Opposite network patterns of integration-segregation in psychedelic and sedated states of consciousness Authors: Rui Dai, Hyunwoo Jang, Anthony G. Hudetz, Zirui Huang, George A. Mashour Published: 2026 ABSTRACT SUMMARYUnderstanding the relationship of conscious states to large-scale brain dynamics remains a central challenge in neuroscience. Here, diverse pharmacological and physiological perturbations—including psychedelics, sleep, and propofol sedation—are used to examine integration-segregation organization of brain networks across altered states of consciousness in humans using functional MRI. Across analyses, psychedelic and sedative states exhibit a robust “mirror-image” pattern of integration and segregation. Psychedelics are characterized by increased large-scale integration and reduced segregation of brain network interactions, whereas sleep and propofol sedation show the opposite configuration. These divergent integration-segregation patterns are indexed by complementary measures of functional connectivity, network topology, and interaction complexity, capturing non-redundant dimensions of large-scale brain organization. Importantly, this mirror-image organization generalizes across multiple spatial levels and reliably differentiates conscious states through unbiased, data-driven methods. Together, these findings demonstrate that psychedelic and sedated states are characterized by systematic and opposite shifts in large-scale integration and segregation, with implications for mechanisms of consciousness. INTRODUCTION The investigation of consciousness has become increasingly tractable using neuroimaging techniques such as functional magnetic resonance imaging (fMRI) that allow measurement of large-scale brain dynamics and their relationship to cognitive and conscious states.1,2 These techniques can be coupled with pharmacological interventions that enable controlled and reversible perturbations of consciousness. Two major classes of psychoactive compounds are particularly useful to advance this approach. Anesthetic agents reliably induce behavioral unresponsiveness and disrupt large-scale cortical communication, providing a well-established model of depressed consciousness.3,4 Psychedelic compounds, in contrast, alter conscious experience in qualitatively different ways, dramatically altering phenomenology—including intensified perception and altered self-experience—while typically preserving wakefulness.5–7 However, these two drug classes are rarely investigated together in systematic studies. A central, long-standing, and open question in the field is whether integration and segregation constitute key determinants of conscious state. Although anesthesia and psychedelics are assumed to produce neural changes in divergent directions, no study has systematically tested whether these shifting neural signatures reflect structured, large-scale transformations along this integration-segregation axis of brain organization. Neuroimaging studies of anesthetic-induced unconsciousness reveal a highly convergent systems-level profile characterized by weakened long-range interactions and constrained neural dynamics. Resting-state fMRI demonstrates robust reductions in communication across higher-order association cortices, with relative preservation of primary sensory systems.8–10 Network-level analyses further indicate a shift toward more locally clustered and modular configurations, accompanied by reduced efficiency of information exchange across distributed systems.11,12 Complementing these spatial findings, information-theoretic approaches reveal a contraction of the brain’s dynamical repertoire, with neural activity becoming increasingly predictable and structurally constrained.13,14 Together, these results depict anesthesia as a transition toward a constrained and fragmented mode of large-scale brain organization that disrupts the conditions supporting conscious processing. In contrast, neuroimaging of psychedelic states suggests a large-scale reorganization in the opposite direction, marked by enhanced integration, reduced segregation, and enriched neural dynamics. Resting-state fMRI reveals increased global connectivity, reduced within-network connectivity, and strengthened cross-network coupling across associative systems such as the default mode, salience, and frontoparietal networks.5,7,15–17 Graph-theoretical and cortical gradient analyses show a collapse of modular structure and flattening of the unimodal-transmodal gradient, indicating more globally distributed and flexible modes of cortical communication.18–20 At the level of neural dynamics, information-theoretic and spectral approaches reveal expanded state repertoires, the emergence of higher-frequency components, and increased signal diversity, indicative of a more flexible and exploratory dynamical regime.14,21–23 Together, these findings stand in direct contrast to anesthesia-induced network contraction and may underlie the profound alterations of conscious experience elicited by psychedelics. Derived from diverse analytical approaches, this body of work converges on a common picture in which conscious states are systematically differentiated by shifts in large-scale integration and segregation.24 Although various studies provide evidence that psychedelics and anesthetics induce opposite network configurations, there has been no rigorous assessment using consistent analytic approaches to integration-segregation balance. Here, we report a systematic test of whether integration and segregation offer a unifying organizational framework for conscious brain states. To do so, we analyzed fMRI data spanning multiple pharmacological and natural perturbations of consciousness, including psychedelic states (lysergic acid diethyamide [LSD], psilocybin, subanesthetic ketamine, and subanesthetic nitrous oxide [N2O]), anesthetic-induced unresponsiveness with propofol, and sleep. Across these conditions, we quantified large-scale brain organization at multiple hierarchical levels (whole brain, unimodal-attentional-transmodal hierarchy, and canonical functional networks) using functional connectivity, network topology, and interaction complexity. These measures were used as complementary indices of inter-regional coupling, large-scale topological integration, and temporal interaction diversity, respectively. Together, this multi-level, multi-metric approach enables a data-driven evaluation of whether shifts in integration and segregation consistently organize diverse conscious states across perturbations, spatial scales, and analytical domains. RESULTS We investigated two classical (i.e., serotonergic) and two non-classical psychedelic agents—LSD (n = 15), psilocybin (n = 7), ketamine (n = 12), and N2O (n = 15)—all known to induce non-ordinary states of consciousness,5,7,20 as well as sleep and sedation, which are associated with diminished consciousness. For sleep, we analyzed non-REM sleep stages N1 (n = 33) and N2 (n = 29). N3 sleep was not analyzed due to insufficient sample size (n = 3). For sedation, we examined the effects of propofol at various effect-site concentrations: 1.9 μg/mL (n = 14), 2.4 μg/mL (n = 26), and 2.7 μg/mL (n = 27). Each altered state of consciousness was compared with its corresponding baseline state of wakefulness. To characterize brain organization across altered states of consciousness, we employed a set of complementary measures that capture distinct aspects of integration and segregation across multiple spatial scales (Figure 1). Functional connectivity quantifies the strength of statistical coupling between brain regions and provides a measure of functional association that can be examined across different spatial scales. Topological integration quantifies the graph-theoretical conditions that support efficient communication across the network, measured here as global efficiency. Interaction complexity captures the diversity and dimensionality of interaction patterns over time and is, therefore, inversely related to dynamical segregation: higher interaction complexity indicates a richer and more flexible repertoire of interaction modes, whereas lower interaction complexity reflects more constrained, stereotyped, and segregated dynamics.24 Interaction complexity should be distinguished from other measures of complexity, such as Lempel-Ziv, which are often applied