Chronic Inhibition of STAT3/STAT5 in Treatment-Resistant Human Breast Cancer Cell Subtypes: Convergence on the ROS/SUMO Pathway and Its Effects on xCT Expression and System xc- Activity

Pharmacologically targeting activated STAT3 and/or STAT5 has been an active area of cancer research. The cystine/glutamate antiporter, system xc-, contributes to redox balance and export of intracellularly produced glutamate in response to up-regulated glutaminolysis in cancer cells. We have previously shown that blocking STAT3/5 using the small molecule inhibitor, SH-4-54, which targets the SH2 domains of both proteins, increases xCT expression, thereby increasing system xc- activity in human breast cancer cells. The current investigation demonstrates that chronic SH-4-54 administration, followed by clonal selection of treatment-resistant MDA-MB-231 and T47D breast cancer cells, elicits distinct subtype-dependent effects. xCT mRNA and protein levels, glutamate release, and cystine uptake are decreased relative to untreated passage-matched controls in triple-negative MDA-MB-231 cells, with the inverse occurring in estrogen-responsive T47D cells. This “ying-yang” effect is linked with a shifted balance between the phosphorylation status of STAT3 and STAT5, intracellular ROS levels, and STAT5 SUMOylation/de-SUMOylation. STAT5 emerged as a definitive negative regulator of xCT at the transcriptional level, while STAT3 activation is coupled with increased system xc- activity. We propose that careful classification of a patient’s breast cancer subtype is central to effectively targeting STAT3/5 as a therapeutic means of treating breast cancer, particularly given that xCT is emerging as an important biomarker of aggressive cancers.

tigation demonstrates that chronic SH-4-54 administration, followed by clonal selection of treatment-resistant MDA-MB-231 and T47D breast cancer cells, elicits distinct subtypedependent effects. xCT mRNA and protein levels, glutamate release, and cystine uptake are decreased relative to untreated passage-matched controls in triple-negative MDA-MB-231 cells, with the inverse occurring in estrogen-responsive T47D cells. This "ying-yang" effect is linked with a shifted balance between the phosphorylation status of STAT3 and STAT5, intracellular ROS levels, and STAT5 SUMOylation/de-SUMOylation. STAT5 emerged as a definitive negative regulator of xCT at the transcriptional level, while STAT3 activation is coupled with increased system x c activity. We propose that careful classification of a patient's breast cancer subtype is central to effectively targeting STAT3/5 as a therapeutic means of treating breast cancer, particularly given that xCT is emerging as an important biomarker of aggressive cancers.

Introduction
Aggressive cancer cells adapt to increased levels of reactive oxygen species (ROS) that accompany their dysregulated metabolism by up-regulating the activity of the plasma membrane antiporter, system x c -, which releases glutamate in exchange for cystine taken up from the extracellular environment. Imported cystine is essential to cancer cells, as it is intracellularly reduced to cysteine for the synthesis of glutathione (GSH), an antioxidant molecule that serves as one of the main mechanisms by which cancer cells effectively maintain redox balance (reviewed in [1]). System x c consists of the xCT (SLC7A11) gene product linked at the plasma membrane to the 4F2 heavy chain (4F2hc, CD98) [2]. In cancer cells, heterodimers of xCT-4F2hc may also complex with a variant of the CD44 cell adhesion molecule (CD44v) [3]. Of these components, xCT most prominently influences the function of system x c -, with high expression levels correlating with higher antiporter activity [2]. It has been suggested that insufficient xCT expression may sensitize cancer cells to ROS-mediated damage [3,4] and potentially curb their rapid proliferation [5] and metastatic potential [6]. It has also been speculated that high levels of xCT may be an indicator of chemo-and radiation therapy resistance [7], making it a target for novel drug development.
Signal transducer and activator of transcription (STAT) proteins, particularly STAT3 and STAT5, are constitutively activated in a large number of human cancers, where their sustained activity contributes to dysregulated proliferation, apoptosis, angiogenesis, and immune surveillance (reviewed in [8]). Therefore, these particular STATs have emerged as pharmacological targets for cancer therapy, and the development of novel STAT3/5 inhibitors is of considerable clinical interest for treating breast cancer, glioma, and leukemia [9][10][11]. In order to both positively and negatively regulate transcription, STATs are activated, homo-or heterodimerize, translocate to the nucleus, and bind to their recognition sites in target gene promoter regions, also potentially interacting with other transcription factors to modulate gene expression [12][13][14]. By regulating the expression of numerous target genes, including, for example, MYC [15,16], BCL-2, BCL-XL, cyclin D1 (CCND1), cyclin D2 (CCND2), p21WAF/Cip1 (CDKN1A), and p27kip (CDKN1B) [17,18], STAT family members dynamically control complex intracellular mechanisms. It has also been shown that STAT3 and STAT5 may oppose each other's actions, producing reciprocal effects at target gene loci [19,20].
