Glucocorticoid receptors are required effectors of TGFβ1-induced p38 MAPK signaling to advanced cancer phenotypes in triple-negative breast cancer

Parental (unmodified) MDA-MB-231 cells are a gift from Dr. Ronald Wegener and were maintained in 10% FBS and 1% P/S. Cells that were used for generation of CRISPR/Cas cell line were authenticated on April 27, 2017, by the University of Arizona Genetics Core and were compared to the ATCC short-tandem repeat (STR) database. Additionally, cells tested negative for Mycoplasma. U2OS cell lines were obtained from Dr. John Cidlowski. These cells were grown in 10% FBS, 1% P/S, 1% glutamax, 0.2 mg/mL Hygromycin, and 200 μg G418. The TNBC cell lines: Hs578T, MDA-MB-468, and HCI-10 cells were maintained in 10% FBS and 1% P/S. All of these cell lines tested negative for Mycoplasma.

All materials and reagents are listed in Supplementary Table 5.

Stable sh14-3-3ζ (TRCN0000029404 and TRCN0000029406) were generated by transducing MDA-MB-231 models with CMV-Neo lentiviral vectors containing target gene shRNA sequences (MISSION TRC library - Sigma). Pools were selected and maintained through culture with 0.75 mg/mL of G418 sulfate (Corning).

Cells were plated in a six-well plate at a density of 1.5 × 10⁵ cells to achieve monolayer growth at 100% confluency. Pretreatment (hormone, growth factor) was performed at various times before the scratch was created, as indicated. Experiments that required hormone (Dex), growth factor (Hepatocyte Growth Factor—HGF), or cytokine (Transforming Growth Factor Beta β1 – TGFβ1) treatment, cells were starved for 18–24 h in Modified Improved Minimum Essential Media (IMEM) containing 1% DCC. A 200-μL sterile pipette tip was used to create each scratch. After scratching was performed, cells were maintained in 37 °C, 5% CO₂ with their respective treatments (see legends). Three independent wells per group were imaged by taking three images of each well at × 10 magnification. ImageJ was used for quantification of the area of the wound, and three biological replicates were used per cell line. Data are shown as a representative of three experimental replicates.

Proliferation of wt-GR and S134A-GR MDA-MB-231 cells in basal medium was measured via MTT assays ((3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide). Using 24-well plates, 2.5 × 10⁴ cells/well were plated. At days 0, 1, 3, and 7, cell proliferation was determined. Sixty microliters of MTT was added per well for a final concentration of 5 mg/mL. Plate was incubated at 37 °C for 3 h. At this point, the medium was removed and solubilization solution (90% v/v dimethyl sulfoxide (DMSO)/PBS) was added to lyse the cells. Absorbance was measured in plate reader at 650 and 570 nm. The 650-nm measurements were subtracted from the 570-nm measurements. Sample means were normalized to day 0 and plot ± standard deviation (SD) using a linear plot.

Cells were plated, and for experiments that required hormone (Dex), growth factor (HGF), or cytokine (TGFβ1) treatment, cells were starved for 18–24 h in Modified Improved Minimum Essential Media (IMEM) containing 1% DCC. Whole-cell lysates were prepared using RIPA lite supplemented with 1 mmol/L PMSF, 5 mmol/L NaF, 0.05 mmol/L Na₃VO₄, 25 mmol/L beta glycerophosphate (BGP), 20 μg/mL aprotinin, 1 complete mini tablet of protease inhibitors (Roche), and 1 tablet of PhosSTOP (Roche). Electrophoresis in an SDS-PAGE gel was used to resolve and separate the proteins. Fifty micrograms of protein per lane was loaded. After electrophoresis, proteins were transferred to polyvinylidene difluoride membranes and probed with the following antibodies: pGR (1:750 - custom made, Pierce Biotechnology), GR (1:1000 - Santa Cruz Biotechnology, sc-1003), p-p38 (1:1000 - Cell Signaling Technology, 4511p), p38 (1:1000 - Cell Signaling Technology, 9212), YWHAZ (14-3-3ζ – 1:3000 – Millipore, AB9746), MAP3K5/ASK1 (1:1000 – Santa Cruz Biotechnology, sc5294), pERK1/2 (1:1000 – Cell Signaling Technology, 9101S), ERK1/2 (1:1000 – Cell Signaling Technology, 9102S), pJNK(1:1000 – Cell Signaling Technology, 9251S), and JNK (1:1000 – Cell Signaling Technology, 9252S) in 1% milk. For densitometric calculations, Fiji software version 2.0 was used to quantify bands representing total and phosphorylated proteins (GR, pS134-GR, p38, p-p38). All samples were normalized to respective controls and fold-change calculated relative to the vehicle condition. Densitometric data were collected for at least two to three independent experiments, and relative values were plotted as the mean ± SEM; the one-way ANOVA post hoc Fisher test was utilized to assess statistical significance.

