EZH2 engages TGFβ signaling to promote breast cancer bone metastasis via integrin β1-FAK activation

Bone metastases occur in 50–70% of patients with late-stage breast cancers and effective therapies are needed. The expression of enhancer of zeste homolog 2 (EZH2) is correlated with breast cancer metastasis, but its function in bone metastasis hasn’t been well-explored. Here we report that EZH2 promotes osteolytic metastasis of breast cancer through regulating transforming growth factor beta (TGFβ) signaling. EZH2 induces cancer cell proliferation and osteoclast maturation, whereas EZH2 knockdown decreases bone metastasis incidence and outgrowth in vivo. Mechanistically, EZH2 transcriptionally increases ITGB1, which encodes for integrin β1. Integrin β1 activates focal adhesion kinase (FAK), which phosphorylates TGFβ receptor type I (TGFβRI) at tyrosine 182 to enhance its binding to TGFβ receptor type II (TGFβRII), thereby activating TGFβ signaling. Clinically applicable FAK inhibitors but not EZH2 methyltransferase inhibitors effectively inhibit breast cancer bone metastasis in vivo. Overall, we find that the EZH2-integrin β1-FAK axis cooperates with the TGFβ signaling pathway to promote bone metastasis of breast cancer.

B reast cancer is the most commonly diagnosed cancer in female individuals worldwide 1 . About 50-70% of breast cancer patients with late-stage disease develop bone metastases that cause skeletal-related events, including pain, pathological fractures, spinal cord compression, hypercalcemia, and other complications 2 . The treatments for bone metastasis are limited and merely palliative; standard antiresorptive agents, chemotherapy, and radiotherapy can delay or lessen skeletalrelated events, but they cannot cure bone metastasis 3 . Exploring the molecular mechanism of bone metastasis comprehensively may provide additional therapeutic strategies for patients with bone metastasis. Breast cancer bone metastasis frequently induces osteolytic lesions, which lead to massive bone resorption and bone fractures 4 . Osteolytic bone resorption causes secretion of several growth factors, including transforming growth factor beta (TGFβ). Bone metastasis is incited by "the vicious cycle", which designates the feed-forward cycle among cancer cells, osteoblasts, and osteoclasts in promoting both uncontrolled tumor growth and osteoclast activity [4][5][6] .
TGFβ plays dual roles in cancer initiation and progression: it works as a tumor suppressor in premalignant cells but induces breast cancer metastasis by enhancing epithelial-mesenchymal transition, angiogenesis, and immunosuppression 7,8 . Studies have well established that TGFβ is a predominant cytokine driving the feed-forward vicious cycle to promote metastatic cancer cell growth in bones 9 . In canonical TGFβ signaling, active TGFβ binds to its receptor, TGFβ receptor type II (TGFβRII), which binds and activates TGFβ receptor type I (TGFβRI) on the cell membrane. TGFβRI phosphorylates downstream signaling molecules Smad2/3, which form a complex with Smad4; the Smad2/3/4 complex is then translocated to the nucleus. The nuclear Smad2/3/4 complex works as transcription factors to turn on the transcription of target genes 10,11 . Noncanonical TGFβ signaling works as a Smad-independent pathway through activation of p38 mitogen-activated protein kinase (MAPK), extracellular signal-regulated kinase (ERK), c-Jun N-terminal kinase (JNK), or phosphoinositide 3-kinase (PI3K)/AKT signaling 12 .
EZH2 is a histone methyltransferase that serves as an enzymatic subunit of the polycomb repressive complex 2 13 . It regulates gene expression through trimethylation of histone H3 at lysine (K) 27 (H3K27me3) or as a transcription co-factor 13 . Overexpression of EZH2 is correlated with metastasis of solid tumors such as prostate and breast cancers 14,15 and is considered a prognostic biomarker of metastasis risk in women with earlystage hereditary breast cancer 16 . It is reported that EZH2 was highly expressed in tissues of renal cell carcinoma obtained from patients who had bone metastases 17 , suggesting that EZH2 promotes cancer cell bone metastasis. However, the function of EZH2 in the vicious cycle of breast cancer bone metastasis is unknown.
Here we found that depletion of EZH2 blocked breast cancer bone metastasis in vivo. Under TGFβ stimulation, EZH2 increased the level of pS465/467-Smad2 and the expression of parathyroid hormone-like hormone (PTHLH, also named parathyroid hormone-related protein, PTHRP), two key effectors of the canonical TGFβ pathway. Mechanistically, EZH2 increases the transcription of integrin β1-encoding ITGB1 that activates a downstream effector, focal adhesion kinase (FAK). Activated FAK phosphorylates TGFβRI and enhances the binding of TGFβRI to TGFβRII to activate the TGFβ signaling pathway. Our study revealed the cooperation between EZH2 and TGFβ signaling in promoting bone metastasis of breast cancer through a methyltransferase-independent mechanism, and demonstrated that targeting FAK may be an effective strategy for treatment of EZH2-induced breast cancer bone metastasis.

