Super Enhancer RNAs Rewrite the Molecular Script of Complex Gene Regulation in the MDD Brain

Abstract

Chromatin conformational changes are key to gene regulation in the brain and have recently been shown to be regulated by super-enhancer RNAs (seRNAs). We examine the role of seRNAs in major depressive disorder (MDD) by profiling the prefrontal cortex from 25 MDD subjects and 25 matched non-psychiatric controls. Our analysis reveals 175 differentially expressed seRNAs, of which 140 are upregulated and 35 are downregulated. Our proximity mapping (±50 kb) links 94 seRNAs to nearby protein-coding genes, while long-range analysis (±500 kb) uncovered 584 seRNA–mRNA interaction pairs, most within 400 kb of genomic range. These interactions formed cis-regulatory hubs and hotspots, with notable pairs involving CEBPA, NR3C1, and DUSP1. Additionally, several genes are found to be co-regulated by distinct seRNAs. Spatial mapping confirms that many altered seRNAs are located near super-enhancer regions. Our results for the first time highlight seRNAs as key modulators of proximal and distal gene networks in MDD, offering novel insights and therapeutic avenues.

Super-enhancer RNAs in the prefrontal cortex reshape chromatin and gene regulation in major depressive disorder, forming cis-regulatory hubs around stress-related genes and emerging as key modulators and potential therapeutic targets.

Introduction

Major depressive disorder (MDD) is a debilitating condition whose underlying molecular mechanisms are not well understood¹. Changes in local chromatin conformation are of great interest in understanding the complex gene regulation process, which is often disrupted in the MDD brain^2–4. A recent understanding of chromatin regulation has shown the role of cis-acting transcriptional units (known as enhancers) to help transcribe non-coding RNAs called enhancer RNAs (eRNAs)⁵. In the genome, such enhancers often form clusters (super enhancers) to drive long-range gene expression⁶. The role of super-enhancer regions in the genome is integrally connected with eRNA functions (known as super-enhancer RNA, or seRNA)^7,8. seRNAs primarily serve as epitranscriptomic switches to stimulate transcription of neighboring genes⁹. Compelling evidence has highlighted their mechanistic roles in long-distance gene regulation through chromatin looping¹⁰. Recent studies have elucidated the involvement of eRNAs in activity-dependent transcriptional processing, facilitating synaptic plasticity¹¹. To date, the functional role of seRNA in the neuropsychiatric disorders remains unclear. Additionally, their epitranscriptomic role in altering gene expression profiles is a relatively emerging area and warrants further research¹². Based on current understanding, we hypothesize that changes in seRNA expression may influence the regulation of local and long-range target gene expression, which will be key to uncovering a previously unknown domain of the gene regulatory landscape in the MDD brain.