Cell migration was measured using 8-μm transwell Boyden Chamber inserts (Corning). Cells were pretreated as indicated by the figure legends. After pretreatment, cells were trypsinized and plated at a density of 5 × 10⁴ cells in the upper chamber of the transwell system in IMEM media. Chemoattractant agents that were added to the lower-well are indicated in the figure legends in addition to 1% DCC. Cells were incubated for the time indicated in figure legends and maintained in 37 °C, 5% CO₂. Cellular invasion assays were done in a similar fashion with the exception that Matrigel (Corning) transwell inserts were utilized; Matrigel inserts were placed for at least 2 h in the incubator before plating the cells. Transwell inserts were imaged using a light microscope. Four pictures per transwell were taken at × 10 magnification. Three biological replicates were used in each condition and data are shown as representative of three experimental replicates. Results are reported as the mean percentage of migration for each well relative to wt-GR ± SD.

Six-well plates were prepared, and a base agarose layer (Invitrogen) was created in the surface of the wells. Cells were trypsinized and counted, and 4 × 10⁴ cells were plated in the top agarose layer in triplicate per treatment condition (indicated in figure legends). Plates were kept at 4 °C for 10 min to allow the agar to solidify. Incubation of plates occurred at 37 °C and 5% CO₂ for 15 days. To quantify colonies, 1 mL of PBS containing 4% formaldehyde and 0.005% crystal violet was added to each well. Cells were incubated for 1 h at room temperature. Four images were taken in each well. Results are reported as colonies counted per field and using mean ± SD.

A single-cell solution was obtained after enzymatic dissociation in 0.25% trypsin-EDTA and strained through a 40-μm sieve (BD Falcon). Cells were plated in ultra-low attachment plates (Corning) at 1 × 10³ cells per well and grown in a serum-free DMEM/F12 phenol-free medium (Corning) containing 1% methylcellulose (Sigma Aldrich), 1% B27 proprietary supplement (Invitrogen), 1% penicillin-streptomycin, 20 ng/ml EGF (Sigma Aldrich), 20 ng/ml basic-FGF (Gibco), and 10 μg/ml heparin (Sigma Aldrich). After 5–7 days, generate secondary tumorspheres were generated; primary spheres were collected and dissociated enzymatically in 0.25% trypsin-EDTA. Single cells were plated as described in conditioned media, which consisted of a 1:1 mixture of DMEM/F12 tumorsphere media (as above) and media from cultured parental cells. The tumorspheres were allowed to grow for 5–7 days before manual counting. Data are presented as the average ± SD of two independent measurements.

Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) data (via cBioPortal – http://​www.​cbioportal.​org/​) were obtained for patients who had the three-gene classifier and microarray data available for MAP3K5 and 14-3-3ζ [31‐33]. After that, we log2 transformed all of the mRNA expression values. Boxplot was utilized to plot the data and compare the expression across groups. Statistical significance was calculated based on one-way ANOVA and Tukey post hoc. All data associated to this analysis are in Supplementary File 1 and 2.

MDA-MB-231 expressing either wt-GR or S134A-GR were serum starved in IMEM containing 1% DCC for 18 h. Cells were treated with vehicle treatment or either 10 ng/mL of TGFβ1 for 6 h. Total RNA was extracted using the Qiagen RNAeasy kit. RNA quality and concentration were assessed by nanodrop. A TruSeq RNA kit was utilized to prepare and generate the mRNA libraries that were sequenced on the Illumina HiSeq2500 in a 75-base paired-end mode. Forty million reads were sequenced on average per sample.

Quality of raw RNA-seq sequences for each sample was assessed with FastQC. Trimmommatic was used to remove low-quality bases and trim adapters. FastQC was used again on each filtered FASTQ files to generate a sequence quality plot. Alignment to the hg38 genome was performed using HISAT2 [34]. Abundance of transcript was estimated with the featureCounts program in the SubRead package. DESeq2 (Version 1.22.2) was used to evaluate differential expression across groups and generate principal component analysis (PCA) plots [35]. For analyses of data using the generalized linear model framework, we utilized the EdgeR package (v3.26.8). Generation of heatmaps was performed with the pheatmap package (Version 1.0.12) using log2 normalized read counts from DESeq2 analysis. The package that was used to perform the gene set enrichment analysis was fgsea (Version 1.8.0). Ingenuity Pathway Analysis was also used to evaluate pathway significance. To do this, we uploaded the differential expression gene analysis obtained from the DESeq2 and EdgeR (Generalize Linear Model) analysis. All packages were used in the R environment (Version 3.5.2) and through R Studio (Version 1.1.383). In general, the cutoff for differential expression was an absolute log2 fold-change of 1.5 and a Benjamini-Hochberg p-adjusted value of 0.05. Figure legends include the criteria for gene selection.

