Dysconnectivity in Schizophrenia Revisited: Abnormal Temporal Organization of Dynamic Functional Connectivity in Patients With a First Episode of Psychosis

Abstract

Background and Hypothesis

Abnormal functional connectivity between brain regions is a consistent finding in schizophrenia, including functional magnetic resonance imaging (fMRI) studies. Recent studies have highlighted that connectivity changes in time in healthy subjects. We here examined the temporal changes in functional connectivity in patients with a first episode of psychosis (FEP). Specifically, we analyzed the temporal order in which whole-brain organization states were visited.

Study Design

Two case-control studies, including in each sample a subgroup scanned a second time after treatment. Chilean sample included 79 patients with a FEP and 83 healthy controls. Mexican sample included 21 antipsychotic-naïve FEP patients and 15 healthy controls. Characteristics of the temporal trajectories between whole-brain functional connectivity meta-states were examined via resting-state functional MRI using elements of network science. We compared the cohorts of cases and controls and explored their differences as well as potential associations with symptoms, cognition, and antipsychotic medication doses.

Study Results

We found that the temporal sequence in which patients’ brain dynamics visited the different states was more redundant and segregated. Patients were less flexible than controls in changing their network in time from different configurations, and explored the whole landscape of possible states in a less efficient way. These changes were related to the dose of antipsychotics the patients were receiving. We replicated the relationship with antipsychotic medication in the antipsychotic-naïve FEP sample scanned before and after treatment.

Conclusions

We conclude that psychosis is related to a temporal disorganization of the brain’s dynamic functional connectivity, and this is associated with antipsychotic medication use.

Introduction

The dysconnectivity hypothesis of schizophrenia proposes that this disorder is due to an abnormal interaction between distant brain regions, rather than a single localized brain dysfunction.^1,2 Neuroimaging studies have provided evidence supporting this idea, showing abnormal structural³ and functional⁴ connectivity in several networks in patients.

More recently, functional magnetic resonance imaging (fMRI) studies have shown that the brain’s slow oscillatory activity and interactions between regions in health are not static, but rather change over time.^5–7 This evolving activity goes through periods of stable dynamics with a particular whole-brain organization or meta-states.^8–10 Opening the temporal domain of brain functional connectivity poses several questions to the dysconnectivity hypothesis of schizophrenia. Studies have shown that machine-learning algorithms are better at identifying schizophrenia from other disorders when the time-varying connectivity data are included, rather than the average or static connectivity matrices.^11–13 This suggests that temporal dynamics have valuable information about schizophrenia’s pathophysiology.

Studies have focused on understanding how specifically is the dynamic functional connectivity abnormal in schizophrenia. We could divide the existing approaches into 3 groups of studies (figure 1A). The first group suggests that dysconnectivity in schizophrenia (localized or global) is a transient rather than a constant state, more prevalent during specific periods of brain activity.^14–16 The second group highlights that functional connectivity within these transient states might not be different in patients compared to controls. Instead, it is the amount of time that patients remain on certain whole-brain meta-states that differs, so when examining “static” (average) functional connectivity differences between groups emerge.^17–20 The third group emphasizes the sequential (temporal) nature of these dynamics, which was not necessarily addressed in the 2 other groups. Rather than focusing on the characteristics of the individual states (group 1) or their number (group 2), they examine the time domain and the changes between these states. They have done this by thinking of the dynamic functional network as linked or multiplex networks,²¹ with networks stacked according to the passing of time, or else by examining the nodal connections evolving in time²² (figure 1B). Unlike the 2 first groups, the characterization of these dynamic changes is not trivial to the order in which time passes.

Fig. 1.

