Prolactin and Risk of Epithelial Ovarian Cancer

Prolactin is a multifactorial hormone, supporting mammary gland growth and development, synthesis of milk, maintenance of milk secretion, and regulation of immune responses in physiologic and pathologic states (1). Several studies have observed associations between higher prolactin levels and increased risk of breast cancer among postmenopausal women (2–6), while modest positive or null associations were noted among premenopausal women (2–7). Moreover, the addition of prolactin, along with other endogenous hormones, to postmenopausal breast cancer risk prediction models resulted in improved model performance (8, 9).

Several lines of evidence suggest that prolactin may have a role in the etiology of ovarian cancer. First, while prolactin is primarily produced in the pituitary gland, it is also synthesized in the ovaries (1). In addition, pregnancy, which has been shown to lower prolactin levels (10, 11), is an established protective factor for both breast and ovarian cancers (12, 13). One previous nested case–control study found a nonsignificant positive association between prediagnostic prolactin levels and ovarian cancer risk after adjusting for multiple covariates including parity. In stratified analyses comparing the top versus bottom quartile, a significant 3.1-fold increased odds of ovarian cancer was observed among women with a body mass index (BMI) ≥25 kg/m² (12). However, this study did not have sufficient sample size to assess heterogeneity of associations by tumor histotype, which has been observed for parity and other established ovarian cancer risk factors (14).

Therefore, we pooled data from case–control studies nested within five prospective cohorts to examine the association of prolactin levels with ovarian cancer risk, overall and by menopausal status, BMI, age, and histotype.

This study included participants from five cohorts, with a total of 703 ovarian cancer cases and 864 matched controls. The Nurses' Health Study (NHS; refs. 3–5, 15), Nurses’ Health Study II (NHSII; refs. 3–5, 15), Women's Health Study (WHS; refs. 16–18), NYU Women's Health Study (NYUWHS; refs. 19, 20), and Northern Sweden Health and Disease Study (NSHDS; refs. 21, 22) have been described in detail previously, and a brief description of each cohort, including total participants, number of cases and controls, sample type, enrollment period, age at enrollment, eligibility criteria and matching criteria, is shown in Table 1. Of note, our study included women from two subcohorts of the NSHDS: the Västerbotten Intervention Programme and the Mammography Screening Project cohorts. Two of these cohorts (NYUWHS and part of NSHDS) were included in the previously published pooled analysis (12).

Briefly, cases were defined as women with incident invasive or borderline (except WHS) epithelial ovarian or peritoneal cancer who had no prior history of cancer, except nonmelanoma skin cancer, before blood collection; all ovarian cancer diagnoses were after blood collection and confirmed by medical record review or cancer registry linkage. Controls were alive and had intact ovaries at the time their matched case was diagnosed. NYUWHS and NSHDS cohorts excluded participants using exogenous hormones at the time of blood draw (which can increase prolactin levels; ref. 23), while NHS, NHSII, and WHS matched on exogenous hormone use at blood collection. All studies minimally matched controls to cases on age, menopausal status, and date at blood collection. NHS, NHSII, and WHS controls were additionally matched for fasting status and postmenopausal hormone use (Table 1).

The overall project plan and the NHS/NHSII study protocol was approved by the institutional review boards of the Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health (Boston, MA), and those of participating registries as required. The WHS study was approved by the Institutional Review Board of Brigham and Women's Hospital. The NYUWHS study was reviewed and approved annually by the Institutional Board of Research Associates of New York University School of Medicine (New York, NY). The NSHDS study was reviewed and approved by the Swedish Ethical Review Authority, Umeå. Participants in NHS and NHSII provided implied consent and participants in WHS, NYUWHS, and NSHDS provided written informed consent to participate. All studies were conducted in accordance with recognized ethical guidelines [either the Declaration of Helsinki (NSHDS) or the U.S. Common Rule (all other studies)].