to neurophysiological recordings such as the electroencephalogram and are regarded as having a direct relationship with differentiated neural activity. Together, these measures dissociate changes in functional association, network topology, and interaction dynamics, enabling a data-driven assessment of integration-segregation across states of consciousness. Global-level patterns of functional connectivity, topological integration, and interaction complexity across conscious states At the global level, states of consciousness were robustly differentiated using summary measures that aggregate functional connectivity across the entire brain, without reference to specific network-network interactions. Across the psychedelic datasets, a general pattern of increased large-scale integration was observed relative to normal wakefulness (Figure 2; Tables S1–S3). Psychedelic states were characterized by elevated whole-brain average functional connectivity, increased between-network functional connectivity, and reduced within-network functional connectivity, indicating enhanced coupling across distributed brain networks. Convergent effects were observed in topological integration and interaction complexity: topological integration was increased, and interaction complexity was elevated under psychedelic conditions, reflecting both greater network integration and more diverse interaction patterns over time. In contrast, states associated with diminished consciousness exhibited the opposite configuration across the same measures. Both non-REM sleep and propofol sedation were characterized by reductions in whole-brain functional connectivity, accompanied by decreased between-network and within-network connectivity. These changes were paralleled by reductions in topological integration and interaction complexity, indicating a breakdown of large-scale integration and a concomitant reduction in the diversity of network interactions. To directly quantify whether psychedelic and diminished-consciousness states change in statistically opposite directions across pharmacological and physiological perturbations of consciousness, we fit a linear mixed-effects model (LME) that pooled observations from all nine altered states within a single unified framework (see STAR Methods). The model included group (psychedelic vs. diminished consciousness) and state (baseline vs. altered) as fixed effects, mean framewise displacement as a motion covariate, and random effects to account for dataset-level heterogeneity and within-subject dependence. The critical state × group interaction term tests whether the two groups show reliably opposite trajectories from baseline to altered state. Across all nine metric × measure combinations (three metrics × three spatial measures; Figure 2), this interaction was positive and significant after Benjamini-Hochberg false discovery rate (BH-FDR) correction (all β > 0, all pFDR <0.05), confirming that the mirror-image organization reflects a robust and generalizable cross-dataset effect spanning coupling strength, topological efficiency, and temporal richness simultaneously. Together, these results reveal a robust mirror-image pattern of large-scale brain organization across altered states of consciousness: psychedelic states consistently shift brain networks toward greater integration and interaction complexity, whereas non-REM sleep and propofol sedation converge on a state of reduced integration and diminished network dynamics. Hierarchy-level patterns of functional connectivity, topological integration, and interaction complexity across conscious states At the level of the cortical functional hierarchy, large-scale brain organization exhibited systematic and graded differences across unimodal, attentional, and transmodal systems (Figure 3; Table S4). We examined interactions both within and between hierarchy levels of functional organization, spanning within-unimodal, within-attention, within-transmodal, unimodal-attention, unimodal-transmodal, and attention-transmodal relationships, across functional connectivity, topological integration, and interaction complexity (Figure 3). Across all three measures, altered states of consciousness exhibited a robust mirror-image pattern across both within-level and between-level interactions along the cortical functional hierarchy. Relative to baseline wakefulness, psychedelic states were associated with increases in hierarchy interactions, whereas non-REM sleep and propofol sedation showed concordant reductions across the same hierarchy contrasts. Within this overall mirror-image organization, psychedelic states tended to show stronger effects on transmodal-related interactions. In functional connectivity, increases were apparent both within transmodal systems and in interactions linking transmodal regions with attentional and unimodal systems, with a similar pattern observed in interaction complexity. In contrast, states associated with diminished consciousness showed the largest reductions at lower levels of the cortical hierarchy. Across non-REM sleep and propofol sedation, reductions were most pronounced within unimodal systems, within attentional systems, and in unimodal-attention interactions, a pattern that was consistently observed across functional connectivity, topological integration, and interaction complexity. The same LME framework was applied at the hierarchy level to formally test the mirror-image effect across all eighteen metric × hierarchy-pair combinations (three metrics × six hierarchy pairs). Across all eighteen combinations, the state × group interaction was positive and significant after BH-FDR correction (all β > 0, all pFDR <0.05; Figure 3). Together, these findings demonstrate that alterations in conscious state are differentially expressed across levels of the cortical functional hierarchy. Psychedelic states tend to show stronger effects on higher-order interactions involving attentional and transmodal systems, whereas diminished states of consciousness are characterized by pronounced reductions at lower hierarchy levels, including unimodal systems and unimodal-attention interactions. Across these hierarchical contrasts, interaction complexity provided a particularly sensitive index of state-dependent differentiation. Network-level patterns of functional connectivity, topological integration, and interaction complexity across conscious states At the level of functional networks, altered states of consciousness continued to exhibit a robust mirror-image pattern in large-scale brain organization (Figure 4; Table S5). We examined pairwise interactions among seven functional networks—visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal, and default-mode—across functional connectivity, topological integration, and interaction complexity. Across all psychedelic conditions, network-level analyses revealed wide-spread increases in inter-network functional connectivity, particularly involving frontoparietal and default-mode networks. These increases were accompanied by corresponding increases in topological integration and interaction complexity (Figure 4A). In contrast, sleep and propofol sedation showed the opposite network-level pattern. Both the deeper stage of non-REM sleep and increasing effect-site concentrations of propofol displayed robust reductions in inter-network functional connectivity, together with parallel decreases in topological integration and interaction complexity across a broad range of network pairs (Figure 4B). The magnitude of these reductions increased systematically with anesthetic concentration. The same LME framework was applied at the network level, fitting separate models for each of the 84 metric × network-pair combinations (3 metrics × 28 network pairs). The state × group interaction was positive and significant after BH-FDR correction across the majority of network pairs for all three metrics (Figure 4C). Taken together, network-level analyses revealed consistent, state-dependent differences in functional connectivity, topological integration, and interaction complexity across large-scale brain networks. By resolving these effects at the level of individual functional networks, these findings demonstrate that the mirror-image reorganization observed at global and hierarchy levels is underpinned by coherent, network-specific patterns of integration and