Differentiating the Effect of Medication and Illness on Brain Volume Reductions in First-Episode Psychosis: A Longitudinal, Randomized, Triple-blind, Placebo-controlled MRI study

Study Design

The study was a triple-blind randomised placebo-controlled trial of antipsychotic-naïve first episode psychosis (FEP) patients (Figure 2). Patients were randomized to one of two groups: one given antipsychotic medication plus intensive psychosocial therapy (MIPT) and the other given a placebo plus intensive psychosocial therapy (PIPT). A third healthy control group who received no intervention was also recruited. For both patient groups, the treatment period spanned six months. MRI and clinical assessments were conducted at baseline, three months, and a final follow-up at 12 months. The randomization phase of the study terminated at 6 months, so patients in either the MIPT or PIPT group could have received antipsychotics and ongoing psychosocial interventions in between the 6 and 12 months into the study. The trial took place at the Early Psychosis Prevention and Intervention Centre, which is part of Orygen Youth Health, Melbourne, Australia. The trial was registered with the Australian New Zealand Clinical Trials Registry in November 2007 (ACTRN12607000608460) and received ethics approval from the Melbourne Health Human Research and Ethics committee. A detailed overview of the research and safety protocols of the study has been previously published²¹.

• Download figure
• Open in new tab

Figure 2

Recruitment flow diagram for the patient group.

Participants

Patients were aged between 15 and 25 and were experiencing a first episode of psychosis, defined as fulfilling Structured Clinical Interview for DSM-5 (SCID) criteria for a psychotic disorder, including schizophrenia, schizophreniform disorder, delusional disorder, brief psychotic disorder, major depressive disorder with psychotic symptoms, substance-induced psychotic disorder or psychosis not otherwise specified. Additional inclusion criteria to minimise risk were: ability to provide informed consent; comprehension of English language; no contraindication to MRI scanning; duration of untreated psychosis (DUP) of less than 6 months; living in stable accommodation; low risk to self or others (score of < 5 on the Brief Psychiatric Rating Scale version 4 (BPRS-4) Suicidality and Hostility subscales); minimal previous exposure to antipsychotic medication (less than 7 days of use or lifetime 1750mg chlorpromazine equivalent exposure; further details provided in Supplementary Table 1). Additional safety measures, including discontinuation criteria, were also applied and are outlined elsewhere²¹. A recruitment flow diagram and final group numbers at each time point are presented in Figure 2. Additional demographic characteristics of the groups are presented in Table 1.

Healthy control participants were aged between 18 and 25, could provide written informed consent, and were psychiatrically, neurologically and medically healthy, as determined by self-report, the SCID, and Comprehensive Assessment of At-Risk Mental States (first four subscales). Additional exclusion criteria included personal or family history (first degree relative) of psychiatric and neurological illness or current use of psychotropic medications.

Randomisation and blinding

A stratified randomisation design was used to allocate patients to either MIPT or PIPT treatment groups. Gender and DUP were included as stratifying factors as they have been associated with treatment outcomes in FEP. DUP was included as a three-level factor (0-30 days, 31-90 days, and >90 days). Computerised randomisation was carried out by an independent person who had no contact with participants and communicated directly to the trial pharmacy. Clinicians, patients, and study assessors remained blinded to treatment allocation throughout the trial. Researchers conducting MRI pre-processing remained blinded to treatment allocation.

Symptomatic and functional measures

The preregistered primary and secondary trial outcome measures were the total Social and Occupational Functioning Assessment Scale (SOFAS) and the BPRS-4 scores respectively²¹. Additional measures included the Scale for Assessment of Negative Symptoms (SANS), Hamilton Depression Scale (HAM-D) and Hamilton Anxiety Scale (HAM-A), and the World Health Organisation Quality of Life Scale -Brief (WHOQoL-BREF). Substance use was measured using the World Health Organisation Alcohol, Smoking and Substance Involvement Screening Test (WHO-ASSIST). Duration of untreated psychosis (DUP) was assessed using a clinical interview.

