Genome-wide Association Analysis of Proinflammatory Cytokines and Gene–lifestyle Interaction for Invasive Breast Cancer Risk: The WHI dbGaP Study

Chronic inflammation may play an important role in the pathogenesis of noninflammatory diseases, including specific types of cancers such as colorectal, liver, and breast cancers, from tumor initiation through progression (1, 2). Activation of innate immunity involves a number of inflammatory cells, growth and transcription factors, chemokines, and proinflammatory mediators, creating a tissue microenvironment high in reactive oxygen and nitrogen species, leading to potential DNA alterations in nearby cells (3). The inflammatory response also elevates the circulating levels of cancer-promoting inflammatory cytokines such as C-reactive protein (CRP) and IL6 (2). These key biomarkers reflect different molecular pathways in the immune cascade in acute and chronic immune responses but may be inter-related in carcinogenesis. For example, IL6, upregulated by macrophages and adipose tissue, has promoted breast tumor initiation and progression (4). CRP, a major acute-phase reactant and a biomarker of chronic low-grade inflammation, partially induced by IL6, has been associated with increased risk of breast cancer (5). The molecular mechanisms of these markers in carcinogenesis have not been confirmed and are partially understood. IL6 regulates estrogen synthesis and aromatase activity responsible for estrogen production in adipose tissue, which is important in postmenopausal breast cancer (6). CRP levels are reduced when COX-2 action (promoting estrogen formation in adipose tissue) is inhibited (7). Thus, IL6 and CRP may provide a link between inflammatory pathways and breast cancer tumorigenesis.

The heritability of CRP and IL6 levels was 25%–40% (8, 9) and 60% (10), respectively, in Europeans on the basis of family and twin studies. Previous genome-wide association studies (GWAS) explain about 5% (11) and <2% (12) of interindividual variability of CRP and IL6 levels, respectively, suggesting that additional genetic contributions to CRP and IL6 concentrations remain largely undetermined. CRP and IL6 levels are also determined by obesity (overall and central; refs. 11–16), lipid metabolism (15), and obesity-related lifestyle factors such as exercise, high-fat diet, smoking, and alcohol (12, 17, 18). Thus, studying whether these lifestyle factors interact with genetic markers to influence CRP and IL6 concentrations may uncover the complicated genotype–phenotype pathway.

Few GWA gene–environment interaction (G × E) studies have examined those pleiotropic cytokines' gene–phenotype relationship with obesity factors. Particularly, genomic studies for CRP have investigated obesity-specific stratifications among GWA-based single-nucleotide polymorphisms (SNP) only (19, 20) or tested the G × E effect by introducing the interaction term into the gene–phenotype association (11). A full genome-wide scan within obesity strata at the initial stage throughout the GWA analysis may help to provide a more profound molecular basis for gene–phenotype pathways that are influenced by environment. In addition, a GWA analysis examining inflammatory markers as binary outcomes (e.g., normal range vs. chronic low-grade inflammation) rather than examining as a continuous variable, could address a nonlinearity issue that has frequently been violated in traditional linear regression and further identify SNPs whose effect reflects an allele-based risk magnitude of chronic low-grade inflammation.

Immune-related etiologic pathways that influence breast cancer may differ by menopausal status, probably owing to the role of sex hormones in mediating the innate and adaptive immune systems (21). Furthermore, in postmenopausal women, inflammatory cytokines and the genetic markers have displayed different associations with breast cancer according to obesity-related factors such as obesity status (overall and visceral; ref. 22), physical activity, and dyslipidemia (23, 24). Thus, our G × E study has focused on the risk for breast cancer among postmenopausal women, a population vulnerable to a high incidence of inflammation (25), obesity, and breast cancer (e.g., 80% of new cases occur in women age 50 years and older; ref. 26).

