Cause, Consequence, or Compass? The Thalamus in Temporal Lobe Epilepsy Surgery Authors: John D. Rolston Published: 2026 ABSTRACT Thalamic organization differentially contributes to clinical conditions in epilepsy. Communications Medicine in press Zhang Q, Javidi SS, Ankeeta A, Sperling MR, Zhang Z, Tracy JI. Communications Medicine in press.Background: The thalamus plays an important role in key clinical conditions of focal temporal lobe epilepsy (TLE), but no investigation has examined whether the same thalamic local and connectome properties shape a patient's status across these different conditions. Methods: This retrospective longitudinal MRI study used resting-state fMRI and structural imaging to identify whole-brain focal/regional and connectome-level features associated with 6 binary clinical conditions of TLE (pre-/postsurgery, seizure and neurocognitive outcomes, pathology, seizure subtype, and SOZ lateralization) in 91 patients across 2 centers (age range: 15-65 years) and 85 matched healthy participants (age range: 18-60 years). Results: Across conditions, relative to all other brain regions, thalamic features exert the strongest influence. Specifically, thalamic focal and connectome intrinsic activity and gray matter volume are robustly associated with postsurgical reorganization. Presurgical thalamic hyperconnectivity predicts poorer seizure control, whereas postsurgical reorganization is not associated with either seizure outcome. Neuropsychological outcomes are subsequently examined and show associations with local ipsilateral thalamic properties. Namely, presurgical organization of the ipsilateral thalamus is associated with better preservation of cognitive performance, whereas postsurgical organization is associated with greater cognitive decline. Conclusions: Our results expand and refine our understanding of the thalamus as a region showing robust and recurrent associations across multiple clinical conditions of TLE. Importantly, we distinguish its role in pre- versus postsurgical brain organization with respect to seizure and neuropsychological outcomes, highlighting its importance for planning and prognosis in epilepsy surgery. Commentary The thalamus is now recognized as both a structural casualty and a therapeutic target in temporal lobe epilepsy (TLE), but its role as a network-level driver of the surgical phenotype has been less systematically interrogated. ENIGMA structural surveys, SEEG recordings, Bertram's extratemporal circuit work, and functional imaging have converged on the idea that thalamocortical embedding shapes both seizure expression and surgical response.1–6 Against this backdrop, Zhang and colleagues offer an innovative coordinated test of how broadly thalamic influence reaches across the clinical phenotypes that matter for surgical decision-making. 7 Across 91 patients with unilateral TLE from 2 centers (Thomas Jefferson in the United States and Jinling Hospital in China), plus matched longitudinally scanned healthy controls, the authors computed nine local and connectome resting-state metrics across 116 brain regions, then asked which regions showed the largest effects across 6 clinical contrasts: presurgical organization, postsurgical reorganization, seizure outcome, seizure subtype, pathology, and lateralization. Through this analysis, they found that the ipsilateral thalamus—not temporal cortex or mesial structures—emerged as the most consistent contributor across all 6 contrasts. Perhaps surprisingly, given the current surgical focus on the anterior, centromedian, and pulvinar nuclei, the mediodorsal nucleus carried most of the signal—a target none of the current neuromodulation programs have prioritized, but whose limbic and prefrontal connectivity would make it a coherent candidate. The core thalamic findings replicated in the independent Jinling cohort, though it is worth noting that this validation sample was restricted to hippocampal sclerosis, while the Jefferson discovery cohort included a broader pathological spectrum. The more provocative findings arise as they dive deeper into individual metrics. While presurgical thalamic connectome features (degree centrality and modal controllability) tracked with seizure outcome, in line with prior research, the authors unexpectedly found that a pair of local thalamic features (amplitude of low frequency fluctuations, ALFF, and gray matter volume, GMV) relate to neuropsychological outcomes in opposite directions before and after surgery. Presurgically, higher ipsilateral thalamic ALFF activity and preserved volume track with better cognitive preservation, whereas postsurgically, the same profile tracks with greater decline. This temporal change is hard to reconcile. It may be that increased low-frequency activity reflects deafferentation of the thalamus, but the association of higher GMV may be related