Cognitive subtypes and brain network differences in antipsychotic-naïve first-episode psychosis

Abstract

Cognitive functioning within psychosis is heterogeneous, and data-driven approaches may help evaluate this heterogeneity. Brain network connectivity within first-episode psychosis (FEP) may provide an additional tool for understanding cognitive heterogeneity. FEP participants and healthy controls (HC) completed the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS), and premorbid intellectual functioning was estimated using The American National Adult Reading Test (AmNART). k-means clustering of RBANS domain scores identified cognitive subgroups within FEP. Resting-state functional connectivity (FC) of the default mode (DMN), dorsal attention (DAN), executive control (ECN), and salience networks (SN) was assessed at baseline (prior to antipsychotic treatment), 6-, and 16 weeks. Voxelwise linear mixed models evaluated group-by-time interactions within each network. Two cognitive subgroups were identified within FEP (impaired and intact). All subjects had average premorbid intellectual functioning. At baseline, the cognitively impaired FEP subgroup showed reduced functional connectivity relative to healthy controls across regions within the DAN and ECN, with no significant differences observed between FEP subgroups (P_FDR < .05). Over time, the impaired subgroup exhibited fluctuating FC in the DAN and ECN, while the intact subgroup showed FC changes only in the ECN (P_FDR < .05). HCs also showed minor changes in the DAN and ECN. Distinct FEP cognitive subgroups show differential changes in FC. Notably, the impaired subgroup’s shift from average premorbid intellectual functioning to impaired cognition suggests potential cognitive deterioration in a subset of psychosis participants. These results underscore the importance of considering both cognitive and neural heterogeneity in FEP.

Introduction

Cognitive impairments are a core characteristic of psychosis and pose a significant challenge to both clinical and functional recovery¹. Overall, cognitive performance in psychosis is typically about one standard deviation below that of healthy controls, based on large meta-analyses^2–6. Deficits span multiple domains, including working memory⁷, attention⁸, processing speed⁹, and executive functioning¹⁰. However, literature suggests that global cognitive deficits are not consistently observed, and evidence indicates the presence of cognitive subgroups within the psychosis population^11,12.

To better characterize cognitive profiles of individuals with psychosis and help reduce the heterogeneity of findings, researchers have increasingly turned to data-driven methods. Cluster analysis is one such method, providing a framework for identifying more homogeneous subgroups within heterogeneous samples. This method has identified a range of cognitive subgroups, including a substantial subset of individuals who remain cognitively intact^11,13–15. These subgroups differ in both neurobiological and clinical characteristics, including structural volume^15,16, functional connectivity^16,17, duration of illness¹⁶, symptoms^16,18, and general functioning¹⁸. While some studies have reported two distinct subgroups^16,17, others have identified three or even four cognitive subgroups among psychosis participants^12,15,18.

Across studies, a common framework has emerged in which cognitive subtypes are often described as “preserved”, “compromised”, and “deteriorated”^19–21. The preserved group typically demonstrates relatively intact cognitive performance and resembles healthy controls, the compromised group shows stable premorbid deficits suggesting early neurodevelopmental alterations, and the deteriorated group displays evidence of cognitive decline related to estimated premorbid ability¹⁹. Such variability points to the possibility that distinct neuropathological mechanisms underlie psychosis subgroups defined by cognitive status. However, most of this research primarily comes from chronic psychosis samples, where individuals have experienced prolonged clinical symptoms. As such, these findings may be limited, as cognitive performance could already be influenced by the long-term effects of antipsychotic medication²².

Psychosis spectrum disorders are also highly heterogeneous at the neural level, involving multiple brain regions and large-scale functional networks. Research suggests that intrinsic connectivity within the default mode (DMN), dorsal attention (DAN), executive control (ECN), and salience (SN) networks plays a crucial role in cognitive functions²³; however, findings in these networks show considerable heterogeneity, including both hypo- and hyperconnectivity within and between network regions^24,25. Recent work has begun to link these cognitive subtypes to distinct patterns of functional connectivity. For example, Lewandowski et al.¹⁷ found that preserved and compromised subtypes differed in frontoparietal connectivity, suggesting that large-scale network organization may underpin differences in cognitive performance. Extending this work, Meda et al.²⁶ reported that cognitive subgroups in first-episode psychosis and chronic psychosis exhibited divergent resting-state connectivity patterns across the DMN, salience, striatal, and executive networks, with more impaired subtypes showing widespread dysconnectivity relative to preserved subgroups, highlighting the neural relevance of cognitive heterogeneity. Similarly, Rodriguez et al., found that cognitive subgroups in first-episode psychosis displayed distinct resting-state functional connectivity patterns across the DMN, SN, frontoparietal, and cerebellar networks, further linking cognitive heterogeneity to neural network organization²⁷. In addition, structural neuroimaging studies have identified gray matter differences across cognitive subtypes, including regional volumetric reductions and morphometric variations in various cortical and subcortical regions^20,28,29, highlighting that cognitive heterogeneity is reflected in both functional and structural neural measures.

