Investigation of psychosocial stress and DNA methylation in genes related to immune response, metabolism, and calcium signaling among African American and White middle-aged adults

Abstract

Chronic psychological stress is an environmental factor associated with chronic disease risk and health disparities. Since environmental stressors alter gene expression and physiologic responses through epigenetic mechanisms, perceived discrimination (PD), or the subjective experience of receiving negative treatment related to personal characteristics may influence DNA methylation (DNAm) of CpGs within genes linked to chronic disease. Using the Illumina 850K EPIC chip and psychosocial stress-associated discrimination scales, including the lifetime, racial, and everyday discrimination scales, we identified novel CpGs and differentially methylated positions (DMPs) associated with PD in the context of age, sex, and poverty status among African American and White adults and ones that overlap previous findings of differentially methylated sites (or genes) with discrimination, inflammation, and chronic disease. Ingenuity Pathway Analysis identified several pathways associated with the DNAm patterns and PD with age, sex, and/or poverty status. With age, the white adipose tissue browning pathway was activated among African American participants. This was related to the differential methylation found in the CACNA1H, PRDM16, and BDNF genes. Among White participants, the opioid signaling pathway was activated and significantly enriched for the genes FGR, POMC, and RPS6KA2. Additionally, CpG sites associated with PD among White participants with poverty status revealed that the netrin signaling, opioid signaling, and calcium signaling pathways were activated and significantly enriched for differentially methylated genes, including NFATC1 and NFATC2. The identified novel DNAm genes associated with PD transduce effects through biological pathways related to inflammation and immune response, white adipose tissue browning, and calcium signaling.

Introduction

Psychosocial stress adversely affects health and health outcomes. These stressors cause biological and physiological responses that are thought to contribute to health disparities and the accelerated aging phenotype [1]. Recent data suggest that perceived discrimination (PD) is a psychosocial stressor that contributes to health disparities in age-associated chronic diseases. For example, discrimination has been associated with cardiovascular disease risk factors [2, 3], metabolism [4], cognition [5, 6], and brain health [7]. Despite accumulating evidence that discrimination adversely affects mental and physical health, the underlying biological mechanisms that drive these effects are largely unknown. One possible biologic transduction pathway considered has been genome-based. There have been some genome-wide association studies and related genetic studies examining individual experiences of psychosocial stress. There is evidence in the literature demonstrating that psychosocial stress is associated with alterations in DNA methylation (DNAm) status [8, 9] and the effects of genetic variants around single-nucleotide polymorphisms on clinical outcomes. Epigenetic mechanisms may be one biologic pathway through which psychosocial stressor discrimination may cause physiological changes. These mechanisms can alter transcriptional programs without changing DNA sequence and are influenced by environmental, lifestyle, and psychosocial factors [10–12]. Of the epigenetic mechanisms, DNAm is the most well-studied epigenetic modification and occurs at CpG sites in the genome. There is substantial evidence that DNAm is influenced by environmental exposures, including lead, diet, and polychlorinated biphenyls, and psychosocial stressors such as maternal deprivation, exposure to early life adversity, and violence (for review, see [13, 14]). Few studies have examined the effect of discrimination on DNAm. In African American youth, experiencing high levels of perceived racial discrimination with a less supportive family environment led to accelerated epigenetic aging [15]. On the contrary, exposure to perceived racial discrimination coupled with a strong supportive family environment mitigated these effects on epigenetic aging [15]. Major life discrimination was associated with nine different DNAm CpG sites in African American mothers enrolled in the Intergenerational Impact of Genetic and Psychological Factors on Blood Pressure (InterGEN) study [16]. Discrimination was also associated with DNAm of select genes in pregnant Latina women and in Ghanaian migrants [17, 18]. These studies suggest a link between discrimination and DNAm. Nevertheless, there is still much to learn about the different experiences of discrimination and their influence in both African American and White adults.

The psychosocial stressor discrimination is multidimensional in nature, via race or gender [19] and can be experienced both short term (everyday) and long term (lifetime) [20]. Different forms of PD have different health-related consequences given the different timescales and nature of the experiences, acute versus chronic exposure. Work from the Forde laboratory compared the risk of developing hypertension among African Americans who reported exposure to discrimination on both the everyday and lifetime discrimination scales. Only the lifetime discrimination scale predicted the risk of developing hypertension [21] . Measures of psychological stress are influenced by endogenous and exogenous factors. These factors may include exogenous environmental-based exposures like housing, occupation, social support, and educational opportunities, as well as endogenous factors like mental health, health behaviors, and coping measures. These factors operate and interact at many levels. Therefore, in this study, we primarily explored the impacts of interpersonal or individual-level PD as measured by the lifetime discrimination scale on DNAm in a cohort of African American and White adults. Secondarily, we explored the relationship of everyday, racial, and gender discrimination with DNAm in this cohort. Here, we identified novel CpG sites and investigated previously reported CpG sites associated with psychosocial stress. We also evaluated the differentially methylated signatures associated with psychosocial stress versus the differentially methylated signatures associated with chronic diseases. Assessing a multitude of discrimination measures in relation to DNAm can provide a broader picture of the impact and complexity of this psychosocial stressor on the epigenome and whether these changes overlap with those observed in chronic disease.

Results

Study baseline characteristics

A cohort of 467 participants had both DNAm and complete psychosocial stress scale data. Of these, 51% were African American adults and 49% were White adults (Table S1 panel A). A fraction of both African American and White participants reported experiencing lifetime or racial discrimination. Both African American (50.0%) and White (49.0%) participants (n = 228) reported (Ldisc+) (Table S1 panel B). African American participants more frequently reported Ldisc compared to White participants (32.7% and 16%, respectively; P-value < .0001). Analysis of Rdisc revealed that 36.7% (n = 172) reported experiencing racial discrimination (Rdisc⁺) (Table S1 panel C). While both African American and White participants reported experiencing racial discrimination, more African American participants reported Rdisc compared to White participants (28.0% to 8.7%, respectively; P-value < .001). Therefore, in this cohort, African Americans more frequently reported Ldisc and Rdisc forms of PD, but it is also notable that White participants reported both lifetime and racial discrimination. There were no significant differences between African American and White participants in age, race, sex, or poverty status for the sources of discrimination (Sdisc), gender (Gdisc), everyday (Edisc), and coping with discrimination (Cdisc) scales (data not shown).

Race stratified age, sex, and poverty status adjusted analysis of DNA methylation and lifetime discrimination with age, sex, and poverty status

Among African American participants, there were 60 CpG sites significantly methylated in the age-adjusted analysis (Model 1). The top 10 hypermethylated CpG sites shown in Fig. 1A (red bars) are associated with the following genes: PRDM16, GFI1, NOD1, PALM, PITX2, and CD300A. The top 10 hypomethylated CpG sites shown in Fig. 1A (green bars) are associated with the following genes: ARHGAP26, DTHD1, NFATC2, FAM160B1, ZSWIM6, MBP, PTPRN2, PTPRE, DYSF, and SNCA. The most differentially methylated CpG sites are listed in Table 1 (top panel).

Figure 1.

Significant differentially DNA-methylated CpG sites for lifetime discrimination among African American and White participants with demographic variables: (A) age, (B) sex, (C) poverty status, and (D) age, sex, and poverty status. Green bars represent CpG sites that are hypomethylated; red bars represent CpG sites that are hypermethylated. Panels (A)–(C) show the top 10 significantly hyper- and hypomethylated CpG sites; panel (D) shows all the significant hyper- and hypomethylated CpG sites.

Table 1.

Most significantly differentially methylated CpG sites associated with lifetime discrimination with age in HANDLS (P-value < .05).

