Altered Temporal Dynamics of Resting-State Functional Magnetic Resonance Imaging in Adolescent-Onset First-Episode Psychosis

Abstract

Background

Dynamic functional connectivity (dFC) alterations have been reported in patients with adult-onset and chronic psychosis. We sought to examine whether such abnormalities were also observed in patients with first episode, adolescent-onset psychosis (AOP), in order to rule out potential effects of chronicity and protracted antipsychotic treatment exposure. AOP has been suggested to have less diagnostic specificity compared to psychosis with onset in adulthood and occurs during a period of neurodevelopmental changes in brain functional connections.

Study Design

Seventy-nine patients with first episode, AOP (36 patients with schizophrenia-spectrum disorder, SSD; and 43 with affective psychotic disorder, AF) and 54 healthy controls (HC), aged 10 to 17 years were included. Participants underwent clinical and cognitive assessments and resting-state functional magnetic resonance imaging. Graph-based measures were used to analyze temporal trajectories of dFC, which were compared between patients with SSD, AF, and HC. Within patients, we also tested associations between dFC parameters and clinical variables.

Study Results

Patients with SSD temporally visited the different connectivity states in a less efficient way (reduced global efficiency), visiting fewer nodes (larger temporal modularity, and increased immobility), with a reduction in the metabolic expenditure (cost and leap size), relative to AF and HC (effect sizes: Cohen’s D, ranging 0.54 to.91). In youth with AF, these parameters did not differ compared to HC. Connectivity measures were not associated with clinical severity, intelligence, cannabis use, or dose of antipsychotic medication.

Conclusions

dFC measures hold potential towards the development of brain-based biomarkers characterizing adolescent-onset SSD.

Introduction

First-episode psychosis often presents with a combination of psychotic and affective symptoms and longitudinal assessments may be necessary to establish diagnosis. This holds treatment and prognostic implications and is especially relevant in adolescent-onset psychosis (AOP) since onset of psychosis during adolescence has been associated with a lower level of diagnostic specificity of clinical presentations than when it occurs in adulthood.^1,2 However, there is still a lack of objective, brain-based biomarkers with diagnostic and prognostic potential in psychiatry.

The Disconnection Hypothesis³ suggests that the primary pathophysiological mechanism underlying psychosis is synaptic in nature. Functional connectivity abnormalities measured using resting-state functional magnetic resonance imaging (rs-fMRI), which quantifies the blood oxygen level-dependent (BOLD) signal of the brain in the absence of a task, have been observed both in schizophrenia-spectrum disorders (SDD) and affective psychoses (AF, bipolar, and depressive disorders with psychotic symptoms), including a diffuse alteration of brain connectivity^4–7 and a reduction in the number of hubs or regions with a significant contribution to brain functional networks.^8,9 However, the reported alterations are not consistent across all studies¹⁰ and they do not permit to disentangle whether the observed rs-fMRI changes are caused by synaptic abnormalities or structural constraints. Methods capable of profiling rs-fMRI dynamics, known as dynamic functional connectomics (dFC), have the potential to better describe the non-stationary¹¹ nature of brain connectomics, which are likely to be especially prominent when mental activity is unconstrained.¹² Additionally, dFC methods are considered less dependent on structural connections and more capable to identify changes specifically related to neural population dynamics.¹³

Most dFC studies to date focused on the spatial organization of different temporal connectivity patterns, activated during the rs-fMRI sequence. These studies reported a higher recurrence of hypo-connected states in patients with SSD compared to healthy controls (HC),^14,15 and to a lesser extent in AF patients.¹⁶ Fewer studies have examined the temporal component of such transitions. Miller et al,¹⁷ for example, reported a smaller number of dFC transitions between meta-states in patients with chronic schizophrenia, and a larger similarity between meta-states, which was more pronounced in cases with more severe psychotic symptoms. A recent study examined adult patients with first-episode psychosis,¹⁸ reporting higher segregation, less efficiency, and greater redundancy in the flow of temporal networks. These findings were associated with antipsychotic doses and were not observed in a small group of antipsychotic naive patients.

