At each public data collection event, eligible participants, in their preferred language (ie, English or Amharic) will receive a full verbal explanation of the study (ie, information about the study, eligibility criteria, compensation, benefits and risks to study participation, steps to ensure data confidentiality, and right to withdraw from study at any time) and written informed consent will be obtained prior to participation. Participants can ask questions about the study to a study team member, and if they agree to participate, sign the informed consent document (Station 1). Proceeding through the remaining stations, participants will complete verbally administered questionnaires (stress measures, health status, and demographics), provide biological samples (ie, finger prick point of care HbA_1c and saliva for DNA), and complete screening tests for obesity, hypertension, diabetes, depression, and anxiety (Table 1).

Data are anonymous, and each participant will receive a unique ID number, which is used on all measures. All study data and tracking will be entered into the REDCap (Research Electronic Data Capture; Vanderbilt University) database, a secure research electronic data management system that provides an interface for validated data entry, audit trails for tracking data manipulation, and export procedures. REDCap is compliant with the HIPAA (Health Insurance Portability and Accountability Act) and satisfies all local, state, and federal regulations for the capture and storage of private health information for research purposes.

Participants who express any distress during verbal administration of questionnaires, those who indicate high levels of stress, and those with abnormal or elevated clinical measures will be referred, and the on-call physician will determine the need for emergency room evaluation. Participants with an abnormal HbA_1c result (≥5.7), elevated BMI (≥25), and/or elevated BP (≥120/80 mm Hg) will be counseled on healthy lifestyle modifications and the need for medical evaluation with a primary care provider. Participants will then be provided with contact information for local primary care clinics. If a participant has a screening test that requires urgent or emergent evaluation (HbA_1c >9, random blood glucose of ≥ 400, and/or BP ≥180/120 mm Hg), the community coordinator will contact the on-call physician for consultation. The on-call physician will speak with the participant and determine the need for emergency room evaluation. Participants who complete the data collection event will receive a US $75 gift card.

To capture both the unique and joint contributions of stress burden for Ethiopian American immigrants, we will measure cumulative stress burden in a robust, multidimensional way [58] which includes validated stress scales assessing immigration-related stress, post-migration stress, and psychosocial perceived stress (recent and chronic stress) in combination with a saliva specimen (biological measure of stress) and qualitative semistructured interviews to capture additional missed experiences of stress and buffers to stress. The scales and interviews will be verbally administered by the trained community research coordinators. We will work with the CAB and a professional translator to translate scales and the interview guide into Amharic. We will work with the CAB to ensure that all measures are culturally appropriate for Ethiopian American immigrant.

Immigration-related stress will be measured using the 21-item Stress of Immigration Survey (SOIS), which assesses 5 domains of stress: language, immigrant status, work issues, yearning for family and home country, and cultural dissonance among immigrants living in the United States [59]. The SOIS has robust psychometric properties with immigrant samples. Post migration, psychosocial, and perceived stress will be assessed using Boen’s stress exposure measure [20] which considers stress and adversity in 6 domains of social life: lifetime traumas; recent stressful life events (last 5 y); financial strain; ongoing chronic strains (eg, health, work, relationship, and housing); everyday discrimination; and major lifetime discrimination. The individual measures have been previously used to assess the association between psychosocial stress and a variety of mental and physical health outcomes (eg, depressive symptoms, chronic illnesses, and physical limitations) [21]. Following the same procedure used by Boen and Hummer [20] and Cuevas et al [27], we will create 2 measures of cumulative exposure across these stress domains: (1) high risk stress burden (numbers of stressors for which the respondent is in the highest quartile and (2) composite total stress exposure across domains of stress (no cutoffs). The 10-item Perceived Stress Scale (PSS-10) is a 10-item scale that is used to assess the degree of perceived stress a person experiences [60]. The items relate to how unpredictable, uncontrollable, and overloaded respondents find their lives in general and do not focus on a specific event. The PSS-10 has been used in multiple languages and has well-established psychometric properties, including in Ethiopian university students [61-64]. A biological measure of stress will be collected with saliva samples or oral-rinse specimens obtained by trained study team members. Nonfasting saliva will be collected from each participant following oral rinsing with tap water. Participants will expectorate into an Oragene-RNA RE-100 Self-Collection Kit (DNA Genotek), and samples will be stored at room temperature for further analysis [65]. Buccal cells for DNA extraction will be collected using DNA Genotek saliva collection kits [66]. Saliva has been used as a noninvasive and stress-free diagnostic alternative to blood to examine associations of stress and resilience factors with DNAm by our groups and others [37,67,68].

