Metabolomic and Lipidomic Profiling Identifies The Role of the RNA Editing Pathway in Endometrial Carcinogenesis

Endometrial cancer (EC) remains the most common malignancy of the genital tract among women in developed countries. Although much research has been performed at genomic, transcriptomic and proteomic level, there is still a significant gap in the metabolomic studies of EC. In order to gain insights into altered metabolic pathways in the onset and progression of EC carcinogenesis, we used high resolution mass spectrometry to characterize the metabolomic and lipidomic profile of 39 human EC and 17 healthy endometrial tissue samples. Several pathways including lipids, Kynurenine pathway, endocannabinoids signaling pathway and the RNA editing pathway were found to be dysregulated in EC. The dysregulation of the RNA editing pathway was further investigated in an independent set of 183 human EC tissues and matched controls, using orthogonal approaches. We found that ADAR2 is overexpressed in EC and that the increase in expression positively correlates with the aggressiveness of the tumor. Furthermore, silencing of ADAR2 in three EC cell lines resulted in a decreased proliferation rate, increased apoptosis, and reduced migration capabilities in vitro. Taken together, our results suggest that ADAR2 functions as an oncogene in endometrial carcinogenesis and could be a potential target for improving EC treatment strategies.

specifically associated with human endometrium 7 and onset and progression of EC 8,9 . Metabolomics defines the end point of cellular processes and hence provides a readout of current physiological status of the system 10 . Thus, metabolomics, lipidomics and glycomics have emerged as promising tools for clinical and translational research 11 . Rapid advancement of metabolomics technologies such as ultra-performance liquid chromatography mass spectrometry (UPLC-MS), enable comprehensive interrogation of the human metabolome and lipidome [12][13][14][15] . Bioinformatics analyses of these data is likely to augment the discovery of new clinical and pharmacological targets 16 .
It is known that changes at the transcriptomic levels generate a new source of complexity that promotes initiation and progression of cancer and other diseases [17][18][19] . The most common RNA editing events are mediated post-transcriptionally by the Adenosine deaminases acting on RNA (ADAR) family of enzymes 20 . The ADAR gene family catalyzes the deamination of adenosine that is converted to an inosine creating a dysregulation of the adenosine/inosine (A/I) ratio in the cell. Adenosine to inosine (A-to-I) editing can lead to amino acid recoding events 21 since inosine is recognized as a guanosine. This family includes three enzymes, ADAR1 (UniProtKB P55265), ADAR2 (UniProtKB P78563) and ADAR3 (UniProtKB Q9NS39); also known as ADAR, ADARB1 and ADARB2, respectively 22 . ADAR1 and ADAR2 are ubiquitously expressed and, differently that the brain specific ADAR3, they show catalytic activity 23 .
The overall goal of this study was to identify altered metabolic pathways in EC. Characterization of the metabolome and lipidome of EC tumors was performed using a high resolution mass spectrometry approach in conjunction with UPLC-MS. Pathway validation was performed with an independent set of samples that for the first time, yielded insights into dysregulation of the RNA editing pathway in endometrial tumors. Alterations of the RNA editing pathway correlated with high histological grade and EC serous subtypes that are indicators of poor prognosis in EC. Finally, the expression of ADAR editing enzymes was modulated in vitro to confirm an important oncogenic role of this pathway in EC cell proliferation, apoptosis and migration. These results are instructive of the role of this pathway in EC tumor progression.

Results
Untargeted metabolomics profiling of EC human tissue samples. To better understand alterations in metabolic pathways occurring in EC, we performed metabolomics/lipidomics untargeted discovery using UPLC-ESI-TOF-MS of tumor and non-tumoral tissue samples. The sample set included a total of 56 samples: 39 EEC tumors from different FIGO stages, including 10 stages IA, 9 stages IB, 10 stages II and 10 stages III; and 17 benign endometrial tissues (see patient details in Table 1). All patients included in the study were postmenopausal women and did not receive any treatment before surgery. Pre-processing of TOFMS data yielded a total of 8,146 features in the positive and 7,558 in the negative electrospray ionization mode, respectively.
