Expression of p52, a non-canonical NF-kappaB transcription factor, is associated with poor ovarian cancer prognosis

The canonical and non-canonical nuclear factor-kappaB (NF-κB) signaling pathways have key roles in cancer, but studies have previously evaluated only the association of canonical transcription factors and ovarian cancer survival. Although a number of in vitro and in vivo studies have demonstrated mechanisms by which non-canonical NF-κB signaling potentially contributes to ovarian cancer progression, a prognostic association has yet to be shown in the clinical context. We assayed p65 and p52 (major components of the canonical and non-canonical NF-κB pathways) by immunohistochemistry in epithelial ovarian tumor samples; nuclear and cytoplasmic staining were semi-quantified by H-scores and dichotomized at median values. Associations of p65 and p52 with progression-free survival (PFS) and overall survival (OS) were quantified by Hazard Ratios (HR) from proportional-hazards regression. Among 196 cases, median p52 and p65 H-scores were higher in high-grade serous cancers. Multivariable regression models indicated that higher p52 was associated with higher hazards of disease progression (cytoplasmic HR: 1.54; nuclear HR: 1.67) and death (cytoplasmic HR: 1.53; nuclear HR: 1.49), while higher nuclear p65 was associated with only a higher hazard of disease progression (HR: 1.40) in unadjusted models. When cytoplasmic and nuclear staining were combined, p52 remained significantly associated with increased hazards of disease progression (HR: 1.91, p = 0.004) and death (HR: 1.70, p = 0.021), even after adjustment for p65 and in analyses among only high-grade serous tumors. This is the first study to demonstrate that p52, a major component of non-canonical NF-κB signaling, may be an independent prognostic factor for epithelial ovarian cancer, particularly high-grade serous ovarian cancer. Approaches to inhibit non-canonical NF-κB signaling should be explored as novel ovarian cancer therapies are needed.


Background
Ovarian cancer is the 5th leading cause of cancer deaths among women in the United States and has the highest mortality among gynecologic cancers, with an overall 5year survival rate lower than 50% [1]. While overall rates of death from cancer have dramatically declined over the past few decades, mortality from ovarian cancer has hovered around 10 women per 100,000 for the last 50 years [2,3]. Long-term survival is poor as patients are most often diagnosed at advanced stage, largely due to nonspecific symptoms and lack of early detection methods [4,5]. Furthermore, while most patients have a dramatic initial response to first-line platinum and taxane-based chemotherapies, the majority will ultimately develop recurrent and treatment-resistant disease [1,6]. This is particularly relevant to high-grade serous ovarian cancer, the most common and deadliest of the five main subtypes when classified by grade and histology: high-grade serous, low-grade serous, endometrioid, clear cell, and mucinous tumors [7][8][9][10]. Targeted treatments aimed at specific pro-tumorigenic gene products, such as vascular endothelial growth factor (VEGF) inhibitors and poly ADP ribose polymerase inhibitors (PARPi), are now being incorporated into the clinic as first-line maintenance therapies and treatment for recurrence [11][12][13][14]. The development of additional treatment modalities to improve ovarian cancer survival requires further understanding of the pathogenesis of this disease.
Activation of nuclear factor-kappaB (NF-κB) signaling has been identified as an important molecular link between inflammation and cancer, and modulates tumor growth, chemotherapy resistance, and immune escape in a variety of malignancies, including ovarian cancer [15][16][17]. NF-κB signaling is mediated by dimerization of a family of five transcription factors that share N-terminal DNA binding Rel homology domains: p50 (and its precursor p105), p52 (and its precursor p100), p65/RelA, RelB, and c-Rel. RelA, RelB, and c-Rel additionally contain C-terminal nuclear localization signals that facilitate nuclear translocation [18]. In homeostatic conditions, homo-and heterodimers of NF-κB members are bound to inhibitor of kappaB proteins (IκB) and sequestered in the cytoplasm. Following upstream NF-κB pathway signaling, inhibitor of kappaB kinase (IKK) proteins phosphorylate IκB, causing its proteasomal degradation and release of the dimer from inhibition, thereby allowing for nuclear translocation and transcriptional activation of anti-apoptotic and pro-proliferative gene targets [18].
NF-κB signaling can be divided into two broad pathways, termed canonical and non-canonical. Activation of canonical NF-κB signaling occurs when proinflammatory ligands bind to cell surface receptors such as tumor necrosis factor receptors or Toll-like receptors, most often resulting in nuclear localization of p65/p50 dimers to initiate transcription of acute phase reactants, inflammatory cytokines, and regulators of apoptosis [18]. Non-canonical NF-κB signaling typically occurs following binding of a ligand to the CD40 receptor, lymphotoxin beta receptor, or B-cell activating factor receptor. Activation of this cell surface receptor results in phosphorylation of NF-κB-inducing kinase (NIK), which then activates IKK to induce processing of cytoplasmic p100 to active p52 with resultant nuclear localization of p52/ RelB dimers [19].
A few studies have evaluated the expression of p65 and other canonical NF-κB pathway members in ovarian cancer, but findings for associations with patient outcomes have been inconsistent [20][21][22][23][24][25]. Associations between high p65 expression and poor ovarian cancer survival have been reported [20,24]; however, the largest study of human tissue found better overall survival for high nuclear p65 expression among 324 high-grade serous cases [21]. Non-canonical NF-κB signaling remains relatively understudied compared to canonical signaling and comparatively little is known about the non-canonical NF-κB pathway and cancer survival. Prior studies have demonstrated increased activation of p52 in breast, lung, prostate, pancreatic, and ovarian cancers [26][27][28][29][30][31][32], and increased expression has been linked to worse prognosis in lung cancer [33]. While pre-clinical in vitro and in vivo studies have demonstrated a variety of mechanisms by which noncanonical NF-κB signaling may contribute to ovarian cancer carcinogenesis and progression, no studies to date have reported associations between non-canonical NF-κB transcription factors and ovarian cancer clinical prognosis [31,32,34,35]. Thus, we undertook this study to evaluate if variation in expression of NF-κB transcription factors, specifically tumor expression of p65 (a major contributor to canonical signaling) and p52 (a major contributor to non-canonical signaling), was associated with survival outcomes among a retrospective cohort of ovarian cancer cases.

