Circular RNA as biomarkers for acute ischemic stroke: A systematic review and meta‐analysis

Abstract Background Rapid diagnosis of acute ischemic stroke (AIS) patients is still challenging, and reliable biomarkers are needed. Noncoding RNAs are important for many physiological activities, among which circular RNAs (circRNAs) have been proven to be more tissue‐specific and conservative. Many recent studies found the potential of circRNAs as biomarkers for many diseases, including cardiovascular diseases, cancers, and ischemic stroke. This systemic review and meta‐analysis aimed to identify circRNAs as potential biomarkers for AIS. Methods This study has been prospectively registered in PROSPERO (Registration No. 11 CRD42021288033). Published literature comparing circRNA expression profiles between AIS and non‐AIS in human and animal models were retrieved from the articles published by January 2023 in major databases. We descriptively summarized the included studies, conducted meta‐analysis under a random effects model, and did bioinformatics analysis including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. Results Totally 23 studies were included, reporting 18 distinctive upregulated and 20 distinctive downregulated circRNAs. Diagnostic meta‐analysis indicated discriminative ability of the circRNAs. Furthermore, circRNA HECTD1, circRNA DLGAP4, circRNA CDC14A, circRNA SCMH1, and circRNA TLK1 were reported with the same regulation trend in more than one study (animal studies included). GO and KEGG enrichment analyses indicated that the target genes of these five circRNAs were enriched in regulating cell proliferation, apoptosis, and oxidative stress. Conclusions This study demonstrates that circRNAs (circRNA HECTD1, circRNA DLGAP4, circRNA CDC14A, circRNA SCMH1, and circRNA TLK1) generally are promising as potential biomarkers for AIS. However, due to the limited number of studies, diagnostic value of individual circRNA could not be validated. More in vitro and in vivo functional studies are needed.


| INTRODUC TI ON
The incidence of cerebrovascular diseases is increasing worldwide, of which acute ischemic stroke (AIS) has high mortality and morbidity, making it one of the top causes of death. 1 The early and accurate diagnosis of AIS and early beginning of treatment greatly improve AIS patients' prognosis. And understanding the molecular mechanisms behind AIS could help to identify applicable diagnostic biomarkers.
Increasing evidence has proved that noncoding RNAs play important roles in many physiological and pathophysiological processes. 2,3 The circular RNA (circRNA) is a subclass of noncoding RNAs, characterized by a covalently closed loop that enables it to escape the degradation by nuclease. 4 Furthermore, circRNAs were found to be more tissue-specific and evolutionally conserved.
Current studies demonstrated that circRNAs functioned through combining with miRNAs or proteins and reacting with RNA polymerase II, thus regulating epigenetic modification and transcription. 3,5 Besides, some circRNAs could encode proteins with an insertion of internal ribosome entry site in the upstream of start codons. 6 Research and meta-analyses found that certain kinds of circRNAs were associated with cancer, 7,8 cardiovascular diseases, 9 and other diseases, which indicated its possible utility in disease diagnosis and treatment. In recent years, several studies explored the differences in circRNA profiles between AIS patients and healthy people. However, there is inconsistency among these studies, concerning research technologies, statistical analysis methods, cut-off value, and so on. Therefore, it could be challenging to integrate individual results and comprehend the function of circRNAs in AIS.
Systemic review and meta-analysis could be effective and optimal approaches to combine results from various studies. Therefore, we did this study on the published and registered research that explored the differently expressed circRNAs in AIS patients and animal models of AIS, hoping to provide information for future clinical application of circRNAs.

| ME THODS
The present study has been prospectively registered in PROSPERO (Registration No. 11 CRD42021288033) and was developed following the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols statement (PRISMA checklist and abstract checklist).

| Searchstrategies
Four major English databases, PubMed, EMBASE, ISI web of science, and Cochrane Library were thoroughly searched to investigate the association between circRNA profiling and AIS. The following terms were used: "circular RNA" or "circRNA," and "stroke" or "cerebral infarction" or "cerebral ischemia" or "cerebral vascular accident" or "brain infarction" or "brain ischemia" or "brain vascular accident" from January 2010 to January 2023. Two authors carefully examined the titles and abstracts of retrieved records, and any disagreements were resolved by consensus.

