Pre-diagnosis aspirin use, DNA methylation, and mortality after breast cancer: a population-based study

Abstract

BACKGROUND:

We hypothesized that epigenetic changes may help to clarify the underlying biologic mechanism linking aspirin use to breast cancer prognosis. Ours is the first epidemiologic study to examine whether global methylation and/or tumor promoter methylation of breast cancer-related genes interact with aspirin use to impact mortality after breast cancer.

METHODS:

Pre-diagnosis aspirin use was assessed through in-person interviews within the population-based cohort of 1,508 women diagnosed with first primary breast cancer in 1996-1997. Global methylation in peripheral blood was assessed by long interspersed elements-1 (LINE-1) and the luminometric methylation assay. Promoter methylation of thirteen breast cancer-related genes was measured in tumor by methylation-specific PCR and Methyl Light. Vital status was determined by the National Death Index through December 31, 2014 (N=237/597 breast cancer-specific/all-cause deaths identified). We used Cox proportional hazards regression to estimate hazard ratios (HRs) and 95% confidence intervals (95%CIs), and the likelihood ratio test to evaluate multiplicative interaction.

RESULTS:

All-cause mortality was elevated among aspirin users with methylated promotor of BRCA1 (HR=1.67; 95%CI=1.26–2.22), but not among those with unmethylated BRCA1 (HR=0.99; 95%CI=0.67–1.45) (p_interaction≤0.05). Decreased breast cancer-specific mortality was found among aspirin users with unmethylated promotor of BRCA1 and PR, and LINE-1 global hypermethylation (HR=0.60, 0.78, and 0.63, respectively; p_interaction≤0.05), although the 95%CIs included the null.

CONCLUSIONS:

Our current study suggests that the LINE-1 global methylation, and promoter methylation of BRCA1 and PR in tumor may interact with aspirin use to influence mortality following breast cancer.

INTRODUCTION

Breast cancer (BC) is the second leading cause of cancer death among women in the United States (US), with 41,760 BC deaths expected to occur in 2019.¹ Laboratory and epidemiological studies^2–5 have shown that aspirin, reduces the risk of breast cancer development due to its ability to inhibit cyclooxygenase-2 (COX-2)-mediated prostaglandin synthesis, which plays a crucial role in inflammation and estrogen biosynthesis.^{6, 7} However, the underlying biological mechanisms and epidemiological findings on aspirin use in relation to prognosis/mortality after BC are limited and inconsistent. We recently reported no association between aspirin use and mortality after breast cancer ⁸, which is consistent with a recent meta-analysis that failed to find an inverse association between pre-diagnostic aspirin use and either all-cause or BC-specific mortality.⁹

BC progression is most likely a multifactorial condition involving complex interactions among genetic/epigenetic and estrogen- and inflammation related factors. We therefore hypothesized a priori that the inconsistent results for aspirin in relation to mortality could be due to differences in the association by DNA methylation status. Epigenetic modifications involve changes in gene function but do not entail a change in DNA sequence.¹⁰ Methylation may occur during embryogenesis, is heritable by somatic cells after cell division,¹¹ but also can be modified during the life course under environmental stimulation and lifestyle modulation.^12–16 Global and tumor gene-specific aberrant DNA methylation have been associated with breast cancer prognosis in previous reports, including our own.^17–19 Aberrant DNA methylation may lead to whole genomic instability and altered gene transcription,¹⁵ which may further induce increased mutation rates of key genes (including COX-2) that aspirin acts on.^{20, 21} Thus, it is biologically plausible that aspirin works in conjunction with DNA methylation state to influence mortality after BC diagnosis.

