Cerebral white matter myelination is associated with longitudinal changes in processing speed across the adult lifespan

Abstract

Myelin’s role in processing speed is pivotal, as it facilitates efficient neural conduction. Its decline could significantly affect cognitive efficiency during ageing. In this work, myelin content was quantified using our advanced MRI method of myelin water fraction mapping. We examined the relationship between myelin water fraction at the time of MRI and retrospective longitudinal change in processing speed among 121 cognitively unimpaired participants, aged 22–94 years, from the Baltimore Longitudinal Study of Aging and the Genetic and Epigenetic Signatures of Translational Aging Laboratory Testing (a mean follow-up duration of 4.3 ± 6.3 years) using linear mixed-effects models, adjusting for demographics. We found that higher myelin water fraction values correlated with longitudinally better-maintained processing speed, with particularly significant associations in several white matter regions. Detailed voxel-wise analysis provided further insight into the specific white matter tracts involved. This research underscores the essential role of myelin in preserving processing speed and highlights its potential as a sensitive biomarker for interventions targeting age-related cognitive decline, thereby offering a foundation for preventative strategies in neurological health.

Gong, Murat, et al. report that higher myelin water fraction values, measured via MRI, correlate with better-maintained processing speeds in ageing adults. Their study highlights myelin’s potential as an early imaging biomarker for cognitive decline.

Graphical Abstract

Graphical Abstract.

Introduction

Myelin, the insulating sheath around axons, is critical for boosting neural impulse conduction and reducing refractory times, which are essential for rapid information processing in the brain.^1,2 This modulation of speed is essential for cognitive functions, ensuring synchronized neural signals across extensive brain networks.³ Salami et al.⁴ revealed that in mice, changing myelination patterns of thalamocortical axons allows uniform sensory signal timing to the cortex, despite varying distances between origins and destinations. This myelination gradient accounts for the 10-fold difference in conduction velocity, facilitating the synchronization of long-distance neural connections.⁴ Furthermore, studies suggest that myelination is not static; it can adapt over time, altering conduction speeds in response to various stimuli, suggesting a fluid interplay between myelin dynamics and neural processing speed (PS).^5-8 This evidence underscores the significance of myelin in maintaining the brain’s communication efficiency and its potential adaptability in response to changing cognitive demands.

Numerous studies have explored the relationship between white matter microstructure and cognitive decline. Shafer et al.⁹ used diffusion tensor imaging (DTI) to observe accelerated white matter decline in early mild cognitive impairment and dementia stages, linking microstructural changes to cognitive impairment. Similarly, using DTI, Jacobs et al.¹⁰ demonstrated a negative correlation between executive function, PS and white matter integrity. However, DTI metrics, while sensitive to microstructure changes, lack specificity to biological details.^11,12 Alternatively, direct investigation into cerebral myelination and PS using relaxometry-based methods reveals new insights. In young children, Chevalier et al.¹³ found that higher myelin volume fraction (VF_M) in specific brain regions correlated with shorter inspection times, reflecting the minimal visual presentation time needed to identify targets. For older adults, Lu et al.¹⁴ and Chopra et al.¹⁵ reported that better-preserved myelin was associated with faster cognitive processing, emphasizing myelin’s role in maintaining cognitive function with ageing. In patients with multiple sclerosis (MS), Abel et al.¹⁶ showed correlations between myelin variations and cognitive test scores, highlighting myelin’s role in neurological conditions.

PS broadly refers to how quickly an individual can perform cognitive operations necessary for completing a task. Research has shown that the slowing of PS with age contributes to declines in higher-order cognitive functions such as memory and executive functioning.^17-21 The role of axon myelination is crucial in this context, as it enhances neural signal transmission speed and reduces refractory periods, facilitating the integration of information across widespread neural networks that support both cognitive and motor functions.^22,23 Intact myelin, therefore, plays a significant role in maintaining cognitive performance by enabling efficient communication within the brain. Despite the compelling evidence of an intimate relationship between myelin status and PS, a critical gap remains in our understanding of the impact of myelination on longitudinal changes in PS across the adult lifespan. The present study sought to fill this gap through an original examination of the relationship between myelin content, measured using state-of-the-art MRI, and retrospective longitudinal changes in PS over several years leading up to the MRI.