State of the art: an overview of dynamic functional connectivity studies in schizophrenia. (A) A summary of the existing approaches. Group 1 characterizes studies that focused on transient dysconnectivity states in schizophrenia. Group 2 represents the literature that studied the amount of time that patients remain on certain meta-states. Group 3 characterizes studies that examine characteristics of temporal dynamics, highlighting the temporal nature of dynamics. (B) In detail, group 3 studies the dynamic functional network by examining the nodal connections evolving in time or multiplex networks. (C) Our approach differs from group 3 in that we study the dynamic functional connectivity by the construction of transition networks where each node is a whole-brain meta-state and the links between nodes are the transitions between meta-states. We then analyze the temporal order in which meta-states were visited.

The present study aims to contribute to our understanding of the temporal dynamics in the dysconnectivity state in schizophrenia by looking at a new aspect of the temporal transitions in brain activity, which is related to the third group of studies described above. For each time period (time window), we examined the whole-brain interactions or brain states, which happen to repeat themselves in time.^9,10 Just as one can build different words and sentences with the same letters by organizing them in an alternative order (figure 1C), we recently showed that the sequence in which the brain explores these brain states in healthy subjects is not random, and is related to general cognition and motor abilities.²³ Some healthy subjects tend to revisit meta-states forming a redundant trajectory, or visit consecutive meta-states that are similar, characteristics that were associated with lower general cognition. We here used this same approach based on network science to study the transitions between brain meta-states from functional MRI data of individuals experiencing a first episode of psychosis (FEP) and healthy controls. Based on their relationship with cognition in healthy controls, we examined 2 main questions:

1. do patients tend to revisit whole-brain configuration meta-states shortly after leaving them, providing a more redundant trajectory than healthy controls?

2. are patients able to consecutively switch from very different brain meta-states and visit all the existing states efficiently?

We were also interested in the clinical characteristics associated with any observed differences in the dynamic changes between whole-brain meta-states. In particular, we explored its relationship with cognition, symptoms severity, and the use of antipsychotics, which are known to disrupt brain connectivity.²⁴ We hypothesized that patients with FEP would present a more redundant dynamic path when visiting different meta-states and a less efficient overall trajectory, when compared to healthy controls. This could be associated with difficulties in cognition and the use of antipsychotic medication. We addressed these questions in 2 first-episode samples and their matched healthy controls, exploring the replicability of our findings. Furthermore, we included repeated measurements of our subjects, which for 1 sample included before and after antipsychotic treatment.

Materials and Methods

Subjects

Our analyses included 2 datasets of resting-state fMRI data. The first included 79 patients admitted with a FEP to the early intervention program ward from the Instituto Psiquiátrico José Horwitz in Santiago, Chile.²⁵ Patients fulfilled the criteria for a psychotic episode according to the Mini-International Neuropsychiatric Interview.²⁶ Twelve patients presented an affective psychosis. Patients did not present any neurological, medical illness, comorbidity with other psychiatric disorders, and psychomotor agitation. The inclusion and exclusion criteria were evaluated by their treating psychiatrists using standardized evaluations at the scanning time. A history of substance use was not considered as an exclusion criterion, but cases with substance-induced psychosis were excluded. Patients were scanned at the earliest possible opportunity. On average, they were treated with antipsychotics for 24.6 ± 13.1 days (mean ± standard deviation) before the baseline scan. A subgroup of 27 patients were rescanned after an average of 111.6 ± 37.7 days. Eighty-three healthy participants without a lifetime history of psychotic disorder, no first-degree family history of psychotic disorders, and no current history of any mental disorder were also included, scanning them at baseline. Thirty-two healthy controls were reassessed at 128.3 ± 47.7 (mean ± standard deviation) days. The study was approved by the Ethics Committee of the Pontificia Universidad Católica de Chile (Ref: 15–297) and Servicio de Salud Metropolitano Norte. Written informed consent was obtained from all participants.

