Lesion-Independent Associations Between Thalamic Nuclei Volumes and Information Processing Speed in Multiple Sclerosis

Background: Cognitive impairment in multiple sclerosis (MS) is driven by both focal inflammation and compartmentalized neurodegeneration, yet the relative effect of lesion-independent thalamic atrophy on information processing speed (IPS) remains unclear. Methods: This retrospective cohort study included 100 participants with MS. Automatic segmentation techniques quantified lesion load and delineated 26 thalamic regions of interest (ROIs). Linear models compared associations between ROI volumes and Symbol Digit Modalities Test (SDMT) performance in lesion-adjusted and unadjusted models. Results: Twenty-one of 26 ROIs showed significant SDMT associations before lesion adjustment; twelve remained significant after adjustment. Lesion-independent associations were observed in the global thalamus, sensory relay nuclei (ventral posterolateral, medial and lateral geniculate), and associative hubs (pulvinar and mediodorsal-parafascicular complex). These processing-associated ROIs exhibited significantly lower lesion-mediated effects (13.4%) than those losing significance after adjustment (34.2%, p < 0.001). Conclusion: Our findings suggest that IPS impairment reflects heterogeneous contributions from focal lesion-driven and chronic neurodegenerative pathology, with nucleus-specific phenotyping potentially informing identification of higher risk individuals.

💡 Research Summary

This retrospective cohort study examined how thalamic nuclei volumes relate to information‑processing speed (IPS) in multiple sclerosis (MS), independent of white‑matter lesion burden. One hundred adult MS patients (mean age 46.2 years, mean disease duration 11 years, median EDSS 2.0) underwent 3 T MRI with high‑resolution T1‑MPRAGE and 3D FLAIR sequences. Lesion segmentation was performed automatically using the LST‑AI toolbox, which employs an ensemble of three 3‑D U‑Net convolutional neural networks trained on a large MS dataset. Total lesion volume was log‑transformed to reduce skewness. Thalamic nuclei were segmented with the HIPS‑THOMAS pipeline, yielding 13 regions per hemisphere: a global thalamus mask, nine major nuclei (anteroventral, ventral anterior, ventral lateral anterior, ventral lateral posterior, ventral posterolateral, central medial, pulvinar, lateral geniculate, medial geniculate), two perithalamic structures (habenula, mammillothalamic tract), and a combined mediodorsal‑parafascicular (MD‑Pf) complex. All volumes were normalized to intracranial volume (ICV) and Z‑scored across participants.

Statistical analysis used two sets of analysis‑of‑covariance (ANCOVA) models. In the first set, each nucleus volume served as the independent variable, with Symbol Digit Modalities Test (SDMT) score as the dependent variable, adjusting for age, sex, and years of education. The second set added total lesion load as an additional covariate to isolate lesion‑independent effects. Multiple comparisons across the 26 nuclei (left and right hemispheres) were controlled with the Benjamini‑Hochberg false‑discovery‑rate (FDR) procedure, with FDR‑adjusted p < 0.05 considered significant.

In the unadjusted models, 21 of the 26 nuclei showed significant positive associations with SDMT performance, indicating that larger thalamic volumes correspond to faster information processing. The strongest effects (p < 0.001) were observed for the global thalamus, pulvinar, ventral posterolateral (VPL), medial geniculate (MGN), lateral geniculate (LGN), and the MD‑Pf complex. After controlling for lesion load, twelve nuclei retained significance: the global thalamus, VPL, pulvinar, MGN, LGN, MD‑Pf, and a few additional nuclei (e.g., ventral anterior, ventral lateral anterior). Notably, standardized regression coefficients (β) were larger in the left hemisphere, especially for the pulvinar (β ≈ 4.79 left vs. 3.49 right), suggesting a modest left‑dominant contribution consistent with the left hemisphere’s role in symbolic mapping and rule‑based attention.

To differentiate lesion‑mediated from lesion‑independent contributions, the authors performed mediation analyses and examined the change in explained variance (ΔR²) when lesion load was added. For the twelve “processing‑associated” nuclei that remained significant after adjustment, lesion load mediated only 13.4 % of the volume‑SDMT relationship, and added 2.4 % to the model’s R². In contrast, the nine “tract‑mediated” nuclei that lost significance after adjustment showed a larger mediation proportion (34.2 %) and a ΔR² of 5.5 % (both p < 0.01). These findings indicate that focal lesion burden drives the association between tract‑mediated nuclei and IPS, whereas processing‑associated nuclei reflect a more intrinsic, neurodegenerative process.

The discussion interprets these results in the context of thalamocortical circuitry. First‑order sensory relay nuclei (VPL, LGN, MGN) continuously transmit somatosensory and visual information, imposing high metabolic demands and making them vulnerable to chronic mitochondrial dysfunction, oxidative stress, and microglial activation. Higher‑order associative nuclei (pulvinar, MD‑Pf) sit at the hub of long‑range cortico‑thalamic loops that support attentional selection, visual‑perceptual integration, and rapid sensorimotor coordination—functions directly probed by the SDMT. Their lesion‑independent atrophy likely reflects smouldering neurodegeneration rather than downstream effects of acute inflammation. Conversely, nuclei linked primarily to motor relay (ventral anterior, ventral lateral anterior/posterior) and limbic regulation (anteroventral, central medial) depend heavily on white‑matter tracts; thus, focal lesion‑induced Wallerian degeneration appears to dominate their volume‑cognition relationship.

Clinically, the study suggests that nucleus‑specific thalamic atrophy may serve as a more sensitive biomarker of cognitive risk than global thalamic volume. Automated segmentation pipelines now make such measures feasible for routine clinical use, potentially enabling early identification of patients at heightened risk for cognitive decline and providing outcome metrics for neuroprotective trials. The authors acknowledge limitations: cross‑sectional design precludes causal inference; the cohort is single‑center, predominantly mildly disabled, and the cognitive assessment is limited to the SDMT. Moreover, lesion burden was quantified as total brain lesion volume rather than region‑specific or tract‑specific load, which could mask more nuanced relationships.

In summary, the paper demonstrates that in MS, atrophy of specific thalamic nuclei—particularly sensory relay and associative hubs—correlates with slowed information processing independently of white‑matter lesion load. This lesion‑independent thalamic degeneration likely represents a chronic, compartmentalized neurodegenerative process that contributes directly to cognitive impairment. These findings refine our understanding of the neuroanatomical substrates of MS‑related cognitive decline and point toward nucleus‑targeted imaging biomarkers for future therapeutic monitoring.