Quantitative MRI offers a unique non-invasive approach for measuring myelin content in vivo. While various MRI methods, such as DTI, magnetization transfer imaging and relaxation times, are sensitive to myelin, they lack specificity to any determinant of white matter tissue due to their sensitivity to various other physiological factors including macromolecular content, axonal degeneration and architectural features, such as fibre fanning and crossing. To overcome these limitations, advanced MRI methods have been developed based on multicomponent relaxometry that provide greater sensitivity and specificity in non-invasive MRI myelin mapping.^24,25 Multicomponent relaxometry has characterized two main water pools in white matter, with distinct relaxation times and fractions. The pool exhibiting the more rapid transverse relaxation and smaller fraction size has been attributed to myelin-bound water, while the more slowly relaxing pool has been assigned to relatively unbound intra- and extracellular water.^25,26 The fraction of the former pool, representing the water trapped between the myelin sheaths, is calculated as the myelin water fraction (MWF) and has been shown to represent a direct measure of myelin content.^24-27 Various MRI techniques for MWF imaging have been introduced and applied in clinical investigations. The advantages and limitations of each method and their application have been well described in recent reviews.^26,28-30 Among them, the Bayesian Monte Carlo multicomponent-driven equilibrium single-pulse observation of T₁ and T₂ (BMC-mcDESPOT) analysis^31-34 stands out. BMC-mcDESPOT is a bicomponent relaxometry MRI method assuming a two-component system consisting of short- and long-relaxation water pools. The short component corresponds to the signal of water trapped within the myelin sheaths, while the long component corresponds to intra-/extracellular water. This analysis explicitly accounted for non-zero echo time (TE), as incorporated into the TE-corrected mcDESPOT signal model.^33,34 BMC-mcDESPOT-derived MWF has revealed myelin deterioration in mild cognitive impairment and dementia³⁵ and its correlation with genetic, metabolic and vascular factors.^36-43

Building on our previous work investigating associations between MWF and other cognitive domains²² using the same study cohort, in this study, we employed linear mixed-effects analyses to investigate the association between myelin content, measured by BMC-mcDESPOT, and retrospective changes in PS over several years. Our final study cohort comprised 121 cognitively unimpaired adults spanning a broad age range from 22 to 94 years. Voxel-wise analyses and regional analyses were conducted in both cerebellar white matter and cerebral lobar white matter. Our hypothesis posits that lower myelin content would be associated with a steeper decline in PS. Through our investigation, we aim to contribute to a deeper understanding of the complex interplay between myelin content and PS, providing valuable insights into the cognitive changes that occur across the adult lifespan. This knowledge may have implications for developing targeted interventions to support healthy brain ageing and potentially inform therapeutic strategies for cognitive disorders associated with myelin abnormalities.

Materials and methods

Participants

The study sample comprised cognitively normal participants drawn from two studies: the Baltimore Longitudinal Study of Aging (BLSA) and the Genetic and Epigenetic Signatures of Translational Aging Laboratory Testing (GESTALT). Both studies adhered to identical inclusion and exclusion criteria, with a shared objective of assessing multiple ageing-related biomarkers. Prior to enrolment in our MRI protocol, stringent eligibility criteria were applied, ensuring the absence of central nervous system diseases (e.g. dementia, stroke, bipolar illness and epilepsy), cardiac disease, pulmonary disease and metastatic cancer. Participants with metallic implants or significant neurological or medical disorders were excluded from the MRI cohort. The research protocols received ethical approval from the local institutional review boards, and all participants provided written informed consent at each visit.

Participants were determined to be cognitively normal if they had ≤ 3 errors on the Blessed Information–Memory–Concentration test⁴⁴ and, where available, a clinical dementia rating⁴⁵ of zero. If participants had > 3 errors on the Blessed Information–Memory–Concentration or a clinical dementia rating > 0, their clinical and neuropsychological data were thoroughly reviewed as part of consensus case conferencing. Visits at which participants were determined to have impairment or dementia were excluded from these analyses. Among seven participants who were flagged for case conferencing (i.e. Blessed Information–Memory–Concentration errors > 3 or clinical dementia rating > 0), six were determined to be cognitively normal as a result of case conferencing.

Neuropsychological testing

Participants were administered the Digit Symbol Substitution Test (DSST) to measure PS over a mean follow-up duration of 4.34 ± 6.34 years (339 visits total).⁴⁶ Participants are required to match specific symbols with corresponding digits within a specified time frame. Higher scores on the DSST indicate faster PS and cognitive efficiency. The test has been extensively used to assess PS and executive function in healthy individuals and clinical populations.^47-49 Further, given concerns about the specificity of sole DSST to measure PS, we have also adopted a PS composite score, as suggested before.¹⁴ PS composite score combines results from the Trail Making Test Part A (TMT A)⁵⁰ and DSST. The TMT A scores were first natural log-transformed and then Z-transformed to standardize the scores. To maintain consistency, we reversed the signs of the Z-scores so that higher composite scores consistently indicated superior PS performance across both TMT A and DSST scores. Similarly, the DSST scores were Z-scored using the baseline mean and standard deviation. The final composite score was obtained by averaging the Z-scores from both measures. In cases where one measure was missing, the PS was recorded as the available score, ensuring that data integrity was maintained throughout the analysis. Among the total 339 cognitive observations, we identified 43 cases with only Trail A, 23 cases with only DSST, 257 cases with both DSST and Trail A and 16 cases with both missing. Figure 1 illustrates the availability of retrospective longitudinal measurements of PS and PS composite anchored at the time point of the MRI scan. The longitudinal PS and PS composite scores of each participant against the age at the assessment are also plotted.

Figure 1.

Overview of available longitudinal PS data. The data availability of longitudinal cognitive measures for individual participants is presented, covering both PS and PS composite scores. The top row delineates the retrospective duration of measurements anchored at the time of the MRI scan. The bottom row shows the Z-scores for longitudinal measurements of both PS and its composite counterpart. Participants are color-coded for the BLSA and GESTALT studies. Additionally, male and female subjects are distinguished by circles and triangles, respectively.

