Intrinsic and extrinsic contributors to subregional thalamic volume loss in multiple sclerosis

Abstract

Objective

To evaluate the intrinsic and extrinsic microstructural factors contributing to atrophy within individual thalamic subregions in multiple sclerosis using in vivo high‐gradient diffusion MRI.

Methods

In this cross‐sectional study, 41 people with multiple sclerosis and 34 age and sex‐matched healthy controls underwent 3T MRI with up to 300 mT/m gradients using a multi‐shell diffusion protocol consisting of eight b‐values and diffusion time of 19 ms. Each thalamus was parcellated into 25 subregions for volume determination and diffusion metric estimation. The soma and neurite density imaging model was applied to obtain estimates of intra‐neurite, intra‐soma, and extra‐cellular signal fractions for each subregion and within structurally connected white matter trajectories and cortex.

Results

Multiple sclerosis‐related volume loss was more pronounced in posterior/medial subregions than anterior/ventral subregions. Intra‐soma signal fraction was lower in multiple sclerosis, reflecting reduced cell body density, while the extra‐cellular signal fraction was higher, reflecting greater extra‐cellular space, both of which were observed more in posterior/medial subregions than anterior/ventral subregions. Lower intra‐neurite signal fraction in connected normal‐appearing white matter and lower intra‐soma signal fraction of structurally connected cortex were associated with reduced subregional thalamic volumes. Intrinsic and extrinsic microstructural measures independently related to subregional volume with heterogeneity across atrophy‐prone thalamic nuclei. Extrinsic microstructural alterations predicted left anteroventral, intrinsic microstructural alterations predicted bilateral medial pulvinar, and both intrinsic and extrinsic factors predicted lateral geniculate and medial mediodorsal volumes.

Interpretation

Our results might be reflective of the involvement of anterograde and retrograde degeneration from white matter demyelination and cerebrospinal fluid‐mediated damage in subregional thalamic volume loss.

Background

Multiple sclerosis (MS) involves extensive and heterogeneous pathological changes in gray matter (GM) at the microscopic level, apart from the distinctive alterations observed in white matter (WM). ¹ , ² This diverse array of pathology ultimately culminates in different patterns of atrophy involving the WM and cortical and deep GM. ¹ , ³ , ⁴ Thalamic atrophy has attracted great interest as an imaging biomarker of neurodegeneration in MS and is known to reliably predict clinical outcomes, especially cognitive performance. ³ , ⁵ , ⁶ The thalamus appears disproportionately affected, ³ , ⁷ , ⁸ , ⁹ yet detectable volume loss is difficult to discern early in the disease course. Moreover, underlying pathological factors contributing to this atrophy are not yet clear, limiting therapeutic intervention. It is important to characterize such neurodegenerative changes in the full context of pathological tissue damage in the MS brain.

Contributions to thalamic atrophy involve intrinsic factors such as thalamic demyelinating lesions and neuronal loss in normal‐appearing (NA) thalamus and extrinsic factors such as neuroaxonal degeneration of thalamocortical and cortico‐thalamic projections and neurodegeneration within the cortex. ¹⁰ The ability to parse out the intrinsic and extrinsic substrates of thalamic atrophy in the living human brain would advance an understanding of disease progression at earlier stages in the disease course than histopathologic studies allow. High‐gradient diffusion MRI has shown promise in detecting microstructural changes related to neuroaxonal degeneration in people with MS, which is hypothesized to underlie atrophy in MS. ¹¹ , ¹² , ¹³ , ¹⁴

The compartment‐based soma and neurite density imaging (SANDI) model generates maps representing three signal fractions, originating from distinct compartments: soma (f _is, referring to cell bodies of various brain cell types, from neuroglia to neurons), neurites (f _in, comprising axons, dendrites and neuroglial processes), and extra‐cellular space (f _ec). ¹⁵ Recent work has shown that high‐gradient diffusion MRI using the SANDI model is sensitive to cellular changes in the cortical and deep GM in people with MS. ¹³ , ¹⁶ , ¹⁷ SANDI measure changes are expected to reflect MS‐related inflammation and neurodegeneration in the WM and GM, which may counterbalance each other in certain situations. In prior work, we found a significant association between reduced thalamic volume and reduced cortical cell body density reflected by SANDI measures, ¹³ supporting the contribution of extrinsic microstructural tissue damage to thalamic neurodegeneration. We hypothesized that thalamic volume loss reflects the net accumulation of cellular and axonal damage throughout the MS brain, making the thalamus a sensitive and appealing landmark for studying progressive neurodegeneration in people with MS.

To characterize clinically relevant microstructural changes that would advance an understanding of the heterogeneous nature of MS pathology and ultimately aid in identifying therapeutic targets, one optimal strategy is to probe the microstructural alterations on a regional basis in individual thalamic subregions, which exhibit diverse functions and neuronal inputs contingent upon their respective locations. ¹⁸ , ¹⁹ , ²⁰

The goal of this work was to dissect the contributions of tissue microstructure to subregional thalamic volume loss, using the in vivo microstructural measures offered by high‐gradient diffusion MRI and SANDI to provide insight into the intrinsic and extrinsic factors underlying atrophy within individual thalamic subregions. We applied the SANDI measures as biomarkers of microstructural neurodegeneration and inflammation and characterized the most salient factors contributing to atrophy of individual thalamic subregions, including intrinsic microstructural changes within specific thalamic subregions and extrinsic microstructural changes in the (dis)connected WM and cortex.

Methods

Participants

This cross‐sectional study was approved by the Institutional Review Board and represents a targeted reanalysis of previously published data on advanced diffusion imaging of the GM in people with MS. ¹³ All participants provided written informed consent. For the initial study, inclusion criteria for people with MS were a diagnosis of clinically definite MS, being relapse free for at least 3 months, and receiving stable disease‐modifying treatment or no treatment for at least 6 months. Exclusion criteria for all included participants were presence of other structural brain disease and contraindication to MRI. Of the initial dataset, we selected all people with MS and HC, of whom clinical data and the full scanning protocol (see “Image Acquisition”) were available. This yielded the inclusion of 41 people with MS scanned between 2015 and 2019 and 34 age‐ and sex‐matched HC. Of people with MS, 32 had relapsing–remitting MS and nine had progressive MS (three primary‐ and six secondary‐progressive MS).