African Americans
Gene symbol	CpG site	β-value	Function
MAD1L1	cg16393715	−0.00092	Polymorphisms are associated with schizophrenia susceptibility [68]
CD300A	cg14181940	0.03483	Modulate immune cell functions [69]
PTPRN2	cg07783094	−0.00858	Associated with childhood obesity [70]
RERE	cg11072627	−0.00043	Role in neurodevelopmental disorder [71]
ARHGAP26	cg21591141	−0.01976	Polymorphisms are associated with Alzheimer’s disease and cardiovascular disease [72]
FAM160B1	cg12417938	−0.01737	Associated with behavioral abnormalities [73]
PRKAR1B	cg25001831	−0.00772	Variants cause neurodevelopmental disorders [74]
GFI1	cg04158367	0.00583	Role in the epigenetic regulation of acute myeloid leukemia [75]
PRDM16	cg00087546	0.00243	Involved in age dependent cardiac hypertrophy, adverse remodeling, mitochondrial dysfunction, and heart failure [76]
ICA1	cg22313239	−0.00382	Variations are associated with type 1 diabetes [77]
Whites
Gene symbol	CpG site	β-value	Function
ARHGAP26	cg19456812	0.00262	Associated with, cancer, Alzheimer’s disease, and cardiovascular disease [32, 72]
CORO2B	cg08290628	−0.01148	Role in diabetic nephropathy [78]
PTPRN2	cg11202265	0.00255	Associated with childhood obesity [70]
MCC	cg02951292	−0.00624	Associated with the predisposition of colon cancer [79]
RERE	cg16958658	0.00216	Role in neurodevelopmental disorder [71]
C7orf50	cg04194173	−0.0095	Role in type 2 diabetes [80]
RPS6KA2	cg20669159	−0.01045	Role in prostate cancer [81]
POU6F2	cg12423935	−0.01181	Role in cancers [82]
SBF2	cg27156515	−0.01055	Role in glioblastoma [83]
PRKAR1B	cg26461009	−0.00591	Variants cause neurodevelopmental disorders [74]

Among White participants, 94 CpG sites were significantly methylated in the age-adjusted analysis. The top 10 hypermethylated CpG sites shown in Fig. 1A-second column (red bars) are associated with the following genes: SORCS1, CD44, NR3C1, PITX2, AHRR, PRDM16, ZXDC, PTPRN2, PTPRE, DYSF, and SNCA. The most differentially methylated CpG sites are listed in Table 1 (top panel).

Among White participants, 94 CpG sites were significantly methylated in the age-adjusted analysis. The top 10 hypermethylated CpG sites shown in Fig. 1A-second column (red bars) are associated with the following genes: SORCS1, CD44, NR3C1, PITX2, AHRR, PRDM16, ZXDC, PTPRN2, HOXD3, and ATP9A. The green bars in Fig. 1A (second column) show the top 10 hypomethylated CpG sites that are associated with the following genes: MUC5B, AHRR, FAM49B, NAALADL2, TOMM20, JARID2, CLYBL, DSE, SMG6, and POU6F2. The most differentially methylated CpGs are listed in Table 1 (bottom panel).

In the sex-adjusted analysis (Model 2), we identified 48 significantly methylated CpG sites. Among African American individuals, the top 10 hypermethylated CpG sites as shown in Fig. 1B (red bars) are associated with the following genes: PRDM16, FGR, MYO5C, GFI1, SLC7A11, PTPRN2, RUNX1, and ZNF423. The top 10 hypomethylated CpG sites shown by the green bars in Fig. 1B are associated with genes NPAS3, PTPRN2, RSPO3, FAM160B1, PTPRE, TSC2, and RASIP1. The most significantly differentially methylated CpGs are listed in Table 2 (top panel).

Table 2.

Most significantly differentially methylated CpG sites associated with lifetime discrimination associated with sex in HANDLS (P-value < .05).

African Americans
Gene symbol	CpG site	β-value	Function
CORO2B	cg12390780	0.00335	Role in diabetic nephropathy [78]
ERICH1	cg11857858	−0.0037	DNA methylation patterns are associated with obesity [84]
MAD1L1	cg25259697	−0.0089	Polymorphisms are associated with schizophrenia susceptibility [68]
PTPRN2	cg26308704	−0.02372	Associated with childhood obesity [70]
PARD3	cg06832449	−0.00292	Role in cancer [85]
NPR1	cg07299819	−0.00172	Associated with renal function [86]
SLC7A11	cg06690548	0.00889	Plays a key role in tumor growth [87]
FAM160B1	cg12417938	−0.01737	Associated with behavioral abnormalities [73]
PRKAR1B	cg22994830	−0.01086	Variants cause neurodevelopmental disorders [74]
GFI1	cg04158367	0.00583	Role in the epigenetic regulation of acute myeloid leukemia [75]
Whites
Gene symbol	CpG site	β-value	Function
WWOX	cg01256026	−0.01122	Loss of function in neurodevelopmental and neurodegenerative disorders [88]
CORO2B	cg08290628	−0.01148	Role in diabetic nephropathy [78]
MCC	cg15391499	0.00159	Associated with the predisposition of colon cancer [79]
MAD1L1	cg10196995	−0.01095	Polymorphisms are associated with schizophrenia susceptibility [68]
PTPRN2	cg00369443	−0.01032	Associated with childhood obesity [70]
RERE	cg01162877	−0.00435	Role in neurodevelopmental disorder [71]
PARD3	cg27298509	−0.00769	Role in cancer [85]
C7orf50	cg26568662	−0.00492	Role in type 2 diabetes [80]
ERICH1	cg16040935	−0.01123	DNA methylation patterns are associated with obesity [84]
RPS6KA2	cg13906646	−0.00441	Role in prostate cancer [81]

There were 112 CpG sites significantly methylated with Ldisc among White participants in this analysis. The top 10 hypermethylated CpG sites in Fig. 1B second column (red bars) are associated with genes: CDC42BPB, SBF2, PACS1, MFHAS1, CTBP2, NIFK-AS1, JPH2, ROD1, and ATP9A. The top 10 hypomethylated CpG sites (green bars) are in the following genes: RASIP1, TRAF1, MTMR2, PTPRN2, SHANK3, PTPRE, and CLYBL. The most significantly differentially methylated CpGs are listed in Table 2 (bottom panel).

This poverty-adjusted analysis (Model 3) identified 28 significantly methylated CpG sites for Ldisc, including poverty status in the model. Among African American participants, the top 10 hypermethylated CpG sites showed in Fig. 1C (red bars) are associated with the following genes: RUNX1, CD300A, AHNAK, MLLT1, and GPR146. The top 10 hypomethylated CpG sites (green bars) are associated with the following genes: MAD1L1, DTHD1, CAPN13, PTPRN2, TPK1, PTPRE, and ERICH1. The most significantly differentially methylated CpGs are listed in Table 3 (top panel).

Table 3.

Most significantly differentially methylated CpG sites associated with lifetime discrimination associated with poverty status in HANDLS (P-value < .05).