Antipsychotic Medication

Patients randomised to the MIPT group received either 1mg risperidone (n=25) or 3mg paliperidone (n=5). To reflect real-world clinical treatment, this starting dose was then increased according to clinical response by the blinded treating clinician. The same procedure was followed for participants in the PIPT group, who received a placebo pill that was identical in taste, appearance, and packaging to the active medication. Additional details can be found in Supplementary materials.

MRI acquisition and pre-processing

A 3-T Siemens Trio Tim scanner located at the Royal Children’s Hospital in Melbourne, Australia, was used to acquire a high resolution structural T1-weighted scan for each participant. Pre-processing of scan was done using the Computational Anatomy Toolbox and Diffeomorphic Anatomical Registration Exponentiated Lie algebra algorithm (DARTEL)²³. Additional details can be found in Supplementary materials.

Statistical analyses

Mixed effects marginal models were used to analyse regional GMV across the three groups (MIPT, PIPT and Healthy control) and three-time points (baseline, three-months and 12-months follow-up). The models were implemented at voxel-level in the Sandwich Estimator Toolbox²⁴ (version 2.1.0). All other statistical analyses were conducted in R-studio (version 1.1.423).

Our primary analyses sought to disentangle the effects of medication and illness (e.g., Figure 1) on total GMV and to map localised changes using VBM. This analysis focused on the baseline and 3-month timepoints, as they fell within the treatment period. For both total and regional GMV, we first tested for baseline differences between groups using an analysis of covariance (ANCOVA), controlling for age at baseline, sex, handedness and total intracranial volume. We then examined longitudinal changes using a marginal model. The voxel-level analysis was implemented in the Sandwich Estimator Toolbox²⁴ (version 2.1.0), which uses ordinary least squares estimators of group-level regression parameters and a modified sandwich estimator for standard errors²⁵. This method allows for robust and accurate estimation of random effects while mitigating problems posed by mis-specification of covariance structure when using traditional mixed-effects models²⁴. The contrast of interest was a group (MIPT, PIPT, Control) by time (baseline, 3-month) interaction. We performed inference using non-parametric bootstrapping (10,000 bootstraps), with statistical significance for total volume assessed at p <0.05, and a p<0.05, family-wise error (FWE)-corrected threshold for voxel-level analyses, as implemented in the Sandwich Estimator Toolbox²⁴. To provide a more complete picture, we also report results surviving a less stringent threshold of p<0.001, uncorrected, with an extent threshold of 10 voxels, but caution that these findings require replication.

Our secondary analysis included the 12-month follow-up timepoint. The treatment period for the trial ended after 6-months. Thus, in principle, clinicians and patients were no longer bound by the treatment protocol after this point. In practice, 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 thus 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 effects of illness and medication exposure on GMV. To ensure that this approach did not substantially influence the results, we repeated the analysis after removing the eight antipsychotic-exposed patients from the PIPT group. The results were largely consistent (see Supplementary material).

For analyses including the 12-month scans, we used the same procedure as in the primary analysis, with the contrast of interest being a linear polynomial contrast examining differences in linear trend between the three groups. We constrain our contrasts in this way, because our hypotheses concern linear interactions between group and time over the follow-up period (as per Figure 1).

We assessed the functional impact of any regions showing a statistically significant group by time interaction by correlating percent change in GMV over time (measured by the first eigenvariate of the region, adjusted for covariates) with percent change in scores on the pre-registered²⁶ primary outcomes of the clinical trial; namely the SOFAS and BPRS-4 total scores. Associations were quantified using Spearman’s rank correlations with the threshold for significance set at p<.025 (Bonferroni-adjusted for two comparisons). Additional exploratory correlations between all available clinical and functional scales were Bonferroni-corrected for ten comparisons (p<.005).

