Cullell et al. Clinical Epigenetics (2022) 14:124 https://doi.org/10.1186/s13148-022-01340-5 RESEARCH Open Access Altered methylation pattern in EXOC4 is associated with stroke outcome: an epigenome‑wide association study Natalia Cullell1,2,3†, Carolina Soriano‑Tárraga4,5,6†, Cristina Gallego‑Fábrega1, Jara Cárcel‑Márquez1, Elena Muiño1, Laia Llucià‑Carol1, Miquel Lledós1, Manel Esteller7,8,9, Manuel Castro de Moura7, Joan Montaner10,11, Anna Rosell10, Pilar Delgado10, Joan Martí‑Fábregas12, Jerzy Krupinski2,13, Jaume Roquer4, Jordi Jiménez‑Conde4† and Israel Fernández‑Cadenas1*† Abstract Background and purpose: The neurological course after stroke is highly variable and is determined by demo‑ graphic, clinical and genetic factors. However, other heritable factors such as epigenetic DNA methylation could play a role in neurological changes after stroke. Methods: We performed a three-stage epigenome-wide association study to evaluate DNA methylation associ‑ ated with the difference between the National Institutes of Health Stroke Scale (NIHSS) at baseline and at discharge (ΔNIHSS) in ischaemic stroke patients. DNA methylation data in the Discovery (n = 643) and Replication (n = 62) Cohorts were interrogated with the 450 K and EPIC BeadChip. Nominal CpG sites from the Discovery (p value < ­10–06) were also evaluated in a meta-analysis of the Discovery and Replication cohorts, using a random-fixed effect model. Metabolic pathway enrichment was calculated with methylGSA. We integrated the methylation data with 1305 plasma protein expression levels measured by SOMAscan in 46 subjects and measured RNA expression with RT-PCR in a subgroup of 13 subjects. Specific cell-type methylation was assessed using EpiDISH. Results: The meta-analysis revealed an epigenome-wide significant association in EXOC4 (p value = 8.4 × ­10–08) and in MERTK (p value = 1.56 × ­10–07). Only the methylation in EXOC4 was also associated in the Discovery and in the Replication Cohorts (p value = 1.14 × ­10–06 and p value = 1.3 × ­10–02, respectively). EXOC4 methylation negatively correlated with the long-term outcome (coefficient = − 4.91) and showed a tendency towards a decrease in EXOC4 expression (rho = − 0.469, p value = 0.091). Pathway enrichment from the meta-analysis revealed significant associa‑ tions related to the endocytosis and deubiquitination processes. Seventy-nine plasma proteins were differentially expressed in association with EXOC4 methylation. Pathway analysis of these proteins showed an enrichment in natural killer (NK) cell activation. The cell-type methylation analysis in blood also revealed a differential methylation in NK cells. † Natalia Cullell, Carolina Soriano-Tárraga, Jordi Jiménez-Conde, Israel Fernández-Cadenas have contributed equally to this article *Correspondence: israelcadenas@yahoo.es 1 Stroke Pharmacogenomics and Genetics, IIB‑Sant Pau, Institut de Recerca de Sant Pau, Hospital Sant Pau, C/Sant Antoni Mª Claret,167, 08025 Barcelona, Spain Full list of author information is available at the end of the article © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. The Creative Commons Public Domain Dedication waiver (http://​creat​iveco​ mmons.​org/​publi​cdoma​in/​zero/1.​0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Cullell et al. Clinical Epigenetics (2022) 14:124 Page 2 of 17 Conclusions: DNA methylation of EXOC4 is associated with a worse neurological course after stroke. The results indi‑ cate a potential modulation of pathways involving endocytosis and NK cells regulation. Introduction genes (PATJ and LOC105372028) have been associated Stroke is a high incidence disease that represents the with long-term functional outcome (at 3 months) [14, first cause of death and disability in adults [1, 2]. More 15] and seven loci with the neurological course using than 70% of stroke survivors need help for their daily the difference between baseline NIHSS and NIHSS activity 5 years after an ischaemic stroke [3]. at 24 h [13]. However, not all the heritability associ- During the acute phase of a stroke, there are dynamic ated with the neurological course has been completely changes in the clinical symptoms that determine the discovered [13] and other heritable factors, such as evolution of the lesion and the associated deficits [4]. epigenetics, could be associated with the post-stroke To measure the outcome of a stroke, two different quan- neurological outcome. It has been demonstrated that tifiable measures are usually considered: the neurologi- epigenetics plays an important role in stroke risk and cal clinical symptoms and the functional independence stroke vascular recurrence [16–21]. Previous Epige- of patients. There is high variability in the neurological nome-Wide Association Studies (EWAS) have iden- and functional outcomes after stroke which is asso- tified 22 CpG sites and 21 loci with altered DNA ciated with several factors, including demographic, methylation associated with stroke risk [16]. Moreo- clinical and genetic factors [5–7]. Different scales are ver, biological age calculated with DNA methylation is widely used to quantify the neurological deficit and the associated with stroke outcome and mortality [20, 21]. functional outcome. The National Institutes of Health Our aim is to study the epigenetic risk factors and bio- Stroke Scale (NIHSS) considers 15 different measures logical mechanisms associated with post-stroke neuro- of neurological worsening: consciousness, eye move- logical course using the difference in baseline NIHSS and ment, vision, coordination, language, sensory function, NIHSS at discharge (ΔNIHSS) as outcome variable. upper and lower limb strength, facial muscle function, and neglect [8]. The modified Rankin scale (mRS) is Materials and methods used to quantify the functional outcome based on the Data availability capacity of patients to be independent in carrying out The DNA methylation data analysed in this study are daily activities [9]. Different variables have been found available in GEO. The Discovery data from BASIC- to be predictors of the early (24 h post-stroke) neuro- MAR are available under the GEO accession number logical outcome: baseline NIHSS, tPA treatment, age, “GSE69138”. The Discovery from the GRECOS together stroke subtype, glucose levels, and systolic blood pres- with the replication cohort could be identified in GEO sure [5–7]. The early neurological outcome has been under the code “GSE203399”. found to explain up to 30% of the long-term outcome (mRS at 3 months) [5]. The neurological evolution of Patient selection stroke patients during hospitalization has also been We included in the Discovery Cohort 738 Caucasian assessed as a good predictor of 30-day and long-term patients with EWAS data who had suffered a stroke mortality [10]. Reznik et al. compared the predictive and had had a blood sample taken during the first 24 h value of NIHSS measured at different time-points: following ischaemic stroke. The Discovery consisted baseline, 24 h, and discharge. They concluded that dis- in patients from the Mar Hospital, who enrolled 662 charge-NIHSS was the best predictor of the 3-month patients as part of the BASICMAR register [22–24], outcome [11]. Thus, the difference between the NIHSS and 76 from the GRECOS study [25]. BASICMAR is a at baseline and the NIHSS at discharge has recently prospective register of patients with ischaemic stroke gained importance as a valid outcome variable [12] as it recruited between 2009 and 2012 [22–24]. The GRECOS covers the entire period of hospitalization. (Genotyping RECurrence Risk of Stroke) study is a pro- Interestingly, genetics seems to play a role in the neu- ject that enrolled 1,494 Caucasian patients with a first rological course. Ibanez L et al. found that 8.7% of the ischaemic stroke and population-based controls between difference between NIHSS at baseline and NIHSS at July 2005 and May 2009 from 23 Spanish Hospitals [25]. 24 h was explained by common single nucleotide poly- The patients included from the BASICMAR and the morphisms (SNPs) [13]. However, only three different GRECOS study were included in previous EWAS [16, 18, Genome-Wide Association Studies (GWAS) have been 19]. From the 738 patients, 725 had registered the main performed in the stroke outcome field [13–15]. Two variable analysed in this study: the NIHSS at baseline and Cullell et al. Clinical Epigenetics (2022) 14:124 Page 3 of 17 the NIHSS at discharge and were included in the EWAS between genotypic and phenotypic sex (Additional file 1: analysis. Figure SI). After this processing, beta values represent- In the Replication Cohort, we included 62 Caucasian ing methylation of each CpG site were normalized using stroke patients from the EPIGENESIS study. The EPI- the Noob function from the minfi package [29]. We also GENESIS study selected ischaemic stroke patients [26] assessed the batch effect by a single value decomposi- with a blood sample collected during the first 6 h follow- tion (SVD) analysis in ChAMP and by performing a mul- ing onset of symptoms to study epigenetics associated tidimensional scaling (MDS) plot where it is shown the with stroke outcome. distance matrix of each sample depending on the batch Differences in demographic and clinical variables (Additional file 1: Figure SII). The proportion of the dif- between the Discovery and Replication cohorts were ferent blood cell types was estimated for each patient, calculated. Differences between groups were tested with and beta values were corrected based on these data using Kruskal–Wallis rank-sum test for nonparametric quanti- the “champ.refbase” function