Germline APOE ε2 homozygosity and reduced risk of breast cancer: a population-based cohort study Authors: Hwamin Woo, Suhyun Han, Jihye Heo, Juhee Cho, Do-Geun Kim, Chi-Hun Kim, Danbee Kang Published: 2026 ABSTRACT AbstractWe evaluated whether Apolipoprotein E (APOE) ε2 homozygosity, a germline variant associated with lower LDL cholesterol, is linked to reduced risk of breast cancer. We analyzed data from 234,857 cancer-free and dementia-free women in the UK Biobank. APOE genotypes were classified as ε2 homozygotes, ε2 heterozygotes, and non-ε2 carriers. Incident breast cancer was assessed via national registry linkage. Cox proportional hazards models were used to estimate adjusted hazard ratios (HRs). Subgroup analyses were conducted by lifestyle, metabolic, and female-specific factors. Over a median follow-up of 12.3 years, 7,961 first primary breast cancers occurred. Compared to non-ε2 carriers, ε2 homozygotes had significantly lower breast cancer risk (HR, 0.64; 95% CI 0.45–0.90). The protective association of ε2/ε2 was stronger in participants with smoking history. APOE ε2 homozygosity was associated with reduced risk of breast cancer, particularly under inflammatory stress. These findings suggest that genetically determined lipid and inflammatory regulation may influence breast cancer susceptibility.Keywords: APOE, breast cancer, lipid metabolism, genetic epidemiology, UK biobank Introduction Lipid metabolism is a key regulator of the tumor microenvironment [1] Lipid substrates for membrane biosynthesis, energy metabolism, and oncogenic signaling are numerous in cancer cells, and many cancers are sensitive to drugs that inhibit lipid uptake or synthesis [1]. Among various malignancies, breast cancer has been particularly shown to exhibit an intricate dependency on lipid metabolism, and as such has been highlighted for potential therapeutic intervention [2, 3]. In this perspective, germline genetic drivers of lipid metabolism could provide novel options for risk stratification and metabolic prevention interventions [4]. Among apolipoprotein family genes, Apolipoprotein E (APOE) is a key genetic determinant for lipid transportation and immune response, with three common isoforms (ε2, ε3, and ε4), that are differentially associated with the plasma lipid levels and systemic inflammation [5]. The ε2 allele is associated with decreased affinity for the low-density lipoprotein (LDL) receptor and lower levels of plasma LDL cholesterol, which may be protective against LDL-sensitive cancers such as breast cancer [6]. Previous research has demonstrated that ε2 homozygosity is associated with a decreased susceptibility for a number of disorders including thromboembolism, aneurysm, and peptic ulcer among others [7, 8]. However, results of studies on the association between APOE polymorphisms and cancer including breast cancer remains inconsistent [9]. Prior pooled analysis reported no overall effect of ε2 or ε4 on breast cancer risk, with significant associations observed only in certain subpopulations [9]. These inconsistencies might be due to grouping ε2 homozygotes together with ε2 heterozygotes in previous studies, which may hide recessive influences that belong to the group of ε2/ε2 (occurring in less than 1% of the population) [7]. Moreover, little research has examined whether these associations are modified by metabolic factors or correlated with circulating lipid levels [10]. We therefore investigated whether APOE genotype, particularly ε2/ε2 homozygosity, confers protection against breast cancer and whether this relationship is modified by metabolic factors. This