The following characteristics of people with MS were extracted from the electronic medical records: age, sex, disease duration, and type of disease‐modifying therapy. A board‐certified neurologist blinded to imaging conducted a standard clinical examination. For all people with MS, the Symbol Digit Modalities Test and Expanded Disability Status Scale scores were reported, as broad measures of cognitive and clinical disability respectively.

Image acquisition

All included participants underwent brain imaging on a 3 Tesla MRI scanner (MAGNETOM Connectom; Siemens, Erlangen, Germany) equipped with a maximum gradient strength of 300 mT/m. A custom‐built 64‐channel phased array head coil was used for signal reception. ²¹ A diffusion‐weighted spin‐echo single‐shot EPI sequence was used for acquiring the diffusion data (2 mm³ isotropic resolution, sagittal slices, echo time/repetition time = 77/3600 ms, simultaneous multi‐slice imaging with slice acceleration factor 2, parallel imaging acceleration factor R = 2, anterior‐to‐posterior phase encoding, total acquisition time 51 min), following a previously established protocol. ²² , ²³ Nondiffusion‐weighted images (b = 0) were acquired every 16 images. For the SANDI analysis, diffusion data acquired at diffusion times of Δ = 19 at eight b‐values (b = 50–350–800–1500 s/mm² in 32 directions and b = 2400–3450–4750–6000 s/mm² in 64 directions) were used. Distortions due to susceptibility effects were corrected by means of five b = 0 images with a reversed‐phase encoding direction. Other imaging included a T1‐weighted multi‐echo magnetization prepared rapid acquisition gradient echo sequence (1 mm³ isotropic resolution, echo time/repetition time/inversion time = 1.15–3.03–4.89–6.75 ms/2530 ms/1100 ms, R = 3, flip angle = 7°, acquisition time 3 min 58 s) and a 3D fluid‐attenuated inversion recovery sequence (0.9 mm³ isotropic resolution, echo time/repetition time/inversion time = 389 ms/5000 ms/1800 ms, R = 2, acquisition time 5 min 47 s).

Diffusion MRI data processing

Diffusion data were preprocessed using an established pipeline. ²² In short, diffusion‐weighted images were corrected for gradient nonlinearity using in‐house software, ²⁴ and any distortions due to susceptibility and eddy current effects were corrected using topup and eddy in the FMRIB Software Library (FSL version 5.0). ²⁵ , ²⁶ , ²⁷ For the SANDI fitting, we used the AMICO software ²⁸ (https://github.com/daducci/AMICO), applying the default diffusivity parameters and a λ ₂ regularization term of 0.005, as previously described. ¹³ The SANDI model generated maps of the intra‐neurite, extra‐cellular, and intra‐soma signal fractions (f _in, f _ec, and f _is) and apparent soma radius (R _s ).

Registration and Segmentation

White matter segmentation

Lesion segmentation of FLAIR hyperintensities was performed using a validated FreeSurfer‐based automatic segmentation tool. ²⁹ A board‐ and subspecialty‐certified neuroradiologist (12 years of experience) manually edited the lesion masks. Lesion volumes were calculated by multiplying lesion area by slice thickness. Lesion masks were nonlinearly transformed to diffusion space using ANTs registration. ³⁰ WM lesion segmentations in diffusion space were subtracted from the WM volume fraction map, yielding a partial volume‐weighted NAWM mask in diffusion space.

Gray matter segmentation

Details regarding GM segmentation are reported in the “Methods Supplement.” Briefly, cortical surface and volumetric reconstruction were performed on the T1‐weighted data using FreeSurfer (version 5.3, http://surfer.nmr.mgh.harvard.edu). A partial volume‐weighted cortical GM mask was generated and divided into 210 cortical GM regions (105 in each hemisphere) based on the Brainnetome atlas. ³¹ Each thalamus was parcellated into 25 individual by subregions by THALAMODEL ³² with an enhanced T1‐weighted image as input. This enhanced T1‐weighted image was generated by in‐house software combining the T1‐weighted, fluid‐attenuated inversion recovery and fractional anisotropy images. ³³ The inclusion of the latter enhances the contrast particularly at the thalamus border adjacent to the WM. The bilateral paratenial, paracentral, and ventromedial subregions were excluded due to their size (smaller than 1 voxel in diffusion space). Eventually, the following bilateral subregions (N = 22) were included in the analysis (Fig. 1): anteroventral, central medial, central lateral, centromedian, laterodorsal, lateral geniculate, lateral posterior, limitans suprageniculate, lateral mediodorsal, medial mediodorsal, medial geniculate, reuniens medial ventral, parafascicular, anterior pulvinar, inferior pulvinar, lateral pulvinar, medial pulvinar, ventral anterior, magnocellular ventral anterior, anterior ventral lateral, posterior ventral lateral, and ventral posterolateral. All volumes are given in mL and corrected for the intracranial volume, yielding normalized volume measures.

Figure 1.

Visualization of thalamic subregions included in our study. Thalamic subregions included in the analysis, shown for right thalamus on axial T1‐weighted scans: AV, anteroventral; CL, central lateral; CM, central medial; CeM, centromedian; LD, laterodorsal; LGN, lateral geniculate; LP, lateral posterior; LSg, limitans suprageniculate; MDl, lateral mediodorsal; MDm, medial mediodorsal; MGN, medial geniculate; MVRe, reuniens medial ventral; Pf, parafascicular; PuA, anterior pulvinar; PuI, inferior pulvinar; PuL, lateral pulvinar; PuM, medial pulvinar; VA, ventral anterior; VAmc, magnocellular ventral anterior; VLa, anterior ventral lateral; VLp, posterior ventral lateral; VPL, ventral posterolateral. *Thalamic subregions showing a significant association between either intrinsic or extrinsic microstructure and its normalized volume, as further highlighted in the results section (Table 5).