Quantitative real-time PCR (qRT-PCR) experiments were conducted as previously described [36]. The cDNA was generated from total RNA extracted from MDA-MB-231 cells expressing either wt-GR or S134A-GR. Cells were plated at 5 × 10⁴ cells/well in six-well plates. After their respective treatments, media was removed, and cells were washed twice with cold 1× PBS. RNA was isolated with trizol, as per the manufacturer’s protocol. The cycling conditions by qPCR were as follows: 10 min of initial denaturation at 95 °C, 10 s denaturation at 95 °C, 10 s at 60 °C for annealing of primers and extension at 72 °C for 5 s for 45 cycles. Relative target gene expression was normalized to the expression of internal control genes, either TATA-binding protein (TBP), Actin (beta-actin), or 18S rRNA and they are shown as the mean value of three biological replicates ± SD. The primers used were MAP3K5-F 5′-AGGTGGTACTCTTTGGTTTTCAAG-3′ and MAP3K5-R 5′-GATACTGTCTAAGGCAAACATCCAG-3′; LEFTY2-F 5′-CTGGACCTCAGGGACTATGG-3′ and LEFTY2-R 5′-TCCCCTGCAGGTCAATGTAC3’; PIK3IP1-F 5′-CCTGGTGCTACGTCAGTGG-3′ and PIK3IP1-R 5′-TCCTGGATTTCTGTCGTGAAG-3′.

RNA-seq data available as RSEM was downloaded for patients with negative status for ER, PR, and HER2 for the TCGA breast cancer provisional dataset using cBioPortal [31, 32]. All expression data was log2 transformed for both genes. PRISM (GraphPad) was employed to plot each value and to execute the Pearson analysis of correlation. All data associated to this analysis are in Supplementary File 3.

ChIP assays were performed as per the manufacturer’s instruction (ChIP-IT express – Active Motif). Following respective treatments as indicated in figure legends, 37% formaldehyde was used for fixation (5 min) and sonication was utilized to shear chromatin for 30 min. Lysates were immunoprecipitated for 4 h with the following antibodies: 2 μL of GR (sc-1003) and equal amount of rabbit IgG. QPCR was used to analyze the resulting DNA and the data is represented as a percentage of input DNA. The primers used were LEFTY2 ChIP F 5′-CCCTCTAGTGGTTACAGGAAGACTC-3′ and LEFTY2 ChIP R 5′-AAAATCTGAGAGCAACTGAAGTGAG-3′; PIK3IP1 CHIP F 5′-GTACAAGTGCCCTGATAGGATTG-3′ and PIK3IP1 CHIP R 5′-CACTTCCCAGAACTGTTTTCAAC-3′. These primers were designed based on previous ChIP-Seq studies that focus on bindings sites of the GR [37, 38].

The METABRIC microarray expression data and survival data were downloaded for the following genes via cBioPortal – http://​www.​cbioportal.​org/​: YWHAZ(14-3-3ζ), DLEU7, JUNB, LBH, SNAI1, C1orf106, CCL20, NRP2, PIK3IP1, SYT8, KRT16, NLRC3, LEFTY1, NKD1, KPNA7, LTBP3, MDFI, SYN1, LEFTY2, MYOZ1, GPR183, MATK, STK19, HRAS, and COL8A2 as available for subjects. First, we log2 transformed all of the mRNA expression values. In the case of the survival analysis for 14-3-3ζ (Fig. 4B), we classified patients into upper 50th percentile and bottom 50th percentile based on the median cutoff for the log2 mRNA expression of 14-3-3ζ across all patients. All data associated to this analysis are in Supplementary File 4. For our gene signature, R programming was used to calculate the average gene expression of all genes from each patient and stratify groups based on the median of the average expression into upper 50th percentile or bottom 50th percentile. Kaplan-Meier plots and logrank test was used to assess the differences in overall survival between the cohorts using PRISM (GraphPad). All data associated to this analysis are in Supplementary File 5.

SurvExpress was used to analyze survival data for the TCGA cohort. The 24 TGFβ1-induced pS134-GR-dependent gene signature (DLEU7, JUNB, LBH, SNAI1, C1orf106, CCL20, NRP2, PIK3IP1, SYT8, KRT16, NLRC3, LEFTY1, NKD1, KPNA7, LTBP3, MDFI, SYN1, LEFTY2, MYOZ1, GPR183, MATK, STK19, HRAS, and COL8A2) was utilized for the SurvExpress analysis. Data was downloaded and analyzed locally in R. PRISM (GraphPad) was used to generate the Kaplan-Meier plots. Logrank test was used to assess the differences in survival between expression cohorts. All data associated to this analysis is in Supplementary File 6.

Cells were lysed in ELB lysis buffer containing the following: 50 mmol/L HEPES, 0.1% nonidet P-40 (NP-40), 250 mmol/L NaCl, 5 mmol/L EDTA, 1× complete protease inhibitors (Roche), 1× PhosSTOP (Roche), 1 mmol/L PMSF, 1 mmol/L NaF, 0.5 mmol/L Na₃PO₄, 25 mmol/L BGP, and 20 μg/mL aprotinin]. 1000 μg of lysate was incubated with 1 μg of specific antibodies overnight at 4 °C in a rotator. Protein G agarose was used to isolate the complexes for 1 h at 4 °C. SDS-PAGE and western blot analysis were used to analyze the immunocomplexes.