The study of first-episode AOP patients has the potential to help understand how the brain functions during rest, without the confounding effects of chronicity¹⁷ and protracted exposure to antipsychotic medication, while focusing on a crucial period in terms of development of brain networks. Gozdas et al,¹⁹ analyzed the progression of static functional connectivity in healthy adolescents, and reported a differentiation process of spatial functional clusters, along with an enhancement of the general efficiency of the network with age. Similarly, Lopez-Vicente and colleagues,²⁰ who specifically analyzed dFC in a sample of healthy youth, also identified an increase in the activation of the same connectivity patterns for longer periods of time over adolescence and observed that older youth spent longer time in more defined or spatially clustered meta-states. However, it remains unclear whether the onset of psychosis during adolescence can impact on physiological changes in brain network dynamics. On the other hand, grouping AOP patients according to their confirmed diagnosis over time may help dissect potentially heterogeneous pathophysiological mechanisms. Studies in AF and SSD psychoses have suggested that they present some degree of specificity in brain phenotype,¹⁴ but this has been subject to limited study and has not been examined from a dFC perspective so far.^21,22

We, therefore, set out to examine potential differences in the temporal component of dFC, analyzed using a graph theory approach,²³ between patients with AOP divided according to diagnosis (AF vs SSD). As a secondary aim, we sought to examine whether these measures were associated with clinical characteristics of our sample, including clinical and functional severity, cannabis use, and dose of antipsychotic medication.

Methods

Data Acquisition

Seventy-nine patients with first-episode AOP aged 10 to 17 years, were recruited at the Department of Child and Adolescent Psychiatry and Psychology of the Hospital Clinic of Barcelona (Spain). Diagnosis of first episode of psychosis was established at first contact with mental health services and defined as the presence of positive psychotic symptoms of <12 months duration with an onset prior to age 18 (for details on baseline recruitment and assessment see²⁴).

Fifty-four age-matched HC were recruited within community settings from the same geographical area. General exclusion criteria were presence of autism spectrum disorders, posttraumatic stress disorders, and drug-induced psychosis; intellectual disability as defined by DSM-5 criteria²⁵; other neurological disorders or history of head trauma with loss of consciousness; pregnancy; and medical or technical counter-indications for MRI. Additional exclusion criteria for HC participants were a current axis I psychiatric diagnosis and having first- and second-degree relatives with any psychotic disorder.²⁶

The study was approved by the local Ethics Review Board. All parents or legal guardians and participants over age 12 signed informed consent or provided their assent prior to inclusion in the study. At baseline, all participants underwent a demographic and clinical assessment by experienced mental health professionals. This included the Kiddie Schedule for affective disorders and schizophrenia semi-structured interview in its Spanish version,²⁷ administered to participants and their parents or legal guardians. The latter was repeated at 6 months follow-up to sub-divide patients into SSD (schizophrenia, schizophreniform, and schizoaffective disorders) and AF (bipolar disorder or depressive disorder with psychotic symptoms), according to DSM-5 criteria.²⁵ Clinical severity at the time of scanning was evaluated using the Positive and Negative Syndrome Scale (PANSS)²⁸ and the General Assessment of Functioning (GAF) Scale. Details on current cannabis use were also recorded (categorized dichotomously as absence or presence of any use, within the 6 months prior to study intake) as well as type and dose of antipsychotic medication at the time of scanning, converted to chlorpromazine equivalents.²⁹

Participants also underwent a cognitive assessment with the Wechsler Intelligence Scale for Children Fourth Edition (WISCIV)³⁰ or Wechsler Adult Intelligence Scale–III, revised,³¹ when older than 16 years. Results were used to obtain their global Intelligence Quotient (gIQ) derived from the verbal comprehension and perceptual reasoning indices.³²