Diffusion tractography

For the generation of masks of WM trajectories connecting thalamic subregions and the cortex, the MRtrix 3.0 software package (Brain Research Institute, Melbourne, Australia, http://www.brain.org.au/software/) was used to generate whole‐brain tractograms of HC. Details regarding diffusion tractography and connectome reconstruction are reported in the “Methods Supplement.” Briefly, for each HC participant, the multi‐shell, multi‐tissue constrained spherical deconvolution probabilistic tracking method ³⁴ was used to generate tractograms, which were converted to trajectory volumes, yielding 25 masks of WM trajectories connecting a thalamic subregion and the unilateral cortex in diffusion space. By concatenating all masks of HC with the use of a population‐specific template, 25 subregional‐specific template WM trajectories of the healthy brain were available per hemisphere. For people with MS, we subsequently transformed the template WM trajectories to native space. WM trajectory masks were masked by the WM lesion masks to obtain both a NAWM and lesional trajectory mask for each thalamic subregion connected to the cortex.

Connectome reconstruction

For the generation of masks of connected cortex, per thalamic subregion, a connectivity matrix of dimension 1 × 10⁵ (i.e., within‐hemisphere connectivity with each cortical atlas region) was generated with a density threshold of 30% to exclude spurious links. ³⁵ The top 10 cortical GM regions most strongly connected to the corresponding thalamic subregion based on data of HC were put together in one mask, yielding one structurally connected cortical GM mask in diffusion space per subregion.

SANDI measure calculations

The most strongly structurally connected cortical GM atlas regions, NAWM and lesional WM trajectory, and subregional thalamic masks in diffusion space were overlaid on the maps for all SANDI microstructural measures (f _in, f _ec, f _is, and R _s ) to calculate mean values of each metric per region. To minimize potential confounding by partial volume averaging effects, we calculated partial volume‐weighted mean values per SANDI measure.

Statistical analysis

Statistical analyses were performed with the use of Python v3.7 and Rstudio v2023.06.1. A full description of the statistical analysis is provided in the “Methods Supplement.” Briefly, Kolmogorov–Smirnov testing and histogram inspection of the variable distribution were performed to test the normality of the data. Participant characteristics were expressed as N (%), Mean (SD), or Median [IQR]. All normalized volumes and mean values of SANDI measures were transformed into Z‐scores based on the data of HC. Comparisons between people with MS and HC were performed using independent samples t‐tests or Mann–Whitney U tests as nonparametric alternative, if appropriate. Within‐subjects analyses were performed by means of repeated measures analysis or paired t‐tests, as appropriate. Univariate linear regression models were performed with the Z‐scores of SANDI metrics predicting the Z‐scores of normalized volumes of thalamic subregions, adjusting for age and sex. A final linear regression analysis with backward stepwise selection was performed with significant contributors to subregional atrophy as independent variables and normalized volume of the corresponding subregion as dependent variable, adjusting for age and sex. All statistical tests were two‐tailed. p‐values were false discovery rate (FDR)‐corrected for multiple comparisons (p _corr) per analysis step with α = 0.05.

Results

Demographics

People with MS had comparable age and sex distributions compared to HC (Table 1). People with MS had a mean disease duration of 9.7 (SD = 6.9) years, had predominantly relapsing–remitting MS (80.5%), and were mildly clinically disabled, reflected by a median Expanded Disability Status Scale score of 2.5 [IQR = 2.0–3.5] and mean Symbol Digit Modalities Test score of 51.0 (12.5). Normalized cortical and deep GM volumes were significantly lower in people with MS compared to HC (p = 4.17 × 10⁻⁹ and p = 0.001, respectively).

Table 1.

Baseline demographics and clinical characteristics of included participants.

	HC (N = 34)	People with MS (N = 41)	HC versus MS a
			p‐value b
Demographics
Age, years	39.1 (14.8)	45.2 (12.9)	0.058
Sex, male/female [number (%)]	14 (41.0)/20 (59.0)	11 (26.8)/30 (73.2)	0.286
Education, years	–	16.2 (2.5)
Disease characteristics
Disease duration, years	–	9.7 (6.9)
MS subtype, RR/SP/PP	–	33 (80.5)/5 (12.2)/3 (7.3)
DMT use, c yes [number (%)]	–	35 (85.4)
Clinical characteristics
EDSS, score	–	2.5 [2.0–3.5]
SDMT, raw score	–	51.0 (12.5)
MR characteristics
WM lesion load, mL	–	4.13 [1.87–10.80]
Cortical GM volume, mL	479.1 (59.0)	421.0 (73.9)	2.11 × 10 −4
Normalized cortical GM volume	0.30 (0.02)	0.27 (0.03)	4.17 × 10 −9
Deep GM volume, mL	44.1 (3.9)	41.9 (5.0)	0.035
Normalized deep GM volume	0.028 (0.002)	0.027 (0.002)	0.001

Variables are reported as mean (SD) or median [IQR] unless otherwise indicated.

DMT, disease‐modifying treatment; EDSS, Expanded Disability Status Scale; GM, gray matter; MR, magnetic resonance; MS, multiple sclerosis; PP, primary progressive; RR, relapsing–remitting; SDMT, Symbol Digit Modalities Test; SP, secondary progressive; WM, white matter.

^a Independent samples t‐test for continuous variables; chi‐square test for categorical variables.

^b p‐values <0.05 are marked in bold.

^c 35 of 41 people with MS were taking MS disease‐modifying therapy (9 dimethyl fumarate, 8 glatiramer acetate, 5 fingolimod, 5 ocrelizumab, 3 interferon beta (1a/1b), 2 natalizumab, 2 rituximab and 1 teriflunomide).

Subregional thalamic atrophy

We first assessed volume loss of thalamic subregions in people with MS compared to HC. The variability in significant volume losses of thalamic subregions was then evaluated by within‐subjects analysis.

Group comparisons

Thalamic subregions that showed volume loss in people with MS compared to HC included the left anteroventral, bilateral lateral geniculate, lateral posterior, medial geniculate, reuniens medial ventral, anterior pulvinar, medial pulvinar and ventral anterior, and right central medial, lateral mediodorsal, medial mediodorsal, inferior pulvinar, magnocellular ventral anterior, posterior ventral lateral and ventral posterolateral (Table S1). The reductions in subregional thalamic volume Z‐scores among people with MS when compared to HC are further highlighted by the right panel of Figure 2, as indicated by the gradient of blue to light blue coloring in the respective subregions.

Figure 2.