In this study, we hoped to identify SNPs explaining additional interindividual variability in cancer-promoting inflammatory biomarkers, including CRP and IL6. We conducted a GWA G × E study to characterize the genetic architecture of the biomarkers that interact with obesity factors. Next, we evaluated whether the identified SNPs in particular behavioral settings are associated with breast cancer risk in the identical behavioral setting. This may avoid bias derived from a different population structure by examining an identical genomic structure of the population for association with inflammatory cytokines and cancer risk simultaneously. We tested an empirical hypothesis that a substantial proportion of the susceptibility of GWA-based SNPs in CRP and IL6 affects breast cancer risk and that obesity lifestyle factors modify the relationship (Supplementary Fig. S1).

Our study included postmenopausal women who enrolled in the Women's Health Initiative (WHI) Harmonized and Imputed GWASs coordinated by the database of Genotypes and Phenotypes (dbGaP). These studies encompass the WHI Observational Studies and Clinical Trials to contribute a joint imputation and harmonization effort to the GWASs. Detailed rationale and study design have been discussed elsewhere (27, 28). The WHI study included women enrolled between 1993 and 1998 at 40 clinical centers across the United States. Eligible women were 50–79 years old, postmenopausal, expected to reside near the clinical centers for at least 3 years after enrollment, and able to provide written informed consent. The Harmonized and Imputed studies involved 6 GWASs (Table 1). Of the 16,088 who reported their race or ethnicity as non-Hispanic white (Supplementary Fig. S2), we excluded 2,714 who had been diagnosed with diabetes at or after enrollment. We also excluded 1,301 whose genetic information was found to be duplicated in the 6 GWASs and/or those with first- and second-degree relatives. In addition, we excluded 1,275 women whose genetic data did not pass the quality assurance (QA) test [outliers based on Principal Components (PCs)], leaving 10,798 (90% of the eligible 12,073) for our GWA G × E analysis. For the association with breast cancer, we next excluded 619 women who had been followed up for < 1 year and/or had been diagnosed with any type of cancer at enrollment, leaving a total of 10,179 women (94% of the 10,798 GWA participants). These women had been followed up through August 29, 2014, with a mean of 16 years follow-up, and 537 of them had developed invasive breast cancer. The Institutional Review Boards of each WHI participating clinical center and the University of California, Los Angeles, approved this study.

The WHI coordinating center collected information by using standardized written protocols and conducted data QA with periodic visits. Participants had completed at enrollment self-administered questionnaires on the following information (listed for only those selected for our study): demographic and socioeconomic (age, education, race/ethnicity, family income, and family history of breast cancer), lifestyle (depressive symptom, smoking, and exercise), dietary [dietary alcohol intake in g/day and % calories from saturated fatty acids (SFA)/day] factors, and reproductive histories (history of hysterectomy, ages at menarche and menopause, oral contraceptive use, and 2 types of exogenous estrogen use: unopposed estrogen-only and opposed estrogen plus progestin). Anthropometric data, including height, weight, and waist and hip circumferences, were measured at baseline by trained staff. Those 17 variables (Table 2) were initially selected on the basis of their association with inflammation and breast cancer among postmenopausal women through the literature review (23, 29, 30) and were confirmed by univariate, stepwise, and multicollinearity regression testing to be analyzed in our study.

Breast cancer development was our cancer outcome of interest. The time between enrollment and breast cancer development, censoring, or study endpoint was measured as the number of days and then converted into years. Breast cancer diagnosis was determined using a centralized review of medical charts by a committee of physicians on the basis of pathology or cytology reports. Cancer cases were coded according to the National Cancer Institute's Surveillance, Epidemiology, and End-Results guidelines (31).

