Integrative Analyses of scRNA-seq, Bulk mRNA-seq, and DNA Methylation Profiling in Depressed Suicide Brain Tissues

Abstract

Background

Suicidal behaviors have become a serious public health concern globally due to the economic and human cost of suicidal behavior to individuals, families, communities, and society. However, the underlying etiology and biological mechanism of suicidal behavior remains poorly understood.

Methods

We collected different single omic data, including single-cell RNA sequencing (scRNA-seq), bulk mRNA-seq, DNA methylation microarrays from the cortex of Major Depressive Disorder (MDD) in suicide subjects’ studies, as well as fluoxetine-treated rats brains. We matched subject IDs that overlapped between the transcriptome dataset and the methylation dataset. The differential expression genes and differentially methylated regions were calculated with a 2-group comparison analysis. Cross-omics analysis was performed to calculate the correlation between the methylated and transcript levels of differentially methylated CpG sites and mapped transcripts. Additionally, we performed a deconvolution analysis for bulk mRNA-seq and DNA methylation profiling with scRNA-seq as the reference profiles.

Results

Difference in cell type proportions among 7 cell types. Meanwhile, our analysis of single-cell sequence from the antidepressant-treated rats found that drug-specific differential expression genes were enriched into biological pathways, including ion channels and glutamatergic receptors.

Conclusions

This study identified some important dysregulated genes influenced by DNA methylation in 2 brain regions of depression and suicide patients. Interestingly, we found that oligodendrocyte precursor cells (OPCs) have the most contributors for cell-type proportions related to differential expression genes and methylated sites in suicidal behavior.

Significance Statement.

Suicide is the second-leading cause of death among young people in different populations. Currently, some candidate targets were identified from suicidal brain tissues with a single-omic-level study. This study aims to integrate multi-omics data from bulk mRNA-seq, snRNA-seq, and DNA methylation data to discover novel cell type–specific DEGs and signaling pathways. We found that 17 key genes were finally found to share changes across transcription, DNA methylation, and 2 different brain tissues (BA11 and BA25). OPCs have the most contributors for cell-type proportions related to differential expression genes and methylated sites in suicidal behavior, which replicated the results of previously published studies on single-nucleus RNA-sequencing. This study provides insights into key cell-type transcripts and CpG biomarkers for the treatment of MDD with suicide.

Introduction

Suicidal behaviors, including suicidal ideation, suicide attempts, and suicide completion, have become a serious public health concern globally due to the economic and human cost of suicidal behavior to individuals, families, communities, and society (Turecki and Brent, 2016b; Gifuni et al., 2021; Brassell et al., 2022; Harmer et al., 2022). Approximately 800,000 people die by suicide every year worldwide, and the number of suicide attempts is estimated to be 20 times that of completed suicides (Brassell et al., 2022; Meda et al., 2022; Miller et al., 2022). Suicide is the second leading cause of death among young people in different populations (Hilario et al., 2021; Kim et al., 2021; Marsden and Tuma, 2022; Watling et al., 2022). From the clinical perspective, the treatment methods for suicidal behavior mainly include physical therapy and psychotherapy intervention. For physical therapy, it mainly includes drug treatment and electroconvulsive therapy (Turecki and Brent, 2016). Regarding drug treatment, a meta-analysis of randomized controlled trials showed that antidepressant drug treatment reduces suicidal ideation in individuals aged 25 years and older (Vitiello et al., 2011; Gibbons et al., 2012; Grunebaum et al., 2012). In participants younger than 24 years, although antidepressant treatment reduced depressive symptoms, it did not always reduce suicidal ideation (Stone et al., 2009; Gibbons et al., 2012). In recent years, clinical studies have found that single and repeated use of ketamine can reduce suicidal ideation, but it has side effects such as transient responses and potential misuse (Fond et al., 2014; Rajkumar et al., 2015). Due to the heterogeneity in the performance and treatment of individuals with suicidal behavior, any treatment plan cannot be effective for all patients, and the development of risk prediction biomarkers and effective targets is of great importance for the prevention and treatment of suicide. The risk of suicidal behavior is multifactorial and includes a range of biological, psychiatric, psychosocial, and cultural factors. However, the underlying etiology and biological mechanism of suicidal behavior remains poorly understood (Docherty et al., 2021; Jiang et al., 2021; Neupane, 2021; Lindner et al., 2022). Currently, postmortem brain tissues obtained from suicide victims offer an important avenue to study the neuropathophysiological mechanisms associated with suicide by offering a snapshot of what happened in the brains that led to suicide and death (Gate et al., 2021; Chang et al., 2022; Jaffe et al., 2022; Kenigsbuch et al., 2022).

With the rapid development of genotyping and sequencing technologies, as well as powerful analytical tools, genome-wide association (GWA), transcriptomic, epigenomic studies, scRNA-seq of postmortem brains has been widely used to investigate genome-wide genetic association, gene expression changes, and epigenetic modifications that influence or mediate the effects of environmental factors on suicidal behaviors (Price et al., 2021; de Witte et al., 2022; Grima et al., 2022). A series of GWA studies, particularly the most recent—as well as the largest and most diverse GWA studies of suicide attempts and suicidal ideation—have identified at least 7 replicable cross-ancestry and 9 replicable ancestry-specific genome-wide significant loci, which provide convincing evidence that ESR1, DRD2, TRAF3, and DCC are highly plausible cross-ancestry risk genes for suicide attempts and suicidal ideation along with genes involved in synapse, dopamingeric pathways, and axon guidance (Kimbrel et al., 2023). In transcriptomic studies, gene expression changes have been reported in postmortem brain tissues of completed suicide, which were involved with altered glial, endothelial, and ATPase activity (Pantazatos et al., 2017); endothelial dysfunction (Kim et al., 2022); GABA and glutamate neurotransmitter system imbalance (Yin et al., 2016); as well as abnormal purinergic signaling in microglia (Punzi et al., 2022). In epigenomic studies, several genes, including PSORS1C3, TAPBP, ATP5G2 (Murphy et al., 2017), GRIK2 (Nagy et al., 2015), and MPP4 (Jeremian et al., 2017), showed differential methylation patterns in brain tissues in epigenome-wide association studies of suicidal behaviors with mood disorders (Fiori and Turecki, 2020). The above-mentioned studies have explored the etiology of mood disorders from the single-omic level. However, from the perspective of systems biology, single-omic-level studies may not be able to explain the complex molecular relationships and regulatory processes involved in the neurobiology of mood disorders. The research method of scientific integration is more conducive to exploring the complexity of brain organization and function from a holistic perspective so as to discover biomarkers and drug targets with clinical potential.

