Epigenome-wide association studies of three social determinants of health and implications for lung functions among survivors of childhood cancer

Characteristics of study population

Study participants included childhood cancer survivors of AA (n=258, median time from diagnosis=25.2 years, interquartile range [IQR]=19.9-32.1 years) and survivors of European ancestry (EA) (n=1,618, median time from diagnosis=27.3 years, IQR=21.1-33.7 years) (Table 1). The median age at primary diagnosis was 9.6 (IQR=4.2-14.4) years in AA and 9.0 (IQR=3.8-14.4) years in EA and median age at DNA sampling was 33.9 (IQR=29.4-39.6) years in AA and 35.3 (IQR=30.0-42.1) years in EA. The proportion of female was 53.1% in AA and 40.4% in EA. More than 60% (61.6% in AA and 61.7% in EA) was overweight (≥25kg/m²). Smoking was significantly different with 70.2% of non-smokers in AA and 58.2% of non-smokers in EA (P<0.0001). Three SDOH factors were significantly different between AA and EA; specifically, AA had the lower educational attainment (P<0.0001), lower personal income (P<0.0001), and lived in neighborhoods with poorer socio-economic and physical environment measured by the area deprivation index (ADI) (P<0.0001) than EA. Primary cancer diagnoses and cancer treatment information for the study population are also listed in Table 1.

Association of DNA methylation sites with SDOH factors

First, EWAS was conducted in EA. The association P-values in -log₁₀ scale were shown for three series of EWAS analyses across the genome in the circos plot (Figure 1), with 337 for educational attainment, 39 for personal income, and 24 CpG sites for ADI reached over epigenome-wide significance level with P<9×10⁻⁸, respectively. There were 13 significant DNAm CpGs commonly associated with all three SDOH factors. Those 13 CpGs were mapped to the five genomic regions on chromosomes 2, 3, 5, 9, and 14 harboring genes including GPR55, CLDND1, CPOX, GPR15, AHRR, PRRC2B, and ELMSAN1, which are previously known to be associated with smoking exposures (Table 2). After additionally adjusting for BMI and smoking, all 13 CpGs remained statistically significant (P<0.05), but with attenuated effect sizes remaining on average 36.8% for educational attainment (range=30.8%-48.8%), 48.3% for personal income (range=38.5-58.8%), and 43.3% for ADI (range=35.1-61.5%). Given SDOH-CpG associations related to and attenuated by smoking, this finding prompted us to hypothesize that health outcomes related to lung functions may be most likely impacted by the SDOH-associated CpGs.

• Download figure
• Open in new tab

Figure 1.

Circos plot for epigenome-wide association studies (EWAS) of three social determinants of health factors. Outer circle (red): EWAS for educational attainment; middle circle (green): EWAS for personal income; inner circle (blue): EWAS for area deprivation index. Each dot depicts -log10 p value for each DNAm CpG site mapped to a chromosome location along the genome. The black lines indicate the epigenome-wide significance level (P=9×10⁻⁸).

In the EA model additionally adjusted by BMI and smoking, there were still three epigenome-wide significant associations between educational attainment and cg04180924 (chr3, CPOX, P=2.0×10⁻⁸), cg04885881 (chr1, intergenic, P=1.3×10⁻⁹), cg06359375 (chr22, HPS4, P=2.0×10⁻⁸). There was one epigenome-wide significant association between personal income and cg04180924 (chr3, CPOX, P=5.4×10⁻⁹), but no single CpG reached epigenome-wide significant level for ADI. We chose to report the models without adjusting for BMI and smoking as the primary results, however, we applied the condition of P<0.05 in the model additionally adjusted by BMI and smoking to filter against all epigenome-wide significant DNAm CpG sites in the corresponding models without BMI and smoking adjustments. Following this filtering step, we had 130 SDOH-CpG associations including educational attainment (N=88), personal income (N=23), and ADI (N=19). The detailed estimates were provided (Supplementary Table 1).

Similarly, EWAS analyses were performed in AA survivors and none of DNAm CpGs reached epigenome-wide significance level (P<9×10⁻⁸). However, for comparison purpose, we presented the AA results for 130 SDOH-CpG associations statistically established in EA survivors (Supplementary Table 2). Among 130 SDOH-CpG associations, 25 associations were also validated in AA with P<0.05; specifically, 15 for educational attainments, 8 for personal income, and 3 for ADI.