Genotyped data generated on several different platforms were extracted from the WHI dbGaP Harmonized and Imputed GWASs (Table 1). The genotypes were normalized to the reference panel GRCh37 and genotype imputation was performed via 1,000 genomes reference panels. (28) SNPs were checked for harmonization with pairwise concordance among all samples across the 6 GWASs. We crosschecked the self-reported race/ethnicity with PCs. If any discrepancy or admixed participants were detected, an additional analysis was performed with a follow-up demographic questionnaire (32). In the initial quality control (QC) process, SNPs were filtered with a missing-call rate of <2% and a Hardy–Weinberg Equilibrium of P ≥ 1E-04. In the secondary QC, we retained SNPs with ${\hat{R}^2} \ge 0.6\ $imputation quality (33). Within high-quality SNPs that included only HapMap3 SNPs with ${\hat{R}^2} \ge 0.9$ (34), we estimated relatedness between samples. To reduce potential confounding due to shared environment, we excluded individuals with a kinship estimate >0.088 on the basis of the KING robust kinship estimator (35). We next computed 10 PCs in each GWAS using the same set of high-quality SNPs, but without relatedness, within the linkage disequilibrium (LD) and excluded outlier samples that fell outside of 6 SD on the basis of Mahalanobis distances. Finally, a total of 21,784,812 autosomal SNPs from 10,798 individuals were examined in this study.

Participants' fasting blood samples at enrollment were drawn by trained phlebotomists. Serum concentrations of IL6 and CRP were measured by using, respectively, the Quantitative Sandwich Enzyme Immunoassay technique (Quantikine HS Immunoassay Kit; R&D Systems, Inc.) and a high sensitivity immunoturbidimetric assay (Kimiya Biomedical Company) read with a Roche analyzer (Roche Diagnostics), with the median interassay coefficients of variation of 12.4% and 2.3%, respectively. About 30% of the phenotypes were replaced by imputation values using an unsupervised splitting of Random Survival Forest imputation (https://github.com/ehrlinger/randomForestSRC/blob/master/R/impute.rfsrc.R; ref. 36). In the overall and the obesity-specific strata of the GWA analysis, sensitivity testing was performed with and without imputed values for IL6 and CRP, producing regression estimates, genomic control inflation factors (λ_GC), Q-Q plots, and Manhattan plots; no significant differences were observed.

Participants' baseline characteristics by breast cancer development were examined by unpaired two-sample t tests for continuous variables and χ² tests for categorical variables; if continuous variables were skewed or had outliers, Wilcoxon rank-sum test was used. A GWA analysis in each study was conducted via multiple linear and logistic regressions, adjusting for age and 10 PCs to produce effect sizes and ORs, respectively, and 95% confidence intervals (CI) of IL6 and CRP concentrations, with additive and minor allele–dominant and –recessive genetic models. In this study, we present the results from only the logistic regressions due to the high genomic control inflation factors (λ_GC > 1.0) from the linear regressions. Multiple testing was corrected by adjusting P values to the genome-wide significance level (P < 5E-08). The inverse variance-weighted fixed-effects meta-analysis was conducted to combine the findings across the 6 GWASs. Heterogeneity among studies was tested via Cochran Q statistic (37) at P < 8E-03 after Bonferroni corrections.

For all the SNPs at the initial stage of analysis, gene–obesity interactions were examined in each study (i) by introducing an interaction term into a regression model, adjusting for 10 PCs, and (ii) within the obesity strata defined by body mass index (BMI; cutoff, 30 kg/m²), waist circumference (WST; cutoff, 88 cm), waist-to-hip ratio (WHR; cutoff, 0.85), metabolic equivalents (MET; cutoff, 10·hours/week), and % calories from SFA (cutoff, 9%). Meta-analysis was performed assuming fixed effects to combine the interaction results across the 6 GWASs. Multiple comparisons were adjusted by the Benjamini–Hochberg method (38). The LDs between the top GWA signals were computed, and regional plots were created using LOCUSZOOM (http://locuszoom.org/). Gene enrichment and functional annotation analyses were performed using DAVID v6.8 (https://david.ncifcrf.gov/).