Normalized volumes across thalamic subregions showing multiple sclerosis‐related volume loss. Z‐score distribution of normalized volumes of thalamic subregions that show multiple sclerosis‐related volume loss is shown as violin plot (left) and as spatial representations on axial slices (right). Significant Z‐score differences are marked by brackets in the violin plot: *p < 0.05, **p < 0.01. AV, anteroventral; CM, central medial; LGN, lateral geniculate; LP, lateral posterior; MDl, lateral mediodorsal; MDm, medial mediodorsal; MGN, medial geniculate; MVRe, reuniens medial ventral; PuA, anterior pulvinar; PuI, inferior pulvinar; PuM, medial pulvinar; VA, ventral anterior; VAmc, magnocellular ventral anterior; VLp, posterior ventral lateral; VPL, ventral posterolateral.

Within‐subject variability in subregions

In people with MS, the within‐subject variability of the normalized volume Z‐scores relative to HC was assessed among the thalamic subregions. Significant differences are depicted in Figure 2.

Intrinsic microstructural changes in thalamic subregions

Then, we assessed alterations in SANDI measures of thalamic subregions in people with MS compared to HC. The variability in significant alterations of thalamic subregions was again evaluated by within‐subjects analysis.

Group comparisons

Among subregions showing MS‐related volume loss, people with MS showed significant reduction in f _is of the left anteroventral, bilateral anterior pulvinar and medial pulvinar, and right lateral geniculate, lateral posterior, medial mediodorsal, inferior pulvinar, posterior ventral lateral, and ventral posterolateral relative to HC (Table S2A). Reductions in subregional thalamic f _is Z‐scores in people with MS versus HC are reflected by the gradient of blue to light blue coloring in the respective subregions in the right panel of Figure 3A.

Figure 3.

Intra‐soma and extra‐cellular signal fraction across thalamic subregions showing multiple sclerosis‐related volume loss and microstructural changes. Z‐score distribution of intra‐soma (A) and extra‐cellular (B) signal fraction of thalamic subregions that show multiple sclerosis‐related microstructural changes are shown as violin plots (left) and as spatial representations in a graph against normalized volumes (center) and on axial slices (right). Significant Z‐score differences are marked by brackets in the violin plot: *p < 0.05, **p < 0.01, ***p < 0.001. AV, anteroventral; CM, central medial; LGN, lateral geniculate; LP, lateral posterior; MDl, lateral mediodorsal; MDm, medial mediodorsal; MGN, medial geniculate; MVRe, reuniens medial ventral; PuA, anterior pulvinar; PuI, inferior pulvinar; PuM, medial pulvinar; VA, ventral anterior; VAmc, magnocellular ventral anterior; VLp, posterior ventral lateral; VPL, ventral posterolateral.

Greater f _ec was observed in people with MS compared to HC in the left anteroventral, bilateral lateral geniculate, and medial pulvinar, and right lateral posterior, medial mediodorsal, medial geniculate, anterior pulvinar, inferior pulvinar, and ventral posterolateral (Table S2B). Again, increases in subregional thalamic f _ec Z‐scores in people with MS versus HC are highlighted by the right panel of Figure 3B, as indicated by the gradient of red to yellow coloring in the respective subregions.

The f _in and R _s of atrophied subregions did not differ significantly between MS and HC and were not explored further (Table S2C,D).

Within‐subject variability in subregions

In people with MS, the within‐subject variability of Z‐scores relative to HC of intrinsic SANDI measures was assessed among the thalamic subregions with evidence for volume loss. Significant differences are depicted in Figure 3A for f _is and Figure 3B for f _ec.

Relationships between intrinsic microstructural changes and subregional thalamic atrophy

We then studied the relationship between subregional thalamic SANDI measures and their normalized volumes to assess whether intrinsic SANDI measures were related to volume loss in people with MS.

Linear models predicting volume loss

Lower f _is and higher f _ec of the left medial pulvinar (p _corr = 0.018, and p _corr = 2.96 × 10⁻⁷, respectively), right medial pulvinar (p _corr = 0.014, and p _corr = 2.68 × 10⁻⁵, respectively), and right medial mediodorsal (p _corr = 0.007, and p _corr = 2.68 × 10⁻⁵, respectively) were significantly associated with lower normalized volume of the corresponding thalamic subregion (Fig. S1A,B). The f _ec of the left anteroventral and lateral geniculate were also negatively associated with corresponding normalized volumes (p _corr = 0.021, and p _corr = 1.87 × 10⁻⁴, respectively). Results are summarized in Table 2.

Table 2.

Summary of linear models involving intrinsic subregional thalamic microstructure predicting subregional thalamic volumes.

Subregion	Intrinsic factor	B (95%CI)	p‐value
Bilat. Pulvinar medial	Decreased subregional f is	0.75 (0.24; 1.27) [left] 0.85 (0.32; 1.38) [right]	0.005 0.003
	Increased subregional f ec	−1.10 (−1.42; −0.79) [left] −1.06 (−1.46; −0.66) [right]	2.69 × 10−8 5.00 × 10−6
R Mediodorsal medial	Decreased subregional f is	0.45 (0.20; 0.69)	6.77 × 10−4
	Increased subregional f ec	−0.64 (−0.89; −0.39)	7.30 × 10−6
L Anteroventral	Increased subregional f ec	−0.47 (−0.81; −0.12)	0.009
L Lateral geniculate	Increased subregional f ec	−0.40 (−0.58; −0.22)	6.79 × 10−5

Normalized volume of the following thalamic subregions could be predicted by either decreased intra‐soma signal fraction (f _is) or increased extra‐cellular signal fraction (f _ec) of the respective thalamic subregion: left and right medial pulvinar, right medial mediodorsal, and left anteroventral and lateral geniculate. Beta‐coefficients (B) with 95% confidence interval (CI) are reported with corresponding uncorrected p‐values surviving FDR correction.

Bilat., bilateral; L, left; R, right.

Relationships between extrinsic white matter disconnection and subregional thalamic atrophy

Regarding WM trajectories connecting thalamic subregions, we initially assessed the alterations in SANDI measures in the NAWM of these trajectories in people with MS compared to HC. Subsequently, SANDI measures in lesional WM were compared to NAWM by within‐subjects analysis. Next, we assessed whether the SANDI measures of the WM trajectories could predict the respective subregional thalamic volume by linear regression models. Significant associations were further investigated by incorporating both NAWM and lesional WM SANDI measures to identify the main contributors to subregional thalamic volume loss.

