Aggressive and recurrent ovarian cancers upregulate ephrinA5, a non-canonical effector of EphA2 signaling duality

Erythropoietin producing hepatocellular (Eph) receptors and their membrane-bound ligands ephrins are variably expressed in epithelial cancers, with context-dependent implications to both tumor-promoting and -suppressive processes in ways that remain incompletely understood. Using ovarian cancer tissue microarrays and longitudinally collected patient cells, we show here that ephrinA5/EFNA5 is specifically overexpressed in the most aggressive high-grade serous carcinoma (HGSC) subtype, and increased in the HGSC cells upon disease progression. Among all the eight ephrin genes, high EFNA5 expression was most strongly associated with poor overall survival in HGSC patients from multiple independent datasets. In contrast, high EFNA3 predicted improved overall and progression-free survival in The Cancer Genome Atlas HGSC dataset, as expected for a canonical inducer of tumor-suppressive Eph receptor tyrosine kinase signaling. While depletion of either EFNA5 or the more extensively studied, canonically acting EFNA1 in HGSC cells increased the oncogenic EphA2-S897 phosphorylation, EFNA5 depletion left unaltered, or even increased the ligand-dependent EphA2-Y588 phosphorylation. Moreover, treatment with recombinant ephrinA5 led to limited EphA2 tyrosine phosphorylation, internalization and degradation compared to ephrinA1. Altogether, our results suggest a unique function for ephrinA5 in Eph-ephrin signaling and highlight the clinical potential of ephrinA5 as a cell surface biomarker in the most aggressive HGSCs.


Results
High EFNA5 expression correlates with poor HGSC patient survival. Unbiased analyses of different OC patient cohorts have revealed survival associations for several genes encoding Eph receptors (EPHA/B) and ephrin ligands (EFNA/B), including contradictory results 16,18,[20][21][22]24 . To systematically investigate all the 14 Eph receptors and 8 ephrin ligands in clinical HGSC, we analyzed overall survival (OS) and progression-free survival (PFS) in patients with 40% highest versus 40% lowest gene expression for all the Ephs and ephrins first using TCGA HGSC mRNA dataset 35  For result validation, we next used seven independent OC datasets (n = 815 HGSC patients) from a publicly available database with pooled mRNA data from various OC studies (curatedOvarianData 36 ). Across these datasets as well as in TCGA, the genes for ephrin ligands were variably expressed; EFNA1, EFNB1 and EFNB2 showing relatively high and EFNA2 low global expression (Supplementary Fig. 2A, B). The overall expression levels of the receptors EPHA1, EPHA2, EPHA4 and EPHB2 were higher than those of EPHA3, EPHA5, EPHA6, EPHA7, EPHA8, EPHA10 and EPHB1 (Supplementary Fig. 2A, B).
With these OC datasets, we performed multivariate analyses for OS, considering also residual tumor after surgery and FIGO stage. The ligand encoding genes EFNA5 and EFNB2 correlated with poor OS (HR > 1.00, p < 0.05) in the combined validation datasets (Fig. 1I, Supplementary Fig. 3A, B). The receptor genes EPHA2, EPHB2 and EPHB4 also correlated with poor OS, whereas EPHA1 correlated with favorable OS (Supplementary Fig. 4A 39 . All ephrin genes except EFNA2 and EFNB1 were expressed variably, EFNA5 being relatively high in all these cell lines ( Fig. 2A; the Cancer Cell Line Encyclopedia, CCLE, https:// porta ls. broad insti tute. org/ ccle). All the six HGSC cells were characterized by particularly high EPHA2, whereas the other Ephs were variably expressed and/or low ( Fig. 2A) in a relatively similar manner as in the patients (see Supplementary Fig. 2).