All statistic information is disclosed in the legends. In general, for experiments that involve two groups, one-paired Student’s t-test was performed. For experiments that involved more than two groups, one-way or two-way ANOVA were utilized. Post-hoc tests within the vehicle or control treatment was performed using Dunnett’s statistical testing. However, if comparison within different groups was required, Tukey’s post-hoc test was utilized. For densitometric analysis, one-way ANOVA and Fisher Least Significant Difference (LSD) post hoc. Statistical tests were performed using either PRISM v7/v8 and R v3.5.

Previous studies demonstrated that GR activation has anti-migratory effects [39]. Conversely, West and colleagues recently reported that GR induces the expression of genes that are essential for migration and invasion [12, 40]. Similar to other steroid hormone receptors, signaling context and the hormonal milieu (i.e., time and concentration of exposure to either agonist or antagonist) may dictate distinct receptor actions in the same biological system [41]. We tested whether GR differentially responds to acute and chronic effects of Dex by exposing GR+ MDA-MB-231 TNBC cells to a range of pharmacologically relevant Dex concentrations at two different time points: 15 min (“acute”) and 6 h (“chronic”) (Fig. 1a). Our data demonstrate that when cells are pretreated with Dex in charcoal-stripped serum-containing media (i.e., steroid hormone free, but growth factor rich) for 15 min and then subjected to scratch wound assays, subsequent Dex-induced migration is inhibited (Fig. 1b). Conversely, when the same cells are instead pretreated with Dex for 6 h in charcoal-stripped serum-containing media, subsequent treatment with Dex stimulates migration (Fig. 1c). A biphasic response of acute vs. chronic Dex treatment was also observed in Hs578T cells (Supplementary Fig. 1A-C).

To explore signaling pathways that may be altered upon chronic (i.e., 6 h) Dex treatment in TNBC cells, we probed existing microarray data from the GSE113571, in which MDA-MB-231 cells were treated with 100 nM Dex for 4 h [12]. Ingenuity Pathway Analysis was used to determine which pathways were significantly upregulated or downregulated by liganded GR. Notably, the TGFβ1 pathway was among the most significantly upregulated (Fig. 1d). We thus evaluated the requirement for GR in TGFβ1-induced MDA-MB-231 cell migration. Cells were pretreated for 18 h with vehicle, RU486 (i.e., a GR antagonist) alone, Dex alone, TGFβ1 alone, or either agent (Dex or TGFβ1) in combination with RU486. Cells were then subjected to scratch wound assays in the presence of the respective treatments. As expected, we observed pro-migratory effects on cells that were chronically treated with 1 μM Dex (18 h). Cells that were treated for 18 h with 10 ng/mL of TGFβ1 were also more migratory relative to vehicle controls. Surprisingly, RU486 abrogated the pro-migratory effects of either Dex or TGFβ1 (Fig. 1e). Similar results were observed in Hs578T TNBC cells (Supplementary Fig. 1D). Our results suggest that TGFβ1-induced TNBC cell migration is mediated by GR. Notably, TGFβ1-induced cell migration occurred in steroid hormone-free conditions in the absence of exogenously added GR ligands.

Previously, we reported that cellular stress (i.e., H₂O_2, hypoxia, nutrient starvation) and stressful growth conditions such as growth in low attachment/suspension and chemotherapeutic agents (i.e., Taxol) increase phosphorylation of GR on Ser134, leading to the increased ligand-induced expression of selected GR target genes (e.g., BRK, HIF2, AhR) that are essential for advanced cancer phenotypes in TNBC [16, 17]. Additionally, we reported that pS134-GR species are elevated in TNBC relative to luminal breast cancer subtypes [16]. Since TGFβ1 expression is also elevated in TNBC and associated with poor outcome, we predicted tight linkage between TGFβ1 signaling and phosphorylation of GR Ser134 [42, 43]. MDA-MB-231 cells were treated with 10 ng/mL TGFβ1 for 30 min, 1 h, and 2 h; phosphorylation of GR Ser134 was induced by TGFβ1 at 1 h and 2 h. Additionally, p38 MAPK activity, as measured by active-site phosphorylation, was also elevated at these timepoints (Fig. 2a and see densitometric analyses of aggregated data from multiple repeats in Supplementary Fig. 2a). Dose-response curves demonstrated robust GR Ser134 phosphorylation over a wide range of TGFβ1 doses (0.1–10 ng/ml) (Fig. 2b and see densitometric analyses of aggregated data from multiple repeats in Supplementary Fig. 2B). We then tested the requirement for p38 MAPK in GR Ser134 phosphorylation in TNBC models; MDA-MB-231 cells were pretreated with either vehicle control (DMSO) or SB203580 (p38 inhibitor) for 30 min prior to TGFβ1 exposure for 1 h. While basal levels of p38 MAPK activity remained somewhat elevated, inhibition of p38 MAPK completely blocked TGFβ1 dependent phosphorylation of GR Ser134 (Fig. 2c and see densitometric analyses of aggregated data from multiple repeats in Supplementary Fig. 2C). The specificity of GR Ser134 phosphorylation by p38 MAPK was further explored using SB203580 and SB202190 (p38 inhibitors), LY294002 (Akt inhibitor), or UO-126 (MEK1/2 inhibitor). MDA-MB-231 and Hs578T TNBC cells were pretreated for 30 min with each kinase inhibitor prior to TGFβ1 treatment; TGFβ1-induced GR phosphorylation was exclusively blocked upon inhibition of p38 MAPK, while inhibition of either AKT or ERK1/2 was without effect (Supplementary Fig. 2D and 2E). Similar to MDA-MB-231 and Hs578T cells, TGFβ1 induced robust phosphorylation of GR Ser134 in a TNBC Patient-Derived Xenograft (PDX) model (HCI-10) (Fig. 2d).

