Blocking immunosuppressive neutrophils deters pY696-EZH2-driven brain metastases

Abstract

The functions of immune cells in brain metastases are unclear because the brain has traditionally been considered “immune privileged.” However, we discovered that a subgroup of immunosuppressive neutrophils is recruited into the brain, enabling brain metastasis development. In brain metastatic cells, enhancer of zeste homologue 2 (EZH2) is highly expressed and phosphorylated at tyrosine-696 (pY696-EZH2) by nuclear-localized Src tyrosine kinase. Phosphorylation of EZH2 at Y696 changes its binding preference from histone H3 to RNA polymerase II, which consequently switches EZH2’s function from a methyltransferase to a transcription factor that increases c-JUN expression. c-Jun upregulates pro-tumorigenic inflammatory cytokines, including granulocyte-colony stimulating factor (G-CSF), which recruits Arg1-positive and PD-L1-positive immunosuppressive neutrophils into the brain to drive metastasis outgrowth. G-CSF-blocking antibodies or immune checkpoint blockade therapies combined with Src inhibitors impeded brain metastasis in multiple mouse models. These findings indicate that pY696-EZH2 can function as a methyltransferase-independent transcription factor to facilitate the brain infiltration of immunosuppressive neutrophils, which could be clinically targeted for brain metastasis treatment.

One-sentence summary:

G-CSF antibodies and immune checkpoint blockade with Src inhibitors block neutrophil infiltration to impede pY696-EZH2 driven brain metastasis.

Introduction

Brain metastasis is the most common malignancy of the central nervous system (1, 2), and the median survival time of patients with brain metastasis is less than 1 year (3, 4). Recent impressive advances in targeted therapy and immunotherapy have led to better control of systemic disease. However, the incidence of brain metastasis associated with disease recurrence is steadily increasing (5–7), which represents an imposing challenge in the era of precision cancer medicine (1). The co-evolution of metastatic cancer cells with the brain microenvironment is critical for metastatic cells’ escaping dormancy and colonizing the brain (8, 9). The main cell types in the brain—astrocytes, microglia, and neurons—have been reported to regulate metastatic cancer cells’ seeding and outgrowth (10–15). However, whether immune cells, especially peripheral adaptive and innate immune cells, are present and function in the brain tumor microenvironment (TME) was unknown for a long time (16, 17). Recently, it has been noticed that various types of immune cells, especially innate immune cells, can be recruited into the brain TME when the blood-brain barrier is compromised by metastatic cancer cells (18, 19).

Among circulating innate immune cells, neutrophils are the most abundant group (20). Neutrophils’ function in inflammatory responses is well characterized, but their function in tumor progression and metastasis is unclear (21–24). In patients with brain metastasis and with glioblastoma, a high ratio of neutrophils to lymphocytes in the peripheral blood was a biomarker of poor prognosis (25, 26). However, the function of neutrophils in brain metastasis development remains controversial (21, 27). A clear answer regarding whether and how neutrophils support brain metastasis could be beneficial for devising effective therapeutic strategies.

In the course of investigating the function of key enhancers of brain metastasis and potential regulators of brain-infiltrating immune cells, we found that enhancer of zeste homologue 2 (EZH2) is highly expressed in patients’ brain metastases. EZH2 is a well-known histone methyltransferase that serves as an enzymatic subunit of the polycomb repressive complex 2 (PRC2), which epigenetically represses gene expression through trimethylation of histone H3 at lysine (K) 27 (H3K27me3). Here, we report that, in brain-metastatic cancer cells, EZH2 is phosphorylated by the oncogenic tyrosine kinase Src at tyrosine (Y) 696 (pY696-EZH2) and that pY696-EZH2 induces, independent of methyltransferase function, the pro-tumorigenic cytokine granulocyte-colony stimulating factor (G-CSF), which recruits programmed death-ligand 1 (PD-L1)-positive immunosuppressive neutrophils into the brain and facilitates brain metastasis. Inhibiting EZH2 phosphorylation with a Src inhibitor, blocking recruitment of neutrophils by inhibiting G-CSF, and targeting immunosuppressive neutrophils with immune checkpoint blockade (ICB) therapies deterred brain metastasis outgrowth in multiple mouse models. Thus, neutrophils’ function in promoting brain metastasis provides a scientific basis for targeting immunosuppressive neutrophils as a therapeutic strategy for treating brain metastasis.

Results

EZH2 promotes brain metastasis in a methyltransferase-independent manner

To identify brain metastasis-specific gene expression changes, we performed a transcriptomic analysis of RNA samples extracted from cultured A375 human melanoma cells and from malignant tissues of mice injected with A375 cells, including subcutaneous xenografts, lung metastases generated through tail vein injection, or brain metastases generated by intracarotid artery (ICA) injection. We found that 590 genes were specifically upregulated in brain metastases compared to subcutaneous tumors, lung metastases, and cultured A375 cells (fig. S1A, data file S1). Additionally, we compared gene expression in metastases from the bones, lungs, and brains of breast cancer patients (GSE14020, GPL570) and found 1263 upregulated genes in brain metastases (fig. S1B). EZH2 was a top (the fifth most) upregulated gene among the 41 genes that were upregulated in both clinical and experimental sets of brain metastases and is clinically targetable (Fig. 1, A and B). Furthermore, RNA sequencing of 24 pairs of primary tumors (breast cancer, lung cancer, and melanoma) and their matched brain metastases in another patient cohort (IRB protocol PA16–1122) validated that EZH2 mRNA expression was significantly higher in brain metastases than in corresponding primary tumors (P < 0.0457, Fig. 1C). Besides, EZH2 mRNA is highly expressed in triple-negative and HER2-positive subtypes of primary breast cancers (fig. S1C), which have high incidences of brain metastasis (28, 29).