study aims to evaluate APOE’s potential as a biomarker for risk stratification and metabolic prevention in breast cancer. Results Among the 234,857 eligible female participants, 198,804 (84.6%) were non-ε2 carriers, 34,530 (14.7%) were ε2 heterozygotes, and 1,523 (0.6%) were ε2 homozygotes. The mean baseline age was approximately 56 years across all groups, and most participants were of White ethnicity. Sociodemographic and behavioral characteristics including smoking status, alcohol status, and physical activity were largely comparable among the three groups, with standardized mean differences (SMDs) < 0.1 for most variables. (Table 1) Table 1Baseline characteristics of study populationNon-ε2carriers (N = 198,804)ε2 heterozygotes (N = 34,530)ε2 homozygotes (N = 1,523)SMDAge, years56.08 (8.01)56.12 (8.01)55.83 (8.20)0.024Ethnicity, n (%)0.089White186,831 (94.0)32,526 (94.2)1,427 (93.7)Asian4,514 (2.3)503 (1.5)16 (1.1)Black3,254 (1.6)830 (2.4)45 (3.0)Mixed3,374 (1.7)533 (1.5)29 (1.9)Unknown831 (0.4)138 (0.4)6 (0.4)Townsend deprivation index, n (%)0.043 < 30th59,647 (30.0)10,376 (30.0)452 (29.7)30th–70th80,314 (40.4)14,028 (40.6)601 (39.5) > 70th58,620 (29.5)10,078 (29.2)470 (30.9)Unknown223 (0.1)48 (0.1)0 (0.0)BMI category, n (%)0.025 < 2579,138 (39.8)13,402 (38.8)579 (38.0) ≥ 25118,969 (59.8)21,008 (60.8)939 (61.7)Unknown697 (0.4)120 (0.3)5 (0.3)Smoking status, n (%)0.031Never119,250 (60.0)20,684 (59.9)946 (62.1)Ever78,586 (39.5)13,669 (39.6)570 (37.4)Unknown968 (0.5)177 (0.5)7 (0.5)Alcohol status, n (%)0.023Never11,735 (5.9)1,959 (5.7)90 (5.9)Ever186,574 (93.8)32,495 (94.1)1,427 (93.7)Unknown495 (0.2)76 (0.2)6 (0.4)MET, minutes /week, n (%)0.023Low (< 600)28,676 (14.4)5,045 (14.6)229 (15.0)Moderate80,463 (40.5)14,003 (40.6)603 (39.6)High (≥ 3000)45,708 (23.0)7,791 (22.6)340 (22.3)Unknown43,957 (22.1)7,691 (22.3)351 (23.0)Laboratory findingLDL cholesterol, mmol/L3.69 (0.86)3.25 (0.78)2.65 (0.89)0.817HDL cholesterol, mmol/L1.59 (0.37)1.63 (0.40)1.64 (0.44)0.088Total cholesterol, mmol/L5.94 (1.12)5.48 (1.03)4.91 (1.24)0.597Triglyceride, mmol/L1.53 (0.84)1.59 (0.90)1.85 (1.06)0.222Glucose, mmol/L5.06 (1.07)5.06 (1.07)5.07 (1.37)0.005Medication use, n (%)Hypertension medication38,508 (19.4)6,418 (18.6)296 (19.4)0.014Diabetes mellitus medication5119 (2.6)877 (2.5)43 (2.8)0.012Lipid modifying agents32,822 (16.5)4,290 (12.4)219 (14.4)0.078Antithrombotic agents20,903 (10.5)3,389 (9.8)162 (10.6)0.018Menarche age, n (%)0.022 > 1370,653 (35.5)12,225 (35.4)519 (34.1) ≤ 13121,792 (61.3)21,200 (61.4)951 (62.4)Unknown6,359 (3.2)1,105 (3.2)53 (3.5)Number of births, n (%)0.048037,229 (18.7)6,360 (18.4)294 (19.3)126,352 (13.3)4,503 (13.0)207 (13.6)286,744 (43.6)15,286 (44.3)623 (40.9) ≥ 347,996 (24.1)8,301 (24.0)396 (26.0)Unknown483 (0.2)80 (0.2)3 (0.2)Menopausal status, n (%)0.048No49,392 (24.8)8,523 (24.7)420 (27.6)Yes118,470 (59.6)20,548 (59.5)868 (57.0)Had a hysterectomy21,851 (11.0)3,803 (11.0)168 (11.0)Other reason8,481 (4.3)1,545 (4.5)62 (4.1)Unknown610 (0.3)111 (0.3)5 (0.3)HRT use, n (%)0.010Never123,580 (62.2)21,367 (61.9)940 (61.7)Ever74,164 (37.3)12,992 (37.6)576 (37.8)Unknown1,060 (0.5)171 (0.5)7 (0.5)First live birth age, n (%)0.055 < 2013,390 (6.7)2,401 (7.0)116 (7.6)20–2446,630 (23.5)8,169 (23.7)382 (25.1)25–2950,415 (25.4)8,872 (25.7)360 (23.6) ≥ 3024,025 (12.1)4,087 (11.8)160 (10.5)Nulliparous or Unknown64,344 (32.4)11,001 (31.9)505 (33.2)Oral contraceptive intake, n (%)0.010Never37,062 (18.6)6,320 (18.3)282 (18.5)Ever160,807 (80.9)28,056 (81.3)1,235 (81.1)Unknown935 (0.5)154 (0.4)6 (0.4)Values are presented as n (%) or mean (standard deviation)BMI, body mass