A non-coding RNA risk pathway in schizophrenia: miR-137 enhances the lncRNA GOMAFU through a pathological transcription network

Abstract

MicroRNAs (miRNAs) and long non-coding RNAs (lncRNAs) regulate broad gene networks through distinct mechanisms, which govern normal brain development and function but are dysregulated in schizophrenia (SCZ). However, how disease-risk miRNAs and lncRNAs co-operate to form pathogenic pathways in SCZ brains remain poorly understood. In this study, we identified a novel miRNA-lncRNA pathway in which the well-recognized SCZ-risk factor miR-137 enhances expression of the SCZ-risk lncRNA GOMAFU in human neuron development. We found significant up-regulation of GOMAFU during differentiation of multiple types of human neurons in vivo and in culture. Interestingly, the accumulation of histone acetylation, which activates numerous neuronal genes, down-regulates GOMAFU in iPSC-derived human neurons through inducing transcription repressors of GOMAFU, represented by the miR-137-target E2F6. We further demonstrated that miR-137 is necessary and sufficient for enhancing GOMAFU expression in a human neuronal progenitor cell (NPC) line and observed co-regulation of MIR137 with GOMAFU during normal human neuronal development and in SCZ brains. Moreover, we identified human NPC transcriptomic changes induced by miR-137 and discovered that miR-137 integrates functional co-operation of histone acetylation and transcription factors to promote GOMAFU expression. Notably, a significant number of miR-137-regulated transcription factors are predicted to bind the GOMAFU promoter and affected in SCZ brains, forming a highly interactive molecular network. Together, these results unveil the SCZ risk miR-137-GOMAFU non-coding RNA pathway connected by SCZ-affected transcription factors, providing a new mode of functional integration of non-coding and coding risk genes of SCZ that contributes to the complex etiology.

Introduction

The contribution of non-coding RNA (ncRNA) abnormalities to the pathogenesis of psychiatric disorders, represented by schizophrenia (SCZ), has been increasingly recognized [1, 2]. Numerous discoveries demonstrated crucial roles of microRNAs (miRNAs), a class of small ncRNAs of 20–24 nucleotides that suppress translation and/or degrade their target mRNAs in the cytoplasm [3, 4], in normal neuronal development and pathogenesis of brain diseases [5, 6]. In addition, a fast increasing volume of discoveries indicated malfunction of long non-coding RNAs (lncRNAs), which carry more than 200 nucleotides without protein coding capacity, in neuropsychiatric diseases [7]. Genome-wide association studies (GWAS) have identified genetic variants in neuropsychiatric disease-associated gene loci that produce miRNAs and lncRNAs [8–10]. Moreover, dysregulation of miRNAs and lncRNAs were detected in the postmortem brain as well as circulating blood of psychiatric disease patients as compared to their healthy controls [11–13].

In contrast to the highly conserved miRNAs, lncRNAs are poorly conserved and often preferentially enriched or uniquely present in the human brain [14, 15]. Nuclear and cytoplasmic lncRNAs can elicit transcriptional and/or posttranscriptional regulation of wide molecular networks [16] through distinct mechanisms from those by miRNAs [17]. Despite the growing lists of documented genetic alterations and dysregulated ncRNAs in the brains of psychiatric disease patients, whether and how disease-associated miRNAs and lncRNAs co-operate to form pathological pathways underlying the complex etiology and/or risk of neuropsychiatric disorders remain largely elusive. Examples of lncRNA modulating miRNA function have emerged in recent years [18, 19], among which sequestration of miRNA by cytoplasmic lncRNAs is the most studied mechanism [20, 21]. The converse possibility that miRNAs also regulate lncRNAs in neuropsychiatric brains has been postulated but rarely explored. In fact, molecular mechanisms that regulate lncRNAs in normal and diseased brains are poorly understood.

One human neuronal lncRNA dysregulated in multiple SCZ cohorts is GOMAFU [22–26]. The lncRNA gene that produces GOMAFU was originally named Myocardial Infarction Associated Transcript (MIAT) due to genetic variants associated with the susceptibility of myocardial infarction [27]. A linkage analysis also identified GOMAFU as a susceptibility gene for paranoid SCZ [28]. In addition, GOMAFU expression was dysregulated in peripheral blood mononuclear cells (PBMCs) isolated from SCZ patients compared to healthy controls and can be regulated by antipsychotic treatment [29, 30]. These observations suggest that GOMAFU abundance in the blood is a biomarker for SCZ diagnosis and treatment outcomes. Of note, the GOMAFU lncRNA is restricted in the nuclei [31] hence unlikely acts to sponge cytoplasmic miRNAs to regulate downstream genes. Instead, GOMAFU was reported to interact with splicing factors and regulate alternative splicing of SCZ risk factors represented by ERBB4 in human iPSC-derived neurons [32, 33]. Our recent study revealed differential dysregulation of GOMAFU in various regions of the SCZ postmortem brains from transcriptomic studies and a primary function of GOMAFU in suppressing interferon-γ (IFN-γ) response gene pathways in human neuronal progenitor cells (NPCs) besides modulating splicing. Moreover, IFN-γ insult, a well-recognized environmental factor of SCZ risk and pathogenesis [26], rapidly shuts down GOMAFU in human NPCs and triggers upregulation of GOMAF targets in neuroinflammatory gene cascades. Nonetheless, molecular mechanisms that regulate GOMAFU in normal human neuron development and cause GOMAFU dysregulation in SCZ brains remain undefined. Whether and how GOMAFU can cooperate with other SCZ risk ncRNAs in human neurons has never been explored.

MIR137HG is a well-recognized risk gene locus associated with multiple neuropsychiatric disorders, including intellectual disability [34], autism spectrum disorders (ASD) [35], and SCZ [36]. MIR137HG produces functional miR-137 that is highly enriched in the brain, which regulates fate selection of neural stem cells (NSCs) [37], neuronal differentiation [38], and synaptic maturation [39]. Single nucleotide polymorphisms (SNPs) within or near the MIR137 locus have been identified to associate with the risk of SCZ in multiple rounds of GWAS [1, 40, 41], which can alter miR-137 abundance in human neurons [39, 42]. Moreover, SCZ-associated MIR137 SNPs alter synaptic gene expression, and dendrite/synapse maturation of human iPSC-derived cortical neurons [42]. Besides SCZ-associated genetic alterations within the MIR137 locus, genetic variants in miR-137 target genes are also causatively associated with major mental illnesses [43, 44]. Together, these findings suggest that miR-137 and its downstream gene networks must be tightly regulated, which reinforces the critical roles of miR-137 in governing normal brain development and function.

