Association of PAI-1 rs1799889 Polymorphism with Susceptibility to Ischemic Stroke: a Huge Meta-Analysis based on 44 Studies

A B S T R AC T Background: the PAI-1 rs1799889 polymorphism has been reported to be associated with susceptibility to ischemic stroke. However, the results of previous studies have been inconsistent or controversial. Hence, we performed a systematic review and meta-analysis to evaluate the association of PAI-1 rs1799889 polymorphism with ischemic stroke risk. Methods: A comprehensive literature search was performed on PubMed, Web of Science, Scopus, SciELO, CNKI, and CBD databases up to November 05, 2019. Pooled odds ratio (OR) with 95% confidence interval (CI) were used to access the strength of this association in fixedor random-effects model. Results: A total of 44 case-control studies with 8,620 cases and 10,260 controls were selected. Pooled data showed a significant association between PAI-1 rs1799889 polymorphism and ischemic stroke risk in the overall populations (GG vs. AA: OR = 0.791, 95% CI 0.633–0.988, p = 0.039; GA vs. AA: OR = 0.807, 95% CI 0.683–0.953, p = 0.012; and GG+GA vs. AA: OR = 0.795, 95% CI 0.637–0.993, p = 0.043). Subgroup analysis by ethnicity revealed a significant association in Asian and Mixed populations, but not in Caucasians. Moreover, stratified analysis by country of origin revealed an increased risk of ischemic stroke in Chinese populations, but not among Dutch (Netherlands) and Swedish. Conclusions: This meta-analysis result suggested that PAI-1 rs1799889 polymorphism was associated with an increased risk of ischemic stroke, especially in Asian and Mixed populations.


INTRODUCTION
Stroke is the second leading cause of death globally and leading cause of long-term disability worldwide (1). It pos es a huge threat to public health and is the leading cause of death in developed and developing countries (2). It is estimated that approximately 70% of new strokes are is chemic in origin, 51% stroke death, and 58% of stroke dis ability-adjusted life years are because of ischemic stroke (3). The exact etiology of ischemic stroke is multifactorial and a complex interaction between modifiable and nonmodifiable conventional risk and genetic factors could be behind the pathogenesis of this disease (4). Several var iants at lowpenetrance and highpenetrance genes have been identified as potential ischemic stroke susceptibility loci. Numerous studies have found that Plasminogen acti vator inhibitor1 (PAI1) also serpin E1 was involved in the pathogenesis of ischemic stroke (5). Therefore, PAI1 gene polymorphisms and its circulating levels may be associat ed with the development of ischemic stroke (5,6).
Human PAI1 gene is located at chromosome 7q21.3-q22, contains 9 exons and spans 12.3 kb (7). PAI1, a se creted singlechain glycoprotein, is one of the early inflammatory response genes, and its expression level changes dramatically in response to many stimuli, in cluding growth factors and endotoxins (8,9). Several polymorphisms within the PAI1 gene have clearly been postulated to modulate the expression of PAI1 (10,11). Among SNPs of the PAI1 gene, rs1799889 (4G/5G) pol ymorphism has been extensively studied in different disease (7,12). PAI1 rs1799889 is an inserted or deleted in the 4G sequence polymorphism in the PAI1 promoter (4G/5G) at 675 bp upstream from the start of transcrip tional start site in the promoter region. Studying the as sociation of PAI1 gene with different disease will help us to understand the mechanism of PAI1 regulation and the role of PAI1 in many physiological and pathological pro cesses (12,13).
Studies have shown that the 4G/4G genotype has been linked to higher PAI1 level, compared with the 5G/5G gen otype, with the heterozygous genotype associated with in termediate levels. In 2003, Chen et al., have reported that PAI1 rs1799889 polymorphism alone is not associated with an increased risk of ischemic stroke. However, they revealed a significant contribution of PAI1 4G/4G geno type with an increased triglyceride and decreased HDL cholesterol levels in the healthy group (14). There are sev eral numbers of epidemiological studies have evaluated association between PAI1 rs1799889 polymorphism and ischemic stroke risk, but their results were inconsistent or even contradictory. For example, Adamski et al., and EsparzaGarcía et al., have reported that PAI1 rs1799889 polymorphism was not associated with an increased risk of ischemic stroke in Polish and Mexican populations, re ceptively (15,16). By contrast, Xu et al. results supported that PAI1 rs1799889 polymorphism might be associated with an increased risk of ischemic stroke in Han Chinese (17). In recent years, some studies already studied po tential associations PAI1 rs1799889 polymorphism with risk of ischemic stroke. Nevertheless, the results of these studies were not always consistent and the sample size of each study was also statistically insufficient. Thus, we performed a metaanalysis to offer a more comprehensive estimation of the association between PAI1 rs1799889 and ischemic stroke susceptibility in globally populations.

