https://orcid.org/0000-0002-9577-226X
Abstract
Breast cancer survivors have increased incidence of age-related diseases, suggesting that some survivors may experience faster biological aging.
Among 417 women enrolled in the prospective Sister Study cohort, DNA methylation data were generated on paired blood samples collected an average of 7.7 years apart and used to calculate 3 epigenetic metrics of biological aging (PhenoAgeAccel, GrimAgeAccel, and Dunedin Pace of Aging Calculated from the Epigenome [DunedinPACE]). Approximately half (n = 190) the women sampled were diagnosed and treated for breast cancer between blood draws, whereas the other half (n = 227) remained breast cancer–free. Breast tumor characteristics and treatment information were abstracted from medical records.
Among women who developed breast cancer, diagnoses occurred an average of 3.5 years after the initial blood draw and 4 years before the second draw. After accounting for covariates and biological aging metrics measured at baseline, women diagnosed and treated for breast cancer had higher biological aging at the second blood draw than women who remained cancer-free as measured by PhenoAgeAccel (standardized mean difference [β] = 0.13, 95% confidence interval [CI) = 0.00 to 0.26), GrimAgeAccel (β = 0.14, 95% CI = 0.03 to 0.25), and DunedinPACE (β = 0.37, 95% CI = 0.24 to 0.50). In case-only analyses assessing associations with different breast cancer therapies, radiation had strong positive associations with biological aging (PhenoAgeAccel: β = 0.39, 95% CI = 0.19 to 0.59; GrimAgeAccel: β = 0.29, 95% CI = 0.10 to 0.47; DunedinPACE: β = 0.25, 95% CI = 0.02 to 0.48).
Biological aging is accelerated following a breast cancer diagnosis and treatment. Breast cancer treatment modalities appear to differentially contribute to biological aging.
In the United States, there are approximately 4 million breast cancer survivors (1). The number of women surviving breast cancer has steadily risen in recent decades as breast cancer incidence rates have increased by 40% and mortality rates have decreased by 43% (2). Compared with cancer-free women, those with a history of breast cancer experience faster functional decline and have higher incidences of age-related diseases (3-7). These observations have led to the hypothesis that breast cancer survivors experience faster rates of aging than cancer-free women (8).
Alterations to DNA methylation (DNAm) occur over the life course and may reflect processes related to aging (9). Blood DNAm profiles have been used to construct a variety of different metrics related to age. The PhenoAge and GrimAge epigenetic clocks were designed by identifying cytosine-phosphate-guanine (CpG) sites where DNAm was indirectly associated with all-cause mortality (10,11). Specifically, the PhenoAge clock uses a set of CpGs where blood DNAm was predictive of a previously developed phenotypic age score (10). The phenotypic age risk score is a composite metric based on 9 clinical biomarkers (albumin, creatinine, glucose, C-reactive protein, alkaline phosphatase, white blood cell count, lymphocyte percent, red blood cell width, and volume) and chronological age. The GrimAge epigenetic clock uses separate sets of CpGs that were predictive of 7 circulating proteins (adrenomedullin, beta-2-microglobulin, cystatin C, growth differentiation factor 15, leptin, plasma activator inhibitor-1, and tissue metalloproteinase-1) and smoking history (pack-years); these DNAm-predicted proteins and smoking history metrics were then combined with chronological age and sex (11). Both the PhenoAge and GrimAge clocks produce estimates of epigenetic age (10,11). The difference between DNAm-predicted epigenetic age and chronological age is termed epigenetic age acceleration (referred to as PhenoAgeAccel and GrimAgeAccel), which is associated with past behavioral, lifestyle, and environmental exposures and predictive of age-related disease and mortality risk (10-22).
In 2022, the Dunedin Pace of Aging Calculated from the Epigenome (DunedinPACE) was developed (12). This metric uses DNAm to predict a previously developed Dunedin Pace of Aging score (23). The Pace of Aging score was developed in a birth cohort with serially collected blood samples and physiological traits over the course of 2 decades in adulthood; 19 trajectories of blood-based biomarkers and physiological traits were derived and combined to represent age-dependent changes in the integrity of the cardiovascular, metabolic, renal, hepatic, immune, dental, and pulmonary systems (23). As DunedinPACE uses DNAm measured at a single timepoint to estimate the collective, time-dependent changes in various biological systems, it is proposed to measure the rate, or pace, of aging (12). Unlike the PhenoAge and GrimAge epigenetic clocks whose epigenetic age estimates are on the same scale as chronological age, the values produced by the DunedinPACE metric are scaled to have a mean of 1, representing the average change that occurs over 1 year. Consequently, the output from DunedinPACE is conceptually distinct from the PhenoAgeAccel and GrimAgeAccel metrics, which represent deviations from chronological age at a single timepoint. However, for convenience, we refer to all 3 as metrics of biological aging. Despite evidence that biological aging metrics are useful estimators of aging phenotypes (13), they have not yet been widely applied in longitudinal designs or studies of breast cancer survivorship.