At intake, all participants underwent a scanning session on a 3Tesla scanner (Magnetom Trio or its upgrade, Magnetom Prismafit, Siemens, Erlangen), at the Magnetic Resonance Image Core Facility of IDIBAPS, Center for Image Diagnosis, Hospital Clínic of Barcelona, which included a structural T1-weighted sequence, used for reference purposes, and then an 8-minute rs-fMRI acquisition with eyes closed. A technician engaged in conversation with the participants before and after the rs-fMRI session to guarantee that they did not fall asleep. Acquisition parameters were as follows: 240 volumes; TR, 2000 milliseconds; TE, 29 milliseconds; matrix size, 480 × 480; slice thickness, 4 mm; acquisition matrix, 80 × 80 mm²; 32 slices; voxel size, 3 × 3 × 4 mm³. All MRI scans were reviewed by an experienced neuroradiologist to rule out structural pathology.

fMRI Preprocessing

Raw rs-fMRI signal preprocessing is crucial for studies assessing dFC, as subject motion can bias dFC results.³³ Preprocessing for motion correction was performed based on the pipeline proposed by Parkes and colleagues,³⁴ which is further described in supplementary material 1.

Network Construction

To analyze dFC, directional graphs were constructed for each subject following the pipeline proposed by Ramirez-Mahaluf et al.²³ Briefly, using the atlas employed by Crossley and colleagues,³⁵ rs-fMRI signals were extracted from similarly sized regions (ROIs), which were used to generate connectivity matrices at each time step, using the multiplication of temporal derivatives (MTD).³⁶ These matrices were classified into more general types (ie, meta-states) and the temporal sequence of these labels was read to generate a directed graph, providing information about how many times during the rs-fMRI the brain jumps from one specific meta-state to the others. figure 1 schematizes this process, which is further explained in next sections.

Fig. 1.

Scheme of the pipeline followed for each subject, and predefined number of meta-states.

Time-Series Extraction and Connectivity Matrix Construction.

Preprocessed EPI volumes of each of the subjects were parceled into a total of 638 similarly sized ROIs, respecting anatomical landmarks,³⁵ so that a time series per region was extracted. To construct connectivity matrices, MTD³⁶ was the method chosen, because of its sensitivity to small variations. The main idea is that 2 connected regions should undergo similar changes, as opposed to those which are not. For a more robust dFC estimation, every 2 consecutive MTD matrices were averaged, obtaining 117 connectivity states characterized by 638 × 638-sized matrices.

Clustering and Definition of Optimal Number of Meta-states.

MTD correlation matrices were then clustered into general types according to their similarity, using K-means. An open question when using this approach^5,37 is the number of clusters or meta-states to consider. We here propose to use a data-driven approach (supplementary material 3), based on the idea that a specific number of meta-states is possible only if the quality of the identified meta-states is better than the one of those identified in a population of randomly generated connectivity matrices. Using this approach, we determined that the different clusters could be identified as meaningfully distinct, both for HC and AOP groups independently, only for the number of meta-states (kⁿ) contained in the range.^4,33 Therefore, our method identifies multiple possible sets of meta-states.

Transition Network Construction and Parameter Extraction.

For each of the sets, we constructed a graph, in order to capture how the brain temporally transitions across the set of identified whole-brain meta-states. For this purpose, labels corresponding to the cluster assigned via K-means clustering (ie, the meta-state type), using K = kⁿ, were assigned to the sequence of connectivity matrices. This vector of labels was then consecutively read, so as to estimate the transition from meta-state activated at time t_i and its consecutive meta-state at time t_i+1. An adjacency matrix (A_kn) was generated, of size kⁿxkⁿ (ie, where kⁿ is the total number of meta-states considered), which quantifies how many times the brain transitions from the whole-brain meta-state k_iⁿ to k_jⁿ in this specific order, for each position (i,j). Each graph was constructed from A_kn, and used to derive the different graph measures. These included modularity (capturing redundancy and temporal segregation of the flow), immobility (capturing the extent of the absence of transitions across different meta-states), global and local efficiencies (reflecting the ability to activate the repertoire of meta-states), transition cost and leap size (a measure of metabolic expenditure and its normalization, respectively), and cost-efficiency (measuring the balance between metabolic expenditure and ability to activate all meta-states). A detailed description of these measures can be found in supplementary material 5.

