Load-dependent functional connectivity deficits during visual working memory in first-episode psychosis

Abstract

Introduction:

Aberrant network connectivity is a core deficit in schizophrenia and may underlie many of its associated cognitive deficits. Previous work in first-episode schizophrenia spectrum illness (FESz) suggests preservation of working memory network function during low-load conditions with dysfunction emerging as task complexity increases. This study assessed visual network connectivity and its contribution to load-dependent working memory impairments.

Methods:

Magnetoencephalography was recorded from 35 FESz and 28 matched controls (HC) during a lateralized change detection task. Impaired alpha desynchronization was previously identified within bilateral dorsal occipital (Occ) regions. Here, whole-brain alpha-band connectivity was examined using phase-locking (PLV) and bilateral Occ as connectivity seeds. Load effects on connectivity were assessed across participants, and PLV modulation within networks was compared between groups.

Results:

Occ exhibited significant load modulated connectivity with six regions (FDR-corrected). HC exhibited PLV enhancement with load in all connections. FESz failed to show PLV modulation between right Occ and left inferior frontal gyrus, lateral occipito-temporal sulcus, and anterior intermediate parietal sulcus. Smaller PLVs in all three network connections during both memory load conditions were associated with increased reality distortion in FESz (FDR-corrected.)

Conclusion:

Examination of functional connectivity across the visual working memory network in FESz revealed an inability to enhance communication between perceptual and executive networks in response to increasing cognitive demands. Furthermore, the degree of network communication impairment was associated with positive symptoms. These findings provide insights into the nature of brain dysconnectivity and its contribution to symptoms in early psychosis and identify potential targets for future interventions.

1. Introduction

Beyond the disruptive impact of positive symptoms in psychosis, cognitive and negative symptoms account for significant disease-related functional and occupational deficits (Milev et al., 2005; Vesterager et al., 2012). Importantly, these impairments are present during very early stages of the illness (Fusar-Poli et al., 2012; Zanello et al., 2009) and may manifest in more subtle ways, exhibiting greatest sensitivity to increases in cognitive load or task complexity. Attempts to isolate the biological concomitants of disease-related cognitive impairments have increasingly focused on measures of network connectivity given conceptualizations of schizophrenia as syndrome of brain dysconnectivity (Pettersson-Yeo et al., 2011). Visual working memory (vWM), an executive function that relies upon coordinated activity across distributed brain regions to support various cognitive operations, has consistently exhibited impairments in schizophrenia (Forbes et al., 2009; Lee and Park 2005). Given the greater impact of interventions when delivered during early stages of the illness (Correll et al., 2018), examination of connectivity within the vWM network during first-episode psychosis is vital to future treatment efforts.

An extensive body of research has revealed a distributed neural network that supports vWM function (Chai et al., 2018). Informed by the multicomponent cognitive model of working memory (Baddeley 2000; Baddeley and Hitch 1974), neuroimaging studies have described a central executive component within the prefrontal cortex responsible for active monitoring and manipulation (Curtis and D’Esposito 2003; Lara and Wallis 2015). This modality independent control system interacts with modality specific subcomponents including the visuospatial sketchpad which function as storage buffers for sensory representations (Christophel et al., 2017; Fuster 1997). Neurophysiological studies have also contributed substantially to our understanding of vWM, identifying a restricted storage capacity limited to about 4 objects by examining a load dependent marker of working memory termed the contralateral delay activity (CDA) (Luck and Vogel 1997; Luria et al., 2016). While the precise storage location of objects in vWM remains a debated topic (Christophel et al., 2017), cortical localization of CDA activity utilizing magnetoencephalography (MEG) (Becke et al., 2015; Robitaille et al., 2010) suggests a primary role for dorsal parieto-occipital regions.

