EEG microstate alterations in psychotic depression: prolonged microstate C duration and associations with psychotic symptoms

Introduction

Depression is a common and debilitating mental disorder characterized by persistent low mood, anhedonia, and cognitive and vegetative symptoms. It is a leading cause of disability worldwide, affecting an estimated 280 million people across all age groups. Major depressive disorder (MDD) represents the clinical diagnosis when these symptoms meet specific duration and severity criteria, and it exhibits substantial heterogeneity in clinical presentation, treatment response, and underlying neurobiology. Major Depressive Disorder with psychotic features (MDD-Psy) represents a severe and distinct clinical subtype characterized by the co-occurrence of affective and psychotic symptomatology. Epidemiologic studies have reported that 9.2 to 30% of patients with MDD exhibit psychotic features, and this proportion increases with the severity of the illness, as indicated by frequent relapses, suicidality, and treatment resistance [1, 2]. This form of depression is associated with significant morbidity and presents unique treatment challenges [3]. Notably, antipsychotic medications, a cornerstone of its pharmacotherapy, have been shown to induce measurable alterations in brain structure during treatment, underscoring the profound neurobiological impact of both the illness and its management [4]. Furthermore, clinical trials targeting this population face specific hurdles, such as higher attrition rates linked to illness severity, highlighting the need for a deeper understanding of its pathophysiology to improve patient care and study design [5]. The clinical stakes are substantial, as MDD-Psy is also implicated in a heightened risk for suicidal behavior within psychotic disorders [6].

Over the past two decades, neuroimaging research has increasingly focused on quantifying alterations in brain dynamics associated with depressive states using objective numerical measures. Functional magnetic resonance imaging (fMRI) studies have consistently reported disrupted functional connectivity within and between large-scale brain networks in MDD, including the default mode network (DMN), the salience network (SN), and the central executive network (CEN) [7–9]. More recently, resting-state fMRI has been used to characterize the temporal dynamics of these networks, revealing altered connectivity variability and network switching in depression [10, 11]. However, the limited temporal resolution of fMRI (on the order of seconds) cannot capture the millisecond-scale neural dynamics that underlie rapid cognitive and affective processes. Electroencephalography (EEG) and magnetoencephalography (MEG) offer complementary temporal precision, and studies employing these modalities have identified abnormalities in oscillatory power, functional connectivity, and signal complexity in MDD [12, 13]. Despite these advances, the distinct pattern of large-scale network dynamics in MDD-Psy—as distinct from non-psychotic MDD—remains poorly understood, largely because most studies have either excluded psychotic depression or pooled it with non-psychotic cases. In parallel, the analysis of resting-state electroencephalogram (EEG) microstates has emerged as a powerful tool for probing the rapid dynamics of large-scale brain networks. This method characterizes the millisecond-order stability and transitions between canonical neuroelectric patterns (classes A, B, C, and D), which reflect fundamental operations of sensory, salience, and attention-related networks [14, 15]. Compared to traditional EEG power spectral analysis or connectivity measures, microstate analysis captures the sequential activation and switching dynamics of whole-brain networks at a subsecond timescale, providing a unique window into the temporal organization of spontaneous brain activity. This approach has proven sensitive to alterations across a range of neuropsychiatric conditions, including schizophrenia, depression, and bipolar disorder. Research has demonstrated that EEG microstates serve as sensitive candidate markers across neuropsychiatric conditions. In schizophrenia spectrum disorders, well-replicated alterations include changes in the duration and expression of microstates C and D, with microstate D specifically proposed as a psychosis-specific correlate observable even in clinical high-risk and first-episode populations [16–19]. Similarly, in non-psychotic Major Depressive Disorder (MDD), studies—particularly in adolescents and young adults—have documented significant microstate abnormalities. Findings include altered temporal parameters of microstates C and D linked to depression severity, anhedonia, and attention deficits [20–22], as well as distinct microstate patterns associated with clinical sub-phenotypes such as suicidal ideation and non-suicidal self-injury (NSSI) [23, 24]. The sensitivity of microstates extends to developmental and environmental factors, with specific parameters (e.g., occurrence of microstate A, B→A transitions) being associated with adverse childhood experiences like emotional abuse in depressed adolescents [25]. The utility of this approach is further validated by its predictive capacity for treatment outcomes in conditions like epilepsy and its application in disorders of consciousness, establishing EEG microstates as a robust transdiagnostic metric of brain network function [26–29].