In particular, the rise of scRNA-seq technology in recent years has been applied to the study of mood disorders to discover independent cell-type–specific gene expression changes. In 2020, Nagy et al. (2020) used single-nucleus transcriptomics technology to identify gene expression differences in the prefrontal cortex tissue of male depression patients and found that nearly 60% of the identified cell types had gene expression dysregulation, among which OPCs and excitatory neurons had significant gene expression changes (Nagy et al., 2020). Subsequent comparison of single-nucleus RNA-sequencing data from the dorsolateral prefrontal cortex of 71 men and women found that cell-type–specific MDD-associated gene expression patterns were similar between males and females, but there were differences in significantly differentially expressed genes (DEGs): more DEGs were found in microglia and parvalbumin interneurons in females, whereas there were more DEGs in males in deep-layer excitatory neurons, astrocytes, and oligodendrocyte precursors (Nagy et al., 2020; Maitra et al., 2023). In this study, bioinformatics methods such as regression and deconvolution were used to integrate data from bulk mRNA-seq, single nucleus RNA sequencing, and DNA methylation in depressed suicide postmortem brains to discover new cell-type–specific DEGs and signaling pathways.

Methods and Materials

Human Postmortem RNA Sequencing Data and Matched Methylomic Profiling Data

To conduct our research, we accessed 4 datasets from the GEO database: a mRNA expression dataset GSE102556, a methylation dataset (GSE88890), and 2 single-cell datasets (GSE144136 and GSE197622) (Labonté et al., 2017; Murphy et al., 2017; Nagy et al., 2020; Rayan et al., 2022). These omics data were derived from tissue in the prefrontal cortex, a brain region where dysfunction is not only associated with stress-induced cognitive deficits but also with suicidal behavior. For example, early candidate gene studies found that the mRNA or protein levels of brain derived neurotrophic factor and Tyrosine Kinase receptor B genes were significantly reduced in the prefrontal cortex tissues of suicide completers, and the promoters or 3’-UTR gene regions of these 2 genes were also differentially methylated in the prefrontal cortex tissues (Dwivedi et al., 2003; Pandey et al., 2008; Penner-Goeke and Binder, 2019). Additionally, the postmortem brain tissues were obtained from the Douglas Bell Canada Brain Bank (DBCBQ; Douglas Mental Health Institute, Montreal, Québec, Canada) with overlapped samples and brain regions. The detailed clinical characteristics and data generation for both datasets were presented in the previous studies (Labonté et al., 2017; Murphy et al., 2017). For the GSE102556 dataset, 6 brain regions from a total of 48 subjects consisting of 37 MDD/suicide patients and 11 nonpsychiatric, sudden-death controls were examined at the transcription level by RNA-sequencing; and the 6 brain regions included the anterior insula (Ant), orbitofrontal (BA11), cyrus cinguli 25 (BA25), dorsolateral prefrontal cortex (BA8/9), nuclei accumbens (Nac), and subiculum (Subic). The GSE88890 dataset included genome-wide patterns of DNA methylation array data from 2 cortical regions, that is, BA11 and BA25, in 20 depressed MDD suicide deaths and 20 nonpsychiatric sudden deaths. We extracted corresponding matched mRNA expression and methylation array datasets from the same subjects in 2 overlapped regions, that is, BA11 and BA25, which resulted in 6 paired transcriptome-methylome datasets from MDD/suicide subjects and 4 paired datasets from control subjects from BA11 tissue as well as 5 paired datasets from MDD/suicide subjects and 3 paired datasets from control subjects from BA25 tissue (Table 1). For the cell-type analysis, we mainly conducted deconvolution analysis with single-cell sequencing data from the cortex regions (BA9) of 34 subjects from 17 MDD and 17 nonpsychiatric controls in the GSE144136 dataset (Nagy et al., 2020). Meanwhile, we examined the effects of fluoxetine-treated rats by analyzing single-nucleus transcript expression data in 2 brain regions, dorsal hippocampal dentate gyri (dorDG) and ventral hippocampal dentate gyri (venDG), for the GSE197622 dataset (Rayan et al., 2022).

Table 1.

Demographic Information of Subjects

Region	Sample ID (GSE102556)	Sample ID (GSE88890)	Phenotype	Age	Gender	Rin	pmi	pH
BA11	14	X9297953165_R03C02	CTRL	47	Male	79	12	6.49
	20	X9297962014_R04C01	CTRL	31	Male	7.1	29.5	6.67
	28	X9297962057_R01C01	CTRL	46	Male	7.7	19.5	6.42
	56	X9297962027_R02C01	MDDS	39	Male	4.8	19	6
	67	X9297962027_R03C01	MDDS	22	Male	7.4	24	6.68
	72	X9297962027_R05C01	MDDS	55	Female	6.9	36	6.79
	105	X9297953165_R02C01	MDDS	39	Male	6.8	18.5	6.37
	153	X9297953165_R05C02	MDDS	55	Female	8.5	2.5	6.5
	162	X9297962014_R02C02	CTRL	79	Female	6.8	7.5	6.4
	183	X9297962014_R03C01	MDDS	64	Male	9.1	6.5	6.25
BA25	14	X9297962043_R06C01	CTRL	47	Male	79	12	6.49
	20	X9283265107_R03C02	CTRL	31	Male	7.1	29.5	6.67
	28	X9297962015_R03C01	CTRL	46	Male	7.7	19.5	6.42
	56	X9297962015_R05C01	MDDS	39	Male	4.8	19	6
	67	X9297962015_R01C02	MDDS	22	Male	7.4	24	6.68
	72	X9297962015_R03C02	MDDS	55	Female	6.9	36	6.79
	105	X9297962043_R02C01	MDDS	39	Male	6.8	18.5	6.37
	153	X9343114119_R03C02	MDDS	55	Female	8.5	2.5	6.5

Note: CTRL, Control (); MDDS, Major Depressive Disorder with Suicide; rin, RNA integrity number; pmi, post-mortem interval.

Data Analyses

Differential Gene Expression (DGE)—

DGE was assessed through a generalized linear model implemented in limma package, with phenotype (MDD vs CTRL) as main factors for BA11 and BA25 brain region separately (Phipson et al., 2016), adjusted for age, sex, RNA integrity number (RIN), and Body Mass Index (BMI) as covariates. An individual gene was defined as differentially expressed if the P value of its t statistic was at most 0.05.