MRI data acquisition

All MRI scans were performed on a 3-T whole-body Philips MRI system (Achieva, Best, The Netherlands) using the internal quadrature body coil for transmission and an eight-channel phased-array head coil for signal acquisition. Following the BMC-mcDESPOT imaging protocol,³³ three sequences were acquired to calculate MWF: 3D SPGR images were acquired with flip angles (FAs) of [2 4 6 8 10 12 14 16 18 20]°, a TE of 1.37 ms and a repetition time of 5 ms. Three-dimensional bSSFP images were acquired with FAs of [2 4 7 11 16 24 32 40 50 60]°, a TE of 2.8 ms and a repetition time of 5.8 ms. The bSSFP images were acquired twice with radiofrequency excitation pulse phase increments of 0 and π to account for off-resonance effects.⁵¹ All SPGR and bSSFP images were acquired with an acquisition matrix of 150 × 130 × 94 and a voxel size of 1.6 mm × 1.6 mm × 1.6 mm. To correct for excitation radiofrequency inhomogeneity, B₁⁺, we used the double-angle method⁵² by acquiring two fast spin-echo images with FAs of 45° and 90°, a TE of 102 ms, a repetition time of 3000 ms and an acquisition voxel size of 2.6 mm × 2.6 mm × 4 mm. All images were acquired with a field of view of 240 mm × 208 mm × 150 mm and reconstructed to voxel size of 1 mm × 1 mm × 1 mm. All MRI scans for BLSA and GESTALT participants were performed with the same MRI system, running the same pulse sequences, at the same facility, and directed by the same investigators.

Regions of interest

For each participant, a whole-brain MWF map was generated from the SPGR, bSSFP and B₁ images using BMC-mcDESPOT.³³ Further, using the FSL software,⁵³ the averaged SPGR image over FAs underwent non-linear registration to the Montreal Neurological Institute (MNI) standard space, with the computed transformation matrix then applied to the corresponding MWF maps. White matter (WM) segmentation was conducted using the FSL FAST algorithm on the MNI template, and WM regions of interest (ROIs) were delineated by applying the WM mask to the MNI structural Atlas. Our ROIs included cerebellar WM and the WM of the four cerebral lobes, namely the frontal, parietal, temporal and occipital lobes. We also investigated whole-brain WM. All ROIs were eroded to reduce partial volume effects and imperfect image registration using a kernel box of 2 voxels × 2 voxels × 2 voxels with the FSL tool fslmaths. These six WM ROIs were defined using the MNI structural atlas in FSL.⁵³ Within each ROI, the mean MWF values were calculated. Further analysis included the examination of additional small structures and white matter tracts, with the results provided in the Supplementary material (see Supplementary csv File). Before calculating the ROI values or conducting voxel-wise statistical analyses, MWF maps underwent smoothing using a boxcar filter with a kernel size of 3 mm × 3 mm × 3 mm. Smoothing was restricted to voxels exhibiting 95% of WM tissue defined using the FSL FAST segmentation to the MNI structural Atlas to prevent enhancing partial volume effects with grey matter and CSF.

Statistical analysis

Linear mixed-effects models were used to investigate associations between regional or voxel-wise MWF values and retrospective changes in PS score or PS composite score. The dependent variable was each of the corresponding longitudinal PS Z-scores. Age at MRI, time at or to MRI, sex, race, years of education, regional or voxel-wise MWF value and MWF × time interaction were included as independent variables. A random intercept was included per participant. The explicit regression model was:

$$\mathrm{PS}{i}_j={\mathrm{\beta }}_0+{\mathrm{\beta }}_{\mathrm{age}}\times \mathrm{ag}{\mathrm{e}}_i+{\mathrm{\beta }}_{\mathrm{ag}{\mathrm{e}}^2}\times \mathrm{ag}{\mathrm{e}}_i^2+{\mathrm{\beta }}_{\mathrm{sex}}\times \mathrm{se}{\mathrm{x}}_i+{\mathrm{\beta }}_{\mathrm{race}}\times \mathrm{rac}{\mathrm{e}}_i+{\mathrm{\beta }}_{\mathrm{EDY}}\times \mathrm{ED}{\mathrm{Y}}_i+{\mathrm{\beta }}_{\mathrm{time}}\times \mathrm{tim}{\mathrm{e}}_{ij}+{\mathrm{\beta }}_{\mathrm{MWF}}\times \mathrm{MW}{\mathrm{F}}_i+{\mathrm{\beta }}_{\mathrm{time}\times \mathrm{MWF}}\times \mathrm{MW}{\mathrm{F}}_i\times \mathrm{tim}{\mathrm{e}}_{ij}+{\mathrm{\epsilon }}_{ij}+b_i,$$

