A Stanford Medicine-led study has found that there are two key systems malfunctioning in the brains of people with psychosis. One is a “filter” that directs attention toward important external events and internal thoughts, and the other a “predictor” composed of pathways that anticipate rewards.
The team used a type of machine learning algorithm called a spatiotemporal deep neural network to analyze brain scan data from children, teens, and young adults with a rare genetic disease called 22q11.2 deletion syndrome (22q11.2DS) who experience psychosis, as well as from individuals with psychosis of unknown origin, and healthy individuals. The results suggested that in psychosis, dysfunction of these two brain systems makes it difficult to know what’s real, manifesting as hallucinations and delusions. Confirming an existing theory of how breaks with reality occur, the findings advance scientists’ understanding of the underlying brain mechanisms and theoretical frameworks related to psychosis. In addition to supporting the scientists’ theory about how psychosis occurs, the new insights could help scientists develop strategies for prevention.
“The brain patterns we identified support our theoretical models of how cognitive control systems malfunction in psychosis,” said senior study author Vinod Menon, PhD, the Rachel L. and Walter F. Nichols, MD, Professor; a professor of psychiatry and behavioral sciences; and director of the Stanford Cognitive and Systems Neuroscience Laboratory. “This work provides a good model for understanding the development and progression of schizophrenia, which is a challenging problem,” added Kaustubh Supekar, PhD, clinical associate professor of psychiatry and behavioral sciences.
Menon, Supekar, and colleagues reported on their studies in Molecular Psychiatry, in a paper titled “Robust and replicable functional brain signatures of 22q11.2 deletion syndrome and associated psychosis: a deep neural network-based multi-cohort study,” in which they concluded, “Collectively, our findings, replicated across multiple independent cohorts, advance the understanding of 22q11.2DS and associated psychosis, underscoring the value of 22q11.2DS as a genetic model for probing the neurobiological underpinnings of psychosis and its progression … The robustness of our findings also further underscores the potential of our novel DNN-based approach to uncover commonalities between neurogenetic and psychiatric disorders.”
During psychosis, patients experience hallucinations, such as hearing voices, and hold delusional beliefs, such as thinking that people who are not real exist. Psychosis can occur on its own and is a hallmark of certain serious mental illnesses, including bipolar disorder and schizophrenia. Schizophrenia is also characterized by social withdrawal, disorganized thinking and speech, and a reduction in energy and motivation.
It is challenging to study how schizophrenia begins in the brain. The condition usually emerges in teens or young adults, most of whom soon begin taking antipsychotic medications to ease their symptoms. When researchers analyze brain scans from people with established schizophrenia, they cannot distinguish the effects of the disease from the effects of the medications. They also do not know how schizophrenia changes the brain as the disease progresses.
To get an early view of the disease process, the Stanford Medicine team studied young people aged six to 39 with 22q11.2 deletion syndrome (a genetic condition with a 30% risk for psychosis, schizophrenia or both). Children with 22q11.2DS have a significantly increased risk of developing schizophrenia in adolescence and adulthood, making the deletion the highest genetic risk factor for schizophrenia after having an identical twin or two parents with the disorder,” the team wrote. “Thus, 22q11.2DS offers a unique opportunity to investigate early brain signatures in a genetic subtype of schizophrenia that may shed light on the neurodevelopmental mechanisms underpinning the emergence and expression of the disorder.” However, the investigators acknowledged, “ … robust and replicable functional brain signatures of 22q11.2DS and 22q11.2DS-associated psychosis remain elusive due to small sample sizes and a focus on small single-site cohorts.”
For their reported study the researchers assembled as much functional MRI (fMRI) brain-scan data as possible from young people with 22q11.2 deletion syndrome, totaling 101 individuals scanned at three different universities. The analysis also included brain scans from several comparison groups without 22q11.2 deletion syndrome: 120 people with early idiopathic psychosis, 101 people with autism, 123 with attention deficit/hyperactivity disorder and 411 healthy controls.
22q11.2 deletion syndrome, which is characterized by deletion of part of the 22nd chromosome, affects one in every 2,000 to 4,000 people. In addition to the 30% risk of schizophrenia or psychosis, people with the syndrome can also develop autism or attention deficit hyperactivity disorder, so these conditions were included in the comparison groups.
Normally, the brain’s cognitive filtering system—aka the salience network—works behind the scenes to selectively direct our attention to important internal thoughts and external events. With its help, we can dismiss irrational thoughts and unimportant events and focus on what’s real and meaningful to us, such as paying attention to traffic so we avoid a collision. The ventral striatum, a small brain region, and associated brain pathways driven by dopamine, play an important role in predicting what will be rewarding or important.
The researchers used stDNN analysis to characterize and compare patterns of brain function in all patients with 22q11.2 deletion syndrome and in healthy subjects. With a cohort of patients whose brains were scanned at the University of California, Los Angeles, they developed an algorithmic model that distinguished brain scans from people with 22q11.2 deletion syndrome versus those without it. The model predicted the syndrome with greater than 94% accuracy. “We found that our stDNN model, which models the underlying dynamic spatiotemporal characteristics of brain activity to distinguish between groups using fMRI timeseries, accurately (>94%) distinguishes individuals with 22q11.2DS from healthy controls in the UCLA cohort, outperforming conventional approaches,” they stated.