The objective of our study was to examine whether pre-diagnosis aspirin use interacts with two global methylation markers in peripheral blood DNA and promoter methylation of a panel of 13 breast cancer-related genes in tumors (APC, BRCA1, CDH1, CYCLIND2, DAPK1, ESR1, GSTP1, HIN, CDKN2A, PR, RARβ RASSF1A, and TWIST1) to influence mortality after BC. To our knowledge, no study has systematically addressed the interplay between aspirin use and DNA methylation in the context of BC progression. However, our BC-focused hypothesis is supported by a recent study of colon cancer, which found interactions between aspirin use and methylation profiles in association with polyp occurrence.¹³ Our study may help to identify women who, may benefit from aspirin use to improve survival after BC diagnosis because of their DNA methylation profile.

METHODS

Study Design.

We used resources from follow-up component of a population-based study, the Long Island Breast Cancer Study Project (LIBCSP). Details of the LIBCSP baseline²² and follow-up ²³ studies have been described in detail elsewhere. Institutional Review Board approval was obtained from participating institutions.

Study Population.

Eligible subjects were English-speaking adult women newly diagnosed with first primary in situ or invasive BC between August 1, 1996, and July 31, 1997, and were residents of either Nassau or Suffolk counties on Long Island, NY at the time of diagnosis. There were no age or race restrictions. Women with BC were identified using rapid case ascertainment through daily/weekly contact with the pathology departments of 31 hospitals in Long Island and New York City. Subjects’ physicians were contacted to confirm BC diagnoses and obtain permission to contact their patients. Participants who completed the baseline interview included 1,508 women with BC (82.1% of eligible BC patients), of which 1,273 had invasive BC. Respondents ranged in age from 20 to 98 years, and self-reported their race as white (93%), African American (5%), or other (2%), consistent with the race distribution in Nassau and Suffolk counties at the time of data collection.²²

Assessment of Pre-Diagnosis Aspirin Use.

Trained interviewers administered the 100-minute standardized baseline questionnaire shortly after diagnosis (mean 96 days).²² All exposure information was truncated to 12 months prior to the date of diagnosis. Participants were shown a card displaying commonly used prescription and over-the-counter medications by brand name; women reporting use of medications containing aspirin for at least once a week for 6 months or longer before diagnosis were defined as ever users, and others were defined as never users (reference group) (aspirin ever/never). Participants with complete responses in this section of the questionnaire included 1,442 women with BC (96%).

Assessment of Other Epidemiologic/Clinical Factors.

At baseline, respondents were also asked about socio-demographic characteristics, medical and medication history, family history of cancer, menstrual history, use of exogenous hormones, reproductive history, body size, physical activity, cigarette smoking, alcohol intake, and other factors before date of diagnosis. Dietary intake in the year prior to the interview was assessed using a validated, 101-item modified Block food frequency questionnaire.²⁴ Medical records were abstracted, at baseline and again about five years later, to obtain information on tumor characteristics and first course of treatment for the first primary BC.²²

Assessment of Gene-Specific Promoter Methylation.

As described previously,²⁵ BC tissue blocks were successfully retrieved for 975 BC patients (67.2%), and of these, adequate tissue was available for laboratory analyses for 859 subjects (89.3%).¹⁸ Most socio-demographic and clinicopathological features were similar between patients with or without tumor block available for methylation analysis.^{25, 26} Tumor DNA was isolated and extracted from the archived formalin-fixed, paraffin-embedded tumor tissue using the method we previously published.²⁵

We selected a panel of 13 genes that were known to play a key role in mammary gland carcinogenesis (APC, BRCA1, CDH1, CYCLIND2, DAPK1, ESR1, GSTP1, HIN, CDKN2A, PR, RARβ, RASSF1A, and TWIST1).^27–35 Briefly, methylation-specific (MSP) polymerase chain reaction (PCR) was used to determine promoter methylation of BRCA1, ER and PR genes.²⁵ This assay outputs dichotomous outcomes (methylated vs. unmethylated) directly with DNA considered methylated if a PCR product using methylated-specific primers was visualized while a PCR product using unmethylated-specific primers is absent, and vice versa. The methylation status of the remaining ten genes were measured by MethyLight assay.^{36, 37} The percentage of methylation was calculated by the 2^−ΔΔCT method, where ΔΔCT = (C_T,Target − C_T,Actin)_sample − (C_T,Target − C_T,Actin)_{fully methylated DNA} and multiplying by 100.³⁸ These continuous values of the ten gene-specific methylation levels were further dichotomized with ≥4% was defined as methylated.^{14–16, 18, 26, 37}

Assessment of Global Methylation.

