Effect of glycogen synthase kinase-3 inactivation on mouse mammary gland development and oncogenesis

Many components of the Wnt/β-catenin signaling pathway have critical functions in mammary gland development and tumor formation, yet the contribution of glycogen synthase kinase-3 (GSK-3α and GSK-3β) to mammopoiesis and oncogenesis is unclear. Here, we report that WAP-Cre-mediated deletion of GSK-3 in the mammary epithelium results in activation of Wnt/β-catenin signaling and induces mammary intraepithelial neoplasia that progresses to squamous transdifferentiation and development of adenosquamous carcinomas at 6 months. To uncover possible β-catenin-independent activities of GSK-3, we generated mammary-specific knockouts of GSK-3 and β-catenin. Squamous transdifferentiation of the mammary epithelium was largely attenuated, however, mammary epithelial cells lost the ability to form mammospheres suggesting perturbation of stem cell properties unrelated to loss of β-catenin alone. At 10 months, adenocarcinomas that developed in glands lacking GSK-3 and β-catenin displayed elevated levels of γ-catenin/plakoglobin as well as activation of the Hedgehog and Notch pathways. Collectively, these results establish the two isoforms of GSK-3 as essential integrators of multiple developmental signals that act to maintain normal mammary gland function and suppress tumorigenesis.


INTRODUCTION
Epithelial malignancies, including those of the breast, are thought to initiate in stem-like cells defined by their capacity to self-renew and persist long enough to sustain and propagate mutations, as well as generate all functional cell types within a given tissue. These 'tumor-initiating cells' (TICs) may arise from normal adult stem cells as a consequence of alterations in the regulation of balance between self-renewal and differentiation or from more committed progenitors that have re-acquired stem cell characteristics during transformation. 1,2 The contribution of distinct stem/progenitor cells to breast cancer heterogeneity remains obscure. With the identification of candidate mammary stem and progenitor populations that vary in their differentiation capacities, the nature of the TIC population and the impact of oncogenes within this critical cellular pool are being elucidated. 1,3 Multiple lines of evidence have shown developmental pathways including Wnt, Hedgehog (Hh) and Notch regulate stem cell homeostasis and have the capacity to induce cancer of various tissues when upregulated. The protein-serine kinases glycogen synthase kinase (GSK)-3α and β are shared components of these networks and act to regulate signals emanating from receptors upon morphogen or ligand ligation. Inhibition of GSK-3 is critical to canonical Wnt signaling leading to increased β-catenin levels associated with elevated proliferation and suppression of differentiation in a number of tissues. 4-7 GSK-3 has been shown to interact with Hh pathway components at several levels to inhibit transcriptional functions of Gli proteins. [8][9][10][11] Several cell line studies have demonstrated that GSK-3 inhibition also modulates Notch signaling, at least in part by phosphorylating the Notch intracellular domain (NICD), thereby preventing nuclear entry and efficient association with its target gene promoters. [12][13][14] Pathological hyperactivation of each of these pathways is a hallmark of many tumors, including breast cancer. Although mutations in β-catenin and other pathway components, such as adenomatous polyposis coli, are some of the most frequent signaling abnormalities contributing to human tumor pathogenesis, they are much less frequent in breast cancer. Despite the rarity of mutations, increased cytoplasmic and nuclear β-catenin levels have been documented in~40% of primary breast cancers including metaplastic breast carcinomas, a rare yet aggressive subset of breast cancer that shares features of both basal-like and triple-negative breast cancers. [15][16][17][18][19][20] The mechanism by which Wnt/β-catenin signaling is activated in these tumors is unknown but may involve excess Wnt ligand production and aberrant receptor activation. [21][22][23] Wnt/β-catenin functions as a stem cell survival factor in continuously self-renewing systems including mammary tissues. 