Identification of novel biomarkers for intracerebral hemorrhage via long noncoding RNA-associated competing endogenous RNA network

Abstract

Intracerebral hemorrhage (ICH) is a leading cause of death and disability worldwide. This study aimed to examine the involvement of long non-coding RNAs (lncRNAs), a group of non-coding transcripts, in ICH as potential biomarkers. An expression profile of patients with ICH using four contralateral grey matter controls (GM) and four contralateral white matter controls (WM) was downloaded from the Gene Expression Omnibus (GEO) database. Co-expressed lncRNAs and mRNAs were selected to create competing endogenous RNA (ceRNA) networks. Key lncRNAs were identified in ceRNA networks, which were validated through Real-time qPCR (RT-qPCR) with peripheral blood samples from patients with ICH. A total of 49 differentially expressed lncRNAs were discovered in different brain regions. The ceRNA network in GM included 9 lncRNAs, 40 mRNAs, and 20 microRNAs (miRNAs), while the one in WM covered 6 lncRNAs, 25 mRNAs, and 14 miRNAs. Six hub lncRNAs were observed and RT-qPCR results showed that LY86-AS1, DLX6-AS1, RRN3P2, and CRNDE were down-regulated, while HCP5 and MIAT were up-regulated in patients with ICH. Receiver Operating Characteristic (ROC) assessments demonstrated the diagnostic value of these lncRNAs. Our findings highlight the potential roles of lncRNA in ICH pathogenesis. Moreover, the hub lncRNAs discovered here might become novel biomarkers and promising targets for ICH drug development.

---

Introduction

Intracerebral hemorrhage (ICH) is the second most predominant subtype of stroke, which is a major threat to public health worldwide due to high morbidity and mortality.¹ Bleeding in the brain can lead to serious damage, affecting several biological processes such as inflammation, apoptosis, and autophagy.² The etiology of ICH is complicated and has not been fully elucidated. The ideal therapeutic strategy for ICH remains challenging.³ Patients with ICH or survivors frequently suffer from catastrophic disabilities due to disease and a heavy economic burden is imposed on families or society. Hence, the identification of novel biomarkers as diagnostic indicators and therapeutic targets is urgent for ICH management.

Long non-coding RNAs (LncRNAs) refer to a class of non-coding (ncRNA) transcripts with a size of more than 200 nucleotides. As an essential regulator, lncRNA enables the promotion or suppression of gene expression via methods including direct coupling with the target genes or recruiting transcription factors.^4,5 Moreover, based on the competing endogenous RNA (ceRNA) hypothesis, lncRNA can serve as a “microRNA (miRNA) sponge” for modulating mRNA levels.⁶ Currently, it is garnering significant interest because of its indispensable role in various physiological and pathological processes. The brain exhibits an abundance of lncRNA, implying its potential ability to govern molecular cascades in the neurological system. Dysfunctional lncRNAs are involved in brain disorders, ranging from brain tumors to neurodegenerative disorders.^7,8 Abnormal expression of lncRNAs has been demonstrated in two ICH rodent models,^9,10 suggesting lncRNA's participation in ICH onset and development. Latest research reported that lncRNA Blnc1 knock-out could relieved nerve injury caused by ICH.¹¹ Silencing lncRNA H19 would suppress ferroptosis observed in brain microvascular endothelial cells from ICH experimental model.¹² LncRNAs MEG3 and Mtss1 were both linked to the inflammatory response after ICH occurrence.^13,14 However, so far, the evidence as to the lncRNA-associated mechinsm is still limited and the comprehensive understanding of lncRNA's function in ICH remains elusive.

Here, with a combination of bioinformatics analysis and experimental validation, we aimed to prospectively provide a systematic dissection of the molecular mechanisms of ICH with respect to the role of lncRNA. We also intended to examine the feasibility of hub lncRNAs as biomarkers for diagnosis, and as promising drug targets of ICH.

Materials and methods

Raw data pre-processing and probe reannotation

Primitive data of GSE24265³ were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/gds/). It contains tissue samples corresponding to four perihematomal brain tissues (PH), four contralateral grey matter controls (GM), and four contralateral white matter controls (WM) obtained from patients with ICH. The platform of the dataset was “GPL570, [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array”. The mRNA and lncRNA were re-annotated using the Encyclopedia of DNA Elements (ENCODE) with ENSEMBL.^15,16

Identification of and differentially expressed (DE) lncRNAs and DEmRNAs

After data normalization and pre-processing with the “Impute” package in “R”, “Limma” package of “R” language (Version X64 3.6.1) was used to detect DEmRNAs and DElncRNAs between PH and GM and PH and WM groups. The software calculated P-value and false discovery rate (FDR) by the student's t-test and the Benjamini–Hochberg method, respectively. |Log2FoldChange (FC)| > 1.0 and P-value < 0.05 were finally set as the cut-off criteria. Probes with no exact corresponding gene symbol were filtered. The results of DElncRNAs were visualized through the “Pheatmap” package.

LncRNA–mRNA co-expression network (CEN) construction

The Pearson correlation coefficient (PCC) was calculated to identify co-expression relationships between DElncRNAs and DEmRNAs. The lncRNA–mRNA pairs showing absolute values of more than 0.95 for PCC and those with P < 0.05, were selected to construct the CENs, which were visualized using the Cytoscape software (Version 3.7.1).

Functional assessments of mRNAs in the CENs

Genome Ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were implemented using an online tool: Database for Annotation, Visualization and Integration Discovery (DAVID, https://david.ncifcrf.gov). The species was limited to Homo sapiens, and values with P < 0.05 were considered significant. The results were visualized using the “GGplot2” package of R.