DNA was also isolated from participant blood samples that were collected by nurse/phlebotomists from 1,102 (73.1%) of women with BC at the time of the baseline interview.^{22, 26} Approximately two-thirds of the samples were obtained prior to chemotherapy. As previously described,³⁹ we used two independent but complementary assays, analysis of long interspersed elements-1 (LINE-1) and the luminometric methylation assay (LUMA) to measure “global methylation content”. The first marker of LINE-1 methylation ⁴⁰ approximates global methylation levels of repetitive elements or transposons, which play key roles in maintaining genomic stability,⁴¹ while LUMA measures levels of 5-methylcytosine (^mC) in the C^mCGG motif, which is over-represented in gene promoters.³⁹ LINE-1 levels were expressed directly as an overall percentage 5^mC status from system. LUMA levels were expressed as a percentage based on the following equation: methylation (%) = [1 − (HpaII ΣG/ΣT)/(MspI ΣG/ΣT)] *100guanine, where ΣG represents the sum of guanine (G) peak heights and ΣT represents the sum of thymine (T) peak heights in HpaII- and MspI-digested DNA samples, respectively.¹⁵

Outcome Assessment.

Women with BC who participated in the baseline interview (n=1,508) continue to be followed to determine mortality, including dates and cause of death, using the National Death Index (NDI).⁴² The International Statistical Classification of Diseases codes 174.9 and C-50.9 were used to identify breast cancer-related deaths. The median duration of follow-up was 17.6 years (range, 0.2-18.4). Among our cohort of 1,266 BC patients with information on aspirin use available, 476 (37.6%) deaths occurred, of which 202 (15.9%) were related to BC.

Statistical Analysis.

We constructed Kaplan-Meier Survival curves ⁴³ and estimated hazard ratios (HR) and 95% confidence intervals (CI) using Cox proportional hazards regression ⁴⁴ for the associations with ever use of aspirin in relation to all-cause and BC-specific mortality after diagnosis over ~18 years of follow-up. Interaction terms with exposure variables and log-time were tested to examine the proportional hazards assumptions.⁴³ No violations of this assumption were observed, indicating that the results between aspirin use and mortality after BC did not vary substantially over time.

Potential confounders of the aspirin-mortality associations were selected based on a directed acyclic graph.⁴⁵ The final minimally sufficient set included age at diagnosis (continuous), race (non-white/white), cigarette smoking (current/former/never smokers) and fruit and vegetable intake (≥35/0-34 servings/week). Because pre-diagnosis aspirin use is associated with less aggressive BC,⁴⁶ stage and treatment are likely causal mediators between pre-diagnosis aspirin use and mortality among women with BC; consequently, to reduce bias,^{45, 47} we did not include stage and treatment in our models.

To examine potential effect measure modification of the aspirin-mortality association by DNA methylation status, unmethylated status of gene-specific methylation was chosen as the reference group. Additionally, global methylation levels were dichotomized at the median.^{22, 26} Given LUMA, a global measurement of promoter methylation, was positively associated with overall gene control,³⁹ we used the level <median as the referent group. In contrast, LINE-1 hypomethylation is hypothesized to represent decreased genomic integrity, and therefore, LINE-1 level ≥median (hypermethylation) was selected as the referent category. We formally evaluated modification on the multiplicative scale using the likelihood ratio test (LRT) by comparing nested models with and without interaction terms with significance criteria of 0.05.⁴⁸ All analyses were completed in SAS 9.4 (Cary, NC).