where PS_ij is the processing speed (PS or PS composite) score of subject i at time point j, age is the age (in years) at MRI, age² is the square of the age (in years) at MRI, MWF_i is the MWF value of subject i, time_ij is the time at or to MRI of subject i at time point j, b_i ∼ N(0, σ_b²) is the random intercept for subject i and ɛ_ij ∼ N(0, σ_ɛ²) is the residual. We note that the main parameters of interest in this analysis are β_MWF, reflecting the adjusted fixed effect between MWF and PS; β_time, reflecting the expectation of annual retrospective longitudinal change in PS; and β_{time × MRI}, reflecting the expectation of the difference in the retrospective longitudinal decline in PS per unit difference in MWF. All ROI-based statistical analyses were corrected for multiple comparisons using the false discovery rate (FDR) method to ensure a significance threshold of P < 0.05. Statistical significance for regional analysis was determined using corrected P < 0.05, while for voxel-wise analysis, significance was defined as uncorrected P < 0.01 with a cluster size exceeding 400 voxels. Detailed method of significant cluster calculation is given in the Supplementary material. Since there is no consensus on the appropriate multiple comparison correction method for longitudinal voxel-wise analysis, we compared our procedure with the Benjamini–Hochberg procedure for FDR-based correction. We concluded that our thresholding approach is more conservative than the Benjamini–Hochberg method, and the detailed results are given in the Supplementary material (Supplementary Figs 1 and 2). To illustrate the trajectory of longitudinal PS decline, we plotted three predicted decline lines by holding other independent variables at the mean and using three different MWF values at the 25th, 50th and 75th percentiles. All ROI-based analyses were conducted in R, version 4.2.0, and voxel-wise analyses were implemented in MATLAB 2022a, using the ‘fitlme’ function.

Finally, we conducted a sensitivity analysis investigating the association between MWF and changes in the PS composite scores derived when both TMT A and DSST measures were available. Additionally, we examined the correlation between the TMT A test itself and MWF to discern the contribution of TMT A to the PS composite. Detailed descriptions and results of these analyses are presented in the Supplementary material (Supplementary Figs 3 and 4 and Supplementary Table 1).

Results

Table 1 provides the demographic characteristics of our final study cohort, which comprised 121 cognitively unimpaired participants aged between 22 and 94 years, after excluding nine participants due to cognitive impairment and eight data sets due to low image quality caused by severe motion artefacts. Detailed demographic information for BLSA and GESTALT separately is also provided indicating similar characteristics (Table 1). Of all participants, 105 had MRI scans within 3 years of the last neuropsychological testing visit, with only one participant having a visit 3.1 years after the MRI scan. In this cohort, the minimum time interval between two consecutive cognitive tests was 0.8 years, with 55 tests conducted between 0.8 and 1 year and 163 tests conducted beyond a 1-year interval. However, due to its longer running status, BLSA has much more extended follow-up visits as compared to GESTALT. No significant age differences were observed between males and females (P > 0.1). The final cohort included 53 females (43.8%) and 24 Black participants (19.8%).

Table 1.

Participant demographics

Characteristic	All participant (n = 121)	BLSA (n = 69)	GESTALT (n = 52)
Age at MRI scan (years), mean (SD)	55.9 (20.8)	56.6 (22.9)	56.4 (18.1)
Sex male, n (%)	68 (55.3%)	36 (52.2%)	31 (59.6%)
Hypertensive, n (%)	32 (26.4%)	18 (26.1%)	14 (26.9%)
BMI, mean (SD)	26.6 (4.19)	25.8 (4.12)	27.7 (4.05)
Education, mean (SD)	16.1 (2.8)	16.7 (2.8)	15.4 (2.8)
Number of cognitive assessments, mean (SD)	2.8 (3.0)	3.6 (3.8)	1.7 (0.46)
Duration of cognitive assessments (years), median (mean, SD)	2.2 (4.3, 6.3)	14.7 (12.6, 8.3)	2.2 (1.9, 0.89)
Race, n (%)
Race, Black	24 (19.8%)	15 (21.7%)	9 (17.3%)
White	82 (67.8%)	45 (65.2%)	39 (75.0%)
API and other	15 (12.4%)	9 (13.0%)	4 (0.07%)

API, Asian and Pacific Islander; BMI, body mass index.

Figure 2 displays the regression coefficient maps illustrating the effects of the MWF × time to MRI, MWF and time to MRI terms incorporated in the linear mixed-effects models for both PS and PS composite. The visualization includes only statistically significant voxels, determined using an uncorrected threshold of P < 0.01 and a minimum cluster size of 400 voxels. We note that multiple comparison corrections to limit the false discovery rate using the Benjamini–Hochberg procedure⁵⁴ yielded similar results (see Supplementary Figs 1 and 2). While limited significant clusters were observed for the fixed-effects MWF term across all brain regions, the longitudinal time to MRI term exhibited notable and consistent negative correlations with both PS and PS composite, indicating that longer periods spanned from initial cognitive assessment to MRI scan were associated with greater declines in PS in various brain areas. Importantly, the MWF × time to MRI interaction terms revealed a pattern of numerous significant positively correlated clusters within several white matter regions for both PS and PS composite. In the results, significant clusters were predominantly found in the anterior brain regions, such as the temporal and parietal lobes, which is particularly evident in the PS composite results. The effect of the MWF × time interaction is evenly distributed across the white matter, with slightly higher values in the deep white matter regions.

Figure 2.