Given the various regions supporting vWM across the brain, it is not surprising that coordinated activity between network nodes, commonly operationalized as functional connectivity, is vital to the operation of this neurocognitive system (Gazzaley et al., 2004; Palva et al., 2010). Within the spectral domain, neurophysiological recordings have identified synchronization within the alpha-band as the primary substrate for communication between posterior visuospatial storage centers and frontal executive centers (Figueira et al., 2020; Pavlov and Kotchoubey, 2022; Wianda and Ross 2019). Importantly, communication between regions of this vWM network exhibit modulation by load with increases in connectivity supporting working memory capacity (Newman et al., 2002).

Historically, studies investigating contributors to working memory impairments in schizophrenia have focused on dysfunction in frontal executive networks (Cannon et al., 2005; Manoach 2003), with less emphasis on storage within posterior cortical regions. Interestingly, impairments in both systems appear to be load dependent during chronic disease stages with evidence for a relatively preserved (or increased) response to low-load condition and deficits becoming apparent as task complexity increases (Leonard et al., 2013; Metzak et al., 2012; Potkin et al., 2009). The exaggerated engagement of executive networks to meet relatively simple task demands and inability to enhance activity in the face of more challenging ones has traditionally been conceptualized as a form of cognitive inefficiency in schizophrenia.

A similar load-dependent deficit was also observed at illness onset in our recent study that found impaired dorsal occipital alpha-band desynchronization during vWM in first-episode patients during high, but not low memory load conditions (Coffman et al., 2020). Examinations of functional connectivity in psychosis have also revealed disruptions in communication across the vWM network (Deserno et al., 2012; Meyer-Lindenberg et al., 2001), though the task-dependency of this deficit is less clear, with several studies suggesting the degree of impairment to be independent of cognitive load (Repovš and Barch 2012). A potential source of this discrepancy may be illness chronicity, with more recent studies of first-episode and early onset psychosis revealing impaired modulation of connectivity by memory load (Nielsen et al., 2017; Schmidt et al., 2013). However, differential effects of cognitive load on connectivity across the vWM network remains understudied during early disease stages, limiting our understanding of pathophysiology within this integral executive network and its potential progression with disease course.

To address this gap in the literature, the current study examined load-dependent modulations of alpha-band connectivity in first-episode schizophrenia spectrum individuals (FESz) to identify network-level dysfunctions contributing to vWM impairments and associated clinical outcomes. MEG was recorded during a lateralized vWM task consisting of low and high memory load conditions (Luck and Vogel 1997), and structural MRI was obtained separately for each participant to facilitate localization of cortical activity. Our previous analysis revealed significant task-related alpha modulation within bilateral dorsal occipital regions that was impaired among FESz (Coffman et al., 2020). The present investigation used these regions dedicated to visual memory storage as seed regions to explore patterns of whole-brain alpha-band phase synchrony during working memory maintenance and its disruption in early psychosis. We predicted an impaired ability to enhance connectivity within the vWM network in response to memory load increases in FESz.

2. Methods

2.1. Participants

Thirty-five FESz and 28 HC were included. A portion (28 FESz and 25 HC) were included in an earlier report of source-localized activity during the vWM task (Coffman et al., 2020). Exclusion criteria included: colorblindness, less than nine years of schooling, history of head injury with sequelae, history of a substance use disorder in the past 5 years, and neurological comorbidity. Participants completed the WASI to estimate premorbid intellect and the MATRICS Cognitive Consensus Battery (MCCB) to assess the impact of psychosis on neurocognitive performance (Green et al., 2004). Global Functioning: Role and Social (GF: Role/Social) scales were used to assess occupational and interpersonal functioning, respectively (Cornblatt et al., 2007).

Diagnosis was confirmed using the Structured Clinical Interview for DSM-IV. Twenty-two FESz were diagnosed with schizophrenia, 5 with schizoaffective disorder (4 depressive and 1 bipolar type), and 8 with psychotic disorder not otherwise specified. The Scale for the Assessment of Positive/Negative Symptoms (SAPS/SANS) were used to assess symptom severity. A 4-dimension symptom model was derived based on SAPS/SANS scores (Kotov et al., 2016). In this model, reality distortion and disorganization dimensions further parse positive symptoms while inexpressivity and apathy/asociality further parse negative symptoms. FESz participated within 2 months of first clinical contact and had less than 2 months of lifetime antipsychotic exposure; 11 were unmedicated at testing.

