Introduction
Schizophrenia is a severe and chronic psychiatric disorder that affects ~1% of the global population and remains a leading cause of long-term disability (Marder & Cannon, Reference Marder and Cannon2019). It manifests through a heterogeneous constellation of symptoms, encompassing positive symptoms, negative symptoms, and pervasive cognitive impairments involving attention, working memory (WM), and executive functioning (Lett et al., Reference Lett, Voineskos, Kennedy, Levine and Daskalakis2014). While antipsychotic drugs have proven effective in reducing positive symptoms, current antipsychotics have limited effects on cognition (Feber et al., Reference Feber, Peter, Chiocchia, Schneider-Thoma, Siafis, Bighelli and Leucht2025; Keefe et al., Reference Keefe, Bilder, Davis, Harvey, Palmer, Gold and Lieberman2007). Among these, WM impairment stands out as a central and persistent deficit (Lett et al., Reference Lett, Voineskos, Kennedy, Levine and Daskalakis2014): it is evident early in the illness (Nuechterlein, Ventura, Subotnik, & Bartzokis, Reference Nuechterlein, Ventura, Subotnik and Bartzokis2014), endures even when other clinical symptoms have remitted, and strongly predicts clinical (Jenkins, Bodapati, Sharma, & Rosen, Reference Jenkins, Bodapati, Sharma and Rosen2018) and functional outcomes (Borgan et al., Reference Borgan, O’Daly, Veronese, Reis Marques, Laurikainen, Hietala and Howes2021; Fu, Czajkowski, Rund, & Torgalsbøen, Reference Fu, Czajkowski, Rund and Torgalsbøen2017) such as occupational attainment and social recovery. Notably, WM impairments remain largely refractory to available treatments, resulting in a persistent illness burden (Park et al., Reference Park, Püschel, Sauter, Rentsch and Hell1999) that may lead to recurrent relapses (Hui et al., Reference Hui, Li, Li, Lee, Chang, Chan and Chen2016) and worsen negative symptoms (Carter et al., Reference Carter, Robertson, Nordahl, Chaderjian, Kraft and O’Shora-Celaya1996; González-Ortega et al., Reference González-Ortega, de los Mozos, Echeburúa, Mezo, Besga, Ruiz de Azúa and González-Pinto2013; Nejad et al., Reference Nejad, Madsen, Ebdrup, Siebner, Rasmussen, Aggernæs and Baaré2013; Pantelis et al., Reference Pantelis, Stuart, Nelson, Robbins and Barnes2001). Despite their clinical significance, WM deficits in schizophrenia show only modest and variable improvement following current cognitive and behavioral interventions, highlighting the urgent need to better delineate their underlying neural mechanisms (Lejeune, Northrop, & Kurtz, Reference Lejeune, Northrop and Kurtz2021).
WM refers to the ability to temporarily store, manipulate, and update information in support of goal-directed behavior and serves as a foundation for higher-order cognition (Baddeley, Reference Baddeley1992; Hartley & Hitch, Reference Hartley, Hitch and Worrell2022). In schizophrenia, WM deficits manifest as difficulties in task switching and coping with complex cognitive demands (Forbes, Carrick, McIntosh, & Lawrie, Reference Forbes, Carrick, McIntosh and Lawrie2009), and are commonly assessed using paradigms such as the n-back task. During WM tasks, extant neuroimaging evidence indicates that patients with schizophrenia show abnormal network-level coordination (Nielsen et al., Reference Nielsen, Madsen, Wang, Liu, Friston and Zhou2017; Repovš & Barch, Reference Repovš and Barch2012; Wang et al., Reference Wang, Liu, Ford, Deng, Zhang, Yang and Palaniyappan2024, Reference Wang, Yang, Yang, Cheng, Tan, Huang and Palaniyappan2025; Yang et al., Reference Yang, Liu, Wang, Tan, Huang, Ouyang and Palaniyappan2025). The triple network, which includes the default mode network (DMN), the frontoparietal control network (FPN), and the salience network (SAL), has been proposed as a key systems-level model for understanding WM, as effective performance requires dynamic coordination among these networks (Menon, Reference Menon2011; Wu & Jiang, Reference Wu and Jiang2020). In healthy individuals, WM task performance typically involves the suppression of the DMN and activation of task-positive networks, with the opposing interaction between these systems playing a key role in balancing internal and external attention (Singh & Fawcett, Reference Singh and Fawcett2008). In contrast, schizophrenia is characterized by disrupted coordination among these large-scale networks (Menon, Palaniyappan, & Supekar, Reference Menon, Palaniyappan and Supekar2023). During the WM task, schizophrenia shows impaired suppression (hyperactivity) of the DMN (Pu et al., Reference Pu, Luo, Palaniyappan, Xue, Yao, Feng and Liu2016; Wang et al., Reference Wang, Xi, Liu, Deng, Zhang, Cao and Palaniyappan2022; Wu & Jiang, Reference Wu and Jiang2020; Yang et al., Reference Yang, Pu, Wu, Chen, Lee, Liu and Palaniyappan2020), reduced FPN (Greenman et al., Reference Greenman, La, Shah, Chen, Berman, Weinberger