24,25 Non-neoplastic glands of MMTV-Wnt1 and those from β-catenin mutant mice where the transgene lacks the N-terminal region containing GSK-3 regulatory sites present an expanded progenitor cell fraction and limited or impaired capacity for functional differentiation in favor of a less committed cellular state. 26,27 In addition, Wnt/β-catenin-responsive cells have recently been shown to be long-lived stem cells that make up a large proportion of the basal compartment able to survive multiple rounds of lobuloalveolar tissue turnover. 28 This pool of vulnerable cells may constitute a vulnerable population for oncogenic mutation leading to generation of TICs. Indeed, stabilization of a β-catenin transgene expressed from its endogenous promoter or via adenomatous polyposis coli mutations or deficiency leads to mammary hyperplasias, accompanied by loss of alveolar structures, [29][30][31][32] whereas overexpression of N-terminally truncated β-catenin results not only in precocious alveolar differentiation but also in formation of mammary adenocarcinomas. [33][34][35] In addition, overexpression of the positive Wnt/β-catenin regulator, casein kinase 2, as well as a putative Wnt/β-catenin target gene, cyclin D1, in the mammary epithelium results in hyperplasias, squamous differentiation and adenocarcinomas. 36,37 High levels of Notch-1 were found to be associated with a poorer outcome in breast cancer patients while rearrangements of Notch-1 and 2 were found in triple-negative breast cancers where the fusion transcripts retained exons that encode the NICD thus maintaining transcriptional outputs. [38][39][40] Increased numbers of breast stem/progenitor cells have been reported upon enhanced Notch signaling. [41][42][43] This expansion of immature cells was predictive of tumor formation suggesting NICD1 expression in the mammary epithelium generates a population of unstable, pre-malignant progenitor cells. 43 Notch activity has also been shown to result in luminal progenitor expansion leading to hyperplasia and tumorigenesis. 41,44,45 Recently, two Notch-2 progenitor populations were identified within the luminal compartment representing unique mammary lineages. 46 Thus, dysregulation of Notch activity may amplify a distinct stem/ progenitor population within the mammary tissues, which may contribute to TIC formation.
Hh ligand overexpression is associated with the basal-like breast cancer subtype and poor outcome in terms of metastasis and breast cancer-related death while expression of Hh receptor Ptch1 and transcriptional activator Gli1 has been detected in invasive carcinomas but not in normal breast epithelium. [47][48][49][50][51] Hh may increase the proliferative capacity and enhance self-renewal of mammary progenitors rather than enlarge the multi-potent stem cell pool, thus perturbing the stem cell vs progenitor cell balance. [52][53][54][55] In concordance, Gli1 overexpression within the mammary epithelium induces mammary tumors with increased levels of Bmi-1, a transcriptional repressor of the polycomb gene family previously implicated in stem cell maintenance. 56,57 To assess the functional consequence of GSK-3 loss from the mammary epithelium and to explore potential β-cateninindependent roles of these protein kinases, both GSK-3 genes were deleted specifically in mammary tissues in the presence or absence of a functional β-catenin gene. Glands of GSK-3 mutant animals assumed epidermal identity via cell-autonomous mechanisms leading to adenosquamous carcinoma formation. Although mammary-selective inactivation of β-catenin initially subverted epidermoid transdifferentiation of GSK-3-null tissue, misregulation of Hh and Notch pathways contributed to the development of adenocarcinomas in the absence of GSK-3 and β-catenin. Thus, GSK-3, by restraining multiple pathways maintains mammary epithelial cell (MEC) function and inhibits tumor formation.