Institutional approval
Approval was previously obtained from the Vanderbilt University Medical Center (VUMC) Institutional Review Board for construction of a tissue microarray (TMA) and collection of clinical data for Gynecologic Oncology patients at VUMC from 1994 to 2004; as all patient data was de-identified and discarded tissue from routine clinical care was evaluated, we obtained a waiver of consent from the VUMC Institutional Review Board committee. Data collection and processing were performed in accordance with standard guidelines, including US Federal Policy for the Protection of Human Subjects and the Declaration of Helsinki.

Ovarian cancer tissue microarray
We evaluated a validated TMA constructed from primary ovarian tumor samples from the VUMC Tissue Repository for Ovarian Cancer (TROC) [36,37]. Briefly, tissue samples were collected from patients seeking care at VUMC at the time of primary staging and/or cytoreductive surgical procedures by the Division of Gynecologic Oncology between 1994 and 2004. Inclusion criteria included age > 18 years and a primary epithelial tumor of the ovary. Patient samples were excluded if they were non-epithelial ovarian tumors or were nonprimary (i.e. metastatic or recurrent) tumor of the ovary (Additional File 1). Hematoxylin and eosin stained sections from identified cases were reviewed and a representative formalin-fixed paraffin-embedded tissue block containing > 80% tumor cells was identified for each case. For each specimen, four 1 mm cores from the representative tumor block were included in the tissue microarray. Tumor cores with < 100 tumor cell nuclei were excluded from analysis. Because ovarian cancer classification has been updated over time, our samples underwent additional pathology review to re-classify cases according to the contemporary 2014 WHO guidelines [8,38]. Among the 196 epithelial samples included the current analysis, there were 177 invasive tumors of serous (N = 131), endometrioid (N = 24), mucinous (N = 8), clear cell (N = 11), and other tumor histotypes (N = 3: a carcinosarcoma, a primary squamous cell carcinoma, and a small cell carcinoma); a total of 19 samples were from non-invasive borderline tumors of either serous (N = 12) or mucinous (N = 7) histology.