In addition to TGFβ1, Hepatocyte Growth Factor (HGF) is a TME factor known to promote cell migration and metastasis [44]. HGF has also been shown to activate p38 MAPK [45]. Given the requirement of p38 MAPK for TNBC motility, we tested the ability of HGF to induce regulated GR phosphorylation [46]. Thus, MDA-MB-231 cells were treated with 50 ng/mL HGF for 0.5–48 h. Robust and persistent phosphorylation of GR Ser134 was observed after 1–2 h (Fig. 2e and Supplementary Fig. 2F). These findings suggest that the TME induces ligand-independent GR Ser134 phosphorylation, nominating GR as an important sensor not only of host/life stress in the form of elevated corticosteroid levels, but for both cellular and local (i.e., TME-derived) cytokine-mediated stress signaling in TNBC cells.

To test the requirement for GR Ser134 in TGFβ1-regulated TNBC cell migration, we employed a CRISPR/Cas9 approach to create MDA-MB-231 cells expressing either wt-GR (control) or a point mutant GR in which Ser134 has been changed to Alanine (S134A-GR clone #1 and clone #2). We confirmed expression of either endogenous WT or S134A GR via western blotting following treatment of cells with 10 ng/mL of TGFβ1. Predictably, we observed equal levels of total GR in all cell lines, but no signal representing GR Ser134 phosphorylation in TNBC cells harboring S134A-GR (Fig. 3a). Using MTT assays as a readout of cell viability and proliferation, we compared wt-GR and S134A-GR cell lines; cells harboring S134A GR remained viable and proliferated at similar rates relative to wt-GR+ controls (Day 1). However, CRISPR knock-in of S134A GR resulted in reduced cell numbers evident at days 3–5 relative to wt-GR+ controls (Fig. 3b). The migration of these models in response to 10 ng/mL TGFβ1 (18 h) was then measured using scratch wound assays. Fetal Bovine Serum (FBS) (10%) was included as a positive control that contains multiple soluble factors capable of stimulating cancer cell migration. Cells harboring either wt-GR or S134A-GR (clone #1) were pretreated with vehicle control, 10 ng/mL TGFβ1, or 10% FBS overnight; cells were then subjected to scratch wounds, re-treated with their respective treatments and allowed to migrate for 18 h. TGFβ1-mediated migration was significantly impaired in cells expressing S134A-GR relative to wt-GR (Fig. 3c). These findings were replicated (clone #1 and clone #2) in transwell migration assays (Fig. 3d). Migration was stimulated with either vehicle or 10 ng/mL TGFβ1 as the chemoattractant (i.e., added to the bottom chamber). Again, robust TGFβ1-induced migratory activity occurred in wt-GR cells cultured in steroid hormone-free media lacking exogenously added GR ligands (i.e., Dex). However, cells harboring S134A-GR (i.e., two independent CRISPR clones) were significantly less migratory (18 h) relative to cells harboring wt-GR (Fig. 3d). Similarly, we tested the invasive ability of wt-GR and S134A-GR clones. Cells were seeded and allowed to invade through Matrigel inserts towards either vehicle or 10 ng/mL TGFβ1 as the chemoattractant. Predictably, the invasive ability of both S134A GR clones was significantly impaired relative to cells expressing wt-GR (Supplementary Fig. 3).

