Global miRNA/proteomic analyses identify miRNAs at 14q32 and 3p21, which contribute to features of chronic iron-exposed fallopian tube epithelial cells

Malignant transformation of fallopian tube secretory epithelial cells (FTSECs) is a key contributing event to the development of high-grade serous ovarian carcinoma (HGSOC). Our recent findings implicate oncogenic transformative events in chronic iron-exposed FTSECs, including increased expression of oncogenic mediators, increased telomerase transcripts, and increased growth/migratory potential. Herein, we extend these studies by implementing an integrated transcriptomic and mass spectrometry-based proteomics approach to identify global miRNA and protein alterations, for which we also investigate a subset of these targets to iron-induced functional alterations. Proteomic analysis identified > 4500 proteins, of which 243 targets were differentially expressed. Sixty-five differentially expressed miRNAs were identified, of which 35 were associated with the “top” proteomic molecules (> fourfold change) identified by Ingenuity Pathway Analysis. Twenty of these 35 miRNAs are at the 14q32 locus (encoding a cluster of 54 miRNAs) with potential to be regulated by DNA methylation and histone deacetylation. At 14q32, miR-432-5p and miR-127-3p were ~ 100-fold downregulated whereas miR-138-5p was 16-fold downregulated at 3p21 in chronic iron-exposed FTSECs. Combinatorial treatment with methyltransferase and deacetylation inhibitors reversed expression of these miRNAs, suggesting chronic iron exposure alters miRNA expression via epigenetic alterations. In addition, PAX8, an important target in HGSOC and a potential miRNA target (from IPA) was epigenetically deregulated in iron-exposed FTSECs. However, both PAX8 and ALDH1A2 (another IPA-predicted target) were experimentally identified to be independently regulated by these miRNAs although TERT RNA was partially regulated by miR-138-5p. Interestingly, overexpression of miR-432-5p diminished cell numbers induced by long-term iron exposure in FTSECs. Collectively, our global profiling approaches uncovered patterns of miRNA and proteomic alterations that may be regulated by genome-wide epigenetic alterations and contribute to functional alterations induced by chronic iron exposure in FTSECs. This study may provide a platform to identify future biomarkers for early ovarian cancer detection and new targets for therapy.

Iron is essential for maintenance of cellular homeostasis and organismal survival 1 . Iron participates in Fenton reactions, yielding reactive oxygen species (ROS) that are highly damaging to macromolecules including proteins, lipids, and nucleic acids 2 . Increased intracellular iron accumulation is a key feature of ferroptosis, a programmed cell death mechanism that is characterized by increased lipid peroxidation 3,4 . In contrast, deregulated expression of mediators involved in iron metabolism (i.e., Transferrin Receptor) leads to increased intracellular labile iron that promotes increased cellular proliferative capacity and cancer pathogenesis 5,6 . In our prior report, in response to supraphysiological levels of NTBI (non-transferrin bound iron, presented as ferric ammonium citrate (FAC)), we performed a proteomics screen utilizing reverse phase protein array (RPPA) to identify global proteomic alterations, which confirmed RAS-and MAPK-dependency in short-term iron-exposed ovarian cancer cell lines 7,8 . High-grade serous ovarian cancer is the deadliest gynecological malignancy in women for which the fallopian tube secretory epithelial cell (FTSEC) type is now considered an established precursor cell 9 . Our recent findings have identified that chronic iron exposure contributes to upregulated EVI1 (a transcriptional regulator, amplified at 3q26.2) and TERT expression, accompanied by increased cell numbers and migration in FTSECs 10 . However, to our knowledge, an assessment of global alterations induced by this mode of iron exposure in FTSECs has yet to be performed and is thus necessary to identify comprehensive cellular changes for which improved treatment regimens may be designed. Furthermore, since miRNAs elicit both oncogenic and tumor suppressive functions by altering the expression of multiple protein targets, a global miRNA screening approach in chronic iron exposed FTSECs is deemed to be an additional highly valuable experimental approach.
Herein, we utilized a multiomics approach using the GeneChip miRNA 4.1 microarray and mass spectrometry-based proteomics to identify deregulated miRNAs and protein targets, respectively, in chronic iron exposed and transformed FTSECs. We applied stringent statistical approaches to identify the most significantly deregulated hits and integrated both of these analyses using Ingenuity Pathway Analysis (IPA). A majority of the identified downregulated miRNAs are located at the 14q32 locus, a highly aberrant chromosomal region in multiple tumor types [11][12][13] . Since 14q32 miRNAs can be regulated by differentially methylated promoter regions (i.e., DLK1-DMR, IG-DMR and MEG3-DMR [14][15][16][17], we investigated whether FAC could epigenetically alter the methylation and acetylation status in genomic regions potentially involved in miRNA regulation. Inhibition of methyltransferases and histone deacetylases using 5-Azacytidine (AZA) and Suberoylanilide Hydroxamic Acid (SAHA), respectively, resulted in rescuing the expression of miR-432-5p, miR-127-3p, with minimal effects on miR-138-5p expression. IPA analyses identified notable proteomic targets of key miRNAs including ALDH1A2 (for miR-138-5p target) and PAX8 (for miR-432-5p, miR-127-3p, and miR-138-5p). Although these targets could not be validated experimentally, TERT RNA was identified to be partially regulated by miR-138-5p. From a functional perspective, overexpression of these miRNAs reversed cell survival induced by chronic iron exposure in FT194 cells.