segregation. To rule out head-motion confounds, we conducted four complementary analyses examining whether motion differences between baseline and altered states could account for the mirror-image pattern (Figure S1; Tables S6–S9). Motion differences between conditions did not align with the psychedelic-versus-sedation distinction underlying the mirror-image pattern, and subject-wise changes in motion rarely correlated with subject-wise changes in brain metrics. All originally significant effects were retained after direct motion covariate adjustment, and the state × group interaction in the LME model remained significant across all three spatial levels when mean framewise displacement was included as an explicit covariate. Data-driven consistency of integration-segregation effects across metrics and spatial scales To assess the consistency of effects across metrics and spatial scales, we next examined whether large-scale network features yielded a common ordering of conscious states. We first pooled the effect size rank of the states by averaging the ranks of Cohen’s d of nine states at each global- and hierarchical-level analysis (Figure 5A). This yielded a clear ordering, with psychedelic states occupying higher ranks (i.e., overall positive Cohen’s d) and sleep and propofol sedation occupying progressively lower ranks (i.e., overall negative Cohen’s d). To evaluate which level or metric best aligned with this overall rank, we assessed the Kendall’s tau between the average rank and within-metric rank (Figure 5B). Across spatial scales, interaction complexity showed the strongest agreement with the average rank, followed by functional connectivity and topological integration, indicating that complexity-based measures most consistently captured the overall ordering of conscious states. To further evaluate whether this ordering can be reproduced in a network-level approach, we performed an unsupervised principal component analysis (PCA) on network-level Cohen’s d values. Projection onto the first principal component revealed a clear separation of psychedelic states from sleep and propofol sedation along a single dominant axis (Figure 5C). PC1 explained 69.8% of the variance in the network-level Cohen’s d matrix, PC2 explained an additional 12.3%, and PC3 explained 8.4% (cumulative variance explained by PC1–PC3: 90.5%), indicating that a single dominant axis captures the majority of between-condition variance. Notably, the ordering along this axis perfectly recapitulated the ordering in average rank obtained from global- and hierarchy-level analyses. Examination of the first principal component revealed that the separation was driven by a coordinated contribution of multiple features, including functional connectivity, topological integration, and interaction complexity, rather than by any single metric or network class (Figures 5D–5F). Together, analyses across multiple metrics and spatial levels consistently revealed mirror-image patterns of integration and segregation across states of consciousness. DISCUSSION This study systematically investigated a broad spectrum of altered conscious states, revealing a striking “mirror image” pattern in network features of integration and segregation. Comparing psychedelic states to sleep and propofol sedation, changes in functional connectivity, network topology, and interaction complexity consistently diverged in opposite directions. Psychedelic states were characterized by increased large-scale interactions across brain networks, whereas sleep and propofol sedation showed convergent reductions across the same metrics. Importantly, this mirror-image pattern was preserved across global, hierarchical, and network-level analyses, indicating that opposite patterns of integration and segregation are a stable feature of altered conscious states rather than a property of any single spatial scale or analytical approach. Although prior research has identified alterations in between-network and within-network integration during psychedelic states,7,22,25,26 sleep,27,28 and sedation,3,8,24,29–32 most studies have examined these effects within a limited set of metrics or spatial scales or studied the states in isolation. Here, we extend this literature by systematically characterizing large-scale brain organization across a broad spectrum of altered states using complementary measures of functional connectivity, network topology, and interaction complexity. By jointly considering these metrics and states, our approach provides an integrated framework for assessing how large-scale patterns of brain interaction reorganize across different modes of consciousness. Crucially, we examined large-scale brain organization across multiple spatial levels, encompassing global measures of whole-brain coupling, an intermediate cortical functional hierarchy defined by a unimodal-attentional-transmodal axis,33 and network-specific analyses based on canonical functional systems. Together, these levels provide complementary perspectives on the reconfiguration of large-scale brain organization across conscious states. This hierarchy level provides an intermediate scale between global measures of large-scale coupling and finer-grained network-specific analyses. The unimodal-attention-transmodal hierarchy captures a graded transition from primary sensory processing to higher-order associative integration and has emerged as a fundamental organizational principle of the human cortex.20,34–38 By quantifying interactions both within and between unimodal, attentional, and transmodal systems, this approach allows state-dependent reconfigurations to be situated along the functional hierarchy while remaining anchored to global and network-level findings. These findings are consistent with prior psychedelic studies reporting enhanced interactions among higher-order systems18,39 and extend them by embedding such effects within a unified hierarchy-level framework that generalizes across pharmacological and physiological perturbations of consciousness. A key strength of the present study lies in the use of multiple, complementary metrics to characterize large-scale brain organization across altered states of consciousness. Functional connectivity provides a sensitive measure of statistical coupling between brain systems but, on its own, cannot distinguish whether increased coupling reflects more efficient integration or simply stronger correlations within a fixed network architecture. Topological integration, derived from graph-theoretical measures, contextualizes connectivity within the broader network structure by characterizing the conditions for efficient large-scale interactions, yet remains relatively insensitive to temporal variability in interaction patterns. Interaction complexity, in contrast, captures the diversity and dimensionality of network interactions over time, but does not directly index connection strength or network efficiency. By combining these metrics, our approach overcomes the limitations inherent to any single measure and provides a more complete characterization of large-scale brain organization. Importantly, these three measures are not interchangeable and are dissociable in certain brain states. For example, two networks may exhibit high mean functional connectivity while simultaneously showing low interaction complexity if their interactions become highly stereotyped or temporally rigid—a condition that may reflect “rigid integration” rather than flexible large-scale communication. In this framework, functional connectivity primarily reflects the strength of inter-regional coupling, topological integration reflects the efficiency of information transfer within the time-averaged network architecture, and interaction complexity captures the temporal richness and flexibility of network dynamics across time. Thus, integration as indexed by graph-theoretical topological integration should be conceptually distinguished from complexity, which more directly characterizes the dynamical repertoire of brain interactions. The convergence of functional connectivity, topological integration, and interaction complexity on the same mirror-image pattern across psychedelic states, sleep, and propofol sedation indicates that the observed effects are not driven by metric-specific biases or analytical choices. Instead, these findings reflect a