Fig. 1. EZH2 promotes brain metastasis in a methyltransferase-independent manner.

(A) Schematic of the microarray analyses. In experimental brain metastases induced by A375 cells, the expressions of 590 genes were upregulated compared with the expressions in lung metastases, subcutaneous tumors, and cultured A375 cells. In a patient breast cancer data set (GSE14020 GPL570), the expressions of 1263 genes were upregulated in brain metastases compared with the expressions in bone and lung metastases. EZH2 was among 41 genes upregulated in brain metastases in both data sets. (B) Heat maps showing expression of 41 commonly upregulated genes (see A) in clinical brain metastases versus lung and bone metastases. (C) EZH2 mRNA expression, represented by RPKM values from RNA-seq, in 24 matched pairs of primary tumors (breast cancer, lung cancer, and melanoma) and brain metastases in a patient cohort (IRB protocol PA16–1122). *P < 0.05, Wilcoxon test. (D) Representative images of immunohistochemistry (IHC) staining of EZH2 in brain metastases and primary tumors from mice injected with B16BL6, MDA-MB-231, HCC1954, or 4T1 cells. Scale bar: 50 μm. IHC score: +, low; ++, medium; +++ high staining. (E) Western blotting of EZH2 and β-actin in lysates of B16BL6-induced brain (Br) metastases (B16Br) versus primary tumors (B16P) and in brain-seeking (Br) versus parental (P) cells (MDA-MB-231, HCC1954, 4T1). (F) Kaplan-Meier curves showing overall survival of mice injected intracardially with 4T1.Ctrl cells and treated with vehicle (4T1.Ctrl, n = 8) or GSK126 (4T1.GSK, n = 7, 150 mg/kg/mouse, 3 times per week, intraperitoneal administration) and mice injected with 4T1.EZH2 knockout cells and treated with vehicle (4T1.EZH2.KO, n = 10). ****P < 0.0001, log-rank test. (G) Representative bioluminescence images (BLI) of mice in the 3 subgroups described in (F). (H) Representative mouse brain BLI (top) and hematoxylin and eosin (H&E) staining (bottom) of brain metastases in 3 subgroups of mice injected intracardially with 4T1 cell sublines: EZH2 knockout cells (4T1.C50) transduced with pLenti (4T1.C50.pLenti, n = 9), wild-type EZH2 cells (4T1.C50.EZH2.WT, n = 8), or cells with an EZH2 methyltransferase-defective H689A mutation (4T1.C50.EZH2.H689A, n = 8). Scale bar: 50 μm. (I) Brain metastasis counts at 3 weeks post-intracardiac injection in the 3 subgroups described in (H). Micrometastases were lesions smaller than 50 μm in diameter; macrometastases were lesions 50 μm in diameter or larger. Results are shown as means ± standard errors of the mean (s.e.m.). *P < 0.05; **P < 0.01, t-test. N.S., not significant.

Consistent with the RNA upregulation of EZH2 in the 3 independent analyses above (Fig. 1, A to C), immunohistochemistry (IHC) analyses detected higher EZH2 protein expression in all brain metastatic lesions induced by ICA injection of B16BL6 mouse melanoma cells, MDA-MB-231 and HCC1954 human breast cancer cells, and 4T1 mouse mammary tumor cells than in their respective primary tumors (Fig. 1D, data file S2). Additionally, Western blotting confirmed that EZH2 protein expression was higher in brain metastatic tissues (B16Br) isolated from mice injected with B16BL6 melanoma cells than in their corresponding primary tumors (B16P) and higher in breast cancer brain-seeking (Br) sublines than in parental (P) breast cancer cells (MDA-MB-231, HCC1954, 4T1) (Fig. 1E).