In addition to neural differences, cognitive subgroups are associated with distinct clinical and functional outcomes. For example, Oomen et al.¹⁸ reported that first-episode psychosis patients with more intact cognitive profiles showed better symptom improvement and higher functional recovery over time, highlighting that cognitive heterogeneity has meaningful implications beyond neuroimaging findings. Similarly, cognitive profiles derived from cluster analysis have been shown to predict differential social and occupational functioning over time in FEP samples³⁰. Together, these findings suggest that cognitive heterogeneity in psychosis reflects both neural and functional differences, though disentangling the respective contributions of cognition and illness processes remains a challenge.

Given the heterogeneity in resting-state functional connectivity (rs-FC) findings, traditional cross-sectional approaches may lack the specificity and sensitivity to identify network-level abnormalities underlying cognitive impairment in psychosis. A longitudinal approach is essential for mapping how functional connectivity changes over time, particularly across cognitive subtypes. Despite growing interest in cognitive heterogeneity, the neurobiological underpinnings of cognitive subtypes in psychosis are still relatively understudied and not yet fully characterized. Understanding these subtypes could be key to defining biotypes with distinct biological characteristics within the psychosis spectrum.

To address these gaps, the current study examined large-scale functional brain networks (DAN, DMN, ECN, and SN) in a large cohort of antipsychotic medication-naïve first episode psychosis (FEP) participants and healthy controls. Initially, we assessed FEP participants and gathered imaging and clinical data before they were exposed to antipsychotic medication. Using cluster analysis, FEP participants were categorized into subgroups based on baseline cognition. Additional imaging data were collected at 6 and 16 weeks after initiating appropriate psychopharmacological treatment. Linear mixed modeling was applied to examine changes in connectivity over time and between-group differences. We hypothesized that the cluster analysis would reveal between 2 and 4 subgroups of FEP participants, and all subgroups would exhibit distinct functional connectivity trajectories and clinical differences relative to healthy controls.

Results

Sample characteristics and demographics

Of the recruited healthy control participants, 1 was excluded after quality control (see “Data processing”). Within the FEP participant sample, 17 were excluded due to quality control, leaving us with a total sample size of 212 (FEP = 81, HC = 131). Supplemental analyses comparing included and excluded FEP participants on demographic and clinical variables indicated no significant differences in age, education, parental occupation, DUP, BPRS baseline scores, sex, smoking status, or cannabis use; although the overall distribution of diagnoses differed (${\chi }^2$(3) = 10.56, p = .014), no specific diagnostic category was over- or underrepresented based on standardized residuals (see Supplementary Table S1). Demographic data are summarized in Table 1. Participants ranged in age from 15 to 42 years old. The three groups did not differ based on age, sex, or parental occupation. Compared to the FEP groups, healthy controls had higher education (F(2,198) = 21.84, p < .001), smoked less per day (F(2, 206) = 16.91, p < .001) and had lower numbers of cannabis users (${\chi }^2$(2) = 91.98, p < .001). In terms of race/ethnicity, chi-square tests with post hoc analyses examining adjusted residuals indicated that there were more Black individuals and less White individuals in the impaired FEP group as compared to the intact FEP group, χ²(4) = 10.50, p < .05. There were also less Black individuals and more Asian and White individuals in the HC group compared to the impaired FEP group, χ²(5) = 31.17, p < .001.

Table 1.

Demographics and clinical measures.

				Statistics
Characteristics	FEP impaired (n = 41)	FEP Intact (n = 40)	HC (n = 131)	F/t/${\chi }^2$	p
Sex (male/female)1	24/17	23/17	75/55	0.01	0.99
Age (years)2	24.35 (5.66)	22.93 (5.64)	24.18 (4.99)	0.95	0.38
Education3	13.00 (2.35)	12.87 (2.10)	15.13 (2.34)	21.84	<.001
Race/Ethnicity	–	–	–	–	–
Black	26	15	27
White	11	23	70
Hispanic/Latino	1	0	6
Asian	1	2	23
Other	2	0	2
Parental occupation4	5.79 (5.46)	4.64 (4.98)	4.51 (4.17)	2.55	0.08
Diagnosis				8.64	0.19
Schizophrenia	21	18	–	–	–
Bipolar disorder with psychosis	3	1	–	–	–
Schizophreniform disorder	2	1	–	–	–
Schizoaffective Disorder	10	7	–	–	–
MDD w/psychotic features	1	1	–	–	–
Brief psychotic disorder	2	1	–	–	–
Psychosis NOS	2	11	–	–	–
Smoking, PPD5	0.20 (0.33)	0.23 (0.45)	0.01 (.05)	16.91	<.001
Number of Cannabis Users6	24/41	24/40	2/128	91.98	<.001
Duration of Untreated Psychosis7	22.79 (40.34)	23.23 (36.62)	–	−0.52	.95
BPRS Baseline Positive Symptoms8	14.90 (3.69)	14.50 (4.55)	–	0.43	.667
BPRS Week 16 Positive Symptoms9	6.64 (3.46)	5.77 (2.21)	–	1.13	0.263
BPRS Baseline Negative Symptoms10	5.63 (2.66)	4.62 (2.63)		1.68	.096
BPRS Week 16 Negative Symptoms11	5.25 (2.36)	4.58 (2.26)		1.11	.272
BPRS Baseline Total Symptoms12	47.84 (10.03)	46.54 (11.32)		0.53	.595
BPRS Week 16 Total Symptoms13	29.75 (5.73)	27.84 (6.34)		1.21	.231
Treatment Reponse14	0.74 (0.36)	0.77 (0.33)	–	−0.40	.686
Quality of Life15	85.78 (15.12)	81.78 (15.27)	94.76 (9.98)	17.37	<.001

