Materials and Methods

This study utilizes the population-based case-control component of the Long Island Breast Cancer Study Project (LIBCSP). Details of the parent study have been published previously [355]. Institutional Review Board approval was obtained from all participating institutions.

Study population. Cases and controls were English-speaking residents of Long Island, NY (Nassau and Suffolk counties). Cases were adult women newly diagnosed with a first primary in situ or invasive breast cancer between August 1, 1996 and July 31, 1997, and were identified using a “super-rapid” network where study personnel contacted (either 2-3 times per week or daily) hospital pathology departments. Controls were identified using Waksberg’s method of random digit dialing [393] for women under 65 years of age, and the Health Care Finance Administration rosters for women 65 years and older. Controls were frequency matched to the expected age-distribution of the case women. There were no upper age or race restrictions for subject eligibility.

The parent LIBCSP respondents included 1,508 cases and 1,556 controls.

Respondents ranged in age from 20 to 98 years of age, 67% were postmenopausal, and the majority self-reported their race as white (94%), followed by black or African American (4%), or other (2%), which is consistent with the racial population distribution of these two counties at the time of data collection [355].

administered a main questionnaire by a trained interviewer about 3 months after diagnosis for cases and 5.5 months after identification for controls. The questionnaire asked about

demographic characteristics, pregnancy history, menstrual history, hormone use, medical history, family history of cancer, body size changes, alcohol use, active and passive cigarette smoking, physical activity, occupational history, and other potential risk factors for breast cancer [355]. LIBCSP researchers have previously reported that breast cancer risk in this population is associated with known reproductive risk factors (increasing age at first birth, few or no children, little or no breastfeeding, late age at menarche) [21], and lifestyle risk factors (increasing alcohol intake and, for postmenopausal breast cancer, decreased physical activity and increased body size) [52, 394].

Approximately 98% of participants (1,479 cases and 1,520 controls) also completed the validated [359, 395, 396] self-administered 101-item modified Block food frequency questionnaire (FFQ). Participants with implausible total energy intake (± 3 standard deviations from the mean) were excluded (n = 36). Thus, 1,463 cases and 1,500 controls were included in our examination of the association between PUFA intake and breast cancer risk.

We estimated PUFA intake by linking responses from the FFQ (i.e., grams per day for each line item) with nutrient values available in the U.S. Department of Agriculture databases for ω-3 and ω-6 PUFAs [397]. The following PUFAs were estimated: (1) ω-3 fatty acids, including alpha-linolenic acid (ALA), docosapentanoic acid (DPA), docosahexanoic acid (DHA), eicosapentanoic acid (EPA); and (2) ω-6 fatty acids, including linoleic acid (LA) and arachidonic acid (AA). An estimate of total PUFA intake was calculated by combining all individual fatty acids. Additionally, an estimate of total ω-3 and ω-6 fatty

acids was obtained by summing each individual fatty acid within category (e.g., total ω-3 = ALA + DPA + DHA + EPA).

Fish and/or seafood intakes were assessed according to the following items recorded in the FFQ: (1) tuna, tuna salad, tuna casserole; (2) shell fish (shrimp, lobster, crab, oysters, etc.); and (3) other fish (either broiled/baked).

Genotyping. Eighteen polymorphisms (in fifteen genes) were selected for this analysis spanning three biologically plausible pathways for PUFA metabolism, including inflammation, oxidative stress, and estrogen metabolism pathways. Variants affecting polyphen prediction (GPX1), transcription factor binding prediction (PTGS-2 rs20417, FAS,

FASL, TNF-α, MPO, CAT, GSTA1, COMT rs737865, CYP17), miRNA binding (PTGS-2

rs5275, GPX1), 3D conformation (PPAR-α, COMT rs4680), or splicing regulation (PPAR-α,

FAS rs2234767, GPX1, GSTP1, COMT rs4680) were considered as putatively functional variants as defined in the NIEHS SNPInfo WebServer [361].

Blood samples collected from subjects at the time of the case-control interview were used as the source of DNA for the genotyping. Genotyping methods have been previously described [65, 366-370, 398]. Briefly, DNA was isolated from mononuclear cells in whole blood which was separated by Ficoll (Sigma Chemical Co., St. Louis, Missouri) in the laboratory of Dr. Regina Santella at Columbia University using standard phenol and chloroform-isoamyl alcohol extraction and RNase treatment [398]. Genotyping for inflammation genes (PTGS-2, FAS, FASL, PPAR-α, TNF-α), used the following assays: Taqman 5’-Nuclease Assay (Applied Biosystems, Foster City, CA) and AcycloPrimeTM

-FP SNP Detection Kit obtained from Perkin Elmer Life Sciences (Boston, Massachusetts, USA) [65, 367, 368]. The same assay was used for aromatase gene (CYP17) with a 10 µM probe

[369, 370]. For oxidative stress genes (CAT, MPO, MnSOD, GPX, GSTA1, GSTP1, COMT), BioServe Biotechnologies (Laurel, Maryland) performed the genotyping using high-

throughput, matrix assisted, laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry of Sequenom, Inc. (San Diego, California). Gene deletions for GSTM1 and

GSTT1 were determined by a multiplex polymerase chain reaction method, with the constitutively present gene β-globulin as an internal positive control [366].