RESULTS
Generation of mice harboring mammary-gland-selective deletion of GSK-3α and GSK-3β To investigate the functions of GSK-3s in the mammary gland, we generated GSK-3α − / − ; GSK-3β Exon2 LoxP/Exon2 LoxP , referred to as GSK-3α − / − ; GSK-3β FL/FL mice with mammary-gland-selective deletion of the floxed GSK-3β alleles achieved by Cre-mediated recombination driven by the whey acidic protein (WAP) promoter yielding WAP-GSK-3 double knockout animals (WAP-DKO). Expression of the WAP-Cre transgene can be detected during the estrus stage of the murine estrus cycle (hence starting at~4 weeks of age) and the WAP promoter is substantially increased during each pregnancy when WAP expression is observed in both ductal and alveolar epithelial cells, particularly starting from 15 days post coitum (dpc 15). [58][59][60] Genomic PCR of WAP-DKO pregnant glands confirmed absence of GSK-3α and loss of GSK-3β (Figure 1a). Cre expression was found in a large proportion of WAP-DKO cells as assessed by immunohistochemistry (IHC) (Figure 2). Immunoblotting of pregnant WAP-DKO whole-gland lysates demonstrated excision of the GSK-3β FL/FL allele (Figure 1b). Unrecombined GSK-3β present in the non-epithelial, stromal compartment (e.g., fat, extracellular matrix, immune cells) of mammary tissues not targeted by WAP-Cre likely accounts for the observed residual GSK-3β protein.
β-catenin is a major GSK-3 target implicated in multiple stages of mammary development as well as in oncogenesis. For direct comparison with GSK-3 DKO animals, we also generated mice with stabilized β-catenin, which have been characterized previously. 31 In this case, endogenous β-catenin was activated in mammary tissues by WAP-Cre recombination of the floxed exon 3 allele of β-catenin (β-catenin Exon3 LoxP/+ , referred to as β-catenin FL-Ex3/+ ) that encodes for the portion of the amino-terminus containing GSK-3 regulatory phosphodegron sites, thus termed WAP-βcatenin Active .
WAP-Cre-mediated inactivation of GSK-3 in the murine mammary epithelium results in squamous transdifferentiation Approximately 81% of primiparous midpregnant (dpc 15) WAP-DKO glands exhibited pronounced epidermoid transdifferentiation (Table 1, Figure 1c), similar to that found in WAP-βcatenin Active , in agreement with previous reports. 31,61 The mosaic distribution of WAP-Cre, reflecting the heterogeneity in the synthetic activities of cells within individual alveoli during estrous and at midpregnancy, likely accounts for the variability in the phenotypes observed and also strongly suggests these effects are cell autonomous. Mammary intraepithelial neoplasia (MIN) was a predominant feature of WAP-DKO glands where lumens of alveoli were filled with atypical cells in a solid pattern. Multilayered structures exhibiting squamous differentiation with central ghost cells (large masses of 'shadow' cells or cellular ghosts lacking nuclear and cytoplasmic details with clear conservation of basic cellular outline) and keratinization were also frequently observed and appeared to arise from within MIN lesions (Figure 1c). In contrast, MIN was rare in WAP-β-catenin Active glands, which were found to contain squamous differentiation as well as prominent keratinization with many of the larger lesions completely replaced by ghost cells (Figure 1c). Whereas lesions in dpc 15 WAP-DKO glands appeared to arise from both the alveolar and ductal secretory epithelium, ducts seemed to be preferentially affected in WAP-β-catenin Active (Figure 1c). Unsurprisingly, the transdifferentiated glands of WAP-DKO and WAP-β-catenin Active females were unable to support nourishment of pups and litters were found dead within 24 h of birth with no evidence of milk spots.
To examine the nature of effects of GSK-3 deletion on mammary gland development, we enriched for primary mouse MECs from GSK-3α − / − ; GSK-3β FL/FL mice and infected them with adenovirus expressing Cre-GFP (Ad-Cre-GFP) or GFP alone (Ad-GFP) as a control. We then introduced paired groups of KO and control MECs into the contralateral cleared mammary fat pads of 3-week-old FVB mice and allowed them to develop for a period of 8 weeks. After infection with Ad-GFP, floxed MECs generated normal mammary epithelial outgrowths upon transplantation ( Figure 1d). In contrast, after infection with Ad-Cre-GFP, squamous foci were once again observed in fat pads transplanted with GSK-3 KO MECs (Figure 1d). Similarly, epidermoid transdifferentiation was also found in outgrowths from β-catenin Active MECs following injection of Ad-Cre-GFP-infected β-catenin FL-Ex3/+ MECs ( Figure 1d). These data provide evidence that effects of mammary tissue-specific ablation of GSK-3 are intrinsic to the mammary epithelium.