It is instructive to contrast EEG microstate analysis with other quantitative neuroimaging approaches, particularly graph-theoretic connectivity methods. Graph-theoretic analysis, often applied to fMRI or source-reconstructed EEG/MEG data, quantifies pairwise synchronization or statistical dependence between brain regions, representing the brain as a network of nodes (regions) and edges (connectivity strength). This approach has revealed disrupted topological properties in depression, such as reduced global efficiency and altered modular organization [30, 31]. However, graph-theoretic methods face several challenges: they are sensitive to the choice of connectivity metric (e.g., Pearson correlation, phase locking value), thresholding procedures, and brain parcellation schemes; moreover, when applied to EEG, the temporal resolution is often reduced by sliding windows of several hundred milliseconds to seconds, blurring rapid dynamics [32]. In contrast, microstate analysis operates directly on the scalp electric field at each millisecond, is entirely data-driven (no parcellation or thresholding required), and captures the sequential activation and switching of whole-brain networks at their native temporal scale. Microstates provide a “global” rather than “pairwise” representation of brain activity, and parameters such as transition probabilities directly quantify the temporal syntax of network dynamics—a feature not easily derived from connectivity matrices. The primary limitations of microstate analysis include lower spatial resolution (scalp topography rather than source space) and the assumption that the electric field remains quasi-stable for tens of milliseconds, which may oversimplify rapid transitions. For the present study, which focuses on rapid network dynamics and altered switching patterns in MDD-Psy, microstate analysis offers a well-suited, hypothesis-driven approach that does not require arbitrary parameter selection and directly tests our hypothesis of reconfigured temporal transitions between large-scale networks. Moreover, large-scale transdiagnostic study protocols explicitly position EEG microstate analysis as a core method for investigating neurophysiological alterations across mood and psychosis spectra [33].

It is important to position microstate analysis within the broader landscape of quantitative EEG candidate markers. While spectral power analysis quantifies oscillatory activity within specific frequency bands (e.g., theta, alpha, beta) and graph-theoretic connectivity measures quantify pairwise synchronization between regions, microstate analysis captures the sequential organization and temporal syntax of whole-brain network states at the millisecond scale [14, 15]. These approaches provide orthogonal information: spectral measures capture “how” the brain oscillates, connectivity measures capture “where” regions synchronize, and microstates capture “when” and “in what order” large-scale networks activate and switch. For the present study, which focuses on rapid network switching dynamics in psychotic depression, microstate analysis offers a hypothesis-driven approach that directly tests our prediction of reconfigured temporal transitions between large-scale networks. However, a critical gap persists at the intersection of these two research fronts. While EEG microstate abnormalities are established in both schizophrenia (a primary psychotic disorder) and non-psychotic MDD, their distinct pattern in MDD-Psy remains “poorly characterized,” as noted in prior work. It is unknown whether the microstate signature of MDD-Psy resembles that of schizophrenia, aligns with non-psychotic MDD, or constitutes a unique profile that could mechanistically explain the co-occurrence of both affective and psychotic symptoms. This knowledge gap limits our ability to identify objective neurophysiological candidate markers for this severe subtype, which could otherwise aid in diagnosis, prognosis, and the development of targeted interventions.

To address this gap, the present study aimed to characterize resting-state EEG microstate dynamics specifically in MDD-Psy, in direct comparison with MDD without psychotic features (MDD-NonPsy). Based on the central role of the salience network (linked to microstate C) in psychosis [16, 17, 19] and its interaction with other large-scale networks in depression [21, 22], we hypothesized that MDD-Psy would be associated with a distinct pattern of aberrant microstate dynamics, particularly involving prolonged salience network activity and reconfigured temporal transitions between networks. Furthermore, we hypothesized that these microstate alterations would correlate with the core dimensions of psychotic symptom severity. By testing these hypotheses, this study seeks to provide novel insights into the large-scale brain network dysfunctions that underpin psychotic depression.

Methods

Results

Discussion

The present study characterized aberrant resting-state EEG microstate dynamics in MDD-Psy, revealing a potentially distinct pattern of large-scale brain network dynamics relative to non-psychotic depression. The principal finding was a significantly prolonged duration of microstate C, a topographic pattern that prior multimodal work has associated with salience network dynamics, in MDD-Psy patients (91.99 ms) relative to MDD-NonPsy controls (85.56 ms). Additionally, psychotic depression exhibited a reconfigured temporal architecture characterized by altered transition probabilities between canonical microstates, including increased B→A transitions and decreased B→C and C→A transitions. Critically, these neurophysiological alterations demonstrated clinically meaningful relationships with psychotic symptom dimensions: microstate C duration correlated positively with thought disturbance severity, while the diminished B→C transition probability showed robust negative correlations with both thought disturbance and hostility. These findings provide empirical support for dysregulated salience network dynamics as a core neurophysiological feature of psychotic depression, linking aberrant brain temporal organization directly to the expression of psychotic symptoms.