Results
EZH2 promotes breast cancer bone metastasis, which cannot be blocked by an EZH2 methyltransferase inhibitor. To explore the function of EZH2 in bone metastasis of breast cancer, we transfected either EZH2 shRNA or control shRNA into the MDA-MB-231 bone-seeking 231-1566 cell subline that expresses GFP and luciferase 7 to generate the EZH2-knockdown cell lines 1566.shEZH2 and its control cell line 1566.shScr, respectively ( Supplementary Fig. 1a). The sublines 1566.shEZH2 and 1566.shScr were injected, separately, into the left ventricles of nude mice. Mice injected with 1566.shEZH cells had significantly longer bone metastasis-free survival (P = 0.0047) and overall survival (P = 0.0024) than did mice injected with 1566.shScr cells (Fig. 1a). Bioluminescence imaging (BLI), X-ray imaging, and hematoxylin and eosin (H&E) staining of bone lesions all showed that mice injected with 1566.shEZH cells had fewer bone metastases than did mice injected with 1566.shScr cells on the same day post-injection ( Fig. 1b). Using the CRISPR/CAS9 system, EZH2-knockout MDA-MB-231 cell subclones (231.KO) and their control clone (231.sgCtrl) were also generated (Supplementary Fig. 1b), and one of the 231.KO subclones (231.KO#1) were stably re-expressed with wild-type EZH2 (231.KO#1.EZH2) or a pLenti control vector (231.KO#1.pLenti) ( Supplementary Fig. 1c). The derived sublines 231.KO#1.EZH2 and 231.KO#1.pLenti were intracardially injected into nude mice, respectively. Mice injected with 231.KO#1.EZH2 cells were treated with a vehicle or GSK126, a potent small-molecule EZH2 methyltransferase inhibitor, whereas the control mice injected with 231.KO#1.pLenti were only treated with a vehicle. The results showed that the vehicle-treated 231.KO#1.EZH2 group had significantly poorer bone metastasis-free survival rates than did the control 231.KO#1.pLenti group (Fig. 1c, d). Unexpectedly, the GSK126-treated 231.KO#1.EZH2 group had similar metastasis-free survival rate to that in the vehicle-treated 231.KO#1.EZH2 group (Fig. 1c, d). The data indicated that EZH2 overexpression increased the incidence of bone metastasis, which cannot be deterred by inhibiting EZH2 methyltransferase function with GSK126. Furthermore, we knocked out EZH2 in the 231-1566 cell subline, generated EZH2-knockout single clones #1 and #2 ( Supplementary Fig. 1d) and mixed them together as the 1566.KO cell subline. We labeled 1566.KO cells and their control cells 1566.Ctrl with GFP and luciferase, and injected them intracardially into mice to generate bone metastases. Mice injected with 1566.KO cells had significantly longer bone metastasis-free survival (P = 0.0016) and overall survival (P < 0.0001) than did mice injected with 1566.Ctrl cells (Fig. 1e, f, Supplementary Fig. 1e), and the data echo that of EZH2 shRNA knockdown in 231-1566 cells (Fig. 1a, b). EPZ-6438 (Tazemetostat) is another potent, and selective EZH2 methyltransferase inhibitor 18 , which is under several clinical trials (ClinicalTrials.gov Identifier: NCT01897571, NCT03009344) for the treatment of advanced solid tumors or lymphomas. Those 1566.Ctrl cell-injected mice were also treated with EPZ-6438 (250 mg/kg, twice/day, oral), or vehicle beginning at day 6 post-injection. Similar to findings from GSK126treated mice bearing 231.KO#1.EZH2 bone metastasis, EPZ-6438 treatment did not deter bone metastasis incidence and progression of 1566.Ctrl cell-injected mice as there was no significant difference in the bone metastasis-free survival, overall survival and BLI images between the vehicle-versus EPZ-6438treated groups (Fig. 1e, f, Supplementary Fig. 1e). Additionally, we pooled five of MDA-MB-231 EZH2 knockout single clones together and labeled them with GFP and luciferase (231.KO mixed) ( Supplementary Fig. 1f), which were intratibially injected into nude mice with 231 cells as controls. Clearly, EZH2 knockout significantly inhibited bone metastasis outgrowth ( Supplementary Fig. 1g, h). The above loss-and gain-of EZH2 function in vivo bone metastasis experiments demonstrated that EZH2 promoted breast cancer bone metastasis and that EZH2's effect on bone metastasis is likely methyltransferaseindependent.
To explore the mechanism of EZH2 promotion of bone metastasis, we first compared the proliferation, migration, and invasion ability of high and low EZH2-expressing MDA-MB-231 cell sublines. High EZH2-expressing cells ( had similar rates of proliferation in two-dimensional cell culture as measured using MMT assays ( Supplementary Fig. 1i). However, high EZH2-expressing cells had greater migration and invasion ability in vitro than did low EZH2-expressing cells (Fig. 1g, Supplementary Fig. 1j Fig. 1c), to test EZH2 methyltransferaseindependent function. Re-expression of the wild-type EZH2 (231.KO#1.EZH2) greatly increased cell migration and invasion compared to the control 231.KO#1.pLenti cells (Fig. 1h, Supplementary Fig. 1n). And re-expression of methyltransferase-dead mutant EZH2 H689A in 231.KO#1 (231.KO#1.H689A) also promoted migration or invasion just like that of re-expression of wild-type EZH2 (231.KO#1.EZH2) (Fig. 1h, Supplementary  Fig. 1n). Similarly, GSK126 treatment didn't have any inhibitory effect on migration or invasion of 231.KO#1.EZH2 cells (Fig. 1h, Supplementary Fig. 1n). These data suggested that EZH2 promoted MDA-MB-231 cell migration and invasion but did not change cell proliferation in two-dimensional cell culture. Furthermore, blocking EZH2's histone methyltransferase function, either genetically by EZH2 H694A mutation or with EZH2 methyltransferase inhibitor GSK126, did not inhibit the migration or invasive ability induced by wild-type EZH2.
To expand the investigation of EZH2's effect on bone metastasis, we also established CRISPR/CAS9-mediated EZH2knockout subclones in 4T1 mouse mammary tumor cells (4T1.KO #1 and #2) ( Supplementary Fig. 1o) and examined their proliferation, migration, and invasion compared with those of the control 4T1 cells. Knocking out EZH2 inhibited 4T1 cell migration and invasion but did not have an apparent effect on cell proliferation ( Supplementary Fig. 1p, q). GSK126 treatment of 4T1 cells did not result in different proliferation, invasion, or migration from untreated 4T1 cells ( Supplementary Fig. 1p-r). Together, our data from both MBA-MD-231 and 4T1 cells showed that EZH2 promoted cancer cell migration and invasion, but this function is unlikely dependent on EZH2's methyltransferase activity.