Differentially Methylated Regions (DMRs)—

DMRs in BA11 and BA25 were identified using the Bumphunter package (Jaffe et al., 2012). We first grouped methylation data, unmethylation data annotation, and phenotypic information of individual subject and merged into 1 dataset using the MethySet function. Then, the grouped dataset was transformed into beta and M values using function ratioConvert. The function mapToGenome was used to integrate the genomic coordinates and annotation information of CpG sites into the ratioConvert object. Finally, the bumphunter function was used to identify the DMRs between the 2 conditions, and the bumphunter algorithm was applied to compute a t statistic approach at each genomic location. The differentially methylated CpG sites were statistically screened in the BA11 and BA25 regions separately, and the P value of each significant CpG site was further corrected using the GLM function.

To demonstrate the changes of the methylated sites and each transcript in the 2 groups, the Manhattan function in the qqman package was used to plot the relationship between the significant level and the chromosomal position of the CpG sites and transcripts. In addition, we observed simultaneous changes in the 2 omics levels in specific chromosomal regions.

Cross-omics Correlation Analysis—

The chromosomal positions of all the CpG sites were retrieved from the GPL13534 platform, which annotates the BeadChip HumanMethylation450_15017482. LiftOver software was then used to transform the GRCh37/hg19 coordinates into the corresponding ones of the GRCh38/hg38 version. Based on the transformed chromosomal coordinates, bed files were generated and analyzed using the Bedtools software (Utah,USA) to obtain matched genes with each CpG site. Then, using the Ensembl function of the biomart package, each CpG site was mapped into the corresponding transcript ID of the Ensembl database.

After the chromosomal regions of CpG sites were mapped into the corresponding transcripts, correlation coefficients between methylation and expression levels of transcripts in each sample were calculated using COR function. By comparing the mean correlation coefficients between the case and control groups, we focused on methylation sites and transcripts with absolute correlation coefficients greater than 0.8, which were defined as highly correlated CpG sites with gene expression levels. We also investigated the CpG sites with opposite correlation coefficients between the 2 groups.

In addition to the correlation coefficients, we examined the methylation level of each differentially methylated site and calculated the fold-changes between cases and controls. In the present study, we set the threshold of fold change >1.5 as hypermethylated sites and a fold change <0.67 as hypomethylated sites. We further studied the fold changes in the corresponding transcripts in which chromosomal regions have hypermethylated CpG sites.

Functional Annotation of Highly Correlated CpGs Across Omics—

Based on the results of the correlation analysis, we retained only all the highly correlated CpG sites. To obtain the genomic functional regions of these CpG sites, the makeGRangesFromDataFram function from the GenomicRanges package was used to transform into GRanges files, including the chromosomal, start position, and end position of each CpG site. Then, GRanges files were input into the function of locateVariants of the VariantAnnotation package and aligned the GRange files with annotated reference genomic function regions provided by the TxDb.Hsapiens.UCSC.hg38 gene annotated package. Finally, each CpG site was annotated with different genomic functional regions, including promoter, 3’untranslated region (3' UTR), and intron. Finally, we summarized the regulatory gene regions at which these highly correlated CpG sites were located and explored the distribution of gene function regions of these highly correlated methylated sites.

Identification and Functional Annotation of Overlapping Genes Cross Omics or/and Cross Brain Regions—

The overlapping genes in the transcriptomics and methylated omics from BA11 or BA25 were obtained for similarity and difference analyses using the VennDiagram Package. KEGG and enriched pathway analyses of these overlapping transcripts were performed using the clusterProfiler package to identify the main enriched pathways for these key genes. Meanwhile, the same methods were used to further sort out overlapping genes between the 2 brain regions and 2 omics levels, which may contribute to the occurrence of MDD with suicide.

Deconvolution Analysis of Bulk RNA Data—

By using the ReadMtx function in the Seurat package to input the single-cell data (GSE 144136), the cell annotation types were referred to in the raw annotation cell types, which mainly included endothelial, excitatory neurons, OPCs, astrocytes, oligodendrocytes, macrophage/microglia, and inhibitory neurons. Then, dimensionality reduction analysis was performed on the cell annotation results by RunUMAP and RunTSNE functions. With regard to the MuSiC2 package, the single-cell data were input as a reference to estimate the cell type ratio of the bulk RNA-seq data. We also performed statistical analysis on the cell types of the normal group and the suicide group, and analyzed them with the music2_prop_t_statistics function to obtain the cell-type–specific differential genes. The clusterProfiler package was used to perform GO annotation analysis and KEGG pathway analysis on the obtained differential genes.

Cell-Type Deconvolution Analysis of Bulk DNA Methylation Data—

First, we collected the significant methylation sites corresponding to the transcripts and the beta value representing the methylation level. Similar to deconvolution analysis of bulk mRNA-seq, 7 broad cell types were obtained from single-cell sequence data, and the single-cell data as a reference were used to estimate the cell type proportions of the methylation data and compare the difference between the normal group and MDD groups of cell types. The music2_prop_t_statistics function was used to analyze the cell-type–specific differential genes, and the clusterProfiler package was applied to annotate the biological function of those differential genes.

DGE of Fluoxetine-Related RNA-seq Data—

The single-nuclear transcript expression data of fluoxetine-treated rats were retrieved from the GEO database (GSE197622). We reconsidered fluoxetine treatment and cell type as bivariate variables and performed multiple comparison analysis to screen the fluoxetine-specific and cell-type–specific differential genes. The VennDiagram package was used to identify the intersection of those genes and performed GO annotation and KEGG pathway analysis on the intersection genes.

Results

Differential Transcription and Methylation Analyses

For the transcriptional analysis, a total number of 1300 significantly differentially expressed transcripts was identified from the transcriptional data of BA11 after correction by multiple comparisons with different variable factors, such as age, sex, and BMI, as covariates. Similarly, 6793 transcripts were found in BA25. The Manhattan plot presents the significance level of the different transcripts of BA11 and BA25 and their chromosomal positional relationship (Fig. 1C and D). Interestingly, some top significant changes in transcripts belong to the lncRNA family, which is associated with suicide and depression as epigenetic modifications (supplementary Table 1).

Figure 1.

