Circulating Biomarkers of Inflammation and Ovarian Cancer Risk in the Nurses' Health Studies

The role of chronic inflammation as a mechanism of ovarian carcinogenesis has been supported by substantial evidence, with epidemiologic studies showing an increased risk for proinflammatory exposures (e.g., endometriosis, pelvic inflammatory disease; refs. 1, 2) and a decreased risk for anti-inflammatory exposures (e.g., aspirin use; refs. 3, 4). Moreover, higher concentrations of circulating biomarkers of inflammation, such as C-reactive protein (CRP) and TNFα, have been linked to an increased risk of ovarian cancer in prospective studies (5–9). However, these biomarkers can be activated by several immune pathways and thus, are relatively nonspecific, leaving the specific immunogenic processes influencing ovarian tumor development unclear.

In this study, we use data from a case–control study nested within two cohort studies, the Nurses' Health Study (NHS) and the Nurses' Health Study II (NHSII), to evaluate the association between five cytokines or chemokines involved in B-cell activation or pathways of T-cell differentiation [B-cell activating factor (BAFF), C-X-C motif chemokine ligand 13 (CXCL13), IL8, soluble(s)IL2-receptor-α(Rα), sIL6Rα] and epithelial ovarian cancer risk. We specifically selected IL8, sIL2Rα, and sIL6Rα to replicate findings from other studies of circulating biomarkers and risk of ovarian cancer (6, 10). BAFF and CXCL13 were chosen to investigate B cells in ovarian cancer, which has yet to be explored in circulation but has shown promising findings in other studies (11). These inflammatory biomarkers may provide clues as to which immune pathways are the most influential for ovarian carcinogenesis.

The NHS was initiated in 1976 and prospectively enrolled 121,700 female registered nurses aged 30–55 years from 11 U.S. states (12). The NHSII was initiated in 1989 and prospectively enrolled 116,429 female registered nurses aged 25–42 years from 14 U.S. states (12). Women in both cohorts completed questionnaires that provided detailed information on demographics, reproductive history, lifestyle factors, and medical history at the time of enrollment and biennially thereafter. A subset of 32,826 NHS and 29,611 NHSII participants provided blood samples and completed a short questionnaire in 1989–1990 and 1996–1999, respectively (13). Of the NHSII participants, 18,521 women provided a sample timed during the luteal phase of the menstrual cycle (e.g., days 20 to 22 in a 28-day cycle), while the other 11,090 provided an untimed sample. Aliquots of plasma, buffy coat, and red blood cells were stored in liquid nitrogen after collection. Study protocols for NHS and NHSII were approved by the institutional review boards of the Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required. Research was conducted in accordance with the Belmont report. Participant return of self-administered questionnaires and blood samples was accepted as implicit consent.

Diagnoses of incident ovarian cancer were identified through follow-up questionnaires or through the National Death Index and confirmed via medical record review or linkage to cancer registries. Histomorphologic characteristics were abstracted for identified cases and confirmed via slide review if tumor blocks were available. Within each cohort, cases were matched to one or two controls with at least one intact ovary at the time of blood collection, based on the following characteristics at blood draw: age (±1 year), month of collection (±1 month), time of day of collection (±2 hours), fasting status (>8 or ≤8 hours), and menopausal status including use of hormone therapy for postmenopausal women (premenopausal, postmenopausal on hormone therapy, postmenopausal not on hormone therapy, unknown). NHSII also matched on the date of the luteal blood draw (±1 day). When available, covariate data were obtained from the questionnaires completed at the time of blood draw; otherwise, these data were obtained from the biennial questionnaire returned nearest to the date of blood draw. For the purposes of this study, we included cases with a confirmed ovarian cancer diagnosis that occurred up until the 2016 (NHS) or 2015 (NHSII) follow-up cycle (and their matched controls), and excluded any case diagnosed within one year of blood draw to reduce the possibility of reverse causation.