Estimates of perceived and actual health status will be obtained using self-reported measures of health status and screening tests for obesity, diabetes, hypertension (high BP), depression, and anxiety. Perceived health status will be assessed with the Patient-Reported Outcomes Measurement Information System (PROMIS) Global v1.2 scale [69-71], a self-reported measure of global, physical, mental, and social health. Participants’ BMI will be measured and used to categorize participants’ weight status. Screening for hypertension will be conducted using measurements of systolic and diastolic BP. BP measurements will be used to categorize participants into categories (normal, elevated, stage 1 hypertension, and stage 2 hypertension). Screening for diabetes will be conducted using a point-of-care capillary blood glucose test (HbA_1c), which will be used to categorize participants (normal blood sugar, prediabetes, or diabetes). Depression and anxiety symptoms will be measured using the 8-item Patient Health Questionnaire (PHQ-8) and the 7-item Generalized Anxiety Disorder (GAD-7) assessments, both widely used and well-validated in different settings and population groups [72-75]. PHQ-8 and GAD-7 have been validated in Ethiopia using different cutoff points (5 or above on the PHQ-8; 10 or above on the GAD-7) [76-78].

To explore the nuances of Ethiopian American immigrant stress burden, trained members of the team will conduct qualitative semistructured interviews [79] with a subset of Aim 1 Ethiopian American immigrants who vary in levels of reported stress on the above measures. These interviews will be conducted to ensure we understand Ethiopian American immigrant experiences with stress that may be missed from standardized measures and to explore potential buffers to the adverse effects of stress (Multimedia Appendix 1 for interview guide reviewed and pilot tested with members of the CAB). We will contact participants by phone, text, or WhatsApp (Meta; depending on preference indicated during the public data collection events) and schedule interested individuals for a ~60-minute web-conference (eg, Zoom and Microsoft Teams) interview. We will use maximal variation sampling [57] to identify potential interviewees of diverse ages, genders, and varied immigration length and status. Data collection will continue until saturation, or informational redundancy [80-82] of major categories and themes is achieved. Specifically, we will operationalize saturation, for the purpose of this study, as the point at which no new substantive information is gained from 3 consecutive interviews. We will conduct a minimum of 10 interviews in each geographic area (Cincinnati and Columbus) and expect that we will reach data saturation within 20-25 total interviews based on our prior experience in similar qualitative projects with immigrant populations.

We will use a genome-wide approach (Illumina EPIC arrays) to identify Cytosine-phosphate-Guanine (CpG) sites associated with stress and chronic illnesses (hypertension, obesity, cardiovascular disease, and diabetes). One-microgram DNA from saliva samples will be treated with bisulfite and submitted to the DNA Sequencing and Genotyping Core for DNAm quantification. The genome-wide DNAm analysis will be performed using the Infinium Human Methylation 850 BeadChip “860K array” (Illumina), [93] which covers 99% RefSeq genes, 333,265 CpG sites located in enhancer regions identified by the ENCODE (Encyclopedia of DNA Elements), 5’ untranslated region (5’UTR), exon, gene, and 3’ untranslated region (3’UTR) regions [94]. The 850K array has recently become an attractive choice for DNAm studies due to its effective cost and high coverage of promoter regions [95]. Samples will be randomly assigned to avoid batch effects due to chips and plates. The MethylationEPIC BeadChip images will be captured using the Illumina iScan. Samples with high-quality bisulfite conversion efficiency (intensity signal >4000) that also pass all GenomeStudio quality control steps based on built-in control probes for staining, hybridization, extension, and specificity will be filtered. Individual probes will be filtered away based on mean detection P>.01. The raw DNAm data will be exported from GenomeStudio and analyzed using Bioconductor and the lumi package.