and Tan2020; Nielsen et al., Reference Nielsen, Madsen, Wang, Liu, Friston and Zhou2017; Pu et al., Reference Pu, Luo, Palaniyappan, Xue, Yao, Feng and Liu2016) coupling, and disrupted interactions among DMN, FPN, and SAL (Braun et al., Reference Braun, Harneit, Pergola, Menara, Schäfer, Betzel and Tost2021; Luo et al., Reference Luo, Pan, Gu, Simmonite, Francis, Liddle and Palaniyappan2020; Pu et al., Reference Pu, Luo, Palaniyappan, Xue, Yao, Feng and Liu2016; Zhou et al., Reference Zhou, Zeidman, Wu, Razi, Chen, Yang and Friston2018). Moreover, while healthy individuals efficiently deactivate DMN regions and recruit task-relevant hubs, patients often fail to suppress the DMN and instead engage task-irrelevant regions during n-back tasks (Huang et al., Reference Huang, Rogers, Anticevic, Blackford, Heckers and Woodward2019; Pu et al., Reference Pu, Luo, Palaniyappan, Xue, Yao, Feng and Liu2016; Williams et al., Reference Williams, Zheng, Tubiolo, Luceno, Gil, Girgis and Van Snellenberg2023; Wu & Jiang, Reference Wu and Jiang2020). Taken together, these alterations indicate inefficient allocation of neural resources, likely due to reconfiguration of connectivity patterns during WM, leading to poorer performance. Against this backdrop, we turn to the FPN, the principal executor of control signals during WM within the triple-network architecture.
Most prior WM studies in schizophrenia have relied on static functional connectivity (sFC), which assumes stationarity of signals and fails to capture the time-varying reconfiguration of large-scale networks. Functional connectivity, however, is not stationary but rather fluctuates over time (Chang & Glover, Reference Chang and Glover2010; Hutchison et al., Reference Hutchison, Womelsdorf, Gati, Everling and Menon2013). These fluctuations (Long, Liu, & Liu, Reference Long, Liu and Liu2023; Preti, Bolton, & Van De Ville, Reference Preti, Bolton and Van De Ville2017) are often referred to as dynamic functional connectivity (dFC) and offer valuable insights into brain organization by revealing subtle temporal dynamics (Allen et al., Reference Allen, Damaraju, Plis, Erhardt, Eichele and Calhoun2014; Fu et al., Reference Fu, Sui, Turner, Du, Assaf, Pearlson and Calhoun2021; Hutchison et al., Reference Hutchison, Womelsdorf, Gati, Everling and Menon2013; Sakoğlu et al., Reference Sakoğlu, Pearlson, Kiehl, Wang, Michael and Calhoun2010) that static FC analyses are unable to capture. Recent work has therefore shifted toward dynamic measures, including state occupancy/transition rates (Braun et al., Reference Braun, Harneit, Pergola, Menara, Schäfer, Betzel and Tost2021; Wang et al., Reference Wang, Yang, Yang, Cheng, Tan, Huang and Palaniyappan2025), dFC (Cassidy et al., Reference Cassidy, Van Snellenberg, Benavides, Slifstein, Wang, Moore and Horga2016), and graph-based variability metrics (Cheng et al., Reference Cheng, Liu, Wang, Yang and Yang2025; Dimitriadis et al., Reference Dimitriadis, Lancaster, Perry, Tansey, Jones, Singh and Linden2021). These studies consistently demonstrate abnormal frontoparietal dynamics in schizophrenia during WM tasks, such as reduced stability of FPN-DMN antagonistic states (Wang et al., Reference Wang, Yang, Yang, Cheng, Tan, Huang and Palaniyappan2025), blunted load-dependent modulation within FPN (Nielsen et al., Reference Nielsen, Madsen, Wang, Liu, Friston and Zhou2017), and inefficient reallocation of resources via salience network driven switching (Luo et al., Reference Luo, Pan, Gu, Simmonite, Francis, Liddle and Palaniyappan2020).
Temporal variability (TV) (Long et al., Reference Long, Cao, Yan, Chen, Li, Castellanos and Liu2020; Long et al., Reference Long, Liu, Chan, Wu, Xue, Pan and Pu2020; Sun et al., Reference Sun, Liu, Rolls, Chen, Yao, Yang and Qiu2019; Zhang et al., Reference Zhang, Cheng, Liu, Zhang, Lei, Yao and Feng2016) is a dynamic connectivity metric that focuses on the changing configurations in functional connectivity strength across time windows. In schizophrenia, TV has been primarily examined in the resting state. Patients consistently show increased variability within large-scale networks such as the DMN and SAL (Long, Liu, et al., Reference Long, Liu, Chan, Wu, Xue, Pan and Pu2020; Rolls, Cheng, & Feng, Reference Rolls, Cheng and Feng2021), and these alterations have been associated with greater symptom severity (Long, Liu, et al., Reference Long, Liu, Chan, Wu, Xue, Pan and Pu2020) and cognitive impairments (Yue et al., Reference Yue, Li, Shi, Lin, Sun and Lu2018; Zhang et al., Reference Zhang, Cheng, Liu, Zhang, Lei, Yao and Feng2016. In this study, we estimate TV for the first time during a WM paradigm in schizophrenia, as a metric to assess load-dependent network flexibility (ΔTV, defined as the difference in variability between high- and low-load WM conditions) during task performance.
