Biological Age Acceleration Is Lower in Women With Ischemic Stroke Compared to Men

Stroke risk and mortality increase with age.¹ Despite advances in prevention and treatment, the aging population in conjunction with accumulating risk factors, contribute to an increasing lifetime risk of strokes.^1,2 Age-specific incidence rates are lower in women than men in the younger and middle-aged groups, but these differences narrow in the older groups, where incidence rates in women are approximately equal to or even higher than in men.^3,4

An extended hypothesis regarding sexual dichotomy in stroke onset is that steroid hormones, in particular estrogen, play a neuroprotective role.⁵ Several studies in animal models suggest that females are protected by endogenous estrogens. However, randomized trials did not find a reduced risk of coronary heart disease in postmenopausal women treated with hormone therapy.^6–8 Similarly, the WEST (Women’s Estrogen for Stroke Trial) did not observe a reduction in mortality or stroke recurrence in postmenopausal women after an ischemic stroke (IS) or a transient ischemic attack when treated with estradiol.⁹ A recent observational study of 5 population-based cohorts, concluded that the potential protective effects of hormone therapy are time dependent on treatment initiation in relation to menopause onset.¹⁰ Currently, the biological mechanisms associated with the lower stroke incidence rates in women in younger and middle-aged groups are poorly understood. However, there is evidence that genetic and epigenetic factors may play a role in sexual disparities in stroke age at onset.¹¹

Epigenetic regulation is a critical driver of cell function and survival that involves many pathways. Differences in DNA methylation (DNAm) have been found in a variety of age-related diseases,¹² as well as in the normal aging process.^13–15 Biological age is a measure of an individual’s age based on different biomarkers. It can be determined via telomere length, genomic instability, or epigenetic marks. An individual’s biological age changes throughout their lifespan and can be modulated by both environmental factors and genetic variation.^14,15 This process captures the biological status of an individual which may differ from their chronological age (time passed since birth). In recent years, Epigenetic clocks have gained popularity, in part, because they accurately measure age-related health and mortality.¹⁶

Epigenetic clocks are biological age-prediction models, based on the DNAm levels of certain cytosine-phosphate-guanine sites (CpGs) directly related to age. Each model uses different approaches, but most of them are based on regression techniques.¹⁷ The 2 more broadly used DNAm age estimators are those developed by Horvath¹⁸ and Hannum et al.¹⁹ Horvath developed a multitissue age predictor which estimates biological age from different tissues or cell types.¹⁸ Simultaneously, Hannum et al¹⁹ developed a blood-specific estimator to calculate the individual’s biological age.¹⁹ Measures of epigenetic age acceleration (EAA) have been developed based on these DNAm age predictors. EAA measures indicate whether an individual’s age is higher than their chronological age, therefore capturing the aging process.²⁰ Two widely used EAA measures are based on blood samples, that are either independent of blood cell counts (cell-intrinsic measured; intrinsic EAA [IEAA]) or enhanced by the changes in blood cell composition (extrinsic measured; extrinsic EAA [EEAA]).²⁰

Several studies have reported the association of EAA measures with diverse conditions. The effects of age acceleration have been observed in brain and blood tissue in Down syndrome²¹ and in the dorsolateral prefrontal cortex in Alzheimer disease.²² Both IEAA and EEAA are higher in Parkinson disease patients compared with controls.²³ However, decelerated epigenetic aging is associated with mood stabilizers in patients with bipolar disorder.²⁴ Age acceleration has also been linked to reproductive aging. Levine et al²⁵ observed an increased EAA in early menopause. Their study indicates that menopause may accelerate biological aging in blood and that chronological age at menopause and EAA share a common genetic signature.²⁵ A previous study in IS observed that IS patients are biologically older than their chronological age at time of onset.²⁶ This phenomenon is not exclusive to IS and has been observed across other age-related diseases.²⁷

Here, we aimed to evaluate EAA in IS female and male patients, to determine if age acceleration explains the differences in age at stroke onset among sexes. Additionally, we aimed to assess the EAA differences in women and men, stratified into stroke subtypes, and evaluate the relation of EEAA to blood protein levels in IS.