of the ChAMP package tative variables, while differences for qualitative variables [30]. This function has implemented the RefbaseEWAS was tested with chi-square test. Statistically significance method, which uses a methylation reference database for was defined with p value < 0.05. each of the major cell types present in blood. All the projects included in this study have been approved by ethics committees and all the patients have Statistical analysis signed informed consent forms. The study was conducted We calculated the difference between baseline NIHSS in accordance with Declaration of Helsinki and European and NIHSS at discharge (ΔNIHSS). First, we evaluated guidelines: requirements of the Spanish Law 3/18 on the which demographic, cardiovascular and other stroke- protection of personal data and the new European Union related variables were associated with ΔNIHSS. We also legislation on personal data, specifically Regulation (EU) assessed whether ΔNIHSS and other variables were 2016/679 of the European Parliament and of the Council associated with the long-term outcome (measured with of April 27, 2016 Data Protection (GDPR). the Rankin scale at 3 months, mRS) in our cohort using bivariate analysis. Then, we included the significant vari- DNA extraction and bisulphite conversion ables (p value < 0.05) in a backward stepwise regression. Whole blood was obtained in EDTA tubes. DNA from The DNA methylation was considered the depend- the GRECOS and EPIGENESIS studies was extracted ent variable and the ΔNIHSS the independent variable. using a Gentra Puregene Blood Kit (Qiagen, Hilden, Ger- We calculated the differential methylation positions many) following the manufacturer’s instructions. DNA using multiple linear regression (lm). First, in the Dis- from BASICMAR was extracted manually using salt pre- covery analysis we considered as covariates the known cipitation in the National Bank of DNA (Carlos III Insti- variables to be associated with the dependent variable tute (ISCIII)). (DNA methylation): sex, age, self-reported smoking hab- EWAS analysis using the EZ DNA Methylation-Gold™ Bisulphite conversion of DNA was performed before its, and the first two principal components (PC) (basic EWAS adjustment). For the PC calculation, we used the Kit (Zymo Research, CA, USA). DNA methylation was function princomp in R from stats package which uses a studied with the 450 K BeadChip (Illumina) in all sam- spectral decomposition strategy to study the correlation ples from the Discovery Cohort and EPIC BeadChip between the methylation beta values. We selected the (Illumina) in all samples from the Replication Cohort. first two columns from the loading matrix (eigenvectors) from the output to adjust the results. To ensure that the Epigenome‑wide association study (EWAS) batch effect was corrected when adjusting by the first two Methylation raw data were processed using R (http://​ PC, we performed an additional analysis adjusting in the www.​cran.r-​proje​ct.​org) and Bioconductor packages lm by the basic EWAS adjustment + batch. We used the (http://​www.​bioco​nduct​or.​org). MethylToSNP package in R to evaluate whether any of We proceed with quality controls (QCs) using ChAMP the CpG sites with a p value < 0.05 had SNP patterns and package in R [27]. CpG quality controls consisted of was not removed during QCs that could cause false posi- the removal of CpG sites with a non-significant detec- tive results [31]. tion p value (p value > 0.05), CpG sites from sex chro- As a secondary analysis, we considered the ΔNIHSS as mosomes, CpG sites with affinity for multiple probes, a dichotomic variable, classifying patients into those with no “CG” probes, probes with bead count < 3 in at least improvement in the outcome (ΔNIHSS ≥ 4) or decline 5% of samples and SNP probes [28]. When performing in the outcome (ΔNIHSS < 4), based on previous studies QCs on samples, we removed samples with more than [32, 33]. The purpose was to assess whether nominal CpG 1% missing CpG sites and samples with discordance sites associated with the continuous ΔNIHSS variable Cullell et al. Clinical Epigenetics (2022) 14:124 Page 4 of 17 were also associated with neurological improvement or using METAL [36] based on a random effect model. decline. We selected this model after reviewing the assump- In the Discovery Analysis, all the significant, with a p tions of random and fixed effect models [37]. Based on value < 2.4 × ­10–07 (based on the threshold estimated by Nikolakopoulou A et al., the random effect model could Saffari A et al. [34, 35]), and nominal (p value < ­10–06) dif- be beneficious on allowing to differ on the true effects ferentially methylated positions (DMP) were analysed in and accounting for unexplained heterogeneity [37]. The the Replication Cohort (Replication Analysis). The CpG association of the nominally CpG sites from the Dis- methylation sites with p value < 0.05 in the Replication covery was evaluated in the meta-analysis. Cohort were considered replicated. The methylation Finally, we considered all the CpG sites that fol- from the replicated CpG sites were analysed in bivariate lowed the next criteria as validated DMPs: 1) at least and backward stepwise regression analyses to identify a nominal association (p value < ­10–06) in the Discov- independent clinical and demographic factors condi- ery Cohort, 2) significant p value (p value < 0.05) and tioning the methylation pattern of that replicated sites. the same effect direction in the Replication Cohort, Finally, the independent variables were used as covariates and 3) an epigenome-wide significant p value (p in a new lm analysis including only the replicated CpG value < 2.4 × ­10–07) in the meta-analysis. sites in the Discovery Cohort (Fig. 1). Moreover, we combined the results from the Dis- covery and Replication Cohorts in a meta-analysis Fig. 1 Scheme of the three-stage EWAS. The three-stage EWAS consisted in a Discovery stage analysed with multiple linear regression with adjustment of variables known to be associated with methylation (age, sex, smoking habit, PC1 and 2), a Replication stage where the 44 nominally associated CpG sites were analysed with multiple linear regression with the same basic adjustment than in the Discovery. The replicated CpG site (cg00039070) was re-analysed in the Discovery including the specific adjustment (covariates associated with the methylation of cg00039070): PC1 and PC2. In the Discovery cohort, differentially methylation region (DMR) and block (DMB) analyses were also performed. Finally, a meta-analysis that merged the Discovery and Replication Cohorts was performed with basic EWAS adjustment. The significant CpG sites in the meta-analysis were evaluated in the pathway/features enrichment analysis, in the differentially methylated cell-type analysis and in the blood–brain correlation analysis Cullell et al. Clinical Epigenetics (2022) 14:124 Page 5 of 17 Feature enrichment and metabolic pathway enrichment used the standard recommendation of a maximum length analyses of 250,000 nucleotides for a block and to only include We explored for enrichment of specific features for sig- blocks with a minimum of 10 regions [27, 30]. nificant (n = 5) or nominal (n = 44) CpG sites from the discovery in comparison with all the CpG sites included Gene expression analysis in the analysis (n = 423,156). Features were classified We studied the correlation between relative EXOC4 into 1st exon, 3’ untranslated region (UTR), 5’UTR, mRNA levels and EXOC4 methylation. We analysed gene body, intergenic region (IGR), transcription start by Real-Time Quantitative Reverse Transcription PCR site (TSS) 1500 and TSS200. In this analysis, we classi- (qRT-PCR) the cDNA isolated from whole blood from fied CpG sites into hypermethylated or hypomethylated 13 healthy controls from the GRECOS study which also in their association with ∆NIHSS. We applied chi-square have available EWAS data. test to evaluate differences in the features between Whole blood in EDTA tubes was obtained, and DNA groups (significant/nominal CpG sites vs. all CpG sites). was extracted using a Gentra Puregene Blood Kit (Qia- We looked for enrichment of metabolic pathways gen, Hilden, Germany) following the manufacturer’s using as input all the CpG sites and the CpG sites that instructions. We used the 7900 Real-Time PCR system were nominally associated (p value < ­10–06) with ΔNIHSS (qRT-PCR) (Applied Biosystems, Foster city, CA, USA) in the meta-analysis. The analysis was done with Meth- ard TaqMan® PCR kit protocol as described previously to quantify EXOC4 expression. We followed a stand- ylGSA [38], an R package specifically designed for path- way analysis from EWAS results. We applied the three [15]. Briefly, the EXOC4 (Hs00253986_m1) probe was functions (methylglm, methylRRA and methylgometh) used and the results were normalized using endogenous from the package (all of them designed to adjust for the controls: Cyclophilin A (PPIA, Hs99999904_m1) and number of CpGs in each gene to reduce possible bias). In Glyceraldehyde-3-Phosphate Dehydrogenase (GAPDH, the main analysis, all the CpG sites from the EWAS meta- Hs99999905_m1). We ran triplicates for the 13 sam- analysis were included together with their p value. Using ples, and we included an external sample as calibrator. the first and second function, we also restricted the anal- Reactions were analysed with applied Biosystems SDS ysis to specific types: CpGs from promoters (TSS1500 7900 system software (Applied Biosystems, Foster city, or TSS200) or located in gene bodies. Using the second CA, USA). We