Briefly, signal intensities of samples that pass quality control will be background-adjusted using out-of-band probes (noob) and normalized using subset-quantile within array normalization (“swan”). Beta values calculated as signal_methylation/(signal_methylation+signal_{unmethylation}) will be used as the measure for DNAm levels, while m values will be used in the analysis. Since DNAm is collected from saliva, and it may be affected by cell composition [96] and cell dilution, we will correct for cell composition using Houseman reference-based deconvolution algorithms (R package “minfi”). We will subsequently control for possible batch effects, as well as hidden structure in DNAm data, using surrogate variable analysis. To test association of DNAm with stress, we will use linear regression models (R limma package), [97] where DNAm level of a CpG locus will be the dependent variable, stress factors in aim 1 are the independent variables and the following covariates will be adjusted: significant surrogate variables from surrogate variable analysis, estimated proportions of saliva cell types, immigration duration, age, and sex, as well as a random effect for recruitment site. Differentially methylated CpG sites with methylation differences of at least 5% and P<.0001 will be identified.

Because exposure to cumulative stress most likely affects methylation of a variety of genes, single CpG site analysis may fail to fully capture the epigenetic pathways impacted by stress. We plan to use weighted gene coexpression network analysis (WGCNA) to identify methylation modules (ie, groups of genes with highly correlated methylation values and that may represent biological pathways) and then relate these modules to stress profiles. We apply the coexpression approach because differentially expressed genes may fail to fully capture the molecular pathways, as genes do not function in isolation [98]. Briefly, WGCNA takes the methylation correlation matrix and transforms it into a weighted undirected network by raising the absolute value of each pair-wise correlation to a power β. The higher the value of β, the more weight is given to high correlations at the expense of low correlations [99]. This weighting scheme preserves the continuous nature of the methylation data and produces highly robust results. Modules are then identified using average linkage hierarchical clustering. In addition, the use of network approaches reduces the burden of multiple testing, as we will test the association of a few hundred modules, representing biological pathways, with the stressors of interest, rather than test hundreds of thousands of methylation sites. TM and LD have conducted WGCNA-related experiments in the past [98,100].

We will calculate for each module the eigengene E that is defined as the first principal component of the methylation values of a given module. The epigenome can be considered as representative of the methylation profiles in a module. The association between the identified modules and cumulative stress can then be evaluated using linear models. Analyses will be adjusted for principal components of methylation variation (calculated from the epigenome-wide methylation data), estimated proportions of saliva cell types, other covariates (eg, immigration duration, age, and sex), and random recruitment site effect. We will also examine what genes and/or CpG sites have a greater contribution to the stressor-module association by looking into their weights on the module eigengene.

Our mediation analysis aims to explore the relationship between the independent variable (stress, X), the mediator (DNAm, M), and the dependent variable (disease-related outcome, Y) [102-104]. The test for mediation will be accomplished with a 3-step model. We will include covariates (age, sex, and immigration duration) in all of the models. However, for simplicity, we will omit them in the following equations: Step 1: logit(Pr(Y=1, x)) = ɑ₁ + C’X, testing for path C’ (total effect) when X predicting Y; Step 2: M = ɑ₂ + AX + ε₂, testing for path A when X predicting M; and Step 3: logit(Pr(Y=1|x, m))= ɑ₃ + BM + CX, multiple regression analysis with X and M predicting Y. The direct effect (path C) is the effect of stress on health outcome, adjusting for the mediator (methylation modules). The mediation effect is the indirect effect (path A*B) [105]. We will estimate the average causal mediation effects [106].

Power analysis is based on the association between DNAm module eigengenes and cumulative stress burden. We expect to analyze about 100 normalized module eigengenes from 100 methylation modules using genome-wide methylation data. With a sample size of 150, for a multiple linear regression model which already includes ~6 covariates with a squared multiple correlation ρ² of 0.1, a sample size of 150 will have 80% power to detect at a significance level of 3.125×10^–5 (=0.05/100 methylation modules × 16 stress subscales) an increase in ρ² of 0.136 due to including 1 additional stress subscale.