The second cohort included 21 patients with non-affective FEP and 15 healthy participants from the Instituto Nacional de Neurología y Neurocirugía, Mexico. The Structured Clinical Interview for DSM-5 was utilized to determine inclusion. Patients were antipsychotic-naïve at baseline. Exclusion criteria included a concomitant medical or neurological illness, current substance abuse, or history of substance dependence (excluding nicotine), comorbidity with other psychiatric disorders, a high risk for suicide, and psychomotor agitation. Fifteen age- and sex-matched healthy controls were also enrolled and assessed in the same manner as the patients. Controls with a history of psychiatric illness or a family history of psychosis were excluded. Fifteen patients were rescanned and reassessed at a 4-week follow-up, alongside twelve healthy controls. The study was approved by the Ethics Committee of the Instituto Nacional de Neurología y Neurocirugía, and written informed consent was obtained from all participants.

Table 1 provides further demographic and clinical information about these 2 samples and supplementary table 1 provides information about the antipsychotic medication. All patients from Mexico were treated with risperidone at the follow-up.

Table 1.

Demographical and Clinical Scales of Chilean and Mexican Cohorts

	Chilean Sample			Mexican Sample
	First-Episode Psychosis	Healthy Control	P-Value	First-Episode Psychosis	Healthy Control	P-Value
Demographical variables
N baseline (male)	79 (64)	83 (56)	.074	21 (11)	15 (7)	.735
N follow-up (male)	27 (20)	32 (19)	.361	15 (8)	12 (6)	.863
Age	20.2 ± 0.3	23.3 ± 0.4	<.001	27.3 ± 1.2	24.2 ± 0.6	.094

Clinical scales and antipsychotic doses in patients
	Baseline	Follow-up		Baseline		Follow-up
PANSS total	69.3 ± 2.1	44 ± 3.1		110.3 ± 4.8		75.9 ± 7.5
PANSS positive symptoms	16.3 ± 0.7	9.4 ± 0.7		29.8 ± 1.2		17.3 ± 2.2
PANSS negative symptoms	20.4 ± 1	13 ± 1.5		26.6 ± 2.1		21.6 ± 2.9
PANSS general symptoms	32.6 ± 0.9	21.6 ± 1.2		54 ± 2.4		37.1 ± 3.1
MATRICS total score	29.9 ± 1.9	33.4 ± 3.1		24.6 ± 2.6		30.3 ± 2.8
Antipsychotic doses (mg)	537.5 ± 32.9	420.2 ± 50.1		0 ± 0		172 ± 17.4

Note: Antipsychotic doses expressed in chlorpromazine equivalents (Leucht et al. 2016). Mean ± SEM.

MATRICS, measurement and treatment research to improve cognition in schizophrenia; PANSS, positive and negative syndrome scale.

Both samples were assessed clinically at both time points using symptom rating scales (PANSS)²⁷ and a neuropsychological assessment (MATRICS Consensus Cognitive Battery).^28,29 Doses of antipsychotics were converted to chlorpromazine equivalents using the defined daily doses (DDD) method.³⁰

Data Acquisition and PreProcessing

Our first sample was scanned in a Philips Ingenia 3-T MRI with a 16-channel coil. Resting-state images were acquired for 8.33 min (195 volumes), while participants had their eyes open, using an EPI acquisition with a TR of 2.5 s, TE 32 ms, and a flip angle of 82°. Forty slices with a continuous descending order were acquired, using a field of view of 220 × 220 mm, and an isotropic voxel size of 2.75 mm. A structural T1-weighted image with a voxel size of 1.0 mm³ isotropic, a minimum TI delay of 965.2, TE = 3.5, TR = 7.7, and a flip angle of 8° was also acquired.

The second sample were scanned in a 3-T Siemens Skyra scanner with a 20-channel radio frequency coil. Resting-state images were acquired during 5.06 min (645 multiband volumes) while participants kept their eyes open. T2*-weighted functional images were acquired using a gradient-echo EPI sequence with TE = 29 ms, TR = 0.46 s, flip angle = 44°, slice thickness = 3 mm, slice gap = 3 mm, the field of view 268 mm, base resolution = 82, multiband acceleration factor = 8, voxel size = 3.3 mm × 3.3 mm × 3.0 mm. A structural T1-weighted image with a voxel size of 1.0 mm³ isotropic, a minimum TI delay of 965.2, TE 3.5, TR 2.3, and a flip angle of 9° was also acquired.

