GRECOS Project (Genotyping Recurrence Risk of Stroke)
The Use of Genetics to Predict the Vascular Recurrence After Stroke
Background and Purpose—Vascular recurrence occurs in 11% of patients during the first year after ischemic stroke (IS) or transient ischemic attack. Clinical scores do not predict the whole vascular recurrence risk; therefore, we aimed to find genetic variants associated with recurrence that might improve the clinical predictive models in IS.
Methods—We analyzed 256 polymorphisms from 115 candidate genes in 3 patient cohorts comprising 4482 IS or transient ischemic attack patients. The discovery cohort was prospectively recruited and included 1494 patients, 6.2% of them developed a new IS during the first year of follow-up. Replication analysis was performed in 2988 patients using SNPlex or HumanOmni1-Quad technology. We generated a predictive model using Cox regression (GRECOS score [Genotyping Reurrence Risk of Stroke]) and generated risk groups using a classification tree method.
Results—The analyses revealed that rs1800801 in the MGP gene (hazard ratio, 1.33; P=9×10−03), a gene related to artery calcification, was associated with new IS during the first year of follow-up. This polymorphism was replicated in a Spanish cohort (n=1.305); however, it was not significantly associated in a North American cohort (n=1.683). The GRECOS score predicted new IS (P=3.2×10−09) and could classify patients, from low risk of stroke recurrence (1.9%) to high risk (12.6%). Moreover, the addition of genetic risk factors to the GRECOS score improves the prediction compared with previous Stroke Prognosis Instrument-II score (P=0.03).
Conclusions—The use of genetics could be useful to estimate vascular recurrence risk after IS. Genetic variability in the MGP gene was associated with vascular recurrence in the Spanish population.
Stroke remains a major cause of death and disability. In the European Union, estimates showed 2 700 000 recurrent stroke cases, and 536 000 incident stroke cases per year.1,2 The risk of stroke recurrence after a first stroke is high, especially at the early stages, being around 6% to 12% within the first year of the initial stroke. Moreover, stroke patients also confront a high risk of developing other vascular diseases, such as acute myocardial infarction and vascular death. Data suggest that within 10 years of ischemic stroke or transient ischemic attack (TIA), around 60% of patients will die and 54% will experience a new vascular event. In the European Union countries, the total annual direct and indirect costs of stroke are estimated at €27 billion.3,4
Unfortunately, we cannot clearly predict individualized stroke recurrence risk. Therefore, identification of patients at highest risk of recurrent stroke, including through the development of specific measures for early detection and prevention, remains a high priority. Several modifiable, lifestyle-associated risk factors, such as physical inactivity, smoking, and alcohol consumption, are implicated in vascular recurrence.5 Moreover, age, hypertension, atrial fibrillation, hyperlipidemia, homocysteine levels, and diabetes mellitus also increase risk of stroke recurrence.6
Two main scores including clinical variables have been generated to predict vascular recurrence after ischemic stroke, such as the Essen Stroke Risk Score (ESRS)7 and the Stroke Prognosis Instrument-II (SPI-II)8; however, the predictive ability of these models is limited and not useful for clinical practice.9
Although genetic background explains some of the interindividual differences in cardiovascular events recurrence, few studies, with a sample size higher than 1000 subjects, have analyzed the role of individual genetic background in the risk of vascular recurrence among first-ever ischemic stroke patients.10–13
We aimed to identify genetic variants associated with vascular and with ischemic stroke recurrence to estimate the ability of a clinical-genetic model to predict cardiovascular recurrence after a first-ever ischemic stroke.
The Discovery cohort (cohort A; n=1494) consisted of consecutive white patients with a first ischemic stroke that were admitted to the emergency room of 1 of the 23 participant Spanish hospitals (online-only Data Supplement) between July 2005 and May 2009. Replication cohort B (n=844) and C (n=461) were recruited at Vall d’Hebron Hospital and 5 more Spanish hospitals as a part of Genot-PA project between November 2000 and May 2005.14 Replication cohort D (n=1683) was composed of European ancestry individuals recruited between 1997 and 2001 in North America and Scotland as a part of the VISP (Vitamin Intervention for Stroke Prevention) clinical trial.15 We used a prospective cohort study design and considered patients with a vascular recurrence as cases and patients without a vascular recurrence as controls. We analyzed 2 end points: (1) ischemic stroke recurrence; and (2) vascular recurrence (new ischemic stroke, myocardial infarction, peripheral vascular disease, or cardiovascular death). This study was approved by local ethics committee, and written informed consent was obtained from all individuals.
