Abstract

   Disrupted genes linked to mental disorders sometimes exhibit
   characteristics of Intrinsically Disordered Proteins (IDPs). However,
   few studies have comprehensively explored the functional associations
   between protein disorder properties and different psychiatric
   disorders. In this study, we collected disrupted proteins for seven
   mental diseases (MDD, SCZ, BP, ID, AD, ADHD, ASD) and a control dataset
   from normal brains. After calculating the disorder scores for each
   protein, we thoroughly compared the proportions and functions of IDPs
   between differentially expressed proteins in each disease and healthy
   controls.

   Our findings revealed that disrupted proteins, particularly in ASD and
   ADHD, contain more IDPs than controls from normal brains. Distinct
   patterns in disorder properties were observed among different mental
   disorders. Functional enrichment analysis indicated that IDPs in mental
   disorders were associated with neurodevelopment, synaptic signaling,
   and gene expression regulatory pathways. In addition, we analyzed the
   proportion and function of liquid-phase-separated proteins (LLPS) in
   psychiatric disorders, finding that LLPS proteins are mainly enriched
   in pathways related to neurodevelopment and inter-synaptic signaling.
   Furthermore, to validate our findings, we conducted an analysis of
   differentially expressed genes in an ASD cohort, revealing that the
   encoded proteins also exhibit a higher proportion of IDPs. Notably,
   these IDPs were particularly enriched in pathways related to
   neurodevelopment, including head development, a process known to be
   disrupted in ASD.

   Our study sheds light on the crucial role of IDPs in psychiatric
   disorders, enhancing our understanding of their molecular mechanisms.

   Keywords: Psychiatric disorders, Intrinsically disordered proteins
   (IDPs), Protein function, Liquid-liquid phase separation (LLPS)

Graphical Abstract

   ga1
   [39]Open in a new tab

1. Introduction

   Over the past two decades, the prevalence of mental disorders has risen
   rapidly, along with an increasing burden on healthcare systems and
   society [40][1]. Due to the complexity of their pathogenesis and the
   presence of unknown etiological factors, effective therapies for mental
   disorders are rarely available. Although genetic variation assessments,
   such as Genome-Wide Association Studies (GWAS), have identified
   numerous genes associated with mental disorders, the underlying
   molecular mechanisms are still not fully understood [41][2], [42][3],
   [43][4].

   In recent years, significant progress has been made in exploring the
   functional properties of intrinsically disordered proteins (IDPs) and
   unraveling their crucial role in mental disorders [44][4], [45][5],
   [46][6].

   Intrinsically disordered regions (IDRs) are biologically active protein
   regions that exhibit high conformational variability but lack stable
   three-dimensional structures [47][7], [48][8], [49][9]. IDPs, which may
   contain IDRs of varying lengths or be completely disordered, play key
   roles in numerous biological processes, such as signaling pathways,
   transcription, translation, and the cell cycle [50][10], [51][11],
   [52][12], [53][13], [54][14]. The precise regulation of abundant IDPs
   in cells ensures the accuracy of signaling pathways. Mutations or
   changes in IDPs could lead to multiple diseases [55][15], [56][16],
   [57][17].

   IDPs play a crucial role in brain development. For example, ZSWIM8, a
   ubiquitin ligase, is essential for the development of the embryonic
   nervous system [58][18]. Its function is attributed to the presence of
   abundant IDRs. These IDRs in ZSWIM8 and Dab1 interact with each other,
   correcting misfolding through the "disorder targets misorder" mechanism
   [59][19]. ZSWIM8 also promotes the proper phosphorylation of Dab1,
   thereby maintaining its regulatory function in the Reelin signaling
   pathway, which is implicated in several neuropsychiatric disorders,
   including Autism Spectrum Disorders (ASD), Schizophrenia (SCZ), Bipolar
   Disorder (BP), Major Depressive Disorder (MDD), and Alzheimer’s disease
   [60][18], [61][20].

   Mutations in IDPs have been identified as major contributors to protein
   aggregation in the brain and are associated with plaque formation in
   patients with neurodegenerative disorders. Well-known IDPs, such as
   alpha-synuclein, amyloid-beta peptide, and Huntington's protein, are
   implicated in the pathogenesis of diseases like Alzheimer's disease,
   Parkinson's disease, and Huntington's disease [62][21], [63][22],
   [64][23], [65][24], [66][25], [67][26], [68][27].

   Several studies have already demonstrated the functional importance of
   disorder properties by investigating one or a few genes in mental
   diseases [69][4], [70][28]. However, the comprehensive analysis of the
   functional roles of IDPs in psychiatric disorders has rarely been
   performed [71][29]. With the advancement of experimental methods, along
   with the rising number of diagnosed psychiatric cases, there is an
   urgent need to conduct in-depth analyses of the disorder properties of
   IDPs and their involvement in the pathophysiology of major psychiatric
   disorders. Understanding these molecular mechanisms will provide
   insights into the development of more targeted therapeutic strategies.