Methods

Participants

Data Availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Discovery Sample

Table 1 summarizes the demographic and clinical data of the study cohorts. A total of N=374 ischemic stroke patients, from the Vall d’Hebron Hospital’s (Barcelona, Spain) GRECOS (Genotyping Recurrence Risk of Stroke) project²⁸ and Mutua Terrassa Hospital’s (Barcelona, Spain) SEDMAN (Dabigatran Study in the Early Phase of Stroke, New Neuroimaging Markers and Biomarkers) project²⁹ were used for the discovery phase. The GRECOS project²⁸ is a project that consecutively recruited patients (N=1494) of European descent admitted to the emergency room of 23 Spanish hospitals, in the first 24 hours after the patients’ first-ever IS, between July 2005 and May 2009. The SEDMAN study²⁹ is an ongoing prospective, multicenter study that has been consecutively enrolling stroke patients from 12 different Spanish sites since June 2016 (https://www.clinicaltrials.gov; Unique identifier: NCT02742480). For the present study, we selected IS cases from both projects. They were recruited within the first 24 hours of a first-ever IS event, >18 years old with data on age, sex, smoking status, diabetes, dyslipidemia, hypertension and stroke subtypes, and available biological material to perform an epigenomic study (Table 1).

A stratification analysis of time of blood extraction within the acute phase of IS was performed. We grouped together samples recruited within early time points, within the first 6 hours of hospital admission (0–6 hours; N=182), and samples labeled as having been obtained within a broader range of up to the first 24 hours (0–24 hours; N=192). No detailed information on 6 to 24 hours was available to differentiate early and late extraction.

Replication Sample

For the replication phase, N=981, IS patients from the BASICMAR prospective register,³⁰ Hospital del Mar (Barcelona, Spain) were used. Blood samples were obtained during the first 24 hours after first IS event; patients were >18 and data on age, sex, smoking status, diabetes, dyslipidemia, and hypertension were available (Table 1).

Cardiovascular and demographic factors were collected from the different cohorts and TOAST (Trial of ORG 10172 in Acute Stroke Treatment)³¹ was used to classify stroke etiology (Table 1).

The study was carried out in accordance with the Declaration of Helsinki and was approved by the respective hospitals’ research ethics committees. Written informed consent was obtained from all participants, who did not receive any economic compensation.

Methylation Study

Epigenome Wide Methylation Assay

Whole blood was obtained during the first 24 hours after a stroke. For the discovery phase, DNA was extracted using the Gentra Pure gene Blood Kid (Qiagen, Hilden, Germany) following the manufacturer’s instructions and stored at −80 ºC until required. DNA from the replication phase was extracted using manual salt precipitation in the Banco Nacional de ADN (Instituto de Salud Carlos III).

Genome-wide DNAm was assessed using the Infinium HumanMethylation450K BeadChip (450K) and Infinium MethylationEPIC BeadChip (EPIC) (Illumina Inc, San Diego, CA), which analyzed more than 450 000 and 850 000 CpG-sites respectively. Samples from the discovery used the 450K and EPIC arrays (450K=113 samples, EPIC=261 samples) and the replication cohort, the EPIC array. DNA methylation β-values, the ratio of methylation ranging from 0 (unmethylated) to 1 (methylated), were determined using the GenomeStudio Software (Illumina). Quality control metrics were examined to determine the success of the bisulfite conversion and array hybridization. Probe filtering was performed by removing those that failed at hybridization (detection P>0.05) and were not represented by a minimum of 3 beads. Samples with poor bisulfite conversion were also removed (> 1% of probes with a detection P>0.01), as described elsewhere.^32,33 Special attention was placed on identifying and removing samples with sex discrepancies between self-reported gender and biological/genetic sex.