Preprocessing of the functional images of both datasets followed a previously published pipeline.³¹ Briefly, this included slice-time correction, realignment, normalization, spatial smoothing with a 6 mm FWHM kernel, and temporal filtering between 0.008 and 0.08 Hz. Management of residual movement was performed using an automated-ICA method.³²

Data Analysis

Identification of Temporal Meta-States.

Our analyses first involved identifying the set of whole-brain configurations or meta-states that the functional brain dynamics visited over time. We first divided the brain into similarly sized regions using a previously established template (638 nodes),³³ and used multiplication of temporal derivatives to measure the functional connectivity between all pairs of regions in non-overlapping temporal windows. Multiplication of temporal derivatives³⁴ has shown high sensitivity to changes in functional connectivity and robustness to noise introduced by head movement, making it appropriate for dynamic functional connectivity analysis. Based on our previous study examining temporal dynamics at relatively short windows with sparse sampling,²³ we used time windows of the length of 2 volumes (5040 ms) for the Chilean cohort and 3 volumes (1380 ms) for the Mexican cohort. We, therefore, obtained a 638 × N time window for each subject.

The similarity between the functional connectivity organization of each subject’s time windows was then clustered using k-means.³⁵ This process assigned each window to a specific meta-state. For example, starting with the original 97 windows in the Chilean sample, our clustering procedure grouped them into a reduced number of discrete states that the subject could repeatedly visit. Considering that there is no certainty about the number of discrete states that brain dynamics may explore, we analyzed a wide range of clusters (35–55).

Finally, we represented the temporal trajectories between visited brain states as a directed graph. Here each meta-state was represented as a node and the transitions between meta-states as directed edges. A weight was assigned to each edge according to the number of times the brain transitioned between a pair of meta-states, in the direction of the edge. supplementary figure 1 summarizes the construction of this directed and weighted network.

Network Parameters Examined.

We explored differences in temporal trajectories between groups using the following metrics on the weighted-directed graphs:

Measures of Redundancy in the Temporal Paths

 (1) modularity of the graph,³⁶ which examines how easy it is to divide a specific network into highly connected subgroups or modules. Considering that the strength of the connections is based on the amount of time the brain transits through those states, a highly modular network would indicate a temporal trajectory characterized by a high likelihood to revisit certain groups of states.

(2) local efficiency of the network,³⁷ a measure closely related to the clustering of a network, describing the level of connectivity between neighbors of a node.

Measures Exploring Transitions Between Different Brain States

 (1) transition cost,²³ a global parameter defined as the distance between 1 meta-state and the next 1 (1—correlation coefficient of their connectivity matrices). Transitions between consecutive states that are very different in their connectivity patterns (low correlation) would have a high cost. This could reflect a higher metabolic cost associated with switching from 1 connectivity pattern to another.

We also analyzed 2 different aspects related to this cost measure: the immobility of the network (number of times the brain dynamics remained in the same meta-state between 2 consecutive windows) and the leap size (the transition cost without considering immobility periods).

(2) global efficiency,³⁷ a global measure based on the shortest paths between states. It provides an indicator of how easy it is, on average, for the brain to transition from 1 state to any other.

Low global efficiency could be due to higher average transition costs or an inefficient organization of the connecting paths. We, therefore, measured the global efficiency of the transition network accounting for transition costs, or its cost-efficiency.³⁸

All these metrics are measured on directed graphs. Directed graphs are relevant for path-length-based metrics such as global, and local efficiency, and cost efficiency. This is because nodes could be connected in one direction but not in opposite direction, unlike undirected graphs, where if 2 nodes are connected, this is for both directions (supplementary figure 2).