Our group has previously reported that women with elevated epigenetic age acceleration are at increased risk of developing breast cancer (17,18). Here, we extend that work using paired blood samples collected an average of 7.7 years apart from women who did and did not develop breast cancer during the intervening years between the blood draws. We use newly generated DNAm data from these paired samples to examine whether intervening breast cancer diagnoses and their accompanying treatments are associated with accelerated biological aging.
Methods
Study population
The Sister Study is a nationwide, prospective cohort of 50 884 women living in the United States, including Puerto Rico, enrolled between July 2003 and March 2009 (24). The study was designed to identify novel environmental and biological factors associated with breast cancer incidence and survival. Women were eligible for enrollment if they were aged between 35 and 74 years, had a sister (full or half) diagnosed with primary breast cancer, and had not themselves been diagnosed with breast cancer. Baseline data were collected using self-completed questionnaires and a 2-part computer-assisted telephone interview. As part of enrollment, trained medical examiners conducted an in-person home visit to collect anthropometric characteristics and biospecimens, including whole blood. Participants are recontacted annually to provide new information on health (including incident breast cancer) and every 2-3 years to update environmental exposure and lifestyle information; annual response rates have been approximately 90%. Written informed consent was collected at the home visit, and the institutional review board of the National Institutes of Health oversees the study. Additional information and procedures for accessing Sister Study data can be found at https://sisterstudy.niehs.nih.gov/English/coll-data.htm.
Between 2013 and 2015, 3738 women who had provided a blood sample at the enrollment home visit were invited to participate in a second home visit (25). Of these women, 2315 (62%) participants provided a second whole-blood sample. By design, approximately half (n = 1146) the women had been diagnosed and treated for breast cancer between the 2 blood draws, whereas the other half (n = 1169) remained breast cancer-free.
DNA methylation assessment and quality control
In 2019, paired blood samples collected at the initial and follow-up home visits from 433 self-identified Black (Hispanic and non-Hispanic) and non-Hispanic White women were selected for DNA methylation profiling. Approximately half (n = 197; 45%) of the women were selected because they had been diagnosed and treated for breast cancer between the blood draws, whereas the other half (n = 236; 55%) had remained breast cancer–free. To improve the racial diversity of the study sample, women who self-reported as Black were oversampled for both groups. Genomic DNA was extracted from whole-blood aliquots using a modified salt precipitation methodology at the National Institute of Environmental Health Sciences (NIEHS) Molecular Genetics Core Facility or using DNAQuik at BioServe Biotechnologies Ltd (Beltsville, MD, USA) (26). One microgram of extracted DNA was bisulfate converted using the EZ DNA Methylation Kit (Zymo Research, Orange County, CA, USA) in 96-well plates. Samples were tested for completeness of bisulfite conversion, and converted DNA was assayed using Illumina’s Infinium MethylationEPIC BeadChips at the Cancer Genomics Research Laboratory of the National Cancer Institute (Bethesda, MD, USA) (27).
The ENmix pipeline (https://bioconductor.org/packages/release/bioc/html/ENmix.html) was used for DNAm data preprocessing as it is reported to outperform alternative pipelines (28-30). This pipeline preprocesses raw DNAm data IDAT files in a stepwise manner that includes background correction (31), REgression on Logarithm of Internal Control probes (RELIC) dye-bias correction (32), interarray normalization, and Regression on Correlated Probes (RCP) probe-type bias correction (33). Samples were excluded if they did not meet quality control measures including bisulfate intensity less than 5000, had greater than 5% of probes with low-quality methylation values (detection P > .000001, <3 beads, or values outside 3 times the interquartile range), or were outliers for their methylation beta value distributions. Leukocyte subset proportions were estimated using previously derived reference panels (34).
Epigenetic clock calculation
For the PhenoAge and GrimAge epigenetic clocks, epigenetic age acceleration represents the difference between a person’s DNAm-predicted epigenetic age and the individual’s chronological age. PhenoAgeAccel and GrimAgeAccel are calculated as the residuals from linear regression models where DNAm-predicted PhenoAge or GrimAge is treated as the dependent variable and chronological age is treated as the independent variable. Thus, age acceleration values should have a mean of zero and should be, by design, uncorrelated with chronological age. The epigenetic age acceleration values based on the PhenoAge and GrimAge epigenetic clocks are called PhenoAgeAccel and GrimAgeAccel and were obtained using an online calculator (https://dnamage.genetics.ucla.edu/home). As incremental improvements in the reliability of the biological aging metrics have been reported when using an intermediate calculation step based on principal component analysis of genome-wide DNAm data, we derived principal component versions of the PhenoAgeAccel and GrimAgeAccel metrics using the approach as described by developers (35).
