7 Apr 2016
April 8, 2016. Considering the heterogeneity of schizophrenia, researchers have been looking for signs beyond symptoms that may divide the diagnosis into more homogeneous subgroups. Such subtypes, whether defined by cognitive profiles, blood markers, brain scans, or genetics, might reflect different etiologies and guide treatment.
On Monday, April 4, at the Schizophrenia International Research Society meeting in Florence, Italy, Nikolaos Koutsouleris of Ludwig Maximilian University in Munich, Germany, gave the morning plenary talk in which he reviewed the use of machine learning as a hypothesis-free way to find patterns in brain images obtained from magnetic resonance imaging (MRI) using multivariate pattern analysis (MVPA). Among other things, he showed preliminary evidence that such an approach could differentiate two brain types in scans from people with schizophrenia.
In general, Koutsouleris had messages for the brain imaging community: He stressed the importance of validating findings in imaging data sets that were not used in training the pattern classifier; the need to find ways to deal with different scanner systems in different centers; and a call to resist the urge to over-purify samples to find signals, as findings would not ultimately be useful in a clinic.
That afternoon, Katherine Burdick of the Icahn School of Medicine at Mount Sinai in New York City led a symposium about finding different ways to differentiate patients based on phenotypes that may be shared across diagnoses. Applying clustering methods to cognitive measures from people with bipolar disorder, Burdick has previously found three subtypes: one with intact cognition; another with impaired cognition in processing speed, attention, and social cognition; and a third with impairments across all domains (Burdick et al., 2014). She presented preliminary data showing that those with cognitive declines after illness onset also had higher cytokine levels in their blood than those without declines. This suggests that neuroinflammation accompanies episodes of mania or depression, and anti-inflammatory treatments may stave off any damaging effects it might have on cognition.
Melissa Green of the University of New South Wales in Sydney, Australia, applied a similar clustering analysis in cognitive measures from people with schizophrenia and bipolar I as part of the Imaging Genetics in Psychosis (IGP) study. This gave two clusters: one in which cognition was spared, and another in which it was deficient. The clusters cut across diagnoses, as those with schizophrenia were split fairly evenly between the two subgroups, but with most of those with bipolar belonging to the spared group. Wondering whether those in the cognitive deficit group had a higher load of genetic risk factors, Green used the Psych Chip to genotype people on risk variants identified by the Psychiatric Genomic Consortium's GWAS published in 2014 (see SRF related news report). This gave some slight non-significant separation between the two groups. A separate cohort of people with schizophrenia and healthy controls who had had their whole genomes sequenced gave a cleaner separation between the cognitively spared and cognitively deficient subgroups, with the latter carrying the highest polygenic risk score.
Researchers have been entranced by the idea that genetics alone could differentiate subtypes, and, indeed, one controversial study claims to have found eight types of schizophrenia (see SRF related news report). Shaun Purcell of the Icahn School of Medicine at Mount Sinai argued that finding substructures within the many common single nucleotide polymorphism (SNP) risk factors for schizophrenia is a plausible but fraught prospect. For example, it's not clear which SNPs matter; heterogeneity in ethnicity of the population could masquerade as subtypes (the suspected problem in the "eight types" study), and researchers lack a way to evaluate whether they've identified a true subtype. "There are many ways this can go wrong," Purcell said.
Taking a conservative approach in a "toy" example, he showed that when data from a genetically homogeneous subset were analyzed, and strict SNP pruning was applied to avoid residual linkage disequilibrium and population stratification, a single type resulted, either with principal component analysis or latent class analysis. Though the jury is still out on this approach, he suggested that SNPs that are found in future GWAS may be more subtype specific. Alternatively, schizophrenia's heterogeneity may not be mediated by genetics at all; rather, genes may get a person to a baseline level of risk from which other factors delineate a specific phenotype.
John McGrath of Queensland Brain Institute in Australia remarked that such biological markers of etiology—sometimes called endophenotypes—have not been as forthcoming as people had hoped. "Endophenotypes were thought to be the royal road to discovery, but they are not simple enough, or obedient enough," he said. He suggested that environmental factors might also help disentangle subtypes.—Michele Solis.