With the top signals, we further conducted a multiple Cox proportional hazards regression in the combined 6 GWASs to obtain HRs and 95% CIs predicting a risk for breast cancer by adjusting for 17 confounding factors (Table 2). The proportional hazards assumptions were tested by a Schoenfeld residual plot and rho. Given that the SNP–cancer pathway was a hypothesis-driven question, the P value was not subject to multiple testing corrections; thus, a two-tailed P < 0.05 was considered statistically significant. During the data QC process, dose2plink script (http://genepi.qimr.edu.au/staff/sarahMe/dose2plink), PLINK1.9/2.0, KING kinship estimator, and R stats package on UCLA's Hoffman2 high performance computing cluster were used. For GWA, phenotype imputation, and SNP–cancer association tests, PLINK1.9/2.0 (glm/interaction/meta-analysis) and R3.5.1. (qqman/manhattan, randomForestSRC, and survival packages) were used.

Distributions of the pro-inflammatory cytokines (CRP and IL6) in the 6 GWASs are displayed in Table 1. Blood-level thresholds were determined as reflecting either the chronic low-grade inflammation status (>3.0 mg/L for CRP) or the fourth quartile of the participants (≥4.4 pg/mL for IL6; refs. 39, 40). Allele frequencies in each study (Supplementary Table S1) were cross-checked with those in Europeans in the 1000 Genomes Project Ensembl GRCh37 (http://grch37.ensembl.org/Homo_sapiens/Info/Index). Participants' baseline characteristics by breast cancer development are shown in Table 2. Women who developed breast cancer were more likely to have greater incomes and a family history of breast cancer, be obese overall and viscerally, smoke ≥15 cigarettes/day, consume more dietary alcohol/day, have experienced menarche at earlier age, have undergone hysterectomy less frequently, and have shorter durations of oral contraceptive and E-only use but longer durations of estrogen + progestin use.

Of 21,784,812 common autosomal SNPs, 88 SNPs reached genome-wide significance (including 4 index SNPs and 3 SNPs, independent of each other) in 10 loci, 7 of which were novel, and 3 loci (SNPs near CRP, HNF1A, and APOE) that were previously observed. The genomic control inflation factor (λ_GC) was 1.0, indicating no evidence of type 1 error inflation. Those top genetic markers did not overlap between the two cytokine phenotypes. Identified loci explained 5.8% and 0.7% of interindividual variability of CPR and IL6, respectively.

In the overall analysis, 82 top signals were detected in 7 loci of CRPP1, CRP, HNF1A-AS1, HNF1A, C12orf43, APOE, and APOC1 (Supplementary Table S2; Supplementary Fig. S3A–S3H). Specifically, 8 SNPs near CRPP1, 1 near CRPP1/CRP, and 11 in intron, 3′-UTR, or 3′ flanking regions of CRP (index: rs2808628 with r² > 0.7, Fig. 1A) were associated with increased risk of chronic inflammation (approximately 20%). On the contrary, the following loci at 12q and 19q displayed about 20% of decreased risk: 23 in an intronic region of HNF1A-AS1 and 31 in intronic or 3′-UTR regions of HNF1A (index: rs2243458 with r² > 0.7, Fig. 1B), 6 in intronic or 3′-UTR regions of C12orf43 (index: rs1169311 with r² > 0.9; Fig. 1C), APOE rs429358, and APOC1 rs5117.

When stratified by BMI (Supplementary Table S3), a number of SNPs near HNF1A-AS1 and HNF1A (18 and 27 SNPs, respectively, with index: rs2243458) and one (rs1169311) in C12orf43 were detected; these partially overlapped those identified from the overall analysis, with about 20% of decreased risk for chronic inflammation, only in the nonobese (BMI < 30) subgroup. In the obese subgroup (BMI ≥ 30), 3 novel SNPs in CRP/RP11–419N10.5 in LD (r² > 0.9) were identified with 70% of increased risk for chronic inflammation. When stratified by WHR or WST, similar patterns were shown for SNPs on inflammation: SNPs near CRP and CRP/RP11–419N10.5 with 35%–50% of increased risk in the viscerally obese subgroup only. SNPs near HNF1A were shown with decreased risk for inflammation, also only in the viscerally obese subgroup (Supplementary Table S3).

