Abstract
Background
A key challenge in the field of neuropsychiatry lies in matching patients to effective treatments. Most studies operate under the canonical assumption that categorical diagnostic groupings (such as schizophrenia and bipolar disorder) and/or pre-existing clinical assessments (such as the PANSS) are the ‘gold standard’ for describing behavioral and therefore neural variation in patients. Attempts to robustly characterize the neural substrates of these symptom measures have yielded limited success, suggesting an inadequate mapping to neurobiologically meaningful variation. Notably, a great deal of behavioral and neural heterogeneity exists even within groups of patients with the same diagnosis. Thus, understanding the mapping between specific symptoms and clinically meaningful variation in neural properties is critical to developing, and ultimately administering, effective individualized treatment.
Methods
Here, we describe a multivariate neurobehavioral framework under which behavioral variation in psychosis spectrum disorders can be mapped to features of specific neural systems in a data-driven way. We leverage neural (fMRI-derived) and behavioral data from 202 healthy controls and 436 patients from the Bipolar & Schizophrenia Consortium for Parsing Intermediate Phenotypes study. The patient sample includes 167 individuals with a formal diagnosis of schizophrenia, 119 with a diagnosis of schizoaffective disorder, and 150 with bipolar disorder with psychosis. We first identify dimensions of behavioral variation in patients by performing a principal component analysis across all behavioral measures. Next, using data-driven multivariate techniques including canonical correlation analysis, we show that variation along these behavioral dimensions relates to variation in the global brain connectivity of specific neural systems in the cortex and subcortex. Throughout our analyses, we use permutation testing to nonparametrically assess for significance, and explicitly test for site effects and robustness using both leave-one-site-out and k-fold cross-validation. Finally, we demonstrate that these brain-behavioral relationships along the psychosis spectrum can be readily mapped to neural/cellular properties such as gene expression.
Results
Importantly, the behavioral dimensions of maximal variation identified via PCA are not parallel to traditional symptom scales from pre-existing clinical instruments, and do not reflect conventional diagnostic boundaries. These behavioral axes are highly stable and robust to site and sample effects. Further, we observe robust neurobehavioral relationships using our data-driven behavioral dimensions that are not present using either traditional diagnostic groups or a priori clinical scales. We further show that this framework can inform the identification of pharmacological targets aimed at treating specific symptom profiles and may assist in selecting behavioral measures that precisely pinpoint neural variation at the individual level.
Discussion
Characterizing how and which specific sets of symptoms map to neural circuitry is a key step towards developing targeted and effective treatments for schizophrenia and psychosis spectrum disorders. Using such a framework, we can identify genetic and molecular targets that are associated with specific dimensions of behavioral variation and develop pharmaceutical agents that may address deficits along these axes. We propose the Neuro-Behavioral Relationships In Dimensional Geometric Embedding (N-BRIDGE) framework as a key step towards unified mapping between the geometry of behavioral variation and the geometry of neural variation in psychiatry.