2026-01-27

106_ Integrative transcriptomic regulation in Schizophrenia

Research Question and Aims

Schizophrenia is a multifaceted neuropsychiatric disease characterized by significant gene expression changes in both the central nervous system and peripheral tissues. Although the majority of research has concentrated on protein-coding mRNAs, recent findings suggest that non-coding RNAs, such as long non-coding RNAs (lncRNAs), circular RNAs (circRNAs), microRNAs (miRNAs), and tRNA-derived fragments (tRFs), play essential roles as regulators of the gene networks altered in schizophrenia^1-3. LncRNAs and circRNAs are part of the majority of the human transcriptome and can regulate expression at multiple levels; i.e., they can function as competing endogenous RNAs (ceRNAs) to sequester miRNAs, which in turn affect mRNA levels. These non-coding RNAs often regulate fundamental biological processes such as the immune response, synapse function, and neurodevelopment, which are crucial for understanding the pathophysiology of schizophrenia^2,4-6.
Recent articles have shown the function of integrated transcriptome regulation in psychiatric disorders. A blood small RNA-seq identified mir-99b-5p as a significant regulator of schizophrenia: it is downregulated in the blood and brains of patients, and its suppression in a mice model triggered microglial inflammation, leading to synaptic pruning and disease-like phenotypes. A regulatory connection exists between an immune pathway (microglial activation via the miR-99b target Zbp1) and synaptic dysfunction due to non-coding RNAs, facilitating the integration of pertinent biological themes (immune/inflammatory response in the brain linking neuroinflammation and synaptic connectivity)¹. Similarly, studies on lncRNA have uncovered dysregulation in immune and developmental processes: for instance, the IFNG-AS1 lncRNA is diminished in the blood of schizophrenia patients and correlates with interferon-γ levels⁷, while other blood lncRNAs (e.g., AC006129.1 and RP5-998N21.4) have been identified as modulators of pro- and anti-inflammatory cytokine pathways through epigenetic mechanisms⁸. Many circRNAs are downregulated in the brains of individuals with schizophrenia, and their host genes are significantly associated with neurodevelopmental activities, such as neurogenesis, neuronal differentiation, and dendritic morphology^9,10. Certain circRNAs (e.g. circHomer1a) seem to modulate synaptic genes through interaction with RNA-binding proteins and are correspondingly reduced in patient neurons and cortices6. Significantly, alterations in non-coding RNAs are also observable peripherally: several circRNAs and lncRNAs are differentially expressed in the blood of schizophrenia patients^11-13, indicating that blood transcriptome profiles may represent facets of the condition and potentially serve as accessible biomarkers. In this perspective, a thorough integrative analysis of coding and non-coding transcripts in schizophrenia may reveal the coordinated regulatory networks that underlie the disorder. We possess a distinctive possibility to utilize paired total RNA sequencing (encompassing mRNAs, lncRNAs, and circRNAs) alongside small RNA sequencing (including miRNAs and tRFs) from 175 schizophrenia patients and 176 matched controls whose totalRNA at baseline (visit 1) was sequenced under the auspices of the DFG “Sequencing Cost in Projects” grant entitled “Longitudinal smallRNAome changes in the transdiagnostic PsyCourse Study”. By combining these datasets, we may transcend discrete differential expression lists to formulate co-expression and co-regulatory networks that elucidate the interactions among lncRNAs, circRNAs, and mRNAs with small RNAs, leading to the dysregulation of pathways in schizophrenia. We will specifically concentrate on competing endogenous RNA (ceRNA) interactions, wherein lncRNAs or circRNAs sequester miRNAs to regulate target mRNAs, and on uncovering common pathway regulation that may connect diverse RNA classes. This methodology is corroborated by initial findings in similar contexts: for instance, integrative network analyses in blood have effectively identified disease-specific transcriptomic modules (e.g., immune and ncRNA-processing gene modules that differentiate subtypes of depression¹⁴), and recent studies on schizophrenia indicate that the construction of lncRNA-miRNA-mRNA networks can reveal disease signatures and prospective biomarkers¹⁵. An integrative transcriptome study is poised to offer new insights into the molecular foundations of schizophrenia and identify potential regulatory RNAs that may serve as treatment targets or blood-based biomarkers.