African Americans
Gene symbol	CpG site	β-value	Function
AHNAK	cg00685014	0.02155	Associated with poor outcome of pancreatic ductal adenocarcinoma prognosis [89]
MAD1L1	cg14004847	−0.03723	Polymorphisms are associated with schizophrenia susceptibility [68]
CD300A	cg05592035	0.04584	Modulate immune cell functions [69]
PTPRN2	cg00638020	−0.01412	Associated with childhood obesity [70]
ERICH1	cg22843129	−0.01487	DNA methylation patterns are associated with obesity [84]
GPR146	cg01414857	0.02571	Associated with altered DNA methylation patterns among the African American patients with colorectal neoplasia [90]
POU6F2	cg19704451	−0.00657	Role in cancers [82]
TPK1	cg07811915	−0.01813	Predictive marker for the anti-tumor effects in gastric cancer [91]
NOP58	cg17930737	0.00667	Prognostic marker in hepatocellular carcinoma [92]
DTHD1	cg08798147	−0.0303	Polymorphisms confer risk for hepatocellular carcinoma [92, 93]
Whites
Gene symbol	CpG site	β-value	Function
WWOX	cg01256026	−0.01122	Loss of function in neurodevelopmental and neurodegenerative disorders [88]
SLC12A7	cg08293408	−0.00551	Role in adrenocortical carcinoma [94]
MAD1L1	cg21998840	−0.00454	Polymorphisms are associated with schizophrenia susceptibility [68]
PTPRN2	cg20030024	−0.01075	Associated with childhood obesity [70]
RERE	cg08548328	−0.00682	Role in neurodevelopmental disorder [71]
SORCS2	cg02447115	−0.00968	Required for social memory [95]
SCARB1	cg23460943	−0.01011	Inherited variants cause early-onset coronary artery disease [96]
PARD3	cg27188994	0.01468	Role in cancer [85]
C7orf50	cg01778643	−0.0085	Role in type 2 diabetes [80]
ERICH1	cg21740055	−0.00896	DNA methylation patterns are associated with obesity [84]

Among White participants, we found 141 significantly methylated CpG sites for Ldisc in the poverty-adjusted model. The top 10 significantly hypermethylated CpG sites shown in Fig. 1C-second column (red bars) are associated with the following genes: SMG6, PARD6G, NEDD4L, BCL3, AHRR, PRKAR1B, SEMA7A, and PARD3. The top 10 most hypomethylated CpGs (green bars) in this group are associated with these genes: WWOX, LINC00578, PARD3, SHANK3, ARHGAP22, JPH2, PACS1, PTPRN2, CD8A, and CLYBL. The most differentially methylated CpGs are listed in Table 3 (bottom panel).

The age-, sex-, and poverty status-adjusted model (Model 4) analysis among African American study participants identified one hypermethylated CpG site, cg22029284 (red bar), in the PRDM16 gene and one significantly hypomethylated CpG site, cg24750513 (green bar), in the gene BIN1 (Fig. 1D) and Table 4 (top panel). The analysis among White participants (Fig. 1D, second column) identified two significantly hypermethylated CpG sites cg12608507 and cg19456812 (red bars) and Table 4 (bottom panel). These CpG sites are in the genes EEF1B2 and ARHGAP26, respectively. There was only one significantly hypomethylated CpG site cg00139947 (green bar), which is in the gene TOP1MT.

Table 4.

Significantly differentially methylated CpG sites associated with lifetime discrimination with age, sex, and poverty status in HANDLS (P-value < .05).

African Americans
Gene symbol	CpG site	β-value	Function
BIN1	cg24750513	−0.000261857	Susceptibility gene for Alzheimer’s disease [28]
PRDM16	cg22029284	0.003540734	Involved in age dependent cardiac hypertrophy, adverse remodeling, mitochondrial dysfunction, and heart failure [76]
Whites
Gene symbol	CpG site	β-value	Function
ARHGAP26	cg19456812	0.002621218	Associated with cancer, Alzheimer’s disease, and cardiovascular disease [32, 72]
EEF1B2	cg12608507	0.002309914	Role in cancer, neurodevelopmental disorders [97, 98]
TOP1MT	cg00139947	−0.005440972	Role in cancer, early-onset Parkinson disease [33, 99]

Analysis of CpGs and candidate genes previously associated with PD

Previously published work by others has identified differentially methylated CpG sites and genes that are associated with PD and with various conditions associated with racial and ethnic health disparities, including blood pressure level, diabetes mellitus, inflammation, and renal function [22–26]. To further validate our findings, we examined whether these previously published differentially methylated genes associated with PD overlapped with genes and their respective CpG sites in our data set [16, 18] (Figs 2–4 and Table S2). This analysis was performed using age, sex, and poverty status as covariates separately in African American and White participants.

Figure 2.

Venn analysis identifying overlaps between genes significantly methylated with lifetime discrimination in HANDLS and previously reported to be significantly methylated with PD and conditions associated with health disparities with age: (A) PD—Venn diagram showing overlap between CpGs/genes differentially methylated in HANDLS and previously reported PD-related genes. (B) Renal function decline—Venn diagram showing overlap between genes differentially methylated in HANDLS with PD and those previously reported renal function decline-related genes. (C) Blood pressure—Venn diagram showing overlap between genes differentially methylated in HANDLS with PD and previously reported blood pressure-related genes. (D) Diabetes—Venn diagram showing overlap between genes differentially methylated in HANDLS with PD and previously reported diabetes mellitus-related genes. (E) Inflammation—Venn diagram showing overlap genes differentially methylated in HANDLS with PD and previously reported inflammation-related genes. *Only the top five methylated genes were represented. ^#Gene with more than one CpG site differentially methylated.

Figure 4.

Venn analysis identifying overlaps between CpG sites/genes significantly methylated with lifetime discrimination in HANDLS and previously reported significantly methylated with PD and conditions associated with health disparities associated with poverty status: (A) PD—Venn diagram showing the overlap between CpGs/genes differentially methylated in HANDLS and previously reported PD-related genes. (B) Renal function decline—Venn diagram showing CpGs differentially methylated with discrimination in HANDLS and previously reported for renal function decline. (C) Blood pressure—Venn diagram showing the overlap between CpGs/genes differentially methylated in HANDLS with discrimination and previously reported blood pressure-related genes. (D) Diabetes—Venn diagram showing the overlap between CpGs/genes differentially methylated in HANDLS with discrimination and previously reported diabetes mellitus-related genes. (E) Inflammation—Venn diagram showing overlap CpGs/genes differentially methylated in HANDLS with PD and previously reported inflammation-related genes. *Only the top five methylated genes were represented. ^#Gene with more than one CpG site.

First, we identified CpG sites that were significantly methylated in our dataset that overlapped with the CpGs previously identified in the published literature as associated with PD. One CpG site, cg16393715 in the gene MAD1L1, that was previously reported to be differentially methylated with PD was significantly associated with the lifetime discrimination instrument in the age-adjusted model among African American participants (Fig. 2 panel A, red circle). For sex and lifetime discrimination in African American participants, we found that CpG sites cg07485685, cg25259697, and cg05411165 were significantly methylated (Fig. 3 panel A, red circle). Two of these CpGs were in the gene ABR, and the other was in the gene ABLIM2. For poverty status and lifetime discrimination among African American participants, CpG sites cg12734526, cg14004847, cg03575939, cg09742952, and cg21089453 were significantly methylated (Fig. 4 panel A, red circle) and associated with genes MAD1L1 and TMEFF1. There was more than one CpG site associated with the MAD1L1 gene. Therefore, among African American participants, we reported overlap with CpG sites that have been previously associated with PD.

Figure 3.

Venn analysis identifying overlaps between genes significantly methylated with lifetime discrimination in HANDLS and previously reported significantly methylated with PD and conditions associated with health disparities associated with sex: (A) PD—Venn diagram showing the overlap between CpGs/genes differentially methylated in HANDLS and previously reported discrimination-related genes. (B) Renal function decline—Venn diagram showing CpGs differentially methylated with discrimination in HANDLS and previously reported for renal functional decline. (C) Blood pressure—Venn diagram showing the overlap between CpGs/genes differentially methylated with discrimination in HANDLS and previously reported blood pressure-related genes. (D) Diabetes—Venn diagram showing the overlap between CpGs/genes differentially methylated in HANDLS with discrimination and previously reported diabetes mellitus-related genes. (E) Inflammation—Venn diagram showing overlap CpGs/genes differentially methylated in HANDLS with discrimination and previously reported inflammation-related genes. *Only the top five methylated genes represented. ^#Gene with more than one CpG site differentially methylated.

