The Role of MicroRNAs in Predicting the Neurological Outcome of Patients with Subarachnoid Hemorrhage: A Meta-analysis

Abstract

Subarachnoid hemorrhage (SAH) is a hemorrhagic cerebrovascular disease with an extremely poor prognosis. The molecular mechanism and biomarkers involved in neurological outcome after SAH still need to be explored. This study assessed the microRNA expression characteristics of SAH patients with different neurological outcomes by meta-analysis. Public databases were searched from database inception until December 2022. The study reported that microRNA expression data in SAH patients with different neurological outcomes were included in the analysis. The differential expression of miRNAs was evaluated by meta-analysis. Overrepresentation analysis was performed for the targeted genes of significant miRNAs. The XGBoost algorithm was used to assess the predictive ability for neurological outcomes with clinical characteristics and significantly expressed miRNAs. Seven studies were finally included in the meta-analysis. The results showed that the levels of miR-152-3p (SMD: − 0.230; 95% CI − 0.389, − 0.070; padj = 0.041), miR-221-3p (SMD: − 0.286; 95% CI − 0.446, − 0.127; padj = 0.007), and miR-34a-5p (SMD: − 0.227; 95% CI − 0.386, − 0.067; padj = 0.041) were significantly lower in SAH patients with good neurological outcomes than in those with poor neurological outcomes. The PI3K-AKT signaling pathway may have an important role in neurological recovery after SAH. Based on the XGBoost algorithm, the neurological outcome could be accurately predicted with clinical characteristics plus the three miRNAs. The expression levels of miR-152-3p, miR-221-3p, and miR-34a-5p were significantly lower in patients with good neurological outcomes than in those with poor outcomes. These miRNAs can serve as potential predictive biomarkers for neurological outcomes.

Graphical Abstract

The molecular mechanism and biomarkers involved in neurological outcome after SAH still need to be explored. Our study analyzed microRNA expression characteristics of SAH patients with different neurological outcomes by meta-analysis. After analyze studies reporting the microRNA expression data in SAH patients with different neurological outcomes, results show that the levels of miR-152-3p, miR-221-3p, and miR-34a-5p were significantly lower in SAH patients with good neurological outcomes than in those with poor neurological outcomes. The PI3K-AKT signaling pathway may have an important role in neurological recovery after SAH. Based on the XGBoost algorithm, the neurological outcome could be accurately predicted with clinical characteristics plus the three miRNAs.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10571-023-01327-7.

Introduction

Subarachnoid hemorrhage (SAH) is an acute and life-threatening hemorrhagic cerebrovascular disease. The incidence of SAH is approximately 6 to 27.63 per 100,000 persons each year worldwide (Chen et al. 2022a, b). SAH patients have an extremely poor prognosis, with 40% dying within the first month. Approximately 30% of the survivors also suffered neurological functional impairments and were dependent on others in daily living (Liu et al. 2021). In addition, SAH affects younger adults than ischemic stroke and accounts for 27% of all stroke-related years of life loss before age 60, which has a serious impact on the productivity of society (Annoni et al. 2022). At present, the prognosis of neurological outcome can be evaluated by clinical syndromes and radiography, but with limited accuracy (Ten Brinck et al. 2022).

In the era of personalized and precision medicine, biomarkers are becoming increasingly important (Chou et al. 2019). miRNAs are able to resist degradation by ribonucleases that have the potential to be biomarkers for disease prediction and diagnosis (Inzaghi et al. 2020; Zaiou 2020). miRNAs are a class of noncoding single-stranded RNA molecules of approximately 22 nucleotides that bind to complementary target mRNAs to reduce mRNA stability and inhibit their translation. miRNAs have been proven to be involved in the progression of various cardiovascular and neurological diseases (Jiang et al. 2020; Song et al. 2022). A recent study confirmed the important regulatory role of miRNAs in the occurrence and development of SAH (Zheng et al. 2022). However, miRNA expression characteristics in SAH patients with different neurological outcomes remain puzzling. This study will comprehensively analyze the miRNA expression characteristics of SAH patients with different neurological outcomes and then explore the potential molecular mechanisms and their predictive ability.