Activation of Wnt/β-catenin signaling in mammary glands lacking GSK-3 More detailed analysis revealed a correlation between the level of β-catenin expression and changes in cell morphology in WAP-DKOs and in WAP-β-catenin Active (Figure 2), as β-catenin was primarily found surrounding keratinized structures. Nuclear β-catenin was never observed in control (GSK-3α − / − ; GSK-3β FL/FL ) virgin glands in which it is normally maintained at a low level and is localized primarily at the membrane ( Figure 2). As assessed by increased BrdU incorporation and Ki67 staining, dpc 15 WAP-DKO glands were actively proliferating ( Figure 2), beyond the level associated with normal mammary growth during pregnancy. Variable expression of cytokeratin 8/18 (K8/18) and cytokeratin 14, markers of luminal and myoepithelial lineages, respectively, was maintained within the MIN lesions ( Figure 2). Keratinizing epithelium of both WAP-DKO and WAP-β-catenin Active was associated with intense expression of cytokeratin 6 (K6) (Figure 2), which was never observed in control glands. On the basis of its expression pattern, which is observed early in embryonic mammary development and in non-proliferating terminal end buds but rare in mature glands, K6 is considered a putative progenitor marker in the mammary epithelium. 27,62,63 However, because K6 is also found in activated keratinocytes in the epidermis, 63 it is difficult to discern whether the observed elevation of K6 in WAP-DKO and WAP-β-catenin Active represents an expansion of immature cells or is a consequence of epidermoid transdifferentiation. Flow cytometric analysis of MECs purified from dpc 15 glands of both strains using CD24 and CD49f mammary stem cell markers did not reveal abnormalities in the stem cell profile where the ratios of luminal (CD24 hi :CD49f + ) to basal (CD24 lo :CD49f hi ) cells were similar to control glands (data not shown). This result is consistent with positive K6 staining reflecting a switch to epidermoid marker expression in transdifferentiated cells rather than aberrations within the mammary stem cell compartment. These observations correlate β-catenin stabilization with proliferation, neoplastic lesion formation and alterations in the differentiation status of the mammary epithelium. Taken together, these results show that the two GSK-3 isoforms, by restraining β-catenin levels and ensuing proliferation, play a key role in maintaining glandular cell fate in the mammary epithelium.
Loss of GSK-3 and activation of β-catenin have differential effects on mammary stem cells To further investigate GSK-3 functions, we employed a mammosphere (MS) assay to measure in vitro stem/progenitor cell frequency in primary MEC preparations. 64 In the absence of attachment to an exogenous substratum or cell-cell adhesion, stem cells are able to survive and proliferate and we found no significant differences in primary or secondary MS formation of GSK-3 DKO MECs generated by infecting Lin − GSK-3α − / − ; GSK-3β FL/FL cells with either Ad-GFP or Ad-Cre-GFP. Interestingly,   Figure 4c). There was no statistical difference in survival curves for these strains (P = 0.2333 log-rank test, P = 0.3706 Wilcoxon test) (Figure 4c). Near-complete loss of GSK-3β in WAP-DKO tumors was confirmed by detection of markedly elevated β-catenin levels by immunoblotting ( Figure 4a). The primary tumors in WAP-DKO and WAP-β-catenin Active were composed of poorly differentiated glandular structures of variable size and large numbers of multilayered squamous epithelial structures ( Figure 4b). The neoplastic cells had a high nuclear-tocytoplasmic ratio (1:1), a moderate amount of granular basophilic cytoplasm and a round to oblong large nucleus with fine chromatin and one to three prominent nucleoli. There was marked anisocytosis (variation in cellular size) and anisokaryosis (variation in nuclear size). Occasional, large multinucleated cells were also present. Nearly half of the tumor volume was composed of multilayered epithelial structures with squamous differentiation and central ghost cells. Some of these structures contained lamellar keratin material and in rare areas, had features reminiscent of normal stratified, keratinized, squamous epithelium (epidermis). Invasion to the regional lymph node was observed occasionally in WAP-DKO and WAP-β-catenin Active tumors (data not shown).