All participants provided voluntary informed consent and procedures were approved by the University of Pittsburgh IRB.

2.2. Visual working memory task

Participants completed the vWM task depicted in Fig. 1. This task, originally developed by Luck and Vogel (1997), utilizes a lateralized attention manipulation to control for visual perceptual processes and isolate evoked neurophysiological responses underlying vWM maintenance including the contralateral delay activity and contralateral alpha suppression. Our previous investigation (Coffman et al., 2020) examined the latter to identify sources of load-dependent alpha power modulation used as connectivity seeds in the current study.

Fig. 1.

An attend left, high load trial of the lateralized change-detection task. Participants were instructed to covertly shift their attention in the direction of the arrow cue and indicate via button-press if there was a change in color between memory and probe array on the attended side. Connectivity was assessed during the 1000 ms retention interval.

Trials began with a fixation cross (500 ms) follow by a centrally presented arrow cue (200 ms). Participants were instructed to covertly shift their attention towards the cued hemifield and, following a brief delay (300 ms), a memory array consisting of 1 (low-load) or 3 (high-load) colored circles was presented in each hemifield (200 ms). A probe array was then presented following a 1000 ms retention phase. Participants were instructed to maintain fixation throughout the trial and indicate via button-press if the color of one of the circles in the attended hemifield changed. Circles subtended 0.65° and could be 1 of 6 possible colors, each with equivalent luminance and contrast. Circles were positioned within a 3° × 7° grid presented 1.5° off fixation. Stimuli were presented in blocks of 75 trials with cue direction, load, trial type (change/no-change) randomized. Only correct trials were included for analysis. Final trial counts during high (HC: 124.5 ± 24.8; FESz: 113.0 ± 24.6) and low (HC: 141.4 ± 26.5; FESz: 131.1 ± 26.9) load conditions did not differ between groups (p’s > 0.05).

2.3. Data acquisition

MEG (Elekta Neuromag) was recorded using 306-channel at a sampling rate of 1000 Hz (online bandpass filter = 0.1–330 Hz). Channels were arranged as triplets (1 magnetometer and 2 gradiometers). Bipolar leads recorded blinks (VEOG), lateral eye movements (HEOG), and cardiac activity (ECG). Four head position indicator coils were placed on the scalp and their position was recorded using a 3D-digitizer (ISOTRAK; Polhemus, Inc., Colchester, VT). Head position was monitored throughout testing and corrected for using Neuromag MaxFilter software (http://imaging.mrc-cbu.cam.ac.uk/meg/Maxfilter_V2.2).

T1-weighted images were obtained for each participant using a Siemens TIM Trio 3 T MRI system with a multi-echo 3D MPRAGE sequence [TR/TE/TI = 2530/1.74, 3.6, 5.46, 7.32/1260 ms, flip angle = 7°, field of view (FOV) = 220 × 220 mm, 1 mm isotropic voxel size, 176 slices, GRAPPA acceleration factor = 2].

2.4. MEG data processing and source localization

Noise originating outside the MEG helmet was removed using the temporal extension of the Signal Space Separation method (TSSS) (Uusitalo and Ilmoniemi, 1997). Data were visually inspected using EEGLAB Toolbox (Delorme and Makeig 2004) and segments or channels (interpolated after pre-processing) containing corrupted data were removed. Adaptive mixture independent component analysis was then performed to isolate and remove (at most) one VEOG, one HEOG, and two ECG components from the data. MEG sensor locations were registered to each participant’s MRI using MRIlab (Elekta-Neuromag Oy, Helsinki, Finland) and Freesurfer (http://www.surfer.nmr.mgh.harvard.edu) was used to tessellate the cortical surface into an icosahedron mesh.