Central aim: To elucidate the integrative transcriptomic regulation of schizophrenia by analyzing how lncRNAs, circRNAs, and mRNAs (from total RNA-seq) are coordinately regulated with miRNAs and tRFs (from small RNA-seq) in blood. We will construct co-expression and ceRNA networks that highlight key regulatory interactions and pathways disrupted in schizophrenia. To achieve this, we propose the following specific aims:
Aim 1: Characterize differential expression of coding and non-coding RNAs in Schizophrenia. We will quantify all expressed mRNAs, lncRNAs, circRNAs (≥200 nt, including back-spliced circRNAs), and small RNAs (miRNAs, tRFs ~15-40 nt) in 175 patients vs 176 controls. Differential expression analysis (patients vs controls) will identify dysregulated transcripts in each RNA category (with adjustment for covariates like age, sex, batch). This provides a foundation of candidate schizophrenia-associated RNAs (including novel or unannotated lncRNAs/circRNAs and tRFs) for network analyses.
Aim 2: Construct co-expression networks linking total RNA and small RNA signatures. Using weighted gene co-expression network analysis (WGCNA) and related methods, we will build co-expression networks to find modules of highly correlated transcripts. This will be done for the total RNA-seq data (clustering mRNAs together with lncRNAs and circRNAs that show similar expression patterns) and for the small RNAs (miRNAs and tRFs). We will then integrate these by correlating module "eigengenes" across data types or by combining datasets in a multi-layer network, to identify points of coordinated regulation, for example, associations between a particular miRNA/tRF module and a mRNA/lncRNA module. This analysis will highlight groups of transcripts deregulated in schizophrenia, which often represent underlying biological processes or cell-type programs.
Aim 3: Identify ceRNA networks and regulatory interactions (lncRNA/circRNA-miRNA/tRF-mRNA). We will explicitly map potential regulatory interactions between small and long RNAs. This involves in silico target prediction for miRNAs and tRFs to identify which mRNAs and lncRNAs/circRNAs harbor binding sites for each small RNA. By overlaying this with our expression data, we will construct a ceRNA network: linking lncRNAs or circRNAs with mRNAs if they share common miRNA/tRF regulators. We will prioritize competing endogenous RNA interactions where a non-coding RNA and an mRNA are positively co-expressed and both are anti-correlated with a shared miRNA, consistent with a sponge/target relationship. Such network analysis will allow us to discover putative regulatory triads (e.g. a schizophrenia-upregulated lncRNA sequestering a miRNA, leading to regulation of immune genes). We aim to identify key hub nodes in these ceRNA networks, for example, a miRNA that regulates many mRNAs in a module, or a lncRNA/circRNA that sponges multiple miRNAs as master regulators of transcriptomic dysregulation.
Aim 4: Pathway enrichment and biological interpretation of integrated networks. For each co-expression or ceRNA module identified, we will perform functional enrichment analysis (Gene Ontology, KEGG pathways, etc.) to determine which biological processes are over-represented. We specifically expect to see modules related to immune/inflammatory responses, synaptic function and neuroplasticity, and neurodevelopmental processes, based on prior schizophrenia findings. For example, immune response gene modules (e.g. cytokine response pathways) may be co-regulated with immunoregulatory lncRNAs or miRNAs and genes associated with synaptic plasticity may overlap with neuron-associated lncRNAs or circRNAs coexpressed in cohorts. Enrichment analysis will allow confirmation that the networks we detect overlap with known disease-associated themes (e.g. altered immune response, synaptic deficit) while also directing potential novel pathways (e.g. metabolic, stress response) highlighted by our data. Thus, these analyses will generate hypotheses for downstream testing regarding biological implications of such coordinated RNA alterations in schizophrenia.
Through these investigations, the project will establish a unified network model of transcriptomic regulation in schizophrenia that applies a systems biological approach to understanding disease-relevant dysregulation with coding and non-coding RNA interactions.