Plasma levels of BAFF, CXCL13, IL8, sIL6Rα, and sIL2Rα were determined using two customized multiplexed (Luminex platform) panels from R&D Systems. Plasma specimens were diluted 1:2 for IL8 and 1:10 for BAFF, CXCL13, sIL6Rα, sIL2Rα. Briefly, Luminex microparticles precoated with analyte-specific antibodies were incubated with diluted serum samples, followed by a biotin–antibody and by a streptavidin–phycoerythin conjugate. The fluorescence intensity of each analyte's microparticles was quantified using a Bioplex 200 (Luminex) System Analyzer (Bio-Rad), and the data were analyzed using BioPlex Manager (v 4.1.1) software. The lower limit of detection was set as the lowest value that the BioPlex Manager software could calculate using the standard curve. The study samples (one sample per subject) were arranged to ensure that samples from matched case–control sets were tested in the same batch, with case–control order within sets determined randomly and with technicians blinded to case–control status. Technicians were also blinded to pairs or trios of blinded quality control (QC) samples that were distributed to ensure even placement (but otherwise, random location) within each batch. Using the blinded QC samples, intra- and inter-assay (only for NHS) coefficients of variation (CVs) were calculated. Intra-assay CVs ranged from 4.4% (IL8) to 15.5% (CXCL13) for NHS and 2.9% (sIL2Rα) to 3.9% (BAFF) for NHSII, and inter-assay CVs ranged from 8.9% (sIL2Rα) to 19.9% (CXCL13) for NHS.

To alleviate concerns related to technical variation across sample runs, batch correction was conducted using the average batch method (14), and included the following variables: study cohort, case–control status plus histology for cases, all cohort matching factors (age at blood draw, month and time of blood draw, fasting status, menopausal status, and hormone therapy use at blood draw), and potential confounders [duration of oral contraceptive use, parity, tubal ligation, family history of ovarian or breast cancer in a first-degree relative, ever use of aspirin, physical activity, smoking status, and body mass index (BMI)]. After batch correction, the extreme Studentized deviate (ESD) many-outlier approach (15) was used to identify outliers. Any participant that was flagged as an outlier for more than two biomarkers was removed from all subsequent analyses (n = 1 control and its matched case). Next, the distribution of each biomarker in the controls was used to determine quartiles for subsequent analyses.

Multivariable conditional logistic regression was used to estimate ORs and 95% confidence intervals (CI) for the association between each biomarker and risk of ovarian cancer. We conducted these analyses separately for NHS and NHSII and pooled across cohorts using random effects meta-analysis to assess between-study heterogeneity. All models were adjusted for the number of pregnancies (continuous), duration of oral contraceptive use (never, <1 year, 1–5 years, ≥ 5 years), family history of ovarian or breast cancer in a first-degree relative (yes, no), tubal ligation (yes, no), BMI (continuous, kg/m²), smoking (never, former, current), aspirin use (ever, never), and physical activity [< 3, 3–8.9, 9–17.9, 18–26.9, ≥ 27 metabolic equivalent task (MET)-hours/week]. Another model was assessed that simultaneously adjusted for all of the biomarkers. We additionally used restricted cubic splines to test potential nonlinearity of the association between each biomarker and ovarian cancer risk (16).

To investigate differences in the risk associations by histotype, we grouped cases according to high-grade serous carcinoma (HGSC) or non-HGSC. Non-HGSC combined low-grade serous, endometrioid, clear cell, and mucinous carcinoma together due to the rarity of these histotypes and similar risk profiles (17). Risk associations were stratified by histotype using polytomous logistic regression (adjusting for matching factors and the above covariates), and heterogeneity was assessed using a likelihood ratio test comparing a model where the effect of each biomarker was assumed to be the same across histotype to a model where the effect was allowed to vary across histotype (18). For these analyses, the covariates were constrained across histotype, and women diagnosed with an other or unknown histotype were excluded (n = 34 case–control sets). Using unconditional logistic regression, stratified analyses were performed by selected inflammation-related exposures (BMI, smoking status, aspirin use) and menopausal status at the time of blood draw, as prior studies have investigated differences in circulating immune-related biomarker risk associations by these factors (7–10, 19). For analyses stratified by menopausal status at blood draw, 41 case–control sets with unknown menopausal status were excluded. A likelihood ratio test comparing models with and without an interaction term for the biomarker and potential effect modifier (e.g., BAFF x BMI) was performed to assess heterogeneity in the risk associations. Because of the number of tests for interaction, the false discovery rate was used to control for multiple comparisons (20).