Subsequent processing was conducted using Brainstorm (Tadel et al., 2011). A low-pass filter (100 Hz; 24dB/oct) was applied. Continuous data were epoched into 2500 ms segments including 800 ms prior to the memory array and 1700 ms following it. Segments were baseline corrected using the 300 ms pre-cue arrow (−800 to −500 pre-memory array) and those with data exceeding 5 pT were rejected. Retention phase saccades were identified using a step function with a 200 ms moving window applied to HEOG (Luck et al., 2014).

Sources of brain activity were localized for each trial using the minimum-norm estimate (Gramfort et al., 2014). The forward solution was modeled as a series of overlapping spheres (one per sensor). The noise covariance matrix was calculated from the baseline period. A fully constrained linear inverse operator with depth weighting applied was used to isolate sources of activity.

2.5. Functional connectivity

Phase locking (PLV) (Lachaux et al., 1999; Mormann et al., 2000) was used to assess functional connectivity. PLV is a common method to examine phase synchronization between two signals based on the assumption that two functionally connected signals will exhibit consistency or ‘locking’ in the timing of their activity. The narrower the distribution of phase differences between two signals (i.e., reduced angular variance), the more robust the connectivity between them is presumed to be. Importantly, measurement of PLV is not confounded by amplitude of the signals.

The present investigation focused on phase synchronization within the alpha-band given its primary role in long-distance communication within the visual system and during working memory (Figueira et al., 2020; Lobier et al., 2018). Connectivity was assessed across the whole brain during the retention phase (200–1200 ms post-memory array). Selection of bilateral dorsal occipital seed regions (Occ) was informed by our previous analysis that revealed significant task-related alpha modulation within these regions (Coffman et al., 2020). The cortical surface of each participant was parcellated using the Destrieux atlas (Destrieux et al., 2010) with our seed regions defined by the middle occipital sulcus, sulcus lunatus, and superior occipital sulcus and gyrus bilaterally.

To compute PLV, source resolved cortical activity was band-pass filtered (8–12 Hz) and a Hilbert transform was used to obtain instantaneous phase data. The PLV between two regions i and j was calculated:

$$\text{PLVij}=\frac{1}{T}\left|\sum _{t=1}^T{e}^{-i(\phi i(t)-\phi j(t))}\right|$$

In the above equation, T is the number of data points in the time series and φ(t) represents the phase of each signal at timepoint t. A plot depicting dynamic changes in alpha power observed over the trial period is depicted in Supplemental Fig. 1.

2.6. Data analysis

Demographic variables (Table 1) were compared between groups. Working memory capacity (K) was calculated as proposed by Cowan (2001):

$$\text{K}=\text{N}\left({\text{h}}^{\wedge }-{\text{f}}^{\wedge }\right)$$

where ĥ represents hit rate, f^ false alarm rate, and N the number of objects. Task performance (Table 2) was subjected to a 2 (group: HC/FE) × 2 (load: high/low) repeated-measures ANOVA. As an initial step in our connectivity analysis, permutation tests were conducted across groups with 1000 permutations comparing PLVs between load conditions for each seed-to-region connection to identify the load-dependent connectivity network engaged by our task (FDR-adjusted; q < 0.05). These regions were then subjected to a group (HC/FE) × load (high/low) × region repeated-measures ANOVA to examine differential group effects on PLV modulation across regions. Bivariate relationships between PLVs from connections exhibiting impaired load modulation among FESz and measures of task performance and clinical assessments were investigated using Spearman’s rank-order correlation (FDR-adjusted; q < .05). Antipsychotic medication effects were examined by comparing PLVs between patients who were and were not prescribed medications at testing.

Table 1.

Group demographic clinical assessment data (Mean ± SD).