Materials and methods
Experimental design and statistical rationale. Understanding the mechanism underlying transformation of FTSECs requires the use of an in vitro model system, such as immortalized FT194 cells, generated via SV40 LTAg and hTERT stable expression, characterized by p53 inactivation, as described previously 18 . We previously reported the generation of oncogenically transformed FT194 cells; briefly, these were produced by stable overexpression of c-Myc T58A , H-Ras V12A , and SV40 LTAg 10 . In addition, we described the generation of a transformed-like FTSEC cell line following chronic iron overload (with 250 nM Ferric Ammonium Citrate (FAC) for > 60 days in culture) which was characterized by alterations in oncogene expression (based on a focused approach) and survivability 10 . Briefly, the FTSECs were seeded at 500 cells/well in 6 well plates and subsequently treated with a range of FAC doses (0, 25 nM, 250 nM, 2.5 µM, 25 µM, or 250 µM), as previously described 10 . Cell growth was continually monitored and cultures propagated in FAC-containing media (with media replenishments every 4 days). Cells treated with 250 nM FAC elicited greater cell numbers (compared with Untreated cells or other iron doses) and was therefore selected for expansion and experimentation along with Untreated control cells which were maintained concurrently. Our prior attempts with mM FAC doses similar to those previously reported 19,20 was highly toxic and therefore not further pursued.
To further characterize global changes in an unbiased manner, we prepared experimental samples from Untreated (UNT), FAC-treated (FAC), control virus (CV), and oncogenic cocktail virus (OCV) infected FT194 cells; these were utilized for both noncoding RNA microarray and proteomic analyses. Specifically, the proteomic study utilized flash frozen cell pellets (UNT/FAC or CV/OCV) of ~ 500,000 cells per replicate, based on the protein extraction yield obtained by the S-trap sample processing approach previously reported by us 21,22 . For LFQ-based quantitation of protein expression, 5 replicates (from the same "batch" of UNT/FAC or CV/OCV cells) per group were utilized based on the expected quantitation precision of our approach obtained for cell lines 22 . For a statistical power of 90% with alpha = 0.05 and n = 5 replicates per group, an effect size of 2.348 would be needed based on using a two-tailed, unpaired t-test. The average coefficient of variation for each group was calculated for LFQ intensities obtained experimentally and then used to determine the fold change to achieve this effect size, which was then compared to the z-score cutoff used in this study (|z-score|> 1). A conventional FDR correction approach (e.g., Benjamini-Hochberg) was not employed given the tendency for decreased sensitivity; however, a combined filtering approach that considers variance and fold change was used (Welch's t-test p < 0.05 and |z-score|> 1), which has been shown to adequately control FDR while maintaining sensitivity 23 .
For the miRNA analyses, total RNA was isolated from 3 replicates (from the same "batch" of UNT/FAC or CV/ OCV cells), quantified by Nanodrop, and then analyzed using the GeneChip 4.1 Array (#902409, ThermoFisher Scientific, Waltham, MA, USA). The miRNA array contained 30,424 mature miRNAs, of which 2,578 were of human origin. To obtain a statistically relevant dataset of differentially expressed miRNA targets, an approach was applied of a > twofold change cutoff along with a FDR-adjusted p value < 0.05 top miRNA targets were selected to generate volcano plots and Venn diagrams.
Mass spectrometry-based proteomic analyses. Suspension trap (S-trap) sample processing of each experimental group (U, F, CV, OCV) was performed as previously described 22 using an approximate 500,000 cell count for each group (n = 5 per group). Tryptic peptide concentrations were normalized based on the original protein concentration measurements determined by a Pierce 600 nm protein assay (Thermo Fisher Scientific). LC-MS/MS analysis of the cell lysate digests was performed using a hybrid quadrupole-Orbitrap instrument (Q Exactive Plus, Thermo Fisher Scientific) coupled to an Ultimate 3000 UPLC system (Thermo Fisher Scientific). Digested samples were first concentrated on a 2 cm × 75 µm ID PepMap C18 trap column (Thermo Fisher Scientific) followed by separation on a 55 °C-heated, 75 cm × 75 µm ID C18 PepMap column (Thermo Fisher Scientific). A 120 min gradient from 2 to 28% B, where B was 0.1% formic acid in 80% acetonitrile:20% water was used to separate peptides, as described in our prior publication 21 . An additional ramp to 40% B over 15 min followed by a wash at 95% B was implemented. For mass spectrometric analysis, data-dependent acquisition (DDA) with a top-10 method was utilized. The full MS spectra were acquired in the m/z range of 375-1200 at 70,000 resolution followed by MS/MS scans at 17,500 resolution. AGC target counts of 1E6 and 5E4 with maximum IT values of 20 and 50 ms for MS1 and MS2 were utilized, respectively. A normalized collision energy of www.nature.com/scientificreports/ 28 and isolation window of 1.6 m/z was employed with charge state exclusion set for unassigned, 1, 6-8, and > 8. Dynamic exclusion was set for 20 s with isotope exclusion enabled and the peptide match setting to preferred. All details of the mass spectrometry data acquisition and LC parameters are embedded in the raw data files, which have been deposited to the ProteomeXchange Consortium via the PRIDE 28 partner repository with the dataset identifier PXD018416. MaxQuant (version 1.6.6.0) was used to search raw files against the Uniprot protein database for Homo sapiens (version UP000005640, 71607 entries). Search parameters included the variable modifications of N-terminal protein acetylation and methionine oxidation as well as the constant modification of cysteine by carbamidomethylation. An additional database of known contaminants provided with MaxQuant was utilized where the first search tolerance was set to 20 ppm followed by a main search tolerance of 4.5 ppm, as described in our earlier work 21,22 . Furthermore, a search strategy using reversed sequences in a decoy database was employed to achieve protein and peptide FDR values of less than 1% 21,22 . Label free quantification (LFQ)-based quantitation was enabled, with a minimum ratio count of 1, and the "match-between-runs" feature using default settings was employed to increase proteomic identification, as described in our earlier work 21,22 .