Efforts in elucidating the function of miR-137 have been focused primarily on its direct mRNA targets derived from coding genes [38, 45]. However, whether and how miR-137 regulated gene cascades control expression of ncRNAs, especially lncRNAs that may also contribute to the etiology and/or risk of SCZ, have not been reported. This is a critical issue in understanding SCZ pathogenic gene networks, especially considering the broad impact of lncRNAs in regulating the transcriptome.

In this study, we discovered that in contrast to the well-known gene silencing activity of miRNAs, miR-137 enhances GOMAFU expression through indirect regulatory mechanisms during human neuron development, forming a novel miRNA-lncRNA pathway indicated in the risk of SCZ. Mechanistically, we identified a miR-137-regulated transcription regulatory network composed of many factors affected in SCZ, which mediate miR-137 induced GOMAFU expression in human neurons. These findings uncover new mechanisms that underlie co-operation of distinct ncRNA classes that form pathological networks through disease affected coding genes, which contribute to the risk and/or pathogenic development of neuropsychiatric diseases.

Materials and methods

Cell culture, transfection, and treatment

Human induced pluripotent stem cell (iPSC) lines derived from three healthy individuals were propagated, periodically authenticated, and differentiated into NPCs and cortical neurons as previously described [46]. Briefly, iPSCs were cultured on irradiated MEFs in human iPSC medium containing DMEM/F12 (Invitrogen), 20% Knockout Serum Replacement (KSR, Invitrogen), 1X Glutamax (Invitrogen), 1X MEM Non-essential Amino Acids (Invitrogen), 100 μM β-Mercaptoenthanol (Invitrogen), and 10 ng/ml human basic FGF (PeproTech). Then, human iPSC colonies were detached from feeder layer with 1 mg/ml collagenase IV for 1 h and suspended in embryonic body (EB) medium containing FGF-2-free iPSC medium supplemented with 2 μM A-83 and 2 μM dorsomorphin for 1 week. Subsequently, the floating EBs were transferred to Matrigel-coated 6-well plates on day 7 to form neural-tube like rosettes and maintained in hNPC medium including DMEM/F12 (Invitrogen), N2 supplement, B27, MEM non-essential amino acids (MEM NEAA) (Invitrogen), 2μg/ml heparin and 2 μM cyclopamine for 14 days. Attached rosettes were mechanically detached and cultured in low-attachment plates for another two days to form neurospheres. For cortical neuron differentiation, neurospheres were dissociated with Accutase at 37 °C for 8 min and placed on Matrigel-coated plates in the neurobasal medium supplemented with B27, 10 ng/ml GDNF (PeproTech), 10 ng/ml BDNF (PeproTech) and 2mM L-glutamine (Invitrogen). For inhibition of HDACs, 4-week-old human iPSC-derived cortical neurons were treated with 40 ng/ml trichostatin A (TSA, Sigma, T8552) for 24 h before harvest. Generation of Human iPSC-derived interneurons were published in previous studies [47].

The human neuroblastoma cell lines BE(2)-M17 and SH-SY5Y were purchased from ATCC and cultured in DMEM/F12 medium containing 10% fetal bovine serum (FBS) (HyClone). A mycoplasma PCR detection test was performed routinely to ensure no contamination. Cells were transfected with 100 pmol of a negative control siRNA, or siSP1 (Thermo Fisher Scientific, s13319), siTBX3 (Thermo Fisher Scientific, 5’-CAGCUCACCCUGCAGUCCA-3’), miRNA mimics, or miRNA inhibitors, including hsa-miR-137 and hsa-anti-miR-137 (Dharmacon, IH-310413-07-0005), using Lipofectamine 2000 (Invitrogen) following manufacturer’s instructions. To induce differentiation, BE(2)-M17 cells were propagated in DMEM/F12 medium containing 10% FBS and 20μm retinoic acid (RA, Sigma, r2625). Fresh medium was replaced every 2 days. BE(2)-17 cells that harbor deletion of the GOMAFU/MIAT promoter was described in our previous report [26].

Western blot

Whole cell lysates were prepared as previously described [48] and separated on SDS-PAGE before transferred to PVDF membranes. After blocking in Tris-buffered saline (TBS) supplemented with 0.1% Tween (TBST) and 5% milk for 1 h at room temperature (RT), membranes were probed with primary antibody at 4 °C overnight. After washing three times in TBST, membranes were incubated with secondary antibodies and subjected to chemiluminescence detection using chemidoc^TM MP imaging system (Bio-Rad Laboratories, Hercules, CA, USA). The density of each protein band was quantified using NIH ImageJ. Primary antibodies used are Anti-E2F6 (Abcam, ab53061, 1:3000), anti-H3K9AC (Sigma,07-352,1:2000) and anti-β-ACTIN (Sigma, A5441, 1:10000).

Analysis of transcription initiation of GOMAFU by FANTOM 5

CAGE (Cap Analysis of Gene Expression) RNA-Seq data sets in RIKEN FANTOM5 (https://fantom.gsc.riken.jp/zenbu/) was analyzed (gencode v19 filtered) to map transcription initiation activities of GOMAFU in human tissues based on reads normalized by Relative Log Expression (RLE).

Luciferase assay

The genomic fragment containing the minimal promoter of GOMAFU was amplified by PCR and inserted into the PGL3 basic luciferase reporter construct. In addition, the ORF of E2F6 (NM_198256.4) was amplified by RT-PCR and cloned into pcDNA3.1 (Invitrogen) as previously described [49]. The Luciferase reporter, the E2F6 plasmid, and the Rinella luciferase plasmid (Promega) were co-transfected using Lipofectamine 2000 (Invitrogen) into SH-SY5Y cells. Cells were harvested 24 h after transfection and subjected to the dual luciferase assay following manufacturer’s instructions (Promega, Madison, WI, USA).