Next, we examined whether EZH2 promoted bone metastasis outgrowth in vivo by increasing tumor cell proliferation or/and inhibiting apoptosis by IHC staining of Ki67 and cleaved caspase 3, which showed that the bone metastasis of EZH2 knockdown 1566.shEZH2 cells had significantly decreased proliferation and increased apoptosis compared to that of 1566.shScr cells ( Supplementary Fig. 2h). Furthermore, we intratibially injected 231.KO#1.EZH2 and 231.KO#1.H689A cells into mice and monitored the bone metastasis outgrowth. We found that 231.KO#1.EZH2 and 231.KO#1.H689A cells induced bone metastasis lesions similarly (Fig. 2f). Together, data from both cell models and in vivo experiments indicated that EZH2 promoted the vicious cycle of breast cancer bone metastasis, which cannot be blocked by EZH2 methyltransferase inhibitor or the EZH2 H689A methyltransferase dead mutation.
Parathyroid hormone-like hormone (PTHLH, also named PTHRP) is an essential mediator of breast cancer bone metastasis, and metastatic cancer cells secrete PTHLH into the bone microenvironment to activate osteolysis 21 . Knockout of EZH2 in both MDA-MB-231 and 4T1 cells reduced their PTHLH mRNA expression under TGFβ treatment as measured by qRT-PCR (Fig. 2g, h), whereas GSK126 treatment of 4T1 cells didn't reduce Pthlh mRNA expression (Fig. 2h). Besides PTHLH, IL-8 is a cytokine that also regulates osteolysis in breast cancer bone metastasis 22 . Knockout of EZH2 in MDA-MB-231 and 4T1 cells also inhibited their IL-8 mRNA expression, but GSK126 treatment didn't change it ( Supplementary Fig. 2i, j). These data indicated that EZH2 facilitates PTHLH and IL-8 expressions, which can mediate the vicious cycle of breast cancer bone metastasis.
pY397-FAK induces TGFβRI tyrosine phosphorylation that enhances its binding to TGFβRII in response to TGFβ. We further explored the mechanism of how EZH2-mediated FAK activation induces pS465/467-Smad2. Smad7 can block the TGFβRI-induced pS465/467-Smad2 29 , but knocking down EZH2 did not change Smad7 expression (Fig. 3d). Thus, we assessed whether FAK regulates TGFβRI and TGFβRII expressions in MDA-MB-231 cells. Knocking down FAK didn't change the protein expressions of TGFβRI or TGFβRII ( Supplementary  Fig. 4a, b). Next, we tested whether FAK can bind to Smad2 to phosphorylate Smad2 by immunoprecipitation (IP) of FAK followed with western blotting of Smad2, which didn't show detectable binding (Supplementary Fig. 4c). Surprisingly, FAK IP brought down TGFβRI, not TGFβRII, with or without TGFβ exposure (Fig. 4a). Additionally, blocking FAK kinase activity by FAKi VS-4718 treatment reduced the binding of FAK to TGFβRI (Fig. 4b). Since TGFβ treatment induces TGFβRI binding to TGFβRII (Supplementary Fig. 4d) and, consequently, increased pS465/467-Smad2 levels, we postulated that pY397-FAK may phosphorylate TGFβRI that increases the binding affinity of TGFβRI for TGFβRII under TGFβ stimulation, leading to activation of Smad2 signaling. To test this, MDA-MB-231 cells were treated by the FAKi VS-4718 or had FAK knockdown by shRNAs (shFAK#21 or shFAK#3), and treated with TGFβ. After collecting cell lysates, we performed TGFβRI IP followed by western blotting of TGFβRII, which showed that FAK inhibition dramatically reduced the binding of TGFβRI to TGFβRII in response to TGFβ stimulation ( Fig. 4c and Supplementary Fig. 4e, f). To explore whether FAK tyrosine kinase can phosphorylate TGFβRI, we performed IP to pull down TGFβRI from MDA-MB-231 cells (231.TGFβRI) or HEK 293FT cells transfected with exogenous FLAG-tagged wild-type TGFβRI (293FT.TGFβRI), then western blotting with anti-phospho-tyrosine antibodies. We detected tyrosine phosphorylation on TGFβRI ( Supplementary Fig. 4g, h), which were reduced by FAKi VS-6063 treatment (Fig. 4d). The data indicated that activated FAK can induce tyrosine phosphorylation of TGFβRI. Next, we performed mass spectrometric analysis to locate the site of FAK-induced tyrosine phosphorylation on TGFβRI. We identified an unreported TGFβRI phosphorylation site at tyrosine 182 (pY182), which is located in the glycine and serine residues enriched-domain (GS domain) 30 (Fig. 4e). Notably, it is known that after TGFβ binding, activated TGFβRII phosphorylates TGFβRI in the GS domain to activate TGFβRI kinase function and the transduction of TGFβ signals 31 .
Structural analysis revealed that Y182 of TGFβRI is highly exposed for potential phosphorylation (Fig. 4f), and it is close to two threonine and one serine sites (T185, T186, S187) in TGFβRI (Fig. 4g), which are bound and phosphorylated by TGFβRII 31,32 . Our protein docking suggests that Y182 of TGFβRI is oriented toward K381 of TGFβRII with a distance of 4.1 Å (Supplementary Fig. 4i left, and Fig. 4g left). Upon phosphorylation, the distance of the negatively charged phosphate group of TGFβRI to positively charged K381 of TGFβRII becomes much closer (2.1 Å), therefore, significantly enhancing binding of TGFβRI to TGFβRII through increased charge-charge interactions with positively charged K381, as well as Mg 2+ coordinated with ATP ( Supplementary Fig. 4i right and Fig. 4g right). Consequently, such increased TGFβRI binding to TGFβRII may further promote the phosphate transfer from the TGFβRII-bounded ATP to T185, T186, and S187 in the GS domain of TGFβRI (Fig. 4g). To determine whether phosphorylation of Y182 at TGFβRI changes the binding affinity of TGFβRI to TGFβRII, we stably expressed FLAG-tagged wild-type TGFβRI (RI-WT-FLAG), a nonphosphorylatable Y182F mutant TGFβRI (RI-YF-FLAG), or a phosphomimic Y182D mutant TGFβRI (RI-YD-FLAG) in HEK 293FT cells. The wild-type TGFβRI, mutant TGFβRI Y182F, or TGFβRI Y182D in these cells were pulled down with an anti-FLAG antibody after TGFβ treatment followed by western blotting of TGFβRII. We found that the non-phosphorylatable TGFβRI-Y182F mutant had reduced binding to TGFβRII, compared with wild-type TGFβRI and TGFβRI-Y182D mutant ( Supplementary Fig. 4j, k); phosphomimic TGFβRI-Y182D mutant had slightly increased binding to TGFβRII, compared with wild-type TGFβRI (Supplementary Fig. 4l). Evidently, the phosphorylation of TGFβRI at Y182 enhanced TGFβRI binding to TGFβRII in response to TGFβ stimulation, which is consistent with our protein docking analysis.