Association of social determinants of health with pulmonary conditions

The incidence of three common pulmonary conditions were examined among EA and AA survivors using 2 or 3 as the minimum Common Terminology Criteria for Adverse Events (CTCAE)-grade (Table 1 and Figure 2). In the unadjusted model with minimum CTCAE-grade set to 2, pulmonary diffusion deficit was significantly higher in AA (25.2% in AA and 18.2% in EA, P=0.03), and restrictive pulmonary deficit was also significantly higher in AA (14.2% in AA and 7.5% in EA, P=0.002), whereas obstructive pulmonary deficit was comparable between the two racial groups (9.8% in AA and 13.1% in EA, P=0.21). In the model adjusting for other covariates, regardless of with or without SDOH factors, race was significantly associated with restrictive pulmonary deficits. However, race was significantly associated with pulmonary diffusion deficits only in the model without adjusting for SDOH factors and became not significant after adjusting for SDOH (Supplementary Table 3), suggesting significant contribution of SDOH factors to association between race and pulmonary diffusion deficits.

• Download figure
• Open in new tab

Figure 2.

Incidence of common pulmonary conditions among survivors of African ancestry and European ancestry

Methylation mediation effect for social determinants of health on pulmonary conditions

Among all 130 SDOH-CpG associations from EWAS in EA survivors, 61 CpGs (29 educational attainment associated CpG sites, 16 personal income associated CpGs sites, and 16 ADI associated CpG sites) were also significantly associated with risk of pulmonary diffusion deficits after adjusting for multiple comparisons (P_FDR<0.05) (Supplementary Table 4). In the mediation analysis, 17 out of 29 educational attainment associated CpGs were identified with significant average causal mediation effects (ACME) after adjusting for multiple comparisons. Using squared pairwise Pearson correlation coefficient r² threshold of 0.05, three independent CpGs including cg04180924 (chr3, CPOX, mediation=32.9%, P_FDR=0.01), cg11205006 (chr3, CPOX, mediation=32.9%, P_FDR=0.01), and cg27470486 (chr17, ACLY, mediation=8.6%, P_FDR=0.04) were obtained by top-down pruning the 17 CpGs sorted by estimated ACME in decreasing order. For the final mediation analysis, a combined score (i.e., summation of DNAm levels of the three CpGs with the same direction of association) was used as the mediating variable and 48.9% of mediation effect for educational attainment on pulmonary diffusion deficits was achieved (Table 3). Similarly, single significant mediator cg08064403 was found with significant partial mediation effect for personal income (mediation=25.9%, P<0.001) and ADI (mediation=24.1%, P<0.001) on pulmonary diffusion deficits (Table 3).

Study population

The St. Jude Lifetime Cohort (SJLIFE) study, a retrospective hospital-based cohort study for survivors of childhood cancer with prospective clinical follow-up, has been described elsewhere.^17,18 Survivors completed demographic and epidemiological questionnaires, received clinical assessments at follow-up visits. A total of 2,052 EA survivors and 370 AA survivors also had DNAm profiling data generated with EPIC array.¹⁵ We subsequently excluded survivors with attained age younger than 25 years old (n=533) to ensure most individuals should have possibly attained the highest level of education (i.e., post-college), and principal components analysis (PCA) outliers (n=13). The remaining 1,618 EA and 258 AA survivors were included in this current study.

Social determinants of health

SDOH factors for cancer survivors at personal and neighborhood levels were considered and collected at the same time as or before DNA sampling. The personal level factors included educational attainment and personal income measures. Neighborhood social determinants included area SES deprivation as a risk factor. Educational attainment and personal income were collected from SJLIFE surveys. Personal educational attainment was categorized into five levels (below high school, high school or training after high school, some college, college graduate, and post-graduate), and household incomes was categorized into five levels (None, <20, 20K to <40K, 40 to <60K, and ≥60K). We evaluated area deprivation of each survivor using the ADI that is comprised of 17 neighborhood-based SES measures including income, employment, education, and housing status collected in American Community Survey.^19,20 Census track and block level of data were created by geocoding the participants’ residential addresses using Geographic Identifiers (GEOIDs) from Census Bureau, then geocoded data was linked to national ADI file. Technically, these measures present SES and physical environment by census blocks in the USA, and each census block receives a percentile ranking with minimum disadvantage in the 1st percentile and maximum disadvantage in the 100th percentile. For ADI, we considered >75th percentile, 40 to 75 percentile, and <40 percentile as high, moderate, and low area SES deprivation, respectively.