Values are mean (SD), unless indicated otherwise.

BPRS Brief Psychiatric Rating Scale, FEP, first episode psychosis, MDD with psychosis, major depressive disorder with psychosis, NOS, not otherwise specified, PPD, packs per day.

Data missing for ¹1 healthy control, ²2 FEP participants, ³11 healthy controls, ⁵63 healthy controls, ⁶1 healthy control, ⁸1 FEP participant, ⁹22 FEP participants, ¹⁰4 FEP participants, ¹¹22 FEP participants, ¹²4 FEP participants, ¹³22 FEP participants, ¹⁴22 FEP participants, ¹⁵4 FEP participants and 31 HC.

⁴Ranks determined from Diagnostic interview for Genetic Studies (1–18 scale); Lower numerical value (higher rank) corresponds to higher socioeconomic status; Data missing for 6 FEP participants.

⁷Recorded in months.

Cluster and clinical analyses

The optimal number of clusters for the data was determined to be two, since it was the cluster solution identified by the largest number of clustering indices (10/30). Of the remaining indices, four supported either a three- or six-cluster solution, two supported a four-cluster solution, and one index each supported a one-, eight-, nine-, or ten-cluster solution. The remaining indices did not identify a clear optimum. The clusterwise bootstrap analysis resulted in Jaccard stability scores of 0.94 (cluster 1) and 0.93 (cluster 2), indicating that the cluster solution was valid and stable.

Examination of the RBANS scores for each cluster (Table 2) indicated that FEP participants in cluster 1 demonstrated low average to below average performance in all five cognitive domains and overall total scores (impaired cluster; n = 41). FEP participants in cluster 2 exhibited low average to average performance across all domains and total scores (intact cluster; n = 40). HC participants (n = 116) demonstrated average total performance across all domains, with the exception of the visuospatial construction domain, in which scores were measured in the low average range. Post hoc comparisons between the three groups on each RBANS domain and total score are described below and reported in the supplementary information (Supplementary Table S2).

Table 2.

One-way ANOVA examining group differences in cognitive performance and premorbid intellectual functioning.

Measure	Impaired FEP (n = 41)		Intact FEP (n = 40)		HC (n = 131)		F	η2
	M	SD	M	SD	M	SD
RBANS
Total	64.07	7.87	88.14	9.80	93.47	10.61	144.19***	.58
Immediate memory	73.40	12.84	95.63	13.79	102.78	14.26	72.88***	.42
Visuospatial constructional	67.18	11.28	83.98	16.88	84.22	13.63	26.09***	.20
Language	76.16	16.40	95.56	11.91	96.46	15.76	30.63***	.23
Attention	70.82	13.41	95.00	11.39	102.82	14.42	89.51***	.47
Delayed memory	68.98	15.41	86.19	10.17	91.22	10.25	60.13***	.37
AmNART	102.72	10.12	109.51	10.10	108.13	15.18	3.38*	.03

AmNART American National Adult Reading Test, FEP first episode psychosis, HC healthy controls, RBANS Repeatable Battery for the Assessment of Neuropsychological Status.

***p < 0.001, *p < 0.05.

Results of the one-way ANOVA examining differences between groups on RBANS domain and total scores (Table 2) indicated a significant effect of group for immediate memory (F(2, 201) = 72.88, p < .001), visuospatial constructional (F(2, 201) = 26.09, p < .001), language (F (2, 201) = 30.63, p < .001), attention (F(2, 201) = 89.51, p < .001, delayed memory (F(2, 201) = 60.13, p < .001), and overall total cognitive score (F(2, 201) = 144.19, p < .001). Post hoc comparisons using Bonferroni correction showed that in comparison to the FEP intact subgroup, the FEP impaired subgroup demonstrated significantly lower performance in the domains of immediate memory (Mean Difference = −22.228, p = <.001), visuospatial constructional (Mean Difference = −16.799, p = <.001), language (Mean Difference = −19.403, p = <.001) attention (Mean Difference = −24.178, p = <.001), delayed memory (Mean Difference = −17.208, p = <.001), and overall total score (Mean Difference = −24.073, p = <.001). Compared to the HC group, the FEP impaired subgroup similarly demonstrated significantly lower performance across all domains, including immediate memory (Mean Difference = −29.384, p = <.001), visuospatial constructional (Mean Difference = −17.046, p = <.001), language (Mean Difference = −20.301, p = <.001), attention (Mean Difference = −31.997, p = <.001), delayed memory (Mean Difference = −22.246, p = <.001), as well as the total score (Mean Difference = −29.399, p = <.001). Compared to the HC group, the intact FEP subgroup had significantly lower mean scores in the domains of immediate memory (Mean Difference = −7.157, p = 0.013), attention (Mean Difference = −7.819, p = .005), delayed memory (Mean Difference = −5.038, p = .047), and overall total score (Mean Difference = −5.326, p = .009).