robust reconfiguration of large-scale brain organization that manifests consistently across multiple, non-redundant dimensions. The observation that psychedelic states increase all three metrics, whereas sedative and sleep states generally decrease them, suggests a coherent systems-level shift spanning coupling strength, topological efficiency, and temporal richness simultaneously, rather than an isolated alteration in any single network property. The relationships described here are conceptual rather than formally causal, as the present analyses do not include directional or effective connectivity measures capable of establishing causal relationships between these organizational levels. Importantly, the mirror-image organization identified here was not specific to a single dataset, compound, or experimental manipulation, but was consistently observed across a diverse set of pharmacological and physiological perturbations of consciousness, including multiple psychedelic agents, sleep, and propofol sedation. Despite substantial heterogeneity in study designs, acquisition protocols, and subject populations, all the altered states manifested the same opposite pattern of large-scale brain organization. Moreover, this distinction did not rely on any predefined ordering of conscious states or theory-driven assumptions. Instead, unsupervised, data-driven analyses of network-level features—most notably PCA—independently recovered a low-dimensional organization that cleanly separated psychedelic states from sleep and anesthetic conditions. Some psychedelic datasets, particularly N2O and subanesthetic ketamine, showed comparatively weaker or more variable effects when examined individually relative to serotonergic psychedelics such as LSD. This variability may reflect differences in subjective intensity or underlying pharmacological mechanisms across altered states, and future work directly comparing distinct pharmacological perturbations will be important. Similarly, more transitional states of consciousness, such as sleep N1, generally showed weaker and less consistent effects across metrics compared with deeper or more stable altered states. This pattern may reflect the graded nature of large-scale brain reorganization across the continuum from normal wakefulness to more profoundly altered states of consciousness. Importantly, however, the multi-dataset framework used here allows directionally consistent effects across heterogeneous datasets to collectively reveal a coherent large-scale organizational pattern that may not be apparent in individual studies alone. Limitations of the study This study has limitations. First, it is important to acknowledge that increased connectivity or integration does not always equate to a greater capacity for consciousness. For example, excessive connectivity can lead to pathological states, such as seizures, where neural synchronization is heightened but consciousness is impaired. However, this was mitigated by using alternative measures such as interaction complexity. Second, the reliance on multiple datasets introduces heterogeneity, which may create confounds due to different experimental settings or populations, including differences in scanner manufacturer, acquisition parameters, baseline conditions, and preprocessing pipelines. Although all primary analyses used within-subject comparisons to reduce sensitivity to between-dataset differences, heterogeneity cannot be fully mitigated and should be considered when interpreting the findings. Third, several datasets, particularly the psilocybin dataset, were limited by relatively small sample sizes, reducing the statistical robustness of individual-dataset inference. Accordingly, these findings should be interpreted cautiously and require replication in larger cohorts. Fourth, although our analyses span multiple spatial scales, they are based on resting-state fMRI and, therefore, cannot directly address content-specific or task-evoked mechanisms of consciousness. Resting-state approaches are inherently limited in their ability to probe stimulus-driven or perceptual content, and therefore, our findings should be interpreted as complementary to task-based investigations rather than a complete description of conscious states. Fifth, serotonergic psychedelics such as LSD and psilocybin may alter the blood-oxygen-level-dependent (BOLD) signal through direct vasoactive mechanisms, making it difficult to fully dissociate neural from vascular contributions through resting-state fMRI alone. Future work combining resting-state fMRI with direct measures of cerebral blood flow (e.g., arterial spin labeling) will be needed to address this limitation. Sixth, the present analyses focused on large-scale cortical functional organization and did not systematically incorporate subcortical structures, including corticothalamic loops and other key subcortical hub nodes that are known to play important roles in regulating conscious state and arousal.40 Accordingly, the present findings may reflect the large-scale cortical expression of altered conscious states while potentially missing primary subcortical driving sources of these integration-segregation patterns. Future studies incorporating detailed subcortical analyses and corticothalamic connectivity will be important for developing a more complete account of consciousness-related brain reorganization. Finally, it should be noted that the opposite patterns reported here are specific to the measures employed in the present study. Other analytical frameworks have reported partially shared signatures across psychedelic and anesthetic states, suggesting that these conditions may exhibit both shared and divergent features depending on the dimension of brain organization being examined. Despite these limitations, the present study advances the understanding of conscious states. By examining consciousness across diverse pharmacological and physiological perturbations, we identify a robust mirror-image organization of large-scale brain dynamics that generalizes across compounds, datasets, experimental contexts, analytic techniques, and spatial scales. This convergence strengthens the interpretability of the findings and supports the hypothesis that integration-segregation network features constitute a fundamental organizational principle of conscious brain dynamics. By situating diverse altered states within a unified systems-level framework, this work advances a more comprehensive understanding of consciousness that accommodates heterogeneity across states while identifying shared, generalizable organizational signatures. RESOURCE AVAILABILITY Lead contact Further information and requests for resources and data should be directed to and will be fulfilled by the lead contact, George A. Mashour (gmashour@med.umich.edu). Materials availability This study did not generate any new materials. Data and code availability Data used in the analyses are publicly available from Zenodo: https://doi.org/10.5281/zenodo.14029241. The natural sleep fMRI dataset is available from OpenNeuro: ds003768. The LSD dataset is available from OpenNeuro: ds003059. The psilocybin dataset is available from OpenNeuro: ds006072. openneuro.org/datasets/ds006072) Data used in the analyses are publicly available from Zenodo: https://doi.org/10.5281/zenodo.14029241. The natural sleep fMRI dataset is available from OpenNeuro: ds003768. The LSD dataset is available from OpenNeuro: ds003059. The psilocybin dataset is available from OpenNeuro: ds006072. openneuro.org/datasets/ds006072) Publicly available software and toolboxes used for analysis and visualization include AFNI (https://afni.nimh.nih.gov), Brain Connectivity toolbox (https://doi.org/10.1016/j.neuroimage.2009.10.003), JASP (v0.16.3; https://jasp-stats.org), MATLAB R2022a (https://www.mathworks.com), Prism (version 9.5.0; https://www.graphpad.com), and PyCharm (v2023.2.1; https://www.jetbrains.com/pycharm). Schematics in Figure 1 and the graphical abstract were created in part with BioRender (https://www.BioRender.com). Custom-built code is available in the Zenodo: https://doi.org/10.5281/zenodo.14029241. Publicly available software and toolboxes used for analysis and visualization include AFNI (https://afni.nimh.nih.gov), Brain Connectivity toolbox (https://doi.org/10.1016/j.neuroimage.2009.10.003), JASP (v0.16.3; https://jasp-stats.org), MATLAB R2022a (https://www.mathworks.com), Prism (version 9.5.0; https://www.graphpad.com), and PyCharm (v2023.2.1; https://www.jetbrains.com/pycharm). Schematics in Figure 1 and the graphical abstract were created in part with BioRender (https://www.BioRender.com). Custom-built code is available in the Zenodo: https://doi.org/10.5281/zenodo.14029241. Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request. STAR★METHODS EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS All participants were healthy adult human volunteers. Eight independent datasets were included: LSD (n = 15, 4 females; mean age ± SD: 30.5 ± 8.0 years), psilocybin (n = 7, 3 females; aged 18–45 years), ketamine (n = 12, 5 females; mean age ± SD: 41.4 ± 8.6 years), nitrous oxide (n = 15, 9 females; mean age ± SD: 24.9 ± 3.5 years), sleep (n = 33, 16 females; mean age ± SD: 22.1 ± 3.2 years), propofol 1.9 μg/mL (n = 14, 6 females; mean age ± SD: 26.7 ± 4.8 years), propofol 2.4 μg/mL (n = 26, 13 females; mean age ± SD: 25.0 ± 4.1 years), and propofol 2.7 μg/mL (n = 27, 20 females; mean age ± SD: 24.4 ± 5.2 years). Within each dataset, participants contributed data from both a baseline wakefulness condition and an altered-state condition using a within-subject design. All studies were approved by the appropriate institutional review boards, and all participants provided written informed consent. Information regarding gender identity, race, ethnicity, and ancestry was not consistently available in the original datasets and therefore was not included in the present analyses. The influence of sex on the reported effects was not formally analyzed because the analyses focused on within-subject state-dependent changes and the sample sizes of the individual datasets were insufficient for adequately powered sex-stratified comparisons. METHOD DETAILS Participants and data acquisition Dataset 1: LSD Fifteen healthy subjects (4 females, mean age ± SD: 30.5 ± 8.0 years) were sourced from the OpenNEURO database.5 The original study was approved by the National Research Ethics Service Committee London–West London and was conducted in accordance with the revised Declaration of Helsinki (2000), the International Council for Harmonisation Good Clinical Practice guidelines, and the National Health Service Research Governance Framework. All participants provided written informed consent. The study design involved two experimental sessions, where participants received either a placebo or 75 μg of intravenous LSD in a counterbalanced order. Subjective experiences were assessed using the 11-Dimensional Altered States of Consciousness (11D-ASC) questionnaire41 at the end of each dosing day. Each participant completed three scans: the first and third were eyes-closed resting-state sessions, while the second involved a music listening task. The first eyes-closed resting-state sessions, under both drug and no-drug (baseline) conditions, were used in this study; music-listening scans were not included. Each scan duration was 7 min. This study utilized a 3T GE HDx scanner. For echo-planar imaging (EPI), the scan settings were as follows: 35 slices, a repetition time/echo of 2000/35 ms, 3.4 mm slice thickness, a 220 mm field of view, and a 90° flip angle. High-resolution T1 images were acquired as well. In addition, only preprocessed data were available in the released dataset, followed by several steps. Initially, the first three volumes of each scan were removed to ensure stability. De-spiking was then conducted to correct signal artifacts, followed by slice time correction to align image acquisition timings. Motion correction was applied to counteract participant movement, and brain extraction isolated brain tissue from other elements in the images. The images were aligned to anatomical scans via rigid body registration and further aligned to a 2mm MNI brain template through non-linear registration. The dataset was then scrubbed using an FD threshold of 0.4, with a maximum of 7.1% of volumes scrubbed per scan, replacing them with the mean of surrounding volumes. Further processing included applying a 6mm kernel for spatial smoothing, filtering the data within the 0.01–0.1 Hz range, removing signal drifts through linear and quadratic de-trending, and eliminating motion- and anatomy-related artifacts through regression. Dataset 2: Psilocybin Seven healthy adults (3 females, aged 18–45 years) were enrolled in a randomized cross-over functional brain mapping study conducted at Washington University in St. Louis. The dataset was obtained from the OpenNeuro database.25 The original study was approved by the appropriate institutional review board, and all participants provided written informed consent. Participants had at least one prior lifetime exposure to a psychedelic substance but no psychedelic use within the six months preceding enrollment. Individuals with a history of psychiatric illness, including depression, psychosis, or substance use disorder, were excluded. For the present study, analyses focused exclusively on resting-state fMRI data acquired during the first non-drug (baseline) session and the first psilocybin dosing session for each participant. Psilocybin was administered orally at a dose of 25 mg, and resting-state fMRI acquisition commenced approximately 60–180 min following ingestion, corresponding to peak drug effects. Functional MRI data were acquired on a 3T Siemens Prisma scanner at Washington University Medical Center. High-resolution structural images (T1-and T2-weighted) were collected at 0.9 mm isotropic resolution. Functional MRI data were acquired using a multiband, multi-echo echo-planar imaging sequence with the following parameters: voxel size = 2 mm isotropic, repetition time (TR) = 1,761 ms, flip angle = 68°, multiband factor = 6, five echo times (TEs = 14.20, 38.93, 63.66, 88.39, 113.12 ms), and in-plane acceleration factor = 2. Each resting-state scan comprised 510 volumes (15 min 49 s). Two resting-state scans were collected per session. During resting-state acquisition, participants were instructed to fixate on a centrally presented white cross on a black background. Dataset 3: Ketamine The research was approved by the Institutional Review Board of Huashan Hospital, Fudan University, and all participants provided written informed consent.7,42 Twelve right-handed individuals (5 females; mean age ± SD: 41.4 ± 8.6 years) were enrolled. Participants had no history of neurological disorders, significant organ dysfunction, or neuropsychiatric medication use, and were classified as American Society of Anesthesiologists physical status I or II. Intravenous ketamine was administered while fMRI scans were performed without interruption throughout the experiment, which spanned 44 to 62 min. A 10-min baseline recording of the conscious state was conducted at the start, with the exception of two participants who had shorter baseline durations of 6 and 11 min. Ketamine was then administered at a rate of 0.05 mg/kg per minute over 10 min (total dose: 0.5 mg/kg), followed by an increased rate of 0.1 mg/kg per minute for another 10 min (cumulative dose: 1.0 mg/kg). Two participants only received the second dose. Once the ketamine infusion was complete, participants naturally regained consciousness. Behavioral responsiveness was monitored every 30 s during the scanning session using a button-press task. Our analysis focused on the 10-min subanesthetic ketamine administration, which is associated with psychedelic experiences. The present study analyzed only the resting-state fMRI data acquired during the 10-min subanesthetic ketamine administration period associated with psychedelic-like subjective effects. This study utilized a 3T Siemens MAGNETOM scanner. For EPI, the scan settings were as follows: 33 slices, a repetition time/echo time of 2000/30 ms, 5mm slice thickness, a 210 mm field of view, a 64 × 64 image matrix, and a 90◦ flip angle. High-resolution T1 images were acquired as well. Dataset 4: Nitrous oxide Twenty healthy participants (10 females, mean age ± SD: 25.6 ± 4.1 years) were included for this study, approved by the Institutional Review Board of the University of Michigan Medical School (HUM00096321), with all participants giving written informed consent.7,20 All participants were classified as American Society of Anesthesiologists physical status I, exclusion