EZH2 increases the FAK upstream ITGB1 expression. FAK signaling is initiated by integrin-mediated cell adhesions. Integrins (such as β1 or β3) can facilitate FAK autophosphorylation at tyrosine 397, which increases the catalytic activity of FAK 24,33 . To understand the underlying mechanism of EZH2-induced pY397-FAK, we investigated whether and how EZH2 regulates expression of integrins β1 or β3. Since EZH2 methyltransferase activity is not required for increasing pY397-FAK (Supplementary Fig. 3j) and we recently reported that EZH2 can function as a transcription co-factor of RNA Pol II to upregulate mRNA transcription 34 , we examined whether EZH2 also regulate RNA Pol II transcription of genes encoding β1, β3, or other genes that may regulate pY397-FAK. We analyzed our chromatin IP sequencing (ChIP-seq) dataset (GSE188640) to compare RNA Pol II occupancy of gene promoters between EZH2-expressing MDA-MB-231 cells and 231.KO#1 cells in which EZH2 was knocked out 34 . The result showed that knocking out EZH2 led to reduced binding of RNA Pol II to promoter regions of at least 470 genes (Supplementary Data 2). Among these, binding of RNA Pol II to ITGB1 (encoding integrin β1) promoter is substantially decreased, whereas binding to ITGB3 (encoding integrin β3) showed little changes ( Fig. 5a and Supplementary Fig. 5a). Consistently, qRT-PCR showed that knockdown and knockout of EZH2 downregulated ITGB1 mRNA expression (Fig. 5b, c), resulting in decreased integrin β1 protein expression (Supplementary Fig. 5b, c), while knockdown of EZH2 had no significant effect on β3 mRNA expression (Supplementary Fig. 5d). Also, ITGB1 promoter-driven luciferase reporter gene assays showed that ITGB1 promoter activity was higher in EZH2-expressing MDA-MB-231 and 231.sgCtrl cells than that in EZH2-null cells (231.KO#1 and 231.KO#2) (Fig. 5d)  ChIP-qPCR using a series of PCR primers that bind to various regions of the ITGB1 promoter from −2.6 kb upstream of (primer P1) to near (primer P5), the ITGB1 transcription start site (Fig. 5e, top). In MDA-MB-231 cells, EZH2 was recruited to the ITGB1 promoter from P1 to P5 loci, and expectedly, in 231.KO#1 cells, binding of EZH2 to these loci of ITGB1 promoter was lost (Fig. 5e, bottom). Our ChIP-EZH2-qPCR assays showed that EZH2 binds to ITGB1 promoter almost at the level of EZH2 binding to HOXA9B, a well-known methyltransferase substrate of EZH2 35 , with no binding to a non-substrate gene promoter    Supplementary Fig. 5e). Markedly, RNA Pol II bound well to P1 to P5 loci within the ITGB1 promoter that overlapped with EZH2 binding loci in MDA-MB-231 cells, and RNA Pol II binding to the ITGB1 promoter was also lost in EZH2-knockout 231.KO#1 cells (Fig. 5f), indicating EZH2 is required for RNA Pol II binding to the ITGB1 promoter. Similarly, shRNAmediated knocking down of EZH2 in 231.shEZH2#3 and 231.shEZH2#4 cells also reduced the RNA Pol II binding at P1 to P4 loci of the ITGB1 promoter compared to control 231.shScr cells (Fig. 5g), which paralleled with the reduced EZH2 binding (Supplementary Fig. 5f). Expectedly, EZH2 knockdown in 231.shEZH2#3 and 231.shEZH2#4 cells reduced both EZH2 binding and H3K27me3 binding to, but increased RNA Pol II binding at, the HOXA9B promoter, compared to control 231.shScr cells; However, H3K27me3 binding to the ITGB1 promoter were similar in control 231.shScr cells versus EZH2knockdown cells (Supplementary Fig. 5g), further indicating EZH2 regulated ITGB1 independent of its methyltransferase function. Taken together, both EZH2 and RNA Pol II bind to the same promoter regions of ITGB1, and EZH2 is likely functioning as a co-factor of RNA Pol II to upregulate ITGB1 transcription independent of its methyltransferase function.
Since integrin β1 is responsible for FAK activation 24 and activated FAK bound to and phosphorylated TGFβRI (Fig. 4a, d, e), we questioned whether integrin β1 can bind to TGFβRI in the same complex. IP integrin β1 followed by western blotting of TGFβRI and reverse IP TGFβRI followed by western blotting of integrin β1 explicitly showed that integrin β1 can bind to TGFβRI in MDA- Fig. 5h, i), which further demonstrated the cross interactions between the TGFβ/TGFβRI pathway and the integrin β1/FAK pathway. Moreover, our molecular docking between TGFβRI/TGF-β/TGFβRII and integrin αVβ1 complex 36,37 using the ClusPro web server showed that TGFβRI interacts with integrin αVβ1 among their ectodomains and this interaction requires TGFβ (Supplementary Fig. 5j). Interestingly, our IP of TGFβRI from untreated MDA-MB-231 cells followed with western blotting of integrin β1 showed that TGFβRI still can bind with integrin β1 without TGFβ ( Supplementary Fig. 5k), conceivably via cytoplasmic domain. The binding between TGFβRI and integrin β1 without TGFβ stimulation was reduced by FAK inhibitor treatment (Supplementary Fig. 5l), implying that activated FAK can mediate the cytoplasmic binding of TGFβRI with integrin β1. These data suggested that the ectodomain of integrin β1 may bind with TGFβRI through TGFβ; whereas the cytoplasmic domain of integrin β1 may bind with TGFβRI through activated FAK ( Supplementary Fig. 5m).