Next, we examined whether there was overlap of differentially methylated CpG sites previously reported with PD with Ldisc in White HANDLS participants. We found that the CpG sites cg08818984, cg07733851, cg03239861, cg04498286, cg11897120, cg18944144, and cg22070462 were significantly methylated with age (Fig. 2 panel A, blue circle). These CpG sites were associated with the two genes SORCS1 and ZXDC. For sex, the analysis showed significant methylation at CpG sites cg01256026, cg09396019, cg10196995, cg23846046, cg00008629, cg02349373, cg12727138, cg18683606, and cg19247475 (Fig. 3 panel A, blue circle). These CpGs are associated with several genes displayed in the box adjacent to the blue circle. There were two CpGs associated with the SRRT gene. The analysis with poverty status revealed that CpG sites cg01256026, cg03943446, cg07300192, cg08599635, cg18353661, cg21998840, cg24297873, cg00219598, cg01805469, cg06502071, cg08293408, cg21942646, and cg22391718 were significantly methylated among White participants (Fig. 4 panel A, blue circle) and associated with six genes. One gene, UGP2, had more than one CpG site. Interestingly, the Venn diagram showed that these significantly differentially methylated CpG sites are race-specific with no significant overlap between African American and White participants for previously reported PD genes with age, sex, or poverty status (Fig. 2 panel A, Fig. 3 panel A, and Fig. 4 panel A).

Analysis of candidate genes previously associated with renal function decline

Examining overlap between previous findings of differentially methylated genes and CpGs with renal function decline [26] in HANDLS participants and lifetime discrimination showed more differentially methylated CpG sites among White compared to African American participants. Among White participants, we identified three CpG sites that remained significantly associated with the lifetime discrimination instrument after adjustment for age (Fig. 2 panel B, blue circle). There were no significant CpG sites identified among African American participants (Fig. 2 panel B, red circle). With sex (Fig. 3 panel B, red circle), among African American participants, we found one CpG site differentially methylated in the gene UCP2. In White participants, we found five CpG sites associated with the four genes listed in the box adjacent to the blue circle (Fig. 3 panel B). The gene LPCAT1 had more than CpG site. With poverty status, four CpG sites were significantly methylated among the White participants (Fig. 4 panel B, blue circle). For the African American participants, with poverty status, there was no significantly methylated CpG site (Fig. 4B, red circle).

Analysis of candidate genes previously associated with blood pressure

Using our candidate gene approach for blood pressure-related genes [22, 24], we examined the overlap in our data set with blood pressure-related genes. We found that among African American participants, CpG sites cg05531134 and cg26915870 were significantly methylated with age and lifetime discrimination (Fig. 2C, red circle). The genes associated with these CpGs are associated with the genes listed in the adjacent box. Among White participants, we found significant methylation at the CpG sites cg02079326, cg04661001, cg13993469, cg18729298, cg23484461, and cg27491887 with age (Fig. 2C, blue circle). These CpGs are associated with the four genes listed in the adjacent box. Two of these genes were associated with more than one CpG site. There was only one CpG site significantly methylated with sex, cg06914593, in African American participants (Fig. 3C, red circle). The associated gene is listed in the adjacent box. There were no CpGs among White participants. Both cg07354679 and cg17084941 were significantly methylated with poverty status among African American study participants (Fig. 4C, red circle). Three CpG sites were significantly methylated with poverty status among White participants, cg04106390, cg14826184, and cg16663063 (Fig. 4C, blue circle). The associated genes are listed in the adjacent boxes. One gene CACNA1H, calcium voltage-gated channel subunit alpha1H, is significantly differentially methylated in both African American and White participants with blood pressure and poverty status. However, the differentially methylated CpGs do not overlap between African American and White participants.

Analysis of candidate genes previously associated with diabetes mellitus

Similarly, using our lifetime discrimination data, we performed a gene candidate approach for previously reported diabetes-related genes [27]. We identified several overlaps between CpG sites associated with discrimination and previously published diabetes-related genes and CpG sites. This analysis among the African American participants revealed 46 significantly methylated CpG sites associated with age (Fig. 2D, red circle). The associated genes are listed in the adjacent box. There were 66 significantly methylated CpG sites associated with age among White participants (Fig. 2D, blue circle) with age. The associated genes are listed in the adjacent box. We identified 42 significantly methylated CpG sites associated with sex (Fig. 3D) in African American participants and 87 CpG sites associated with sex (Fig. 3D) in White participants. There was one CpG site that overlapped between African American and White participants (Fig. 3D). It is associated with the gene PTPRE, protein tyrosine phosphatase receptor type E, which is involved in signaling molecules that regulate cell growth, differentiation, mitosis, and cellular transformation.

In this analysis, there were 22 significantly methylated CpG sites for diabetes-related genes that were also found to be associated with poverty status among African Americans (Fig. 4D, red circle). Among White participants, we identified 115 CpG sites associated with poverty status that overlapped with CpG sites in diabetes-related genes.

The age-, sex-, and poverty status-adjusted analyses for lifetime discrimination using CpG sites and genes associated with diabetes mellitus identified only two significantly methylated CpG sites, cg22029284 (PRDM16) and cg24750513 (BIN1) among African American study participants that were also differentially methylated with diabetes mellitus. There were only three significantly methylated CpG sites, cg00139947 (TOP1MT), cg12608507 (EEF1B2, SNORA41, SNORD51, NDUFS1), and cg19456812 (ARHGAP26) associated with m among White participants (Table S28H). We did not identify CpGs that overlapped with age + sex + poverty status for the other diseases or conditions examined in our study.

Analysis of candidate genes previously associated with chronic low-grade inflammation

When we examined previously reported candidate genes associated with inflammation [25], we identified among African American participants, five significantly methylated CpG sites associated with age and lifetime discrimination (Fig. 2E, red circle). There were 13 associated with age among White participants (Fig. 2E, blue circle); one of these CpGs overlapped with African American participants. This CpG is associated with the gene islet cell autoantigen 1, which is associated with insulin-dependent diabetes and autoimmune disease. We noted four significantly associated with sex among African American participants (Fig. 3E, red circle). We found 16 significantly methylated CpGs associated with sex among White participants (Fig. 3E, blue circle), but none of these overlapped with those found among African American participants. There was only one CpG site cg17930737 associated with poverty status for African American participants (Fig. 4E, red circle). Among White participants, we found 19 significantly methylated CpG sites, but none of these overlapped with the CpG site among African American participants (Fig. 4E, blue circle). Overall, our results in the HANDLS cohort identified a significant number of CpG sites that were differentially methylated with PD as measured by Ldisc that overlap with differentially CpG sites and genes suspected to play a role in conditions associated with health disparities.

Functional analysis of the CpG sites significantly methylated with PD as measured by Ldisc

We performed Ingenuity Pathway Analysis (IPA) to examine which molecular and cellular pathways may be influenced by the CpG methylation patterns associated with PD with age, sex, poverty status, age + sex, and age + sex + poverty status. We imported the genes annotated to the significantly differentially methylated CpG sites into IPA for functional enrichment and pathway analysis. With age and lifetime discrimination, among African American participants, the functional analysis revealed that the white adipose tissue browning pathway was activated (Fig. 5A). This was related to the differential methylation found in the BDNF, CACNA1H, PRDM16,and PRKAR1B genes among African Americans with age (Fig. 5A). The IPA analysis revealed that for White participants, for CpG sites associated with PD with age, the opioid signaling pathway was activated and significantly enriched for the genes FGR, POMC, RPS6KA2, and PRKAR1B (Fig. 5B).

Figure 5.

