Intersection of DNA Repair Pathways and the Immune Landscape Identifies PD-L2 as a Prognostic Marker in Epithelial Ovarian Cancer

Simple Summary Here, we examined the interaction between DNA repair proteins and immune biomarkers and their association with survival in 181 cases of epithelial ovarian carcinoma (EOC). We used a panel of 12 antibodies for immunocytochemistry staining of tissue microarray (TMA) consisting of 181 cases. Applying standard statistical methods, we detected that PD-L2 expression was associated with decreased survival in ovarian cancer. This is the first demonstration that increased expression of PD-L2 may serve as a marker for decreased progression-free survival (PFS). Therefore, further investigation into PD-L2 based immunotherapy as a strategy to treat ovarian cancer is warranted. Abstract Background: Targeting DNA repair and immune checkpoint pathways has been the focus of multiple clinical trials. In this study, we explore the association between DNA repair proteins, immune response markers, and clinical outcome in women with EOC. Methods: Immunohistochemical analysis of TMA with 181 EOC samples was used to determine expression levels for DNA repair proteins (PARP, PTEN, p53, H2Ax, FANCD2, and ATM) and immune-markers (CD4, CD8, CD68, PD-L2, PD-L1, and FOXP3). Biomarker expression was correlated to clinical data. Prognostic discriminatory ability was assessed per the combination of biomarkers. Results: Tumor immunity biomarkers correlated with HRD biomarkers. High PD-L2 was significantly associated with high expression of CD8 (r = 0.18), CD68 (r = 0.17), and FOXp3 (r = 0.16) (all, p < 0.05). In a multivariate analysis, PD-L2 (hazard ratio (HR) 1.89), PARP (HR 1.75), and PTEN (HR 1.96) expressions were independently associated with decreased progression-free survival (PFS), whereas PD-L1 (HR 0.49) and CD4 (HR 0.67) were associated with improved PFS (all, p < 0.05). In 15 biomarker combinations, six combinations exhibited a discriminatory ability of >20% for the 4.5-year PFS rate, with four based on PD-L2 (PARP, PTEN, CD4, and PD-L1, 20.5–30.0%). Conclusions: Increased PD-L2 expression is a prognostic marker of decreased survival in EOC. Interaction between tumor DNA repair and microenvironment determines tumor progression and survival.


Introduction
Most women with ovarian cancer present with advanced disease at diagnosis and despite an initial response to treatment, the majority experience recurrence and die of chemoresistant disease. Given the poor response of recurrent disease to traditional chemotherapy, there is a need to develop alternative treatment strategies. Two potential therapeutic targets, i.e., the DNA repair pathway and the immune checkpoint pathway, are being vigorously investigated and both are in clinical trials in ovarian and other cancers.
Exploiting the vulnerability of DNA repair deficient ovarian cancer, PARP inhibitors olaparib, niraparib, and rucaparib have been granted FDA approval for primary or secondary maintenance or treatment of recurrent ovarian cancer [1,2]. Meanwhile, the programmed cell death 1 (PD-1) and its ligand (PD-L1) pathway, which inhibits anticancer response, have been investigated in many tumor types. PD-1 expression in ovarian cancer has been associated with unfavorable prognosis [3]. In preclinical models, disruption of the PD-1 pathway has been shown to enhance tumor immune response [4], suggesting that PD-1/PD-L1 has a role in ovarian cancer treatment. While PD-1 is a transmembrane receptor expressed at the cell surface of T cells, B cells, monocytes, NK cells, and dendritic cells [5], inducible expression of PD-L2 is mainly through Th2-associated cytokines, on the surface of macrophages, dendritic cells, and other immune and non-immune cells [6]. PD-L1 and PD-L2 compete for binding to PD-1, and PD-L2 has two-six-fold higher affinity than PD-L1 [7]. PD-L2 is generally expressed at lower levels than PD-L1 [8], but PD-L2 expression is significantly higher during Th2 responses [9].
Targeting both DNA repair and immune checkpoint pathways has emerged as a concept based on the unique characteristic of DNA repair deficient tumors as genetically unstable and characterized by production of tumor-specific neoantigens. This concept has been recently investigated in clinical trials and a recent molecular subclassification of samples from a trial of niraparib and pembrolizumab in platinum resistant ovarian cancer yielded important results [10]. Nonetheless, it is still largely unknown which biomarkers predict tumor response to immune treatment or combined therapy.
Here, we investigate the correlations between selected DNA repair proteins, the immune checkpoint pathway, and survival in patients with sporadic epithelial ovarian cancer.