	HC (n = 28)	FE (n = 35)	t/χ2	p
Female/Male	10/18	9/26	0.74	.39
Education (years)	13.9 ± 3.0	12.9 ± 2.2	1.46	.15
Age	21.6 ± 4.3	23.3 ± 4.9	1.39	.17
WASIa	109.0 ± 8.4	108.8 ± 14.3	0.06	.96
MCCB-Total	49.5 ± 6.7	40.9 ± 14.0	2.95	.004
SESb	34.0 ± 14.7	30.7 ± 13.2	0.92	.36
PSESc	48.7 ± 12.5	43.7 ± 13.4	1.48	.14
GF: Roled	9.05 ± 0.2	6.01 ± 2.4	6.72	<.001
GF: Sociale	9.02 ± 0.2	5.56 ± 1.9	9.61	<.001
SANSf/SAPSg:
Inexpressivity	-	11.0 ± 7.0
Asociality/Apathy	-	11.7 ± 6.2
Reality Distortion	-	10.3 ± 7.6
Disorganization	-	3.2 ± 4.1
Medicated/Unmedicated	-	24/11
Medication (CPZh mg/day)	-	202.6 ± 164.4

^aWechsler Abbreviated Scale of Intelligence.

^bSocioeconomic status.

^cParental socioeconomic status.

^dGlobal Functioning: Role scale.

^eGlobal Functioning: Social scale.

^fScale for the Assessment of Negative Symptoms.

^gScale for the Assessment of Positive Symptoms.

^hChlorpromazine equivalent dose.

Table 2.

Task performance and connectivity measures by working memory load.

	HC (n = 28)		FE (n = 35)
	Low Load	High Load	Low Load	High Load
Performance
Working Memory Capacity (K)	0.95 ± .05*and	2.22 ± .46*and	0.86 ± .16*and	1.78 ± .63*and
Hit Rate (%)	92.3 ± .18.8*	77.2 ± 122*and	89.8 ± .9.4*	67.8 ± .15.4*and
False Alarm Rate (%)	8.8 ± .13.0and	9.9 ± .8.3and	18.3 ± .18.8and	18.0 ± .14.6and
RT (ms)	739.2 ± 139.4*and	840.1 ± 136.6*	870.1 ± 169.5*and	918.8 ± 178.8*
Connectivity (PLVa)
Across region	.085 ± .02*	.093 ± .03*	.089 ± .02*	.092 ± .02*
Right Occb – Left IFGc	.095 ± .02*	.104 ± .03*	.100 ± .02	.102 ± .02
Right Occ – Right IFG	.089 ± .02*	.096 ± .03*	.095 ± .02*	.099 ± .02*
Right Occ – Left AIPSd	.083 ± .01*	.093 ± .02*	.087 ± .02	.087 ± .02
Right Occ – Left OTSe	.079 ± .02*	.087 ± .02*	.083 ± .02	.084 ± .02
Right Occ – Right OTS	.073 ± .01*	.078 ± .02*	.075 ± .01*	.080 ± .01*
Left Occ – Left MFGf	.089 ± .02*	.099 ± .04*	.092 ± .03*	.098 ± .03*

^*Significant difference between task condition (p < .05).

^AndSignificant differences between groups (p < .05).

^aPhase Locking Value.

^bOccipital seed region.

^cInferior frontal gyrus.

^dAnterior intermediate parietal sulcus.

^eLateral occipital temporal sulcus.

^fmiddle frontal gyrus.