DNAm-Based Biological Age Calculation

We used the online calculator (http://dnamage.genetics.ucla.edu/)¹⁸ to evaluate 2 DNAm-based biological age predictors, Horvath multitissue epigenetic clock,¹⁸ Hannum blood-based epigenetic clock,¹⁹ and EAA measures.²⁰

DNAm age (in units of years) estimates the number of years that had passed since birth. Horvath estimator (Horvath-DNAmAge) was developed using 353 CpGs from multiple tissues from children and adults,¹⁸ whereas Hannum estimator (Hannum-DNAmAge) was developed using 71 CpGs from blood samples from adults.¹⁹ CpGs from both epigenetics clocks are present in the 450K and EPIC arrays.

EAA measures are the residual from regressing DNAm age on chronological age. Positive EAA indicates a higher epigenetic/biological age than expected from chronological age, whereas negative EAA values indicate lower epigenetic/biological age than expected. We have calculated EAA based on Horvath (Horvath-EAA) and EAA based on Hannum (Hannum-EAA) epigenetic clocks. Additionally, intrinsic and extrinsic calculations of EAA were also used. IEAA is defined as independent of blood cell counts, whereas EEAA is enhanced by blood cell count information.²⁰

Functional Study

A total of 26 samples from the discovery cohort had available data on protein levels, as part of the ISSYS (Investigating Silent Strokes in Hypertensives: a Magnetic Resonance Imaging Study) project.³⁴ The ISSYS, is an observational cross-sectional and longitudinal study aimed to determine the relationship of stroke risk and cognitive decline in a cohort of hypertensive patients.³⁴ We leveraged the availability of this data to perform a small pilot function study. All samples from the discovery cohort with available protein data were selected, descriptive characteristics of these samples are shown in Table 1.

Blood samples from a subsection of the discovery cohort, N=26 IS, were analyzed by a highly multiplex assay of 1305 proteins (SOMAScan, SomaLogic, Inc, Boulder, Co). Descriptive characteristics of samples used in the functional study are shown in Table 1. The assay transforms the measurement of proteins into the measurement of relative fluorescent units. As part of the quality control process samples and proteins with a high percentage of failed measures were removed. We evaluated the relationship between EAA measures with protein levels. Proteins with nominal association P<0.05 were selected for gene set analysis. The WEB-Based Gene Set Analysis Toolkit (WebGestal)³⁵ was used to identify enriched gene sets. We evaluated the overlap between our gene set and the Gene Ontology’s nonredundant biological process database, using the Gene Set Enrichment Analysis method.

Statistical Analyses

The R statistical computing environment³⁶ (v.3.6.3) was used to perform Illumina’s Infinium Methylation BeadChip quality control and preprocessing, as well as all statistical analyses and figures. Univariate analyses were calculated using chi-squared and the Mann-Whitney U test; for the multivariate analysis, we used generalized linear models. Multivariate linear models were used to adjust for covariates in each individual cohort, based on the differences found among the groups compared (Table 1). Discovery cohort, model 1: accounts for smoking status and dyslipidemia; replication cohort, model 2: accounts for smoking status and hypertension. Pearson correlation test was used to measure the linear correlation between chronological and biological age. To analyze the differences between stroke subgroups, we used the ANOVA statistical test. Finally, Spearman correlation test was used to assess the relationship between the EAA measures and SOMAScan protein levels.

Missing Data

There were no missing data, for age, sex, smoking status, diabetes, dyslipidemia, hypertension in either the discovery or replication cohort. Stroke subtype data were only available in the discovery cohort; consequently, no stratification analysis was performed on the replication cohort.

Reporting Checklist for Cohort Study

We used the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) cohort reporting guidelines.³⁷

Results

There was a significant correlation between chronological age (the age at the time of stroke onset), with Horvath-DNAmAge (P=6.83×10^-48), and with Hannum-DNAmAge (P=5.83×10^-24), indicating the accuracy of the epigenetic estimators used (Figure 1A).