Study Population
All patients underwent surgical staging or debulking for epithelial ovarian, fallopian tubal, or primary peritoneal carcinoma. All pathology specimens were collected and reviewed, and tumors were classified according to the World Health Organization (WHO) criteria [11]. The medical records of the patients were retrospectively reviewed under an approved Institutional Review Board protocol (Oregon Health and Sciences IRB 921) in accordance with relevant guidelines and regulations, and informed consent was obtained as required by the IRB. The review included outpatient and inpatient treatment, including surgery and chemotherapy. Survival information for overall survival (OS) and progressionfree survival (PFS) were also collected.

Tissue Microarrays
Ovarian cancer tissue microarrays (TMAs) were constructed using paraffin-embedded archival tissue from 181 eligible patients with epithelial ovarian cancer, as described previously [12,13]. Hematoxylin-eosin (HE) staining was used to select morphologically representative regions and core biopsies were sampled from 3 distinct areas of each tumor to account for tumor heterogeneity. Triplicate 0.6 mm cores were punched from the individual donor formalin-fixed, paraffin-embedded blocks, and transferred to the TMA paraffin-embedded receiver blocks. One section from each TMA was stained with H&E to confirm the presence of the tumor by light microscopy [13].
The rationale of this cutoff analysis was to provide a mechanistic approach to detect a more predictive cutoff for survival as compared with our prior analysis where the biomarker expressions were only dichotomized as 0-1 vs. 2-3. In this way, we identified 2 biomarkers (ATM and FANCD2) that a higher cutoff (0-2 vs. 3) predicted PFS better as Cancers 2021, 13,1972 4 of 13 compared with the historical cutoff (0-1 vs. 2-3). Likewise, we identified two biomarkers (H2Ax and PD-L2) that a lower cutoff (0 vs. 1-3) predicted PFS better vs. the historical cutoff. For consistency purposes, a similar methodology was undertaken for cell counts in our study.
The age cutoff was 70, per prior studies. High-grade serous ovarian cancer was defined as grade 2-3 serous carcinoma, as described before [15]. PFS was defined as the time interval between the initial surgery and the first recurrence/progression of disease, or death from ovarian cancer. OS was the interval between surgery and death from all cause. Data were censored at the last follow-up for patients without these survival events.

Statistical Analyses
The primary step of the analysis was to examine the correlation of examined biomarkers and clinical and pathological factors. The secondary step of the analysis was to assess the prognostic impact of the examined biomarkers.
Continuous variables were expressed as a mean (±SD). Categorical and ordinal variables were expressed as a number (%). Spearman's correlation coefficient was used to examine the statistical significance of biomarkers expressed with correlation coefficient value. For survival analysis, the Kaplan-Meier method was used to construct the survival curves, and statistical difference between the curves was assessed by a log-rank test in univariate analysis. A Cox proportional hazard regression model was fitted to identify the independent prognostic markers for survival (PFS or OS) in the multivariate analysis. In this study, covariates with p-value of less than 0.20 in univariate analysis were chosen for the model construction. This relatively liberal cutoff of the covariate selection was due to the small sample size in our study. The effect size for survival outcome measures was expressed by adjusted-hazard ratio (HR) with corresponding 95% confidence interval (CI).
For a sensitivity analysis, the effects of biomarker combination on survival were examined. Specifically, among the significant biomarkers for PFS on multivariate analysis, combination patterns of any two markers were assessed. Then, among four patterns in each biomarker combination of the two, the absolute difference between the lowest and highest survival rate at the specific point estimate, namely the discriminatory ability for survival rate, was computed. A p-value of less than 0.05 was considered to be statistically significant. All analyses were based on the two-tailed hypothesis. IBM SPSS Statistics, version 25.0 (Armonk, NY, USA) was used for the analysis.

Study Population and Cohort Characteristics
There were 181 cases examined for analysis. Patient's characteristics at baseline are summarized in Table 2. The mean age of the study population was 61.6 years, and the majority of the patients had ovarian cancer (75.1%), grade 3 tumors (86.2%), and stage III-IV disease (90.0%). The most common histology type was high-grade serous (n = 141) in this study population. Complete cytoreduction was recorded in only about one-third of cases (34.1%). All patients received platinum-based first line chemotherapy.