RNA isolation, quantification, RNA-seq Library preparation, high‑throughput sequencing and RNA-seq analyses

RNA was isolated from three independent cultures of BE(2)-M17 cells harboring overexpression of miR-137 using TRIzol following manufacturer’s instructions (Invitrogen). After DNase treatment, an Agilent 2100 Bioanalyzer was used to determine the quality of RNA. PolyA-RNA libraries were sequenced on an Illumina HiSeq platform (Admera Health, LLC, South Plainfield, NJ, USA) with a read length configuration of 150 PE for 40 M total reads per sample (20 M in each direction). Paired-end reads were aligned to human genome assembly version (GRCh38/hg38) using TopHat2 version 2.1.0 [50] with default parameter in three replicates from independent experiments. Differential expression gene (DEG) analysis was carried out by Cuffdiff. DEGs with FDR < 0.05 were considered significant changes [51]. RT-qPCR was carried out for quantification of specific RNA species using primers listed in Table S1. The Protein Annotation Through Evolutionary Relationship (PANTHER) was used to perform the Gene Ontology (GO) analysis.

Identification of transcription factors predicted to bind the GOMAFU promoter and targeted by miR-137

Putative transcription factor (TF) binding sites at the human GOMAFU promoter were identified by JASPAR [52]. Relative profile score threshold was set as 80%. The mRNAs that encode GOMAFU-binding TFs and harbor predicted miR-137 target sites were identified by miRDB (https://mirdb.org/). The predicted functional influence on GOMAFU by each miR-137 regulated TF identified in RNA-seq was calculated as follows:

Total binding score on GOMAFU promoter in JASPAR x log2 Fold change caused by miR-137.

Identification of histone modification at the promoter of GOMAFU

H3K9AC ChIP-seq reads at the promoter of GOMAFU in human middle frontal cortex (ENCFF649HJS) and adult hippocampal layers (ENCFF920RBG) were obtained by analyzing the Encyclopedia of DNA Elements (ENCODE) project (https://www.encodeproject.org/).

Statistical analysis

Sample size was determined through simulation-based power analysis based on our preliminary data. All Statistical analyses were performed using Prism 9.4 (GraphPad Software). Data normality was assessed using the Shapiro-Wilk test. If data were normally distributed, an unpaired two-tailed t-test was used to determine differences between two groups; otherwise, the Mann-Whitney U test was applied. One-way ANOVA with Tukey’s post hoc test was carried out for multiple-group comparisons. All results were presented as mean ± SEM. *denotes p < 0.05, **denotes p < 0.01, ***denotes p < 0.001, ****denotes p < 0.0001, and ns denotes no significant difference.

Results

GOMAFU transcription is upregulated during differentiation of multiple types of human neurons

The risk of SCZ begins early in embryonic brain development [53]. We first examined expression of the SCZ risk lncRNA GOMAFU in NPCs (PAX6 + ), excitatory neurons (NEUROD2 +), and interneurons (GAD1 + ) in the single cell RNA-seq data sets derived from human fetal cortex [54]. As shown in Fig. 1A, GOMAFU is significantly upregulated in both excitatory neurons and interneurons as compared to NPCs. The upregulation of GOMAFU in vivo was recapitulated by RT-qPCR in culture upon differentiation of NPCs derived from human induced pluripotent stem cells (hiPSCs) into mature 2D cortical neurons (Fig. 1B) primarily composed of glutamatergic neurons [46]. Unlike GOMAFU, the NBAT1 lncRNA, which was previously reported to regulate neuronal differentiation [55], was not up-regulated (Fig. 1B), hence providing a specificity control that indicates not all lncRNAs are upregulated during human neuron differentiation. In addition, up-regulation of GOMAFU during differentiation of hiPSC-derived NPCs into 2D cortical neurons was positively correlated with the increased expression of the GRIN1 mRNA (Fig. 1C), which encodes the ionotropic glutamate receptor NMDA1, a well-characterized marker for neuronal differentiation [56]. In contrast to neuronal expression, GOMAFU was negligible in hiPSC-derived astrocytes (Fig. 1S). Besides hiPSC-derived glutamatergic neurons, upregulation of GOMAFU was also detected during differentiation of hiPSC-derived parvalbumin⁺ GABAergic interneurons based on RNA-seq [47] (Fig. 1D). NBAT1 was not found in the interneurons and the lncRNA HOTAIRM1 was downregulated. Moreover, upregulation of GOMAFU, but not NBAT1, was detected during differentiation of the human dopaminergic neuroblastoma cell line BE(2)-M17 by RNA-seq (Fig. 1E) [57] and confirmed by RT-qPCR (Fig. 1F). Hence, unlike the differential expression/regulation of other lncRNAs in neuronal subtypes, differentiation-programed selective upregulation of GOMAFU is a common phenomenon in various major subtypes of human neurons.

Fig. 1. GOMAFU is upregulated during differentiation of hiPSC-derived neurons.

A Violin plot of GOMAFU expression based on single cell RNA-seq in NPCs, excitatory neurons and interneurons in the prefrontal cortex of human fetal brain (PFC). TPM, transcript per million (GSE104276). B RT-qPCR of GOMAFU and NBAT1 in hiPSC-derived NPCs and cortical neurons undergone differentiation of four weeks in culture. Results are derived from multiple hiPSC lines from heathy individuals (*denotes p < 0.05, student t-test, n = 5–6. NS denotes no significant difference). C Linear regression analysis of RT-qPCR units of GRIN1 mRNA and GOMAFU (n = 12) in hiPSC-derived cortical neurons undergone 4–6 weeks of differentiation. D Expression levels of GOMAFU and HOTAIRM1 in hiPSC-derived NPCs and eight-week-old GABAergic interneurons quantified by RNA-seq (GSE145073) of 3 biological samples were calculated by DEseq2. Fold change of each RNA was normalized to that in NPCs (**denotes FDR < 0.01. NS denotes no significant difference). E Expression of GOMAFU and NBAT1 in BE(2)-M17 cells 10 days after differentiation based on RNA-seq (GSE181729) of 3 biological samples (**denotes FDR < 0.01). F RT-qPCR analysis of GOMAFU in BE(2)-M17 cells undergone 0 and 10 days of differentiation (**denotes p < 0.01, student t-test, n = 6).

In the 5th edition of the RIKEN FANTOM project (FANTOM5) dataset, which uses cap analysis of gene expression sequencing (CAGE-seq) as a surrogate measurement for transcription initiation [58], CAGE-seq reads of GOMAFU are detected in human brain embryonic neural stem cells (NSCs), which are increased in neurons isolated from human fetal brains. In contrast, only a few GOMAFU CAGE-seq reads are found in astrocytes isolated from cerebellum and cerebral cortex (Fig. S2A). Moreover, high levels of CAGE-seq reads of GOMAFU are found in adult brain regions that contain distinct types of neurons, except in cerebellum (Fig. S2B). Although GOMAFU was initially found as a product of the MIAT gene associated with myocardial infarction [27], FANTOM5 reads of GOMAFU in the heart were much lower than that in the brain (Fig. S2B). Together, these results suggest selective upregulation of transcription initiation of GOMAFU during human neuron development, which remains at high levels in mature brain neurons.

