Circulating Non-coding RNAs as Potential Biomarkers for Ischemic Stroke:A Systematic Review

Recent studies have demonstrated that dysregulated non-coding RNAs (ncRNAs) are involved in the pathogenesis of ischemic stroke (IS) including neuroinammation, apoptosis, atherosclerosis and angiogenesis. However, discrepant results make it dicult to apply ncRNAs into clinical practice. 0.80 (95%CI: 0.76-0.83) for AUC, also a moderate diagnostic value.


Introduction
Stroke is a prevalent disease worldwide, with high disability and mortality. It has caused serious sociodemographic and socioeconomic burden in recent years, and the burden continues to increase (1). Ischemic stroke (IS) is a major subtype of stroke, for which thrombolysis (through the administration of tissue plasminogen), and thrombectomy (the surgical removal of emboli), are the main therapeutic methods. However, the number of patients bene ting from these therapies is relatively low because of the narrow therapeutic window, hemorrhagic transformation, and reperfusion injury. As a medical emergency with high morbidity, there is a lack of serum biomarkers for the diagnosis and prevention of IS (2,3). Recent studies have found that non-coding RNAs (ncRNAs) show differential expression during IS and play widespread roles in the different phases of the disease (4)(5)(6).
Recent studies have demonstrated that ncRNAs are involved in the pathogenesis of IS, including neuroin ammation, apoptosis, atherosclerosis, and angiogenesis (5,6). NcRNAs are de ned as nonprotein-coding transcripts and are classi ed into many types according to their length and connection, e.g. microRNAs (miRNAs), long non-coding RNAs (lncRNAs), and circular RNAs. lncRNAs are a type of linear RNA over 200 nucleotides in length, whereas miRNAs are 22 nucleotide long ncRNAs (7). As major members of the ncRNA family, lncRNAs and miRNAs often exhibit cell and tissue speci city. They also play extensive roles in pathophysiological events by regulating the expression of mRNA and protein (8)(9)(10). MiRNAs regulate gene expression at the post-transcriptional level by binding to the 3 prime untranslated region of mRNA, whereas lncRNAs exert their effects at the transcriptional and posttranscriptional levels (7,11). Recently, emerging studies have begun to reveal that ncRNAs can accumulate and be expressed in mammalian brain and in neuronal cell lines and play several roles in physiological and pathological activities in the brain (12).
To date, multiple dysregulated miRNAs and lncRNAs (upregulated or downregulated) have been identi ed from the middle cerebral artery occlusion model (MACO), the oxygen glucose deprivation/re-oxygenation cell model (OGD/R), and from the blood of IS patients. Stroke signi cantly alters the expression pro le of miRNAs and lncRNAs, indicating that these ncRNAs have the potential to be diagnostic and predictive biomarkers for IS (13)(14)(15)(16). However, discrepant reports make it di cult to apply ncRNAs into clinical practice. Moreover, not all ncRNAs are suitable as potential biomarkers for IS. The diagnostic potential of ncRNAs is yet to be determined; we performed this meta-analysis to evaluate and elucidate the diagnostic value of ncRNAs in IS.

2.1Search Strategy
We systematically searched the literature in four databases: PubMed, Web of Science, EMBASE, and the Cochrane Library, up to December 31, 2020. The literature on ncRNAs was retrieved using the following search strategies: (microRNA OR miRNA OR long non-coding RNA OR lncRNA OR non-coding RNA OR ncRNA) AND (diagnosis OR speci city OR sensitivity OR receiver operating characteristics OR ROC) AND (ischemic stroke OR cerebral infarction OR brain infarction OR cerebrovascular disease). We also checked references in these articles to identify further relevant studies. All studies identi ed were inspected by two independent reviewers, and disagreements, if any, were discussed until a consensus was reached.

2.2Inclusion Criteria
The eligible studies should meet the following criteria: (1) Studies were focused on the diagnostic performance of lncRNAs or miRNAs in IS; (2) Describing peripheral blood detection (plasma, serum, and whole blood); (3) All the patients were diagnosed as IS according to radiological imaging (CT or MRI) and neurological examination; (4) The su cient data including the sample size of case and controls, sensitivity (SEN), speci city (SPE) and area under curve (AUC) were available. In addition, the exclusion criteria are as follows: (1) Literature not written in English; (2) Reviews, meeting records or letters; (3) Studies not conducted in humans; (4) Not describing peripheral blood detection of lncRNAs and miRNAs.

2.3Data Extraction and Quality Assessment
Two researchers extracted information from included literature covering rst author, publication year, ethnicity, method, sample size of case and control, sample source, and necessary data including SPE, SEN and AUC. Quality assessment of these studies were carried out by two independent reviewers according to the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) (17), which contain 4 domains: patient selection, index test, reference standard, and ow and timing. Each domain is used for bias assessment and the rst 3 domains were applied to evaluate clinic applicability of these studies.
Data availability statement All data generated or analyzed during this study are included in this published article (and its supplementary information les).