In order to define a generic metabolomic profile of EC, initially, we combined all tumor samples from EEC patients in one group (n = 39) and compared them against matched control tissues (n = 17). Inherent differences in metabolomic profiles were visualized using descriptive Principal Component Analysis (PCA) plot that showed clear separation between tumors and controls (Fig. 1A). Subsequently, t-statistics was used to select 80 metabolites that showed significant variation (adjusted p-value < 0.05) between the two study groups and fold change (FC) values over 2 and below 0.5 (Supplementary Table 1). We confirmed the putative identity of a subset of 42 metabolites using tandem mass spectrometry (MS/MS) ( Table 2). An example of the fragmentation pattern of two verified metabolites is shown in Supplementary Figures 1 and 2. We found a significant dysregulation in the lipid metabolism, with an important number of glycerophosphocholines (PCs), phosphatidylserine (PSs), phosphatidylethanolamines (PEs), phosphatidylinositols (PIs), and phosphatidylglycerol (PGs) that were upregulated in the endometrial tumor tissue (detailed in Table 2 Few studies have used high-throughput approaches to study the changes in metabolomic and lipidomic profiles that underscore EC progression. Hence, in order to understand the metabolic phenotype associated with EC development, we interrogated profile differences in 29 tumor tissues restricted to the uterine cavity (FIGO stages I and II) compared to 10 tumors showing lymph node dissemination (FIGO stage III). We found a set of features to be significantly dysregulated (adjusted p-value < 0.05) among the different FIGO stages (Supplementary Table 2). The PCA plots showing the segregation of the groups and the heat map showing the expression of several features in normal endometrium and in different FIGO stages of the tumor are represented in Fig. 1B and C. A subset of seven metabolites was verified by MS/MS (Table 3). The dysregulation observed in lipids (two PCs and three PEs) in tumor tissues compared to controls was significant along tumor progression. Additionally, we observed that arachidonic acid and UDP-N-acetyl-D-galactosamine are important contributors in the tumor progression, as they appeared to be upregulated in advanced compared to early FIGO stages (Supplementary Figure 3).

Increased expression of ADAR family of enzymes in human EC tumors.
Among the metabolites identified in EC tissues, we were particularly interested in the dysregulation of nucleoside inosine since the relative abundance of this metabolite was significantly higher in EC tumors. To our knowledge, the underlying impact of alterations of endogenous levels of inosine has never been investigated in EC. Dysregulated levels of adenosine and inosine (A/I ratio) can be attributed to the modulation of the A-to-I editing pathway 24 . Consequently, we studied the status of this pathway in EC by analyzing the expression level of members of the A-to-I editing enzymes family (ADAR1 and ADAR2) in three independent sets of samples by immunohistochemistry (IHQ) ( Table 4). The first set included the evaluation of 20 EEC samples and their corresponding paired healthy tissues; the second set included 36 EEC tumors diagnosed at different histological grades (low grade, n = 12; intermediate grade, n = 13; high grade, n = 11); and the third set allowed to differentiate between histological subtypes, and so, 78 EEC and 29 NEEC tissue samples were included. Although ADAR1 and ADAR2 staining was clearly localized in the nucleus of epithelial, stromal, and endothelial cells, we specifically analyzed the staining of the epithelial tumor cells (Fig. 2B,D,F and Supplementary Figure 4B). Interestingly, no staining was observed in cells undergoing mitosis (Supplementary Figure 4A). Our results demonstrated that ADAR1 and ADAR2 were both significantly increased in tumor samples compared to healthy endometrial tissue ( Fig. 2A), confirming that the RNA editing pathway is activated in EC carcinogenesis. ROC analysis for ADAR1 and ADAR2 expression yielded an AUC of 0.79 and 0.90 respectively, emphasizing the differences observed between control and EC samples (Supplementary Figure 5). Moreover, ADAR expression increased progressively with the presence of poor prognostic factors, such as tumors presenting high grade or a NEEC histology ( Fig. 2C and E). These data suggested for the first time an activation of the RNA editing pathway in patients with EC, which positively correlates with aggressive disease, resulting in a dysregulation of the nucleosides/nucleotides balance in the tumor tissue.