To compare the effects of ephrinA1 and ephrinA5 in EphA2 activation/phosphorylation, we used the epithelial HGSC cell lines OVCAR3 and OVCAR4 29 , which both expressed these three genes, for EFNA1 and EFNA5 gene knockdown. Using specific pooled siRNAs, the ephrins were silenced with > 70% efficiency (Fig. 2B, C). In both cell lines, EphA2 was constitutively phosphorylated at Y588 and S897 amino acid residues (Fig. 2D)   www.nature.com/scientificreports/ of the ligand-mediated signaling, EFNA5 silencing left EphA2-pY588 essentially unaltered in OVCAR3 and even increased this tyrosine phosphorylated receptor in OVCAR4 (Fig. 2D, E). Compared to control cells, the oncogenic EphA2-pS897 was increased after silencing of either EFNA1 or EFNA5 (Fig. 2D, F), whereas total EphA2 was less affected, or increased after EFNA5 silencing in OVCAR3 (Fig. 2D, G). These results suggest that while endogenous ephrinA1 mediates the canonical EphA2-Y588 phosphorylation, ephrinA5 has a different function with potential to instead limit this tumor-suppressive EphA2-pY588.
EphrinA5 is an inefficient activator of EphA2-pY588 signaling and receptor internalization compared to ephrinA1. To better understand the unexpected contribution of ephrinA5 to EphA2 signaling, OVCAR3 and OVCAR4 were treated with dimeric (Fc-tagged) and monomeric (His-tagged) recombinant ephrinA1 and ephrinA5 for 120 min. Both ephrinA1 and ephrinA5 were detected bound to the cells after the treatment ( Supplementary Fig. 5B). As described above, EphA2 was constitutively phosphorylated both at Y588 and S897 residues in OVCAR3 and OVCAR4 (Fig. 3A). While treatment with ephrinA1 (Fc-and His-tagged) increased the tumor-suppressive EphA2-pY588 in these cells, as expected upon ligand-mediated receptor activation, this phosphorylation was not altered by ephrinA5 (Fc-and His-tagged) in OVCAR3 and remained lower in OVCAR4, compared to the respective ephrinA1 (Fc-and His-tagged) treated cells (Fig. 3A, B). The oncogenic EphA2-pS897 was reduced by ephrinA1 (Fc-and His-tagged) in OVCAR4, and by ephrinA5-His treatment in both cells (Fig. 3A, C). Total EphA2 remained unaltered/unaffected upon treatments with the exogenous ligands under these experimental conditions, except for a reduction upon ephrinA1-His treatment in OVCAR4 ( Fig. 3A; Supplementary Fig. 5C).
To elucidate putative differences in EphA2 internalization occurring after activation by ephrinA1 or eph-rinA5, we assessed the localization of this receptor by immunofluorescence in HGSC cells after treatment with the dimeric Fc-tagged ephrins for 45 min (treatment time adjusted to efficient EphA2 activation by the dimeric ephrins 40 ). Prominent cell surface localization of EphA2 detected in control cells was lost and the total EphA2 intensity reduced in OVCAR3, OVCAR4 as well as in the third, more mesenchymal HGSC cell line, OVCAR8 29 , after ephrinA1 treatment (Fig. 3D, E), indicating EphA2 internalization and degradation. Markedly, ephrinA5 treatment had minor effects on EphA2 localization and resulted in only slightly reduced signal intensity in OVCAR3 and OVCAR8 (Fig. 3D, E). Of note, EphA2 intensity remained significantly higher in these HGSC cells after ephrinA5 treatment compared to ephrinA1 treatment ( EphrinA5 is expressed in the cancer cells and strongly associates with the most aggressive HGSC subtype of ovarian cancer. To   To investigate the relevance of this ephrin ligand among the different OC subtypes, we assessed ephrinA5 expression in 392 epithelial OC and benign tissue samples by IHC of tissue microarrays (TMAs). See examples for grading from negative to strong expression in Fig. 4B. Among all epithelial OC subtypes and benign tissues, HGSC comprising the most aggressive tumors had the highest overall ephrinA5 expression ( Fig. 4C; p = 9.00 × 10 −6 ), supporting the strong association between ephrinA5 and OC malignancy.

