A JUN N-terminal kinase inhibitor induces ectodomain shedding of the cancer-associated membrane protease Prss14/epithin via protein kinase CβII

Serine protease 14 (Prss14)/epithin is a transmembrane serine protease that plays essential roles in tumor progression and metastasis and therefore is a promising target for managing cancer. Prss14/epithin shedding may underlie its activity in cancer and worsen outcomes; accordingly, a detailed understanding of the molecular mechanisms in Prss14/epithin shedding may inform the design of future cancer therapies. On the basis of our previous observation that an activator of PKC, phorbol 12-myristate 13-acetate (PMA), induces Prss14/epithin shedding, here we further investigated the intracellular signaling pathway involved in this process. While using mitogen-activated protein kinase inhibitors to investigate possible effectors of downstream PKC signaling, we unexpectedly found that an inhibitor of c-Jun N-terminal kinase (JNK), SP600125, induces Prss14/epithin shedding even in the absence of PMA. SP600125-induced shedding, like that stimulated by PMA, was mediated by tumor necrosis factor-α–converting enzyme. In contrast, a JNK activator, anisomycin, partially abolished the effects of SP600125 on Prss14/epithin shedding. Moreover, the results from loss-of-function experiments with specific inhibitors, short hairpin RNA–mediated knockdown, and overexpression of dominant-negative PKCβII variants indicated that PKCβII is a major player in JNK inhibition– and PMA-mediated Prss14/epithin shedding. SP600125 increased phosphorylation of PKCβII and tumor necrosis factor-α–converting enzyme and induced their translocation into the plasma membrane. Finally, in vitro cell invasion experiments and bioinformatics analysis of data in The Cancer Genome Atlas breast cancer database revealed that JNK and PKCβII are important for Prss14/epithin-mediated cancer progression. These results provide important information regarding strategies against tumor metastasis.

Prss14/epithin (also known as matriptase, suppression of tumorigenicity 14, membrane-type protease-1), a typical member of the type II transmembrane serine protease family, plays important roles in cancer progression and metastasis (1,2). Overexpression of Prss14/epithin is found in various epithelial cancer types (2). Particularly, Estrogen Receptor (ER) 4 breast cancer patients with a poor prognosis express higher levels of Prss14/epithin (3). In our earlier report involving pathological examinations, we showed that the prognosis of post-surgery esophageal cancer patients with higher Prss14/epithin expression is very poor (4). Transgenic expression of Prss14/epithin in the mouse causes spontaneous squamous cell carcinomas (5). In contrast, reduced Prss14/epithin expression in a transgenic mouse model clearly impairs tumor progression and metastasis (6). Moreover, down-regulated Prss14/epithin inhibits ErbB2induced prostate cancer cell invasion (7), and specific inhibitors of Prss14/epithin protease activity impair tumor growth and metastasis (8). We have reported previously that knockdown of Prss14/epithin reduces cell migration (9,10) and the metastatic ability of breast cancer cells such as 4T1 (9) and EO771 (11). When a specific mAb raised against the activation loop portion of Prss14/epithin is injected into a PyMT breast cancer mouse model, metastasis is reduced significantly (11).
Prss14/epithin has a multidomain structure with a short cytoplasmic N terminus, a single transmembrane region, a sperm protein, enterokinase, and agrin domain (SEA), two complement subcomponent C1r/C1s domains (CUB), four low-density lipoprotein (LDLRA) receptor class A repeats, and a C-terminal serine protease domain (12,13) and undergoes multiple processing steps before it is fully mature. Its full-length form with a molecular mass of 110-kDa converts to a 92-kDa form after being processed at Gly-149, and this conversion is essential for ectodomain shedding of this protease (14). Previously, we showed that phorbol 12-myristate 13-acetate (PMA) and TGF␤ can induce ectodomain shedding of Prss14/epithin (15,16) by inducing actin cytoskeletal remodeling, which results in filamin-dependent translocation of the protease to the cell-cell contact (17). Filamin appears to be essential for the translocation and shedding of Prss14/epithin (17). Prss14/epithin shedding events induced by PMA or TGF␤ occur mainly through the action of tumor necrosis factor-␣-converting enzyme (TACE) (15,16). Although it is still debated whether shedding occurs prior to activation, in the case of TGF␤-induced shedding, shedding precedes activation of the protease (10,16) suggesting that shedding is possibly required for activation of the protease. Because PMA, a diacylglycerol mimic, acts as activator of the PKC family, and some PKC isoforms have already been recognized for being involved in ectodomain shedding of other transmembrane proteins (18,19), there is no doubt regarding the involvement of PKCs in ectodomain shedding of Prss14/epithin. However, it is still a mystery which PKC isoforms and other intracellular signaling pathways are involved in shedding of Prss14/epithin. Considering that ectodomain shedding of the protease can profoundly affect the extracellular environment in favor of cancer metastasis (20 -22), identification of such signaling pathways will be useful for establishing therapeutic approaches. Moreover, because of the broad biological effects of PKC family members, pinpointing the PKC isoforms responsible for shedding may also be critical for specific targeting of Prss14/ epithin ectodomain shedding.
In this study, we show that PKC␤II is a key molecule required for shedding of Prss14/epithin induced by PMA and a JNK inhibitor. PMA treatment and JNK inhibition can increase phosphorylation and translocation of PKC␤II to the plasma membrane, which could be the essential step for ectodomain shedding of Prss14/epithin. Indeed, inhibition of PKC␤II and JNK reduces in vitro cell invasion. Finally, bioinformatics analysis revealed that the levels of signaling molecules are correlated with better or worse patient survival. Thus, our finding can provide important information about new therapeutic approaches for cancer patients with high expression of Prss14/epithin.

