The Brain Beneath the Label |
Schizophrenia risk may reflect two brain pathways with different cortical and subcortical balance.
Early language learning relies more on subcortical systems; later learning engages cortical control.
Genetic studies suggest schizophrenia includes distinct pathways with different brain system balance.
John Nash won the Nobel Prize in Economic Sciences in 1994. By then, his most important mathematical work was decades behind him, his psychosis had largely receded, and he had spent years doing mathematics at Princeton in what most clinicians would call partial remission. His story, immortalized in A Beautiful Mind, struck millions as a triumph of resilience.
It struck Michael Halassa, a neuroscientist and psychiatrist at Tufts, as a puzzle.
Halassa writes in the context of a broader research program he calls algorithmic psychiatry, which argues that mental illness is best understood at the level of how the brain builds and updates models of the world. His recent essay describes the dissonance between Nash's trajectory and the clinical picture that dominates psychiatric training: patients who arrive at their first psychotic break already cognitively compromised, who decline further despite adequate treatment, and who never return to baseline. Did Nash really have schizophrenia? For Halassa, the question is less a diagnostic quibble than an invitation to look more carefully at what computational processes, and what developmental conditions, lie beneath that label.
It was Halassa’s new essay that sent me there. When I read it, it introduced me to a paper by Watson and colleagues in Molecular Psychiatry, one that made me return to work of my own that had always felt like it belonged to a different conversation.
One Label, Two Biological Stories
The Watson et al. paper begins with a longstanding paradox. Schizophrenia is associated clinically with lower educational attainment, yet the aggregate genetic signal showed almost no relationship with education, or even a slight positive one. The same variants that increased schizophrenia risk seemed, at the population level, to be associated with staying in school longer.
The resolution: two distinct genetic components were canceling each other out. Using genomic structural equation modeling, Watson and colleagues formally separated them. One component—SZspecific—correlates negatively with both IQ and educational attainment. A second, shared with bipolar disorder, correlates positively with educational attainment and points toward genes governing synaptic signaling. Brain expression analyses found the shared component most active in cortical regions, particularly frontal cortex, while the schizophrenia-specific component extends further into subcortical regions including the caudate and hippocampus. The difference is not a clean cortical-versus-subcortical split—it is a difference in degree, one component more cortically dominant, the other reaching further subcortically.
What we call schizophrenia appears to compress patients with meaningfully different underlying biology into a single label. Halassa’s puzzle about Nash begins to dissolve: if patients vary continuously in the proportions of these two components they carry, Nash’s late-arriving, episodic psychosis is not a paradox. It is exactly what you would expect somewhere in that distribution.
Reading Watson’s paper, I felt an odd sense of recognition. The architecture they were describing—a tension between cortical and subcortical systems emerging across development—resembled something I had encountered in a completely different domain: language learning.
The Same Architecture, a Different Outcome
In our work on bilingualism, we have been asking a related question: how does the balance between cortical and subcortical systems shift as the brain learns language?
In 2007, Ping Li and I proposed the Sensorimotor Hypothesis: the neural systems involved in language learning depend critically on when that learning begins. Early acquisition is organized subcortically, with the basal ganglia playing a central role. Children learn through sensorimotor engagement and procedural memory. Later acquisition shifts toward cortical systems: attention, working memory, and executive control. The dopamine system mediates that transition.
In 2018, Kelly Vaughn and I published a study examining whether genetic variation in dopamine functioning could predict bilingual proficiency in Spanish-English bilinguals. Two variants, one influencing subcortical dopamine in the striatum, the other influencing prefrontal dopamine levels, revealed a striking three-way interaction with age of acquisition. For early second language learners, higher subcortical dopamine predicted the highest proficiency. For late second language learners, balanced cortical dopamine, neither too stable nor too flexible, predicted the best outcomes. And crucially, the highest scores were balanced scores: genuine fluency in both languages without one dominating the other.
When we published those findings, they felt oddly homeless. Dopamine belonged to a different conversation—schizophrenia, psychiatric risk, clinical populations. Watson et al. do not invoke dopamine directly; their characterization focuses on synaptic signaling and remains agnostic about mechanism for the schizophrenia-specific component. But the pattern their brain expression analyses reveal—one component more anchored in frontal cortex, the other extending subcortically—is consistent with the developmental architecture our work describes. The inference is ours, not theirs. What their findings make possible is the question: could the same dopamine-mediated shift from subcortical to cortical processing that shapes language trajectories also differentiate these two genetic pathways to psychosis?
What the Developmental Moment Determines
Becoming balanced in two languages is not simply a matter of exposure or effort. It reflects a developing brain meeting the right input at the right moment, with the right neurochemical conditions in place. The same logic may apply to psychosis: what a genetic liability produces depends on when in development it expresses itself, and in which systems.
This is what the label “schizophrenia” has been concealing—not just clinical heterogeneity, but a fundamental difference in developmental timing. Watson et al. provide a genetic probe for that difference. Our work on bilingualism suggests the underlying architecture may be broader than psychiatry has recognized.
None of this resolves what John Nash had. But it reframes the question. The more productive ask is not which diagnostic box he belonged in, but which biological process, expressing at which developmental moment, shaped his mind’s trajectory. The answers illuminate not only the brains that fracture, but the ones that learn to hold two worlds at once.
Halassa, M. M., Frank, M. J., Garety, P., Ongur, D., Airan, R. D., Sanacora, G., Dzirasa, K., Suresh, S., Fitzpatrick, S. M., & Rothman, D. L. (2025). Developing algorithmic psychiatry via multi-level spanning computational models. Cell reports. Medicine, 6(5), 102094. https://doi.org/10.1016/j.xcrm.2025.102094
Vaughn, K. A., & Hernandez, A. E. (2018). Becoming a balanced, proficient bilingual: Predictions from age of acquisition & genetic background. Journal of neurolinguistics, 46, 69-77. https://doi.org/https://doi.org/10.1016/j.jneuroling.2017.12.012
Watson, C.J., Zvrskovec, J., Merola, G.P. et al. Splitting schizophrenia: divergent cognitive and educational outcomes revealed by genomic structural equation modelling. Mol Psychiatry (2026). https://doi.org/10.1038/s41380-026-03444-3