In relation to breast cancer risk with the top signals, the loci at 12q, including HNF1A-AS1, HNF1A, and C12orf43 were associated, while the loci at 1q (CRPP1, CRP, and RP11–419N10.5) were not. In particular, CRP-decreasing SNPs near HNF1A in the overall analysis were associated with 13% increased risk for breast cancer in the identical overall group (Table 3). CRP-decreasing SNPs near HNF1A-AS1, HNF1A, and C12orf43 in the nonobese (BMI < 30) subgroup displayed a slightly decreased risk without reaching statistical significance, but in the counterpart obese subgroup (BMI ≥ 30), approximately 25% increased risk for breast cancer was observed (Table 4). In addition, CRP-decreasing SNPs near HNF1A in the viscerally obese subgroup showed a 23% increased risk for breast cancer in the same viscerally obese subgroup (Table 4).

Similar genetic associations for the CRP phenotype were observed when stratified by physical activity and % calories from SFA (Table 5). SNPs near HNF1A-AS1 and HNF1A (index: rs2393776 with r² > 0.7) were associated with 35% decreased risk for chronic inflammation (except for few SNPs HNF1A-AS1 with increased risk) in the less-fat diet subgroup, whereas SNPs near CRPP1 and CRP (index: rs2808628) were associated with 20% increased risk for chronic inflammation in the physically inactive and high-fat diet subgroups. No associations of those SNPs with breast cancer were detected (Supplementary Table S4).

Three of the most significant signals were detected in two loci of DEC1 and MAPK1 (Table 6; Supplementary Fig. S3I–S3L). In all participants and in the nonviscerally obese and high-fat diet subgroups, rs149109490 near DEC1 was strongly associated with increased IL6 levels, reflecting high risk for chronic low-grade inflammatory status (OR = 3.61; P = 3.42E-08; Q = 0.078). Two MAPK1 SNPs (rs56398890 and rs9607320) in LD (r² > 0.9) displayed 25% decreased risk in the physically active subgroup only. None of the top SNPs were associated with breast cancer risk (Supplementary Table S5).

Gene–ontology annotation analyses (Supplementary Tables S6 and S7) indicated CRP involvement in the acute-phase inflammation response; HNF1A, APOC1, and APOE in glucose, lipoprotein, triglyceride, and cholesterol metabolisms; and DEC1 and MAPK1 in the regulation of mitogen-activated protein kinase/PI3K/ERK 1/2 cascades, apoptotic process, cellular response to DNA damage stimulus, and cell differentiation and migration.

A growing number of population-based genomic studies have emphasized the role of environmental factors in modifying gene–phenotype pathways. To our knowledge, this is the first study to characterize genetic determinants of inflammatory cytokines and subsequently their genomic associations with breast cancer development in an identical postmenopausal population, at the genome-wide level, by incorporating obesity factors as an effect modifier. A number of novel top-GWA signals have been detected in relation to chronic low-grade inflammation on the basis of CRP and IL6 concentrations. Some of the top loci for their associations with chronic inflammation and breast cancer would not have been detected without the incorporation of the obesity factors.

Many genes annotated to the CRP-related variants mainly clustered in the innate and adaptive immune functions (CRP) or the glucose and lipid metabolisms (HNF1A and APOC1; refs. 11, 13). In particular, the CRP gene encodes a member protein of the pentaxin family involving host defense functions during the acute-phase response to tissue infection to promote phagocytosis and the complement system by interacting with the DNA in circulating cells (41). Several CRP and CRPP1 SNPs detected in our study are located at 1q21 to 25 region, which contains genes encoding proteins with immune- and inflammation-associated functions (42). Furthermore, the CRP-related SNPs were mainly clustered in the 3′-UTL or 3′ flaking region (e.g., rs1205) of the CRP gene involved in the post-transcriptional process by regulating mRNA stability, localization, and translational efficiency; thus, those SNPs in this region may affect CRP production (19). The CRP gene is a single-copy gene. Within about 16-kb upstream and downstream from the CRP gene, only one other sequence is CRPP1, a pseudogene, with 50%–80% homology to CRP (43). CRP-related SNPs near CRPP1 have thus far been reported only in African Americans (14). We found a different set of CRPP1 SNPs that were novel in non-Hispanic whites.