As expected, β-catenin was elevated specifically in WAP-DKO tumors as compared with adjacent normal mammary tissue (Figure 4a). Tumor expression of ducto-luminal (K8/18) and ductobasal (K14) cells was maintained along with K6 and EGFR ( Figure 5). A high level of proliferation was evident in all tumors based on Ki67 staining ( Figure 5).
We found~10% rate of pulmonary metastasis in WAP-βcatenin Active , whereas none of the WAP-DKO tumors were ever found to metastasize (Table 1). Further assessment of metastatic potential showed Lin − tumor cells from two of eight WAP-β-catenin Active were able to regenerate tumors when injected into syngeneic recipients suggesting their weak malignant propensity (data not shown). In contrast, none of WAP-DKO Lin − tumor cells grew following transplantation.
Tumor incidence was also noted in 20% of glands lacking three of the four GSK-3 genes, that is, of genotype WAP-Cre; GSK-3α − / − ; GSK-3β FL/+ ; or WAP-Cre; GSK-3α − /+ ; GSK-3β FL/FL (referred to as ¾ WAP-DKO) arising with a median latency of 14 months, which was statistically significantly different from WAP-DKO (P = 0.0015 log-rank test, P = 0.0075 Wilcoxon test) (Table 1, Figure 4c). No tumors were observed in animals with either just the α or β isoform of GSK-3 missing. The markedly different morphology of ¾ WAP-DKO tumors (Figure 4b) and their low incidence may reflect these being spontaneous age-related tumors previously reported in the FVB strain rather than attributable to loss of GSK-3. This may also be true of the single observed case of tumor formation in virgin WAP-DKOs (latency~13.6 months). However, we cannot exclude the possibility that loss of three of the four alleles causes weak tumor predisposition.
From these data, we conclude that depletion of all four alleles of GSK-3 from the mouse mammary epithelium leads to rapid formation of adenosquamous carcinomas indicating GSK-3 functions to inhibit tumorigenesis within this compartment.
To determine whether simultaneous ablation of GSK-3 and β-catenin affected the propensity of MECs to form MSs, GSK-3α − / − ; GSK-3β FL/FL ; β-catenin FL-Ex2-6/FL-Ex2-6 MECs were infected with either Ad-Cre-GFP or Ad-GFP. Whereas MS formation was unaffected by Ad-GFP, no spheres were ever observed in TKO MECs (Figure 6c). Analysis of the CD24:CD49f stem cell profile of these TKO MECs did not reveal any aberrations (Figure 6c). To evaluate the possibility that membrane/adhesion functions of β-catenin are necessary for sphere formation, we subjected β-catenin FL-Ex2-6/FL-Ex2-6 MECs to the MS assay and found sphereforming ability was normal (data not shown). Thus, in this shortterm assay, loss of GSK-3 and β-catenin abrogates stem cell functionality without perturbation to the stem cell compartment.