RESULTS

Ever use of aspirin was reported by 301 of 1,442 BC patients (20.9%) (Table 1). Most women in the population-based LIBCSP cohort of BC patients were white (93%), postmenopausal (68%), former or current smokers (55%) and consumed <34 servings/week of fruits and vegetables (64%). Compared with never users, aspirin ever users were more likely to be older, postmenopausal, overweight/obese at diagnosis and consume more fruit/vegetables.

Table 1.

Distribution of selected participant characteristics, overall and stratified by pre-diagnosis aspirin use among women with breast cancer, LIBCSP, 1996-1997

Characteristics	No. of Women with Breast Cancer (%)
	Overall (N=1,442)	Aspirin users (N=301)	Aspirin non-users (N=1,441)
	n (%)	n (%)	n (%)
Age at diagnosis (years)
<35	39 (2.6)	5 (1.7)	34 (3.0)
35-44	173 (12.0)	21 (7.0)	152 (13.3)
45-54	379 (26.3)	67 (22.3)	312 (27.3)
55-64	356 (24.7)	75 (25.2)	281 (24.6)
65-74	345 (23.9)	95 (31.6)	250 (21.9)
≥75	150 (10.4)	38 (7.1)	112 (9.8)
Race
White	1,354 (94.0)	289 (96.0)	1,065 (93.5)
Black	62 (4.3)	11 (3.7)	51 (4.5)
Other	24 (1.7)	11 (0.3)	23 (2.0)
Menopausal status
Premenopausal	451 (31.9)	73 (24.9)	378 (33.8)
Postmenopausal	962 (68.1)	220 (75.1)	742 (66.3)
BMI at diagnosis (kg/m2)
<25	655 (45.9)	120 (40.4)	535 (47.4)
25-30	454 (31.8)	97 (32.7)	257 (31.6)
>30	317 (22.2)	80 (26.9)	237 (21.0)
Smoking
Never	645 (44.7)	137 (45.5)	508 (44.5)
Current	276 (19.1)	52 (17.3)	224 (19.6)
Past/former	521 (36.1)	112 (37.2)	409 (35.9)
Fruits and vegetables intake
0-34 servings/week	910 (64.3)	178 (59.5)	732 (65.5)
35+ servings/week	506 (35.7)	121 (40.5)	385 (34.5)

As previously reported,⁸ among women with BC, over ~18 years of follow-up the hazard of mortality for aspirin users, compared to never users, was decreased by 13% for BC-specific mortality (HR=0.87, 95% CI=0.59-1.29), but increased by 21% for all-cause mortality (HR=1.21, 95% 0=0.99-1.48) (Supplemental Table 1). When stratified by global methylation status, we observed significant multiplicative interaction between aspirin use, LINE-1 global methylation, and BC-specific mortality among women with BC (p_interaction≥0.05) (Table 2). Specifically, in women with LINE-1 hypomethylation, ever aspirin use was associated with a higher risk of BC-specific mortality (HR=1.45, 95% CI=0.86–2.42; p_interaction=0.05) whereas in women with LINE-1 hypermethylation, ever aspirin use was associated with a lower risk of BC-specific mortality (HR=0.63, 95% 0=0.33–1.20). There was no apparent interaction with LINE-1 for all-cause mortality (p_interaction=0.44), and we also found no statistically significant interaction between aspirin use and LUMA methylation in relation to BC-specific (p_interaction=0.38) or all-cause mortality (p_interaction=0.94). Similar patterns of association were also identified among women with hormone receptor positive or invasive breast cancer (supplemental tables 2 and 3).

Table 2.

Multivariable-adjusted^a hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between pre-diagnostic aspirin use and mortality (breast cancer-specific and all-cause) stratified by global methylation status (measured by LINE-1 and LUMA^b) among all women (with blood sample), in the Long Island Breast Cancer Study Project (N=1,026), 1996-1997.