A positive reciprocal feedback role between CRP and adiposity pathophysiology has been suggested (15, 20), and CRP-related genes were correlated with BMI/WST/WHR-related genes (13). Consistent with previous findings (19, 44), we found that CRP/CRPP1 SNPs were related to elevated CRP levels (i.e., greater risk for chronic inflammation) among those who were obese or had obesity lifestyles, indicating that obesity factors play a role in regulating the effect of CRP/CRPP1 genes on CRP levels. Of note, the CRP/CRPP1 gene–phenotype relationships were more profound in women with central obesity, suggesting an important role of adipose tissue distribution in CRP gene–phenotype pathways. Furthermore, the novel SNPs we found near CRP/RP11–419N10.5 were associated with up to 70% elevated risk for chronic inflammation only in women with overall and central obesity, but the biological implications of those SNPs on RP11–419N10.5, a processed pseudogene, are unclear.

Hepatic nuclear factor 1 alpha (HNF1A) is a key regulator of CRP protein. It binds to promotor regions of the CRP gene to regulate CRP synthesis in liver cells (45). Thus, HNF1A is required for CRP gene expression (46). This implicates the functional role of HNF1A SNPs in altering the expression of its target gene, CRP; HNF1A mutations led to loss of function in reducing CRP levels (47). This provides a biologically plausible basis for our findings that SNPs in HNF1A/HNF1A-AS1 were associated with reduced risk for chronic inflammation (i.e., decreased CRP levels). Furthermore, similar to previous literature (48, 49), we found that the decreased effects of the HNF1A SNPs on CRP levels were more profound in the low-fat diet subgroup, conferring the regulating role of CRP on expression of proprotein convertase subtilisin/kexin type-9, a key regulator of low-density lipoprotein (LDL) metabolism, via the p38MAPK–HNF1A pathway in hepatic cells, as recently reported by in vivo studies (50, 51). Also, we found novel SNPs at C12orf43, which is located downstream of HNF1A in a tail-to-tail manner (52), that were associated with a reduced risk for chronic inflammation (i.e., decreased CRP levels); this finding warrants further biological mechanistic study of a link between the particular gene, its mutation, and CPR phenotype.

For many genes discovered by GWASs for metabolic traits, such as chronic inflammation, the mechanism by which the encoded proteins affect disease risk is controversial. Some intergenic or intronic SNPs may affect the function of transcriptional control structures, including enhancers and silencers (53). Of the 88 top markers in our study, 27 in HNF1A/HNF1A-AS1/C12orf43 displayed the increased risk for breast cancer. Their functional significance and the molecular mechanisms of the target genes that mediated breast carcinogenesis remain unclear. The HNF1A gene regulates tissue-specific expression of multiple genes in the liver, pancreas (45), proximal tubule of the kidney (54), and epithelial cells of the intestine (55), and its mutation has been associated with maturity-onset diabetes of the young type 3 (56), hepatocellular adenoma (HCC; ref. 57), and endometrial (58) and pancreatic carcinomas (59). Whether the HNF1A gene and its mutation act as a tumor suppressor (59) or an oncogene (55, 60) on the specific types of tumor (HCC and pancreatic cancer) has not been determined. For example, one study (61) reported that HNF1A silencing in HCC cells led to overexpression of several genes encoding growth factor receptors, components of translational machinery, cell cycle, and angiogenesis regulators, which promote cell proliferation and suppress apoptosis. Another study (60) presented high HNF1A gene expression in pancreatic cancer stem cells and ductal adenocarcinoma. Our study is the first to show that SNPs near HNF1A/HNF1A-AS1 in overall and viscerally obese subgroups were associated with increased risk for breast cancer, calling for study of the biological molecular mechanisms in the gene–breast cancer pathways.