Construction of LncRNA-associated ceRNA network

To further probe the regulatory role of lncRNAs, the interactions between DElncRNAs and miRNAs were predicted. The DElncRNAs in PH vs. GM and PH vs. WM groups were characterized for predicting targeted miRNAs using the mircode database (http://www.mircode.org/, approximately 770 lncRNA–miRNA pairs). The putative miRNA–mRNA interactions were identified using miRTarBase (Version 8.0, http://miRTarBase.cuhk.edu.cn/, approximately 322135 miRNA–mRNA pairs), miRDB (Version 6.0, http://www.mirdb.org/, approximately 938564 miRNA–mRNA pairs) and TargetScan (Version 7.2, http://www.targetscan.org/, approximately 1048576 miRNA–mRNA pairs). Genes identified in all three databases were intersected with the DEmRNAs positively correlated with DElncRNAs to recognize the final mRNA targets. The lncRNA–miRNA–mRNA triplet consisted of DElncRNA, predicted miRNA and final mRNA target. A regulatory network was then established and visualized via Cytoscape (Version 3.7.1).

Topological analysis of ceRNA network

The degree of each node in the ceRNA network was assessed using the “Centiscape2.2” plug-in of Cytoscape. The top 10 nodes with the highest node degree were extracted to create sub-networks and visualized using the “cytoHubba” plug-in. The lncRNA with a node degree > 5 was considered as a hub lncRNA. The information regarding the relationships of hub lncRNA–miRNAs–mRNAs was then collected to generate hub lncRNA-associated sub-networks.

Patients and blood samples

Six hub lncRNAs in the two ceRNA networks were further validated using Real-time qPCR (RT-qPCR). A total of 33 patients with ICH from the Department of Neurology at the First Affiliated Hospital of China Medical University and 15 control subjects were enrolled for this study. The inclusion criteria are described in our previous manuscript.¹⁷ The peripheral blood was collected for RNA isolation and subsequent analysis. This study was approved by the Ethics Committee of the First Affiliated Hospital of China Medical University, Shenyang, Liaoning Province, China (no. [2012]38-1). All participants provided written informed consent according to the Declaration of Helsinki (as revised in 2013).

RT-qPCR

Total RNA was extracted from the blood samples using TRIzol reagent (Thermo Fisher Scientific, USA, 15596026) according to the manufacturer's instructions. First-strand cDNA was synthesized using a Revert Aid First Strand cDNA Synthesis Kit (Jizhen biology, China, AORT-0060). Resultant cDNA was added as a template, and the amplification reactions were performed using AceQ qPCR SYBR Green Master Mix (Jizhen Biology, China, Q121-02) on Bio-Rad CXF96 PCR detection system (Bio-Rad, Hercules, CA, USA). GAPDH was employed as an endogenous control. The relative gene expression levels were calculated using the comparative cycle threshold (2−ΔΔCt) method. The primer sequences are listed in Table S1 (ESI†). Area under the curve (AUC) of the Receiver Operating Characteristic (ROC) was calculated via GraphPad Prism 7.0. P < 0.05 with the AUC > 0.80 were set as the cutoff standards.

Statistical analysis

For RT-qPCR, all values are presented as mean ± SEM. The student's t-test was used to evaluate the differences between the ICH and the control groups, and statistical significance was set at P < 0.05. All statistical analyses were conducted using SPSS for Windows version 16.0 and GraphPad Prism 7.0.

Results

Differentially expressed lncRNAs

Since the Affymetrix HG-U133 Plus 2.0 array consists of probe set-targeted lncRNA genes,¹⁵ we acquired DElncRNAs in patients with ICH via probe reannotation. Based on our cut-off criteria, a volcano diagram shown in Fig. 1a was constructed. There are a total of 1316 DEgenes in the PH compared to those in the GM, in which 477 up-regulated genes, including 11 lncRNAs, and 839 down-regulated genes, including 15 lncRNAs, were observed. Fig. 1b shows a total of 790 DE genes discovered in the PH compared to those in the WM, in which 391 up-regulated genes, including 4 lncRNAs, and 399 down-regulated genes, including 22 lncRNAs, were observed. The heatmaps shown in Fig. 1c and d display all DElncRNAs in PH vs. GM and PH vs. WM groups.

Fig. 1 Differentially expressed genes in ICH. (a) The volcano plot of the differently expressed genes in PH vs. GM of GSE24265. (b) The volcano plot of the differently expressed genes in PH vs. WM of GSE24265. Black dots represent genes which are not altered. The red dots represent upregulated genes while the green dots represent downregulated ones. (c) Heatmap of differently expressed lncRNAs in PH vs. GM of GSE24265. (d) Heatmap of differently expressed lncRNAs in PH vs. WM of GSE24265.

Co-expression of DElncRNAs and DEmRNAs

To further explore the potential role of the DElncRNAs in ICH development, DElncRNA–DEmRNA networks were constructed based on correlation analysis.^18,19 Only strong correlations with |PCC| with a value more than 0.95 along with a P-value of less than 0.05, were selected to construct the network. The GM-associated CEN included 24 lncRNAs and 635 mRNAs, exhibiting 1230 co-expression relationships, whereas the WM-associated CEN included 26 lncRNAs and 353 mRNAs, exhibiting 619 co-expression relationships. Fig. 2a and b. demonstrate the CENs in the GM and the WM, respectively.

Fig. 2 The co-expression networks. (a) Co-expression relationship between DElncRNAs and DEmRNAs in PH vs. GM. (b) Co-expression relationship between DElncRNAs and DEmRNAs in PH vs. WM. Red V shapes indicate lncRNAs and blue ellipses indicate mRNAs. Gray edges indicate lncRNA–mRNA interactions.

Functional annotation of mRNAs in CENs

To elucidate the molecular functions of the lncRNAs, GO enrichment and KEGG pathway analyses focusing on mRNA in CENs were employed. Complete lists of the outcomes are presented in Tables S2–S5 (ESI†). The bar plot shown in Fig. 3a and Table 1 demonstrates that the mRNAs in the GM-related CEN were associated with “chemical synaptic transmission”, “cell junction”, and “syntaxin-1 binding”. The KEGG results shown in Fig. 3b demonstrate the enrichment of these mRNAs primarily in “Axon guidance”. As shown in Fig. 3c and Table 2, “positive regulation of cell migration”, “microvillus”, and “cytoskeletal protein binding” were all involved in biological processes associated with the WM-related CEN mRNAs. Fig. 3d shows the enrichment of these mRNAs in the KEGG pathways, including “metabolic pathways”.

