Study population and study design
For this study, baseline data of the longitudinal, multicentre Research on Obesity and Diabetes among African Migrants (RODAM) study were used. Details on the RODAM study have been published before [14] and will be briefly described here.
In the period 2012–2015, 6385 Ghanaian men and women were recruited. The participants resided in rural Ghana, urban Ghana, and in the European cities of London, Amsterdam, and Berlin. The majority of participants were of Akan ethnicity, and Ghanaians in Europe were first-generation migrants.
Of participants aged 25 years and older with complete data on physical examination and blood samples profile, 736 participants were selected for DNA methylation profiling. These participants were selected based on a case-control design, including about 300 cases with non-drug-treated diabetes, 300 controls without diabetes, and 135 controls with neither diabetes nor obesity. After excluding participants with sex discordances (n = 11), duplicates (n = 8), and those not meeting the quality control thresholds (n = 12), 713 eligible participants remained. Of these participants, those without aldosterone and renin measurements were excluded (n = 644, excluding all participants residing in London and Berlin as they were not included in aldosterone and renin analysis), as was one participant with an outlier in renin concentration (77 pg/mL), which was > 3 times the SD of the log-transformed renin concentration. This resulted in 68 participants in the current analysis, all residing in rural and urban Ghana, or Amsterdam (Additional file 1: FigureS1).
Phenotypic measurements
Data collection procedures for questionnaire and physical examination were highly standardised across the different study locations. Questionnaires were used to collect data on gender, age, level of education, and length of stay in Europe. Physical examination was performed using validated devices. Weight was measured in light clothing without shoes with a SECA 877 scale to the nearest 0.1 kg. Height was measured without shoes using a SECA 2017 portable stadiometer to the nearest 0.1 cm. Anthropometric measures were taken twice and the mean of the two measures was used in analyses. Body mass index (BMI) was calculated by dividing the weight in kilograms by the height in metres squared. After at least 5 min of rest, three blood pressure (BP) readings were taken in a sitting position, with a cuff fixed on the left arm. The mean of the second and third reading was used in the analyses. Hypertension was defined as having a BP of systolic ≥ 140 mmHg and/or diastolic ≥ 90 mmHg and/or the use of BP-lowering medication. BP-lowering medication was categorised based on Anatomical Therapeutic Chemical classification of medication that participants brought with them to the research location.
Venous blood samples were collected in the sitting position, after an overnight fast of at least ten hours. All samples were collected around the same time of the day (morning) to control for the influence of circadian rhythm. After samples collection, samples were immediately processed, aliquoted and cryopreserved, before storage in − 80˚C freezers. Samples collected in Ghana were shipped to Europe while kept frozen at − 80 ˚C. Fasting plasma glucose concentrations were measured using the hexokinase method by colorimetry, in the laboratory of the Institute of Tropical Medicine and International Health, Berlin, Germany. Diabetes mellitus was defined according to self-reported diabetes and/or fasting glucose ≥ 7.0 mmol/L. Aldosterone and renin concentration (pg/mL) were measured in heparin plasma samples, and analyses were performed at the Department of Internal Medicine of the Erasmus MC, Rotterdam, the Netherlands. All samples, including those from Ghana, were analysed in the same laboratory, to prevent inter-laboratory differences affecting the results. After thawing of the samples, renin concentration was determined by an immunoradiometric assay (Cisbio, Saclay, France) using an active site-directed radiolabelled antibody binding to renin only. The lower detection limit of this assay was 2 pg/mL. Aldosterone concentrations were measured by solid-phase radioimmunoassay (Demeditec Diagnostics, Kiel, Germany), with a lower detection limit of 12 pg/ml. Aldosterone-to-renin ratio (ARR, pg/mL/pg/mL) was calculated by dividing aldosterone by the renin concentration. The distributions of aldosterone, renin, and ARR were assessed using histograms and the Shapiro-Wilkinson test. To ensure normal distribution of the traits, aldosterone concentration was transformed using Box-Cox transformation, and renin and ARR were log-transformed. Renin, aldosterone and ARR were chosen for analysis, because of their relevance in the context of SS and salt-sensitive hypertension.
DNA methylation profiling, processing, and quality control
Previous RODAM publications have elaborated upon the DNA methylation profiling, processing, and quality control on whole blood samples [15], and this process will be summarised here. In the lab of Source BioScience, Nottingham, UK, the Zymo EZ DNAm™ kit was used for bisulfite conversion of DNA. Using the Infinium® HumanMethylation450 BeadChip, the converted DNA was amplified and hybridised, hereby quantifying DNAm levels of approximately 485,000 CpG sites. Methylation levels were measured based on the intensities of the methylated and unmethylated probes for each CpG site on the array. These intensities were expressed as methylation Beta values, which is a value between zero (unmethylated) and one (methylated). A log2 ratio of the intensities of methylated versus unmethylated probes was calculated, which is referred to as M values. Using R statistical software (version 4.2.0), quality control was performed using the MethylAid package (version 1.30.0), using default thresholds of 10.5 for methylated and unmethylated intensities, 11.75 for overall quality control, 12.75 for bisulfite conversion, 12.50 for hybridisation control, and 0.95 for detection p-value. Functional normalisation of the raw 450 K data was conducted using the minfi package (version 1.42.0). After the removal of probes annotated to the X and Y chromosomes, known to involve cross-hybridisation or to contain common single-nucleotide polymorphisms (SNPs) with a minor allele frequency of ≥ 5%, a set of 429,449 CpG sites remained for analysis [16]. Blood cell type distribution was estimated based on the method of Houseman et al. [17].