Distribution of transcripts and methylation sites on chromosomes. (A) BA11 methylation Manhattan plot. (B) BA25 methylation Manhattan plot. (C) BA11 transcriptome plot. (D) BA25 transcriptome plot.

For the methylation analysis, 3177 differential CpGs were identified at the genome-wide level, with significant changes in the BA11 region. In the BA25 region, 1304 differential CpG sites were identified with significant changes (P < .01). In Figure 1A and B, the Manhattan plot shows a significant relationship between the of CpG sites and chromosomal position. The most significant changes in CpG sites in BA11 and BA25 are listed in Table 2. In BA11 tissue, these significant changes in CpG loci were mainly distributed on chromosomes 8, 11, and 17 (Fig. 1A), whereas in BA25 tissue, these significant changes in CpG loci were distributed on chromosomes 10 and 5 (Fig. 1B).

Table 2.

The Top 20 Differentially Methylated Sites in the BA11 and BA25 Regions of Suicidal MDD Patients and Healthy Controls^a

Brain Region	Seq_ID	CpG	Chromosome	Position	Function	Flanking gene	Bio	P
BA11	1	cg16292768	8	27467783	Intron	CHRNA2	Neuronal acetylcholine receptor subunit alpha-2	.0000026
	2	cg05914894	11	36055687	Intron	LDLRAD3	Low density lipoprotein receptor class A domain containing 3	.0000161
	3	cg02508651	17	46604554	Non-coding RNA	NSF	Vesicle-fusing ATPase	.0000189
	4	cg26258213	10	127220321	Intron	DOCK1	Dedicator of cytokinesis protein 1	.0000199
	5	cg12734688	1	48308390	Intron	SPATA6	Spermatogenesis-associated protein 6	.0000296
	6	cg00083790	12	132944010	Intron	ZNF605	Zinc finger protein 605	.0000332
	7	cg02959939	21	47813025	—	—	—	.0000344
	8	cg25621518	5	6372336	Exon	MED10	Mediator of RNA polymerase II transcription subunit 10	.000066
	9	cg01761758	17	76850277	Non-protein coding RNA	LINC00868	—	.0000702
	10	cg18257839	15	63369284	Intron	CA12	Carbonic anhydrase 12	.0000736
	11	cg00152041	20	61274867	Intron	CDH4	Cadherin-4	.0000852
	12	cg11965936	11	69811698	Intron	FGF3	Fibroblast growth factor 3	.0000988
	13	cg07196992	4	78098654	Intron	FRAS1	Extracellular matrix protein FRAS1	.000100045
	14	cg01996297	6	33970166	—	—	—	.000100555
	15	cg24327877	6	131222276	Intron	AKAP7	A-kinase anchor protein 7 isoform gamma	.000100718
	16	cg00394844	19	18343302	Intron	PGPEP1	Pyroglutamyl-peptidase I	.000105844
	17	cg20704247	8	144409764	Intron	CPSF1	Cleavage and polyadenylation specificity factor subunit 1	.000108626
	18	cg16033053	10	28034794	Intron	ENSG00000233472	—	.000111832
	19	cg16300509	10	102280155	Intron	GBF1	Golgi-specific brefeldin A-resistance guanine nucleotide exchange factor 1	.000114731
	20	cg07033643	14	74420178	Intron	SYNDIG1L	Synapse differentiation-inducing gene protein 1-like	.000115666
BA25	1	cg09432202	6	163768551	Intron	ENSG00000235538	—	3.78E-05
	2	cg15212354	10	422330	Intron	DIP2C	Disco-interacting protein 2 homolog C	5.56E-05
	3	cg19781637	5	77935323	—	—	—	.000114315
	4	cg15643796	1	10421972	—	—	—	.000119521
	5	cg04882341	4	170679129	—	—	—	.000166889
	6	cg08169875	2	77340582	Intron	LRRTM4	Leucine-rich repeat transmembrane neuronal protein 4	.000169684
	7	cg03238595	19	44598377	—	—	—	.000172248
	8	cg24041637	1	233462711	—	—	—	.000210215
	9	cg10810860	8	144639591	Intron	ARHGAP39	Rho GTPase-activating protein 39	.000215068
	10	cg08389497	10	73820491	Non-coding RNA	CAMK2G	Calcium/calmodulin-dependent protein kinase type II subunit gamma	.000235872
	11	cg20526654	3	72495865	—	—	—	.000257424
	12	cg25202111	10	99519374	—	—	—	.000262818
	13	cg05653714	3	184349359	Intron	CLCN2	Chloride channel protein 2	.000281551
	14	cg12568756	2	218717561	Intron	TTLL4	Tubulin polyglutamylase TTLL4	.000292654
	15	cg05708074	12	122326393	Intron	CLIP1	CAP-Gly domain-containing linker protein 1	.000309344
	16	cg16765706	14	21109008	Intron	ENSG00000178107	—	.000343509
	17	cg26079320	1	166809407	—	—	—	.000360808
	18	cg08073312	3	137483555	Intron	SOX14	Transcription factor SOX-14	.000368467
	19	cg03691958	11	113345618	Intron	TTC12	Tetratricopeptide repeat protein 12	.00039959
	20	cg26116980	18	77170314	Intron	ENSG00000265844	—	.000418198

^a Chromosome based on GRCh38/hg38.

Genomic Correlation of Differentially Methylated Regions

The Bumphunter method identified 1 and 5 significantly DMRs (P < .05) between the depression with suicide and the normal control groups in BA11 and BA25, respectively (Table 3). The DMR in the BA11 region is located on chromosome 8 (143751796–143751801) and contains 2 CpG sites located in the gene region of IQANK1, which encodes the Q motif and the ankyrin repeat domain-containing protein 1. This gene was increased in BA11 tissues with depression and suicide compared with healthy controls (FC = 1.51). Interestingly, this DMR was also shared between the BA11 and BA25 regions. In the BA25 tissue, we identified 6 DMR regions, of which 3 were located on chromosome 8. The top 2 DMR regions were located on chromosomes 2 (30669597–30669863) (P = .0057) and 8 (2075209–2075777) (P = .0013). The former DMR is an intergenic gene whose downstream flanking gene is CAPN13, and the latter DMR is located in MYOM2.

Table 3.