| Selectionstandards
Eligible studies had to meet the following inclusion criteria: (a) cir-cRNA expression studies on AIS patients with approximate ageand sex-matching healthy controls for comparison; (b) circRNA expression studies on rodents of stroke model with age-and sex-matching healthy mice/rats for comparison; (c) using realtime quantitative reverse transcription polymerase chain reaction (qRT-PCR), circRNA microarray, and/or RNA-sequencing technologies; and (d) reporting selection criteria of differentially expressed circRNAs.
For participant inclusion, patients must: (a) be confirmed of AIS with magnetic resonance imaging or computer tomography circRNA CDC14A, circRNA SCMH1, and circRNA TLK1 were reported with the same regulation trend in more than one study (animal studies included). GO and KEGG enrichment analyses indicated that the target genes of these five circRNAs were enriched in regulating cell proliferation, apoptosis, and oxidative stress. For animal experiments, animals used should: (a) be adult, and have detailed reports of feeding condition; (b) have no gene-edition, medication, or any other treatment; (c) have histological and/or behavioral confirmation of brain damage after occlusion; (d) have specimens acquired within 24 hours after occlusion.

| Dataextractionandqualityassessment
Information from the full texts and supplementary files of the selected studies were collected, which included: first author, publication year, selection of AIS patients and healthy controls, research technologies, cut-off criteria and fold changes (FC), names, and analysis of the significantly dysregulated circRNAs and so forth. When values were not provided in articles or supplementary materials, the extraction of statistical data from graphs was performed with WebPlotDigitizer (Version 4.4). Besides, the accuracy of this method was verified with studies providing both graphs and detailed values. The cut-off threshold was set as |FC| > 1.5 or <0.6, and p < 0.05.
Because there is no scale to evaluate all the used platforms, the new quality assessment scale from Li′s meta-analysis 9 was adopted, which consisted of five parts: (a) course design, (b) detailed description of the samples, (c) description and representation of case and control, (d) annotation of the platform and naming convention of circRNAs, and (e) raw data processing and data analysis. Full score for each part was 2, 0 for undescribed part, while 1 for incomplete description. Furthermore, Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) 10 was adopted to evaluate the included studies for the diagnosis value of circRNAs. The QUADAS-2 consists of four main domains, patient selection, index test, reference standard, and flow and timing, each of which is assessed in terms of risk of bias as "low," "high," or "unclear." The assessment was done by two qualified members of our team, and consistency was reached for final score.

| Statisticalanalysis
Data collection and organization were done with Excel (2019 Microsoft). Statistical analysis of the diagnostic tests was executed STATA 15.1. Q-tests and I 2 statistics were used to estimate the heterogeneity with either p < 0.10 or I 2 > 50% suggesting the existence of substantial heterogeneity. A random-effects model was applied to quantify the pooled sensitivity, specificity, log diagnostic odds ratio (LDOR), and area under curve (AUC), with corresponding 95% confidence intervals (CIs). Spearman correlation analysis was conducted to verify the threshold effects. For the test of robustness, a sensitivity analysis was performed to attain more accurate results.
Publication bias was evaluated with Deek's funnel plots. All tests were two-tailed and p < 0.05 was considered statistically significant. F I G U R E 1 Flow diagram of literature inclusion. Identification, screening, eligibility extraction, and inclusion steps of studies were depicted.

| Analysisforsecondaryoutcome
The circRNAs reported in several studies were further selected for bioinformatic analysis. According to ceRNA theory, the targeted miRNAs for circRNAs were predicted with CircInteractome database, and the targeted mRNA/genes were predicted with miRDB database. The competing endogenous RNAs (ceRNA) network for the first-ranking miRNA and the top 20 mRNAs were demonstrated with Cytoscape (version 3.9.0). Furthermore, we used Gene  Table 1 showed the information of the included studies, and the quality evaluation of the included studies was shown in Tables S1 and S2. Most studies lacked or were insufficient in describing the prespecified thresholds, standard reference tests and annotating the circRNAs.
Among these studies, 10 had divided experiments into exploration section, with microarray or RNA sequencing, and verification section, with qRT-PCR. These studies chose random or the top ranking circRNAs differentially expressed in sequencing data for further PCR verification. Table 2 showed details into circRNAs verified with PCR of the included studies, including FC, P, and the originating genes and positions of the circRNAs. In total, 18 distinctive upregulated and 20 distinctive downregulated circRNAs were reported in human studies, among which the upregulated ones, circ HECTD1 and circ CDC14A, and the downregulated one, circ DLGAP4, were reported in more than one study. These circRNAs spread across the autosomes, and two were in X chromosome, hsa_circ_0007290 in FUNDC1 gene and hsa_circ_0090002 in PHKA2 gene. Since circRNAs were proven to be highly conserved in different species, animal experiments could be illustrative. Twelve distinctive upregulated and 14 distinctive downregulated circRNAs were reported in animal studies. Circ HECTD1, circ RBM33, circ DLGAP4, circRNA TLK1, and circ SCMH1 were shown to be differentially expressed both in human and rodent. However, circ RBM33 were reported to be upregulated in AIS patients' plasma, but downregulated in the blood and brain tissue of mice. While circRNA CDC14A was reported in two independent human studies.