Examination of the AmNART scores indicated that all three groups demonstrated average performance, suggesting intact premorbid intellectual functioning (Table 2). One-way ANOVA results showed a significant effect of group on AmNART scores (F(2, 196) = 3.38, p < .05). However, post hoc comparisons indicated no significant differences between the groups (all p > .05).

In terms of clinical differences (see Table 1), the two FEP groups did not show any significant differences based on diagnosis, quality of life, duration of untreated psychosis, BPRS positive, negative, or total symptoms at baseline or week 16, or treatment response. Compared to both FEP groups, healthy controls reported a higher quality of life (F(2, 174 = 17.37, p < .001).

FC group differences at baseline

Following the exclusion of participants with excessive head motion (≤50% of data remaining after censoring or volume-to-volume motion ≥0.5 mm), as well as those with missing data or artifacts, the final count of participants with at least one rs-fmri timepoint was 212 (81 FEP and 131 HC).

Baseline group differences revealed one cluster within the DAN (right superior parietal lobule) where significant group differences were observed between the HC group and the impaired FEP subgroup (HC > Impaired FEP; Fig. 1; partial $\eta$² = .054). Within the ECN, the impaired FEP subgroup demonstrated significantly lower FC compared to the HC in two separate clusters located within the right middle frontal gyrus (Fig. 1; partial $\eta$² = .090 and.056). There were no significant baseline differences between the FEP subgroups in either the DAN or ECN.

Fig. 1. Baseline DAN and ECN connectivity differences across groups.

Violin plots illustrating baseline group differences in DAN and ECN resting state functional connectivity. Each dot corresponds to one participant. Group differences were found between the impaired FEP subgroup and healthy controls in (a) one DAN cluster and (b) two ECN clusters. Impaired impaired FEP subgroup, intact intact FEP subgroup, HC healthy control group, R right.

Functional connectivity trajectories

Linear mixed models through 3dLME indicated two clusters of significant group × time interactions in the DAN within the right posterior central gyrus and right superior parietal lobule (P_FWE < .05; Table 3). Follow-up tests for both clusters (Fig. 2) showed that the impaired group demonstrated decreasing functional connectivity from baseline to week 6 (Cohen’s d = 0.59), and baseline to week 16 (d = 0.39). In contrast, the HC group showed increasing functional connectivity from baseline to week 6 (d = 0.18) and decreasing functional connectivity from week 6 to 16 (d = 0.17). The intact FEP subgroup showed no change over time in either of the two significant clusters for this network.

Table 3.

Regions exhibiting significant correlations between group status and longitudinal functional connectivity.

		MINI coordinates
Seed (Network)	Location	x	y	z	Cluster size
RIPS (DAN)	Post Central Gyrus, R	43	−32	52	73
	Superior Parietal Lobule, R	26	−53	60	72
RPPC (ECN)	Middle Frontal Gyrus, R	32	59	25	124
	Middle Frontal Gyrus, R	44	12	59	107

ECN executive control network, L left, MNI Montreal Neurological Institute, R right, RIPS right intraparietal sulcus, RPPC right posterior parietal cortex.

Fig. 2. Longitudinal changes in DAN connectivity by group.

Clusters in the right postcentral gyrus and right superior parietal lobule, where significant group-by-time interactions were found using 3dLME in the dorsal attention network. Each line graph represents the change in mean functional connectivity for each group at each time point. Error bars represent the standard error of the mean. Analyses were controlled for age, sex, and framewise displacement. Impaired impaired FEP subgroup, Intact intact FEP subgroup, HC healthy control group, R right.

Additionally, two clusters of significant group x time interactions were identified in the ECN, both within the right middle frontal gyrus (P_FWE < .05; Table 3). Follow-up tests for cluster one (Fig. 3) showed that functional connectivity of the impaired FEP subgroup significantly increased from baseline to week 16 (d = 0.68). The intact FEP subgroup showed FC increases from baseline to week 6 (d = 0.57) and then FC decreases from week 6 to 16 (d = 0.64). The HC group showed significant decreases in functional connectivity from baseline to week 6 (d = 0.24) and again from baseline to week 16 (d = 0.21). In cluster two, follow-up tests indicated that the impaired FEP subgroup showed FC increases from both baseline to week 16 (d = 0.82) and week 6 to 16 (d = 0.52). In contrast, the intact FEP subgroup showed increases from baseline to week 6 (d = 0.38) and baseline to week 16 (d = 0.70). The HC group demonstrated decreases in FC from baseline to week 16 (d = 0.23). See supplemental material (Supplemental Table S3) for more information.