To test whether EZH2 promotes brain metastasis in vivo, we generated EZH2-knockout 4T1 mammary tumor cells (4T1.EZH2.KO) by CRISPR/Cas9 and corresponding control (4T1.Ctrl) cells (fig. S1D), then transduced them with a luciferase-green fluorescent protein (GFP) plasmid (Luc.GFP). We injected 4T1.EZH2.KO or 4T1.Ctrl cells individually into female BALB/c mice via the left cardiac ventricle. Mice injected with 4T1.EZH2.KO cells had significantly longer survival (P < 0.0001, Fig. 1F) and fewer and smaller metastases in the brain and other organs than did mice injected with 4T1.Ctrl cells, as shown by in vivo bioluminescence imagining (BLI) of luciferase (Fig. 1G and fig. S1E). Mice injected with EZH2-high-expressing 4T1.Ctrl cells were also treated with GSK126, a potent small-molecule EZH2 inhibitor that selectively blocks EZH2 methyltransferase function by competitively binding to the methyl donor S-adenosylmethionine (30). Unexpectedly, GSK126 treatment did not affect mouse survival or brain metastasis growth (Fig. 1, F and G, and fig. S1E), even though GSK126 effectively decreased H3K27me3 expression and increased the expression of HOXA9, which gene is well known to be epigenetically repressed by EZH2 (31), in brain metastasis lesions (fig. S1F). To determine whether methyltransferase function is required for EZH2-mediated brain metastasis, we stably re-expressed the wild-type (WT) EZH2 (4T1.C50.WT), a methyltransferase-defective EZH2-H689A mutant (4T1.C50.H689A), or the pLenti control vector (4T1.C50.pLenti) (fig. S1G) in a 4T1 subline that had CRISPR/Cas9-mediated EZH2-knockout (4T1.C50) (fig. S1D). Each of the 3 4T1.C50-derived sublines (fig. S1G) were intracardially injected into BALB/c mice, which were monitored for brain metastasis by both BLI and IHC. Mice injected with 4T1.C50.WT cells developed brain metastasis, while mice injected with 4T1.C50.pLenti cells rarely did (Fig. 1, H and I). This indicated that cells with high EZH2 expression have an advantage in specifically metastasizing to the brain. In addition, mice injected with 4T1.C50.H689A cells developed both micro- and macro-lesions of brain metastases, similar to mice injected with 4T1.C50.WT cells (Fig. 1, H and I). Blocking EZH2 histone methyltransferase function, either genetically or with the EZH2-methyltransferase inhibitor GSK126, failed to reduce the incidence or the outgrowth of brain metastases. These data suggest that EZH2’s function in enhancing brain metastases is histone methyltransferase-independent.

Src phosphorylates EZH2 at Y696, which reduces EZH2 methyltransferase function

Post-translational modifications can induce EZH2’s non-methyltransferase oncogenic functions (32, 33). Our mass spectrometry analysis identified an unreported phosphorylation site in the C-terminal SET domain of EZH2 at tyrosine 696 (pY696) that matches a Src phosphorylation motif (Fig. 2, A and B) and is highly conserved across eukaryotes (fig. S2A). Src is a non-receptor tyrosine kinase that is dysregulated in multiple cancer types. Importantly, Src is hyperactivated in brain metastases of breast cancer patients and promotes tumor-cell extravasation into the brain parenchyma (34). We tested whether Src binds to and phosphorylates EZH2 at Y696 to enable EZH2 to promote brain metastasis by a methyltransferase-independent function. Indeed, using glutathione S-transferase (GST)-tagged Src (GST-Src), EZH2 proteins were pulled down from MDA-MB-231 cell lysates (fig. S2B). Src is known as a cell membrane-bound cytoplasmic protein (35), whereas EZH2 is a nuclear protein. To determine the subcellular location of the Src-EZH2 interaction, MDA-MB-231 and HCC1954 cell lysates were separated into cytoplasmic and nuclear fractions. Src proteins were readily detected in the nucleus alongside EZH2 by Western blotting (fig. S2C). Co-localization of Src and EZH2 in the nucleus was also shown by immunofluorescence (IF) staining (Fig. 2C). The Src/EZH2 complexes were detected in the nuclear fractions of both MDA-MB-231 human breast cancer cells and 4T1 mouse mammary tumor cells by immunoprecipitation (IP) of Src followed by Western blotting of EZH2 and vice versa (Fig. 2D, and fig. S2D). Src bound to EZH2 mostly in the nucleus but did not bind to SUZ12 and EED, other PRC2 components, suggesting that the nuclear Src/EZH2 complex is not associated with PRC2 (Fig. 2E).

Fig. 2. Src phosphorylates EZH2 at Y696, which reduces EZH2 methyltransferase function.

(A) Mass spectrometry analysis showing EZH2 tyrosine (Y)-696 phosphorylation. (B) The Src phosphorylation motif in EZH2 and known Src substrates. Y696 is in the EZH2 C-terminal methyltransferase SET domain. (C) Fluorescent microscopy images showing EZH2 and Src nuclear co-localization in MDA-MB-231 cells. Scale bar: 50 μm. (D) Immunoprecipitation (IP) of Src from MDA-MB-231 nuclear fractions followed by Western blotting (WB) for EZH2, and vice versa. (E) Src protein complexes immunoprecipitated from MDA-MB-231 cytoplasmic and nuclear fractions were blotted with the indicated antibodies. (F, G) Synthetic EZH2 proteins were incubated with [γ-³²P]-ATP and Src protein immunoprecipitated from 231.Src cell lysates (F) or synthetic Src (G), and then EZH2 phosphorylation was detected by autoradiography. (H) Endogenous pY696-EZH2 was immunoprecipitated from MDA-MB-231 and 4T1 cell lysates using an anti-pY696-EZH2 antibody, followed by Western blotting for EZH2. (I) 231.pLenti and 231.Src cell lysates were immunoprecipitated using a pY696-EZH2 antibody, followed by Western blotting for EZH2. (J, K) Western blotting for the indicated proteins in cell lysates of 231.pLenti and 231.Src cells (J) and of 231 cells with Src knockdown by shRNAs sh#648 and sh#1579 or with scrambled control shRNA (sh.Scr) (K). (L) The 231.3C13 cells were infected with control lentiviruses (pLenti), lentiviruses expressing wild-type EZH2 (WT), lentiviruses expressing EZH2 with Y696 mutated to phenylalanine (YF), and lentiviruses expressing EZH2 with Y696 mutated to aspartic acid (YD). Expression of the indicated proteins in these 4 groups was detected by Western blotting.