Since we ended up with several dFC descriptors estimations, one per each of the meta-state sets and subsequent graph, we reduced them into a single measure by computing the area under the curve of each of them with respect to the kⁿ within the feasible interval.

dFC Statistical Analysis

To test potential differences in extracted graph measures of AOP patients relative to HC, ANOVA tests were carried out to test the effect of group, adjusting for covariates (age, sex and frame-wise displacement, scanner model, and cannabis use). Interactions between these covariates and groups were also tested and were retained in the final model when significant or when variables were unequally distributed between groups. We then repeated the same analysis by further dividing AOP patients between SSD and AF.

To examine the relationship between these descriptors and clinical dimensions, general linear models were used to assess the effect of clinical variables (ie, PANSS positive and negative symptoms and GAF scores; current antipsychotic dose converted to chlorpromazine equivalents; gIQ) on each graph descriptor within the whole AOP sample, adjusting for the same covariates as in the previous experiment, and controlling for the interaction of each of the clinical variables with diagnosis. We also tested the association of gIQ in both the AOP and the HC groups. All P-values were corrected for multiple comparisons using false discovery rate. Given an upgrade during the study period, potential scanner effects were also examined (see supplementary material 4 for details).

Results

Sociodemographic and Clinical Variables

Sociodemographic and clinical measures are described in table 1. The 3 groups did not differ in terms of age, while there was a significant difference in sex distribution between AF and SSD groups, and a trend level increase of the frame-wise displacement in AF compared to SSD patients. gIQ was significantly different across the 3 groups, with significantly higher scores in HC compared to both SSD and AF, but also significantly higher scores in the AF group compared to the SSD group. A trend-level higher rate of cannabis use was observed in the SSD compared to the AF group. No significant difference was found in chlorpromazine equivalents of antipsychotic dose between AOP subgroups. PANSS negative scores were significantly higher in the SSD group compared to the AF group, while there was no significant between-group difference in either the PANSS positive subscale or the GAF.

Table 1.

Socio-demographic and Clinical Characteristics of the Sample

	HC (N = 54)	SSD (N = 36)	AF (N = 43)	P-value	Post hoc
Age	15.70 ± 2.42	15.70 ± 1.82	15.40 ± 1.38	.762	—
Sex (Female)	N = 27 (50.00%)	N = 12 (33.30%)	N = 28 (65.10%)	.019	SSD < AF (**)
Global intelligence quotient	107 ± 11.8	85.8 ± 13.9	94.4 ± 14.2	<.001	HC > SSD (**); HC > AF (**); SSD < AF (**)
Cannabis use	N = 8 (20.5%)	N = 12 (33.3%)	N = 5 (11.6%)	.063	SSD > AF (*)
FD	0.09 ± 0.07	0.08 ± 0.05	0.11 ± 0.08	.054	SSD < AF (*)
Scanner (Trio)	N = 36 (83.7%)	N = 31 (86.7%)	N = 29 (69.00%)	.121	—
Current antipsychotic dose	—	261 ± 187	299 ± 204	.405	—
PANSS positive symptoms	—	19.2 ± 4.63	20.0 ± 7.94	.581	—
PANSS negative symptoms	—	19.7 ± 6.89	15.8 ± 6.65	.013	—
GAF	—	45.8 ± 14.8	45.3 ± 15.2	0873	—

Note: HC = Healthy Controls; SSD = Schizophrenia-Spectrum Adolescent-onset First-Episode Psychosis patients; AF = Affective Adolescent-onset First-Episode Psychosis patients; PANSS = Positive and Negative Syndrome Scale; GAF = General Assessment of Functioning Scale; (**) P < .05; (*) 0.09 > P > .05.