   For this purpose, we meticulously compiled proteins associated with
   seven psychiatric disorders—Anxiety Disorder (AD), Attention Deficit
   Hyperactivity Disorder (ADHD), Autism Spectrum Disorder (ASD), Bipolar
   Disorder (BP), Major Depressive Disorder (MDD), Intellectual Disability
   (ID), and Schizophrenia (SCZ)—using stringent criteria. The resulting
   database includes differentially expressed proteins from seven disease
   datasets and one control dataset (BRAIN). We then calculated the
   disorder properties of these proteins using the state-of-the-art
   disorder prediction platform, RIDAO [72][30].

   We conducted a comprehensive comparison of the disorder properties
   between the disease datasets and the controls. Using protein
   interaction analysis, Gene Ontology (GO), and the Kyoto Encyclopedia of
   Genes and Genomes (KEGG), we explored the functional roles of IDPs in
   psychiatric disorders. Additionally, we compared the proportions and
   functions of liquid-liquid phase-separated (LLPS) proteins between
   psychiatric disorders and control data. Finally, we validated our
   findings by analyzing differentially expressed gene data from an ASD
   cohort study.

   We found a higher prevalence of intrinsically disordered regions (IDRs)
   in ASD and ADHD compared to normal human brain proteins (p-value <
   0.05). There are distinct IDR patterns across different psychiatric
   disorder datasets. IDPs associated with psychiatric disorders were
   significantly enriched in pathways related to neuron projection
   development, head development, cell morphogenesis, brain development,
   and synaptic transmission regulation. Additionally, our LLPS analysis
   revealed that ASD and ADHD datasets contained more LLPS proteins than
   controls, with these proteins also enriched in neurodevelopment and
   synaptic signaling pathways. A similar pattern was observed in an ASD
   cohort study, where proteins exhibited higher IDP proportions and
   greater involvement in head development and inter-synaptic signaling.

   Our research aims to explore the neurobiological connections between
   IDPs and mental diseases, providing new insights into the pathogenesis
   and treatment of psychiatric disorders.

2. Materials and methods

2.1. Datasets

   To create a dataset related to mental disorders, we searched the
   DisGeNET [73][31] and GeneCards [74][32] databases using the following
   keywords: ‘anxiety disorder’, ‘attention deficit/hyperactivity
   disorder’, ‘autism spectrum disorder’, ‘bipolar disorder’,
   ‘intellectual disability’, ‘major depressive disorder’, and
   ‘schizophrenia’. Genes meeting the following criteria are included: (i)
   gene-disease association score > 0.1 in the DisGeNET database; (ii)
   gene score > 1 in the GeneCards database.

   The above step selected 2388 differentially expressed genes associated
   with mental disorders. We focus on genes expressed in the brain, given
   the primary involvement of the brain in the occurrence of psychiatric
   disorders. To ensure this, we applied the following criteria: (i) genes
   expressed in either the Cerebellum, Cerebral Cortex, Hippocampus, or
   Lateral Ventricle tissue according to the Human Protein Atlas [75][33];
   (ii) genes expressed in ≥ 75 % of sample slices in at least 3 of 6
   brain donors from the Allen Human Brain Atlas [76][34]; or (iii) genes
   with expression RPKM (reads per kilobase of transcript per million
   mapped reads) values > 1 in ≥ 75 % of samples from at least one
   developmental stage according to the Allen Human Brain Atlas [77][34].
   [78]Fig. 1 illustrates the gene selection pipeline for psychiatric
   diseases.

Fig. 1.

   [79]Fig. 1
   [80]Open in a new tab

   The data collection processes. This figure shows how we collect the
   disease dataset.

   We mapped the selected genes to their corresponding proteins using
   UniProtKB [81][35], resulting in a mental disorder-related protein
   dataset containing 2189 non-redundant proteins, referred to as the
   MENTAL dataset.

   For comparison, we also constructed a control dataset using
   Brain-Specific Proteome data from the Human Protein Atlas [82][33]. The
   BRAIN dataset contains genes expressed in normal brains, selected based
   on the following criteria: (i) Genes with minimal expression signals in
   the human brain; (ii) Genes with immunohistochemical evidence of brain
   expression; (iii) Genes that can be mapped to proteins by the UniProt
   database; (iv) Genes not found in the MENTAL dataset. In the Human
   Protein Atlas, proteins are categorized based on their relative
   expression levels into three groups: (i) Elevated in brain； (ii) Low
   tissue specificity but expressed in brain； and (iii) Elevated in other
   tissues but expressed in brain. To ensure a fair comparison, we
   randomly selected 2188 non-redundant proteins from the Brain-Specific
   Proteome data, matching the distribution of expression categories in
   the MENTAL dataset. Finally, the MENTAL dataset comprised 2189
   proteins, while the BRAIN dataset contained 2188 proteins.