Clinical database
Demographic and clinical information, including stage, grade, treatment history, and patient outcomes were obtained by manual data abstraction from the VUMC electronic medical record (EMR) and collected using a REDCap database built for the VUMC TROC [39]. Tumor stage was classified based on the International Federation of Gynecology and Obstetrics (FIGO) system, with early stage including stages I and II and late stage including III and IV [40]. Low-grade tumors included low-grade serous tumors and grade 1 and 2 endometrioid and mucinous tumors; high-grade tumors included high-grade serous carcinomas, grade 3 endometrioid and mucinous tumors, clear cell carcinomas, and carcinosarcomas. Date of diagnosis was defined as the earliest date of pathologic confirmation, either by cytologic evidence of adenocarcinoma from ascites or from pathologic evaluation of tissue specimen. Date of disease progression or recurrence after staging and/or debulking surgery in the first-line treatment setting was defined by earliest evidence of increasing measurable disease, as determined by RECIST imaging criteria, development of new or enlarged lesions on physical exam, confirmed ovarian cancer on biopsy of new lesions, or disease progression explicitly stated within the EMR [41]. Date of death was ascertained from VUMC EMR, Tumor Registry, OnCore Clinical Trial Research Databases, and the National Technical Information Services Death Master File of deaths reported to the Social Security Administration. Outcomes were captured through 11/29/2018. Patients without visible pelvic extension of disease at the time of primary surgery were categorized as "not applicable" for debulking. Among patients with advanced pelvic disease, 5 had unresectable disease and were not debulked, and 24 had unknown residual disease. For those who underwent debulking with outcomes known, debulking was characterized in accordance with current guidelines based on manual review of operative reports as optimal (not otherwise specified, no residual disease, residual disease ≤ 1 cm maximum), or sub-optimal (> 1 cm) [42]. Six samples were obtained during interval debulking after neoadjuvant chemotherapy, while 190 samples were obtained during primary tumor reductive surgery prior to any adjuvant chemotherapy and were classified as "chemotherapy naïve." All chemotherapy data was collected from first-line treatment. Among patients who had dates of first-line platinum chemotherapy available, platinum sensitivity was defined as a progression-free interval of greater than 6 months after completion of first-line chemotherapy, while platinum resistance was defined by a progression-free interval of less than 6 months following completion of first-line chemotherapy. Accordingly, patients with resistant or refractory disease (demonstrated progression at the time of first-line therapy) were grouped together for this study. Patients for whom the date of completion of first-line platinum chemotherapy was unknown were categorized as unknown for response to platinum treatment.

Immunohistochemistry (IHC)
We conducted immunostaining for p52 with anti-p52 (C-5, Santa Cruz; sc-7386) on TMA slides utilizing a previously published protocol [33]. Staining of p65 was performed under automated conditions by the Vanderbilt Translational Pathology Shared Resource with mouse monoclonal anti-p65 (F-6, Santa Cruz; sc-8008). Human tonsil, lung parenchyma, and lung adenocarcinoma, were used as positive controls. True negative controls were not feasible as NF-κB signaling is ubiquitous in human tissue. However, we selected commercially available antibodies that have been well-studied and validated in prior IHC studies of NF-κB transcription factors in a variety of human tissues [33,43]. Whole-slide imaging and semi-quantitative measurement of the percentage of epithelial tumor cells with positive cytoplasmic and nuclear expression was performed using the automated Ariol SL-50 Platform in the Digital Histology Shared Resource of the VUMC. Cytoplasmic and nuclear H-scores were calculated for each sample by multiplying the percentage of cells staining positive (0-100) by staining intensity (weak: 1, moderate: 2, or strong: 3), to yield an H-score ranging from 0 to 300. Low and high expression of all markers was dichotomized a priori by median values so that our exposure of interest, expression of p52 or p65, was not defined by our outcome of interest, disease prognosis.

Statistical analysis
Normality of continuous variables, including H-scores for IHC quantification, was visually assessed using histograms; based on this and our sample sizes, we generally used non-parametric statistical approaches. Monotonic correlations between continuous variables were calculated using Spearman's rank correlation coefficient, and the Kruskal-Wallis test was used to compare two or more groups of non-normal data, such as differences in NF-κB expression in relation to clinical covariates. The chi-squared test was used to assess relationships between categorical variables. In line with prior studies of NF-κB pathway effectors in ovarian cancer tissue, we also performed analyses where nuclear and cytoplasmic staining were combined [23]; if either was higher than the median, then expression was considered to be high. Overall survival (OS) was defined as the interval between date of diagnosis to either date of death or censored at the date of last contact. Progression-free survival (PFS) was defined as the interval from date of diagnosis to date of first disease progression after initial staging and/or cytoreductive surgery in the first-line treatment setting, death, or censored at the date of last contact. To minimize potential bias from loss to follow-up, all events after were censored and time was truncated at 15-years post-diagnosis. Survival functions were visualized using Kaplan-Meier plots and differences were assessed by Log-Rank tests. Associations with survival outcomes were quantified using hazard ratios (HR) and 95% confidence intervals (CI) calculated from Cox proportionalhazards regression with calendar time as the time-scale. Adjustment for clinical covariates included known prognostic factors as well as significant associations from our bivariate analysis: age at diagnosis, stage, histologic subtype, grade, and platinum and/or taxane chemotherapy treatment. Maximal adjustment additionally included race, residual disease after debulking, and response to platinum therapy. Combined staining indices enabled us to employ mutually adjusted regression models to disentangle the effects of canonical and non-canonical NF-κB expression. To evaluate the robustness of our findings among all cases (N = 196), we conducted sensitivity analyses among chemotherapy naïve cases (N = 190), invasive cases (N = 177), serous cases (N = 143), late stage cases (N = 137), invasive serous cases (N = 131), late stage serous cases (N = 119) and high-grade serous cases (N = 118). Analyses were conducted in R version 3.4.4 using the ggplot2 and survminer packages, and with the SAS System for Windows (SAS version 9.4) [44][45][46][47]. A p-value ≤ 0.05 was interpreted as statistically significant, with a Bonferroni corrected significance threshold used when considering multiple comparisons.