The resulting Protein-Groups text file generated by MaxQuant was edited by removing the reverse and contaminant sequences as well as proteins only identified by modification (similarly described in our earlier work) 21 . The file was then uploaded into Perseus (version 1.6.1.1) 21 twice for separate analysis of FAC-treated FT194 cells (F) relative to Untreated FT194 cells (U), and oncogenic cocktail virus infected FT194 cells (OCV) relative to control virus infected cells (CV). Each file was then analyzed whereby LFQ values were log 2 -transformed and proteins were removed that had missing values in more than just 2 out of the 5 replicates, similarly described in our earlier work 21 . The imputation function was utilized where missing values were replaced using width parameters of 0.3 for both and downshift parameters set to 1.8 and 1.75 for F vs. U and OCV vs. CV, respectively 21 . The average ratio of treatment over control was then calculated in Excel along with a Welch's t-test (p-value < 0.05) and z-score (z-value > 1), similarly described in our earlier work 21 . These filtered lists containing protein identification and average ratio of each comparison were then uploaded to Ingenuity Pathway Analysis (IPA) in order to determine upstream regulator overlap and activity, over-represented canonical pathways, as well as other biological and disease functions (p < 0.05, Fisher's exact test), similarly described in our earlier work 21 . Additionally, differentially expressed miRNAs (described below) were uploaded into IPA and paired against the proteins identified from proteomic analysis, which are known (experimentally determined) or predicted (moderate or high confidence) downstream targets of the miRNAs, through the miRNA Target Filter function. Paired miRNAprotein targets were filtered to include those in which the observed miRNA up-or down-regulation resulted in down-or up-regulation of the protein target, respectively. The corresponding network was reconstructed in IPA to demonstrate the potential regulatory role of each selected miRNA on the protein expression profile obtained.
MicroRNA microarray. Total RNA was isolated from FT194 cells maintained in 250 nM FAC for 104 days (p = 31) along with parental untreated FT194 cell line cultured simultaneously. In addition, total RNA was isolated from CV and OCV-infected FT194 cell lines (p = RV + 12). RNA isolation was performed using the RNeasy Kit (#74106, QIAGEN, Valencia, CA, USA) according to manufacturer's instructions. Total RNA was quantified by Nanodrop and then analyzed using the GeneChip 4.1 Array (#902409, ThermoFisher Scientific, Waltham, MA, USA). The miRNA array contained 30,424 mature miRNAs, of which 2,578 were of human origin. To obtain a statistically relevant dataset of differentially expressed miRNA targets, an approach was applied of a > twofold change cutoff along with a non-adjusted p-value < 0.05; top miRNA targets were selected to generate volcano plots and Venn diagrams. miRNA array and proteomics data were combined in Ingenuity Pathway Analysis (IPA) to associate miRNAs with top proteomic hits. Thirty-five miRNAs had 28 protein targets identified in the proteomic screen in FAC-treated (compared to Untreated) FT194 cell samples whereas 45 miRNAs had 74 protein targets identified in the proteomic screen in transformed OCV (compared to control virus transfected), both with a cutoff of > fourfold change to focus on the "top hits" of biological relevance. The experimental strategy is depicted in Fig. 1.
Protein isolation, SDS-PAGE, and western blotting. Cell lysates were collected for separation on appropriate percentage SDS-PAGE gels and protein expression analyzed via western blotting using previously The western blots were developed at multiple exposures onto film, which were all scanned using a Flatbed Scanner (HP Scanjet 5590) and inserted as images into Powerpoint, without any manipulation (no contrast alterations) apart from cropping to within 6 band widths above and below the band of interest. Each developed blot (representing a specific antibody application) are presented in one powerpoint slide with space in-between, delineating different antibody applications to the same blot. The EVI1 and Pan-Actin antibodies were previously optimized in our laboratory 31 . The use of DNMT1 32 , Acetyl Histone H3 33 , and PAX8 34 were utilized based on prior publications and ALDH1A2 was utilized based on data available by Cell Signaling Technology. EVI1 siRNA in FT194 cells. siRNA transfection in FT194 cells was performed as previously reported 10 .
Briefly, cells were seeded in six-well plates at a density of 500,000 cells/well or at 1,000,000 cells/dish in 60 mm dishes followed by overnight adherence. ON-Target Plus non-targeting control siRNA (#D-001810-10-20, Dharmacon (Lafayette, CO, USA)) or EVI1-targeting siRNA (siB, custom designed as described previously) 10,31,35 were transfected into cells using RNAiMax (#13778-075, Invitrogen, Carlsbad, CA, USA). Cells were recovered 24 h post transfection; cell lysates and miRNAs were collected at 48 h post-transfection for western blotting and real-time PCR, respectively.

Bioinformatics of EVI1 binding site in miR-138-5p-1 promoter region. The UCSC Genome
Browser (www. genome. ucsc. edu, Human Dec. 2013 (GRCh38/hg38 Assembly) was utilized to obtain the genomic sequence (5000 bp upstream) of the promoter region for miR-138-5p-1 (located at 3p21.32) and for miR-138-5p-2 (located at 16q13); miRBase was utilized to obtain the pre-miRNA sequence for these miRNAs. Previous research identified the DNA sequences that EVI1 binds to; EVI1 N-terminus binds to the sequence GAC AAG ATA 36,37 while the C-terminus binds to the sequence GAA GAT GAG 38,39 . Overall, the consensus EVI1 binding sequence is TGA CAA GATAA 36,39 . Thus, these reported EVI1 binding sequences were aligned with the promoter regions for both miR-138-5p-1 and miR-138-5p-2 using the Genomatix software suite (version 3.11, http:// www. genom atix. de/ cgi-bin/ diali gn/ diali gn. pl). www.nature.com/scientificreports/ Statistical analyses. Data from real-time PCR, densitometry, and cell counting studies were analyzed using the Graphpad Prism software, version 6.04 (La Jolla, CA, USA). Error bars represent the mean ± SD and p-values were determined through the non-parametric Student's t-test for which "ns" represents non-significant values, * indicates p ≤ 0.05, ** indicates p ≤ 0.01, *** indicates p ≤ 0.001, and **** indicates p ≤ 0.0001. Fold changes and percent reductions were calculated from the average of at least three independent experiments.