calculated the fold change of the average and third function, we were able to perform a secondary expression (using relative quantification (RQ) values) analysis filtering out CpG sites based on their p value, from EXOC4 and endogenous controls. selecting the list of nominally associated CpG sites from the meta-analysis. More details for the different func- We used data from SOMAscan® Assay (SomaLogic) to Proteomic and pathway analysis tions and the options that we selected are specified in the Supplementary methods. We included in the analy- cg00039070 methylation. Briefly, the SOMAscan® Assay find differentially expressed proteins associated with ses the three available pathway databases in MethylGSA: Gene Ontology, KEGG and Reactome. We considered uses plasma samples to bring data for 1305 proteins using significant associations when the p value was < 0.05 and a short single-stranded DNA sequence (SOMAmer rea- the Q value (false discovery rate (FDR) adjusted p value) gents) that binds to target proteins and allows their quan- was < 0.05. tification [39]. controls for which proteomic (from SOMAscan®) and In this analysis, we included 26 stroke patients and 20 Differentially methylated regions (DMRs) and blocks (DMBs) DNA methylation data (from EWAS) were available. We evaluated whether differential methylation regions For the proteomic assay, blood from patients collected (DMRs) and blocks (DMBs) were associated with in EDTA tubes was centrifuged at 3.000 g for 10 min to ΔNIHSS using the Bumpunther algorithm implemented until they were analysed with SOMAscan®. The results obtain plasma. Plasma samples were frozen at − 80 °C in the ChAMP package [27, 30]. A DMR was defined in our analysis with the standard parameters: a segment were processed as described in [40]. The analysis was with a minimum of seven CpG sites with a maximum adjusted by case/control state. length of 300 nucleotides. DMRs should be separated by We investigated differentially expressed proteins a minimum of 1,000 base-pairs to be considered different in association with cg00039070 methylation. Stroke DMRs. For the DMB calculation, a block was considered patients were included from the EPIGENESIS (N = 7) a large cluster generated from open sea regions. Each and GRECOS (N = 19) cohorts. Controls were included region located in an open sea was collapsed into a unit, from the ISSYS (Investigating Silent Stroke in hYperten- calculating the mean methylation and mean position. We sives) cohort. It is an observational prospective study in Cullell et al. Clinical Epigenetics (2022) 14:124 Page 6 of 17 hypertensive participants to determine the prevalence A more detailed explanation for these two methods of silent or magnetic resonance imaging (MRI)-defined is included in the Supplemental Methods (Additional brain infarcts and cognitive impairment. This cohort file 1). comprises 1000 non-demented individuals, aged 50 to 70 years, and diagnosed with essential hypertension at Results least one year before inclusion in the ISSYS study [41]. Discovery stage All the proteins associated with cg00039070 methyla- A total of 643 patients and 423,156 CpG sites passed tion (p value < 0.05) were analysed using over-represen- QCs in the Discovery analysis (Additional file 1: Figure tation analysis (ORA) to find enrichment of metabolic SI). The median NIHSS at baseline and discharge was 8 pathways using the Wilcoxon Rank-Sum Test in the Gene and 3, respectively, and the median ΔNIHSS at discharge Ontology database (Biological Process) in WebGestalt was positive, indicating an improvement in the neu- [42, 43]. rological status of these patients (Table 1). We wanted to assess which variables were independently associ- ated with the ΔNIHSS in our cohort. With this purpose, Differentially methylated cell‑type (DMCT) analysis we studied the association of different demographic, The cellular component of the blood tissue was estimated cardiovascular and stroke-related variables, includ- and analysed to determine whether the differential meth- ing the long-term outcome (measured with the Rankin ylation was specific to one cell type using EpiDISH [44]. scale at 3 months) with ΔNIHSS. Then, we explored The CellDMC function was used to identify differentially which variables were independently associated with the methylated cell types associated with the ΔNIHSS based mRS, to be sure that the main variable in our analysis on the proportion of B cells, CD4 + and CD8 + T cells, (ΔNIHSS) was a good predictor of long-term outcome NK, neutrophils and monocytes. We studied the nomi- in our cohort. We performed bivariate and backward nally associated CpG sites in the meta-analysis to find stepwise regression analyses. NIHSS at baseline (p enrichment of cell-type differential methylation. value < 2.2 × ­10–16), mRS (p value < 2.2 × ­10–16), treatment with rTPA (p value = 2.5 × ­10–02) and atrial fibrillation (p value = 1.75 × ­10–02) were independently associated with Tissue‑specific signal detection ΔNIHSS in a multivariate analysis (Additional file 1: Table We used eFORGE (experimentally derived Functional SI), whereas ΔNIHSS (p value < 2.2 × ­10–16), baseline element Overlap analysis of ReGions from EWAS) [45] NIHSS (p value < 2.2 × ­10–16), age (p value = 4.72 × ­10–10), to estimate tissue-specific signals from the significant sex (p value = 1.3 × ­10–02) and smoking habit (p and nominal results from the EWAS meta-analysis. We value = 4.4 × ­10–02) were found to be associated with included analyses with all the different functional ele- the 3 months of mRS in the stepwise regression analysis ments. A p value < 0.05 was considered a nominal asso- (Additional file 1: Table SII). ΔNIHSS at discharge and ciation and a Q value < 0.05 according to FDR adjustment mRS at 3 months were negatively correlated, indicating was considered statistically significant. A more com- that a worsening in the neurological course measured plete description of this tool is provided in Supplemental with ΔNIHSS was associated with a worse long-term Methods (Additional file 1). outcome measured with mRS. ΔNIHSS was used as the independent variable to cal- culate differential methylation positions using multiple Blood–brain epigenetic correlation linear regression (lm) (Fig. 1). In order to compare the methylation from the significant In the Discovery EWAS with basic adjustment (sex, findings in the meta-analysis between blood and brain, age, smoking habits and the first two PCs), we identified we used the Blood Brain DNA Methylation Comparison a total of five epigenome-wide (p value < 2.4 × ­10–07) CpG Tool [46]. We analysed the correlation of cg00039070 sites and 44 nominally CpG sites (p value < ­10–06) associ- methylation and the four brain regions included in the ated with the ΔNIHSS (Fig. 2, Table 2). When the batch tool. variable (defined by the two cohorts included in the dis- Using Blood–Brain Epigenetic Concordance (BECon) covery) was used as covariate, the results did not change [47], we also investigated the concordance in the remarkably, and all the 44 CpG sites remained significant cg00039070 methylation between three brain regions with a p value < 0.05 (Additional file 1: Table SIII). Thus, (Brodmann area (BA) 10, BA20 and BA7) and blood the observed batch effect (Additional file 1: Figure SII) using the three metrics available in this tool. was corrected when adjusting by PCs. Using Methyl- ToSNP, none of the evaluated CpG sites was predicted to have SNP patterns. Cullell et al. Clinical Epigenetics (2022) 14:124 Page 7 of 17 Table 1 Demographic and clinical data from the Discovery and Replication Cohorts Discovery Replication P value GRECOS BASICMAR Subjects(n) 59 584 62 Sex 0.429 Male, n (%) 49 (17%) 322 (55.2%) 32 (51.6%) Female, n (%) 10 (83%) 262 (44.8%) 30 (48.4%) Age in years, median (IQR) 71 (15) 77 (14) 77 (9.5) 0.22 NIHSS at baseline, median (IQR) 3 (6.5) 5 (8) 15 (11.5) < 2.2 × ­10–16* NIHSS at discharge, median (IQR) 1 (2) 3 (6) 4 (12) 2.62 × ­10–06* ΔNIHSS, median (IQR) 1 (4) 1 (4) 5 (11) 6.6 × ­10–04* mRS 90 days, median (IQR) 1 (3) 2 (3) 1 (3) 5.4 × ­10–04* Presence of HTN, n (%) 37 (62.7%) 236 (40.1%) 40 (64.5%) 1.04 × ­10–03* Smoking, n (%) 13 (22%) 168 (28.8%) 6 (9.67%) 3 × ­10–07* Presence of AF, n (%) 3 (5%) 201 (34.4%) 25 (40.3%) 2.86 × ­10–02* Presence of DM, n (%) 17 (28.8%) 421 (72.1%) 8 (12.9%) 9.5 × ­10–11* Treatment with rtPA, n (%) 9 (15.3%) 91 (15.6%) 53 (85.5%) < 2.2 × ­10–16* TOAST CES, n (%) 0 (0%) 231 (53.9%) 29 (46.7%) < 2.2 × ­10–16* LAS, n (%) 26 (44.1%) 153 (39.6%) 17 (27.4%) SVS, n (%) 11 (18.6%) 194 (33.2%) 0 (0%) Und, n (%) 21 (35.6%) 4 (0.7%) 13 (2.1%) Other, n (%) 1 (1.7%) 2 (0.34%) 0 (0%) NA 0 (0%) 0 (0%) 3 (4.8%) Main clinical characteristics of patients included in the analysis from the Discovery and Replication Cohorts. P value column indicates differences for each variable between Discovery (both GRECOS and BASICMAR together) and Replication Cohorts IQR interquartile range; ΔNIHSS = NIHSS at baseline – NIHSS at discharge; HTN: hypertension; AF atrial fibrillation; DM: diabetes mellitus; CES cardioembolic stroke; LAS large artery stroke; SVS small vessel stroke; und undetermined stroke *Statistically significant p values for differences between cases and controls. NA: not available We found a differential feature enrichment (p the 38 CpG sites that could be evaluated in the Replica- value = 9.23 × ­10–03) when comparing nominal CpG sites tion Cohort, one site was significant (p value < 0.05): (n = 44) with all the CpG sites included in the analysis cg00039070 (p value = 1.14 × ­10–06, coefficient = − 4.91 (n = 