Included patients from cohort A met the following criteria: First episode of persistent focal neurological deficit of <1 hour (TIA) or >1 hour with neuroimaging (computed tomography or magnetic resonance imaging) confirming an ischemic stroke, ruling out other pathology for both TIA and ischemic stroke. We excluded individuals who have previously been diagnosed with a stroke, individuals with modified Rankin Scale (mRS) at discharge ≥4 and with a life expectancy <1 year at the time of inclusion, and patients participating in a clinical trial of secondary prevention of stroke. Patients with a presence of diseases that are associated with a life expectancy of <1 year, such as cancer or AIDS in advanced stages, were excluded. On the basis of their experience and information provided by other physicians, the neurologists decided whether to include or exclude each patient from the study.
We enrolled the patients during the first 24 hours after stroke and performed a follow-up for at least 1 year to record the occurrence of ischemic stroke recurrence or global vascular recurrences. Ischemic stroke recurrence was defined by the occurrence of a new ischemic stroke after first-ever stroke. Vascular recurrence was defined by a new ischemic stroke, myocardial infarction, peripheral vascular disease, or cardiovascular death. We used standard questionnaires to capture follow-up clinical and demographic data for each participant during phone calls performed every 3 months and performed Morisky–Green tests for treatment compliance.16
Replication cohorts (cohorts B–C) were recruited as a part of the Genot-PA study,14 clinical data regarding ischemic stroke were assessed in the emergency room or during hospitalization, and vascular recurrence data and follow-up data were obtained by reviewing the clinical records and by telephone interview. The VISP trial was a multicenter clinical trial that enrolled patients aged ≥35 years with a nondisabling cerebral infarction defined as an ischemic brain infarction not because of embolism from a cardiac source. The trial was designed to determine whether daily intake of a multivitamin tablet reduced ischemic stroke recurrence events. The recurrence was detected by telephone contacts and in-clinic visits every 3 months for up to 2 years. Patients were excluded if the presence of cancer, pulmonary disease, or another illness was observed which, in the opinion of the study physician, would limit the life expectancy of the patient to <2 years.
The VISP cohort differed from the discovery cohort A with regard to the B vitamin intervention and exclusion of patients with TIAs or cardioembolism.
We identified 256 single nucleotide polymorphisms (SNPs) in 115 genes selected from the literature in May 2010 (Methods in the online-only Data Supplement) related to angiogenesis, stroke, hypertension, and other pathways associated with stroke (Table I in the online-only Data Supplement).
Genotyping was performed using SNPlex technology (Applied Biosystems, Foster City, California). For quality control, 2 HapMap samples (NA10860/NA10861) were included, and their genotype concordance was verified. The 256 SNPs reached the minimal call rate.
VISP samples (cohort D) were genotyped at the Johns Hopkins Center for Inherited Disease Research, using the Illumina HumanOmni1-Quad_v1-0_B BeadChip (Illumina, San Diego, CA; Methodology in the online-only Data Supplement).
Sample size calculation was performed using Ene 2.0 software (http://gsk.com/). One hundred ten subjects were needed to detect SNP frequencies >0.30 in the experimental group (patients with recurrence) and <0.15 in the reference group (patients without recurrence) to achieve statistical power of 80% with P=0.05.
Statistical analysis was performed using SPSS software, v.15 (IBM, Chicago, Illinois). Statistical significance for each SNP in the discovery cohort A was assessed by Kaplan–Meier curves using log-rank test. Multivariable analysis was performed using Cox regression in cohort A. Statistical significance for each SNP in the replication cohorts B, C, and D was assessed by χ2 analysis or Fisher exact test.
