STAT5 confers lactogenic properties in breast tumorigenesis and restricts metastatic potential

Abstract

Signal transducer and activator of transcription 5 (STAT5) promotes cell survival and instigates breast tumor formation, and in the normal breast it also drives alveolar differentiation and lactogenesis. However, whether STAT5 drives a differentiated phenotype in breast tumorigenesis and therefore impacts cancer spread and metastasis is unclear. We found in two genetically engineered mouse models of breast cancer that constitutively activated Stat5a (Stat5a^ca) caused precancerous mammary epithelial cells to become lactogenic and evolve into tumors with diminished potential to metastasize. We also showed that STAT5a^ca reduced the migratory and invasive ability of human breast cancer cell lines in vitro. Furthermore, we demonstrated that STAT5a^ca overexpression in human breast cancer cells lowered their metastatic burden in xenografted mice. Moreover, RPPA, Western blotting, and studies of ChIPseq data identified several EMT drivers regulated by STAT5. In addition, bioinformatic studies detected a correlation between STAT5 activity and better prognosis of breast cancer patients. Together, we conclude that STAT5 activation during mammary tumorigenesis specifies a tumor phenotype of lactogenic differentiation, suppresses EMT, and diminishes potential for subsequent metastasis.

INTRODUCTION

The STAT family of transcriptional factors comprises STAT1, 2, 3, 4, 5A, 5B, and 6 [1]. STAT5a is the principal STAT family member expressed in the mammary glands at all stages of mammary development except during involution [2]. It is usually phosphorylated and activated by Janus kinase 2 (JAK2) following prolactin and other cytokine hormones binding to their cognate receptors. Phosphorylated STAT5a proteins then homodimerize (or sometimes heterodimerize with another family member), translocate to the nucleus, and transactivate genes that regulate lobuloalveolar formation and milk production as well as cell survival, as we and others have reported [3–6].

Although not often mutated, STAT5 (referring to STAT5a/b as most commercial antibodies cannot distinguish these two highly homologous proteins) is known to be activated in breast atypia [7], ductal carcinoma in situ (DCIS) [8], and between 20 and 70% of human breast cancers [9–11]. pSTAT5 is found primarily in estrogen receptor α positive (ERα+) mammary tumors, but also in a subset of HER2+ tumors as well as triple-negative breast cancer (TNBC) [9–11]. The oncogenic potential of this lactogenic protein has been tested in animal models. STAT5a activation alone does not generally transform mammary cells [6, 8] although there are a few reports suggesting that transgenic expression of activated or even wild-type Stat5a in mice may lead to mammary tumors in a small proportion of the mice with a long latency [12–14]. On the other hand, constitutively activated Stat5a (Stat5a^ca) can potently accelerate mammary tumorigenesis initiated by an oncogene such as ErbB2 (HER2), likely via transactivation of anti-apoptosis genes that lower the apoptosis anticancer barrier often activated during tumor initiation [8].

While STAT5 can instigate breast tumor formation, STAT5 and nuclear STAT5 were detected less in high-grade breast cancer than in low-grade tumors, less in invasive carcinomas than in the prognostically more favorable secretory carcinoma subtype, and less in metastases than in primary tumors [9–11, 15, 16]. Higher STAT5 or pSTAT5 levels are also correlated with better patient overall and disease-specific survivals [10, 11, 15, 16]. These association studies suggest that STAT5 may be a favorable prognostic factor. In accordance, a few in vitro studies also showed that activation of PRL-JAK2-STAT5 signaling in cultured breast tumor cells can induce alveolar differentiation and suppress migration, invasion, and epithelial to mesenchymal transition (EMT) [17–19], but in vivo evidence confirming these observations is lacking. On the contrary, in two experimental metastasis mouse models, STAT5 was found to stimulate triple-negative breast cancer bone metastasis [20] and prostate cancer lung metastasis [21]. Furthermore, in clinical studies, high STAT5 activity was associated with advanced stage and poor outcome of prostate [22] and hepatocellular carcinoma [23]. Besides, STAT5 has been reported to promote EMT in several cancers of non-breast origin [21, 23–26]. Nonetheless, it remains unclear whether STAT5 activation drives alveolar and lactogenic differentiation during breast tumorigenesis and whether STAT5 affects metastasis of breast cancer in vivo.

In this study, we investigated the causal roles of STAT5a (the predominant isoform of STAT5) in alveolar and lactogenic differentiation during mammary tumorigenesis and progression in two different mouse models of breast cancer and in human breast cells cultured in vitro and as xenografts. We also studied potential mechanisms and their implications in breast cancer patients. These studies have uncovered a dichotomous function of STAT5 in breast tumorigenesis and metastasis.