Numerous extracellular ligands, such as interleukin 6 and prolactin, induce canonical STAT3 and STAT5 singling cascades in relevant cell types [21,22]. ROS also modulate STAT activation in normal and cancer cells, and in turn, STAT3 and STAT5 modulate ROS production (reviewed in [23]). This latter effect may be due to their more recently described ability to translocate to the mitochondria, where they modulate cellular respiration and metabolism [24]. Given that system x c is intricately involved in redox homeostasis, we previously examined whether representative STAT family members regulate xCT expression, demonstrating by ChIP analysis that STAT3 and STAT5A are able to directly bind to the xCT promoter in MDA-MB-231 human breast cancer cells [25]. Our findings are computationally supported by several lines of evidence: xCT is listed as a STAT5A target gene based on published ChIP-chip, ChIP-seq, and other transcription factor binding site profiling studies assembled through the Ma'ayan Laboratory of Computational Systems Biology on-line database, Harmonizome (http://amp.pharm.mssm.edu/Harmonizome/dataset/CHEA+Transcription+Factor+Targets) [26], microarray studies indicate a link between STATs and changes in xCT mRNA levels [27,28], and PathwayNet predicts a transcriptional connection between xCT and STAT5A (http://pathwaynet.princeton.edu/predictions/geneset/?network=human-transcriptional-regulation&geneset=21547%2C13375). We have also shown that acutely blocking STAT3 and STAT5 phosphorylation using SH-4-54, a novel, structurally unique small molecule inhibitor that effectively targets interactions with the pTyr-SH2 domain to block both STAT3 and 5 dimerization and subsequent DNA-binding, also providing a promising means of targeting brain cancer stem cells [10], significantly up-regulates xCT expression and system x c activity in several human breast cancer cell lines [25]. This effect was coupled with increased ROS production [25] and potential NRF2 pathway activation, which has been computationally linked with STAT-mediated signaling through the JAK pathway via the NRF2-ome [29]. Interestingly, we observed that treating cells with a commercially available STAT5 inhibitor that targets the SH2 domain of STAT5 [30], inhibiting IFNα-stimulated STAT5 tyrosine phosphorylation without affecting STAT3 in human Burkitt's lymphoma Daudi cells, increased xCT mRNA and protein levels without affecting intracellular ROS levels [31], suggesting that STAT5 could directly affect transcription at the xCT gene locus by acting as a repressor. The current investigation examined the chronic effects of blocking STAT3/5 signaling in triple-negative MDA-MB-231 and estrogen receptor alpha (ERα)-positive T47D cells. Resistant clones representing sub-lines derived from both parental passage-matched wild-type breast cancer cell subtypes were selected based on their survival in response to chronic SH-4-54 treatment. Only a small number of individual treatment-resistant clones were obtained, confirming the overall potency of SH-4-54 and the importance of these STAT family members in breast cancer cells. SH-4-54-resistant clones were characterized based on their STAT3/5 activation profile (phospho-STATs relative to total protein), as well as changes in xCT mRNA levels, which were definitively confirmed by RNA-sequencing (NextGeneration analysis), as well as xCT protein levels and system x c activity. In addition, the stability of genotypic changes in the absence of further SH-4-54 treatment was assessed in vivo using murine xenografts. The overall goals of the current investigation were (1) to determine a potential mechanism by which blocking the activity of STAT3 and STAT5 affects system x c -, given the dynamic involvement of these particular transcription factors with mitochondrial function, redox balance, and the regulation of other key factors associated with cellular metabolism, which are all processes potentially interconnected with xCT expression, and (2) how these changes ultimately affect the genetic profile of different cancer cell types. Findings reported here may be of therapeutic interest for clinically applying STAT3/5 inhibitors to target cancers in which xCT expression is upregulated, including gliomas and aggressive breast cancers.

Materials and Methods
Cell Lines, Culture, and Production of SH-4-54-Resistant Cell Lines Both human cell lines were utilized in accordance with institutional biosafety guidelines. MDA-MB-231 and T47D human breast cancer cells lines were cultured according to the in vitro culture specifications outlined by ATCC. For clonal selection, cells were plated at several different densities into 10-cm dishes in either DMEM or RPMI supplemented with 10% fetal bovine serum to support the optimal growth of MDA-MB-231 or T47D cells, respectively. Media was changed every 2-3 days to administer SH-4-54 from a freshly thawed aliquot. After one or two months of continuous drug selection for T47D or MDA-MB-231 cells, respectively, individual clones were isolated by "picking" them using sterile cloning discs (Scienceware) presoaked in trypsin-EDTA. Each individual clone was transferred into one well of a 48-well plate and cultured to confluence in the presence of SH-4-54 prior to reseeding into a larger well format. For experiments, cells were plated into 6-well tissue culture-treated plates at 2.5x10 5 cells/well 24 hours prior to manipulation. Untreated parental MDA-MB-231 or T47D cells, referred to as wild-type counterparts, were passaged in parallel. All cells were determined to be mycoplasma-free. Cell viability was assessed using trypan blue exclusion during cell count determination.
Drugs SH-4-54, a novel small molecule STAT3/5 inhibitor [10], was reconstituted in DMSO at a 25 mM stock. Individual aliquots were stored at -20°C, and cells were treated with vehicle or an appropriate concentration of drug (initially at 10 μM, followed by a 5 μM maintenance dose). Recombinant human prolactin (Cedarlane) was used at 100 ng/ml. Capsazepine (Cayman Chemical), paclitaxel (Sigma-Aldrich), and bleomycin (Sigma-Aldrich) were reconstituted in DMSO and used at final concentrations corresponding to 25 μM, 0.1 nM, and 1500 mU, respectively.

Detection of SUMOylated Proteins by Immunoprecipitation
To detect SUMOylated proteins, cell extracts were prepared using lysis buffer supplemented with protease inhibitors plus N-ethylmaleimide (Pierce, Thermo Fisher Scientific) following the detailed protocol of Park-Sarge and Sarge (2009) [33]. Immunoprecipitation reactions (100 μg of total protein/1 reaction) were carried out with either normal rabbit IgG (DA1E; #3900, Cell Signaling Technology) as a negative control or total STAT5 antibody (sc-1081X, Santa Cruz Biotechnology). Immunocomplexes were isolated using protein A magnetic beads (#8687, Cell Signaling Technology) followed by standard Western blotting with either total STAT5, SUMO-1 (C9H1; #4940, Cell Signaling Technology), SUMO-2/3 (18H8; #4971, Cell Signaling Technology), or calnexin antibody as a negative control. Non-immunoprecipitated input was run in parallel on the same blots as a positive control.

ROS Assays
Intracellular ROS levels were measured at 24 hours post-plating using CM-H 2 DCFDA (Thermo Fisher Scientific) as described previously [31]. Each of 3 independent experiments included a negative control of unlabeled cells to verify staining specificity. Results were normalized to cell number obtained by staining with crystal violet.

Tumour Xenografts
Animal studies were reviewed and approved by the McMaster University Animal Research Ethics Board, complying with the Canadian Council of Animal Care guidelines. 4-6 week old female immunocompromised BALB/c nu/nu mice (Charles River) were implanted subcutaneously with 21-day release, 0.25 mg pellets of 17β-estradiol (Innovative Research of America) three days prior to subcutaneous injection with either MDA-MB-231 wild-type or MDA-MB-231 clone #2 (SH-4-54-resistant) cells prepared in sterile phosphate-buffered saline at 4×10 6 cells in 100 μL per animal. The viability of cells to be injected was assessed via trypan blue exclusion staining. Pellet implants are routinely carried out to promote the growth of MDA-MB-231 cells in our xenograft models [34]. Two animals were included for each group. Tumour size was evaluated every 3 days, and animals were sacrificed on Day 25 post-injection, when tumours in the wild-type group reached a mean volume of 794 mm 3 . Tumours were excised, flash-frozen in liquid nitrogen, and stored at -80°C.