Continuous variables were compared by ANOVA, Mann–Whitney, or Kruskal–Wallis tests. We generated a predictive score based on Cox regression β-coefficients using the data from cohort A, using a forward stepwise procedure, with a Pvalue of 0.05 as the threshold for entry.
Receiver operating characteristic curves were plotted, and predictive capacity was calculated by measuring the area under the curve (AUC). The different AUCs were compared using z test17 from MedCalc version 188.8.131.52 (Mariakerke, Belgium).
The decision tree algorithm was generated with SPSS classification tree test (SPSS software; IBM). The association of the nodes with ischemic stroke recurrence was tested with the χ2 test using SPSS (SPSS software; IBM).
The clinical variables included in the GRECOS score (Genotyping Reurrence Risk of Stroke) were those statistically (P<0.05) associated with stroke recurrence after Cox regression, and the SNPs included in the GRECOS score were those statistically associated with stroke recurrence after replication analysis (P<0.05).
In discovery cohort A, new ischemic strokes were observed in 6.2% of patients during the first year of follow-up and global vascular recurrence was observed in 10.2%. Clinical and demographic data are available based on their availability in at least 80% of the cohort (Table II in the online-only Data Supplement). Eleven variables were nominally associated with recurrence of new ischemic stroke events (P<0.1) and were included in the Cox regression analysis (Table).
Ischemic Stroke Recurrence Association Studies
In the discovery stage (cohort A, n=1494), 17 SNPs were associated with the occurrence of a new ischemic stroke under a dominant–recessive model (P<0.05). However, none of these SNPs remained significant after replication stages (Table III in the online-only Data Supplement).
Under an additive model, 60 SNPs were associated (P<0.05) with the occurrence of a new ischemic stroke in the discovery stage (cohort A; Table IV in the online-only Data Supplement). After replication stages (in cohort B, n=844 and cohort C, n=461; 476 cases and 829 controls), only rs1800801 in MGP gene was associated with the risk of new ischemic stroke after a genetic additive model: discovery cohort: P=9×10−03; replication cohort (cohorts B and C): 1.48 odds ratio (95% confidence interval, 1.04–1.81); P=0.025. Replication analysis within VISP cohort (cohort D; n=1683) revealed that rs1800801 was not associated with stroke recurrence in a non-Spanish cohort (P>0.05). However, the Kaplan–Meier curves performed with prospective cohort A showed a clear cumulative risk of recurrent ischemic strokes during the time of follow-up (Figure 1). The rs1800801 SNP was only imputed in the cohort D using MACH version 1.0.16.
Vascular Recurrence Association Study
Under a dominant–recessive model and under the additive model, 13 and 53 SNPs, respectively, were associated with vascular recurrence. None of them remained significant after replication stages (Table III in the online-only Data Supplement).
Ischemic Stroke Recurrence Clinical Predictors
Univariate analysis in cohort A identified 11 clinical variables (Table) nominally associated with the recurrence of new ischemic stroke events (P<0.1) from the 50 analyzed (Table II in the online-only Data Supplement). We included these 11 variables in a Cox regression analysis, along with rs1800801, to determine whether rs1800801 was independently associated with ischemic stroke recurrence and to generate the clinical genetics score. We selected the median age (71 years) as cut off for the analysis.
After Cox regression, only TIA at enrollment, median age >71 years, and previous acute myocardial infarction or angina before enrollment were associated with ischemic stroke recurrence during the first year after the first ischemic stroke or TIA. When we introduced the rs1800801 SNP in the Cox regression, this SNP was a predictor of ischemic stroke recurrence (rs1800801; hazard ratio, 2.26 [1.462–3.488]; P=1.27×10−4).
Ischemic Stroke Recurrence Score Generation
A pilot predictive score from the Cox regression model was generated to determine the usefulness of the combination of clinical and genetic variables to predict ischemic stroke recurrence. We included age, TIA at enrollment, previous acute myocardial infarction, or angina and rs1800801 (G allele) in the following formula based on the independent variables obtained by Cox regression: Score=(1.85×age>71 years)+(2×inclusion TIA)+(3.6×AMI/angina)+(2.26×rs1800801 (G allele)). The coefficient for each variable was the hazard ratio value from the COX regression. This combined genetic and clinical model (the GRECOS score) predicts ischemic stroke recurrence (P=3.2×10−09) with an AUC of 0.684 (0.569–0.675).
