Associations between high-dimensional datasets, each comprising many features, can be discovered through multivariate statistical methods, like Canonical Correlation Analysis (CCA) or Partial Least Squares (PLS). CCA and PLS are widely used methods which reveal which features carry the association. Despite the longevity and popularity of CCA/PLS approaches, their application to high-dimensional datasets raises critical questions about the reliability of CCA/PLS solutions. In particular, overfitting can produce solutions that are not stable across datasets, which severely hinders their interpretability and generalizability. To study these issues, we developed a generative model to simulate synthetic datasets with multivariate associations, parameterized by feature dimensionality, data variance structure, and assumed latent association strength. We found that resulting CCA/PLS associations could be highly inaccurate when the number of samples per feature is relatively small. For PLS, the profiles of feature weights exhibit detrimental bias toward leading principal component axes. We confirmed these model trends in state-ofthe-art datasets containing neuroimaging and behavioral measurements in large numbers of subjects, namely the Human Connectome Project (n ≈ 1000) and UK Biobank (n = 20000), where we found that only the latter comprised enough samples to obtain stable estimates. Analysis of the neuroimaging literature using CCA to map brain-behavior relationships revealed that the commonly employed sample sizes yield unstable CCA solutions. Our generative modeling framework provides a calculator of dataset properties required for stable estimates. Collectively, our study characterizes dataset properties needed to limit the potentially detrimental effects of overfitting on stability of CCA/PLS solutions, and provides practical recommendations for future studies.Significance StatementScientific studies often begin with an observed association between different types of measures. When datasets comprise large numbers of features, multivariate approaches such as canonical correlation analysis (CCA) and partial least squares (PLS) are often used. These methods can reveal the profiles of features that carry the optimal association. We developed a generative model to simulate data, and characterized how obtained feature profiles can be unstable, which hinders interpretability and generalizability, unless a sufficient number of samples is available to estimate them. We determine sufficient sample sizes, depending on properties of datasets. We also show that these issues arise in neuroimaging studies of brain-behavior relationships. We provide practical guidelines and computational tools for future CCA and PLS studies.

Cognitive control is a cognitive and neural mechanism that contributes to managing the complex demands of day-to-day life. Studies have suggested that functional impairments in cognitive control associated brain circuitry contribute to a broad range of higher cognitive deficits in schizophrenia. To examine this issue, we assessed functional connectivity networks in healthy adults and individuals with schizophrenia performing tasks from two distinct cognitive domains that varied in demands for cognitive control, the RiSE episodic memory task and DPX goal maintenance task. We characterized general and cognitive control-specific effects of schizophrenia on functional connectivity within an expanded frontal parietal network (FPN) and quantified network topology properties using graph analysis. Using the network based statistic (NBS), we observed greater network functional connectivity in cognitive control demanding conditions during both tasks in both groups in the FPN, and demonstrated cognitive control FPN specificity against a task independent auditory network. NBS analyses also revealed widespread connectivity deficits in schizophrenia patients across all tasks. Furthermore, quantitative changes in network topology associated with diagnostic status and task demand were observed. The present findings, in an analysis that was limited to correct trials only, ensuring that subjects are on task, provide critical insights into network connections crucial for cognitive control and the manner in which brain networks reorganize to support such control. Impairments in this mechanism are present in schizophrenia and these results highlight how cognitive control deficits contribute to the pathophysiology of this illness.

Evidence has been accumulating for an immune-based component to the etiology of psychotic disorders. Advancements in diffusion magnetic resonance imaging (MRI) have enabled estimation of extracellular free water (FW), a putative biomarker of neuroinflammation. Furthermore, inflammatory processes may be associated with altered brain levels of metabolites, such as glutathione (GSH). Consequently, we sought to test the hypotheses that FW is increased and associated with decreased GSH in patients with first episode schizophrenia (SZ) compared to healthy controls (HC). SZ (n=36) and HC (n=40) subjects underwent a multi-shell diffusion MRI scan on a Siemens 3T scanner. 1H-MR spectroscopy data were acquired using a GSH-optimized MEGA-PRESS editing sequence and GSH/creatine ratios were calculated for DLPFC (SZ: n=33, HC: n=37) and visual cortex (SZ: n=29, HC: n=35) voxels. Symptoms and functioning were measured using the SANS, SAPS, BPRS and GSF/GRF. SZ demonstrated significantly elevated FW in whole-brain gray (p=.001) but not white matter (p=.060). There was no significant difference between groups in GSH in either voxel. However, there was a significant negative correlation between DLPFC GSH and both whole-brain and DLPFC-specific gray matter FW in SZ (r=−.48 and −.47, respectively; both p<.05), while this relationship was nonsignificant in HC and in both groups in visual cortex. These data illustrate an important relationship between a metabolite known to be important for immune function – GSH – and the diffusion extracellular FW measure, which provides additional support for these measures as neuroinflammatory biomarkers that could potentially provide tractable treatment targets to guide pharmacological intervention.

Blunted and exaggerated neuronal response to rewards are hypothesized to be core features of schizophrenia spectrum disorders (SZ) and bipolar disorder (BD), respectively. Nonetheless, direct tests of this hypothesis, in which response between SZ and BD is compared in the same study, are lacking. Here we examined the functional correlates of reward processing during the Incentivized Control Engagement Task (ICE-T) using 3T fMRI. Reward-associated activation was examined in 49 healthy controls (HCs), 52 recentonset individuals with SZ, and 22 recent-onset individuals with Type I BD using anterior cingulate (ACC), anterior insula, and ventral striatal regions of interest. Significant group X reward condition (neutral vs. reward) interactions were observed during reward anticipation in the dorsal ACC (F(2,120) = 4.21, P = 0.017) and right insula (F(2,120) = 4.77, P = 0.010). The ACC interaction was driven by relatively higher activation in the BD group vs. HCs (P = 0.007) and vs. individuals with SZ (P = 0.010). The insula interaction was driven by reduced activation in the SZ group relative to HCs (P = 0.018) and vs. people with BD (P = 0.008). A composite of reward anticipation-associated response across all associated ROIs also differed significantly by diagnosis (F(1,120) = 5.59, P = 0.02), BD > HC > SZ. No effects of group or group X reward interactions were observed during reward feedback. These results suggest that people with SZ and BD have opposite patterns of activation associated with reward anticipation but not reward receipt. Implications of these findings in regard to Research Domain Criteria-based classification of illness and the neurobiology of reward in psychosis are discussed.

The current study extends the mand-for-information literature by examining a method to teach mand-for-information frames, targeting 2 frames for the "How?" mand ("How do I?" and "How many?"). Using separate behavior chains to target the 2 frames, we taught 3 children with autism to emit mands for information with 1 behavior chain and assessed generalization with the remaining behavior chains. Behavior chains that the participants were unable to perform independently and that produced a desirable outcome for the participant (e.g., tornado water) were used to contrive the relevant motivating operation. For all 3 participants, mands for information generalized across motivating operations and response topographies.