Regression coefficient maps from the linear mixed-effects models. Regression coefficient maps illustrating the effects of MWF × time to MRI, MWF or time to MRI terms in the linear mixed-effects models for (A) PS composite and (B) PS. Only statistically significant voxels are displayed (uncorrected P < 0.01, cluster size > 400 voxels). Limited significant clusters were observed for the fixed-effects MWF term, whereas the longitudinal time to MRI term exhibited significant negative correlations with PS and PS composite across most brain regions. Importantly, the MWF × time to MRI interaction terms revealed wide regions of positively correlated significant clusters within cerebral WM for both PS and PS composite.

To examine regional details, six ROI analyses were conducted to assess the association between regional MWF and PS. Figures 3 and 4 show representative longitudinal decline curves of PS scores and PS composite scores, respectively, obtained from the linear mixed-effects regression models. The curves illustrate the longitudinal changes in PS at low, median and high MWF values, represented by red, blue and green lines, respectively. Notably, lower MWF values were consistently associated with steeper declines in PS across the majority of ROIs in both figures. These associations were statistically significant for PS (Fig. 3), as indicated in Table 2, and significant for PS composite (Fig. 4) in all ROIs, except the occipital and temporal lobe regions, as shown in Table 3.

Figure 3.

Predicted longitudinal decline curves for PS from the linear mixed-effects models. (A)–(F) Corresponding representative longitudinal PS score decline curves obtained from the linear mixed-effects regression models. The magenta, yellow and green lines represent the longitudinal changes in PS at low, median and high MWF values, respectively. It is evident from the curves that lower MWF values were associated with steeper declines in PS. These associations, indicated by the ${\mathrm{\beta }}_{\mathrm{time}\times \mathrm{MWF}}$ term, are significant for all ROIs (whole brain: β = 0.673, P^BH = 6.51E-04; frontal: β = 0.679, P^BH = 4.51E-04; occipital: β = 0.422, P^BH = 1.23E-02; parietal: β = 0.549, P^BH = 1.80E-03; temporal: β = 0.552, P^BH = 1.21E-03; cerebellum: β = 0.589, P^BH = 1.08E-02).

Figure 4.

Predicted longitudinal decline curves for PS composite from the linear mixed-effects models. (A)–(F) Corresponding representative longitudinal PS composite score decline curves obtained from the linear mixed-effects regression models. The magenta, yellow and green lines represent the longitudinal changes in PS composite scores at low, median and high MWF values, respectively. It is evident from the curves that lower MWF values were associated with steeper declines in PS composite scores. These associations, indicated by the ${\mathrm{\beta }}_{\mathrm{time}\times \mathrm{MWF}}$ term, are significant for all ROIs, except in the occipital lobe and temporal lobe (whole brain: β = 0.346, P^BH = 4.35E-02; frontal: β = 0.380, P^BH = 4.35E-02; occipital: β = 0.206, P^BH = 1.42E-01; parietal: β = 0.292, P^BH = 4.41E-02; temporal: β = 0.246, P^BH = 7.16E-02; cerebellum: β = 0.422, P^BH = 4.35E-02).

Table 2.

Regression coefficients and P-values of relevant variables of the linear mixed-effects regression models examining the relationship between MWF and PS

		Whole brain	Frontal	Occipital	Parietal	Temporal	Cerebellum
MWF	β =	1.65	0.67	1.46	2.07	2.27	2.16
	P =	0.454	0.756	0.407	0.291	0.236	0.320
	PBH =	0.545	0.756	0.545	0.545	0.545	0.545
Time to MRI	β =	−0.0448*	−0.0412*	−0.0503*	−0.0462*	−0.0488*	−0.0527*
	P =	4.70E-11	3.73E-09	1.24E-13	1.60E-11	2.81E-13	5.85E-15
	PBH =	5.63E-11	3.73E-09	3.72E-13	2.40E-11	5.62E-13	3.51E-14
MWF × time	β =	0.673*	0.679*	0.422*	0.549*	0.552*	0.589*
	P =	2.17E-04	7.51E-05	1.23E-02	1.20E-03	6.05E-04	9.01E-03
	PBH =	6.51E-04	4.51E-04	1.23E-02	1.80E-03	1.21E-03	1.08E-02
Age	β =	−0.0415*	−0.0428*	−0.0414*	−0.0406*	−0.0406*	−0.0410*
	P =	2.04E-12	4.07E-12	5.85E-14	1.86E-12	5.88E-13	4.22E-14
	PBH =	2.45E-12	4.07E-12	1.75E-13	2.45E-12	1.18E-12	1.75E-13
Age2	β =	−5.67E-04*	−5.93E-04*	−5.55E-04*	−5.53E-04*	−5.46E-04*	−5.52E-04*
	P =	9.64E-04	6.62E-04	1.03E-03	1.10E-03	1.23E-03	9.12E-04
	PBH =	1.23E-03	1.23E-03	1.23E-03	1.23E-03	1.23E-03	1.23E-03
Sex	β =	−0.134	−0.141	−0.134	−0.124	−0.131	−0.144
	P =	0.237	0.214	0.236	0.279	0.244	0.201
	PBH =	0.279	0.279	0.279	0.279	0.279	0.279

The results are presented for lobar cerebral WM and cerebellar WM.