to their methodology. Some of the apparent reversals reflect a quiet shift in the definition of these variables: the presurgical predictor is a static level, while the postsurgical predictor is a within-patient change score, and these are not the same quantity even when computed from the same metric. They also use a partial least squares structural equation model, which combines ALFF and GMV into a shared latent profile, despite the observation that the 2 share little variance. Their relative weight within the composite shifts between pre- and postsurgery models. The interpretation that “the local profile” relates to cognition, therefore, obscures whether the signal is being carried by intrinsic activity, by structural integrity, or by something different at the two timepoints. Three other caveats deserve discussion. First, the headline that the thalamus “wins” across all conditions warrants scrutiny. The Automated Anatomical Labeling atlas that the authors use to parcellate the brain treats the thalamus as a single bilateral parcel, while heavily subdividing the cortex. Larger parcels with broader anatomical connectivity are systematically favored in whole-brain rankings of mean effect size. While the thalamus is almost certainly a hub, the magnitude of its supremacy may be in part cartographic. Second, while they document both postsurgical changes in the thalamus and neurocognitive changes, the presence of both does not mean that the thalamic changes drive the cognitive changes, as the article implies. The observed thalamic changes are most parsimoniously explained by disconnection from the resected temporal lobe rather than by active reorganization. The mediodorsal nucleus, specifically, is connected to the amygdala, entorhinal, and perirhinal cortices. Resecting these inputs would be expected to produce the observed pattern of decreased thalamic GMV, consistent with anterograde degeneration or diaschisis, alongside altered intrinsic activity. 8 That these thalamic features predict postoperative cognitive decline may simply reflect that more disconnection produces both more thalamic change and more cognitive loss, with neither causing the other. Third, the predictive performance is more modest than the framing suggests. The seizure outcome classification achieved an area under the curve of approximately 0.78 to 0.80 in-sample with 5-fold cross-validation on 54 patients, and the independent cohort unfortunately lacks Engel outcomes for external validation. More importantly, the reported absence of a stage-by-outcome interaction—good- and poor-outcome groups differed at baseline but did not diverge across surgery—indicates that the thalamic signal is better characterized as a marker of preexisting epileptogenic distribution than as a prognostic biomarker in the actionable sense. This is a useful refinement of the earlier hubness finding, 4 but it does not yet rise to the level of a clinical decision tool. For example, it could reflect that some of these patients have networks wider than the diagnosed TLE. For the epilepsy surgeon, the practical question is how to use these results. The predictive ceiling here is below what would warrant denying surgery to a candidate who otherwise meets criteria, and the postsurgical signal, by definition, cannot inform decisions made beforehand. The cognitive finding is more clinically suggestive—if a patient's preoperative thalamic ALFF and GMV profile genuinely flags vulnerability to language and verbal memory decline, that information could one day inform the choice between resection and minimally invasive ablation and presurgical counseling. 9 But this requires prospective validation in cohorts with standardized neuropsychological batteries and known dominance, and ideally tract-resolved disconnection mapping to disentangle thalamic cause from thalamic consequence. The more promising translational direction, suggested but not pursued by these data, is mechanistic. If the mediodorsal nucleus is genuinely central to TLE network expression—and Zhang and colleagues join several converging lines of evidence that it is—the obvious next step is to ask how thalamic neuromodulation of this nucleus may be useful. Pairing these imaging biomarkers with intracranial recordings from the mediodorsal nucleus and, if warranted, prospective trials of mediodorsal neuromodulation, would test whether the correlations identified here translate into mechanism and intervention. Zhang and colleagues’ contribution is methodological as much as empirical: by interrogating multiple imaging metrics across multiple coupled clinical conditions within a single cohort, they show how much can be missed by the usual one-metric, one-endpoint design. The harder work—turning these associations into decisions about who gets which intervention, and where in the thalamus to deliver it—remains ahead.