2.2. Intrinsic disorder prediction

   We used the state-of-the-art RIDAO platform [83][30] to predict the
   disorder properties of each protein in the datasets. RIDAO is a fast
   and efficient tool that integrates six well-established disorder
   prediction tools into a unified platform. Residues with scores above
   0.5 are classified as disordered, while scores below 0.5 indicate
   ordered. In this study, protein regions with a continuous length of
   ≥ 30 disordered residues were classified as intrinsically disordered
   regions (IDRs), and proteins containing at least one IDR were
   classified as intrinsically disordered proteins (IDPs) [84][29]. Based
   on the presence or absence of IDRs, we categorized mental
   disease-related proteins into two groups: 'Low Disorder' and 'High
   Disorder'.

2.3. Binding sites prediction

   The ANCHOR software was used to identify the binding regions [85][36],
   [86][37]. ANCHOR uses pairwise energy estimation methods to predict
   disordered binding regions by identifying fragments within disordered
   regions that interact with globular protein partners to gain
   stabilizing energy. Disordered binding sites were predicted for each
   protein in the seven disease datasets and the BRAIN dataset.

2.4. Protein-protein interaction (PPI) network construction

   We used the UniProt ID as input, set ‘Protein Query’ as the data
   source, and applied a confidence score threshold of ‘0.90′ to completed
   the protein-protein interaction (PPI) networks using the STRING plugin.
   The construction of network edges is based on known interactions in the
   database, including those from curated databases, experimentally
   validated interactions, and predicted interactions such as gene
   neighborhoods, gene fusions, and gene co-occurrence. Additional factors
   such as text mining, gene co-expression, and protein homology are also
   considered. The network nodes represent the input proteins, with
   non-interacting nodes excluded.

   We applied the degree algorithm from the CytoHubba plugin to identify
   the top 50 pivotal proteins from each dataset. These proteins were then
   used to construct interaction networks through the STRING plugin. To
   capture a broader range of interactions, we lowered the confidence
   score threshold to 0.40.

2.5. Protein functional annotation

   We applied Metascape [87][38] to perform GO and KEGG analysis on the
   genes in each dataset, using the following parameters: a minimum
   overlap of 3, a p-value cutoff of 0.05, and a minimum enrichment of 3.
   We identified the top 5 most significantly enriched pathways for each
   dataset in terms of Biological Process, Molecular Function, Cellular
   Component. For the KEGG analysis, we determined the top 10 pathways
   with the most significant gene enrichment in each dataset. R version
   4.2.3 was used to visualize these observations.

2.6. Prediction of LLPS-prone proteins

   We predicted the LLPS scores of psychiatric disease-associated proteins
   using three state-of-the-art machine learning and deep learning
   predictors: PSPHunter [88][39], PSPredictor [89][40], and PredLLPS_PSSM
   [90][41], [91][42]. The final LLPS prediction score was obtained by
   averaging the scores from these three predictors. We applied a
   threshold of 0.5, based on the criteria of all three predictors, to
   classify proteins as either LLPS or non-LLPS. Proteins with scores
   equal to or above this threshold were considered as LLPS proteins.

2.7. A ASD cohort containing 1141 patients

   We selected data from a cohort study that included 1141 patients with
   ASD [92][43], [93][44]. From this study, we identified 391 genes with
   statistical significance (FDR < 0.1) ([94]Supplementary Table 1). These
   genes were mapped to their corresponding proteins using UniProtKB
   [95][35], and we filtered the results to retain only proteins with a
   'Reviewed' status and a protein existence level of 'Protein level'. For
   duplicate mappings meeting these criteria, we retained the protein with
   the longest sequence. The disorder properties of the screened proteins
   were assessed using the RIDAO platform [96][30] and IDPs and ordered
   proteins were categorized based on the criteria described in
   [97]Section 2.2. For the identified IDPs, we performed GO functional
   enrichment analysis using MetaScape [98][45] and listed the top 5 terms
   across biological processes, cellular components, and molecular
   functions.

2.8. Statistical analysis

   The statistical analysis was conducted using R version 4.2.3. Normality
   tests and tests for variance homogeneity were performed on each
   dataset, revealing that none of the datasets followed a normal
   distribution and that variances were not equal across groups.
   Therefore, protein sequence lengths between groups were compared using
   the Kruskal-Wallis Test, proportional comparisons were assessed using
   the chi-squared test, and pairwise comparisons were adjusted using the
   Bonferroni correction. The correlation between the proportion of
   disordered residues in disrupted proteins and their LLPS scores was
   assessed using the Pearson correlation coefficient.

3. Results & discussion

3.1. Acquisition and characterization of differentially expressed genes in
mental disorders

   As described in the methods section, we created two datasets: the
   MENTAL dataset, consisting of disease-related proteins, and the BRAIN
   dataset, serving as the control. The disease dataset includes seven
   subsets: the anxiety disorders (AD) dataset: 68 proteins; the attention
   deficit hyperactivity disorder (ADHD) dataset: 307 proteins; the autism
   spectrum disorders (ASD) dataset: 143 proteins; the bipolar disorder
   (BP) dataset: 602 proteins; the major depression (MDD) dataset: 512
   proteins; the intellectual developmental disorders (ID) dataset: 812
   proteins; the schizophrenia (SCZ) dataset: 939 proteins. The list of
   these proteins is in [99]Supplementary Table 1.