Group Comparison in dFC measures

We observed no significant differences in the dFC measurements surviving multiple-comparisons correction between the entire AOP and HC. When dividing the AOP group into AF and SSD, and comparing the 3 groups, we found that SSD patients showed larger modularity, but reduced global efficiency in the transitions, cost, and leap size, when compared to both AF and HC groups. SSD also displayed larger immobility and decreased cost-efficiency when compared to HC. AF patients showed neither significant differences nor any trend level difference in these measures compared to HC. Estimated means and confidence intervals for the area under the curve for each of the dFC measures per group are described in table 2 and figure 2. We did not find any effect of age, sex, age or sex group interaction, cannabis use, frame-wise displacement, or scanner on the findings.

Table 2.

Estimated Means and Comparison of dFC Measures between Groups

	µHC ± SE (CI 95%)			Statistics
	HC	SSD	AF	F, corrected P-value	Post hoc
Q	4.32 ± 0.067 ([4.18, 4.45])	4.58 ± 0.065 ([4.45, 4.71])	4.35 ± 0.056 ([4.24,4.46])	F = 6.44 P = .006	SSD > HC (**); P = .007; D = 0.76 SSD > AF (**); P = .016; D = 0.73
GE	0.291 ± 0.020 ([0.251, 0.330])	0.211 ± 0.019 ([0.173,0.250])	0.273 ± 0.017 ([0.239, 0.306])	F = 6.05 P = .006	SSD < HC (**); P = .007; D = −0.71 SSD < AF (**); P = .022; D = 0.70
LE	0.731 ± 0.020 ([0.690, 0.771])	0.690 ± 0.020 ([0.651, 0.730])	0.725 ± 0.017 ([0.691, 0.758])	F = 1.77 P = .176	—
CO	170 ± 8.29 ([154,187])	137 ± 8.07 ([121, 153])	163 ± 6.94 ([149, 177])	F = 5.99 P = .006	SSD < HC (**); P = .007; D = −0.73 SSD < AF (**); P = .022; D = −0.64
Im	467 ± 9.58 ([448, 486])	502 ± 9.32 ([483, 520])	472 ± 8.02 ([456, 488])	F = 5.18 P = .010	SSD > HC (**); P = .010; D = 0.54 SSD > AF (**); P = .022; D = 0.58
LS	3.31 ± 0.13 ([3.04, 3.57])	2.78 ± 0.13 ([2.52, 3.04])	3.18 ± 0.11 ([2.96, 3.41])	F = 6.07 P = .006	SSD < HC (**); P = .007; D = −0.82 SSD < AF (**); P = .022; D = −0.91
CE	[10.5 ± 0.265]10−3 ([10.02,11.10]10−3)	[9.84 ± 0.258]10−3 ([9.33,10.40]10−3)	[10.36 ± 0.220]10−3 ([9.91,10.80]10−3)	F = 2.75 P = .080	SSD < HC (*); P = .050; D = −0.58

Note: HC = Healthy Controls; SSD = Adolescent-onset First-Episode Schizophrenia-Spectrum Disorders; AF = Adolescent-onset First-Episode Affective Psychosis patients; Q = Modularity; GE = Global Efficiency; LE = Local Efficiency; CO = Transition Cost; Im = Immobility; LS = Leap Size; CE = Cost-Efficiency; µ = mean; CI = Confidence Intervals; SE = Standard Error; (**) P < .05; (*) .09 > P > .05.D stands for Cohen’s d.

Fig. 2.

Estimated means and confidence intervals for the area under curve of each of the dynamic measures per group: HC (healthy controls, ), SSD (schizophrenia-spectrum disorder adolescent-onset psychosis (AOP), ) and AF (affective AOP, ). Uncorrected P-values for each group comparison are shown in each graph.

Relationship Between dFC Measures and Clinical Characteristics

We observed no association surviving multiple-comparisons correction between any network parameter and PANSS positive or PANSS negative and GAF scores in the AOP group. Similarly, we found no significant association between gIQ and dFC measures in either the AOP or the HC group. We also observed no association between graph measures and antipsychotic doses converted to chlorpromazine equivalents, cannabis use, frame-wise displacement (see also supplementary table S1), or scanner.