index; MET, total metabolic equivalent task for all activity; HDL, high density lipoprotein; LDL, low density lipoprotein; HRT, hormone-replacement therapy; SMD, standardized mean difference Baseline characteristics of study population Values are presented as n (%) or mean (standard deviation) BMI, body mass index; MET, total metabolic equivalent task for all activity; HDL, high density lipoprotein; LDL, low density lipoprotein; HRT, hormone-replacement therapy; SMD, standardized mean difference During the follow-up period (median duration = 12.3 years), 7,961 breast cancers were identified. The 15-year cumulative incidence of these cancers was 3.52% in non-ε2 carriers, 3.50% in ε2 heterozygotes, and 2.20% in ε2 homozygotes (Fig. 1). In multivariable-adjusted Cox models (Table 2), ε2 homozygotes had a significantly lower risk of breast cancers compared with non-ε2 carriers (hazard ratio [HR], 0.64; 95% confidence interval [CI], 0.45–0.90). Kaplan–Meier survival data between APOE ε2 and breast cancer incidence probability Table 2Comparisons of breast cancer outcome according to type of APOE ε2Non-ε2 carriers(N = 198,804)ε2 heterozygotes(N = 34,530)ε2 homozygotes(N = 1,523)Breast cancerNumber of cases6,7571,17133Incidence rate, per 1,000 person-years2.832.831.79Model 1 HR (95% CI)Reference1.00 (0.94–1.06)0.64 (0.45–0.89)Model 2 HR (95% CI)Reference1.00 (0.94–1.06)0.64 (0.45–0.90)Model 1: Adjusted for ageModel 2: Adjusted for age, ethnicity (White, Asian, Black, Mixed, Unknown), Townsend deprivation index (< 30th, 30th–70th, > 70th, Unknown), body mass index, smoking status (Never, Ever, Unknown), alcohol status (Never, Ever, Unknown), Total Metabolic Equivalent Task (MET) minutes per week for all activities (Low, Moderate, High, Unknown), hypertension medication use (Yes/No), diabetes medication use (Yes/No), lipid modifying agents use (Yes/No), antithrombotic agent use (Yes/No), menarche age (> 13, ≤ 13, Unknown), number of births (0, 1, 2, ≥ 3, Unknown), menopausal status (No, Yes, Had a hysterectomy, Other reasons, Unknown), HRT use (Never, Ever, Unknown), first live birth age (< 20, 20–24, 25–29, ≥ 30, Nulliparous or Unknown), and oral contraceptive intake (Never, Ever, Unknown) Comparisons of breast cancer outcome according to type of APOE ε2 Model 1: Adjusted for age Model 2: Adjusted for age, ethnicity (White, Asian, Black, Mixed, Unknown), Townsend deprivation index (< 30th, 30th–70th, > 70th, Unknown), body mass index, smoking status (Never, Ever, Unknown), alcohol status (Never, Ever, Unknown), Total Metabolic Equivalent Task (MET) minutes per week for all activities (Low, Moderate, High, Unknown), hypertension medication use (Yes/No), diabetes medication use (Yes/No), lipid modifying agents use (Yes/No), antithrombotic agent use (Yes/No), menarche age (> 13, ≤ 13, Unknown), number of births (0, 1, 2, ≥ 3, Unknown), menopausal status (No, Yes, Had a hysterectomy, Other reasons, Unknown), HRT use (Never, Ever, Unknown), first live birth age (< 20, 20–24, 25–29, ≥ 30, Nulliparous or Unknown), and oral contraceptive intake (Never, Ever, Unknown) In subgroup analyses (Table 3), the protective association of ε2 homozygosity was more pronounced among participants who had ever smoked (P for interaction = 0.006). Among women who had never smoked, HRs were 1.00 and 0.91, while among ever smokers, the HRs were 1.00 and 0.24. For other stratifying variables including age, ethnicity, physical activity, LDL-C levels, use of lipid-modifying agents, use of antithrombotic agents, individual metabolic components, and reproductive factors, the associations were broadly consistent