Aberrant brain network properties are increasingly recognized to be shaped by environment factors, with childhood trauma (CT) emerging as a major environmental risk factor for psychosis (Kraan et al., Reference Kraan, van Dam, Velthorst, de Ruigh, Nieman, Durston and de Haan2015). Evidence suggests that exposure to CT substantially elevates the risk of schizophrenia, with some studies estimating up to a threefold increase compared with nonexposed individuals (Aas et al., Reference Aas, Henry, Andreassen, Bellivier, Melle and Etain2016;Croft et al., Reference Croft, Heron, Teufel, Cannon, Wolke, Thompson and Zammit2019; De-Nardin et al., Reference De-Nardin, Muratori, Ribeiro, Huguet and Salgado2022). This risk follows a dose–response pattern, whereby earlier, more severe, and more frequent trauma confers proportionally greater vulnerability (Croft et al., Reference Croft, Heron, Teufel, Cannon, Wolke, Thompson and Zammit2019; De-Nardin et al., Reference De-Nardin, Muratori, Ribeiro, Huguet and Salgado2022). Beyond illness onset, CT has been linked to worse clinical outcomes, including greater symptom severity (Ruby et al., Reference Ruby, Rothman, Corcoran, Goetz and Malaspina2017), poorer treatment response (Kilian et al., Reference Kilian, Asmal, Phahladira, Plessis, Luckhoff, Scheffler and Emsley2020), and persistent cognitive impairments (S. Kilian et al., Reference Kilian, Asmal, Chiliza, Olivier, Phahladira, Scheffler and Emsley2018; Wells et al., Reference Wells, Jacomb, Swaminathan, Sundram, Weinberg, Bruggemann and Weickert2020) such as deficits in WM (Popovic et al., Reference Popovic, Schmitt, Kaurani, Senner, Papiol, Malchow and Falkai2019). Recent evidence also indicates that CT may exert long-lasting neurodevelopmental effects, altering large-scale brain networks that support cognition and thereby shaping trajectories of vulnerability to psychosis (Lipner et al., Reference Lipner, O’Brien, Pike, Ered and Ellman2023; Liu et al., Reference Liu, Mendonça, Johnson, O’Reilly, Bartmann, Marlow and Wolke2021). Notably, converging evidence suggests that childhood trauma affects prefrontal and frontoparietal control systems, which are highly sensitive to developmental adversity and are central to executive control and WM (Gerin et al., Reference Gerin, Viding, Herringa, Russell and McCrory2023; Philip et al., Reference Philip, Sweet, Tyrka, Carpenter, Albright, Price and Carpenter2016, Reference Philip, Sweet, Tyrka, Price, Carpenter, Kuras and Niaura2013).
Against this background, this study integrates CT, TV, and WM performance into a unified framework. We hypothesize that CT will affect WM via altered TV. Specifically, we focus on the triple-network systems (DMN, FPN, and SAL), with particular emphasis on the FPN as the core control hub that is likely to be disrupted in the presence of childhood trauma. Based on prior evidence and theoretical considerations, we aimed to: (1) determine whether childhood trauma affects triple-network TV in patients and control subjects; (2) examine whether ΔTV within the FPN affects WM performance and symptom burden in patients with schizophrenia; and (3) test whether altered TV of the FPN mediates the relationship between trauma exposure and WM deficits in schizophrenia. Establishing these relationships will clarify the neurophysiological mechanism by which exposure to CT may affect core cognitive functions and symptoms in schizophrenia. A schematic overview of the research design is shown in Figure 1.
Schematic illustration of the study design and analytic pipeline. Note: n-back task-based fMRI data were collected from patients with SZ and HCs. Temporal variability (TV) of dynamic functional connectivity (dFC) was assessed under different task conditions, and ΔTV was calculated as the difference in TV between the 2-back and 0-back conditions. Participants were further stratified by the presence or absence of childhood trauma, and subgroup differences in ΔTV, as well as associations with clinical symptoms and cognitive performance, were examined.
Materials and methods
Participants
A total of 190 patients with schizophrenia were recruited from the Second Xiangya Hospital of Central South University. A total of 117 healthy controls (HCs) were recruited from Changsha and its surrounding areas. All participants underwent brief, unstructured interviews conducted by psychiatrists and provided written informed consent before taking part in the study. The study was approved by the Ethics Committee of the Second Xiangya Hospital of Central South University. All procedures were carried out in strict accordance with the Declaration of Helsinki. Inclusion criteria for the patient group were right-handedness and a confirmed diagnosis of schizophrenia according to DSM-5. Exclusion criteria for patients were: (1) a primary psychiatric diagnosis other than schizophrenia; (2) a history of serious physical illness or neurological conditions; (3) current or past substance abuse or dependence (except for nicotine use); and (4) metal implants (e.g. electronic devices) or other contraindications to MRI. HCs were screened using the same general exclusion criteria, with the additional requirement that none met DSM-5 criteria for any psychiatric illness.