Inhibition of ADAR2 reduces viability, increases apoptosis and reduces migration capabilities in human EC cell lines. Next, in order to determine the possible role of the RNA editing pathway in EC   The inhibition of ADAR2, but not ADAR1, resulted in a significant reduction of cell viability and proliferation compared to controls in the three cell lines used in this study (Fig. 3A). Furthermore, the ratio of apoptotic cells significantly increased in the three cell lines upon silencing of ADAR2. Similarly, proliferation assay performed using cells treated with siRNA-ADAR1, did not result in significant changes in apoptosis (Fig. 3B). Finally, the migration capabilities of the EC cell lines were interrogated after inhibiting ADAR1 and ADAR2 in a wound healing assay. Knockdown of ADAR2 expression resulted in significant reduction of the migration rate in HEC-1A and RL95-2 cell lines. Remarkably, similar changes were not observed when the same cell lines were treated with siRNA-ADAR1 (Fig. 3C).  In order to further confirm the functional role of ADAR2 in EC cell lines, we also conducted functional assays silencing ADAR2 expression with two new and different siRNAs (siRNA-ADAR2_B and siRNA-ADAR2_C), independent from the siRNA-ADAR2 used in Fig. 3. We corroborate a significant decrease of cell viability, a significant increase in apoptosis rate and a significant decrease of wound healing rate of the 3 EC cell lines induced by the silencing of ADAR2 expression (Supplementary Figure 8). These results clearly demonstrate that changes in the expression of ADAR2 leads to significant changes in the aggressive behavior of EC cell lines, specifically on cell viability and apoptosis, and cell migration capabilities of EC cell lines.

Discussion
Recently, few studies have reported alterations in the metabolomic or proteomic phenotype of EC in serum 25 , urine 26 or in other sample types 27,28 , underscoring the clinical translational relevance of these technologies in furthering the personalized medicine initiative. In this study, we used a global metabolomics profiling approach in order to understand the metabolic changes that take place in EC carcinogenesis and during tumor progression (Fig. 4).
Our study reveals an array of metabolites dysregulated in EC, some of which have been previously described in endometrial carcinogenesis. Glycerophospholipid class of metabolites was found to be upregulated in tumor tissues, including PCs, PEs, and PIs. Lipid biosynthesis and catabolism is known to be altered in several diseases, including cancer [28][29][30][31][32] . Consistent with our findings, Trousil et al. 28 also observed an increase (up to 70%) in PC levels in EC tissues. We also observed a downregulation of the acylamido analogs of endocannabinoids such as palmitamide, stearamide and oleamide in EC. This downregulation has been previously reported 33 . We also observed a decreased proportion of picolinic acid in tumor samples. Picolinic acid and quinolinic acid are the end products of the Kynurenine pathway and have been shown to have anti-tumoral and pro-tumoral activity respectively 34 . Moreover, the activity of one of the main enzymes of the pathway, indoleamine 2,3-dioxygenase (IDO) has also been studied in EC 35 . Decreased levels of Glu Phe Arg Trp and inosine as well as upregulation of 3-Deoxyvitamin D3 and UDP-N-acetyl-D-galactosamine were also found in our EC tumor sample set compared to control tissues. We also analyzed metabolomic changes that underscore tumor progression. Our data suggest significant changes in the lipidome including PC   UDP-N-acetyl-D-galactosamine and arachidonic acid at advanced stages of EC. Further studies would be needed to fully understand the scope and impact of these alterations in cancer progression. The dysregulation of inosine was further studied to dissect the functional implications of the RNA editing pathway in EC. The dysregulation of inosine is indicative of a possible imbalance in the I/A ratio, which was also reported by Trousil et al. 28 in studies with EC tissue compared to normal endometrium. The A-to-I conversion is the most common type of RNA editing found in mammals mediated by the ADAR enzymes. Although, to date, the RNA editing pathway and the expression and function of the ADAR gene family has not been interrogated in EC, it has been reported that the expression of ADAR enzymes is upregulated in many cancers 18,19 , including breast 17 and esophageal squamous cell carcinoma 36,37 . The ADAR family is comprised of three members: ADAR1 and ADAR2, that are present in most human tissues; and ADAR3, that is brain specific 23 . Changes in editing frequencies have been described in other diseases including prostate, lung, kidney and testis tumors while reduced RNA levels of ADAR1, ADAR2 and ADAR3 have been observed in brain tumors 38,39 . ADAR enzymes are also involved in physiological events such neuronal development, immune response, cell response to viruses and regulation of miRNA expression among others 40,41 .