The GRECOS score had higher predictive capacity than the previously reported SPI-II score or ESRS score, although the difference with ESRS was not significant (AUC-GRECOS score: 0.684 versus AUC-SPI-II: 0.601, P=0.03; and AUC-GRECOS score: 0.684 versus AUC-ESRS: 0.622, P=0.11). The 3 clinical variables included in the GRECOS score alone were slightly but not significantly different from the classical scores (AUC-clinical variables alone: 0.622 versus AUC-SPI-II: 0.601, P=0.9); AUC-clinical variables alone:0.622 versus AUC-ESRS: 0.622 P=0.9). As shown by AUC values, neither genetic data nor clinical data alone significantly increased the predictive capacity compared with the SPI-II or ESRS scores, only the combination of clinical and genetic variables improves the capacity of prediction.
We included the 3 clinical variables and the rs1800801 SNP in the classification tree algorithm implemented in the SPSS software (Figure 2). We observed that 12.6% of patients harboring allele G of rs180081 (risk allele) and with previous angina or acute myocardial infarction (Node 7) suffered a new ischemic stroke during follow-up (Figure 2). Interestingly, only 1.9% of patients without previous angina/acute myocardial infarction, without TIA at enrollment, and harboring allele A of rs180081 (Node 8) suffered a new ischemic stroke.
The association of all nodes with ischemic stroke recurrence was significant (P value=5.58×10−5). In addition, when we analyzed the node with the highest risk (node 7) versus the others, the results were significant (P value=0.023), as were the results when we compared the node with the lowest risk (node 8) versus the others (P value=2.8×10−4).
The clinical-genetic score generated with 3 clinical variables (age >71 years, TIA at the enrollment and previous acute myocardial infarction/angina) and rs1800801 (GRECOS score) improved AUC results versus previously reported clinical scores SPI-II or ESRS, with a statistically significant improvement in relation to the SPI-II score. However, when we used only the clinical variables, the GRECOS score did not show statistically significant differences versus the SPI-II score, indicating that the inclusion of genetic risk factors in clinical scores can improve the prediction of stroke recurrence. The scores generated to predict vascular recurrence in stroke patients have low predictive values, questioning the usefulness of these risk scores in daily clinical practice,18 afterward the increase of the prediction using genetic biomarkers could be important for a clinical implementation.
Furthermore, the GRECOS classification tree indicates a major patient risk group (Node 7) harboring rs180081 allele G (risk allele) and with previous angina/acute myocardial infarction that showed a 12% risk of ischemic stroke recurrence, which is double the mean of all ischemic stroke patients (6%). In contrast, we observed a minor patient risk group with a recurrence risk of only 1.9%, one third of the risk for all the patients. If these results are replicated in an independent study, it could be useful from a clinical point of view. A neurologist could modify the treatment of this group of high-risk patients, for example, by increasing the antiaggregation doses or using double antiaggregation (aspirin and clopidogrel).
An international replication did not find any association between rs1800801 and new recurrent ischemic strokes, suggesting that this SNP may be associated with stroke recurrence only in the Spanish population. However, the VISP project (cohort D) was a clinical trial in which the patients were treated with vitamin interventions, introducing a potential confounding factor that could influence the genetic results. Other differences among VISP and the Spanish cohorts are the higher frequency of genetically influenced clinical risk factors, including hypertension and diabetes mellitus (Table V in the online-only Data Supplement), compared with the GRECOS cohort. Moreover, VISP enrolled participants between 1997 and 2001, before the majority of individuals enrolled in the Spanish populations and during a time of transition in secondary stroke prevention. Another problem might be the sample size needed for the replication and the fact that in the VISP project patients with TIA or cardioembolism were not enrolled. Additional studies will be needed to determine whether the association of rs1800801 SNP with ischemic stroke recurrence is observed in other populations.