Group comparisons

Compared to HC, people with MS showed lower f _in, f _is, and R _s , and higher f _ec in the NAWM of trajectories connecting thalamic subregions showing MS‐related volume loss and the cortex (Table S3A–D).

Within‐subject variability

Within people with MS, lesional WM showed higher f _is, f _ec, and R _s , and lower f _in compared to the corresponding NAWM (Table S3A–D).

Linear models predicting volume loss

The lower f _in observed in the WM trajectories for all 22 thalamic subregions showing MS‐related volume loss was significantly associated with lower normalized volume of the corresponding thalamic subregions (Fig. S2). The strongest associations were seen for f _in of WM trajectories of the left and right anterior pulvinar (p _corr = 5.96 × 10⁻⁴, and p _corr = 9.23 × 10⁻⁶, respectively), left and right medial pulvinar (p _corr = 0.003, and p _corr = 1.33 × 10⁻⁴, respectively), and left medial geniculate (p _corr = 1.85 × 10⁻⁶). Results are summarized in Table 3. After FDR correction, these significant associations seemed to be driven specifically by f _in in NAWM of the thalamocortical trajectories rather than f _in of lesional WM for the left and right lateral geniculate (p _corr = 1.16 × 10⁻⁴, and p _corr = 0.024, respectively), left and right anterior pulvinar (p _corr = 0.034, and p _corr = 0.009, respectively), left medial geniculate (p _corr = 0.006), and right posterior ventral lateral and ventral posterolateral (p _corr = 0.024, and p _corr = 0.012, respectively).

Table 3.

Summary of linear models involving extrinsic subregional thalamic microstructure in white matter predicting subregional thalamic volumes.

Subregion	Extrinsic factor	B (95%CI)	p‐value
L Anteroventral	Decreased WM f in	0.26 (0.12; 0.40)	5.24 × 10−4
	Increased WM f ec	−0.33 (−0.49; −0.18)	9.89 × 10−5
R Central medial	Decreased WM f in	0.17 (0.04; 0.30)	0.010
	Increased WM f ec	−0.19 (−0.33; −0.05)	0.010
Bilat. Lateral geniculate	Decreased WM f in	0.36 (0.24; 0.43) [left] 0.31 (0.16; 0.46) [right]	1.58 × 10−8 1.46 × 10−4
	Increased WM f ec	−0.34 (−0.44; −0.23) [left] −0.24 (−0.40; −0.09) [right]	9.87 × 10−8 0.003
Bilat. Lateral posterior	Decreased WM f in	0.17 (0.04; 0.30) [left] 0.25 (0.13; 0.37) [right]	0.011 1.50 × 10−4
	Increased WM f ec	−0.19 (−0.34; −0.03) [left] −0.22 (−0.35; −0.10) [right]	0.021 7.17 × 10−4
R Mediodorsal lateral	Decreased WM f in	0.21 (0.11; 0.31)	2.27 × 10−4
	Increased WM f ec	−0.23 (−0.33; −0.13)	6.29 × 10−5
R Mediodorsal medial	Decreased WM f in	0.20 (0.10; 0.31)	4.27 × 10−4
	Increased WM f ec	−0.24 (−0.34; −0.14)	2.23 × 10−5
Bilat. Medial geniculate	Decreased WM f in	0.52 (0.33; 0.72) [left] 0.25 (0.10; 0.41) [right]	2.52 × 10−6 0.002
	Increased WM f ec	−0.58 (−0.80; −0.37) [left] −0.25 (−0.39; −0.10) [right]	6.21 × 10−6 0.002
Bilat. Medial ventral reuniens	Decreased WM f in	0.14 (0.01; 0.27) [left] 0.30 (0.13; 0.47) [right]	0.040 0.001
	Increased WM f ec	−0.28 (−0.49; −0.07) [right]	0.010
Bilat. Pulvinar anterior	Decreased WM f in	0.38 (0.19; 0.58) [left] 0.49 (0.32; 0.66) [right]	2.71 × 10−4 8.39 × 10−7
	Increased WM f ec	−0.45 (−0.68; −0.22) [left] −0.47 (−0.65; −0.28) [right]	3.95 × 10−4 6.92 × 10−6
R Pulvinar inferior	Decreased WM f in	0.21 (0.04; 0.38)	0.017
	Increased WM f ec	−0.21 (−0.38; −0.05)	0.013
Bilat. Pulvinar medial	Decreased WM f in	0.40 (0.16; 0.64) [left] 0.49 (0.28; 0.70) [right]	0.002 3.26 × 10−5
	Increased WM f ec	−0.42 (−0.70; −0.13) [left] −0.48 (−0.69;−0.27) [right]	0.006 5.11 × 10−5
Bilat. Ventral anterior	Decreased WM f in	0.16 (0.01; 0.32) [left] 0.24 (0.10; 0.38) [right]	0.042 0.002
	Increased WM f ec	‐0.27 (−0.44; −0.09) [left] −0.17 (−0.33; −0.01) [right]	0.003 0.037
R Ventral anterior magnocellular	Decreased WM f in	0.19 (0.02; 0.36)	0.027
R Ventral lateral posterior	Decreased WM f in	0.23 (0.13; 0.32)	2.03 × 10−5
	Increased WM f ec	−0.20 (−0.30; −0.09)	7.44 × 10−4
R Ventral posterolateral	Decreased WM f in	0.24 (0.14; 0.35)	3.62 × 10−5
	Increased WM f ec	−0.22 (−0.33; −0.11)	4.03 × 10−4

Normalized volume of the following thalamic subregions could be predicted by either decreased intra‐neurite signal fraction (f _in) or increased extra‐cellular signal fraction (f _ec) of the structurally connected white matter of the respective thalamic subregion: left anteroventral, left and right lateral geniculate, lateral posterior, medial geniculate reuniens medial ventral, anterior and medial pulvinar, and ventral anterior, and right central medial, lateral and medial mediodorsal, inferior pulvinar, magnocellular ventral anterior, posterior ventral lateral, and ventral posterolateral. Beta‐coefficients (B) with 95% confidence interval (CI) are reported with corresponding uncorrected p‐values surviving FDR correction.