Treatment with a clinically applicable FAK inhibitor blocks EZH2-induced breast cancer bone metastasis. Our above findings indicated that EZH2, via upregulating integrin β1 transcription, activated FAK, which activated the TGFβ/Smad2 pathway to increase bone metastasis. Our findings prompted us to test FAK inhibitor for treatment of bone metastases of high EZH2 expressing breast cancers. For bone metastasis outgrowth model, GFP-and luciferase-labeled MDA-MB-231 cells, which have relatively high EZH2 expression among tested breast cancer cell lines ( Supplementary Fig. 6a), were intratibially injected into nude mice. We treated these mice with FAKi VS-6063 (50 mg/kg, twice a day, oral gavage), which is currently tested in clinical trials for treating patients with advanced lymphoma or solid tumors (ClinicalTrials.gov Identifier: NCT04439331, NCT03875820). To validate that EZH2-induced breast cancer bone metastasis outgrowth is independent of its methyltransferase function, a group of mice was treated with EZH2 methyltransferase inhibitor GSK126 (100 mg/kg, once a day, i.p. injection). The resulting bone metastasis outgrowth were detected using BLI, which confirmed that GSK126 treatment did not block tumor outgrowth in the bones (Fig. 6a). Excitingly, treatment with the FAKi VS-6063 significantly impeded the outgrowth of bone tumors compared to the control group (P = 0.0442) (Fig. 6a) and did not induce significant side effects ( Supplementary Fig. 6b-e). IHC staining of bone metastasis showed that FAKi VS-6063, but not GSK126, significantly reduced pS465/467-Smad2 (P = 0.0066) level and PTHLH expression (P = 0.0074) in the bone metastases and both drugs effectively inhibited their targets (Fig. 6b, c and Supplementary Fig. 6f). Finally, we examined the GSE2603 dataset and validated that EZH2 expression was negatively correlated with bone metastasis-free survival in breast cancer patients (r = −0.2394, P = 0.03) (Fig. 6d), suggesting that high EZH2 expression in primary breast tumors produces a high risk of developing bone metastasis in patients. We also examined the correlation between the expression of EZH2 and the downstream effector PTHLH in bone metastasis tissues obtained from breast cancer patients in the GSE14020 dataset. We found that EZH2 mRNA expression was positively correlated with PTHLH mRNA expression in patients' bone metastases (r = 0.4630, P = 0.05) (Fig. 6e) but not in metastases to other organ sites (e.g., lung, liver, brain metastases; r = 0.2452, P = 0.097) ( Supplementary  Fig. 6g). This unique effect of EZH2 in promoting bone metastasis may result from dramatically higher TGFβ expression in bone metastasis than that in metastases of other organs and in primary mammary tumors ( Supplementary Fig. 6h), i.e., the enriched TGFβ effectively activate the EZH2/integrin β1/FAK/ p-Smad2 axis to upregulate PTHLH in bone metastasis. Most importantly, the clinical data confirmed that EZH2 high expression can increase PTHLH expression that promotes bone metastasis in patients.

Discussion
As described herein, we revealed a mechanism of how EZH2 promotes breast cancer bone metastasis. Specifically, EZH2 works as a transcription co-factor of RNA Pol II to increase ITGB1 gene transcription; the increased integrin β1 induces phosphorylation of Y397 on FAK leading to FAK activation; activated pY397-FAK phosphorylates TGFβRI at Y182 that increases TGFβRI's binding affinity for TGFβRII in response to TGFβ exposure, thereby triggering pS465/467-Smad2 that induces the downstream effector PTHLH; PTHLH accelerates osteolysis leading to more TGFβ release, and thus driving the feed-forward vicious cycle of breast cancer bone metastasis outgrowth (Fig. 7). Since FAK and TGFβ enhances epithelial-mesenchymal transition and cell migration 8,38 , EZH2-induced FAK/TGFβ signaling activation is also an underlying mechanism of the strong migration and invasion ability of EZH2 high expressing breast cancer cells. Activation of TGFβ signaling by FAK-induced phosphorylation has a critical and distinct effect in enhancing bone metastasis, partially due to the TGFβ-enriched bone microenvironment.
Although the function of TGFβ signaling in the bone metastasis of breast cancer is well-known, the cross-talk between the integrin/FAK and TGFβ pathways is not well-documented. Integrin β6 and β8 were reported to bind with TGFβ latencyassociated peptide (LAP) and convert the latent TGFβ to the active form of TGFβ 39,40 . And it was reported that TGFβ activated FAK through integrin β3 or β1 and leading to p38 MAPK activation in renal cell carcinoma and hepato-carcinoma cells 41,42 . In the present study, we found that integrin β1-FAK is involved in the classical TGFβ/Smad-dependent pathway rather than the p38 MAPK pathway. Also, our data on the binding between integrin β1 and TGFβRI suggested that TGFβ may activate FAK through the TGFβRI-integrin β1 complex as targeting integrin β1 inhibited FAK activating under TGFβ treatment. Administering FAKi and genetically rendering FAK deficient in breast cancer cells abrogated the interaction between TGFβRI and TGFβRII and thereby blocked phosphorylation of Smad2 and expression of its downstream effector PTHLH. Our study demonstrated that the integrin/FAK and TGFβ/TGFβRI/ TGFβRII pathways can cross talk and critically cooperate in driving the feed-forward vicious cycle of breast cancer bone metastasis.
It was reported that FAK and integrin β3 bind to TGFβRII in stellate hepatic cells 43 and breast cancer cells 44 . Here, we uncovered that in EZH2 high expressing breast cancer cells, FAK and integrin β1 bound to TGFβRI rather than TGFβRII and that FAK phosphorylated TGFβRI. Little is known about tyrosine phosphorylation sites in TGFβRI and their functions, although several serine phosphorylation sites have been reported in TGFβRI 45 . Our mass spectrometry analysis identified a previously unreported tyrosine phosphorylation site at Y182 in the GS domain of TGFβRI. Our protein structure analysis and IP/western blotting experiments showed that the Y182 of TGFβRI is important for regulating the binding of TGFβRI to TGFβRII and subsequent TGFβ/Smad2 pathway activation. Interesting, Y182 is conserved in the GS domains of several activin type I receptors of the TGFβ superfamily, such as ALK4 and ALK7, but not in the GS domains of bone morphogenetic protein (BMP) type I receptors of the TGFβ superfamily, such as ALK2. However, it was reported that R206H mutation at the GS domain of ALK2, led to constitutive activation of this receptor and BMP signaling 46 , which can crosstalk to TGFβ signaling in bone formation 47 . This report and our findings indicated that modifications of the GS domain, either the activating mutation or the phosphorylation, is critical for TGFβ receptor activation and indicates the intricate regulations of downstream pathway.
