Researchers at the University of Illinois Urbana-Champaign have traced a genetic mutation found in psychosis patients to a specific breakdown in the brain’s learning machinery, offering one of the clearest links yet between a single gene variant and the synaptic failures thought to drive schizophrenia symptoms. The study, published in Molecular Psychiatry, shows that an increased copy number of the glycine decarboxylase gene, known as GLDC, depletes a chemical the brain needs to form new memories and update beliefs. That finding now sits alongside a growing body of work connecting faulty learning circuits to hallucinations, cognitive rigidity, and other hallmarks of the disorder.
How a Single Gene Starves a Key Brain Region
The research began with two patients at McLean Hospital who carried extra copies of the GLDC gene and had been diagnosed with psychosis. To test whether the genetic change was more than a coincidence, the team engineered mice carrying the same GLDC variant and tracked what happened in the dentate gyrus, a hippocampal subregion essential for pattern separation and new learning.
The results were direct. Extra GLDC copies boosted the enzyme that breaks down glycine, slashing extracellular glycine levels in the dentate gyrus. Glycine is a co-agonist of NMDA receptors, which act as molecular gatekeepers for long-term potentiation, or LTP, the process by which synapses strengthen during learning. With less glycine available, LTP in the dentate gyrus was suppressed. In the engineered animals, that synaptic deficit translated into measurable behavioral changes that resembled negative and cognitive symptoms seen in human psychosis.
In a university news release, co-author Uwe Rudolph noted that the mutation led to reduced social interaction and other abnormalities in the mouse model. First author Maltesh Kambali and colleagues used a combination of molecular assays, electrophysiology, and behavioral testing to bridge the gap between human genetics and mechanistic neuroscience, showing how a single copy-number increase can propagate from DNA to synapses to complex behavior.
One limitation deserves attention: direct measurements of glycine levels in the human dentate gyrus do not yet exist. The chain of evidence runs from patient genotyping through an animal model to electrophysiology, not from living human brain tissue. That gap means the findings are suggestive rather than conclusive for people, and future imaging or cerebrospinal fluid studies will need to close it. Still, by tying a defined genetic lesion to a specific hippocampal computation, the work strengthens the case that disrupted plasticity is a core feature of schizophrenia rather than a side effect of chronic illness or medication.
Dopamine, Hallucinations, and Broken Prediction Signals
The GLDC study addresses the hippocampal side of schizophrenia’s learning deficit, but a parallel line of research points to a second circuit where learning goes wrong: the corticostriatal loops that rely on dopamine to signal prediction errors. A learning-based circuit model published separately proposes that excess dopamine in these loops generates hallucination-like biases by distorting how the brain weighs sensory evidence against prior expectations. In that framework, dopamine no longer simply encodes surprise; instead, it exaggerates the influence of prior beliefs, so that ambiguous or noisy inputs are interpreted as confirming what the brain already expects.
That model was grounded in converging data. In rodents, direct recordings from dopamine neurons and optogenetic stimulation showed that artificially boosting dopamine at key moments pushed animals to rely more heavily on prior cues, even when those cues were misleading. In humans, behavioral tasks that pitted expectations against sensory evidence revealed similar shifts toward prior-driven perception in participants with higher dopaminergic tone. Together, these findings suggest that dopamine dysregulation can be understood as a miscalibration of learning signals rather than a global chemical imbalance.
Earlier clinical evidence supports the same idea from the patient side. Task-based measures of perceptual bias in unmedicated schizophrenia patients showed that the strength of the bias tracked with striatal dopamine release, a relationship confirmed through pharmacological manipulation with amphetamine. In plain terms, the more dopamine flooding the striatum, the more the brain trusted its own internal predictions over incoming sensory data, tipping perception toward hearing or seeing things that were not there. That link between neurotransmitter levels and computational behavior provides a mechanistic bridge between molecular pathology and the lived experience of hallucinations.
What connects these two circuits is a shared computational problem. In the hippocampus, reduced glycine weakens the ability to encode new patterns. In the striatum, excess dopamine inflates the weight the brain gives to old patterns. Both failures push the system toward rigidity, making it harder for a person to update beliefs when the world changes. The clinical picture of schizophrenia (persistent delusions, resistance to contradictory evidence, and difficulty adapting to new situations) maps neatly onto this combination of impaired encoding and overconfident prediction.