Results and discussion

In this study, we determined the seRNA expression profile in well-matched dorsolateral prefrontal cortex (dlPFC) from 25 non-psychiatric controls and 25 MDD subjects (Supplementary Methods and Supplementary Table 1). Following normalization, differential expression of seRNAs between control and MDD groups was determined using an independent two-sample t-test. Statistical significance was defined by nominal p-values (p ≤ 0.05), and 175 significantly differentially expressed seRNAs (p ≤ 0.05) were identified in the MDD group; of these, 140 were significantly upregulated, and 35 were downregulated (Supplementary Data 1), regardless of any expression threshold. A volcano plot for overall differential regulation and an expression heatmap for the 25 top-ranking seRNAs are shown in Fig. 1A, B, respectively. These seRNAs were largely unaffected by covariates such as age, post-mortem interval, sex, and brain pH, and positive alcohol and antidepressant toxicology (Supplementary Fig. 1, Supplementary Data 2A–C). To assess the potential influence of demographic covariates, differential expression analysis was repeated, including age and sex as confounders in the linear model. In our findings, the overall direction and magnitude of fold changes (in log2FC) were highly consistent between adjusted and unadjusted analyses (Supplementary Fig. 2), suggesting the observed transcriptional patterns were not primarily driven by clinical covariates. The results from the covariate-adjusted model following linear regression analysis have now been included in the revised manuscript and provided in Supplementary Data 3, which shows a direct comparison between the adjusted and unadjusted analyses. Together, these results demonstrate that the primary transcriptional patterns observed in this study are robust to covariate adjustment. For proximity analysis, mRNA profiling was done simultaneously using the same RNA samples, ensuring robust and biologically meaningful insights into the spatial and functional interactions between seRNAs and mRNAs. The mRNA expression profiling revealed changes in 6662 genes in the MDD group (Supplementary Fig. 3A); of these, 75 were upregulated and 160 were downregulated, with p < 0.01 (Supplementary Data 4). The top 50 significantly altered genes are displayed in Supplementary Fig. 3B as a clustered heatmap. Next, the pairing between seRNAs and mRNAs was determined based on their genomic proximity (i.e., within ±50 kb) or their overlap with the corresponding seRNA locus, while also accounting for the ability of seRNA to regulate genes over long distances (±500 kb). For proximity analysis, we identified genes within 50 kb of the seRNA locus, evaluated their regulatory potential based on consistent co-expression in the MDD brain, and prioritized those targets potentially influenced by cis. In the long-range analysis, we extended the locus window to 500 kb to include distal genes regulated by complex genomic features associated with seRNAs that were upregulated in the MDD brain. The proximity analysis revealed 94 seRNAs, indicating a putative association between seRNAs and their target coding genes. The pairing in the genomic context is illustrated as a circular plot (Fig. 1C), whereas the top 25 proximity pairs between seRNAs and mRNA are mapped as a composite arc plot (Fig. 1D). These pairs were associated with genes such as CEBPA, NR3C1, LPCAT1, ID2, and DUSP1, all of which are crucial for neuronal functions and plasticity.

Fig. 1. Differential regulation of seRNAs and their genomic pairing with mRNAs.

A Volcano plot illustrating the magnitude (log₂ fold change) and significance (−log₁₀ p-value) of differential seRNA expression, with top-ranking candidates highlighted. B Expression heatmap of the 25 highest-ranking seRNAs, showing distinct expression profiles across samples. C Circos plot depicting genomic pairing between seRNAs and coding genes, defined by proximity (±50 kb) or locus overlap, identifying 94 significant seRNA–mRNA pairs. D Composite arc plot of the top 25 proximity pairs, including genes such as CEBPA, NR3C1, LPCAT1, ID2, and DUSP1, which are involved in neuronal function and plasticity. E Alluvial plot generated using a dynamic genome overlap algorithm (GenomicRanges, ±500 kb window) illustrating 584 pairwise interactions between significantly upregulated seRNAs and differentially expressed mRNAs, with flow widths representing interaction strength. F Genomic distance plot showing the distribution of absolute distances (in base pairs) between each lncRNA and the transcription start site (TSS) of its nearest protein-coding gene, providing a visual representation of potential cis- and trans-regulatory relationships within a 500 kb search window. G The qPCR expression profile of three seRNAs using dlPFC of MDD and control subjects. Fold change was calculated following the delta delta CT method and is shown via the bar diagrams. Data was normalized against U6. Student’s t-test was used to determine statistical significance (p ≤ 0.05). Data are presented as the mean ± SEM.

Based on current understanding, the regulatory reach of seRNA may not be limited to the proximity of its target genes; instead, it can extend over several hundred kilobases across the genome¹³. With this, we used a dynamic genome overlap algorithm to screen for protein-coding genes with a search window of ±500 kb from the seRNA loci. Using the GenomicRanges package from the R library, we determined a set of 584 pairwise interactions between significantly upregulated seRNAs and differentially expressed mRNAs (Supplementary Data 5). Although we did not perform a formal co-expression network analysis, we conducted this exploratory analysis that considered the known regulatory properties of seRNAs. To gain insight into their potential roles in both cis- and trans-acting gene regulation over long genomic distances, we visualized this putative regulatory relationship using an alluvial plot (Fig. 1E); the flow widths in the figure reflect interaction strength. This displays the long-range association between seRNAs and coding genes across the genome. This representation highlights seRNAs that may function as regulatory hubs by linking to multiple target genes, helping to prioritize key regulatory relationships and visualize the structure of distal gene regulation driven by seRNAs.