   We then conducted functional enrichment analysis to examine the roles
   of the differentially expressed genes in the MENTAL dataset. We
   observed that proteins associated with psychiatric disease were
   significant enrichment in processes related to behavior (GO:0007610),
   head development (GO:0060322), neural projection development
   (GO:0031175), trans-synaptic signaling regulation (GO:0099177), and
   brain development (GO:0007420). These processes are closely related to
   nervous system formation and function, as well as intraneuronal
   communication and neurotransmitter release. Furthermore, proteins in
   MENTAL dataset are strongly related to transcription, signal
   transduction, protein-protein interactions, and gene regulatory
   mechanisms. They play crucial roles in dendritic, axonal, postsynaptic
   and presynaptic synapses. These findings are consistent with previous
   studies [100][46], [101][47], [102][48]. KEGG pathway enrichment
   analysis was also performed on this dataset, with the detailed results
   shown in [103]Figs. 2B and [104]2C.

Fig. 2.

   [105]Fig. 2
   [106]Open in a new tab

   Differential Gene Function Enrichment and Protein Interaction Network
   Analysis in MENTAL and BRAIN datasets. (A) Top 5 Biological Process,
   Cellular Component, and Molecular Function terms of genes in the BRAIN
   dataset. (B) Top 5 Biological Process, Cellular Component, and
   Molecular Function terms of genes in the MENTAL dataset. (C) Top 10
   KEGG pathways in the MENTAL dataset were ranked by gene enrichment
   significance. (D) Protein-protein interaction networks of the top 100
   proteins in the protein interaction network of the MENTAL dataset
   calculated by the degree algorithm. The inner circle lists the top 10
   proteins in clockwise order.

   We also constructed a protein interaction network of protein in the
   MENTAL dataset to obtain the potential hub proteins and their
   interactions. The key hub proteins identified included AKT1, TP53, and
   ACTB. [107]Fig. 2D depicts the comprehensive data of the protein
   interaction network associated with mental disorders.

3.2. Comparison of protein intrinsic disorder properties between MENTAL
diseases and controls

   We used the RIDAO platform [108][30] to predict protein disorder
   properties of proteins in the MENTAL and the BRAIN dataset. The
   resulting mean disorder profile (MDP) score were used to calculate the
   percentage of IDPs in each dataset. [109]Table 1 presents the details
   of the top 15 proteins with the highest percentage of protein disorder
   in the MENTAL dataset.

Table 1.

   Top 15 proteins with the highest percentage of disordered residues in
   the MENTAL dataset.
   Protein name Protein ID Disease Protein length (aa) Amino acid residues
   predicted to be disordered, aa (%) Longest disordered region (aa)
   Function
   Stathmin STMN1_HUMAN MDD/ MENTAL 149 149(100) 149 tubulin binding
   Neuromodulin NEUM_HUMAN BP/ MDD/ MENTAL/ SCZ 238 238(100) 238
   calmodulin binding
   Complexin−2 CPLX2_HUMAN ADHD/ BP/ MDD/
   MENTAL/ SCZ 134 134(100) 134 calcium-dependent protein binding
   Neurogranin NEUG_HUMAN BP/ MENTAL/ SCZ 78 78(100) 78 calmodulin binding
   PRKC apoptosis WT1 regulator protein PAWR_HUMAN BP/ MENTAL/ SCZ 340
   340(100) 340 actin binding
   Protein phosphatase 1 regulatory subunit 1B PPR1B_HUMAN BP/ MDD/MENTAL/
   SCZ 204 204(100) 204 protein kinase inhibitor activity
   Complexin−1 CPLX1_HUMAN BP/ ID/ MDD/
   MENTAL/ SCZ 134 133(99.25) 123 SNARE binding
   Transcription factor SOX−2 SOX2_HUMAN ID/ MENTAL 317 306(96.53) 154 DNA
   binding
   Protein PRRC2A PRC2A_HUMAN MENTAL/ SCZ 2157 2082(96.52) 1856 RNA
   binding
   Neurosecretory protein VGF VGF_HUMAN BP/ MDD/ MENTAL/
   SCZ 615 592(96.26) 345 growth factor activity
   B-cell CLL/lymphoma 9 protein BCL9_HUMAN BP/ MENTAL/ SCZ 1426
   1372(96.21) 943 beta-catenin binding
   Transcription factor 12 HTF4_HUMAN ID/ MENTAL 682 640(93.84) 575 bHLH
   transcription factor binding
   Microtubule-associated protein 6 MAP6_HUMAN MENTAL/ SCZ 813 760(93.48)
   500 calmodulin binding
   Methyl-CpG-binding protein 2 MECP2_HUMAN AD/ ADHD/ ASD/ BP/ ID/ MDD/
   MENTAL/ SCZ 486 452(93.00) 333 chromatin binding
   Charged multivesicular body protein 2b CHM2B_HUMAN MDD/ MENTAL 213
   196(92.02) 153 cadherin binding
   [110]Open in a new tab