Results miRNA and proteomic profiling of chronic FAC-treated and oncogenically transformed
FTSECs. Elucidating molecular mechanisms involved in initiation and progression of ovarian cancer is essential towards developing novel therapeutic strategies. We have previously identified a subset of genes (including EVI1, located at 3q26.2 in HGSOC) that were altered following long-term iron exposure in FTSECs 10 . However, to acquire a comprehensive understanding of the iron-induced alterations, a global mass spectrometry-based proteomics analysis was performed in FAC-exposed and Untreated FT194 cells. We then compared the resultant proteomic alterations to OCV-and CV-infected FT194 cells, as generated previously 10 . For FAC-exposed FT194 cells compared to Untreated cells, 4402 total proteins were identified with 3968 quantifiable proteins after filtering. In OCV-infected compared to CV-infected FT194 cells, 4,691 total proteins were identified with 4148 quantifiable proteins after filtering. The average coefficient of variation for the LFQ values of CV-infected, OCV-infected, FAC-treated, and Untreated FT194 cells was 25.4, 21.4, 21.9, and 19.6%, respectively. The median coefficient of variation for the LFQ values of CV-infected, OCV-infected, FAC-treated, and Untreated FT194 cells was 16.8, 13.7, 15.0, and 11.7%, respectively. Statistically significant "top hits" were obtained using Welch's t-test (p < 0.05) and z-score (z-score > 1) which identified 700 protein targets in the OCV (relative to CV) cells and 459 protein targets in iron-exposed FTSECs (relative to Untreated). To achieve the effect size corresponding to a statistical power of 0.9 based on our experimental conditions, a fold-change of ~ 1.7-1.8 would be needed (assuming the global average of the coefficient of variation determined for each group), which is consistent with the implemented z-score cutoff. Additional restrictions were applied to this data set to identify the protein targets that had LFQ intensity ratio of ≥ 2 or ≤ 0.5. Using this strategy, we thus identified 622 targets for OCV (relative to CV) and 243 targets for FAC-exposed FTSECs (relative to Untreated) (Fig. 1). The list of total quantifiable as well as differentially expressed proteins are provided as supplemental Tables 1 and 2.
Since alterations in miRNAs can contribute to cancer pathogenesis by targeting multiple protein targets 40,41 , we performed microRNA array profiling using the GeneChip miRNA 4.1 Array. We compared chronic FAC-exposed FTSECs relative to Untreated cells as well as OCV-relative to CV-infected FTSECs, which identified a total of 65 and 78 unique non-coding RNAs, respectively, including mature miRNAs and snoRNAs (7 snoRNAs were repeated as duplicates in the dataset resulting in a total of 71 and 79 ncRNAs identified in FAC/UNT and OCV/ CV analyses, respectively), with at least a twofold change (p value > 0.05, displayed as a heat map, volcano plot, and Venn diagram (Fig. 2a-e). There were 42 upregulated, 29 downregulated miRNAs in FAC-treated (relative to Untreated) and 45 upregulated, 34 downregulated miRNAs in OCV (relative to CV). miRNAs downregulated in FAC-treated (relative to Untreated) cells showed higher fold change (X-axis) and statistical significance (p-value at Y-axis) compared to OCV cells (relative to CV) (Fig. 2c, d). Since the quantity of changes are more numerous in the OCV versus CV FTSEC comparison (relative to FAC versus Untreated FTSEC), this suggests that the combination of p53 inactivation, c-Myc T58A expression, and H-Ras V12A expression (within the OCV cocktail) may mediate increased neoplastic cellular alterations.
Since alterations in protein expression can be regulated post-transcriptionally by miRNAs, we integrated the top miRNA hits with the protein targets using IPA to identify associations between the two analyses. Integrated proteomics and microarray analyses identified 35 miRNAs in FAC-treated FT194 cells (relative to Untreated) and 45 miRNAs in OCV cells (relative to CV cells) with the following characteristics: (1) > twofold change and (2) with direct protein targets that were altered with respect to miRNA levels ( Fig. 3a, b). Twenty of the 35 miRNAs (57.14%) altered in FAC-treated cells (relative to Untreated) were located at chromosome 14q32 (Fig. 3a), while only 9 out of the 45 miRNAs (20%) altered in OCV cells (relative to CV) were located at this region (Fig. 3b). This region, amongst others, also harbors common fragile sites (Tables 1 and 2), possibly rendering the chromosomal loci susceptible to replication stress, which is known to impact genomic stability in many cancers 42 . Interestingly, the 14q32 locus contains a cluster of 54 miRNAs, one of the largest miRNA clusters in the human genome 43 . Many of these miRNAs appear to be downregulated in multiple cancer types associated with tumor suppressive properties 11,27,[44][45][46][47][48] and oncogenic properties 49 . From the miRNA profiling, we identified that two miRNAs from this cluster, miR-432-5p and miR-127-3p, were 97.9-and 111.7-fold downregulated, respectively, in FAC-treated cells in contrast to only 11.7-fold downregulated in OCV cells. These results were validated by real-time PCR in FAC-treated FT194 cells (Fig. 3c, d), which showed a 99.9% reduction for miR-432-5p and miR-127-3p (p value < 0.0001) in FAC-treated relative to Untreated FT194 cells. Additionally, among other highly dysregulated miRNAs from other genomic regions such as 3p21.31, we identified that miRNA-138-5p (located at 3p21) was 16.3-fold downregulated with chronic FAC-exposure in FT194 cells (Fig. 3a) and validated via qPCR to be 90.1% reduced (p value < 0.0001, Fig. 3e). miR-138-5p also appears to be commonly downregulated in multiple cancers [50][51][52][53] .