Fig. 3. Longitudinal changes in ECN connectivity by group.

Clusters in the right middle frontal gyrus (both clusters) where significant group-by-time interactions were found using 3dLME in the executive control network. Each line graph represents the change in mean functional connectivity for each group at each time point. Error bars represent standard error of the mean. Analyses were controlled for age, sex, and framewise displacement. Impaired impaired FEP subgroup, Intact intact FEP subgroup, HC healthy control group, R right.

Discussion

The present study aimed to identify cognitive subgroups within first-episode psychosis (FEP) and examine whether these subgroups showed distinct patterns of rest-state functional connectivity. Using a data-driven clustering approach, we identified two FEP subgroups: one with overall preserved (intact) cognitive performance and another with generally deteriorated (impaired) performance, relative to their premorbid intellectual functioning estimates. Results of the one-way ANOVAs indicated that the impaired FEP subgroup demonstrated significantly lower total RBANS scores and performed worse across all domains than the FEP preserved subgroup and HC group. The preserved FEP group demonstrated lower performance in some, but not all, cognitive domains as compared with the HC group. Interestingly, our FEP subgroups did not differ on clinical measures, including positive, negative, or total psychosis symptoms, or treatment response. A recent systematic review by Carruthers et al.¹¹ found that cognitive subgroups in psychosis spectrum disorders are not consistently distinguished by most non-cognitive clinical characteristics, including positive symptoms, suggesting that cognitive heterogeneity may reflect a dimension of illness that is at least partially independent of symptom severity.

Regardless of the best-fitting cluster solution, studies generally report a cognitively spared subtype of FEP participants who have average performance across all domains^28,31, as well as the existence of a subgroup with severe cognitive deficits^32–35. Our results support these findings, as well as a handful of studies that have found a two-cluster solution to identify cognitive subgroups of psychosis participants^16,36. For example, Fernandez Linsenbarth et al.¹⁶, used a k-means cluster approach and identified a two-cluster solution. Interestingly, this study found that the cognitive subgroups differed based on clinical characteristics such as positive and negative symptoms, a pattern not replicated in our findings. This discrepancy may be due to sample differences, as our study included only FEP participants, whereas the Fernandez-Linsenbarth et al. study examined both chronic schizophrenia and FEP participants. Another study by Amoretti and colleagues³⁶ also identified a two-cluster solution using the Partition Around Medoids (PAM) algorithm with clinical differences related to symptoms and duration of untreated psychosis. Again, our divergent clinical findings may relate to sample characteristics, as this study had both a higher cognitively intact subgroup and a lower mean duration of untreated psychosis.

Additionally, while premorbid intellectual functioning did not significantly differ between groups and was within the average range across the sample, the marked cognitive deficits observed in the impaired FEP subgroup suggest a decline from a previously intact baseline and support the possibility of illness-related cognitive decline occurring early in the course of psychosis. It is also important to note that the intact subgroup demonstrated average to low performance across domains but still performed below healthy controls, indicating relative rather than normative cognitive preservation. These results complement other studies that have used current neuropsychological functioning and estimated premorbid intellectual functioning to characterize cognitive decline in psychosis participants^19,20,37,38. For example, Woodward et al.²⁰, identified deteriorated and compromised subgroups of psychosis participants with current impaired neuropsychological functioning, with each subgroup demonstrating unique patterns of brain structure abnormalities. Additionally, work by Menkes³⁷ and Weickert¹⁹ has classified subgroups of participants as cognitively deteriorated, compromised, and intact, suggesting that additional subgroup structure (e.g., a chronically impaired trajectory) may exist beyond a two-cluster solution and may not be fully captured by all analytic approaches. Nevertheless, these results highlight that cognitive decline is not a universal finding across the psychosis spectrum disorders.

Importantly, these findings should also be considered in the context of broader evidence indicating that cognitive vulnerabilities in psychosis often emerge early in development, with impairments observed in childhood^39–42 and further differentiation in clinical high risk samples^43–47. Across these stages, findings suggest substantial inter-individual variability in both the severity and pattern of cognitive functioning. This heterogeneity further supports models proposing the existence of distinct cognitive subgroups with different developmental courses rather than a singular pattern of progressive decline.

The most compelling and interpretable findings in the current study were the baseline resting-state functional connectivity differences, which revealed a clear neural distinction between cognitively impaired FEP participants and healthy controls. Specifically, the impaired subgroup showed reduced baseline functional connectivity within the dorsal attention network (DAN; right superior parietal lobule) and executive control network (ECN; right middle frontal gyrus). These network disruptions were not observed in the intact FEP subgroup and did not differ between FEP subgroups, suggesting that disrupted connectivity within attention and executive control networks may serve as a neural signature of early cognitive dysfunction in psychosis. Notably, these networks are more closely associated with domain-specific attentional and executive processes^48,49, which may be more sensitive to early dysfunction than global estimates of intellectual functioning, such as premorbid IQ. Our baseline results also align with prior findings demonstrating that frontoparietal dysconnectivity is more pronounced in cognitively impaired participants and persists even after controlling for clinical symptoms and medication burden, supporting the interpretation that such network alterations reflect specific biological correlates of cognitive subtype rather than general illness severity¹⁷.