DunedinPACE was developed by identifying CpGs where DNAm predicts the previously developed Pace of Aging score (12). DunedinPACE values are always positive; values greater than 1 represent faster rates of aging, whereas values less than 1 represent slower rates of aging. The DunedinPACE epigenetic clock was calculated using the methylAge function within the ENmix R package.
Breast cancer characteristics and therapies
Incident breast cancers and dates of diagnosis of Sister Study participants are self-reported via follow-up questionnaire or direct contact with the Sister Study. Women who report an incident breast cancer are asked to share a personal copy of their pathology report and are recontacted 6 months after their diagnosis to obtain permission to contact their health-care providers for retrieval of medical records. Medical records were available for 1112 (97%) of the women diagnosed with breast cancer between the first and second blood draws.
Medical records abstraction was performed to collect information on breast cancer therapies and tumor characteristics. Breast cancer therapy information included details on breast cancer–associated surgeries, chemotherapy, radiation therapy, and endocrine therapy. Information on tumor characteristics included estrogen receptor (ER) status and tumor stage.
Statistical analysis
Of the 433 women in the sample population, 12 were excluded because one of their samples failed DNAm quality control, and another 4 were excluded because one or more of their samples were an extreme outlier in biological aging (defined as outside 4 SDs from the mean); in total, 417 (190 breast cancer survivors, 227 cancer-free participants) women remained in the analytic sample. Because the biological aging metrics have different means and distributions, the baseline and follow-up biological age metrics were pooled and standardized as a group to have a mean of zero and a SD of 1 to allow for comparison across the metrics. Unlike the PhenoAgeAccel and GrimAgeAccel metrics, DunedinPACE is bounded to positive values, so we also conducted a sensitivity analysis using a log-transformed version of DunedinPACE. Characteristics of the sample population at baseline and follow-up, stratified by breast cancer status, were reported using means and SDs for continuous measures and counts and percentages for categorical measures.
In the primary analyses, associations between breast cancer status and biological aging were examined using linear regression models where the biological aging metric at follow-up was treated as the dependent variable and breast cancer status as an independent variable. All models were adjusted for the baseline biological aging metric, baseline chronological age (years), follow-up time (years), and self-reported race (White, Black). Supplemental analyses were conducted using the principal component–based PhenoAgeAccel and GrimAgeAccel metrics, adjusting for change in leukocyte proportions between the baseline and follow-up blood draws and using models treating the difference in the biological aging metrics between the 2 blood draws as the dependent variable and breast cancer status as the independent variable.
To explore differential associations by tumor invasiveness (ductal carcinoma in situ [DCIS] vs invasive) and ER status (positive vs negative), associations were examined by comparing cancer-free participants with the subset of survivors diagnosed with the breast cancer characteristics of interest. Associations of breast cancer status with the biological aging metrics were also examined stratified by self-reported race. To examine persistence of associations with biological aging, we also stratified the case group by time of breast cancer diagnosis (greater than 4 years before the second blood draw vs within 4 years) and, in a case-only analysis, examined associations with the biological aging metric at follow-up with time since diagnosis. Statistical tests for etiologic heterogeneity by breast cancer characteristics were performed using a case-only approach with 2-sided P values calculated from logistic regression models adjusted for the previously mentioned covariates, treating the breast cancer characteristic as dependent variables (eg, DCIS vs invasive) and biological aging metrics at the second timepoint as independent variables (36).
To examine differential associations with breast cancer treatments, we first compared cancer-free participants to breast cancer survivors, grouped by the combinations of treatments received. Subsequent analyses to determine associations with individual treatment classes were conducted using a case-only analysis. Specifically, linear regression models were constructed with biological aging metrics at the second timepoint as the dependent variable and with chemotherapy (yes, no), radiation therapy (yes, no), or endocrine therapies (yes, no) included separately or together as independent variables. In addition to the covariates listed above, supplemental treatment analyses were conducted with further adjustment for tumor stage (0, I, II, III) and ER status (positive, negative) and stratified by self-reported race and time of the breast cancer diagnosis.
Results
Sample characteristics
Compared with women who remained breast cancer–free, the women who were diagnosed with breast cancer before the second blood draw were slightly older at enrollment (cancer-free participants: mean = 55 [SD = 9] years; breast cancer survivors: mean = 57 [SD = 9] years) and were less likely to self-identify as being of Black race (cancer-free participants: 35%; breast cancer survivors: 24%) (Table 1). Overall, the mean time between blood draws was 7.7 (SD = 1) years. Women who remained breast cancer–free had a slightly shorter time interval between blood draws than the women who developed breast cancer (cancer-free participants: mean = 7.5 [SD = 1] years; breast cancer survivors: mean = 7.8 [SD = 1]). At baseline, approximately two-thirds of the women were postmenopausal, and by the end of study follow-up, nearly all women had transitioned through menopause. PhenoAgeAccel and GrimAgeAccel at baseline were marginally higher in women who went on to develop breast cancer than women who remained cancer-free (median unstandardized PhenoAgeAccel, cancer-free participants = −1.3, breast cancer survivors = −0.2; median unstandardized GrimAgeAccel, cancer-free participants = −0.5, breast cancer survivors = −0.4; Supplementary Figure 1, available online).