To test whether Src phosphorylates EZH2, Src kinases were immunoprecipitated from the lysates of MDA-MB-231 cells stably overexpressing Src (231.Src) to perform a kinase assay. These Src proteins phosphorylated recombinant EZH2 substrate (Fig. 2F). Moreover, recombinant Src directly phosphorylated recombinant EZH2 proteins in an in vitro kinase assay (Fig. 2G). Also, we generated a pY696-specific EZH2 antibody that detected endogenous pY696-EZH2 in both MDA-MB-231 and 4T1 cells (Fig. 2H and fig. S2E). This antibody immunoprecipitated more endogenous pY696-EZH2 from 231.Src cells than from control 231.pLenti cells (Fig. 2I and fig. S2F). These data confirmed that Src directly phosphorylates EZH2 at the Y696 site.

Serine/threonine phosphorylation of EZH2 at S21, T311, or T487 suppresses EZH2 methyltransferase function, but O-GlcNAcylation of EZH2 at S75 activates it (36–39). We explored whether Src-mediated phosphorylation at Y696 affects H3K27me3 by EZH2. 231.Src cells with high endogenous pY696-EZH2 (Fig. 2I, bottom) showed reduced H3K27me3 compared to control 231.pLenti cells (Fig. 2J). Knocking down Src in MDA-MB-231 cells with short hairpin RNAs (shRNAs: sh#648 and sh#1579) or inhibiting Src in 231.Src and 231.Src527F cells (which express a constitutively active Src.527F mutant) with a Src inhibitor (saracatinib) increased H3K27me3 (Fig. 2K, and fig. S2, G and H). Also, we stably re-expressed EZH2 WT, a non-phosphorylatable Y696F mutant (YF), a phospho-mimic Y696D mutant (YD), or the control vector (pLenti) in the 3C13 subline of MDA-MB-231 EZH2-knockout cells (231.EZH2.KO), in the C50 subline of 4T1 EZH2-knockout cells (4T1.C50), and in EZH2.shRNA-transduced MCF7 cells (MCF7.shEZH2) (Fig. 2L, and fig. S2, I, J, and K). The 231.3C13.YD, 4T1.C50.YD, and MCF7.shEZH2.YD cells had dramatically lower H3K27me3 expression than did their corresponding WT or YF cells (Fig. 2L, and fig. S2, J and K), indicating that phosphorylation at Y696 by Src suppressed EZH2 methyltransferase function.

pY696-EZH2 facilitates brain metastasis through increased secretion of G-CSF

To assess whether and how pY696-EZH2, with its reduced methyltransferase activity, functions in promoting brain metastasis, we intracardially injected nude mice with 231.3C13.pLenti, 231.3C13.WT, 231.3C13.YF, or 231.3C13.YD cells and monitored brain metastasis formation and mouse survival. Mice injected with 231.3C13.WT or 231.3C13.YD cells had shorter brain metastasis-free survival than did mice injected with 231.3C13.pLenti or 231.3C13.YF cells (Fig. 3A). Additionally, 231.3C13.YD cells dramatically enhanced brain metastasis outgrowth compared to 231.3C13.YF cells after ICA injection into mice (Fig. 3B, data file S3). This outgrowth was not inhibited by GSK126 treatment (fig. S3A), indicating that EZH2-Y696D drives brain metastasis by a methyltransferase-independent function. However, differences in bone-metastasis-free survival or mammary tumor growth were insignificant (P > 0.05) between 231.3C13.YF or 231.3C13.YD cell-injected mice (fig. S3, B and C). 231.3C13.YD cells showed no advantages in proliferation, colony formation, migration, or matrigel invasion compared to 231.3C13.YF cells in vitro (fig. S3, D, E, and F), while the EZH2 knockout cell line 231.3C13 exhibited reduced cell migration and matrigel invasion, compared with control MDA-MB-231 cells (fig. S3F). These discordant in vivo and in vitro data indicate that pY696-EZH2 may promote brain metastasis by modifying the brain TME.

Fig. 3. pY696-EZH2 facilitates brain metastasis through the increased secretion of G-CSF.