In NHS/NHSII as well as WHS and a subset of NSHDS plasma samples, prolactin was measured by microparticle enzyme immunoassay, using the ARCHITECT chemiluminescence immunoassay system (Abbott Diagnostics; ref. 3). Prolactin was measured in serum samples from NYUWHS and the other NSHDS plasma samples by Luminex Xmap multiplex bead-based technology using a kit from Linco/Millipore, as described in ref. 12. For all studies, matched case–control sets were assayed together in the same batch. All values were above the limit of detection. Coefficients of variation calculated from blinded replicate samples were <8% across all studies and assay batches. Previous studies within NHS and NHSII evaluated the within-person reproducibility of prolactin measurements among participants who donated two plasma samples 3 years apart. The intraclass correlation (ICC) was 0.53 among postmenopausal women (24). For premenopausal women, the ICC across three blood collections in the follicular phase was 0.41 and luteal phase was 0.64 (25).

In all cohorts, participants provided information on key covariates, such as oral contraceptive (OC) use, parity, menopausal status, and postmenopausal hormone therapy (HT) use, via a self-administered questionnaire. Menopausal status was defined as premenopausal, postmenopausal, or unknown; unknown menopausal status included those who selected not sure or had a missing value on the questionnaire, had a hysterectomy or were on hormone therapy (see Supplementary Table S1 for detailed information on menopausal status definitions by cohort). Questionnaires were given at study enrollment and at varied follow-up intervals. Details about the blood collection were assessed at the time of blood draw. All data were collected prospectively, with the exception of the NSHDS cohort where reproductive variables, such as parity and menopausal status, were collected retrospectively among ovarian cancer cases and at the same time for matched controls.

We used the generalized extreme studentized deviate many-outlier procedure to identify outlier prolactin values, after natural log transformation, among samples included in each assay run (26). We excluded one outlier (121.5 ng/mL) from NYUWHS, one outlier (81.5 ng/mL) from NSHDS using microparticle enzyme immunoassay, and three outliers (>115.6 or <1.6 ng/mL) from NSHDS using the Luminex assay. To account for variability in prolactin measurements due to interlaboratory variability, we recalibrated prolactin levels using an average-batch correction procedure (3, 27) on log-transformed prolactin levels, using the NHS/NHSII samples as the reference batch. Following recalibration, we converted the prolactin values back to their original scale (ng/mL) and pooled data across studies.

We used unconditional multiple logistic regression to calculate odds ratios (OR) and 95% confidence intervals (CI) for ovarian cancer risk according to quartiles of prolactin concentration based on the distribution of prolactin among all control women. Analyses were first conducted separately by cohort and results combined using random effects meta-analysis. Because there was no between-cohort heterogeneity (P = 0.96), we pooled the data for subsequent analyses. The models were adjusted for cohort (NHS, NHSII, NYUWHS, WHS, NSHDS from microparticle enzyme immunoassay, NSHDS from Luminex assay), variables measured at the time of blood draw including matching factors [age (continuous), fasting status (yes, no), menopausal status (premenopausal, postmenopausal), HT use (yes, no, unknown)] as well as ovarian cancer risk factors measured at the time of blood draw [BMI (continuous), OC use (never, ever, unknown), tubal ligation (no, yes), parity (0, 1, 2, 3, 4+, or unknown births), family history of breast or ovarian cancer (no, yes), hysterectomy status (no, yes, unknown)]. Tests for trend were calculated by including the prolactin quartile median value in the model as a continuous variable. Given that prolactin is related to menopausal status (lower in postmenopausal versus premenopausal women) and HT use (higher in HT users vs. nonusers; ref. 28) and studies did not match on HT use among women with unknown menopausal status, we limited analyses to women with a known menopausal status.

We used unconditional polytomous logistic regression to examine potential heterogeneity in risk estimates for type I (low-grade serous, endometrioid, clear cell, and mucinous) versus type II (high-grade serous, Brenner, transitional, carcinosarcoma) subtype and for time between blood draw and diagnosis <10 versus ≥10 years, conducting a likelihood ratio test comparing a model allowing the association of prolactin to vary versus not across histotype or time between blood draw and diagnosis (29). We conducted stratified logistic regression analyses to explore whether the association between prolactin and ovarian cancer risk differed by participant characteristics at blood draw, including menopausal status (premenopausal, postmenopausal), age (<50, 50–54, 55–59, 60+ years), BMI (<25, 25+ kg/m²), and HT use at time of blood collection among postmenopausal women (not current, current use) based on prior studies of prolactin in relation to breast and ovarian cancer risk that observed different associations by these factors (3, 12).