Histone acetylation downregulates GOMAFU in human neurons through inducing transcription repressors targeted by miR-137

Histone acetylation induces transcription of numerous genes during neuronal development [59]. We next explored acetylation of the 9^th lysine residue of histone 3 (H3K9ac), which is associated with increased chromatin accessibility and transcription activation [60], at the GOMAFU promoter in the human brain. As shown in Fig. 2A, chromatin immunoprecipitation sequencing (ChIP-seq) reads of H3K9ac are enriched surrounding the GOMAFU/MIAT transcription start site (TSS) in both the human cortex middle frontal area 46 and hippocampus (ENCODE). To test whether histone acetylation regulates GOMAFU, we inhibited histone deacetylases (HDACs) by Trichostatin A (TSA) in human iPSC-derived cortical neurons, which is known to induce chromatin decondensation and transcriptional activation of numerous neuronal genes [61, 62]. Increased H3K9ac was confirmed on immunoblot (Fig. 2B-C). Surprisingly, TSA-induced histone acetylation caused a significant reduction of the GOMAFU lncRNA detected by 5’ and 3’ primer sets in RT-qPCR (Fig. 2D). In contrast, TSA-treatment increased the NEUROD1 mRNA (Fig. 2E) consistent with a previous report [63, 64], without changing the lncRNA NBAT1 (Fig. 2F).

Fig. 2. Histone acetylation regulates GOMAFU in hiPSC-derived cortical neurons.

A H3K9ac ChIP-seq reads surrounding the GOMAFU promoter region in the middle frontal area 46 and layer of hippocampus of the human brain. Dashed line and arrow indicate the transcription start site (TSS). B Representative immunoblot of H3K9ac upon 24-hr TSA-treatment of hiPSC-derived four-week-old cortical neurons. C Quantification of H3K9ac intensity on immunoblot normalized to ACTIN. (*denotes p < 0.05, student t-test, n = 3). D–F RT-qPCR analysis of GOMAFU using 5’ end and 3’end primer sets (D), NEUROD1 (E) and NBAT1 (F) in human iPSC-derived four-week-old cortical neurons subjected to mock- or TSA-treatment for 24 hr (**denotes p < 0.01, ***denotes p < 0.001, NS denotes no significant difference, student t-test, n = 6). G Top: A schematic of a previously published luciferase reporter construct harboring functional minimal GOMAFU promoter. E2F6 binding sites are marked by red triangles with the sequences indicated below. Bottom: Luciferase reporter activities from the pGL3 construct carrying the GOMAFU promoter (pGL3-GOMAFU) co-transfected with a E2F6 construct or the parent control vector (****denotes p < 0.0001, student t-test, n = 16). H RT-qPCR analysis of E2F6 mRNA in human iPSC-derived four-week-old cortical neurons treated with TSA. (*denotes p < 0.05, student t-test, n = 9). I Top: Representative immunoblot of E2F6 after TSA-treatment of hiPSC-derived four-week-old cortical neurons. Bottom: Quantification of E2F6 intensity on immunoblot was normalized to that of ACTIN (*denotes p < 0.05, student t-test, n = 5–7).

One possibility is that TSA may induce transcription repressors that selectively downregulate GOMAFU transcription in human neurons. Consistent with this idea, we found multiple consensus sequence motifs for binding E2F6, a well-characterized polycomb transcription repressor [65, 66], within the functional core promoter of GOMAFU defined in our previous study [26] based on JASPAR (http://jaspar.genereg.net), a database of known TF-binding sites derived from experimentally based literature. The two E2F6 sites near the GOMAFU TSS that display highest scores in JASPER are shown in Fig. 2G. Moreover, enriched ChIP-seq reads of E2F6 were found immediately upstream of the endogenous GOMAFU TSS in multiple human cell types (Fig. S3). Indeed, activity of the GOMAFU promoter was significantly suppressed by E2F6 in a luciferase reporter assay (Fig. 2G). Furthermore, the E2F6 mRNA and protein levels were induced by TSA in human iPSC-derived cortical neurons (Fig. 2H and I), opposing to TSA-induced GOMAFU downregulation (Fig. 2D). Hence, E2F6 is an example of TSA-induced transcription repressors of GOMAFU.

Of note, the 3’ untranslated region (3’UTR) of human E2F6 contains an miR-137 target site (Fig. 3A) experimentally validated in non-neuronal cells [67]. We transfected miR-137 mimics into the human NPC model BE(2)-M17 cell line and observed significant reduction of E2F6 protein levels (Fig. 3B and C). To explore whether miR-137 targets additional transcription factors (TFs) to regulate GOMAFU gene transcription, we searched JASPAR to identify putative TFs that bind the GOMAFU promoter area [33]. As shown in (Fig. 3D), 274 human TFs in JASPAR were expressed in BE(2)-M17 cells based on transcriptomic profiling [57] and have predicted binding sites on the GOMAFU promoter, among which 45 TFs are predicted direct targets of miR-137 (https://mirdb.org/). The miR-137-targeted TFs were ranked by their predicted binding scores to the GOMAFU promoter, and the top 15 TFs are shown in Fig. 3E. E2F6, which suppresses GOMAFU promoter activity (Fig. 2), is among the top hits. These results suggest that GOMAFU transcription in human neurons may be regulated by multiple miR-137-targeted TFs.

Fig. 3. Putative GOMAFU promoter-binding transcription factors are predicted targets of miR-137, represented by the transcription repressor E2F6.

A A previously reported miR-137 target site in the 3’UTR of the E2F6 mRNA. B miR-137 suppresses E2F6 protein expression in BE(2)-M17 cells. A representative immunoblot of E2F6 protein transfected with miR-137-mimic or control miR (con-miR) for 48 h. C Quantification of E2F6 intensity on immunoblot was normalized to that of ACTIN and graphically displayed (*denotes p < 0.05, student t-test, n = 6). D Overlap of human TFs predicted to bind the GOMAFU promoter and TFs predicted as direct targets of miR-137 by miRDB. E Top 15 putative targets of miR-137 are predicted to bind promoter of GOMAFU.