Clinical assessment
The severity of clinical syndromes in schizophrenia was evaluated by trained psychiatrists using the Positive and Negative Syndrome Scale (PANSS) (Kay, Fiszbein, & Opler, Reference Kay, Fiszbein and Opler1987). The PANSS comprises 30 items divided into three subscales: positive symptoms, negative symptoms, and general psychopathology. It assesses symptom severity over the past month, with each item rated on a 7-point scale and offers a thorough evaluation of the patient’s mental state.
Following previous research (Shafer & Dazzi, Reference Shafer and Dazzi2019), we derived PANSS five-factor scores, which provide a more nuanced dimensional structure. These include negative symptoms, positive symptoms, disorganization (often termed cognitive), excitement/activity (resistance), and affect (often termed depression-anxiety).
Assessment of childhood trauma
Childhood trauma exposure was assessed using the Chinese version of the Childhood Trauma Questionnaire (CTQ) (Bernstein et al., Reference Bernstein, Stein, Newcomb, Walker, Pogge, Ahluvalia and Zule2003), a validated (Zhao et al., Reference Zhao, Zhang, Li, Zhou, Li and Yang2005) self-reported instrument that has been widely used in previous research (Huang et al., Reference Huang, Liu, Cao, Yang, Wu and Long2021; Li et al., Reference Li, Liu, Zhu, Zhang, Tang and Wang2014; Lu et al., Reference Lu, Jian, Dong, Gao, Zhang, Chen and Liu2020; Yang et al., Reference Yang, Sheng, Ge, Zhang, Huang, Cui and Zhang2022). The CTQ consists of 25 clinical items organized into five subscales: emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect, each consisting of five items. Responses are scored on a 5-point Likert scale ranging from 1 (never) to 5 (very often), reflecting the frequency of the reported experiences. Based on the cutoff criteria used in earlier studies (Huang et al., Reference Huang, Liu, Cao, Yang, Wu and Long2021; Lu et al., Reference Lu, Gao, Huang, Li and Xu2016; Lu et al., Reference Lu, Gao, Wei, Wang, Hu, Huang and Li2017; Yang et al., Reference Yang, Cheng, Mo, Bai, Shen, Liu and Xu2017), participants were considered to have experienced childhood trauma if their scores exceeded the predefined thresholds for any subscale. The cutoffs were as follows: emotional abuse (≥13), physical abuse (≥10), sexual abuse (≥8), emotional neglect (≥15), or physical neglect (≥10).
Imaging data acquisition and preprocessing
The MRI imaging data were acquired from a Siemens 3-T scanner with the following parameters: 32 slices, 64 × 64 matrix, 5 mm slice thickness with no inter-slice gap, 90° flip angle, field of view of 240 × 240 mm2, repetition time (TR) of 2000 ms, echo time (TE) of 30 ms, and a total of 253 time points collected.
Preprocessing of the data was performed using the DPARSF software (Chao-Gan & Yu-Feng, Reference Chao-Gan and Yu-Feng2010; Yan, Wang, Zuo, & Zang, Reference Yan, Wang, Zuo and Zang2016) (http://rfmri.org/DPARSF), adhering to a standardized processing pipeline. The initial five volumes were discarded, after which slice-timing correction, motion realignment, spatial normalization, and temporal band-pass filtering (0.01–0.10 Hz) were applied. To control for nuisance signals, covariates including white matter and cerebrospinal fluid signals were regressed out. Subsequently, experienced researchers manually evaluated image quality. To ensure the reliability of the data, mean framewise displacement (FD) was computed for each subject. Datasets exhibiting excessive head motion (mean FD > 0.2 mm) or poor image quality, as determined by visual inspection, were excluded from further analyses. After image quality control, a total of 162 individuals with schizophrenia and 99 HCs were included in the study. For each participant, time series were extracted from 264 regions of interest (ROIs) based on Power et al.’s (Reference Power, Cohen, Nelson, Wig, Barnes, Church and Petersen2011) atlas. These ROIs were assigned to nine functional networks according to previous publications: default mode network (DMN), salience network (SAL), visual network (VIS), subcortical network (SUB), auditory network (AUD), frontoparietal network (FPN), cingulo-opercular network (CON), sensorimotor network (SM), and attention network (ATT).
n-back task and performance
We employed the ‘n-back’ paradigm to assess WM function, utilizing both 0-back and 2-back conditions in this study. In the 0-back task, participants were instructed to press the right button whenever the letter ‘X’ or ‘x’ appeared, and the left button for all other letters. In the 2-back task, they were required to press the right button when the current letter matched the one presented two trials earlier, regardless of case (e.g. ‘D’ and ‘d’ were considered the same); otherwise, they pressed the left button.