EZH2 is a classic epigenetic protein that silences tumor suppressors through H3K27me3 13 . Recently, its noncanonical functions in the development of various cancers are gaining increasing attentions. For example, EZH2 can methylate non-histone substrates, such as Jarid2, STAT3, RORα, and PLZF, to regulate their transcription function or protein stability [48][49][50][51] . EZH2 also has functions independent of its histone methyltransferase activities. For example, EZH2 forms a complex with RelA and RelB to activate nuclear factor κB signaling in estrogen receptor-negative breast cancer cells 52 and activates androgen receptor gene transcription through binding at the androgen receptor promoter in prostate cancer cells 53,54 . We recently reported that EZH2 can function as a methyltransferase-independent transcription factor to upregulate c-JUN expression that induced G-CSF to facilitate the brain infiltration of immunosuppressive neutrophils 34 . In the present study, we found that EZH2 upregulated ITGB1 transcription in breast cancer cells by functioning as a transcriptional co-factor of RNA Pol II to facilitate its binding to the ITGB1 promoter. Thus, ITGB1 is a substrate regulated by EZH2 methyltransferase-independent activities.
EZH2 is highly expressed in various human malignancies and regulates tumor progression. Therefore, it is regarded as an attractive therapeutic target in cancer patients 55 . However, targeting EZH2 with methyltransferase inhibitors has not always proven to be beneficial in clinical trials [55][56][57][58] , partially because of the EZH2 methyltransferase-independent functions in promoting cancer development as mentioned above. In the present study, we found that small-molecule EZH2 inhibitors cannot block MDA-MB-231 cell-induced bone metastasis. However, targeting EZH2 downstream effector FAK with clinically applicable kinase inhibitors have striking effects on blocking breast cancer bone metastasis.
EZH2 inhibitors were reported to inhibit breast cancer lung metastasis in mouse models 59,60 , whereas EZH2 inhibitor EPZ-6438 failed to block but promoted bone metastasis in experimental models 61 . These contradictory effects of EZH2 on lung metastasis versus bone metastasis suggest that EZH2 inhibitors' effects are depended on the tumor microenvironment. Bone metastasis is reported to have significantly lower EZH2 activity compared to the lung metastases 61 , suggesting that targeting EZH2 methyltransferase activity by EZH2 inhibitors in the unique bone microenvironment might not yield inhibitory efficacies on bone metastasis. In addition, the methyltransferaseindependent EZH2-Integrin β1-FAK-TGFβ pathway identified in this study might be more predominant in the TGFβ-enriched bone metastasis microenvironment, but not in the lung metastasis microenvironment of lower TGFβ levels. Therefore, the efficacy of targeting EZH2 methyltransferase varies in different metastasis organs.
Since EZH2 plays distinct functions in different types of cancer and in metastases of different organs, targeting downstream effectors of EZH2, or EZH2's enzyme function should be carefully evaluated. We found that FAK is a downstream effector of EZH2 in the vicious cycle of breast cancer bone metastasis. Thus, treatment with a FAKi combined with standard antiresorptive agents, chemotherapy, or radiotherapy may provide added benefit to breast cancer patients who suffer from bone metastasis.
Western blotting and immunoprecipitation. Western blotting and immunoprecipitation (IP) were performed as described previously 64 . Briefly, for western blotting, cells were lysed in lysis buffer (5 M urea, 10% sodium dodecyl sulfate [SDS], DNase-free water: 1:1:1) and then sonicated. The lysates were collected for western blotting analysis. Proteins were separated using SDS-polyacrylamide gel electrophoresis and transferred onto a polyvinylidene difluoride membrane. After each membrane was blocked with 5% milk for 1 h, it was probed with various primary antibodies overnight at 4°C followed by incubation with secondary antibodies for 1 h at room temperature before being visualized with enhanced chemiluminescence reagent. For IP, cells were washed twice with phosphatebuffered saline and scraped into IP lysis buffer (1% Triton X-100, 150 mM NaCl, 10 mM Tris, pH 7.4, 1 mM EGTA, 1 mM EDTA, 0.5 mM sodium orthovanadate, 0.4 mM phenylmethanesulfonyl fluoride, 0.5% NP-40). The total cell lysates were precleared via incubation with protein G-linked agarose beads (Sigma; #1124323300) for 2 h at 4°C. After preclearing, lysates were incubated with the primary antibody overnight at 4°C and then with protein G-linked agarose beads for 1 h at 4°C. The beads were washed with IP buffer three times, and the protein immunocomplex was extracted from agarose and detected using SDSpolyacrylamide gel electrophoresis and western blotting.
Cell proliferation assays. Cell proliferation was measured using a 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. Ten thousand MDA-MB-231 cells or 1000 4T1 cells per well (four wells per sample) were seeded in a 24-well plate, and the cell growth was examined via staining with MTT (Thermo Fisher Scientific; #M6496). The resulting intracellular purple formazan was solubilized using dimethyl sulfoxide and measured its absorbance on a plate reader at 570, measure also at 650 nm as reference wavelength, using a Gen5 microplate reader (BioTek); calculate the signal sample as OD570 minus OD650.