	Breast Cancer-Specific Mortality								All-Cause Mortality
Global marker Aspirin categories	Deaths	PYb	HR	95% CI	Deaths	PYb	HR	95% CI	Deaths	PYb	HR	95% CI	Deaths	PYb	HR	95% CI
LINE-1 methylation		< Median (78.65)				≥ Median				< Median (78.65)				≥ Median
Never users	62	5877	1.00	reference	61	5875	1.00	reference	156	5877	1.00	reference	144	5875	1.00	reference
Ever users	21	1429	1.45	(0.86, 2.42)	11	1670	0.63	(0.33, 1.20)	54	1429	1.32	(0.98, 1.79)	48	1670	1.06	(0.77, 1.47)
p interaction	0.05								0.44
LUMA methylation		< Median (0.56)				≥ Median				< Median (0.56)				≥ Median
Never users	47	3905	1.00	reference	75	7717	1.00	reference	107	3905	1.00	reference	189	7717	1.00	reference
Ever users	16	1122	1.10	(0.62, 1.96)	16	1946	0.88	(0.51, 1.53)	39	1122	1.08	(0.76, 1.54)	61	1946	1.27	(0.96, 1.68)
p interaction	0.38								0.94

^aAdjusted for baseline age, race, fruit and vegetable intake, cigarette smoking

^bLINE-1, long interspersed elements-1; LUMA, luminometric methylation assay

^cPY, person-years

As shown in Table 3, the association between aspirin use and mortality among women with BC was modified significantly by the tumor promoter methylation status of BRCA1 and PR (p_interaction≤0.05), but not with the other 11 gene promoters considered (although effect estimates were not estimated when cell counts were <5). Specifically, in aspirin users with unmethylated tumor BRCA1 promoter, BC-specific mortality was lower (HR=0.60, 95% CI=0.25–1.45), but the corresponding effect estimate in women with methylated BRCA1 was above the null value (HR=1.16, 95% CI=0.69–1.95; p_interaction=0.04). For all-cause mortality, a positive association was observed in women with a methylated BRCA1 (HR=1.67, 95% CI=1.26–2.22), but not among women with an unmethylated BRCA1 (HR=0.99, 95% CI=0.67–1.45; p_interaction=0.02). For tumor methylation of the PR, we observed a higher risk of BC-specific mortality among women with a methylated PR promoter (HR=1.63, 95% CI=0.68–3.90), whereas the corresponding hazard for an unmethylated PR promoter was below the null (HR=0.78, 95% CI=0.46–1.33; p_interaction=0.03). No statistically significant interaction was observed between aspirin use and PR promoter methylation on all-cause mortality (p_interaction=0.19). Corresponding findings restricted to hormone receptor positive or invasive breast cancer did not vary substantially from those among women overall (supplemental tables 2 and 3). Consistent patterns of the associations were also observed after a sensitivity analysis by adjusting for global methylation (supplemental table 4).

Table 3.

Multivariable-adjusted^a hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between pre-diagnostic aspirin use and mortality (breast cancer-specific and all-cause) stratified by gene methylation status (methylated vs. unmethylated tumors) among all women (with tissue sample), in the Long Island Breast Cancer Study Project (N=821), 1996-1997.