We also investigated the potentially confounding effects of DUP, concomitant medication, and substance use on GMV change in each region showing a significant group x time interaction. Further methodological details can be found in Supplementary materials.

Demographics and clinical characteristics

There were no significant differences between the patient and control samples in sex or handedness, but the patients were, on average, 1.9 years younger and had 2 years less education (Table 1). At baseline, the two patient groups (PIPT and MIPT) did not significantly differ in age (t = -0.937, p = 0.353), education (t = -1.51, p= 0.122), sex (χ² = 0.020, p = 0.887), handedness (χ² = 0.001, p = 0.972), BPRS (t = 1.40, p = 0.166) or SOFAS (t = 0.384, p = 0.703) score.

Baseline differences in total and regional grey matter volume

No significant baseline difference in total GMV was detected between patients (collapsed across treatment groups) and healthy control participants (F = 0.297; p = 0.588), nor were any voxel-level regional differences detected following whole-brain FWE-correction. At an uncorrected threshold (k > 10, p < 0.001; Supplementary Table 3), patients showed reduced GMV within the right postcentral gyrus, right supramarginal gyrus, right frontal pole, right insula, middle temporal gyrus, and left hippocampus. As expected, no significant total (F = 0.530, p = 0.470) or voxel-level GMV differences were found between the MIPT and PIPT groups at baseline at either corrected or uncorrected thresholds.

Disentangling medication-related and illness-related brain changes in the first three months of treatment

No significant group by time interaction in total GMV was detected between the three groups (F = 0.387, p = 0.689). Using VBM to map regional changes, a significant group by time interaction was identified within the right pallidum (p < 0.05, FWE-corrected; Figure 3a). From baseline to 3 months, GMV in this region remained stable in controls, decreased in the PIPT group, and increased in the MIPT group (Figure 3b). Post-hoc contrasts revealed that compared to baseline, pallidal GMV significantly decreased in PIPT patients (t = 2.34, p = 0.021), increased in MIPT patients (t = -2.20, p = 0.029), and did not change in controls (t =-0.142, p = 0.888). Further post-hoc contrasts conducted at the 3-month timepoint showed that the PIPT had significantly less pallidal GMV than the MIPT (t = -2.26, p = 0.012), with neither group differing from controls.

• Download figure
• Open in new tab

Figure 3

(A): Red = Location of the cluster within the right pallidum where significant group x time interaction (p < 0.05, FWE-corrected) was detected. (B): The principal pallidal GMV eigenvariate for each group at baseline and 3-month follow-up, adjusted for model covariates. Error bars show 95% confidence intervals. (C): The association between percentage change (%Δ) in total Brief Psychiatric Rating Scale score (BPRS; y-axis) and percentage change in pallidal GMV volume within the two treatment groups.

Supporting a putative neuroprotective influence of antipsychotics (e.g., Figure 1d), greater increase in pallidal GMV over 3 months was associated with a greater reduction in symptom severity, as indexed by the BPRS Total score (ρ = -0.418; p = 0.017; Figure 3c). There was no significant association with SOFAS total score (ρ = -0.002; p = 0.998), and none of the exploratory correlations with ancillary measures survived correction for multiple comparisons. At an uncorrected threshold, we found a negative correlation between right pallidal volume and BRPS positive symptom change scores that was comparable in magnitude to the association with BPRS total (ρ = -0.431; p= 0.012), suggesting that the relationship between pallidal volume and symptom change may be specifically related to positive symptoms.

At an uncorrected threshold (k > 10, p < 0.001), we identified interactions between group and time in the left inferior temporal cortex, precuneus, lateral occipital, cerebellum and right orbitofrontal cortex (Figure 4). These effects were consistent with a unmodified illness-related effect within lateral occipital cortex (Figure 4a, Figure 1b); a putative neurotoxic effect within cerebellum (Figure 4b, Figure 1a); and medication-induced hypertrophy within the inferior temporal cortex, precuneus, and orbitofrontal cortex (Figure 4c-e, Figure 1d).