In relation to increased risk for chronic low-grade inflammation on the basis of IL6, we found one novel SNP near DEC1, whose expression was suppressed in esophageal cancer (62), suggestive of a candidate tumor suppressor. DEC1 mutations' functional implications in other tumor cells have not been determined. In our study, the SNP near DEC1 had a substantial suppressive effect on breast cancer development, although that effect did not reach statistical significance.

Because of the restrictions of the available data, our study focused on two key inflammatory biomarkers (CRP and IL6) and examined obesity factors at screening in cross-section fashion. In addition, our study was confined to non-Hispanic white postmenopausal women, so the generalizability of our results to other populations is limited. All identified loci contributed to 6.3% of interindividual variability in proinflammatory cytokines. Future DNA methylation studies may address this low heritability. in addition, the chronic inflammation pathway involved in breast cancer among postmenopausal women may differ by the cancer's molecular subtype and the exogenous estrogen intake; our study could not delineate such differences due to insufficient statistical power. Finally, some meta-analysis results indicated pleiotropic effects of inflammatory cytokines.

In conclusion, our study suggests that a number of newly identified top signals may exert their effect on chronic low-grade inflammation by interacting with obesity factors and may have implications for breast carcinogenesis in postmenopausal women. Our findings may contribute to better understanding of the molecular genetic associations between proinflammation and cancer and suggest potential intervention strategies, such as body weight control, for women who carry the inflammatory-risk genotypes, thus reducing their risk for breast cancer.

No disclosures were reported.

S.Y. Jung: Conceptualization, resources, data curation, formal analysis, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. P.A. Scott: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. J.C. Papp: Conceptualization, resources, software, supervision, validation, writing-original draft, writing-review and editing. E.M. Sobel: Conceptualization, resources, data curation, supervision, investigation, writing-original draft, writing-review and editing. M. Pellegrini: Conceptualization, resources, supervision, writing-original draft, writing-review and editing. H. Yu: Conceptualization, resources, supervision, investigation, writing-original draft, writing-review and editing. S. Han: Formal analysis, visualization, methodology, writing-original draft, writing-review and editing. Z.-F. Zhang: Conceptualization, resources, supervision, validation, visualization, methodology, writing-original draft, writing- review and editing.

This study was supported by the National Institute of Nursing Research of the NIH under Award Number K01NR017852 and a University of California Cancer Research Coordinating Committee grant (CRN-18-522722). This work used computational and storage services provided by UCLA Institute for Digital Research and Education's Research Technology Group. Part of the data for this project was provided by the WHI program, which is funded by the National Heart, Lung, and Blood Institute, the NIH, and the U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C. The datasets used for the analyses described in this manuscript were obtained from dbGaP at http://www.ncbi.nlm.nih.gov/sites/entrez?db=gap through dbGaP accession (phs000200.v11.p3).

Program Office: National Heart, Lung, and Blood Institute, Bethesda, MD: Jacques Rossouw, Shari Ludlam, Dale Burwen, Joan McGowan, Leslie Ford, and Nancy Geller.

Clinical Coordinating Center: Fred Hutchinson Cancer Research Center, Seattle, WA: Garnet Anderson, Ross Prentice, Andrea LaCroix, and Charles Kooperberg.

Investigators and Academic Centers: JoAnn E. Manson, Brigham and Women's Hospital, Harvard Medical School, Boston, MA; Barbara V. Howard, MedStar Health Research Institute/Howard University, Washington, DC; Marcia L. Stefanick, Stanford Prevention Research Center, Stanford, CA; Rebecca Jackson, The Ohio State University, Columbus, OH; Cynthia A. Thomson, University of Arizona, Tucson/Phoenix, AZ; Jean Wactawski-Wende, University at Buffalo, Buffalo, NY; Marian Limacher, University of Florida, Gainesville/Jacksonville, FL; Robert Wallace, University of Iowa, Iowa City/Davenport, IA; Lewis Kuller, University of Pittsburgh, Pittsburgh, PA; and Sally Shumaker, Wake Forest University School of Medicine, Winston-Salem, NC.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.