RESULTS AND DISCUSSION

STAT5a activation causes alveolar differentiation of precancerous mammary lesions initiated by ErbB2

We previously reported that mammary intraductal injection of a Rous Sarcoma virus-derived vector (RCAS) can be used to deliver genes into the mouse mammary epithelia made susceptible to infection by transgenic expression of tva (encoding the receptor for RCAS) using the MMTV LTR promoter [27]. RCAS-mediated expression of a gene encoding a constitutively activated version of Stat5a (Stat5a^ca) drove alveolar differentiation and milk production in the absence of a pregnancy but rarely, if ever, induced precancerous lesions or tumors after many months [6, 8]. On the other hand, two weeks after intraductal injection of both RCAS-FLAG-Stat5a^ca and RCAS-HA-ErbB2^ca, precancerous lesions (defined as at least 3 layers of atypical cells growing into lumen [8]) were detectable by H&E staining (Fig. S1A) and 31% of these lesions were positive for both HA and FLAG (Fig. S1B, S1C). To test whether STAT5a affects the differentiation status of precancerous cells, we stained these lesions for two differentiation markers associated with milk production, β-casein and perilipin-2 (PLIN2, or adipophilin). We found abundant amounts of β-casein and PLIN2 in lesions positive for both ErbB2^ca and STAT5a^ca, but much less expression in lesions positive for ErbB2^ca only (Fig. 1A, S1D). These data demonstrate that STAT5a activation in precancerous lesions specifies an alveolar and lactogenic fate to the cells, similar to the impact of the STAT5 activation in the normal epithelial cells [6, 28].

Figure 1. STAT5a^ca causes lactogenic differentiation in ErbB2^ca-induced precancerous lesions and tumors, and it blocks lung metastases.

(A) Representative images of immunochemistry (IHC) staining for FLAG in STAT5a^ca and two milk production markers, β-casein and PLIN2, in precancerous lesions and tumors. Scale bars: 50 μm. N = 5 for tumor comparison; N = 7 for precancerous lesion comparison.

(B) H&E images of a pulmonary metastatic nodule from a ErbB2^ca-initiated primary tumor and a benign lung from a mouse bearing a ErbB2^ca/STAT5a^ca primary tumor. Scale bars: 50 μm.

(C) Summary of the lung metastatic incidence in mice infected by RCAS-ErbB2^ca alone and in mice bearing ErbB2^ca+/STAT5a^ca+ tumors. The Fisher’s exact test was used and *P < 0.05.

STAT5 activation leads to a functionally differentiated tumor phenotype

We previously reported that STAT5a activation accelerated tumorigenesis initiated by ErbB2^ca based on the observation that MMTV-tva female mice co-injected with RCAS-HA-ErbB2^ca and RCAS-FLAG-Stat5a^ca developed mammary tumors significantly faster than age-matched MMTV-tva mice injected with RCAS-ErbB2^ca alone [8]. These data predicted that the small proportion of co-infected cells (approximately 8% [8]) were much more competitive in tumorigenesis than cells expressing ErbB2^ca only. Immunohistochemistry for HA and FLAG on the tumors from the co-infected mice detected double positivity in 7 of 13 (53.8%) tumors while the remaining 6 tumors (46.2%) expressed ErbB2^ca only (Fig. S1E, S1F). And 5 of the 7 double-positive tumors exhibited near uniform positivity for FLAG, compared to moderate proportion scores in precancerous lesions (Fig. S1G). As expected, Kaplan-Meier tumor-free survival curves of mice bearing ErbB2^ca+/STAT5a^ca+ tumors separated well from those of mice bearing ErbB2^ca+/STAT5a^ca- tumors, which became inseparable from that of age-matched mice infected by RCAS-ErbB2^ca alone (Fig. S2A). In accordance with our previous report of STAT5 primarily promoting tumorigenesis via suppressing the apoptosis anticancer barrier (rather than proliferation) [8], ErbB2^ca+/STAT5a^ca+ tumors produced increased levels of BCL2 and BCL-xL (BCL2L1) (Fig. S2B, S2C), pro-survival proteins reported to be STAT5 targets [29, 30] although this transcriptional control has been controversial [31]. We next examined whether STAT5 activation also specifies a differentiated mammary tumor phenotype. Immunohistochemistry confirmed robust expression of both β-casein and PLIN2 in all 5 STAT5a^ca+ tumors examined while 3 of 5 STAT5a^ca- tumors didn’t show any β-casein expression, and 4 of 5 exhibited low levels of PLIN2 (Fig. 1A). Together, while confirming our previous observation that STAT5 activation accelerates ErbB2-initiated tumorigenesis likely by suppressing the anticancer barrier, these data show that STAT5 induces a functionally differentiated phenotype of tumors initiated by ErbB2 activation.