• Download figure
• Download PowerPoint

As expected, we observed a large statistically significant difference in chronological age at stroke onset, between men and women (77.8±10 years in women and 71.8±10 years in men, P=1.57×10^-5); conversely, both groups showed the same biological age, Horvath-DNAmAge, (64.3±7.8 years women, 64.4±8.7 years men; P=0.75) and Hannum-DNAmAge (64.9±11 years women, 66.7±12 years men; P=0.78) (Table 2) (Figure 1B).

Sex Dichotomy of Age Acceleration in All Stroke Patients

When assessing EAA measures, women showed negative acceleration values across the board (Horvath-EAA=−0.89; Horvath-IEAA=-0.66; Hannum-EAA=−0.55; Hannum-IEAA=−0.49; Hannum-EEAA=−0.64), whereas men all showed positive values (Horvath-EAA=0.93; Horvath-IEAA=0.57; Hannum-EAA=0.89; Hannum-IEAA=0.40; Hannum-EEAA=1.24) (Table 2, Figure 2). Differences in EAA measurements between female and male IS patients reached nominal statistical significance in all analyses, after adjustment for confounding factors, Horvath-EAA (P=6.93×10^-3), Horvath-IEAA (P=4.16×10^-02), Hannum-EAA (P=1.90×10^-02), and Hannum-EEAA (P=1.34×10^-2) except for in Hannum-IEAA (P=9.25×10^-2) (Table 2). A stratification analysis by time of blood extraction revealed that differences in age acceleration are present in samples collected exclusively within the first 6 hours of hospital admission, as well as in samples from within the first 24 hours (Table S1).

• Download figure
• Download PowerPoint

In the replication cohort (N=981), we observed the same trends as in the discovery cohort. Chronological age at IS onset was different in women than men patients (women=72.2±13 years, men=72.8±13 years, P=5.71×10^-16), whereas DNAm age calculated by Horvath and Hannum methods were the same (Horvath: women=68.13±11.3 years, men=69.53±11.7 years, P=0.163; Hannum: women=60.11±10 years, men=61.25±10.2 years, P=0.165) (Figure 3-A). Results from Hannum-EEAA measures were replicated, women=-0.57, men=0.79, nominal P=0.02) (Figure 3-B). For the other acceleration measures, we found the same direction of effects as that observed in the Discovery cohort; however, the results did not reach statistical significance. (Table 3, Figure 3)

• Download figure
• Download PowerPoint

Sex Dichotomy of Age Acceleration in Stroke Subtypes

Next, we investigated the relationship between EAA measures in sexual dichotomy in age at onset of IS subtypes in the discovery cohort (N=374). The ANOVA test revealed differences in chronological age at stroke onset among IS subgroups (P=7.58×10^-6), but not in DNAm age (Horvath-DNAmAge, P=0.65; Hannum-DNAmAge, P=0.10) (Table S2). When interrogating each individual TOAST subtype, we identified that women had negative EAA values and men positive values in 2 subtypes, cardioembolic stroke (CE) and undetermined stroke (Und). Differences that reached statistical significance in most cases. For CE, Horvath-EAA, P=1.52×10^-2; Horvath-IEAA, P=2.95×10^-3, and Hannum-EAA P=5.08×10^-4, Hannum-IEAA P=1.80×10^-3 and Hannum-EEAA P=1.50×10^-5. For undetermined IS, Horvath-EAA P=4.08×10^-2, Hannum-EAA P=1.20×10^-2, and Hannum-IEAA P=4.80×10^-2. In large artery atherosclerosis, both women and men had positive acceleration values, except for Hannum-IEAA which were both negative. For small vessel disease, both women and men had negative acceleration values. There were no statistical differences in acceleration between sexes in either large artery atherosclerosis or small vessel disease (Table S2).