Discussion

In the present study, we initially compared whole-brain dFC between AOP and HC groups and observed no significant differences. When dividing the AOP group based on diagnosis, we observed changes in patients with SSD relative to both patients with AF and HC. Our results suggest that at illness onset, SSD patients showed greater redundancy in temporal trajectories, reflected through greater modularity in the flow between meta-states, and longer periods of immobility. We also identified greater difficulty in activating the full repertoire of possible meta-states, as reflected by reduced global efficiency. These changes were observed alongside alterations in measures associated with metabolism, as we found a decrease in the global metabolic demands and in those only associated with transitions. This not only suggests that there is a reduction in metabolic expenditure due to increased immobility rates, but also due to such transitions predominantly occurring between more similar meta-states. It is worth noting that the described measures are highly correlated, and from the graph theory perspective, an optimal increase in modularity with a subsequent decrease in the global efficiency could lead to a favorable decrease in metabolic demands. The ability to equally transition from one meta-state to any other may come with larger metabolic demands, especially if the meta-states involved are very dissimilar. Assigning a larger probability to transitions between more similar meta-states may come with conveniently lower metabolic demands, so that the brain would be transitioning in a cost-efficient manner.³⁸ In fact, Ramirez-Mahaluf et al²³ reported a positive association between cost-efficiency of transitions and gIQ in HC, confirming that networks displaying abnormal dFC may still be cost-efficient. However, in our study, this was not the case for SSD participants, who also displayed reduced cost-efficiency when compared to HC, confirming that these alterations reflect actual impairments. These results are consistent with dFC findings in a sample of adult patients with chronic schizophrenia,¹⁷ and recently, in a first-episode psychosis sample.¹⁸ Thus, our results further support the idea that these changes are already present at the beginning of the clinical course of the disease.²

In contrast with the SSD group and contrary to our prediction, we found no evidence of significant dFC alterations in AF patients compared to HC. AF patients differed from SSD patients by displaying preserved dynamic fluidity (immobility), a greater variety in mental transitions (modularity), and an associated higher metabolic expenditure (cost and leap size). Previous studies in chronic patients reported that AF patients displayed intermediate alterations between patients with SSD and HC.¹⁶ This contrasts with our findings and raises the possibility that dFC abnormalities in AF patients may manifest later over the course of disease. Such possibility would be in keeping with the suggestion that neurodevelopmental mechanisms are more pronounced in the pathophysiology of SSD compared to AF³⁹ but would need to be tested systematically in a sample of patients assessed at different illness stages.

Of particular interest is the relationship between antipsychotic treatment and dFC measures. Previously, Lottman and colleagues⁴⁰ examined dFC in first-episode patients with schizophrenia while unmedicated and after 6 weeks of treatment, describing the dwell and fraction of time spent in differently connected general meta-states. They found that before treatment, patients with schizophrenia tended to spend less time in poorly connected meta-states than HC, but such dwell time stabilized after treatment. The authors tentatively attributed the initial reduction of time in hypo-connected meta-states to glutamatergic hyperactivity that stabilized after treatment. This hypothesis could also apply to our findings for SSD. In fact, Ramirez-Mahaluf et al¹⁸ reported in their adult FEP sample, which predominantly included SSD patients, similar findings to those in our SSD group, and an association between such measures and dose of antipsychotic treatment.

However, in our sample, which had received lower doses of antipsychotic treatment, we did not observe this association. Furthermore, while doses of medication were statistically equivalent between SSD and AF groups, we only observed dFC impairments in SSD participants. Our results, therefore, extend the findings by Ramirez-Mahaluf et al¹⁸ and Lottman et al,⁴⁰ by suggesting that changes in dFC in SSD patients with first-episode psychosis with an onset during adolescence are unlikely to be simply an epiphenomenon of antipsychotic treatment.