   AD, Anxiety Disorder; ADHD, Attention Deficit Hyperactivity Disorder;
   ASD, Autism Spectrum Disorder; BP, Bipolar Disorder; MDD, Major
   Depressive Disorder; ID, Intellectual Disability; SCZ, Schizophrenia

   The proportion of IDPs in each dataset was as follows: ASD (75.40 %),
   ADHD (73.40 %), ID (66.70 %), BP (65.60 %), MENTAL (65.70 %), SCZ
   (65.30 %), BRAIN (66.20 %), MDD (64.20 %) and AD (56.50 %). As shown in
   [111]Table 2, we observed that the percentage of IDPs in the ASD and
   ADHD dataset was higher than that in the BRAIN dataset (P-value < 0.05,
   [112]Fig. 3B).

Table 2.

   Comparison of disorder properties in different mental diseases.
   Protein length
     __________________________________________________________________

   Intrinsic disorder estimates
     __________________________________________________________________

   Name N Mean SE Maximum Median Overall disordered Proteins with long
   regions of
   amino acids (%) disorder (%)
   BRAIN 2188 575.8 10.9 4870 441 32.08 % 66.20 %
   AD 68 695.8 72.8 2768 479.5[113]^e 33.87 %[114]^e 56.50 %
   ADHD 307 995.1 54.3 8797 710.0[115]^e 33.89 %[116]^e 73.40 %[117]^e
   ASD 143 1033 68.3 4967 708.0[118]^e 33.25 %[119]^e 75.40 %[120]^e
   BP 602 753.1 29.2 8797 560.5[121]^e 29.77 %[122]^e 65.60 %
   MDD 512 687.8 28.6 6306 485.5[123]^e 27.85 %[124]^e 64.20 %
   ID 812 921.1 30.2 8797 634.5[125]^e 30.84 %[126]^e 66.70 %
   SCZ 939 754.2 23.2 8797 541.0[127]^e 30.31 %[128]^e 65.30 %
   P-value[129]^a P < 0.001[130]^c P < 0.001[131]^d P < 0.001[132]^d
   MENTAL 2189 791.9 15.6 8797 559.0[133]^e 30.43 %[134]^e 65.70 %
   P-value[135]^b P < 0.001[136]^c P < 0.001[137]^d P > 0.05
   [138]Open in a new tab
   ^a

   P-value for disease-specific comparison with BRAIN dataset
   ^b

   P-value of MENTAL compared to BRAIN dataset
   ^c

   Kruskal–Wallis Test.
   ^d

   chi-square test
   ^e

   Values different from BRAIN dataset

Fig. 3.

   [139]Fig. 3
   [140]Open in a new tab

   Characterization and analysis of differentially expressed genes in
   psychiatric disorders. (A) Petal plots of the seven psychiatric disease
   datasets, with the number in each region representing the number of
   genes included in the corresponding disease intersection. (B)
   Percentage of proteins containing disordered residues of each length in
   the seven disease datasets. (C) Percentage of proteins containing at
   least one disordered binding site in the seven disease datasets and the
   control dataset. (D) Percentage of disordered residues of each length
   in the seven disease datasets relative to the total number of
   disordered residues.

3.3. Binding sites prediction in MENTAL diseases and controls

   We employed ANCHOR to predict disordered binding sites within the
   protein sequences of each dataset. Proteins with at least one
   disordered binding site were selected for analysis. We found that the
   ASD and ADHD datasets contained significantly more proteins with
   disordered binding sites compared to the BRAIN dataset (Bonferroni
   p − value < 0.05). The number of disordered binding sites in the ASD
   and ADHD was significantly higher than that in the MDD dataset
   (Bonferroni p − value < 0.05).

   Proteins in the ASD and ADHD datasets exhibited a higher tendency to
   bind and interact with other proteins compared to those in the BRAIN
   and MDD datasets. This trend was closely associated with a higher
   prevalence of IDPs in the ASD and ADHD datasets. A summary of the
   disordered binding sites for each dataset can be found in
   [141]Supplementary Table 2 and [142]Fig. 3C.

3.4. Comparison of enrichment analysis between 'Low Disorder' and the 'High
Disorder' proteins

   After comparing the proportions of IDPs between mental disorders and
   healthy controls, we further investigated the potential functional
   pathways associated with IDPs in psychiatric diseases. To simplify the
   analysis, we categorized the proteins into two groups: 'Low Disorder'
   and 'High Disorder', based on the length of their disordered regions.

   We conducted GO and KEGG functional enrichment analyses on both the Low
   Disorder and High Disorder datasets. We identified and listed the top 5
   items for Biological Process, Cellular Component, and Molecular
   Function in GO, as well as the top 10 pathways in KEGG. This analysis
   aims to further validate the pivotal role of disordered proteins in
   psychiatric disorders and explore their key functions.