Proteomic Study of Age Acceleration

Finally, for the 26 IS samples studied via SOMAScan, we interrogated all EAA measures for each individual protein level (1305). We observed that 42 protein levels correlated with Hannum-EEAA (P<0.05), although none of them surpassed correction by multiple comparison (Table S3). Gene Ontological analysis on nonredundant biological processes showed an over-representation on 21 pathways (Table S3). The ERK (extracellular signal-regulated kinase) 1 and ERK1 cascade pathway was the only one surpassing multiple comparison (P=2.164×10^-3, false discovery rate Q value=0.016) (Table S4 and S5).

Discussion

Data from the FHS (Framingham Heart Study), a long-term longitudinal cohort study aimed to unravel the underlying causes of cardiovascular diseases, suggest that stroke incidence declines over time. The age-adjusted incidence of stroke per 1000 person-years has reduced by 2.3 points in men and 1.1 points in women.³⁸ The lifetime risk of a first stroke at 65 has decreased 5 points in men and 1.9 points in women.³⁸ Although encouraging, this data indicates that the decline in stroke incidence is not equal among men and women, the GCNKSS (Greater Cincinnati/Northern Kentucky Stroke Study) also observed a significant decline in stroke incidence in men but not in women.³⁹ Hence, the importance of finding specific biological mechanisms implicated in stroke risk in women. Targeted prevention strategies, whether lifestyle or pharmacological, focused on the biological aging processes unique to women, are key.

In this study, we interrogated EAA measures, including Horvath-EAA, Horvath-IEAA, Hannum-EAA, Hannum-IEAA, and Hannum-EEAA in men and women who have had an IS event. Women, consistently, showed a significant decrease in age acceleration, at a nominal level, compared to men. Furthermore, the same differences were only observed in CE and Und subtypes.

Previous studies have identified DNAm-based biomarkers of aging (DNAmAge) as a better estimate of stroke risk, outcome, and recurrence than chronological age.^26,40–42 Soriano-Tárraga et al²⁶ in a cohort of age-matched individuals (N=123) reported that IS patients were biologically older than controls. In the same study, they observed a statistically significant association between age acceleration in IS patients but not in controls. Authors described age acceleration as the average difference between DNAm age and chronological age (delta-age), whereas, here, we use a more robust method based on the residuals resulting from regressing DNAmAge on chronological age. Nevertheless, the result indicates a tendency for age acceleration in stroke patients in comparison to control subjects.

As expected, women's average chronological age at stroke onset was older than men, but no differences were found in DNAmAge. This suggests that at the time of the first IS, both men and women, have the same biological status. We observed that the age equalization was due to a shift in the biological age of women. Women show a negative age acceleration, they are, on average, younger than their chronological age. Whereas men show a positive age acceleration.

We observed decreased age acceleration measures (Horvath-EAA and IEAA and Hannum-EAA and EEAA) in women, but not in men. These differences, although present in the replication cohort, did not reach statistical significance, except for Hannum-EEAA. We think that, despite women having decreased EAA measures and men positive ones, and delta values between both groups similar to the ones present in the discovery cohort, the large SD of the measures have limited the ability to reach statistical significance.

IEAA and EEAA apply only to blood samples.²⁰ IEAA measures indicate cell-intrinsic aging independent of age-related changes in blood cell counts, whereas EEAA, which is based on Hannum algorithm, is associated with age-dependent changes in blood cell counts and indicates immune system aging.^20,43 EEAA is more strongly related with lifestyle and socioeconomic factors (diet, body mass index, education, income, etc) than IEAA.^20,43 This suggests that sexual dichotomy of age at stroke onset might be influenced by a combination of intrinsic biological process and lifestyle factors.

A biological process specifically captured in blood samples is the menopause. EAA has been associated with a younger age at menopause in a 3 cohort metanalysis,²⁵ but not in saliva and buccal epithelium samples. Thus, blood samples are specially equipped to capture the specific aging process that occurs in women undergoing menopause. Specifically, authors observed that menopause accelerates epigenetic aging. This acceleration experienced by postmenopausal women could be one of the reasons why, in the older age range, there is an increased incidence of strokes in women.