Activation of STAT5a during tumor initiation restricts the metastatic potential of HER2+ mammary tumors in mice

To test whether STAT5a-caused functional differentiation in tumor cells affects their metastatic potential, we compared the tumor burden in lungs by H&E. Figure 1B shows representative images of H&E staining of lungs at euthanasia when the primary ErbB2^ca+ tumors with or without STAT5a^ca reached 1.5 cm in diameter. Twelve of 30 (40 %) mice bearing ErbB2^ca-only mammary tumors harbored nests of tumor cells in the lungs. In contrast, only 1 of 16 (6.25%) mice bearing ErbB2^ca+/STAT5a^ca+ tumors presented tumor nests in the lungs (Fig. 1C). This 6-fold decrease of lung metastases suggests that the activation of STAT5 restricts the metastatic potential of ErbB2-initiated mammary tumors.

STAT5a overexpression leads to a more functionally differentiated tumor phenotype of ER+ breast cancer initiated by PIK3CA^H1047R

To test whether this STAT5 impact on tumor cell differentiation and metastasis is broadly applicable among a wide range of breast tumors, we tested the effects of STAT5 activation in mammary tumors initiated by PIK3CA, the proto-oncogene mostly commonly mutated in human breast cancer, especially in the ER+ subtype [32]. PIK3CA is too large to be expressed using RCAS, but we have reported that intraductal injection of lentivirus (FUCGW) carrying PIK3CA^H1047R, leads to ER+ mammary tumors in mice [33], while others have reported similar tumors using transgenic expression of mutated PIK3CA [34, 35]. Therefore, we intraductally injected FUCGW-PIK3CA^H1047R alone or together with FUW-FLAG-Stat5a^ca into the mammary glands of FVB female mice, and compared the features of the resulting mammary tumors by H&E staining and immunostaining. All 8 tumors initiated by PIK3CA^H1047R alone were well differentiated ductal carcinoma with extensive glandular formation, in accordance with our previous report of tumors induced by FUCGW-PIK3CA^H1047R [33]. All 8 tumors in mice co-injected with both viruses produced both PIK3CA^H1047R and STAT5a^ca, highlighting their collaborative roles in mammary tumorigenesis (Fig. 2). Besides, all STAT5a^ca+ tumors exhibited advanced pathological features including high-grade nuclear pleomorphism, high mitotic rates (>15 mitoses per 10 HPF’s), extensive areas of necrosis, and some metaplastic areas of squamous differentiation (Fig. 2). (This increase in aggressive pathological features was not observed in comparing tumors initiated by ErbB2^ca vs. ErbB2^ca/Stat5a^ca.) However, these PIK3CA^H1047R/STAT5a^ca tumors remained ERα+ (Fig. 2). Consistent with our observation in tumors initiated by both ErbB2^ca and Stat5a^ca, all 8 PIK3CA^H1047R/STAT5a^ca tumors exhibited robust expression of both β-casein and PLIN2, indicative of elevated lactogenic differentiation, while all 6 PIK3CA^H1047R only tumors showed negligible β-casein and PLIN2 expression (Fig. 2). However, we could not assess the impact of STAT5a^ca on metastasis in this model, since mice bearing PIK3CA^H1047R-only tumors or PIK3CA^H1047R/STAT5a^ca tumors did not develop detectable lung metastases at the ethnical end point of primary tumors reaching 2.0 cm in diameter. Together, these data suggest that STAT5 activation causes a more functionally differentiated phenotype in mammary tumors initiated by PIK3CA activation although the tumors with higher STAT5 activity appear more aggressive histopathologically.

Figure 2. STAT5a^ca causes functional differentiation of mammary tumors initiated by PIK3CA^H1047R.

Representative images of tumor sections stained by H&E and for the indicated markers. Scale bars: 100 μm. N = 6-8.

Activation of STAT5a restricts the metastatic potential of human breast cancer cells in vitro and in vivo

To determine whether STAT5 activation affects the metastatic ability of human breast tumor cells, we stably introduced STAT5a^ca using a lentiviral vector into two human breast lines and compared their migratory and invasive capacity with their corresponding control cells. STAT5a^ca decreased the migration and invasion abilities of MCF10A-PIK3CA^H1047R by approximately 70-80% (Fig. 3A, 3B, S3A), and those of MCF7 by approximately 50% (Fig. S3B) while STAT5a^ca didn’t affect the growth of MCF10A cells (Fig. S3C). Furthermore, we compared two MCF10A cell sublines in a 3D Matrigel culture. While both MCF10A-PIK3CA^H1047R and MCF10A-PIK3CA^H1047R/STAT5a^ca grew similarly shaped spheres initially (day 4), with time, the former developed invasive protrusions while the latter remained spherical (Fig S3D). These in vitro data suggest that STAT5a hinders breast cancer metastasis.

Figure 3. STAT5a^ca restricts the metastatic potential of human breast tumor cells, and higher STAT5a levels are associated with better prognosis of patients.

