{"paper_id":"17ff611f-778c-4f89-8b56-0b5e46bcb0db","body_text":"Convergent Multimodal Evidence of Cortical Excitation-\nInhibition Imbalance in Psychosis \nIoana Varvari MD 1,2,3,a,†; Max Doody BMBCh 1,†; Zilin Li Msc 1; Dominic Oliver Phd 1,3,4; \nPhilip McGuire Phd1,2,3; Matthew M. Nour Phd1,2,5 , Robert A. McCutcheon Phd1,2,3,4,b \n†These authors contributed equally to this work.  \nAuthor affiliations:  \n1. Department of Psychiatry, University of Oxford, Oxford, UK. \n2. Oxford Health NHS Foundation Trust, Oxford, UK. \n3. Oxford Health Biomedical Research Centre, Oxford, UK. \n4. Department of Psychosis Studies, King’s College London, London, UK. \n5. Max Planck UCL Centre for Computational Psychiatry and Ageing Research, London, \nUK. \nCorresponding Author:  \naioana.varvari@psych.ox.ac.uk  \nbrobert.mccutcheon@psych.ox.ac.uk  \n \nOffice address: Warneford Hospital, Warneford Ln, Headington, Oxford, OX3 7JX \n \nRunning Title: Cortical disinhibition in psychosis  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nAbstract  \nPsychosis is increasingly understood as a disorder of disrupted cortical excitation-inhibition \nbalance, yet robust non-invasive translational biomarkers remain lacking. The resting-state \nfMRI Hurst exponent (HE) and EEG aperiodic spectral exponent are promising \ncomplementary biomarkers, with lower values in each proposed to reflect a shift towards \ncortical hyperexcitability, but they have not been jointly examined in psychosis, and the \nspatial and molecular architecture of HE alterations remains poorly defined. We therefore \ntested for convergent systems-level signatures across independent cohorts and modalities, \nusing resting-state fMRI (107 patients, 53 controls) and EEG (547 patients, 363 controls). \nWhole-brain and regional HE were estimated using wavelet methods, and EEG aperiodic \nexponents were quantified using spectral parameterisation. Compared with healthy controls, \nindividuals with psychosis showed reduced whole-brain HE and widespread regional \nreductions. Regional HE case-control differences were associated with cortical gene-\nexpression patterns, with enrichment for potassium channel and GABA receptor pathways, \nand correlated with noradrenergic, muscarinic, serotonergic, glutamatergic and dopaminergic \nreceptor density maps, but not with cortical thickness or symptom or cognitive measures. In \nthe independent EEG cohort, psychosis was similarly associated with a reduced aperiodic \nspectral exponent. Together, these findings provide cross-modal evidence for altered cortical \nresting-state dynamics in psychosis, consistent with a shift towards cortical hyperexcitability. \nIntegration with receptor-density and transcriptomic maps implicates biologically plausible \nmolecular pathways and supports HE and EEG aperiodic activity as scalable translational \nbiomarkers in psychosis. \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nIntroduction \nExcitation - inhibition (E/I) balance is a fundamental property of neuronal circuits and is \nessential for normal brain function. Excitatory (glutamatergic) and inhibitory (GABAergic) \ninputs dynamically interact to regulate neuronal firing and network activity across brain \nstates. GABAergic inhibitory interneurons play a central role in maintaining this balance and \nprevent pathological extremes such as hyperexcitable or silent brain states (1). Disruption of \nE/I balance has been proposed as a core pathophysiological mechanism underlying psychosis, \npotentially accounting for features observed at a range of temporal and spatial scales. \nPerturbations of E/I balance can alter mesoscopic neuronal oscillations detectable using EEG, \nwhich in turn may contribute to large-scale network dysconnectivity that is measurable with \nfMRI (2-6). These disturbances may manifest clinically as positive (3), cognitive (2, 4, 7), \nand negative (8) symptoms of psychosis, which contribute to the functional and social \nimpairments associated with the disorder (9). \nDespite broad recognition of E/I imbalance, most evidence linking microcircuit dysfunction \nto large network disruption is largely derived from ex vivo, in silico, and in vivo animal \nstudies (2-4, 10). Direct measurement of E/I shifts in humans remaining challenging, and \ntranslational evidence in patients with psychosis is comparatively limited. Traditional \nnoninvasive neuroimaging metrics such as EEG power spectra, functional magnetic \nresonance imaging fMRI, magnetic resonance spectroscopy (MRS) metabolites, provide \nindirect, relatively nonspecific proxies with limited validation against microcircuit \nmechanisms (10-12). This gap highlights the need for noninvasive translational biomarkers. \nTwo emerging candidates are the Hurst Exponent (HE) for rs-fMRI (13-16) and the aperiodic \n1/f spectrum for the EEG (17), both predicted by computational microcircuit models and \nempirically linked to changes in E/I shifts across animal and human studies.  The HE \nasks, “Does current brain activity depend on its past?’’ (13) , quantifying the persistence of \ntemporal dependencies, with lower values indicating reduced inhibition and increased cortical \nexcitability (14, 18). The aperiodic 1/f slope captures the broadband spectral structure of \nneural activity, with lower values similarly indicating a shift toward reduced inhibitory drive \n(19). Together, these different metrics capture E/I shifts from complimentary perspectives, \noffering a translational framework for psychosis research. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nTo date, these metrics have not been comprehensively characterised in psychosis. In this \nstudy, we investigated the HE and the aperiodic 1/f spectrum as proxy measures of E/I \nbalance in two independent clinical psychosis datasets. We estimated HE using robust \nwavelet-based methods to capture long-range temporal dependencies, and quantified the \naperiodic 1/f slope (14). Our aims were to test case-control differences in these measures, \ncharacterize their spatial patterns, and examine their associations with clinical features and \nputative biological substrates. \nMaterials and methods  \nPopulation \nfMRI data were obtained from the  Human Connectome Project for Early Psychosis (HCP-\nEP) release 1.1 and EEG data were obtained from the  Bipolar Schizophrenia Network on \nIntermediate Phenotypes-2 (BSNIP2). The HCPEP dataset provided data from 183 patients \nwith psychosis within three years from illness onset and 68 demographically matched healthy \ncontrols, diagnosed with the Structured Clinical Interview for DSM-5 (SCID-5-RV) . The \nBSNIP2 dataset provided data from 582 patients with a diagnosis of longstanding (more than \n5 years from 1\nst admission) schizophrenia or schizoaffective disorder and 379 healthy \ncontrols, diagnosed with the Structured Clinical Interview for DSM-IV (SCID-IV) . Detailed \ninclusion and exclusion criteria are described elsewhere (20, 21). All participants provided \nwritten informed consent under institutional review board approved procedures.  \nData collection \nHCP-EP fMRI data were acquired on a 3T Siemens Connectome Skyra scanner using \nmultiband echo - planar imaging (TR = 0.8 s, TE = 33 ms, flip angle = 52°, voxel = 2 mm, 60 \nslices, multiband factor = 8). Participants completed four 5 min eye open runs. \nPsychopathology was assessed with the Positive and Negative Syndrome Scale (PANSS) and \ncognitive performance was measured using the NIH Toolbox Cognitive Battery. BSNIP2 \nEEG data were recorded using the Compumedics Neuroscan 64-channel Quik-Cap with a \nsampling rate 1000 Hz and no online band-pass filtering. Psychopathology was assessed with \nPANSS, and cognitive functioning with the Brief Assessment of Cognition in Schizophrenia \n(BACS) battery. Detailed data collection protocols are described elsewhere (20, 21). \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nData preprocessing \nfMRI data were preprocessed with fMRIPrep v25.0 (22), then denoised using wavelet \ndespiking (BrainWavelet toolbox v2.0, threshold = 10) (23) and 13 parameter nuisance \nregression (6 motion parameters, their first derivatives, and cerebrospinal fluid signal) \nfollowing prior work validating HE estimation (14). Rs - time series were parcellated into \n360 cortical and 66 subcortical regions, using the HCPex atlas v1.1. (24) EEG data were \npreprocessed with an MNE based pipeline  including average re-referencing, notch and 0.5 - \n40 Hz band-pass filtering, bad-channel interpolation, Independent Component Analysis \n(ICA) based ocular artefacts removal and 2 second epochs segmentation (25). Power spectral \ndensities (PSDs) were averaged per participant to yield subject-level spectra. Aperiodic 1/f \ncomponents were fit using Spectral Parameterization (FOOOF) across 1 - 40 Hz (26). See \neSection 1.1 for detailed preprocessing details. \nData analysis  \nRs-fMRI analyses were implemented in Python v 3.12.  Regional HE  was computed with a \nwavelet-based maximum likelihood estimator (14). A whole-brain HE was derived by \naveraging regional HEs. To minimise site effects, regional HE and whole brain HE were \nharmonized using ComBat, with age, sex, and diagnosis as covariates of interest (27). We \nimplemented automated quality control assessments for quantity of missing data and outliers \nand excluded participants with mean Framewise Displacement (mFD) > 0.25 mm (see \nsupplement for details).  \nGroup differences in whole brain and regional HE were assessed using ordinary least squares \nregression (OLS) adjusted for covariates (HE = \nβ /i1  + β /i1 Phenotype + β /i1 Age + β /i1 Sex + \nβ /i1 FD + ε ). Additionally, phenotype-by-age and phenotype-by-sex interaction models tested \nmoderation effects.  Group difference effect sizes were quantified using Cohen’s d. The \npotential effect of antipsychotic exposure was assessed by comparing whole-brain HE \nbetween medicated and unmedicated patients using an independent-samples t-test, and by \ncalculating the Pearson correlation between chlorpromazine-equivalent dose and whole-brain \nHE. To relate regional HE to PANSS symptoms and cognitive performance, we used 10 fold \ncross validation (CV) partial least squares regression (PLSR) models across multiple \npredictor-outcome configurations (Supplementary Table 1). Cognitive performance was \nsummarised using a composite score derived from the NIH Toolbox and BACS batteries. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nWe then investigated the nature of the regional case-control differences. Parcel-wise effects \nwere examined for enrichment within canonical Yeo 7 networks, to contextualise spatial \npatterns (28). To explore potential biological correlates, regional HE was Pearson correlated \nacross subjects with similarly but independently harmonised cortical thickness (CT). The \nspatial distribution of regional HE effects were then related to gene expression profiles from \nthe Allen Human Brain Atlas (29, 30) employing 10 fold CV PLSR, after preparing the gene \nexpression data via the abagen pipeline (31), followed by ontology enrichment with GOrilla \n(32). Association of the regional HE effects with receptor density distributions were also \nexamined using PLSR and the Hensen Atlas (33). In these analyses, parcel-level CT, gene \nexpression, or receptor density were used as predictors to estimate spatial pattern of the case-\ncontrol difference in HE. Model interpretation and feature contribution were assessed using \nvariable importance in projection (VIP) scores. See eSection 2 for methodological details. \nEEG analyses were implemented in Python (v3.6.8). Following quality control assessments \noutlier or poor-fit spectra were excluded (see eSection 2). For each participant, all valid \nsessions were included. Preprocessing and PSD derivation were performed separately for \neach run, after which a single mean 1/f per participant was calculated. Group-level \ndifferences in 1/f aperiodic exponent were assessed using OLS regression controlling for age, \nsex, offset and site. The 1/f offset is less mechanistically understood than the HE, but can be \ninterpreted as the overall broadband (baseline) level of neuronal activity. Offset effects were \nsimilarly tested i.e. controlling for age, sex, 1/f, and site. \nThroughout, significance was assessed using Benjamini Hochberg False Discovery Rate \n(FDR) correction (q < .05 two tailed) and case-control permutation testing ( n = 10,000). \nDetailed QC and analysis details are provided in eSections 1.2-1.3.  \nResults \nDemographic characteristics \nOf 183 participants in the HCP-EP rs-fMRI cohort, 23 were excluded (8 for unavailable \nimaging, 14 for excessive motion [mFD > 0.25], and 1 outlier > 3 SD), leaving 160 for \nanalysis. Including the outlier showed unchanged results (Supplementary Table 3). In the \nBSNIP2 EEG cohort 63 of 961 participants were excluded (44 poor spectral fits [R² < 0.8 or \nexponent/offset out of range], 7 missing age/site data). Demographics are summarised in \nTable 1.  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nrs-fMRI analyses  \nWhole brain and regional Hurst Exponent group differences \nEarly psychosis participants showed lower whole brain mean rs-fMRI HE values than \ncontrols. Group differences remained significant after adjusting for age, sex, and mean FD ( β  \n= -0.031; 95% CI = [-0.048, -0.015]; Cohen’s  d = -0.68, perm P < 0.001) (Fig. 1A). HE did \nnot differ significantly between patients taking antipsychotic medications and those who were \nmedication free (p=.66), and did not correlate with chlorpromazine equivalent dose ( r\nρ  = \n−0.15, P = 0.2) (Fig. 1B-C). No diagnosis × age (β  = −0.0000; P = 0.89) or diagnosis × sex (β  \n= 0.0016; P = 0.92) interactions were found.  \nThere were also widespread cortical HE reductions, at the regional level in early psychosis \n(223/360 [62%] regions nominally significant, 117 [33%] FDR corrected), which were \nstrongest in somatomotor, insular-opercular, midcingulate, dorsolateral prefrontal, and \nsuperior parietal regions (Fig. 2A). There were additional reductions in, 29/66 [44%] regions \n(22 [33%] FDR corrected), particularly in the thalamus bilaterally (Fig. 2B). The 20 most \nsignificant cortical and subcortical regions are listed in Supplementary Table 2.   \nClinical correlates \nAcross all 10 fold CV PLSR models, there were no statistically significant associations \nbetween HE values and symptoms or cognitive scores (all P > 0.05). \nNetwork enrichment via Y eo 7 functional network atlas \nPsychosis related HE reductions were not uniformly distributed across resting state networks. \nThe Somatomotor Network showed the strongest enrichment ( Δβ  = 0.019; perm P = 0.001, \nFDR q < 0.05). While not statistically significant, overlap with the salience network was also \nnumerically overrepresented ( Δβ  = 0.006; perm  P =  0.12), while there was an \nunderenrichment of the default mode network ( Δβ  = -0.005; perm P = 0.02, FDR q > 0.05) \n(Fig. 2C).  \nExploratory analyses of regional HE and biological substrates  \nWe then examined the relationship between HE and cortical thickness. After additional \nstructural MRI QC, 145/160 participants were included (13 excluded for missing CT; 2 \nexcluded as outliers as values >3SD from group mean). Across 360 cortical parcels, HE \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nshowed widespread reduction, whereas CT reductions were more limited (18 significant; 9 \nFDR corrected). We examined correlations across individuals per parcel, and across parcels. \nNo significant correlation between HE and CT across individuals after FDR correction ( q > \n0.05). Adjusted for covariates, regional HE and CT effects, correlated modestly across \ncortical parcels but did not reach statistical significance when accounting for spatial \nautocorrelation via permutation testing ( r\nρ  =0.149, perm P = 0.13) (Supplementary Fig. 3). \nSensitivity analysis including outliers was unchanged (Supplementary Table 3). \nTo investigate the relationship between HE and putative neuromodulator and neurotransmitter \npathways, we compared the HE spatial distribution to monoaminergic, cholinergic, \nglutamatergic, GABAergic, and endocannabinoid targets from 19 receptor density maps (33). \nA 10 fold CV PLSR model showed that receptor distribution patterns significantly predicted \nregional HE effects. This was statistically significant when assessed against 10,000 random \npermutations of case-control status (r\nρ  = 0.69, perm P = 0.01; Fig. 3A). We next examined the \nmodel, with bootstrap (B = 10,000) derived VIP scores identifying NET, VAChT, 5-HT/i1 , 5-\nHT/i1 A, NMDAR, and D/i1  (VIP > 1) as the strongest contributors (Fig. 3B). \nTo investigate the relationship between HE and gene expression maps, we compared the HE \nspatial distribution with regional gene expression profiles derived from the Allen Human \nBrain Atlas. Using a 10 folds CV PLSR model, we found that cortical gene expression \npatterns predicted cortical HE patterns. This was statistically significant when tested against \n10,000 random permutations of case-control status (r\nρ  = 0.55, perm P = 0.02) (Supplementary \nFig. 5). We next examined the model, with bootstrap (B = 10,000) derived VIP scores \nshowing FDR corrected enrichment ( q < 0.05) for synaptic signaling, ion transport, and \ncentral nervous system development. Enriched cellular components included GABA \nreceptors and voltage-gated sodium channels, while molecular functions were dominated by \npotassium and calcium channel activity, transmembrane transporter and actin biding (Fig. \n3D). Potassium-channel genes showed the strongest enrichment, with  KCNB2 exhibiting the \nstrongest spatial correspondence with HE maps (Fig. 3C), indicating that regions with lower \nKCNB2 expression exhibited larger illness-related differences ( r\nρ  = -0.46, perm P = 0.004; \nSupplementary Fig. 7). \nEEG findings \nIndividuals with chronic psychosis showed a lower 1/f exponent than healthy controls, after \nadjustment for age, sex, site, and offset (β  = -0.101; 95% CI = [-0.160, -0.042]; Cohen’s d = -\n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n0.22; perm P < 0.001; Fig. 4). The reduced spectral exponent in psychosis indicates flatter \naperiodic slopes, consistent with a shift in E/I balance toward excitation. No significant \neffects of age, sex, or medication were detected. We also examined the aperiodic 1/f offset, \nwhich was significantly lower in healthy controls than in chronic psychosis (\nβ  = -0.179; 95% \nCI = [-0.303, -0.054]; perm P  = 0.005; Cohen’s d = -0.18; Supplementary Fig. 8). PLSR \nmodels for symptoms and cognition found similar results as with rs-fMRI (all P > 0.05). \nDiscussion \nIn this multimodal investigation integrating rs-fMRI and EEG across two large, independent \ncohorts, we found convergent evidence for a shift in cortical excitation-inhibition (E/I) \nbalance toward excitation in psychosis. The rs-fMRI data showed that there were widespread \nreductions in the HE in people with Early Psychosis. Complementing this, the EEG data \npointed to a flattening of the aperiodic 1/f spectral slope in people with chronic psychosis, a \npattern similarly associated with reduced inhibitory drive, as previously reported in \nSchizophrenia (34). Together, these results suggest that E/I imbalance is as a systems-level \nfeature of psychosis, detectable across different neuroimaging modalities and illness stages. \nThe rs-fMRI findings revealed widespread but not uniform HE reductions. The strongest \neffects were localized to somatomotor, insular-opercular, midcingulate, dorsolateral \nprefrontal, and parietal cortices - regions central to sensory integration, cognitive control, and \nsalience processing. Thalamic reductions further highlight a potential disruption in \nthalamocortical loops, which are consistently implicated in psychosis (35). This is consistent \nwith emerging work on sensorimotor gating deficits (36), somatomotor network \ndysconnectivity evidence (37), and loss of inhibitory control within thalamo-cortical loops \n(4). Parallel EEG slope flattening supports multimodal convergence.  \nThese results align with a mechanistic understanding of psychosis rooted in cortical \ndisinhibition. Convergent postmortem (38, 39), animal (40), and computational (19, 41) work \nimplicates deficient GABAergic inhibition, particularly within parvalbumin-pozitive \ninterneurons, as a key driver of microcircuit instability. When inhibitory feedback is \ncompromised, pyramidal neurons exhibit increased baseline firing, high-frequency power \nbecomes accentuated, and the temporal persistence of network activity collapses: precisely \nthe features captured by a reduced HE and flattened 1/f slope (2, 3, 42). \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nPotassium channels, which critically shape neuronal resting potentials and spike-frequency \nadaptation, are of particular relevance to psychosis. Dysfunction in these channels can \nincrease baseline excitability while impairing the precision of task-evoked responses. KV3 \nfamily channels are most classically linked to parvalabumin-pozitive interneuron physiology \n(43, 44),  and canonical PV-linked KV3 genes were present in the enriched set, although the \nmost prominent association was with KCNB2 (a KV2-family delayed rectifier enriched in \npyramidal neurons) (45). This finding suggests that HE may capture regional differences in \nintrinsic neuronal electrophysiological properties, including those related to pyramidal cell \nion-channel expression. \nDespite capturing case-control differences, neither HE nor the aperiodic slope robustly \npredicted the severity of symptoms or cognitive impairments. One interpretation is that, \nalthough they index a shared microscale substrate, E/I imbalance represents a stable, trait-like \nneural vulnerability, while behavioural expression depends on secondary modulatory \ninfluences such as dopaminergic tone and network level compensation (46, 47).  \nAlternatively, the mismatch may reflect the distal relationship between microscale neuronal \ndynamics and clinical phenotypes, confounded by limitations of clinical scales, which show \nmodest reliability (48). Futhermore, medication is likely to shift symptom scores whereas HE \nabnormalities were robust to antipsychotic medication status and dosage, thereby potentially \ndisrupting any brain-behaviour correlation.  \nReceptor density mapping further implicated noradrenergic, cholinergic, serotonergic, \nglutamatergic and dopaminergic systems, of which the noradrenergic and cholinergic markers \nwere the most stable This is consistent with evidence that reduced cholinergic tone weakens \ninhibitory control. These findings are timely given renewed interest in muscarinic \nmodulation, supported by the clinical efficacy of the M1/M4 agonist xanomeline-trospium \n(KarXT) in improving psychotic symptoms. Neuroimaging markers of inhibitory tone could \nhelp identify individuals most likely to benefit from such interventions (49). Similarly, the \nidentification of potassium-channel enrichment supports ongoing efforts to develop Kv3 \nagonists as potential treatments for restoring PV interneuron function. \nBeyond mechanistic insights, these findings have clear translational potential. Because HE \nand the aperiodic slope quantify intrinsic, task-free neural dynamics, they are efficient, \nnoninvasive and suitable for longitudinal use. As biomarkers sensitive to inhibitory tone, they \ncould potentially inform early detection and patient stratification. Their applicability across \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nspecies offers a pathway to mechanistically informed, cross-species biomarker development, \nan essential step toward precision psychiatry.  \nThis study was limited by its cross-sectional design, which restricted causal inference and \nprevented modelling of illness-stage effects because time since onset was not available at \nindividual subject level. The HE and aperiodic 1/f spectrum are indirect measures of E/I \nbalance, and further concurrent pharmacological challenge studies would strengthen their \nbiological specificity. Although no relationship was seen with antipsychotic exposure, \nanalyses in medicaton-naive cohorts would be of value. Finally, transcriptomic analyses rely \non postmortem samples from a relatively small sample of individuals that may not generalize \nacross populations. \nFuture studies could validate these markers using multimodal and pharmacological designs. \nCombining HE and aperiodic slope in the same cohorts with perturbation using E/I \nmodulators could clarify their relationship. Longitudinal work following individuals across \nillness stages will determine whether these indices track progression or treatment response. \nLarge, harmonized datasets are also needed to test their reliability and clinical utility as \nbiomarkers of cortical E/I imbalance and treatment response prediction. \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n \nData availability \nThe data analysed in this study were obtained from the Human Connectome Project–Early \nPsychosis (HCP-EP) and the Bipolar and Schizophrenia Network on Intermediate Phenotypes \n2 (BSNIP-2) datasets. These datasets are available through the NIMH Data Archive (NDA) \nand/or dbGaP under controlled access and are subject to data use agreements. Derived data \nsupporting the findings of this study are available from the corresponding author upon \nreasonable request. \nAknowedgments \nResearch was funded by the Wellcome Trust (224625/Z/21/Z), Brain and Behaviour Research \nFoundation (28891), and Academy of Medical Sciences (SGL023\\1009). IV , RAM, PM, and \nDO are supported by the NIHR Oxford Health Biomedical Research Centre. The views \nexpressed are those of the author(s) and not necessarily those of the NIHR or the Department \nof Health and Social Care. \nConflict of Interests  \nIV , ZL have no competing interests to declare. RAM has received speaker/consultancy fees \nfrom Angelini Pharma, Boehringer Ingelheim, Bristol Myers Squibb, Janssen, Karuna, \nLundbeck, Newron, Otsuka, and Viatris, and co-directs a company that designs digital \nresources to support treatment of mental ill health. MMN is a Principal Applied Scientist at \nMicrosoft AI.  \n \n \n \n  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nReferences \n1. Isaacson Jeffry S, Scanziani M. How Inhibition Shapes Cortical Activity. Neuron. \n2011;72(2):231-43. \n2. Yizhar O, Fenno LE, Prigge M, Schneider F, Davidson TJ, O’Shea DJ, et al. \nNeocortical excitation/inhibition balance in information processing and social dysfunction. \nNature. 2011;477(7363):171-8. \n3. Uhlhaas PJ, Singer W . Abnormal neural oscillations and synchrony in schizophrenia. \nNature Reviews Neuroscience. 2010;11(2):100-13. \n4. Anticevic A, Lisman J. How Can Global Alteration of Excitation/Inhibition Balance \nLead to the Local Dysfunctions That Underlie Schizophrenia? Biological Psychiatry. \n2017;81(10):818-20. \n5. Menon V . Large-scale brain networks and psychopathology: a unifying triple network \nmodel. Trends in Cognitive Sciences. 2011;15(10):483-506. \n6. Nour MM, Liu Y , El-Gaby M, McCutcheon RA, Dolan RJ. Cognitive maps and \nschizophrenia. Trends in Cognitive Sciences. 2025;29(2):184-200. \n7. Murray JD, Anticevic A, Gancsos M, Ichinose M, Corlett PR, Krystal JH, et al. \nLinking Microcircuit Dysfunction to Cognitive Impairment: Effects of Disinhibition \nAssociated with Schizophrenia in a Cortical Working Memory Model. Cerebral Cortex. \n2012;24(4):859-72. \n8. Liu Y , Ouyang P, Zheng Y , Mi L, Zhao J, Ning Y , et al. A Selective Review of the \nExcitatory-Inhibitory Imbalance in Schizophrenia: Underlying Biology, Genetics, \nMicrocircuits, and Symptoms. Frontiers in Cell and Developmental Biology. 2021;9. \n9. Handest R, Molstrom I-M, Gram Henriksen M, Hjorthøj C, Nordgaard J. A \nSystematic Review and Meta-Analysis of the Association Between Psychopathology and \nSocial Functioning in Schizophrenia. Schizophrenia Bulletin. 2023;49(6):1470-85. \n10. McCutcheon RA, Keefe RSE, McGuire PK. Cognitive impairment in schizophrenia: \naetiology, pathophysiology, and treatment. Molecular Psychiatry. 2023;28(5):1902-18. \n11. Ahmad J, Ellis C, Leech R, V oytek B, Garces P, Jones E, et al. From mechanisms to \nmarkers: novel noninvasive EEG proxy markers of the neural excitation and inhibition \nsystem in humans. Translational Psychiatry. 2022;12(1). \n12. Cohen Kadosh R. Rethinking excitation/inhibition balance in the human brain. Nature \nReviews Neuroscience. 2025;26(8):451-2. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n13. Hardstone R, Poil S-S, Schiavone G, Jansen R, Nikulin VV , Mansvelder HD, et al. \nDetrended Fluctuation Analysis: A Scale-Free View on Neuronal Oscillations. Frontiers in \nPhysiology. 2012;3. \n14. Trakoshis S, Martínez-Cañada P, Rocchi F, Canella C, You W, Chakrabarti B, et al. \nIntrinsic excitation-inhibition imbalance affects medial prefrontal cortex differently in autistic \nmen versus women. eLife. 2020;9. \n15. Nishio M, Ellwood-Lowe ME, Woodburn M, McDermott CL, Park A T, Tooley UA, et \nal. The Development of Neural Inhibition across Species: Insights from the Hurst Exponent. \nThe Journal of Neuroscience. 2025;45(41). \n16. Xie K, Royer J, Rodriguez\n/i1 Cruces R, Horwood L, Ngo A, Arafat T, et al. Temporal \nLobe Epilepsy Perturbs the Brain/i1 Wide Excitation/i1 Inhibition Balance: Associations with \nMicrocircuit Organization, Clinical Parameters, and Cognitive Dysfunction. Advanced \nScience. 2025;12(9). \n17. V oytek B, Knight RT. Dynamic Network Communication as a Unifying Neural Basis \nfor Cognition, Development, Aging, and Disease. Biological Psychiatry. 2015;77(12):1089-\n97. \n18. Nishio M, Ellwood-Lowe ME, Woodburn M, McDermott CL, Park A T, Tooley UA, et \nal. 2024. \n19. Gao R, Peterson EJ, V oytek B. Inferring synaptic excitation/inhibition balance from \nfield potentials. NeuroImage. 2017;158:70-8. \n20. Glasser MF, Smith SM, Marcus DS, Andersson JLR, Auerbach EJ, Behrens TEJ, et al. \nThe Human Connectome Project's neuroimaging approach. Nature Neuroscience. \n2016;19(9):1175-87. \n21. Tamminga CA, Ivleva EI, Keshavan MS, Pearlson GD, Clementz BA, Witte B, et al. \nClinical Phenotypes of Psychosis in the Bipolar-Schizophrenia Network on Intermediate \nPhenotypes (B-SNIP). American Journal of Psychiatry. 2013;170(11):1263-74. \n22. Esteban O, Markiewicz CJ, Blair RW, Moodie CA, Isik AI, Erramuzpe A, et al. \nfMRIPrep: a robust preprocessing pipeline for functional MRI. Nature Methods. \n2018;16(1):111-6. \n23. Patel AX, Bullmore ET. A wavelet-based estimator of the degrees of freedom in \ndenoised fMRI time series for probabilistic testing of functional connectivity and brain \ngraphs. NeuroImage. 