Characteristics of the sample population at baseline (2003-2009) and at follow-up (2013-2015)
| Breast cancer–free controls (n = 227) | Breast cancer cases (n = 190) | |||
|---|---|---|---|---|
| Characteristic | Baseline | Follow-up | Baseline | Follow-up |
| Age, mean (SD), y | 55 (9) | 63 (9) | 57 (9) | 65 (9) |
| Body mass index, mean (SD), kg/m2 | 28 (7) | 29 (7) | 29 (6) | 29 (6) |
| Physical activity, mean (SD), h/wk | 13 (8) | 10 (16) | 13 (7) | 9 (11) |
| Alcohol intake, mean (SD), drinks per wk | 2.4 (4) | 3.2 (6) | 1.9 (3) | 3.0 (5) |
| Smoking pack years, mean (SD), y | 4.9 (10) | 5.0 (10) | 6.4 (14) | 6.6 (15) |
| Stress, mean value (SD)a | 2.8 (3) | 4.0 (3) | 2.3 (3) | 3.5 (3) |
| Systolic blood pressure, mean mmHg (SD) | 115 (13) | 118 (13) | 115 (13) | 119 (13) |
| Diastolic blood pressure, mean mmHg (SD) | 72 (9) | 74 (8) | 73 (8) | 73 (8) |
| Menopause status, No. (%) | ||||
| Premenopausal | 76 (33) | 7 (4) | 55 (29) | 9 (5) |
| Postmenopausal | 151 (67) | 187 (96) | 135 (71) | 181 (95) |
| Race, No. (%) | ||||
| Black, Hispanic or non-Hispanic | 79 (35) | — | 46 (24) | — |
| White, non-Hispanic | 148 (65) | — | 144 (76) | — |
| Educational attainment, No. (%) | ||||
| High school or GED or less | 32 (14) | — | 22 (12) | — |
| Some college | 70 (31) | — | 50 (26) | — |
| College graduate or more | 125 (55) | — | 118 (62) | — |
| Breast cancer–free controls (n = 227) | Breast cancer cases (n = 190) | |||
|---|---|---|---|---|
| Characteristic | Baseline | Follow-up | Baseline | Follow-up |
| Age, mean (SD), y | 55 (9) | 63 (9) | 57 (9) | 65 (9) |
| Body mass index, mean (SD), kg/m2 | 28 (7) | 29 (7) | 29 (6) | 29 (6) |
| Physical activity, mean (SD), h/wk | 13 (8) | 10 (16) | 13 (7) | 9 (11) |
| Alcohol intake, mean (SD), drinks per wk | 2.4 (4) | 3.2 (6) | 1.9 (3) | 3.0 (5) |
| Smoking pack years, mean (SD), y | 4.9 (10) | 5.0 (10) | 6.4 (14) | 6.6 (15) |
| Stress, mean value (SD)a | 2.8 (3) | 4.0 (3) | 2.3 (3) | 3.5 (3) |
| Systolic blood pressure, mean mmHg (SD) | 115 (13) | 118 (13) | 115 (13) | 119 (13) |
| Diastolic blood pressure, mean mmHg (SD) | 72 (9) | 74 (8) | 73 (8) | 73 (8) |
| Menopause status, No. (%) | ||||
| Premenopausal | 76 (33) | 7 (4) | 55 (29) | 9 (5) |
| Postmenopausal | 151 (67) | 187 (96) | 135 (71) | 181 (95) |
| Race, No. (%) | ||||
| Black, Hispanic or non-Hispanic | 79 (35) | — | 46 (24) | — |
| White, non-Hispanic | 148 (65) | — | 144 (76) | — |
| Educational attainment, No. (%) | ||||
| High school or GED or less | 32 (14) | — | 22 (12) | — |
| Some college | 70 (31) | — | 50 (26) | — |
| College graduate or more | 125 (55) | — | 118 (62) | — |
Stress measured using the short version of the Perceived Stress Scale to assess the subject’s internal appraisal of stress in the last 30 days.