Hence, we asked if the expression of ADAR enzymes had any correlation with EC initiation and progression. Our findings not only elucidate ADAR1 and ADAR2 enzymes to be significantly upregulated in EC tumor tissues compared to healthy endometrium, but also demonstrate a significant correlation between ADARs expression and the malignancy of the tumor. We found that the ADARs expression increased progressively with tumor grade. More importantly, our data showed a significant increase in the expression of ADAR1 and ADAR2 in the most aggressive subtype of EC, the NEEC, that have the worst predicted survival 3 .
The role of the RNA editing pathway in EC was further investigated by knocking-down the expression of ADAR1 and ADAR2 in HEC-1A, RL95-2 and Ishikawa EC lines. Our results demonstrate the impact of decreased ADAR2 expression on an array of cellular functions in EC cell lines including a significant decrease of cell proliferation and viability, increased apoptosis rate, and reduced migration capabilities in vitro. Similar to our observations, silencing of ADARs in breast cancer cell lines led to less cell proliferation and more apoptosis 17 and overexpression of editing enzymes accelerated growth rate and colony formation in esophageal squamous cell carcinoma in vitro 36 . Taken together, our results strongly suggest, for the first time, that the RNA editing gene family, specifically ADAR2, may play an important role promoting EC carcinogenesis.
In conclusion, a global molecular profiling approach using high resolution mass spectrometry has been useful, not only to describe changes in the metabolome and lipidome in human endometrial carcinogenesis and EC progression but also led to the discovery of an important alteration of the RNA editing pathway in EC. We further demonstrated the role of this pathway in proliferation and viability, apoptosis, and migration of EC cells, leading us to conclude that the activation of the RNA editing pathway is an oncogenic process in EC. This study opens several avenues for further investigations of ADAR2 as possible target for the development of therapeutic approaches for the treatment of EC patients.   > 50  20  -20  36  --78  29  -< 50  ----------Collection center   VHUH  20  -20  7  --78 29 - Validation phase. Patients split in 3 new cohorts were enrolled in VHUH or in University Hospital Arnau de Vilanova of Lleida following the approval of the CREC at each participating institution. Tissue samples were embedded in paraffin block for individual slides or tissue microarray (TMA) construction in VHUH. A description of the clinical and pathological characteristics of the tissues is detailed in Table 4.

Tissue metabolomics using Ultra Performance Liquid Chromatography coupled to Quadrupole Time-Of-Flight Mass Spectrometry (UPLC-QTOF-MS).