3. Results

3.1. Task performance

Main effect of group on working memory capacity (K) was observed (F_1,60 = 10.23, p = .002, η² = 0.15) with FESz exhibiting reduced K relative to HC. As predicted by the method for calculating K, a main effect of task load was also present (F_1,60 = 289.22, p < .001, η² = 0.83) with a larger K on high relative to low-load trials. In addition, there was an interaction between group and task (F_1,60 = 7.28, p = .009, η² = 0.11) with a larger group difference observed during high (t₆₀ = 3.03, p = .002, d = 0.78) compared to low (t₆₀ = 2.89, p = .005, d = 0.74) load condition. Main effects of group (F_1,60 = 6.78, p = .01, η² = 0.10) and task (F_1,60 = 107.16, p < .001, η² = 0.64) on RT were also present with slower RTs among FESz and during the high load condition. An interaction between group and task (F_1,60 = 13.10, p = .001, η² = 0.18) on RTs driven by slower responses among FESz during the low (t₆₀ = 3.25, p = .002, d = 0.83) compared to high (p = .06) load condition relative to HC. Performance measures broken down by group and task condition are presented in Table 2.

3.2. Functional connectivity

Initial analyses to identify the load-dependent vWM across all participants revealed significant PLV modulation between occipital seed regions and 6 regions (Fig. 2) including right hemisphere Occ with the bilateral inferior frontal gyri (IFG), left anterior intermediate parietal sulcus (AIPS), and bilateral lateral occipito-temporal sulci (OTS) as well as the left hemisphere Occ with left middle frontal gyrus (MFG) (FDR-adjusted; q’s < 0.05). PLVs for the 6 connectivity pairs comprising our load-dependent vWM network broken down by group and task condition are presented in Table 2.

Fig. 2.

The visual working memory network including left and right occipital seeds (A) and the regions with which they exhibited load dependent enhancement in connectivity (B). Regions in green exhibited load-modulated connectivity with the left hemisphere seed and those in red with the right hemisphere seed. Group differences in load modulation of connectivity were observed between the right occipital seed and left inferior frontal gyrus (C), left anterior intermediate parietal sulcus (D), and left occipital temporal sulcus (E). *p < .05.

ANOVA to examine effects of group and region on load-dependent modulation of connectivity found no main effect of group across region or condition (p = .79). There was, however, a differential effect of memory load on group PLVs (F_1,61 = 4.89, p = .03, η² = 0.07) driven by more robust load-dependent PLV enhancement among HC (t₂₇ = 3.93, p = .001, d = 0.74) relative to FESz (t₃₄ = 2.79, p = .009, d = 0.47). There was also a three-way interaction involving group, task, and region (F_5,305 = 2.49, p = .03, η² = 0.04). While HC exhibited significant enhancement with load across all connectivity pairs (Right Occ-Left IFG: t₂₇ = 4.78, p < .001, d = 0.90; Right Occ-Right IFG: t₂₇ = 2.81, p = .009, d = 0.53; Right Occ-Left AIPS: t₂₇ = 5.08, p < .001, d = 0.96; Right Occ-Left OTS: t₂₇ = 4.06, p < .001, d = 0.77; Right Occ-Right OTS: t₂₇ = 2.27, p = .031, d = 0.43; Left Occ-Left MFG: t₂₇ = 2.20, p = .04, d = 0.42), FESz failed to exhibit such modulation between the right hemisphere Occ and left IFG (p = .22), AIPS (p = .83), and OTS (p = .29) (Fig. 2). FESz did exhibit significant modulation at the remaining connection pairs (Right Occ-Right IFG: t₃₄ = 2.39, p = .02, d = 0.40; Right Occ-Right OTS: t₃₄ = 3.41, p = .002, d = 0.58; Left Occ-Left MFG: t₃₄ = 2.19, p = .04, d = 0.37).

3.3. Correlations with performance and clinical measures

Correlations between PLV values and memory capacity (K) and clinical assessment measures (MCCB, GF: Role/Social, and symptom dimensional scores) for connectivity pairs exhibiting deficits in load-dependent modulation in FESz were performed. Larger PLVs during both memory load conditions for all 3 connectivity pairs were associated with reduced reality distortion scores in FESz (High Load Left IFG: ρ = −0.49, q = 0.04; Low Load Left IFG: ρ = −0.48, q = 0.04; High Load Left AIPS: ρ = −0.59, q < 0.01; Low Load Left AIPS: ρ = −0.50, q = 0.03; High Load Left OTS: ρ = −0.61, q < 0.01; Low Load Left OTS: ρ = −0.46, q = 0.048) (Fig. 3). Remaining correlations did not achieve significance following FDR correction but are presented in Supplemental Table 1.