IPA of significantly methylated CpGs associated with lifetime discrimination among African American and White participants. (A) Top canonical pathways for the age-associated CpG sites associated with PD in African American participants. (B) Top canonical pathways for the age-associated CpG sites associated with PD in White participants. (C) Top canonical pathways for the poverty status-associated CpG sites associated with PD in White participants. Blue bars represent the pathways that are inactivated (Z-score <−2). The P-value for each pathway is represented by the bar and is expressed as a −log (P-value).

Results from the functional analysis for the CpG sites associated with PD among White participants with poverty status revealed that netrin signaling, opioid signaling, and calcium signaling pathways were the top pathways and that there was a significant enriched for the following genes: ABLIM2, CACNA1H, NFATC1, NFATC2, PRKAR1B, AP2A2, CACNA1H, POMC, PRKAR1B, RPS6KA2, CACNA1H, NFATC1, NFATC2, and PRKAR1B (Fig. 5C). The analyses for sex, age + sex, and age + sex + poverty status revealed no significant pathways among African American or White participants.

Identification of differentially methylated CpGs associated with other forms of PD and age, race, sex, and poverty status

The age-adjusted analyses revealed 93 CpG sites were significant for Sdisc (Fig. S1 panel A and Table S3), 6 CpG sites were significant for Ldisc with age (Fig. S1 panel B and Table S4), 4 CpG sites were significant for Gdisc (Fig. S1 panel C, Table S5), and 261 CpG sites were significant for Rdisc (Fig. S1 panel D, Table S6) and no significant CpG sites for Edisc (Fig. S1 panel E) or Cdisc (Fig. S1 panel F). No significant CpG sites were found for any of the discrimination scales in race-adjusted analyses (Fig. S2 panels A–F).

The sex-adjusted analyses revealed 84 CpG sites significant for Sdisc (Fig. S3 panel A, Table S7), 7 CpG sites for Ldisc (Fig. S3 panel B, Table S8), 5 CpG sites for Gdisc (Fig. S3 panel C, Table S9), and 283 CpG sites for Rdisc (Fig. S3 panel D, Table S10), but no significant CpG sites for Edisc (Fig. S3 panel E) and Cdisc (Fig. S3 panel F).

The poverty-adjusted analyses revealed 87 CpG sites significant for Sdisc (Fig. S4 panel A, Table S11), 14 CpG sites for Ldisc (Fig. S4 panel B, Table S12), 5 CpG sites for Gdisc (Fig. S4 panel C, Table S13), and 327 CpG sites for Rdisc (Fig. S4 panel D, Table S14), but no significant CpG sites for Edisc (Fig. S4 panel E) or Cdisc (Fig. S4 panel F).

Results from the multivariable linear regression models adjusted for both age and sex showed that 97 CpG sites were significant for Sdisc (Fig. S5A, Table S15), 6 CpG sites were significant for Ldisc (Fig. S5 panel B, Table S16), 5 CpG sites were significant for Gdisc (Fig. S5 panel C, Table S17), and 283 CpG sites were significant for Rdisc (Fig. S5 panel D, Table S18), but no CpG sites for Edisc (Fig. S5 panel E) and Cdisc (Fig. S5 panel F).

The age, sex, and poverty status analyses identified 104 CpG sites significant for Sdisc (Fig. S6 panel A, Table S19), 20 CpG sites for Ldisc (Fig. S6 panel B, Table S20), 5 CpG sites for Gdisc (Fig. S6 panel C, Table S21), and 361 CpG sites for Rdisc (Fig. S6 panel D, Table S22), but no CpG sites significant for the Edisc (Fig. S6 panel E) and Cdisc (Fig. S6 panel F).

We examined the overlap of differentially methylated CpG sites that were significant in the different discrimination scales. This analysis showed that there was no overlap in CpG sites between discrimination scales in our age, sex, age + sex, and age + sex + poverty status analyses (Fig. S7 panels A, B, D, and E). We found that one CpG site cg04118610, located in the Adhesion G Protein-Coupled Receptor L3 (LPHN3) gene, was significant in all the discrimination scales for the poverty-adjusted model (Fig. S7 panel C).

Differentially methylated positions associated with lifetime and racial discrimination with age, sex, and poverty status

Next, we used the dmpFinder function for the two discrimination scales, Ldisc and Rdisc, to identify the differentially methylated positions (DMPs) associated with each scale. This analysis revealed a significant number of CpGs associated with lifetime and racial discrimination, with age + sex + poverty status at P < .05. To select the most robustly associated DMPs, we used a P < .01 cut-off. The more stringent DMP analysis demonstrated that among the 16 771 DMPs that were significant at a P < .01 for Ldisc, 4356 (25.9%) CpG sites were hypermethylated. The top hypermethylated CpGs based on β-values and associated genes included TMEM154-Transmembrane Protein 154 (cg16203592), EIF3A-Eukaryotic Translation Initiation Factor 3 subunit A (cg24623109), IQ motif containing-IQCK (cg03615426), and Copper Chaperone for Superoxide Dismutase CCS (cg24851651). In addition, there were 12 415 (74.0%) CpG sites that were hypomethylated. The top hypomethylated genes were LOC101927924-long intergenic non-protein coding RNA 1856 (cg07929412), SLN-Sarcolipin (cg20221740), GPR1-AS-Chemerin chemokine-like receptor 2 (cg22444562), TMEM63B-Transmembrane protein 63B (cg25069157), LPHN3 Adhesion G protein-coupled receptor L3 (cg04118610), OR2T5-Olfactory Receptor family 2 subfamily T member 5 (cg27191320), WWOX-WW domain containing oxidoreductase (cg02171206), PAWR-Pro-apoptotic WT1 regulator (cg11258982), and TMEM9B-Transmembrane Protein 9 domain family member B (cg15570860).

For Rdisc, we found that among the 33 990 DMPs that were significant, 15 046 (44.2%) CpG sites were hypermethylated. The most hypermethylated genes were TMEM154 (cg16203592), LOC101928708 (cg03146160), EIF3A (cg24623109), CORO2B-coronin, actin-binding protein, 2B (cg15917410), and IQCK (cg03615426). There were 18 944 (55.7%) hypomethylated CpG sites. The most hypomethylated genes included SLN (cg20221740), TMEM9B (cg15570860), OR2T5 (cg27191320), WWOX (cg02171206), and PAWR (cg11258982).

Gene network analysis regulatory effects

To better understand the regulatory effects of the pathways identified in our IPA analysis, we performed gene network analysis. We focused on the genes that were significantly enriched in the pathway analysis at a P-value < .05 (Fig. 5A–C). For the white adipose tissue browning pathway, we focused on BDNF (brain-derived neurotrophic factor), CACNA1H (calcium voltage-gated channel subunit alpha1H), PRDM16 (PR/SET domain 16), and PRKAR1B (protein kinase cAMP-dependent type I regulatory subunit beta) because they were differentially methylated. Our results showed that for the age-associated Ldisc CpG sites in African American participants, downregulation of the genes BDNF (Fig. 6A), CACNA1H (Fig. 6B), and PRDM16 (Fig. 6C) was predicted to activate nitric oxide. Although the PRKAR1B gene was differentially methylated with age in African American participants, it did not predict the activation of any gene or molecule in the white adipose tissue browning pathway.

Figure 6.

Gene network analysis showing the regulatory effects in the white adipose tissue browning and opioid signaling pathways: (A) BDNF,(B) CACNA1H, and (C) PRDM16 genes in the white adipose tissue browning pathway among African American participants and (D) POMC in the opioid signaling pathway among White participants. The relationship among the genes is represented by solid lines representing direct associations and dotted lines representing indirect associations. The darker shaded shapes indicate a higher predicted inhibition, while a lighter shaded shape indicates a lower predicted inhibition.