Anchorage-independent growth in soft agar (i.e., an in vitro measure of cell transformation) is a hallmark of cancer cells that is indicative of the ability to grow and survive independently of basement membrane attachment. TNBC cells expressing S134A-GR exhibited impaired soft-agar colony formation relative to cells expressing wt-GR (Fig. 3e). TNBC cells harboring wt-GR, but not S134A-GR (two clones), formed colonies in media containing steroid hormone-free fetal bovine serum (DCC) with either vehicle control or GR agonists (e.g., Dex). However, while Dex or TGFβ1 did not appreciably increase basal soft-agar colony formation over that observed in DCC alone, the GR antagonist, RU486, blocked soft-agar colony formation, demonstrating the GR-dependence of this cancer cell biology in 3D conditions (Fig. 3e). Similar results were observed in tumorsphere assays in which TNBC cells were plated at limiting dilution in ultra-low attachment dishes as an in vitro readout of cancer stem-like cell behavior [47]; while addition of Dex to sphere media was without effect, MDA-MB-231 cells harboring S134A mutant GR failed to form tumorspheres relative to cells expressing endogenous wt-GR (Fig. 3f). Impaired cell migration as measured by chemotaxis in transwell migration assays (Fig. 3g) as well as attenuated tumorsphere formation were also observed using the previously described GR-null/low U2OS osteosarcoma cell line engineered to stably express S134A-GR relative to wt-GR (Fig. 3h) [15]. These data demonstrate that pS134-GR is critical for TGFβ1-mediated cell migration and basal growth in suspension/3D (anchorage-independent growth and tumorsphere formation), even in the absence of exogenously added GR agonists.

14-3-3 proteins are a conserved family of regulatory molecules that bind to diverse signaling proteins in eukaryotic cells. Galliher-Beckley et al., first reported that 14-3-3ζ interacts with GR in a pS134-GR-dependent manner in U2OS osteosarcoma cells [15]. To evaluate the role of 14-3-3ζ in GR+ TNBC, we probed 14-3-3ζ mRNA expression levels across breast cancer subtypes in the TCGA dataset and stratified patients based on the expression levels of ER, PR and HER2 receptors as determined by clinical immunohistochemistry analyses of primary tumors. Data from METABRIC were obtained from cBioPortal. The expression of 14-3-3ζ mRNA in patients with ER−/HER2- (TNBC) or Her2+ breast cancer was higher when compared to patients with ER+/HER2- breast cancer (Fig. 4a). Similarly, high levels of 14-3-3ζ mRNA, as separated by the median cutoff of mRNA expression, correlate with poor overall survival in patients within the METABRIC cohort (Fig. 4b). Accordingly, we observed abundant expression of 14-3-3ζ in several breast cancer cell lines, including MDA-MB-231, Hs578T, and MDA-MB-468 (TNBC) cells as well as U2OS osteosarcoma cells (positive control). Weak 14-3-3ζ expression was observed in non-tumorigenic MCF10A cells (Fig. 4c). To test the requirement for 14-3-3ζ in TGFβ1-mediated migration, we generated MDA-MB-231 cells stably expressing two distinct 14-3-3ζ-targeted shRNAs or a non-targeting vector control (shcontrol); 14-3-3ζ knockdown was confirmed in two separate pools by Western blotting (Fig. 4d; upper panel). TGFβ1 induced robust cell migration in shcontrol but not sh14-3-3ζ TNBC models (Fig. 4d). We confirmed that 14-3-3ζ/GR interaction occurs in TNBC models (Fig. 4e); MDA-MB-231 cells were subjected to treatment with either Veh, TGFβ1, Dex, TGFβ1+Dex, or H₂O₂ (a potent inducer of GR Ser134 phosphorylation). Similar to that observed in H₂O₂-treated U2OS osteosarcoma cells [15], co-immunoprecipitation assays revealed modest TGFβ1 or Dex-regulated interaction between GR and 14-3-3ζ that was further enhanced following combinatorial treatment of cells with both TGFβ1 and Dex. These data suggest that 14-3-3ζ and pS134-GR function in the same signaling pathway and/or cooperate to induce altered gene expression and subsequent migration of TGFβ1-responsive TNBC cells.

Our data (Figs. 1, 2, 3, and 4) suggest that significant ligand-independent gene expression occurs via phosphorylation of GR Ser134. To further expand our understanding of the impact of pS134-GR on TNBC cell behavior, we performed RNA-seq studies in MDA-MB-231 cells expressing either wt-GR or S134A-GR (clone #1). Cells were treated with either vehicle, to capture basal homeostatic (i.e., ligand-independent) GR actions or with TGFβ1, as a regulated input to p38 MAPK activation and subsequent GR phosphorylation (6 h). Principal component analysis demonstrated intrinsic differences between both vehicle-treated and TGFβ1-treated models (Fig. 5a). Remarkably, in the absence of a specific stimulus, the basal transcriptome of cells expressing S134A GR differs dramatically from that of cells expressing wt-GR. These data suggest that GR Ser134 acts as a homeostatic “sensor” of numerous cell-intrinsic and cell-extrinsic signaling inputs to p38 MAPK, accounting for significant ligand-independent or “basal” differences between vehicle-treated models. Volcano plots were used to further illustrate the effect of S134A-GR single point mutation in vehicle or TGFβ1-treated cells. Namely, vehicle-treated models clearly exhibited a broad array of transcription-wide differences, while fewer changes in mRNA expression were attributed to TGFβ1 treatment between wt and S134A GR+ models, as measured using the generalized linear model framework (GLM) in EdgeR (Fig. 5b). GLM analysis identified a limited number of differentially expressed genes in response to TGFβ1 when comparing wt-GR+ and S134A-GR+ models. Notably, LEFTY1, COL1A1, and OLFM2, known mediators of TGFβ1 signaling, were identified as differentially expressed (red dots) using the GLM approach (Supplementary Table 1A) [48, 49]. Ingenuity Pathway Analysis (IPA) of this limited gene subset identified ERBB2 as well as TGFBR1/TGFβ1 (i.e., both receptor and cognate ligand) signaling as potential S134A GR-regulated pathways (Supplementary Table 1B). Consistent with these results, comparison of the response to TGFβ1 between wt and S134A GR models also revealed a subset of ~ 80 TGFβ1-induced genes in cells harboring wt-GR that were not regulated by TGFβ1 in cells expressing S134A-GR (Fig. 5c green rectangle and Supplementary Fig. 4A). Importantly, ligand-dependent GR activation (i.e., in Dex-treated cells) remained similar across these models as measured by qPCR analysis of two well-characterized GR target gene (DUSP1 and SGK1) mRNAs, indicating that ligand-bound S134A-GR is not transcriptionally impaired (Fig. 5d). Further studies are needed to evaluate the role of Ser134 in the global transcriptional response of liganded GR.