Fig. 3.

Correlations between reality distortion dimension scores derived from the Scale for the Assessment of Positive Symptoms (SAPS) and connectivity between the right occipital seed region and the left inferior frontal gyrus (IFG), left anterior intermediate parietal sulcus (AIPS), and occipito-temporal sulcus (OTS) during high and low memory load conditions.

3.4. Medication effects

There was no effect of medication status or interactions involving it on functional connectivity (p’s > 0.2). Medication dose calculated using chlorpromazine equivalents did not correlate with any connectivity measure among those on an antipsychotic at time of testing (p’s > 0.2).

4. Discussion

Brain dysconnectivity is increasingly recognized as a pathophysiologic hallmark of schizophrenia responsible for various disease-related outcomes. The present study examined functional connectivity in the schizophrenia spectrum at first psychotic episode to identify the integrated neural network contributing to vWM impairment and its associated disease-related morbidity during early stages of the illness. Not only did FESz exhibit a load-dependent deficit in vWM capacity relative to HC, they were also unable to enhance connectivity within a portion of the load-dependent vWM network in response to increasing task demands. Furthermore, FESz exhibiting reduced PLV between within this network experienced greater symptoms of reality distortion. These findings enhance our understanding of brain dysconnectivity and its contribution to symptoms in early psychosis and provide potential targets for future interventions.

Traditionally, impaired network connectivity has been viewed as an intrinsic deficit in schizophrenia observed across cognitive states (Meyer-Lindenberg et al., 2001; Repovš and Barch 2012). However, the preserved PLVs among FESz within each task condition in the present study suggests a specific impairment in the modulation of vWM network communication to meet task demands rather than a broad disruption of brain connectivity. In fact, while a statistical difference was not observed between groups, FESz exhibited larger mean PLVs during the low load condition and smaller mean PLVs during the high load condition relative to HC. This pattern of results is consistent with models of cognitive inefficiency described by neuroimaging studies of working memory in schizophrenia (Metzak 2012; Potkin et al., 2009) as well as more recent neurophysiological examinations of selective attention in first-episode psychosis (Sklar et al. 2020, 2021).

One explanation for inconsistency with past studies of functional connectivity in schizophrenia might relate to the illness chronicity of the sample. Recent studies in FESz have found similar load-dependent disruptions rather than across-the-board deficits (Manivannan et al., 2019; Nielsen et al., 2017; Schmidt et al., 2013) suggesting a progressive nature of dysconnectivity in schizophrenia,. Further emphasizing the importance of treatment during early disease course, some of these studies observed an amelioration of deficits in network connectivity with antipsychotic treatment and shorter durations of untreated psychosis (Manivannan et al., 2019; Nielsen et al., 2017). Although no effect of medication was observed in the present study, this may be related to the relatively short window between first clinical contact and testing.

Beyond a general disruption across the vWM network, impaired PLV enhancement among FESz appeared to be restricted to certain regional connections. Our selection of sensory-perceptual seed regions yielded load dependent interactions with additional subcomponents of the canonical working memory network (Baddeley 2000; Baddeley and Hitch 1974) including the MFG, IFG, and AIPS. While less frequently discussed in the context of WM, the inferior occipito-temporal region identified by our analysis (OTS) has consistently exhibited enhanced retention phase activation (Miller et al., 1996; Ranganath 2006) and likely supports the perceptual processing associated with the more dorsal occipital seed region. Interestingly, despite a historic emphasis on executive disruptions associated with impaired MFG activity in schizophrenia (Minzenberg et al., 2009; Niendam et al., 2012; Potkin et al., 2009), our data suggests load-depended disruptions restricted to connectivity between visual perceptual regions and the IFG, AIPS, and OTS in FESz. This finding, however, may be a result of our vWM task which provided less of a challenge to central executive systems during a retention phase that was free of distractor interference or memory updating and manipulation requirements.