EphrinA5 increases during disease progression in patient-derived HGSC cells. To investigate
the changes in ephrinA5 expression upon disease progression, considering the prominent protein expression particularly in the cancer cells, we collected ascites cells longitudinally from a HGSC patient at three time points including diagnosis, interval (after three rounds of platinum chemotherapy) and relapse. The isolated cells were 84-94% positive for the nuclear HGSC marker PAX8 (Fig. 5A). Immunofluorescence of ephrinA5 coupled with the cell surface marker CD44 allowed us to investigate the expression of this ligand in the samples enriched for cancer cells. EphrinA5 was essentially undetectable in this diagnosis sample and expressed at low levels in HGSC cells from the chemotherapy-treated interval time point of the same patient (Fig. 5B). Notably, cells from the relapse time point had higher expression of ephrinA5 compared to cells from the earlier diagnosis and interval stages of this patient's longitudinal samples (Fig. 5B, C).
To uncover more comprehensive expression changes of the ephrins and Eph receptors upon HGSC progression, we analyzed RNA sequencing data of ascites samples from four HGSC patients at the times of diagnosis and relapse of the disease. These samples were characterized by high expression of the epithelial markers KRT7 and EPCAM, and lower to very low expression of immune cell (CD45, CD3) and fibroblast (FAP, PDGFRB) markers (Fig. 5D). In contrast to decreased EFNA3 expression, EFNA5 was significantly increased from the time of diagnosis to relapse ( Fig. 5E; EFNA3: p = 0.015, EFNA5: p = 0.038). In these same HGSC samples, EFNB2 was likewise increased, whereas EFNA1, EPHA1, EPHA2, EPHA4, EPHB2 and EPHB4 expression remained essentially unaltered ( Fig. 5E; EFNB2: p = 0.005). These results correlate with the above survival associations, where low EFNA3 showed association to short PFS and high EFNA5 and EFNB2 were associated with poor OS (see Fig. 1). www.nature.com/scientificreports/ Altogether, these data suggest that the uniquely acting ephrinA5 can contribute to the abysmal HGSC patient outcome upon upregulation in the malignant HGSC cells and during progression of the disease.

Discussion
The tumor-suppressive signaling elicited by EphA2-ephrinA complexes is often halted in aggressive cancers via EphA2 receptor overexpression coupled with ephrinA ligand downregulation 12 . Yet, the epithelial OC cells display variable ephrinA/EFNA expression (see Fig. 2A) and high mRNA expression of EFNA5, encoding the ephrinA5 ligand, has been repeatedly associated to poor survival in OC patients 18,20,21 . Here we show that ephrinA5 protein levels are high specifically in the most aggressive HGSC and further upregulated transcriptionally upon disease progression. Functionally, we describe an unusual signaling pattern for this ligand in HGSC cells, whereby it leaves unaffected or even impairs the canonical, tumor-suppressive tyrosine phosphorylation and subsequently induced internalization and degradation of EphA2 receptor. Unbiased analyses of independent OC cohorts have revealed survival associations of individual Eph and ephrin genes, resulting in an incomplete understanding of the clinical importance of the distinct receptors and ligands in this disease 16,18,[20][21][22]24 . Here we report that, in line with the canonical function of ephrins in promoting tumor-suppressive Eph receptor signaling 12 , low EFNA3 mRNA levels are associated to poor OS and short PFS in the TCGA HGSC dataset. In contrast, high EFNA5 expression is associated to poor survival in HGSC. In agreement with previous reports [16][17][18][20][21][22] , our systematic analysis of all the 22 genes for Eph receptors (EPHs) and ephrin ligands (EFNs) concurs with the association of high EFNB2, EPHA2, EPHB2 and EPHB4 with poor OS. For the Ephs/ephrins with previously reported inconsistent findings 16,18,22,24 , our analysis supports the correlation of low EPHA1 mRNA expression with poor OS, whereas the analyzed data lacked significant association for EFNA1.