A nonspecific systemic inflammatory biomarker, CRP, has also been measured in this study population and its association with ovarian cancer risk has been published elsewhere (5). We combined the data on CRP with the data on the inflammatory biomarkers in this study to perform an exploratory analysis assessing biomarker interactions, and to assess whether additional adjustment for CRP using the clinical cut-off points (<1, 1–<10, and ≥10 mg/L) influenced our results.

Besides the control identified by the ESD method as an outlier for more than two biomarkers, there were 26 participants with outlying values for a single biomarker (BAFF, n = 1; CXCL13, n = 9; and IL8, n = 16). We repeated the main analyses for BAFF, CXCL13, and IL8 excluding the corresponding outlier case–control sets to assess whether these extreme values distorted the findings.

Blood draws were conducted between 1.3 and 24.1 years before diagnosis (median = 12.2 years). We explored whether excluding women diagnosed within two years of collection (n = 8 cases) and their matched controls would have any impact on our findings, and we additionally repeated the analyses stratified by three categories of time since collection (1–<9 years, n = 97; 9–<15 years, n = 94; ≥15 years, n = 108).

A total of 299 ovarian cancer cases (242 from NHS and 57 from NHSII) and 334 controls (242 from NHS and 92 from NHSII) were included in this study. In comparison to NHS, NHSII participants were younger, more likely to be premenopausal, and have a higher BMI at blood draw (Table 1). NHS cases were also more likely to be diagnosed with HGSC than NHSII (71% vs. 51%, respectively). For the majority of the inflammatory biomarkers, the concentrations were similar or higher among cases versus controls, and the biomarkers were weakly to modestly correlated among controls (Pearson r = −0.08–0.39).

In the analyses to investigate potential between-study heterogeneity, the only suggestive indication of heterogeneity was for the association of IL8 with ovarian cancer (P_{heterogeneity} = 0.08; Supplementary Table S1). However, the sample size for NHSII was small, producing imprecise estimates. Therefore, we pooled NHS and NHSII data for all analyses. No statistically significant evidence of nonlinearity was observed in the associations between each biomarker and risk of ovarian cancer (P > 0.05). Women with the highest (Q4) versus lowest quartile (Q1) levels of CXCL13 had a 72% increased risk of epithelial ovarian cancer (OR = 1.72; 95% CI = 1.04–2.83; P_trend = 0.007; Table 2). This positive association remained after adjustment for levels of the other biomarkers under study (OR_Q4vsQ1 = 1.96, 95% CI = 1.11–3.45; P_trend = 0.007). No statistically significant associations were observed for the other biomarkers.

There were minimal changes to the findings after excluding women who were diagnosed one to two years after blood collection and their corresponding matched controls (e.g., CXCL13 OR_Q4vsQ1 = 1.81, 95% CI = 1.08–3.02; P_trend = 0.003). When examining the CXCL13 and ovarian cancer risk by time since blood draw, the association was most pronounced among the women with a blood draw ≥15 years prior to diagnosis (OR_Q4vsQ1 = 2.46, 95% CI = 1.03–5.84; P_trend = 0.03), although the positive association with CXCL13 was present irrespective of time since blood draw (Supplementary Table S2; P_{heterogeneity} = 0.91). After removal of records with additional outlier values for BAFF, CXCL13, or IL8 and their corresponding matches, the risk associations for those biomarkers were similar to the overall findings, and the conclusions remained the same.

The assessment of heterogeneity by histotype (HGSC vs. non-HGSC; Table 3) showed that the positive association with CXCL13 was more pronounced in magnitude for non-HGSC (OR_Q4vsQ1 = 5.32, 95% CI = 1.26–22.35; P_trend = 0.03) than for HGSC (OR _Q4vsQ1 = 1.56, 95% CI = 0.83–2.93; P_trend = 0.04), although the associations were not statistically significantly different (P_{heterogeneity} = 0.19). A similar suggestive pattern by histotype was present for some of the other biomarkers (BAFF, sIL6Rα), but no statistically significant heterogeneity in associations was observed (P_{heterogeneity} = 0.57 and 0.28, respectively).