Longitudinal analyses revealed several significant group × time interactions across both the DAN and ECN. However, these trajectories varied across clusters and subgroups, and the magnitude and direction of change were not always consistent. In healthy controls, functional connectivity showed small-magnitude bidirectional changes over time across both networks. Some degree of variability in resting-state connectivity over time in healthy controls is expected and has been reported in prior longitudinal rs-fMRI studies⁵⁰. The intact FEP subgroup showed patterns of longitudinal change that were generally comparable to those observed in healthy controls, and the impaired subgroup tended to show declining DAN connectivity over time. ECN trajectories were more variable, with some evidence of early increases, followed by later declines in the intact subgroup. Longitudinal rs-FC studies involving antipsychotic-naïve FEP participants show that following the initiation of psychopharmacological treatment, rsFC strength in differential brain areas progressively increases and correlates with symptom improvement⁵¹. For the intact FEP group, the relative pattern of increasing ECN connectivity over time may indicate a partial medication-induced compensatory mechanism. However, given that no significant differences in treatment response were observed between our subgroups during the current follow-up period, it is possible that a longer observation window is needed to capture more robust or delayed effects of medication on network connectivity and clinical outcomes.

Conclusion

This study had several strengths and limitations. We included a large sample of medication-naïve participants with first-episode psychosis, and since FEP participants were administered the cognitive test battery at baseline, their performance is free from the potential confounding effects of antipsychotic medication. Moreover, with most participants having no prior exposure to antipsychotics before the scan, we were able to conduct the study with minimal concerns about medication-related imaging confounds at the baseline visit. One limitation of the current study is the focus on a small set of cognitive domains, excluding others, such as motor speed and social cognition, which are also commonly affected in psychosis. Next, the RBANS combines working memory and processing speed into a single attentional domain, which may obscure the distinct contribution of processing speed, a domain known to be significantly impaired in psychosis^2,4,5,9. Future research should examine whether these findings extend to a broader range of cognitive domains. Additionally, single-word-reading tests such as the NART and AmNART, are commonly used to estimate premorbid intellectual functioning⁵² but are influenced by educational experience⁵². Participants in our FEP sample had relatively high educational attainment (approximately one year beyond high school), which is consistent with some prior studies^20,37. While we cannot fully rule out that these differences contributed to group effects, the observed cognitive impairments are unlikely to be solely explained by education. We also assessed positive symptoms using the BPRS, which may be less sensitive than more targeted measures of positive symptomatology, potentially limiting our ability to detect subgroup differences in positive symptoms. Another limitation of this study is that the longitudinal connectivity patterns were complex and sometimes inconsistent across regions and subgroups, limiting the ability to draw clear inferences about network change over time. The longitudinal analyses were also conducted in a modest sample, which may limit statistical power to detect smaller effects. Although linear mixed-effects models maximize the use of available data, replication in larger samples with a longer follow-up period will be important to confirm the stability and meaning of these findings. Additionally, we did not control for the possible confounding effects of nicotine use in the sample, and we elected not to exclude first-episode psychosis participants based on cannabis use, as this would have introduced sample bias and limited the generalizability of the results.

Overall, these findings underscore that first-episode psychosis is marked by meaningful cognitive heterogeneity and corresponding differences in baseline network connectivity. Distinct patterns may reflect early markers of cognitive impairment or preserved functioning. These findings highlight the need for a more nuanced approach in both research and clinical practice to account for individual variability in cognitive profiles within this population.

Materials and methods

Participants

First-episode psychosis participants were recruited through outpatient clinics, inpatient units, and the emergency department at the University of Alabama at Birmingham (UAB). All study procedures were approved by the UAB Institutional Review Board, and written informed consent was obtained prior to participation. Participants with first-episode psychosis were excluded if they had major neurological or medical conditions, a history of significant head trauma, substance use disorders (excluding nicotine or cannabis) within one month of imaging, were pregnant or breastfeeding, had MRI contraindications, or had more than five days of lifetime antipsychotic exposure. Participants with first-episode psychosis were considered medication-naïve if they had <5 days of lifetime antipsychotic exposure prior to the baseline scan, consistent with our prior work^53–56. First-episode psychosis was defined as the presence of a first lifetime episode of a diagnosable psychosis-spectrum disorder in individuals with no prior history of psychotic illness. Twenty-three first-episode psychosis participants had received any antipsychotic medication, of whom 19 had only 1-2 doses (risperidone, haloperidol, or ziprasidone) for acute symptom control. There was no exclusion based on maximum duration of symptoms (e.g., duration of untreated psychosis) prior to study entry. Healthy controls (HCs) were recruited and were matched on age, sex, and parental socioeconomic status (SES). In addition to the exclusion criteria listed above, HCs were excluded if they had a personal history of psychiatric illness or if a first-degree relative had a psychotic diagnosis. Diagnoses were established using DSM-5 criteria by two board-certified psychiatrists, based on comprehensive clinical and historical data. Given the longitudinal nature of the study (ClinicalTrials.gov Identifiers: NCT02034253, NCT03442101), follow-up assessments over several months contributed to the final consensus diagnoses. These included schizophrenia, schizophreniform disorder, psychosis not otherwise specified (NOS), brief psychotic disorder, schizoaffective disorder, bipolar disorder with psychosis, and major depressive disorder with psychosis.