For each measure, we calculated the area under the curve for the entire range of the number of clusters used in the k-clustering algorithm of the meta-states. We then used this value for the linear mixed-effect model (described below).

Clinical Analyses

Both samples included repeated measurement data for some (but not all) participants. We, therefore, used a linear mixed-effect model, which provided the greatest power and flexibility as it allowed us to inform our model from all the available measurements. We first looked at differences between patients and controls, using the following model:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{G}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{c}\sim \mathrm{a}\mathrm{g}\mathrm{e}+\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{r}+\mathrm{f}\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{e}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e}-}\\ \mathrm{d}\mathrm{i}\mathrm{s}\mathrm{p}\mathrm{l}\mathrm{a}\mathrm{c}\mathrm{e}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}+\mathrm{c}\mathrm{a}\mathrm{s}\mathrm{e}+(1|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{j}\mathrm{e}\mathrm{c}\mathrm{t})\end{array}}$$ (1)

Subject variable was defined as a random effect to account for repeated measures. Gender, age, and a measure of within-scanner movement (framewise displacement³⁹), were included as covariates of no interest.

Another mixed-effect model was constructed to explore the clinical importance of potential differences in these metrics. These included only patients, using the following model:

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{l}{\displaystyle \mathrm{G}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{c}\sim \mathrm{a}\mathrm{g}\mathrm{e}+\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{r}+\mathrm{f}\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{e}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e}-\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{p}\mathrm{l}\mathrm{a}\mathrm{c}\mathrm{e}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}}\\ +\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{S}\mathrm{S}\mathrm{p}+\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{S}\mathrm{S}\mathrm{n}{\displaystyle +\mathrm{M}\mathrm{A}\mathrm{T}\mathrm{R}\mathrm{I}\mathrm{C}\mathrm{S}+\mathrm{A}\mathrm{n}\mathrm{t}\mathrm{i}\mathrm{p}\mathrm{s}\mathrm{y}\mathrm{c}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{i}\mathrm{c}}\\ +(1|\mathrm{s}\mathrm{u}\mathrm{b}\mathrm{j}\mathrm{e}\mathrm{c}\mathrm{t})\end{array}}\end{array}}$$ (2)

We also included in the model above the variable affective for the Chile sample.

Supplementary tables 2 and 3 provide demographic and clinical data organized according to the presence (or not) of repeated measurements.

Although our analysis did not explicitly examine changes before and after treatment (for which we had less data), by including the PANSS, cognitive functioning, and antipsychotic medication in the mixed-effect analysis, our analyses could inform changes in time related to these variables. Analyses looking at changes before and after treatment in the participants with complete follow-up data are reported in supplementary figures 3 and 4 and supplementary tables 4 and 5.

Multiple comparisons were managed using false discovery rates for each of these analyses. For example, the analyses of the clinical features included 7 graph metrics and 4 variables of interest (28 comparisons). All data analyses were done with MATLAB (MathWorks, Natick, USA).

Results

We first present the result from our larger Chile sample, and then the replication sample from Mexico.

Patients Present Redundant and Segregated Temporal Dynamics

Previously, we showed that transition networks in the healthy brain could be decomposed into clusters of nodes with high connectivity between themselves.²³ This meant that brain dynamics evolved with periods visiting a specific group of global state configurations, eventually moving on to another set. Q values from Newman modularity algorithm were significantly higher for patients than controls, suggesting a more segregated network (F_1,216 = 8.67, P_FDR = .013, figure 2A, supplementary table S6).

Fig. 2.