Quantitative Real-Time PCR (qPCR) cDNA was prepared using Superscript III and oligo dTs (Thermo Fisher Scientific) from total RNA isolated from cell pellets (Qiagen RNeasy kit). Following DNase treatment (Ambion), cNDA was used as template for qPCR on a BioRad CFX Connect Real-Time System using primers listed in Table 1 and SsoAdvanced Universal SYBR Green Supermix (BioRad). Reference gene primers to amplify either β-actin or RNA polymerase II (RPII) have been described previously [32] and are also listed in Table 2. The 2 -[Δ][Δ]Ct method was used to calculate relative mRNA levels [35], and results are presented as fold changes of relevant controls.

Statistical Analyses
Results represent the mean ± SEM of at least three independent replicates for each experiment. Statistical differences between relevant groups were established by either t-test (denoted by stars) or 1-way ANOVA coupled with a Tukey's post-test (denoted by different letters) using GraphPad Prism software. Results were considered significant at p <0.05. Immunoblots depict a representative image of three independent experiments. biological replicate was split between two lanes to mitigate lane effects, with reads being subsequently combined during analysis using the Tuxedo protocol as previously described [36]. The RNA-sequencing dataset was submitted to the National Institutes of Health (NIH) Sequence Read Archive (SRA) as SRA accession SRP078574. The Galaxy Project, a web-based platform for biomedical data analysis, was used to perform several of the steps outlined in the Tuxedo protocol [37,38]. The FastQC tool (http://www. bioinformatics.babraham.ac.uk/projects/fastqc/) was applied to evaluate the quality of sequencing results in FASTQ format. Tophat was applied to align reads from individual biological replicates to the human GRCh38/hg38 assembly. Cufflinks created assembled transcripts, which were combined with the corresponding GRCh38/hg38 reference transcriptome annotation using Cuffmerge, creating a transcriptome assembly. Cuffdiff was used to assess transcript abundance in fragments per kilobase of transcript per million mapped reads (FPKM) and differential gene expression (DEG) by merging the transcriptome assembly with individual aligned reads created by Tophat. The false discovery rate-adjusted p-value (q-value) was set to >0.05.

RNA-Sequencing Analysis
The Tuxedo suite, a bioinformatics pipeline that uses FPKM data to normalize the number of reads to library size and gene length [36], was used to investigate DEGs between the two different MDA-MB-231 cells lines and between the two different T47D cell lines. It is worth noting that other software packages and pipelines for differential expression analysis that employ alternative normalization and statistical methods are available. Although no consensus on a "best practice" currently exists for RNA-sequencing, it has been noted that FPKM normalization leads to overly-conservative reports of DEG using the Tuxedo pipeline compared to other methods such as DEseq [39]. Therefore, it is possible that additional, less stringently selected DEGs could be identified in our study using alternative differential expression analysis. One advantage of the Tuxedo pipeline and FPKM data is the ability to perform further analysis of alternative splicing events, which may be of value for future investigations.
The Bioconductor package CummeRbund was used to visualize Cuffdiff output files in RStudio (version 0.99.467 [44][45][46]. This platform facilitated preparation of scatter and volcano plots for each pairwise comparison, as well as an expression level plot and heatmap comparing experimental groups. The Lander/Waterman equation (C = LN/G) was applied to calculate mean genome coverage: C denotes coverage, G corresponds to genome/transcriptome length (for RNA-sequencing), L denotes average read length, and N stands for the average number of reads [40]. The base coverage tool in Galaxy (https://toolshed.g2.bx.psu.edu/view/devteam/ basecoverage/b8a9e718caa3) was used to derive the total length of annotated transcripts executed on the most recent human genome assembly (GRCh38), which was applied to calculate coverage.
DAVID, a web-based bioinformatics tool, was applied to perform ontological and KEGG pathway enrichment analyses to functionally interpret gene sets [41,42]. The list of DEGs for each pairwise comparison obtained from RNA-sequencing was imported into the "functional annotation" tool, with homo sapiens as the reference species. Enriched KEGG pathways and Gene Ontology Biological Processes (GO:BP) terms were identified with the Expression Analysis Systematic Explorer (EASE) threshold (maximum EASE score/p-value) set to a default of 0.1, which is used by DAVID to identify significant gene enrichment. Fold-enrichment representing the ratio of the proportion of input genes relative to the number of genes represented by a particular term or pathway within the reference human genome was also reported.
qPCR data was used to validate RNA-sequencing results (see section on qPCR above). For each of the 18 target genes selected for validation, pairwise comparisons were based on foldchanges calculated relative to appropriate controls (wild-type counterparts of each cell type).
To determine the experimental asymmetrically distributed SEM for each mean, which is necessary for linearly representing data derived from an exponential analysis, SEMs derived from each ΔCT value were used to calculate upper and lower 2 −ΔΔCT values [43]. Linear regression was used to test the overall correlation between RNA-seq and qPCR results, with α set to 0.05 [44].