We identified 28 protein targets in FAC-treated relative to Untreated cells and 74 protein targets in OCV relative to CV cells out of the total statistically significant "top hits" (p < 0.05 and z-value > 1), which were associated with characteristics of gynecological cancers (Fig. 4a, b). In Fig. 4c, we validated two of these "top hits", namely CRYAB and ITGA2 via western blotting, which have roles as molecular chaperones 54 and in adhesion to the extracellular matrix 55 , respectively.
The number of identified proteins and miRNAs from the OCV FT194 cells (relative to CV) were higher as compared to FAC-exposed cells (relative to Untreated), which suggests that the extent of neoplastic alterations www.nature.com/scientificreports/ induced by the oncogenic cocktail (hTERT, LTAg, c-Myc T58A , and H-Ras V12A ) is more extensive than that induced by iron exposure (see Experimental Procedures). These "hits" were not validated due to the expectation that oncogenic transformation of FTSECs would lead to a comparatively larger array of alterations recapitulating more closely tumorigenic profiles of ovarian tumors relative to chronic iron exposed FTSECs, which is the main focus of this study. www.nature.com/scientificreports/ As the protein targets from the FAC versus Untreated analyses were found to be involved in multiple signaling and molecular pathways ( Fig. 4d and Table 3), we further investigated the mechanism of FAC-induced dysregulation in FT194 cells, specifically focusing on analyzing the mechanism by which FAC alters miR-138-5p, miR-432-5p, and miR-127-3p levels to potentially contribute towards increased tumorigenesis.

FAC-induced epigenetic regulation of PAX8.
Integrated microarray and proteomics analyses were used to identify the most common protein targets of miR-432-5p, miR-127-3p, and miR-138-5p in FAC-treated cells relative to Untreated (Fig. 5a-c and supplementary Fig. S1) as well as in OCV (relative to CV) FT194 cells (supplementary Fig. S2-S4). Via IPA, the analyses identified PAX8 (Paired Box 8) as a potential common target of miR-432-5p, miR-127-3p, and miR-138-5p involved in ovarian cancer pathophysiology (Fig. 5d). Global proteomic analyses demonstrated that PAX8 was increased 1.7-fold in FAC-exposed FT194 cells, which was validated by western blotting and densitometric analyses (Fig. 6a, left and right panels). Since iron-overload conditions are associated with epigenetic changes in various human tissues 56,57 and epigenetic modification of PAX8 is reported in ovarian cancer 58,59 , we next investigated whether FAC-induced increase of PAX8 could be altered with AZA and/or SAHA treatment. As shown in Fig. 6a, we observed a 95.6%-fold reduction in PAX8 protein following combinatorial treatment of SAHA and AZA. These results suggest that the increased PAX8 protein is epigenetically regulated as a result of chronic iron treatment in FT194 cells. www.nature.com/scientificreports/ FAC-induced epigenetic regulation of miRNAs at 14q32 and 3p21. miRNAs are also regulated by epigenetic mechanisms [60][61][62] and miRNAs at chromosome 14q32 appear to be transcribed as a polycistronic miRNA cluster under control of epigenetic mechanisms 63,64 . In specific cancers, there is evidence to support hypermethylation at the 14q32 differentially methylated CpG regions (DMRs) which leads to cancer development 27,65 ; furthermore, promoter hypoacetylation, which can recruit HDACs through methyl CpG binding proteins can also regulate gene expression at this locus 66 . In view of these regulatory mechanisms and the  identification of twenty 14q32 miRNAs that were altered in FAC-exposed FT194 cells, we hypothesized that inhibition of DNA methylation with AZA and/or inhibition of histone deacetylases with SAHA may alter the expression of miR-432-5p and miR-127-3p (refer to Fig. 2 for "top hits" identified in the screening approach). Since miR-138-5p is associated with epigenetic changes 67,68 , we therefore investigated whether AZA and SAHA www.nature.com/scientificreports/ treatments could alter miR-138-5p levels. Inhibition of DNA methyltransferase with AZA was validated by western blotting for DNMT1, which showed reduced protein levels (Fig. 6a). In addition, HDAC inhibition was validated by demonstrating increased acetyl histone H3 levels via western blotting (Fig. 6a). Although SAHA treatment was more potent alone compared to AZA alone, the combination of AZA with SAHA, resulted in a further fold-increase of 164.8, 48.8 and 2.3 in miR-432-5p, miR-138-5p, and miR-127-3p levels, respectively, as measurement via real-time PCR (Fig. 6b-d). These results can be explained by the ability of the HDAC inhibitor, SAHA, to inhibit not only HDAC but also DNMT1 protein expression via inhibition of MAPK and thus the DNA methylation status 69 . HDAC inhibitors can also target DNMTs for degradation via the ubiquitin-proteasome-pathway by an Hsp90 chaperone mediated mechanism 70 . On the other hand, AZA specifically inhibits only DNMTs 71 .

FAC-induced miR-138-5p downregulation is mediated independently of EVI1.