423,156) (Additional file 1: Table SIV). Specifically, in the Discovery and p value = 1.14 × ­10–02, coeffi- we found that nominal CpG sites that were hypermethyl- cient = − 2.58 in the Replication Cohort) (Table 2). This ated in association with ΔNIHSS tend to be located in the CpG site was located in the body of the EXOC4 gene, a body of genes rather than in IGR. Hypomethylated nomi- member of the exocyst complex. The effect was the same nally CpG sites were found with higher probability in the in both cohorts: higher methylation was associated with TSS200 in comparison with all the CpG sites analysed. stroke worsening measured with ΔNIHSS (Fig. 3A). The differential feature enrichment could suppose differ- We analysed the association of the cg00039070 meth- ent effects in gene expression. ylation with demographic and cardiovascular risk vari- The DMR analysis revealed 50 regions associated (p ables in the Discovery Cohort. We found sex, batch, value < 0.05) (Additional file 1: Table SV) with ΔNIHSS, PC1, PC2, baseline NIHSS and diabetes mellitus associ- and the DMB analysis showed a total of 323 blocks asso- ated with cg00039070 methylation in bivariate analyses. ciated with ΔNIHSS (Additional file 1: Table SVI). After backward stepwise regression, only PC1 and PC2 remained significant. We re-analysed by lm the associa- Replication stage tion of ΔNIHSS with cg00039070 methylation using only All the CpG sites with a nominal p value (< ­10–06) in the PC1 and PC2 as covariates. Methylation of cg00039070 Discovery Analysis (n = 44) were evaluated in the Repli- in EXOC4 maintained its association with ΔNIHSS with cation Cohort. The median ΔNIHSS also had a positive a p value of 2.27 × ­10–06 and a coefficient of -4.77. value in this cohort (Table 1). Six CpG sites could not be evaluated in the Replica- tion Cohort because they fell to pass QCs (Table 2). From Cullell et al. Clinical Epigenetics (2022) 14:124 Page 8 of 17 Fig. 2 Manhattan plots for the EWAS analyses. Manhattan plot in the Discovery Analysis (A), Replication Analysis (B) and meta-analysis (C). The green dots represent CpG sites nominally associated with ΔNIHSS in the Discovery Cohort. The red and blue lines represent the epigenome-wide and nominal significance threshold, respectively. CpG sites are annotated in the plot if they are nominally associated with NIHSS in the Discovery Cohort and are statistically significant in the corresponding analysis EWAS meta‑analysis (Table 2). From these, all but one presented consistent The meta-analysis combining effect sizes from the Dis- effect direction in both Discovery and Replication analy- covery and Replication Cohorts revealed two CpG sites ses (Table 2). with epigenome-wide significant association, includ- The meta-analysis considering the ΔNIHSS as a dicho- ing the CpG site in EXOC4 (p value = 4.44 × ­10–08, tomic variable (ΔNIHSS ≥ 4 vs ΔNIHSS < 4) showed coefficient = − 5.47), and 12 CpG sites with nominal all but two CpG sites from the 38 analysed in the meta- association. analysis to be significantly associated with ΔNIHSS (p value < 0.05). The CpG site in EXOC4 was also associ- ated with the dichotomic ΔNIHSS with the same effect Table 2 EWAS summary statistics in the Discovery, Replication and meta-analyses Discovery Replication Meta-analysis CpG CHR BP Gene Feature COEFF P COEFF P COEFF P Direction (UCI,LCI) (UCI,LCI) (UCI,LCI) cg00039070 7 133,482,303 EXOC4 Body − 4.91(− 6.86,− 2.96) 1.14E-06 − 2.58(− 4.62,− 0.54) 1.30E-02 − 5.474(− 5.4738,− 5.4742) 4.44E-08 −− cg05979619 2 111,947,000 MERTK Body 9.72E-10 2.15E-01 1.54E-07 Cullell et al. Clinical Epigenetics − 6.21(− 8.16,− 4.26) − 1.26(− 3.24,0.72) − 5.250(− 5.2495,− 5.2501) −− cg10977834 4 77,075,687 CCNI 5’UTR​ 4.85(6.80,2.90) 1.53E-06 1.56(3.55,− 0.43) 1.25E-01 4.875(4.8749,4.8747) 1.11E-06 +  + cg08119231 1 220,747,874 MOSC2 TSS1500 5.01(6.96,3.06) 7.07E-07 0.87(2.83,− 1.09) 3.91E-01 4.871(4.8717,4.8704) 1.11E-06 +  + cg21574204 6 28,081,180 ZNF165 5’UTR​ 4.72(6.67,2.77) 2.88E-06 1.27(3.24,− 0.70) 2.09E-01 4.847(4.8470,4.8470) 1.21E-06 +  + cg01149259 3 46,979,906 NBEAL2 1stExon 4.68(6.63,2.73) 3.48E-06 1.42(3.41,− 0.57) 1.62E-01 4.768(4.7684,4.7682) 1.86E-06 +  + cg18831371 5 176,610,324 GPRIN1 TSS200 5.35(7.30,3.40) 1.25E-07 0.32(2.17,− 1.53) 7.48E-01 4.752(4.7527,4.7522) 2.02E-06 (2022) 14:124 +  + cg26050512 20 58,981,011 TH1L TSS1500 4.83(6.78,2.88) 1.71E-06 − 0.09(− 1.93,1.75) 9.30E-01 4.563(4.5626,4.5624) 4.83E-06 −+ cg21404878 15 72,286,264 BRUNOL6 3’UTR​ 4.94(6.43,2.51) 1.03E-06 0.02(2.19,− 2.15) 6.69E-01 4.540(4.5407,4.5393) 5.61E-06 +  + cg09548897 16 291,277 AXIN1 Body − 4.69(− 6.64,− 2.74) 3.37E-06 − 0.73(− 2.67,1.21) 4.69E-01 − 4.499(− 4.4986,− 4.4989) 6.80E-06 −− cg14659930 3 114,409,454 ZBTB20 5’UTR​ 4.59(6.48,2.58) 5.24E-06 0.19(2.08,− 1.70) 3.78E-01 4.453(4.4544,4.4511) 8.48E-06 +  + cg00939347 10 606,320 DIP2C Body − 5.32(− 7.27,-3.37) 1.45E-07 − 0.5(− 2.42,1.42) 6.22E-01 − 4.450(− 4.4501,− 4.4506) 8.50E-06 −− cg07925823 16 68,264,601 SLC7A6 5’UTR​ 4.76(6.71,2.81) 2.43E-06 1.4(3.39,− 0.59) 1.68E-01 4.443(4.4430,4.4428) 8.81E-06 +  + cg20383948 21 45,478,223 COL18A1 Body 5.17(7.12,3.22) 3.15E-07 0.97(2.93,− 0.99) 3.38E-01 4.431(4.4307,4.4305) 9.28E-06 +  + cg09741713 12 62,078,134 FAM19A2 5’UTR​ 4.66(6.63,2.73) 3.82E-06 − 1.12(− 3.08,0.84) 5.20E-01 4.340(4.3408,4.3393) 1.42E-05 +  + cg12103149 6 30,213,396 TRIM26 TSS200 4.56(6.58,2.68) 6.06E-06 0.08(1.98,− 1.82) 3.79E-01 4.331(4.3312,4.3311) 1.45E-05 +  + cg10156941 17 1,563,002 PITPNA TSS200 4.67(6.62,2.72) 3.63E-06 0.12(1.90,− 1.66) 9.03E-01 4.349(4.3489,4.3488) 1.46E-05 +  + cg15765398 21 44,990,079 IGR − 5.27(6.51,2.61) 1.92E-07 0.89(2.85,− 1.07) 1.60E-01 − 4.328(− 4.3267,− 4.3288) 1.51E-05 + - cg12349416 11 4,184,868 IGR 4.58(6.89,2.99) 5.73E-06 0.43(2.32,− 1.46) 4.84E-01 4.321(4.3219,4.3207) 1.55E-05 +  + cg14482313 12 52,233,105 KRT7 TSS200 4.60(6.55,2.65) 5.19E-06 0.62(2.55,− 1.31) 5.40E-01 4.304(4.3047,4.3037) 1.68E-05 +  + cg02996131 6 152,637,463 SYNE1 TSS1500 4.60(6.61,2.71) 5.08E-06 0.65(2.58,− 1.28) 8.63E-01 4.252(4.2532,4.2516) 2.11E-05 +  + cg19935850 4 41,751,422 IGR 4.99(6.94,3.04) 7.94E-07 − 0.14(− 1.91,1.63) 8.86E-01 4.182(4.1825,4.1808) 2.90E-05 −+ cg10363284 4 8,005,411 ABLIM2 Body − 5.54(-7.49,− 3.59) 4.55E-08 0.29(2.18,− 1.60) 7.76E-01 − 4.179(− 4.1782,− 4.1790) 2.94E-05 +− cg18707780 15 99,733,585 LYSMD4 TSS200 4.46(6.41,2.51) 9.75E-06 − 0.41(− 2.31,1.49) 6.85E-01 4.154(4.1539,4.1538) 3.42E-05 −+ cg20648899 6 93,416,423 EPHA7 Body 4.53(6.60,2.70) 7.17E-06 − 0.2(− 2.03,1.63) 8.54E-01 4.007(4.0083,4.0057) 6.14E-05 +  + cg07987148 20 46,690,251 TP53RK TSS1500 4.54(6.56,2.66) 6.79E-06 − 0.41(− 2.28,1.46) 9.89E-01 3.980(3.9813,3.9790) 6.88E-05 - + cg07475390 1 14,114,109 IGR 4.68(6.53,2.63) 3.52E-06 0.7(2.62,− 1.22) 2.67E-01 3.959(3.9593,3.9580) 7.55E-05 - + cg00347584 11 47,261,474 NR1H3 Body 4.47(6.47,2.57) 9.50E-06 − 1.26(− 3.24,0.72) 9.87E-01 3.909(3.9098,3.9081) 9.27E-05 +  + cg13114315 21 36,174,169 DOPEY2 Body 4.48(6.43,2.53) 8.92E-06 0.49(2.41,− 1.43) 6.30E-01 3.843(3.8438,3.8418) 1.22E-04 +  + cg14414100 9 19,547,532 SLC24A2 Body 4.63(6.59,2.69) 4.55E-06 − 0.56(− 2.46,1.34) 9.40E-01 3.772(3.7733,3.7717) 1.62E-04 +  + cg18795809 4 10,456,907 ZNF518B 5’UTR​ 4.68(6.63,2.73) 3.58E-06 0.34(2.22,− 1.54) 7.36E-01 3.771(3.7708,3.7703) 1.64E-04 +  + cg03732020 11 47,261,417 NR1H3 Body 4.64(6.49,2.59) 4.35E-06 − 0.01(− 1.29,1.27) 5.76E-01 3.551(3.5523,3.5506) 3.83E-04 −+ cg24978805 7 3,687,557 SDK1 Body − 4.84(− 6.79,− 2.89) 1.67E-06 − 0.35(− 2.24,1.54) 7.30E-01 − 3.184(− 3.1840,− 3.1845) 1.45E-03 −− Page 9 of 17 Cullell et al. Clinical Epigenetics Table 2 (continued) Discovery Replication Meta-analysis CpG CHR BP Gene Feature COEFF P COEFF P COEFF P Direction (2022) 14:124 (UCI,LCI) (UCI,LCI) (UCI,LCI) cg08526825 16 2,752,228 SRRM2 TSS200 4.65(− 7.22,− 3.32) 4.13E-06 1.43(3.42,-0.56) 8.41E-01 3.051(3.0506,3.0504) 2.28E-03 −+ cg25354926 4 818,823 CPLX1 Body − 4.49(− 6.44,− 2.54) 8.52E-06 0.95(2.91,− 1.01) 3.47E-01 − 2.784(− 2.7840,− 2.7844) 5.37E-03 +− cg04886221 1 27,343,238 SYTL1 5’UTR​ 4.61(6.55,2.65) 4.97E-06 0.17(1.97,− 1.63) 6.80E-01 2.532(2.5320,2.5319) 1.14E-02 −+ cg22363670 7 86,643,853 GRM3 TSS200 4.49(6.44,2.54) 8.55E-06 − 0.3(− 2.16,1.56) 7.65E-01 2.409(2.4095,2.4092) 1.59E-02 - + cg06933752 11 65,122,316 FAU TSS200 5.08(7.03,3.13) 4.90E-07 − 0.06(− 1.80,1.68) 9.51E-01 2.199(2.1995,2.1994) 2.80E-02 −+ cg04311230 6 159,693,649 SOD2 TSS1500 4.52(6.68,2.78) 7.40E-06 – – – – – cg04349420 8 48,501,719 IGR 4.73(6.66,2.76) 2.79E-06 – – – – – cg07797073 1 2,051,352 PRKCZ Body 4.71(− 6.61,− 2.71) 3.02E-06 – – – – – cg11491381 20 3,082,857 AVP Body − 4.66(6.55,2.65) 3.85E-06 – – – – – cg18862005 2 177,076,135 IGR 4.60(6.56,2.66) 5.10E-06 – – – – – cg25794823 18 63,949,347 HMSD TSS200 4.61(− 8.16,− 4.26) 4.90E-06 – – – – – –6 Summary statistics for the 44 CpG sites with nominal association (p value < ­10 ) in the Discovery Analysis. Significant CpG sites in the meta-analysis are written in bold letters in the CpG site and p value columns. Numbers in italics in the p value column for the meta-analysis indicate CpG sites with more significant p values in the meta-analysis compared with the Discovery Analysis CpG: CpG site ID; CHR: chromosome where the CpG site is located; BP: specific chromosomal position for the CpG site; Gene: Gene annotation (from the Illumina Manifest File); Feature: Genomic location of the CpG site. It could reside in the 5’ untranslated region (UTR), between 0 and 200 nucleotides from the transcription start site (TSS), TSS200, or between 200 and 1500 nucleotides from the TSS, TSS1500. It could be also