After~10 months, multiparous WAP-TKO dams developed mammary tumors (Table 1, Figure 7a). This temporal incidence was statistically significantly different from both WAP-DKO and WAP-β-catenin Active tumor latency (P o 0.0001 log-rank and Wilcoxon tests) (Figure 7a). Unlike WAP-DKOs, which developed adenosquamous carcinomas with a prominent squamous component, acinar adenocarcinomas formed in WAP-TKOs with rare and multifocal squamous foci present in some tumors (Figure 7c). Both mammary lineage markers continued to be expressed in these tumors along with K6, SMA, EGFR and Ki67 (Figure 7e). GSK-3β expression was virtually undetectable in these tumors (Figure 7b) and laser-capture micro-dissection demonstrated efficient excision at the β-catenin locus in both the adeno and squamous components of tumors (data not shown). Interestingly, we found elevated levels of γ-catenin/plakoglobin (Figure 7b), an armadillo   were allowed to form spheres and subsequently treated with escalating doses of γ-secretase inhibitor (Figure 7d). Even at low doses of γ-secretase inhibitor, a reduction in the number of tumorsphere was observed, whereas secondary tumorsphere formation was completely abrogated (Figure 7d). Taken together, these findings provide evidence that in the absence of GSK-3, strong selective pressure exists for increases in γ-catenin along with induction of major developmental signaling pathways, Hh and Notch, which may underlie β-catenin-independent tumorigenic effects of GSK-3 disruption.

DISCUSSION
GSK-3 is a primary negative regulator of β-catenin and we anticipated similar phenotypes of GSK-3 inactivation compared to genetic activation of β-catenin. Given extensive literature demonstrating that activation of Wnt signaling in mammary epithelium induces squamous differentiation, 31,36,37,61,66 it was not surprising that the loss of GSK-3 also conferred squamous potential on mammary cells. This phenotype is cell-autonomous and although strikingly similar to that observed upon direct stabilization of β-catenin, at the stem cell level GSK-3 also exhibits distinct functions. Elevation of β-catenin results in the generation of a CD24 hi population not found in the absence of GSK-3 and this may contribute to the malignant character of β-catenin tumors. We speculate that MEC response is largely determined by the specific level of β-catenin and although a lower threshold may be required for epidermoid transdifferentiation common to both WAP-DKOs and WAP-β-catenin Active glands, a stronger β-catenin signal is required for changes associated with malignancy.
GSK-3 inhibits mammary tumorigenesis by restraining signaling via multiple pathways Deletion of GSK-3 and the ensuing elevation in β-catenin levels resulted in adenosquamous carcinomas in WAP-DKOs. Although not previously reported, stabilization of endogenous β-catenin in the mammary epithelium also led to tumors in the present study. These WAP-β-catenin Active tumors were histologically similar to those of WAP-DKOs and developed with similar kinetics suggesting that stabilization of β-catenin is the key requisite step for WAP-DKO tumor formation. Thus, elevation of β-catenin may permit MECs to re-enter the cell cycle resulting in MIN lesions that progress to adenosquamous carcinomas.
Analysis of β-catenin-independent tumors formed in the absence of GSK-3 revealed markedly elevated expression of γ-catenin/plakoglobin, suggesting a potential compensation mechanism for loss of β-catenin signaling and/or for its role in cell adhesion. This is consistent with recent findings in embryonic stem cells where γ-catenin was found to be responsive to GSK-3 inhibition and can activate TCF target genes when the levels of its expression reach a critical threshold. 67 However, despite their interactions with common cellular partners, β-catenin has well-documented oncogenic potential whereas γ-catenin was characterized as having tumor suppressor activity that may be independent of its role in mediating cell-cell adhesion. 68   Human tumor analysis has shown loss or reduced expression of γ-catenin is associated with poor clinical outcome and increased tumor formation and metastasis. [71][72][73][74] The capacity of γ-catenin to suppress motility and migration may also underlie the lack of metastasis observed in WAP-TKOs and possibly the observed lack of MS formation in TKO MECs.