3.4.1. Gene ontology functional analysis

   In our study, we observed significant differences in the molecular
   functions of these two classes of proteins. Low Disorder proteins are
   implicated in molecular transmembrane transport, protein-protein
   interactions, and oxidoreductase activity. In contrast, High Disorder
   proteins exhibit a strong preference for intermolecular interactions,
   particularly those involving proteins and DNA. Most of these
   interactions are related to signaling, gene expression, and
   transcriptional regulation.

   In terms of biological processes, High Disorder proteins are more
   likely to be involved in neuron projection development, head
   development, cell morphogenesis, brain development, and the modulation
   of chemical synaptic transmission compared to Low Disorder proteins.
   Both categories of proteins play crucial roles in synaptic signaling.
   Additionally, these proteins are important components of the synapse in
   both Low Disorder and High Disorder groups, underscoring the critical
   role of synapses in mental disorders.

   We performed functional enrichment of the ASD and ADHD datasets, which
   have significantly more IDPs compared to the BRAIN dataset. The results
   showed that the molecular functions of IDPs in ASD involved chromatin
   binding, DNA binding, DNA-binding transcription factors binding, etc.,
   which were closely related to the regulation of gene expression, while
   the molecular functions of IDPs in ADHD were mainly related to the
   activities of various ion transmembrane transport proteins. In terms of
   biological processes, IDPs in ASD are associated with a variety of
   biological processes including learning and memory, chemical synaptic
   regulation, behavior, head development, and cognition, while IDPs in
   ADHD are mainly involved in the regulation of membrane potential, the
   regulation of ionic transmembrane transporter, and synaptic signaling,
   etc., which correspond to the molecular functions of IDPs in ADHD. The
   details of the GO analysis were shown in [143]Fig. 4.

Fig. 4.

   [144]Fig. 4
   [145]Open in a new tab

   GO functional enrichment of disease genes in different subgroups. (A)
   Top 5 Biological Process, Cellular Component, and Molecular Function
   terms for genes in the Low Disorder dataset. (B) Top 5 Biological
   Process, Cellular Component, and Molecular Function terms for genes in
   the High Disorder dataset. (C) Top 5 Biological Process, Cellular
   Component, and Molecular Function terms for genes in the IDPs in ASD.
   (D) Top 5 Biological Process, Cellular Component, and Molecular
   Function terms for genes in the IDPs in ADHD.

3.4.2. KEGG pathway enrichment analysis

   The KEGG enrichment analysis showed no significant differences between
   Low Disorder and High Disorder proteins, nor in comparison to the KEGG
   pathways enriched in the MENTAL dataset. Both groups were similarly
   enriched in pathways associated with neurodegenerative diseases,
   cancer, neuroactive ligand-receptor interactions, as well as signaling
   pathways such as PI3K-Akt, MAPK, and cAMP. The top 10 KEGG pathways
   enriched for Low Disorder and High Disorder proteins are illustrated in
   [146]Figs. 5C and [147]5D.

Fig. 5.

   [148]Fig. 5
   [149]Open in a new tab

   Kyoto Encyclopedia of Genes and Genomes (KEGG) and Protein-Protein
   Interaction (PPI) analysis in the Low Disorder and the High Disorder
   datasets. (A) Protein-protein interaction networks of the top 50
   proteins in the LOW Disorder dataset, sorted in descending order by
   their degree scores. (B) Protein-protein interaction network among the
   top 50 proteins in the High Disorder dataset. (C) Top 10 KEGG pathways
   in the Low Disorder dataset were sorted by the number of enriched
   genes. (D) Top 10 KEGG pathways in the High Disorder dataset.

3.5. Comparison of PPI network between the 'Low Disorder' and the 'High
Disorder' proteins

   It has been reported that IDPs are important in protein interactions
   due to their high flexibility. To investigate whether IDPs could affect
   the occurrence of psychiatric disorders through protein-protein
   interactions, we constructed protein interaction networks for proteins
   in Low Disorder and High Disorder datasets, identifying key proteins in
   the protein interaction networks.

   The Low Disorder Protein Interaction Network consists of 619 nodes and
   1455 edges. We filtered the top 50 nodes using the degree algorithm,
   resulting in a reconstructed network of 50 nodes and 970 edges, with an
   average of 19.4 edges per node. The top three important proteins
   identified are ACTB, AKT1, and KRAS.

   Similarly, the High Disorder Protein Interaction Network contains 605
   nodes and 1707 edges. The reconstructed network includes 50 nodes and
   1426 edges, averaging 28.5 edges per node. Key proteins here are TP53,
   CREB1, and CREBBP.

   We briefly describe these key node proteins, especially those in the
   High Disorder Protein Interaction Network. The first one is KRAS, which
   is a significant protein in the Low Disorder network, mediates the
   RAS/MAPK pathway, influencing neuronal growth, proliferation, and
   central nervous system development [150][49], [151][50]. In the High
   Disorder network, TP53 encodes the P53 protein, an IDP with
   intrinsically disordered regions crucial for its multifunctionality.
   The C-terminal domain of P53 interacts with GSK3, a protein linked to
   schizophrenia [152][51], [153][52], [154][53].