INCLUSION AND EXCLUDING CRITERIA
The inclusion criteria for the gene association studies in this metaanalysis were as follows: 1) studies with casecontrol or cohort design; 2) fulltext published stud ies; 3) studies evaluated the association between PAI1 rs1799889 polymorphism and ischemic stroke risk; and 4) provided the genotype distribution in both cases and controls for estimating an odds ratio (OR) with 95% con fidence interval (CI). Additionally, studies were excluded if one of the following criteria was fulfilled: 1) studies without detailed raw data regarding PAI1 rs1799889 pol ymorphism; 2) case only studies; 3) family-based, sibling, twins and linkage studies; 4) abstracts, review, letters, comments, conference editorials, presentations, case re ports, case series previous metaanalyses; 5) duplicates or overlapping studies. If the authors published two or more studies using the same data (with overlapping data), the newest publication or the publication with the largest sample size was included. There was no any limitation by ethnicity, race, placed or geography area. Moreover, nonEnglish publications were translated and included in the metaanalysis.

DATA EXTRACTION
Two authors (HN and MJA) systematically extracted data from all eligible studies using a standardized form. Then, they have checked the data extraction results and reached consensus. If different results were generated, the two authors carried out discussions until a consensus was reached or a third author was invited to resolve the disagreement and then a final decision were made by the majority of the votes. The collected data were: first au thor's name, publication year, country of origin, ethnic ity (Caucasian, Asian, African, Mixed populations), total numbers of cases and controls, genotypes frequencies of cases and controls, minor allele frequencies (MAFs) and HardyWeinberg equilibrium test in control subjects.

STATISTICAL ANALYSIS
An ethical approval was not necessary as this study was a metaanalysis based on previous studies. The strength of the associations PAI1 rs1799889 (4G/5G) polymorphism and susceptibility to ischemic stroke was measured by odds ratios (ORs) with 95% confidence intervals (CIs). The statistical significance of the pooled OR was determined using the Ztest. Pooled estimates of the OR were obtained by calculating a weighted average of OR from each study. The pooled ORs was calculated under all five genetic mod els, i.e., allele (G vs. A), homozygote (GG vs. AA), hetero zygote (GA vs. AA), dominant (GG+GA vs. AA) and reces sive (GG vs. GA+AA). Betweenstudies heterogeneity was assessed by a Chisquared Qtest and I 2 statistics (P < 0.05). The heterogeneity between studies was estimated by Co chran's χ2 based Qstatistic test, in which it was consid ered to be statistically significant at P ≤ 0.01. In addition, I 2 test was used to quantify the effect of heterogeneity, with the range of 0 to 100%, and 0-40% meant no risk of heterogeneity, 30-60% meant a low risk of heterogeneity, 50-90% meant substantial heterogeneity and 75-100% meant considerable heterogeneity. Accordingly, when be tweenstudy heterogeneity existed a randomeffects mod el weighted (the DerSimonianLaird method) was applied to give a more conservative result; otherwise, a fixedef fects model weighted (the MantelHaenszel method) method was selected. HardyWeinberg equilibrium (HWE) of the genotype distribution in controls was conducted by Pearson's χ2 test, in which it was considered to be sta tistically significant at P ≤ 0.05. A subgroup analysis by Fig. 1 The study selection and inclusion process.  ethnicity, country of origin, and source of controls was performed to explore potential sources of betweenstudy heterogeneity (18,19). To check the stability of the pooled data, a sensitivity analysis was performed by omitting each individual study in turn from the all selected stud ies and reanalyzing the remainder. Moreover, sensitivity analysis was performed by excluding HWEviolating stud ies. The potential publication bias was explored visually by Egger's linear regression test and Begg's quantitative test (20). The asymmetric plot of Egger's test and the Pvalue of Begg's test less than 0.05 were considered a significant publication bias. All statistical analyses were performed using Comprehensive MetaAnalysis (CMA) Software ver sion 2.0 (Biostat, Englewood, NJ). All tests were twosided, and the P values of < 0.05 were considered statistically significant.   Table 1. More over, the distribution of genotypes in the controls was in agreement with HardyWeinberg equilibrium (HWE) for all selected studies, except for seven studies (Table 1).