Ingenuity Pathway Analysis and Gene Set Enrichment Analysis (GSEA) tools were then used to evaluate the impact of GR Ser134 modification on ligand-independent basal (Veh) and TGFβ1-induced cellular pathways/behaviors. IPA analysis revealed that cellular migration is strongly represented in the genes that are primarily regulated in cells expressing wt-GR, but not in cells expressing S134A GR (i.e., gene sets associated with cell migration are decreased in S134A-GR TNBC cells regardless of treatment; Fig. 5e and Supplementary Fig. 4B). This finding further strengthens our hypothesis that phosphorylation of GR on Ser134 is critical for migration of TNBC cells. Interestingly, the TGFβ1 pathway itself is significantly downregulated in S134A-GR cells, and consistent with these results, this is also readily apparent in TGFβ1-treated cells, as was independently identified via both the upstream regulator analysis in the IPA platform and METACORE (Fig. 5f and Supplementary Fig. 4C). In sum, the transcriptomes of cells harboring either wt-GR or S134A-GR are remarkably different basally, perhaps because GR Ser134 acts as an important ligand-independent sensor of multiple homeostatic intrinsic (basal) and extrinsic (TME-derived factors) cellular stress inputs to the p38 MAPK pathway [15, 16].

Using GSEA analysis for the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways study, we identified the MAPK pathway as significantly downregulated in cells harboring S134A-GR (Fig. 6a and Supplementary Table 2). Interestingly, we found that there are also basal differences in the MAPK KEGG module (Fig. 6a). This indicates that one or more key components of the MAPK pathway are likely to be regulated by GR Ser134 phosphorylation. Indeed, TGFβ1-induced activation of p38 MAPK, as measured by via active-site phosphorylation, appears to be arrested in cells expressing S134A GR relative to cells expressing wt-GR (Fig. 6b and see densitometric analyses of aggregated data from multiple repeats in Supplementary Fig. 5A). Similarly, 14-3-3ζ knockdown resulted in greatly decreased TGFβ1-induced p38 activation with no change in total p38 expression levels (Fig. 6c and see densitometric analyses of aggregated data from multiple repeats in Supplementary Fig. 5B). Consequently, sh14-3-3ζ cells also exhibited less pS134-GR. To confirm these findings, we tested for phosphorylation of p38 MAPK in response to TGFβ1 and found that it is greatly attenuated in both S134A GR+ clones (Fig. 6d). These data suggest that pS134-GR is an important regulator of one or more kinases upstream of p38 MAPK (i.e., that are required for p38 activation). We thus probed our RNA-seq data within the MAPK pathway KEGG signature, thereby nominating the mitogen-activated protein kinase kinase kinase 5 (MAP3K5, also called MEKK5 or ASK1) as a possible factor that integrates these pathways. MAP3K5 interacts with 14-3-3ζ to facilitate activation of downstream kinases MEK3/6 and p38 MAPK [50]. Indeed, MAPK3K5 protein levels were significantly downregulated in two clones harboring S134A-GR relative to wt-GR (Fig. 6e). Moreover, in an independent experiment using S134A-GR clone #1, we observed downregulation of MAP3K5 at both the mRNA and protein levels (Supplementary Fig. 5C). Although known to be associated with other cancer types, to our knowledge, MAP3K5 has not been implicated in breast cancer. Similar to expression of pS134-GR and 14-3-3ζ (above), MAP3K5 mRNA levels are elevated in TNBC relative to other breast cancer subtypes represented in the METABRIC (Supplementary Fig. 5D), and MAP3K5 and GR mRNA expression levels are strongly correlated in TNBC patients (Fig. 6f) [16]. Notably, inhibition of MAP3K5 with the selective inhibitor, selonsertib, blocked phosphorylation of GR (Fig. 6g) and inhibited TGFβ1-induced migration (Fig. 6h) in both MDA-MB-231 and Hs578T cells. Similarly, inhibition of p38 MAPK using SB203580, halted TGFβ1-mediated migration in both TNBC models (Fig. 6h). Taken together, these results indicate that phosphorylation of GR on Ser134 is critical for the expression of key MAPK pathway intermediates. Specificity controls indicated that serum-induced activation of JNK and p42/p44 MAPKs (ERK1/2) were unaffected in TNBC cells harboring S134A GR (Supplementary Fig. 5E), suggesting that pS134-GR is a highly selective regulator of the p38 MAPK “module” characterized by the three kinase cascade initiated by MAP3K5 in TNBC cells. While the TGFβ1-induced activation of SMADs is thought to occur independently of p38 MAPK, we observed a slight attenuation of TGFβ1-induced SMAD2 phosphorylation (i.e., primarily apparent after 2 h) in cells harboring S134A-GR (Supplementary Fig. 5F). These data suggest that pS134-GR may potentiate sustained SMAD2 phosphorylation and activation, perhaps in part via regulation of the signaling components required for intact p38 MAPK signaling (Fig. 6). We were unable to detect a physical interaction between GR and SMADs (not shown); further studies are needed to address weather GR crosstalk with the TGFβ1 signaling occurs at the level of SMAD regulation.