criteria included any history of drug abuse, psychosis, and other medical conditions as detailed in the trial registry (https://www.clinicaltrials.gov/ct2/show/NCT03435055). Three participants were excluded due to excessive head motion and two due to incomplete data, leaving a final sample of 15 healthy participants. This study utilized a within-subjects design with two conditions: a baseline condition and a condition with the administration of subanesthetic nitrous oxide (35%). Each condition included a 6-min resting-state scan and a 3-min passive viewing scan. Only the resting-state scans acquired before and during nitrous oxide administration were included in the present study. Neuroimaging data were acquired using a 3T Philips Achieva scanner. For EPI, the scan settings were as follows: 48 slices, a repetition time/echo time of 2000/30ms, 3 mm slice thickness, a 200 mm field of view, and a 90◦ flip angle. High-resolution T1 images were acquired as well. Subjective experiences were assessed both prior to the scanning session, using an 11D-ASC questionnaire,41 and after the scanning session to specifically evaluate experiences during the nitrous oxide administration. Dataset 5: Sleep Thirty-three healthy participants (16 females, mean age ± SD: 22.1 ± 3.2 years) were sourced from the OpenNEURO database.43,44 The original study was approved by the Pennsylvania State University Institutional Review Board, and all participants provided written informed consent. The dataset includes three non-REM sleep stages (N1, N2, and N3), in addition to an awake resting-state condition. These stages were identified using electroencephalogram signatures analyzed by a registered polysomnographic technologist. This study utilized a 3T Siemens Prisma scanner. For EPI, the scan settings were as follows: 35 slices, a repetition time/echo time of 2100/25 ms, 4 mm slice thickness, a 240 mm field of view, and a 90◦ flip angle. High-resolution T1 images were acquired as well. Only awake (n = 33), N1 (n = 33) and N2 (n = 29) were included in the analysis due to the limited number of subjects in N3 (n = 3). REM sleep was not available in the publicly available dataset. Dataset 6: Propofol sedation (1.9 μg/mL doses) Fifteen healthy participants (6 females, mean age ± SD: 26.7 ± 4.8 years) were included for this study, approved by the Institutional Review Board of the Medical College of Wisconsin.3,45 All participants provided written informed consent, were classified as American Society of Anesthesiologists physical status I or II. One participant was excluded due to excessive movement, leaving 14 participants for analysis. Behavioral responsiveness was measured using the Observer’s Assessment of Alertness/Sedation (OAAS) scale. During baseline and recovery, participants were fully responsive to verbal cues, indicated by an OAAS score of 5. In the light sedation phase, participants responded lethargically to verbal commands, corresponding to an OAAS score of 4, while deep sedation was marked by the absence of a response, with OAAS scores ranging from 1 to 2. Individual propofol target plasma concentrations varied (light sedation: 0.98 ± 0.18 μg/mL; deep sedation: 1.88 ± 0.24 μg/mL), reflecting personal differences in sensitivity to the anesthetic. Propofol infusion rates were manually adjusted using STANPUMP to maintain steady sedation, balancing drug accumulation and elimination. Throughout the study, participants were monitored according to ASA standards, including electrocardiogram, blood pressure, pulse oximetry, and end-tidal CO2, with supplemental oxygen provided via nasal cannula. Resting-state data were collected across four 15-min scans, each representing a different condition: baseline consciousness, light sedation, deep sedation, and recovery. Data from baseline consciousness, light sedation, and deep sedation conditions were used. This study utilized a 3T GE Signa 750 scanner. For EPI, the scan settings were as follows: 41 slices, a repetition time/echo time of 2000/25 ms, 3.5 mm slice thickness, a 224 mm field of view, and a 77° flip angle. High-resolution T1 images were acquired as well. Dataset 7: Propofol sedation (2.4 μg/mL doses) Twenty-six healthy participants (13 females, mean age ± SD: 25.0 ± 4.1 years) were included for this study, which was approved by the Institutional Review Board of the University of Michigan, and all participants provided written informed consent.37,46,47 All participants were classified as American Society of Anesthesiologists physical status I, exclusion criteria included any history of drug abuse, psychosis, and other medical conditions. Before the study, participants fasted for eight hours. On the experiment day, a preoperative evaluation, including a physical exam, was conducted by an anesthesiologist. Two anesthesiologists continuously monitored vital signs, including breathing, heart rate, end-tidal CO2, pulse oximetry, and ECG. Noninvasive arterial pressure was recorded using an MR-compatible monitor. Participants received 0.5 mL of 1% lidocaine for local anesthesia before intravenous cannula insertion, and oxygen was delivered at 2 L/min through a nasal cannula. Propofol, chosen for its minimal impact on cerebral blood flow and precise titration capability, was administered via target-controlled bolus and infusion, based on the Marsh pharmacokinetic model using STANPUMP (http://opentci.org/code/stanpump). Dosages increased in 0.4 μg/mL increments until participants showed no behavioral response, with the target concentration maintained for an average of 21.6 ± 10.2 min before recovery. Resting-state data from conscious baseline and deep sedation periods were used in this study. Behavioral responsiveness was assessed by a rubber ball squeeze task, with responses quantified using the BIOPAC MP160 system and AcqKnowledge software. Sixty motor response trials, spaced 90 s apart, were conducted during scanning sessions. Between trials, participants engaged in mental imagery tasks such as imagining playing tennis or navigating a space. Further experimental details are available in prior publications.37,46,47 Four fMRI sessions were conducted as part of the protocol: a 15-min conscious baseline, a 30-min session during propofol infusion, a 30-min session post-propofol infusion, followed by a 15-min recovery baseline. Data from conscious baseline and deep sedation periods were used in this study. This study utilized a 3T Philips scanner. For EPI, the scan settings were as follows: 28 slices, a repetition time/echo time of 800/25 ms (MB factor of 4), 4 mm slice thickness, a 220 mm field of view, and a 76° flip angle. High-resolution T1 images were acquired as well. Dataset 8: Propofol sedation (2.7 μg/mL doses) Thirty healthy participants (20 females, mean age ± SD: 24.4 ± 5.2 years) with complete scan data were included in this study, which was approved by the Institutional Review Board of the University of Michigan, and all participants provided written informed consent.38 All participants were classified as American Society of Anesthesiologists physical status I, exclusion criteria included any history of drug abuse, psychosis, and other medical conditions. Three participants were excluded due to excessive movement and MRI technical issues, leaving 27 participants for analysis. The anesthetic procedure was similar to Dataset-6, with propofol manually adjusted to achieve effect-site concentrations of 1.5, 2.0, 2.5, and 3.0 μg/mL. Each level was held for 4 min to titrate the dosage and determine the threshold for loss of responsiveness (LOR). To minimize head motion artifacts, the concentration was maintained one step higher than the LOR threshold for about 32 min (e.g., if LOR occurred at 2.0 μg/mL, 2.5 μg/mL was maintained). In rare cases, if participants remained responsive at 3.0 μg/mL, the concentration was raised to a maximum of 4.0 μg/mL. The infusion was then stopped, and participants engaged in behavioral tasks, rest, or listened to music. Eight fMRI scans, each lasting 16 min, were conducted over a 2.5-h session. These included baseline scans (Rest1 and Music1), LOR scans (Rest2 and Music2), and recovery scans (Rest3 and Music3). Between each scan, participants had 1–5 min breaks. Resting-state scans required participants to lie still with eyes closed, while music-listening involved tracks from Jazz, Rock, Pop, and Country