Chronic health conditions

We will focus on the following three common CHC related to lung function: obstructive pulmonary deficit, pulmonary difficult deficit, and restrictive pulmonary deficit. Clinical evaluations of these common conditions have been documented in previous publications for SJLIFE cohort.^17,18 CHCs with severity grading based on CTCAE version 4.0: 0 (no problem), 1 (mild), 2 (moderate), 3 (severe/disabling), 4 (life-threatening), or 5 (death).¹⁸ Only incident CHC with grade ≥2 and occurred after blood draw for DNAm profiling were considered as an adverse event in the current study.

DNA methylation profiling

Blood samples collected from each participant at their follow-up visits were banked in the St. Jude Biorepository. Methylation profiling were performed using the Infinium® MethylationEPIC BeadChip. Genomic DNA (250ng per sample) were treated with bisulfite using the Zymo EZ DNA Methylation Kit under the following thermos-cycling conditions: 16 cycles: 95°C for 30 sec, 50°C for 1 hour. Following bisulfite treatment, DNA samples were desulphonated, column purified, then eluted using 12 ul of elution buffer (Zymo Research Corp.). Treating DNA with sodium bisulfite selectively converted cytosine to uracil but preserves 5-methylcytosine, which was immune from deamination. Bisulfite-converted DNA (4 µl) was then processed by following the Illumina Infinium Methylation Assay protocol, which included hybridization to MethylationEPIC BeadChips, single base extension assay, and staining and scanning using the Illumina HiScan system.

DNA methylation bioinformatic analysis

Methylation raw intensity data were analyzed in R (version 3.6.1) using minfi package.²¹ The processed methylation data was described as a β-value, which is a continuous variable ranging between 0 (no methylation) and 1 (full methylation). In any sample, a probe with a detection p value more than 0.01 was assigned missing status. Any sample or probe with more than 5% missing values were excluded from downstream analysis. Non-specific or cross-reactive probes or probes with SNPs nearby the CpG site were also be excluded. The methylation data were normalized by quantile normalization method implemented in minfi package.²¹ M-value (logit transformation of β-value) were calculated and subsequently used in regression analyses.²² The epigenome-wide methylation scores were used to impute six lymphocyte cell subtype proportions (lymphocyte, monocyte, and granulocyte, CD4, natural killer, and B cells).^23,24 A PCA were performed to quantify latent structures or batch effects in the data. We use the array annotations provided by Illumina to map probes to their corresponding genes.

Statistical analysis

The EWAS between methylation level at each CpG-site (dependent variable, continuous) and each SDOH factor (independent variable, categorical) were evaluated using multiple linear regression with covariate adjustments including sex, attained age, race, cancer treatment exposures, lymphocyte subtype proportions, significant genetic principal components and methylation principal components (model 1). Health behaviors (BMI and smoking) were additionally adjusted to evaluate the contribution of each SDOH factor on methylation level independent of health behaviors (model 2). However, because health behaviors are intermediate variables and potentially on the causal pathway from SDOH factors to DNAm, we would avoid the over adjustment²⁵ and treated epigenome-wide significant results from model 1 as the primary results with a post filtering step with the criteria of P<0.05 in model 2 to derive more credible SDOH-associated CpGs considering the effect did not completely disappear in model 2. We used R package CpGassoc²⁶ for the linear regression analysis and used P<9×10⁻⁸ as the epigenome-wide significance threshold.²⁷ EWAS results were visualized with circos plot (Circos v.0.69).²⁸

Multivariable logistic regression models were used to assess the associations between the adjusted M-value for each SDOH-associated CpG (by the same set of covariates specified in aforementioned EWAS model) and each of pulmonary condition (binary outcome) with covariates including sex, age, BMI, smoking, and radiation to the chest (a treatment exposure significantly associated with pulmonary conditions). Only CpG-CHC associations with the significant false discovery rate (FDR) adjusted p-value (P_FDR< 0.05) were regarded as statistical significance and were advanced as candidate CpGs for the downstream mediation analysis.

In the mediation analysis, each of the candidate CpGs was considered as a hypothesized mediating variable, each SDOH factor as the exposure, status of each pulmonary condition (binary) as the outcome, and adjusted for sex, age, BMI, smoking, and radiation to the chest. Individually significant mediator CpGs will be checked for pair-wise correlations. All independent significant mediator CpGs (squared Pearson correlation coefficient r²<0.05) were combined into a methylation risk score for a final mediation analysis. Mediation analysis was carried out using Mediation R package.²⁹ All statistical analyses were performed by using R.3.6.3³⁰ or SAS 9.4 (SAS Institute Inc., Cary, NC, USA) and all statistical tests were two-sided.