Results
Phospho-STAT3 and Phospho-STAT5 Levels Are Altered by Chronic SH-4-54 Treatment in MDA-MB-231 and T47D Treatment-Resistant Clones Western blotting was performed on selected MDA-MB-231 and T47D SH-4-54-resisistant clones following long-term drug treatment. Phospho-STAT3 levels were significantly lower in MDA-MB-231 resistant clones relative to wild-type MDA-MB-231 cells, with the inverse occurring in T47D resistant clones in which phospho-STAT3 levels increased by approximately 3-fold relative to their respective untreated wild-type counterpart ( Fig 1A). Given that SH-4-54 inactivates both STAT3 and STAT5, its effect on phospho-STAT5 was also examined in resistant clones of both cell types. Interestingly, although the band for phospho-STAT5, which is expected to migrate at 95 kDa, was not detected after rapid exposure in wild-type MDA-MB-231 cells, two other bands at approximately 130 and 200 kDa were observed, with the latter notably absent in the SH-4-54-resistant clone ( Fig 1B). A similar higher molecular weight banding pattern for phospho-STAT5 was observed in T47D cells. However, in contrast to MDA-MB-231 cells, the band at approximately 200 kDa markedly increased in overall intensity in the SH-4-54-resistant T47D clone ( Fig 1B). A faint band at 95 kDa corresponding to uninduced, basal levels of phospho-STAT5 was detected in T47D wild-type cells (Fig 1B, indicated by the arrow), which was absent in its corresponding clone. This band matched the canonical migration of phospho-STAT5 obtained in response to prolactin treatment of T47D cells ( Fig 1C). In addition to the 95 kDa phospho-STAT5 band induced by PRL stimulation of T47D cells, a prominent 130 kDa and very weak 200 kDa band were detected after prolonged exposure (Fig 1C), which matched the banding pattern observed in MDA-MB-231 wild-type cells treated with PRL ( Fig 1D). While MDA-MB-231 cells are more resistant to canonical PRL signaling via the activation of STAT5 via JAK2 phosphorylation due to lower PRL receptor numbers compared to T47D cells [45], we have previously shown that STAT5 does become phosphorylated in response to a 100 ng/ml PRL treatment in this cell line, with maximal levels of phospho-STAT5 obtained after 24 hours [32]. Interestingly, the type of complex STAT5 banding pattern observed in the present study has been linked with SUMOylation, a particular type of post-translational modification that has been shown to significantly influence the predicted molecular weight of STAT5 during early lymphoid development [46,47]. Given that SUMOylation/de-SUMOylation is a cyclical process, we examined whether this particular STAT5 modification could be the underlying mechanism resulting in non-canonical STAT5 protein migration in breast cancer cells resistant to SH-4-54. We therefore added N-ethylmaleimide to the lysis buffer in subsequent experiments, which served to inhibit de-SUMOylation by blocking the action of SUMO proteases [33], thereby enhancing the retention of SUMOylated bands in protein extracts.
In addition to changes in the phosphorylation status of STAT3 and STAT5, a lower cell count was obtained for SH-4-54-resistant MDA-MB-231 and T47D clones over 48 hours relative to their respective untreated wild-type counterparts ( Fig 1E). Lower counts were not due to changes in cell viability, given that trypan blue exclusion, which identifies non-viable cells, did not indicate any significant differences between the wild-type and SH-4-54-resistant cell lines (data not shown). Selection in SH-4-54 also significantly changed the morphology of MDA-MB-231 resistant clones compared to their wild-type counterpart (Fig 1F). No significant morphological changes were visually observed between wild-type and SH-4-54-resistant T47D cells (Fig 1G).
xCT Expression and System x c -Activity Is Differentially Regulated in MDA-MB-231 and T47D Clones Resistant to SH-4-54 Two resistant clones from each cell type were selected for further analysis, denoted as clone #1 and #2. qPCR revealed that xCT mRNA levels were approximately 2-fold lower in the two MDA-MB-231 STAT3/5 inhibitor-resistant clones relative to wild-type cells, with an inverse 2 to 3-fold increase occurring in T47D clones relative to their wild-type counterpart (Fig 2A). Given that we previously showed SH-4-54 to acutely increase xCT promoter activity in MDA-MB-231 cells, we examined its chronic effects at the transcriptional level in a representative resistant clone. Relative to wild-type MDA-MB-231 cells, xCT promoter activity was significantly reduced following long-term SH-4-54 treatment (Fig 2B). Importantly, the changes in xCT mRNA levels were reflected at the protein level, densitometrically corresponding to a 2-fold decrease in MDA-MB-231 SH-4-54-resistant clones relative to wild-type cells, while xCT protein levels increased significantly by 2.5-fold in T47D clones in response to chronic SH-4-54 treatment (Fig 2C). xCT is the functional component of system x c -, and as such, changes in protein levels corresponded to significant down-or up-regulation of its activity, including cystine uptake ( Fig 2D) and glutamate release (Fig 2E), in both sets of MDA-MB-231 and T47D STAT3/5 inhibitor-resistant clones, respectively.
xCT Expression in Tumours Isolated from Murine Xenografts Remains Low in Animals Subcutaneously Injected with MDA-MB-231 Clones Resistant to Chronic SH-4-54 Treatment Relative to Wild-Type Cells To ascertain whether chronic SH-4-54-induced repression of system x c is sustained in vivo without further drug administration, which would eliminate any effects of in vitro culture conditions and indicate permanent cellular changes at the gene level, we injected cells of one of the representative MDA-MB-231 resistant clones into nude mice in parallel to wild-type cells. Mean tumour growth rates (n = 2 for each cell type) revealed that wild-type-induced tumours grew at a faster rate compared to tumours initiated from SH-4-54-resistant clones ( Fig 3A). qPCR demonstrated that xCT mRNA levels were reduced in tumours isolated from animals injected with SH-4-54 resistant cells compared to those injected with wild-type cells ( Fig 3B). Western blot analysis of protein isolated from each tumour using standard lysis conditions confirmed changes occurring at the mRNA level, showing that xCT expression appears to be lower in animals initially injected with the SH-4-54-resistant clonal cells compared to those receiving wild-type cells (Fig 3C). In contrast, 95 kDa phospho-STAT5 levels were higher in tumours initiated from SH-4-54-resistant clones, while phospho-STAT3 levels did not change when normalized to levels of total STAT3 (Fig 3C). These results support that activated STAT5 may continue to repress xCT expression in vivo, and that the changes in  (Fig 4A and 4B) highlighting overall similarities and differences in gene expression between wild-type and SH-4-54-resistant clones within a given comparison. Volcano plots generated by plotting log2(p-value) against log2(fold-change) of individual genes (Fig 4C and 4D) highlight the DEGs in each comparison. Density plots (Fig 4E and 4F) illustrate expression level distributions, with DEGs represented by regions of non-overlapping groups. Heatmaps (Fig 4G and 4H) provide a visual representation of the expression level [in log10(FPKM+1)] for all DEGs across the two different comparisons. A total of 4227 DEGs were identified in the MDA-MB-231 wild-type versus MDA-MB-231 STAT3/5 inhibitor-resistant clone comparison, while 752 DEGs were identified in the T47D wild-type versus T47D STAT3/5 inhibitor-resistant clone comparison (S1 Table). All results obtained for GO:BP terms and KEGG pathway analysis are listed in S2 Table. A subset of 18 genes was chosen to validate fold-changes in DEGs obtained by RNA-sequencing via qPCR analysis (Table 3). Linear regression confirmed high overall concurrence between the two methods ( Fig 4I).  Table 5, with a substantial number containing putative STAT3 or STAT5 binding sites in their promoter regions based on the Weizmann Institute of Science Human Gene Database (GeneCards). Among the list in Table 5 is MYC, an important oncogene and known STAT3/5 target gene [15,16], which clones in which xCT protein levels were significantly higher compared to WT T47D cells. A corresponding densitometric analysis in graphical format is presented in the left panel, corresponding to analysis of at least 3 independent blots. The activity of system x c was inversely affected, with relevant changes in (D) cystine uptake and (E) glutamate release demonstrating high concordance with xCT expression levels. Data represent the mean of three independent experiments (±SEM) calculated relative to appropriate controls. A star (*) denotes statistically significant differences determined by a t-test (p<0.05). Different letters a, b, or ab in panels A, C, and D correspond to statistical differences between groups (p < at least 0.05), as determined by One-way ANOVA and a Tukey's post-test.    (Tables 2 and 3). qPCR validations for several representative STAT3/5 target genes identified by RNA-sequencing are graphically represented in Fig 5, including BCAR3 (Fig 5A), GSTM3 (Fig 5B), MYC (Fig 5C), AKT1 (Fig 5D), CCND1 (Fig 5E), CCND2 (Fig 5F), CSF1 (Fig 5G), PANX1 (Fig 5H), SOCS3 (Fig 5I), SUMO3 (Fig 5J), and SLC7A11 (xCT; Fig 5K), confirming the overall specificity of results obtained by RNA-sequencing for a range of fold-changes. A complete list of 18 DEGs validated by qPCR is presented in Table 3.