	Breast Cancer-Specific Mortality								All-Cause Mortality
Gene promoter Aspirin categories	Unmethylated				Methylated				Unmethylated				Methylated
	Deaths	PYb	HR	95% CI	Deaths	PYb	HR	95% CI	Deaths	PYb	HR	95% CI	Deaths	PYb	HR	95% CI
APC
Never users	45	4607	1.00	reference	62	4049	1.00	reference	121	4607	1.00	reference	127	4049	1.00	reference
Ever users	8	1074	0.79	(0.37, 1.67)	15	1006	1.03	(0.58, 1.84)	38	1074	1.19	(0.85, 1.68)	47	1006	1.48	(1.08, 2.04)
p interaction	0.12								0.39
BRCA1
Never users	34	3601	1.00	reference	76	5548	1.00	reference	104	3601	1.00	reference	156	5548	1.00	reference
Ever users	6	1157	0.60	(0.25, 1.45)	19	1147	1.16	(0.69, 1.95)	32	1157	0.99	(0.67, 1.45)	60	1147	1.67	(1.26, 2.22)
p interaction	0.04								0.02
CDH1
Never users	92	7753	1.00	reference	7	413	1.00	reference	217	7753	1.00	reference	18	413	1.00	reference
Ever users	21	1986	0.89	(0.55, 1.44)	N.A.	N.A.	N.A.	N.A.	81	1986	1.35	(1.06, 1.72)	N.A.	1.21	N.A.	N.A.
p interaction	N.A.								N.A.
CYCLIND2
Never users	75	6272	1.00	reference	24	1444	1.00	reference	183	6272	1.00	reference	52	1444	1.00	reference
Ever users	16	1641	0.85	(0.49, 1.47)	5	466	0.73	(0.27, 1.96)	58	1641	1.21	(0.91, 1.61)	26	466	1.56	(1.02, 2.39)
p interaction	0.99								0.49
DAPK
Never users	81	7157	1.00	reference	18	1009	1.00	reference	196	7157	1.00	reference	39	1009	1.00	reference
Ever users	15	1715	0.76	(0.44, 1.32)	6	391	0.89	(0.33, 2.40)	68	1715	1.32	(1.01, 1.72)	16	391	1.25	(0.73, 2.15)
p interaction	0.54								0.99
ESR1
Never users	58	5033	1.00	reference	52	4017	1.00	reference	141	5033	1.00	reference	118	4017	1.00	reference
Ever users	12	1220	0.87	(0.47, 1.62)	13	1084	0.98	(0.52, 1.83)	46	1220	1.37	(0.99, 1.90)	46	1084	1.27	(0.92, 1.76)
p interaction	0.91								0.99
GSTP1
Never users	53	6084	1.00	reference	46	2082	1.00	reference	155	6084	1.00	reference	80	2082	1.00	reference
Ever users	18	1508	1.30	(0.76, 2.23)	N.A.	N.A.	N.A.	N.A.	59	1508	1.47	(1.09, 1.98)	25	599	0.97	(0.65, 1.45)
p interaction	N.A.								0.98
HIN
Never users	29	3095	1.00	reference	70	5071	1.00	reference	82	3095	1.00	reference	153	5071	1.00	reference
Ever users	11	763	1.58	(0.77, 3.27)	10	1343	0.53	(0.27, 1.02)	33	763	1.63	(1.09, 2.44)	51	1343	1.14	(0.85, 1.53)
p interaction	0.32								0.86
P16
Never users	96	8090	1.00	reference	6	317	1.00	reference	231	8090	1.00	reference	8	317	1.00	reference
Ever users	20	1954	0.88	(0.54, 1.43)	N.A.	N.A.	N.A.	N.A.	80	1954	1.35	(1.06, 1.72)	4	50	2.76	(0.82, 9.27)
p interaction	N.A.								0.52
PR
Never users	92	8042	1.00	reference	18	1107	1.00	reference	232	8042	1.00	reference	28	1107	1.00	reference
Ever users	17	2038	0.78	(0.46, 1.33)	8	266	1.63	(0.68, 3.90)	75	2038	1.25	(0.97, 1.60)	17	266	2.14	(1.18, 3.90)
p interaction	0.03								0.19
RARB
Never users	60	5957	1.00	reference	39	2209	1.00	reference	161	5957	1.00	reference	74	2209	1.00	reference
Ever users	16	1500	0.98	(0.56, 1.72)	5	556	0.61	(0.24, 1.52)	64	1500	1.43	(1.07, 1.89)	20	556	1.12	(0.71, 1.75)
p interaction	0.97								0.83
RASSF1A
Never users	9	1259	1.00	reference	90	6907	1.00	reference	26	1259	1.00	reference	209	6907	1.00	reference
Ever users	3	279	1.76	(0.49, 6.33)	18	1828	0.75	(0.45, 1.26)	11	279	1.91	(0.99, 3.69)	73	1828	1.22	(0.95, 1.58)
p interaction	0.99								0.96
TWIST1
Never users	82	7083	1.00	reference	17	1084	1.00	reference	196	7083	1.00	reference	39	1084	1.00	reference
Ever users	14	1743	0.65	(0.37, 1.15)	7	364	1.71	(0.67, 4.34)	68	1743	1.34	(1.03, 1.73)	16	364	1.20	(0.67, 2.15)
p interaction	0.23								0.99