We further analyzed the distribution of genomic distances between each seRNA and the start site of its nearby protein-coding gene, identified within a 500 kb window using a dynamic overlap approach under the same assumptions as described above. A genomic distance plot was generated by calculating the absolute distance (in base pairs) between each seRNA and the transcription start site (TSS) of its nearest gene (Fig. 1F). This plot provides insights into the distance between protein-coding genes and their associated seRNAs, helping to determine whether these seRNAs are likely to act in cis (locally) or trans (distally). In our dataset, the normalized distance distribution was skewed toward larger distances (most values > 0.8, corresponding to <400 kb), indicating that most of the seRNA-gene pairs are distally located within several kilobases of each other. This enrichment of long-range interactions supports a predominant role for distal cis- and trans-regulatory activity, suggesting that the 500 kb window used for identifying dynamic overlaps effectively captured meaningful biological associations. The genomic distance plot also aided in identifying regulatory hotspots and setting thresholds (e.g., 10–100 kb) for more refined downstream analyses.

The cross-platform validation of seRNA expression data was another objective that we wanted to confirm. We aimed to evaluate the reliability of expression data obtained from a probe-based microarray compared with primer-specific qPCR in the dlPFC of MDD subjects. Using qPCR, we randomly tested three upregulated seRNAs (identified in the microarray analysis) in the dlPFC of MDD and control subjects. Our fold change-based expression analysis determined significant (p ≤ 0.05) expression changes of the three seRNAs (CATG00000097857.1, AC007879.2, and RP11-245J9.5) with more than 1.5 fold expression upregulation in the dlPFC of MDD subjects compared to matched controls. We present the expression data as a color-coded bar diagram in Fig. 1G. The qPCR-specific oligo primer sequences used for the three seRNAs are provided in Supplementary Data 6.

We were also interested in revealing a genome-wide overview of regulatory interactions between seRNAs and their nearby protein-coding genes. To achieve this, we created a circos plot (Fig. 2A) using the seRNA and long target gene interaction data, encompassing all 584 pairwise connections across the genome. Our analysis revealed that several upregulated seRNAs were closely linked to genes exhibiting significant expression changes, indicating a strong potential for cis-regulatory activity. Several chromosomes exhibited dense clusters of connections, highlighting regulatory hotspots. These hotspots could be crucial for understanding the potential positive regulatory influence of seRNAs on multiple genes within the same genomic neighborhood. From the circos plot, it is evident that some genes are linked to multiple seRNAs, suggesting redundant or cooperative regulation, while certain seRNAs connect to widely spaced genes, indicating a broader regulatory reach across the genome. Overall, the data support the idea that seRNAs play a central role in organizing local transcriptional programs and may function as key regulators in long-distance gene regulation focused on specific genomic regions.

Fig. 2. Genome-wide visualization and network analysis of seRNA–protein-coding gene interactions in the MDD brain.

A Circos plot showing all 584 pairwise genomic connections between seRNAs and nearby protein-coding genes within a ±500 kb window. Connections reveal genomic regions with dense regulatory interactions, including multi-seRNA–gene links and seRNAs connecting to widely spaced targets. B An interaction network constructed from seRNAs and significantly upregulated protein-coding genes (p ≤ 0.1), with nodes representing seRNAs or genes and edges indicating genomic proximity-based associations. Node size is proportional to the number of connections, and the spring layout highlights potential regulatory hubs and co-regulatory clusters. C Horizontal arc plot of the top 20 statistically significant seRNA–gene interactions, with seRNAs (blue) and target genes (red) placed on opposite axes. Arcs are colored by −log₁₀(p-value) to indicate statistical significance. D Chromosome clustering map of 24 seRNAs forming significant pairs with target genes, illustrating the genomic distribution of cis-regulatory modules across the genome. E Dot heatmap of the top 50 seRNA–gene interactions, where dot color reflects association significance (−log₁₀ p-value) and dot size indicates fold change of the target gene expression. F UpSet plot of intersections among protein-coding gene sets linked to individual seRNAs. Each seRNA is defined as a set of horizontal bars for proximal or associated target genes. Left horizontal bars indicate set size (number of genes per seRNA). Upper vertical bars indicate intersection size for specific seRNA combinations, as defined by connected dots in the matrix.