With age, the differential methylation of the Pro-opiomelanocortin (POMC) gene (Fig. 6D) was associated with activation of the opioid signaling pathway in White participants. With poverty status among White participants, the opioid signaling pathway was enriched for the following genes: FGR-proto-oncogene (FGR), ribosomal protein S6 kinase A2(RPS6KA2), adaptor-related protein complex 2 subunit alpha 2 (AP2A2), PRKAR1B, and CACNA1H (Fig. S8). Downregulation of the genes FGR (Fig. S8A) and RPS6KA2 (Fig. S8B) was predicted to activate the genes BCL2-associated agonist of cell death (BAD), glycogen synthase 3 beta (GSK3B), ETS transcription factor ELK1 (ELK1), and AKT serine/threonine kinase 1 (AKT1). We found that downregulation of the gene CACNA1H (Fig. S8E) was predicted to activate adenosine triphosphate. Genes AP2A2 (Fig. S8C) and PRKAR1B (Fig. S8D) on the other hand did not predict the activation of any gene or molecule in the opioid signaling pathway.

Our results showed that for the poverty status-associated Ldisc CpG sites in White participants, downregulation of the genes NFATC1 (Fig. S9A) and NFATC2 (Fig. S9C) was predicted to activate adenosine triphosphate and MAPK3 (mitogen-activated protein kinase 3) gene and to inhibit the genes PPP3R1 (protein phosphatase 3 regulatory subunit B, alpha), RCAN1 (regulator of calcineurin 1), MYH1 (myosin heavy chain 1), HDAC1 (histone deacetylase 1), NFATC3 (nuclear factor of activated T cells 3), and MEF2D (myocyte enhancer factor 2D) in the calcium signaling pathway. For the netrin signaling pathway, our results showed that downregulation of the genes NFATC1 (Fig. S9B) and NFATC2 (Fig. S9D) did not predict the activation of any gene or molecule that inhibited the genes PPP3R1 and NFATC3.

Overall, our results demonstrate that Ldisc adjusted for age and poverty is associated with white adipose tissue browning, opioid signaling, netrin, and calcium signaling pathways through differential methylation of pathway-enriched genes. Psychosocial stress may have a role in important signaling pathways through differential DNAm in both African American and White adults.

Discussion

Our data support the idea that differential DNAm is a relevant biological pathway through which psychosocial stress “gets under the skin.” We identified novel, significantly differentially methylated CpGs and DMPs associated with a psychosocial stressor, PD, among African American and White adults. Many CpGs identified are involved in pathways that play fundamental roles in various chronic disease health disparities. Using our data set, we identified overlap with previously reported CpGs that were differentially methylated with exposure to PD or the presence of diseases and conditions associated with health disparities. This suggests that psychosocial stress as a toxic exposure may influence DNAm of CpGs that play roles in susceptibility to chronic disease. Functional and regulatory gene network analysis of significantly differentially methylated CpG sites with PD identified three important pathways: the white adipose tissue browning, the opioid signaling, and the calcium signaling pathways. Gene network analysis identified that PD and age were associated with white adipose tissue browning in African American participants, and PD and poverty were associated with the opioid signaling pathway in White participants. These network analyses also identified the calcium signaling pathway among White participants living in poverty and reporting perceived psychosocial stress. The identification of these pathways brings to the forefront the importance of stress in the activation of molecular signaling mechanisms that influence health and disease.

The race-stratified EWAS analysis for Ldisc demonstrated that these variables are significantly associated with DNAm and revealed several novel genes and CpG sites associated with PD. Many of these genes as referenced in Tables 1–4 have functions related to neurodegenerative diseases, including Alzheimer’s disease, metabolic diseases, including diabetes mellitus types 1 and 2 and obesity, cancer, and neurodevelopmental disorders. The CpGs reported in Tables 1–4 are nominally significant findings (P < .05) and should be considered exploratory. The primary conclusions of the study are based on analyses that survived multiple-testing correction. The age-, sex-, and poverty-adjusted models were included primarily as sensitivity analyses to evaluate the stability of psychosocial stress-associated methylation signals under different covariate adjustments and to explore potential pathways through which social and demographic factors may influence these associations. Our primary interpretation focuses on the fully adjusted model, which provides the most conservative assessment of the relationship between perceived stress and DNAm. Findings from partially adjusted models should therefore be considered exploratory and hypothesis-generating pending validation in independent cohorts. The significantly differentially methylated CpG sites associated with lifetime discrimination and age + sex + poverty status differed between African American and White participants. In African American participants, the differentially methylated CpG site (cg24750513) is associated with the BIN1 gene, which produces the Bridging Integrator 1 protein. It is an important susceptibility gene for late-onset Alzheimer’s disease second only to APOE4 [28]. BIN1 may modulate APP metabolism and synaptic transmission as well as influence Tau protein function and neurotoxicity. In our study, the CpG within BIN1 was hypomethylated. It is interesting to note that hypomethylation of the BIN1 promoter was associated with preclinical AD [29]. Given the significant disproportionate burden of dementia among African Americans, this finding may deserve further investigation. The other differentially methylated CpG among African American participants is in the PRDM16 gene, which is involved in several molecular functions, including DNA-binding transcription activator activity and histone methyltransferase activity. It is also important in brown fat cell differentiation, energy homeostasis, as well as glucose and lipid metabolism (for review, see [30]). The differential methylation of these CpGs within these genes suggests that the pathway through which discrimination is transduced as a risk factor for obesity and cognitive decline involves the epigenetic alteration, DNAm, highlighting the importance of the epigenome in health disparities [5, 31].

Among White participants, three CpG sites were differentially methylated in the age + sex + poverty status adjusted analysis. These CpGs are in the following genes: RhoA GTPase-activating protein 26 (ARHGAP26), eukaryotic translation elongation factor 1β 2 (EEF1B2), and DNA topoisomerase l mitochondrial (TOP1MT). ARHGAP26 is involved in tumorigenesis as well as tumor promotion and invasion for selected cancers and as a biomarker of overall survival in others (for review, see [32]). EEF1B2, a subunit of eukaryotic translation elongation factor 1, plays a role in de novo protein synthesis. Mutations in EEF1B2 lead to defects in the translational balance of protein synthesis and risk of neurodevelopmental disorders, including intellectual impairment and disability. TOP1MT is a human mitochondrial topoisomerase involved in DNA replication, chromatin remodeling, genomic stability, and DNA metabolism (for review, see [33]). In the EWAS models, among White participants, discrimination was associated with differential methylation of CpGs in genes that are associated with cancer, genomic stability, and neurodevelopmental disorders.

To validate our hypothesis that psychosocial stress influences health outcomes through differential DNAm of CpGs located in disease-relevant genes, we examined the overlap between our data and previously reported differentially methylated CpG sites and genes associated with discrimination stress, blood pressure, diabetes, inflammation, and renal function decline. Our study replicates many of the CpGs previously reported to be differentially methylated with discrimination [16, 18]. In addition, we found that CpGs differentially methylated with discrimination were also differentially methylated in chronic conditions that have disparate health outcomes for African Americans and other marginalized US populations in age-, sex-, and poverty status-adjusted analyses. For example, we identified several overlapping CpGs associated with discrimination in our study with those previously published. Our results demonstrated a significant number of CpG sites that were differentially methylated in these diseases with age, sex, and poverty status in both African American and White participants. Interestingly, the CpGs and genes differentially methylated with discrimination in our study have notable overlap with genes/CpGs that play important roles in diseases and conditions that are critical areas of health disparities, including diabetes mellitus, cardiovascular disease, cancer, depression, end stage renal disease. Additionally, these overlaps were also present among White participants suggesting that discrimination is a health risk for African American as well as White participants. The IPA analysis based on the significantly methylated CpGs linked to discrimination exposure among the African American and White participants revealed that, in the age-adjusted analysis among African American participants, the white adipose tissue browning pathway was activated, while among White participants, the opioid signaling pathway was activated. The poverty status-adjusted analysis revealed an over-representation of the netrin signaling, opioid signaling, and calcium signaling pathways among White participants. These molecular pathways play critical roles in human health and disease and are particularly relevant to health disparities. White adipose tissue, which stores food calorie energy as triglycerides, provides thermal insulation, and mechanical protection, is an endocrine organ that plays a role in immunity and inflammation among other physiologic processes (for review, see [34]). Netrin-1 plays a role in multiple physiologic processes playing important roles in cardiovascular and renal disease as well as cancer (for review, see [35]). In cardiovascular disease, Netrin-1 serves as a central modulator of atherosclerosis. Plasma levels of netrin are negatively correlated with plaque density and the extent of arterial wall inflammation [36, 37].