Fig. 3 Functional analyses of the mRNA in CENs. Each category only presented top 10 terms (P < 0.05). (a) GO enrichment analysis of mRNA in GM-associated CEN. (b) KEGG pathway analysis of mRNA in GM-associated CEN. (c) GO enrichment analysis of mRNA in WM-associated CEN. (d) KEGG pathway analysis of mRNA in WM-associated CEN.

Table 1 Functional enrichments analysis of mRNAs in GM-related CEN

Functional enrichments	Terms	Description	P-value
GO_BP	GO:0007268	Chemical synaptic transmission	8.25 × 10^−16
	GO:0007399	Nervous system development	1.03 × 10^−7
	GO:0034765	Regulation of ion transmembrane transport	3.87 × 10^−6
	GO:0007269	Neurotransmitter secretion	5.58 × 10^−6
	GO:0007528	Neuromuscular junction development	2.75 × 10^−5
	GO:0014047	Glutamate secretion	2.75 × 10^−5
	GO:0035584	Calcium-mediated signaling using intracellular calcium source	3.38 × 10^−5
	GO:0071805	Potassium ion transmembrane transport	4.82 × 10^−5
	GO:0060078	Regulation of postsynaptic membrane potential	6.15 × 10^−5
	GO:0051899	Membrane depolarization	1.34 × 10^−4
GO_CC	GO:0030054	Cell junction	2.37 × 10^−24
	GO:0014069	Postsynaptic density	2.87 × 10^−17
	GO:0005886	Plasma membrane	7.11 × 10^−16
	GO:0045211	Postsynaptic membrane	1.69 × 10^−14
	GO:0043005	Neuron projection	6.59 × 10^−11
	GO:0045202	Synapse	2.94 × 10^−10
	GO:0030425	Dendrite	1.98 × 10^−09
	GO:0030424	Axon	9.67 × 10^−08
	GO:0008076	Voltage-gated potassium channel complex	1.78 × 10^−07
	GO:0048786	Presynaptic active zone	2.48 × 10^−7
GO_MF	GO:0017075	Syntaxin-1 binding	9.28 × 10^−7
	GO:0005509	Calcium ion binding	5.42 × 10^−6
	GO:0030276	Clathrin binding	2.73 × 10^−4
	GO:0005248	Voltage-gated sodium channel activity	3.88 × 10^−4
	GO:0005249	Voltage-gated potassium channel activity	5.85 × 10^−4
	GO:0005515	Protein binding	0.001334181
	GO:0005245	Voltage-gated calcium channel activity	0.001822166
	GO:0015459	Potassium channel regulator activity	0.001822166
	GO:0004890	GABA-A receptor activity	0.003022365
	GO:0004871	Signal transducer activity	0.003121369
KEGG	hsa04360	Axon guidance	5.20 × 10^−7
	hsa05033	Nicotine addiction	8.84 × 10^−6
	hsa04727	GABAergic synapse	8.86 × 10^−6
	hsa04723	Retrograde endocannabinoid signaling	1.29 × 10^−5
	hsa04725	Cholinergic synapse	3.84 × 10^−5
	hsa04728	Dopaminergic synapse	4.87 × 10^−5
	hsa05032	Morphine addiction	3.81 × 10^−4
	hsa04724	Glutamatergic synapse	7.51 × 10^−4
	hsa04080	Neuroactive ligand-receptor interaction	0.002170521
	hsa04921	Oxytocin signaling pathway	0.002762355

---

Table 2 Functional enrichments analysis of mRNAs in WM-related CEN

Functional enrichments	Terms	Description	P value
GO_BP	GO:0030335	Positive regulation of cell migration	0.007315187
	GO:0043086	Negative regulation of catalytic activity	0.012517416
	GO:0048146	Positive regulation of fibroblast proliferation	0.017480309
	GO:0006096	Glycolytic process	0.024449498
	GO:0030833	Regulation of actin filament polymerization	0.026705051
	GO:0009611	Response to wounding	0.028977114
	GO:0060326	Cell chemotaxis	0.032017746
	GO:0009440	Cyanate catabolic process	0.036582608
	GO:0009749	Response to glucose	0.036916882
	GO:0035556	Intracellular signal transduction	0.037199119
GO_CC	GO:0005902	Microvillus	5.10 × 10^−4
	GO:0070062	Extracellular exosome	7.01 × 10^−4
	GO:0005737	Cytoplasm	0.015363395
	GO:0030424	Axon	0.018197632
	GO:0042470	Melanosome	0.034243687
	GO:0005634	Nucleus	0.041779378
	GO:0045177	Apical part of cell	0.04458957
	GO:0005768	Endosome	0.048244172
GO_MF	GO:0008092	Cytoskeletal protein binding	0.001713596
	GO:0004536	Deoxyribonuclease activity	0.013387921
	GO:0042802	Identical protein binding	0.032099848
	GO:0004572	Mannosyl-oligosaccharide 1,3-1,6-alpha-mannosidase activity	0.035926247
	GO:0004674	Protein serine/threonine kinase activity	0.042230075
KEGG	hsa01100	Metabolic pathways	0.003198261
	hsa00270	Cysteine and methionine metabolism	0.031772014
	hsa04915	Estrogen signaling pathway	0.034758077
	hsa01230	Biosynthesis of amino acids	0.042510774