(A&B) Dot plots showing the migration and invasion of MCF10A cells carrying the indicated genes. Values in Y-axis are normalized to the PIK3CA^H1047R group. The number of biological replicates is indicated by the filled circles. The Student’s t test was used. ***P < 0.001, ****P < 0.0001.

(C) Representative images of bioluminescence signals in the soft organs (upper panel: ovaries, spleen, kidneys; lower panel: liver, brain, lungs) and bones from mice implanted with MCF10A cells carrying PIK3CA^H1047R (P) or PIK3CA^H1047R/STAT5a^ca (P/S).

(D) Quantification of combined signals in (C). The Student’s t test was used. * P < 0.05. Each filled circle represents a mouse. P, PIK3CA^H1047R (N = 8); P/S, PIK3CA^H1047R/STAT5a^ca (N = 4).

(E) Representative images of bioluminescence signals in the NSG mice at 6 weeks post tail vein injection of MCF10A cells carrying PIK3CA^H1047R or PIK3CA^H1047R/STAT5a^ca.

(F) Quantification of the metastasis burden. Each filled symbol represents a mouse. P, PIK3CA^H1047R (N = 8); P/S, PIK3CA^H1047R/STAT5a^ca (N = 6). Values in Y-axis are photon flux from lungs of individual animals.

(G) Kaplan-Meier plot of metastasis-free survival of patients stratified according to tumor STAT5A mRNA expression in the EMC-MSK dataset (N = 557). The patients were assigned into high or low STAT5A group using the median of STAT5A mRNA expression as the cutoff.

(H&I) Kaplan-Meier plots of relapse-free survival (G) and overall survival (H) of patients stratified according to tumor STAT5A mRNA expression in the METABRIC Dataset (N = 1 903). The patients were assigned into high or low STAT5A group using the median of STAT5A mRNA expression as the cutoff.

Next, we tested whether STAT5a^ca affects human breast cancer metastasis in vivo. We intraductally implanted MCF10A-PIK3CA^H1047R cells and MCF10A-PIK3CA^H1047R/STAT5a^ca cells into the #4 mammary glands of female NSG mice. IVIS imaging and tumor palpation indicate that STAT5a^ca does not significantly impact the growth of these MCF10A-PIK3CA^H1047R cells (Fig. S4A, S4B), in accordance with our in vitro data (Fig. S3C). When tumors reached 2.0 cm in diameter, we harvested soft organs (ovaries, spleen, kidneys, liver, brain, and lungs) and bones for ex vivo IVIS-imaging to quantify metastasis. We found that 6 of the 8 mice (75%) implanted with MCF10A-PIK3CA^H1047R showed metastases in distant organs (Fig. 3C), which was confirmed by H&E (Fig. S4C), while none of the 4 mice implanted with MCF10A-PIK3CA^H1047R/STAT5a^ca had a detectable metastatic signal in any organ (Fig. 3C, 3D). To determine the stage at which STAT5 influences metastasis, we tail vein-injected MCF10A-PIK3CA^H1047R cells and MCF10A-PIK3CA^H1047R/STAT5a^ca cells into NSG mice, and compared the mice for pulmonary metastatic burden over a window of 6 weeks by in vivo IVIS imaging. As shown in Fig. 3E and 3F, the two groups developed lung metastasis overtime without discernable differences, suggesting that STAT5a does not significantly impact metastatic development after intravasation. Taken together, these in vitro and in vivo results indicate that STAT5 activation restricts metastatic potential of human breast cancer cells primarily by suppressing local migration, invasion and possibly intravasation.

High STAT5A level predicts favorable outcome in breast cancer patients

By immunohistochemical staining of breast cancer specimens, STAT5 activation has been correlated with ER+ breast cancer with a more favorable prognosis [10, 11], but this potential biomarker has not been studied in large public datasets or within specific breast cancer subtypes. Therefore, to further explore the prognostic potential of STAT5 in breast cancer patients, we performed bioinformatic analyses to compare outcome of patients bearing STAT5-low vs. STAT5-high breast tumors in two publicly available datasets. Due to the scarcity of STAT5 and pSTAT5 protein data in public datasets, we compared tumors with high vs. low mRNA levels of STAT5A (which has Spearman correlation of 0.37 with STAT5A protein levels in TCGA based on LinkedOmics [36]). In the EMC-MSK dataset [37, 38], we found that higher STAT5A mRNA levels were correlated with a better metastasis-free survival when all patients were included (p = 0.0251; Fig. 3G). When the patients were separated into ERα+, HER2+ and TNBC subtypes, the association was highly significant among the ERα+ breast cancer patients but was not detected in HER2+ patients or triple-negative patients (Fig. S5A). Consistent with the finding, in the METABRIC Database [39, 40], higher STAT5A expression was associated with better relapse-free survival among all breast cancer patients (Fig. 3H), and also in Luminal A, and Luminal B patients, but not among other subtypes (Fig. S5B). In addition, higher STAT5A expression was associated with better overall survival among all breast cancer patients in METABRIC Database (Fig. 3I),