Topological and economical properties of transition networks and transition network examples of patients with first-episode psychosis and healthy controls. The plots show the residuals of the area under the curve for topological and economical properties in the transition graph for healthy controls and patients with a first episode of psychosis. (A) modularity (Q value). (B) local efficiency. (C) global efficiency. (D) Transition cost. (E) Leap size. (F) Immobility. (G) Cost-efficiency. The circles and error bars mark the mean and standard error of the mean. The bottom figures show transition networks from 2 subjects. Column (H) shows healthy control and column (I) the patient. The layout of the graph is such that the distance between nodes is proportional to their transition cost (1—correlation of connectivity matrices). The color of the nodes denotes the degree, the arrow of links represents the direction, and the width of links is the number of transitions between the connected nodes (weight). Self-connections of the nodes represent consecutive periods (time windows) in which the brain remained in the same meta-state (ie, immobility). The temporal path depicted in the transition graph of the healthy control spans longer distances (larger leap size), reaching other nodes rapidly (higher global efficiency), with apparently less redundant paths (lower local efficiency). In the transition network of the patient the nodes are less connected, which implies going through more intermediate nodes to visit meta-states (lower global efficiency). For color, please see the figure online.

We then examined whether dynamic trajectories in patients were more redundant, frequently returning to recently visited meta-states, measuring the local efficiency. Patients’ transition networks had significantly higher local efficiency (F_1,216 = 5.12, P_FDR = .029, figure 2B, supplementary table S6) than healthy controls, showing a higher redundancy.

Less Efficient Trajectories in Patients With Psychosis

The global efficiency of transition networks describes how long brain dynamics traverse on average to get from a state to another. This metric was lower for patients (F_1,216 = 9.3, P_FDR = .013), reflecting less efficient trajectories between brain states. Since edges of the network refer to time periods (changes between consecutive time windows), this could be interpreted as patients being slower to visit on average different meta-states compared to healthy subjects (figure 2C, supplementary table S6).

We then analyzed the costs of transitions, which we conceptualized as how much the brain had to change its connectivity pattern between a meta-state and the following one.²³ The transition cost was significantly lower in patients than in controls (F_1,216 = 6.45, P_FDR = .021, figure 2D, supplementary table S6). Further examination of the lower costs found in patients, we found that leap size was significantly lower compared to controls (F_1,216 = 5.36, P_FDR = .029, figure 2E, supplementary table S6). Additionally, immobility was significantly higher in patients (F_1,216 = 4.52, P_FDR = .035, figure 2F, supplementary table S6). Thus, lower transition costs in patients were driven by dynamics that remained in the same meta-state more frequently and had a reduced capacity to move between very different meta-states.

Cost-efficiency (global efficiency normalized by the cost) in patients’ network was also lower than in controls’ (F_1,216 = 7.6, P_FDR = .015, figure 2G, supplementary table S6).

Figure 2H–I illustrate these dynamic differences in a representative healthy control (figure 2H) and patient (figure 2I).

Associations Between Abnormal Dynamic and Clinical Characteristics

Higher antipsychotic doses in patients were associated with more segregated and redundant trajectories. We found a positive association between the dose of antipsychotics and modularity (Q value) of the transition network (segregation; F_1,97 = 11.58, P_FDR = .021, figure 3A, supplementary table S7), as well as with its local efficiency (redundancy; F_1,97 = 7.5, P_FDR = .041, figure 3B, supplementary table S7).

Fig. 3.

Transition network properties are related to antipsychotic doses and negative symptoms. Scatter plots (A–D) show significant associations between topological characteristics of the transition networks in patients and antipsychotic doses. (A) modularity (Q value). (B) local efficiency. (C) global efficiency. (D) Cost-efficiency. Scatter plot (E) shows a significant association between the patient’s leap size and negative symptoms. The error bar around the linear regression mark is 95% CI.

Antipsychotic doses were associated with changes in the path length-based measures, including a negative association between dose and global efficiency (F_1,97 = 10.73, P_FDR = .021, figure 3C, supplementary table S7). We also found a negative association between the cost-efficiency of the transition networks and doses (F_1,97 = 8.66, P_FDR = .038, figure 3D, supplementary table S7).