across subgroups, with no statistically significant interactions. Table 3Subgroup analysis according to type of APOE ε2Non-ε2 carriers (N = 198,804)ε2 heterozygotes (N = 34,530)ε2 homozygotes (N = 1,523)P for interactionEthnicity0.332White (N = 220,784)Reference1.01 (0.95–1.07)0.63 (0.44–0.89)Asian (N = 5,033)Reference0.79 (0.44–1.43)1.72 (0.23–12.80)Black (N = 4,129)Reference0.68 (0.38–1.24)0.00 (0.00-Inf)Age0.205 < 50 (N = 57,971)Reference1.00 (0.87–1.15)0.42 (0.18–1.03)50–65 (N = 137,243)Reference1.00 (0.93–1.08)0.83 (0.57–1.23) ≥ 65 (N = 39,643)Reference0.98 (0.84–1.13)0.23 (0.06–0.90)Smoking status0.006Never (N = 140,880)Reference1.00 (0.92–1.08)0.91 (0.63–1.32)Ever (N = 92,825)Reference1.00 (0.91–1.11)0.24 (0.10–0.58)MET, minutes/week0.430 < 600 (N = 33,950)Reference1.08 (0.93–1.26)0.37 (0.12–1.15)600–3000 (N = 95,069)Reference1.05 (0.96–1.15)0.47 (0.25–0.88) ≥ 3000 (N = 53,839)Reference0.96 (0.84–1.10)0.87 (0.45–1.67)LDL-C0.401 < 4.138 (N = 165,127)Reference0.98 (0.91–1.05)0.67 (0.47–0.97) ≥ 4.138 (N = 58,305)Reference1.11 (0.95–1.30)0.63 (0.16–2.50)Lipid modifying agents0.912No (N = 197,526)Reference0.99 (0.93–1.06)0.63 (0.44–0.92)Yes (N = 37,331)Reference1.03 (0.87–1.22)0.66 (0.27–1.58)Antithrombotic agent use0.130No (N = 210,403)Reference1.01 (0.95–1.08)0.69 (0.49–0.98)Yes (N = 24,454)Reference0.90 (0.73–1.10)0.17 (0.02–1.20)Metabolic syndromeAbdominal obesity0.114No (N = 90,838)Reference1.07 (0.96–1.19)0.44 (0.22–0.89)Yes (N = 143,509)Reference0.96 (0.89–1.04)0.71 (0.48–1.07)High triglyceride0.571No (N = 152,620)Reference0.98 (0.90–1.06)0.60 (0.37–0.98)Yes (N = 71,054)Reference1.04 (0.93–1.15)0.74 (0.46–1.19)Low HDL-C0.040No (N = 158,541)Reference1.05 (0.97–1.13)0.68 (0.45–1.01)Yes (N = 44,875)Reference0.84 (0.72–0.98)0.68 (0.32–1.42)High blood pressure0.435No (N = 84,905)Reference0.94 (0.85–1.05)0.70 (0.40–1.24)Yes (N = 149,952)Reference1.03 (0.95–1.11)0.60 (0.39–0.93)High fasting glucose0.987No (N = 227,955)Reference1.00 (0.94–1.06)0.64 (0.45–0.90)Yes (N = 6,902)Reference0.97 (0.67–1.40)0.60 (0.08–4.31)Menarche age0.557 > 13 (N = 83,397)Reference0.99 (0.89–1.11)0.42 (0.20–0.87) ≤ 13 (N = 143,943)Reference1.00 (0.92–1.08)0.65 (0.43–1.00)Number of births0.8480 (N = 43,883)Reference0.98 (0.85–1.13)0.47 (0.20–1.13)1 (N = 31,062)Reference0.95 (0.80–1.13)0.40 (0.13–1.23)2 (N = 102,653)Reference1.00 (0.91–1.10)0.71 (0.43–1.19) ≥ 3 (N = 56,693)Reference1.05 (0.92–1.19)0.79 (0.42–1.46)Menopause status0.348No (N = 58,335)Reference0.96 (0.83–1.09)0.47 (0.21–1.04)Yes (N = 139,886)Reference1.03 (0.96–1.12)0.71 (0.47–1.09)HRT use0.260Never (N = 145,887)Reference1.04 (0.96–1.13)0.70 (0.46–1.07)Ever (N = 87,732)Reference0.94 (0.85–1.04)0.55 (0.31–0.98)First live birth age0.698 < 20 (N = 15,907)Reference1.01 (0.79–1.30)0.29 (0.04–2.05)20–24 (N = 55,181)Reference1.07 (0.94–1.22)0.91 (0.50–1.64)25–29 (N = 59,647)Reference0.95 (0.84–1.08)0.65 (0.32–1.30) ≥ 30 (N = 28,272)Reference1.07 (0.90–1.27)0.91 (0.38–2.19)Oral contraceptive intake0.846Never (N = 43,664)Reference1.00 (0.87–1.15)0.51 (0.21–1.22)Ever (N = 190,098)Reference1.00 (0.93–1.07)0.67 (0.46–0.97)HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; HRT, hormone-replacement therapyAdjusted for age, ethnicity (White, Asian, Black, Mixed, Unknown), Townsend deprivation index (< 30th, 30th–70th, > 70th, Unknown), body mass index, smoking status (Never, Ever, Unknown), alcohol status (Never, Ever, Unknown), Total Metabolic Equivalent Task (MET) minutes per week for all activities (Low, Moderate, High, Unknown), hypertension medication use (Yes/No), diabetes medication use (Yes/No), lipid modifying agents use (Yes/No), antithrombotic agent use (Yes/No), menarche age (> 13, ≤ 13, Unknown), number of births (0, 1, 2, ≥ 3, Unknown), menopausal status (No, Yes, Had a hysterectomy, Other reasons, Unknown), HRT use (Never, Ever, Unknown), first live birth age (< 20, 20–24, 25–29, ≥ 30, Nulliparous or Unknown), and oral contraceptive intake (Never, Ever, Unknown) Subgroup analysis according to type of APOE ε2 HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; HRT, hormone-replacement therapy Adjusted for age, ethnicity (White, Asian, Black, Mixed, Unknown), Townsend deprivation index (< 30th, 30th–70th, > 70th, Unknown), body mass index, smoking status (Never, Ever, Unknown), alcohol status (Never, Ever, Unknown), Total Metabolic Equivalent Task (MET) minutes per week for all activities (Low, Moderate, High, Unknown), hypertension medication use (Yes/No), diabetes medication use (Yes/No), lipid modifying agents use (Yes/No), antithrombotic agent use (Yes/No), menarche age (> 13, ≤ 13, Unknown), number of births (0, 1, 2, ≥ 3, Unknown), menopausal status (No, Yes, Had a hysterectomy, Other reasons, Unknown), HRT use (Never, Ever, Unknown), first live birth age (< 20, 20–24, 25–29, ≥ 30, Nulliparous or Unknown), and oral contraceptive intake (Never, Ever, Unknown) Discussion In this large cohort study, germline APOE ε2 homozygosity is associated with a significantly reduced risk of breast cancer. Women with ε2 homozygosity had a 36% lower risk of breast cancer compared to non ε2 carriers. In particular, the magnitude of the association was strongest among ever smokers. Previous case–control studies and meta-analyses showed contradictory associations between APOE polymorphisms and cancer risk [9, 11]. In contrast, this study was designed as a hypothesis-driven analysis focusing on APOE ε2 homozygosity as a natural genetic model of altered lipid metabolism, which may explain the differences from prior investigations. First, most prior studies combined heterozygotes and homozygotes to increase statistical power. However, since ε2 and ε4 have opposite effects in many metabolic pathways [7, 12], the presumed protective benefit of ε2 is not fully expressed in the heterozygous state (and, hence, must act either recessively or co-dominantly) [13]. Second, the analysis was site-specific of breast cancer rather than all cancer to account for the heterogeneous etiologies among cancer types. Homozygosity for ε2 was associated with lower risk of breast cancer, but did not significantly correlate with cancers not involving primarily lipid metabolism, such as prostate cancer [14]. However, our finding was in agreement with recent observations suggesting that breast cancer depends on an external lipid supply for tumor growth, membrane biosynthesis, and intracellular signaling [6]. APOE was selected a priori based on its central role in lipid transport and immune regulation, and its well-established effect on circulating LDL cholesterol, which is directly relevant to lipid-dependent tumor biology. According to the previous study, the role of LDL-driven signaling in tumor proliferation, ferroptosis resistance, and inflammatory remodeling of the tumor microenvironment has been suggested [3, 6]. This literature supports a lipid–tumor axis in which lower LDL availability may constrain tumor-promoting metabolic programs. APOE ε2, which reduces LDL receptor affinity and lowers circulating LDL cholesterol, fits within this mechanistic framework. A key finding of this study was that the protective effect of APOE ε2 was modified by smoking status. Among ever-smokers, ε2 homozygotes had a significantly