To examine the robustness of the results to the choice of prolactin quartile cut-off points, we repeated the analyses using quartiles based on cohort-specific control distributions of prolactin. We also conducted sensitivity analyses excluding cases diagnosed within two years of their blood draw (n = 5), and excluding the Luminex-assayed samples included in the previous publication (n = 496; ref. 12). Finally, we conducted secondary analyses among all women and among women with an unknown menopausal status. All P values were two-sided and were considered statistically significant if less than 0.05. Analyses were conducted using SAS, version 9.4 (SAS Institute Inc., Cary, NC).

The combined data from the 5 nested case–control studies was 1,567 women, including 703 cases and 864 controls (Supplementary Table S2). Of these, 613 cases and 758 controls had a known menopausal status at blood collection and were included in primary analyses (Table 2). Mean age at time of blood collection (a matching factor) was highest in NHS (57.4 years for cases and 57.5 years for controls) and lowest in NHSII (44.3 years for cases and 44.5 years for controls). For cases, the average number of years between blood collection and diagnosis ranged from 6.5 (NHSII) to 11.5 years (NHS). Characteristics of cases and controls were largely similar, although cases were more likely to be nulliparous and have a family history of breast or ovarian cancer.

In analyses conducted separately by cohort, there were suggestions of higher risk of ovarian cancer among women with higher prolactin levels in all studies, although none of the P values for linear trend were statistically significant (P_trend ≥ 0.24, Supplementary Table S3). In pooled analyses, we observed a statistically significant trend of higher ovarian cancer risk among women with higher prolactin levels (OR = 1.34; 95% CI = 0.97–1.85 comparing quartile 4 vs. 1; P_trend = 0.045; Table 3). This association was similar in analyses using study-specific quartiles of prolactin concentration (comparable OR = 1.33; 95% CI = 0.97–1.83; P_trend = 0.07). In addition, in sensitivity analyses excluding cases whose blood draw was within 2 years of their diagnosis, the association was similar (comparable OR = 1.34; 95% CI = 0.97–1.85; P_trend = 0.045). ORs comparing the highest versus lowest quartile of prolactin were similar in premenopausal and postmenopausal women, although not statistically significant [OR, top vs. bottom quartile = 1.38 (P_trend = 0.32) and 1.41 (P_trend = 0.08), respectively; P_{heterogeneity} = 0.91].

We assessed potential heterogeneity in associations by tumor histotype (type I vs. type II), time between blood draw and diagnosis (<10 vs. ≥10 years), BMI (<25 vs. ≥25 kg/m²), age group (<50, 50–54, 55–59, ≥60 years), and HT use at time of blood collection among postmenopausal women (not current vs. current; Table 4). Notably, higher prolactin was associated with increased ovarian cancer risk among overweight/obese women (OR, top vs. bottom quartile = 2.68; 95% CI = 1.56–4.59; P_trend < 0.01), while no association was observed among normal weight women (P_trend = 0.98; P_{heterogeneity} < 0.01). We did not observe any significant variation in the association by histology, years between blood draw and diagnosis, age, or HT use (P_{heterogeneity} ≥ 0.56). However, there were suggestions that the positive prolactin–ovarian cancer risk association was stronger if blood collection was more proximate to diagnosis, in women ≥55 years old, and postmenopausal women not using HT. We conducted sensitivity analyses excluding 180 cases and 316 controls that had been included in the previous publication (12). Results were slightly attenuated compared with those in the primary analysis (Supplementary Table S4).