The rs1800801 variant is located in the promoter region of the MGP gene. This gene encodes a matrix Gla protein, which is an inhibitor of bone formation. The MGP -7A>G (rs1800801) polymorphism has been associated with higher risk of coronary artery calcification.19,20 The functional relevance of this SNP is controversial: it is located at a possible transcription start site,19 but its effects on matrix Gla protein expression and serum levels are unclear.19–21 Several authors hypothesized that MGP -7A>G is not a functional variant and that the amino acid change Thr83Ala, in linkage disequilibrium with MGP -7A>G (D′=0,97), might be responsible for matrix Gla protein changes.19 The Thr83Ala substitution changes protein polarity and could affect its capacity to bind to calcium, leading to calcium deposition in the arterial wall19; however, the effect of rs1800801 has not yet been established.
Matrix Gla protein may play a protective role in atherosclerosis progression.21 It has been found in plaque deposits, and its expression is increased in atherosclerotic plaques. Moreover, a knockout mouse model showed intense arterial calcification and death within the first 2 months.22
MGP is a mineralization-associated gene, as it is OPN, which encodes osteopontin, a protein previously reported to be associated with poor prognosis after stroke by our group.23 Both MGP and OPN genes are upregulated through the ERK1/2 (extra cellular signal–regulated kinases 1/2) pathway24 and are expressed at calcification sites within atherosclerotic lesions.25
Previous studies have found SNPs associated with vascular recurrence in stroke.10–13 Two studies performed with Asian population10,11 observed associations of ANRIL and NINJ2 polymorphisms with vascular recurrence. SNP rs10757278 of the ANRIL gene was not associated with stroke recurrence in our cohort. However, we did not check NINJ2 SNPs because previously unpublished results did not find any association between this locus and ischemic stroke in Spanish population. These 2 studies analyzed <5 SNPs and were performed in Asian populations; thus, it is possible that these polymorphisms are only genetic risk factors for Asian populations. Two other studies have been performed in white population.12,13 One of these studies recently found an association of 3 CRP polymorphisms with recurrent stroke. In our study, we did not analyze these 3 polymorphisms because of the selection of the genes, and polymorphism for our project was performed previously to the publication of the mentioned study.12
Only in 1 article,13 the potential role of genetics to predict recurrent strokes has been calculated. In this article, the authors did not find any improvement in the use of the genetics to predict the occurrence of new recurrent ischemic strokes compared with the use of clinical factors. However, the authors only analyzed SNPs associated with nonrecurrent ischemic stroke from previous studies. These results are in concordance with our results, indicating that genetic risk factors associated with recurrent strokes may be different than genetic risk factors associated with nonrecurrent strokes. In contrast, in our study with different SNPs selection criteria, we have observed a prediction improvement when the genetic risk factors are included in a score. The differences observed in the 2 studies can be explained regarding the strategy followed by the 2 groups. In our study, we have found the SNPs associated with stroke recurrence, as a first stage, and second, we have performed the score with the significant genetics variables; however, the study of Achterberg et al13 performed the genetics score with SNPs associated previously with ischemic stroke without checking the association with recurrent strokes. Based on our results, we think that genetics could be useful for stroke recurrence prediction when more studies will be performed in the field, including genome-wide association studies.
Replication cohorts were not recruited prospectively as it has been performed with the discovery cohort. Prevention strategies in stroke have changed during the last 10 years but these data are difficult to quantify, but we make the assumption that cases with recurrences and cases without recurrences recruited in the same time period received comparable treatment and prevention strategies. Subanalyses depending on the subtype of ischemic stroke were not performed because of the limited sample size. It would be interesting to test the GRECOS score in the other replication cohorts. However, one of the main clinical variables associated with stroke recurrence is previous myocardial infarction (before first stroke), and these data were not included in the replication cohorts. The matrix Gla protein polymorphism was not significant after Bonferroni correction. However, we think that Bonferroni penalizes gene candidate studies too harshly, and the SNP was replicated in independent cohorts. Bonferroni is a useful multivariable test correction in several situations; however, the Bonferroni method is conservative and has low power.26,27 It has been suggested that Bonferroni is a useful method in situations in which several subgroups of analysis have been tested, such as multistratification of the cohort (selection of women only, younger subjects only, etc) or in omics studies with no selection of genes, proteins, or clinical factors with a biologically plausible association with the end point of the study.26 We think replication is the most important phase of the study to validate the results obtained in gene candidate analysis.28
We did not count the number of SNPs discarded during the first search of SNPs. The data could be valuable for future studies, but unfortunately we do not have this information. Economic studies could estimate the usefulness of including this SNP among the clinical tests.