Analytic Plan

We will employ a fully computational analytical pipeline to integrate the paired total RNA-seq and small RNA-seq datasets. All analysis will be conducted in a high-performance computing environment using state-of-the-art bioinformatics tools and statistical methods, ensuring reproducibility through scripted workflows. An overview of the approach is as follows:
Data processing and quality control: Raw sequencing reads from PAXgene blood RNA will undergo quality trimming and adapter removal (e.g. using FastQC and Trim Galore for both paired-end total RNA reads and small RNA reads). For total RNA-seq, reads will be aligned to the human genome (e.g. with STAR or HISAT2) and expression quantified for known genes/transcripts using a tool like featureCounts. This will yield counts for mRNAs and annotated lncRNAs. In parallel, we will detect circRNAs by leveraging reads spanning back-splice junctions (using specialized algorithms such as CIRI or find_circ). For small RNA-seq, we will align reads to miRNA reference sequences (e.g. using miRDeep2 to identify known and novel miRNAs) and also map reads to tRNA databases (GtRNAdb) to identify tRNA-derived fragments (tRFs). A tool like MINTmap will be used for precise tRF annotation. Rigorous QC will be implemented at each step (acceptable read quality, mapping rates and library complexity, expression data will be appropriately normalized (i.e. DESeq2 variance stabilizing transform or TPM for visualization) and batch effects in RNA-seq data will be corrected when necessary.
Differential expression: For each category of RNA, we will compare diagnosis (schizophrenia vs. control) on a group basis. Using statistical models (DESeq2 or limma-voom), differentially expressed mRNAs, lncRNAs, circRNAs, miRNAs and tRFs will be obtained (adjusted for covariates of age, sex, RIN and technical batch). Transcripts with false discovery rate (FDR) below a significance threshold (e.g. q < 0.05) will be considered significantly dysregulated in schizophrenia. This will produce an initial catalog of candidate RNAs of interest for example, we anticipate on the order of hundreds of mRNAs and lncRNAs showing altered expression, and dozens of miRNAs and tRFs, based on sample size and prior studies. These results provide independent validation of the contribution of each RNA class (e.g. confirming that many miRNA and tRFs are indeed differently expressed, as suggested by our prelimanary analysis) and guide inclusion of the most variable/dysregulated transcripts in network analyses.
Co-expression network construction (WGCNA): To uncover groups of RNAs that function in concert, we will construct weighted co-expression networks using WGCNA or similar algorithms. First, for the total RNA-seq data, we will use the expression matrix of all reliably detected mRNAs, lncRNAs, and circRNAs (filtered for sufficient counts/variance) to create a correlation network. The adjacency will be defined by pairwise expression correlations between genes across the samples, and a soft thresholding power will be chosen to approximate scale-free topology. Modules of highly co-expressed transcripts will be identified via hierarchical clustering and dynamic tree cutting. Each module is summarized by its eigengene (first principal component of expression). Similarly, we will build a co-expression network for the small RNAs (miRNAs and tRFs), identifying modules of co-regulated small RNAs, an approach successfully applied in our prior small RNA study1. The integration comes by relating these two networks: we will calculate correlations between each small-RNA module eigengene and each mRNA/lncRNA module eigengene to find cross-modal associations. For example, we might find that a module of pro-inflammatory genes (highly co-expressed mRNAs and lncRNAs) correlates strongly with a module of downregulated miRNAs/tRF that target those genes. Significant module-module correlations (p < 0.05) will indicate putative regulatory links between the long and small RNA datasets. Additionally, we may augment this by constructing a unified network including all RNA types: e.g., using approaches like MOFA or by incorporating miRNA/tRF as additional nodes connected to their predicted targets. This would directly yield mixed modules containing both small and long RNAs. The network analysis will also identify hub genes within each module which are likely key drivers of the module's function.
ceRNA network and correlation analysis: On the basis of the co-expression results, we will then directly validate ceRNA interactions. miRNA target prediction databases (TarBase, miRDB, TargetScan, etc.) will provide candidate predicted miRNA-mRNA and miRNA-lncRNA/circRNA interactions. For tRFs, which have not yet been extensively characterized, we will predict tRFs with corresponding targets by searching for complementary 3' sequences in the 3'UTRs or gene regions of interest. Based on these predicted interactions, we will create a bipartite network of miRNAs/tRFs and potential target mRNAs/lncRNAs/circRNAs. Additionally, we will create a bipartite network of lncRNAs/circRNAs and mRNAs that share regulating miRNAs/tRFs to characterize a ceRNA network. We will overlay our expression data onto these networks to filter for plausible interactions: for a miRNA-mRNA interaction, we expect an inverse correlation (miRNA high, target mRNA low and vice versa) if the miRNA represses the target; for a ceRNA pair (lncRNA-mRNA sharing a miRNA), we expect a positive correlation (both are released from repression or co-regulated). We will thus retain and highlight interactions that are supported by expression correlations in our cohort (e.g. Pearson correlation with p < 0.01). The outcome will be a refined lncRNA/circRNA-miRNA-mRNA regulatory network for schizophrenia. This network will undergo graph theoretical investigation to ascertain high degree sub-networks and nodes, for example, we'll determine hub miRNAs and hub lncRNAs/circRNAs (that interact with many miRNAs and mRNAs, suggesting sponge or scaffold potentials). For example, an upregulated lncRNA in schizophrenia may bind/sequester a brain-expressed miRNA such as miR-137 (a known SZ linked miRNA) to regulate a set of synaptic genes. Thus constructing a triangular ceRNA module and by obtaining its predicted binding with ce-expression data, we will be more confident of regulatory relationships.
Pathway and functional enrichment analyses: In order to take meaning from the networks we will obtain, we will run enrichment analysis on gene sets from our findings. For each co-expression module of interest (especially those that differ between patients and controls or correlate with clinical traits), we will test enrichment of Gene Ontology (GO) terms, KEGG pathways, and other curated gene sets (e.g. cell-type specific markers, schizophrenia risk gene lists). Similarly, for each key miRNA or tRF, we will examine whether its predicted target genes are enriched for particular pathways. We anticipate modules reflecting known schizophrenia-associated processes: for example, a module downregulated in patients may be enriched for synaptic transmission and neuroplasticity genes, aligning with the synaptic dysfunction hypothesis, while an upregulated module might be enriched for immune/inflammatory response genes, consistent with immune activation in schizophrenia. Prior studies support these expectations e.g. Mahmoudi et al. found that circRNA host genes downregulated in schizophrenia cortex are enriched in neuronal development and immune pathways that frequently emerge from blood transcriptomic studies of psychiatric diseases16. We will also look for enrichment of less well-studied processes (such as metabolic pathways, RNA processing, etc.) if they arise, as these could represent novel insights. Any ceRNA modules we identify will likewise be examined for functional coherence: do the mRNAs in a ceRNA sub-network participate in a common pathway or cellular function? This could point to mechanistic units, e.g. a set of co-regulated genes in a neurotransmitter signaling pathway all modulated by a common miRNA and its lncRNA sponges. Through pathway analysis, we will generate testable hypotheses about how the coordinated RNA changes contribute to pathophysiology of schizophrenia. These findings will guide the biological interpretation of our networks.