^aAdjusted for baseline age, race, fruit and vegetable intake, cigarette smoking

^bPY, person-years; N.A. not applicable because cell sizes were less than 5

DISCUSSION

In our population-based cohort of women with breast cancer, all-cause mortality after BC was elevated among aspirin ever-users with methylated tumor promotor of BRCA1, but not those with unmethylated tumors. BC-specific mortality was lower among aspirin users with unmethylated tumor promotor of BRCA1 and PR, and hypermethylation of LINE-1, although the corresponding 95% CIs included the null value. These interactions between aspirin use and DNA methylation status were statistically significant on a multiplicative scale (p<0.05). Our findings suggest that the association between aspirin use and mortality after BC may depend upon methylation profiles and warrant further investigation.

Improved survival with pre-diagnosis aspirin use among women with BC has been observed in several epidemiologic studies,^{49, 50} but not others.^{8, 51} Explanations for the inconsistent findings between studies are unclear. However, consideration of DNA methylation profiles as potential modifiers of the aspirin-mortality association may provide new biologic insights. Aspirin is an effective analgesic, antipyretic, and anti-inflammatory drug. It targets the COX-2 enzymes to interrupt prostaglandin and estrogen synthesis, which further modulate apoptosis, cell proliferation, angiogenesis, and immune surveillance.^{52, 53} Numerous studies, including our own, have demonstrated positive correlations between aberrant DNA methylation of BC-related gene promoters and lower survival rates among BC patients.^{18, 19, 25} The epigenetic inactivation of inflammation and hormone related genes by gene-specific promoter methylation and genomic instability by global methylation may be regulated by exogenous factors with overlapping pathways. Thus, it appears biologically plausible that the association between pre-diagnosis aspirin use and mortality may be altered by aberrant methylation changes.

We are first to report the multiplicative interactions between aspirin use and tumor promoter methylation of BRCA1 and PR on BC prognosis. BRCA1 is a tumor suppressor gene and part of the DNA repair complex, which plays an important role in maintaining genomic stability and controlling cell-cycle checkpoints.²⁵ PR is a member of the nuclear receptor family, which is involved in normal breast development and tumorigenesis by mediating the physiological effects of progesterone and affecting cellular proliferation/differentiation.⁵⁴ Inactivation (silencing) of BRCA1 and PR is thought to occur via gene promoter hypermethylation,^{28, 55} which appears to impact the efficiency of DNA repair processes, drive dysregulated cell proliferation, and reduce chromosomal stability.^{56, 57} Further, laboratory studies have shown the interaction ^58–60 of DNA repair pathways with the inflammation and estrogen pathways through which aspirin mainly operates. Collectively, it is plausible that a tumor environment characterized with BRCA1 and PR promotor methylation may contribute to low sensitivity to aspirin exposure for the host, which would result in little benefit from pre-diagnosis aspirin use on breast cancer prognosis and overall survival. Additionally, methylation of BRCA1 has previously been found to correlate with large tumor size and the presence of axillary node metastases.²⁵ Thus, it is plausible that tumor promoter methylation can be linked to worse BC prognosis, which pre-diagnostic use of aspirin is not able to reverse.