Bilat., bilateral; L, left; R, right.

Higher f _ec observed in the WM trajectories of thalamic subregions showing MS‐related volume loss was significantly associated with lower normalized volume of the corresponding thalamic subregion (except for the left reuniens medial ventral and right magnocellular ventral anterior; Fig. S3). The strongest associations were again seen in f _ec of WM trajectories of the left and right anterior pulvinar (p _corr = 9.85 × 10⁻⁴ and p _corr = 5.07 × 10⁻⁵, respectively), left and right medial pulvinar (p _corr = 0.008 and p _corr = 2.25 × 10⁻⁴, respectively), and left medial geniculate (p _corr = 2.87 × 10⁻⁵). Results are summarized in Table 3. Again, after FDR correction, specifically the f _ec in NAWM of the thalamocortical trajectories was negatively associated with subregional thalamic volumes rather than f _ec in lesional WM for the left anteroventral (p _corr = 0.031), left lateral geniculate (p _corr = 0.001), and left medial geniculate (p _corr = 0.001).

The f _is and R _s of WM trajectories of thalamic subregions showing MS‐related volume loss did not show significant associations with subregional thalamic volumes.

Relationships between extrinsic structurally connected cortical microstructure and subregional thalamic atrophy

Then, we assessed the alterations in SANDI measures in the cortex structurally connected to the thalamic subregions in people with MS compared to HC. Next, we examined the associations between cortical SANDI measures and normalized volumes of the corresponding subregions by linear regression models.

Group comparisons

The f _is of cortical regions that were structurally connected to thalamic nuclei showing MS‐related volume loss was lower in people with MS compared to HC (Table S4A–D). The f _in, f _ec, and R _s did not differ between MS and HC.

Linear models predicting volume loss

The f _is in the cortex that was structurally connected to the left lateral posterior, left ventral anterior, and right medial mediodorsal showed significant positive associations with normalized volumes of the corresponding thalamic subregions (p _corr = 0.045 for all; Fig. S4). Results are summarized in Table 4.

Table 4.

Summary of linear models involving extrinsic subregional thalamic microstructure in structurally connected cortex predicting subregional thalamic volumes.

Subregion	Extrinsic factor	B (95%CI)	p‐value
L Lateral posterior	Decreased cortex f is	0.24 (0.07; 0.40)	0.006
L Ventral anterior	Decreased cortex f is	0.26 (0.08; 0.43)	0.006
R Mediodorsal medial	Decreased cortex f is	0.17 (0.06; 0.29)	0.005

Normalized volume of the following thalamic subregions could be predicted by decreased intra‐soma signal fraction (f _is) of the structurally connected cortical regions of the respective thalamic subregion: left lateral posterior, left ventral anterior, and right medial mediodorsal. Beta‐coefficients (B) with 95% confidence interval (CI) are reported with corresponding uncorrected p‐values surviving FDR correction.

L, left; R, right.

Contributions of intrinsic versus extrinsic microstructure to subregional thalamic atrophy

Of all thalamic subregions presenting MS‐related volume loss, the normalized volumes of the left anteroventral and lateral geniculate, bilateral medial pulvinar, and right medial mediodorsal showed associations with subregional intrinsic microstructure (either f _is or f _ec) as well as extrinsic microstructure of the corresponding WM trajectories (either f _in or f _ec) and/or of the connected cortical regions (f _is). For each of these subregions, we examined the contributions of intrinsic and extrinsic microstructure to subregional thalamic atrophy through linear regression models.

Linear models predicting volume loss

The left anteroventral subregion appears an atrophy‐prone subregion showing strong associations between extrinsic microstructural measures and thalamic volume, as its normalized volume was significantly associated with the f _ec of its connected WM trajectory (p _corr = 6.59 × 10⁻⁴). Lower normalized volume of the left lateral geniculate was associated with lower f _in of its WM trajectory as well as higher f _ec of the left lateral geniculate itself (p _corr = 3.53 × 10⁻⁶, and p _corr = 0.005, respectively). For the left medial pulvinar, the f _ec of the subregion itself (p _corr = 2.98 × 10⁻⁶) was negatively associated with normalized volume, whereas the f _ec and f _in of the WM trajectories and the f _is of the left medial pulvinar itself were not. The same held for normalized volume of the right medial pulvinar, which could only be predicted by the f _ec of the subregion itself (p _corr = 0.021). For the right medial mediodorsal, f _is of structurally connected cortex was added to the model, as this thalamic subregion also showed a significant association between its normalized volume and connected cortical regions. In the right medial mediodorsal subregion, lower f _is within the subregion itself and, to a lesser extent, lower f _is of its structurally connected cortex and lower f _in of its WM trajectory were associated with lower normalized volume (p _corr = 0.003, p _corr = 0.034, and p _corr = 0.005, respectively). Results are summarized in Table 5.

Table 5.

Summary of final linear models predicting subregional thalamic volumes.

Subregion	Extrinsic or intrinsic factor	B (95%CI)	p‐value
L Anteroventral	Increased WM f ec	−0.33 (−0.49; −0.18)	9.89 × 10−5
	Increased subregional f ec	–	–
	Decreased WM f in	–	–
R Mediodorsal medial	Decreased subregional f is	0.36 (0.16; 0.56)	0.001
	Decreased cortex f is	0.12 (0.03; 0.21)	0.014
	Decreased WM f in	0.15 (0.06; 0.24)	0.001
	Increased subregional f ec	–	–
	Increased WM f ec	–	–
Bilat. Pulvinar medial	Increased subregional f ec	−1.55 (−2.03; −1.06) [left] −0.72 (−1.24; −0.21) [right]	1.49 × 10−7 0.008
	Decreased subregional f is	–	–
	Decreased WM f in	–	–
	Increased WM f ec	–	–
L Lateral geniculate	Increased subregional f ec	−0.24 (−0.38; −0.10)	0.001
	Decreased WM f in	0.28 (0.19; 0.37)	3.53 × 10−7
	Increased WM f ec	–	–

Normalized volume of the left anteroventral subregion could only be predicted by the extra‐cellular signal fraction (f _ec) of its white matter (WM) trajectory, whereas the WM intra‐neurite signal fraction (f _in) and the subregional f _ec could not. For the right medial mediodorsal subregion, subregional f _is as well as, though to a lesser extent, of its structurally connected cortex and the WM f _in could predict normalized volumes. For the left and right medial pulvinar subregions, the subregional f _ec was significantly associated with normalized volumes, whereas the WM f _ec and f _in as well as the subregional intra‐soma signal fraction (f _is) were not. Normalized volume of the left lateral geniculate subregion was associated with the WM f _in as well as subregional f _ec. Beta‐coefficients (B) with 95% confidence interval (CI) are reported with corresponding uncorrected p‐values surviving FDR correction.