JNK inhibition increases Prss14/epithin shedding
To investigate signaling pathways involved in PMA-induced Prss14/epithin ectodomain shedding, we first sought to test three main MAPK pathways (extracellular signal-regulated kinase, p38, and JNK) (23) by employing commonly used specific inhibitors in the absence or presence of PMA in 427.1.86 cells. As seen in Figs. 1, A and B (see Fig. S1 for the full-size blot), in the absence of any pathway-specific inhibitors, PMA slightly increased the shed form of Prss14/epithin (Epi-S') in the conditioned medium but decreased the amount of protein (Epi-S) remaining in the cell lysate. When three inhibitors were used (PD98059 for extracellular signal-regulated kinase, SB203580 for p38, and SP600125 for JNK), SP600125 significantly increased the levels of Epi-S' regardless of the presence of PMA (Fig. 1B). The effects of SP600125 on Epi-S' appearance in the medium (and disappearance of Epi-S in the cell lysate) were in a dose-and time-dependent fashion (Fig. 1C).
Because we already know that PMA-induced Prss14/epithin shedding is mediated by TACE (15), we tested whether SP600125-induced shedding is also mediated by TACE (Fig. 1,  D and E). Use of a TACE-specific inhibitor and knocking down of TACE expression significantly reduced the amount of Epi-S' in conditioned medium while retaining the amount of cellular Epi-S. In the experiments with 427.1.86 cell line, we rarely observed the smaller activated form in the same blot (data not shown).
We also tested SP600125-induced shedding in other cell types: T47D human breast cancer and 4T1 mouse breast cancer cells. T47D cells showed that SP600125 induced Epi-S' and aEpi-S (30-kDa activated form) in a dose-dependent manner (Fig. 1F). The time-and dose-dependent increases in Epi-S' and aEpi-S in 4T1 cells are shown in Fig. 1G. The T47D and 4T1 breast cancer cell lines clearly revealed the active form, aEpi-S (with enhanced exposure of the same blot) following the appearance of Epi-S, suggesting that shedding is a general phenomenon and may precede the activation event.
To investigate whether SP600125-induced Prss14/epithin shedding is indeed due to its inhibitory effect on JNK activity, the downstream effects of SP600125 were tested (Fig. 2). SP600125 abolished phosphorylation of p-c-Jun, a target of JNK. It is generally agreed that anisomycin is a JNK activator involved in protein stability and/or transcription (24). To confirm that the effect of SP600125 is specifically mediated by JNK inhibition and not by possible off-target effects of the inhibitor, we attempted to suppress the inhibitory effect by pretreatment with anisomycin and test the effects of SP600125. When anisomycin was used together with SP600125, shedding of Prss14/ epithin and phosphorylation of c-Jun were partially affected, suggesting that anisomycin can interfere with SP600125-induced shedding ( Fig. 2A). In addition, when expression of JNK1 and JNK2 was reduced by specific siRNAs (reductions relative to control samples with nontargeting siRNA were 63.9% for JNK1 and 48.4% for JNK2), ectodomain shedding of Prss14/ epithin to the medium was increased 3.0-fold for JNK1 knockdown and 4.8-fold for JNK2 knockdown) (Fig. 2B). These results indicate that JNK activity inhibits Prss14/epithin shedding.
To investigate the mechanisms of SP600125-induced Prss14/ epithin shedding in more detail, we applied cycloheximide, which is generally considered an inhibitor of protein synthesis. As seen in Fig. 2C, cycloheximide abolished SP600125-induced Prss14/epithin shedding. This strongly suggests that Prss14/ epithin shedding requires de novo synthesis of labile protein(s). When new transcription was interfered with by actinomycin D or ␣-amanitin pretreatment, SP600125-induced Prss14/epithin shedding was slightly reduced but severely affected by treatment with both reagents together, suggesting that at least some new transcription is required (Fig. 2D).
Previously, we showed that actin rearrangement induced by PMA is essential for translocation and shedding of Prss14/epithin (17). SP600125, similar to PMA, induced actin to rearrange to form cortical actin filaments. This rearrangement was abolished by additional anisomycin treatment (Fig. S2).