There was a negative association between negative symptoms and leap size (F_1,97 = 8.08, P_FDR = .038), with lower negative symptoms associated with a higher capacity to transit between remarkably different brain states (figure 3E, supplementary table S7).

There were no significant associations between dynamic functional networks and cognitive measures.

Transition Networks in Antipsychotic-Naïve Patients With First Episode Psychosis

We then explored case-control differences using a second dataset of resting-state fMRI of 21 antipsychotics-naïve patients with FEP (Mexico dataset). 15 patients were rescanned 1 month later after receiving treatment. Patients were similar in demographics as shown in table 1. The scanning sequence used was different from the first cohort, with a higher temporal resolution, and a smaller temporal window size used. However, as supplementary figure S5 shows, healthy controls from this sample displayed a similar non-trivial temporal organization in their brain dynamics to the sample from Chile and the Human Connectome Project.²³ It also displayed similar within-subject reliability in metrics such as leap size (supplementary figure S6 and supplementary table S8).²³

We did not find any significant differences in the configuration of transition networks between patients and controls (supplementary figure S7 and supplementary table S9). However, exploring clinical features we found a positive association between local efficiency and doses of antipsychotics (F_1,28 = 4.31, P_unc = .047, figure 4A, supplementary table S10) and a trend-level negative association between global efficiency and doses of antipsychotics (F_1,28 = 3.996, P_unc = .055, figure 4B, supplementary table S10), partially replicating the main results of the Chilean dataset. In addition, exploring changes before and after treatment in the patients with complete follow-up data, we found that after treatment modularity (Q) was significantly higher (P_unc = .0347, supplementary figure 4A, supplementary table S5) and global efficiency was significantly lower (P_unc = .0182, supplementary figure 4B, supplementary table S5).

Fig. 4.

Transition network properties are related to antipsychotic doses in the Mexican sample. Scatter plots show the significant associations between topological characteristics of the transition networks in patients and antipsychotic doses. (A) local efficiency. (B) global efficiency. The error bar around the linear regression mark is 95% CI.

Discussion

By using a novel method that focuses on the order in which certain brain configurations are visited over time, we here contribute further to our understanding of dysconnectivity in schizophrenia in the temporal domain. We found that patients with psychosis present a higher redundancy and segregation in their temporal trajectories than controls. We also found that patients are less efficient in visiting different brain states in time. These changes were associated with the antipsychotic dose received, possibly through the dopaminergic blockage. Finally, the association between redundant and inefficient trajectories and antipsychotic doses was replicated on a smaller sample of antipsychotic-naïve patients treated for 1 month with lower doses of antipsychotics.

Patients presented differences in several functional dynamic metrics. These results add up to previous findings showing that patients present differences in the transition probability between brain states and in the amount of time they remain in several brain states relative to controls.^15,17,18 Together, they suggest that the dysconnectivity hypothesis should be further expanded to consider this abnormal evolving activity, a state of temporal disorganization. In other words, schizophrenia might not only be linked to problems in the building blocks of the brain’s interacting activity, but also in the way they are sequentially visited. The latter would be akin to creating aberrant sentences with normal words by organizing them in a nonsensical sequence.

How do we interpret our findings? One can see a global common picture emerging of a slowed-down system, which probably has difficulties in varying its dynamic repertoire, and tends to repeat itself. That is how we found an increase in modularity and local efficiency in the trajectories, which increases the redundancy in the dynamics; a decrease in its cost, which makes it less able to switch plastically to different states; and a lower global efficiency, which describes a landscape of possible dynamics which is visited in a slower way. Such an interpretation complements the traditional implications of the dysconnectivity hypothesis. The temporal dysconnectivity view might give a fuller picture of a complex clinical presentation, providing possible explanations for things such as processing speed difficulties,⁴⁰ the tendency to perseverate,⁴¹ or catatonic repetition.⁴² However intuitive, some aspects of the data do not fully support this explanation, particularly the lack of associations between dynamics and global cognition in patients.