The analyses stratified by BMI and menopausal status revealed differences in the association between CXCL13 and risk of ovarian cancer across these strata (Fig. 1; Supplementary Table S3). Among overweight and obese women, the OR comparing ovarian cancer risk among women with the highest vs. lowest quartile levels of circulating CXCL13 was 2.44 (95% CI = 1.09–5.44), while among normal weight women, the comparable OR was 1.13 (95% CI = 0.61, 2.09; P_interaction = 0.10). In addition, a positive association for CXCL13 and risk of ovarian cancer was evident among postmenopausal women (OR_Q4vsQ1 = 2.45, 95% CI = 1.24, 4.84) but not among premenopausal women (OR_Q4vsQ1 = 0.72, 95% CI = 0.32, 1.62; P_{heterogeneity} = 0.04). Most of the other biomarkers were positively associated with risk among overweight and obese women but not among normal weight women (Supplementary Table S3), although statistically significant heterogeneity was not evident (P_{heterogeneity} > 0.10). Besides CXCL13, no other biomarkers demonstrated differences in associations by menopausal status. Generally, the associations between each biomarker and risk of ovarian cancer were stronger in magnitude among never users of aspirin compared to ever users, although after FDR correction, no heterogentiy was observed (P_{heterogeneity} > 0.12). A positive association for IL-8 and ovarian cancer risk was observed only among never users of aspirin (OR_Q4vsQ1 = 5.43, 95% CI = 1.65, 17.88; P_trend = 0.004) but not among ever users (OR_Q4vsQ1 = 0.88, 95% CI = 0.51, 1.53; P_{heterogeneity} = 0.12). No differences were found by smoking status for any of the biomarkers.

After restricting the dataset to participants that also had data on CRP (n = 548; 262 cases and 286 controls), no interactions were present for CRP with any of the biomarkers (P_interaction ≥ 0.2). CRP was weakly correlated with the inflammatory biomarkers (Pearson r = 0.01–0.20), and ORs for these biomarkers and ovarian cancer risk were similar to those in the primary analysis after further adjustment for CRP (Supplementary Table S4).

In the NHS and NHSII cohorts, we observed an increased risk of ovarian cancer for women with high circulating levels of CXCL13, but no association for BAFF, IL8, sIL2Rα, and sIL6Rα. There was also evidence to suggest that the association for CXCL13 was more pronounced in magnitude for non-HGSC, obese and overweight women, and postmenopausal women, although only menopausal status showed statistically significant variation. CRP, which has been associated with ovarian cancer risk in other studies (5–9), was not a confounder or effect modifier for any of the biomarkers we assessed.

The most notable finding in the present study is the positive association between CXCL13 and ovarian cancer risk. CXCL13 is a chemokine that is a chemoattractant for B cells that preferentially promotes the migration of B cells to lymphoid tissues (21). CXCL13 and its G protein–coupled receptor, C-X-C chemokine receptor 5 (CXCR5), have been implicated in the development of several inflammatory conditions and autoimmune diseases (21). In the context of cancer, the CXCL13:CXCR5 signaling axis has been shown to have both antitumor immune surveillance (e.g., promotion of tertiary lymphoid structures, regulation of cytolytic T-cell activity) and tumor immune evasion (e.g., suppression of T-effector cell activity, recruitment of immunosuppressive T-regulatory, and myeloid-derived suppressor cells) roles in disease initiation and progression (21, 22). CXCL13 is also a marker of germinal center formation (23), suggesting that CXCL13 is associated with an active antibody response against a developing cancer. CXCL13 has been seen to act as a tumor growth–promoting factor for cancers of B-cell origin, including non-Hodgkin lymphoma (NHL; refs. 24, 25). Higher CXCL13 levels have been observed in the serum or plasma of cancer cases compared with controls (26–29) as well as in tumor versus adjacent normal tissue (26, 30). High CXCL13 expression (either in serum or tumor tissue) has also been associated with worse prognosis or unfavorable tumor characteristics (e.g., advanced stage) in many cancers (27, 29, 31–35), although improved outcomes have also been observed (36–39).