While the aberrant salience hypothesis provides a valuable transdiagnostic framework, it risks obscuring the phenomenological specificity of psychotic depression if applied generically. A critical cognitive dimension that distinguishes affective psychosis from primary psychotic disorders is the role of maladaptive emotion regulation, particularly rumination, as a pathogenic mechanism. In MDD-Psy, delusional ideation and thought disturbance often emerge not de novo but through the rigid entrenchment of negative self-referential processing, wherein ruminative loops become so perseverative that they distort reality testing [3, 45]. Our microstate findings can be reinterpreted within this affective neuroscience framework, offering a mechanistically richer account than schizophrenia-derived models alone.

The prolongation of microstate C duration in MDD-Psy may reflect the salience network’s pathological engagement with internally generated ruminative content. In this view, the salience network does not merely misattribute significance to external neutral stimuli (as classically proposed for schizophrenia); rather, it becomes chronically occupied by the self-salience of depressive rumination, assigning aberrant significance to internally generated negative thoughts and autobiographical memories. This internal hyper-focus would manifest neurophysiologically as prolonged stability of the salience network configuration (microstate C), effectively “locking” processing resources into a closed self-referential loop. This interpretation converges with recent electrophysiological evidence demonstrating that individuals with high ruminative tendencies exhibit decreased frontal EEG complexity (frequency-specific entropy reduction), indicating a restricted, perseverative information-processing mode [46]. The reduced neural complexity observed in ruminators parallels the prolonged temporal stability of microstate C in our MDD-Psy patients—both suggesting a loss of flexible, dynamic network exploration in favor of rigid, self-focused states.

The diminished B→C transition probability acquires a specific meaning within this ruminative framework: it reflects the suppression of external visual information processing by internally directed attention. When the salience network is monopolized by ruminative self-focus, perceptual inputs from the visual network (microstate B) fail to gain access to salience evaluation and reality monitoring. This creates a functional decoupling in which the external world is progressively excluded from conscious processing, providing a neurophysiological substrate for the detachment from reality and delusional misinterpretation characteristic of psychotic depression. This mechanism aligns with findings that opposing emotion regulation strategies—rumination versus cognitive distraction—exhibit markedly distinct spectral coherence architectures, with ruminative states reorganizing inter-regional communication to prioritize internal over external information streams [47].

The neuroscientific interpretation of diminished B→C transitions warrants careful dissection of whether this represents an active pathological defense mechanism or a structural breakdown in network coordination secondary to chronic rumination. We propose that both processes operate sequentially and may be distinguishable along a temporal and severity gradient.

Initially, the B→C decoupling likely reflects an active, albeit maladaptive, defensive process. In the context of severe depressive affect, the salience network may deliberately withdraw from processing external visual input to protect and sustain internally directed ruminative focus. This is consistent with the phenomenology of psychotic depression, wherein patients describe an overwhelming, involuntary preoccupation with negative self-referential thoughts that actively excludes external reality. From a regulatory perspective, this represents a failed emotion regulation strategy: rather than flexibly deploying attention to modulate affect, the system becomes rigidly entrenched in a self-perpetuating loop of internal processing. The negative correlation between B→C transition probability and hostility severity (ρ = −0.357) supports this interpretation—higher hostility, often reflecting externalized frustration with reality intrusion, is associated with more complete visual-salience decoupling, suggesting an active defensive exclusion of the external world.

However, with chronicity and severity, this functional withdrawal may progress to structural breakdown in salience network gating capacity. The robust correlation between B→C transitions and thought disturbance (ρ = −0.393)—the core psychotic symptom dimension encompassing delusional ideation and conceptual disorganization—suggests that prolonged decoupling is not merely strategic but reflects a degraded capacity to re-engage external input even when adaptive. This aligns with Aydın & Akın’s (2022) finding of decreased frontal EEG complexity in chronic ruminators: reduced entropy indicates a loss of dynamic range in information processing, paralleling the rigid, non-flexible microstate dynamics we observe. The salience network, chronically deprived of external visual input, may undergo use-dependent plasticity changes that progressively impair its gating function—a mechanism akin to learned non-use in motor systems, but applied to perceptual-salience integration.