In secondary analyses conducted in the full population, including women with an unknown menopausal status, results were largely attenuated and we did not observe a statistically significant association between quartile of prolactin concentration and ovarian cancer risk (OR = 1.15; 95% CI = 0.85–1.55 comparing quartile 4 vs. 1; P_trend = 0.23; Supplementary Table S5). This attenuation was due to a strong and statistically significantly lower risk of ovarian cancer among women whose menopausal status was unknown (comparable OR = 0.21; 95% CI = 0.07–0.61; P_trend = 0.01, P_{heterogeneity} = 0.02 comparing unknown vs. postmenopausal status and P_{heterogeneity} = 0.03 comparing unknown vs. premenopausal status).

This large prospective analysis of five nested case–control studies observed a modest positive association between prediagnosis prolactin levels and ovarian cancer risk. This positive association was stronger among overweight/obese women and we observed similar associations between premenopausal and postmenopausal women. This is consistent with a study conducted by Clendenen and colleagues (12), which examined the relation between prediagnostic prolactin levels and risk of ovarian cancer in a case–control study nested within three prospective cohorts, two of which were included in this study. Among 230 cases and 432 controls, there was a nonsignificant positive association between prolactin and ovarian cancer risk, that was statistically significant among women with a BMI ≥ 25 kg/m². Of note, the association in our study remained suggestive, although not statistically significant, when excluding the cases and controls included in the study by Clendenen and colleagues Furthermore, among patients with ovarian cancer, serum prolactin levels are elevated compared both with benign controls and individuals with other cancer types (e.g., lung, breast, pancreas), and 80% of ovarian tumors, across all histotypes, expressed the prolactin receptor (10). Overall, these data support a potential role of prolactin specifically in the development of ovarian tumors, particularly in a high adiposity environment.

Several mechanisms may contribute to this association, including increased cellular proliferation, angiogenesis, cell mortality, and inflammation, as well as reduced apoptosis and immune function (1, 10, 30–32). An experimental study looking at the human prolactin-like gene, Prl2c2, reported that the gene was significantly amplified in ovarian cancer tissue, and overall survival was lower in patients with ovarian cancer with higher expression of prolactin receptors (32). Furthermore, deletion of the prolactin receptor in ovarian cancer cell lines led to a significant reduction of cellular proliferation. In an immortalized normal ovarian epithelial cell line, acute prolactin exposure led to increased Ras pathway activation, and chronic exposure led to uncontrolled growth and colony formation, as well as other morphologic changes indicative of transformation (10). In addition, inflammatory cytokines such as TNFα and IL1β (33, 34), which can induce expression of prolactin receptor in fibroblasts (35), are involved in ovarian carcinogenesis. Furthermore, prolactin itself, can enhance macrophage production of these cytokines in an autocrine fashion (36).

Our results further suggest that prolactin may play a larger role in ovarian cancer development among women with high adiposity. Prolactin production can occur in adipose tissue (37) and can act as an adipokine through suppression of adiponectin and alterations in leptin. Specifically, prolactin appears to increase lipoprotein lipase (LPL) in mouse mammary glands (38); LPL is responsible for hydrolyzing triacylglycerols (TAG) into component fatty acids. However, whether prolactin impacts lipolysis or lipogenesis in adipose tissue is unclear, with mixed results in experimental studies (39). Interestingly, both prolactin and TAGs have been associated with a decrease in risk of type II diabetes (15, 40) but, conversely an increase in ovarian cancer risk (41), suggesting a complex interplay between endogenous factors that appear to protect against diabetes but increase ovarian cancer risk. Notably, high TAGs were found in highly aggressive epithelial ovarian cancer cell lines, with levels increasing under induced detachment stress (42), consistent with prior findings that high TAGs were specifically associated with ovarian tumors that were rapidly fatal (i.e., death within three years of diagnosis; ref. 41). Future research should focus on the metabolic pathways of prolactin in metabolism, particularly with respect to TAGs, in relation to ovarian carcinogenesis.