lower breast cancer risk than ε2 heterozygotes and non-ε2 carriers. This could be explained by the interaction of lipid peroxidation, inflammation and oxidative stress, which are increased among smokers. APOE regulates lipid-induced immunity, and has been linked to a less pro-inflammatory phenotype in ε2 versus ε3 or ε4 [6]. Smokers with ε2 homozygosity may be less susceptible to smoke-induced pro-inflammatory routes that facilitate tumor proliferation or development [15]. The stronger protective association among ever smokers in our study suggests that APOE ε2 may buffer against smoke-induced inflammatory pathways that support breast tumor development [15]. This study has several limitations. First, while the UK Biobank offers extensive genotypic and phenotypic information, its volunteer-based sampling might restrict the potential generalizability of such findings to more diverse populations, especially those of non-European origin. Secondly, even after controlling for several potential confounders, residual confounding as a result of unmeasured factors and other exposures such as dietary, specific hormonal exposures, and environmental carcinogens cannot be eliminated. Third, we evaluated APOE genotype as a germline factor; however, tumor subtype, stage, and molecular data were not available, limiting the possibility of a more detailed phenotypic characterization. However, the present study has a number of strengths, including its large sample size, prospective design, APOE genotyping data standardization, and linkage to validated national cancer registry data. In summary, homozygosity for APOE ε2 was associated with lower breast cancer incidence, particularly among participants exposed to inflammatory stress such as smoking. These findings suggest that modulation of lipid and immune pathways may contribute to breast cancer susceptibility. Further studies are needed to elucidate how APOE-driven lipid and immune pathways interact to influence breast cancer risk. Methods Data sources and study population This study used data from the UK Biobank, under application number 70034. The data comprised 502,410 participants between the ages of 37 and 73 years recruited from 2006 to 2010. Participants with prevalent dementia (N = 213) or any cancer (N = 44,816) at baseline were excluded. Dementia was defined using hospital inpatient and death registry data based on ICD-10 codes (F00–F03, G30). Participants with dementia were excluded because APOE genotype is strongly associated with dementia risk, which could confound the association of interest. Cancer was identified through national cancer registry linkage and pre-baseline diagnosis records. Among 457,411 participants, 234,857 women who had APOE genotyping data were included in our analysis. Baseline was defined as each participant’s initial assessment date, which served as the start of follow-up. Anonymized data and samples were donated by all the participants for testing and analysis in health-related research, as well as for recontacting to conduct sub-studies, with participants consenting to UK Biobank use of their health-related information. All subjects gave written informed consent and the study was approved by the Northwest Multicenter Research Ethics Committee (11/NW/0382). APOE genotypes UK Biobank participants were genotyped using UK BiLEVE array (Affymetrix, Santa Clara, CA, USA) or UK Biobank Axiom array, of which both arrays overlap 95% in marker content [7]. APOE gene has three major alleles (ε4, ε3, and ε2) dependent on variants at two single nucleotide polymorphisms (rs429358 and rs7412). We included genotyped