Consistent with Ramirez-Mahaluf et al,¹⁸ we observed no significant association between dFC descriptors and gIQ in AOP. However, contrary to our prediction, we failed to observe this association in the HC group, which was previously reported in a study²³ using a composite cognitive measure obtained via principal component analysis. This methodological difference may partially explain the difference in the results. The gIQ may suboptimally characterize the variability present in our sample when compared to a composite measure tailored to the sample. This could also suggest that dFC descriptors could correlate differently to each of the original cognitive dimensions. We also observed no significant association between dFC descriptors and measures of clinical severity. Taken together, our results might suggest that dFC measures may perform better at capturing underlying pathophysiological mechanisms, rather than directly reflecting specific clinical or cognitive domains. In view of the results and to understand what these particular dFC measurements represent, future studies should focus on their correlation with measures closer to pathophysiology, such as neurometabolites, and/or structural and effective connectivity.

Several methodological considerations need to be acknowledged when interpreting the findings of this study. First, we cannot rule out that the lack of case–control findings in AF or the association of dFC measures to clinical and demographical measures are due to insufficient power to detect small effects. A larger sample size per group would help to interpret our findings. Second, the choice of method for analyzing rs-fMRI data may have influenced our results: Unlike other approaches, this method is able to capture changes that take place in the temporal dimension of rs-fMRI; however, these are also summarized in mean measures which span across time and structures. This can be advantageous, as it enables one to easily describe a structure that is organized in the form of a complex network; however, average measures may also mask other biologically relevant information. It is also important to acknowledge that there are various approaches to dFC analysis in the literature which follow different methodological choices throughout the pipeline, ranging from the definition of the regions from which to extract the BOLD signal, to the method to compute and identify the different connectivity patterns and meta-states. One of the strengths of our study is the data-driven choice of the number of meta-states. However further studies should be carried out to quantify the influence of other methodological decisions on the quality and robustness of the result. Finally, we chose to focus on a sample that was clinically representative of youth with a first episode of psychosis, therefore use of cannabis was not an exclusion criterion, although drug-induced psychosis was. There was a sex imbalance between patient groups, which is consistent with an earlier age onset of SSD in males, while all analyses included sex as covariate, and no sex effects were detected. Patients had a recent onset of the disease and had received overall a short exposure to medication, although we examined the effect of antipsychotic medication quantitatively to measure its effects on our findings. It is extremely challenging to scan adolescents with a first episode of psychosis naïve to antipsychotic (or other sedative) medications and is not an option in our clinical setting.

To conclude, during rest, adolescents with SSD transitioned between mental meta-states in a less random manner and with reduced metabolic cost relative to HC, possibly reflecting inefficiency in the activation of different connectivity patterns. These alterations are already present at illness onset, are not likely to be simply an epiphenomenon of antipsychotic treatment, and are independent of clinical severity, suggesting that a combination of the described dFC measures holds potential as a brain-based biomarker, specifically characterizing youth with SSD.

Funding

Grants from the Spanish Ministry of Health, Instituto de Salud Carlos III supported by ERDF Funds from the European Commission (PI18/0242/PI21/0391). GS also received grants from the Spanish Ministry of Health, Instituto de Salud Carlos III supported by ERDF Funds from the European Commission (PI18/00976; 21/00330), the Alicia Koplowitz Foundation, the Fundació Clínic Recerca Biomèdica and Brain and Behavior Research Foundation (NARSAD Young Investigator Award 26731). GP is supported by ICREA Academia programme. This work has been performed thanks to the 3T Equipment of Magnetic Resonance at IDIBAPS (project IBPS15-EE-3688 cofounded by MCIU and by ERDF). The funding sources had no role in study design, interpretation of results, report ,writing or in the decision to submit the article for publication.

Conflict of Interest

IB and GS have received speaker fees from Angelini Pharma, DI has received travel support from Otsuka-Lundbeck and Janssen. This support has been unrelated to the topic of the current article. None of the other authors have any conflicts of interest to disclose.