QUANTITATIVE DATA SYNTHESIS
The summary of the metaanalysis of the association of be tween PAI1 rs1799889 polymorphism and ischemic stroke are shown in Table 2. Pooled data revealed that there was a significant association between PAI1 rs1799889 poly morphism and an increased risk of ischemic stroke in the overall population under three genetic models, i.  Fig 2B). More over, we have performed subgroup analyses by ethnicity and country of origin. Subgroup analysis by ethnicity showed that there was a significant association between PAI1 rs1799889 polymorphism and ischemic stroke risk in

BETWEEN-STUDY HETEROGENEITY TEST
As shown in To explore the potential sources of heterogeneity, subgroup analyses by ethnicity, country of origin and HWE was performed. The results suggested that the above mentioned factors did not contribute to be tweenstudy heterogeneity in the current metaanalysis.

SENSITIVITY ANALYSIS
A sensitivity analysis was used to test the effects of each study on pooled ORs. There were no significant differ ences observed upon removal of any of the studies, sug gesting that our findings were statistically robust and reliable. Moreover, we performed sensitivity analysis by excluding the HWEviolating study (Figure 3). When this study was excluded, the results were not changed in overall population and also by subgroup analyses, indi cating that our metaanalysis was statistically robust and reliable.

PUBLICATION BIAS
Begg's funnel plot and Egger's test were inspected to evaluate the possible publication bias in this metaanalysis. Results of publication bias were shown in Table 2 and Figure 4. The shape of the funnel did not show any obvious asymmetry in all of the genetic models. Moreover, Egger's test was statis tically revealed that there was no a significant bias under all five genetic models in the overall populations all five genetic models, i.e., allele (P Beggs = 0.112; P Eggers = 0.859), homozygote (P Beggs = 0.198; P Eggers = 0.307), heterozygote (P Beggs = 0.107; P Eggers = 0.267), dominant (P Beggs = 0.172; P Eggers = 0.841), and recessive (P Beggs = 0.723; P Eggers = 0.876).