PKC␤II is involved in PMA-and SP600125-induced Prss14/epithin shedding
We tried to identify the PKC isoforms involved in PMAand/or SP600125-induced Prss14/epithin shedding by using PKC inhibitors, siRNAs, and dominant-negative (DN) forms PKC␤II in Prss14/epithin shedding induces cell invasion (Fig. 3). Treatment with a broad-spectrum PKC inhibitor, Go6976, and a PKC␤ selective inhibitor (␤i) significantly inhibited PMA-and SP600125-induced shedding of Prss14/epithin in 427.1.86 cells (Fig. 3A). Both inhibitors decreased Epi-S' in the medium, whereas Epi-S levels in the cell lysate were similar to those of Epi-S'. Knockdown studies using PKC␣ and ␤ siRNAs were then carried out (Fig. 3B). Only PKC␤, but not PKC␣ siRNA, affected SP600125-and PMA-induced shedding. Because PKC␤I and PKC␤II use the same message, the levels of PKC␤I and PKC␤II proteins were decreased when cells were treated with PKC␤ siRNA. PMA treatment also reduced PKC levels, which is known as PMA-induced down-regulation following activation (25). However, treatment with 5 M SP600125 for 2 h did not severely affect the expression levels of these PKC isoforms. From these results, it was concluded that SP600125-and PMA-induced Prss14/epithin shedding was pri-marily mediated by PKC␤, not by PKC␣ (Fig. 3B). Next, DN forms of PKC␤ isoforms were tested. Overexpression of dnPKC␤II cDNA inhibited PMA-or SP600125-induced Prss14/epithin shedding whereas overexpression of dnPKC␤I cDNA did not (Fig. 3C). The PKC␤ inhibitor also reduced SP600125-induced Prss14/epithin shedding in PC3 and MCF7 cells (Fig. 3D). These results indicate that PKC␤II is critical for PMA-and SP600125-induced Prss14/epithin shedding.

SP600125 induces activation of PKC␤II and membrane translocation
Signaling further downstream of PKC␤II, such as activation and translocation of PKC␤II, was then investigated with SP600125 in 427.1.86 cells (Fig. 4). SP600125 increased phosphorylation of PKC␤II at serine 660 (a hallmark of the active form) whereas additional anisomycin pretreatment did not  (Fig. 4A). In all samples, the total amounts of PKC␤II were not changed upon treatment. In addition, the activity of PKC␤II in immunoprecipitated samples treated with SP600125 peaked after 1 h (Fig. 4B) and was reduced by anisomycin (Fig. 4C).