MiR-137 enhances GOMAFU expression, forming a novel SCZ risk ncRNA pathway in human neurons

During differentiation of human iPSC-derived cortical neurons that harbor increased GOMAFU expression (Fig. 1B), both the miR-137 host gene (MIR137HG) transcript and the mature miR-137 are significantly up-regulated (Fig. 4A and B). In addition, co-upregulation of MIR137HG and GOMAFU is detected upon induced differentiation of the human neuroblastoma cell line BE(2)-M17 that harbor transcriptomic changes recapitulating differentiation of human iPSC-derived NPCs [26] (Fig. 4C). Interestingly, MIR137HG transcript levels are positively correlated with GOMAFU levels in multiple regions of postmortem adult brains (Fig. S4). Furthermore, both MIR137HG and GOMAFU are significantly reduced in the SCZ frontal and temporal cortex regions as compared to healthy controls in the PsychENCODE Consortium data set (Fig. 4D) [11], suggesting functional cooperation of MIR137 and GOMAFU deficiency in SCZ. Given the neuronal specific expression of MIR137HG and GOMAFU, the small fold changes likely underestimate the actual capacity of dysregulation.

Fig. 4. MiR-137 enhances GOMAFU expression to form a non-coding RNA pathway that regulates human neuron gene expression.

A, B RT-qPCR analysis of miR-137 host gene (MIR137HG) transcript (A) and the mature miR-137 (B) in human iPSC-derived NPCs and four-week-old cortical neurons. The fold change of each RNA in human iPSC-derived cortical neurons as compared to that in NPCs is shown (*denotes p < 0.05, **p < 0.01, student t-test, n = 5–6). C RNA-seq analysis shows the expression of GOMAFU and MIR137HG during BE(2)-M17 cells differentiation based on fragments per kilobase of exon per million fragments mapped (FPKM). (**denotes FDR < 0.01, ***FDR < 0.001). D MIR137HG and GOMAFU are significantly downregulated in the frontal and temporal cortex regions derived from SCZ based on RNA-seq dataset [11]. E, F RT-qPCR analysis of GOMAFU in BE(2)-M17 cells transfected with anti-miR-137 (E) or miR-137 mimic (F) for 48 h (*denotes p < 0.05, student t-test, n = 6). G RT-qPCR analysis of NBAT1 in BE(2)-M17 cells transfected with miR-137 mimic for 48 h (NS denotes no significant difference, student t-test, n = 7). (H, I) RT-qPCR of GRIN1, GRIA2 and MAP2 mRNAs and GOMAFU in the parental BE(2)-M17 cells transfected with control miR, or miR-137 (H) or harboring deletion of the GOMAFU promoter (ΔGOMAFU) undergone ten days of differentiation (I) (*denotes p < 0.05, **denotes p < 0.01, ***denotes p < 0.001, NS denotes no significant difference, student t-test, n = 6). For MAP2 in (I), NS denotes no significant difference, Mann-Whitney U test, n = 6.

We next questioned whether and how miR-137 may regulate GOMAFU. Upon transfection of anti-miR-137 into BE(2)-M17 cells that express low levels of miR-137 recapitulating that in human NPCs, a significant reduction of GOMAFU was detected (Fig. 4E), suggesting the functional requirement of endogenous miR-137 for GOMAFU expression from early stages of human neuron development. Reciprocally, we transfected exogenous miR-137-mimic to recapitulate the developmental upregulation of miR-137 and found a robust increase of GOMAFU (Fig. 4F), but not the lncRNA NBAT1 (Fig. 4G). Hence, an acute increase of miR-137 alone is sufficient to induce GOMAFU in human neuronal progenitor cells. In addition, acute treatment of miR-137 mimic also induced the GRIN1 mRNA, which is co-upregulated with GOMAFU during human neuron differentiation (Fig. 1C), and the mRNA encoding the glutamate receptor GRIA2 (Fig. 4H) that is also upregulated upon neuronal differentiation [56]. Conversely, CRISPR-Cas9 mediated ablation of GOMAFU expression [26] ameliorates miR-137-induced GRIN1 and GRIA2 in differentiated BE(2)-M17 cells (Fig. 4I). These data are consistent with the hypothesis that miR-137 and GOMAFU form a novel ncRNA pathway to enhance expression of genes induced during differentiation of NPCs and play important functions in human neurons.

Identification of transcriptional networks regulated by miR-137 that induce GOMAFU, which contain numerous SCZ-affected factors

To elucidate transcriptomic changes caused by miR-137 and molecular mechanisms by which miR-137 induces GOMAFU, BE(2)-M17 cells were transfected with miR-137 mimic or control miR. RNA-seq analysis identified 4479 differentially expressed genes (DEGs, FDR < 0.05) by miR-137, including 2081 upregulated and 2398 downregulated that are indicated in red and blue, respectively (Fig. 5A, Table S2). These DEGs were cross-compared with miRDB-predicted human mRNA targets of miR-137 that are expressed in BE(2)-M17 cells (Fig. 5B). Among the 925 predicted direct targets of miR-137, 494 are significantly downregulated, 385 remain unchanged, and only 46 are upregulated. These results support the predominant suppressor function of miR-137 on its direct targets. Nonetheless, both up-regulated and down-regulated DEGs are far beyond the predicted direct targets of miR-137 (Fig. 5B), consistent with the growing evidence that miRNAs regulate the transcriptome through complex gene cascades that involve broad indirect mechanisms [68, 69]. Interestingly, a substantial number of miRDB-predicted human miR-137 targets do not overlap with miRDB-predicted mouse targets (Fig. S5A). In addition, only a small fraction of miR-137 regulated DEGs in mouse NSCs [70] overlap with miR-137 DEGs in the human BE(2)-M17 cells (Fig. S5B). These data suggest that despite the identical miR-137 in human and mice, direct targets and genes indirectly regulated by miR-137 are species-specific. Importantly, miR-137 regulated human DEGs significantly overlap with the top ranked SCZ risk loci/genes identified by the PGC3 GWAS [1] (Fig. 5C), providing strong evidence that miR-137 indeed regulates SCZ risk genes in the BE(2)-M17 cell model of human NPCs. Moreover, we compared the DEGs caused by miR-137 overexpression (OE) with the RNA-seq data sets derived from miR-137 KO human iPSC-derived forebrain neurons [71]. Importantly, significant numbers of reversely overlapping DEGs were identified (Fig. 5D and Fig. S6A). The top hits, represented by PSD-95/DLG4 and RPS13 that are altered in SCZ [72, 73], were validated by RT-qPCR (Fig. 5E and F).