We further narrowed the list of 584 interaction pairs to include significantly upregulated mRNAs, based on the idea that the unilateral influence of upregulated seRNAs in inducing target gene expression was a key factor. A seRNA–gene interaction network was constructed based on genomic proximity (500 kb) following the adjustment of significance level (p ≤ 0.01) of the upregulated genes in MDD brain (Fig. 2B). In the network, nodes represent seRNAs and nearby protein-coding genes, with edges indicating potential cis-regulatory associations between them. Node sizes are proportional to the number of connections, highlighting potential regulatory hubs as central coordinators of the regulatory network. For clarity, the network was visualized using a spring layout to reveal topological organization and potential co-regulatory clusters. This network visualization provides a biological overview of how the upregulated seRNAs may be involved in modulating nearby gene expression and structuring the cis-regulatory landscapes in the MDD brain. It also offers a systems-level perspective on seRNA–gene interactions, valuable for understanding intricate gene regulation landscapes in the context of MDD. To better understand the genomic proximity and differential expression analysis of target genes, we constructed a horizontal arc plot to visualize the top 20 statistically significant interactions between seRNAs and their nearby protein-coding genes (Fig. 2C). In the plot, the nodes are separated along two axes: the seRNA (blue) and its gene targets (red). Arcs between them represent potential regulatory interactions, colored by −log10(p-value) to reflect statistical significance. This visualization highlights the strength and scope of seRNA-driven gene regulation, revealing both highly significant targets and broader, weaker associations within a local genomic neighborhood. The pairing determined a significant positive association between seRNA-RP1-25719.2 with TIAM2, seRNA-CNPY3 with TRERF1, seRNA-CAML3-AS1 with GDI2, seRNA-RP11-875011.2 with SORBS3, seRNA-RP11-70F11.11 with BCDIN3D, and several more.

Next, we attempted to interpret how seRNAs may influence gene regulation in their local environment as cis-regulators of nearby protein-coding genes. This could be a key mechanism in structuring the local gene regulatory network, which is primed to adapt to the changes occurring in the MDD brain. Further highlighting the genomic clustering pattern, 24 seRNAs forming significant pairs with targeted coding genes (Supplementary Table 2) were used to construct a chromosome clustering map across the genome (Fig. 2D). To our understanding, discovering novel interactions could help define a functional cis-acting module responsible for crucial changes in the MDD brain. The dot heatmap (Fig. 2E) summarizes the top 50 seRNA–gene interactions based on statistical association and gene expression changes. The plot highlights seRNAs with multiple strong associations, suggesting their role as cis-regulatory hubs, and identifies genes that may be key regulatory targets. This visualization helps prioritize candidate seRNAs and genes for downstream functional studies and reveals patterns of co-regulation and local transcriptional control mediated by seRNAs.

Additionally, we aimed to highlight the regulatory overlap of seRNAs with their long-distance target genes, demonstrating that certain coding genes can be coregulated by multiple seRNAs. With this idea, we used the seRNA-gene overlap datasets to generate an UpSet plot visualizing the overlap of protein-coding genes associated with different seRNAs (Fig. 2F). Each seRNA was treated as a set containing its nearby or potentially regulated genes. The plot reveals how these gene sets intersect across seRNAs. In the plot, the horizontal bars represent the total number of genes linked to each individual seRNA (as shown with the numbers on the leftmost side of the plot), while the vertical bars with their connected dots indicate the size of gene sets shared between specific combinations of seRNAs. Finding such a shared regulation is quite interesting and reflects functional redundancy, cooperative control, and involvement in common biological pathways in the MDD brain.

Next, we sought to determine how genetic variation might influence seRNA-target gene regulation and contribute to MDD pathogenicity. To this end, we conducted comprehensive integrative analyses, including GWAS, Bayesian colocalization, and eQTL analyses, focusing on 584 seRNA-target gene pairs located within a long-range genomic window.

We retrieved chromosomal segments corresponding to 126 seRNA pairs and extracted summary statistics for major depressive disorder (MDD) from the PGC, GWAS Catalog, and FinnGen databases. Among these resources, only the PGC dataset identified genome-wide significant susceptibility loci (Supplementary Data 7), several of which overlapped the chromosomal regions harboring the 126 seRNA pairs (Supplementary Data 8). In contrast, no genome-wide significant variants were detected within these regions in either the GWAS Catalog (Supplementary Data 9 or FinnGen datasets (Supplementary Data 10).