located in the body of the gene, in the 3’UTR or in an intergenic region (IGR); COEFF (UCI,LCI): Effect size for the association of the CpG site with the ∆NIHSS, with information for the upper (UCI) and lower (LCI) 95% confidence intervals; P: p value for the association of the CpG site with the ∆NIHSS In bold are the significant CpG sites in the meta-analysis Page 10 of 17 Cullell et al. Clinical Epigenetics (2022) 14:124 Page 11 of 17 Fig. 3 EXOC4 methylation in association with ΔNIHSS and with EXOC4 expression. A Correlation between EXOC4 methylation, calculated as β values of methylation (Y-axis) and ΔNIHSS (X-axis) for all the patients included in the Discovery Analysis. B Correlation between EXOC4 methylation, calculated as β values of methylation (Y-axis) and ΔEXOC4 expression (X-axis) assessed in 13 controls with EXOC4 expression and DNA methylation data direction (p value = 4.56 × ­10–04, coefficient: − 3.63) (Table 3). The analyses were performed including as (Additional file 1: Table SVII). input the list of all the CpG sites (with the information of their p values) or the list of only the nominally asso- ciated CpG sites from the meta-analysis. As the feature Pathway enrichment analysis enrichment analysis indicated an over-representation We identified different significantly enriched pathways of CpG sites from the gene body and promoter regions, after FDR adjustment (Q value < 0.05) using the methyl- we decided also to test the pathway enrichment analysis RRA function from methylGSA but only one significant depending on the CpG site location: gene body or gene pathway (p value < 0.05) and any significant pathways promoter (TSS1500 and TSS200). when using methylglm and methylgometh, respectively Table 3 Pathway enrichment analysis from meta-analysis EWAS results Function Database Method Sign CpG cut-off Type CpGs ID Description Size P value Q value mglm GO N/A No Promoter GO:0,030,100 Regulation of endocytosis 242 6.38E-04 5.36E-01 methylRRA​ GO GSA Nominal All GO:0,016,358 Dendrite development 206 3.76E-04 1.62E-02 methylRRA​ GO GSA Nominal All GO:0,048,588 Developmental cell growth 205 1.28E-03 2.75E-02 methylRRA​ GO GSA No All R-HSA-5688426 Deubiquitination 206 1.83E-02 1.83E-02 methylRRA​ Reactome GSA No Promoter R-HSA-5688426 Deubiquitination 206 4.57E-03 1.83E-02 methylRRA​ Reactome GSA Nominal Promoter R-HSA-5688426 Deubiquitination 206 4.72E-03 1.89E-02 methylRRA​ Reactome ORA Nominal All R-HSA-211859 Biological oxidations 203 3.08E-02 3.18E-02 methylRRA​ Reactome ORA Nominal All R-HSA-72203 Processing of Capped Intron- 203 3.08E-02 3.18E-02 Containing Pre-mRNA methylRRA​ Reactome ORA Nominal All R-HSA-162906 HIV infection 205 3.11E-02 3.18E-02 methylRRA​ Reactome ORA Nominal All R-HSA-68882 Mitotic anaphase 209 3.17E-02 3.18E-02 methylRRA​ Reactome ORA Nominal All R-HSA-2555396 Mitotic metaphase and anaphase 210 3.18E-02 3.18E-02 methylRRA​ KEGG ORA Nominal All 4510 Focal adhesion 200 3.03E-02 3.05E-02 methylRRA​ KEGG ORA Nominal All 4144 Endocytosis 201 3.05E-02 3.05E-02 Description of the significant pathways obtained from MethylGSA analysis using EWAS results from the meta-analysis Function: Indicates which of the three MethylGSA functions was used; Database: The pathway database used in the analysis; Method: Method used (ORA or GSA) when methylRRA was selected; Sign CpG cut-off: “No” indicates that the analysis was performed including the results from all the CpG sites and “Nominal” indicates that the analysis was performed using the nominally associated CpG sites; type CpGs: “All” indicates that the analysis was performed for all the CpG types, while “promoter” indicates that in the analysis only CpG sites from promotors were analysed; ID: Identifier for the specific pathway from each database; Description: Detail of the pathway; Size: number of genes included in the gene set; and p value: enrichment p value for each gene set; Q value: FDR corrected p value Cullell et al. Clinical Epigenetics (2022) 14:124 Page 12 of 17 The endocytosis pathway was found to be enriched only the nominal CpG sites; and 3) including only CpG using the methylglm function (in the Gene Ontology sites in promoters but also with all the CpG site types database) and the methylRRA function (in the KEGG (Table 3). Different pathways involving cell cycle and database). This pathway was enriched when all the CpG development were also overrepresented (Table 3). sites were considered but also when only the suggestive CpG sites were included. When CpG sites were filtered Gene expression and proteomic analyses out based on their location, the endocytosis pathway Only one CpG site (cg00039070 in EXOC4) met the was significant in two situations: when all the CpG pre-established criteria for being considered significant. sites were included independently of their location This CpG site was located in the gene body, 10 kb down- and when CpG sites from the promoter regions were stream a predicted enhancer. Given that the effect of gene selected (Table 3). The deubiquitination pathway was expression in gene body is variable, we wanted to assess also found to be enriched in different situations with whether this CpG site was affecting EXOC4 expression. the methylRRA function: 1) using Gene Ontology and We found a negative correlation between cg00039070 Reactome databases; 2) including all the CpG sites and methylation and EXOC4 mRNA levels (Spearman cor- relation: -0.469) although it was not significant (p value = 0.091) (Fig. 3B). To study the effect of EXOC4 CpG site methylation at Table 4 SOMAscan results the proteomic level, we evaluated the proteins differen- Protein t value P value tially expressed by EXOC4 methylation using SOMAs- can array data from 46 subjects (Additional file 1: Table IFNA7 − 3.59 8.52 × ­10–04 SVIII). We found 79 differentially expressed proteins C8A.C8B.C8G − 3.35 1.67 × ­10–03 (p value < 0.05) in association with cg00039070 meth- BMP1 − 3.09 3.52 × ­10–03 ylation. The most significant associations were for IGFBP4 − 2.85 6.73 × ­10–03 IFNA7 (p value = 8.52 × ­10–04) and C8A, C8B, C8G (p IL17A − 2.60 1.28 × ­10–02 value = 1.67 × ­10–03) (Table 4). The pathway enrichment CGA.FSHB − 2.57 1.36 × ­10–02 analysis using WebGestalt showed that the most signifi- TGFB1 − 2.55 1.43 × ­10–02 cantly associated pathway related to significant proteins IFNL1 − 2.51 1.58 × ­10–02 was NK cell activation (p value = 7.13 × ­10–04) (Table 5). LCN2 − 2.49 1.66 × ­10–02 SMAD3 − 2.48 1.70 × ­10–02 DMCT analysis and tissue specificity analysis BMPER − 2.47 1.74 × ­10–02 We looked for differentially methylated cell types based SERPINE2 − 2.46 1.78 × ­10–02 on the 14 CpG sites significantly or nominally associ- IL23R − 2.44 1.89 × ­10–02 ated with ΔNIHSS in the meta-analysis. We only identi- IL18R1 − 2.44 1.90 × ­10–02 fied significant CpG sites in NK (in NBEAL2 and SLC7A6 genes) and B cells (in NBEAL2 gene). However, the methylation and protein levels measured by SOMAscan® Assay Summary statistics for the top significant p value < 0.01) association of EXOC4 Table 5 Pathway enrichment analysis for significant proteins associated with EXOC4 methylation Gene Set Description Size Expect Ratio P value GO:0,030,101 NK cell activation 85 0.21814 27.505 7.13 × ­10–04 GO:0,042,110 T cell activation 452 11.600 8.6207 1.40 × ­10–03 GO:0,050,673 Epithelial cell proliferation 372 0.95468 9.4272 3.10 × ­10–03 GO:0,002,521 Leukocyte differentiation 496 12.729 7.8560 3.30 × ­10–03 GO:0,070,661 Leukocyte proliferation 281 0.72115 11.093 4.54 × ­10–03 GO:0,001,819 Positive regulation of cytokine production 418 10.727 8.3898 8.24 × ­10–03 GO:0,002,250 Adaptive immune response 382 0.98035 8.1604 4.53 × ­10–06 GO:0,002,285 Lymphocyte activation involved in immune response 172 0.44141 13.593 4.57 × ­10–06 GO:0,018,212 Peptidyl-tyrosine modification 389 0.99831 8.0135 5.17 × ­10–06 GO:0,002,449 Lymphocyte mediated immunity 238 0.61079 9.8233 2.9 × ­10–05 Description of the significant pathways from Gene Ontology enriched among proteins significantly associated with ΔNIHSS methylation Size: number of genes included in the gene set; expect: ratio of enrichment expected by chance in the gene set; Ratio: observed ratio for each specific gene set and p value: enrichment p value for each gene set Cullell et al. Clinical Epigenetics (2022) 14:124 Page 13 of 17 results were more significant in NK cells (NBEAL2: p were epigenome-wide significant (p value < 2.4 × ­10–07) value = 5.99 × ­10–09; SLC7A6: p value = 8.65 × ­10–08) than in the meta-analysis of the Discovery and Replication in B cells (NBEAL2: p value = 2.59 × ­10–07) (Additional cohorts, but only one CpG site (cg00039070) located in file 1: Figure SIII). There were not any CpG site in EXOC4 the body of the EXOC4 gene accomplished all the pre- differentially methylated in specific cell types. established criteria to be considered significant. We also looked for any tissue-specific regulatory com- EXOC4, also known as SEC8, encodes for a subunit in ponent from the 14 CpG sites significantly and nominally the exocyst complex, a protein complex involved in the associated with ΔNIHSS in the meta-analysis. Using as tethering of secretory vesicles to the plasma membrane functional element the 15 chromatin state marks from [48]. Different functions are attributed to the exocyst the Roadmap project, we identified several brain tissue complex, including but not limited to, exocytosis, cell signals nominally enriched (p < 0.05), but not significant growth cytokinesis and neuronal development [48, 49]. It after multiple comparison adjustment in our results is highly expressed in