P^BH is the BH-adjusted P-values. Asterisk on the β coefficients indicates P and P^BH < 0.05; P and P^BH values < 0.05 are also bolded.

Table 3.

Regression coefficients and P-values of relevant variables of the linear mixed-effects regression models examining the relationship between MWF and PS composite

		Whole brain	Frontal	Occipital	Parietal	Temporal	Cerebellum
MWF	β =	−1.00	−1.38	−0.72	0.09	−0.38	0.35
	P =	0.640	0.511	0.674	0.961	0.837	0.868
	PBH =	0.961	0.961	0.961	0.961	0.961	0.961
Time to MRI	β =	−0.0488*	−0.0457*	−0.0539*	−0.0499*	−0.0525*	−0.0538*
	P =	1.45E-12	5.52E-10	4.01E-17	1.95E-13	3.81E-16	4.08E-19
	PBH =	1.74E-12	5.52E-10	1.20E-16	2.92E-13	7.61E-16	2.45E-18
MWF × time	β =	0.346*	0.380*	0.206	0.292*	0.246	0.422*
	P =	2.18E-02	1.16E-02	1.42E-01	2.94E-02	5.96E-02	2.15E-02
	PBH =	4.35E-02	4.35E-02	1.42E-01	4.41E-02	7.16E-02	4.35E-02
Age	β =	−0.0460*	−0.0471*	−0.0450*	−0.0445*	−0.0449*	−0.0441*
	P =	8.92E-15	2.63E-14	4.96E-16	1.65E-14	3.03E-15	6.30E-16
	PBH =	1.34E-14	2.63E-14	1.89E-15	1.98E-14	6.07E-15	1.89E-15
Age2	β =	−5.67E-04*	−5.86E-04*	−5.50E-04*	−5.38E-04*	−5.45E-04*	−5.30E-04*
	P =	7.29E-04	5.48E-04	8.98E-04	1.17E-03	1.01E-03	1.15E-03
	PBH =	1.17E-03	1.17E-03	1.17E-03	1.17E-03	1.17E-03	1.17E-03
Sex	β =	0.0198	0.0199	0.0187	0.0239	0.0232	0.0230
	P =	0.857	0.855	0.864	0.829	0.832	0.833
	PBH =	0.864	0.864	0.864	0.864	0.864	0.864

The results are presented for lobar cerebral WM and cerebellar WM.

Asterisk on the β coefficients indicates P and P^BH < 0.05; P and P^BH values < 0.05 are also bolded.

While no statistically significant fixed effects were found between MWF and PS in any ROI (Tables 2 and 3), longitudinal time to MRI terms were significantly associated with all ROIs for both PS and PS composite, consistent with the voxel-wise analysis. Our results also revealed age-related differences and age declines in all analyses, with significantly lower PS values observed with older ages (Tables 2 and 3). While sex differences in PS were not significant, it is likely due to the relatively limited sample size in this study as compared to the previous study where significant sex differences were found for a variety of cognitive tasks in a much larger sample size.⁵⁵

Finally, our sensitivity analysis conducted using a conservative definition of the composite PS score, that is, composite PS score derived only when both TMT A and DSST measures are available, indicates that the associations between MWF or MWF × time and composite PS score exhibit similar trends as described above. Interestingly, the results of the association between MWF × time and conservative PS composite score exhibit greater size effect and significance. However, the association between MWF × time and TMT A alone revealed significance in very limited voxel clusters. Detailed descriptions of the results of this sensitivity analysis are presented in the Supplementary material (Supplementary Figs 3 and 4 and Supplementary Table 1).

Discussion

The intricate relationship between the microstructural integrity of white matter, with a particular emphasis on myelin, and cognitive function is a burgeoning focal point in contemporary neuroscientific investigations.^22,56-58 Central to the efficient transmission of neural signals, white matter and its essential component, myelin, ensure coordinated neural communication across diverse circuits. Any degeneration of myelin, whether attributed to the natural ageing process or to specific pathological conditions, can detrimentally hinder this neural transmission, subsequently manifesting as clinical cognitive impairment.^22,35,59 Recent advances in neuroimaging have included the adoption of sophisticated modalities like quantitative MRI, allowing for an in-depth assessment of myelin content using parameters such as MWF, relaxation times and DTI metrics. A study of note revealed that a reduction in myelin content directly correlates with steeper cognitive decline in cognitive domains including attention, memory, executive function and verbal fluency during normative ageing.²² In a separate study of patients with MS, Abel et al.⁶⁰ explored the potential links between myelin water measures within the normal-appearing white matter areas and cognitive deficits in patients diagnosed with MS. Their findings revealed the association between myelin water metrics and cognitive capabilities in areas such as PS, verbal memory and word retrieval among patients with MS. In contrast, these associations were not evident in their cohort of healthy controls, aligning with the absence of a cross-sectional correlation between myelin and other cognition domains.²² However, as per our current understanding, no existing research explicitly demonstrated the association of specific measurements of myelin content with the longitudinal dynamics of PS.