Methylation Regions in the BA11 and BA25^a

Seq_ID	Chromosome	Start	End	Gene symbol	Value	Area	Cluster	Index start	Index end	L	Cluster L	P
BA11_1	chr8	143751796	143751801	IQANK1	−0.15793041	0.315860825	171507	207552	207553	2	10	.00906879
BA25_1	chr2	30669597	30669863	—	−0.13578541	0.407356232	93593	47009	47011	3	12	.005758724
BA25_2	chr8	2075209	2075777	MYOM2	−0.12100563	0.3630169	164336	192790	192792	3	4	.013387803
BA25_3	chr8	143751796	143751801	IQANK1	−0.13326407	0.26652813	171507	207552	207553	2	10	.014273761
BA25_4	chr8	1365049	1365659	DLGAP2	−0.10969437	0.329083121	164092	192111	192113	3	10	.022148939
BA25_5	chr13	43597657	43597736	ENOX1	−0.11820323	0.23640646	48434	290890	290891	2	6	.037702417

^a The bold characters represent overlapping regions between BA11 and BA25.

Integrated Analysis of Methylation and Transcriptomic Data

To more intuitively show the expression changes of the mRNA transcripts corresponding to the methylation sites, we mapped the chromosomal positions of the methylation sites and each transcript and determined the overlapping of chromosomal regions of the methylation sites and the transcripts. We calculated the correlation coefficient between the matched methylation levels and transcript expression levels. Here, we have shown highly correlated CpG sites that have the opposite correlation relationship and a larger difference in correlation coefficients in the case and control samples (Table 4). Fifteen methylation sites were filtered out in the BA25 part. The most significant methylation site, cg03462556, is located in the SLC16A9 gene, and the correlation coefficients in the MDD with suicide and healthy controls were 0.99 and −0.96, respectively, indicating that cg03462556 likely regulates the expression of the SLC16A9 gene directly and that the abnormal expression of the SLC16A9 gene could further mediate the occurrence of depression with suicidal behavior. For the BA11 region, we did not find any highly methylated sites that had a higher conversely correlated relationship between the suicide and control groups; cg17977470 in the TMEM88 gene had the highest difference in correlation coefficients (cases: r = −0.7; controls: r = 0.95).

Table 4.

Correlation Coefficients Between Methylation and Expression Levels of BA11 and BA25 Regions

Seq_ID	CpG	Gene_symbol	Ensembl_gene_id	Chromosome	Correlation_Control	P value_control	Correlation_MDD	P value_MDD
BA11_1	cg17977470	TMEM88	ENSG00000167874	17	0.948925289	.004740912	−0.65980585	.60000938
BA11_2	cg01290710	HIPK4	ENSG00000160396	19	−0.703659967	3.72E-08	0.855472097	6.89E-10
BA11_3	cg20037968	CARD11	ENSG00000198286	7	0.656621585	.369201006	−0.853767051	.00235136
BA11_4	cg12267236	PTGFRN	ENSG00000134247	1	0.855233761	.000623981	−0.639358679	1.83E-08
BA11_5	cg14092536	PTGES2	ENSG00000148334	9	0.8424328	.000541436	−0.648549089	5.90E-08
BA11_6	cg16903122	KLF17	ENSG00000171872	1	0.830488917	3.61E-08	−0.605647773	2.33E-11
BA11_7	cg06759255	TMEM88	ENSG00000167874	17	0.927817849	.076350645	−0.500603773	.12574782
BA11_8	cg15658543	CARD11	ENSG00000198286	7	0.6020029	.06829912	−0.784237835	.29764647
BA11_9	cg22863798	PTGES2	ENSG00000148334	9	−0.590679307	.000510823	0.773734462	5.56E-08
BA25_1	cg03462556	SLC16A9	ENSG00000165449	10	0.995372893	.032890446	−0.960859014	.00764052
BA25_2	cg10162673	ACAD9	ENSG00000177646	3	0.980216299	.000548044	−0.936690779	6.69E-06
BA25_3	cg22003284	KIF6	ENSG00000164627	6	−0.965002098	.001246527	0.919116769	.00012615
BA25_4	cg19535609	LRRC43	ENSG00000158113	12	−0.999999548	.007079487	0.876993627	8.55E-05
BA25_5	cg17334018	UNC5C	ENSG00000182168	4	0.886992089	.010077041	−0.981183168	1.88E-05
BA25_6	cg16755630	GABRB1	ENSG00000163288	4	0.952405279	.00253271	−0.912125836	1.67E-05
BA25_7	cg08860287	SLC7A5	ENSG00000103257	16	−0.999296624	.018135309	0.861405282	.00154594
BA25_8	cg04837071	NOXA1	ENSG00000188747	9	−0.98601094	.001183872	0.86074114	.00308855
BA25_9	cg09213964	LRRC43	ENSG00000158113	12	−0.92735654	.005969537	0.91375274	5.30E-05
BA25_10	cg20978193	SCNN1D	ENSG00000162572	1	−0.99237863	.107748625	0.847503548	.00108108
BA25_11	cg09552892	MMRN2	ENSG00000173269	10	0.896169973	.034142133	−0.938561759	3.22E-05
BA25_12	cg04738965	ZIC1	ENSG00000152977	3	0.99360233	.088212048	−0.821019255	.00071461
BA25_13	cg22797991	ATP10A	ENSG00000206190	15	−0.999800408	.018293452	0.806339301	5.93E-05
BA25_14	cg11110864	CERT1	ENSG00000113163	5	0.965953123	.000838712	−0.838449669	2.27E-06
BA25_15	cg05092932	SLC29A3	ENSG00000198246	10	−0.900490718	.005456505	0.895755569	5.45E-07

From the perspective of the effect of the extent of methylation on the expression level, we identified 85 CpG sites in BA25 that met the criteria for hypermethylation and 17 hypomethylation sites. After further investigation of the altered expression levels of these transcripts mapped by these hypermethylated CpG sites, 3 transcripts, ENSG00000139549, ENSG00000142185, and ENSG00000005379, were highly upregulated (FC > 1.5), and 5 transcripts were significantly downregulated (FC < 0.67). Two corresponding transcripts (ENSG00000269891 and ENSG00000134201) mapped with hypomethylated sites showed 2.4- and 2.0-fold change regulation in the brain tissues of suicide with depression, while the ENSG00000139537 transcript had a downregulated expression (FC = 0.39). No transcripts mapped with the 17 hypermethylation sites, and 15 hypomethylated sites reached a significantly different expression level in the brain tissues of suicide patients with depression compared with healthy controls screened in the BA11 region (supplemental Table 2).