Quantitative reverse transcription-polymerase chain reaction. Total RNA in cells was isolated using TRIzol reagent (Thermo Fisher Scientific; #15596026) and then reverse-transcribed using an iScript cDNA Synthesis Kit (Bio-Rad; #1708891). Chromatin IP-quantitative PCR. Chromatin IP (ChIP)-quantitative PCR (qPCR) was performed as described previously 34 . Briefly, cells were fixed with 37% formaldehyde (final, 0.5%), treated with glycine (final, 125 mM), washed, resuspended in ChIP lysis buffer (50 mM Tris, pH 8.1, 10 mM EDTA, 1% SDS, 1% protease inhibitor cocktail, 10 mM phenylmethanesulfonyl fluoride), and sonicated. Lysates containing soluble chromatin were incubated with antibodies against EZH2, RNA Pol II, H3K27me3, or IgG overnight at 4°C and then incubated for an additional 2 h at 4°C with added protein G-linked agarose beads. The agarose bead-bound complexes were then washed, and the protein-chromatin complexes were extracted from the agarose beads with elution buffer. Reversal of the cross-linking of protein and DNA was performed by incubating the elution buffer with 10 mg/mL RNase A and 5 M NaCl overnight at 65°C followed by incubation with 0.5 M EDTA, 1 M Tris, pH 6.5, and 10 mg/mL proteinase K for 1 h at 45°C. Co-precipitated DNA was purified using a QIAquick spin column (QIAGEN), and 2 μL of DNA was analyzed via qPCR using specific primers for the ITGB1 promoter, which is Triple co-culture assay and TRAP staining. Triple co-culture assay and TRAP staining were performed as described previously 7 . Murine RAW 264.7 preosteoclasts (3 × 10 4 cells/well) were seeded directly into the wells of six-well coculture plates, and MC3T3 cells (3 × 10 4 cells/well) were seeded into Millicell Hanging Cell Culture Inserts (Millipore) in the six-well co-culture plates. The next day, MC3T3 cells attached to the membranes of the inserts, and luciferase/green fluorescent protein (GFP)-labeled (GFP + ) MDA-MB-231, 231.sgCtrl, 231.KO#1, or 231.KO#2 cells (3 × 10 4 cells/well each) or 4T1, 4T1.KO#1, or 4T1.KO#2 cells (500 cells/well each) were added on top of the MC3T3 cell layer in triplicate and treated with 5 ng/mL TGFβ, 2 μM GSK126, or a vehicle. Co-culture assays were performed in Dulbecco's modified Eagle's medium/high-glucose medium supplemented with 10% FBS and that was changed every 2 days. TRAP staining of osteoclasts was performed on day 6 using a leukocyte acid phosphatase kit (Sigma-Aldrich; #387 A). TRAP + multinucleated cells were scored as mature osteoclasts and quantified. MC3T3 cells and GFP + tumor cells were trypsinized from the inserts and calculated GFP + cell numbers using flow cytometry.
Flow cytometry. For detecting TGFβRI on cell surfaces, breast cancer cells were seeded at the same density and collected when they reached 80-90% confluence. One million cells in each sample were washed twice in fluorescence-activated cell sorter (FACS) buffer (1% bovine serum albumin in phosphate-buffered saline), resuspended in 100 μL of FACS buffer, and stained with 1 μg of an anti-TGFβRI antibody or mouse anti-IgG antibody for 1 h, washed twice in FACS buffer, and stained with 1 μg of an APC anti-mouse secondary antibody for 1 h. Afterward, cells were washed with cold phosphate-buffered saline twice and analyzed using a FACSCanto II flow cytometer (BD Biosciences). For a triple co-culture assay, MC3T3 cells and GFP + tumor cells were trypsinized from the inserts in tubes and washed twice in cold FACS buffer, resuspended in 400 μL of FACS buffer, and analyzed using a FACSCanto II flow cytometer.
Migration and invasion assays. For a migration assay, breast cancer cells (30,000 cells/well) in FBS-free medium were placed on the top side of the uncoated membrane of transwell inserts and allowed to migrate through the pores, to the bottom side of the membrane. For an invasion assay, the top side of the membranes of transwell inserts were coated with 14.3% Matrigel for 1 h, and then breast cancer cells (30,000 cells/well) were loaded the same way as for the migration assay. A medium containing 10% FBS was added to the lower compartment of transwell as a chemical attractant. After culturing for 18 h in a migration assay or 24-30 h in an invasion assay, the migrated and invaded cells at the bottom side of the membrane of transwell inserts were stained with crystal violet and counted under a light microscope.
Reverse-phase protein array. RPPA analysis of MDA-MB-231, 231.sgCtrl, 231.KO#1, and 231.KO#2 cells after treatment with a vehicle or 5 ng/mL TGFβ for 2 h was performed at the MD Anderson Functional Proteomics Reverse Phase Protein Array (RPPA) Core. Briefly, cellular proteins were denatured using 1% SDS, serially diluted, and spotted on nitrocellulose-coated slides. Each slide was probed with a validated primary antibody plus a biotin-conjugated secondary antibody. The signal obtained was amplified using a Dako Cytomation-catalyzed system and visualized in a DAB colorimetric reaction. Slides were scanned on a flatbed scanner to produce 16-bit tiff images. Spots from tiff images were identified and the density was quantified by Array-Pro Analyzer software. Each dilution curve was fitted using a logistic model ("SuperCurve Fitting" developed at MD Anderson) 65 and normalized according to median polish. The heatmap included was generated in Cluster 3.0 (http://bonsai.hgc.jp/~mdehoon/software/cluster/ software.htm) as a hierarchical cluster using Pearson Correlation and a center metric. The resulting heatmap was visualized in Treeview (http://www.eisenlab.org/ eisen/) and presented as a high resolution.bmp format.
Mass spectrometry. Liquid chromatography/tandem-mass spectrometry was used to identify phosphorylation sites of TGFβRI. HEK 293FT cells were transfected with pRK5-TGFβRI-FLAG plasmid (Addgene, #14833) using Lipofectamine 3000 transfection kit; HEK 293FT cell lysates were immunoprecipitated with an anti-FLAG antibody. After protein gel electrophoresis, TGFβRI band was excised from the gels and subjected to tryptic digestion. (Matrix Science) run on an in-house server. Search results were trimmed to a 1% FDR for strict and 5% for relaxed condition using Percolator. For the trypsin, up to two missed cleavages were allowed. MS tolerance was set 10 ppm; MS/MS tolerance 0.8 Da. Carbamidomethylation on cysteine residues was used as fixed modification; oxidation of methionine as well as phosphorylation of serine, threonine, and tyrosine was set as variable modifications.