Characteristics of the sample population at baseline (2003-2009) and at follow-up (2013-2015)
| Breast cancer–free controls (n = 227) | Breast cancer cases (n = 190) | |||
|---|---|---|---|---|
| Characteristic | Baseline | Follow-up | Baseline | Follow-up |
| Age, mean (SD), y | 55 (9) | 63 (9) | 57 (9) | 65 (9) |
| Body mass index, mean (SD), kg/m2 | 28 (7) | 29 (7) | 29 (6) | 29 (6) |
| Physical activity, mean (SD), h/wk | 13 (8) | 10 (16) | 13 (7) | 9 (11) |
| Alcohol intake, mean (SD), drinks per wk | 2.4 (4) | 3.2 (6) | 1.9 (3) | 3.0 (5) |
| Smoking pack years, mean (SD), y | 4.9 (10) | 5.0 (10) | 6.4 (14) | 6.6 (15) |
| Stress, mean value (SD)a | 2.8 (3) | 4.0 (3) | 2.3 (3) | 3.5 (3) |
| Systolic blood pressure, mean mmHg (SD) | 115 (13) | 118 (13) | 115 (13) | 119 (13) |
| Diastolic blood pressure, mean mmHg (SD) | 72 (9) | 74 (8) | 73 (8) | 73 (8) |
| Menopause status, No. (%) | ||||
| Premenopausal | 76 (33) | 7 (4) | 55 (29) | 9 (5) |
| Postmenopausal | 151 (67) | 187 (96) | 135 (71) | 181 (95) |
| Race, No. (%) | ||||
| Black, Hispanic or non-Hispanic | 79 (35) | — | 46 (24) | — |
| White, non-Hispanic | 148 (65) | — | 144 (76) | — |
| Educational attainment, No. (%) | ||||
| High school or GED or less | 32 (14) | — | 22 (12) | — |
| Some college | 70 (31) | — | 50 (26) | — |
| College graduate or more | 125 (55) | — | 118 (62) | — |
| Breast cancer–free controls (n = 227) | Breast cancer cases (n = 190) | |||
|---|---|---|---|---|
| Characteristic | Baseline | Follow-up | Baseline | Follow-up |
| Age, mean (SD), y | 55 (9) | 63 (9) | 57 (9) | 65 (9) |
| Body mass index, mean (SD), kg/m2 | 28 (7) | 29 (7) | 29 (6) | 29 (6) |
| Physical activity, mean (SD), h/wk | 13 (8) | 10 (16) | 13 (7) | 9 (11) |
| Alcohol intake, mean (SD), drinks per wk | 2.4 (4) | 3.2 (6) | 1.9 (3) | 3.0 (5) |
| Smoking pack years, mean (SD), y | 4.9 (10) | 5.0 (10) | 6.4 (14) | 6.6 (15) |
| Stress, mean value (SD)a | 2.8 (3) | 4.0 (3) | 2.3 (3) | 3.5 (3) |
| Systolic blood pressure, mean mmHg (SD) | 115 (13) | 118 (13) | 115 (13) | 119 (13) |
| Diastolic blood pressure, mean mmHg (SD) | 72 (9) | 74 (8) | 73 (8) | 73 (8) |
| Menopause status, No. (%) | ||||
| Premenopausal | 76 (33) | 7 (4) | 55 (29) | 9 (5) |
| Postmenopausal | 151 (67) | 187 (96) | 135 (71) | 181 (95) |
| Race, No. (%) | ||||
| Black, Hispanic or non-Hispanic | 79 (35) | — | 46 (24) | — |
| White, non-Hispanic | 148 (65) | — | 144 (76) | — |
| Educational attainment, No. (%) | ||||
| High school or GED or less | 32 (14) | — | 22 (12) | — |
| Some college | 70 (31) | — | 50 (26) | — |
| College graduate or more | 125 (55) | — | 118 (62) | — |
Stress measured using the short version of the Perceived Stress Scale to assess the subject’s internal appraisal of stress in the last 30 days.
Among the breast cancer survivors, diagnoses occurred an average of 3.6 years after the initial blood draw and 4 years before the second (Supplementary Figure 2, available online). A majority of the tumors were ER positive (83%) and classified as invasive (71%; stage I, II, or III) (Supplementary Table 1, available online). In addition to surgery, 68% of women received endocrine therapies, 64% received radiation therapy, and 35% received chemotherapy (Supplementary Table 1, available online).
Breast cancer status and biological aging over time
Linear regression models were used to assess differences in the biological aging metrics at the second sample collection by breast cancer status. Models were adjusted for the baseline biological aging metrics, baseline chronological age, follow-up time, and self-reported race. All 3 biological aging metrics were standardized to have a mean of zero and a SD of 1; thus, the β values produced by the linear regression models represent estimated effect of the intervening breast cancer diagnosis and treatment on the difference in the standardized biological aging metrics between the breast cancer survivors and cancer-free participants.