---

Construction and topological analysis of ceRNA network

We next attempted to elucidate the potential regulatory mechanism of these lncRNAs in ICH We selected DEmRNAs that were positively correlated with DElncRNAs and constructed the lncRNA-associated ceRNA network. The Venn diagrams shown in Fig. 4a and d indicate the final number of mRNA targets for the networks. For “PH vs. GM”, the ceRNA network was composed of 9 lncRNAs, 40 mRNAs, 20 microRNAs (miRNAs), and 128 edges, shown in Fig. 4b and Table 1. For in “PH vs. WM”, the ceRNA network was composed of 6 lncRNAs, 25 mRNAs, 14 miRNAs, and 71 edges, shown in Fig. 4e and Table 3. We further assessed the topological features of each gene in the ceRNA network. As shown in Table 4, in the GM-related ceRNA network, the degree scores of five lncRNAs were over five, while WM-related ceRNA network showed two lncRNAs with degree scores of over five. Fig. 4c and f show the top 10 nodes with high degree scores in the two ceRNA networks. Interestingly, MIAT appeared in both lists of key nodes, implying its significance. Finally, Fig. 5a–g. display the hub lncRNA-associated subnetworks in GM and WM.

Fig. 4 The lncRNA-associated ceRNA network. (a) Venn diagram of mRNA targets in GM-related ceRNA network. (b) GM-related ceRNA network. (c) Network of the top 10 hub nodes in GM. (d) Venn diagram of mRNA targets in WM-related ceRNA network. (e) WM-related ceRNA network. (f) Network of the top 10 hub nodes in WM; Red V shapes indicate lncRNAs. Yellow triangles indicate miRNAs. Blue ellipses indicate mRNAs. Gray edges indicate lncRNA–miRNA–mRNA interactions. Different colors represent the different degrees of nodes.

Table 3 The details of the ceRNA networks

	GM-related ceRNA	WM-related ceRNA
lncRNAs	RRN3P2, LY86-AS1, MIAT, LINC00323, ALMS1-IT1, DLX6-AS1, JAZF1-AS1, CRNDE, LINC00507	HCP5, MIAT, RBM26-AS1, AC091133, FTX, EPN2-AS1
miRNAs	hsa-miR-10a-5p, hsa-miR-17-5p, hsa-miR-27a-3p, hsa-miR-590-5p, hsa-miR-137, hsa-miR-363-3p, hsa-miR-301b-3p, hsa-miR-107, hsa-miR-429, hsa-miR-20b-5p, hsa-miR-23b-3p, hsa-miR-212-3p, hsa-miR-129-5p, hsa-miR-490-3p, hsa-miR-193a-3p, hsa-miR-24-3p, hsa-miR-1297, hsa-miR-449c-5p, hsa-miR-125b-5p, hsa-miR-375	hsa-miR-375, hsa-miR-17-5p, hsa-miR-20b-5p, hsa-miR-22-3p, hsa-miR-23b-3p, hsa-miR-125b-5p, hsa-miR-107, hsa-miR-125a-5p, hsa-miR-27a-3p, hsa-miR-490-3p, hsa-miR-425-5p, hsa-miR-217, hsa-miR-301b-3p, hsa-miR-363-3p, hsa-miR-24-3p, hsa-miR-129-5p, hsa-miR-216b-5p, hsa-miR-507, hsa-miR-212-3p, hsa-miR-140-5p, hsa-miR-1297, hsa-miR-429, hsa-miR-590-5p, hsa-miR-876-3p, hsa-miR-10a-5p
mRNAs	CSRNP3, ELAVL2,TMEM170B, AGO2, FMNL2, MOAP1, FAXC, MAP3K9, CPEB3, SLC12A5, ITPR1, CUL3, ATP13A3, PSD3, ZNF280B, GNB5, EPHA4, TNFAIP3, BMPER, YBX1, UNC5D, CRY2, CELF2, LHFPL4, RUNX1T1, BCL11B, NEURL1B, WDR82, BLCAP, STAT3, REEP1, KLHDC3, CBX6, LITAF, OTUD7B, GABBR2, FNIP1, ELAVL4, DDN, TUBB2A	ELAVL4, CPOX, ANKIB1, PTPDC1, ZNF532, LPGAT1, RUFY2, PIP4K2A, SP1, BRWD1, TMEM136, PGM2L1, SUN2, KMT2C, DDX5, RBPJ, BCOR, ERBB3, FOXO3, UBN2, KIAA1109, AMOTL2, DPYSL2, EXPH5, CEP97, ZNF566, GAB1, BACH2, FXR1, SMAD5, TRIM23, SEMA6A, PXK, NLN, SLC12A5, VEZF1, RNF11, AMOT, KCTD10, ATG14, HIP1, CKS2, TET2, H2AFX, KAT6B, TUBB2A, MBTD1, CBX4, N4BP1, PSPC1, MAP3K3, RBBP6, SECISBP2L, MTUS1, BAZ2B, PIM2, CNOT4, KLF13, AKAP1, CADM1, MAP3K12, MARCKSL1, CDK19, RBL2, DUSP10, TJAP1, GPATCH8, UBXN2A, POGZ, HOXA10, FOXN2

---

Table 4 The key genes in ceRNA networks (Degree > 5)

Network	Gene name	Gene type	Degree
GM-related ceRNA	MIAT	lncRNA	14
	DLX6-AS1	lncRNA	13
	LY86-AS1	lncRNA	12
	hsa-miR-17-5p	miRNA	12
	RRN3P2	lncRNA	10
	hsa-miR-27a-3p	miRNA	10
	hsa-miR-107	miRNA	10
	hsa-miR-20b-5p	miRNA	10
	hsa-miR-363-3p	miRNA	10
	hsa-miR-129-5p	miRNA	9
	hsa-miR-23b-3p	miRNA	8
	hsa-miR-24-3p	miRNA	8
	hsa-miR-10a-5p	miRNA	7
	hsa-miR-1297	miRNA	6
	CRNDE	lncRNA	6
WM-related ceRNA	hsa-miR-17-5p	miRNA	12
	HCP5	lncRNA	11
	hsa-miR-20b-5p	miRNA	11
	MIAT	lncRNA	11
	hsa-miR-24-3p	miRNA	7

---

Fig. 5 Sub-networks of hub lncRNA candidates. (a) MIAT-associated ceRNA network (in GM). (b) LY86-AS1-associated ceRNA network. (c) DLX6-AS1-associated ceRNA network. (d) RRN3P2-associated ceRNA network. (e) CRNDE-associated ceRNA network. (f) HCP5-associated ceRNA network. (g) MIAT-associated ceRNA network (in WM). Blue ellipses indicate mRNAs. Gray edges indicate lncRNA–miRNA–mRNA interactions.