Methods

Literature Search

Two investigators searched public literature databases, including PubMed, EmBase, Cochrane Library, China National Knowledge Infrastructure, Wanfang, China Science and Technology Journal Database, SinoMed, Gene Expression Omnibus, and ArrayExpress. The search was conducted from database inception until December 2022 with a brief Boolean strategy (“subarachnoid hemorrhage”) AND (“miRNA” OR “miR” OR “microRNA” OR “noncoding RNA”). No language restriction was added.

Inclusion and Exclusion Criteria

The inclusion criteria were as follows: (1) the study contained SAH patients; (2) the study reported the neurological outcome of SAH patients; and (3) the study reported the miRNA expression level in peripheral blood or cerebrospinal fluid according to different neurological outcome groups. The exclusion criteria were as follows: (1) animal model research or in vitro research; and (2) duplicate studies. Comments, case reports, letters, and conference abstracts were also excluded. Two researchers independently screened the literature for those that met the inclusion criteria, and disputes were agreed upon through mutual discussion.

Data Extraction

Data were extracted from the literature, including the first author’s name, publication year, study location, sample size (number of patients), average age, percentage of females, sample sources, miRNA detection method, judgment of neurological outcome, and follow-up. The miRNA names were standardized according to the miRbase database (Release 22.1). The quality of the included studies was evaluated using the Quality Assessment of Diagnostic Accuracy Studies version 2 (QUADAS-2) tool.

Statistical Analysis

The standard mean difference (SMD) and its 95% confidence interval (CI) were used to combine the microRNA expression results. The I² statistic was used to assess the heterogeneity among studies. When I² ≥ 50%, it was considered that there was significant heterogeneity, and results based on a random-effects model were adopted; otherwise, the results based on a fixed-effects model were adopted. The p values were corrected for multiple tests using the Benjamini & Hochberg (BH) adjustment. A p value < 0.05 was considered statistically significant.

Overrepresentation Analysis

The targeted genes of miRNAs with statistically significant differences were screened using the miRWalk database (version 3.0, http://mirwalk.umm.uni-heidelberg.de/). The targeted genes were filtered and confirmed in all three subdatabases (TargetScan, miRDB and miRTarBase) (Sticht et al. 2018). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis and Over-Representation Analysis for WikiPathways were performed for targeted genes using the R package “clusterProfiler” (Version 4.6.0) (Wu et al. 2021).

Prediction Models for Neurological Outcome in SAH Patients

The eXtreme Gradient Boosting (XGBoost) algorithm was used to predict neurological outcomes (Chen and Guestrin 2016). The dependent variable included patients’ clinical characteristics and significant expression miRNAs. Although the XGBoost algorithm did not require normalization of continuous variables, the miRNA expression data collected from each study were still internally normalized in this study. Eighty percent of the data were randomly selected as the training set to train the model, and the remaining 20% were used as the test set to verify the predictive ability. Fivefold cross validation was used in model training to lessen the probability of model overfitting. DeLong's test was used for model comparison. The R packages “xgboost” (Version 1.6.0.1) and “tidymodels” (Version 1.0.0) were used in the analysis.

Results

A total of 395 studies in English were obtained through the literature database search, and 284 were obtained after removing duplicates. A total of 537 studies in Chinese were obtained, and 489 were obtained after removing duplicates. Twelve datasets were obtained from the GEO and ArrayExpress databases. The 785 studies were screened based on titles and abstracts, and 747 irrelevant studies were excluded. Subsequently, 38 studies were adjusted in full-text, and 31 studies were excluded for the following reasons: studies that did not report neurological outcomes (22 studies), animal or in vitro research (5 studies), and reviews (4 studies). Seven studies were finally included (Bache et al. 2017; Bache et al. 2020; F. Chen et al. 2022a, b; Chen et al. 2020; Lai et al. 2017; Pedrosa et al. 2022; Zhao et al. 2022) (Fig. 1).

Fig. 1.

Flowchart illustrating the selection of studies included in this work

The included studies were published in the last five years. The studies were located in China, Denmark, and Spain. Three studies used cerebrospinal fluid as the sample source. Three studies detected miRNA expression with a microRNA array or panel. The results obtained by microRNA array or panel were more reliable than those obtained by predetermined miRNA types. All studies adopted a modified Rankin Scale (mRS) score ≤ 2 as the criterion for good neurological outcome. However, the follow-up periods were different among studies, including at three-, six-, and nine-month follow-up timepoints (Table 1).