The entire task-based fMRI session consisted of four 40-second blocks of the 0-back task and four 40-second blocks of the 2-back task. A 2-second instruction cue was presented before each task block, and a 20-second fixation period followed it. Participants were asked to fixate on a central crosshair during the fixation. The cue and fixation were not part of the task blocks. All stimuli were white letters presented at the center of the screen, each displayed for 500 ms with an interstimulus interval of 1500 ms. The stimulus sequence was pseudorandomized throughout the session.
To assess the performance on the n-back task, we calculated both accuracy (ACC) and response time (RT) measures. Specifically, six indices were derived: target accuracy (target ACC), overall accuracy in the 0-back condition (0-back ACC), overall accuracy in the 2-back condition (2-back ACC), as well as response times for target trials (target RT), 0-back trials (0-back RT), and 2-back trials (2-back RT). Target ACC was defined as the proportion of correct responses to target stimuli. Overall ACC in the 0-back and 2-back conditions reflected the proportion of correct responses across all trials within each condition, including both target and nontarget trials. Response times were calculated only for correct trials and averaged within each condition.
Temporal variability
The method for calculating TV followed established procedures in previous literature (Long, Liu, et al., Reference Long, Liu, Chan, Wu, Xue, Pan and Pu2020; Sun et al., Reference Sun, Zhao, He, Chang, Wang, Wei and Qin2022; Zhang et al., Reference Zhang, Cheng, Liu, Zhang, Lei, Yao and Feng2016. To evaluate the TV of dFC under different task conditions, we segmented the preprocessed ROI time series to retain only the task execution periods (i.e. blocks). Each block lasted 40 seconds (20 TRs); task instruction and fixation periods were excluded. This segmentation produced four blocks for the 0-back condition and four for the 2-back condition.
For each block, pairwise Pearson correlation coefficients between all ROIs were computed to construct a dFC matrix. The TV of dFC was then calculated across blocks for each condition at three spatial scales: intra-network, inter-network, and whole-brain. Specifically, the intra-network TV for a given network (e.g. the DMN) was defined as:
$$ {TV}_{(DMN)}=1-\overline{corrcoef\left({F}_{i\;(DMN)},{F}_{j\;(DMN)}\right)},i,j=1,2,3,4;i\ne j, $$
where Fi (DMN) is a vector representing the dFC between all ROI pairs within the DMN in the ith block. Similarly, the inter-network TV of dFC between two networks (e.g. the DMN and SAL) was computed as:
$$ {\displaystyle \begin{array}{l}{TV}_{\left( DMN- SAL\right)}=1-\overline{corrcoef\left({F}_{i\left( DMN- SAL\right)},{F}_{j\left( DMN- SAL\right)}\right)},\\ {}i,j=1,2,3,4;i\ne j,\end{array}} $$
where Fi (DMN-SAL) is a vector representing the dFC between all ROIs in the DMN and all ROIs in the SAL in the ith block, where the overline denotes the average of the Pearson correlation coefficients computed across all unique block pairs. The TVs for all other individual networks and network pairs were calculated using the same method. Consequently, for each condition (0-back and 2-back), the above analyses produced a total of 9 intra-network TV values (1 per network) and 36 inter-network TV values (1 per unique network pair).
To further characterize task-related changes in TV of dFC, we computed the difference in TV between the 2-back and 0-back conditions for each participant. This difference metric, denoted as ΔTV, was calculated as:
$$ \varDelta TV={TV}_{2- back}-{TV}_{0- back} $$
where TV2-back and TV0-back represent the TV values derived from the 2-back and 0-back tasks, respectively. The ΔTV thus reflects the individual-level modulation of dynamic brain connectivity variability induced by increased WM load. The ΔTV was calculated for all intra-network and inter-network TVs.
Statistics
In this study, based on the classification criteria described above, participants with schizophrenia and HCs were further stratified into subgroups with and without a history of childhood trauma. This resulted in 110 schizophrenia patients with childhood trauma and 52 without, as well as 34 HCs with childhood trauma and 65 without.
To compare demographic characteristics between groups, independent samples t-tests were used for age, years of education, head motion (measured by mean framewise displacement, FD), and CTQ score, while chi-square tests were conducted for sex distribution.
In the HC group, we compared ΔTV between individuals with and without childhood trauma. The analysis focused on the triple network, covering both within-network (DMN, SAL, and FPN) and between-network (DMN-SAL, DMN-FPN, and SAL-FPN) connections. An analysis of covariance (ANCOVA) was performed with sex, age, and head motion as covariates. The same ANCOVA was performed in the schizophrenia group to compare those with and without trauma. Results were considered significant at p < 0.05 after FDR correction, which was applied across the six ΔTV measures within each group separately.
In patients with schizophrenia, partial correlation analyses were performed to examine the associations between the six previously mentioned ΔTV and clinical symptoms, including PANSS total score, PANSS subscale scores, and PANSS five-factor scores, while controlling for age, sex, and head motion. Additionally, partial correlation analyses were conducted in both the schizophrenia and HC groups to assess the associations between these ΔTV and n-back task performance, controlling for the same covariates. All correlation results were corrected for multiple comparisons using FDR correction, applied across the six ΔTV measures for each outcome variable within each group, with p < 0.05 considered statistically significant.