| Sensitivityanalysisandmeta-regression
Because of the overall high heterogeneity in results with the random effects model, we further performed sensitivity analysis. We postulated that sample size was the main bias affecting the results, however, the results of meta-analysis excluding two studies with the smallest sample size was still high in heterogeneity (sensitivity: I 2 = 99.37%, p < 0.001; specificity: I 2 = 98.46%, p < 0.001).
Furthermore, meta-regression suggested that standard test, index test, and subject descriptions all had no significant contribution to heterogeneity.

| Bioinformaticsanalysis
CircRNAs that were reported to have the same trend of regulation in more than one human study or at least one human and one animal study were selected, which were circRNA HECTD1, circRNA DLGAP4, circRNA TLK1, circRNA SCMH1, and circRNA CDC14A.

| DISCUSS ION
Acute ischemic stroke is the second leading cause of death and disability worldwide, and timely removal of blockade was crucial for improving patients' outcomes. Therefore, it is imperative to diagnose AIS early, which currently depends largely on physicians' experience, and some accurate and practical biomarkers are expected. The circRNA is a class of noncoding RNAs and recently has been proved by many studies to be multifunctional in many diseases. Research found that brain has the most tissue-specific circRNAs, indicating their roles in regulating bioprocesses of the brain. 11,12 Several metaanalyses have shown that circRNAs are promising as diagnostic biomarkers and treatment targets in cardiovascular diseases and cancers, [7][8][9] and some individual studies have already explored their roles in AIS. Therefore, we hoped to update the comprehensive knowledge of circRNA profile in AIS through this systemic review and meta-analysis, which to the best of our knowledge, was the first to integrate circRNA studies in AIS patients and animal models. Parkinson's disease, all by regulating the most probable target, miR-143, but different downstream target genes. [16][17][18] We did KEGG analysis for miR-143 alone and found that MAPK is the most enriched pathway, which is a well-known pathway involved in proinflammation, apoptosis, and growth. Two recent human studies proved that circRNA CDC14A, which is made by reverse splic- Previous studies revealed that activation of the PI3K/Akt signaling pathway could inhibit cerebral cell apoptosis, attenuating ischemia/ reperfusion (I/R) injury. 28,29 MAPK signaling pathway regulates various basic cellular processes such as cellular proliferation, differentiation, migration, metabolism, and apoptosis. Mounting evidence indicated that activation of p38 MAPK and ERK1/2 MAPK was associated with inflammatory reaction, and abnormal blood-brain barrier. 30,31 Several included studies explored the feasibility of circRNA as diagnostic biomarkers for acute ischemic stroke. Peng et al. 15  This study has the following limitations. Because of the small number of studies exploring circRNAs in AIS and various nomenclature methods, we uniformly utilized the originating gene symbols to represent different circRNAs. Although there were previous studies that adopted this method, deviations might exist among circRNAs of different IDs. Besides, because the raw data of some studies were not public and the primary results of RNA sequencing could be complicated and probably not robust, we only did further diagnostic analysis and bioinformatic analysis of differentially expressed circRNAs detected by qRT-PCR. In addition, there is high heterogeneity regarding the outcomes of included studies, therefore pooling these data had a risk due to inherent uncertainty. Finally, the functions of circRNAs were still relatively unclear and there is still no practical use of circRNA in clinic, more laboratory experiments were needed to validate the current results.

| CON CLUS ION
In conclusion, this systemic review and meta-analysis summarizes circRNA profiling studies of acute ischemic stroke. Results illustrated the general promising performance of circRNAs as biomarkers for ischemic stroke. Five most likely important circRNAs, which were circRNA HECTD1, circRNA DLGAP4, circRNA CDC14A, circRNA SCMH1, and circRNA TLK1, could be the primary targets for future research.

CO N FLI C TO FI NTE R E S TS TATE M E NT
No of the authors have any disclosures.

DATAAVA I L A B I L I T YS TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.