We have previously reported that telomerase reverse transcriptase (TERT) can be transcriptionally regulated by EVI1 (genomically amplified at chromosome 3q26) in FAC exposed FT194 cells 10 . Prior work has shown that miR-138 levels are inversely correlated with TERT as a result of direct binding of miR-138 to the 3′-UTR of TERT [72][73][74] . Therefore, we hypothesized that the EVI1 may upregulate TERT transcripts in a miR-138-dependent manner. Mature miR-138 originates from two primary transcripts: pri-miR-138-1 (encoded on chromosome 3p21, Genecard) and pri-miR-138-2 (encoded on chromosome 16q13, Genecard). Indeed, we identified predicted EVI1 binding sites ~ 2500 bp upstream of miR-138-5p-1 ( Fig. 7a and supplementary Fig. S5); specifically, the C-terminal binding site, GAA GAT GAG, was 100% aligned in this region, while the N-terminal and consensus sequences were imperfectly aligned (5 out of 9 nucleotides and 5 out of 11 nucleotides, respectively). There was also a potential binding site ~ 1000 bp upstream of miR-138-5p-2, but the binding sites were < 100% aligned, of which 8 out of 9 nucleotides aligned for the N-terminal binding site, 5 out of 9 nucleotides aligned for the C-terminal binding site, and 8 out of 11 nucleotides aligned for the consensus sequence (results not shown). To determine whether EVI1 is potentially involved in the direct regulation of miR-138-5p-1 expression in chronic iron-exposed FT194 cells, we reduced EVI1 levels using an siRNA transfection (siB) approach and validated the efficiency of knockdown via western blotting (Fig. 7b). We then determined the levels of miR-138-5p via real-time PCR analysis. Although we confirmed that the levels of miR-138-5p were reduced following chronic FAC exposure (p < 0.0001), as noted previously (see Fig. 6d), its levels were not significantly altered upon siB (EVI1 knockdown) treatment in FAC-exposed FT194 cells relative to FAC-exposed parental cells (p = 0.8509) (Fig. 7c). These results indicate that miR-138-5p expression is regulated in an EVI1-independent manner and that other factors likely contribute to the downregulation of miR-138-5p under these conditions.

miR-138-5p partially regulates transcript expression of the stem cell marker TERT but not ALDH1A2 protein. Cancer stem cells (CSCs or tumor initiating cells) represent a subpopulation of cells
that are responsible for tumor initiation and progression 75,76 . Dysregulated iron homeostasis in CSCs may also aggravate cancer phenotypes 77,78 . CSCs are notably characterized by the expression of stem cell markers including elevated activity by aldehyde dehydrogenase (ALDH, represented by multiple isoforms) 79 , a superfamily of metabolic markers which serves as a potential poor prognostic factor of cancer 80 . One family member, ALDH1A2, is aberrantly expressed in acute lymphoblastic leukemia cells 81 and ovarian cancer 82,83 . Interestingly, the proteomics analyses identified a 3.2-fold increase in ALDH1A2 levels upon FAC exposure in FTSEC cells and IPA analyses suggests that ALDH1A2 is a potential direct target of miR-138-5p (Fig. 5c). Indeed, prior published work supports miR-138-5p involvement in ALDH1A2 regulation 84,85 . To investigate whether ALDH1A2 is a target of miR-138-5p in human FTSECs, we transfected FAC-exposed FT194 cells with miR-138-5p mimic. Although overexpression of miR-138-5p was confirmed by real-time PCR (Fig. 8a) and western blotting demonstrated that ALDH1A2 protein was increased in iron-exposed FTSECs, its levels were not altered by miR-138-5p overexpression (Fig. 8b). These results suggest that miR-138-5p is not a regulator of ALDH1A2 expression in FTSECs. Another stem cell marker TERT 86,87 has also been reported to be directly regulated by miR-138-5p 74,88 .
Since chronic iron exposure can induce TERT mRNA levels 10 and TERT sequence alignment with miR-138-5p mature sequence has shown direct binding (Fig. 8d) 72 , we analyzed whether TERT increase is mediated via miR-138-5p. Indeed, we observed that miR-138-5p mimic transfection partially rescued the TERT mRNA levels by 25% (Fig. 8e), suggesting that FAC-induced increase in TERT transcript levels could be partially regulated via miR-138-5p.
miR-432-5p, miR-127-3p, and miR-138-5p overexpression does not alter PAX8 protein but reduces cell numbers in FAC-treated FT194 cells. PAX8 (a member of paired box PAX gene family) is a positive marker of FTSECs 89 and is elevated in a variety of tumors [90][91][92][93][94][95] . Our prior findings demonstrated that PAX8 was upregulated following chronic iron exposure 10 ; we now validate this observation and further show that it is epigenetically regulated, also supported by prior literature 59 . Furthermore, PAX8 appears to be a common target of miR-432-5p, miR-127-3p, and miR-138-5p via the IPA analyses; target analyses using bioinformatic programs (Targetscan, miRDB, miRNA.org, miRpath v.3 and miRmap) identified PAX8 to be a target of the aforementioned miRNAs in a subset of these databases (data not shown). However, western blot analyses did not identify any marked changes in PAX8 protein expression in FTSECs upon overexpression of these miRNAs (Fig. 8b, c). Thus, it remains to be experimentally determined whether these miRNAs interact directly with PAX8 to regulate its expression.