To examine whether ΔTV in the triple network mediates the association between childhood trauma and n-back performance in patients with schizophrenia, we conducted six separate mediation analyses using the PROCESS macro for SPSS, specifying Model 4. In each model, the total score of the CTQ was used as the independent variable (X), and n-back target accuracy was entered as the dependent variable (Y). Each of the six ΔTV related to the triple network was entered separately as a mediator (M). Sex, age, and mean FD were included as covariates. The significance of indirect effects was assessed using nonparametric bootstrapping (5,000 resamples), with statistical significance determined by 95% bootstrap confidence intervals.
Supplementary analyses
To further explore group-level differences in temporal variability between patients with schizophrenia and HCs, we conducted additional analyses using a subsample matched for age and sex (115 SZ and 85 HCs). Specifically, ΔTV, TV0-back, and TV2-back were compared across all functional networks using ANCOVA, with sex, age, and mean FD included as covariates.
Results
Demographic, clinical, head motion, and CTQ characteristics
As shown in Table 1, within both the SZ and HC groups, there were no significant differences in age, sex, years of education, or head motion between individuals with and without childhood trauma.
Comparison of demographic, clinical, head motion, and CTQ characteristics between CT+ and CT− individuals within the SZ and HC groups
Note: SZ, schizophrenia; HC, healthy control; CTQ, Childhood Trauma Questionnaire; CT+, individuals with childhood trauma; CT−, individuals without childhood trauma; SD, standard deviation; Mean FD, mean framewise displacement. One patient in the CTQ+ group did not complete the PANSS, resulting in a slightly reduced sample size (n = 109) for that variable.
A significant difference in total PANSS scores was observed between schizophrenia patients with and without childhood trauma, with the CT+ group showing higher scores than the CT− group (F = 0.279, p = 0.036).
ANCOVA results for ΔTV differences within groups
All p-values reported are corrected for multiple comparisons using the false discovery rate (fdr) method. As shown in Table 2 and Figure 2, within the HC group, ANCOVA revealed significant differences in ΔTV between participants with and without childhood trauma for the network with DMN (F = 7.641, p-fdr = 0.011), between the DMN and SAL (F = 8.706, p-fdr = 0.011), between the FPN and DMN (F = 10.246, p-fdr = 0.011), and between FPN and SAL (F = 7.981, p-fdr = 0.011) networks, with CT– individuals showing positive mean ΔTV values (higher flexibility with 2-back load vs. 0-back) and CT+ individuals showing negative mean values (lower flexibility with 2-back load vs. 0-back). No significant within-network changes were observed for SAL or FPN. All analyses controlled for age, sex, and head motion. No significant ΔTV differences were observed between CT+ and CT− individuals with schizophrenia across all triple-network pairs, indicating that the connectivity-related effects of trauma were not prominent in the presence of schizophrenia.
ANCOVA results for ΔTV differences within HC
Note: ANCOVA, analysis of covariance; ΔTV, change in temporal variability of dynamic functional connectivity between the 2-back and 0-back conditions; HC, healthy control; CT+, individuals with childhood trauma; CT−, individuals without childhood trauma; SD, standard deviation; fdr, false discovery rate; DMN, default mode network; SAL, salience network; FPN, frontoparietal network.
(a) Within-group effects of childhood trauma on ΔTV across the triple network: left, HCs (CT+ vs. CT−); right, SZ (CT+ vs. CT−). (b) Partial correlation results between ΔTV and clinical and task performance in schizophrenia. Note: ΔTV, change in temporal variability of dynamic functional connectivity between the 2-back and 0-back conditions; HC, healthy control; SZ, schizophrenia; CT+, individuals with childhood trauma; CT−, individuals without childhood trauma; DMN, default mode network; SAL, salience network; FPN, frontoparietal network; SumN, negative symptoms; SumP, positive symptoms; SumG, general psychopathology; SumS, specific symptoms; ACC, accuracy; RT, response time; ‘*’ indicates corrected p < 0.05.
Partial correlation analysis
As shown in Figure 2, partial correlation analyses controlling for age, sex, and head movement revealed several significant associations between ΔTV and both clinical and task performance variables within the schizophrenia group. Specifically, increased ΔTV within the FPN (i.e. higher flexibility at 2-back state) was significantly associated with lower PANSS positive symptoms (r = −0.211, p-fdr = 0.046) and lower excitement symptoms (r = −0.246, p-fdr = 0.012) based on the PANSS five-factor model, while showing a positive association with target accuracy (r = 0.303, p-fdr = 0.002) on the n-back task. ΔTV between the FPN and SAL network (i.e. higher 2-back related flexibility) was positively correlated with target accuracy (r = 0.211, p-fdr = 0.024), 0-back accuracy (r = 0.309, p-fdr = 0.002), and 2-back accuracy (r = 0.26, p-fdr = 0.018). Furthermore, both ΔTV between the FPN and DMN (r = 0.242, p-fdr = 0.015) and within DMN (r = 0.228, p-fdr = 0.018) were positively associated with target accuracy. ΔTV within the SAL network was positively correlated with 0-back accuracy (r = 0.236, p-fdr = 0.021).