We speculated that pS134-GR-regulated gene sets important for TGFβ1 signaling may predict prognosis and disease progression in patients with breast cancer. Using our transcriptome (RNA-seq) data, we further evaluated all genes that were uniquely upregulated in cells expressing wt-GR but not S134A GR (i.e., those genes whose TGFβ1-induced regulation requires an intact GR Ser134). Thus, we compared the expression of genes between the vehicle and TGFβ1-treated groups for each TNBC cell line. Using a cutoff of log2 fold-change of + 1.5 and a Benjamini-Hochberg p value of 0.05, we identified all TGFβ1-upregulated genes relative to vehicle controls. A total of 90 genes were upregulated by TGFβ1 in MDA-MB-231 cells expressing either wt-GR or S134A GR. Of these TGFβ1-induced genes, 38 were upregulated only in cells harboring wt-GR, and 8 were significantly upregulated only in cells expressing S134A-GR, while 44 genes were upregulated in both cell lines (Fig. 7a). A heatmap approach was used to plot the log2 normalized expression values of the 38 genes upregulated by TGFβ1 in cells expressing wt-GR but not S134A-GR (Fig. 7b and Supplementary Table 4). We further identified a 24-gene cluster of TGFβ1-regulated genes that require GR Ser134 using the same threshold of log2 fold-change of > 1.5 and a Benjamini-Hochberg p value of 0.05 (Supplementary Table 3 and 4). Representative genes (LEFTY2, PIK3IP1) were further validated by qPCR in TGFβ1-treated (10 ng/mL; 6 h) cells. As predicted by our RNA-seq data, we observed TGFβ1-induced upregulation of both LEFTY2 (~ 4.3 fold increase) and PIK3IP1 (~ 200 fold increase) in cells expressing wt-GR, but not in cells expressing S134A-GR (Fig. 7c). Additionally, using ChIP assays, we demonstrated robust recruitment of wt-GR species to glucocorticoid response element (GRE)-containing promoter regions of both LEFTY2 and PIK3IP1 in response to TGFβ1 alone. However, consistent with our qPCR results, recruitment of S134A GR to these regions was significantly diminished (LEFTY2) or failed to occur (PIK3IP1) in CRISPR models expressing phospho-mutant GR (Fig. 7d). These data indicate that in the absence of ligand, pS134-GR is recruited to both LEFTY2 and PIK3IP1 in response to TGFβ1. Further studies are needed to evaluate the impact of pS134-GR on the global cistrome in the context of TNBC.

To further evaluate the importance of the pS134-GR gene signature in breast cancer patients, we used the METABRIC dataset to calculate the average expression of the above defined 24 pS134-GR-induced genes for each patient tumor and stratified patient populations based on a median cutoff for the average expression of the gene signature. Using Kaplan-Meier curves for analyses of overall survival, we observed that patients whose breast tumors exhibited an average expression of the gene signature within the upper 50th expression percentile experienced poor survival relative to patients whose tumors fell within the lower 50th percentile (Fig. 7e). This difference was significant with a log rank p value of 0.0008 (Fig. 7e). To confirm our findings in the METABRIC dataset, we employed the SurvExpress tool with the TCGA breast cancer dataset [51] and observed similar results with significant separation based on overall survival (Fig. 7f). These results suggest that the presence of pS134-GR, as measured using a 24-gene signature, confers poor prognosis in breast cancer patients, regardless of subtype. More studies are needed to further explore the utility of tracking either pS143-GR or pS134-GR gene signatures as potential biomarkers of advanced cancer behaviors (i.e., p38 MAPK-mediated tumor cell dissemination) in breast cancer patient subpopulations.