They validated the model in additional groups of people with or without the genetic syndrome who had received brain scans at UC Davis and Pontificia Universidad Católica de Chile (PUC), showing that in these independent groups, the model sorted brain scans with 84% to 90% accuracy. “We found that stDNN accurately (84–90%) distinguishes individuals with 22q11.2DS from healthy controls in a robust and consistent manner across two independent cohorts (UC Davis and PUC cohorts) without additional training, outperforming conventional approaches,” they added.
The researchers then used the model to investigate which brain features play the biggest role in psychosis. Prior studies of psychosis had not given consistent results, likely because their sample sizes were too small. Comparing brain scans from 22q11.2 deletion syndrome patients who had and did not have psychosis, the researchers showed that the brain areas contributing most to psychosis are the anterior insula—a key part of the salience network or “filter”—and the ventral striatum—the “reward predictor.” This, they found, was true for different cohorts of patients. “Robust distinguishing features of 22q11.2DS-associated psychosis emerged in the anterior insula node of the salience network and the striatum node of the dopaminergic reward pathway,” they wrote.
And when comparing the brain features of people with 22q11.2 deletion syndrome and psychosis with those features in people with psychosis of unknown origin, the model found significant overlap, indicating that these brain features are characteristic of psychosis in general. And these brain patterns matched what the researchers had previously theorized was generating psychosis symptoms. “Additionally, the brain signatures of 22q11.2DS-associated psychosis overlap with those of idiopathic early psychosis in the salience network and dopaminergic reward pathway, providing substantial empirical support for the theoretical aberrant salience-based model of psychosis,” they wrote.
Thoughts that are not linked to reality can capture the brain’s cognitive control networks, Menon said. “This process derails the normal functioning of cognitive control, allowing intrusive thoughts to dominate, culminating in symptoms we recognize as psychosis.”
A second mathematical model, trained to distinguish all subjects with 22q11.2 deletion syndrome and psychosis from those who have the genetic syndrome but without psychosis, selected brain scans from people with idiopathic psychosis with 77.5% accuracy, again supporting the idea that the brain’s filtering and predicting centers are key to psychosis. Furthermore, this model was specific to psychosis, and could not classify people with idiopathic autism or ADHD. “Taken together, these results establish the distinctness of brain features associated with psychosis spectrum symptoms in 22q11.2DS and provide evidence that they specifically overlap with idiopathic early psychosis and not with idiopathic autism and idiopathic ADHD,” they stated.
“It was quite exciting to trace our steps back to our initial question—‘What are the dysfunctional brain systems in schizophrenia?’—and to discover similar patterns in this context,” Menon noted. “At the neural level, the characteristics differentiating individuals with psychosis in 22q11.2 deletion syndrome are mirroring the pathways we’ve pinpointed in schizophrenia. This parallel reinforces our understanding of psychosis as a condition with identifiable and consistent brain signatures.” However, these brain signatures were not seen in people with the genetic syndrome but no psychosis, holding clues to future directions for research, he added.
In addition to supporting the scientists’ theory about how psychosis occurs, the findings have implications for understanding the condition—and possibly preventing it. “One of my goals is to prevent or delay development of schizophrenia,” Supekar said. The new findings supporting those of the team’s prior research on which brain centers contribute most to schizophrenia in adults suggests there may be a way to prevent it, he suggested. The authors further noted, “These findings are critical for improving the accuracy of psychosis risk prediction in individuals with 22q11.2DS and identifying brain-circuit targets for clinical interventions to mitigate the risk of developing psychosis.” And the overlap, or commonality, between functional brain signatures in 22q11.2DS-associated psychosis and idiopathic early psychosis also “… points to shared underlying brain mechanisms, which could serve as early biomarkers for psychosis and potential brain targets for effective interventions.”
Supekar continued, “In schizophrenia, by the time of diagnosis, a lot of damage has already occurred in the brain, and it can be very difficult to change the course of the disease. What we saw is that, early on, functional interactions among brain regions within the same brain systems are abnormal. The abnormalities do not start when you are in your 20s; they are evident even when you are 7 or 8.”
The researchers plan to use existing treatments, such as transcranial magnetic stimulation or focused ultrasound, targeted at these brain centers in young people at risk of psychosis, such as those with 22q11.2 deletion syndrome or with two parents who have schizophrenia, to see if they prevent or delay the onset of the condition or lessen symptoms once they appear.
The results also suggest that using functional MRI to monitor brain activity at the key centers could help scientists investigate how existing antipsychotic medications are working. Although it’s still puzzling why someone becomes untethered from reality—given how risky it seems for one’s well-being—the “how” is now understandable, Supekar said. “From a mechanistic point of view, it makes sense,” he said.