The distinction between defense mechanism and structural breakdown is further illuminated by the asymmetric transition reconfiguration. The reciprocal increase in C→B transitions (while B→C decreases) suggests that information flow is not uniformly blocked but pathologically rerouted: the salience network actively projects internally generated signals back to visual processing, potentially generating the vivid, self-referential visual imagery and reality-distorting percepts reported in psychotic depression. This unidirectional rerouting—outflow without inflow—mirrors the assortativity reduction documented in ruminative brain networks by Aydın (2023), wherein diminished network resilience reflects impaired reciprocal coordination between systems. The C→A decrease (salience-to-auditory disengagement), in contrast, may represent a more primary structural deficit, as auditory processing lacks the compensatory back-projection seen in the visual domain.

Synthesizing these perspectives, we interpret the B→C transition decrease as a dynamic process with two phases: (i) an early, functional phase of active defensive withdrawal driven by overwhelming ruminative self-focus, and (ii) a later, structural phase of degraded salience gating capacity resulting from chronic disuse and network plasticity maladaptation. In our cross-sectional sample, patients likely represent varying points along this continuum, with the severity-dependent correlations (stronger for thought disturbance than hostility) suggesting that more psychotic individuals have progressed further toward structural breakdown. Longitudinal studies tracking microstate dynamics from acute depressive episodes through chronic psychotic depression would be essential to validate this proposed temporal trajectory and to determine whether early intervention targeting ruminative processes could prevent the progression from functional defense to structural network damage.

Furthermore, the asymmetric reconfiguration of transitions—increased C→B but decreased C→A—resonates with models of global network dysregulation during persistent negative self-referential thought. Rather than reflecting a uniform disconnection, this pattern suggests a pathological rerouting in which the salience network aberrantly projects signals back to visual processing (C→B increase) while disengaging from auditory processing (C→A decrease). This asymmetry may mirror the inter-hemispheric and inter-network coordination deficits documented in ruminative profiles. Recent graph-theoretic analyses of resting-state EEG have shown that frequent rumination is associated with reduced assortativity and disrupted segregation-integration balance—topological alterations indicating diminished network resilience and impaired flexible coordination between large-scale systems [48]. In our data, the altered transition syntax (particularly the reciprocal B→C/C→B dysregulation) may represent the temporal dynamic counterpart to these topological changes: a rigid, perseverative pattern of network switching that prevents the flexible reallocation of processing resources required for adaptive emotion regulation.

Collectively, this reframing positions the microstate alterations in MDD-Psy within a coherent affective-cognitive pathophysiology: psychotic symptoms in depression emerge when maladaptive rumination, sustained by prolonged salience network activity and reinforced by decoupled visual-salience transitions, becomes so entrenched that it overrides reality testing. This perspective shifts the narrative away from a generic “schizophrenia-like psychosis” model and firmly anchors our findings in the specific domain of affective psychosis, where the co-occurrence of mood disturbance and reality distortion is mechanistically linked to dysregulated self-referential processing rather than primary sensory-limbic disconnection.