Table 1.

The characteristics of included studies

Author	Year	Study location	Number of patients, N	Age, year	Male, N (percentage)	Time point of sample collection	Sample source	Detection method	Neurological outcomes judgement	Follow-up
Nian-sheng Lai	2017	China	60	Age < 55: n = 20; Age ≥ 55: n = 40	28 (46.7%)	72 h after SAH	Serum	PCR	mRS ≤ 2	9 months
Bache Søren	2017	Danmark	37	57 ± 13	18 (48.6%)	Day 5 after SAH	CSF	Pick & Mix microRNA PCR panels	mRS ≤ 2	3 months
Bache Søren	2020	Danmark	126	Discovery: 59 ± 11; Replication: 56 ± 12	34 (27.0%)	Day 3–14 after SAH	CSF	Pick & Mix microRNA PCR panels	mRS ≤ 2	3 months
Ming Chen	2020	China	55	51.92 ± 10.77	31 (56.4%)	24 h after SAH	Blood	qPCR	mRS ≤ 2	6 months
Chunyan Zhao	2022	China	176	47.26 ± 4.8	69 (39.2%)	Day 10 after SAH	Serum	qPCR	mRS ≤ 2	3 months
Feihu Chen	2022	China	171	47.37 ± 4.96	67 (39.2%)	36 h after SAH	Serum	qPCR	mRS ≤ 2	3 months
Leire Pedrosa	2022	Spain	57	57 ± 13	19 (33.3%)	Day 3 after SAH	CSF	nCounter Human v3 miRNA Expression Assay Kit	mRS ≤ 2	3 months

CSF cerebrospinal fluid, mRS modified Rankin scale, SAH subarachnoid hemorrhage, qPCR Quantitative polymerase chain reaction

With regard to the design quality of the included studies, one study did not specify the inclusion and exclusion criteria for SAH patients (Lai et al. 2017). It had an unclear impact on patient selection. One study included only SAH patients who received endovascular coil embolization(F. Chen et al. 2022a, b). Another study included patients who received hyperbaric oxygen therapy (Zhao et al. 2022). These specific characteristics of patient selection may have an unclear bias in the analysis. Four studies detected predetermined miRNA types. It is possible that positive outcomes will likely be reported. The other 3 studies reported results based on arrays or panels. Therefore, the overall quality of the studies was medium (Fig. 2).

Fig. 2.

The risk of bias and applicability concerns of included studies

Meta-analysis showed the results of miRNAs that were reported in three or more studies. Based on the fixed effect model, the levels of miR-152-3p (I² = 21.7%, SMD: − 0.230; 95% CI − 0.389, − 0.070; p = 0.005, adjusted p = 0.041), miR-221-3p (I² = 0%; SMD: − 0.286; 95% CI − 0.446, − 0.127; p < 0.001; adjusted p = 0.007), and miR-34a-5p (I² = 2.9%; SMD: − 0.227; 95% CI − 0.386, − 0.067; p = 0.005, adjusted p = 0.041) were significantly lower in SAH patients with good neurological outcomes than in those with poor neurological outcomes (Table 2). The level of miR-193b-3p (I² = 92.3%, SMD: 0.791; 95% CI 0.503, 1.079; p < 0.001; adjusted p < 0.001) was significantly higher in patients with good neurological outcomes; however, the pooled results had high obvious heterogeneity. Based on the random-effect model, there was no significant difference in miR-193b-3p levels between patients with good neurological outcomes and those with poor outcomes (SMD: 0.363; 95% CI − 0.687, 1.413; p = 0.498). miR-221-3p (SMD: − 0.286; 95% CI − 0.446, − 0.127; p < 0.001; adjusted p = 0.014) also showed significant differences based on a random-effects model (Table 2).

Table 2.