• Download figure
• Open in new tab

Figure 4

Red clusters indicate anatomical locations where significant group by time interactions (p < 0.001, uncorrected) were detected between baseline and 3 months. Bottom row shows the nature of the interaction. A) left lateral occipital cortex; B) left cerebellum; C) left inferior temporal cortex; D) left precuneus; E) right orbitofrontal. Abbreviations: PIPT = placebo plus intensive psychosocial therapy, MIPT = antipsychotic medication plus intensive psychosocial therapy.

Potential confounds

No statistically significant associations between percentage change in pallidal volume between baseline and 3 months and DUP, concomitant medication use, or substance use were detected. Details of results can be found in Supplementary materials.

Long-term effects of medication and illness on grey matter volume

We next considered MRI measures at the 12-month follow-up to differentiate the long-term effects of medication and illness. Patients retained at the 12-month follow up did not significantly differ in baseline age (t = 1.7858, p = 0.08), sex (χ² = 0.087, p = 0.767), education (t = 0.652, p = 0.518), BPRS (t = -0.076, p = 0.940) or SOFAS (t = 0.780, p = 0.439) from those who did not complete the 12-month follow up scan.

No statistically significant differences in linear trend for total GMV over the 12-month follow-up period were detected (F = 1.60, p = 0.192). At the voxel-level, no statistically significant regional differences were identified at the FWE-corrected threshold. At p < 0.001 (k=10) uncorrected, there were differences in linear trend within the bilateral dorsolateral superior frontal gyrus, right superior and middle orbito-frontal gyrus, and left superior medial frontal gyrus, right cerebellar crus, right middle temporal gyrus, temporal pole and cerebellar VIII (k > 10, p < 0.001; Figure 5). These effects were consistent with unmodified illness-related changes (Figure 1b) within the bilateral dorsolateral superior frontal gyrus (Figure 5a-b), right superior orbito-frontal gyrus (Figure 5c), middle orbito-frontal gyrus (Figure 5d), and left superior medial frontal gyrus (Figure 5e); a putative neurotoxic effect (Figure 1a) within the right cerebellar crus I (Figure 5f); and medication-induced hypertrophy (Figure 1d) within the right middle temporal gyrus (Figure 5g), temporal pole (Figure 5h) and cerebellar VIII (Figure 5i). The results were largely consistent when eight individuals within the PIPT group who were exposed to antipsychotic medication were removed from the analysis (Supplementary Figure 1).

• Download figure
• Open in new tab

Figure 5

Red clusters indicate anatomical locations where differences in linear trend (p < 0.001, uncorrected) were detected between baseline and 12 months. Bottom row of each panel shows the nature of the interaction. A) right dorsolateral superior frontal gyrus; B) left dorsolateral superior frontal gyrus; C) right superior orbito-frontal gyrus; D) right middle orbito-frontal gyrus; E) left superior medial frontal gyrus; F) right cerebellar crus I; G) right middle temporal gyrus; H) right temporal pole; I) right cerebellar VIII; Abbreviations: PIPT = placebo plus intensive psychosocial therapy, MIPT = antipsychotic medication plus intensive psychosocial therapy.

Additionally, we assessed whether the changes seen within pallidal cluster detected in the primary analysis persisted at 12-month follow-up. The differences between the three groups were not statistically significant (Supplementary Figure 2).

Assessing specificity of findings to grey matter

To assess the specificity of our findings to grey matter, we repeated the above primary and secondary analyses in white matter. For the primary analysis, we found a significant group by time interaction within a small area of the left cerebellar lobule IX white matter (k = 9, p < 0.05, FWE-corrected; Supplementary Figure 3a). From baseline to 3 months, white matter volume in this region increased in the controls (t = -2.34, p = 0.021), remained stable in the PIPT group (t = 0.216, p = 0.830), and decreased in the MIPT group (t = 2.239, p = 0.027). This pattern of results is consistent with a putative neurotoxic effect of antipsychotics (e.g., Figure 1a). The change in volume within this cluster was not correlated with change in BPRS-4 or SOFAS.

