SIRS 2012—The Challenge of Predicting Psychosis With Brain Imaging
9 May 2012. Although the first psychotic episode in schizophrenia may seem to come on suddenly, it typically reflects the culmination of subtle shifts in behavior, called the “prodrome.” A cadre of brain imagers at the 2012 Schizophrenia International Research Society meeting in Florence, Italy, presented their work on detecting the brain changes behind psychosis onset, with a focus on people at high risk for developing psychosis. Of people showing attenuated psychotic symptoms or a brief, yet quickly resolved episode of psychosis, about a third develop full-blown psychosis within three years, according to a recent analysis (Fusar-Poli et al., 2012). Studying a population enriched for imminent psychosis offers researchers a chance to capture valuable before-and-after snapshots of the brain, and to discern differences between those who transition to psychosis and those who don’t. What’s more, extracting accurate predictors of psychosis would guide early treatment decisions, which are currently complicated by the high number of high-risk people who do not ever develop psychosis.
The challenge lies in the large amount of variability in any single measure of brain scan data. Combining different measures in a multimodal approach might offer greater accuracy, and several researchers tried to pull together structural neuroanatomy, brain activity, neurochemistry, and connectivity findings. The multimodal approach poses a tremendous burden of time and cost, however, so it may still be a while before psychiatry sees a clinical application of imaging.
Signs of psychosis in dopamine
In a symposium Sunday afternoon, Oliver Howes of Imperial College, London, United Kingdom, reviewed his work on signs of enhanced dopamine levels in schizophrenia and in high-risk patients. His new meta-analysis of 50 positron-emission tomography (PET) or single-photon emission computed tomography (SPECT) studies on the subject (see SRF related news story) finds that presynaptic increases in dopamine distinguish schizophrenia from controls, but not measures of dopamine receptor or transporter availability. Elevated dopamine also marks high-risk subjects, who show increased dopamine synthesis capacity in the striatum compared to controls (Howes et al., 2009), and Howes said that this has recently been replicated in a second cohort. Intriguingly, higher dopamine levels also distinguish those who later transition to psychosis (see SRF related news story), suggesting that overactive dopamine synthesis or release precedes illness. This measure shows some specificity for psychotic illness, because healthy people who have hallucinations without any functional downsides show dopamine levels similar to controls (Howes et al., 2012).
Stefan Borgwardt of the University of Basel, Switzerland, followed with a review of other measures that distinguish high-risk subjects from controls, including brain volume reductions (Mechelli et al., 2011); similar to the pattern with dopamine, people who later develop psychosis show even more pronounced reductions. He suggested that following brain changes, rather than abnormalities, was important, and integrating the relationships between different brain measures rather than focusing on a single measure may increase predictive power.
A multimodal look at the prodromal brain
Getting an integrative look at the brain to better define a high-risk state with better accuracy for transition to psychosis was the focus of a symposium on Monday afternoon. Paolo Fusar-Poli of King’s College, London, United Kingdom, described his efforts to put together how brain activity varies with neurochemistry, specifically glutamate. Last year he reported that in a group of high-risk patients, glutamate levels in the thalamus (measured with magnetic resonance spectroscopy, or MRS) were inversely correlated with fMRI signals in the dorsolateral prefrontal cortex or the orbitofrontal cortex evoked during a verbal fluency task in which subjects generated a word beginning with a given letter; the corresponding correlations in the controls were in the opposite direction (Fusar-Poli et al., 2011). Extending this kind of analysis to dopamine, Fusar-Poli reported new results showing a direct correlation between striatal levels of dopamine and activation of the inferior frontal gyrus.
Another method in the multimodal genre is to consider all structural changes in the brain using voxel-based morphometry analysis (VBMA), which gives an unbiased view of all the voxels in a brain scan, rather than a priori choosing some region of interest. In a VBMA published last year, Fusar-Poli found that those who later developed psychosis showed at baseline reductions in the inferior frontal gyrus and the superior temporal gyrus. He is now working to integrate these structural findings with fMRI results in people imaged at the time of their first episode of psychosis. In the same Monday multimodal symposium, Stefan Borgwardt, presented other data highlighting the insula, which VBMA found to have a smaller volume in people designated with a high-clinical-risk mental state for only three months (and so had a higher transition-to-psychosis probability) compared to those in that category for five years (with a lower chance of transition).