To our knowledge, this is the first study investigating the association between circulating CXCL13 and ovarian cancer risk. Our findings are consistent with studies in lung cancer (40, 41) and lymphoma (42, 43) that noted an increased cancer risk for individuals with higher levels of circulating CXCL13. The few studies (44–47) that have examined CXCL13 in the context of ovarian cancer measured CXCL13 levels in other types of biological specimens (e.g., ascites fluid, tumor tissue) and focused on other cancer-related endpoints (e.g., survival, tumor characteristics). Three studies (44–46) noted better overall survival for tumors expressing high levels of CXCL13, particularly for TP53-mutant serous ovarian tumors and advanced stage ovarian cancers. Auer, and colleagues (47) found that CXCL13 expression in ascites fluid of women with HGSC differed by disease spread, with a higher expression noted in tumors with an upper abdominal or miliary disease spread.

The magnitude of the association between CXCL13 and ovarian cancer risk was stronger for non-HGSC than HGSC, although after FDR correction, no statistically significant heterogeneity was detected by histotype. This finding is similar to those from recent studies of inflammatory-related exposures or biomarkers, including but not limited to CRP (9), smoking status (17, 48), and endometriosis (1, 17), for which the association with ovarian cancer risk was also more pronounced for non-HGSC than HGSC. However, risk associations with other inflammatory or immune-related factors, including chlamydia seropositivity, aspirin use, and talcum powder use, have been similar across histotype or were more pronounced among serous tumors (2–4, 49–51). Larger studies are needed to fully explore potential heterogeneity by histotype.

This study found a statistically significant interaction for the CXCL13 association and menopausal status in relation to ovarian cancer risk, although the confidence intervals overlapped at all quartiles, and we recommend caution in the interpretation of these findings. The positive association between CXCL13 and risk was more pronounced among postmenopausal than premenopausal women. Likewise, although not statistically significant, we observed a more pronounced association for CXCL13 among overweight and obese women in this study. Although little work has been conducted on CXCL13 in this context, studies have shown that the association between other circulating inflammatory biomarkers (e.g, CRP, IL6, and TNFα) and ovarian cancer risk was more pronounced or found exclusively among postmenopausal or overweight and obese women (7–9). Both obesity and menopause have been associated with higher levels of circulating inflammatory biomarkers (52, 53) as well as other factors that may further exacerbate inflammation, such as hormonal changes and a greater number of comorbid conditions (54–56). There is also evidence that estrogen levels, which are known to vary by menopausal status and obesity, are inversely correlated with CXCL13 expression (57). It is possible that these groups of women may have a greater proinflammatory state in which the mechanism underlying the association of CXCL13 with ovarian cancer risk could be enhanced. Although this represents a plausible explanation for our findings, the mechanism is currently unknown and warrants further study.

No associations with risk of ovarian cancer were observed for the other inflammatory biomarkers. It is possible that we did not have adequate power to detect an association for these biomarkers, which had lower levels within our study population. Prior studies on circulating inflammatory or immune markers and ovarian cancer risk have primarily focused on CRP, IL6, and TNFα (5, 7–9), but two studies (6, 10) investigated a variety of immune markers including some of those in this study. In the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (6), the authors noted an increased risk of serous ovarian cancer for increasing levels of IL8, and although not statistically significant, an inverse association for sIL6R. Another study (10) combined data from three prospective cohorts, the New York University Women's Health Study, the Northern Sweden Health and Disease Study, and the Italian Hormones and Diet in the Etiology of Cancer Study, and investigated a variety of inflammatory cytokines and cytokine modulators in ovarian cancer, including sIL2Rα and sIL6R. Although not statistically significant, a positive association was observed for increasing levels of sIL2Rα while an inverse association was present for sIL6R. Larger studies are needed with sufficient power to explore these biomarkers in more detail overall and by histotype.