(A) Brain metastasis-free survival curves and representative bioluminescence images (BLI) of the brains of mice intracardially injected with 231.3C13-derived sublines: empty vector (pLenti, n = 10), EZH2.WT (WT, n = 10), EZH2.Y696F (YF, n = 9), and EZH2.Y696D (YD, n = 9). *P < 0.05, log-rank test. (B) Quantification of BLI photon changes (normalized to Day 1 post-injection) and representative BLI of mice that received intra-carotid artery (ICA) injection of 231.3C13.YF or 231.3C13.YD cells. Data are means ± s.e.m. *P < 0.05, **P < 0.01, t-test. (C) Cytokine array of 231.3C13.YF and 231.3C13.YD cells. (D) Quantification and representative imaging of brain BLI signals of mice that received ICA injection of 231.3C13.YF cells and subcutaneous treatment with vehicle (n = 10) or G-CSF (n = 9). Data are means ± s.e.m. *P < 0.05, t-test. (E) Quantification and representative imaging of brain BLI signals of mice that received ICA injection of 231.3C13.YD cells and intraperitoneal (i.p) treatment with either IgG (n = 6) or anti-G-CSF antibody (n = 8). Data are means ± s.e.m. *P < 0.05, t-test. (F) Kaplan-Meier curves showing overall survival of mice bearing brain metastases of 4T1.C50.EZH2.YD cells transduced with a scrambled shRNA (4T1.C50.YD.sh.Scr., n = 11) or shCSF3 #441 (4T1.C50.YD.shCSF3, n = 9) or 4T1.C50.EZH2.WT cells (n = 11). **P < 0.01; ****P < 0.0001, log-rank test. (G) Representative IHC staining and quantification of Ki-67- and TUNEL-positive cells in the brain metastases of mice that were injected with 4T1.C50.YF.vector or 4T1.C50.YF.CSF3 cells. Scale bar: 50 μm. Data are means ± s.e.m. *P < 0.05, t-test.

We next studied how pY696-EZH2-expressing cancer cells communicate with the TME. Our cytokine array analysis showed that 231.3C13.YD cells secreted higher amount of G-CSF, interleukin 1-alpha (IL-1α), and serpinE1 than did 231.3C13.YF cells (Fig. 3C). Also, CSF3, IL1A, and SERPINE1 mRNA expression was higher in 231.3C13.YD and 231.3C13.WT cells, as well as in 4T1.C50.YD and 4T1.C50.WT cells, than in their respective pLenti or YF cells (fig. S3, G and H). Serpins are known to facilitate brain metastasis cell survival (10), and IL-1α regulates G-CSF expression (40–42). Therefore, we tested whether G-CSF enhances brain metastasis in vivo. Nude mice were ICA-injected with 231.3C13.YF cells and were treated subcutaneously (s.c.) with recombinant G-CSF or vehicle as controls. G-CSF treatment markedly enhanced brain metastasis outgrowth compared to vehicle treatment (Fig. 3D). In contrast, intraperitoneal (i.p.) administration of a G-CSF-blocking antibody in nude mice that had undergone ICA injection of 231.3C13.YD cells diminished brain metastasis outgrowth compared to the immunoglobulin G (IgG) control treatment (Fig. 3E). Furthermore, we knocked down CSF3 (encoding G-CSF) with shRNA in 4T1.C50.YD cells (4T1.C50.YD.shCSF3) (fig. S3I). BALB/c mice that had been ICA injected with 4T1.C50.YD.shCSF3 cells showed significantly longer survival than did mice injected with 4T1.C50.YD.shScr control cells or 4T1.C50.WT cells (P < 0.0001, Fig. 3F). Conversely, overexpressing CSF3 in 4T1.C50.YF cells (4T1.C50.YF.CSF3) increased proliferation (Ki-67 IHC) and reduced apoptosis (TUNEL staining) in brain metastatic lesions from BALB/c mice (Fig. 3G, and fig. S3J). Thus, data from 2 preclinical brain metastasis models (human and mouse) showed that pY696-EZH2 can increase G-CSF secretion, which promotes brain metastasis.

Neutrophil recruitment by G-CSF creates an immunosuppressive brain TME

G-CSF can mediate neutrophil recruitment to the TME in various solid tumors (43–46). Neutrophils can either inhibit or promote cancer progression and formation of a pre-metastatic niche, which is regulated by cytokines released from cancer cells (20). G-CSF mainly activates a pro-tumor and pro-metastasis program in neutrophils (22, 23, 45, 47, 48). However, whether neutrophils are present in brain metastases and, if so, how they function are unknown. Using IF staining for the neutrophil marker S100A8 (49, 50), we detected brain-infiltrating neutrophils in mice that had been ICA-injected with breast cancer cells, and more brain-infiltrating neutrophils in mice injected with 231.3C13.YD cells than in those with 231.3C13.YF cells (Fig. 4, A and B top). Administering G-CSF to mice that had been ICA-injected with 231.3C13.YF cells dramatically increased neutrophil infiltration of the whole brain compared to vehicle controls (Fig. 4, A top and B top), whereas administering a G-CSF-blocking antibody to mice that had been ICA-injected with 231.3C13.YD cells decreased neutrophil infiltration of the whole brain compared to IgG controls (Fig. 4, A bottom and B bottom). Inhibiting EZH2 methyltransferase function with GSK126 had no significant effect on neutrophil brain infiltration in 231.3C13.YD cell-injected mice (Fig. 4, A middle and B top, P = 0.1151). Neutrophils from brain tissue were also readily detected by flow cytometry (fig. S4A). G-CSF treatment of 231.3C13.YF-injected nude mice resulted in increased brain-infiltrating neutrophils detected by CD45+ CD11b+ Ly6G+ (22, 51) (Fig. 4C) and increased neutrophil population in the peripheral blood detected by complete blood count (Fig. 4D).

Fig. 4. Neutrophil recruitment by G-CSF creates an immunosuppressive brain TME.