The generation of a score using clinical and genetic variables to predict ischemic stroke recurrence has revealed its potential to be useful in the future as a diagnostic tool during the follow-up of ischemic stroke patients.
Accurate estimates of vascular recurrence risk would allow clinicians to individualize secondary preventive treatment and to perform an intense follow-up of the major risk patients. This study is a potential first step in the development of personalized medicine tools for vascular recurrence after a first ischemic stroke.
Sources of Funding
This study was funded by Marato TV3, by the Spanish stroke research network (INVICTUS-PLUS) and by Generacion project (PI15/01978) Instituto de Salud Carlos III. Dr Fernández-Cadenas is supported by the Miguel Servet program (CP12/03298), Instituto de Salud Carlos III. Several groups participate in the International Stroke Genetics Consortium and the Spanish Stroke Genetics Consortium. The Hospital del Mar is supported by Spain’s Ministry of Health (III FEDER, RD12/0042/0020). Study recruitment and collection of data sets for the VISP trial were supported by a grant (R01 NS34447; PI James Toole) from the National Institute of Neurological Disorders and Stroke (NINDS). Genome-wide association studies genotyping (U01 HG004438l; PI David Valle), funded by the National Human Genome Research Institute and the Genomics and Randomized Trials (GARNET) Network (U01HG00516-03; co-PI Michèle M. Sale and Bradford B. Worrall), and genetic data cleaning was provided by the GARNET Coordinating Center (U01HG005157; PI Bruce S. Weir).
Guest Editor for this article was Michael Brainin, MD, Dr (Hon).
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.014322/-/DC1.
- Received June 20, 2016.
- Revision received January 26, 2017.
- Accepted February 7, 2017.
- © 2017 American Heart Association, Inc.
- Di Carlo A,
- Launer LJ,
- Breteler MM,
- Fratiglioni L,
- Lobo A,
- Martinez-Lage J,
- et al
- Rosamond W,
- Flegal K,
- Furie K,
- Go A,
- Greenlund K,
- Haase N,
- et al
- Di Carlo A
- Burn J,
- Dennis M,
- Bamford J,
- Sandercock P,
- Wade D,
- Warlow C
- Hankey GJ,
- Jamrozik K,
- Broadhurst RJ,
- Forbes S,
- Burvill PW,
- Anderson CS,
- et al
- Weimar C,
- Goertler M,
- Röther J,
- Ringelstein EB,
- Darius H,
- Nabavi DG,
- et al
- Kernan WN,
- Viscoli CM,
- Brass LM,
- Makuch RW,
- Sarrel PM,
- Roberts RS,
- et al
- Thompson DD,
- Murray GD,
- Dennis M,
- Sudlow CL,
- Whiteley WN
- Hsieh YC,
- Seshadri S,
- Chung WT,
- Hsieh FI,
- Hsu YH,
- Lin HJ,
- et al
- Zhang W,
- Chen Y,
- Liu P,
- Chen J,
- Song L,
- Tang Y,
- et al
- Williams SR,
- Hsu FC,
- Keene KL,
- Chen WM,
- Nelson S,
- Southerland AM,
- et al
- Achterberg S,
- Kappelle LJ,
- de Bakker PI,
- Traylor M,
- Algra A
- Lemmens R,
- Smet S,
- Thijs VN
- Herrmann SM,
- Whatling C,
- Brand E,
- Nicaud V,
- Gariepy J,
- Simon A,
- et al
- Farzaneh-Far A,
- Davies JD,
- Braam LA,
- Spronk HM,
- Proudfoot D,
- Chan SW,
- et al
- Perneger TV
- Lin YT,
- Lee WC
- Dichgans M,
- Markus HS