Meta-analysis results based on the fixed effect model and random effect model of all miRNAs reported in three or more studies

miRNAs	No. of study	I2	Fixed effect model			Random effect model
			SMD (95%CI)	p-value	Adjusted p-value	SMD (95%CI)	p-value	Adjusted p-value
let-7b-5p	5	0.0%	− 0.175 (− 0.335, − 0.016)	0.031	0.137	− 0.175 (− 0.335, − 0.016)	0.031	0.240
miR-125b-5p	5	27.4%	− 0.072 (− 0.231, 0.087)	0.374	0.611	− 0.175 (− 0.437, 0.088)	0.193	0.528
miR-130a-3p	5	42.5%	− 0.086 (− 0.245, 0.073)	0.289	0.497	− 0.140 (− 0.422, 0.143)	0.332	0.631
miR-152-3p	5	21.7%	− 0.230 (− 0.389, − 0.07)	0.005	0.041	− 0.247 (− 0.482, − 0.011)	0.040	0.249
miR-15b-5p	5	41.3%	− 0.051 (− 0.210, 0.109)	0.534	0.747	− 0.051 (− 0.210, 0.109)	0.534	0.719
miR-16-5p	5	42.9%	0.013 (− 0.147, 0.172)	0.876	0.918	0.003 (− 0.215, 0.222)	0.976	0.976
miR-185-5p	5	42.0%	− 0.057 (− 0.216, 0.102)	0.482	0.712	− 0.057 (− 0.216, 0.102)	0.482	0.719
miR-19a-3p	5	7.9%	− 0.059 (− 0.218, 0.100)	0.470	0.712	− 0.059 (− 0.218, 0.1)	0.470	0.719
miR-204-5p	5	0.0%	− 0.202 (− 0.361, − 0.042)	0.013	0.082	− 0.202 (− 0.361, − 0.042)	0.013	0.204
miR-21-5p	5	0.0%	− 0.150 (− 0.309, 0.009)	0.064	0.227	− 0.178 (− 0.387, 0.031)	0.094	0.418
miR-221-3p	5	0.0%	− 0.286 (− 0.446, − 0.127)	0.000	0.007	− 0.286 (− 0.446, − 0.127)	0.000	0.014
miR-222-3p	5	63.8%	− 0.035 (-0.195, 0.124)	0.663	0.762	− 0.116 (− 0.463, 0.232)	0.514	0.719
miR-223-3p	5	0.0%	0.142 (− 0.017, 0.301)	0.079	0.227	0.142 (− 0.017, 0.301)	0.079	0.410
miR-25-3p	5	54.0%	0.047 (− 0.113, 0.206)	0.566	0.747	0.045 (− 0.281, 0.37)	0.788	0.873
miR-26a-5p	5	19.9%	− 0.111 (− 0.27, 0.048)	0.171	0.407	− 0.111 (− 0.27, 0.048)	0.171	0.528
miR-27a-3p	6	90.6%	0.039 (− 0.107, 0.185)	0.601	0.747	0.097 (− 0.419, 0.612)	0.714	0.851
miR-27b-3p	5	0.0%	− 0.085 (− 0.237, 0.068)	0.278	0.497	− 0.085 (− 0.237, 0.068)	0.278	0.593
miR-29a-3p	5	0.0%	− 0.142 (− 0.301, 0.017)	0.080	0.227	− 0.147 (− 0.332, 0.038)	0.121	0.467
miR-29b-3p	5	23.3%	− 0.085 (− 0.238, 0.067)	0.273	0.497	− 0.101 (− 0.316, 0.114)	0.358	0.631
miR-29c-3p	5	43.8%	− 0.102 (− 0.262, 0.057)	0.208	0.46	− 0.174 (− 0.456, 0.109)	0.228	0.543
miR-30a-5p	5	0.0%	− 0.086 (− 0.245, 0.073)	0.287	0.497	− 0.086 (− 0.245, 0.073)	0.287	0.593
miR-30d-5p	5	44.9%	0.042 (− 0.117, 0.202)	0.602	0.747	− 0.054 (− 0.336, 0.229)	0.709	0.851
miR-34a-5p	5	2.9%	− 0.227 (− 0.386, − 0.067)	0.005	0.041	− 0.249 (− 0.470, − 0.028)	0.027	0.240
miR-34b-3p	5	0.0%	− 0.119 (− 0.278, 0.04)	0.144	0.372	− 0.119 (− 0.278, 0.040)	0.144	0.495
miR-361-5p	5	37.9%	0.003 (− 0.156, 0.162)	0.969	0.969	− 0.025 (− 0.287, 0.236)	0.849	0.908
miR-451a	5	76.7%	0.012 (− 0.148, 0.172)	0.882	0.918	− 0.07 (− 0.56, 0.419)	0.778	0.873
miR-92a-3p	5	73.1%	− 0.038 (− 0.198, 0.122)	0.639	0.762	0.091 (− 0.385, 0.567)	0.708	0.851
miR-93-5p	5	0.0%	0.011 (− 0.148, 0.17)	0.888	0.918	0.011 (− 0.148, 0.17)	0.888	0.918
miR-210-3p	4	60.4%	0.372 (0.038, 0.707)	0.029	0.137	0.343 (− 0.187, 0.873)	0.205	0.528
miR-193b-3p	3	92.3%	0.791 (0.503, 1.079)	0.000	0.000	0.363 (− 0.687, 1.413)	0.498	0.719
miR-502-5p	3	81.8%	− 0.39 (− 0.813, 0.033)	0.070	0.227	− 0.512 (− 1.623, 0.599)	0.367	0.631