(A) Representative immunofluorescent staining images of S100A8 for neutrophils (red: Fluor594) and DAPI for nuclei (blue). Scale bar: 50 μm. (B) Quantification of S100A8-positive cells per field of view in the brains of the indicated groups of mice. Data are means ± s.e.m. *P < 0.05, **P < 0.01, t-test. N.S., not significant. (C) Flow cytometry results showing numbers of CD11b+ Ly6G+ cells in the brains of healthy mice (normal brain) or mice that received intracarotid artery (ICA) injection of 231.3C13.YF cells and treatment with vehicle or G-CSF. Data are means ± s.e.m. **P < 0.01, ***P < 0.001, t-test. (D) Absolute numbers of neutrophils in the peripheral blood of healthy mice and of mice injected with 231.3C13.YF and treated with vehicle or G-CSF, according to a complete blood count. Data are means ± s.e.m. *P < 0.05, t-test. (E) Co-immunofluorescence analysis of Arginase 1 (Arg1) and neutrophil marker S100A8 in the brains and the peripheral blood of mice bearing brain metastases. Scale bar: 20 μm. (F) Representative flow cytometry plots and quantification of PD-L1 expression in CD45+ CD11b+ Ly6G+ cells isolated from the brain and the peripheral blood of mice bearing brain metastases. Data are means ± s.e.m. ****P < 0.0001. (G) Mice were treated with anti-CD8 antibody to deplete CD8+ T cells or with IgG antibody as controls, then intracranially injected control 4T1.C50.YD.shScr or CSF3-knockdown 4T1.C50.YD.shCSF3 cells. Kaplan-Meier curves showing overall survival of mice bearing brain metastases of 4T1.C50.YD.shScr or 4T1.C50.YD.shCSF3 after CD8+ T cell depletion (anti-CD8) or not (IgG). **P < 0.01; ***P < 0.001; N.S., not significant; log-rank test.

The number of brain-infiltrating neutrophils was reduced in the brains of BALB/c mice injected with 4T1.C50.YD.shCSF3 cells, compared to 4T1.C50.YD.Scr control cells, as shown by S100A8+ staining of brain tissues and flow cytometric detection of CD11b+ Ly6G+ cells (fig. S4, B and C), although CD11b+ F4/80+ macrophages were similar between the two groups (fig. S4D). In contrast, the number of brain-infiltrating neutrophils (S100A8+ or CD11b+ Ly6G+) was significantly increased in BALB/c mice injected with CSF3-overexpressing 4T1.C50.YF.CSF3 cells compared to control cells (4T1.C50.YF.vector) (fig. S4, E and F, P < 0.0001 and P = 0.0020). The increased population of brain-infiltrating neutrophils paralleled increased proliferation and reduced apoptosis in the brain metastatic lesions induced by 4T1.C50.YF.CSF3 cells (Fig. 3G).

Neutrophils are phenotypically heterogeneous and functionally adaptable (52). We observed that the subgroup of S100A8+ brain-infiltrating neutrophils expressed Arginase 1 (Arg1), which is known to induce immune anergy and immunosuppressive T-regulatory cell expansion (52), whereas circulating neutrophils from matched mice rarely did (Fig. 4E, and fig. S4G). Brain-infiltrating neutrophils (CD11b+ Ly6G+) also expressed the T cell checkpoint ligand PD-L1, which was barely detectable in circulating CD11b+ Ly6G+ cells (Fig. 4F). These data suggest that brain metastatic tumor cells induced Arg1 and PD-L1 upregulation in brain-infiltrating neutrophils. Indeed, adding conditioned medium from 4T1 mammary tumor cells to bone marrow-derived neutrophils (CD11b+ Ly6G+) induced the membrane PD-L1 expression on neutrophils (fig. S4H). Functionally, when these tumor-educated neutrophils were co-cultured with anti-CD3/CD28 antibody-pretreated, activated T cells, they inhibited CD8+ T cell proliferation (fig. S4I). Next, we investigated whether T cell inhibition by the G-CSF-recruited neutrophils is critical for pY696-EZH2-mediated brain metastasis outgrowth. We treated BALB/c mice with anti-CD8 antibody to deplete CD8+ T cells or with IgG antibody as controls, then intracranially injected control 4T1.C50.YD.shScr and 4T1.C50.YD.shCSF3 cells. For control IgG antibody-treated groups, mice injected with 4T1.C50.YD.shCSF3 cells showed significantly longer survival than mice injected with 4T1.C50.YD.shScr control cells did (P = 0.0028, Fig. 4G), consistent with the ICA injection data (Fig. 3F). However, when CD8+ T cells were depleted by anti-CD8 antibody, 4T1.C50.YD.shCSF3 cell-injected mice had no survival advantage compared to C50.YD.shScr cell-injected mice (Fig. 4G). Thus, the survival advantage of 4T1.C50.YD.shCSF3-injected mice versus 4T1.C50.YD.shScr-injected mice is CD8+ T cell dependent.