Survival Statistics
The median follow-up time of the patients who were alive at the last visit was 4.7 years (interquartile range 2.9-6.7). Nearly 90% of the study population had recurrence, progression, or death from ovarian cancer (87.7%). There were 140 deaths recorded during the follow-up. The median estimated PFS and OS for the whole cohort was 1.8 (95% CI 1.4-2.1) and 3.4 (95% CI 2.9-3.8) years, respectively. The 4.5-year PFS rate of the entire cohort was 19.7% (95% CI 14.0-26.1). Of note, 4.5-year survival estimates were based on the median follow-up of censored cases.

Prognostic Biomarkers
Independent characteristics for PFS were examined in a multivariate analysis (Table  4). Three biomarkers were found to be prognostic for decreased PFS (PARP, PTEN, and PD-L2). Specifically, moderate-strong expression of PARP was associated with decreased PFS as compared with negative-mild expression (4.5-year rates, 16.1% vs. 25.6%, HR 1.75, 95% CI 1.19-2.59); moderate-strong PTEN expression was associated with decreased PFS

Survival Statistics
The median follow-up time of the patients who were alive at the last visit was 4.7 years (interquartile range 2.9-6.7). Nearly 90% of the study population had recurrence, progression, or death from ovarian cancer (87.7%). There were 140 deaths recorded during the follow-up. The median estimated PFS and OS for the whole cohort was 1.8 (95% CI 1.4-2.1) and 3.4 (95% CI 2.9-3.8) years, respectively. The 4.5-year PFS rate of the entire cohort was 19.7% (95% CI 14.0-26.1). Of note, 4.5-year survival estimates were based on the median follow-up of censored cases.

Prognostic Biomarkers
Independent characteristics for PFS were examined in a multivariate analysis (Table 4). Three biomarkers were found to be prognostic for decreased PFS (PARP, PTEN, and PD-L2). Specifically, moderate-strong expression of PARP was associated with decreased PFS as compared with negative-mild expression (4.5-year rates, 16

Biomarker Combination Patterns and Survival Outcome
According to the results of the PFS analysis (Table 3), the combination pattern of six biomarkers were examined (i.e., PARP, PTEN, PD-L1, PD-L2, CD4, and CD8). Any two biomarkers were chosen, and a total of 15 combinations were assessed for PFS (Figure 2).
Of those, six combinations exhibited a discriminatory ability of >20% point for the 4.5-year PFS rate, of which four combinations were based on PD-L2 (PARP, PTEN, CD4, and PD-L1). The large discriminatory ability for the 4.5-year PFS rate was seen in the combination of PD-L2 and CD4 (30.0%, p = 0.021, Figure 3A) followed by the combination of PD-L1 and PARP (29.1%, p < 0.01, Figure 3B), the combination of PD-L2 and PARP (27.3%, p = 0.05, Figure 3C), and the combination of PD-L2 and PD-L1 (24.3%, p = 0.01, Figure 3D).    Table  3. Then, the 4.5-year progression-free survival (PFS) rates were computed, and discriminatory ability defined as the difference between the lowest and highest rates was estimated and shown by red dots and line (displayed as the secondary axis for intra-group difference). A log-rank test was used for p-values (lowest vs. highest). Subgroups with fewer than ten cases were not interpreted for analysis due to insufficient number. Bars Table 3. Then, the 4.5-year progression-free survival (PFS) rates were computed, and discriminatory ability defined as the difference between the lowest and highest rates was estimated and shown by red dots and line (displayed as the secondary axis for intra-group difference). A log-rank test was used for p-values (lowest vs. highest). Subgroups with fewer than ten cases were not interpreted for analysis due to insufficient number. Bars   Among the 15 combination patterns, there were two patterns with an estimated 4.5-year PFS rate that exceeded 40% (Figure 2). The highest one was the combination of negative PD-L2 and negative PARP expressions (44.4%, Figure 3C), followed by negative PD-L2 and positive CD4 cell counts (41.1%, Figure 3A). This was followed by positive PD-L1 and negative PARP expressions (38.5%, Figure 3B) and negative PD-L2 and negative PD-L1 expression (37.7%, Figure 3D). In contrast, there were three patterns that the estimated 4.5-year PFS rate was <10% (Figure 2). These included negative PD-L1 and positive PARP expressions (9.4%, Figure 3B), negative CD8 cell count and positive PTEN expression (9.2%), and positive PTEN and positive PARP expressions (7.1%).