CI confidence interval, SMD standard mean difference

Overrepresentation Analysis

Based on the meta-analysis results, miR-152-3p, miR-221-3p, and miR-34a-5p were statistically significant. The 194 target genes of the above miRNAs were selected (Supplementary Table 1). The top three pathways in the KEGG enrichment results were microRNAs in cancer, the PI3K-AKT signaling pathway, and breast cancer (Fig. 3A). The top three pathways in the WikiPathways were the PI3K-AKT signaling pathway, EGFR tyrosine kinase inhibitor resistance, and breast cancer pathway (Fig. 3B). The results suggested that the PI3K-AKT signaling pathway may have an important role in neurological recovery after SAH (Table 3).

Fig. 3.

Overrepresentation Analysis of genes targeted by significant miRNAs. A KEGG enrichment; B WikiPathways enrichment

Table 3.

Predictive accuracy of different variants for neurological outcome in SAH patients based on XGBoost algorithm

Datasets	Variants	True positive	False positive	False negative	True negative	AUC (95%CI)
Training set	Clinical characteristics# + miR-152-3p, miR-221-3p, miR-34a-5p	113	3	2	84	0.974 (0.951–0.997)
	Clinical characteristics + miR-221-3p	106	9	9	78	0.909 (0.869–0.950)
	Clinical characteristics	102	21	13	66	0.823 (0.769–0.877)
	miR-152-3p + miR-221-3p + miR-34a-5p	99	56	16	31	0.609 (0.549–0.669)
Testing set	Clinical characteristics + miR-152-3p, miR-221-3p, miR-34a-5p	21	2	8	20	0.817 (0.813–0.821)
	Clinical characteristics + miR-221-3p	25	7	4	15	0.772 (0.767–0.777)
	Clinical characteristics	26	9	3	13	0.744 (0.739–0.749)
	miR-152-3p + miR-221-3p + miR-34a-5p	26	17	3	5	0.562 (0.456–0.668)

AUC area Under the (Receiver Operating Characteristic, ROC) Curve, CI confidence interval

^#Clinical characteristics included patients’ age, sex, delayed cerebral ischemia occurrence, and the World Federation of Neurosurgical Societies classification

Prediction Models for Neurological Outcome

Neurological outcome was predicted with clinical characteristics and miRNA expression of patients. The clinical characteristics included age, sex, delayed cerebral ischemia (DCI) occurrence, and the World Federation of Neurosurgical Societies (WFNS) classification, which was most frequently reported in the original data from the included studies. The XGBoost algorithm was trained with only clinical characteristics, only three miRNAs, characteristics plus miR-221-3p, and characteristics plus the above three miRNAs. The results showed that in the training set, the model based on characteristics plus three miRNAs was significantly better than that based on characteristics plus miR-221-3p (p = 0.001) and that based on characteristics only (p < 0.001). However, the model based on miRNAs showed very low accuracy. In the testing set, overfitting still exists. Generally, clinical characteristics plus the three miRNAs could improve the predictive ability for neurological outcome compared to characteristics only.