   CREB1 and CREBBP are also vital IDPs in the High Disorder network.
   CREB1 encodes a transcription factor, while CREBBP encodes the
   CREB-binding protein, central to eukaryotic transcriptional regulation.
   The pKID domain of CREB interacts with CBP's KIX domain to trigger gene
   expression [155][54], [156][55]. A meta-analysis found higher
   expression levels of CREB1 and CREBBP in the Brodmann Area 10 samples
   from schizophrenia patients compared to controls, and downregulation of
   CREB1 is associated with MDD, SCZ, and BP [157][28]. [158]Figs. 5A and
   [159]5B illustrate the protein interaction networks for both datasets.

3.6. Results of LLPS prediction

   We used the average prediction scores from three LLPS predictors to
   classify psychiatric disease-related proteins and control proteins into
   LLPS and non-LLPS proteins. The proportions of LLPS proteins in each
   dataset were as follows: ASD (66.20 %), ADHD (59.54 %), ID (45.01 %),
   BP (51.17 %), MENTAL (50.59 %), SCZ (50.59 %), BRAIN (47.41 %), MDD
   (51.91 %), and AD (38.28 %). We compared the LLPS proportions between
   psychiatric disease-related proteins and control proteins. The results
   indicated significant differences (P < 0.05), with ASD and ADHD showing
   significantly higher proportions of LLPS proteins compared to the
   control dataset (P < 0.05), while ID had a significantly lower
   proportion of LLPS proteins (P < 0.05) ([160]Table 3).

Table 3.

   Comparison of LLPS protein percentage in different psychiatric
   disorders.
   Name N Mean SE Maximum Median LLPS protein percentage（%）
   BRAIN 2188 575.8 10.9 4870 441 47.41
   AD 68 695.8 72.8 2768 479.5[161]^e 38.25
   ADHD 307 995.1 54.3 8797 710.0[162]^e 59.54[163]^e
   ASD 143 1033 68.3 4967 708.0[164]^e 66.20[165]^e
   BP 602 753.1 29.2 8797 560.5[166]^e 51.17
   MDD 512 687.8 28.6 6306 485.5[167]^e 51.91
   ID 812 921.1 30.2 8797 634.5[168]^e 45.01[169]^e
   SCZ 939 754.2 23.2 8797 541.0[170]^e 50.59
   P-value[171]^a P < 0.001[172]^c P < 0.001[173]^d
   MENTAL 2189 791.9 15.6 8797 559.0[174]^e 50.59
   P-value[175]^b P < 0.001[176]^c P < 0.05[177]^d
   [178]Open in a new tab
   ^a

   P-value for disease-specific comparison with BRAIN dataset
   ^b

   P-value of MENTAL compared to BRAIN dataset
   ^c

   Kruskal–Wallis Test.
   ^d

   chi-square test
   ^e

   Values different from BRAIN dataset

3.7. Proportion of IDPs for ASD differential proteins in a cohort study

   To validate our findings using real cohort data, we selected a cohort
   of 1141 individuals with ASD. We analyzed 391 statistically significant
   genes identified in this cohort (FDR < 0.1) ([179]Supplementary Table
   1). We found that the proteins encoded by these genes had a higher
   proportion of IDPs (76.68 %), which closely aligns with the proportion
   of IDPs in our ASD dataset (75.40 %) and was significantly higher than
   in the BRAIN dataset (P < 0.05) ([180]Supplementary Table 4).
   Furthermore, functional analysis of these IDPs revealed significant
   enrichment in pathways such as regulation of trans-synaptic signaling,
   modulation of chemical synaptic transmission, head development,
   chromatin remodeling, and cell morphogenesis ([181]Supplementary Figure
   1). These pathways closely resemble those observed in the “High
   Disorder” proteins and LLPS proteins in psychiatric disorders, with
   head development, modulation of chemical synaptic transmission, and
   cell morphogenesis being significantly enriched in all three protein
   types. These results suggest that the differentially expressed proteins
   in the ASD cohort contain a higher proportion of IDPs, consistent with
   our findings, and further emphasize the critical role of IDPs in head
   development and synaptic signaling in ASD.

4. Discussion

   Despite the extensive research efforts over the past few decades to
   uncover the pathophysiology of mental diseases [182][3], [183][54],
   [184][55], [185][56], the precise molecular mechanisms of psychiatric
   disorders remain unclear. Previous studies have shown that changes in
   protein disorder properties may be associated with the development of a
   variety of diseases [186][15], [187][17], [188][57]. Due to the
   diversity and complexity of IDPs, current research on psychiatric
   disorders still focuses on neurotransmitters [189][58], neuronal
   pathways [190][59], and gene mutations [191][56], and few studies have
   explored the role of IDPs in psychiatric disorders. Our study is the
   first to systematically explore functional roles of IDPs in psychiatric
   disorders.