Baseline characteristics
Among 196 cases from the Vanderbilt TROC TMA, clinical characteristics followed expected distributions (Table 1). Cases had a median age of diagnosis of 58.3 years, median progression-free survival of 1.6 years, and median overall survival of 3.6 years. The majority of cases evaluated were White (91.8%) with advanced stage (70.3%), high-grade (73.5%), serous histology (72.9%), and most were treated with chemotherapy (77.6%). Among 152 patients who received first-line chemotherapy, 150 patients (98.6%) had documentation of receiving platinum agents, and 147 patients (96.7%) had documentation of receiving both platinum and taxane agents. Two patients were noted to have completed first-line chemotherapy, but did not have specific therapeutic agent information available and were assumed to have had at least platinum therapy. A total of 44 patients either declined chemotherapy or did not have any information regarding possible first-line chemotherapy available, the majority of which (66%) had stage I disease.

Associations with clinical characteristics
Regardless of whether cytoplasmic or nuclear staining was assessed, median p52 H-scores were significantly higher among high-grade, advanced stage, and serous tumors, and among patients who were treated with chemotherapy ( Table 2, p < 0.05). Similarly, regardless of whether cytoplasmic or nuclear staining was assessed, median p65 H-scores were significantly higher among high-grade, serous, and high-grade serous tumors (p < 0.05). Following Bonferroni correction for multiple comparisons, expression of p52 and p65 remained significantly higher among high-grade and serous cases (Pvalue ≤ 0.003571). When combined indices were assessed, the same associations with p52 remained significant as in compartmentalized analyses; combined high p52 expression was significantly associated with high grade, advanced stage, serous tumors, and receipt of platinum and/or taxane chemotherapy (all p ≤ 0.002, data not shown). As in compartmentalized analyses, combined high p65 expression was significantly associated with serous tumors (p < 0.001); associations with high-grade and high-grade serous tumors approached significance (p = 0.057, p = 0.076, respectively, data not shown). In all analyses, expression of p52 and p65 was not significantly associated with race, residual disease following surgical debulking, or platinum sensitivity.

Analysis with survival outcomes
Kaplan-Meier analysis indicated that higher than median nuclear p52 expression was associated with significantly worse PFS and OS (Log-Rank p < 0.001) and that higher than median cytoplasmic p52 expression was similarly associated with significantly worse PFS and OS (Log-Rank p < 0.001) (Fig. 2a-b). In contrast, neither cytoplasmic nor nuclear expression of p65 were significantly associated with worse PFS or OS by Kaplan-Meier analysis.
To evaluate if effects of the canonical and noncanonical NF-κB pathways were independent, we first combined cytoplasmic or nuclear staining for each factor and then included both p52 and p65 in mutually adjusted regression models (Table 3). Similar to analyses of separate cytoplasmic and nuclear staining, adjustment for age, stage, histologic subtype, grade, and chemotherapy treatment did not attenuate significance of associations for cytoplasmic or nuclear p52 with PFS (HR 1.53, 95% CI 1.08-2.19, p = 0.018) or OS (HR 1.61, 95% CI 1.12-2.33, p = 0.011). In contrast, higher cytoplasmic or nuclear p65 was not significantly associated with PFS or OS after the same statistical adjustment. When both p52 and p65 were included in mutually adjusted regression models, survival associations with p52 retained significance (PFS: HR 1.70, 95% CI 1.21-2.38, p = 0.002; OS: HR 1.64, 95% CI 1.16-2.32, p = 0.005) while associations with p65 remained non-significant.