PKC␤II in Prss14/epithin shedding induces cell invasion
The location of PKC␤II after SP600125 treatment was in the plasma membrane (Fig. 4, D and E). Because levels of endogenous PKC␤II protein were insufficient to investigate cellular localization on a small scale, cellular location was investigated using overexpression of PKC␤II. Thus, cells were transfected with PKC␤II cDNA, and then cellular localization of PKC␤II was determined by specific antibody staining using immunofluorescence microscopy and stylized image transformation. Indeed, SP600125 induced translocation of PKC␤II to the membrane (Fig. 4D). PKC␤II in SP600125-treated cells clearly located in the cell-to-cell contact among clustered cells or in the plasma membrane of an isolated cell. When the fractions of cells with PKC␤II appearing in the membrane were calculated, almost all cells, if not all cells, in PMA-treated samples showed complete membrane localization, but only 20% of cells in SP600125-treated samples showed clear membrane localization. Using subcellular fractionation, as measured by the ratio of phosphorylated PKC␤II over E-cadherin of each fraction, the peak of PKC␤II in the membrane fraction was obtained at 30 min (Fig. 4E).
In Fig. 4F, we show the appearance of phosphorylated TACE upon PMA or SP600125 treatment. Phosphorylation at threonine 735 of TACE, which reflects activation of the enzyme, was increased by PMA and SP600125. From these results, we concluded that TACE phosphorylation is common downstream of PMA and SP600125.

PKC␤II plays critical role in PMA-or SP600125-induced cell invasion
Previously, we and others showed that Prss14/epithin is important for cell invasion (10,26,27). Serum is also an important factor for inducing Prss14/epithin shedding. Consistently, the invasiveness of 427.1.86 cells was decreased in both Prss14/ epithin knockdown cells and TACE inhibitor-treated cells (Fig.  5A). To study the role of PKC␤II in cell invasion, we generated PKC␤II-specific knockdown cell lines (Fig. 5B). As seen Fig. 5C, the degrees of 427.1.86 cell invasion induced by the serum gradient was increased upon SP5200125 or PMA treatment. However, the invasive nature of PKC␤II knockdown cells under the same condition was abolished, even in the presence of SP5200125 or PMA (Fig. 5C).
From these results, we established a model of how SP600125 induces ectodomain shedding of Prss14/epithin. SP600125 suppresses the inhibitory effect of JNK on PKC␤II, and the SP600125mediated increase in PKC␤II activity may require a labile, not yet

PKC␤II in Prss14/epithin shedding induces cell invasion
identified regulator of PKC␤II (Fig. 5D). PKC␤II can be activated after PMA or JNK inhibition, either directly or indirectly, affecting TACE activation. PKC␤II, TACE, and Prss14/epithin are translocated to the membrane; then cleavage of Prss14/epithin occurs at the membrane, increasing cell invasion.

PKC␤II and JNKs as prognostic markers in metastatic breast cancer
To evaluate the clinical significance of PKC␤II and JNKs in cancer patients, we utilized the publicly available TCGA breast cancer RNA-Seq database (Fig. 6). The expression levels of Prss14/epithin (ST14), PKC␤ (PRKCB), and three JNKs (MAPK8, MAPK9, and MAPK10) in more aggressive ER-negative (ER Ϫ ) and less aggressive ER-positive (ER ϩ ) breast cancer patients were compared individually in the plot (Fig. 6A). ST14 and PRKCB levels were higher in ER Ϫ patients, as expected. Among the MAPKs, MAPK9 (JNK2) exhibited significant differences between two groups. MAPK9 levels were higher in ER ϩ patients.
In a survival curve analysis (Fig. 6B), using a combination of ST14 and PRKCB, ST14 high and PRKCB high group patients showed significantly lower survival compared with ST14 low PRKCB low and ST14 low PKRCB high group patients, suggesting that ST14 is a stronger parameter when used with PRKCB. MAPK9 did not affect patient survival in a statistically significant way, although ST14 low MAPK9 low appeared to have the best survival.

Discussion
In this study, we showed upstream signaling of induced Prss14/epithin shedding and the significance and application of the resulting biology. Prss14/epithin shedding was induced by JNK inhibition by SP600125 as well as PMA, which shares downstreammoleculessuchasPKC␤IIandTACEthroughphosphorylation and translocation (see the summary model in Fig. 5C).

JNK inhibition in addition to PMA can mediate Prss14/epithin shedding through PKC␤II and TACE
SP600125-or PMA-induced shedding is mediated by TACE. We already showed that PMA-and TGF␤-induced Prss14/epithin shedding depends on TACE (15,16). Although Prss14/ epithin shedding can be blocked by serine protease inhibitors such as ecotin (17,28), we still believe that shedding is mediated by TACE, as tested by specific inhibition and knockdown experiments (Fig. 1, D and E). It is likely that shedding precedes activation, as shown in Fig. 1, F and G. The activated form of 30-kDa appeared more slowly in the medium than the shed form. In addition, we hardly observed the 30-kDa form in cell lysates (data not shown). In the case of TGF␤-induced Prss14/ epithin shedding, we showed that the kinetics of Prss14/epithin activity were slower than those of Prss14/epithin shedding (16).