Despite what may be discerned from the volume of literature linking GSK-3 and Wnt, this protein kinase acts as a nexus in the control of several critical cellular homeostatic mechanisms, in addition to those supplied by Wnt/β-catenin. Our results suggest that GSK-3 plays an important role in mediation of Hh signaling as loss of GSK-3 resulted in elevated Gli1 expression and associated signaling. In addition, enhanced Notch signaling in the absence of GSK-3 may contribute to tumor formation through dysregulation of TIC control. We anticipate that effects of GSK-3 on Wnt, Notch and Hh signaling are ongoing, however, different proportions of the total GSK-3 pool may participate in the regulation of these pathways. GSK-3 is known to localize to different cellular compartments and may be sequestered in multi-vesicular bodies as recently proposed. 75 Hence, distribution of GSK-3 may be dependent on cellular context and the events integrated by GSK-3 and the relative strength of their output will determine observed cellular responses. Whereas WAP-TKO tumors showed significant activation of Hh and γ-catenin/plakoglobin, these pathways were essentially unchanged in WAP-DKO tumors suggesting GSK-3 predominantly functions within the β-catenin destruction complex to regulate the Wnt pathway when β-catenin is intact.

Generation of mouse strains used in the study
All animals were housed at the Toronto Centre for Phenogenomics (www. phenogenomics.ca). GSK-3α − / − and GSK-3β Exon2 LoxP/Exon2 LoxP (referred to as GSK-3β FL/FL ) mutants and their wild-type littermates were generated as previously described. 76 WAP-TKO WAP-Cre mice were also mated to β-catenin Exon 3 LoxP/+ (referred to as β-catenin FL-Ex3/+ ) transgenics (obtained with permission from Makoto M. Taketo and Derek van der Kooy, University of Toronto) previously described 78 to generate WAP-β-catenin Active mice.

BrdU injection
Intraperitoneal injections of mice were performed using BrdU (BD Pharmingen, Mississauga, ON, Canada) at 100 μg/g body weight.
Incorporation of BrdU was detected 4 h post injection using IHC as described in 'Histology and immunostaining of tissue sections'.
Mammary epithelial single-cell preparation, enrichment and flow cytometry analysis Histology and immunostaining of tissue sections Mammary glands (thoracic) were harvested at indicated developmental time points. Tumors were harvested at tumor endpoint. Tissues were fixed in 4% paraformaldehyde overnight at 4°C and placed in 70% ethanol until they were paraffin-embedded. Five micrometer paraffin sections were stained with hematoxylin and eosin. Tissue sections were deparaffinized in two 5-min changes of xylenes, rehydrated in graduated ethanols and then washed in phosphate-buffered saline (PBS). With the exception of EGFR staining where 1 mM EDTA (pH 8) was used, antigen retrieval was performed using 10 mM citrate buffer (pH 6) in a microwave at 10 min boiling following 10 min sub-boiling. Immunoblotting Fifty milligram of tissue or tumor was lysed in 1 ml of RIPA lysis buffer supplemented with complete protease inhibitor tablet (Roche, Laval, QC, Canada) and phosphatase inhibitor cocktail (Sigma) using TissueLyser (Qiagen, Mississauga, ON, Canada). Lysates were cleared by centrifugation for 15 min at 4°C and protein concentration determined by Lowry assay (Bio-Rad, Mississauga, ON, Canada). Protein lysates (10-30 μg) containing SDS sample buffer were boiled for 5 min and loaded onto an 8 or 10% SDS-PAGE gel, followed by semi-dry transfer onto polyvinylidene difluoride membrane (Millipore, Etobicoke, ON, Canada). Blocking was performed for 1 h at RT in 5% non-fat milk in TBST and membranes were probed with primary antibodies (Supplementary Table 1) overnight at 4°C. Following washes with TBST, membranes were incubated with appropriate HRP-conjugated secondary antibodies (Bio-Rad) for 45 min at RT, washed and exposed to film (Kodak, Burnaby, BC, Canada) using ECL reagent (Pierce, Thermo-Fisher Scientific, Burlington, ON, Canada).