rs429358 and rs7412 variants, which were all consistent with Hardy–Weinberg equilibrium (each P > 0.05). Individuals were classified into one of six prevalent APOE genotypes, according to their allelic status at rs429358 and rs7412: ε4/ε4, ε4/ε3, ε4/ε2, ε3/ε3, ε2/ε3, or ε2/ε2. As in prior investigations, indeterminate ε2/ε4 or ε1/ε3 genotypes were classified as ε2/ε4 because the ε1 allele is infrequent [7]. Participants were divided into three groups: non-ε2 carriers (ε3/ε3, ε3/ε4, and ε4/ε4), ε2 heterozygotes (ε2/ε3 and ε2/ε4), and ε2 homozygotes (ε2/ε2). Non-ε2 carriers (ε3/ε3, ε3/ε4, and ε4/ε4) were the reference group for comparisons. Incidence of breast cancer The main outcome was the first incidence of breast cancer. Breast cancer diagnoses were identified through linkage to national cancer registries in England, Scotland, and Wales with measurements based on the International Classification of Diseases, Tenth Revision (ICD-10) codes. Breast cancer was identified as ICD-10 code C50. Only first primary breast cancers after baseline were considered as events. Participants were followed from baseline to date of first breast cancer diagnosis, death, or administrative censoring date (June 30, 2021), whichever came first. Covariates Baseline covariates included age, ethnicity (White, Asian, Black, Mixed, Unknown), socioeconomic status measured by the Townsend Deprivation Index (< 30th, 30th–70th, > 70th, Unknown), smoking status (Never, Ever, Unknown), alcohol status (Never, Ever, Unknown), total Metabolic Equivalent Task (MET) minutes per week for all activity (Low, Moderate, High, Unknown), hypertension medication use (Yes/No), diabetes medication use (Yes/No), lipid modifying agents use (Yes/No), antithrombotic agent use (Yes/No), menarche age (> 13, ≤ 13, Unknown), number of births (0, 1, 2, ≥ 3, Unknown), menopausal status (No, Yes, Had a hysterectomy, Other reasons, Unknown), hormone replacement therapy (HRT) use (Never, Ever, Unknown), age at first live birth (< 20, 20–24, 25–29, ≥ 30, Nulliparous or Unknown), and oral contraceptive intake (Never, Ever, Unknown). For subgroup analyses, participants were stratified by ethnicity (White, Asian, Black), age group (< 50, 50–65, ≥ 65 years), smoking status (Never vs. Ever), total MET minutes per week (< 600, 600–3000, ≥ 3000), LDL-C level (< 4.138 vs. ≥ 4.138 mmol/L), use of lipid-modifying agents (Yes/No), use of antithrombotic agents (Yes/No), and individual metabolic components (Yes/No). Metabolic components were classified based on clinical cutoffs for abdominal obesity, triglycerides, HDL cholesterol, blood pressure, and fasting glucose. Metabolic syndrome was defined according to modified criteria from the Adult Treatment Panel III and the International Diabetes Federation. Abdominal obesity was defined as a waist circumference ≥ 80 cm; high triglycerides as ≥ 1.694 mmol/L; low HDL cholesterol as < 1.293 mmol/L; high blood pressure as systolic ≥ 130 mmHg, diastolic ≥ 85 mmHg, or current use of antihypertensive medication; and high fasting glucose as ≥ 5.550 mmol/L after ≥ 8 h of fasting, or current use of antidiabetic medication. Participants were also stratified by female reproductive factors including menarche age (> 13 vs. ≤ 13), number of births (0, 1, 2, ≥ 3), menopausal status (No vs. Yes), HRT use (Never vs. Ever), age at first live birth (< 20, 20–24, 25–29, ≥ 30), and oral contraceptive intake (Never vs. Ever). Statistical analysis Baseline characteristics were summarized as means (standard deviations) or counts (percentages), as appropriate. Differences in baseline covariates among the three APOE