PKC␤II in Prss14/epithin shedding induces cell invasion
SP6200125 was partially affected by anisomycin, a known JNK activator and a general inhibitor of protein synthesis. Knocking down JNK1 or JNK2 expression using siRNAs also induced Prss14/epithin shedding (Fig. 2B). SP6200125-induced Prss14/ epithin shedding was also affected by cycloheximide in addition to the transcription inhibitors ␣-amanitin and actinomycin D.
These observations suggest that JNK plays a role in Prss14/ epithin shedding through labile protein degradation or diminished protein stability.
Among conventional PKC isotypes that are known for being activated by PMA, PKC␤II was found to be involved in PMAand SP600125-induced shedding of Prss14/epithin (Fig. 3).

PKC␤II in Prss14/epithin shedding induces cell invasion
Experiments with isotype-specific inhibitors, specific knockdown, and DN forms revealed that PKC␤II is the key player in Prss14/epithin shedding induced by JNK inhibition or PMA exposure. JNK inhibition alone is sufficient to induce translocation of PKC␤II to the membrane. Inhibition of JNKs increases phosphorylation of PKC␤II at serine 660 (Fig. 4A). Serine 660 of PKC␤II resides in a C-terminal hydrophobic region and controls PKC␤II folding and stability (29), activating PKC␤II. SP600125, in fact, increased the activity of PKC␤II (Fig. 4, B and C) and induced TACE activation similar to PMA (Fig. 4D). These results are incorporated in the model presented in Fig. 5C.
In our earlier studies, we showed that actin rearrangement from stress fibers to the cortical actin form is essential for Prss14/epithin shedding (17). This actin rearrangement involves filamin, which can function as a vessel to bring proteins together (30). Prss14/epithin and TACE are brought to membranes by filamin upon PMA activation and interact together in complexes (17). Thus, TACE can cleave Prss14/epithin at the membrane. Because SP600125 induces cortical actin formation, and PKC␤II translocates to the membrane, we suspect that PKC␤II may also interact with filamin. However, to date, there are no data to support this idea, although PKC␣ and PKC⑀ isoforms have been shown to bind to filamin (31,32). There is evidence that actin cytoskeleton remodeling is associated with JNK activity. Actin stress fibers and the associated shear stress mediate inhibition of JNKs in vascular endothelium (33). TGF␤-induced actin stress fibers promote activity of JNK in human mesangial cells (34). Inhibition of JNK induces rearrangement of actin stress fibers to cortical actin filaments in 427.1.86 cells by interaction (Fig. S2). Therefore, it is conceivable that inhibition of JNK may lead to interactions to form complexes of F-actin, filamin, and PKC␤II.

Implication of PKC␤ and JNK as prognosis markers and as therapeutic targets for metastatic cancer
The biological significance of Prss14/epithin shedding is intriguing at this point; although activity clearly takes important parts in tumorigenesis and homeostasis of normal epidermal barrier (6,(35)(36)(37). From the results obtained using PKC␤II knockdown cell lines that had lost the ability to induce invasion (Fig. 5, A and B), we now understand that PKC␤II-mediated shedding is necessary for cell invasion.
In our earlier careful analysis of TCGA breast cancer patients, we showed that Prss14/epithin (ST14) is an excellent prognostic marker for highly metastatic ER Ϫ breast cancer (3). In this study, we show that the expression profiles of PRKCB and ST14 are reasonably well correlated (Fig. 6A). The levels of PRKCB were higher than those of ST14 in ER Ϫ patients, and MAPK9 (JNK2) levels were lower in the same group. In the

PKC␤II in Prss14/epithin shedding induces cell invasion
survival analysis (Fig. 6B), PRKCB collaborated with ST14 in terms of the poorest survival of patients. ST14 low PRKCB low patients have the best survival. Taking these analyses into consideration, we propose that PRKCB and MAPK9 can be additional prognosis markers and can be used in precision medicine for breast cancer patients.
JNK has previously been considered a target for cancer therapy using kinase inhibitors (38). However, there are complications and complexities when using such applications (39). It is important to note that JNK negatively regulates PKC␤ downstream of Prss14/epithin shedding and that JNK2 expression is lower in more aggressive cancers. Therefore, these observations are critical when considering JNK inhibition as a therapeutic approach, at least in breast cancer.