In this retrospective longitudinal study, we observed significant associations between white matter myelin content measured by MWF and changes in PS. The regional analysis in Table 2 and Fig. 3 confirms that lower MWF values were significantly associated with steeper declines in PS measured by DSST across all examined brain regions, encompassing the whole brain, frontal, occipital, parietal and temporal lobes, as well as the cerebellum. PS is a fundamental cognitive function that enables an individual to perform tasks efficiently and accurately.²¹ While it involves distributed neural networks, certain brain regions are particularly implicated in PS tasks. Indeed, various regions of the brain, each with their unique functional specializations, collectively contribute to cognitive PS. The frontal lobe, notably the prefrontal cortex, plays a pivotal role in executive functions,^61,62 decision-making^63,64 and tasking switching,⁶⁵ all of which influence PS. The parietal lobe, with its involvement in spatial orientation and information integration,⁶⁶ also plays a part in rapid information processing.⁶⁷ Being the primary visual processing center,⁶⁸ the occipital lobe contributes to how quickly visual information can be processed and relayed. Meanwhile, the temporal lobe, responsible for auditory processing, memory and semantic comprehension, could indirectly and directly affect the speed at which information is understood and retained.⁶⁹ Lastly, the cerebellum, although traditionally associated with motor control, has been increasingly recognized for its contribution to cognitive processes and could play a role in refining the speed and efficiency of cognitive functions.⁷⁰ In sum, these brain areas collaborate, creating a cohesive network that enables the nimble and adept processing of data. Consequently, the detected associations between alterations in PS and regional white matter myelin content align with this paradigm.

Similar significant associations were found between MWF values and PS composite in all brain regions except occipital and temporal white matter (Table 3; Fig. 4). The DSST, since its inception by Wechsler,⁷¹ has gained widespread use as a measure of PS across a plethora of neuropsychological and cognitive research studies. However, the DSST taps functions beyond PS. It intricately interweaves multiple cognitive dimensions, such as working memory, visuo-manual coordination and fluid intelligence, as highlighted by various studies.^72-74 This multifarious nature of DSST metrics was emphasized by Baudouin et al.,⁷⁵ who posited that within older adult cohorts, the DSST taps not only PS but also executive function and visuo-perceptual integration.^75,76 In contrast, TMT A primarily assesses visual attention, visuo-spatial scanning and speed. To better capture the unique aspects of PS, we examined a composite score comprised of both DSST and TMT. We observed stronger associations between MWF and longitudinal changes in DSST-derived PS, compared with the composite measure, which may reflect the additional contribution of other aspects of cognition in DSST measurement, especially executive function. DSST is a sensitive but not specific cognitive task and may be particularly sensitive to a range of neural networks or white matter tracts influenced by myelin content.

To understand the contribution of TMT A to the PS composite score, we conducted a voxel-wise analysis examining the longitudinal changes in PS measured by TMT A and the cerebral myelination measured by MWF at the time of MRI, utilizing the same statistical approach. Surprisingly, our findings revealed that TMT A alone did not yield widespread significant clusters compared with DSST alone. This unexpected result may be attributed to the distinct cognitive domains assessed by these two tests (Supplementary Fig. 3). A relevant previous study²² has shown that MWF correlates with longitudinal changes in executive function but not attention. Given that DSST assesses PS, executive function and attention, while TMT A tests attention and PS but does not directly assess executive function, it is plausible that the correlation between myelination and DSST is maintained, while the correlation between myelination and TMT A is attenuated. Additionally, we considered a conservative PS composite score, applicable only when the average of TMT A and DSST exists. ROI and voxel-wise analyses both demonstrate widespread and strong correlation between PS decline and MWF (Supplementary Fig. 4 and Supplementary Table 1). However, we note that excluding participants who do not have both measures might also introduce a bias. In this case, participants who were able to undergo both tests might represent a cognitively healthier sample where the MWF associations are more robust. By including participants who had at least one of the measures, we might be obtaining results that are better generalizable.

The absence of significant associations between MWF and PS composite in the occipital and temporal lobes warrants further examination. While it is acknowledged that the occipital lobe is central to visual processing and the DSST and TMT A tasks are indeed sensitive to visual inputs, these tasks may tap primarily into broader cognitive networks rather than the specialized visual processing functions of the occipital lobe. Similarly, while the temporal lobe plays a key role in auditory processing, its contribution to these tasks might be more complex than a direct link. It is essential to consider that while both tasks rely on vision, they may not necessarily engage the same depth or type of visual processing as the specific functions of the occipital lobe. Furthermore, these tasks might not be as directly linked to the specialized auditory functions of the temporal lobe. Future exploration into the unique neural substrates, connectivity patterns and the nature of myelination across regions could shed light on the observed disparities in associations with MWF.