Fig. 5. Identification of transcriptomic alterations and gene pathways induced by acute increase of miR-137 in the human NPC cell line BE(2)-M17.

A Volcano plot indicates differentially expressed genes (DEGs) in BE(2)-M17 cells that harbor acute overexpression of miR-137. Red and blue dots represent significant DEGs up- or down-regulated by miR-137 with FDR < 0.05. Grey dots represent genes without significant changes in expression. B Venn diagram shows an overlap between predicted human miR-137 targets by miRDB and miR-137 regulated DEGs in BE(2)-M17 cells. C Venn diagram shows an overlap between schizophrenia genetic risk loci and miR-137 regulated DEGs in BE(2)-M17 cells. D Venn diagram shows an overlap between miR-137 regulated DEGs in BE(2)-M17 cells and reversely regulated DEGs in miR-137 knockout (137KO) hiPSC-derived forebrain neurons. For (B-D), p-values by Chi-Squared test are indicated. E, F Expression levels of DLG4 (E) and RPS13 (F) by miR-137 OE from RNA-seq of 3 biological samples are calculated by DEseq2 (dashed bars, *** denotes FRD < 0.001) and quantified by RT-qPCR (solid bars, *denotes p < 0.05, ** denotes p < 0.01, student t-test, n = 6), as representative reversely regulated genes by miR-137 OE and 137KO. (G-H) GO analysis of top pathways enriched of miR-137 upregulated DEGs (G) and downregulated DEGs (H). I RNA-seq analysis shows miR-137 enhances histone deacetylases (HDACs) while suppresses histone acetyltransferases (HATs) in BE(2)-M17 cells. The * denotes genes affected in SCZ [11]. J-M RT-qPCR of HDAC1 (J), HDAC10 (K), HDAC5 (L) and HAT1 mRNAs (M) in BE(2)-M17 cells transfected with miR-137 mimics or control miR. For (J), (K) and (M), * denotes p < 0.05, ** denotes p < 0.01, student t-test, n = 6. For (L), * denotes p < 0.05, Mann-Whitney U test, n = 6.

To identify gene pathways regulated by miR-137 in BE(2)-M17 cells, we performed Gene Ontology (GO) analysis on DEGs caused by miR-137 OE. Interestingly, top ranked pathways enriched of miR-137 upregulated DEGs are primarily indicated in the assembly and function of neuronal synapses and circuitry (Fig. 5G), suggesting that the well-known roles of miR-137 in governing neuronal development and synapse function are primarily mediated by its indirect targets. In contrast, miR-137 downregulated DEGs are enriched in pathways indicated in cell cycle DNA replication initiation and regulation of chromatin remodeling (Fig. 5H). Interestingly, STRING analysis of reversely overlapping DEGs in miR-137 OE and KO identified large cooperating gene clusters (Fig. S6B-C), including those governing cell cycle and nervous system development that are top ranked GO regulated by miR-137 OE (Fig. 5G and H), as well as ribosome components known to be affected in SCZ [74]. These data provide transcriptomic level support of functional overlap of miR-137 in BE(2)-M17 cells and human iPSC-derived NPCs.

Further analysis revealed that miR-137 upregulates multiple HDACs but downregulates histone transferases (HATs) (Fig. 5I), validated by RT-qPCR (Fig. 5J-M). The reciprocal effects of miR-137 on HDACs and HATs are expected to reduce histone acetylation, a well-known event in cell differentiation. Given the facts that induced histone acetylation downregulates GOMAFU in hiPSC-derived cortical neurons (Fig. 2D), reduction of histone acetylation by miR-137 likely contributes to miR-137-induced GOMAFU expression (Fig. 4).

We next questioned which TFs are functionally regulated by miR-137 to enhance GOMAFU expression. We identified 141 human TFs in the Jasper database that are regulated by miR-137 in BE(2)-M17 cells (Fig. 6A, Table S3), including 81 TFs downregulated and 60 TFs upregulated [75]. Among these miR-137 regulated TFs, 123 TFs are predicted to bind the GOMAFU promoter region, many are dysregulated in SCZ iPSC-derived NPCs (50), neurons (47), or both (31) (GSE128333) [75] as compared to controls as shown in Fig. 6B. The overlaps between miR-137-regulated TFs predicted to bind GOMAFU and TFs affected in NPCs or neurons derived from SCZ iPSCs are significant (Fig. S7). We ranked miR-137 up- and down-regulated GOMAFU-binding TFs based on their total binding score and fold changes caused by miR-137 (Fig. 6C and D). Notably, a majority of the top-ranked miR-137 regulated GOMAFU-binding TFs are affected in SCZ as compared to healthy controls (read * in Fig. 6C and D), in iPSC-derived NPCs, neurons or SCZ postmortem brains [11, 75, 76].

Fig. 6. Acute increase of miR-137 suppresses transcription factors affected in schizophrenia brains that form a network to regulate GOMAFU.

A Volcano plot indicates differentially expressed transcription factors (TFs) predicted to bind the GOMAFU promoter upon transfection of miR-137 mimic in BE(2)-M17 cells. Blue and red dots represent reduced and increased TFs that bind GOMAFU promoter upon miR-137 overexpression, respectively. Grey triangles indicate TFs regulated by miR-137 but not predicted to bind the GOMAFU promoter. B Upset plot shows TFs regulated by miR-137 that are affected in SCZ iPSC derived NPCs and Neurons as compared to those derived from healthy controls [75]. RNA-seq datasets were obtained from GSE128333. C, D Top 15 TFs downregulated (C) or upregulated (D) by miR-137 ranked by log2 Fold changes x binding scores. Red stars mark genes affected in SCZ. (E-H) RT-qPCR analysis of miR-137-targeted GOMAFU-binding TFs in BE(2)-M17 cells upon siRNA knockdown of SP1 E, F and TBX3 (G–H) and effects on GOMAFU. **denotes p < 0.01, ***denotes p < 0.001, student t-test, n = 5. I Venn diagram shows significant overlap between TFs affected in SCZ (RNA-seq datasets were obtained from Gandal et al. [11]) and miR-137 regulated TFs predicted to bind the GOMAFU promoter (p < 2.2e-16, Chi-Squared test). J STRING analysis of miR-137 regulated TFs affected in SCZ brains that are predicted to bind the GOMAFU promoter, which form a highly interactive molecular network. Nodes filling color was assigned as upregulated (red) or downregulated (blue) in SCZ patient brain, with log2 fold changes indicated by color code on right top corner.