Experimental validation of hub lncRNAs

Since one of our objectives was to determine key lncRNAs with diagnostic value, we further performed RT-qPCR for the aforementioned hub lncRNAs. Considering that MIAT appeared in two networks, six final candidates were considered. The results shown in Fig. 6. indicate that the expressions of LY86-AS1, DLX6-AS1, RRN3P2, and CRNDE were reduced remarkably in the ICH group, while the expressions of MIAT and HCP5 were enhanced. ROC results shown in Fig. 6. demonstrate that based on our criteria, MIAT (AUC = 0.9293, P < 0.0001), LY86-AS1 (AUC = 0.9010, P < 0.0001), DLX6-AS1 (AUC = 0.8162, P = 0.0005), RRN3P2 (AUC = 0.8505, P = 0.0001), CRNDE (AUC = 0.8909, P < 0.0001), and HCP5 (AUC = 0.8020, P = 0.0008) might serve as potential markers for discriminating patients with ICH from the healthy controls.

Fig. 6 Validations of hub lncRNAs. RT-qPCR results of key lncRNA candidates in peripheral blood from patients with ICH compared to healthy people. (a) Relative expression of MIAT. (b) The ROC of MIAT. (c) Relative expression of LY86-AS1. (d) The ROC of LY86-AS1. (e) Relative expression of DLX6-AS. (f) The ROC of DLX6-AS1. (g) Relative expression of RRN3P2. (h) The ROC of RRN3P2. (i) Relative expression of CRNDE. (j) The ROC of CRNDE. (k) Relative expression of HCP5. (l) The ROC of HCP5 ** P < 0.01, *** P < 0.001.

Discussion

A set of lncRNAs could modulate brain development, and the perturbation of these molecules disturbs the cross-talk of miRNA with their mRNA targets, thereby leading to brain disorders, including Huntington's disease, Alzheimer's diseases, and ischemic stroke.^20–23 Here, we primarily concentrated on lncRNAs in ICH and examined the following: (1) lncRNA-associated molecular cascades involved in ICH; (2) potential lncRNA–miRNA–mRNA interactions, and (3) hub lncRNAs that could be considered as biomarkers. We mined lncRNA data by re-annotating probes in a microarray expression profile, filtered DElncRNAs, and established ceRNA networks. To the best of our knowledge, this is the first report focusing on the lncRNA-associated ceRNA network in ICH.

Bleeding within the brain parenchyma will harm both GM and the proximal or distal WM. Owing to the distinct cellular architecture, the blood supply pattern and corresponding function of WM and its response to ICH may vary from that of GM.^24,25 The discrepancy associated with DElncRNAs in the PH versus the GM and the WM observed here could be related to this phenomenon. According to our GO and KEGG analyses, the mRNAs in GM-related CEN were enriched in biological terms, including “chemical synaptic transmission”, “cell junction”, and “syntaxin-1 binding”. The pooled erythrocytes caused by bleeding in hemorrhagic stroke would lyse to produce hemoglobin, contributing a dose dependent suppression of synaptic transmission in hippocampal CA1 neurons.²⁶ Interestingly, syntaxin-1 is a key component of the synaptic vesicle docking/fusion machinery.²⁷ Given that vascular disruption is the underlying reason of ICH, the role of endothelial cell-to-cell junction on maintaining vascular homeostasis might be associated with ICH occurrence.^28,29 However, the mRNAs in WM-related CEN were associated with different cellular processes, including “positive regulation of cell migration”, “microvillus”, and “cytoskeletal protein binding”. When ICH occurred, the neurogenesis in the subventricular zone and hippocampal subgranular zone would be triggered to generate new neuronal cells and migrate to damaged brain regions. This neurogenetic activity is usually accompanied with cytoskeletal protein remodeling and assembling.^30,31 The occurrence of endothelial microvilli could be viewed as an indicator to assess capillary injury in ischemic stroke.³² KEGG analysis revealed that the mRNA in the GM-related CEN were enriched in biological categories regarding neuronal synapses involved in primary disease processes, including “Axon guidance” and “Dopaminergic synapse”, whereas the mRNAs in WM-related CEN were linked to the metabolic processes, including “metabolic pathways” and “cysteine and methionine metabolism”. As mentioned above, the different structures of the two regions of the brain might explain these results. The abundance of neuronal bodies and fibers, glial cells, and blood vessels in the GM possibly activated a series of molecular cascades, such as axon loss or repair, directly in the initial phase of ICH. The myelin sheaths are later destroyed due to blood volume, which leads to the development of WM lesions, affecting other cellular processes indirectly, such as usual metabolic and energy pathways.³³ Accordingly, it is reasonable to speculate that the lncRNAs may be involved in many molecular cascades during different stages of ICH.

According to the ceRNA theory, lncRNA and mRNA could competitively interact with mutual miRNA, constituting a lncRNA–miRNA–mRNA association. Compared to the limited reports on lncRNA, many studies have demonstrated the pathophysiological role of miRNA in ICH. Based on our findings, 20 and 25 miRNAs were predicted to build the ceRNA network for the GM and the WM, respectively. Our previous work demonstrated that miR-27a-3p could protect against post-ICH complications in the capillary endothelial cells of the brain.³⁴ Similarly, as regulators of inflammatory responses observed in the aftermath of ICH, miR-140-5p and miR-590-5p might relieve ICH-induced brain injury.^35,36 An increase in miR-129-5p levels inhibited the HMGB1-RAGE axis, affecting revascularization in the rat models of ICH.³⁷ Nevertheless, whether lncRNA participates in the aforementioned molecular processes and if yes, what its exact role definitely require further examination.