This study has several limitations. First, we only had data on self-reported menopausal status rather than a biological measure of menopausal status. Consequently, menopausal status data were missing for 90 cases and 106 controls and our findings differed dramatically within this group, showing an unexpectedly strong inverse association. This may be due to residual confounding by HT use as cases and controls with unknown menopausal status were not matched on this factor in NHS, NHSII, or WHS. Furthermore, hysterectomy was more common (and more likely to be missing) among controls versus cases with unknown menopausal status. Given the high prevalence of elective bilateral oophorectomy in the United States (39% of women undergoing hysterectomy for benign conditions from 1998–2006; ref. 43), it is possible that some controls in this group were not at risk of ovarian cancer at the time of case diagnosis (i.e., did not report the concurrent oophorectomy). Second, the ICC for prolactin measures is modest (∼0.5), which will attenuate the observed association with ovarian cancer risk. Furthermore, the blood sample types varied by cohort and two different assays were used, this added variability may also have attenuated the risk estimates. However, there was no significant heterogeneity by cohort or when using study-specific quartile cut-off points, suggesting that the analytic approaches to address batch-to-batch variation (27) reduced the impact of using different assays on the pooled risk estimates. Our study also has a number of important strengths. Notably, pooling data from five nested case–control studies provided a sufficient sample size to examine potential heterogeneity in associations of pre-diagnostic prolactin levels with ovarian cancer risk by histotype as well as by several participant characteristics.

In summary, we observed a modest trend of increasing ovarian cancer risk with higher prolactin levels, which was stronger among women with a higher BMI. This, in conjunction with prior epidemiologic and experimental studies, suggests that prolactin may play a role in ovarian carcinogenesis, with a possible mechanistic links related to obesity. Future research should evaluate the relationship of prolactin with ovarian cancer risk in larger populations with known menopausal status and assess the biologic interplay of prolactin and TAGs in ovarian carcinogenesis.

M.S. Rice reports currently being an employee of Sanofi. J.E. Buring reports grants from NIH during the conduct of the study. L.D. Kubzansky reports grants from Department of Defense during the conduct of the study. I-M. Lee reports grants from NIH during the conduct of the study. P.M. Sluss reports Consultant Ansh Labs LLC. A. Zeleniuch-Jacquotte reports grants from NIH/NCI during the conduct of the study. S.S. Tworoger reports grants from NIH/NCI during the conduct of the study. No disclosures were reported by the other authors.

C.A. Hathaway: Formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.S. Rice: Formal analysis, writing–review and editing. M.K. Townsend: Conceptualization, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. S.E. Hankinson: Conceptualization, resources, data curation, methodology, writing–review and editing. A.A. Arslan: Resources, data curation, writing–review and editing. J.E. Buring: Resources, data curation, writing–review and editing. G. Hallmans: Resources, data curation, writing–review and editing. A. Idahl: Resources, data curation, writing–review and editing. L.D. Kubzansky: Resources, data curation, writing–review and editing. I-M. Lee: Resources, data curation, writing–review and editing. E.A. Lundin: Resources, data curation, writing–review and editing. P.M. Sluss: Resources, data curation, writing–review and editing. A. Zeleniuch-Jacquotte: Resources, data curation, writing–review and editing. S.S. Tworoger: Conceptualization, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We would like to thank the Channing Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital, as the home of the Nurses' Health Study and the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY. The authors assume full responsibility for analyses and interpretation of these data. We thank the Department of Biobank Research at Umeå University, Västerbotten Intervention Programme, the Northern Sweden MONICA study, and the County Council of Västerbotten for providing data and samples and acknowledge the contribution from Biobank Sweden, supported by the Swedish Research Council.

This work is supported by the following grants: NIH grants U01 CA186107 to M. Stampfer, P01 CA87969 to M. Stampfer and S.S. Tworoger, R01 CA49449 to C. Byrne, U01 CA176726 to W.C. Williet, R01 CA67262 to S.E. Hankinson (NHS/NHSII); VR 2017–00650 to G. Hallmans, A. Idahl, and E.A. Lundin (NSHDS); NIH grants UM1 CA182934 to A. Zeleniuch-Jacquotte, P30 CA016087, P30 ES000260 (NYUWHS); NIH grants HL043851, HL080467, and HL099355 to J.E. Buring, and CA047988 and CA182913 to J.E. Buring and I. Lee (WHS). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. This investigation was supported by grants from Lion's Cancer Research Foundation, Umeå University.

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