the brain and is enriched in axon (Additional file 1: Table SIX). growth cones and dendritic branches [48]. We found that the methylation pattern identified in association with stroke outcome was enriched for brain specific regula- Blood and brain correlation tory signals. We also investigated specifically the corre- We looked for correlation in the cg00039070 methylation lation of the methylation in cg00039070 between blood between blood and brain. Using the “Blood Brain DNA and brain using different tools. The results indicated Methylation Comparison Tool”, we did not find any cor- that the effect of the EXOC4 methylation in brain could relation in cg00039070 between any of the four brain be specific for some brain regions and its effect could be regions included in this tool (prefrontal cortex, entorhi- exacerbated in patients with stroke. nal cortex, superior temporal gyrus and cerebellum) and The pathway enrichment analysis showed that the reg- blood (Additional file 1: Figure SIV). ulation of DNA methylation in stroke outcome could be Using BECon, we found a negative correlation (− 0.48) mediated by regulation of the endocytosis and the deu- between the cg00039070 methylation in blood and BA20 biquitination. The endocytosis is the process by which (in the temporal cortex). This correlation was classified extracellular material is entered to the cell. This process in the highest correlation percentile (90%) (consider- has been showed to be interconnected with the exocy- ing all the CpG sites included in the tool) (Additional tosis in the regulation of different processes such as cell file 1: Figure SV). The BA7 region (in the parietal cortex) polarity [50]. Gachet et al. [51] demonstrated the affecta- presented an intermediate correlation (in the 50–75% tion of endocytosis in mutated SEC8 (EXOC4) yeasts and percentile) (Additional file 1: Figure SV). We found a a relationship with the cytokinesis process. Later, Jose lack of correlation in the prefrontal cortex that was also et al. [52] described the exocyst complex as a key network observed using the “Blood Brain DNA Methylation Com- hub which is regulating and coordinating both endocyto- parison Tool”. sis and exocytosis and the balance among both processes. The ubiquitin proteasome pathway is involved in the Discussion degradation of proteins and is key in the maintenance of In this study, our aim was to analyse whether the neuro- the correct neuronal and synaptic function. After stroke, logical course could be associated with epigenetic modifi- different pathological pathways are activated in response cations. With this objective, we studied the genome-wide to the neuronal injury such as mitochondrial autophagy, DNA methylation pattern associated with ΔNIHSS at oxidative stress and inflammatory response [53]. All discharge by EWAS. DNA methylation is probably the these processes are related to the ubiquitin proteasome most studied epigenetic variation, consisting of the addi- system. The specific role of the ubiquitin proteasome tion of a methyl group to a cytosine, mainly in the con- system in physiological and pathological processes after text of cytosines and guanines (CpG sites). We selected stroke is still in investigation but it has been suggested as ΔNIHSS at discharge as the main variable in the EWAS a potential target for new drugs [53]. because it was independently associated with Rankin at Higher methylation in cg00039070 from the EXOC4 3 months. The results from our EWAS suggest that the gene seems to be associated with a decrease in the neurological course of stroke patients measured as the expression of EXOC4 gene, based on our gene expres- difference between NIHSS at baseline and NIHSS at dis- sion results. The higher methylation in cg00039070 iden- charge has an impact on DNA methylation in specific tified in patients with a worse stroke outcome could be CpG sites. mediated by the decrease in the EXOC4 expression. Our From the 44 candidate CpG sites identified in the Dis- proteomic analysis also indicated a decrease in different covery Analysis, two CpG sites, located in genes bodies, protein levels linked to higher cg00039070 methylation. Cullell et al. Clinical Epigenetics (2022) 14:124 Page 14 of 17 The results from the pathway analysis from proteins dif- at baseline and at 24 h) [13]. Their functional annotation ferentially expressed linked with EXOC4 methylation, strongly suggested GRIA1 and ADAM23 associated with suggested that the inflammatory pathway, regulated by ΔNIHSS. Both genes are also involved in excitotoxicity NK cells, could be involved in the regulation of stroke processes. Both results support a role of excitotoxicity in outcome by methylation. The results from the differen- processes related to stroke neurological outcomes modu- tial methylation studied by cell type also supported the lated by genetic and epigenetic variations. Despite clini- involvement of this pathway. It showed some associations cal trials using drugs to modulate excitotoxicity processes for the CpG sites identified in our study, especially in having failed, progress has been made in clarifying the NK cells. NK cells are innate immune cells that infiltrate mechanisms that explain this failure [62]. ischaemic stroke lesions in human brains [54]. The func- One in vivo study in EXOC4 mutant’s drosophila tion of NK cells is regulated by activation and inhibitory showed that apart from this gene being involved in glu- receptors located in the cell surface. Thus, the endocyto- tamate receptor trafficking, it is also required for regulat- sis is also important for the NK cell receptors trafficking ing synaptic microtubule formation and synaptic growth, which is key to modulate the dynamic function of NK thus suggesting that EXOC4 methylation could be alter- cells [55]. ing different processes in the synapsis [63]. The Sec8 protein, encoded by EXOC4, has been seen to The excitotoxicity, neuroinflammatory and the synaptic control the synaptic targeting and the insertion of gluta- regulation are pathways that have been suggested to be mate receptors in the synapsis, controlling the directional pathological mediators of ischaemic brain damage [64] movement of glutamate receptors to the post-synaptic and could be potentially regulating the link between the membrane [49]. Another possible hypothesis for the methylation in EXOC4 and the stroke outcome. association of EXOC4 methylation and stroke outcome is the affectation of the glutamate receptors dynamism. Limitations In stroke, the release of the glutamate neurotransmit- The first limitation is the difference in sample size and ter is associated with ischaemic cell death in a process clinical features between the Discovery and Replication known as excitotoxicity. Briefly, the glutamate neuro- Cohorts. However, we looked for which clinical variables transmitter is increased because of the ischaemic insult were associated with the methylation of EXOC4 and we [56] and over-activates two kind of glutamate receptors: did not find any. Therefore, there is no reason to believe the N-methyl-D-aspartate receptor (NMDAR) and the that they would affect EXOC4 methylation in the Rep- α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid lication Cohort. Despite the differences between both receptor (AMPAR) [56, 57]. Activation of the synaptic cohorts, we have been able to replicate the results, which NMDAR leads to pro-survival signalling [58], while acti- reinforces the plausible implication of EXOC4 methyla- vation of extra-synaptic NMDAR induces a downstream tion in stroke outcome. neurotoxic cascade [59] that finally causes delayed neu- Another limitation is the use of whole blood to study ronal death. Both NMDAR and AMPAR have been DNA methylation in association with stroke outcome. reported to be associated with the Sec8 subunit of the However, other epigenomic, transcriptomic and prot- exocyst complex [49, 60], involved in the targeting of eomic studies on stroke [16, 65] have also used blood these receptors to the post-synaptic membrane [49, 60]. samples as it is also a relevant tissue in stroke outcome. Considering that our results indicated that EXOC4 meth- Additionally, the blood and brain tissues have been found ylation is associated with a decrease in the expression of to have a 0.86 correlation in global methylation [66]. For EXOC4 and with worse neurological course, we hypoth- that reason, we performed the analysis and then we cor- esize that Sec8 could be regulating the trafficking of syn- related the methylation results in brain and blood tissues. aptic glutamate receptors related to cell survival in stroke Finally, we were not able to find a significant correlation [58]. between EXOC4 mRNA levels and EXOC4 methylation, Moreover, apart from the differential methylation despite a trend being observed. Probably, the sample size identified in EXOC4, we found another gene, GRM3, of for the transcriptomic analysis was not large enough to nominally associated with ΔNIHSS in the discovery, obtain significant results. that encodes glutamate metabotropic receptor 3, also involved in excitotoxicity processes. This gene was asso- Conclusions ciated with memory impairment in a genetic study in We have expanded the knowledge about biological Alzheimer disease patients [61]. mechanisms regulating post-stroke outcome and high- A recently published GWAS has identified seven loci lighted the relevance of DNA methylation in explaining associated with stroke outcome measured through the variability in functional outcome. Despite a small sample NIHSS scale (calculating the difference between