The current findings both converge with and diverge from prior microstate research across psychotic disorders, highlighting the unique neurophysiological signature of psychotic depression. Our observation of prolonged microstate C duration in MDD-Psy aligns with studies reporting increased microstate C contribution in clinical high-risk and first-episode psychosis [17], and is consistent with the hypothesis that microstate C alterations reflect aberrant salience network functioning [15]. However, this pattern contrasts with findings in treated schizophrenia, where microstate C parameters may normalize or show negative correlations with negative symptoms [39], suggesting that medication status and clinical phase critically influence microstate dynamics. The altered transition probabilities we observed, particularly the reduced B→C and C→A transitions, support recent transdiagnostic evidence of impaired network switching between visual, salience, and auditory networks across psychotic and mood disorders [40]. This finding extends earlier work demonstrating microstate syntax disturbances in first-episode schizophrenia [16], suggesting that transition abnormalities represent a shared feature of psychosis spectrum conditions. Notably, our direct comparison of MDD-Psy and MDD-NonPsy reveals a specificity not captured in previous studies, as microstate D—frequently reported as decreased in schizophrenia and high-risk states [17, 41]—showed only nominal differences between our depression groups, reinforcing that the microstate C prolongation rather than D reduction distinguishes psychotic from non-psychotic depression. This distinction is further supported by resting-state fMRI data from psychotic depression demonstrating decreased frontoparietal network connectivity [42], which may represent a complementary imaging correlate of the altered temporal dynamics observed here. The correlation between the B → C transition probability and psychotic symptom severity (ρ = -0.393 for thought disturbance) represents a medium effect size and was the most robust finding in our exploratory analyses, surviving both FDR and Bonferroni correction. An important methodological consideration is whether the observed symptom-microstate correlations in the combined sample merely reflect the strong diagnostic group differences in both variables (i.e., MDD-Psy patients have higher BPRS scores and altered microstates compared to MDD-NonPsy patients). Our supplementary analyses directly address this concern. First, when diagnostic group and other covariates were controlled in partial correlation models, the B → C transition remained significantly associated with thought disturbance and hostility, demonstrating that this association is not purely an artifact of group dichotomy. Second, within-group correlations in the MDD-Psy sample alone replicated the direction and approximate magnitude of the B → C and C → B associations, confirming that these relationships reflect genuine symptom severity variation within the psychotic subtype. However, the microstate C duration association with thought disturbance did not survive covariate adjustment, suggesting that this weaker finding was indeed partially confounded by group membership. We have therefore conservatively interpreted the microstate C duration correlation as tentative, while presenting the B → C transition as the most robust and replicated finding across all analytical approaches. This association is theoretically meaningful: a reduced propensity to switch from visual processing (microstate B) to salience detection (microstate C) may reflect a fundamental disruption in how external perceptual information is integrated into the salience network in psychotic depression, potentially contributing to the misattribution of salience characteristic of psychotic symptoms. The positive correlation between microstate C duration and thought disturbance (ρ = 0.243) was weaker and did not survive Bonferroni correction; we therefore interpret this as a tentative, hypothesis-generating finding that requires replication in larger, medication-naïve samples. It is important to emphasize that all correlation analyses in this study were exploratory and should not be overinterpreted as establishing causal or clinically predictive relationships. Collectively, our results position MDD-Psy as exhibiting a hybrid microstate pattern: sharing certain features with primary psychotic disorders while maintaining distinct characteristics that differentiate it from both schizophrenia and non-psychotic depression.

The present study advances the field through several methodological innovations that differentiate it from prior work. First, our direct comparison of MDD-Psy and MDD-NonPsy groups, rather than the conventional patient-control designs employed in schizophrenia research [16, 17], isolates neurophysiological features specific to psychotic features rather than depression broadly. The correlation analyses revealed a differential pattern of association strength between microstate dynamics and psychotic symptom dimensions. The most robust finding was the negative correlation between the B → C transition probability and both thought disturbance and hostility severity. This association survived both FDR and Bonferroni correction, representing a medium effect size (|ρ| ≈ 0.35–0.39) that is unlikely to reflect spurious correlation inflation. Theoretically, a reduced propensity to switch from visual network processing (microstate B) to salience network engagement (microstate C) may reflect a fundamental disruption in how external perceptual information is integrated into the salience evaluation system in psychotic depression. This “decoupling” of visual input from salience attribution could mechanistically contribute to the misattribution of salience and reality distortion characteristic of psychotic symptoms [15, 43]. The reciprocal increase in C → B transitions further supports this interpretation, suggesting a bidirectional dysregulation in which the salience network not only fails to appropriately receive visual input but also aberrantly redirects signals back to visual processing.

In contrast, the positive correlation between microstate C duration and thought disturbance, while nominally significant after FDR correction, did not survive Bonferroni correction and represents a small effect size (ρ = 0.243). We therefore interpret this association as tentative and hypothesis-generating rather than conclusive. It is possible that prolonged salience network stability contributes to thought disturbance severity, but the modest effect size and failure to withstand the most stringent multiple comparison correction suggest that this relationship requires replication in larger, ideally medication-naïve samples before it can be considered a reliable candidate marker. Notably, the absence of a significant correlation between microstate C duration and hostility further supports the specificity of the B → C transition as the more robust psychophysiological marker.

This differential robustness has important implications for the theoretical interpretation of our findings. The transition-based measures (B → C, C → B, C → A) capture the dynamic syntax of network interactions—the sequential rules governing how large-scale brain configurations switch and influence one another [16, 19]. These dynamic parameters may be more sensitive to the pathophysiological disruptions underlying psychosis than static measures such as mean duration, which reflect only the temporal stability of a single network state. Our results align with emerging transdiagnostic evidence that impaired network switching, rather than isolated network hyperactivity or hypoactivity, constitutes the core neurophysiological signature of psychosis spectrum conditions [40]. The medium effect sizes observed for transition correlations (|ρ| = 0.35–0.39) are consistent with prior microstate-symptom associations in first-episode schizophrenia and clinical high-risk populations [16, 19], suggesting that our findings are not outliers but rather fit within the established effect size range for this literature.