At an uncorrected threshold (k > 10, p < 0.001), interaction effects were detected within the white matter of the right cerebellar lobule V and crus II. These effects were consistent with a putative neurotoxic effect and an unmodified illness-related change, respectively (Supplementary Figure 3b-c).

No significant interactions were detected when including the 12-month time point. At an uncorrected threshold (k > 10, p < 0.001), an interaction effect consistent with a unmodified illness-related change was detected within left frontal white matter (Supplementary Figure 4).

Illness-related volumetric reductions in FEP

Pallidal volume in the PIPT group declined over the first three months of illness. This decline was not associated with substance use or concomitant medication, which would be consistent with an illness-related effect. In contrast, MIPT patients showed an increase in pallidal volume over time. Thus, antipsychotics appear to prevent or perhaps even reverse illness-related volume loss in this part of the brain.

Using less stringent criteria for significance, we found evidence for illness-related GMV reductions in visual cortex within the first three months of illness, and further reductions in prefrontal cortex by the 12-month time point. Frontal white matter reductions were also identified at the 12-month time point. These changes were observed in both the PIPT and MIPT groups, suggesting that they are unmodified effects of illness.

The pallidum is the primary output structure of the striatum, and disturbances of fronto-striato-thalamic circuits have long been implicated in the pathogenesis of psychosis²⁸. The function of these circuits is heavily modulated by dopamine, and their disruption is apparent in diagnosed patients²⁹, patients’ unaffected first-degree relatives²⁹ and individuals experiencing an at-risk mental state for psychosis³⁰. Functional connectivity within this circuit also correlates with the severity of psychotic-like experiences in non-clinical samples³¹. Thus, one hypothesis that may explain our findings is that altered signalling from the striatum triggers early volumetric loss in the pallidum^28,32, which subsequently spreads to affect functionally-related prefrontal areas. This interpretation aligns with evidence that smaller pallidal volume in anti-psychotic naïve patients is associated with more severe psychiatric symptomology³³, and our own finding that increased pallidal volume over the first three months correlates with improved symptom outcome. Moreover, our finding of possible long-term reductions in prefrontal cortex grey and white matter may explain why prefrontal dysfunction is so commonly reported in patients with established illness. However, we caution that the precise mechanisms underlying volumetric changes in psychosis remain a topic of debate^8,9. Here, we show that some of these changes cannot be attributed to medication and other confounding factors, but more subtle influences such as differences in hydration, physical and mental activity, and stress levels cannot be ruled out. Targeted mechanistic studies are required before we can draw strong inferences about pathophysiological mechanisms.

Evidence for putative antipsychotic-induced neuroprotection

The increase of pallidal volume seen in MIPT patients, together with the correlation between increased pallidal GMV and symptom improvement between baseline and three months, are consistent with a putative neuroprotective effect of atypical antipsychotics (Figure 1d). Larger pallidal volumes have been widely reported in medicated³⁴ but not antipsychotic naive³⁵ patients. Volume increases within the basal ganglia were initially thought to be induced only by typical antipsychotics, but more recent studies have demonstrated a similar enlargement following exposure to atypical antipsychotics³⁶. Human studies have also shown reduced volume loss in patients taking atypical antipsychotics compared to patients receiving typical antipsychotics¹⁶, and work in animals indicates that atypicals can exert several neuroprotective effects¹⁷, including induction of neurogenesis¹⁸, and protection against oxidative stress¹⁹, as well as positive effects on cognition³⁷. Our results are in line with this work and suggest that atypicals prevent illness-related volume loss occuring early in the illness. However, the molecular mechanisms by which atypical medications might protect grey matter structures in humans are poorly understood. Atypicals are characterized by relatively high affinities for both serotonin and dopamine receptors³⁸. Patients in our study received risperidone or its molecularly similar active metabolite paliperidone. Both medications are antagonists for 5HT₂ receptors in addition to showing high affinity for D₂ receptors³⁹. Rodent studies using risperidone have demonstrated cell proliferation⁴⁰, increased levels of brain-derived neurotrophic factor⁴¹, and the promotion of antioxidant defence⁴².