A 2014 study identified that a 1% decrease in the global methylation levels of long interspersed nucleotide element 1, a member of nonlong terminal repeat retrotransposons, associated with a 1.2-fold increase in stroke risk in men, but not in women in a Chinese population.⁴⁴ Long interspersed nucleotide element 1 was a popular way to interrogate global DNAm levels before the popularization of DNAm arrays. Although this does not represent a biological age calculation, it is an indication of potential sex differences in stroke patients regarding DNA methylation levels, denoting DNAm as a new and relevant field of study for sex dichotomy in strokes.

The same behavior observed for all IS was observed in cardioembolic and Und subtypes. Women with CE and Und subtypes present a negative age acceleration and men present a positive one. CE, showed an increasing temporal trend in patients of white ethnicity, from 1993 to 2015.⁴⁵ Authors attributed those changes, among others, to population aging, thus increasing the risk of factors contributing to atrial fibrillation (AF). In fact, cardioembolic patients are the oldest among all the subtypes in our cohort. As AF is the leading cause of CE, perhaps, we are capturing a biological signature related to AF. Thériault et al⁴⁶ identified a biological age proxy, from gene expression profiles in blood, associated with the presence of AF mainly with permanent AF which is better at discriminating AF presence than traditional risk factors. As far as we know, there are no studies in the association of DNAm biological age and AF, but previous studies have observed a relationship between biological age measures and the risk of cardiovascular death.⁴⁷

A total of 26 samples of the discovery cohort were previously analyzed via SOMAScan, allowing us to perform a small-scale functional assessment of DNAmAge and acceleration measures. We identified 42 proteins whose blood levels correlated with Hannum-EEAA measures. Although, these results did not reach statistical significance after multiple comparison, likely due to the low number of samples studied, this protein-set has an over-representation of proteins involved in the ERK1 and ERK2 cascade. The ERK1 and ERK2 cascade is a central signaling pathway that regulates a broad range of cellular processes, such as proliferation, differentiation, survival, as well as stress response and apoptosis. An in vitro work observed that higher levels of estrogen in females protect them against aging via up-regulating antioxidant and longevity-related genes, through the activation of the ERK1 and ERK2 cascade.⁴⁸ The possible implications of our results require a more in-depth study starting with a larger sample to be confirmed.

In further studies, it would be interesting to study longitudinal changes in DNAmAge in women before and after menopause and at the time of stroke onset, to determine how age acceleration relates to stroke risk after menopause.

Limitations

GRECOS, SEDMAN, and BASICMAR projects are prospective studies, which recruited patients from a relatively small area, including mostly patients of European descent. The homogeneity of our sample could present a source of bias and limit the generalization of our results to other population groups. The estimated Horvath and Hannum-DNAm ages are lower than the chronological age, differing from what has been observed in previous studies.²⁶ We attribute the difference to a lower number of CpGs used in the calculations, due to quality control removal. Despite that, both measures were consistent with the estimated biological age and, more importantly, the lack of differences between sexes. IEAA and EEAA measures are dependent on both EAA values and Biological Age values, which limits our ability to perform correction for multiple comparisons. Nonetheless, we have performed a replication effort in a larger independent cohort to validate the observed results of the discovery cohort.

Conclusions

This is the first study looking at the differences between DNAmAge and EAA between men and women in strokes. Here, we observed that women present decreased values of EAA, indicating that their biological age is lower than expected at the time of IS onset, contrary to the increased EAA of men. These differences seem to be specific to CE and Undetermined stroke subtypes, although studies with bigger sample sizes should be performed to confirm these results. This study provides initial evidence that sex-specific underlying biological mechanisms are captured by biological age estimates. Understanding the sex-related complexities of IS is a must in proper patient management and care.

Article Information

Acknowledgments

We thank the Spanish Stroke Genetics Consortium and the Redes Temáticas de Investigación Cooperativa en Salud (RETICS) Network INVICTUS (RD16/0019/0002, RD16/0019/0010, RD16/0019/0011, RD16/0019/0021)

Supplemental Material

Tables S1–S5

STROBE checklist

Footnotes