The robust association between salience network microstate dynamics and psychotic symptoms also resonates with studies linking microstate C to aberrant salience processing across psychiatric disorders [15, 43] and with treatment response prediction in schizophrenia [39] and depression [44]. Our findings underscore the utility of microstate analysis as a translational candidate marker tool that bridges the gap between phenomenology and pathophysiology in affective psychosis, offering a temporal precision that complements structural and functional neuroimaging studies in psychotic depression [4]. Furthermore, the clinical stratification using both KDAS and BPRS dimensions represents a more nuanced approach than prior studies relying solely on diagnostic categories, echoing recent calls for dimensional assessment across psychiatric disorders [3, 45].

The most important limitation of this study is the inability to fully disentangle disease-specific microstate alterations from the effects of antipsychotic medication. Patients with psychotic depression were more frequently treated with antipsychotics (69.1%) than non-psychotic patients (19.0%), reflecting clinical necessity. Antipsychotic medications are known to induce measurable alterations in brain structure and functional connectivity in patients with psychotic depression [4], raising the possibility that observed microstate differences could partially reflect pharmacological rather than pathophysiological effects. However, several considerations argue against medication as the sole explanation. First, in supplementary ANCOVA models, the group differences in microstate C duration and transition probabilities remained significant after controlling for antipsychotic medication status as a binary covariate (all p-corr < 0.05), and there was no significant group × medication interaction (F = 1.34, p = 0.249; Supplementary Table S2). Second, the medication-naïve subsample, though underpowered (n = 14), showed a consistent direction of effects with the full sample (Supplementary Table S2). Third, prior microstate research in psychosis populations has demonstrated that microstate D—rather than microstate C—shows the strongest sensitivity to medication status and clinical phase [17], suggesting that microstate C prolongation may reflect a more stable, disease-related characteristic. Fourth, the robust correlation between microstate C duration and thought disturbance severity, which remained significant after controlling for medication status, would not be expected if the finding were a pure pharmacological artifact. Nevertheless, we acknowledge that without systematic dose recording (e.g., chlorpromazine equivalents) or a prospective medication-naïve design, medication effects cannot be fully excluded. Future longitudinal studies enrolling drug-naïve patients prior to treatment initiation are essential to confirm these findings as trait markers of psychotic depression. Second, illness duration and educational level were not systematically recorded or controlled in this study. Both variables may influence EEG microstate dynamics: longer illness duration may be associated with neuroprogressive alterations in large-scale network function, while educational level may proxy for cognitive reserve and premorbid functioning. Although the two groups were matched for age and sex, the absence of these covariates precludes a fully adjusted analysis. Future prospective studies should record illness duration from illness onset and years of education to determine whether the observed microstate alterations are moderated by clinical chronicity or cognitive reserve. Third, the absence of a healthy control group limits our ability to contextualize the observed microstate alterations relative to normative brain dynamics. We cannot determine whether the prolonged microstate C duration and altered transition probabilities observed in MDD-Psy represent a specific marker of psychotic features within depression, a general feature of all MDD patients, or a deviation from healthy brain function shared across psychiatric conditions. This case-case design was chosen to directly isolate the neurophysiological features distinguishing psychotic from non-psychotic depression, but a three-group design incorporating healthy controls is essential to disentangle disease-specific from subtype-specific alterations. We have therefore tempered our conclusions to avoid claims of specificity relative to healthy function. Fourth, the cross-sectional design precludes establishing causal relationships between microstate dynamics and psychotic symptoms, nor can we determine whether these alterations represent state-dependent markers of acute psychosis or trait vulnerability markers [39]. Fifth, our sample was recruited from a single clinical setting with specific demographic characteristics, which may limit generalizability to other populations or cultural contexts. Sixth, the 19-channel EEG system, while standard for clinical microstate analysis, offers limited spatial resolution compared to high-density systems. An important caveat in interpreting our findings is the limited spatial resolution of 19-channel scalp EEG. While the association between microstate C and the salience network is supported by converging evidence from EEG-fMRI simultaneous recordings and source-localization studies [15, 18, 36], our scalp topography data cannot directly confirm this network attribution. High-density EEG systems or combined EEG-fMRI approaches would be needed to more precisely localize the neural generators of the observed microstate alterations. We have therefore used cautious phrasing throughout the manuscript (e.g., ‘putatively associated with,’ ‘commonly linked to’) to reflect this indirect mapping. Additionally, the 100 Hz sampling rate limits temporal resolution to 10 ms, which may introduce quantization effects in transition probability estimation, particularly for microstate durations approaching the lower end of the typical 60–120 ms range. While our GFP peak detection algorithm (minimum inter-peak interval: 10 ms) and the observed median durations (~ 86–92 ms) suggest that this resolution is adequate for capturing dominant microstate dynamics, future studies employing higher sampling rates (e.g., 500–1000 Hz) could refine transition probability estimates and reduce potential discretization artifacts.