Christopher Chaddock of King’s College, London, United Kingdom, continued with the dopamine theme, but this time in the context of prediction errors. Dopamine neurons produce a teaching signal by firing when experience differs from what is expected. Abnormalities in prediction error signaling have been proposed to underlie psychosis, such that dopamine would inappropriately tag innocuous stimuli as salient, and psychosis would emerge as a cognitive explanation for these abnormally learned associations (Kapur, 2003).
Abnormal prediction error signaling has been found in people with schizophrenia, and Chaddock asked whether this emerges even earlier in the prodrome. While undergoing fMRI, 15 high-risk and 18 control subjects learned to choose a stimulus that predicted a money reward. Controls showed increased fMRI activity in the striatum (a recipient of dopamine inputs) during unexpected reward trials, and decreased activity during unexpected neutral, non-rewarded trials. The inverse was observed in the high-risk group, however, with striatal activity suppressed during unexpected reward trials, and boosted during unexpected neutral ones. The magnitude of the abnormal reward signal correlated with severity of psychotic symptoms, and was highest for those who had transitioned to full-blown psychosis. Adding a PET scanning dimension, Chaddock also reported that those with the highest levels of presynaptic dopamine exhibited the greatest abnormal response-to-reward prediction errors.
Machines at work
While group differences between those who transition and those who don’t are helpful for understanding psychosis onset, it doesn’t mean they’ll be able to predict an individual’s fate. To move toward something more predictive, Nikos Koutsouleris of the University of Pennsylvania, Philadelphia, discussed the application of machine learning to pull patterns from brain imaging data that foretell psychosis. This means thinking of disease as a disturbance of relationships, rather than discrete abnormalities, he said.
Koutsouleris has had some success in training pattern recognition algorithms to discriminate between those who develop psychosis and those who don’t using neuroanatomical features (see SRF related news story) and cognitive features (Koutsouleris et al., 2011). He presented preliminary evidence showing no additional benefit from combining these measures, however. In a sample of 31 high-risk subjects, with 14 transitioning to psychosis, he reported that the classifier could recognize those who transitioned based solely on neuroanatomical features with 64 percent accuracy, and based only on cognitive features with 80 percent accuracy. Combining the two resulted in accuracy of 76 percent, correctly identifying 71 percent of those who transitioned (an increase in sensitivity compared to cognitive alone) and 81 percent of those who didn’t (a decrease in specificity compared to cognitive alone). Koutsouleris said that his classifiers did not do so well in distinguishing controls from those with chronic schizophrenia, suggesting that the brain state in chronic schizophrenia is much more heterogeneous than in the prodrome, perhaps reflecting the different disease courses, lifestyles, and medication histories that emerge with chronic disease.
Stephen Lawrie of the University of Edinburgh, United Kingdom, focused on relating morphological features—cortical folding, in particular—to functional data in a different kind of high-risk cohort consisting of people at genetic risk, who have two or more first- or second-degree relatives with schizophrenia. Prompted by the increase in gyrification index (the ratio of the length of the inner fold over that of the outer fold) in prefrontal cortex found in genetic high-risk subjects who developed schizophrenia (Harris et al., 2007), Lawrie proposed that this reflected an abnormally high degree of local connectivity. To test this, he studied the relationship between gyrification index and functional connectivity, measured via fMRI during an executive function task. Prefrontal gyrification (folding) correlated positively with local connectivity between medial and lateral regions of the prefrontal cortex, but correlated negatively with long-range connectivity between prefrontal cortex and thalamus; these correlations were not seen in controls (Dauvermann et al., 2012). Lawrie suggested that in schizophrenia, local connections are increased at the expense of long-range connections.
Lawrie also described an effort to infer direction of the signals flowing through the brain in these high-risk subjects, using dynamic causal modeling. This approach pulled out a role for signals originating in the thalamus and moving to inferior frontal gyrus (IFG), and high-risk individuals with psychotic symptoms or who had transitioned to schizophrenia showed greater connection strength between thalamus and IFG than did high-risk people without symptoms or healthy controls. Finally, using a graph theory to get a global view of the brain’s network of connections (see SRF related news story), Lawrie reported similar network structures in both high-risk individuals and controls, but that those who transition exhibited a higher degree of connectivity and clustering of regions.—Michele Solis.