Discussion
Inhibitors of immune checkpoints have recently shown therapeutic success in several tumor types, yet a significant number of patients do not respond to this class of drugs. To date, no immune checkpoint inhibitor (ICI) has been approved by the U.S. Food and Drug Administration for the treatment of patients with ovarian cancer. In order to successfully apply ICIs to the treatment of ovarian cancer, it is essential to define molecular and other predictors of response to these agents. Several recent observations have suggested that tumors with defects in homologous recombination (HR) pathway display genomic instability, high tumor mutational burden, and an increased repertoire of neoantigens, are more immunogenic, leading to adaptive upregulation of PD-L1 by tumor and sensitization to ICI [16]. Another important observation is that PARP inhibition leads Among the 15 combination patterns, there were two patterns with an estimated 4.5-year PFS rate that exceeded 40% (Figure 2). The highest one was the combination of negative PD-L2 and negative PARP expressions (44.4%, Figure 3C), followed by negative PD-L2 and positive CD4 cell counts (41.1%, Figure 3A). This was followed by positive PD-L1 and negative PARP expressions (38.5%, Figure 3B) and negative PD-L2 and negative PD-L1 expression (37.7%, Figure 3D). In contrast, there were three patterns that the estimated 4.5-year PFS rate was <10% (Figure 2). These included negative PD-L1 and positive PARP expressions (9.4%, Figure 3B), negative CD8 cell count and positive PTEN expression (9.2%), and positive PTEN and positive PARP expressions (7.1%).