Symptom severity was assessed using the 18-item Brief Psychiatric Rating Scale, with a 4-item positive symptom subscale (conceptual disorganization, suspiciousness, hallucinations, and unusual thought content). Baseline Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) was used to assess cognition⁵⁷. The RBANS⁵⁸ is a well-standardized neuropsychological test battery composed of 12 subtests that form the following 5 age-adjusted domain scores: immediate memory, visuospatial/constructional, language, attention, and delayed memory. For reference, RBANS index and total scores have a mean of 100 and standard deviation of 15, with qualitative ranges defined as follows: average (90-109), low average (80-89), below average 70-79), and impaired (<70). Premorbid intellectual functioning and quality of life were also assessed at baseline using the American National Adult Reading Test (AmNART)⁵⁹ and the Quality of Life Scale⁶⁰, respectively.

Participants with first-episode psychosis in this longitudinal study were enrolled in a 16-week trial of oral risperidone using a flexible dosing protocol. Risperidone was selected due to its frequent use in FEP treatment⁶¹, its availability as a generic medication, and its relative affordability among second-generation antipsychotics in the U.S. The initial dose ranged from 0.5 to 1 mg, with titration in 1–2 mg increments based on clinical response and side effects. If FEP participants experienced intolerable side effects, as judged by the study physician, they were switched to aripiprazole, which was initiated at 2-5 mg and adjusted in 2.5–10 mg increments.

Clinically indicated adjunctive medications, including antidepressants, were allowed following the initial MRI session. Among the medication-naïve FEP participants, 58 took the following concomitant medications during the trial: sertraline (30), trazodone (27), lorazepam (11), bupropion (9), fluoxetine (7), escitalopram (7), propranolol (6), hydroxyzine (3), prazosin (3), diphenhydramine hydrochloride (3), valproic acid (1), lithium (2), lisdexamfetamine (1), amphetamine-dextroamphetamine (1), buspirone (1), melatonin (1), amitriptyline (1), ranitidine (1), citalopram (1), buprenorphine/naloxone (1), levothyroxine sodium (1), metformin (1), and diazepam (1). Twenty-nine participants were treated with more than one concomitant medication during the study period. Medication adherence was assessed via pill counts at each visit. The time between neuroimaging and antipsychotic initiation was less than 24 h

Treatment response was quantified as the percentage reduction in scores on the BPRS positive symptoms subscale from baseline to week 16, after adjusting for the minimum possible score, with greater percent change reflecting greater improvement. Response data were available for 59 FEP participants.

$$treatmentresponse\%=\frac{BPR{S}_{\mathit{\left(}Baseline\right)}-BPR{S}_{\left(Week16\right)}}{BPR{S}_{\left(Baseline\right)}-4}\times 100$$

The current dataset has participant overlap with our recent papers^{53–56,62–68}. Data for NCT 034420101 is deposited in the NDA data archive and shared per NIMH agreement.

Data acquisition

Imaging was acquired via a 3 T whole-body Siemens MAGNETOM Prisma MRI Scanner equipped with a 20-channel head coil. A high-resolution structural T1-weighted scan was used for anatomical reference, to guide voxel placement, and for voxel tissue segmentation. Resting state functional MRI data were acquired in opposing phase encoding directions (anterior > posterior and posterior > anterior; TR = 1550 ms; TE = 37.80 ms; flip angle = 71°, FOV = 104 mm²; multi-band acceleration factor = 4; voxel size = 2 mm³; 225 volumes). Participants underwent ~12-min resting fMRI scan in which they were asked to keep their eyes open and stare passively ahead.

Data preprocessing

The first 10 volumes on each fMRI scan were discarded to allow for signal equilibration, and field inhomogeneities were corrected in FSL’s topup and merged, resulting in a single 4-dimensional image of 430 total volumes⁶⁹. Data preprocessing was conducted in the CONN functional connectivity toolbox version 20.b⁷⁰. Preprocessing consisted of slice-timing and motion correction, structural coregistration, normalization to the Montreal Neurological Institute (MNI) space, low-bandpass filtering (0.008 < f < 0.08 Hz), and spatial smoothing with a 4-mm full-width-at-half-maximum Gaussian kernel, and artifact detection and removal with ART-based scrubbing (composite volume-to-volume motion > 0.5 mm and intensity >3 Standard Deviations). Framewise displacement and percentage of censored data were calculated. Motion outliers were identified using the Artifact Detection Tools toolbox (NeuroImaging Tools and Resources Collaboratory) and then censored (composite volume-to-volume motion >0.5 mm and intensity >3 SDs). Participants with less than 50% of usable data remaining after censoring were excluded from functional connectivity analyses. Additionally, six motion parameters derived from rigid-body realignment and their derivatives, as well as the first component time series derived from CSF and white matter masks using aCompCor within the CONN toolbox and corresponding derivatives, were regressed out from the signal.