Discussion
Iron is an important dietary component that is critical in maintenance of various cellular functions; however, it elicits activity as a mutagenic factor through its participation in Fenton reactions whereby it is involved in generating ROS that may promote DNA damage such as oxidation of DNA bases 99 . In this manner, iron may contribute to the pathophysiology of cancer. Deregulated iron levels and expression of key mediators of iron metabolism are established features of multiple tumors 100 including an established tumor "addiction" to iron 101 . Indeed, multiple studies have linked the exposure to supraphysiological levels of iron with an increased incidence of cancer including renal tumor formation in a rat model exposed to ferric nitrilotriacetate 102 , iron overload patients (e.g., hemochromatosis) with an increased risk of developing liver tumors 102,103 . On the other hand, dietary iron (in excess) can enhance tumor formation in mice harboring abnormalities in Adenomatous polyposis coli (APC) whereas iron chelation could hinder tumor development 104 . Collectively, these reports implicate the potential tumor promoting activities of iron in various experimental systems. Although a recent report identified that chronic iron exposure in human pancreatic ductal epithelial cell line supported epithelial-mesenchymal transition (EMT) and tumorigenesis through a p53-dependent mechanism 105 , the role of chronic iron overload in ovarian cancer initiation by mediating transformation of fallopian tube secretory epithelial precursor cells (FTSECs) remains unclear 106 . Iron sources in the pelvic cavity has been suggested to originate from ovulation, retrograde menstrual reflux, and the rupture of follicles [107][108][109] ; furthermore, a link between hemochromatosis and ovarian cancer has been reported 110 . However, there is currently limiting data regarding the contribution of iron to high grade serous ovarian tumor initiation. Recently, we reported that long-term FAC exposure (at 250 nM) to FTSECs leads to cellular changes that are reminiscent of those identified in HGSOC including alterations in EVI1, β-catenin, and c-Myc protein expression, together with functional  www.nature.com/scientificreports/ changes including increased cell numbers and migratory potential 10 . The work reported herein extends these findings to uncover miRNA and protein level changes under these chronic iron exposure conditions in FTSECs. Further, it is well recognized that aberrant expression of miRNAs is a characteristic of ovarian tumors with potential to serve as biomarkers and/or aid as potential diagnostic tools 41,111,112 ; however, the contribution of iron in altering their expression has not yet been reported in precursors to ovarian tumors. Thus, to comprehensively assess the iron-induced molecular changes in FTSECs, we performed an integrated miRNA and protein analysis approach in chronic iron-exposed and transformed FTSECs. To our knowledge, this is the first study to utilize a multi-omics approach to assess miRNA and protein level changes in FAC-exposed and transformed FTSECs. Herein, we have identified a subset of dysregulated miRNAs along with their corresponding protein targets via our integrated experimental approach following chronic iron exposure in FTSECs. Furthermore, we identified that several miRNAs were epigenetically dysregulated which may therefore be potentially associated with Analyses of oncogenetically transformed FTSECs (which harbors p53 inactivation, c-Myc T58A mutant, and H-Ras V12A mutant) revealed a higher number of altered miRNAs and protein targets relative to long-term iron exposed cells (see Fig. 2c, d). This suggests that these OCV cells are likely to be more extensively transformed than FAC-treated cells. This is not surprising since iron may only be one of many contributing factors mediating tumorigenesis. Indeed, profound alterations of the cellular regulatory networks may indicate the concept of "multiple events" needed to promote propagation of the tumor 113 . Our work presented herein encourages future investigations, including 3-D organoid culture and in vivo mouse xenografts to further explore the contribution of iron dysregulation to tumor initiation and/or the metastatic process. Two-dimensional in vitro culture models lack the complexity of the tumor microenvironment and thus limit physiological relevance 114 . Although murine models have been developed to investigate HGSOC pathophysiology 115,116 , these in vivo model systems pose challenges in terms of studying iron derived from the reproductive organs due to a lack of menstruation in mice. Therefore, alternative strategies such as intra-bursal or intraperitoneal delivery using iron dextran may be needed. Three-dimensional organoid cultures may also provide an alternative approach to investigate the response of FTSECs to exogenous iron 117 . However, our initial attempts to maintain our oncogenically transformed and chronic iron exposed FTSECs in spheroid culture failed to produce spheroids (data not shown). Thus, implementation of alternative 3-D methods including hanging drop method in future studies could be implemented 118 .
Interestingly, there exists literature supporting the association between iron and miRNAs. In one aspect, miRNAs can behave as iron sensors to suppress expression of proteins associated with iron regulation 119 . On the other hand, the role of iron in deregulating miRNAs is comparatively understudied. Although cytosolic iron can regulate miRNA biogenesis by altering miRNA precursor processing 120 , the role of iron-induced miRNA regulation in cancer requires further investigation. Herein, we have identified global dysregulation of miRNAs following chronic iron exposure. There is evidence implicating iron in the regulation of epigenetic control. Specifically, iron deprivation alters DNA methylation and histone deacetylation in preadipocytes 121 whereas brain iron overload reduces DNA methylation 57 . However, the specific underlying mechanism through which iron modulates these events remains to be explored. In the microarray analysis presented herein, we identified 20 out of 35 miRNAs (including miR-432-5p and miR-127-3p) that were down-regulated in FAC-treated FT194 cells and also part of a large imprinted miRNA cluster at 14q32 locus. Interestingly, there is evidence implicating epigenetic modification in the regulation of the miRNAs located within this cluster 63,64 . miR-138-5p, although located at a different chromosomal region at chromosome 3p21, was also identified to be markedly down-regulated following iron exposure. Current evidence supports miR-138-5p as a tumor suppressor in various cancer types 88,[122][123][124] . This locus harbors a tumor suppressor gene cluster comprised of 8 genes within a ~ 120 Kb spanning region 125 whose expression can be regulated epigenetically 126 . Besides miR-138-5p, 4 miRNAs (miR-135a-1, miR-let-7 g, miR-1226, miR-564) out of total 16 miRNAs at this locus have been reported to perform tumor suppressive functions in various cancers [127][128][129][130][131] . Furthermore, upregulation of six miRNAs