MIAT was highlighted in our topological estimation of both ceRNA networks. MIAT, namely myocardial infarction-associated transcript, was initially reported to be associated with myocardial infarction in the Japanese population.³⁸ The main function of this lncRNA involves gene regulation at the post-transcriptional level, which is implemented via the ceRNA machinery.³⁹ It is a critical regulator in neurological disorders, especially neurovascular illnesses. Accumulating evidence has illustrated its pathophysiological role in ischemic stroke. In line with our findings in ICH patients, enhanced MIAT expression was observed in peripheral blood leukocytes of ischemic stroke patients, which was positively correlated with infarct volume. Hence, patients with higher MIAT expression might have a relatively poor prognosis.⁴⁰ In contrast, the neuroprotective ability of this lncRNA was confirmed by another group; overexpressed MIAT dampened neuronal apoptosis via miR-211/GDNF in a neonatal rat model of hypoxic-ischemic injury.⁴¹ The controversial conclusion reflects its complicated role under different stroke-like pathological conditions. Therefore, we hypothesize that MIAT might participate in ICH progression in various ways, and the ceRNA networks would be conducive to the advancement of our understanding of the underlying mechanisms.

The function of hub lncRNAs discovered here in ICH remains unclear, demonstrating a huge knowledge gap. However, several disorders are associated with these genes. DLX6-AS1 could serve as an oncogene and facilitates tumorigenesis by binding to miRNA targeting sites for competing endogenous RNAs, especially in neuroblastoma.^42,43 Sponge miR-149-30 promotes apoptotic activity, contributing to impairments of neurons of cerebral ischemia/reperfusion mice.⁴⁴ Similarly, accumulating evidence has concentrated on the role of HCP5 in oncogenesis, including glioma, gastric cancer, and ovarian cancer.^45,46 In terms of neurological disorders, enhanced HCP5 expression was associated with the aggravation of cerebral ischemic reperfusion injury by sponging miR-652-3p.⁴⁷CRNDE is located in specific brain regions of humans and mice that are tightly linked to gliomagenesis.^48,49 Knocking out LncRNA CRNDE alleviates brain damage, resulting from hypoxic-ischemic or traumatic brain injury by promoting autophagy or nerve repair.^50,51 Besides, as immune-related lncRNAs, LY86-AS1 variants are involved in the onset of autoimmune blistering disease of the skin and type 2 diabetes mellitus.^52,53 The expressions of gender-associated LY86-AS1 are negatively correlated with the Braak stage of Alzheimer's disease.⁵⁴ In accordance with these reports, our hub lncRNAs might manage neuron survival via couples of pathways in the scenario of ICH, such as promoting cell apoptosis or regulating inflammatory response. Accordingly, it is greatly worthy to arrange more researches on this topic. In addition, as a quick and cheap sampling method, blood can be routinely acquired in clinical tests, which was employed for the validation of hub lncRNA candidates here. We observed altered expressions of MIAT, LY86-AS1, DLX6-AS1, RRN3P2, CRNDE, and HCP5 in the blood of patients with ICH, implying their involvement in ICH. ROC curves, together with the AUCs, are useful tools for evaluating the test accuracy and for selecting potential biomarkers.^55,56 Based on our ROC results, these key lncRNAs may be employed as molecular indicators for early screening and monitoring of hemorrhage stroke.

There are several limitations of the current study. Firstly, the sample size of GSE24265 for analysis is relatively small, thereby, more clinical samples are necessary to confirm the expression pattern of hub lncRNAs in the future. Moreover, the relationship between lncRNA and other factors should be investigated for further elucidating the ICH mechanism, such as vascular risk factors. Finally, although we have utilized three databases together with our co-expressed network analysis to acquire mRNA targets for the interested lncRNAs, the incomplete coverage of various interactions will probably produce some biased prediction for lncRNA–miRNA or miRNA–mRNA pairs, particularly those unknown interactions.⁵⁷ On the one hand, using diverse computational models, including machine learning-based models and models with no requirement of uncovering ncRNA pairs, might be a promising approach for lncRNA–miRNA–mRNA selection.^57,58 On the other hand, to probe the relationship between hub lncRNAs or miRNAs and ICH using different computational tools, such as Laplacian Regularized Least Squares for LncRNA–Disease Association (LRLSLDA),⁵⁹ Inductive Matrix Completion for MiRNA–Disease Association prediction (IMCMDA)⁶⁰ or Matrix Decomposition and Heterogeneous Graph Inference for miRNA–disease association prediction (MDHGI),⁶¹ would become our task in the future as well. Additionally, capitalizing on hypergeometric distribution models may be also helpful for evaluating the mRNA–lncRNA interaction simultaneously instead of detecting lncRNA–mRNA interaction only.

Conclusions

We identified DElncRNAs in various regions of the brain by analyzing a published dataset of ICH. Based on co-expression interactions and the ceRNA theory, the lncRNA-related ceRNA networks were constructed. Six lncRNAs: MIAT, LY86-AS1, DLX6-AS1, RRN3P2, CRNDE, and HCP5 may be utilized as biomarkers for ICH progression. Our work may contribute to an increased understanding of the pathogenesis of ICH and elucidate the role of novel lncRNAs as potential targets for a breakthrough in ICH diagnosis and treatment.

Data availability

This study was carried out using publicly available data from Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/gds/) with GSE24265. All data generated or analysed during this study are included in this published article.