To evaluate whether super-enhancer–associated seRNAs mediate the effects of MDD risk variants through cis-regulatory mechanisms, we performed colocalization analyses between MDD GWAS signals and enhancer quantitative trait loci (enhancer eQTLs) within seRNA-associated super-enhancer regions. For each seRNA, we defined a ±500 kb cis-window centered on the corresponding super-enhancer and extracted all overlapping single-nucleotide polymorphisms (SNPs) with available MDD GWAS summary statistics and enhancer eQTL association results. Linkage disequilibrium (LD) was estimated using the 1000 Genomes Project Phase 3 European reference panel.

We applied a SuSiE-based fine-mapping framework to both GWAS and enhancer eQTL summary statistics, followed by Bayesian colocalization analysis to assess whether the two traits share a common causal variant. For each cis-eQTL or enhancer eQTL signal, we estimated the posterior probabilities of five competing hypotheses (PP.H0–PP.H4), with particular emphasis on PP.H4, which represents the probability that the GWAS and enhancer eQTL associations are driven by the same causal variant. Across the 60 cis-eQTL signals located within seRNA-associated super-enhancer regions (Supplementary Data 11 and 12), we observed consistently low posterior support for colocalization, with PP.H4 values close to zero (Supplementary Data 13). Instead, the posterior probability was predominantly driven by PP.H3, indicating the presence of both GWAS and enhancer eQTL association signals within the same loci but attributable to distinct causal variants. These results suggest that, within the tested cis-regulatory windows, MDD GWAS risk variants do not colocalize with enhancer eQTLs linked to super-enhancers containing seRNAs.

We next investigated whether genes regulated by these super-enhancers might instead mediate the effects of GWAS risk variants on MDD. To this end, we conducted a systematic colocalization analysis between GWAS and gene-level eQTL signals within ±500 kb of the target gene regions associated with seRNA-defined super-enhancers. GWAS summary statistics were integrated with eQTL data based on hg19 genome coordinates, and LD was again estimated using the 1000 Genomes Project Phase 3 European reference panel. For each locus, GWAS fine-mapping was performed using SuSiE-RSS, with the effective sample size defined as the median across all variants within the region. Using the same LD structure, eQTL fine-mapping was performed independently for each gene using SuSiE-RSS.

Because SuSiE-based colocalization may be unstable in the presence of weak or heterogeneous eQTL signals, we additionally applied an approximate Bayes factor (ABF)–based colocalization framework using coloc.abf as a complementary analysis. Quantitative traits were modeled assuming standardized phenotypes (sdY = 1), consistent with common practice in eQTL studies. Posterior probabilities for the five competing hypotheses (PP.H0–PP.H4) were extracted, with PP.H4 indicating evidence for shared causal variants between GWAS and eQTL signals. This analysis identified multiple genes within super-enhancer target regions that exhibited strong colocalization with MDD GWAS loci, with posterior probabilities of shared causality reaching PP.H4.abf ≈0.9 (Supplementary Data 14). Collectively, these findings support the presence of shared genetic signals linking MDD risk variants to downstream gene regulation mediated by super-enhancers, providing evidence that seRNAs may function as regulatory intermediaries connecting GWAS risk loci to gene expression programs relevant to MDD.

Placing these results in a broader context, accumulating evidence suggests that genetic risk for complex neuropsychiatric disorders is enriched in regulatory genomic regions, such as enhancers and super enhancers. The presence of these non-coding variants is likely to influence downstream gene expression through regulatory mechanisms rather than through direct changes in protein-coding sequences. In this context, our findings on seRNA-mediated regulation provide a seRNA-mRNA regulatory axis that serves as a key node through which MDD-associated polygenic risk can be interpreted. Our findings also show promise as accessible biomarkers for brain-specific neuropsychiatric conditions, as the presence of such non-coding RNAs in cerebrospinal fluid and exosomes has been documented. Moreover, comparing seRNA regulatory modules across similar brain conditions, including schizophrenia and bipolar disorder, may help to uncover shared or disorder-specific regulatory networks, which could be a key to convergent pathogenic mechanisms in the neuropsychiatric brain. Finally, the integrative analytic framework developed in our study may offer a blueprint for mining regulatory non-coding RNAs in other neuropsychiatric conditions.