As a transcription factor, STAT5 potentially regulates expression of multiple genes in breast tumors. Therefore, we next asked whether a STAT5 gene expression signature is correlated with patient outcome. While a STAT5-dependent gene set in the normal mouse mammary gland has been reported [41], a STAT5 gene signature in human breast cancer cells is not yet available. We performed RNAseq on a panel of human breast cancer cell lines carrying vector vs. lenti-STAT5a^ca, and generated a STAT5A gene signature consisting of 105 genes that are upregulated in any 4 out of 6 pairs (Supplementary Table 1). This STAT5A signature exhibited a significant correlation with STAT5A mRNA levels in METABRIC and TCGA breast cancer datasets (Fig. S6A, S6B). In the TCGA dataset, higher STAT5A signature scores were associated with better disease-free survival, although STAT5A mRNA levels were not associated with disease-free survival (Fig. S6A). Likewise, in the METABRIC dataset, this gene signature was associated with better relapse-free survival although it lacked power to correlate with outcome in breast cancer subtypes (Fig. S6C). Since EMC-MSK dataset is based on Microarray and does not include many of the genes in our signature, we did not attempt correlation analysis. Together, these bioinformatic data indicate that STAT5 expression levels and activity can serve as a biomarker for favorable prognosis for breast cancer patients, especially patients with ERα+ breast cancer. However, the power of prediction is relatively modest, and STAT5a mRNA and its gene signature do not always agree possibly due to several reasons, including uncertain correlation between STAT5a gene expression levels and protein activity levels (which the gene signature) and potential contributions of STAT5b and other STAT family members to the gene signature levels in patients.

STAT5a alters protein levels that regulate epithelial-mesenchymal transition (EMT)

To gain insight into mechanisms by which STAT5 activation blocks breast cancer metastasis, we used the reverse-phase protein array (RPPA) to identify proteins expressed differentially between murine mammary tumors initiated by PIK3CA^H1047R vs. PIK3CA^H1047R/Stat5a^ca. Among 243 antibodies detecting total and phosphorylated proteins, 90 were different between these two groups of tumors (Fig. S7). The top 4 down-regulated proteins in the PIK3CA^H1047R /STAT5a^ca tumor cohort include a prominent EMT driver, Slug (SNAI2) and a key EMT marker, Vimentin (VIM) (Fig. 4A). We confirmed by Western blotting that STAT5a^ca suppressed these two proteins, and we also detected decreased levels of another EMT regulator, ZEB1 and the mesenchymal marker N-cadherin (CDH2), as well as increased levels of the epithelial marker E-cadherin (CDH1) and the milk protein β-casein in murine mammary tumors (Fig. 4B). We validated these data in human breast cancer cells by comparing MCF10A-PIK3CA^H1047R cells vs. MCF10A-PIK3CA^H1047R/STAT5a^ca cells (Fig. 4C), and we additionally detected STAT5a-mediated suppression of two more EMT regulators, Snail (SNAI1) and TWIST (Fig. 4C), consistent with previous studies in cell cultures [19].

Figure 4. STAT5a^ca suppresses breast cancer EMT.

(A) The top 4 proteins decreased in PIK3CA^H1047R/STAT5a^ca+ tumors detected by RPPA. N = 7.

(B & C) Western blotting for the indicated proteins.

(D) Real-time PCR analysis for the indicated genes and miRNA. Data were normalized to the paired MCF10A-PIK3CA^H1047R group. mRNA was normalized to human acidic ribosomal protein (hARP). miR-200c was normalized to miR-103a. Student’s t test was used and n = 4. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001

(E) ChIPseq data showing STAT5 binding to both the promoter and gene body region of Vim, and the gene body of Zeb1 and Zeb2.

(F) Model depicting the dichotomous impact of STAT5 on breast cancer formation and metastasis. While promoting tumorigenesis by lowering the apoptosis anticancer barrier, STAT5 also defines an alveolar differentiation fate (white dots), preventing EMT and metastasis.

To determine if STAT5 regulates the transcription of these genes, we first performed real-time PCR on these human cancer cells. Consistent with the findings in proteins, mRNA levels of CDH2, VIM, SNAI1, SNAI2, TWIST1 and ZEB1 were decreased by STAT5a^ca, but CDH1 and ZEB2 were not affected (Fig. 4D). Interestingly, we also found an increase of miR200c, a microRNA which was reported to suppress the EMT programs via a ZEB/miR-200 feedback loop [42]. Importantly, the analysis of published ChIPseq data comparing mammary gland cells with different gene dosages of Stat5a and Stat5b at lactation day 1 [43] revealed that multiple EMT-related genes, including Vim and Zeb1, Zeb2, had STAT5 bound to their promoters or gene bodies (Fig. 4E, S8, S9), suggesting that STAT5 may regulate EMT via transactivation of EMT regulators.