Calcium is a vital second messenger in cells that is critical in many aspects of cell physiology and function, including cellular differentiation and proliferation, metabolism, apoptosis, as well as gene expression. Calcium signaling plays a role in cancer, neurodegeneration, including Alzheimer’s disease, cardiovascular disease, including hypertension and obesity, among other disease states. Opioid signaling functions in pain and pain modulation, and the neuroendocrine, reproductive, cardiovascular, and immune systems [38, 39, 40]. The netrin, opioid, and calcium signaling overrepresented in our results are known to be methylated in neurological disorders such as PTSD [41], opioid use disorder [42], and cancers, including breast cancer [43]. Furthermore, the regulatory effect analysis of these pathways in our study demonstrated that most of the methylated genes exert inhibitory effects in these pathways increasing the risk for unhealthy outcomes. For example, inhibition of the white adipose tissue browning pathway is associated with obesity and health risks associated with metabolic syndrome since the browning pathway reduces risk for obesity and metabolic syndrome [44, 45].

Our study has several notable strengths and limitations. Our cohort is diverse consisting of community-dwelling middle-aged African American and White participants. Our study considers the influence of potential covariates such as age, sex, and poverty status in discrimination studies. Our genome-wide DNAm analysis was performed using the latest version of EPIC compared to previous studies on discrimination, which used Illumina HumanMethylation450 array to quantify the methylation of CpG sites. Our study has limitations such as the non-inferring of causality between discrimination and DNAm. We have not included other social determinants of health that may influence the effects of discrimination on health outcomes nor have we done any mediation analysis. Future work in this area should model coping as a covariate or mediator that might attenuate the association of discrimination as a stressor with epigenetic outcomes. In addition, this is a cross-sectional study. The sample size in our study is relatively small for an EWAS focusing on conducting race-stratified analyses. The DMP analysis was designed as a complementary exploratory approach to support interpretation of the EWAS findings. Because DNAm is strongly influenced by demographic and socioeconomic variables, careful adjustment for age, sex, race, and poverty status was necessary to minimize confounding. We therefore focused our primary interpretation on covariate-adjusted EWAS findings and considered DMP results primarily as supportive evidence. Furthermore, nominal significance thresholds are insufficient for genome-wide inference and should be interpreted as hypothesis-generating rather than definitive evidence of association. Further validation of the biological relevance of identified CpG sites with pathways of white adipose tissue browning and opioid signaling, and additional functional studies are necessary. Additionally, smoking status was not included in the primary EWAS models because smoking may act as both a potential confounder and a behavioral consequence of chronic psychosocial stress. Adjusting for smoking could therefore remove part of the biological signal through which psychosocial stress affects health.

Conclusions

The main findings of this study are that perceived psychosocial stress is linked to consistent patterns of DNAm involving biological pathways related to immune function, metabolism, calcium signaling, and stress-responsive molecular networks. Although individual CpG sites varied across adjustment models and racial groups, several common themes appeared. First, only a small number of CpGs remained significant after full adjustment, indicating that many observed associations may be influenced by demographic and socioeconomic factors. Second, most psychosocial stress-related methylation signatures were race-specific, suggesting that the biological effects of chronic psychosocial stress might differ among population groups. Third, pathway analyses revealed convergence on a few biologically relevant networks, including white adipose tissue browning, opioid signaling, and calcium signaling pathways, despite limited overlap at the individual CpG level. Finally, the overlap between psychosocial stress-related loci and loci previously linked to diabetes, cardiovascular disease, inflammation, and renal disease suggests that epigenetic changes may be one mechanism by which psychosocial stress influences chronic disease risk and health disparities.

One of the major gaps in our understanding of age-related metabolic and chronic disease is the identification of the biologic transduction pathways through which environmental stressors and social factors influence the risk of disease. Our data suggest DNAm is likely one of the biological pathways involved. Identifying differential DNAm patterns and genes associated with PD in population groups highlights the biological effect of social adversities such as PD on health outcomes in White and African American adults. We have identified differential methylation in specific CpGs within genes whose functions are central to disease and conditions that have been linked to exposure to discrimination, dementia, cancer, cardiovascular disease, renal disease, diabetes, and obesity. We also identified important molecular pathways through which discrimination stress may increase disease risk providing potential targets for further research and interventions to improve the health outcomes of all Americans.

Materials and methods

The HANDLS cohort

The Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study is a population-based longitudinal study initiated in 2004 designed to disentangle the relationship between self-identified race and socioeconomic status in age-related health disparities. HANDLS is a fixed cohort of 3720 community-dwelling participants initially aged 30–64 years old who self-identified as either African American or White. The study includes participants recruited from 13 pre-determined neighborhoods in Baltimore, MD, USA using area probability sampling based on data from the US Census 2000 as described previously [46]. Participants were recruited using a four-way factorial cross of 5-year age groups, sex, race, and poverty status. Race was self-reported as either African American or White. Poverty status was determined by reported household income and categorized as above or below the 125% Federal Poverty Line as determined by the US Department of Health and Human Services in 2004.

HANDLS DNA methylation cohort

This cohort included participants who had blood DNA available from the initial visit during 2004–9. As previously described [47, 48], 508 participants were randomly sampled using a factorial design across race, sex, and poverty status. Ultimately, 487 had DNAm data that passed the quality control, of which 467 had discrimination scale data available.

Ethics

All participants of the HANDLS study provided written informed consent. The HANDLS study is approved by the National Institutes of Health Institutional Review Board. Protocol number: 09-AG-N248.

Perceived discrimination measures

We employed six psychosocial stress measures in the present study: (i) sources of discrimination (Sdisc) [49], a 10-item inventory assessing experiences of discrimination based on gender, race, ethnicity, income, age, religion, physical appearance, sexual orientation, health status, and disability using a four-point scale ranging from 1 (not at all) to 4 (a lot); (ii) lifetime discrimination burden (Ldisc) [50], a two-item inventory assessing overall lifetime difficulties experienced due to discrimination using a four-point scale ranging from 1 (not at all) to 4 (a lot); (iii) gender discrimination (Gdisc) [19], a five-item inventory assessing gender discrimination experienced at school, at work, seeking employment, searching for housing, receiving medical care, and by police and courts using a binary response (no vs. yes); (iv) racial discrimination (Rdisc) [19], a six-item inventory assessing racial discrimination experienced at school, at work, seeking employment, searching for housing, receiving medical care, and by police and courts using a binary response (no vs. yes); (v) everyday discrimination (Edisc) [51], a nine-item inventory assessing the frequency of daily experiences with bias unrelated to race or ethnicity using a six-point scale ranging from 1 (almost every day) to 6 (never); and (vi) coping with discrimination (Cdisc), a four-item inventory assessing discrimination coping strategies using a binary response (yes vs. no) [19, 52]. Further details concerning, these scales were published previously [7]. We applied the Shapiro–Wilk normality test [53, 54] to these scales using the R package mvShapiroTest (https://cran.rproject.org/web/packages/mvShapiroTest/index.html).

