Functional Connectivity in Antipsychotic-Treated and Antipsychotic-Naive Patients With First-Episode Psychosis and Low Risk of Self-harm or Aggression

Key Points Question Are there different patterns of brain connectivity in antipsychotic-treated and antipsychotic-naive patients with psychosis, and how do these patterns evolve over time? Findings In this secondary analysis of a triple-blind, longitudinal, placebo-controlled, randomized clinical trial in antipsychotic-naive patients, patients showed widespread brain dysconnectivity at baseline relative to healthy controls. Patients treated with therapy and placebo or with therapy and antipsychotics showed evidence of circuit-specific and treatment-specific normalization of connectivity over time. Meaning In this study, some prominent baseline connectivity differences in first-episode psychosis normalized with both psychosocial therapy and placebo early in the illness course, with antipsychotics exerting circuit-specific effects.


eMethods 1. Additional Details
The trial took place at the Early Psychosis Prevention and Intervention Centre, which is part of Orygen Youth Health, Melbourne, Australia. The randomization phase of the study terminated at 6 months, so patients in either the MIPT or PIPT group could have received antipsychotic medication and ongoing psychosocial interventions in between 6 and 12 months into the study.

eMethods 2. Trial Safety Procedures
To further ensure safety, several discontinuation criteria were applied in the clinical trial.
These were operationally defined as any of the following 1,2 : increased risk to self or others (score of ≥5 on the BPRS-4 Suicidality or Hostility subscales, maintained for 1 week); increase in positive psychotic symptom severity (2-point increase on the BPRS-4 subscale of Conceptual Disorganisation, Hallucinations, Unusual Thought Content, or Suspiciousness) maintained for at least 1 week not due to substance use; decrease in overall functioning (20point drop in SOFAS score from baseline maintained for 1 month); request by the participant for the introduction of antipsychotic medication; failure to satisfactorily recover 3 months after study entry (a score of 5 or more on the BPRS-4 Hallucinations, Suspiciousness, and Unusual Thought Content subscales or a score of 4 or greater on Conceptual Disorganisation); or becoming pregnant. All participants gave written informed consent after having the study fully explained to them, parental consent was also obtained for participants under the age of 18.

eMethods 3. Antipsychotic and Concomitant Medication Details
The current study took place in the context of a larger clinical trial examining functional and clinical outcome 1 , in which the analyses included patients in the PIPT group who switched to antipsychotics. In the current study, we only included patients in the PIPT group who remained on placebo to more clearly delineate the effects of antipsychotic exposure.
Four patients within the PIPT group were switched to open-label antipsychotic medication before the 3-month MRI scan and were excluded from the primary analysis. Two additional individuals were exposed to antipsychotics at doses lower than the minimal exposure limit for inclusion in our study. Excluding these two additional participants from the analysis yielded a qualitatively similar pattern of results. After termination of the randomization phase at 6 months, five additional patients in the PIPT group were exposed to antipsychotic medication between the 3-month and 12-month scans. Mean cumulative dose and rates of exposure for both patient groups at each timepoint are provided in Supplementary Table 1. Concomitant   medications were permitted during the trial, except for additional antipsychotics or mood stabilisers. Rates of concomitant medication use between the two treatment groups were not significantly different (Supplementary Table 2). To ensure that non-antipsychotic psychotropics did not have a large impact on our primary findings, we recomputed our analyses, this time including a binary nuisance covariate for each of the three medication classes. The findings remain largely the same (Supplementary Figure 5 1 Number of patients in the placebo group who were exposed to antipsychotic medication at amounts greater than mandated by the minimal study inclusion criterion (lifetime 58mg olanzapine equivalents exposure).

eMethods 5. Image Processing and Quality Control
A total of 202 rs-fMRI datasets were acquired in this study. Raw images were first put through an automated quality control procedure (MRI-QC) 3 , which resulted in the exclusion of 4 scans due to large artefacts. All remaining images were then processed using a standardised pipeline (fmriprep 4 ). Briefly, the pipeline included slice time correction, nonlinear spatial normalisation to MNI space, brain tissue segmentation, susceptibility distortion estimation and resampling to 2mm. We used previously suggested stringent criteria for excluding and an additional 5 datasets on the basis of excess motion 5 . Automated ICA-based artefact removal (ICA-AROMA) 6 was applied, then averaged signals from the white matter, CSF, and entire brain were removed from voxelwise time series via linear regression, prior to detrending and high-pass filtering ( > .005 Hz). At each stage of pre-processing, quality control (QC-FC) metrics, FC matrices and carpet plots were visualised to ensure the processing step was having the desired effect of mitigating the impact of noise variables. A total of 193 scans survived our quality control procedure. A full quality control report can be accessed here: https://sidchop.github.io/STAGES_rs-fMRI/. To generate whole-brain FC matrices, we parcellated each individual's normalised scans into 300 cortical 7 and 32 subcortical regions 8 . We screened all regions for insufficient BOLD signal intensity by first calculating each region's mean BOLD signal across all 193 scans. We sorted the regional BOLD intensity values from largest to smallest and found the ''elbow'' of this distribution using the pairwise differences. This led us to exclude 16 cortical parcels, located across the orbitofrontal cortex and temporal pole. In the remainder of the analysis, we therefore used 316 regions. FC was estimated as the Pearson correlation between each pair of regional time series for each individual.