2.4Statistical Analysis
The extracted data were analyzed using STATA 15.0 and p < 0.05 was considered statistically signi cant. Pooled sensitivity (SEN), speci city (SPE), positive likelihood ratios (PLR), negative likelihood ratios (NLR), diagnostic odds ratios (DOR), and corresponding 95% con dence intervals (95% CIs) were calculated using a random effects model. In addition, the summary receiver operator characteristic (SROC) curve and the area under the SROC curve (AUC) were analyzed to assess the diagnostic performance of lncRNAs and miRNAs for IS. A summary of the PLR and NLR, as well as the Fagan's nomogram, were used to evaluate the clinical applicability of ncRNAs for the diagnosis of IS. Cochran's-Q and I-squared (I 2 ) tests were applied for assessment of heterogeneity, with p < 0.1 or I 2 > 50% indicative of signi cant heterogeneity. ROC plane, sensitivity analysis, bivariate boxplot, and meta-regression were further conducted to explore the potential sources of heterogeneity. A group of analysis with a goodnessof-t, a bivariate normality, a in uence analysis and a outlier detection were used to validate our results. A funnel plot was used to examine publication bias.

3.1Study Characteristics
We searched 1458 articles in total from PubMed (904), Web of Science (389), EMBASE (152), and the Cochrane Library (13), with 634 and 824 articles for miRNAs and lncRNAs, respectively. Of these, 1364 records (566 and 798 for miRNAs and lncRNAs, respectively) were excluded after reading the title and abstract due to: duplication; not describing miRNAs or lncRNAs with IS; not a human model; not describing peripheral blood detection; or studies based on databases. Subsequently, 94 articles remained (66 and 28 for miRNAs and lncRNAs, respectively). After reading the whole text, 71 articles were excluded because they did not provide the necessary data, leaving 23 articles ultimately included for miRNAs and lncRNAs . A owchart of the whole selection process is shown in Figure 1.
Of these remaining articles, 15 were miRNA studies and included 1687 IS patients and 1267 controls. There were 27 sets of miRNAs in total, from which single and combined miRNAs (double, triple, and pentuple) were described, in terms of their diagnostic value for IS, respectively in 20 and 7 (3 sets of double, 2 sets of triple, and 2 sets of pentuple miRNAs) studies (Table 1). Meanwhile, 8 studies were conducted for lncRNAs, encompassing 741 qualifying patients and 774 corresponding controls. There were 18 sets of lncRNAs performed in total, from which single and combined lncRNAs (double and triple) were described in 13 and 5 (3 sets of double and 2 sets of triple lncRNAs) studies respectively ( Table 2) (Figure 8d). After excluding these 4 studies, the pooled estimates did not change, whereas the sensitivity and speci city heterogeneity decreased ( Figure  S5). To con rm this result, we conducted further analysis for heterogeneity using a goodness-oft, bivariate normality, in uence analysis and outlier detection, and found similar results ( Figure S6). We read these 4 studies again and carried out further meta-regression analysis on the bias of sample size (≤50 or 50), sample source (serum, plasma or whole blood) and publication year (in 5 years or not).
And we found that sample size and publication year may accounted for part of heterogeneity, with respectively p < 0.01 and p < 0.05 for sensitivity and speci city in sample size, and p < 0.01 for speci city in publication year (Figure 8b). Publication bias existed across the included studies on miRNAs, as identi ed by Deek's funnel plot (p < 0.05) ( Figure 9).
As for lncRNAs, signi cant heterogeneity exist among the 8 lncRNA studies in forest plots (SEN: I2=56.69, SPE: I2=74.97) ( Figure 5). Similarly, ROC plane, sensitivity analysis, bivariate boxplot, and metaregression analysis were carried out. ROC plane showed the signi cant heterogeneity did not result from threshold effect (Figure 10a). In uence analysis showed no study exert in substantial role in the results (Figure 10c). Three studies (Li 2019, Wang 2017, and Wang 2019) were out of the boxplot and may be the source of heterogeneity (Figure 10d). After excluding the 3 studies, the pooled estimates were similar to the previous results, whereas heterogeneity of sensitivity and speci city decreased (SEN: I 2 =27.71, SPE: Figure S7). Subsequently, a further analysis with a goodness-of-t, a bivariate normality, a in uence analysis and a outlier detection for heterogeneity showed the same result ( Figure S8). Other source of heterogeneity indicated by meta-regression analysis may be sample size (>50 or not) with a P<0.01 for sensitivity and a P<0.001 for speci city (Figure 10b). No signi cant publication bias was found in Deek's funnel plot with a P of 0.91 ( Figure 11).