2016;142:14-26. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n24. Huang C-C, Rolls ET, Feng J, Lin C-P. An extended Human Connectome Project \nmultimodal parcellation atlas of the human cortex and subcortical areas. Brain Structure and \nFunction. 2021;227(3):763-78. \n25. Gramfort A, Luessi M, Larson E, Engemann DA, Strohmeier D, Brodbeck C, et al. \nMNE software for processing MEG and EEG data. NeuroImage. 2014;86:446-60. \n26. Donoghue T, Haller M, Peterson EJ, Varma P, Sebastian P, Gao R, et al. \nParameterizing neural power spectra into periodic and aperiodic components. Nature \nNeuroscience. 2020;23(12):1655-65. \n27. Fortin J-P, Cullen N, Sheline YI, Taylor WD, Aselcioglu I, Cook PA, et al. \nHarmonization of cortical thickness measurements across scanners and sites. NeuroImage. \n2018;167:104-20. \n28. Thomas Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, \net al. The organization of the human cerebral cortex estimated by intrinsic functional \nconnectivity. Journal of Neurophysiology. 2011;106(3):1125-65. \n29. Shen EH, Overly CC, Jones AR. The Allen Human Brain Atlas. Trends in \nNeurosciences. 2012;35(12):711-4. \n30. Hawrylycz MJ, Lein ES, Guillozet-Bongaarts AL, Shen EH, Ng L, Miller JA, et al. \nAn anatomically comprehensive atlas of the adult human brain transcriptome. Nature. \n2012;489(7416):391-9. \n31. Markello RD, Arnatkevi\nč iū tė  A, Poline J-B, Fulcher BD, Fornito A, Misic B. 2021. \n32. Eden E, Navon R, Steinfeld I, Lipson D, Yakhini Z. GOrilla: a tool for discovery and \nvisualization of enriched GO terms in ranked gene lists. BMC Bioinformatics. 2009;10(1). \n33. Hansen JY , Shafiei G, Markello RD, Smart K, Cox SML, Nørgaard M, et al. Mapping \nneurotransmitter systems to the structural and functional organization of the human \nneocortex. Nature Neuroscience. 2022;25(11):1569-81. \n34. Hasanaj G, Jaeger I, Karsli B, Schulz E, Boudriot E, Roell L, et al. Periodic and \nAperiodic Alterations of Resting\n/i1 State EEG in Schizophrenia Spectrum Disorders: \nCognitive and Clinical Insights. European Journal of Neuroscience. 2025;62(7). \n35. Anticevic A, Halassa MM. The thalamus in psychosis spectrum disorder. Frontiers in \nNeuroscience. 2023;17. \n36. San-Martin R, Castro LA, Menezes PR, Fraga FJ, Simões PW, Salum C. Meta-\nAnalysis of Sensorimotor Gating Deficits in Patients With Schizophrenia Evaluated by \nPrepulse Inhibition Test. Schizophrenia Bulletin. 2020;46(6):1482-97. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n37. Keyvanfard F, Schmid A-K, Nasiraei-Moghaddam A. Functional Connectivity \nAlterations of Within and Between Networks in Schizophrenia: A Retrospective Study. Basic \nand Clinical Neuroscience Journal. 2023;14(3):397-410. \n38. Lewis DA. Cortical circuit dysfunction and cognitive deficits in schizophrenia – \nimplications for preemptive interventions. European Journal of Neuroscience. \n2012;35(12):1871-8. \n39. Kaar SJ, Angelescu I, Marques TR, Howes OD. Pre-frontal parvalbumin interneurons \nin schizophrenia: a meta-analysis of post-mortem studies. Journal of Neural Transmission. \n2019;126(12):1637-51. \n40. Sohal VS, Zhang F, Yizhar O, Deisseroth K. Parvalbumin neurons and gamma \nrhythms enhance cortical circuit performance. Nature. 2009;459(7247):698-702. \n41. Gonzalez-Burgos G, Cho RY , Lewis DA. Alterations in Cortical Network Oscillations \nand Parvalbumin Neurons in Schizophrenia. Biological Psychiatry. 2015;77(12):1031-40. \n42. Bartos M, Vida I, Jonas P. Synaptic mechanisms of synchronized gamma oscillations \nin inhibitory interneuron networks. Nature Reviews Neuroscience. 2007;8(1):45-56. \n43. Kaar SJ, Nottage JF, Angelescu I, Marques TR, Howes OD. Gamma Oscillations and \nPotassium Channel Modulation in Schizophrenia: Targeting GABAergic Dysfunction. \nClinical EEG and Neuroscience. 2023;55(2):203-13. \n44. Faulkner IE, Pajak RZ, Harte MK, Glazier JD, Hager R. V oltage-gated potassium \nchannels as a potential therapeutic target for the treatment of neurological and psychiatric \ndisorders. Frontiers in Cellular Neuroscience. 2024;18. \n45. Mohamed ZA, Li J, Wen J, Jia F, Banerjee S. The KCNB2 gene and its role in \nneurodevelopmental disorders: Implications for genetics and therapeutic advances. Clinica \nChimica Acta. 2025;566. \n46. Howes O, McCutcheon R, Stone J. Glutamate and dopamine in schizophrenia: An \nupdate for the 21stcentury. Journal of Psychopharmacology. 2015;29(2):97-115. \n47. McCutcheon RA, Krystal JH, Howes OD. Dopamine and glutamate in schizophrenia: \nbiology, symptoms and treatment. World Psychiatry. 2020;19(1):15-33. \n48. Geck S, Roithmeier M, Bühner M, Wehr S, Weigel L, Priller J, et al. COSMIN \nsystematic review and meta-analysis of the measurement properties of the Positive and \nNegative Syndrome Scale (PANSS). eClinicalMedicine. 2025;82. \n49. Kaul I, Sawchak S, Walling DP, Tamminga CA, Breier A, Zhu H, et al. Efficacy and \nSafety of Xanomeline-Trospium Chloride in Schizophrenia. JAMA Psychiatry. 