Mediation analyses
Given the observed associations between ΔTV, childhood trauma in the absence of schizophrenia, and n-back performance in schizophrenia, mediation analyses were performed to test potential indirect effects in schizophrenia. As shown in Figure 3, higher CTQ scores were associated with lower target accuracy (β = −0.0032, p = 0.045) and reduced ΔTV within the FPN (β = −0.0017, p = 0.015). In turn, greater ΔTV within the FPN predicted better target accuracy (β = 0.6104, p = 0.002). The indirect effect of CTQ on target accuracy through ΔTV within the FPN was significant (β = −0.0010, 95% CI: [−0.0022, −0.0001]), indicating that childhood trauma influenced WM performance in schizophrenia, in part, via altered ΔTV within the FPN.
Mediation effect among CTQ score, ΔTV within the FPN, and target accuracy. Note: ΔTV, change in temporal variability of dynamic functional connectivity between the 2-back and 0-back conditions; FPN, frontoparietal network; CTQ, Childhood Trauma Questionnaire; ACC, accuracy.
Supplementary analyses
To further examine the spatial distribution of load-specific TV changes, we conducted ANCOVA comparing schizophrenia and HC groups on ΔTV, TV0-back, and TV2-back, controlling for age, sex, and head motion. As summarized in Supplementary Table S1, patients with schizophrenia exhibited significantly higher temporal variability than HCs across multiple network pairs, including between the VIS and DMN, between the VIS and AUD, between the visual and CON, and between the VIS and SM. These findings suggest that abnormal temporal variability during the 2-back task in schizophrenia extends beyond triple-network regions to widespread circuits involving the visual network and its interactions with higher-order control and sensory systems. In addition, a single between-network connection between VIS and AUD showed a significant group difference in ΔTV. Except for the connections reported in Supplementary Tables S1 and S2, all other tested ΔTV, TV0-back, and TV2-back connections did not show significant differences between the schizophrenia and HC groups after FDR correction.
Discussion
In HCs, individuals with childhood trauma (CT+) showed significantly lower ΔTV in triple-network connections than those without trauma exposure (CT−) within DMN, DMN-SAL, FPN-DMN, and FPN-SAL), indicating reduced load-dependent flexibility with CT exposure. On average, CT+ participants exhibited negative ΔTV values, whereas CT− participants showed positive values, indicating that early adversity is linked to diminished network flexibility across cognitive load conditions. This pattern suggests that early adversity reduces the brain’s capacity to flexibly reconfigure large-scale networks. In schizophrenia, while trauma did not have a group-level effect on triple-network TV, a higher load of trauma affected WM performance by disrupting frontoparietal TV, and was associated with a higher burden of positive and excitement symptoms.
Converging evidence supports this interpretation. Previous research has shown that childhood maltreatment alters connectivity within and between large-scale networks relevant for WM (Philip et al., Reference Philip, Sweet, Tyrka, Carpenter, Albright, Price and Carpenter2016, Reference Philip, Sweet, Tyrka, Price, Carpenter, Kuras and Niaura2013), compromising adaptive coordination for cognitive control. However, task-based neuroimaging studies directly testing the impact of childhood trauma remain relatively rare. By contrast, resting-state studies consistently report adversity-related alterations in intrinsic functional connectivity, especially between the DMN, SAL, and FPN. For example, a recent systematic review reported that childhood maltreatment is linked to reduced resting-state connectivity between the anterior insula and the dorsal anterior cingulate cortex, alongside heightened amygdala coupling within the triple network (Gerin et al., Reference Gerin, Viding, Herringa, Russell and McCrory2023). Additionally, our previous study (D. Huang et al., Reference Huang, Liu, Cao, Yang, Wu and Long2021) found that higher CTQ scores predicted reduced temporal stability across large-scale networks. These alterations were mostly found in the FPN and DMN, as well as in CON and VIS. These results suggest that early-life adversity reduces the stability and flexibility of key intrinsic networks, even in the absence of task demands. This finding aligns with our observation of reduced ΔTV during WM load in HCs.
Why are these effects detectable in healthy individuals but muted in schizophrenia? Mechanistically, childhood adversity can induce enduring neurodevelopmental alterations that disrupt network maturation (Murphy et al., Reference Murphy, Nasa, Cullinane, Raajakesary, Gazzaz, Sooknarine and Roddy2022). This may disproportionately impact late-maturing frontoparietal and salience systems (Tomoda et al., Reference Tomoda, Nishitani, Takiguchi, Fujisawa, Sugiyama and Teicher2024). By contrast, schizophrenia is characterized by baseline disruptions in triple-network dynamics, including inefficient DMN suppression during cognition and weakened FPN coupling. These disease-related dysconnectivity may overshadow incremental trauma effects (a ceiling/occlusion scenario) (Menon et al., Reference Menon, Palaniyappan and Supekar2023). This interpretation aligns with triple-network models and recent work emphasizing dysregulated dynamics during WM in schizophrenia (Supekar et al., Reference Supekar, Cai, Krishnadas, Palaniyappan and Menon2019; Wang et al., Reference Wang, Yang, Yang, Cheng, Tan, Huang and Palaniyappan2025).