The considerable strengths of this study include the prospective design of NHS/NHSII, prediagnostic biomarker measurements minimizing the possibility of reverse causation, and detailed questionnaire data to control for potential confounders. Despite these strengths, this study is not without limitations. A few of the samples had biomarker levels that were below the limit of detection but greater than the lowest reportable standard, although this occurred in <20% of participants. Another potential limitation was that the biomarkers were measured at one time point prior to diagnosis, which may not reflect the entire prediagnostic time period or the time period most influential to carcinogenesis. Studies in other populations have shown high-to-moderate reproducibility of the same biomarkers from samples collected one to five years apart [e.g., intraclass correlation coefficient (ICC)>0.40, with the majority of studies reporting an ICC>0.70; refs. 58–60]. With a median time from collection to diagnosis of 12.2 years in this study, we cannot be certain of the temporal stability for those cases with blood collected more distally from diagnosis. Moreover, measuring cytokines several years prior to diagnosis may indicate an underlying chronic inflammatory condition that could have impacted cancer development. Furthermore, the inflammatory biomarkers were measured in peripheral blood, and it is unclear how this correlates with the immune microenvironment at the tissue site of origin for ovarian cancer (e.g., ovary, fallopian tube). We also had a relatively small sample size, which limited the precision of the effect estimates and precluded a separate evaluation of risk associations for individual non-HGSC histotypes.

In summary, this study supports prior literature implicating inflammation in the etiology of ovarian cancer and is the first to provide evidence that circulating CXCL13 is associated with risk of ovarian cancer, and that this risk association is stronger among postmenopausal women. These findings warrant future study assessing this association in other populations; if replicated, CXCL13 may represent a novel biomarker for ovarian carcinogenesis. Likewise, given the effects of CXCL13 on B-cell activation, further research investigating B cells in circulation and the tumor microenvironment may elucidate additional mechanistic clues by which inflammation and immunity impact ovarian cancer development.

B.M. Birmann reports grants from NCI and American Institute for Cancer Research (AICR) outside the submitted work. J.R. Conejo-Garcia reports grants from NCI during the conduct of the study, as well as personal fees from Leidos and grants and personal fees from Anixa Biosciences and Compass Therapeutics outside the submitted work; in addition, J.R. Conejo-Garcia has a patent for CD277 antibodies pending to Compass Therapeutics and a patent for FSH CAR T cells issued to Anixa Biosciences. L.D. Kubzansky reports grants from Department of Defense during the conduct of the study. L.I. Magpantay reports grants from NIH and Department of Defense during the conduct of the study. O. Martinez-Maza reports other from Moffitt Cancer Center during the conduct of the study. S.S. Tworoger reports grants from NIH/NCI and DOD/CDMRP/OCRP during the conduct of the study. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

L.C. Peres: Formal analysis, investigation, methodology, writing–original draft, writing–review and editing. M.K. Townsend: Formal analysis, investigation, methodology, writing–review and editing. B.M. Birmann: Investigation, methodology, writing–review and editing. J.R. Conejo-Garcia: Investigation, writing–review and editing, data interpretation. Y. Kim: Investigation, methodology, writing–review and editing. L.D. Kubzansky: Conceptualization, resources, funding acquisition, investigation, methodology, writing–review and editing. L.I. Magpantay: Resources, data curation, investigation, methodology, writing–review and editing. O. Martinez-Maza: Resources, data curation, investigation, methodology, writing–review and editing. S.S. Tworoger: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing.

We would like to acknowledge the Channing Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital, as the home of the Nurses' Health Study. In addition, we would like to thank 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. This study was supported by the Department of Defense, W81XWH-17-01-0153 (to L.D. Kubzansky). The Nurses' Health Studies are supported by NIH grants UM1 CA186107, P01 CA87969, R01 CA49449, U01 CA176726, and R01 CA67262. L.C. Peres is supported by NIH/NCI R00 CA218681. Additionally, laboratory resources used for this work were supported by the Pendleton Charitable Trust, the McCarthy Family Foundation, and the Moffitt Foundation.

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.