Distribution on the Genomic Region of Those Highly Correlated CpG Sites

We further analyzed all the methylation sites with a correlation coefficient between methylation level and transcript expression greater than 0.8 and annotated the gene functional regions of these methylation sites. The results of annotated genomic functional regions of highly correlated CpG sites showed that the majority of these CpG sites were mainly distributed in the promoter and intron regions, with BA11 having 40 methylation sites in the intron region and 19 sites in the promoter region and BA25 having 591 CpG sites in the intron region and 461 sites in the promoter region (supplemental Table 3). We also determined the correlation coefficients of these methylated sites and found that 47.5% of the DNA methylation sites in the promoter region were negatively correlated with the corresponding transcript expression level. Approximately 47.2% of CpG sites within the intron regions were negatively correlated with mRNA expression, indicating that the positions of CpG sites in the gene regions were not associated with gene expression. To summarize the results of hypermethylated or hypomethylated CpG sites, gene expression is regulated not only by DNA methylation modification but also by other epigenetic modifications, including ncRNA, histone modification, and epitranscriptome modification.

Identification of Key Genes From Cross-Tissue and Cross-Omics Analysis and Pathway Enrichment Analysis

By analyzing the similarities and differences between DEGs and genes mapped using differentially methylated CpG sites, the results of the cross-omics study are shown in the Venn diagram (Fig. 2A and B). We found that 146 genes in BA11 and 1283 genes in the BA25 brain region showed both differentially methylated DNA and dysregulated mRNA expression. Further KEGG and functional annotation analysis of these genes showed that the genes in the BA11 region were mainly enriched in calmodulin-dependent protein kinase activity, calmodulin binding, and calcium channel activity, whereas the key genes in BA25 were mainly enriched in GTPase regulator activity, calcium ion transmembrane transporter activity, and calmodulin binding (Fig. 3).

Figure 2.

Overlapped differentially expressed and methylated genes in BA 11 and BA25 tissues. (A–B) The intersected transcripts sharing with transcriptome and methylation levels in BA11 and BA25 regions. (C) Seventeen key genes were identified across different omics and tissues.

Figure 3.

The key genes and enriched pathways from cross omics and tissue analysis. (A–D) GO enrichment analysis and KEGG pathway analysis of BA11 and BA25 regional differential transcripts. The results show that these transcripts are associated with ion channel receptor activities (red boxes).

Finally, we also found that there were 17 key genes shared with 2 regions of BA11 and BA25 (Fig. 2C), and the results of gene function further demonstrated that 6 out of 17 genes were directly associated with ion channel–related biological functions (Table 5).

Table 5.

GO and KEGG Pathway Analysis Showed that BA11 and BA25 Shared Genes^a

Seq_ID	GENE	ENSG	CpG	Description	Expression_BA11_FC	Top_Methylation_BA11_FC	Top_DMP_BA11_pval	DEG_BA11_pval	Expression_BA25_FC	Top_Methylation_BA25_FC	Top_DMP_BA25_pval	DEG_BA25_pval
1	G2AN	ENSG00000089597	cg05882782,cg25217772,cg25217772,cg08151654,cg26417427,cg19589057	Glucosidase II alpha subunit	1.0721	0.8062	0.0071	0.0422	0.8554	1.1990	0.1525	0.2839
2	KIF17	ENSG00000117245	cg03626238,cg14035553,cg14035553,cg05653329,cg10530135,cg27312359	Kinesin family member 17	1.3175	0.9933	0.0242	0.0148	1.8211	0.9860	0.3410	0.3862
3	CCNI	ENSG00000118816	cg17080371,cg16316472,cg16316472,cg02639808,cg10560514,cg17512133	Cyclin I	1.1700	0.9329	0.0168	0.0139	1.1881	0.9856	0.1684	0.2458
4	AKAP9	ENSG00000127914	cg02582136,cg08524932,cg08524932,cg22127703,cg00334498,cg00502697	A-kinase anchoring protein 9	0.8671	0.9361	0.0151	0.0086	0.9243	1.0023	0.7927	0.3496
5	CPLX2	ENSG00000145920	cg15622912,cg18328894,cg18328894,cg01252455,cg19885761,cg01938009	Complexin 2	1.2339	0.9533	0.0044	0.0333	1.5097	1.1729	0.2682	0.0376
6	PTGES2	ENSG00000148334	cg22863798,cg14092536,cg14092536,cg01311102,cg16020551,cg13833419	Prostaglandin E synthase 2	1.1576	1.0094	0.0113	0.0198	1.3177	1.0622	0.3050	0.1709
7	PHKG2	ENSG00000156873	cg02091607,cg07029862,cg07029862,cg15970621,cg00027499,cg00467202	Phosphorylase kinase catalytic subunit gamma 2	1.1703	1.0035	0.0467	0.0418	1.2511	0.9907	0.4781	0.2859
8	FOXH1	ENSG00000160973	cg19792544,cg01906600,cg01906600,cg00845765,cg13623135,cg22340053	Forkhead box H1	2.2476	0.9684	0.0249	0.0212	3.3984	0.9822	0.4371	0.0814
9	LOXHD1	ENSG00000167210	cg07931960,cg16481525,cg16481525,cg14850601,cg14165909,cg15132710	Lipoxygenase homology PLAT domains 1	1.9221	0.9755	0.0468	0.0219	2.1435	1.0622	0.0217	0.2604
10	LRP1B	ENSG00000168702	cg10499042,cg21340845,cg21340845,cg26413307,cg21484213,cg09618674	LDL receptor related protein 1B	0.8386	0.9897	0.0115	0.0120	1.2359	0.9905	0.5336	0.2740
11	IRX2	ENSG00000170561	cg05659097,cg15941948,cg15941948,cg05903444,cg23164183,cg09524455	Iroquois homeobox 2	0.1018	1.0616	0.0227	0.0089	0.2731	1.0050	0.2199	0.3988
12	CHRNA7	ENSG00000175344	cg04785227,cg20861607,cg20861607,cg21883683,cg23854667,cg08481732	Cholinergic receptor nicotinic alpha 7 subunit	1.5010	0.8955	0.0379	0.0319	1.4663	0.8666	0.6103	0.3852
13	LRRC75A	ENSG00000181350	cg22912701,cg10498052,cg10498052,cg18017908,cg08783616,cg26744387	Leucine rich repeat containing 75A	1.3350	0.9841	0.0236	0.0170	1.5347	0.9936	0.9896	0.1520
14	CAMK1D	ENSG00000183049	cg20683445,cg12768145,cg12768145,cg12273284,cg09926867,cg26169081	Calcium/calmodulin dependent protein kinase ID	1.2527	0.9633	0.0357	0.0238	1.5639	1.0799	0.9687	0.2604
15	SCN5A	ENSG00000183873	cg16874743,cg00378492,cg00378492,cg18800918,cg12926589,cg09290614	sodium voltage-gated channel alpha subunit 5	1.7448	0.9961	0.0180	0.0360	0.2746	0.9914	0.2656	0.3847
16	C1QTNF12	ENSG00000184163	cg00211609,cg15582176,cg15582176,cg13440692,cg10208113,cg25594899	C1q and TNF related 12	6.4427	0.8108	0.0174	0.0449	Inf	1.0319	0.7701	0.2233
17	GRM7	ENSG00000196277	cg04023483,cg21032008,cg21032008,cg23055496,cg13231680,cg18399935	Glutamate metabotropic receptor 7	1.1591	0.7147	0.0125	0.0280	1.4010	0.9774	0.4663	0.4068