STAT5 Is De-SUMOylated in SH-4-54-Resistant MDA-MB-231 Clones
In response to changes in xCT expression, intracellular ROS levels could also change due to the resulting effects on system x c activity. DCFDA analysis revealed that in MDA-MB-231 SH-4-54-resistant clones, ROS levels were approximately 3-fold higher than in their wild-type counterpart, with the inverse occurring in T47D cells (Fig 6A). Given that inhibition of STAT3 activity has been linked with sensitizing resistant cells to chemotherapy [51], we examined the  xCT Is Differentially Regulated by STAT3/STAT5 and ROS/SUMOylation in Breast Cancer Cell Subtypes effects of paclitaxel, a chemotherapeutic drug commonly used to treat breast cancer, as well as bleomycin and capsazepine, two ROS-inducing agents, on the proliferation of MDA-MB-231 cells. While low doses of paclitaxel and bleomycin significantly reduced the number of both wild-type and SH-4-54-resistant cells at 72 hours post-treatment, the effects of capsazepine were only observed in the latter group (Fig 6B), suggesting that this agent likely has a distinct mechanism of action. In response to changes in ROS levels, cells may also undergo changes in SUMOylation/de-SUMOylation cycles [52,53]. It has been reported that STAT5 may be regulated through SUMOylation, which is reflected by incremental 40 kDa changes in its protein migration pattern relative to the canonical 95 kDa band [46,47]. In Fig 1B, we observed that detection of phospho-STAT5 by standard Western blotting produced an unexpected migration pattern in both MDA-MB-231 and T47D cells that included bands at approximately 130 and 200 kDa in addition to the expected band at 95 kDa. This was similar to banding patterns associated with STAT5 SUMOylation reported by others [47]. Using lysis buffer conditions that included Nethylmaleimide, which blocks the action of de-SUMOylating enzymes [33], we confirmed the presence of these bands in human breast cancer cells, demonstrating that levels of 95 kDa phospho-STAT5 increased modestly, but significantly, in MDA-MB-231 SH-4-54-resistant clones while decreasing in T47D SH-4-54 resistant clones relative to their respective wild-type counterparts ( Fig 6C). Furthermore, the intensity of bands at approximately 130 and 200 kDa decreased by 2-fold in MDA-MB-231 clones relative to wild-type cells while increasing in T47D SH-4-54-resistant cells (Fig 6C). Under these specific lysis conditions, anti-SUMO-1 and Table 4. Subset of select genes reported by others to be regulated by STAT3 or STAT5 [16,20,[48][49][50]. These genes were identified by RNA-sequencing to be differentially expressed in MDA-MB-231 and T47D SH-4-54-resistant clones relative to wild-type (untreated) cells.    (Fig 6D), suggesting that changes in the SUMOylation of total STAT5 may affect STAT5 phosphorylation, which may be related to phospho-STAT3 inactivation in SH-4-54-resistant cells derived from this cell line. Consistent with the notion of SUMO-2/3 modulating STAT5 activation, RNAsequencing revealed that SUMO-3 mRNA levels were significantly down-regulated in MDA-MB-231 SH-4-54-resistant clones relative to wild-type cells (S1 Table), which was validated by qPCR (Fig 5J). In order to definitively establish that STAT5 is indeed differentially SUMOylated, immunoprecipitations were carried out, demonstrating co-migration of total STAT5 and SUMO-2/3 ( Fig 6E). Furthermore, STAT5 was less SUMOylated in MDA-MB-231 STAT3/5 inhibitor-resistant clones than in wild-type cells (Fig 6E, compare lanes 3 and 6), with no changes in the overall intensity of the IgG heavy chain across groups.