Regarding the different BC-specific mortality profiles among aspirin users by LINE-1 global methylation level, our findings are consistent with our a priori hypothesis. LINE-1 serves as a surrogate for overall cellular DNA methylation status^{40, 61} A low level of LINE-1 methylation (hypomethylation) has been associated with increased chromosomal instability that may be associated with poor prognosis in epithelial cancers, including BC.^62–64 Therefore, we could infer that in the presence of LINE-1 hypermethylation (and improved genomic stability), aspirin may be able to function normally or even better in decreasing mortality due to BC. But in an environment of LINE-1 hypomethylation (and genomic instability), the risk reduction benefit from aspirin is not observed.

There are several limitations of this study. First, tumor tissue and blood specimens were not available for all BC patients identified in the parent LIBCSP study, leading to potential selection bias. However, we did not observe considerable differences in key socio-demographic and clinicopathological features between BC patients with and without available tissue or blood data.^{25, 26} Second, the low prevalence of some methylation biomarkers (i.e., low frequency of CDKN2A and CDH1 methylation) in our study sample, as reflected in the wide confidence intervals, may have contributed to our inability to detect modest interactions; thus, these results should be interpreted with caution. Third, our findings are dependent on self-reports of aspirin use, which could result in potential non-differential misclassification bias and recall error. However, our information on aspirin use was based on comprehensive in-person interviews with several memory aids, which likely improve recall. Our interview approach also resulted in similar effect sizes for the aspirin-breast cancer incidence association ⁴⁶ as reported by other studies ^{4, 65, 66} Fourth, we did not have information available regarding the dosage of aspirin, and we were unable to consider aspirin use patterns (i.e., frequency and duration), due to small cell sizes after stratification by methylation status. Fifth, our follow-up study was restricted to an examination of pre-diagnosis NSAID use. However, there is substantial uncertainty about the appropriate timing of aspirin use that will largely impact survival.⁵⁰ Finally, our LIBCSP study population is comprised primarily of white women and, therefore, generalizability of our findings to more diverse populations must be examined in future studies. However, our study results are still largely generalizable to those at highest risk of developing breast cancer in the US – white, postmenopausal women.⁶⁷

Our study also has multiple strengths including epigenetic biomarker assessments of DNA methylation in both tumor tissue and blood samples. The latter were collected from our population-based sample of U.S women within a few months following diagnosis of their first primary BC. The archived tumor tissue is from the pathology blocks of the first primary BC, during a time period (the mid-1990s) when it was not customary to treat breast cancer patients with chemotherapy prior to surgery. Participants were followed for 18+ years after diagnosis using the NDI, which provides high-quality ascertainment of vital status.⁶⁸ To the best of our knowledge, ours is the first study to examine the potential modification by tissue promotor and/or global methylation on the association between pre-diagnosis aspirin use and mortality after BC diagnosis.

In conclusion, among a population-based cohort of women diagnosed with first primary BC, we are first to report significant heterogeneity of the aspirin-mortality association by BRCA1 and PR promoter methylation, and LINE-1 global methylation profiles. These findings, if confirmed: may provide new biological insights on the association between aspirin use and BC prognosis; may impact clinical decision making by identifying a subgroup of BC patients, using epigenetic markers, for whom pre-diagnosis aspirin use impacts subsequent mortality; and may help refine risk reduction strategies to improve survival among women with BC. Future research designed to replicate our findings should include a larger sample size to allow examination of patterns of aspirin use, and an enlarged panel of genes to explore the role of genetic predisposition in driving overall genetic instability on survival after breast cancer diagnosis.

Precis for use in the Table of Contents:

Our findings suggest that the association between aspirin use and mortality after BC may depend upon patients’ DNA methylation profiles. Our study may help to identify women who, may benefit from aspirin use to improve survival after BC diagnosis because of their DNA methylation profile.