NIHSS size, we had enough statistical power to obtain results Cullell et al. Clinical Epigenetics (2022) 14:124 Page 15 of 17 that support the hypothesis of the excitotoxicity, neu- Declarations roinflammatory and synapsis regulation pathways play- Ethics approval and consent to participate ing a significant role in stroke and indicate that further The study was approved by the ethics committees of all the participant hospi‑ research is needed in this field to confirm this pathway as tals (the promoter committees were: Comité Ético de Investigación Clínica del a future therapeutic target. Parc de Salut Mar, Comitè Ètic d’Investigació Clínica de l’Hospital Universitari Vall d’Hebron and Comitè ètic de l’Hospital MútuaTerrassa). All participants provided written informed consent to participate. Abbreviations Consent for publication AMPAR: α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor; Not applicable. DALYs: Disability-adjusted life-years; DMBs: Differential methylation blocks; DMCT: Differentially methylated cell types; DMP: Differentially methylated Competing interests positions; DMRs: Differential methylation regions; EWAS: Epigenome-wide The authors declare no competing interests. association study; GSA: Gene set enrichment analysis; GWAS: Genome-wide association studies; MRI: Magnetic resonance imaging; mRS: Modified Rankin Author details Scale; NK: Natural killer; NMDAR: N-methyl-d-aspartate receptor; NIHSS: 1 Stroke Pharmacogenomics and Genetics, IIB‑Sant Pau, Institut de Recerca de National Institutes of Health Stroke Scale; ORA: Over-representation analysis; Sant Pau, Hospital Sant Pau, C/Sant Antoni Mª Claret,167, 08025 Barcelona, PC: Principal components; QCs: Quality controls; TSS: Transcription start site; Spain. 2 Neurology, Hospital Universitari MútuaTerrassa/Fundacio Docència UTR​: Untranslated region. i Recerca MutuaTerrassa, Terrassa, Spain. 3 Facultat de Medicina, Universitat de Barcelona, Barcelona, Spain. 4 Neurology, Hospital del Mar, Neurovascular Supplementary Information Research Group, IMIM, Universitat Autònoma de Barcelona/DCEXS‐Universitat Pompeu Fabra, Barcelona, Spain. 5 Department of Psychiatry, Washington The online version contains supplementary material available at https://​doi.​ University School of Medicine, Saint Louis, Missouri, USA. 6 NeuroGenomics org/​10.​1186/​s13148-​022-​01340-5. and Informatics, Washington University School of Medicine, Saint Louis, Mis‑ souri, USA. 7 Cancer Epigenetics & Biology Program (PEBC), L’Hospitalet, Spain. 8 Additional file 1. Supplemental Methods. Department of Physiological Sciences II, School of Medicine, Universitat de Barcelona, Barcelona, Spain. 9 Institució Catalana de Recerca I Estudis Avançats (ICREA), Barcelona, Spain. 10 Neurovascular Research Laboratory, Vall d’Hebron Acknowledgements Institut de Recerca (VHIR), Barcelona, Spain. 11 Department of Neurology, Hos‑ We thank the International Stroke Genetics Consortium, the Spanish pital Universitario Virgen Macarena Sevilla, Instituto de Biomedicina de Sevilla, Stroke Genetics Consortium, the International Stroke Genetics Consor‑ IBiS/Hospital Universitario Virgen del Rocío/CSIC, Universidad de Sevilla, Sevilla, tium, the Global Alliance for Stroke acute and long-term outcome, RETICS Spain. 12 Neurology, Hospital de La Santa Creu i Sant Pau, Barcelona, Spain. Network INVICTUS (RD16/0019/0002, RD16/0019/0010, RD16/0019/0011, 13 Centre for Bioscience, School of HealthCare Science, Manchester Metropoli‑ RD16/0019/0021) and the RICORDS Stroke network. tan University, Manchester, UK. Author contributions Received: 20 May 2022 Accepted: 13 September 2022 I.FC. and J.JC. are responsible for the study design and the analytical accuracy of the manuscript. N.C. performed statistical analyses and prepared the manuscript. C.ST. performed statistical analyses and reviewed the manuscript. C.GF., J.CM., L.LLC, M.LL. and E.M. reviewed the manuscript and contributed to data interpretation. M.E. and M.C. performed the genotyping. J.M. and References A.R. collected samples included in the study. J.MF., J.K. and J.R contributed to 1. O’Donnell MJ, Chin SL, Rangarajan S, Xavier D, Liu L, Zhang H, et al. Global data interpretation. All the authors reviewed and approved the manuscript and regional effects of potentially modifiable risk factors associated with content. acute stroke in 32 countries (INTERSTROKE): a case-control study. Lancet Lond Engl. 2016;388(10046):761–75. Funding 2. Vos T, Lim SS, Abbafati C, Abbas KM, Abbasi M, Abbasifard M, et al. Global EPIGENESIS project (Carlos III Institute/Fondo Europeo de Desarrollo Regional burden of 369 diseases and injuries in 204 countries and territories, (FEDER)- PI17/02089, Marató TV3 and Fundació MútuaTerrassa), MAESTRO 1990–2019: a systematic analysis for the global burden of disease study project (Carlos III Institute/FEDER—PI18/01338), iBioStroke project (Eranet- 2019. The Lancet. 2020;396(10258):1204–22. Neuron, European research grants), the EPINEXO project- PI20/00678 (Carlos III 3. Stefan S, Bo N, Jesper P, Teresa U. Long-term survival and function after Institute/FEDER), SEDMAN Study (Boehringer Ingelheim), APHAS Study (Pfizer/ stroke. Stroke. 2019;50(1):53–61. Bristol Myers), Fondo Europeo de Desarrollo Regional (FEDER), 2017SGR-1427 4. Wouters A, Nysten C, Thijs V, Lemmens R. Prediction of outcome in (AGAUR), the RETICS Network INVICTUS + and the RICORDS Stroke network. patients with acute ischemic stroke based on initial severity and improve‑ J. Cárcel-Márquez is supported by an AGAUR Contract (agència de gestió ment in the first 24 h. Front Neurol. 2018;9:308. d’ajuts universitaris i de recerca; FI_DGR 2019, grant number 2019_FI_B 00853) 5. Heitsch L, Ibanez L, Carrera C, Binkley MM, Strbian D, Tatlisumak T, et al. co-financed by Fons Social Europeu (FSE). I. Fernandez is the recipient of a Early neurological change after ischemic stroke is associated with 90-day research contract from the Miguel Servet Program (CP12/03298) from the Car‑ outcome. Stroke. 2021;52(1):132–41. los III Institute. M. Lledós is supported by a PFIS Contract (Contratos Predocto‑ 6. Torres-Aguila NP, Carrera C, Muiño E, Cullell N, Cárcel-Márquez J, Gallego- rales de Formación en Investigación en Salud): FI19/00309 C. Gallego-Fabrega Fabrega C, et al. Clinical variables and genetic risk factors associated is supported by a Sara Borrell Contract (CD20/00043) from Instituto de Salud with the acute outcome of ischemic stroke: a systematic review. J Stroke. Carlos III and Fondo Europeo de Desarrollo Regional (ISCIII-FEDER). E. Muiño 2019;21(3):276–89. is supported by a Río Hortega Contract (CM18/00198) from the Instituto de 7. Heuschmann PU, Wiedmann S, Wellwood I, Rudd A, Carlo AD, Bejot Y, Salud Carlos III. et al. Three-month stroke outcome: the European Registers of Stroke (EROS) investigators. Neurology. 2011;76(2):159–65. Availability of data and materials 8. Taylor-Rowan M, Wilson A, Dawson J, Quinn TJ. Functional Assessment The data sets used and/or analysed during the current study are available for acute stroke trials: properties, analysis, and application. Front Neurol. from the corresponding author on reasonable request. 2018;9:191. https://​doi.​org/​10.​3389/​fneur.​2018.​00191. Cullell et al. Clinical Epigenetics (2022) 14:124 Page 16 of 17 9. Banks JL, Marotta CA. Outcomes validity and reliability of the modified for the analysis of Infinium DNA methylation microarrays. Bioinformatics. rankin scale: implications for stroke clinical trials: a literature review and 2014;30:1363–9. synthesis. Stroke. 2007;38:1091–6. 30. Morris TJ, Butcher LM, Feber A, et al. ChAMP: 450k chip analysis methyla‑ 10. Kim YD, Song D, Kim EH, Lee KJ, Lee HS, Nam CM, et al. Long-term mortal‑ tion pipeline. Bioinformatics. 2014. https://​doi.​org/​10.​1093/​bioin​forma​ ity according to the characteristics of early neurological deterioration in tics/​btt684. ischemic stroke patients. Yonsei Med J. 2014;55(3):669–75. 31. LaBarre BA, Goncearenco A, Petrykowska HM, Jaratlerdsiri W, Bornman 11. Reznik ME, Yaghi S, Jayaraman MV, McTaggart RA, Hemendinger M, MSR, Hayes VM, et al. MethylToSNP: identifying SNPs in Illumina DNA Mac Grory BC, et al. Baseline NIH stroke scale is an inferior predictor of methylation array data. Epigenetics Chromatin. 2019;12(1):79. functional outcome in the era of acute stroke intervention. Int J Stroke. 32. Tissue plasminogen activator for acute ischemic stroke. N. Engl. J. Med. 2018;13(8):806–10. 1995, 333, 1581–1588. 12. Meyer L, Stracke CP, Jungi N, Wallocha M, Broocks G, Sporns PB, et al. 33. Meyer L, Stracke CP, Jungi N, Wallocha M, Broocks G, Sporns PB, Mae‑ Thrombectomy for primary distal posterior cerebral artery occlusion gerlein C, Dorn F, Zimmermann H, Naziri W, et al. Thrombectomy for stroke: the TOPMOST study. JAMA Neurol. 2021;78(4):434–44. primary distal posterior cerebral artery occlusion stroke: the TOPMOST 13. Ibanez L, Heitsch L, Carrera C, Farias FHG, Dhar R, Budde J, et al. Multi- study. JAMA Neurol. 2021;78:434–44. ancestry genetic study in 5876 patients identifies an association between 34. Campagna MP, Xavier A, Lechner-Scott J, Maltby V, Scott RJ, Butzkueven excitotoxic genes and early outcomes after acute ischemic stroke. H, Jokubaitis VG, Lea RA. Epigenome-wide association studies: current MedRxiv.2020 knowledge. Strateg Recomm Clin Epigenetics. 2021;13:214. 14. Söderholm M, Pedersen A, Lorentzen E, Stanne TM, Bevan S, Olsson M, 35. Saffari A, Silver MJ, Zavattari P, Moi L, Columbano A, Meaburn EL, Dud‑ et al. Genome-wide association meta-analysis of functional outcome bridge F. Estimation of a significance threshold for epigenome-wide after ischemic stroke. Neurology. 