SP1 is a top hit among down-regulated TFs by miR-137, which was found dysregulated in distinct regions of SCZ postmortem brains [76]. RNA-seq revealed downregulation of SP1 along with increased expression of MIR137HG upon differentiation of BE(2)-M17 cells (Fig. S8A and B). In addition, multiple experimentally validated miR-137 target sites [77] are present in the human SP1 mRNA 3’UTR (Fig. S8C). Consistent with SP1 regulating GOMAFU transcription, ChIP-seq reads of SP1 are found enriched at the 5’ of the GOMAFU gene (Fig. S8D). Indeed, knockdown (KD) of SP1 by siRNA (Fig. 6E) significantly increased GOMAFU expression (Fig. 6F), indicating SP1 is a repressor TF of GOMAFU. Similarly, siRNA KD of TBX3 (Fig. 6G), another SCZ-affected TF [23] directly targeted and repressed by miR137 during neuronal differentiation [78], also resulted in significantly increased GOMAFU expression (Fig. 6H). These results, together with miR-137-supressed E2F6 (Fig. 3), provided a proof of principle that miR-137 suppresses multiple repressor TFs that bind to the GOMAFU promoter, which collectively enhance GOMAFU transcription.

Finally, we identified 200 TFs (Fig. 6I) in the JASPER database that are affected in the postmortem frontal cortex of a SCZ cohort [11] that harbor deficiency of both MIR137HG and GOMAFU (Fig. 4D). Importantly, 52 of these 200 SCZ-affected TFs are on the list of miR-137 DEGs (Fig. 6I) identified in our RNA-seq and predicted to bind the GOMAFU promoter (Table.S4). Moreover, STRING analysis revealed a highly interactive network formed by these 52 SCZ-affected miR-137-regulated TFs that are likely involved in GOMAFU regulation (Fig. 6J). Together, these data suggest that miR-137-regulates a transcriptional regulatory network affected in SCZ, which apparently contributes to dysregulation of GOMAFU in SCZ brains.

Discussion

Our studies discovered a novel SCZ risk ncRNA pathway, in which miR-137 enhances the lncRNA GOMAFU through a transcriptional molecular network that integrates epigenetic modulation of histone acetylation and numerous SCZ-affected transcription factors. A comprehensive working model that describes the mechanisms by which miR-137 promotes GOMAFU transcription and potential pathological impacts due to dysregulation of this ncRNA pathway in SCZ brains is illustrated in Fig. 7.

Fig. 7. Working model for the miR-137-GOMAFU SCZ risk non-coding RNA pathway.

MiR-137 suppresses transcription repressors of GOMAFU, many of which are dysregulated in SCZ brains. In addition, miR-137 shifts the balance of histone acetylation enzymes that lead to histone hypoacetylation that also suppress transcription repressors of GOMAFU. By such converged mechanisms, miR-137 enhances GOMAFU transcription during human neuron differentiation. Moreover, malfunction of miR-137 dysregulates GOMAFU through SCZ-affected transcription regulatory networks, forming a SCZ risk non-coding RNA pathway.

Numerous examples demonstrated that disturbance of miRNAs or lncRNAs leads to malfunction of the brain. In contrast to the highly conserved miRNAs, lncRNAs arise later during evolution, often exist specifically in the primate lineage, many are uniquely expressed in the human brain [79]. Thus, lncRNAs have been postulated as a mechanism that underlies the evolution of cognitive sophistication in humans but also fragility of neuropsychiatric diseases [80], which often affect early brain development [81]. The marked upregulation of GOMAFU upon early differentiation of human NPCs into major types of human neurons (Fig. 1) suggests increased functional requirement of this lncRNA and broad impacts of its dysregulation in early SCZ pathogenesis. This is reinforced by the key roles of GOMAFU in suppressing IFN-γ response pathways in human neural progenitor cells that are most prone to IFN-γ insult [26], a well-recognized environmental risk of SCZ due to maternal immune activation caused by viral infection [26]. However, although the rs1894720 within GOMAFU was reported to be associated with the risk of paranoid SCZ [28], GOMAFU/MIAT was not identified in SCZ risk loci from the PGC3 GWAS data set [1]. Rather, the significant dysregulation of GOMAFU and GOMAFU-regulated gene networks stand out from transcriptomic analysis of various SCZ postmortem brain regions identified from our previous report [26]. These data suggest that GOMAFU-dysregulation by SCZ risk genes and its impacts on downstream etiological molecular pathways, instead of general genetic alterations within GOMAFU, contribute to SCZ risk and/or pathogenesis.

Despite the growing list of aberrantly expressed lncRNAs in neuropsychiatric diseases, molecular mechanisms that regulate disease-associated lncRNAs are poorly understood. The observation that histone-acetylation induces TF repressors, represented by E2F6, to suppress GOMAFU (Fig. 2) offers an intriguing mechanism to integrate epigenetic modulation and neuronal TFs. Such a mechanism allows specific genes to be upregulated upon the global reduction of histone acetylation and chromatin condensation during cell differentiation [82], which can regulate even broader gene networks when employed by regulating lncRNAs represented by GOMAFU. Given the potent and broad effects of E2F6 in regulating the transcriptome, E2F6 may suppress GOMAFU via directly binding to its promoter as well as indirectly regulating other factors to control GOMAFU expression. Of note, histone hyperacetylation is reported in SCZ iPSC-derived neurons as well as in SCZ postmortem brains [83]. Whether this may contribute to the reported GOMAFU deficiency in adult SCZ postmortem superior temporal gyrus [33] and the frontal and temporal cortex [11] is an intriguing possibility. On the contrary, GOMAFU is aberrantly increased in other clinically relevant SCZ brain regions [22–26], indicating that GOMAFU abundance must be tightly controlled in distinct brain regions and/or neuronal cell types, although precise mechanisms underlying GOMAFU up-regulation in these SCZ brain regions remain the next challenge.