Discussion

This study evaluated differential miRNA expression between different neurological outcomes in SAH patients by meta-analysis. The results showed that the expression levels of miR-152-3p, miR-221-3p, and miR-34a-5p were significantly lower in patients with good neurological outcomes. KEGG and WikiPathways enrichment analyses indicated that the PI3K-AKT signaling pathway may play an important role in neurological recovery. Based on the XGBoost algorithm, the neurological outcome could be accurately predicted with clinical characteristics plus the three miRNAs.

According to the target genes of significant miRNAs obtained from the meta-analysis, the PI3K-AKT signaling pathway may be important in neurological recovery after SAH by overrepresentation analysis. This pathway is considered to be mainly associated with neuroimmune inflammation and neural cell death after SAH (Tian et al. 2022). A review described the association of the PI3K/AKT signaling pathway with the sequelae of SAH. It is suggested that this pathway affects early brain injury, vasospasm, and neurological injury. By activating the PI3K-AKT pathway cascade, the above clinical manifestations can be alleviated (Jin et al. 2022). The stimulated PI3K-AKT signaling pathway could also improve the inflammatory response, which was characterized by reduced neutrophil infiltration and microglia activation, to reduce secondary brain injury (Zhu et al. 2018). However, the target genes included PIK3R1 and PTEN, which have opposite effects in the pathway. This study only qualitatively suggested that target genes are enriched in the PI3K-AKT pathway, and the detailed characteristics of the PI3K-AKT pathway in the neurological recovery process still require quantitative analysis by gene expression profiling.

This study showed differential miRNA expression between different neurological outcomes in SAH patients. The results provided a basis for the adoption of miRNAs as biomarkers for prediction or even as drug targets for disease treatment. However, there is still a long way to go for the clinical application of miRNAs. First, the miRNA regulatory network is extremely complex in that one miRNA can regulate multiple target genes and vice versa. Second, the inconsistency among the results of different studies also makes the explanation of potential mechanisms more confusing. There was a marked contrast between the high proportion of positive outcomes for predetermined miRNA studies and the low proportion of significant genes in microRNA array results. We believe that the results from microarrays are more reliable than those from predetermined miRNA studies. Fortunately, using the XGBoost algorithm, microarray-based gene expression data are able to predict neurological outcomes in SAH patients when combined with clinical characteristics. This provided a basis for further miRNA array studies on neurological outcome in SAH patients.

There were still several limitations in this study. First, the pooling analysis of samples from cerebrospinal fluid and peripheral blood may have an impact on the results. Although miRNAs can easily cross the blood‒brain barrier (Ghasempour et al. 2022; Ozansoy et al. 2023; Wang et al. 2022), the local concentration of miRNAs may still be affected by the distribution of cells that release or absorb specific miRNAs (Thind & Wilson 2016). The difference in the miRNA expression profile between cerebrospinal fluid and peripheral blood could not be analyzed in this study. Second, the different follow-up periods, different sample collection time points, and female gender effects for mRS outcomes may also bring heterogeneity that makes the results inaccurate. Due to the small number of included studies, this study did not conduct a stratified/subgroup analysis based on the above factors. Third, the original data used in training the machine learning model were derived from studies based on Danish and Spanish populations. Whether these variables still have sufficient predictive power in other populations still needs to be confirmed by research. Fourth, miRNAs are widely present in the human body. This study indicated that the expression levels of miR-152-3p, miR-221-3p, and miR-34a-5p were significantly lower in patients with good neurological outcomes. However, these miRNAs were not unique to SAH, which is also a limitation of this study.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by AA, BL and YC. The first draft of the manuscript was written by ZY and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by Appropriate Technology Extension Project of the Health and Health of the Autonomous Region in 2021(Clinician-led application and promotion of bedside intensive ultrasound technology) [SYTG-202172].

Data Availability

The datasets generated during and/or analysed during the current study are not publicly available but are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors have no relevant financial or non-financial interests to disclose.

Ethics Approval

This study was performed in line with the principles of the Declaration of Helsinki.