This study using multiple in vivo models as well as multiple cell lines demonstrates a dichotomous role of STAT5 in cancer initial development and subsequent metastasis (Fig. 4F). We previously published that STAT5 accelerates precancerous lesion progression by lowering the apoptosis anti-cancer barrier (without a significant impact on cell proliferation) [8, 44]. The data presented in this study further strength this previous finding and show a diminished impact of STAT5 on growth of established breast tumor cells especially if PIK3CA is activated. On the other hand, our ErbB2 and PIK3CA mouse model data demonstrate that STAT5 also programs precancerous and cancerous cells into a functionally differentiated state that is characterized by the expression of alveolar markers and milk proteins (Fig. 1, 2). These data provide in vivo support to previous reports of JAK2-STAT5 signaling in promoting alveolar differentiation of cultured breast cancer cells [18, 19]. Of note, it remains to be determined whether the gain of these alveolar features truly caused these precancerous cells to gain or lose stem/progenitor properties compared to other precancerous cells that did not acquire STAT5 activation. It has been reported that certain alveolar cells marked by the expression of the whey acidic protein gene, a transcriptional target of STAT5, can gain multipotency following a reproductive cycle [45, 46], and that they are probably the cells of origin of tumors in mice transgenic for ErbB2 [47].

Using both in vitro and in vivo experiments, we further demonstrate that STAT5 suppresses EMT and restrains early steps of metastasis including migration and invasion although exerting no significant impact on tumor cells that have already disseminated (Fig 3). It is conceivable that a more differentiated epithelial state may counter EMT activators and hinder migration, invasion, and metastasis, but it remains to be determined how much functional differentiation per se contributes to blocked metastasis. Furthermore, while STAT5 regulates multiple EMT-related genes likely via transcriptional activation (Fig. 4), the crucial EMT regulators directly controlled by STAT5 remain to be elucidated. In addition, as multiple previous studies have reported STAT5’s promotion of EMT and metastasis [20, 21, 23–26], the molecular mechanism for these apparently opposing roles of STAT5 on EMT in different tissue contexts remains to be investigated.

In conclusion, while it promotes breast cancer formation at least partly by suppressing the apoptosis anticancer barrier and may be a chemoprevention target [8, 44], STAT5 activation initiates a lactogenic network in precancerous and cancerous cells, suppresses EMT, and impairs cancer migration, invasion, and dissemination to distant organs, consequently serving as a potential biomarker in breast cancer prognosis (Fig. 4F).

MATERIAL AND METHODS

Animals

Wild-type FVB/N mice and MMTV-tva transgenic mice (on the FVB background) were bred in house. NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ (NSG) mice were purchased from the Jackson lab (Bar Harbor, ME, USA). The animal experimentation protocol was approved by the BCM institutional Animal Care and Use Committee, and conducted under the guidelines described in the NIH Guide for the Care and Use of Laboratory Animals.

Virus production and titration

The plasmid RCAS-HA-ErbB2^ca carrying the HA-tagged constitutively activated ErbB2 mutant gene was reported in our previous work [27]. The plasmid RCAS-FLAG-Stat5a^ca [6] carries the FLAG-tagged activated mutant of STAT5a, which was subcloned from the pMX-Stat5a1*6 [48]. The plasmid FUCGW-PIK3CA^H1047R [33] carries a constitutively active mutant of PIK3CA along with a GFP marker. The plasmid FUW-Stat5a^ca was constructed by subcloning Stat5a1*6 to the FUW vector and adding a mCherry cassette by infusion cloning. The production and titration for both avian retrovirus and lentivirus were performed as described before [27, 49, 50]. All mice were randomized at the time of experiments.

Mouse tumor model generation

Virus and breast cells were delivered into mouse mammary glands by intraductal injection as previously described [50–52]. To generate HER2+ mammary precancerous lesions and tumors, 10-week-old MMTV-tva (MA) female mice were intraductally injected via the #4 teat with RCAS-HA-ErbB2^ca (5 × 10⁴ IUs per gland) alone or together with RCAS-FLAG-Stat5a^ca (10⁷ IUs per gland). The injected mammary glands were harvested 2 weeks after virus injection for precancerous lesion studies. Tumors were scored when the diameter reached ~2-3mm and can be detected by palpation. The length and width of tumors were measured with a caliper twice per week until the tumor size reached the ethical endpoint (1.5-2 cm). To generate ER+ mammary tumors, 10-week-old female mice in FVB/N background were intraductally injected with FUCGW-PIK3CA^H1047R (10⁷ IUs per gland) alone or together with FUW-FLAG-Stat5a^ca (7 × 10⁷ IUs per gland). For the spontaneous metastasis model, 10-week-old NSG mice were intraductally implanted with luciferase-tagged MCF10A-PIK3CA^H1047R cells or MCF10A-PIK3CA^H1047R/STAT5a^ca cells (30 000 cells per gland). The lesion burden was determined biweekly by bioluminescence for 16 weeks post cell implantation when >50% of mice developed palpable tumors. The metastasis burden across organs was determined by bioluminescence at harvest. For the experimental metastasis model, these two groups of cells were tail vein-injected into 10-week-old NSG mice (1 × 10⁶ cells per mouse). The metastasis burden was determined weekly by bioluminescence for 6 weeks following tail vein injection.