In patients with schizophrenia, higher ΔTV within the FPN was associated with lower PANSS positive scores and lower excitement scores, while also relating to better n-back target accuracy (age, sex, and head motion controlled). Because ΔTV indexes load-dependent change in temporal variability from 0-back to 2-back, this pattern indicates that a more adaptive increase in FPN variability under higher cognitive load (i.e. larger ΔTV) is linked to reduced positive/excitatory symptom burden and improved performance.
Our mediation analysis demonstrated that ΔTV within the FPN partially mediates the association between CT and WM performance in schizophrenia. Specifically, higher CTQ scores were associated with reduced ΔTV, which in turn predicted poorer target accuracy. This pattern suggests that trauma diminishes the capacity of FPN to flexibly adjust under cognitive load, thereby constraining executive resource allocation, underscoring the vulnerability of this network to early-life stress. Thus, reduced ΔTV constitutes a partially mediating pathway through which trauma relates to WM deficits.
There are several limitations in our study. First, the cross-sectional design limits inferences about temporal precedence and causality. Second, CT was assessed via retrospective self-report (CTQ), which is susceptible to recall bias and current state influences. Third, because the n-back task included only two load levels (0- and 2-back), we were unable to examine brain-state characteristics under higher cognitive loads. The present cohort primarily consisted of individuals with early-onset, first-episode schizophrenia, which has been described in our previous work. Although this characteristic may, to some extent, reduce the confounding effects of long-term disease progression, the effect of antipsychotic medication use, and untreated illness duration cannot be ruled out in the current study. Longitudinal studies that include medication-naive individuals or patients at early stages of treatment will be necessary to further disentangle the respective contributions of illness-related factors, treatment effects, and network dynamic alterations. In addition, the prevalence of childhood trauma differed between patients with schizophrenia and HCs. Although this difference may have influenced group-level characteristics, our primary analyses examined trauma effects using within-group stratification (CT+ vs. CT−), which helps to mitigate the impact of unequal trauma prevalence across diagnostic groups.
Quantifying load-dependent temporal variability offers a systems-level perspective on WM dysfunction in schizophrenia. Demand-sensitive FPN flexibility, together with its coordination with the DMN and SAL, appears to link CT to cognitive inefficiency. The mediation result situates reduced FPN ΔTV on the pathway from early adversity to poorer performance of WM, consistent with a mechanism in which insufficient flexibility leads to inefficient resource allocation. These findings delineate a new mechanistic pathway, whereby early adversity impairs WM through disruptions of dynamic brain-network function.
To conclude, we underscore the central role of dynamic frontoparietal flexibility in mediating the effects of childhood trauma on WM performance in schizophrenia. By quantifying load-dependent temporal variability (ΔTV), we reveal a systems-level mechanism linking early adversity to cognitive inefficiency. Importantly, reduced ΔTV within the FPN may reflect a failure to adapt under cognitive load, a deficit that can be targeted with neuromodulation or psychotherapeutic means. These results extend prior connectivity and controllability studies by identifying ΔTV as a sensitive marker of executive dysfunction and trauma-related vulnerability. Taken together, our study delineates a novel mechanistic pathway through which early-life stress disrupts cognitive control via impaired dynamic network modulation, offering new targets for intervention and the development of mechanistic markers of environmental risks in schizophrenia.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726103869.
Acknowledgments
The authors would like to thank all patients who served as research participants.
Author contribution
DH, LP, and YL: Conceptualization, methodology, investigation, writing – original draft, and writing – review. DH and YL: Data curation and formal analysis. ZL: Investigation. MS: Writing – review. WT, XL, JY, FW, WH, and JY: Data curation.
Funding statement
This work was supported by a grant from the National Natural Science Foundation of China (82071506 to ZL and 82201692 to YL) and China Scholarship Council (Grant No. 202406370144) for DH to train at the Douglas Research Centre. LP’s research is supported by the Canada First Research Excellence Fund, awarded to the Healthy Brains, Healthy Lives initiative at McGill University (through a New Investigator Supplement to LP) and Monique H. Bourgeois Chair in Developmental Disorders and the Graham Boeckh Foundation. He receives a salary award from the Fonds de recherche du Québec-Santé (FRQS 366934).
Competing interests
LP reports personal fees for serving as chief editor from the Canadian Medical Association Journals, speaker honorarium from Janssen Canada and Otsuka Canada, SPMM Course Limited, UK; book royalties from Oxford University Press; investigator-initiated educational grants from Otsuka Canada outside the submitted work, in the last 5 years. No other competing interests were declared.
Open access