has been identified in different cancer types, including miR-4271 132 Figure 10. Proposed model for 14q32 miRNA regulation with combination of DNMT inhibition and HDAC inhibition. We previously reported that chronic iron overload contributes towards oncogenic transformative events, which potentially recapitulate early events in HGSOC transformation. Microarray analyses identified markedly reduced levels of miR-432 and miR-127 (located at 14q32) in FAC-treated FT194 cells. We propose that the FAC-induced reduction in the miRNA levels was a result of epigenetic alterations, specifically in their methylation and acetylation status. Inhibition of DNA methyltransferases (using AZA) and HDACs (using SAHA) led to a reversal of the levels of these specific miRNAs in these chronic iron exposed FTSECs. We previously reported that chronic FAC treatment of FT194 cells notably increased TERT mRNA levels 10 as well as ALDH1A2 protein (from our proteomic analyses, validated via western analysis). We now identified that overexpression of miR-138-5p (located at 3p21) could antagonize the FAC-induced TERT transcript levels; however, miR-138-5p expression did not modulate ALDH1A2 protein. The above described three miRNAs also commonly alter Pax8 protein expression, which may contribute to FAC-induced changes in FT194 cells. Green arrows denote partial regulation. www.nature.com/scientificreports/ The miR-138-5p down-regulation was partially reversed via inhibition of methylation and histone deacetylases, suggesting that iron can regulate miRNA expression by mediating epigenetic changes in FTSECs. Out of the 15 miRNAs (from a total of 35 miRNAs) identified from the integrated microarray analysis (see Table 1) that are not located at 14q32, 9 are reported to be epigenetically regulated (miR-34a 25,142 , miR-145 143 , miR-182 144 , miR-31 145 , miR-708 146 , miR-34c 147 , miR-125b 148 , miR-615 149 , and miR-17 150 ). Similarly, miR-17, which is a part of the 17-92 miRNA cluster located at chromosome 13q31, can also be regulated by epigenetic mechanisms 150 . We also identified 5 miRNAs at chromosome 19q13 to be altered in OCV transformed FT194 cells (relative to CV) via proteomics analysis. Notably, chromosome 19 contains a large imprinted miRNA cluster comprising 46 miRNAs, which can also be regulated via epigenetic mechanisms 151,152 . Collectively, our miRNA analyses have verified epigenetic regulation as a common mechanism underlying miRNA expression, which is also supported by literature for miRNA cluster regulation at different loci in the human genome.
As shown in Fig. 6b, inhibition of methylation and histone deacetylation rescued the levels of miR-432-5p and miR-127-3p in chronic iron exposed FTSECs. This suggests that iron may regulate the activities of epigenetic regulators. There is also evidence that differential miRNA regulation may occur within this locus as independent miRNA regulation of 14q32 at unique miRNA promoters by nuclear receptors may be responsible 153 , although there is also evidence supporting transcription of this miRNA cluster as a polycistronic transcript 43 . Since we observed differential expression of 14q32 miRNAs via our microarray analysis, this may implicate independent regulatory mechanisms for multiple miRNAs within this region.
PAX8 (a member of paired box PAX gene family) is a positive marker of FTSECs 89 and is elevated in a variety of tumors [90][91][92][93][94][95] . Our prior findings demonstrated that PAX8 was upregulated following chronic iron exposure 10 ; we now validate this observation and further show that it is epigenetically regulated, also supported by prior literature 59 . Furthermore, PAX8 appears to be a common target of miR-432-5p, miR-127-3p, and miR-138-5p via the IPA analyses; target analyses using bioinformatic programs (Targetscan, miRDB, miRNA.org, miRpath v.3 and miRmap) identified PAX8 to be a target of the aforementioned miRNAs in a subset of these databases (data not shown). However, western blot analyses did not identify any marked changes in PAX8 protein expression in FTSECs upon overexpression of these miRNAs. Future studies are needed to experimentally validate whether PAX8 is a target of these miRNAs.
Our earlier findings demonstrated that long-term iron exposure in FTSECs leads to altered expression of EVI1 variants 10 . Although we identified an EVI1 binding site within the promoter region of miR-138-5p through our bioinformatics analysis (see supplementary Fig. S5), EVI1 knockdown did not alter miR-138-5p levels implicating the involvement of other regulatory mechanisms.
The stem cell marker, ALDH1A2 80,154,155 is reported to be a direct target of miR-138-5p, as identified via a proteomics screen in zebrafish embryos 84 but as of yet has not been confirmed in humans. As shown in Fig. 8b, overexpression of miR-138-5p in iron exposed FTSECs did not result in an alteration in ALDH1A2 expression; thus, this suggests that ALDH1A2 is not regulated by miR-138-5p in FTSECs. However, another stem cell marker TERT 86,87 , which has already been demonstrated to directly regulate miR-138-5p 74,88 , appears to be at least partially regulated by miR-138-5p (see Fig. 8d).
Although the quantitation accuracy of both the miRNA transcriptomic and proteomic datasets was validated through orthogonal methods such as qPCR and western blot analysis of selected targets, the miRNA target filtering through IPA is based on previously experimentally determined interactions or computational prediction (e.g., TargetScan). Potential regulatory miRNAs will need to be validated in future studies based on the expression pairing of miRNAs and proteins relevant to specific cancer-related pathways identified in our highconfidence transcriptomic and proteomic datasets. Further improvement can also be made in the proteomicsbased methodology, which includes fractionation of the proteome and application of DIA approaches to enhance proteome coverage. In addition to deeper proteome coverage to reveal further differentially expressed proteins, global-scale phosphoproteomic analysis can be employed to identify altered signaling pathways, collectively providing detailed global-scale insight into changes of the molecular landscape associated with iron exposure and oncogenic mechanisms in FTSECs.

Conclusion
Overall, the methodological approach utilized herein served to identify miRNAs and protein targets associated with long-term FAC treatment in FT194 cells. To our knowledge, this is the first study elucidating a comprehensive set of alterations at the miRNA and protein levels in chronic iron exposed FTSECs, which may identify pathways that may contribute to increased tumorigenic potential. In future studies, epigenetic mapping of chronic iron-exposed cells could be performed to further our understanding of the changes induced by long term iron treatment in fallopian tube precursors.

Data availability
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE 28 partner repository with the dataset identifier PXD018416. The RNA microarray data have been deposited to the Gene Expression Omnibus (GEO) 156 with the dataset identifier GSE150622.