Discussion
System x c provides an important adaptive mechanism that helps stabilize the cellular consequences of altered metabolism in aggressive cancer cells by clearing excess ROS, ensuring longterm survival. STAT3 and STAT5 control the expression of numerous genes related to cancer cell metabolism, proliferation, and survival, integrating signals from diverse extracellular stimuli including cytokines, growth factors, hormones, and ROS. Various cytokines and growth factors induce ROS production [54,55], and in turn, ROS inactivate tyrosine phosphatases [56], thereby potentially up-regulating the phosphorylation/activation of STAT3, STAT5, and other important signaling molecules [57]. It has also been shown that oxidative stress and ROS production may interfere with the canonical JAK/STAT signaling pathway [58,59], and that STAT3 and STAT5 may have mitochondrial functions [24]. STAT3 and STAT5 are therefore intricately and complexly involved in oxidative metabolism in cancer cells (reviewed in [23]), although literature conflicts regarding their regulation by ROS as well as their influence on ROS production. In an extension of our previous investigation that examined the effects of acute STAT3/5 inhibition on system x c in human breast cancer cells, representative subtypes of resistant breast cancer cells clonally selected through long-term treatment with the novel STAT3/5 inhibitor, SH-4-54, were assessed for changes in xCT expression and system x c activity relative to untreated wild-type cells using functional and RNA-sequencing-based analyses. Triple-negative, metastatic MDA-MB-231 cells were chosen as representatives of basal B (claudin-low, ER-negative, PR-negative, HER2-negative) human breast cancer cells that have an intermediate response to chemotherapy [60] and express high levels of xCT [61]. This cell xCT Is Differentially Regulated by STAT3/STAT5 and ROS/SUMOylation in Breast Cancer Cell Subtypes Fig 5. 11 genes selected for graphical representation of relative qPCR fold-changes indicate significant differences, which confirmed the status of 18 total DEGs identified by RNA-sequencing (refer to Table 3). For each group, data represents the mean of 3 independent biological replicates, each analyzed in duplicate, with error bars indicating the SEM calculated using the 2 -[Δ][Δ]Ct method. Data represent the mean of three independent experiments (±SEM) calculated relative to appropriate controls. A star (*) denotes statistically significant differences determined using a t-test (p<0.05). line also constitutively expresses high levels of phospho-STAT3 and is more resistant to stimulus-induced STAT5 activation. In contrast, T47D cells represent luminal A (ERα-positive, PRpositive, HER2-negative) cells that are largely responsive to both endocrine and chemotherapy [60]. T47D cells have low levels of constitutive phospho-STAT3 and phospho-STAT5, but respond rapidly and robustly to stimuli that induce both STAT3 and STAT5-mediated signaling, such as prolactin [21]. Both represent cell lines that do not express the HER2/neu receptor, and therefore cannot be clinically targeted using trastuzumab, a monoclonal antibody that interferes with the HER2/neu receptor used to treat HER2-positive metastatic breast cancer [62]. Application of these particular cell types provided a starting point to launch our investigation into potential treatment resistance potent STAT3/5 inhibitors. Chronically blocking STAT3/5 phosphorylation and subsequent DNA binding with SH-4-54 resulted in the isolation of resistant MDA-MB-231 clones that proliferated more slowly and displayed down-regulated xCT mRNA and protein levels accompanied by a decrease in system x c activity. While the SH-4-54-resistant clones derived from wild-type T47D cells also grew more slowly over a 48 hour period relative to their wild-type counterparts, xCT expression was up-regulated concomitant with functional increases in cystine uptake and glutamate release. Importantly, differential changes in SLC7A11 (xCT) mRNA levels were validated by RNA-sequencing, further confirming that STAT3/5 signaling is central to the transcriptional regulation of xCT in human breast cancer cells. In addition to these changes, the morphology of MDA-MB-231 cells was affected by chronic SH-4-54 treatment, with cells adopting a more rounded phenotype relative to the typical spindle shape associated with wild-type cells. The morphology of SH-4-54-resistant T47D clones did not change significantly, with cells retaining the cluster-like formation typical of wild-type T47D cells. Classification of human breast cancer cell lines has established an association between biological characteristics, including morphology and invasiveness, and transcriptionally defined subtypes [63]. While basal B cells are more "mesenchymal-like" and appear less differentiated, luminal cells form tight cell-cell junctions, appearing more homogenous and, therefore, more differentiated [63]. Based on their altered morphology, it is tempting to speculate that the MDA-MB-231 SH-4-54-resistant cells are more luminal than basal-like, which remains to be experimentally evaluated. We previously showed that SH-4-54, which is able to block both phospho-STAT3 and phospho-STAT5, induces rapid ROS production in human breast cancer cells (within 4 hours of treatment), and that increases in SH-4-54-induced xCT promoter activity may be abolished by pre-treating MDA-MB-231 cells with N-acetylcysteine. These acute changes (within 24 hours) are likely due to the effect of SH-4-54 blocking phospho-STAT3, given that WP1066, a small molecule inhibitor know to block JAK2/STAT3 signaling by degrading JAK2 protein, produced identical effects [31]. We therefore speculated that short-term changes in xCT promoter activity are likely due to increased NRF-2 pathway activation in response to increased levels of intracellular ROS, resuling in up-regulated xCT transcriptional activity. Interestingly, while a commercially available STAT5 inhibitor that blocks phospho-STAT5 without targeting the activation of STAT3 also upregulated xCT promoter activity, xCT mRNA levels, and levels of MDA-MB-321 clones relative to their WT counterpart (indicated by the arrow), with an overall decrease in the intensity of bands at approximately 130 and 200 kDa, while in T47D clones, the opposite occurred. (D) Western blots of lysates derived from MDA-MB-231 WT and clones probed with SUMO-1, SUMO-2/3, and total STAT5 antibodies revealed similar banding patterns, with common bands observed at approximately 95, 130, and 200 kDa. Lower overall band intensities were observed in SH-4-54-resistant clones compared to their WT counterpart, with decreased levels of phospho-STAT3. (E) A representative set of IPs confirmed that SUMO-2/3 co-migrated with total STAT5 in MDA-MB-231 cells, and that levels of SUMOylated STAT5 were lower in SH-4-54-resistant clones than in WT cells (compare lanes 3 and 6). Data represent the mean of three independent experiments (±SEM) calculated relative to appropriate controls. A star (*) denotes statistically significant differences determined using a t-test (p<0.05). doi:10.1371/journal.pone.0161202.g006 xCT Is Differentially Regulated by STAT3/STAT5 and ROS/SUMOylation in Breast Cancer Cell Subtypes xCT protein, it did not acutely alter cystine uptake or ROS production, suggesting that STAT5 could be acting as a direct transcriptional repressor in an ROS-independent manner [31]. This supported our previous finding that the STAT5 inhibitor completely abrogated STAT5A binding to the xCT promoter region containing a GAS/STAT site [31]. In light of results produced in the current investigation following chronic STAT3/5 inhibition with SH-4-54 in MDA-MB-231 cells, we provide further evidence that the effect of STAT5 on xCT expression occurs at the gene level, inhibiting xCT transcriptional activity and expression. As a consequence, system x c activity is also decreased in selected MDA-MB-231 clones, leading to higher ROS levels. Interestingly, chronic treatment of T47D cells with SH-4-54 produced clones in which phospho-STAT3 levels increased, culminating in up-regulated xCT expression, increased function of system x c -, and lower levels of intracellular ROS, which is consistent with increased cystine uptake countering any potential increases in ROS by promoting GSH synthesis. These results suggest that STAT3 and STAT5 have distinct functions in regulating the overall expression of xCT in breast cancer cells.