Statistical analyses
Association between renin, aldosterone, and ARR and DNA methylation
To assess differentially methylated positions (DMPs), multivariate linear regression was performed between renin concentration, aldosterone, and ARR (independent variables) and DNA methylation M values (dependent variable), using the Limma package (version 3.52.0). Methylation M values were used in statistical analysis to ensure normal distribution, whereas methylation Beta values were used to facilitate interpretation and visualisation. Models were adjusted for sex, age, BMI, diabetes mellitus, hypertension, estimated blood cell type counts (CD8 + T lymphocytes, CD4 + T lymphocytes, natural killer cells, B cells, monocytes, granulocytes), and technical covariates (hybridisation batch and array position), because of correlation with DNA methylation in the principal components analysis, as well as because of an overrepresentation of participants with diabetes and high BMI in the sample. Model fit was assessed using QQ plots, as well as the genomic inflation factor lambda. Because of improvement in model fit after bias and inflation correction using the R package bacon (version 1.24.0) [18], we applied this adjustment to all analyses (lambda with inflation correction for renin 1.046, aldosterone 0.995, ARR 1.026) (Additional file 1: FigureS2). False discovery rate (FDR) adjustment of the p-values was applied, to reduce the risk of false positive findings in multiple testing. An FDR-adjusted p-value of < 0.05 was considered epigenome-wide significant.
To identify the contribution of the top DMPs to the variance in renin and aldosterone concentration, linear regression was performed using z-standardised methylation M values of the identified DMPs as independent variable and the (untransformed) trait of interest as dependent variable. The R squared statistics from the linear regression analyses with and without covariates were used to calculate trait variance explained by each DMP. Similar methods were used to assess the explained variance in systolic and diastolic BP, with the z-standardised M values of the top DMPs associated with renin, aldosterone, and ARR as independent variables in the linear regression model, adjusted for age, sex, BMI, diabetes mellitus, blood cell distribution, and technical variates.
As several types of BP-lowering medication can interfere with the RAAS system, a sensitivity analysis excluding those on BP-lowering medication (n = 10) was performed, to assess the impact of medication use on the association.
For the top DMPs per trait, we extracted the median methylation Beta values with accompanying interquartile ranges, and stratified these per geographical location, to examine whether differences in methylation levels existed between participants residing in rural and urban Ghana, and Amsterdam. These median values were compared between the location using nonparametric the Kruskal–Wallis test. A two-sided p-value < 0.05 was considered statistically significant.
Differentially methylated regions
To assess whether DNA methylation of genomic regions, rather than on single positions, was associated with the traits of interest, the R package bumphunter (version 1.38.0) was used to assess differentially methylated regions (DMRs), using similar models as used in the DMP analysis. Methylation M value cut-offs of 0.2 was chosen for the input, which limited the analysis to 100 candidate regions and 20% difference in regression coefficient beta between candidate probes. The analysis was run with bootstrapping with 500 permutations. DMRs with more than two CpGs, and a family-wise error rate (FWER) < 0.2 were considered statistically significant. DMRs were visualised using coMET package (version 1.27.2).
Replication
We used the EWAS atlas [19] to extract all CpGs previously reported to be associated with BP, systolic blood pressure (SBP), diastolic blood pressure (DBP), and hypertension and performed a candidate DMP analysis. Additionally, we tried to replicate CpG sites significantly associated with eGFR in a large, multi-ethnic, meta-analysis of EWAS [20]. Lastly, we extracted the probes previously annotated to RAAS-related genes (ACE, AGT, REN, CYP11B2, HDS11B2, and NR3C2) [21] and performed a candidate-gene DMP analysis on these for each of our traits of interest. For these candidate-gene analyses, the same models, including adjustment for bias and inflation, were used as for the main analysis. All CpGs with an FDR-adjusted p-value of < 0.05 were considered to be statistically significant.
Biological relevance
Gene expression levels for the epigenome-wide significant DMPs were assessed using the publicly available iMethyl database, in which the DNA methylation levels of CD4 + T lymphocytes as well as gene expression levels, denoted per kilo base of transcript per million mapped fragments (FPKM), are reported for 100 apparently healthy individuals living in Japan [22].
Enrichment analysis using the Kyoto Encyclopedia of Genes and Genomes (KEGG) catalogue in package R missMethyl (version 1.30.0) was performed to gain insight into the function and biological pathways of our findings. The top 5000 CpGs with the smallest p-values per trait were used as input.
The gaphunter gfunction of the minfi package was used to examine whether the significant DMPs were potentially under the influence of genetic variation. The function was run with a threshold of 0.05, reflecting a gap of 5% in beta value, suggestive of genetic influence.