Quantifying Inflexible Learning Across Diagnoses
Computational psychiatry research published in Brain has put numbers on this rigidity. Using formal modeling of learning-rate dynamics in a large sample, researchers found that learning parameters are altered in schizophrenia in ways that can be mapped onto candidate circuits, including both corticostriatal and hippocampal pathways. Patients showed reduced capacity to adjust how quickly they learned from unexpected outcomes, a deficit that could plausibly arise from either weakened NMDA-dependent plasticity or distorted dopamine prediction-error signals.
The same study reported that similar, though not identical, learning deficits appear in depression, suggesting that inflexible updating may be a shared vulnerability across psychiatric conditions rather than a feature unique to any one diagnosis. That transdiagnostic angle complicates the standard clinical picture. Most drug development for schizophrenia still targets dopamine D2 receptors, a strategy rooted in the decades-old dopamine hypothesis. If the disorder also involves a glycine deficit in the hippocampus and a broader failure of learning-rate calibration, then D2 blockade alone addresses only one node in a distributed problem.
Glycine-site agonists and NMDA receptor modulators have been tested in clinical trials with mixed results, in part because they act broadly on receptors throughout the brain. The GLDC findings offer a more specific upstream target: the enzyme that degrades glycine in the first place. In principle, modulating that enzyme in a regionally selective way could restore plasticity in hippocampal circuits without globally overactivating NMDA receptors, which carry a risk of excitotoxicity. Translating that idea into a safe, targeted therapy will require new delivery strategies and careful dose-finding studies.
Genetic Risk Narrows the Search
Large-scale human genetics has been steadily narrowing the list of genes that carry real weight in schizophrenia risk. A major exome-sequencing study published in Nature identified rare coding variants in ten genes that confer substantial risk for the disorder. Several of those genes, including GRIN2A, encode proteins directly involved in NMDA receptor signaling, the same pathway disrupted by the GLDC mutation. By highlighting a small set of high-impact variants among thousands of candidates, the exome work provides a roadmap for focusing mechanistic experiments on the most promising molecular levers.
These converging genetic results strengthen the argument that glutamatergic pathways sit near the core of schizophrenia biology. While common variants of small effect are scattered across the genome, rare damaging mutations in NMDA-related genes appear again and again in patients with psychosis. The GLDC copy-number increase fits naturally into that picture: it does not alter the receptor itself, but it changes the availability of a co-agonist that the receptor needs to function properly. For clinicians and drug developers, that distinction matters because it expands the list of viable targets beyond receptors to include transporters, enzymes, and other regulators of synaptic microenvironments.
From Campus Labs to Clinical Translation
The GLDC work underscores the role of research universities in connecting basic neuroscience to clinical questions. At the University of Illinois Urbana-Champaign, investigators have drawn on campus-wide strengths in neurobiology, computation, and engineering, supported by institutional resources highlighted on the university’s strategic communications site, to build animal models that mirror human genetic findings. Those models, in turn, allow teams to test how specific mutations reshape circuit function and behavior in ways that cannot be probed directly in people.
For patients and families, the immediate impact of such studies may seem distant. Yet each mechanistic link (from GLDC to glycine depletion, from dopamine surges to biased learning rates, from exome variants to disrupted NMDA signaling) helps replace vague notions of “chemical imbalance” with testable, circuit-level hypotheses. As more of these links are mapped, they offer a scaffold for designing treatments that are both more precise and more personalized, targeting the particular pathways that are disrupted in a given individual rather than applying the same broad-spectrum drugs to everyone.
Schizophrenia remains a heterogeneous and challenging illness, and no single gene or circuit will explain all of its forms. But by tracing how specific mutations derail the brain’s learning machinery, researchers are beginning to show how delusions and hallucinations can emerge from ordinary computations pushed slightly, and then catastrophically, off course. The hope is that future therapies informed by this work will do more than blunt symptoms. They will restore the brain’s capacity to learn from the world as it is, rather than as a broken prediction system insists it must be.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