Discussion
Inhibitors of immune checkpoints have recently shown therapeutic success in several tumor types, yet a significant number of patients do not respond to this class of drugs. To date, no immune checkpoint inhibitor (ICI) has been approved by the U.S. Food and Drug Administration for the treatment of patients with ovarian cancer. In order to successfully apply ICIs to the treatment of ovarian cancer, it is essential to define molecular and other predictors of response to these agents. Several recent observations have suggested that tumors with defects in homologous recombination (HR) pathway display genomic instability, high tumor mutational burden, and an increased repertoire of neoantigens, are more immunogenic, leading to adaptive upregulation of PD-L1 by tumor and sensitization to ICI [16]. Another important observation is that PARP inhibition leads to an accumulation of cytosolic double-stranded DNA, and thereby activates the cytoplasmic DNA sensor cyclic guanine monophosphate/adenosine monophosphate synthase/stimulator of interferon genes (cGAS/STING) [17]. This pathway triggers activation of a signaling cascade connecting cGAS to signal transducers, including STING and TBK1, and eventually to transcription factors (mainly IRF3 and nuclear factor-κB) that collectively induce a type I interferon response, and thereby turn "cold" non-T cell-inflamed tumors into "hot" T cell-inflamed tumors. The same principle molecular mechanism was the rationale for the promising antitumor activity seen in a recent trial (TOPACIO/KEYNOTE-162), using niraparib and pembrolizumab in platinum resistant ovarian cancer [18]. Furthermore, a recent large collaborative initiative utilized the tumor samples from this same trial to analyze immune predictors of tumor response. The mutational signature reflecting the combination of defective DNA repair and a positive immune score was found to be associated with an improved outcome. In contrast, the absence of either defective DNA repair or positive immune score was associated with no response to combination treatment. In addition, single-cell spatial analysis revealed prominent interactions of CD8 + T-cells with PD-L1 + macrophages and PD-L1 + tumor cells as mechanistic determinants of response [19].
In light of these significant contributions for elucidating the biomarkers of ICI sensitivity, we set out to determine the connections between immune and DNA markers in a cohort of 181 patients with sporadic ovarian cancer. As the BRCA1 immunocytochemical staining was overwhelmingly and repeatedly positive in the majority of our samples, we selected six other DNA repair protein markers, based on our prior experience [14]. As expected, the current cross analysis of DNA repair markers and immune markers revealed positive correlations between several markers from the two groups. Of particular interest is that PD-L2 expression was associated with overexpression of CD8, CD68, and FOXp3 suggesting that PD-L2 may induce response of different classes of T cells as well as macrophages.
Survival outcomes depend on the cancer itself and also on the tumor microenvironment of HGSOC which is rich in immune cells and tumor-associated macrophages [20]. An abundance of cytotoxic CD8 + T cells as well as CD4 + in ovarian cancer is a prognostic indicator of greater PFS and OS [21][22][23][24][25]. In our series, we show similar results with high numbers of CD4 and CD8 cell counts associated with prolonged PFS. In the multivariate analysis, the expressions of PARP, PTEN, and PD-L2 were associated with decrease survival. PARP expression is thought to increase the capacity of cells to repair DNA breaks accounting for decreased survival. PTEN is traditionally considered to be a tumor suppressor and its loss is associated with poor prognosis. However, its expression in HGSOC is heterogeneous and effects of PTEN may vary depending on its cytoplasmic or nuclear localization which was not considered in this study. Interestingly, in our study PD-L1 was found to be an independent prognostic factor associated with improved PFS. While several studies have indeed suggested that PD-L1 is a good prognostic (and therapeutic) target in ovarian cancer, the results are inconsistent and many other studies have found no correlation between PD-L1 expression by ICH and PFS or OS in ovarian cancer [26].
A novel finding in this study is the correlation of PD-L2 expression with decreased PFS. PD-1 ligands, PD-L1 and PD-L2, bind to inhibitory molecule PD-1 and together play a key role in the induction of immune tolerance in the tumor microenvironment. While the PD-1/PD-L1 pathway (pembrolizumab, nivolumab/avelumab, atezolumab) has been extensively studied in cancer, including ovarian, the role of PD-L2 in cancer and immunity is less clear and the therapeutic role for PD-L2 inhibition has not been broadly explored. Nevertheless, our results corroborate work of few other studies [9]. Prior studies have found an impaired survival in patients with PD-L2 expressing tumors, yet this trend has rarely reached statistical significance, and in the majority of studies, PD-L2 was expressed in only a minority of cases [27][28][29]. In a cohort of 70 patients with ovarian cancer, the majority of the cases were negative or weakly positive and although PD-L2 expression was correlated with an impaired survival, this did not reach statistical significance [30].
These results highlight the need to explore the potential therapeutic role of PD-L2 as a single factor or in combination. Yearley et al. [31] described a positive response of PD-L2+ /PD-L1 tumors to pembrolizumab, suggesting that the PD-1-PDL-2 axis may function independently of PD-L1 [31,32]. Moreover, Ahmad et al. characterized a CD8 and CD4 T cell PD-L2 specific T cells and showed that they do not cross react with PD-L1specific T cells [32]. They postulated that PD-L2 specific T cells activated by vaccination can exhibit anticancer immunity by directly killing target cells, or by indirectly altering the microenvironment through cytokine release [32].
According to our combinatorial analysis, the lack of PD-L2 expression combined with the lack of PARP expression were the strongest predictor of increased survival. Additionally, positive PD-L2 expression together with either CD4 or PARP expression was associated with increased PFS and points to known synergistic mechanisms that can be further explored. All combinatorial patterns suggest that exploring these leads may confirm the known effects of these combinatorial pairs and possibly reveal novel mechanisms of PD-L1 and PD-L2 interactions.
Our study has several limitations. First, we only selected to test a limited number of DNA repair proteins based on their known prognostic relevance in ovarian cancer and availability of reliable antibodies, yet the list is not comprehensive. Second, this is a descriptive study based on immunocytochemistry and though we tested several antibodies to determine the optimal antibody, the results should be validated by additional techniques in additional cohorts. In addition, BRCA1 antibody had suboptimal performance while BRCA2 antibody was not tested because of its known heterogenous immunochemical results. Finally, the utility of biomarker cutoffs proposed in the analysis has not be validated in an independent cohort and warrants further validation study.

Conclusions
In conclusion, we report a previously unrecognized association of PD-L2 with decreased survival in ovarian cancer. To the best of our knowledge, this is the first demonstration that increased expression of PD-L2 may serve as a marker for decreased progressionfree survival, providing justification for further investigation into the PD1/PD-L2 axis and PD-L2 based immunotherapy as a strategy to treat ovarian cancer.