Discussion
Several studies have reported aberrant expression of miRNAs and lncRNAs in IS patients (41)(42)(43)(44)(45)(46)(47)(48)(49)(50). However, the results of these studies were inconsistent and have not yet reached an consensus. Some are signi cantly up-regulated, by contrast others are down-regulated in the circulating blood of IS patients. The discrepant results among these studies make it di cult to apply ncRNAs to clinical practice. Therefore, we reviewed 15 studies of miRNAs and 8 studies of lncRNAs and performed a systemic metaanalysis to clarify confusions about the diagnostic value of circulating ncRNAs, which would give suggestions for clinical applicability.
Our meta-analysis revealed a moderate diagnostic performance for blood-based miRNAs for IS. Single miRNAs and combined miRNAs showed no signi cant difference in diagnostic accuracy as measured by AUC in subgroup analysis ( Figure S1 and Figure S2 in supplementary materials). Clinical applicability for diagnosis is dependent on the summary of PLR and NLR and Fagan's nomogram. In summary of the PLR and NLR, PLR >10 and NLR <0.1 represent for a higher diagnostic accuracy (51). Among the miRNAs,miR-107 from the study by Yang et al. (29) was in this scope (Figure 4a), hence miR-107 could be more bene cial in clinical diagnostics and deserves further research. In their study, circulating miR-107, miR-128b, and miR-153 are higher in IS patients than healthy controls, but miR-107 showed the greatest diagnostic value with the higher speci city and sensitivity. Yang et al. found miRNA-107 promoted excitatory neurotoxicity by regulating glutamate transporter-1 expression in I/R injury rat model (52). Additionally, the study by Cheng et al. has found that miRNA-107 participated in neuronal apoptosis process by p53 pathway in OGD-treated HT-22 hippocampal neuron cell (53). Recent study by Li et al. have demonstrated that miRNA-107 involved in angiogenesis after stroke by targeting Dice (45). In conclusion, miR-107 played an important role in the pathophysiological process of stroke. Hence, miR-107 is a novel candidate biomarker for the differentiation of IS patients from healthy individuals and needed further intensive study. Additionally, the diagnostic odds ratios of miRNAs was 16 , which further con rmed miRNAs as a promising diagnostic tool for IS.
Multiple studies have reported that single lncRNAs possess good diagnostic performance for clinical use (34,(36)(37)(38)(39)(40). Whereas Deng et al. (33) found that the combination of linc-DHFRL1-4, SNHG15 and linc-  (Figure 7a), whereas the combined lncRNAs were not in this range. Until now the the number of studies about combined lncRNAs is limited, which maybe contributed to the drawbacks. Hence, more and more larger studies are needed to access the diagnostic accuracy of combined lncRNAs for IS.
To our knowledge, this is the rst comprehensive meta-analysis to explore the potential value of ncRNAs in the diagnosis of IS. In our study, we not only used ROC plane, sensitivity analysis, bivariate boxplot, and meta-regression to identify the source of heterogeneity but also performed a goodness-of-t, bivariate normality, in uence analysis, and outlier detection to con rm the results. Moreover, both miRNAs and lncRNAs were included in our study, providing new potential biomarkers for clinical practice.
Additionally, we also assessed single and combined ncRNAs to explore the best diagnostic biomarkers for IS.
However, several limitations of this meta-analysis should be emphasized. First, signi cant heterogeneity existed among the involved studies and in the subgroup analysis. ROC plane were carried out to explore whether heterogeneity caused by threshold effect or not. And the results showed atypical shoulder arm, suggesting nil threshold effect in miRNAs and lncRNAs. Bivariate boxplot analysis revealed that there were respectively 4 studies and 3 studies contributed to the signi cant heterogeneity in miRNAs and lncRNAs analysis. After excluding these studies, the diagnostic e ciency were remain the same, while heterogeneity was decreased. Additionally, meta-regression analysis of miRNAs indicated that sample size and publish year may also account for high heterogeneity. Other factors, such as the variance of the measuring apparatus and the use of different cut-off values between studies may also be responsible for this heterogeneity. Second, nearly all the studies included in this meta-analysis analyzed data from Asian cohorts, and there is a lack of data regarding ncRNAs and IS in other ethnicities. Third, other sample types, such as cerebrospinal uid, may also be useful as biomarkers, and should be investigated in future studies. Fourth, as previously mentioned, most studies included in this analysis were retrospective case-control studies, as such they were limited by the different sample sizes and inclusion criteria between studies. Lastly, Deek's funnel plot suggested that publication bias exists for miRNAs. Studies without full texts, such as case reports, conference abstracts, and non-English literature, were excluded from our study, which likely contributed to this publication bias to some degree. Furthermore, negative results are less likely to be published, leading to the exaggeration of results in terms of diagnostic potency. Hence, the results here need to be interpreted with caution.
In conclusion, our study revealed that blood circulating ncRNAs could be a moderately effective candidate biomarker for the diagnosis of IS. Furthermore, while combined lncRNAs showed more accurate diagnostic properties than single lncRNAs, some single miRNAs (e.g. miR-107) showed better diagnostic performance and require more attention. Further research with more extensive statistics should be conducted to con rm our analysis in the future. Studies regarding other types of ncRNAs, such as circular RNAs, are also needed to better understand the diagnostic potential of ncRNAs.   Deek's funnel plot of miRNAs Deek's funnel plot of lncRNAs

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.