The voxel-wise analysis illustrated in Fig. 2 reveals intriguing insights about the influence of certain smaller but pivotal regions on the changes in PS. Interestingly, our findings revealed that the most anterior brain regions and white matter tracts exhibit the strongest associations between MWF and changes in PS and PS composite, suggesting that the brain regions most susceptible to neurodegeneration are indeed the most affected by ageing and poor myelination. This pattern is in line with the retrogenesis hypothesis (first in–last out), in which posterior brain regions are spared from degeneration as compared to anterior brain regions. Notably, white matter tracts generally show significant associations between higher MWF and reduced decline in PS. This correlation is further supported by the additional regional analysis detailed in the Supplementary File. For instance, the corpus callosum and the superior longitudinal fasciculus are two examples of important white matter tracts. Both corpus callosum and superior longitudinal fasciculus play a crucial role in shaping PS within the human brain. The corpus callosum serves as a major connector between the cerebral hemispheres,⁷⁷ facilitating interhemispheric communication and the integration of diverse cognitive functions. The efficient functioning of this tract is paramount for the swift transfer of information and coordinated processing, which underpin tasks requiring rapid cognitive responses.⁷⁸ Likewise, the superior longitudinal fasciculus, linking frontal and parietal cortices, supports the integration of sensory and motor information necessary for quick decision-making and execution of motor actions.⁷⁹ The highly myelinated structures of these white matter tracts aid in speeding up neural transmission, thus contributing significantly to PS during cognitive endeavours.⁸⁰ While the PS composite showed less pervasive distribution of significant associations, the superior longitudinal fasciculus maintained a significant association in the extended regional analysis, as presented in the Supplementary File.

An important strength of our study is the investigation of myelin content rather than a non-specific measure of white matter tract integrity such as FA or MD computed using conventional DTI. Our study explicitly demonstrated the association between a specific measurement of myelin content and the longitudinal dynamics of PS. While previous studies based on DTI had demonstrated the importance of white matter tract integrity in PS,^9,10 hypotheses regarding specific mechanisms underlying such associations could not be examined due to the limitations of DTI. Our study highlights myelin content as a specific biological phenomenon underlying these associations and adds to a growing body of research investigating myelin in relation to cognition in ageing²² as well as a variety of neurodegenerative disorders, including MS.¹⁶

Our study faces certain limitations. The BLSA study benefits from a longer duration and more frequent follow-up visits compared with the GESTALT study, resulting in fewer longitudinal assessments for GESTALT participants. Moreover, the representation of subjects aged 50–65 years was significantly lower than in other age groups, potentially impacting the analysis of myelin’s role in PS decline during normal ageing. The focus of two studies on an older population means that older participants have more retrospective cognitive measures than younger counterparts. Furthermore, our current cohort did not allow for a stable estimation of random slopes per subject in our statistical models, inevitably not fully accounting for individual variability. There is also an imbalance in gender and racial representation within our sample, highlighting the need for research on more diverse cohorts. Additionally, relaxometry-derived parameters including MWF values are affected by various physiological factors, such as iron content,⁸¹ fibre orientation,⁸² diffusion,^83-86 exchange,^87-92 off-resonance effects,^93-95 magnetization transfer,^96-99 J-coupling,^90,100,101 spin locking,^102-104 internal gradients^105-107 and magnetization spoiling,^108-110 which can introduce complexity and variability into the data.^81,82,87 The importance of these effects in a particular experiment will depend both on the specifics of the sample or subject under investigation and on the details of the pulse sequence, including the selection of parameters such as TE, repetition time, FA and gradient durations and amplitudes. These factors can introduce additional variability and limitations to the accuracy and interpretation of MWF measurements. These factors are not accounted for in the BMC-mcDESPOT approach or other existing MWF measurement techniques. Moreover, mcDESPOT has received criticisms for its instability in the accurate determination of MWF when water exchange is included in the signal model,^87,111,112 calling for further improvements and developments. Furthermore, as MWF is the fraction of water trapped within the myelin sheaths to the total water content, its measurement can be biased by water content changes in the intra- or extracellular space due to, for example, axonal degeneration or inflammation, which may lead to an artificial underestimation of myelin content. Therefore, external references should be used to mitigate this issue. Future work should aim to enhance the specificity of novel MRI imaging biomarkers while preserving sensitivity.

Conclusion

Our findings suggest that MWF serves as a powerful indicator for the nuances of PS, especially in relation to specific brain regions and their white matter tracts. The marked associations between these neural pathways and PS underscore the intricate orchestration of the brain’s architecture in ensuring efficient cognitive performance. Beyond shedding light on these relationships, this study also highlights the need for deeper investigations, especially when interpreting the subtle differences in associations across various cognitive assessments. Unpacking these complexities could offer further insights into the neurobiological underpinnings of cognitive functions, particularly in conditions where myelin integrity might be compromised. Moreover, as neuroimaging techniques continue to evolve, future studies might benefit from even more granular assessments, enabling a fuller understanding of the dynamic interplay between white matter content and cognitive speed. The present research, therefore, not only contributes to the existing body of knowledge but also sets the stage for more intricate and expansive explorations across the adult lifespan.

Supplementary material

Supplementary material is available at Brain Communications online.

Funding

This work was supported by the Intramural Research Program of the National Institute on Aging of the National Institutes of Health.

Competing interests

All authors declare no conflict of interest.

Data availability

Data from the BLSA and GESTALT can be obtained through formal requests on the study website. Analysis codes for this paper including statistical modelling and figure generation can be found at https://github.com/mrpadunit/ProcessingSpeedAndMyelin.