In addition to confirming changes in xCT expression at the mRNA level, RNA-sequencing revealed changes in the expression levels of numerous known STAT3 and STAT5 target genes. For example, MYC, a well-known hallmark of aggressive cancers, was validated by qPCR to follow the same pattern of differential gene expression as xCT in MDA-MB-231 and T47D SH-4-54-resistant clones relative to their respective wild-type counterparts. Another noteworthy putative STAT5 target gene, BCAR3 (breast cancer anti-estrogen resistance 3), which encodes a SRC homology 2 domain by which it may direct cellular signaling to increase cell motility and estrogen-independent proliferation in human breast cancer cells [64], was confirmed by qPCR to be significantly down-regulated by approximately 2-fold in MDA-MB-231 SH-4-54-resistant clones while being significantly up-regulated by at least 2-fold in T47D clones. Specific types of breast cancers initially depend on estrogen for tumour growth and disease progression, largely responding well to anti-estrogen therapies including tamoxifen. However, these breast cancers often become treatment-resistant, and BCAR3 was identified as being involved in the development of anti-estrogen resistance [65]. The STAT3/5 status may induce changes in estrogen-responsiveness and the ERα-status of breast cancer cells, which is supported by the finding that STAT5 may regulate ERα [66]. The altered genotype of T47D SH-4-54-resistant clones reported here may therefore contribute to an estrogen-independent phenotype, warranting further investigation.
In cancer cells, mitochondria produce ROS as a consequence of up-regulated metabolic activity, mediate apoptosis, and are also targeted by ROS-induced damage we have previously shown that STAT3 is able to bind to the xCT promoter [31], demonstrating that inhibition of STAT3 activation may directly affect system x c to affect cellular redox balance. However, chronic suppression of STAT3 activity in SH-4-54-resistant MDA-MB-231 clones could also be further disrupting ROS levels through mitochondrial effects, including STAT3-mediated regulation of the electron transport chain (ETC) (reviewed in [31]). One of several protective means by which mitochondrial quality may be maintained in oxidatively stressed cells, represented by the STAT3/5-resistant MDA-MB-231 clones characterized in the current investigation, is through the autophagy/lysosome pathway (reviewed in [67]). Lysosomal responses to ROS that are generated from perturbations in cellular metabolism, including changes in the activity of system x c -, are closely linked with redox homeostasis. Interestingly, the lysosome was an enriched KEGG pathway identified through further analysis of the MDA-MB-231 RNA-sequencing comparison. Numerous important genes associated with lysosomal activity, most notably cathepsin B, D, F, K, S, and Z (CTSB to CTSZ), which play an important role in cellular protein turnover [68], were significantly up-regulated in SH-4-54-resistant MDA-MB-231 cells (S1 and S2 Tables). In keeping with the autophagy/lysosome pathway playing a potential pro-survival role in STAT3/5 inhibitor-resistant MDA-MB-231 clones, which could equip them to "manage" increased levels of ROS as a consequence of permanent xCT downregulation, KEGG pathway analysis of the MDA-MB-231 comparison identified changes in the apoptosis pathway (S2 Table), which included significant down-regulation of FAS (FAS cell surface death receptor), FADD (FAS-associated via death domain), CYCS (cytochrome C), DFFA (DNA fragmentation factor alpha), and DFFB (DNA fragmentation factor beta; caspaseactivated DNase) (S1 Table). From the RNA-sequencing results that indicate an overall downregulation of pro-apoptostic genes, which will need to be functionally confirmed in future experiments, we speculate that MDA-MB-231 SH-4-54-resistant clones have "adapted", to a certain extent, to higher levels of intracellular ROS. This is supported by the observation that, while MDA-MB-231 SH-4-54-resistant clones proliferated more slowly than their wild-type counterpart, their overall survival was not significantly affected.
A key finding that emerged from our study was that STAT5 SUMOylation/de-SUMOylation mediates, at least in part, differential STAT5 activation in MDA-MB-231 and T47D SH-4-54-resistant cells. SUMOylation/de-SUMOylation is a post-translational process that is influenced by fluctuating ROS levels (reviewed in [52,53]). Indeed, a shift toward de-SUMOylation may be a central mechanism by which cells attempt to regain homeostasis in response to oxidative stress [52]. SUMOylation changes the properties of the protein that is targeted by SUMO moieties, in turn affecting diverse cellular processes including DNA replication and repair, mitosis, signal transduction, nuclear transport, as well as the regulation of transcription [69]. The STAT5 protein harbors several SUMOylation sites at its C-terminal end [46]. Importantly, STAT5 activation has been shown to be influenced by SUMOylation during early lymphoid development, with SUMO-STAT5 antagonizing STAT5 tyrosine phosphorylation, thereby impairing STAT5 nuclear signaling [46,47]. We have shown that in MDA-MB-231 SH-4-54-resistant clones, increased STAT5 phosphorylation is linked with de-SUMOylation, most likely via the action of two nearly identical proteins collectively referred to as SUMO-2/3 [53]. In support of this notion, RNA-sequencing validated by qPCR showed that SUMO-3, a ubiquitin-like protein covalently attached as a monomer or polymer to lysine residues on target proteins [53], was significantly down-regulated in SH-4-54-resistant MDA-MB-231 clones. Covalent substrate attachment of SUMO-3 to a target protein requires several steps, including prior ATP-dependent activation and transesterification of the SUMO-specific, heterodimeric E1 ligase complex SAE1-UBA2 (SAE2) [70] and subsequent linkage to the E2 ligase UBC9, which is encoded by the UBE2I gene [71]. Others have shown that UBC9 plays a key role in STAT5 activation during lymphopoiesis [47]. Importantly, RNA-sequencing revealed that UBE2I expression was also significantly down-regulated in MDA-MB-231 SH-4-54-resistant clones (S1 Table). Following conjugation of the SUMO moiety to UBC9, an E3 ligase transfers it onto the substrate protein, and SUMO-specific proteases then de-SUMOylate the target, which can then be activated and undergo nuclear translocation as part of a cycle that is summarized in Fig 7. While most SUMOylated proteins undergo cyclical SUMO conjugation/de-conjugation, oxidative stress may influence this process (reviewed in [53]). Indeed, ROS have been shown to inactivate the SUMO cycle by inducing the formation of a disulfide bridge between UBA2 and UBC9, resulting in overall net substrate protein de-SUMOylation [72]. This fits well with our finding that in MDA-MB-231 SH-4-54 resistant clones, in which ROS levels are higher than their wild-type counterpart, STAT5 is de-SUMOylated, while the inverse occurs in T47D resistant clones that display lower ROS levels than their parental cells. Numerous other SUMO protein substrates have been identified in mammalian cells, including, for example, DNMT3A/B, FAS, HIF1α, JUN, PCNA, and p53 (reviewed in [53]), with several of these targets also being identified in our RNA-sequencing analysis (S1 Table). It is therefore possible that changes in the expression of these and other cellular factors is an indirect effect of chronic  (3) de-SUMOylation of STAT5 (details shown), which (4) enables STAT5 to be phosphorylated. Activated STAT5 translocates to the nucleus, (5) where it functions as a transcriptional repressor by binding to a GAS/STAT site xCT Is Differentially Regulated by STAT3/STAT5 and ROS/SUMOylation in Breast Cancer Cell Subtypes STAT3/5 inhibition in MDA-MB-231 cells that involves ROS-mediated de-SUMOylation, which remains to be explored.