Sensitivity analysis
Moreover, because ovarian cancer is a diverse clinical entity, we examined the robustness of our findings from analyses conducted after excluding specific subsets of cases ( Table 4). Regardless of which cases were retained, results were materially unaltered with significant associations with PFS and OS for p52, in-

Discussion
This retrospective cohort study was conducted to test our hypothesis that both canonical and non-canonical NF-κB transcription factors influence epithelial ovarian cancer survival outcomes. We measured cytoplasmic and nuclear expression of p52 and p65 in tissue samples from 196 primary ovarian tumors from the Vanderbilt TROC, and tested associations with patient outcomes. We found that high cytoplasmic and nuclear expression of p52 was associated with worse PFS and OS in both crude and multivariable adjusted models. Nuclear p65 was only associated with worse PFS in unadjusted models. High cytoplasmic and nuclear p52 was independently associated with higher risk of disease progression and death, even in regression models that included adjustment for clinical covariates and p65 expression. Further, these associations remained significant when analyses were limited to high-grade serous ovarian cancer cases.
In line with our findings, prior studies have shown constitutive activation and overexpression of p65, a major marker of canonical NF-κB signaling, in many human epithelial cancers including ovarian, and have also demonstrated increasing expression of p65 along the continuum of borderline ovarian tumors to invasive carcinomas [43,[48][49][50][51][52][53][54]. However, evaluations of the relationship between p65 expression and ovarian cancer prognosis have had inconsistent findings [20][21][22][23][24][25]. Three small studies with 33, 68, and 85 epithelial cases found that those with high cytoplasmic, nuclear, or total p65 expression, respectively, tended to have worse PFS and/ or OS [20,22,23]. In the largest of these three studies, the unadjusted OS association was significant, but was attenuated in multivariable adjusted models [23]. In a larger study of 114 serous ovarian cancer cases, those with high nuclear p65 expression had significantly worse PFS [24]. However, in another study among 324 highgrade serous ovarian cancer cases, high nuclear expression was associated with significantly better OS [21]. Reasons for these discordant findings likely include differences as to whether multivariable adjustment was conducted, what clinical covariates were adjusted for, and how p65 expression was categorized. For example, the threshold for dichotomizing expression as positive or high ranged from 21% of cells with moderate staining to 76% of cells with any staining, and was even selected to optimize survival differences in one study [20][21][22]24]. In the current study, we dichotomized expression based on median values, and found that higher cytoplasmic p65 was not significantly associated with PFS or OS, while higher nuclear p65 was only significantly associated with worse PFS in crude models. The association between higher nuclear p65 and worse OS missed significance in crude models (p = 0.052). Following multivariable adjustment, neither cytoplasmic nor nuclear p65 were associated with survival outcomes. Critically, in models with mutual adjustment for both p52 and p65 expression, associations with OS and PFS were attenuated for p65, indicating that prior associations with the canonical pathway may be mediated by p52 and the non-canonical NF-κB pathway.
To our knowledge, this is the first study to assess the relationship between expression of p52, a major component of the non-canonical NF-κB pathway, and epithelial ovarian cancer prognosis, particularly high-grade serous ovarian cancer. We found that higher p52 was associated with higher risk of disease progression and death, and that these associations were robust to multivariable adjustment, including for expression of p65. While canonical and non-canonical NF-κB signaling are known to have some cross-talk, our mutually adjusted analysis indicates that the effects of p52, a primary contributor to the non-canonical pathway, on ovarian cancer outcomes are independent of canonical NF-κB signaling through p65 expression. We also found that correlations between p52 and p65 expression levels were low, suggesting that these two pathways are not highly inter-connected in epithelial ovarian tumors. While there are common factors known to regulate both pathways in cancer cells to support tumorigenesis, such as IKKα, in general there is limited information about how the two pathways may overlap in ovarian cancer or in other cancer cell types. By contrast, there is stronger evidence showing that each pathway can individually influence cancer cell behavior and promote tumorigenesis across the spectrum from borderline tumors to invasive carcinomas, dependent on factors such as tumor type, localization, and microenvironment [43,55,56]. While this is the first study to demonstrate an association with clinical disease prognosis, several additional lines of mechanistic evidence support a pro-tumorigenic role for non-canonical NF-κB signaling independent of the canonical pathway [35]. We previously found that downregulation of cyclooxygenase-1 (COX-1) expression, an established pro-tumor mediator