eMethods 6. Further Details on Statistical Analysis
Mixed-effects marginal models were used to analyse brain-wide FC changes across the three groups (MIPT, PIPT and controls) and three timepoints (baseline, 3-months and 12-months) 9 .
At each of the 99,856 edges linking 316 regions, we computed ordinary least squares estimators of group-level regression parameters and a robust-covariance estimator to account for within-subject correlations. This method allows for robust and accurate estimation of random effects while mitigating problems posed by misspecification of covariance structure when using traditional mixed-effects model 9 . The approach also allows for brain-wide nonparametric computation of -values at each edge using wild-bootstrapping (10,000 bootstraps) 10 . All code used to analyse data and generate figures can be accessed here: https://github.com/sidchop/Stages_rs-fMRI/.
The Network Based Statistic (NBS) was used to perform family-wise error-corrected (FWE) inference at the level of connected-components of edges showing a common effect, resulting in a substantial boost in statistical power compared to mass univariate analysis 11 . The NBS procedure involves setting a primary component-forming threshold, , which is applied to both the observed data, and the bootstrap-generated null data. The choice of this threshold is arbitrary; more lenient thresholds will be sensitive to weaker differences distributed over a large number of edges while more stringent thresholds will be sensitive to stronger effects possibly extending over smaller subsets of edges. We report here results for set to < 0.05 We evaluated NBS results using a Bonferroni-corrected threshold of < .016, adjusted for three key contrasts. This first contrast addressed our first aim and simply tested for baseline differences between healthy controls and patients, collapsed across MIPT and PIPT groups. The second and third contrasts addressed our second study aim. The second contrast was designed to isolate differential FC changes over time in the antipsychotic-naïve patient (PIPT) group compared to the healthy control group. This contrast was defined as a group by time interaction examining changes in PIPT patients (excluding 3-month scans of three patients who were exposed to antipsychotic medication during the treatment period) compared to controls. Two individuals were exposed to antipsychotics at doses lower than the minimal exposure limit for inclusion in our study. Excluding these participants from the analysis yielded a qualitatively similar pattern of results. The third contrast was designed to isolate the specific effects of antipsychotic treatment by examining differential All contrasts were adjusted for age, sex, and mean framewise displacement (head motion).
All continuous covariates were centred. Age was split into between-and within-subject components, with both included as covariates in the design matrix 9 .
To comprehensively delineate changes in FC across 49,928 different connections, present the results at three different scales: (1) the individual edge level, embedded in the spatial layout of the brain (e.g., Fig 1a; panel 1); (2) a level in which different regions are aggregated in one of 10 canonical brain networks, and we show the proportions of affected edges both within and between these networks (e.g., Fig 1a; panel 2); and (3) the level of individual brain regions, to identify specific brain areas which had a high number of significant connections (e.g., Fig 1a; panel 3).

Analysis of network level effects.
To determine whether the observed FC changes showed any network-specificity, we calculated the proportion of edges within a given NBS component that fell within each of seven canonical brain networks 29 (e.g. Fig 1a - linear trend across all three timepoints. We constrain our contrasts in this way, because our hypotheses concern linear interactions between group and time over the follow-up period. The treatment period for the trial ended after 6-months and four PIPT patients commenced antipsychotic medication in this intervening period, in addition to the four patients who had commenced at the 3-month timepoint. Thus, between the 3-month and 12-month scan, a total of eight PIPT patients commenced antipsychotics, whereas all MIPT patients continued medication with varying degrees of exposure. We therefore specified a covariate quantifying cumulative exposure to antipsychotics (olanzapine equivalent, milligrams) for all eight patients within the PIPT group who were exposed to medication at the 3-month or 12-month timepoint. This procedure allowed us to statistically adjust for antipsychotic exposure in the PIPT group when attempting to disentangle the long-term change in FC in antipsychoticnaïve and antipsychotic-treated patients.

eMethods 7. Demographic and Sample Characteristics
We have previously reported some demographics and clinical characteristics of this cohort 14 , but there are subtle differences in current sample compared to the previously reported sample (Table 1). Briefly, we detected no significant differences at baseline between patients and controls, but the patients were on average, younger and less educated. At baseline, the two patient groups (PIPT and MIPT) did not significantly differ in age, education, sex, handedness, BPRS or SOFAS scores. Additionally, the two patient groups did not Using -thresholds of < .01 and < .001, we observe that the strongest FC reductions in patients are concentrated in the limbic network and striatum, while the strongest FC increases occur between the thalamus and visual network (Sup. Fig 3).  Fig 3). eAppendix 3. Antipsychotic-Related Results (Baseline to 3 mo) Using -thresholds of < 0.01 and < 0.001, we find evidence that the strongest medication-related FC increases are concentrated in the visual, somatomotor, and attentional networks (Sup. Fig 3). This result aligns with our CCA of antipsychotic-naive changes over time, in which increased ∆FC in sensory networks is associated with improved functioning and symptoms over the first 3 months of illness.