Bilat., bilateral; L, left; R, right.

Discussion

In this cross‐sectional study, we sought to parse out the intrinsic and extrinsic contributions of MS‐related pathology to thalamic neurodegeneration. We systematically examined SANDI microstructural measures within and outside the thalamus and explored their relationships to volumes of thalamic subregions. In our cohort of people with predominantly relapsing–remitting MS, normalized volumes of the posterior and medial subregions showed more severe atrophy compared to more anteriorly and ventrally located subregions. Looking at microstructural changes, cell body density was lower and extra‐cellular space was increased within thalamic subregions, with regional patterns mimicking those of neurodegeneration. The NAWM of projections to and from atrophy‐prone thalamic subregions showed lower axon density and increased extra‐cellular space, and the structurally connected cortex showed lower cell body density in association with lower subregional volumes. Several subregions demonstrated predominantly extrinsic or intrinsic microstructural contributions to thalamic volume. The left anteroventral subregion showed significant associations between increased extra‐cellular of the connected WM and lower normalized volume. On the other hand, the normalized volumes of the bilateral medial pulvinar were predicted by increased extra‐cellular space within the pulvinar subregions themselves. The heterogeneity of intrinsic and extrinsic contributions to thalamic atrophy was exemplified in the left lateral geniculate and right medial mediodorsal subregions, showing contributions of both intrinsic cell body density changes and extrinsic cell body and axon density changes in WM and connected cortex to their normalized volumes.

MS‐related volume losses of thalamic subregions were predominantly seen in subregions located in medial and posterior parts of the thalamus. Atrophy and shape changes in thalamic subregions situated immediately adjacent to the cerebrospinal fluid compared to those located relatively distant to the cerebrospinal fluid have been reported in both the healthy aging and MS brain, and support the hypothesis that cerebrospinal fluid‐mediated factors might contribute to neurodegeneration. ¹⁸ , ¹⁹ , ²⁰ Interestingly, atrophy‐prone thalamic subregions showed higher f _ec and lower f _is, reflecting an increased extra‐cellular space and greater cell body loss. Relative to their location, the reduction in f _is was most prominent in posterior subregions, particularly the lateral posterior subregions, which act in concert with the pulvinar subregions as key components of the visual pathway. ³⁶ The lower f _is observed in subregions located close to the ventricles supports the “ependymal‐in” gradient concept of thalamic damage in MS, ³⁷ analogous to the “pial‐in” cortical gradient demonstrated for cortical damage. ¹⁸ , ¹⁹ , ³⁸ , ³⁹ Subependymal neuroaxonal loss has been associated with the presence of subependymal perivascular B‐cell infiltrates and soluble inflammatory factors in the cerebrospinal fluid, damaging thalamic tissue directly or indirectly by microglia activation. ³⁷ Along with these findings from histopathology, our data provide in vivo imaging support of an “ependymal‐in” gradient of MS‐specific thalamic tissue damage, with one possible explanation suggesting it may result from intrathecal B‐cell immunity, which may help to improve early stratification of people with MS who would benefit from more personalized treatment.

With regard to extrinsic contributions of MS‐related pathology to thalamic neurodegeneration, people with MS showed lower f _is, f _in, and R _s and higher f _ec in the NAWM compared to HC, reflecting a reduction in both cell bodies (density and size) and axons. In WM lesions, higher f _is, f _ec, and R _s , and lower f _in were seen compared to NAWM. This finding is consistent with the heterogeneity of MS pathology within lesions, which encompasses active and mixed active/inactive lesions, that is, ongoing inflammation as reflected in higher R _s and f _is and demyelination as reflected in lower f _in and higher f _ec. ⁴⁰ Histopathologic correlation studies are needed to understand the precise microstructural substrates of tissue damage in MS. Lower axon density and increased extra‐cellular space in the connected WM trajectories were associated with subregional thalamic volume loss, which is supported by histopathology. ⁴¹ The stronger relationship of NAWM microstructural changes compared to lesions with thalamic volume may reflect the contribution of diffuse WM changes that predominate within WM trajectories to and from the thalamus and suggest that axonal damage is a strong extrinsic driver of thalamic atrophy, as summarized in Table 3.

The f _is of connected cortical regions was significantly lower in people with MS compared to HC. Reductions in f _is were significantly associated with normalized volumes of the lateral posterior, ventral anterior, and medial mediodorsal subregions. Cortical microstructural alterations as reflected by the SANDI measures were heterogeneous and nonuniform in space, possibly due to focal demyelination, which occurs most frequently in the motor, cingulate and frontal cortices, and temporal pole. ⁴² These regions, among others, are connected to the lateral posterior, ventral anterior, and medial mediodorsal subregions. ⁴³ Diffuse neurodegeneration, which develops in a nonrandom distribution and follows distinct anatomical patterns, ⁴ may also contribute to the observed relationships between loss of f _is in specific cortical regions and the connected thalamic subregions. The variation of SANDI measures throughout the cortex and their modulation by focal demyelination merit investigation in future studies.