Resources needed

v1_tstlt
v1_interv_date v1_sex
v1_age
v1_marital_stat
v1_partner
v1_no_bio_chld
v1_no_adpt_chld
v1_stp_chld
v1_liv_aln
v1_school
v1_prof_dgr
v1_ed_status
v1_curr_paid_empl
v1_disabl_pens
v1_spec_emp
v1_wrk_abs_pst_5_yrs
v1_cur_work_restr
v1_cntr_brth
v1_cntr_brth_m
v1_cntr_brth_f
v1_cntr_brth_gmm
v1_cntr_brth_gfm
v1_cntr_brth_gmf
v1_cntr_brth_gff
v1_cur_psy_trm
v1_outpat_psy_trm
v1_age_1st_out_trm
v1_daypat_inpat_trm
v1_age_1st_inpat_trm
v1_dur_illness
v1_1st_ep
v1_Antidepressants
v1_Antipsychotics
v1_Mood_stabilizers
v1_Tranquilizers
v1_Other_psychiatric
v1_adv
v1_medchange
v1_ever_smkd
v1_age_smk
v1_no_cig
v1_alc_pst12_mths
v1_alc_5orm
v1_lftm_alc_dep
v1_evr_ill_drg
v1_pst6_ill_drg
v1_scid_dsm_dx
v1_scid_dsm_dx_cat
v1_scid_ever_halls
v1_scid_ever_delus
v1_scid_ever_psyc
v1_scid_age_fst_psyc
v1_scid_yr_fst_psyc
v1_panss_sum_pos
v1_panss_sum_neg
v1_panss_sum_gen
v1_panss_sum_tot
v1_idsc_sum
v1_ymrs_sum
v1_cgi_s
v1_gaf
gsa_id
v1_smRNAome_id
v2_smRNAome_id
v3_smRNAome_id
v4_smRNAome_id