Future directions should encompass longitudinal investigations assessing microstate stability across illness phases and treatment interventions to differentiate state versus trait markers, following patients from acute psychotic depression through remission, alongside multimodal integration combining EEG microstates with fMRI connectivity measures to provide complementary temporal and spatial resolution, elucidating how millisecond-level network dynamics relate to longer-timescale functional connectivity disruptions observed in psychotic depression [15, 42]. Machine learning approaches should be developed to validate microstate parameters as diagnostic candidate markers, potentially enabling early identification of psychotic features in depression and predicting treatment response, as demonstrated in recent schizophrenia and epilepsy studies [39, 26, 49], while transdiagnostic extensions examining microstate patterns across schizophrenia, bipolar disorder, and psychotic depression could clarify whether the observed microstate C prolongation represents a psychosis spectrum marker or a depression-specific phenomenon [40, 41, 43]. Additionally, sleep EEG protocols should be implemented, as recent work demonstrates that microstate abnormalities differ between wake and sleep states in schizophrenia, potentially revealing novel candidate markers and therapeutic targets [50], and intervention studies manipulating salience network function through pharmacological or neuromodulatory approaches could test whether normalizing microstate C dynamics reduces psychotic symptoms, building on treatment prediction models in depression [44]. Microstate analysis and graph‑theoretic connectivity offer complementary views: the former captures temporal dynamics of global brain configurations, while the latter quantifies pairwise connectivity and network topology [30–32]. Beyond graph-theoretic connectivity, future studies should explicitly incorporate interhemispheric synchronization and signal complexity metrics to fully capture the neurophysiological dynamics of psychotic depression. Interhemispheric synchronization measures—including spectral coherence, phase-locking value, and mutual information—are essential given the documented interhemispheric dysregulation in ruminative states [47] and the potential hemispheric asymmetry in microstate transitions observed in our data (e.g., differential C→B vs. C→A patterns). These metrics would directly test whether the disrupted visual-salience coupling (B→C) is accompanied by systematic interhemispheric decoupling, particularly within frontoparietal regions implicated in salience evaluation. Furthermore, signal complexity measures—such as multiscale entropy, Lempel-Ziv complexity, and frequency-specific complexity (FSC)—should be applied to the same 19-channel datasets to quantify the reduced neural complexity associated with rigid, perseverative microstate dynamics. The documented decrease in frontal EEG complexity among chronic ruminators suggests that microstate prolongation and entropy reduction may represent complementary markers of the same pathological rigidity [46]. Integrating these approaches would provide a more comprehensive neurophysiological portrait of psychotic depression, bridging global network dynamics (microstate transitions), interhemispheric coordination, and local information-processing capacity within a unified analytical framework.

Microstate analysis and graph‑theoretic connectivity offer complementary views: the former captures temporal dynamics of global brain configurations, while the latter quantifies pairwise connectivity and network topology [30–32]. Recent studies have successfully applied graph‑theoretic analysis to resting‑state EEG in depression, revealing altered network properties [51]. A natural next step is to apply graph‑theoretic connectivity analysis to the same 19‑channel EEG dataset to examine whether prolonged microstate C duration relates to altered salience network connectivity, and whether altered transition probabilities (e.g., reduced B→C) correspond to specific inter‑network changes. Such integration would deepen understanding of network‑level correlates in psychotic depression.

Recent studies have successfully applied graph‑theoretic analysis to resting‑state EEG in depression, revealing altered network properties [48]. A natural next step is to apply graph‑theoretic connectivity analysis to the same 19‑channel EEG dataset to examine whether prolonged microstate C duration relates to altered salience network connectivity, and whether altered transition probabilities (e.g., reduced B→C) correspond to specific inter‑network changes. Such integration would deepen understanding of network‑level correlates in psychotic depression.

