The pulvinar is no ordinary relay. While early textbook accounts cast the thalamus as a simple switching station funneling sensory signals to the cortex, the pulvinar sits at the top of the hierarchy as a higher-order hub, coordinating communication between distant cortical networks. It is densely interconnected with visual, attentional, and association cortices, and it has been implicated in attentional filtering, visual salience assignment, and the integration of information across brain-wide networks. Because cortico-cortical pathways in schizophrenia have repeatedly shown disruption in imaging studies, and because the pulvinar orchestrates precisely such long-range coordination, the research team led by Yuanjun Xie and Qiang Hu reasoned that individualized deviations in pulvinar-cortical connectivity might carry a signature of clinical heterogeneity that conventional group-average analyses wash out.

To test that idea, the team assembled resting-state functional MRI data from 746 healthy controls and 387 patients with schizophrenia drawn from multiple scanning sites—a scale that matters, because site-specific differences in scanners and protocols can masquerade as biological signals if not carefully handled. Rather than comparing patients to controls as groups, the researchers built a normative model from the healthy reference data. Normative modeling is a technique borrowed in spirit from growth charts in pediatrics: instead of asking whether a group differs on average, it estimates, for each individual, how far that person’s brain features deviate from the expected range given the healthy population. Each patient thus received an individualized deviation profile describing where and how strongly their pulvinar-cortical functional connections strayed from the norm.

The next challenge was dimensional reduction guided by clinical relevance. Raw connectivity deviation maps contain thousands of features, far too many to cluster meaningfully without the solution being dominated by noise. The team therefore applied a symptom-guided feature selection strategy, retaining the connections whose deviations tracked symptom measures, and then ran unsupervised clustering within this clinically enriched, low-dimensional feature space. Unsupervised clustering is deliberately blind to diagnosis labels; it simply asks whether patients fall into natural subgroups based on their brain deviation patterns. The answer was yes: two partially separable biotypes emerged, and critically, they were not arbitrary mathematical artifacts but carried clinical meaning.

That meaning was anchored most strongly in negative symptom severity. The two biotypes differed significantly on this clinical dimension, and the difference survived statistical adjustment for scanning site and other available covariates—a crucial robustness check in multisite psychiatry research, where demographic and technical confounds can easily manufacture spurious subgroups. In other words, the brain-based split corresponded to a real, measurable difference in how ill patients were in the negative-symptom domain, the domain that most strongly predicts long-term functional outcome and that has proven most resistant to existing antipsychotic medications.

Where in the brain did the symptom-relevant deviations live? The connections that drove the biotype separation showed a structured cortical distribution, concentrated predominantly in the somatomotor, visual, and ventral attention networks. This distribution is intriguing on several counts. The somatomotor network’s involvement resonates with the well-documented motor abnormalities in psychosis, including soft neurological signs and psychomotor slowing that often precede overt illness. The visual network’s prominence fits with the pulvinar’s role as the principal higher-order nucleus of the visual system, and the ventral attention network’s contribution aligns with the salience-processing disturbances long proposed to underlie psychotic experience. Together, the pattern suggests that schizophrenia’s heterogeneity is written, at least in part, into how the pulvinar dialogues with specific cortical systems rather than being a diffuse, nonspecific connectivity decline.

To add molecular context to the circuit-level findings, the team performed imaging-transcriptomic analyses, which link spatial patterns of brain deviation to the spatial expression maps of genes across the cortex. These analyses were explicitly exploratory, but they yielded coherent profiles. Biotype 1 was associated with enrichment of biological processes involving translation, protein targeting, and synaptic organization—pathways pointing toward the machinery that builds and maintains synapses. Biotype 2 showed a distinct profile involving neurodevelopment, synaptic remodeling, and metal-ion homeostasis, hinting at altered developmental and plasticity programs and at the regulation of ions such as zinc and copper that shape synaptic function. The two molecular fingerprints, like the two connectivity patterns, were related but distinguishable, consistent with the notion of partially separable biologies under one diagnostic umbrella.

The researchers also stress-tested their solution. Site-exclusion analyses, in which the clustering was repeated while leaving out individual scanning sites, indicated that the biotype solution was partially stable, although the degree of clinical separation varied from site to site. That variation is an honest caveat: it signals that while the framework captures something real, its sensitivity is not yet uniform across acquisition environments, and multisite harmonization remains a live challenge for connectomic psychiatry.

Perhaps the most clinically provocative—and most cautiously framed—result came from a projection into an independent cohort of patients receiving repetitive transcranial magnetic stimulation, a noninvasive brain stimulation treatment sometimes used for resistant symptoms. When the biotype assignments derived from the pulvinar-cortical deviation framework were applied to this treatment cohort, the researchers observed nominal, uncorrected differences in negative symptom improvement at two weeks between biotypes. However, those differences were not sustained at the four-week follow-up, and the authors are explicit that the finding should be considered hypothesis-generating rather than evidence that the biotypes predict treatment response. Still, the exercise sketches a tantalizing path forward: if brain-defined subgroups could eventually be shown to respond differently to specific interventions, biotyping could move from descriptive science toward personalized treatment selection.

The study’s authors are careful about what their work does and does not establish. The findings support a symptom-guided, pulvinar-centered framework for characterizing schizophrenia’s heterogeneity, but independent replication using prespecified models is required before the biotypes can be considered clinically or biologically valid. That caution is warranted in a field where data-driven subtyping has a history of solutions that fail to travel across datasets. Yet the combination of a large sample, an individualized normative approach, clinical anchoring, covariate-adjusted robustness, molecular context, and a treatment-cohort projection represents an unusually complete methodological pipeline. If the two pulvinar-cortical biotypes hold up under independent replication, clinicians may one day supplement a symptom-based diagnosis with a circuit-based one—distinguishing, at the level of thalamocortical wiring, the different illnesses that today share a single name.

Subject of Research: Pulvinar-cortical functional connectivity deviations defining symptom-relevant schizophrenia biotypes

Article Title: Pulvinar-cortical connectomic deviations identify symptom-relevant biotypes in schizophrenia

Article References: Xie, Y., Liu, W., Guan, M., Yang, R., Wu, D., Ma, H., Wang, Z., & Hu, Q. (2026). Pulvinar-cortical connectomic deviations identify symptom-relevant biotypes in schizophrenia. Translational Psychiatry. https://doi.org/10.1038/s41398-026-04495-4

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04495-4

Keywords: schizophrenia, pulvinar, thalamus, connectomics, normative modeling, resting-state fMRI, negative symptoms, biotypes, imaging transcriptomics, transcranial magnetic stimulation, brain networks, psychiatry