Recurrence and development of treatment resistance in HGSC patients remains an unmet clinical need and thus understanding these processes at the functional level is essential to provide new therapeutic opportunities. www.nature.com/scientificreports/ We show here that ephrinA5 protein as well as mRNA expression in ascites-derived HGSC cells was increased in post-treatment samples upon disease progression. Previous studies have explored the interactions of EphA2 with ephrinA1 and ephrinA5 and the corresponding crystal structures for these Eph/ephrin complexes have been resolved [41][42][43] . However, the signaling mechanisms elicited by endogenous ephrinA5 in cancer cells as well as the differences between ephrinA1-and ephrinA5-mediated EphA2 receptor activation have remained elusive. Further studies will be of interest to fully understand the complex Eph/ephrin signaling mechanisms and outputs, with possible heterotypic Eph-Eph and Eph-growth factor receptor crosstalk, alternative binding of ephrinA ligands to EphB receptors (including ephrinA5-EphB2 interaction, as occurs in neural development) and vice versa (e.g. ephrinB2-EphA4, as reported in the context of monocyte adhesion to endothelial cells), proteolytic ligand and/or receptor cleavages, as well as ephrin ligand crosstalk with other RTKs 28,44,45 . Nonetheless, our current results provide evidence indicating that ephrinA5 can hinder the tumor suppressive signaling normally coupled with degradation of EphA2, thus allowing the oncogenic receptor to function at the HGSC cell membranes. We have recently described a strategy to sensitize HGSC cells to chemotherapy by blocking the platinum-induced, pro-tumorigenic EphA2-pS897 signaling and simultaneously restoring EphA2 phosphorylation at Y588 29 . Disrupting the balance of the canonically acting ephrinAs and the tumor promoting ephrinA5 could likewise help to restore the tumor-suppressive signaling through EphA2-Y588 phosphorylation in order to reduce tumor malignancy and development of the increasingly therapy resistant relapses. In this study we describe a specific link between high ephrinA5 protein expression and HGSC, the most aggressive epithelial OC subtype, and show that ephrinA5 increases upon disease progression and associates to poor survival. We further report a non-canonical signaling function for ephrinA5 in HGSC cells, whereby, opposite to the tumor-suppressive signaling elicited by EphA2-ephrinA1 complexes, ephrinA5 even limits the canonical activation of EphA2 through phosphorylation at Y588. Our findings suggest the potential use of ephrinA5 as an indicator of disease subtype and progression stage, as well as its relevance as a putative survival biomarker in HGSC.  Treatment-naïve, primary HGSC tumor was collected at the time of debulking surgery and subsequently formalin-fixed and paraffin-embedded to later be used for immunohistochemical stainings.

Methods
Fresh patient-derived ascites fluid was collected at the time of diagnosis and longitudinally at interval and relapse-stages and processed for ex vivo cultures as described by Moyano-Galceran et al 29 . Longitudinal ascitesderived cells were simultaneously grown on glass coverslips and stained by immunofluorescence to assess cancer cell purity.

RNA sequencing. Ascites-derived cells from diagnosis and relapse-stage (n = 4 samples/clinical time point)
were used for paired-end RNA-sequencing. Total RNA was extracted from fresh ascites-derived cells by using RNeasy kit (Qiagen) with DNase I treatment according to manufacturer's instructions. RNA quality and concentration were tested by Bioanalyzer 2100 (Agilent). Paired-end 100 bp RNA-seq producing around 60M reads was carried out on Illumina HiSeq4000 platform and the data was processed using SePIA, a comprehensive RNA-seq data processing workflow 46 . Read pairs were trimmed using Trimmomatic, and trimmed reads were aligned to the reference genome (GRCh38.d1.vd1) using STAR (version 2.5.2b), allowing up to 10 mismatches, and all alignments for a read were output 47 . Gene level expression was quantified as log2(TMP + 1), where TPM is transcript per million as calculated by eXpress (version 1.5.1-linux_x86_64) 48 . See the section Data Availability for information on data deposition at the European Genome-phenome Archive.
Immunohistochemistry. The  Image analysis and statistics. Quantitative assessment of immunoblots was performed using ImageJ.