Notably, pallidal volume had normalized in both MIPT and PIPT patients by 12 months (Supplementary Figure 3). While it is possible that this normalisation reflects differences in illness characteristics between patients who did and did not complete the 12 months follow-up, we found no significant difference in baseline demographic and clinical characteristics between these two groups. Nonetheless, it is possible that patients completing the 12-month assessment followed distinct illness trajectories after enrolment into the study. An alternative explanation is that early pallidal changes reflect an acute illness effect with subsequent normalisation reflecting a compensatory process. It is also possible that intensive psychosocial therapy and engagement with clinical services was sufficient to normalize volumes⁴³.

Evidence of putative neurotoxicity related to antipsychotics

We found no evidence for antipsychotic-related neurotoxicity in GMV at whole-brain-corrected thresholds. At the less stringent, uncorrected threshold, we observed consistent GMV changes in the cerebellum at 3 months and at 12 months that are suggestive of a putative neurotoxic effect (Figure 1a). Similar results surviving whole-brain correction were observed in cerebellar white matter. While previous naturalistic⁴ and experimental¹⁵ studies have demonstrated an association between antipsychotics and loss of both total and hemispheric volume, our study is distinct in several important ways. First, our study included a placebo control and healthy control, allowing us to experimentally isolate the effect of atypical antipsychotic medication. Second, we examined patients during a relatively short time span of one year, while other studies have examined longer periods⁴. Third, the mean cumulative dosage of antipsychotics in our study, while still an effective dose⁴⁴, is considered low. Forth, all participants in our study were scanned with the same scanner, reducing confounding variance introduced in multi-scanner studies. Finally, all patients within our study received an evidence-based psychosocial intervention which may have associated neuroprotective effects⁴³. In light of these differences, our results should be interpreted as providing suggestive evidence of potential neurotoxicity in early illness stages that is predominantly expressed in the cerebellum. However, we note that there was no association between volume change within the cerebellum and change in functional or symptom outcome scores, in either grey or white matter. Longer-term follow-up would be required to determine the extent to which further volume-loss emerges with additional antipsychotic exposure.

Strengths and limitations

The strengths of this study include a prospective randomised control trial design, antipsychotic naïve patients, triple blinding to treatment, and the inclusion of a healthy control group as a reference for characterizing normative change over time. We also used robust non-parametric inference to model longitudinal changes in GMV²⁴. In order for this study to satisfy ethical concerns, our inclusion criteria meant that patients who entered the study posed low risk of harm to self or others, lived in stable accommodation, and had a short DUP. Additionally, patients who did not improve in clinical symptomology or functioning were removed from the trial, which contributed to attrition. It is therefore possible that our final patient cohort represents a sub-sample of individuals with a lower severity of illness, and that the changes we report here are a conservative estimate of those that would be observed in a more heterogeneous sample. However, we note that the mean baseline SOFAS scores of our patients were comparable to epidemiologically representative cohorts of FEP patients⁴⁵ and that the mean baseline BPRS score of patients within our study (57.6) would classify them as ‘markedly ill’⁴⁶. Additionally, there were no differences in baseline clinical or demographic characteristics between patients who did and did not complete the study. Thus, there are no obvious differences between our cohort and many FEP samples reported in the literature, beyond the strict safety requirements of our study. Nonetheless, we cannot rule out the possibility that patients who remained in the study have a form of psychotic illness that is perhaps less severe and/or progressive than those who did not complete. A final limitation is that we only examined risperidone and paliperidone. It remains to be seen whether our results generalize to other atypical antipsychotics.