DNA methylation quantification

We used 250 ng of DNA isolated from blood from each participant sample, and DNA was treated with sodium bisulfate using a Zymo EZ-96 DNAm kit according to the manufacturer’s protocol (ZYMO Research, Orange, CA, USA). We followed the standard protocols provided by the manufacturer for all kits with the suggested input/elution volumes [55]. After the bisulfate treatment, we performed genome-wide DNAm analysis using Illumina Infinium HumanMethylation EPIC BeadChip, the latest version of EPIC that efficiently quantifies the methylation of 866 836 CpG sites on the human genome [56, 57]

Quality control and normalization

Respective IDAT files generated from the DNAm quantification were subjected to quality control and data preprocessing using the Bioconductor package minfi [58]. Data analysis of the samples that passed quality control was performed in R by integrating minfi [58], limma [59], missMethyl [60, 61], IlluminaHumanMethylationEPICanno.ilm10b2.hg19 [61], and IlluminaHumanMethylationEPICmanifest [62] using raw methylation values. Background correction, dye-bias equalization, and normalization of the β scores were performed using the normal-exponential out-of-band correction method [63] in the minfi package used to perform background correction, dye-bias equalization, and normalization of the β scores. Single nucleotide polymorphisms (SNPs) associated probes were removed using the function dropLociWithSnps in minfi [58], which drops the probes that contain a SNP at the CpG interrogation or the single nucleotide extension. Technical variabilities in methylation measurements and batch effects from the plate were removed using ComBat [64].

Association analysis of DNAm and psychosocial stress scales

To test the association between methylation at CpG sites across the genome and the different psychosocial stress scales (Sdisc, Rdisc, Ldisc, Gdisc, Edisc, Cdisc), we used the cpg.assoc function in CpGassoc package [65] implemented in R available at Barfield, R., Conneely, K., Kilaru, V (2024). _CpGassoc: Association Between Methylation and a Phenotype of Interest_. R package version 2.70, <https://CRANR-project.org/package=CpGassoc>.

Scale scores were log transformed if they were not normally distributed according to the results of the Shapiro–Wilk normality test (Ldisc and Rdisc). We performed an EWAS using CpGassoc package separately for African American and White participants to determine whether CpG sites are differentially methylated among the participants who reported lifetime discrimination compared to those who did not. We used multivariable linear regression to model the relationship between DNAm levels and psychosocial stress burden in each scale with age, sex, and poverty status with expected −log10(P-values) versus observed using the following models: Model 1 CpG ∼ Lifetime discrimination (Ldisc) + age, Model 2 CpG ∼ Lifetime discrimination (Ldisc) + sex, Model 3 CpG ∼ Lifetime discrimination (Ldisc) + poverty status, and Model 4 CpG ∼ Lifetime discrimination (Ldisc) + age + sex + poverty status. These models were examined on African American and White participants separately. We applied Holm–Bonferroni correction for multiple testing to the results for each discrimination scale. Significant results were determined using a false discovery rate of < 0.05.

Leukocyte cell-type proportions were estimated using the EstimateCellCounts() function in the minfi package with the FlowSorted.Blood.EPIC reference panel, a library for reference-based deconvolution of whole-blood biospecimens to estimate the composition of cell types assayed using the Illumina HumanMethylationEPIC BeadArray [66]. A major limitation of our study is that the estimated cell-type fractions were not included in the primary EWAS models because psychosocial stress may influence immune-cell composition, and adjustment could potentially remove biologically relevant variation that may lie on the causal pathway linking psychosocial stress exposure and DNAm.

Identification of significant differentially methylated positions

The methylation level for each CpG site was estimated based on β values ranging from 0 (the CpG site is unmethylated) to 1 (the CpG site is fully methylated). Initially, DMPs were estimated for the overall population agnostic to self-identified race using the dmpFinder function in the minfi package [58] for all discrimination scales. Using a linear regression test for continuous phenotypes, the dmpFinder function computes association between the methylation and a phenotype of interest. Identified DMPs from the dmpFinder function were prioritized using the following criteria: (i) a P-value < .05, indicating nominal association with each phenotype of interest; (ii) a Δbeta value of ±0.1, signifying a relatively large differential methylation, as previously described [47, 67]. We also performed a race-stratified DMP analysis for the Ldisc scale to identify race-specific DMPs in African American and White study participants. We used these DMPs to identify overlaps between our discrimination data and previously identified DMPs and CpGs associated with PD, inflammation, diabetes, blood pressure, renal function decline, and inflammation. We found these previously identified DMPs and CpGs by conducting a targeted but not systematic literature review of studies that examined DNAm among minoritized populations in the context of PD assessed by the same instruments used in our study and health conditions associated with health disparities found in high frequency in our middle-aged African American and White adults. To identify the significant DMPs, we used the cut-off P-values < .05 and < .01.

Functional pathway identification

We performed a Venn diagram comparison using the Venny 2.1.0 tool available at https://bioinfogp.cnb.csic.es/tools/venny/ to obtain the common and unique CpG sites that were associated significantly with each of the psychosocial stress scales. We uploaded the list of significantly differentially methylated CpG sites that were unique for each psychosocial stress scale for the covariates age, sex, poverty status, age + sex, and age + sex + poverty status containing gene identifiers and corresponding P-values obtained from cpg.assoc, into the IPA software (Qiagen). We performed the “core analysis” function included in the software to interpret biological processes, canonical pathways, upstream transcriptional regulators, gene networks, diseases, and functions for each CpG site significant for each psychosocial stress scale. Each gene identifier for the CpG site was mapped to its corresponding gene object in the Ingenuity Pathway Knowledge Base (IPKB). Networks were drawn based on the IPA-IPKB (https://digitalinsights.qiagen.com/IPA).

Author contributions

Kumaraswamy Naidu Chitrala (Conceptualization [equal], Formal Analysis [lead], Investigation [lead], Methodology [lead], Writing – original draft [equal], Writing – review & editing [equal]), Danielle L Beatty Moody (Conceptualization [equal], Data curation [equal], Writing – review & editing [equal]), Nicolle A. Mode (Data curation [equal], Formal Analysis [equal], Writing – review & editing [equal]), Botong Shen (Formal Analysis [equal], Methodology [supporting], Writing – review & editing [equal]), Nicole Noren Hooten (Methodology [equal], Visualization [equal], Writing – original draft [equal], Writing – review & editing [equal]), Alan B Zonderman (Data curation [equal], Formal Analysis [equal], Investigation [equal], Methodology [equal], Project administration [equal], Writing – original draft [equal], Writing – review & editing [equal]), Ngozi Ezike (Data curation [equal], Investigation [equal], Methodology [supporting], Writing – original draft [supporting]), and Michele Kim Evans (Conceptualization [lead], Funding acquisition [equal], Methodology [equal], Project administration [equal], Resources [lead], Supervision [lead], Writing – original draft [equal], Writing – review & editing [equal])

Conflicts of interest

None declared.

Funding

This research was supported [in part] by the Intramural Research Program of the National Institutes of Health (NIH). This project was funded by the Intramural Research Program of the National Institute on Aging (NIA), the National Institutes of Health (NIH), Baltimore, MD, project numbers AG000519 and AG000988. The contributions of the NIH author(s) are considered Works of the United States Government. The findings and conclusions presented in this paper are those of the author(s) and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.

Data Availability

Data are available to researchers after approval of a research proposal and after they have agreed to confidentiality as required by the National Institutes of Health Institutional Review Board. Policies and forms are available at: https://handls.nih.gov.