Compared with women who remained breast cancer–free, women with an intervening breast cancer had higher biological aging metrics at the follow-up visit (PhenoAgeAccel: β = 0.13, 95% confidence interval [CI] = 0.00 to 0.26; P = .04; GrimAgeAccel: β = 0.14, 95% CI = 0.03 to 0.25; P = .01; DunedinPACE: β = 0.37, 95% CI = 0.24 to 0.50; P < .001) (Figure 1). Breast cancer associations were similar using either the standardized original DunedinPACE or log-transformed DunedinPACE metrics (Supplementary Table 2, available online). Associations were largely similar in linear regression models treating the difference in biological aging metrics between the 2 timepoints as the dependent variable (Supplementary Table 3, available online), and associations appeared stronger using the principal component–based PhenoAgeAccel and GrimAgeAccel metrics (Principal Component [PC]-PhenoAgeAccel, β = 0.21, 95% CI = 0.07 to 0.34; P = .003; PC-GrimAgeAccel: β = 0.19, 95% CI = 0.07 to 0.31; P = .002; Supplementary Table 4, available online). Associations with all 3 biological age metrics were attenuated after adjustment for changes in leukocyte composition between the baseline and follow-up blood draws (PhenoAgeAccel: β = 0.07, 95% CI = −0.05 to 0.18; P = .26; GrimAgeAccel: β = 0.08, 95% CI = −0.0 to 0.17; P = .11; DunedinPACE: β = 0.29, 95% CI = 0.17 to 0.41; P < .001; Supplementary Table 5, available online). No differences in biological aging were observed between survivors diagnosed with DCIS vs those diagnosed with invasive breast cancers (Supplementary Figure 3, available online). There was limited evidence of heterogeneity by ER status, with somewhat stronger associations with GrimAgeAccel for women diagnosed with ER-negative tumors (ER negative: β = 0.28, 95% CI = 0.07 to 0.49; P = .009; ER positive: β = 0.11, 95% CI = −0.01 to 0.22; P = .07; Pheterogeneity = .12) and somewhat stronger DunedinPACE associations for women diagnosed with ER-positive tumors (ER positive: β = 0.39, 95% CI = 0.25 to 0.52; P < .001; ER negative: β = 0.25, 95% CI = 0.01 to 0.50; P = .04; Pheterogeneity = .25) (Figure 2, A). Associations with PhenoAgeAccel or GrimAgeAccel were similar for self-reported White and Black women (Figure 2, B). However, stronger associations were observed among Black women when measured by DunedinPACE (Black women: β = 0.47, 95% CI = 0.22 to 0.72; P < .001; White women: β = 0.32, 95% CI = 0.17 to 0.48; P < .001; Pheterogeneity = .36).
Case-control analysis of biological aging and breast cancer status. Results from multivariable linear regression models treating the biological aging metric at the second timepoint as the dependent variable and breast cancer status as an independent variable. Referent group was the women who remained breast cancer–free. Models adjusted for baseline age, baseline biological aging metric, follow-up time, and race. CI = confidence interval; DunedinPACE = Dunedin Pace of Aging Calculated from the Epigenome; GrimAgeAccel = Age acceleration based on the GrimAge epigenetic clock; PhenoAgeAccel = Age acceleration based on the PhenoAge epigenetic clock.
Case-control analysis of biological aging and breast cancer status, stratified by self-reported race and estrogen receptor (ER) status. Results from multivariable linear regression models treating the biological aging metric at the second timepoint as the dependent variable and breast cancer status as an independent variable. Models are stratified by race (White vs Black) and tumor ER status (positive vs negative). Referent groups were the women who remained breast cancer–free. Models adjusted for baseline age, baseline biological aging metric, follow-up time, and race. DunedinPACE = Dunedin Pace of Aging Calculated from the Epigenome; GrimAgeAccel = Age acceleration based on the GrimAge epigenetic clock; het = heterogeneity; int = interaction; PhenoAgeAccel = Age acceleration based on the PhenoAge epigenetic clock.
Biological aging and time since diagnosis
To examine the persistence of associations, the breast cancer survivors were stratified by time of breast cancer diagnosis relative to the second blood draw (>4 years before the second blood draw vs within 4 years). Compared with cancer-free participants, associations with the biological aging metrics were similar for women diagnosed with breast cancer greater than 4 years before the follow-up visit as those diagnosed closer to the second blood draw (Figure 3). In a case-only analysis, there were no associations between the biological aging metrics at the second timepoint and continuous years since diagnosis (Supplementary Figure 4, available online).
Case-control analysis of biological aging and breast cancer status, stratified by timing of breast cancer diagnosis. Results from multivariable linear regression models treating the biological aging metric at the second timepoint as the dependent variable and breast cancer status as an independent variable. Models are stratified by timing of breast cancer diagnosis: greater than 4 years until the second blood draw vs within 4 years. Referent groups were the women who remained breast cancer–free. Models adjusted for baseline age, baseline biological aging metric, follow-up time, and race. DunedinPACE = Dunedin Pace of Aging Calculated from the Epigenome; GrimAgeAccel = Age acceleration based on the GrimAge epigenetic clock; PhenoAgeAccel = Age acceleration based on the PhenoAge epigenetic clock.