As with impairments in complex cognitive functions such as working memory, positive symptoms of schizophrenia are thought to be mediated by aberrant communication across a distributed cortical network (Rotarska-Jagiela et al., 2010; Supekar et al., 2019). Reality distortion symptoms have been associated with disruptions within the saliency network and its communications with other cognitive control systems (Palanivappan et al., 2011; Taylor et al., 2007). The observed correlations between reality distortion and connectivity between regions located outside this traditional network may be related to the selection of sensory-perceptual regions as a seed for connectivity analyses. Subsequent examination of item scores comprising the reality distortion dimension (see Supplemental Table 2) suggests the observed correlations were driven by reported hallucinations rather than delusions, a finding consistent with previously described hallucination networks consisting of sensory cortices (Allen et al., 2008). Importantly, the direction of these correlations suggests those individuals experiencing more severe symptoms of reality distortion are the least likely to exhibit robust communication within the vWM network, particularly during more challenging conditions. An alternative explanation for the observed correlations is that individuals experiencing more severe reality distortion were more distracted during the task. However, the lack of correlation between reality distortion and task performance (p’s > 0.1) suggests that individuals with a higher symptom burden were not any less engaged.

4.1. Limitations

The lack of significant correlations between alpha-band PLVs and task performance or MCCB scores is a limitation of the present study. However, the study’s focus on cognitive load-dependent deficits in FESz biased the identification of a vWM network toward those connections sensitive to modulation by memory load as opposed to connections exhibiting the largest PLVs, per se. When correlations between PLV and K difference scores in FESz are examined, a relationship between enhanced left hemisphere MFG connectivity and increased K is observed (ρ = .35, p = .04). Furthermore, follow-up correlations with MCCB domain scores reveals a relationship between left IFG enhancement and working memory subscale scores (ρ = .47, p = .004). While these correlations do not survive correction for multiple comparisons, they are consistent with a priori expectations regarding relationships between vWM performance and connectivity within a neural network subserving it. It is possible that a more demanding high load condition (e.g., 4 objects to retain in memory rather than 3) might yield more numerous and robust associations.

Although the vWM task (Luck and Vogel 1997) provides well-validated measures of working memory capacity across load conditions, its design limited our ability to isolate dynamic network interactions tied to distinct memory processes. Given the brief stimulus presentation intervals, it would be difficult to reliably demarcate encoding or retrieval periods across participants. Furthermore, assessing alpha-band connectivity, an established carrier frequency for long-range communication in vWM networks, requires longer time windows. Fortunately, the task design does allow us to assess the load-dependent nature of PLV enhancement observed during the retention interval. There was no effect of task load or interactions involving it on PLVs within our network prior to presentation of the memory array (p’s > 0.3), supporting our conclusion that the reported effects observed during the retention period were indeed load-dependent.

Finally, the lack of established white matter tracts between the identified contralateral connectivity pairs suggests the absence of a direct, structural connectome underlying this network. This does not, however, preclude the presence of indirect structural connections between regions which have been shown to significantly contribute to observed functional relationships (Røge et al., 2017; Stam et al., 2016).

5. Conclusions

In contrast to previous reports of generalized disruptions in psychosis, findings from the present study emphasizes the task dependent nature of disease related impairments during early illness stages. Despite the overall preservation of network connectivity in FESz, they were unable to modulate this system level response to increasing task complexity which may have profound effects on daily functioning in real-world situations. Longitudinal investigations would help reveal progressive deterioration of connectivity within this network that might lead to more profound deficits in task performance or global functioning. Furthermore, thorough examinations of structural connectivity would help identify additional anatomic structures supporting the indirect connectivity observed and assist in the development of targeted interventional therapeutic approaches.