Linear models linking thalamic volume loss with intrinsic and extrinsic microstructural metrics revealed that subregions located in specific areas of the thalamus appeared differentially susceptible to neurodegeneration, which may be explained by several mechanisms. ⁷ The pulvinar and mediodorsal subregions showed predominantly intrinsic contributions to neurodegeneration, with the thalamus itself serving as the primary site of injury, ⁷ potentially driven by the “surface‐in” gradient concept mentioned above. ¹⁸ , ¹⁹ , ³⁸ , ³⁹ The medial mediodorsal subregion also showed contributions from extrinsic microstructural alterations in the cortex and connected WM. The anteriorly located subregions showed predominantly extrinsic contributions arising from loss of f _in in connected WM, raising the possibility of both retrograde and anterograde degeneration due to axonal transection as major contributors of the subregional thalamic degeneration observed here. ⁴¹ , ⁴⁴ , ⁴⁵ , ⁴⁶ In previous histopathological work, thalamic demyelination appeared to not correlate with thalamic volume. ¹⁰ Our observation of intrinsic microstructural alterations in these subregions, when compared to HC, in combination with the lack of an association with their respective volumes further underscores the role of extrinsic tissue damage in contributing to thalamic neurodegeneration. Also, cognitive impairment, for which the cortex acts as major locus, has been linked specifically to localized volume loss of anterior and superior thalamic subregions, suggesting a shared pathological mechanism that is supported by our findings. ¹⁸ , ¹⁹ Interestingly, volumes of the lateral geniculate subregion appeared driven by a combination of intrinsic and extrinsic microstructural factors. The lateral geniculate subregion is a key region in MS due to its central role in vision and is often used as region of interest to study trans‐synaptic degeneration in the optic pathway. ⁴⁷ , ⁴⁸ , ⁴⁹ This subregion would be an interesting target to evaluate the temporal associations between microstructural contributors to thalamic neurodegeneration in combination with measures of visual function. In summary, the relative contributions of intrinsic and extrinsic microstructural factors to tissue damage in the context of subregional thalamic volume loss depend on the subregions' function and location.

Limitations of this study include its cross‐sectional nature, which makes it difficult to infer causal and temporal relationships. The diffusion MRI acquisitions were limited in spatial resolution, and partial volume effects may have biased results. To minimize potential confounding by partial volume averaging effects in our analyses, we calculated partial volume‐weighted averages of the regions of interest. Differences in f _ec reflecting extra‐cellular space between MS and HC were not greater for subregions neighboring the ventricles compared to subregions more laterally positioned, indicating that our disease‐specific results were unlikely to be driven purely by greater partial volume effects in MS participants. The resolution of the diffusion MRI data did not allow us to obtain more detailed spatial information regarding microstructure in small thalamic subregions, which is why we had to exclude three subregions. Among the remaining subregions, the variability of microstructural SANDI measures was higher in smaller (e.g., magnocellular ventral anterior) compared to larger subregions (e.g., medial pulvinar), diminishing potential differential effects between HC and MS. In addition, it is possible that the segmentation of the thalamus using THALAMODEL may have more accurately determined the external boundaries of thalamus rather than inter‐subregion boundaries, thereby potentially introducing bias for which subregions' volumes correlate with various microstructural outcomes. Also, we did not take into account focal thalamic lesions, characterized by demyelination with varying degrees of inflammatory changes, ⁵⁰ when considering the individual subregions because thalamic lesions were difficult to segment based on our acquired sequences. Prior work, though, did not find a relationship between focal atrophy and thalamic lesions. ¹⁸ , ⁵¹ Finally, due to the cross‐sectional design of our study, we cannot draw conclusions regarding the order of pathological events driving subregional thalamic volume loss, which are being addressed in ongoing longitudinal studies.

In summary, our study supports the heterogeneous contribution of extrinsic and intrinsic tissue damage, potentially reflective of the involvement of both anterograde and retrograde degeneration from WM demyelination and cerebrospinal fluid‐mediated damage to subregional thalamic volume loss. The microstructural alterations assessed by a high‐gradient diffusion MR approach may reveal in vivo biomarkers for early assessment of the efficacy of neuroprotective therapies, which modulate a combination of pathophysiological mechanisms resulting in WM demyelination and cerebrospinal fluid‐mediated damage.

Author Contributions

E.A.K. contributed to the conception and design of the study, acquisition, and analysis of data, and drafting the manuscript and figures. E.S.K, A.W.R, H.L., and F.L.C. contributed to the acquisition and analysis of data and drafting the manuscript. M.M.S. contributed to drafting the manuscript. S.Y.H. and E.C.K. contributed to the conception and design of the study, acquisition, and drafting the manuscript and figures.

Conflict of Interest

E.A.K., E.S.K., A.W.R., H.L., and F.L.C. report no conflicts of interest; M.M.S. serves on the editorial board of Neurology and Frontiers in Neurology and Multiple Sclerosis Journal, receives research support from the Dutch MS Research Foundation, Eurostars‐EUREKA, ARSEP, Amsterdam Neuroscience, MAGNIMS and ZonMW and has served as a consultant for or received research support from EIP Pharma, Atara Biotherapeutics, Biogen, Celgene/Bristol Meyers Squibb, Genzyme, MedDay and Merck; S.Y.H. has received consulting fees and research grants from Siemens Healthineers; E.C.K. has received consulting fees from EMD Serono, Genentech, INmune Bio, Myrobalan Therapeutics, OM1, and TG Therapeutics, and received research funds from Abbvie, Biogen, and Genentech.

Supporting information

Table S1. Normalized volumes of thalamic subregions in included participants.

Table S2. Intra‐soma (A), extra‐cellular (B) and intra‐neurite (C) signal fraction and apparent soma radius (D) in thalamic subregions showing MS‐related atrophy.

Table S3. Intra‐soma (A), extra‐cellular (B) and intra‐neurite (C) signal fraction and apparent soma radius (D) in white matter trajectories connecting thalamic subregions showing MS‐related atrophy and the cortex.

Table S4. Intra‐soma (A), extra‐cellular (B) and intra‐neurite (C) signal fraction and apparent soma radius (D) in cortical regions connected to thalamic subregions showing MS‐related atrophy.

Figure S1. Subregional thalamic volume and microstructure.

Figure S2. Subregional thalamic volume and intra‐neurite signal fraction in WM.

Figure S3. Subregional thalamic volume and extra‐cellular signal fraction in WM.

Figure S4. Subregional thalamic volume and intra‐soma signal fraction in structurally connected cortical regions.

Funding Statement

This work was funded by National Institutes of Health grants K23‐NS096056, P41‐EB015896, P41‐EB030006, R01‐NS118187, and U01‐EB026996; McCourt Foundation; Dutch MS Research Foundation and Fund Girn managed by the King Baudouin Foundation.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author, upon reasonable request.