Breast cancer therapies and biological aging
Comparisons of cancer-free participants to breast cancer survivors grouped by the different combinations of treatments received suggested that surgery had little association with the biological aging metrics but that the women receiving other treatments, particularly radiation, had increases in all 3 biological aging metrics (Supplementary Figure 5, available online).
We used case-only analysis to assess the individual associations of chemotherapy, radiation therapy, and endocrine therapy on the biological aging metric at the follow-up visit. After adjustment for the baseline biological aging metric, baseline chronological age, follow-up time, and self-reported race and in models that simultaneously included variables for receiving any of the 3 types of nonsurgical therapy, increases in biological aging were most pronounced for women who received radiation therapy (PhenoAgeAccel: β = 0.39, 95% CI = 0.19 to 0.59; P < .001; GrimAgeAccel: β = 0.29, 95% CI = 0.10 to 0.47; P = .002; DunedinPACE: β = 0.25, 95% CI = 0.02 to 0.48; P = .03; Figure 4). Associations with radiation therapy were stronger using the PC-based PhenoAgeAccel and GrimAgeAccel metrics (PC-PhenoAgeAccel: β = 0.48, 95% CI = 0.26 to 0.70; P < .001; PC-GrimAgeAccel: β = 0.38, 95% CI = 0.18 to 0.57; P < .001; Supplementary Table 6, available online). Adjusting for changes in leukocyte composition attenuated association estimates with radiation (PhenoAgeAccel: β = 0.17, 95% CI = 0.01 to 0.33; P = .04; GrimAgeAccel: β = 0.10, 95% CI = −0.03 to 0.24; P = .13; DunedinPACE: β = 0.11, 95% CI = −0.07 to 0.28; P = .23) but strengthened associations between endocrine therapy and DunedinPACE (β = 0.28, 95% CI = 0.12 to 0.44; P = .001; Supplementary Table 7, available online). Associations were relatively unchanged by adjustment for tumor ER status and stage (Supplementary Table 8, available online). Radiation therapy also had the largest associations when the therapy types were treated as independent variables in separate models (Supplementary Table 9, available online). When associations were stratified by self-reported race, chemotherapy associations with GrimAgeAccel were appreciably stronger in Black women (Black women: β = 0.39, 95% CI = 0.04 to 0.73; P = .03; White women: β = −0.01, 95% CI = −0.22 to 0.21; P = .96; Supplementary Table 10, available online). Chemotherapy associations with GrimAgeAccel were also stronger in women diagnosed within 4 years of the second blood draw visit than those diagnosed greater than 4 years before the second blood draw (within 4 years: β = 0.24, 95% CI = −0.03 to 0.50; P = .08; >4 years: β = −0.14, 95% CI = −0.41 to 0.12; P = .29; Supplementary Table 11, available online). In contrast, endocrine therapy associations with DunedinPACE were stronger in women diagnosed greater than 4 years before the second blood draw than those diagnosed at closer timepoints (within 4 years: β = −0.04, 95% CI = −0.36 to 0.28; P = .79; >4 years: β = 0.45, 95% CI = 0.08 to 0.83; P = .02; Supplementary Table 11, available online).
Case-only analysis of biological aging and breast cancer therapies. Results from multivariable linear regression models treating the biological aging metric at the second timepoint as the dependent variable and whether or not the women received chemotherapy, radiation therapy, or endocrine therapy as part of treatment as independent variables. All 3 treatment variables are included in the model simultaneously. Models adjusted for baseline age, baseline biological aging metric, follow-up time, and race. C = chemotherapy; DunedinPACE = Dunedin Pace of Aging Calculated from the Epigenome; E = endocrine therapy; GrimAgeAccel = Age acceleration based on the GrimAge epigenetic clock; PhenoAgeAccel = Age acceleration based on the PhenoAge epigenetic clock; R = radiation therapy.
Discussion
Using paired blood samples collected an average of 7.7 years apart from 417 women, we examined whether an intervening breast cancer diagnosis and treatment were associated with biological aging. Compared with women who remained cancer-free, breast cancer survivors had increases in 3 different DNAm-based metrics of biological aging. In a case-only analysis to evaluate associations with receipt of chemotherapy, radiation therapy, and endocrine therapies, the greatest increases in biological aging were found among women treated with radiation.
Breast cancer survivors experience higher rates of various age-related diseases, giving rise to the hypothesis that cancer survivors experience faster rates of aging (3-7). Case-control studies have provided some support for this hypothesis, finding that breast cancer survivors exhibit acceleration in various age-related biological endpoints including shorter telomeres, lower telomerase activity, higher DNA damage, more inflammation, and greater cellular senescence (37-39). Alterations to DNAm have also been proposed to underlie aging (9); the epigenetic clocks used in this study were designed to reflect age-related disease and mortality risk and aging rates (10-12).