in ovarian cancer, resulted in reduced expression of multiple noncanonical NF-κB signaling components, including RELB, NFKB2 (p100/p52), CHUK (IKKα), and MAP3K14 (NF-κB-inducing kinase); in contrast, expression of canonical NF-κB pathway members was not changed by COX-1 knockdown [57]. Moreover, overexpression of p52 promotes lung tumorigenesis in mice, and associations have been demonstrated between non-canonical NF-κB signaling, p52-gene targets, and worse prognosis in human lung adenocarcinoma cases [33]. Further studies of in vitro and in vivo models of ovarian cancer have shown that non-canonical NF-κB signaling promotes cell growth and tumorigenicity as well as cancer stem cell self-renewal [31,32,34,35]. Aberrant expression of tissue transglutaminase in ovarian tumors has been implicated as an upstream regulator of p52 expression and non-canonical signaling, ultimately contributing to disease progression and intraperitoneal metastasis [31]. Over-expression of NIK has also been demonstrated in human ovarian cancer cell lines, leading to downstream activation of non-canonical signaling and increased tumorigenicity [32]. Known gene targets of noncanonical NF-κB signaling include important mediators of cell proliferation, such as cyclin D1; cell survival, such as Bcl-x; and tumor invasion and vascular growth, such as MMP-9 and VEGF, all of which are frequently deregulated in cancer cells, contribute to epithelial-tomesenchymal transition, and have been associated with poor prognosis [58][59][60].
Additionally, NF-κB pathway activation is known to contribute to inflammation in the tumor microenvironment  [18,19,61]. Studies in the breast cancer context have indeed demonstrated associations between expression of NF-κB and increased stromal tumor contribution [54]. Although this study focused on epithelial expression of NF-κB transcription factors, future studies should utilize tissue microarrays enriched for stromal compartments to analyze associations of NF-κB signaling with immune infiltration, as well as NF-κB signaling within immune or stromal cells. Given increasing interest in immunotherapy for treatment of cancer, these data may provide insights into mechanisms of disease progression or potential treatment strategies combining immunotherapy with NF-κB modulators. While the canonical NF-κB pathway has been targeted in several solid tumor types using various non-specific NF-κB inhibitors such as bortezomib, thymoquinone, and curcumin, the clinical potential of targeting non-canonical signaling in cancer remains unknown [62][63][64].
Strengths of this study include assessment of both nuclear and cytoplasmic p52 and p65 staining in tumors, multivariable adjusted regression models, and mutual adjustment to parse out separate effects of canonical and non-canonical NF-κB pathway member expression. We additionally performed sensitivity analysis to assess the robustness of our findings, which were materially unaltered among chemotherapy naïve, invasive, serous, late stage, invasive serous, late stage serous, and high-grade serous cases. Our primary limitation was sample size; with only 196 tumor samples stained, we were unable to evaluate associations among all five major subtypes, which are known to have distinct etiologies and prognoses; we were able to conduct analyses among the most clinically common subtype, high-grade serous ovarian cancer, which demonstrated significant associations between p52 expression and disease prognosis [7]. Another possible limitation is that nuclear expression was markedly lower than cytoplasmic expression. However, we dichotomized based on median values, and created combined indices for high nuclear or cytoplasmic staining as in prior studies of NF-κB elements to maximize the generalizability of our results [23]. Further, while nuclear p52 is generally considered to represent noncanonical NF-κB pathway activity, p52 may form heterodimers with all other NF-κB pathway members [65]. While immunohistochemical detection of RelB could also provide information on non-canonical NF-κB pathway activity, a prior study of RelB in ovarian tissue only demonstrated cytoplasmic staining [20]. Further, RelB is highly unstable and is known to preferentially complex with p52 [66]. Therefore, this study focused on evaluation of associations between epithelial ovarian cancer survival and expression of p52, a major component of the non-canonical pathway about which there is limited information. Finally, we recognize that our study population was predominantly White and from a single center, potentially further limiting the generalizability of our study.

Conclusions
In conclusion, we have shown that expression of p52, a major mediator of non-canonical NF-κB signaling, is an important prognostic factor for high-grade serous ovarian cancer, and that this association is independent of canonical NF-κB signaling through p65. While prior work has demonstrated potential mechanisms by which non-canonical NF-κB signaling may contribute to ovarian cancer pathogenesis, this data provides a crucial clinical link that expression of a non-canonical NF-κB transcription factor is associated with ovarian cancer prognosis, and interventions to inhibit non-canonical NF-κB signaling should be explored as novel therapies to limit ovarian cancer progression and optimize survival outcomes.