Overall, our findings demonstrate that seRNAs are changed in the MDD brain, both in their expression and genomic context, and point to a potential role in spatial gene regulation. However, exploring complex chromosomal conformation following 3D chromatin analysis will be an important future step to validate these long-range associations⁵. Here, we would like to highlight that all cis-regulatory and long-distance regulatory interactions are interpreted primarily as observational phenomena rather than as evidence of specific underlying mechanisms.

Furthermore, we also note that in our cohort, age, post-mortem interval, sex, and brain pH were not different between controls and MDD subjects. This reduces the likelihood of confounding effects on the seRNA expression. Therefore, we compared the two groups using a t-test for initial screening. While we agree that including a linear model would have strengthened the rigor of our data analysis, we acknowledge this as a limitation of the present study. To our knowledge, this is the first study to identify such a unique class of non-coding RNA with a transformative role in gene regulation, particularly in the context of MDD¹⁴. These insights may help guide future studies focused on the regulatory landscape of MDD and identify new avenues for therapeutic development.

Methods

Human postmortem brain samples

The study was approved by the Institutional Review Board of the University of Alabama at Birmingham. The study design used human postmortem brain samples, accounting for variability in tissue quality and availability. Samples from dlPFC were obtained from the Maryland Brain Collection at the Maryland Psychiatric Research Center, Baltimore, MD. The cohort comprised 50 individuals, including 25 diagnosed with MDD and 25 non-psychiatric controls (hereafter referred to as controls). The demographic and clinical characteristics of the subjects are shown in Table S1. The psychiatric diagnosis was determined by psychological autopsy as described earlier¹⁵ using Diagnostic Evaluation After Death (DEAD)¹⁶ and the Structured Clinical Interview for the DSM-V (SCID)¹⁷. The interviews were done by a trained psychiatric social worker. Two psychiatrists independently reviewed the write-up from this interview, as well as the SCID that was completed from it, as part of their diagnostic assessment of the case. Diagnoses were made from the data obtained in this interview, medical records from the case, and records from the Medical Examiner’s office. The two diagnoses were compared, and discrepancies were resolved by means of a consensus conference. Control subjects were verified as free from mental illnesses using these consensus diagnostic procedures. We carefully selected the samples to match the groups based on certain key factors, including age, sex, PMI, pH, and RNA integrity number. Further, to reduce the investigator bias during data analysis, we have implemented a blinded approach. Furthermore, to mitigate batch effects, batch randomization was applied during sample processing.

After removal from the cranium, the brains were cut into six major pieces (four cerebral cortical lobes, basal ganglia-diencephalon, and lower brain stem-cerebellum), rapidly frozen on dry ice, and stored at −70 °C until dissection. During dissection, the frontal lobes were sliced into 1-mm to 1.5-mm thick coronal sections at a temperature between 0 °C and 10 °C. To keep the samples frozen, the dissections were performed on a metal plate over a container filled with dry ice. The prefrontal cortical samples were cut out of the coronal sections by a fine microdissecting (Graefe) knife under a stereomicroscope with low magnification. The dorsomedial prefrontal cortex (Brodmann’s area 9) was taken just dorsal to the frontopolar area, including the most polar portion of the superior and partly the middle frontal gyrus between the superior and intermediate frontal sulci. In the sections of the dissected cortical area, the gray and white matter were separated. The tissues were chopped into smaller pieces and stored at −80 °C until use.

All tissues from the control and suicide groups were screened for evidence of neuropathology by experienced neuropathologists at each brain collection program. The tissues were examined histologically. Fixed sections of PFC were screened with hematoxylin and eosin (H&E) staining and an antibody to glial fibrillary acid protein. Alzheimer’s disease, infarcts, demyelinating diseases, or atrophy disqualified subjects from the study. In addition, in each case, screening for the presence of human immunodeficiency virus (HIV) was done in blood samples, and all HIV-positive cases were excluded. Toxicology data were obtained by the analysis of urine and blood samples. Toxicology and the presence of antidepressants were examined by analysis of urine and blood samples from these subjects. Brain pH was measured as described previously¹⁸.