From our findings, it appears that selectively further destabilizing the cellular redox status may be critical to produce clinically meaningful outcomes linked to chronic treatment with SH-4-54 and other potent STAT3/5 inhibitors. Others have shown that a low dose of capsazepine generates significant levels of ROS, thereby sensitizing colorectal cancer cells to apoptosis [73]. Interestingly, capsazepine effectively reduced the number of SH-4-54-resistant MDA-MB-231 clones without affecting their wild-type counterpart. In contrast, treatment with bleomycin, as well as the chemotherapeutic agent paclitaxel, which interferes with the breakdown of microtubules during mitosis and is used to treat breast carcinomas [74], significantly reduced cell numbers in both resistant and wild-type MDA-MB-231 cells. Relevantly, it has been shown in acute myeloid leukemia cells that chemotherapeutic drugs ROS-dependently inhibit the SUMO-conjugating enzyme UBC9, and that pro-oxidants or inhibition of the SUMO pathway by anacardic acid are able to initiate apoptosis in chemoresistant patient samples and leukemic stem cells [75]. Inhibiting STAT3/5 activity with SH-4-54 in conjunction with further raising intracellular ROS levels beyond the cellular detoxification capacity while simultaneously affecting the SUMO pathway may be a viable therapeutic strategy to target treatment-resistant cancer cells. In support of this notion, inhibiting constitutively activated STAT3 resensitizes drug-resistant lymphomas and myelomas to cisplatin, fludarabine, adriamycin, and vinblastine-induced apoptosis [51], and pairing the cyclooxygenase inhibitor ibuprofen with pharmacological system x c inhibition using sulfasalazine synergistically improves its antitumor efficacy in a murine sarcoma model [76]. However, understanding the underlying mechanism of drug action may be a critical component in selecting effective therapeutic combination treatments. Capsazepine may have a unique mechanism of action distinct from bleomycin, which produces ROS during its induction of DNA strand breaks [77]. Interestingly, we speculate that the action of bleomycin may not be a relevant mode of inducing ROS, as expression of glutathione peroxidase 7 (GPX7), which plays a role in protecting esophageal epithelial cells against oxidative DNA damage and double-strand breaks [78], was identified by RNA-sequencing to be significantly up-regulated in MDA-MB-231 SH-4-54-resistant clones relative to wild-type cells (S1 Table). We have previously shown that at 25 μM, capsazepine effectively inhibits glutamate release through system x c in MDA-MB-231 cells [79], and as chronic SH-4-54 treatment does not eliminate xCT expression entirely, capsazepine may therefore be blocking any remaining system x c activity, which would explain our findings.
Our study may have a significant impact on determining the therapeutic efficacy of novel STAT3/5 small molecule inhibitors such as SH-4-54 aimed at treating breast cancer and other aggressive cancers in which the ying-yang effects of STAT proteins play a central role, such as leukemias. SH-4-54 is highly effective at inducing cell death in glioblastomas [10] and in MDA-MB-231 and T47D cells (based on our observations during the selection of treatmentresistant clones). Nevertheless, treatment-resistant cells do emerge, as is the case with most drug regimens aimed at eradicating cancer cells. We have shown that in representative SH-4-54-resistant subtypes of breast cancer cells, shifts in cellular redox balance and changes in the SUMOylation profile may prove therapeutically beneficial, as they may be more vulnerable to further induction of ROS production. This is particularly relevant given that the SUMO cycle is dysregulated in multiple myeloma, has been associated with an adverse outcome in cancer patients [80], and inhibition of a SUMOylation-dependent transcriptional program induces in the xCT promoter, thereby (6) reducing xCT mRNA and (7) xCT protein levels. This ultimately destabilizes the cellular redox balance by (8) limiting the activity of system x c at the plasma membrane, reducing the import of cystine concomitant with the export of glutamate. doi:10.1371/journal.pone.0161202.g007 xCT Is Differentially Regulated by STAT3/STAT5 and ROS/SUMOylation in Breast Cancer Cell Subtypes death of MYC-over-expressing cancer cells [81]. Consistent with reports by others [75], our study demonstrates that cancer cells exhibiting a de-SUMOylated profile produce smaller tumours in murine xenografts. Importantly, these tumours also express lower levels of xCT coupled with higher levels of phosophorylated STAT5 compared to wild-type MDA-MB-231 cells in which STAT5 is SUMOylated, even in the absence of continued SH-4-54 treatment.

Conclusion
The current study has identified that inhibition of constitutive STAT3 phosphorylation with a novel and potent STAT3/5 inhibitor, SH-4-54, induces ROS, thereby affecting the SUMO pathway by favoring de-SUMOylation and subsequent phosphorylation of STAT5. Activated STAT5 is then able to serve as a transcriptional repressor at the xCT gene locus, reducing xCT expression and destabilizing one of the important redox balancing mechanisms of an aggressive triple-negative breast cancer subtype by limiting cystine uptake through system x c - (Fig 7).
This ultimately renders relevant target cells more susceptible to oxidative stress. In contrast, chronic SH-4-54 treatment of representative ERα-positive cells that initially respond to STAT5 signals results in STAT3 activation and up-regulation of xCT, potentially leading to a more aggressive cancer subtype.
Supporting Information S1