Cell culture and stable line generation

Breast MCF10A sublines stably expressing PIK3CA^H1047R and/or Stat5a^ca were generated by transducing luciferase-tagged cells with lentiviruses carrying FUCGW-PIK3CA^H1047R [33] or matched vector FUCGW and/or FUW-Stat5a^ca-mCherry or matched vector FUW-mCherry. In brief, cells were incubated with viral particles (M.O.I.=10) and 10 μg/mL polybrene, centrifuged at 1 100 xg for 30 min at 37°C, and washed with PBS. After passage, cells were FACS-sorted for GFP and/or mCherry. 3D culture assay was performed as published [53]. In brief, 5 000 MCF10A cells in media containing 2% growth factor-reduced Matrigel (BD Biosciences, Franklin Lakes, NJ, USA) were plated to 8-well chamber slides precoated with a base layer of 100% Matrigel. 3D growth was imaged at day 4 and day 7. Breast cancer cell lines MCF7, BT20, MDA-MB-231, T47D were transduced with FUW-Stat5a^ca-mCherry or matched vector FUW-mCherry and FACS-sorted to derive sublines.

Immunohistochemistry (IHC) and immunofluorescence (IF)

Immunochemistry staining was performed using the Vectastain Elite ABC system (Vector Laboratories, Burlingame, CA, USA) and developed using DAB as chromogen (Agilent Dako, Santa Clara, CA, USA). The antibodies used in this study include anti-HA (#3724, 1:4 000, Cell Signaling Technology, Danvers, MA, USA), anti-FLAG (#8146, 1:400, Cell Signaling Technology), anti-β-casein (sc-166530, 1:600, Santa Cruz Biotechnology, Dallas, TX, USA), and anti-BCL-xL (sc-8044, 1:200, Santa Cruz Biotechnology). For IF staining, the anti-PLIN2 (NB110-40877, 1:600, Novus Biologicals, Littleton, CO, USA) primary antibody was used. Sections were incubated with primary antibodies overnight at 4°C, followed by 30 min incubation with Alexa Fluor 568-conjugated anti-rabbit antibodies (#A11011, 1:200, Invitrogen, Waltham, MA, USA). The slides were mounted with VECTASHIELD Antifade Mounting Medium with DAPI (Vector Laboratories).

Western blotting

Murine tumors were ground in a mortar with liquid N₂, and proteins were extracted in lysis buffer (MPER) containing protease inhibitor. Proteins from MCF10A and MCF7 were extracted by cell lysis in RIPA buffer (Bio-Rad, Hercules, CA, USA). 30 to 50 μg of protein per lane were separated in SDS-PAGE and transferred onto nitrocellulose membrane (Bio-Rad) for incubation with antibodies and detection using an ECL chemiluminescent kit (ProteinSimple, San Jose, CA, USA). The antibodies used include antibodies against BCL-2 (sc-7382, 1:1 000, Santa Cruz Biotechnology), BCL-xL (#sc-8044, 1:1 000, Santa Cruz Biotechnology), E-cadherin (Cell Signaling Technology, #3195, 1:1 000), N-cadherin (#610920, 1:1 000, BD Biosciences), Vimentin (#5741, 1:1 000, Cell Signaling Technology), Snail (#3879, 1:1 000, Cell Signaling Technology), Slug (#9585, 1:1 000, Cell Signaling Technology), FLAG (#8146, 1:1 000, Cell Signaling Technology), β-casein (#MA1-46056, 1:1 000, Invitrogen), TWIST (#sc-81417, 1:1 000, Santa Cruz Biotechnology), ZEB1 (#70512, 1:1 000, Cell Signaling Technology), and β-actin (#A5441, 1:5 000, Sigma-Aldrich, St. Louis, MO, USA). At least three independent biological repeats were performed.

Transwell migration and invasion assay

Transwell migration and invasion assays were performed as we described previously [54]. The MCF-10A cells were seeded at 20 000 cells per well and incubated 16 hours for migration assay or seeded at 50 000 cells per well for 20 hours for invasion assay. MCF-7 cells were seeded at 50 000 cells and incubated for 16 hours for migration assays and 100 000 cells per well for 48 hours for invasion assays. At least three independent experiments were performed.