Immunofluorescence images were obtained using Zeiss AxioImager.Z1 microscope outfitted with an ApoTome optical sectioning device and a 20 × Plan Apochromat air objective. Images were processed using Zen 2012. Quantification was performed with CellProfiler 3.0.0., by using DAPI-stained nuclei as primary objects, then propagating the cytoplasmic area from nuclei using a membrane-localizing marker to map the edges of cytoplasm, and finally forming the analysis area by subtracting the nuclear area from each cytoplasm. Mean signal intensities were then obtained from each image and averaged for each treatment/condition. The experimental protein and RNA analyses were performed at least in triplicates, as indicated in the corresponding figure legends, and the statistical significance was determined using two-sided Student t-test. p values are depicted as *p < 0.05; **p < 0.01; ***p < 0.001. For the TMA scoring, a 4-point scale (0: negative, 1: weak, 2: intermediate, and 3: strong) was used to evaluate the maximum intensity of ephrinA5 staining in each core. The scoring was performed independently and in a blinded manner by two investigators, and in case of disagreement, a consensus score was established. Correlation analyses from TMA data were performed through Pearson chi-square test.
The 2011 Agilent 244 K microarray-sequenced HGSC TCGA dataset 35 was used for the survival association analysis. Overall survival (OS) was defined as the interval from the date of initial surgical resection to the date of last follow-up or death, in months. Progression-free survival (PFS) was defined as the timeframe from the date of initial surgical resection to the date of progression, recurrence, or last follow-up, in months. Cases that were healthy (n = 8), from different primary site than ovary/fallopian tube (n = 4), low-grade (n = 87), or with no information regarding histology or neoadjuvant chemotherapy (n = 48) were excluded from the analysis. Differences in 5-year and 13-year OS, as well as PFS in 40% highest versus 40% lowest Eph-receptor/ephrin expressing patients were estimated using log-rank test. The 40% cutoff was adopted to include the maximum number of samples while minimizing the noise produced by ambiguous samples. K-means clustering was also used for analyzing 5-year OS. Cox multivariate analysis integrating the variables age at diagnosis (median cutoff of 59 years), residual tumor after surgery (no vs yes) and FIGO stage (I-IIIB vs IIIC-IV) was also performed to evaluate 5-year OS. TCGA statistics were performed in SPSS, versions 24.0 and 26.0.
The curatedOvarianData, a manually curated data collection of microarray data for altogether 2970 OC patients from 23 studies with documented clinical metadata, was used to validate our Eph-ephrin survival association results in a broader OC setting 36 . The datasets were filtered to include only those that had information www.nature.com/scientificreports/ on vital status, time to death/last follow up, residual tumor after surgery and FIGO stage, and contained at least 60 cases and expression for at least 1000 genes. The TCGA cohort (both the RNA sequencing and microarray datasets) in curatedOvarianData was excluded to generate the validation cohort. From the selected datasets (n = 7), cases that complied with sample type (tumor), histology type (serous), stage (late) and grade (high) were included in the analyses. The batch corrected mRNA expression of the 8 ephrin ligands and 14 Eph receptors in these filtered datasets (n = 815 patients) was plotted in a heat map using Heatmapper 50 and compared to the expression from the TCGA cohort (including both RNA sequencing and microarray data). Multivariate analyses considering residual tumor after surgery (optimal vs suboptimal debulking) and FIGO stage (I to IV) were performed using RStudio. The hazard ratios with 95% confidence interval (CI) and the corresponding significance for all Eph-ephrin associations to OS were obtained and plotted in forest plots using RStudio. The Affymetrix U133 + 2 mRNA sequencing data from the Cancer Cell Line Encyclopedia (CCLE) was obtained from the dataset uploaded on 29th of September 2012 and used for the Eph/ephrin expression analysis in the cell lines. The probe IDs are collected in Table 1.

Data availability
All data generated or analyzed during this study are included in this published article and its supplementary information file. The bulk RNA sequencing data that support the findings of this study are available at European Genome-phenome Archive: EGAD00001006456 (RNA sequencing data of HGSC samples, under the study: EGAS00001004714). Publicly available data were obtained through The Cancer Genome Atlas (http:// cance rgeno me. nih. gov/), Bioconductor (http:// www. bioco nduct or. org/ packa ges/2. 12/ data/ exper iment/ html/ curat edOva rianD ata. html) and The Cancer Cell Line Encyclopedia (https:// porta ls. broad insti tute. org/ ccle). Further details are available from the corresponding author on reasonable request.