One of the strengths of our study is the inclusion of a second sample that replicated some of our results, namely those related to antipsychotic medication. The replication sample was small, and therefore was underpowered to find medium or small effects, such as those possibly related to being a case. Patients received low-dose antipsychotic medication for a month only, making it even more interesting to see that a replication with antipsychotic dose and local efficiency was found. These associations with antipsychotic doses must be interpreted carefully, since the dopamine-blocking ability differs between the different antipsychotic medications used, and conversion to common standards (chlorpromazine equivalent) might not necessarily eliminate all this variation. Functional connectivity changes observed after antipsychotic initiation have been associated with symptomatic improvement and described as a normalization of the connectivity.^43,44 However, disruptions in the functional network organization in healthy subjects have also been reported after pharmacologically induced decreased dopaminergic tone.^45,46 Our replicated findings showing that antipsychotics are associated with a disrupted temporal organization do not necessarily contradict the proposed localized normalization of connectivity also associated with them, as our method examines brain properties that are not seen by usual “static” analyses. We previously found a significant association between cognition and temporal dynamics in healthy subjects,²³ of which we did not replicate within patients in this study. Nevertheless, one could hypothesize that the antipsychotics’ limited effect on cognition,⁴⁷ and possibly its deleterious effect at high doses, might be related to their lack of beneficial effect on temporal dynamics.

From a dynamic point of view, the relationship between antipsychotic use and network dynamics has been described using attractor-based computational frameworks.^48,49 Acute psychosis, particularly high positive symptoms, has been associated with unstable attractors in brain activity. On the contrary, D2 antagonists would lead to more stable states. Our results support these predictions,⁴⁹ suggesting that dopamine modulation generates an over-stability of meta-states resulting in slower transitions between them, with more redundant trajectories and difficulties switching to novel (different) states.

The main limitation of this study is the small sample size of antipsychotic-naïve patients in our replication sample. In addition, patients were recruited in tertiary centers, which could bias the sample towards more severe cases usually treated with higher doses of antipsychotics, and limit the generalizability of the results. The samples also differed in the inclusion of patients with affective psychosis, symptoms severity, and age. Although these limitations were addressed by incorporating these variables in the mixed-effect models, the lack of homogeneity between samples must be taken into account when interpreting the results.

In conclusion, by using a novel method focusing on the temporal changes of the brain’s functional configurations, we found that FEP patients presented a temporal disorganization in their brain dynamics. Namely, their trajectories were more redundant, presented higher segregation, and visited consecutively more similar brain states than healthy controls. In addition, we found that these changes were associated with antipsychotic use.

Supplementary Material

Supplementary material is available at https://academic.oup.com/schizophreniabulletin/.

Funding

This work was funded by the Agencia Nacional de Investigación y Desarrollo from Chile (ANID), through its programs ANILLO PIA ACT1414 and PIA ACT192064, FONDECYT postdoctorado (Ref: 3190311 to Juan P. Ramirez-Mahaluf), and FONDECYT regular (Ref: 1200601 to Nicolas A.Crossley, Ref: 1180932 to Tomas Ossandon, Ref: 1180358 to Juan Undurraga). Camilo de la Fuente-Sandoval is supported by Consejo Nacional de Ciencia y Tecnología of Mexico, (CONACyT) grant 320662, and National Institutes of Health grant R01 MH110270. Pablo León-Ortiz, Francisco Reyes-Madrigal and Camilo de la Fuente-Sandoval are supported by CONACyT’s Sistema Nacional de Investigadores.

Acknowledgments

Drs Reyes-Madrigal, León-Ortiz, and Mora-Durán have received speaking fees from Janssen (Johnson and Johnson). Dr Crossley has received personal fees from Janssen outside the submitted work. All other authors reported no competing interests.

Data Availability

In-house Matlab scripts and data are available from the authors upon request.