In this study, we found that miR-137 is necessary and sufficient to drive GOMAFU transcription during early differentiation of human neurons (Fig. 4). This provides a new mode of functional interplay between small and long non-coding RNAs, beyond the well-known roles of lncRNAs in sequestration of miRNAs [84]. Such results expanded the known function of miR-137 in directly silencing coding genes that suppress neuronal and synaptic development [38, 45] to enhancing the lncRNA-governed molecular networks that also modulate brain development and function. We recognize that more efficient neuronal differentiation, a known roles of miR-137 [85], may also indirectly contribute to enhanced GOMAFU expression. Nonetheless, the co-upregulation of MIR137HG and GOMAGU during human neuron differentiation and their co-deficiency in SCZ postmortem cortex strongly suggests the functional connection of these two SCZ-associated ncRNAs in the risk and early pathogenesis of SCZ.

The abundance of miR-137 varies in different brain regions, particularly enriched in the hippocampus [86]. Similar to the up- or downregulation of GOMAFU in various SCZ brain regions [26], genetic alterations associated with SCZ risk are reported to increase or reduce MIR137/miR-137 expression [42, 86–88]. In addition, either deficiency or overexpression of miR-137 in mouse brains cause neuropsychiatric related phenotypes [45, 89]. Unfortunately, no data set is available that quantify mature miR-137 and GOMAFU in the same SCZ cohorts that harbor altered GOMAFU [26]. Hence, whether hyperactivity of miR-137, which is thought to be predominantly associated with SCZ phenotypes [90], leads to aberrant upregulation of GOMAFU in SCZ neurons remains a future challenge. Furthermore, how hyperfunction of the miR-137-GOMAFU pathway impacts SCZ risk and/or pathogenesis is a critical outstanding question to be addressed by future studies.

To date, most studies have focused on the roles of miR-137 in suppressing its direct target mRNAs [44]. However, the transcriptomic alterations caused by miR-137 in mouse neural stem cells and miR-137 KO mouse brains are far beyond its direct targets [45, 70]. In our study, acute increase of exogenous miR-137 in the BE(2)-M17 human NPC model cell line leads to downregulation of 53% (494/925) predicted miR-137 direct mRNA targets (Fig. 5B). This result is consistent with the gene silencing roles of miRNAs. However, these miR-137 downregulated direct targets only represent ~ 20% (494/2398) of miR-137 downregulated DEGs. Thus, the vast majority of miR-137 downregulated DEGs, plus the large number of miR-137 upregulated DEGs, are due to indirect regulation by orchestrated gene network changes. Importantly, these miR-137 regulated DEGs significantly overlapped with risk genes of SCZ (Fig. 5C), suggesting mechanisms by which miR-137 contribute to SCZ risk. Interestingly, miR-137 upregulated DEGs are strikingly enriched in pathways that underly synaptic development and plasticity (Fig. 5G), indicating that the well-known function of miR-137 in controlling neuronal function is largely mediated by its indirect functional targets. On the contrary, miR-137-downregulated DEGs are enriched in regulating basic molecular and cellular processes, including chromatin remodeling. We identified downregulation of HATs upon miR-137 exposure, accompanied by upregulation of HDACs (Fig. 5I-M). These reciprocal changes predict reduced histone acetylation, which contributes to the enhanced GOMAFU gene transcription, given the negative regulation of GOMAFU upon HDAC inhibition (Fig. 2).

The miR-137 DEGs in the BE(2)-M17 human NPC model contains many TFs (Fig. 6), consistent with the previously reported regulation of TFs by miR-137 overexpression [70]. Hence, miR-137 enhances GOMAFU gene expression in human neural progenitor cells through regulating the functional integration of TFs that bind GOMAFU promoter regions and histone acetylation mediated chromatin modification. However, only a few miR-137 regulated human TFs in BE(2)-M17 cells are regulated in mouse NSCs by miR-137 [70]. In addition, histone acetylation enzymes regulated by miR-137 in BE(2)-M17 cells are not regulated by miR-137 in mouse NSCs. Thus, although human and mouse harbor identical mature miR-137, transcriptional regulators controlled by miR-137 are species-specific.

Although the human TFs directly targeted by miR-137 can repress GOMAFU, represented by E2F6 (Fig. 3), SF1 and TBX3 (Fig. 6), among the miR-137 regulated TFs in BE(2)-M17 cells, only 30 are predicted direct targets. In addition to inhibiting suppressor TFs, miR-137 could also enhance TFs apparently via indirect mechanisms, which potentially may act as activators for GOMAFU transcription. This possibility still waits for future studies. Regardless of how miR-137 achieves sophisticated regulation of these TFs, a significant number of miR-137-regulated TFs predicted to bind the GOMAFU promoter are dysregulated in SCZ iPSC-derived NPCs and neurons (Fig. 6B and Fig. S7), as well as in the prefrontal cortical area (Fig. 6J) that harbor co-deficiency of both MIR137HG and GOMAFU. Importantly, these SCZ-affected TFs form an interacting network, suggesting their functional cooperation that connects malfunction of miR-137 to GOMAFU dysregulation. Moreover, given the aberrant histone acetylation in SCZ [83], miR-137 dysfunction in SCZ affects a transcription network that integrates pathological epigenetic alterations with disease-affected TFs, which converge on dysregulation of GOMAFU.

Conclutions and perspectives

In summary, our studies demonstrated that the well-recognized SCZ risk factor miR-137 enhances expression of the SCZ risk lncRNA GOMAFU through a transcription regulatory network that integrates cooperative actions between disease affected epigenetic mechanisms and transcription factors. These findings provide a new model for the functional interplay between distinct classes of ncRNAs beyond the traditional view in which lncRNAs act as sponges to sequester miRNAs. Moreover, our studies revealed new mechanisms that underlie the broad biological pathways regulated by miR-137 through coding and noncoding genes that cooperatively contribute to the SCZ etiology. At this point, how distinct SCZ-associated genetic variations in MIR137HG may affect the abundance of functional miR-137 and the miR-137-GOMAFU pathway remains elusive. Moreover, how the differential abundance of miR-137 in various brain regions regulates the SCZ affected transcriptional regulator network identified here and GOMAFU, especially in response to synaptic activity changes that are dysregulated in SCZ, are more challenging tasks for future studies.

Author contributions

TP, YZ and XJ designed, performed experiments and write the manuscript, YL, LK and FW performed experiments, WZ and BY helped design the experiments and composition of the manuscript, YF designed the experiments and wrote the manuscript.

Data availability

All genome-wide sequencing datasets have been deposited to Gene Expression Omnibus (GEO) repository with the accession ID GSE283922.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

All methods were performed in accordance with the guidelines and regulations by Emory School of Medicine, approved by the Institutional Biosafety Committee (B6-264-11R25). The informed consent was obtained from all participants.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-025-03709-5.