Reagents and chemicals: Solvents using chloroform, ACN, water and methanol were purchased from Fisher Optima grade, Fisher Scientific (New Jersey, USA). High purity formic acid (99%) was purchased from Thermo-Scientific (Rockford, IL, USA). Ammonium formate, debrisoquine and 4-nitrobenzoic acid (4-NBA) were purchased from Sigma-Aldrich (St. Louis, MO, USA). For metabolomics analysis, endometrial tissue samples were prepared following the procedure previously described. 42 Fresh frozen tissue sections were homogenized on ice using a buffer containing 50% methanol and internal standards (1 mg/ml debrisoquine in distilled water, 1 mg/ml of 4-nitrobenzoic acid in methanol, 10 µg/ ml of phosphatidic acid in 50% methanol-water, 0.1 µg/ml of lysophosphatidylcholine in 50% methanol-water). Protein precipitation was done by adding a 1:1 ratio of acetonitrile (ACN). Samples were centrifuged, the supernatant (Supernatant 1) was transferred to a fresh vial and dried under vacuum and the pellet was resuspended in prechilled dichloromethane:methanol (3:1). After sonication and centrifugation, the supernatant (Supernatant 2) was transferred to a fresh vial and dried under vacuum and the pellet was kept for protein estimation. Supernatant 1 and Supernatant 2 were finally resuspended in a buffer containing methanol:ACN:water in a ratio 50:25:25 for LC-MS analysis. The extraction procedure (from aqueous to semi-polar and lastly non-polar solvent) allows the isolation of a wide range of metabolites. In parallel, the pellet was resuspended with RIPA buffer and centrifuged in order to quantify the protein amount using Bradford method 43 . Resuspended pellets (5 µl) were injected onto Acquity UPLC CSH 1.7 µm, 2.1 × 100 mm column (Waters Corp.) or in a BEH C18 1.3 µm, 2.1 × 50 mm column. The mobile phase gradient consisted of ACN/water (60/40) containing 10 mM ammonium formate and 0.1% formic acid (Solvent A) and IPA/ACN (90/10) containing 10 mM ammonium formate and 0.1% formic acid (Solvent B). UPLC separation was performed at a flow rate of 0.4 ml/min for 20 min. Two different gradients were used in order to analyze the lipidome or the metabolome of the sample. MS data acquisition was performed using ESI-QTOF MS within the mass range of 50 to 1200 mass-to-charge ratio (m/z) in positive and negative electrospray ionization modes on a SYNAPT G2 Si (Waters Corporation, USA). The capillary voltage used was 3.2 kV and a sampling cone voltage of 30 V in negative mode and 20 V in positive mode. The desolvation gas flow was set to 750 l/h, and the temperature was set to 350 °C. The cone gas flow was 25 l/h, and the source temperature was 120 °C. Accurate mass was maintained by infusion of LockSpray interface with Leucine Enkaphalin (556.2771 [M + H] + and 554.2615 [M − H] − ). Data were acquired in TOF MS centroid mode and also in continuum mode for the mass range of 50 to 1200 mass-to-charge ratio (m/z) with MS scanning at a rate of 0.3 seconds. Total protein concentration for each sample was used to normalize any inconsistencies that would arise during tissue sampling. Subsequently, these data were normalized to internal standard to correct for analytical inconsistencies that could potentially occur during MS batch acquisition.
MS data were pre-processed using the XCMS software 44 . In order to determinate the identification of the metabolites based on the mass and charge, the following databases were used: Human Metabolome Database (www.hmdb.ca), Madison Metabolomics Consortium Database (mmcd.nmrfam.wisc.edu), LIPID MAPS (www. lipidmaps.org), KEGG (www.kegg.jp/kegg), and Metlin (metlin.scripps.edu). Multivariate data analysis was performed using Metaboanalyst 3.0 web tool 45,46 and a sub-set of metabolites were verified by tandem mass spectrometry (MS/MS) and using Mass Fragments software (Waters Corp.).  Wound healing assay. To evaluate cell migration capabilities, a total of 8 × 10 4 cells (HEC-1A); 5 × 10 4 cells (Ishikawa), and 2 × 10 5 cells (RL95-2) per well were seed in a p24 (3 replicates per condition). Wound was generated 72 h post-transfection. Pictures were taken every 8 h using an Olympus FSX100 microscope. Wound healing area was measured using Image J software at the different time points. Three experiments were carried out independently.
All experiments were performed in accordance with the relevant guidelines and regulations.
Statistical analysis. The SPSS statistical package version 23 for Windows ® and the GraphPad Prism version 6 were used to perform the statistical analyses and ROC analysis. Each value represents the mean of at least 3 replicates with the corresponding standard deviation. We analyzed the normality of each data set and we used, according to the sample distribution, parametric or no parametric tests. For the IHQ analysis, a Wilcoxon signed rank test was used to compare tumors from controls; and a Kruskal-Wallis and Mann-Whitney tests were applied when comparing protein expression according to grades and histological subtypes, respectively. For the functional analysis, means of the different groups were compared by Kruskal-Wallis followed by Dunn's multiple comparisons test (in case of significance). We considered significant p-values < 0.05. (*p-v < 0.05; **p-v < 0.01; ***p-v < 0.001; ****p-v: 0.0001).