   This paper analyzes the differences in the proportions and functions of
   IDPs across seven psychiatric disorders and normal brain controls. We
   found that psychiatric disorder datasets contained more IDPs than the
   control dataset, suggesting a potential association between IDPs and
   psychiatric disorders, particularly in ASD and ADHD. Previous studies
   indicate that ASD and ADHD share certain common neurological
   characteristics [192][60], [193][61], with their underlying
   neurobiological pathways showing remarkable similarities [194][62],
   [195][63], [196][64]. However, our further analysis revealed functional
   differences in IDPs between ASD and ADHD, which may be linked to their
   distinct pathogenic mechanisms. Continued research on IDPs in ASD and
   ADHD will help clarify these differences and provide insights for the
   selection of clinical treatments.

   Among the 15 most disordered proteins in our dataset, PPR1B encodes for
   protein phosphatase 1 regulatory subunit 1B, a regulator of kinase or
   phosphatase associated with glutamate and dopamine receptor activation.
   As a target of dopamine, PPR1B is of great relevance to the treatment
   of neurological and mental disorders [197][65], [198][66], [199][67],
   [200][68]. Furthermore, we observed that most IDPs were associated with
   pathways or structures involving chromatin, nuclear membrane,
   cytoskeleton, neurons, and other structures. This finding is consistent
   with the established role of IDPs in synaptic and neuronal processes
   [201][69], [202][70]. We also found that some IDPs are proven targets
   of psychiatric drugs, providing insights for further elucidating of the
   mechanism of action of drugs and discovering new psychiatric drug
   targets. These IDPs drug targets are listed in [203]Supplementary Table
   3.

   In addition, our results showed that the proportion of IDPs and LLPS in
   ASD- and ADHD-related proteins was significantly higher than in the
   control dataset. Further comparative analyses revealed a strong
   correlation between the proportion of disordered residues in disrupted
   proteins and their LLPS scores (r = 0.71, [204]Fig. 6A). Consistent
   with the functional enrichment results of 'High Disorder' proteins
   ([205]Fig. 4B), LLPS proteins were also significantly enriched in
   pathways related to neuron projection development, head development,
   cell morphogenesis, brain development, and the modulation of chemical
   synaptic transmission ([206]Fig. 6B). It has been demonstrated that the
   phase separation ability of LLPS is closely linked to its IDR region
   [207][25], [208][71], and truncation of the IDR can disrupt the
   protein's phase separation function [209][71], [210][72], [211][73],
   [212][74]. Therefore, we hypothesize that mutations in IDPs may
   contribute to psychiatric disorders by affecting liquid-liquid phase
   separation.

Fig. 6.

   [213]Fig. 6
   [214]Open in a new tab

   Correlation analysis between LLPS Scores and Disordered Residues (%)
   and functional enrichment of LLPS in mental disease-related proteins.
   (A) Correlation analysis between the Disordered Residues (%) in
   psychiatric proteins and LLPS Scores of psychiatric proteins. (B) GO
   functional enrichment of LLPS proteins associated with psychiatric
   disorders.

   Our study has certain limitations. Although databases such as DisGeNET
   and GeneCards provide rich gene-protein and disease-gene likelihood
   data, there is still a possibility of missing newly identified genes
   linked to mental disorders. Furthermore, changes in gene expression
   levels were not considered in the diseases we studied because the
   databases used did not have information on gene expression levels.

   Future work could investigate the aggregation properties of IDRs in
   psychiatric disease-associated proteins. This analysis would include
   parameters such as radius of gyration, end-to-end distance, polymer
   scaling index, and aggregate sphericity. We could also explore the
   impact of conformational changes of IDRs in mental disease-associated
   proteins on the function of full-length proteins. Furthermore, we could
   investigate how these conformational changes of IDRs in mental diseases
   are linked to cellular function, localization, amino acid sequences,
   evolutionary conservation, and disease variation, together with a newly
   released molecular model for generating IDR conformational assemblies
   [215][75], [216][76].

   In summary, our study contributes to exploring the neurobiological
   mechanisms underlying psychiatric disorders, providing a novel
   perspective from the standpoint of protein disorders. We discuss the
   specific roles and mechanisms of IDPs in psychiatric disorders.

CRediT authorship contribution statement

   Qingzhen Hou: Writing – review & editing, Writing – original draft,
   Methodology, Funding acquisition, Formal analysis, Conceptualization.
   Sankar Basu: Writing – review & editing, Methodology. Dongdong Qiao:
   Writing – review & editing, Writing – original draft, Methodology, Data
   curation. Na Zhou: Validation, Supervision, Methodology. Yaqing Yang:
   Writing – original draft, Formal analysis, Data curation. Ruotong Liu:
   Writing – original draft, Validation, Supervision, Conceptualization.
   Xixi Song: Writing – review & editing, Writing – original draft,
   Visualization, Conceptualization. Guangchun Hu: Writing – review &
   editing, Writing – original draft, Resources, Formal analysis, Data
   curation. Xinwu Zhang: Writing – review & editing, Writing – original
   draft, Methodology, Formal analysis, Data curation, Conceptualization.

Declaration of Competing Interest

   The authors declare that there are no conflicts of interest regarding
   the publication of this paper.

Acknowledgments