2019;92(12):e1271–83. association studies. Genet Epidemiol. 2018;42:20–33. 15. Mola-Caminal M, Carrera C, Soriano-Tárraga C, Giralt-Steinhauer E, Díaz- 36. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of Navarro RM, Tur S, et al. PATJ low frequency variants are associated with genomewide association scans. Bioinformatics. 2010;26:2190–1. worse ischemic stroke functional outcome. Circ Res. 2019;124(1):114–20. 37. Nikolakopoulou A, Mavridis D, Salanti G. Demystifying fixed and 16. Soriano-Tárraga C, Lazcano U, Giralt-Steinhauer E, Avellaneda-Gómez random effects meta-analysis. Evid Based Ment Health. 2014;17:53–7. C, Ois Á, Rodríguez-Campello A, et al. Identification of 20 novel loci https://​doi.​org/​10.​1136/​eb-​2014-​101795. associated with ischaemic stroke. Epigenome-wide association study. 38. Ren X, Kuan PF. MethylGSA: a bioconductor package and shiny app Epigenetics. 2020. https://​doi.​org/​10.​1080/​15592​294.​2020.​17465​07. for DNA methylation data length bias adjustment in gene set testing. 17. Davis Armstrong NM, Chen W-M, Brewer MS, Williams SR, Sale MM, Wor‑ Bioinform Oxf Engl. 2019;35:1958–9. rall BB, et al. Epigenome-wide analyses identify two novel associations 39. Candia J, Cheung F, Kotliarov Y, Fantoni G, Sellers B, Griesman T, et al. with recurrent stroke in the vitamin intervention for stroke prevention Assessment of variability in the SOMAscan assay. Sci Rep. 2017;7:14248. clinical trial. Front Genet. 2018;9:358. 40. Simats A, García-Berrocoso T, Ramiro L, Giralt D, Gill N, Penalba A, et al. 18. Gallego-Fabrega C, Carrera C, Reny J-L, Fontana P, Slowik A, Pera J, et al. Characterization of the rat cerebrospinal fluid proteome following PPM1A methylation is associated with vascular recurrence in aspirin- acute cerebral ischemia using an aptamer-based proteomic technol‑ treated patients. Stroke J Cereb Circ. 2016;47(7):1926–9. ogy. Sci Rep. 2018;8(1):7899. 19. Gallego-Fabrega C, Carrera C, Reny J-L, Fontana P, Slowik A, Pera J, et al. 41. Riba I, Jarca CI, Mundet X, et al. Cognitive assessment protocol design TRAF3 epigenetic regulation is associated with vascular recurrence in in the ISSYS (investigating silent strokes in hYpertensives: a magnetic patients with ischemic stroke. Stroke. 2016. https://​doi.​org/​10.​1161/​ resonance imaging study). J Neurol Sci. 2012;322:79–88. STROK​EAHA.​115.​012237. 42. Nguyen T-M, Shafi A, Nguyen T, Draghici S. Identifying significantly 20. Soriano-Tárraga C, Mola-Caminal M, Giralt-Steinhauer E, Ois A, Rodríguez- impacted pathways: a comprehensive review and assessment. Campello A, Cuadrado-Godia E, et al. Biological age is better than chrono‑ Genome Biol. 2019;20(1):203. logical as predictor of 3-month outcome in ischemic stroke. Neurology. 43. Liao Y, Wang J, Jaehnig EJ, Shi Z, Zhang B. WebGestalt 2019: gene 2017;89(8):830–6. set analysis toolkit with revamped UIs and APIs. Nucleic Acids Res. 21. Soriano-Tárraga C, Giralt-Steinhauer E, Mola-Caminal M, Ois A, Rodríguez- 2019;47(W1):W199-205. Campello A, Cuadrado-Godia E, et al. Biological age is a predictor of 44. Zheng SC, Breeze CE, Beck S, Teschendorff AE. Identification of differen‑ mortality in ischemic stroke. Sci Rep. 2018;8(1):4148. tially methylated cell types in epigenome-wide association studies. Nat 22. Roquer J, Rodríguez-Campello A, Gomis M, Jiménez-Conde J, Cuadrado- Methods. 2018;15(12):1059–66. Godia E, Vivanco R, et al. Acute stroke unit care and early neurological 45. Breeze CE, Reynolds AP, van Dongen J, Dunham I, Lazar J, Neph S, deterioration in ischemic stroke. J Neurol. 2008;255(7):1012–7. Vierstra J, Bourque G, Teschendorff AE, Stamatoyannopoulos JA, et al. 23. Soriano-Tárraga C, Giralt-Steinhauer E, Mola-Caminal M, Vivanco-Hidalgo EFORGE v2.0: updated analysis of cell type-specific signal in epig‑ RM, Ois A, Rodríguez-Campello A, et al. Ischemic stroke patients are bio‑ enomic data. Bioinform Oxf Engl. 2019;35:4767–9. logically older than their chronological age. Aging. 2016;8(11):2655–65. 46. Hannon E, Lunnon K, Schalkwyk L, Mill J. Interindividual methylomic 24. Soriano-Tárraga C, Jiménez-Conde J, Giralt-Steinhauer E, Mola-Caminal variation across blood, cortex, and cerebellum: implications for M, Vivanco-Hidalgo RM, Ois A, et al. Epigenome-wide association study epigenetic studies of neurological and neuropsychiatric phenotypes. identifies TXNIP gene associated with type 2 diabetes mellitus and Epigenetics. 2015;10:1024–32. sustained hyperglycemia. Hum Mol Genet. 2016;25(3):609–19. 47. Edgar RD, Jones MJ, Meaney MJ, Turecki G, Kobor MS. BECon: a tool for 25. Fernández-Cadenas I, Mendióroz M, Giralt D, Nafria C, Garcia E, Carrera interpreting DNA methylation findings from blood in the context of C, et al. GRECOS project (genotyping recurrence risk of stroke): the use brain. Transl Psychiatry. 2017;7: e1187. of genetics to predict the vascular recurrence after stroke. Stroke. 2017. 48. Mei K, Guo W. The exocyst complex. Curr Biol CB. 2018;28(17):R922-5. https://​doi.​org/​10.​1161/​STROK​EAHA.​116.​014322. 49. Gerges NZ, Backos DS, Rupasinghe CN, Spaller MR, Esteban JA. Dual 26. Del Río-Espínola A, Fernández-Cadenas I, Giralt D, Quiroga A, Gutiérrez- role of the exocyst in AMPA receptor targeting and insertion into the Agullõ M, Quintana M, et al. A predictive clinical-genetic model of tissue postsynaptic membrane. EMBO J. 2006;25(8):1623–34. plasminogen activator response in acute ischemic stroke. Ann Neurol. 50. Martin-Urdiroz M, Deeks MJ, Horton CG, Dawe HR, Jourdain I. The exo‑ 2012;72(5):716–29. cyst complex in health and disease. Front Cell Dev Biol. 2016. https://​ 27. Tian Y, Morris TJ, Webster AP, Yang Z, Beck S, Feber A, Teschendorff AE. doi.​org/​10.​3389/​fcell.​2016.​00024. ChAMP: updated methylation analysis pipeline for illumina beadchips. 51. Gachet Y, Hyams JS. Endocytosis in fission yeast is spatially associated Bioinformatics. 2017;33:3982–4. with the actin cytoskeleton during polarised cell growth and cytokine‑ 28. Zhou W, Laird PW, Shen H. Comprehensive characterization, annota‑ sis. J Cell Sci. 2005;118:4231–42. tion and innovative use of infinium DNA methylation beadchip probes. 52. Jose M, Tollis S, Nair D, Mitteau R, Velours C, Massoni-Laporte A, Royou Nucleic Acids Res. 2017. https://​doi.​org/​10.​1093/​nar/​gkw967. A, Sibarita J-B, McCusker D. A quantitative imaging-based screen 29. Aryee MJ, Jaffe AE, Corrada-Bravo H, Ladd-Acosta C, Feinberg AP, Hansen reveals the exocyst as a network hub connecting endocytosis and KD, Irizarry RA. Minfi: a flexible and comprehensive Bioconductor package exocytosis. Mol Biol Cell. 2015;26:2519–34. Cullell et al. Clinical Epigenetics (2022) 14:124 Page 17 of 17 53. Li Y-C, Wang Y, Zou W. Exploration on the mechanism of Ubiquitin pro‑ teasome system in cerebral stroke. Front Aging Neurosci. 2022. https://​ doi.​org/​10.​3389/​fnagi.​2022.​814463. 54. Gan Y, Liu Q, Wu W, Yin J-X, Bai X-F, Shen R, Wang Y, Chen J, La Cava A, Poursine-Laurent J, et al. Ischemic neurons recruit natural killer cells that accelerate brain infarction. Proc Natl Acad Sci USA. 2014;111:2704–9. 55. Masilamani M, Peruzzi G, Borrego F, Coligan JE. Endocytosis and intracel‑ lular trafficking of human natural killer cell receptors. Traffic Cph Den. 2009;10:1735–44. 56. Achzet LM, Davison CJ, Shea M, Sturgeon I, Jackson DA. Oxidative stress underlies the ischemia/reperfusion-induced internalization and degrada‑ tion of AMPA receptors. Int J Mol Sci. 2021. https://​doi.​org/​10.​3390/​ijms2​ 20207​17. 57. Chamorro Á, Dirnagl U, Urra X, Planas AM. Neuroprotection in acute stroke: targeting excitotoxicity, oxidative and nitrosative stress, and inflammation. Lancet Neurol. 2016;15(8):869–81. 58. Lai TW, Zhang S, Wang YT. Excitotoxicity and stroke: Identifying novel targets for neuroprotection. Prog Neurobiol. 2014;115:157–88. 59. Szydlowska K, Tymianski M. Calcium, ischemia and excitotoxicity. Cell Calcium. 2010;47(2):122–9. 60. Sans N, Prybylowski K, Petralia RS, Chang K, Wang Y-X, Racca C, et al. NMDA receptor trafficking through an interaction between PDZ proteins and the exocyst complex. Nat Cell Biol. 2003;5(6):520–30. 61. Jablensky A, Morar B, Wiltshire S, Carter K, Dragovic M, Badcock JC, et al. Polymorphisms associated with normal memory variation also affect memory impairment in schizophrenia. Genes Brain Behav. 2011;10(4):410–7. 62. Choi DW. Excitotoxicity: still hammering the ischemic brain in 2020. Front Neurosci. 2020;14:579953. 63. Liebl FLW, Chen K, Karr J, Sheng Q, Featherstone DE. Increased synaptic microtubules and altered synapse development in Drosophila sec8 mutants. BMC Biol. 2005. https://​doi.​org/​10.​1186/​1741-​7007-3-​27. 64. Jayaraj RL, Azimullah S, Beiram R, Jalal FY, Rosenberg GA. Neuroin‑ flammation: friend and foe for ischemic stroke. J Neuroinflammation. 2019;16:142. 65. Kamtchum-tatuene J, Jickling GC. Blood biomarkers for stroke diagnosis and management. Neuromol Med. 2019;21:344–68. 66. Braun PR, Han S, Hing B, Nagahama Y, Gaul LN, Heinzman JT, Grossbach AJ, Close L, Dlouhy BJ, Howard MA, et al. Genome-wide DNA methylation comparison between live human brain and peripheral tissues within individuals. Transl Psychiatry. 2019;9:47. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub‑ lished maps and institutional affiliations. Ready to submit your research ? Choose BMC and benefit from: • fast, convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations • maximum visibility for your research: over 100M website views per year At BMC, research is always in progress. Learn more biomedcentral.com/submissions