2024;81(8)\n. \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nFigure legends  \nFigure 1 Whole Brain Group Differences and Medication effects on HE. (A) Whole Brain \nHE (±95% CI) for healthy controls (HC; n = 53) and first-episode psychosis patients \n(FEP; n = 107). (B) Whole Brain HE in unmedicated ( n = 28) and medicated ( n = 79) \npatients, shown as box plots with median and interquartile range. (C) Association between \nchlorpromazine (CPZ) equivalent dose residuals and Whole Brain HE residuals among \nmedicated patients (n = 70). Pearson correlations are displayed, with a fitted linear regression \nline. Hurst residuals represent individual Whole Brain HE values adjusted for covariates (age, \nsex, and head motion) using linear regression.  \nFigure 2 Spatial Nature of Illness Effects . (A) Cortical OLS β  maps showing group \ndifferences in regional Hurst exponent between healthy controls (HC) and first-episode \npsychosis patients (FEP). (B) Subcortical OLS β  maps for the same contrast, displayed on \ncoronal and axial slices. For A and B, warm colours indicate regions where the HE is lower in \nFEP than HC, reflecting increased excitation. (C) Yeo -7 network enrichment analysis \nshowing the weighted difference between network-specific \nβ  values and the global mean. \nPoints represent observed effects with 95% confidence intervals from 10.000 permutations; \nfilled blue circles indicate networks with permutation P < 0.05. \nβ  values represent OLS effect \nestimates for HC - FEP contrasts. Maps are thresholded using the Benjamini - Hochberg false \ndiscovery rate (FDR) correction at q < 0.05, which was applied on permutation p results. \nFigure 3 Biological Substrates of Illness Effects . (A) Observed versus predicted HE ( β ) \nfrom the primary PLSR model based on receptor expression; each point represents out-of-\nfold predictions from 10-fold cross-validation ( n = 360). The solid line shows the best-fit \nregression, and the dashed line indicates the identity line (y = x). (B) Receptors with mean \nVIP values > 1 were considered to contribute above average to the model. Confidence \nintervals (CI) illustrate receptor importance stability estimates across 10,000 bootstrap \nresampling; (C) KCNB2 expression (left cortex). AHBA expression mapped to HCPex and z-\nscored across parcels; red = relatively higher, blue = relatively lower, 0 = parcel-mean \nexpression. (D) Gene sets surviving FDR corrected ( q  < 0.05) gene enrichment analyses. \nBars show enrichment scores across gene categories, with colour indicating FDR corrected \nsignificance. \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n \nFigure 4 Whole Brain Group Differences and Medication effects on 1/f Aperiodic \nExponent.  (A) 1/f Aperiodic Exponent (±95% CI) for healthy controls (HC; n = 363) and \nschizophrenia patients (SZ; n = 547). (B) 1/f Aperiodic Exponent in unmedicated ( n = 109) \nand medicated (n = 438) patients, shown as box plots with median and interquartile range. C) \nAssociation between chlorpromazine (CPZ) equivalent dose residuals and 1/f Aperiodic \nExponent residuals among medicated patients ( n = 402). Pearson correlations are displayed, \nwith a fitted linear regression line. 1/f Aperiodic Exponent residuals represent individual 1/f \nvalues adjusted for covariates (age, sex, and head motion) using linear regression.  \n  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nTables and Figures \nCharacteristic \nHCP - \nEP \nPatients \n(n=107) \nHCP - EP \nControls  \n(n=53) \np \nValue \nBSNIP \nPatients \n(n=547) \nBSNIP \nControls \n(n=363) \np \nValue \nAge \nMean years \n(SD) \n22.6 \n(3.6) \n24.7 (4.2) 0.004 40.0 (11.6) 34.1 (12.1) \n< \n0.001 \nMale sex  \nN (%) \n66 \n(61.7%) \n34 (64.2%) 0.9 \n296 \n(54.1%) \n142 \n(39.1%) \n< \n0.001 \nPANSS total \nmean (SD)  \n49.78 \n(11.06) \nN/A N/A \n 66.88 \n(21.14) \n   N/A    N/A \nCognition (NIH \ntoolbox \ncomposite) \nmean (SD) \n101.21 \n(12.95) \n113.22 \n(7.88) \n0.001 N/A N/A  N/A \nCognition \n(BACS) \nMean (SD) \nN/A N/A N/A \n46.9 \n(6.54) \n54.2 \n(5.38)   \n< \n0.001  \nMedicated  \nN (%) \n79 \n(73.8%) \nN/A N/A \n438 \n(80.1%) \nN/A N/A \n \nTable 1. Legend: SD = standard deviation, no = number, PANSS = Positive and Negative \nSymptoms Scale, N/A= not applicable, cognition total = standardised scored of the total NIH \ntoolbox cognitive battery in HCPEP and BACS battery in BSNIP2. The p values were \ncalculated using independent-samples t-tests for continuous variables (age, cognition) and \nchi-square tests for categorical variables (sex, ethnicity). For ethnicity, the p values were \ncalculated between proportion of white and ethnical minorities between cases and controls. \n  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n \nFigure 1 \n  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\nFigure 2 \n  \n \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n \nFigure 3 \n  \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint \n\n \nFigure 4 \n.CC-BY 4.0 International licenseavailable under a \nwas not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprint (whichthis version posted April 6, 2026. ; https://doi.org/10.64898/2026.03.31.715583doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}