3.1 Clinical Data and Behavioral Results
We initially identified 108 patients and 122 comparison subjects with complete imaging, demographic, and medication information. Behavioral response as measured by the number of button presses for each run was recorded for each subject. The average behavioral response for patients and comparison subjects was 0.92 (ratio of total button presses/total number of tones). The response rate was normally distributed reflecting errors of omission and commission. To ensure that subjects had similar behavioral response and similar performance for each run, we only included subjects who completed the behavioral response (i.e., button press total) within one standard deviation (+/− 0.17) of the average. This reduced our sample to 79 patients and 114 comparison subjects. Each site lost participants with this behavioral criterion.
The demographic information and clinical variables are presented in . Subjects were represented from all four sites (patients, comparison subjects): IOWA (21, 45), MGH (14, 15), MINN (21, 24), and UNM (23, 30). Patients differed from comparison subjects in relation to IQ as measured by the Wide Range Achievement Test Third Edition (Reading subtest) (WRAT3) (t188 = −4.19, P < 0.01), but maternal and paternal education were not significantly different between groups (P = 0.17, P = 0.32, respectively).
Table 1 This table displays demographic information for patients with schizophrenia and healthy comparison subjects. The Scale for the Assessment of Negative Symptoms (SANS) and the Scale for the Assessment of Positive Symptoms (SAPS) Global Rating ranged from (more ...)
All patients included in this sample were taking antipsychotics at the time of the scan. The medication information and chlorpromazine equivalents are presented in . Seven patients were taking typical antipsychotics. Five of these seven patients were prescribed benztropine. Seventy-two patients were taking atypical antipsychotics. Three of the 72 patients were prescribed benztropine, and one patient was prescribed amantadine. Patients taking typical and atypical antipsychotics were not significantly different with the Simpson Angus Scale (P = 0.38), Barnes Akathisia Scale (P = 0.25), and the Abnormal Involuntary Movement Scale (P = 0.59). The chlorpromazine equivalents approached statistical significance between the typical and atypical groups (P = 0.07).
The patient medication information is displayed by type of antipsychotic, number of patients and chlorpromazine equivalents. The chlorpromazine equivalents had trend level significance between typical and atypical groups (p = 0.07).
3.2 Component Selection
ICA identified 20 temporally cohesive but spatially distributed components. Visual inspection determined that three components were related to artifact. Six components correlated with spatial maps for cerebral spinal fluid and white matter. The remaining eleven components represented motor (cortical and subcortical), attention (fronto-parietal and inferior frontal), default mode, and visual (occipital lobe) functions. The two basal ganglia/motor networks had similar activation patterns in the caudate, lentiform nucleus, thalamus and cerebellum. To facilitate discussion, we have labeled these components based on their site of maximal activation: motor/caudate and motor/posterior cingulate. The attention networks included the right and left fronto-parietal networks as well as the inferior frontal frontal networks. The default mode networks were divided into two anterior components and one posterior component. The two anterior default mode networks are also labeled after their site of maximal activation: the anterior default mode network/medial frontal and the anterior default mode network/caudate. The axial views of the selected components are shown in and the top five brain regions within each component based on a random effects one-sample t test are presented in .
Figure 1 The axial slices (z = + 30, +10, −10 mm from anterior commissure/posterior commissure plane) of the components of interest: a) unimotor/bitemporal, b) posterior default mode network, c) inferior frontal, d) motor/posterior cingulate, e) anterior (more ...)
The top five brain regions within each component that show the highest significance from a one-sample t test (P < 1.0 × 10−10, false discovery rate).
The mean time course of each component was regressed with the SPM5 design matrix for a temporal sort. Consistent with the experimental motor demands of the experimental paradigm, the unimotor/bitemporal component had the highest correlation with the experimental design (r2 = 0.47). The posterior default mode network had the next highest correlation (r2 = 0.14). The regression values for each component are presented in .
Component statistics included regression from the temporal sort, group differences from a two-sample t-test, and Spearman correlations between component beta weights and chlorpromazine equivalents.
3.3 Group Differences
We assessed group differences in beta weights with a two-sample t-test. Eight of the eleven selected components had significant differences between patients with schizophrenia and healthy comparison subjects. These components included the unimotor/bitemporal (t191 = 2.07, P = 0.04), posterior default mode (t191 = 2.89, P < 0.01), motor/posterior cingulate (t191 = 2.41, P = 0.017), anterior default mode/medial frontal (t191 = 2.39, P = 0.018), motor caudate (t191 = −2.65, P < 0.01), anterior default mode/caudate (t191 = 2.06, P = 0.04), right fronto-parietal (t191 = −4.05, P < 0.01), and bimotor (t191 = 5.06, P < 0.01) components. These results are presented in and . The patients had positive modulation of the bimotor component while the comparisons had negative modulation of this same component. The different activation patterns may be related to the motor and default mode (cingulate, inferior parietal lobule) anatomy of this particular component.