Reverse phase protein array (RPPA) analysis

Reverse phase protein array assays were carried out as reported previously [55]. Protein lysates were prepared from tumors with modified tissue protein extraction reagent (TPER; Pierce, Thermo Scientific, Appleton, WI, USA) and a cocktail of protease and phosphatase inhibitors. Then 243 antibodies against total or phosphorylated protein were used. The full list of antibodies is at https://www.bcm.edu/centers/cancer-center/research/shared-resources/cprit-cancer-proteomics-and-metabolomics/reverse-phase-proteinarray.

Real-time PCR

Cells were lysed with TRI Reagent (Sigma-Aldrich). Total RNA was purified from cells using Direct-zol RNA Microprep Kit (Zymo Research, Irvine, CA, USA). cDNA synthesis from mRNA was carried out using a SuperScript III First-Strand Synthesis System (Invitrogen), and Takyo No Rox SYBR MasterMix dTTP Blue (Eurogentec, Fremont, CA, USA) was used for real-time PCR assay. For microRNA, the miRCURY LNA miRNA PCR Starter Kit (QIAGEN, Germantown, MD, USA) was used to construct cDNA and also real-time PCR assay. All qPCR assays were conducted in a CFX96 qPCR Instrument (Bio-Rad). Four independent repeats were performed. mRNA was normalized to human acidic ribosomal protein (hARP). miR-200c was normalized to miR-103a. The primers used for the PCR are:

CDH1: 5’-CAGGTCTCCTCTTGGCTCTG-3’ and 5’-GACCGGTGCAATCTTCAAAA-3’; CDH2: 5’-GTGCATGAAGGACAGCCTCT-3’ and 5’-AGCTTCTCACGGCATACACC-3’; VIM: 5’-CGAAAACACCCTGCAATCTT-3’ and 5’-CTGGATTTCCTCTTCGTGGA-3’; SNAI1: 5’-CTTCTCTAGGCCCTGGCTG-3’ and 5’-CATCTGAGTGGGTCTGGAGG-3’; SNAI2: 5’-TCGGACCCACACATTACCTT-3’ and 5’-TGACCTGTCTGCAAATGCTC-3’; TWIST1: 5’-GGACAAGCTGAGCAAGATTCA-3’ and 5’-CGGAGAAGGCGTAGCTGAG-3’; ZEB1: 5’-AGCAGTGAAAGAGAAGGGAATGC-3’ and 5’-GGTCCTCTTCAGGTGCCTCAG-3’; ZEB2: 5’-TTTCAGGGAGAATTGCTTGA-3’ and 5’-CACATGCATACATGCCACTC-3’; hARP: 5’-CACCATTGAAATCCTGAGTGATGT-3’ and 5’-TGACCAGCCCAAAGGAGAAG -3’.

Bioinformatic studies of clinical data from public datasets, ChIPseq data study, and statistical analyses

Clinical data were collected from the Erasmus Medical Center/Memorial Sloan-Kettering (EMC-MSK) dataset [37, 38], the METABRIC dataset [39, 40] and the TCGA dataset [56, 57]. To generate STAT5A signature, we built 6 pairs of human breast and cancer cell lines with or without STAT5a^ca overexpression (MCF10A, MCF10A-PIK3CA^H1047R, MCF7, BT20, MDA-MB-231, T47D) for RNAseq analysis (Novogene, China), and fold change of gene expression was calculated based on the ratio of fpkm values in STAT5a^ca overexpressed samples vs vector controls. We generated a STAT5A gene signature consisting of 105 genes that were induced over 2-fold by STAT5a^ca overexpression in at least 4 of the 6 cell lines and calculated STAT5A gene signature scores using z-score method. Survival R package was used to compare Kaplan-Meier survival curves using the log rank test.

ChIPseq data of STAT5 sequentially knockout mice at lactation day 1 were previously published [43]. The EMT-related gene list is derived from EMTome by Vasaikar et al [58]. Raw public data were downloaded with accession number GSE37646 and GSE40930. STAT5A ChIPseq raw data (GSE40930) were mapped to mm10 with BWA (0.7.4). We called significant peaks using MACS2 (2.2.7.1) with default parameters. To find out EMT genes bound by STAT5A, bedtools (2.28.0) was applied to intersect significant peak regions and gene promoters or gene bodies. We utilized deeptools (3.5.1) to generate normalized bigwig files from bam files with RPGC normalization method. RNA-seq data (GSE37646) were mapped to mm10 using STAR (2.5.2b) and then counted using HTSeq (2.7).

For tumor growth of HER2+ tumors, Kaplan-Meier survival curves were analyzed by Prism 9, and P values were calculated using the log-rank test. For assessment of spontaneous metastasis, the P value was calculated by Fisher’s exact test. A two-tailed Student’s t test was applied for all other experiments. Results are expressed as the mean ±SD.

Availability of data and materials

Available upon request to liyi@bcm.edu.