Since blocking G-CSF in MDA-MB-231 cell-injected nude mice also partially inhibited brain metastasis, we tested whether tumor-associated neutrophils directly promote MDA-MB-231 cell growth, independent of T cells, as shown for other tumor cells (53, 54). After culturing bone marrow-derived neutrophils in MDA-MB-231 cell-conditioned medium, mCherry-expressing MDA-MB-231 (231.mCherry) cells were cultured with or without tumor-associated neutrophils in G-CSF-supplemented medium (fig. S4J). Indeed, 231.mCherry cell growth was dramatically increased with neutrophil co-culture compared to that without neutrophils, indicating that tumor-associated neutrophils also promote MDA-MB-231 cell growth without T cells.

pY696-EZH2 upregulates CSF3 by increasing c-Jun expression

To obtain mechanistic insights into how pY696-EZH2 induces cytokine secretion in cancer cells, we compared protein expression profiles of pY696-EZH2-high versus pY696-EZH2-low cancer cells by reverse phase protein array (RPPA). This analysis identified 13 proteins that are more highly expressed in pY696-EZH2-high 231.3C13.WT and 231.3C13.YD cells than in pY696-EZH2-low 231.3C13.pLenti and 231.3C13.YF cells (Fig. 5A). Among them, phospho-Ser73-c-Jun (pS73-c-Jun) is known to transcriptionally upregulate IL1A and CSF3 (55–57). Also, 231.3C13 and 4T1.C50 cells that re-expressed EZH2 WT or YD had higher total c-Jun protein expression than did cells that re-expressed YF (Fig. 5B, and fig. S5A). Functionally, knockdown of c-Jun by small interfering RNA (siRNA) and selective inhibition of the c-Jun N-terminal kinase (JNK) by SP600125 both reduced CSF3 and IL1A mRNA expression in MDA-MB-231 and 4T1 cells (Fig. 5, C and D, and fig. S5, B, C, D, and E).

Fig. 5. pY696-EZH2 upregulates CSF3 by increasing c-Jun expression.

(A) Reverse-phase protein array heatmap showing differentially expressed proteins in 231.3C13.pLenti, 231.3C13.WT, 231.3C13.YF, and 231.3C13.YD cells (2 biological replicates). (B) Western blotting showing the expressions of the indicated proteins in 231.3C13.pLenti, 231.3C13.WT, 231.3C13.YF, and 231.3C13.YD cells. (C) Western blotting showing c-Jun expression (top) and qRT-PCR analysis of IL1A and CSF3 mRNA expression (bottom) in MDA-MB-231 cells transfected with c-Jun-targeting siRNAs (#61, #78) versus controls. n = 3. Data are means ± s.e.m. **P < 0.01, ***P < 0.001, t-test. (D) Results of qRT-PCR for CSF3 and IL1A mRNA expression in MDA-MB-231 cells treated with DMSO control vehicle or the JNK inhibitor SP600125 (SP, 40 μM, 24 h). n = 3. Data are means ± s.e.m. **P < 0.01, ***P < 0.001, t-test. (E) qRT-PCR results showing c-JUN mRNA expression in 231.3C13.pLenti, 231.3C13.WT, 231.3C13.YF, and 231.3C13.YD cells. n=3. Data are means ± s.e.m. ****P < 0.0001, t-test. (F) qRT-PCR analysis of EZH2 and c-JUN mRNA expression in MDA-MB-231 cells harboring EZH2 shRNA #3 or #4. n = 3. Data are means ± s.e.m. **P < 0.01, ****P < 0.0001, t-test. (G) Western blotting showing the indicated proteins in MDA-MB-231 cells transduced with lentiviruses harboring control vector (231.pLenti) or constructs expressing wild-type Src (231.Src). (H) Western blotting showing the indicated proteins in MDA-MB-231 cells transduced with lentiviruses harboring a control shRNA with a scrambled sequence (sh.Scr) or Src-targeting shRNAs (shSrc #648 and shSrc #1579). (I) qRT-PCR analysis of c-JUN mRNA expression in 231.pLenti, 231.Src.sh.Scr, 231.Src.shEZH2 #3, and 231.Src.shEZH2 #4. n = 3. Data are means ± s.e.m. ****P < 0.0001, t-test. (J) c-JUN promoter activity measured by dual-luciferase reporter assay in MDA-MB-231 cells transduced with lentiviruses harboring control vector (231.pLenti) or expressing EZH2 (231.EZH2) (left) and in in 231.3C13.pLenti, 231.3C13.WT, 231.3C13.YF, and 231.3C13.YD cells (right). The Firefly luciferase signal was normalized to the Renilla luciferase signal. n >= 3. Data are means ± s.e.m. *P < 0.05, **P < 0.01, t-test.

Since phospho-JNK and total JNK protein expression in all 4 of the 231.3C13 sublines (pLenti, WT, YF, and YD) were similar (fig. S5F), the increase in pS73-c-Jun expression in 231.3C13.WT and 231.3C13.YD cells is unlikely to be attributable to the upstream kinase JNK but may result from the increased total c-Jun expression induced by pY696-EZH2. Indeed, c-JUN mRNA expression was higher in 231.3C13.WT and 231.3C13.YD cells than in 231.3C13.pLenti and 231.3C13.YF cells (Fig. 5E), and EZH2 knockdown reduced c-JUN mRNA expression in MDA-MB-231 cells (Fig. 5F). Upstream of EZH2, Src-overexpressing 231.Src cells had highly increased c-Jun and pS73-c-Jun protein expression compared with controls (Fig. 5G), while knocking down Src or inhibiting the Src kinase with saracatinib suppressed c-Jun expression in MDA-MB-231 and 4T1 cells (Fig. 5H, and fig. S5G). Src-mediated induction of c-JUN mRNA in 231.Src cells was blocked by shRNA-mediated depletion of EZH2 (Fig. 5I and fig. S5H), indicating that Src enhances c-JUN mRNA expression via pY696-EZH2. Furthermore, c-JUN promoter activity was higher in pY696-EZH2-high cells (231.EZH2, 231.3C13.WT, and 231.3C13.YD) than in control cells, as shown by luciferase reporter assays (Fig. 5J). Collectively, these results show that pY696-EZH2 induces transcriptional upregulation of c-JUN, which upregulates IL1A and CSF3 expression.