genres, played in random order. During behavioral testing, participants were prompted to squeeze a rubber ball every 10 s for 96 cycles, following an audio cue delivered through headphones. Grip strength was measured using the BIOPAC MP160 system. Behavioral transitions during propofol administration were identified by missed and completed squeezes, marking the onset and recovery of responsiveness. Only resting-state data from conscious baseline and deep sedation periods were used in this study; music-listening runs were excluded from all analyses. This study utilized a 3T Philips scanner. For EPI, the scan settings were as follows: 40 slices, a repetition time/echo time of 1400/30 ms (MB factor of 4), 2.9 mm slice thickness, a 220 mm field of view, and a 76° flip angle. High-resolution T1 images were acquired as well. fMRI data preprocessing Except for Dataset-1 (LSD), which relied on publicly released preprocessed OpenNEURO data, all datasets were preprocessed using a unified AFNI-based pipeline. The procedure encompassed several steps: First, the initial two frames of each scan were removed to ensure signal stability. This was followed by slice-timing correction to adjust for temporal differences in the acquisition of slices. Second, head motion correction and realignment were performed. Head motion was assessed using frame-wise displacement (FD), calculated as the Euclidean Norm of the six motion parameters. Frames where the FD exceeded 0.5 mm, along with the preceding frame, were excluded from the analysis. For the LSD dataset, the scrubbing threshold reported in the original study (FD > 0.4 mm) was applied. Third, T1 anatomical images were coregistered for precise alignment, followed by spatial normalization into Talairach space48 and resampling to 3 mm isotropic voxels to standardize image coordinates. Fourth, time-censored data underwent band-pass filtering between 0.01 and 0.1Hz using AFNI’s 3dTproject. Simultaneously, linear regression was applied to eliminate unwanted components such as linear and nonlinear drift, the six head-motion parameters and their temporal derivatives (12 regressors total), as well as mean time series from white matter and cerebrospinal fluid. Fifth, spatial smoothing (6mm Gaussian kernel) was performed. Finally, each voxel’s time series was normalized to zero mean and unit variance to ensure data consistency. Global signal regression was not applied in the in-house preprocessing pipeline. Multi-level functional connectivity analysis Functional connectivity (FC) was quantified as the temporal correlation between regional BOLD time series. Using a functional connectivity–based parcellation scheme,49,50 time series were extracted from 400 cortical regions of interest (ROIs), and Pearson correlation coefficients were computed between all pairs of ROIs to generate a 400 × 400 functional connectivity matrix for each participant and condition. At the global level, whole-brain average functional connectivity was computed as the mean of all unique pairwise correlation co-efficients across the full ROI-to-ROI matrix, excluding self-connections. In addition, global within-network and between-network functional connectivity were calculated by averaging correlations among ROIs belonging to the same functional network and between ROIs belonging to different networks, respectively, providing complementary indices of functional segregation and integration. At the functional hierarchy-level, brain regions were grouped into unimodal, attention, and transmodal systems based on canonical network affiliation.50 Unimodal regions comprised visual and somatomotor networks, attention regions comprised dorsal and ventral attention networks, and transmodal regions comprised default mode and frontoparietal control networks. Within-hierarchy functional connectivity was calculated by averaging pairwise correlations among ROIs belonging to the same hierarchical category, whereas between-hierarchy functional connectivity was calculated by averaging correlations between ROIs belonging to different hierarchical categories. At the network level, network-specific functional connectivity was examined by averaging ROI-to-ROI correlations within and between the seven predefined functional networks (visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal, and default mode50). This yielded a 7 × 7 network-level functional connectivity matrix for each participant and condition, in which diagonal elements represent within-network connectivity and off-diagonal elements represent connectivity between specific network pairs. These network-level connectivity estimates were used to characterize how global and hierarchical connectivity changes are distributed across individual network–network interactions. Topological integration analysis Topological integration was quantified using normalized weighted global efficiency. For a weighted network, global efficiency is defined as the average inverse shortest path lengths between all node pairs: Eg=1N(N-1)∑i≠jN1d(i,j) where N is the number of nodes and d(i,j) denotes the shortest path length between nodes i and j. Shortest paths were computed using the Floyd-Washall algorithm (distance_wei_floyd function in Brain Connectivity Toolbox51) with a logarithmic weight-to-distance transformation, where stronger function connections correspond to shorter distance. To control for average connectivity changes, all efficiency results reported in the main text were normalized with the corresponding value from a weight-distribution-preserving random null model generated with randomize_matrix function52): Egnorm=EgEgrand To ensure consistency with the multi-scale functional connectivity analyses described above, topological integration was evaluated across three spatial levels: global, functional hierarchy, and network level. Whole-brain efficiency was computed using all nodes and non-negative edges in the functional connectivity matrix. Within-network or within-hierarchy efficiency was calculated by first generating a smaller functional connectivity matrix composed of nodes within certain functional network or hierarchy, accompanied by the steps described above. Between-network or between-hierarchy efficiency was calculated by (1) constructing a smaller functional connectivity matrix composed of nodes belonging to two networks/hierarchies of interest, (2) obtaining the distance matrix, (3) setting values of within-network or within-hierarchy pairs as NaN, and (4) calculating the average inverse of the remaining values. Random null model generation was done before distance matrix calculation. Global-level within-network and between-network analyses were done by averaging seven within-network and 21 between-network efficiency values, respectively, each obtained by the process described above. Interaction complexity analysis Interaction complexity was used to characterize the effective dimensionality of dynamic interactions between brain regions, independent of static connectivity strength. For each participant and condition, regional time series were segmented into non-overlapping sliding windows of 5 TRs. Within each window, a matrix M representing the interaction of two sets of nodes was constructed by computing the cross-product of the windowed time series X1 and X2 (time × region): M=X1⊤X2 This yields an interaction matrix capturing covariance-like pairwise interactions between regions of two subsets within a short time window. M can be rectangular if the numbers of nodes in set 1 and 2 differ. Self-interaction can also be calculated when set 1 = set 2, yielding a symmetric square matrix. The interaction matrix was then mean-centered by subtracting its global mean. To quantify the complexity of the rectangular matrix M, singular value decomposition (SVD) was applied to each mean-centered interaction matrix: M=UΣV⊤ where Σ is a diagonal matrix containing the singular values sk, which captures the strength of orthogonal interaction modes. The effective dimensionality of network interactions was quantified as the normalized entropy of the singular value spectrum: C=-1log2K∑kpklog2pk,pk=sk∑jsj where K is the number of singular values. Thus, lower C indicates dominance of certain interaction modes (i.e., low-dimensional structure) while higher C indicates a more uniform contribution of all modes. In this sense, reduced interaction