Cell culture
Cells were maintained in DMEM with 10% FBS, 100 units/ml penicillin, and 100 g/ml streptomycin. For drug treatments, 90% confluent cells were serum-starved for 4 h and then treated with 0.

PKC␤II in Prss14/epithin shedding induces cell invasion Subcellular fractionation
Cells were plated in a 100-mm dish at 2.5 ϫ 10 6 cells and harvested with Tris buffer containing 10 mM Tris-HCl (pH 8.0), 10 mM NaCl, 1 mM KH 2 PO 4 , 5 mM NaHCO 3 , 1 mM CaCl 2 , and 0.5 mM MgCl 2 supplemented with protease and phosphatase inhibitors. After sonication on ice, nuclei and debris were eliminated by centrifugation at 7,500 rpm for 5 min. Supernatants containing cytosol and plasma membrane were centrifuged at 34,000 rpm for 60 min (Sorvall Ultracentrifuge OTD-combi). Cytosol supernatants were collected, and crude plasma membrane pellets were resuspended with RIPA buffer and centrifuged at 34,000 rpm for 15 min. Plasma membrane supernatants were collected. To verify cross-contamination, ␤-actin and E-cadherin were detected by Western blot analysis.

Immunocytochemistry
Immunocytochemistry was performed as described previously (16). F-actin and PKC␤II were detected with rhodamineconjugated phalloidin (Molecular Probes) and anti-PKC␤II polyclonal antibody, respectively. Cells labeled with anti-PKC␤II polyclonal antibody were incubated with FITC-conjugated anti-rabbit IgG (Santa Cruz Biotechnology). Cells were observed with a fluorescence microscope (Axio-observer Z1m, Zeiss, Germany). Fluorescence images were captured and processed with Axio-vision imaging software for microscopy.

Immunoprecipitation, PKC␤II activity assay, and TACE phosphorylation
PKC␤II activity was determined by modifying an immune complex kinase assay method described elsewhere (41). Briefly, cells were lysed with Tris lysis buffer containing 20 mM Tris-HCl (pH 7.5), 1 mM EDTA, 1 mM EGTA, 150 mM NaCl, 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM ␤-glycerol phosphate, protease inhibitors, and phosphatase inhibitors. The cell lysates were precipitated with anti-PKC␤II polyclonal antibody and protein A-Sepharose beads (GE Healthcare). PKC␤II kinase activity was determined by ADP-Glo TM kinase assay kit (Promega). Luciferase values were detected using a Glomax 96 microplate luminometer (Promega). Statistical analyses were performed using Sigma Plot software. Comparisons between two groups were performed by Student's t test, and results are expressed as means Ϯ S.D. To investigate phosphorylation of TACE, cell lysates were analyzed by Western blot analysis using anti-phospho-TACE antibody (Thr-735) and anti-total TACE antibody.

Cell invasion assay
The invasion assay was performed using BioCoat Matrigel invasion chambers (Corning) according to the manufacturer's instructions. 427.1.86, 427-PKC␤II-KD4, and 427-EpiKD (9) cells were incubated with serum-free medium for 12 h. Cells (2 ϫ 10 5 ) were seeded on the upper side of a BioCoat Matrigel invasion chambers. The lower chamber was filled with DMEM containing 2% FBS with PMA (1 M), SP600125 (5 M), or TAPI-0 (20 M). After 24 h, cells on the lower surface of the membrane were fixed with 100% methanol for 10 min and stained with 0.2% crystal violet for 5 min. Invading cells were counted under an Axioimager M1 in five random fields. The total number of cells was divided by the number of counted fields in each assay.

Analysis of TCGA datasets
The Cancer Genome Atlas (TCGA) breast cancer patient data were downloaded using the Broad Institute TCGA Genome Data Analysis Center (2016) web portal, which was developed for automated analyses of TCGA data for general users (42). For comparison of gene expression between ER Ϫ and ER ϩ groups, box plots were generated using GraphPad Prism 7. Comparisons were analyzed by unpaired two-tailed Student's t test. For the 5-year survival rate, Kaplan-Meier survival analysis was performed using TCGA breast cancer data from patients who had not lost contact for five years. p values were calculated using a log rank (Mantel-Cox) test, and the hazard ratio was determined by the Mantel-Haenszel method.

Data availability
All data are contained in this manuscript.