −0.22, p < 0.01), episodic memory (β = −0.65, p < .001), language (β = −0.43, p = .02) and visuomotor speed (β = 0.20, p < .001) such that the derived network as a whole was more pronounced in those that have better cognitive performance.

Table 3.

Subject Specific Factor Scores Associated With Cognitive Performance (N = 119)

Neuropsychological test	βeta	p Value
Global Cognition*	−0.56	<.001
Digit Symbol Substitution	−0.22	<.01
Trail Making Test (B-A)	−0.004	.70
Free Recall	−0.65	<.001
Boston Naming*	−0.43	.02
Trail Making Test—A	0.20	<.001
Digit Span	−0.08	.77

^*Global cognition (N = 101) and naming (N = 117). Bold values indicate statistical signifiance.

Discussion

This study identified a GM covariance network associated with parental longevity in older adults—which also varied as a function of cognitive performance. Key nodes in this GM covariance network were middle frontal gyrus (including rectus), inferior frontal gyrus, middle temporal gyrus (including hippocampus), thalamus, and the insula (see Figure 1; regions highlighted in red). GM volume in these brain regions was relatively greater in those who were enriched for exceptional longevity than in those who were not enriched for exceptional longevity. Significant associations between the participant-specific expressions of this longevity network and several cognitive measures were also observed, including measures of attention, episodic memory, language, visuomotor speed, and global cognition.

To our knowledge, this is the first study that has (a) generated a GM covariance network associated with parental longevity, (b) utilized a sample of solely older adults, (c) examined whole-brain structural data, and (d) used a different definition of longevity. Our results are both consistent and an extension of prior studies of the brain substrates associated with parental longevity. Previous research has shown that parental longevity (defined as having at least 1 parent lived ≥85 years) was associated with less white matter disease as well as subcortical brain changes (19,20). These studies, however, primarily included middle-aged adults and analyzed self-selected regions of interest or global brain volumes. Therefore, the widespread network of cortical and subcortical volume associated with parental longevity in this study offers a more comprehensive picture of brain health in those enriched with parental longevity.

The widespread nature of the derived parental longevity brain network—and the fact that it also varied as function of several cognitive measures—are consistent with current literature indicating better attention, executive function, and memory processes in those with parental longevity (19,20). Yet, cognitive performance in those that were enriched and not enriched with parental longevity was not significantly different in this study (see Table 1). Thus, some of the variability in this brain network can be attributed to parental longevity and some to cognitive performance—yet future studies are needed to examine this issue further. Longitudinal studies are also needed to determine if this parental longevity brain network is associated with slower cognitive decline or resilience to decline as suggested by others (11,15). One study focusing primarily on the relationship between cognition and parental longevity, for example, demonstrated that children of long-lived parents experienced slower cognitive decline—defined by multiple cognitive domains (11). These findings also make sense because the brain regions that contributed to our parental longevity network have been linked to several cognitive domains: sensorimotor processing, attention, working memory, episodic memory, language, motor control, and balance (37–41). A better understanding of the brain regions associated with parental longevity as well as their associations with cognitive performance can inform the development of multimodal interventions that will allow for healthy brain aging.

There are several important strengths of this research. The LonGenity cohort is a genetically homogeneous and physically healthy cohort (see Table 1) that is dementia free at baseline—thereby decreasing the influence of potential confounders such as genetics or overall health. Our whole-brain and data-driven statistical approach is also resistant to between-subject variability and collinearity and removes the need for multiple comparison adjustments—thereby offering a more comprehensive view of brain health in those enriched for parental longevity than previous studies. The administration of a comprehensive battery of neurophysiological tests also granted us the ability to correlate the individual expressions of our derived brain network with a wide range of cognitive functions that undergo age-related changes, including episodic memory and attention.

There are several important limitations of this research as well. Our cross-sectional study design does not allow for the derivation of causal inferences. Although we were able to identify some statistically significant associations, our sample size was limited, and future studies in larger samples with longitudinal follow-up will be needed to expand our observations. Note for example that when we tried to compare our results with different cutoffs for parental longevity (eg, ≥85 years), the overall patterns were statistically significant and topographically very similar to what was observed with our original cutoff (≥95 years)—but the stability of the voxels that contributed to this pattern was not statistically reliable (see Supplementary Figure 1). This typically happens when the analysis is underpowered, or the pattern associated with your outcome is not driven by a particular set of regions (widespread) as was the case here. The above average physical and cognitive health of study participants also limits the generalizability of our results—as does the exclusion of those with neurological impairments and missing cognitive performance data (in 20 participants). Moreover, information regarding childhood environment and resources was not available but could have contributed to our results. APOE-4 was only available in a subset and therefore not included in our statistical model but will be examined in the future. Finally, we note that although age was adjusted for in our statistical model, it is possible that it contributed to our brain pattern as those enriched for parental longevity were on average 2.4 years younger than those who were not enriched for longevity.

In summary, a widespread GM covariance network in healthy older adults that explored the relationship between parental longevity and overall brain health was derived with multivariate statistical methods. Individual expression of this network was also associated with measures such as overall cognition, episodic memory, language, visuomotor speed, and attention. In other words, some of the variability in this brain pattern can be attributed to parental longevity and some to cognitive performance. The preservation of certain areas differentiated by parental longevity status could help in discovering the tools for healthy brain aging. The regions that were included in our network have the potential to be altered through interventions—such as the hippocampus which has been linked to an increase in size due to exercise and improved cognitive health by means of participation in leisure activities.

Funding

This work was supported by the National Institute on Aging (NIA) grants: R01AG062659, R01AG057548, R01AG061155, K76AG083274, and R01AG057909—and a predoctoral diversity supplement 3R01AG062659-04S1.

Conflict of Interest

H.B. is a consultant for Connective Leadership.

Data Availability

All data and codes used to analyze these data are available upon request.

Author Contributions

H.B. and N.D. conceptualized and developed the questions, methods, and analyses described in this manuscript. N.D. wrote the first draft of the manuscript and analyzed the data under the supervision of H.B. N.B. and S.M. initiated, provided resources, and directed the cohort and study procedures. S.M., S.A., H.B., and N.D. curated the data. All authors revised the manuscript and had final responsibility for the decision to submit for publication.