^a Bold characters represented genes associated with ion channels.

Cell Type Transcriptomic Changes From Deconvolution Analysis

To explore the cell types associated with suicide and depression, we deconvoluted the bulk mRNA-seq of test samples from single-cell transcriptome data. With the help of Seruat and MuSiC2 software packages, we found that among 7 broad cell types, OPC cells have the most significant difference of cell type ratios in the suicide group compared with the control (P = .00324), indicating that suicidal behavior may have some links with cell-type transcriptome changes with OPCs (Fig. 4). In addition, we identified 312 differential genes associated with cell type specificity (supplementary Table 4). Further functional annotation analysis found that these genes were mainly enriched in glutamatergic system–related signaling pathways, such as GABAergic synapses, cGMP-PKG signaling pathways, and glutamatergic synapses (see Fig. 5 for details). Among them, studies have shown that MAPK1, GABRB1, SLC1A3, and other genes in this pathway are related to depression. In addition, we analyzed the cell-type enrichment of 17 genes obtained from cross-omics and cross-tissue analysis and found that 13 genes, including KIF17, CCNI, AKAP9, CAMK1D, and GRM7, were significantly enriched in OPCs cells.

Figure 4.

Overview of cell types characterized in the dlPFC. (A) TSNE plot colored by the broad cell types. (B) UMAP plot colored by the broad cell types.

Figure 5.

Functional analysis of key genes and differential cell types for bulk RNA deconvolution analysis. (A–D) The proportion of cells and different cell types in bulk RNA deconvolution analysis. (E–F) GO enrichment analysis and KEGG pathway analysis of bulk RNA deconvolution analysis.

Cell Type Analysis of Methylated Sites With Deconvolution Analysis

Based on the significant methylation sites obtained from the correlated analysis, we further explored the cell-type–specific distribution of these methylation sites. Similar to the deconvolution analysis of bulk mRNA-seq, we have found that these significant changes of cell type proportions present in endothelial (P = 1.94E-20), excitatory neurons (P = 3.51E-12), OPCs (P = 1.51E-10), astrocytes (P = 2.96E-8), oligodendrocytes (P = 1.66E-7), and macrophage/microglia (P = .00057) from statistical analysis. At the same time, 1095 differential genes related to cell type specificity were identified (supplementary Table 5). Further functional enrichment analysis also demonstrated that these methylated genes associated with cell type were mainly enriched in the dopamine signaling pathway, such as Hippo signaling pathway, Dopaminergic synapse, and Wnt signaling pathway (see Fig. 6 for details).

Figure 6.

Functional analysis of key genes and differential cell types for methylation deconvolution analysis. (A–G) The proportion of cells and different cell types in methylation deconvolution analysis. (H–I) GO enrichment analysis and KEGG pathway analysis of methylation deconvolution analysis.

Key Genes and Pathways Analyzed From Data of Antidepressant Treatment

To further understand the effect on antidepressant treatment, we analyzed single-cell transcriptional data for genes in the dorDG and venDG brain regions. Our analysis revealed 1405 cell- and drug-related differential genes in the dorDG region and 1362 in the venDG region. Of these genes, 118 transcripts were found to be overlapping genes sharing 2 brain regions, including NPW, Qdpr, and Aldoc (Fig. 7A). The biological annotation analysis showed that these overlapping transcripts were mainly enriched in pathways related to calcium-dependent exocytosis, calcium-dependent phospholipid binding, ion channel complexes, and glutamate receptor activity (Fig. 7B and C). Our findings confirm our previous results and suggest that these pathways may play a crucial role in the development of depression.

Figure 7.

Key genes and enrichment pathways for cross-tissue analysis of drug single-cell data. (A) The intersected genes of drug single-cell data in dorDG and venDG regions. (B–C) GO enrichment analysis and KEGG pathway analysis of dorDG and venDG regional differential genes. The results show that these genes are associated with ion channel receptor activities (red boxes).

Discussion

High-throughput technologies, including microarray and next-generation sequencing, provide a hypothesis-free approach to investigate the expression and methylation levels of the whole genome and to identify changes in transcripts and CpG methylated sites (Grubb and Caliari, 2021; Yang et al., 2021a, 2021b). Genome-wide expression analysis and DNA methylation profiles have been conducted using postmortem brain samples of suicide completers and suicides with psychiatric disorders from different cohorts (Fagerberg et al., 2014; Edwards et al., 2017). Some novel genes, such as PSORS1C3, TABP, ATP5G2, and POU5F1, have been found to be differentially methylated in the orbitofrontal cortex and anterior cingulate cortex (Fiori and Turecki, 2020). The limited availability of postmortem brain samples remains a challenge with a genome-wide strategy for neurobiological studies of suicidal behavior. Thus, we conducted a series of integrative multi-omics data analyses to identify potential key genes shared with different omics layers and/or tissues. To perform an integrated analysis of transcriptional and methylation data in the present study, we paired 2 sets of separate transcriptomic and methylomic data together from the same postmortem brain tissues and then mapped each significant CpG site into the corresponding transcripts based on their unified corresponding chromosome positions and further correlated the transcriptional and methylation levels from paired data sets.

From the correlation analysis, several genes, such as GABRB1 encoding gamma-aminobutyric acid receptor and solute carrier family genes (SLC16A9 and SLC7A5), have been found to be direct evidence in previously published studies and have also shown from current study the opposite relationship of correlated level in the case and control groups (Patel et al., 2021; Su et al., 2022).