Author contributions

Conceptualization: X. X. and C. Y.; formal analysis: C. Y., J. W., S. X., X. L. and X. X.; funding acquisition: X. X.; methodology: X. L. and X. X.; resources: C. Y., J. W. and X. X.; supervision: X. X.; visualization: C. Y., J. W. and X. X.; writing – original draft: C. Y., J. W., X. L., S. X. and X. X.; writing – review&editing: X. X.

Conflicts of interest

We have no conflict of interests.

Acknowledgements

This work was kindly supported by National Natural Science Foundation of China (81601129) and Science Foundation of Shenyang (F16-205-1-48).

References

1. A. I. Qureshi, A. D. Mendelow and D. F. Hanley, Lancet, 2009, 373, 1632–1644.
2. R. F. Keep, Y. Hua and G. Xi, Lancet Neurol., 2012, 11, 720–731.
3. A. Rosell, A. Vilalta, T. Garcia-Berrocoso, I. Fernandez-Cadenas, S. Domingues-Montanari, E. Cuadrado, P. Delgado, M. Ribo, E. Martinez-Saez, A. Ortega-Aznar and J. Montaner, PLoS One, 2011, 6, e16750.
4. K. T. Lee and J. W. Nam, BMB Rep., 2017, 50, 226–231.
5. Z. Li, X. Li, C. Chen, S. Li, J. Shen, G. Tse, M. T. V. Chan and W. K. K. Wu, Cell Proliferation, 2018, 51, e12483.
6. L. Salmena, L. Poliseno, Y. Tay, L. Kats and P. P. Pandolfi, Cell, 2011, 146, 353–358.
7. P. Wan, W. Su and Y. Zhuo, Mol. Neurobiol., 2017, 54, 2012–2021.
8. D. Liu, Z. Zou, G. Li, P. Pan and G. Liang, Cancer Control, 2020, 27, 1073274819897977.
9. J. M. Kim, J. Moon, J. S. Yu, D. K. Park, S. T. Lee, K. H. Jung and K. Chu, Ann. Clin. Transl. Neurol., 2019, 6, 2014–2025.
10. C. Hanjin, L. Tao, L. Pengfei, Y. Ali, Z. Huajun, L. Jiekun, W. Yang and T. Tao, Cell. Physiol. Biochem., 2018, 45, 1284–1301.
11. L. Xie, Y. Wang and Z. Chen, Life Sci., 2021, 267, 118942.
12. B. Chen, H. Wang, C. Lv, C. Mao and Y. Cui, Bioengineered, 2021, 12, 4004–4015.
13. B. Xie, M. Qiao and J. Xuan, Med. Sci. Monit., 2021, 27, e929435.
14. J. X. Chen, Y. P. Wang, X. Zhang, G. X. Li, K. Zheng and C. Z. Duan, Brain Res. Bull., 2020, 162, 20–29.
15. X. Zhang, S. Sun, J. K. Pu, A. C. Tsang, D. Lee, V. O. Man, W. M. Lui, S. T. Wong and G. K. Leung, Neurobiol. Dis., 2012, 48, 1–8.
16. X. Xu, Y. Hao, S. Xiong and Z. He, Interdiscip. Sci., 2020, 12, 447–460.
17. X. Wang, S. Ma, M. Mao, C. Li, X. Shen, S. Xu and J. Yang, BioMed Res. Int., 2019, 2019, 7505260.
18. Y.-H. Shi, X.-W. He, F.-D. Liu, Y.-S. Liu, Y. Hu, L. Shu, G.-H. Cui, R. Zhao, L. Zhao, J.-J. Su and J.-R. Liu, Gene, 2019, 697, 184–193.
19. J. Wang, Y. Zhu, F. Jin, L. Tang, Z. He and Z. He, J. Int. Med. Res., 2016, 44, 419–432.
20. J. Aprea, S. Prenninger, M. Dori, T. Ghosh, L. S. Monasor, E. Wessendorf, S. Zocher, S. Massalini, D. Alexopoulou, M. Lesche, A. Dahl, M. Groszer, M. Hiller and F. Calegari, EMBO J., 2013, 32, 3145–3160.
21. J. S. Sunwoo, S. T. Lee, W. Im, M. Lee, J. I. Byun, K. H. Jung, K. I. Park, K. Y. Jung, S. K. Lee, K. Chu and M. Kim, Mol. Neurobiol., 2017, 54, 1577–1586.
22. M. A. Faghihi, F. Modarresi, A. M. Khalil, D. E. Wood, B. G. Sahagan, T. E. Morgan, C. E. Finch, G. St Laurent, 3rd, P. J. Kenny and C. Wahlestedt, Nat. Med., 2008, 14, 723–730.
23. J. Li, M. Hao, B. Yang, T. Shi, Y. Zhang, J. Feng and J. Chen, Medicine, 2020, 99, e22964.
24. C. Tao, X. Hu, H. Li and C. You, Front. Hum. Neurosci., 2017, 11, 422.
25. Y. B. Jiang, K. Y. Wei, X. Y. Zhang, H. Feng and R. Hu, CNS Neurosci. Ther., 2019, 25, 1113–1125.
26. S. Yip and B. R. Sastry, Brain Res., 2000, 864, 1–12.
27. G. Lao, V. Scheuss, C. M. Gerwin, Q. Su, S. Mochida, J. Rettig and Z. H. Sheng, Neuron, 2000, 25, 191–201.
28. R. F. Keep, A. V. Andjelkovic, J. Xiang, S. M. Stamatovic, D. A. Antonetti, Y. Hua and G. Xi, J. Cereb. Blood Flow Metab., 2018, 38, 1255–1275.
29. G. Bazzoni and E. Dejana, Physiol. Rev., 2004, 84, 869–901.