Figure 2 A two-sample t-test revealed significant differences between the patients with schizophrenia (SP) and the healthy comparison subjects (HC) in the betas of eight of the eleven selected components. The positively modulated components are in green/dark and (more ...)
3.4 Antipsychotic Dose
We assessed the correlations between the chlorpromazine equivalents and beta weights for the eleven selected components. These results are shown in and . The Spearman’s rank correlations within the patient group (79 observations) revealed significant correlations with the unimotor/bitemporal (rho = −0.32, P = 0.0039), motor/caudate (rho = −0.22, P = 0.046), posterior default mode network (rho = 0.26, P = 0.020), and anterior default mode network (rho = 0.24, P = 0.030). The unimotor/bitemporal component was significant after Bonferonni correction for multiple comparisons (P < 0.0045). The Spearman’s rank correlations were also significant within these networks with cumulative dose years for antipsychotics for the unimotor/bitemporal (rho = −0.34, P = 0.0025), motor/caudate (rho = −0.22, P = 0.046), and the posterior default mode network (rho = 0.23, P = 0.044). The anterior default mode network was not significant with dose years of antipsychotics (P = 0.59).
Figure 3 This figure is the scatter plot and linear fit for the chlorpromazine equivalents and the following components: A) unimotor/bitemporal network (rho = −0.32, P = 0.0039), B) motor/caudate (rho = −0.22, P = 0.046), C) posterior default mode (more ...)
3.5 Antipsychotic Type
We assessed differences between patients on typical and atypical antipsychotics with a two-sample t-test in the betas from the four components that had significant correlations with the chlorpromazine equivalents. Patients on typical antipsychotics had significantly less positive modulation in the motor/caudate network relative to the patients on atypicals (t77 = 2.01, P = 0.048). Patients on typical antipsychotics had a trend towards less negative modulation in the anterior default mode network/caudate network relative to patients on atypical antipsychotics (t77 = −1.92, P = 0.058). These results are shown in . The betas from the unimotor and posterior default mode network did not have any significant between patients on typical and atypical antipsychotics (P = 0.16, P = 0.85, respectively).
Figure 4 Patients on typical antipsychotics had significantly less positive modulation in the motor/caudate network relative to the patients on atypicals (t77 = 2.01, P = 0.048). Patients on typical antipsychotics had trend a trend towards less negative modulation (more ...)
3.5 Symptom and Movement Ratings
We performed pairwise correlations with symptom ratings and the top two components from the ICA temporal sort: the unimotor/bitemporal network and the posterior default mode network. We did not find a significant relationship with the unimotor/bitemporal network and positive (P = 0.57), negative (P = 0.084), and disorganized symptoms (P = 0.40). We also did not find a significant relationship with the posterior default mode network and positive (P = 0.62), negative (P = 0.21), and disorganized symptoms (P = 0.91). We did not find any significant relationships with pairwise correlations between the betas for the unimotor/bitemporal and posterior default mode network and the Abnormal Involuntary Movement, Simpson Angus Scale, and the Barnes Akathisia Scale (P > 0.22).
3.6 Site Differences
We performed a one-way ANOVA with site to assess for site differences in the top two components from the ICA temporal sort, the unimotor/bitemporal and posterior default mode networks. We also used a one-way ANOVA to assess for site differences on demographic variables such as age, IQ, duration of illness and chlorpromazine equivalents. Site had a significant effect on the unimotor/bitemporal network (F3, 189 = 54.25, P < 0.001) and the posterior default mode network (F3, 189 = 8.29, P < 0.001). Subjects at MGH had less positive and negative modulation. Site differences were not significant for age (F3, 189 = 2.38, P = 0.071), IQ as measured by the WRAT-3 (F3, 186 = 0.39, P = 0.76), duration of illness (F3, 72 = 0.51, P = 0.68), and chlorpromazine equivalents (F3, 75 = 1.03, P = 0.38). We repeated the correlations between the betas for the unimotor/bitemporal networks and chlorpromazine equivalents without MGH. The correlations between the betas for the unimotor/bitemporal and chlorpromazine equivalents remained significant (rho = −0.43, P < 0.001, 65 observations).