TGFβRI-TGFβRII complex and Integrinβ1-TGFβRI complex modeling by molecular docking. High-resolution crystal structures of the cytoplasmic domain of TGFβRI (PDB ID: 1ias) and the kinase domain of TGFβRII (AMPPNP, an ATP analog, bound state, PDB ID: 5e92) were obtained from Protein Data Bank. A phosphoryl group was added to TGFβRI Y182 using PyTMs 66 . The complex docking models between TGFβRI/pY182-TGFβRI and TGFβRII were built with ClusPro web-based server (http://cluspro.bu.edu/) 67 . The residues that were used as attraction restraint included T185, T186, and S187 of TGFβRI, and AMPPNP601 of TGFβRII was retained. The top ten low-energy docked models from each restraint were downloaded from the server and then visualized and analyzed with PyMOL. Two final models were selected on the basis of substrate recognition and phosphotransfer mechanism from ATP hydrolysis to TGFβRI. Atomic-resolution structures of the extracellular domains of the TGFβRI/TGF-β1/TGFβRII complex (PDB ID: 3KFD) and the extracellular domains of integrin αVβ1 (PDB ID: 3VI3) were downloaded from Protein Data Bank. Structures were uploaded, and molecular docking between the two complexes was then performed with the ClusPro web server (http://cluspro.bu.edu/) 67 . All the cartoon structural presentations in this manuscript were generated and displayed with PyMOL.
Animal experiments. All animal experiments were carried out in accordance with protocols (00001397-RN02) approved by the MD Anderson Institutional Animal Care and Use Committee. The study is compliant with all relevant ethical regulations regarding animal research. Athymic NCr nu/nu mice (strain: #002019) were obtained from Jackson Lab. The mice were exposed to a 12-h light/12-h dark cycle at 22-24°C with 50-60% humidity, bred as specific pathogen-free mice, and given free access to food and water. The number of mice used in each experimental group was determined via power analysis or based on prior experience with metastatic animal models, and mice were grouped randomly for each experiment. All sample sizes were listed in the corresponding figure legend or figures. All mice used were the same age (8 weeks) and had similar body weights. Two different injection models were used for bone metastasis studies. (1) Intracardiac injection model A: 1 × 10 5 cells of the 231-1566 sublines or EZH2-knockout MDA-MB-231 sublines were injected into the left ventricle in anesthetized female athymic NCr nu/nu mice. GSK126 was dissolved in 20% cyclodextrin (Captisol; CyDex Pharmaceuticals) and adjusted to a pH level of 4.0 to 4.5 with 1 N acetic acid following the instructions described by McCabe et al. 56 . GSK126 was administrated to the mice via intraperitoneal (i.p.) injection three times a week at a dose of 150 mg/kg after 5 days of injections. (2) Intracardiac injection model B: 1 × 10 5 cells of the 231-1566 control (1566.ctrl) or 231-1566 EZH2 knockout subclone mixture (1566.KO) were injected into the left ventricle in anesthetized female athymic NCr nu/nu mice. The first mouse group injected with 1566.ctrl were treated with EPZ-6438. EPZ-6438 was dissolved in 0.5% NaCMC + 0.1% Tween 80 in water and administrated to the mice via oral gavage, two times per day at a dose of 250 mg/kg after 6 days of injections. The second mouse group injected with 1566.ctrl and the third mouse group injected with 1566.KO were treated with vehicle as the same schedule as the first group. (3) Intratibial injection model A: 2 × 10 5 MDA-MB-231 cells or 231.KO.mixed cells were injected into a tibia in anesthetized female athymic NCr nu/nu mice. (4) Intratibial injection model B: 2 × 10 5 MDA-MB-231 cells were injected into a tibia in anesthetized female athymic NCr nu/nu mice. VS-6063 was prepared in a vehicle (0.5% hydroxypropyl methylcellulose with 0.1% Tween 80) and administered to the mice via oral gavage (50 mg/kg) twice a day after 18 days of injection, and GSK126 was administered to them via i.p. injection every day at a dose of 100 mg/kg after 18 days of injections. Development of bone metastases and tumor burdens were monitored using bioluminescence imaging (BLI), and endpoints includes weight loss, loss of mobility, and other signs of distress. After anesthetized mice were intraperitoneally injected with 75 mg/kg Dluciferin, BLI was performed using a Xenogen IVIS 200 imaging system (Perki-nElmer). Analysis of bone metastasis was performed using living image software by measuring the photon flux in the hindlimbs of mice. The photon flux curves were normalized according to the signal on the day when mice were given the drug GSK126 or vehicle. Bone metastasis-free survival curves showed the time point at which each mouse experienced bone metastasis development according to threshold BLI signals in the hindlimbs. X-ray images of hindlimbs of mice were obtained using an IVIS Lumina XR system (PerkinElmer).
Immunohistochemistry staining and scoring system. Standard immunohistochemistry (IHC) staining was performed as described previously 68 . The immunoreactive score (IRS) was used to quantify the IHC staining, ranging from 0 to 12 as a result of multiplication of positive cell proportion scores (0-4) and staining intensity scores (0-3). IHC staining and statistical analysis results were independently evaluated by two pathologists blinded to the experimental groups.
Statistics and reproducibility. All quantitative experiments were performed using at least three independent biological repeats, and the results are presented as means ± standard deviation (S.D.) or means ± standard error of the mean (S.E.M.). One-way analysis of variance (multiple groups) or t-tests (two groups) were used to compare the means for two or more samples using the Prism 8 software program (GraphPad Software). Survival was analyzed using Kaplan-Meier curves and logrank tests. P-values < 0.05 (two-sided) were considered statistically significant. For IHC score, ten visual fields from different areas of each tumor were evaluated by two pathologists independently (blinded to experiment groups). For migration, invasion, and TRAP + osteoclasts staining experiments, more than three visual fields were evaluated, and three independent biology repeats Fwere performed. Representative images of micrographs were shown to represent reproducible data from various experiments using micrographs.