Because women who subsequently develop cancer may already have elevated biological aging metrics before diagnosis (17,18), in the current study, we adjusted for values at initial blood draw before examining associations between intervening breast cancer and biological aging. Using this approach, we found that at the follow-up visit the women who had been diagnosed and treated for breast cancer between the blood draws had higher epigenetic age acceleration (PhenoAgeAccel and GrimAgeAccel) and aging rates (DunedinPACE) than the breast cancer–free participants. Small differences in the biological aging metrics were observed based on the survivor’s race and tumor ER status, and adjustment for changes in leukocyte composition between samplings generally reduced association estimates. We note the associations appeared stronger in the analyses of the principal component–based metrics. In the analysis of the persistence of the associations between breast cancer and accelerated biological aging, women diagnosed greater than 4 years before the second blood draw had increases that were similar to those diagnosed closer. Further, there was no relationship between any of the biological aging metrics and time since diagnosis. These findings provide support for the hypothesis that breast cancer survivors have lasting increases in biological aging that persist for years after diagnosis and treatment.
Previously published studies have found evidence that cancer treatments affect DNAm profiles. A study using pre- and postchemotherapy samples collected from 93 women with early-stage breast cancer approximately 4 months apart reported differences in blood DNAm profiles (40). Although the study did not investigate associations with biological aging metrics, investigators found that more than 4% of loci on the Infinium HumanMethylation450 BeadChip had statistically significant changes from chemotherapy. Notably, few changes remained statistically significant after adjustment for leukocyte composition suggesting that chemotherapy effects on the blood epigenome may be primarily mediated by changes in proportions of different white blood cell subtypes. Another study of 72 breast cancer patients treated with chemotherapy and radiation reported that compared with pretreatment samples, posttreatment samples collected approximately 18 months later had increases in both PhenoAgeAccel and GrimAgeAccel (41). Finally, a longitudinal study of head and neck cancer patients undergoing chemotherapy and radiation found dramatic increases in PhenoAgeAccel immediately after finishing therapy with a return to pretreatment levels within 6 months (42).
The large and heterogenous sample of breast cancer survivors in our study allowed us to assess biological aging associations with different treatment classes and to examine persistence of associations over a longer time course. Of the 3 treatment classes, only radiation had positive associations with all 3 biological aging metrics. In sensitivity analyses, these associations remained remarkably robust to adjustment or stratification by tumor ER status, stage, race, and time since diagnosis. However, associations with radiation were attenuated when models were adjusted for change in leukocyte composition between the blood draws. These findings lend support to the hypothesis that radiation treatment has long-term consequences for leukocyte composition and biological aging metrics and may further inform discussion of limiting radiation treatment for breast cancer when possible (43). We also note that although clinical decisions on extending the duration of endocrine therapy are complex (44), we find little evidence of association with PhenoAgeAccel or GrimAgeAccel, the 2 metrics with the strongest associations with age-related disease risk (10,11).
All women enrolled in the Sister Study cohort have a family history of breast cancer, which may limit the generalizability of our findings. Small sample size limited our ability to examine specific treatment-related factors such as specific chemotherapy agents or radiation field or evaluate dose-response relationships. We also cannot exclude the possibility that the observed changes in biological aging we observed were driven, at least in part, by postdiagnosis changes in nontreatment-related factors such as stress and lifestyle changes (25,45). Despite these limitations, this study’s strengths include its prospective design, longitudinal collection of blood samples from breast cancer survivors and cancer-free women, and its examination of 3 different epigenetic metrics of biologic aging.
In summary, we find that women diagnosed and treated for breast cancer experience greater increases in biological aging than women who remained breast cancer–free. In contrast with earlier studies that examined relatively short-term changes, we find the increases can be detected years after treatment and that these increases were most strongly associated with a history of radiation therapy.
Data availability
Code and a limited dataset for replication purposes can be requested via the Sister Study website: https://sisterstudy.niehs.nih.gov/English/coll-data.htm.
Author contributions
Jacob K Kresovich, PhD (Conceptualization; Formal analysis; Investigation; Methodology; Writing—original draft; Writing—review & editing), Katie M. O’Brien, PhD (Formal analysis; Methodology; Supervision; Writing—review & editing), Zongli Xu, PhD (Data curation; Methodology; Writing—review & editing), Clarice R. Weinberg, PhD (Conceptualization; Data curation; Supervision; Writing—review & editing), Dale P. Sandler, PhD (Conceptualization; Data curation; Funding acquisition; Investigation; Supervision; Writing—review & editing), and Jack A. Taylor, PhD MD (Conceptualization; Data curation; Supervision; Writing—original draft; Writing—review & editing).
Funding
This research was supported by the Intramural Research Program of the National Institutes of Health (Z01-ES049033, Z01-ES049032, Z01-ES044005).
Conflicts of interest
The authors report no competing interests.
Acknowledgements
The study sponsor had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication.