30. X. Xu, L. E. Wicki-Stordeur, J. C. Sanchez-Arias, M. Liu, M. S. Weaver, C. S. W. Choi and L. A. Swayne, Front. Cell. Neurosci., 2018, 12, 124.
31. C. Lei, S. Lin, C. Zhang, W. Tao, W. Dong, Z. Hao, M. Liu and B. Wu, Neuroscience, 2013, 230, 86–93.
32. Z. Süle, E. Mracskó, E. Bereczki, M. Sántha, T. Csont, P. Ferdinandy, F. Bari and E. Farkas, Life Sci., 2009, 84, 935–939.
33. P. J. Derry, A. T. T. Vo, A. Gnanansekaran, J. Mitra, A. V. Liopo, M. L. Hegde, A. L. Tsai, J. M. Tour and T. A. Kent, Front. Cell. Neurosci., 2020, 14, 603043.
34. T. Xi, F. Jin, Y. Zhu, J. Wang, L. Tang, Y. Wang, D. S. Liebeskind, F. Scalzo and Z. He, J. Biol. Chem., 2018, 293, 20041–20050.
35. S. Wang, Y. Cui, J. Xu and H. Gao, Inflammation, 2019, 42, 1869–1877.
36. Q. Guo, H. Su, J. B. He, H. Q. Li and J. J. Sha, Biochem. Biophys. Res. Commun., 2018, 504, 61–67.
37. X. L. Ma, S. Y. Li and F. Shang, Biomed. Pharmacother., 2017, 93, 238–244.
38. Y. Ohnishi, T. Tanaka, R. Yamada, K. Suematsu, M. Minami, K. Fujii, N. Hoki, K. Kodama, S. Nagata, T. Hayashi, N. Kinoshita, H. Sato, H. Sato, T. Kuzuya, H. Takeda, M. Hori and Y. Nakamura, Hum. Genet., 2000, 106, 288–292.
39. C. M. Da, C. Y. Gong, W. Nan, K. S. Zhou, Z. L. Wu and H. H. Zhang, Biomed. Pharmacother., 2020, 129, 110359.
40. M. Zhu, N. Li, P. Luo, W. Jing, X. Wen, C. Liang and J. Tu, J. Stroke Cerebrovasc. Dis., 2018, 27, 326–337.
41. E. Y. Li, P. J. Zhao, J. Jian, B. Q. Yin, Z. Y. Sun, C. X. Xu, Y. C. Tang and H. Wu, Cell Cycle, 2019, 18, 156–166.
42. H. Y. Zhang, M. Q. Xing, J. Guo, J. C. Zhao, X. Chen, Z. Jiang, H. Zhang and Q. Dong, Cancer Cell Int., 2019, 19, 313.
43. Q. Zheng, X. Gu, Q. Yang, Q. Chu, Y. Dai and Z. Chen, Clin. Chim. Acta, 2021, 517, 1–8.
44. X. Hu, Z. Xiang, W. Zhang, Z. Yu, X. Xin, R. Zhang, Y. Deng and Q. Yuan, Aging, 2020, 12, 23096–23113.
45. J. Zhao, X. Song, T. Xu, Q. Yang, J. Liu, B. Jiang and J. Wu, Front. Genet., 2020, 11, 607722.
46. W. Chen, K. Zhang, Y. Yang, Z. Guo, X. Wang, B. Teng, Q. Zhao, C. Huang and Z. Qiu, Int. J. Biol. Sci., 2021, 17, 623–634.
47. J. D. Gao, R. J. Li, P. L. Ma, L. L. Yu, J. T. Li and H. T. Tian, J. Biol. Regul. Homeostatic Agents, 2020, 34, 893–900.
48. B. C. Ellis, P. L. Molloy and L. D. Graham, Front. Genet., 2012, 3, 270.
49. A. Zottel, N. Šamec, A. Videtič Paska and I. Jovčevska, Cancers, 2020, 12, 1842.
50. C. H. Fu, F. F. Lai, S. Chen, C. X. Yan, B. H. Zhang, C. Z. Fang and G. H. Wang, Mol. Cell. Biochem., 2020, 472, 1–8.
51. M. Yi, X. Dai, Q. Li, X. Xu, Y. Chen and D. Wang, Cell Cycle, 2019, 18, 2332–2343.
52. A. Salviano-Silva, T. D. J. Farias, V. Bumiller-Bini, M. S. Castro, S. C. Lobo-Alves, H. Busch, C. Pföhler, M. Worm, M. Goebeler, N. van Beek, A. Franke, M. Wittig, D. Zillikens, R. C. de Almeida, J. E. Hundt, A. B. W. Boldt, S. Ibrahim, D. G. Augusto, M. L. Petzl-Erler, E. Schmidt and D. Malheiros, Exp. Dermatol., 2021, 30, 831–840.
53. L. Saeidi, H. Ghaedi, M. Sadatamini, R. Vahabpour, A. Rahimipour, M. Shanaki, Z. Mansoori and F. Kazerouni, Mol. Biol. Rep., 2018, 45, 2601–2608.
54. M. Cao, H. Li, J. Zhao, J. Cui and G. Hu, Neurobiol. Aging, 2019, 81, 116–126.
55. G. Heller, V. E. Seshan, C. S. Moskowitz and M. Gonen, Biostatistics, 2017, 18, 260–274.
56. K. Soreide, H. Korner and J. A. Soreide, Ann. Surg., 2011, 253, 27–34.
57. X. Chen, C. C. Yan, X. Zhang and Z. H. You, Briefings Bioinf., 2017, 18, 558–576.
58. X. Chen, D. Xie, Q. Zhao and Z. H. You, Briefings Bioinf., 2019, 20, 515–539.
59. X. Chen and G. Y. Yan, Bioinformatics, 2013, 29, 2617–2624.
60. X. Chen, L. Wang, J. Qu, N. N. Guan and J. Q. Li, Bioinformatics, 2018, 34, 4256–4265.
61. X. Chen, J. Yin, J. Qu and L. Huang, PLoS Comput. Biol., 2018, 14, e1006418.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/d1mo00298h

---