Data were missing for some genetic polymorphisms, primarily due to laboratory failures. Thus, the final sample sizes for our examination of gene-environment interactions are PTGS-2 rs20417 and rs5275 (n = 2,106), PPAR-α rs1800206 (n = 1,815), FAS rs2234767 (n = 2,106), FAS rs1800682 (n = 2,095) FASL rs763110 (n = 2,110), TNF-α rs1800629 (n = 2,088), MnSOD rs4880 (n = 2,063), MPO rs2333227 (n = 2,078), CAT rs1001179 (n = 2,068), GPX1 rs1050450 (n = 2,074), GSTM1 deletion (n = 1,925), GSTP1 rs1695 (n = 2,040), GSTT1 deletion (n = 1,946), GSTA1 rs3957356 (n = 2,075), COMT rs4680 (n = 2,084), COMT rs737865 (n = 2,064), and CYP17 rs743572 (n = 2,044).

Tests for Hardy-Weinberg equilibrium (HWE) among the controls were conducted. Only PTGS-2 rs20417 and MPO polymorphisms deviated significantly from HWE (p < 0.05). However, the observer agreement in 8% of the randomly selected was high (PTGS-2

rs20417 kappa statistic = 0.99, MPO kappa statistic = 0.91), and the failure rate of the assay was less than 1% for both polymorphisms. Also, the genotype frequencies for both PTGS-2

rs20417 and MPO polymorphisms were reported to be similar to those observed in other studies [362, 399].

Statistical analyses. All analyses were conducted using SAS version 9.2 (SAS Institute, Cary, NC). Unconditional logistic regression was used to estimate odds ratios

(ORs) and 95% confidence intervals (95% CI) for the association between PUFA intake and breast cancer risk. All PUFA intake estimates (i.e., total PUFA, total ω-3, ALA, DPA, DHA, EPA, total ω-6, LA, AA, ratio of ω-3/ω-6) were categorized as quartiles, according to the distribution among controls. Quartiles were selected over other possible methods of

categorization (e.g., tertiles, quintiles, linear, splines) because the shape of the dose-response between PUFAs and breast cancer risk was best captured using these cut-points. Similarly, fish intake was categorized using quartiles according to the distribution among those controls who reported consuming fish (i.e., tuna, shell fish, other fish); non-consumers of fish were considered the referent group. Tests for linear trend were not conducted, since the relation between any of the PUFA measures and breast cancer risk was not strictly monotonic [373].

Interactions between total ω-3 and total ω-6 intake, and between the ω-3/ω-6 ratio and the eighteen genetic polymorphisms, in association with breast cancer risk were assessed on the additive (common referent) and multiplicative scales. Additive interaction was

evaluated using relative excess risk due to interaction (RERI), with 95% CI estimated using the Hosmer and Lemeshow method [374]. Multiplicative interactions were evaluated by comparing nested models using the Likelihood Ratio Test (LRT) [373]. Total ω-3, total ω-6, and ratio of ω-3/ω-6 were dichotomized at the median for use in the interaction models. Similarly, in order to maximize cell sample sizes, genotypes were dichotomized according to a dominant model and categorized into “high” and “low” risk groups based upon the function of the variant allele, which was determined using the existing literature (see Supplemental

Table 3.6) [219, 399-412].

We also considered effect modification of the association between PUFA intake and breast cancer risk by: menopausal status (post- vs. pre-menopausal status); and dietary

supplement use (yes/no). However, little or no heterogeneity was observed with either of these covariates, and thus the results are not shown. We also considered potential

heterogeneity across breast cancer subtypes, defined by hormone receptor status (any hormone receptor positive breast cancer vs. no hormone receptor positive breast cancer), by constructing polytomous regression models; however, no differences in the association with PUFA intake were observed across hormone receptor subtype, and thus the results are not shown.

All models were adjusted for the frequency matching factor five-year age group. Other potential confounders (including total energy intake, non-steroidal anti-inflammatory drugs (NSAID), family history of breast cancer, income, body mass index, alcohol use, fruit and vegetable intake, and physical activity) were identified using directed acyclic graph (DAG) [373]. The only covariates that changed the estimates by more than 10% were total energy intake for PUFA intake, and energy intake and NSAID use for fish intake. It is possible that chronic NSAID users experience gastrointestinal problems (e.g., stomach ulcers, reflux) which may subsequently influence diet, including fish consumption [413]. Thus, all PUFA models were adjusted for age and energy intake, and all fish intake models included age, energy intake and NSAID use.