From cross-omics and cross-tissue analyses, a total of 17 genes have been identified to share with transcription, methylation, and different tissues. Interestingly, 6 genes, including G2AN, IRX2, CHRNA7, CAMK1D, SCN5A, and GRM7, encode proteins associated with glutamine receptors and ion channels. Meanwhile, the enriched pathways of genes from the cross-omics analysis were related to calcium ion channel–related activities in BA11 and BA25. Thus, altered mRNA expression of ion channel–related genes regulated by CpG methylation in the brain directly mediates the development of depression and suicide, and these receptor-related genes may provide potential targets for treating depression.

Previous studies have shown that DNA methylation in different genomic regions affects different gene functions based on gene flanking sequences. We explored the gene distribution of methylation sites that were highly correlated with expression, and our results showed that not all the methylation sites in promoter regions were negatively correlated with mRNA gene expression; the methylation sites within the gene body were also not completely positively correlated with expression. These results indicate that DNA methylation of the gene body, in addition to promoter regions, may also contribute to gene expression regulation in the brain, which is consistent with some previous reports (Aapola et al., 2000; Guo et al., 2011a, 2011b; Xie et al., 2012).

Our results of the hypermethylated and hypomethylated sites have demonstrated that gene expression dysregulation in the brain is not directly associated with methylation levels of the methylated islands. Gene expression abnormalities in the disease state might be caused by multiple epigenetic modifications, including DNA methylation, microRNAs, and histone modifications. Recently, Lutz et al. (2021) also reported that histone marks, as well as DNA methylation sites, together with transcriptomic changes have consistently presented across multiple epigenetic mechanisms in subjects with a history of ELA and controls (Lutz et al., 2021). Accumulating evidence from such studies indicates that complex mechanisms of genomic regulation may contribute to the disease and healthy state of the brain.

Deconvolution of mRNA and methylation data showed that OPCs have the changes of cell type specificity in suicide brain tissues, and the expression of 13 genes obtained by cross-omics and cross-tissue integration analysis also have been found distributed into OPCs. Therefore, we replicated the results reported by Nagy et al. (2020) with an independent sample cohort (Nagy et al., 2020). Currently, it was found that loss of OPCs is directly associated with the emergence of depression-like behaviors (Birey et al., 2015). For example, chronic psychosocial stress can lead to chronic loss and transient proliferation of OPCs, abnormal differentiation of oligodendrocytes, and severe hypomyelination in a mouse model of RSDS, associated with the combined stress response of depression (Kokkosis et al., 2022). Du et al. (2019) showed that minocycline can exert antidepressant effects by inhibiting microglial activation, promoting OPCs maturation and remyelination (Du et al., 2019). Additionally, OPCs are now recognized as a distinct glial cell type implicated in brain plasticity through integration of synaptic activity and mediation of long-term potentiation (Ge et al., 2006; Birey et al., 2017). Combined with the results of key genes enrichment analysis, we speculated that suicide accompanied by depression may be related to the abnormal expression of glutamatergic receptors and ion channel–related signaling pathways in OPCs, thereby changing synaptic tactile activity and brain plasticity.

Our analysis of drug single-cell transcriptional data indicates that differential genes affected by drugs are mainly enriched in pathways related to calcium ion channels, glutamate nerve energy, and GABA neurons. Studies have shown that the upregulation of calcium/calmodulin-dependent protein kinase type II in the lateral habenula of a depression animal model is reversed by antidepressants. Additionally, it can reverse depressive symptoms by downregulating calcium/calmodulin-dependent protein kinase type II levels, blocking its activity, or its target molecule glutamate receptor GluR1 (Li et al., 2013). Tang et al. (2020) discovered that the hyperactivation of extrasynaptic GluN2B receptors is associated with the antidepressant effects of ketamine and that the interaction between GluN2B and calcium/calmodulin-dependent protein kinase IIα is crucial for GluN2B localization and activity (Tang et al., 2020). Our findings suggest that calcium ion channels and glutamatergic nerves may be potential pathways for antidepressants to exert their efficacy, which supports the results of the cross-omics integration analysis in the previous article.

In recent years, several single-omics data have been generated from different study groups (Schubert et al., 2018; Barh et al., 2021; Yao et al., 2021; Li et al., 2022). Through the integration of different omics data, it is possible to detect and discover new biomarkers and targets, which are difficult to find with single-omics data, and cross-omics data have been applied to the study of emotional disorders. For example, Ju et al. (2019) conducted integration analysis of transcriptomics and methylation data in the MDD patients and successfully mined out 1 methylated site located in the CHN2 gene, which can be a possible candidate biomarker of antidepressant treatment response (Ju et al., 2019). Regarding multi-omics research methods, in addition to conventional regression analysis, overlapping analysis, and correlation analysis, advanced analysis methods are emerging, such as the deconvolution analysis method used in this study, and machine learning methods that have become more popular in recent years (Wang et al., 2018). The field of multi-omics analysis holds immense potential for development as well as significant challenges: (1) data availability: published omics data are stored in public and free databases, such as the TCGA database in the field of tumor research, which is convenient for researchers to obtain and share; (2) develop more accurate algorithms and software, disclose detailed analysis procedures, and promote the progress of molecular psychiatry; (3) generation of multi-omics datasets from more individuals, including clinical and phenotypic data.

In conclusion, our study identified some important dysregulated genes influenced by DNA methylation in 2 brain regions of depression and suicide patients. Interestingly, we found that OPCs have the most contributors for cell-type proportions related to differential expression genes and methylated sites in suicidal behavior, which replicated the results in the previously published studies on single-nucleus RNA-sequencing. This study provides insights into key cell-type transcripts and CpG biomarkers for the treatment of MDD with suicide.

Interest Statement

The authors have no conflict interest.

Ethics Statement

Anonymized data were collected from a publicly available database and did not require ethics committee approval.

Data Availability

The authors declare that all data supporting the findings of this study are available within the paper and its supplementary information files.

Author Contributions

Yalan Zhou (Data curation [Equal], Writing—original draft [Equal]), lan Xiong (Data curation [Equal], Formal analysis [Equal], Writing—original draft [Equal]), Jianhua Chen (Data curation [Equal], Formal analysis [Equal], Writing—original draft [Equal]), and Qingzhong Wang (Data curation [Equal], Formal analysis [Equal], Methodology [Equal], Writing—original draft [Equal])