DISCUSSION
The PAI1 rs1799889 polymorphism association to ischemic stroke was first described by Catto et al. in 1997 (44). Since several epidemiological studies have been evaluated asso ciation between PAI1 rs1799889 polymorphism and risk of ischemic stroke (17,45). However, the results of these studies remain contradictory. It is clear that a single study may fail to demonstrate a complicated genetic relationship completely because of small sample size, which has low statistical power. Larger studies could overcome these dis advantages. Therefore, we performed a comprehensive me taanalysis of all eligible studies evaluated the association of PAI1 rs1799889 polymorphism with risk ischemic stroke.    In the current metaanalysis, we have selected a total of 44 eligible casecontrol studies with 8,620 ischemic stroke cases and 10,260 controls to evaluate the association of PAI1 rs1799889 polymorphism with ischemic stroke risk. Our pooled data showed that PAI1 rs1799889 polymor phism was significantly associated with an increased risk of ischemic stroke in the overall population. Moreover, subgroup analyses revealed that PAI1 rs1799889 polymor phism was associated with significantly increased risk of ischemic stroke in Asian and mixed populations, but not in Caucasians. When stratified analysis by country of ori gin performed a significant association was found among Chinese population, but not in Dutch (Netherlands) and Swedish. This finding indicated that the carriers with the 4G allele of the PAI1 rs1799889 polymorphism in Asians and mixed populations might be predisposed to ischemic stroke, but not in Caucasian populations. Moreover, this finding suggested a possible influence among environ mental exposures and different genetic backgrounds in development of ischemic stroke in different populations. Therefore, more studies are warranted to further vali date genetic background difference in the effect of PAI1 rs1799889 polymorphism in susceptibility to ischemic stroke, especially in Caucasians. Cao et al., in a metaanal ysis of eleven casecontrol studies with 1,358 cases and 1,134 controls evaluated the association of PAI1 rs1799889 polymorphism with susceptibility to ischemic stroke in the Chinese population. Their results showed a signifi cant association between PAI1 rs1799889 polymorphism and ischemic stroke risk. However, their metaanalysis results reliability and the number of studies are consid erably smaller than that needed to receive the robust con clusions (45). Here, we have extended the metaanalysis with a more relevant recently published studies and sub group analysis by ethnicity. Moreover, Hu et al., in me taanalysis of 39 studies with 8,336 cases and 14,403 con trols evaluated PAI1 polymorphisms with risk of stroke. Their results revealed a significant association between PAI1 rs1799889 polymorphism and an increased risk of ischemic stroke in adult, but not pediatric. Their strati fied analysis showed a significant association in Asians, but not Caucasians. Moreover, they found that PAI1-844 G>A, but not 11,053 T>G polymorphism was associated with an increased risk of ischemic stroke and a tendency  of PAI1 rs1799889 polymorphism towards a decreased risk of hemorrhagic stroke (50).
Betweenstudy heterogeneity is a common issue in a metaanalysis on genetic association (51)(52)(53). It could be attributable to differences in several factors such as environmental factors, including criteria or methodolog ical factors in design and conduct of the studies (54,55). Thus, identifying the potential sources of heterogeneity is one of the most important goals of metaanalysis. When all the eligible studies were pooled in this metaanalysis, there was significant betweenstudy heterogeneity under all genetic models. However, after subgroup analyses by ethnicity the heterogeneity not effectively disappeared or decreased, which indicated that ethnicity did not play a crucial role in the existence of betweenstudy heteroge neity in the current metaanalysis.
The current metaanalysis had some advantages. First, this was the most comprehensive and accurate me  taanalysis to evaluate association of PAI1 rs1799889 pol ymorphism with ischemic stroke, which involved Asian, Caucasian, mixed populations. Second, the current me taanalysis search not restricted to studies published in indexed journals. Third, we have evaluated the association under all five genetic models. Forth, there was no evidence of publication bias by Begg's funnel plot and Egger's test in this metaanalysis. Finally, sensitivity analysis confers the reliability and stability of our pooled data. However, some limitations of this metaanalysis should be mentioned. First, the sample size of the included studies was not large enough by ethnicity among African and Mixed popula tions. Therefore, there was a lack of statistical power to bet ter calculate association of PAI1 rs1799889 polymorphism with risk of stroke among African and Mixed populations. Second, all included studies were published in English or Chinese which may be brought some bias. Third, in this metaanalysis betweenstudy heterogeneity was detected under all five genetic models in the overall population and by subgroup analyses, which may be distorting the pooled data. Finally, our results were based on singlefactor esti mations without adjustment for other risk factors such as age, gender, and environmental factors.
In summary, this metaanalysis result revealed that PAI1 rs1799889 polymorphism was significantly associat ed with an increased risk of ischemic stroke, especially in Asian populations. Moreover, there was a significant as sociation between PAI1 rs1799889 polymorphism and is chemic stroke risk. Future studies with large sample sizes and well designs in the Mixed and African populations and genegene and geneenvironment interaction studies are warranted to confirm these findings.