RNA isolation

Total RNA was isolated from postmortem brain tissue using TRIzol (Invitrogen) as described earlier¹⁹ and detailed in the Supplemental section. RNA quality and integrity were assessed using an Agilent Bioanalyzer 2100, with samples exhibiting an RNA integrity number (RIN) ≥ 7 selected for downstream analyses.

seRNA expression microarray

The total RNA extracted from dlPFC obtained from controls and MDD subjects was analyzed using a custom-designed human super-enhancer lncRNA microarray platform. The array was formatted to quantify super-enhancer-encoded lncRNAs with their corresponding protein-coding genes. The array was designed in an 8 × 15 K format, containing 14,873 distinct 60-mer oligonucleotide probes targeting 7753 human selncRNAs and 7040 associated mRNAs. Probe content was supported by a comprehensive annotation framework that integrated two primary curated transcriptomic resources. This framework focuses on public databases such as RefSeq, UCSC known genes, GENCODE, lncRNAdb, and NRED, and includes experimentally validated gold-standard lncRNAs reported in the literature. Furthermore, the genomic coordinates of the se-lncRNAs were obtained from the dbSUPER database.

Data preprocessing, cleaning, and normalization

The raw microarray data were preprocessed and normalized using the GeneSpring GX v12.1 software (Agilent Technologies). Following quantile normalization of microarray data, only the se-lncRNAs were retained for further analysis if they were flagged as present or marginal in ≥50% of samples^20,21. Expression data for each sample were log2-transformed to ensure a consistent distribution of intensity values across all samples, thereby facilitating uniformity in downstream analysis.

Differential expression analysis

Following normalization, differential expression of seRNAs between control and MDD groups was determined using an independent two-sample t-test. Statistical significance was defined by nominal p-values (p ≤ 0.05), and significant seRNAs were carried forward as candidate transcripts for further analysis and exploration. To understand the effect of covariates, differential seRNA expression analysis was also performed using a linear modeling framework based on limma. Specifically, the model included control and MDD groups as the primary factor of interest and incorporated age (as a continuous variable) and sex (as a categorial variable) as covariates to control for potential confounding effects on seRNA transcriptomic datasets. The linear model was fitted separately for each seRNA, and empirical Bayes moderation was applied to stabilize the variance estimate for each seRNA. Further, the differential expression results from the covariate-adjusted analysis were compared with those from the previously unadjusted analysis to assess the robustness of the observed transcriptional signals.

In silico gene proximity analysis

Gene proximity analysis was conducted using genomic coordinates derived from our experimental data, including both seRNA and mRNA datasets generated in this study. Genomic coordinates for seRNAs were identified based on microarray and bioinformatics analyses, while mRNA gene annotations were derived from the same dataset, ensuring consistency in genome build (e.g., hg38). Proximity was defined as a genomic locus (e.g., all coding genes under investigation) located within ±50 kb of seRNA loci (we refer to as the specified threshold distance, e.g., ±50 kb) to determine the spatial relationship between seRNA and gene loci, including upstream, downstream, or bidirectional associations. Computational tools were employed to assess proximity on the generated datasets. Bedtools was used to calculate genomic distances and identify the nearest genes to each seRNA locus, using the ‘bedtools closest‘ command. Additional R-based software packages, such as the GenomicRanges, enabled the identification of overlaps and nearest neighbors between seRNAs and mRNA genes, facilitating detailed proximity analysis and integration with downstream statistical assessments. Since all data originated from this study, mRNA expression levels were incorporated into the annotation process to explore potential correlations between seRNA proximity and gene expression. Statistical tests were performed to validate whether the observed proximity patterns exceeded random expectations. The enrichment beyond random expectation was evaluated using permutation testing across the genome with random genomic reshuffling. For genome-wide analyses, Circos plots and other visualization tools were employed to illustrate proximity patterns comprehensively.