Statistical analyses

Cluster analysis

RBANS domain scores (immediate memory, visuospatial/constructional, language, attention, and delayed memory) for first episode psychosis participants were entered in a cluster analysis using the K-means cluster approach⁷¹. The squared Euclidean distance was used as the dissimilarity measure to calculate the distance between scores. The optimal number of clusters was determined using the NbClust package in R⁷², which evaluates the agreement between 30 clustering indices. The range for the number of clusters was set to 2–10. The stability of the cluster solution was assessed using a nonparametric bootstrapping procedure, resampling the data with replacement 1000 times. Jaccard cluster similarity scores, representing the proportion of subjects grouped together in both the original and resampled data, were calculated. Mean Jaccard scores exceeding 0.75 indicate that the clusters are valid and stable^73,74.

Clinical analyses

One-way ANOVA tests were conducted to examine group differences between FEP subgroups and HC on AmNART scores and RBANS cognitive domain and total scores. Significant omnibus effects were followed up with post-hoc pairwise comparisons. Bonferroni corrections were applied to these pairwise comparisons within each ANOVA to account for multiple testing. Given three groups (further described below), this resulted in three pairwise comparisons per outcome (impaired FEP vs. intact FEP, impaired FEP vs. HC, intact FEP vs. HC). All p-values reported are two-sided. Similarly, one-way ANOVA or ${\chi }^2$ tests were conducted to examine group differences in demographics (e.g., age, sex, education, race/ethnicity, parental occupation) and clinical variables (e.g., BPRS, quality of life, diagnosis, substance use, smoking). Significant effects were followed up using post hoc pairwise comparisons, with Bonferroni corrections applied within each variable to account for three group comparisons.

Functional connectivity analyses

The regions of interest (ROIs) in the functional connectivity (FC) analyses were the posterior cingulate cortex (default mode network), right intraparietal sulcus (dorsal attention network), posterior parietal cortex (executive control network), and anterior cingulate cortex (salience network). These ROIs were derived using CONN’s independent component analysis of the Human Connectome Project dataset. For all three time points (baseline, week 6, and week 16), residual time series from each ROI were extracted, correlated with every other voxel in the brain, and z-transformed using CONN’s seed-to-voxel analysis. The resulting voxelwise FC maps were analyzed using AFNI’s 3dLME linear mixed-effects model, incorporating two within-subjects factors (time and framewise displacement), one between-subjects factor (group), and two covariates (age and sex). Random effects were specified at the subject level, including random intercepts and random slopes for framewise displacement (FD), allowing for between-subject variability and individual differences in the impact of head motion on connectivity estimates⁷⁵. Utilizing a mixed-effects model is an advantageous approach, as it leverages all available observations without using listwise deletion, thereby preserving data. The mask option was applied to restrict all analyses to the networks of interest, implementing a small-volume correction. These network-specific masks were created using AFNI's 3dMean by averaging all subject-level z-transformed FC maps for each network at baseline. The resulting average maps were thresholded at a z-value of 0.20, cluster-corrected (100 voxels), and binarized to generate the final masks. Any significant group × time interactions were further examined by extracting individual data from the surviving clusters identified by 3dLME and then submitted to separate linear mixed models to assess change over time (separately for each group) and pairwise differences between timepoints and one-way ANOVAS at each timepoint followed by pairwise analyses using the Statistical Package for Social Sciences (IBM Corp. Released 2023. IBM SPSS Statistics for Macintosh, Version 29.0.2.0 Armonk, NY: IBM Corp). To account for multiple comparisons, AFNI’s 3dClustSim was applied to each network using the spatial autocorrection function (-acf option) to control the family-wise error (FWE) rate at p < .05 and applied a false discovery rate correction (FDR; p = 0.05) for all post-hoc analyses.

Author contributions

Dr. Lahti had full access to all the data in the study and takes responsibility for the integrity and accuracy of the data analysis. Study concept and design: A.C.L. Acquisition of data: A.C.L. Analysis and interpretation of data: All authors. Drafting of the manuscript: All authors. Statistical analysis: H.N.P., J.O.M. Obtained funding: A.C.L. Administrative, technical, or material support: A.C.L. Study supervision: A.C.L.

Funding

This work was supported by the National Institutes of Health (R01MH102951 and R01MH113800).

Data availability

The datasets generated during and/or analyzed during the current study are available in the NIMH Data Archive, https://nda.nih.gov/edit_collection.html?id=2916, NCT 034420101.

Competing interests

Authors declare no conflicts of interest, including relevant financial interests, activities, relationships, and affiliations. The funding agency had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41537-026-00771-w.