complexity reflects a more segregated interaction repertoire, whereby network dynamics become constrained to stereotyped patterns rather than flexibly exploring a diverse range of interaction modes. This procedure yielded a time-resolved complexity estimate C for each sliding window. Complexity values were then averaged across windows to obtain a single interaction complexity estimate per participant and condition. Identical windowing parameters and normalization procedures were applied across datasets and experimental conditions. To ensure consistency with the multi-level framework applied to functional connectivity and topological integration, interaction complexity was evaluated across three spatial levels: global, functional hierarchy, and network-level. Whole-brain complexity was computed by using whole ROIs as set 1 and set 2, yielding a 400 × 400 M. Within-network and within-hierarchy complexity was calculated by using nodes within a given functional network or hierarchy as set 1 and set 2, yielding a smaller symmetric M. Between-network or between-hierarchy complexity was calculated by choosing two networks or hierarchies of interest as set 1 and set 2, yielding a rectangular M. Global-level within-network and between-network complexity was assessed by averaging seven within-network and 21 between-network complexity values, respectively. The short sliding window (5 TRs) was selected to maximize sensitivity to rapid fluctuations in interaction structure while retaining sufficient temporal information for stable estimation. Unlike computing entropy directly on a single time-averaged FC matrix, this time-resolved approach captures moment-to-moment fluctuations in interaction modes. Specifically, each value in a static FC matrix summarizes coupling between two regions across the entire scan into a single number, such that its SVD entropy is primarily sensitive to interaction modes that recur consistently throughout the scan. In contrast, the time-resolved approach distinguishes states in which interaction modes remain diverse and flexible across time (high complexity) from those that are stereotyped and constrained (low complexity). Notably, these two quantities can dissociate: two networks may exhibit few dominant interaction modes at short timescales while still yielding high SVD-entropy of the static FC matrix if those mode patterns shift randomly over time. This measure extends the integration–segregation framework introduced in Jang et al.24 by characterizing the temporal richness of network interactions rather than their static coupling strength. Composite rank of conscious states across analyses For each analytical metric (functional connectivity, topological integration, and interaction complexity), effect sizes (Cohen’s d) of all conscious states relative to their corresponding baseline conditions were computed at the global and functional hierarchy levels, yielding nine analyses (three global and six hierarchical) per metric. For each analysis-metric pair, nine states were ranked according to the effect size value (not the magnitude or absolute Cohen’s d), with higher ranks generally indicating positive effects, while lower ranks often indicate negative effects. To obtain an overall, data-driven ordering of conscious states, a total of 27 rankings were averaged across analysis-metric pairs. Agreement between individual and average rankings To quantify the consistency between individual rankings and the composite ranking, we assessed rank agreement using Kendall’s tau. For 27 analysis-metric pairs, Kendall’s tau with composite ranking was computed, providing a measure of concordance that captures the extent to which each metric and spatial level aligns with the analysis-general ordering of conscious states. Principal component analysis of network-level effects We performed principal component analysis (PCA) on network-level effect size values. For each conscious state, its feature vector was constructed by concatenating the upper triangular entries of a 7 × 7 matrix from three metrics, yielding 84 unique features of nine states. PCA was then applied to this 84 × 9 matrix. The matrix was not centered because zero conveys a meaningful context (i.e., no effect). The first two principal components were used for visualization and interpretation, capturing the dominant axes of variance across states. Loadings of principal components To interpret the contributions of individual network-level features to the dominant axis of variance, we examined the loadings of the first principal component (PC1). PC1 loadings were reorganized and visualized in the form of three upper triangular matrices, revealing which network interaction and metric contributed more strongly to discriminating among conscious states. QUANTIFICATION AND STATISTICAL ANALYSIS All statistical analyses were performed to compare each altered state of consciousness (psychedelic, sleep, and sedation) with its corresponding baseline condition using paired two-sided t-tests. Analyses were conducted separately for each analytical metric—functional connectivity, topological integration, and interaction complexity—and at each spatial scale, including the global level, functional hierarchy level, and network level, as described above. Because baseline measures can differ across datasets due to variations in experimental protocols, acquisition parameters, and preprocessing pipelines, all statistical comparisons were performed relative to each participant’s own baseline within each dataset. This within-subject design minimized the influence of inter-dataset variability and ensured that observed effects reflected condition-specific changes rather than baseline differences across studies. To facilitate quantitative comparison across metrics, spatial scales, and conscious states, effect sizes were computed for all analyses using Cohen’s dz for paired samples. Statistical significance was assessed using Benjamini–Hochberg false discovery rate (FDR) correction to control for multiple comparisons, with significance defined as FDR-adjusted p < 0.05. All statistical tests were two-sided. To test the mirror-image hypothesis across datasets within a unified framework, linear mixed-effects models were additionally fit for each metric × spatial scale combination, as described below. Linear mixed-effects model To test the mirror-image hypothesis within a single unified statistical framework, we fit a linear mixed-effects model (LME) for each metric × spatial scale combination. For each dataset, the dependent variable was the subject-level metric value at baseline and altered state, yielding two observations per subject. Each observation was labeled with a state variable and a binary group variable (psychedelic or diminished states ofconsciousness), with group remaining constant within each dataset. The model was specified as: metric∼state×group+FD_mean+(1+state∣dataset)+(1∣subject_id) where state and group were fixed effects, and FD_mean (mean framewise displacement, computed as the Euclidean norm of the six motion parameters averaged across non-censored frames) was included as a covariate to control for residual head-motion differences between conditions. For non-sleep datasets, state was coded as 0 (baseline) and 1 (altered state). For the sleep dataset, state was treated as an ordinal variable with three levels: 0 (wakefulness), 1 (N1), and 2 (N2), capturing the progressive deepening of sleep within a single dataset. The random-effects structure included a by-dataset random intercept and random slope for state, allowing baseline-to-altered-state trajectories to vary across datasets, and a by-subject random intercept to account for repeated measurements from the same individuals across sleep stages. The state×group interaction term is the primary test of interest: a significant positive coefficient indicates that psychedelic and diminished states of consciousness change in reliably opposite directions across datasets. Data were pooled across eight independent datasets spanning nine altered states (LSD, psilocybin, ketamine, nitrous oxide, Sleep N1, Sleep N2, propofol at 1.9, 2.4, and 2.7 μg mL−1); Sleep N1 and N2 were derived from the same dataset and modeled as ordinal state levels within a unified sleep trajectory. Supplementary Material SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.117830.