The most critical limitation is the potential confounding effect of antipsychotic medication. MDD-Psy patients were more frequently medicated with antipsychotics (69.1% vs. 19.0%), and these agents are known to alter EEG microstates non-specifically. However, three lines of evidence argue against medication as the sole explanation: (i) the group differences in microstate C duration and transition probabilities remained significant after controlling for medication status in ANCOVA (all p-corr < 0.05); (ii) no significant group × medication interaction was observed (F = 1.34, p = 0.249); and (iii) the medication-naïve subsample showed consistent effect directions, albeit underpowered (Supplementary Table S2). Nonetheless, without systematic dose recording or a drug-naïve design, medication effects cannot be fully excluded. Additional limitations include the cross-sectional design (precluding causal/state-trait inferences), single-site recruitment, 19-channel spatial resolution, and the absence of a healthy control group. The 100 Hz sampling rate limits temporal resolution to 10 ms, which may introduce quantization effects in transition probability estimation, particularly for microstate durations approaching the lower end of the typical 60–120 ms range. While our GFP peak detection algorithm (minimum inter-peak interval: 10 ms) and the observed median durations (~ 86–92 ms) suggest that this resolution is adequate for capturing dominant microstate dynamics, future studies employing higher sampling rates (e.g., 500–1000 Hz) could refine transition probability estimates and reduce potential discretization artifacts.

Conclusions

Psychotic depression represents a severe clinical subtype of major depressive disorder associated with distinct alterations in large-scale brain network dynamics. Within this case-case comparison, patients with MDD-Psy exhibited prolonged salience network activity (indexed by microstate C duration) and reconfigured temporal transitions between the salience, visual, and auditory networks relative to patients with MDD-NonPsy. The most robust finding was the altered transition probability from the visual network to the salience network (B → C), which showed medium-effect-size correlations with core psychotic symptom dimensions including thought disturbance and hostility, surviving both FDR and Bonferroni correction. These findings suggest that dysregulated dynamic coupling between perceptual processing and salience evaluation may constitute a neurophysiological correlate of psychotic features within depression.

However, these results should be interpreted with considerable caution. First, the cross-sectional design and the high rate of antipsychotic medication use in the MDD-Psy group preclude a tentative causal attribution to the disease process alone; while supplementary analyses controlling for medication status preserved the group differences, a medication-naïve design is required to confirm these findings as trait markers. Second, the absence of a healthy control group limits our conclusions to differences between MDD subtypes rather than abnormalities relative to normative brain function; we cannot determine whether the observed microstate alterations are specific to psychotic depression, generalizable to all MDD patients, or shared with other psychiatric conditions. Third, illness duration and educational level were not systematically recorded, precluding full adjustment for potential clinical and developmental confounders. Fourth, the correlation between microstate C duration and thought disturbance, while nominally significant, represented a small effect size and did not survive the most stringent multiple comparison correction; this association should be considered tentative and hypothesis-generating. Finally, microstate analysis captures global network configuration dynamics but does not resolve the underlying spectral power shifts or local oscillatory mechanisms; integration with complementary EEG signal processing approaches will be essential to move from descriptive correlation toward hypothesized mechanism.

At present, our findings contribute to a hypothesized mechanistic framework for psychotic depression by identifying reconfigured temporal syntax between large-scale brain networks as a candidate neurophysiological signature distinguishing MDD-Psy from MDD-NonPsy. Validation in longitudinal, medication-naïve samples with healthy controls is required before clinical translation can be considered. Future longitudinal studies enrolling medication-naïve patients, incorporating healthy controls, and combining microstate analysis with spectral and complexity markers are required to validate whether these dynamic network alterations can serve as reliable, potentially clinically informative candidate markers for the identification and prognostication of psychotic features in depression.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Yu Yang: Conducted EEG data acquisition and preprocessing, performed microstate analysis, contributed to data interpretation, and drafted the manuscript. Dexin Liao: Contributed to study design, conducted clinical assessments and EEG recordings, participated in data analysis and interpretation, and revised the manuscript. Yudiao Liang: Conducted clinical assessments, recruited participants, and contributed to data collection. Yanping Feng: Conducted clinical assessments, coordinated participant enrollment, and contributed to data collection. Shanhong Wang: Conducted clinical assessments, managed clinical data, and contributed to participant follow-up. Sha Zhang: Conducted clinical assessments, assisted with EEG data acquisition, and contributed to data collection. Kezhi Liu: Conceived and designed the study, secured funding, supervised all aspects of the research, contributed to data interpretation, and revised the manuscript critically for intellectual content. Kezhi Liu is the corresponding author and accepts full responsibility for the work and the conduct of the study, had access to the data, and controlled the decision to publish. All authors read and approved the final manuscript.

Funding

This project was supported by the Key Science and Technology Program of Zigong City (2023-NKY-02-11).

Data availability

The datasets generated and analyzed during the current study are not publicly available due to patient confidentiality and ethical restrictions but are available from the corresponding author on reasonable request, subject to approval by the institutional ethics committee of Zigong Mental Health Center.

Declarations