groups were assessed using absolute standardized differences (SMDs), with values > 0.2 indicating meaningful differences. The primary analysis evaluated a single prespecified genetic exposure and a single primary outcome. Incidence rates of first primary breast cancer were calculated per 1,000 person-years. Cox proportional hazards models were used to estimate HRs and 95% CIs for cancer incidence by APOE genotype, using non-ε2 carriers (ε3/ε3, ε3/ε4, and ε4/ε4) as the reference group. Model 1 was adjusted for age, and Model 2 was adjusted for age, Townsend Deprivation Index, lifestyle factors (smoking status, alcohol status, MET minutes per week for all activity), body mass index, medication use (hypertension medication use, diabetes medication use, lipid modifying agents use, antithrombotic agent use), and female reproductive factors as defined in the Covariates section. To evaluate potential effect modification, subgroup analyses were stratified by ethnicity, age, smoking status, LDL-C level, lipid modifying agents use, antithrombotic agent use, individual metabolic components, and female reproductive factors. Interaction terms between APOE genotype group and each stratifying variable were included in separate models. P-values for interactions were derived from likelihood ratio tests comparing models with and without the interaction term. Subgroup analyses were considered exploratory. Statistical significance was defined as a two-sided P < 0.05. All analyses were conducted using R version 4.4.1 (The R Foundation for Statistical Computing, Vienna, Austria; RRID:SCR_001905), including the ‘survival’ package (RRID:SCR_021137) [16]. Acknowledgements This research was supported by the “Korea National Institute of Health” (KNIH) research project (2025-ER1001-01) and Korea Brain Research Institute Intramural fund given from Ministry of Science and Technology to Korea Brain Research Institute (24-BR-04-01). The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. Author contributions DK, CHK and DGK conceived idea for the study. DK designed the research. DK, HW, SH, and JH accessed and verified the data. HW, and SH conducted analyses. DK provided statistical expertise. DK, HW, and JH wrote the first draft of the paper. All authors contributed to the interpretation of the results and critical revision of the manuscript for important intellectual content and approved the final version. DK is the guarantor of the work. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. Data availability Data may be obtained from a third party and are not publicly available. Data are available through the UK Biobank (http:/www.ukbiobank.ac.uk) upon application and with permission of UKBB’s Research Ethics Committee. Code availability Code is available upon request. Declarations Ethics approval and consent to participate UK Biobank has obtained ethical approval from the Northwest Multicenter Research Ethics Committee (11/NW/0382). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Hwamin Woo BS and Suhyun Han BS have contributed equally to this work as first authors. Do-Geun Kim, Chi-Hun Kim and Danbee Kang contributed equally to this work. Contributor Information Do-Geun Kim, Email: kimvet0911@kbri.re.kr. Chi-Hun Kim, Email: chi-hun.kim@hotmail.com. Danbee Kang, Email: dbee.kang@skku.edu. References Associated Data Data Availability Statement Data may be obtained from a third party and are not publicly available. Data are available through the UK Biobank (http:/www.ukbiobank.ac.uk) upon application and with permission of UKBB’s Research Ethics Committee. Code is available upon request.