mRNA expression analysis using seRNA expression microarray platform

The Arraystar seRNA microarray also included a dedicated set of probes (7,040 mRNA probes) targeting protein-coding mRNA transcripts. This feature of the microarray chip (Human MicroArray v5.0; 8 × 15 K, Arraystar) enabled the parallel assessment of coding gene expression for the analysis of potential seRNA targets of seRNAs from the same expression panel. The probes were designed to be transcript-specific and mapped to the exonic regions of the transcript to achieve maximum detection accuracy. The annotation for those selected transcripts is supported by established reference databases such as RefSeq, UCSC knownGene, and GENCODE. The microarray assay for coding genes (mRNA) was performed using the same RNA samples used for seRNA. Sample preparation and microarray hybridization followed the manufacturer’s standard protocols with minor modifications. Total RNA from each sample was amplified and transcribed into fluorescently labeled complementary RNA (cRNA) using the Arraystar Flash RNA Labeling Kit (Arraystar). The labeled cRNAs were hybridized onto the same array chip as described previously. Following hybridization, slides were washed thoroughly to remove non-specifically bound cRNA, and the arrays were scanned using the Agilent Scanner G2505C. A differential expression analysis of mRNAs was performed further as described in the following section.

Data preprocessing, cleansing, and normalization

Agilent Feature Extraction software was used to analyze the acquired array images. Raw intensity values were subjected to quantile normalization using the GeneSpring GX v12.1 software package (Agilent Technologies) to ensure uniform data distribution. Quantile-normalized expression levels were log2-transformed for subsequent analysis.

Differential expression analysis

Differentially expressed mRNAs between the two groups were identified using a Fold Change filtering method. We assessed statistical significance using an independent two-sample t-test with a nominal p-value threshold of 0.05. Genes showing significantly different expression levels (p-value ≤ 0.05) were then selected for further analysis and exploration in our study. The outline of the workflow followed in the data analysis process has been provided in Supplementary Fig. 4.

Integrated genomic association analysis

GWAS and eQTL data sources

Chromosomal segments corresponding to 126 super-enhancer–associated seRNA pairs were retrieved. Summary statistics for major depressive disorder (MDD) were obtained from three large-scale resources: the Psychiatric Genomics Consortium (PGC), the GWAS Catalog, and the FinnGen database. Genome coordinates were harmonized to the hg19 reference genome.

Definition of cis-regulatory windows and SNP extraction

For each seRNA, a ±500 kb cis-window centered on the corresponding super-enhancer was defined. All single-nucleotide polymorphisms (SNPs) within these windows with available MDD GWAS summary statistics and enhancer quantitative trait loci (enhancer eQTL) association results were extracted. Linkage disequilibrium (LD) matrices were constructed using the 1000 Genomes Project Phase 3 European reference panel.

Fine-mapping and colocalization analyses

Fine-mapping of both GWAS and enhancer eQTL summary statistics was performed using a SuSiE-based regression framework. Bayesian colocalization analysis was subsequently conducted to evaluate whether GWAS and enhancer eQTL signals shared a common causal variant. Posterior probabilities for five competing hypotheses (PP.H0–PP.H4) were estimated, with PP.H4 representing shared causality.

For gene-level analyses, eQTL signals within ±500 kb of target genes associated with seRNA-defined super-enhancers were analyzed. GWAS fine-mapping was conducted using SuSiE-RSS, with the effective sample size defined as the median across variants in each region. Gene-level eQTL fine-mapping was performed independently using the same LD structure.

Approximate Bayes factor–based colocalization

To complement SuSiE-based colocalization and mitigate potential instability arising from weak or heterogeneous eQTL signals, an approximate Bayes factor (ABF)–based colocalization analysis was performed using the coloc.abf framework. Quantitative traits were modeled assuming standardized phenotypes (sdY = 1). Posterior probabilities for all hypotheses were extracted, with PP.H4.abf indicating evidence for shared causal variants.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.D. obtained the funding, conceptualized the study, and supervised the study design and implementation. B.R. performed the study and subsequent analyses. Y.D. and B.R. co-wrote the paper. Y.D. edited and finalized the paper.

Peer review

Peer review information

Communications Biology thanks Qiyang Li, Masaki Nishioka and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Mengtan Xing. A peer review file is available.

Funding

This work was supported by a National Institute of Mental Health grant R56MH138596 to Dr. Dwivedi. The funder had no role in the design, execution, analysis, or interpretation of the data.

Data availability

All data generated or analyzed during this study are included in this published article [and its online supplementary information files]. Raw RNA-sequencing data generated during this study are available in the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi) with accession number GSE332986. Any additional requests for information can be directed to and will be fulfilled by the corresponding author.

Code availability

This paper does not report the original code.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-10306-9.