Cortico-cognition coupling in treatment resistant schizophrenia

Abstract

Background: Brain structural alterations and cognitive dysfunction are independent predictors for poor clinical outcome in schizophrenia, and the associations between these domains remains unclear. We employed a novel, multiblock partial least squares correlation (MB-PLS-C) technique and investigated multivariate cortico-cognitive patterns in patients with treatment-resistant schizophrenia (TRS) and matched healthy controls (HC).

Method: Forty-one TRS patients (age 38.5 ± 9.1, 30 males (M)), and 45 HC (age 40.2 ± 10.6, 29 M) underwent 3T structural MRI. Volumes of 68 brain regions and seven variables from CANTAB covering memory and executive domains were included. Univariate group differences were assessed, followed by the MB-PLS-C analyses to identify group-specific multivariate patterns of cortico-cognitive coupling. Supplementary three-group analyses, which included 23 non-affected first-degree relatives (NAR), were also conducted.

Results: Univariate tests demonstrated that TRS patients showed impairments in all seven cognitive tasks and volume reductions in 12 cortical regions following Bonferroni correction. The MB-PLS-C analyses revealed two significant latent variables (LVs) explaining > 90% of the sum-of-squares variance. LV1 explained 78.86% of the sum-of-squares variance, describing a shared, widespread structure-cognitive pattern relevant to both TRS patients and HCs. In contrast, LV2 (13.47% of sum-of-squares variance explained) appeared specific to TRS and comprised a differential cortico-cognitive pattern including frontal and temporal lobes as well as paired associates learning (PAL) and intra-extra dimensional set shifting (IED). Three-group analyses also identified two significant LVs, with NARs more closely resembling healthy controls than TRS patients.

Conclusions: MB-PLS-C analyses identified multivariate brain structural-cognitive patterns in the latent space that may provide a TRS signature.

Keywords: Brain structure; Cognition; MRI; Partial least squares correlation; Treatment-resistant schizophrenia.

Fig. 1

Decomposition of input multiblock cross-correlation matrix into additive components using MB-PLS-C framework. A) The input matrix is composed of four disjoint data blocks with multivariate measures of brain structure, cognition and covariates (30 HC, 30 TRS). B-C) The first two components corresponding to the significant latent variables, LV1 and LV2, describe correlations in the latent space between cortical, cognitive and covariate measures across patients and healthy controls (right side of the equation).

Fig. 2

The first latent cortico-cognitive pattern. A) Group-specific cortical saliences (weights of cortical variables) in the healthy controls (HC) and TRS patients. B) Common between-group cognitive saliences. Reliable contributions from all cognitive variables, except IED-IS and PAL-SC. C) A bar graph of group-wise cortical saliences in the HC (green) and TRS (red) groups with 95% confidence intervals (black lines) and color-coded lobe information. All reliable cortical saliences are negative. D) Regional Normalized structure salience difference between the HC and TRS groups. A positive value (green) indicates stronger regional salience in the HCs and a negative value (red) shows stronger TRS saliences. E) LV1 Cortico-cognitive correlations in the training and test samples, and the mean, and standard deviation of the distribution of correlation coefficients from Monte Carlo cross-validation. F) Post hoc latent cortico-cognitive correlations after LV-specific covariate adjustments. Significant moderate correlations in both groups. Non-significant Fisher’s r-to-z test suggests no between-group differences in LV1 cortico-cognitive coupling. G-H) The effect of sample-size on salience strength. The cortico-cognitive patterns converge after 20 samples per group.

Fig. 3

The second latent cortico-cognitive pattern. A) Group-specific cortical saliences (weights of cortical variables) in the healthy controls (HC) and TRS patients showing a differential cortical pattern between groups. B) Common between-group cognitive saliences. Only PAL and IED related variables contributed significantly to the pattern. C) A bar graph of group-wise cortical saliences in the HC (green) and TRS (red) groups with 95% confidence intervals (black lines) and color-coded lobe information. All reliable cortical saliences are negative in TRS patients and positive in HCs. D) Regional Normalized structure salience difference between the HC and TRS groups. A positive value (green) indicates stronger regional salience in the HCs and a negative value (red) shows stronger TRS saliences. E) LV2 cortico-cognitive coupling in the training and test samples, and the mean, and standard deviation of the distribution of correlation coefficients from Monte Carlo cross-validation. Generalizable cortico-cognitive correlations only in the TRS group. F) Post hoc latent cortico-cognitive correlations after LV-specific covariate adjustments. Significant moderate correlations in the TRS group. Significant Fisher’s r-t-z test suggests reliable cortico-cognitive coupling only in the TRS group. G-H) The effect of sample-size on salience strength. The cortico-cognitive patterns converged after 20 samples per group.

Fig. 4

The first latent cortico-cognitive pattern from three-group MB-PLS-C. A) Group-specific cortical saliences (weights of cortical variables) in the HC, TRS and NAR groups. B) A radial bar graph of group-wise cortical saliences with 95% confidence intervals (black lines) and color-coded lobe information. All reliable cortical saliences are negative. C) Cognitive saliences from averaged cognitive performance across both groups. IED-IS and PAL measures contributed unreliably to the pattern, whereas similar contributions to the cognitive pattern arose from IED-ES and SWM measures. D) The effect of sample-size on salience strength.

Fig. 5

The second three-group latent cortico-cognitive pattern. A) Group-specific cortical saliences (weights of cortical variables). B) A bar graph of group-wise cortical saliences with 95% confidence intervals (black lines) and color-coded lobe information. All reliable cortical saliences are negative in the TRS group and positive in the HC and NAR groups. C) Cognitive saliences from averaged cognitive performance across both groups. Only IED and PAL measures contributed reliably to the pattern. D) The effect of sample-size on salience strength. The cortical and cognitive patterns start converging after 15 samples per group.