pY696-EZH2 directly binds to RNA Pol II to transcriptionally upregulate c-JUN

Next, we investigated how pY696-EZH2 induces transcriptional upregulation of c-JUN. Since pY696-EZH2 suppressed EZH2 methyltransferase function (Fig. 2L, and fig. S2, J and K), which could de-repress gene expression, we tested whether pY696-EZH2 upregulates c-JUN via modulating the PRC2 histone methyltransferase complex. We first found that 231.3C13 sublines (pLenti, WT, YF, and YD) had similar amounts of EED and SUZ12 proteins (fig. S6A); thus, pY696-EZH2 does not alter EED or SUZ12 expression. Second, the binding of EZH2-WT, EZH2-YF, and EZH2-YD with EED and SUZ12 was similar among HEK 293FT (293FT) cells expressing these EZH2 variants (fig. S6B), indicating that pY696-EZH2 does not change EZH2 binding to EED or SUZ12 in the PRC2 complex. Third, transiently depleting EED or SUZ12 in 231.3C13.YD cells by siRNA had no obvious effect on c-Jun expression (fig. S6, C and D), showing that the PRC2 complex is not functionally involved in c-Jun upregulation by pY696-EZH2.

To explore the mechanisms of pY696-EZH2-induced c-JUN transcriptional upregulation, we detected EZH2 binding to c-JUN by chromatin IP (ChIP) using a series of PCR primers that bind to various regions of the c-JUN promoter. In 231.pLenti cells, EZH2 is recruited to c-JUN promoter from −300 base pairs (bp) upstream (primer P2) of the c-JUN transcription start site (TSS) to +400 bp downstream (primer P4) of the TSS (Fig. 6A). In 231.EZH2 cells that exogenously overexpressed EZH2, EZH2 binding to P3 (at the TSS) and P4 of c-JUN and the promoter sites of HOXA9B (an EZH2 epigenetic substrate, used as a positive control) was increased (Fig. 6A); whereas in EZH2-knockdown 231.shEZH2 cells, EZH2 binding at P3, P4, and HOXA9B was decreased (Fig. 6B). Additionally, 231.3C13.WT cells had higher EZH2 binding at the P4 and P5 sites than did 231.3C13.pLenti cells (fig. S6E).

Fig. 6. pY696-EZH2 directly binds to RNA Pol II to transcriptionally upregulate c-JUN.

(A) Top: The locations of the primers for qPCR of c-JUN gene. TSS: transcription start site. Bottom: EZH2 was immunoprecipitated from 231.pLenti and 231.EZH2 cells, and EZH2 binding to c-JUN was detected by qPCR using the indicated primers. HOXA9B, a well-known EZH2 target gene, was used as a positive control. (B) EZH2 was immunoprecipitated from 231.shScr and 231.shEZH2 #4 cells, and EZH2 binding to c-JUN was detected by qPCR using the indicated primers. (C) RNA polymerase II (RNA Pol II) was immunoprecipitated from 231.EZH2 cells, and then RNA Pol II binding to c-JUN was detected by qPCR using the indicated primers. (D) EZH2 was immunoprecipitated from 231.3C13.YF and 231.3C13.YD cells, and EZH2 binding to c-JUN was detected by qPCR using the indicated primers. (E) RNA polymerase II (RNA Pol II) was immunoprecipitated from 231.3C13.YF and 231.3C13.YD cells, and then RNA Pol II binding to c-JUN was detected by qPCR using the indicated primers. All fold-enrichment values in the ChIP-qPCR analyses (A-E) were normalized to IgG (red line). n = 3. Data are means ± s.e.m. *P < 0.05, **P < 0.01, ****P < 0.0001, t-test. (F) EZH2 protein complexes were immunoprecipitated from 231.3C13.YF and 231.3C13.YD cells by EZH2 antibodies, followed by Western blotting of RNA Pol II and vice versa. (G) EZH2 binding to its histone H3 substrate in 231.3C13.YF and 231.3C13.YD cells. Endogenous H3 was immunoprecipitated from 231.3C13.YF and 231.3C13.YD cell lysates, followed by Western blotting of EZH2. (H) HEK 293FT cells were transfected with an empty vector (pLenti) or EZH2.WT (WT), EZH2.Y696F (YF), or EZH2.Y696D (YD) plasmids. Immunoprecipitation for EZH2 was followed by Western blotting of histone H3 and vice versa.

RNA polymerase II (RNA Pol II) also bound to P2, P3, and P4 loci within and flanking the c-JUN promoter; these loci overlapped with EZH2 binding loci (Fig. 6C). Both EZH2 and RNA Pol II showed increased binding at P3 and P4 loci in 231.3C13.YD cells compared with 231.3C13.YF cells (Fig. 6, D and E), whereas H3K27me3 had negligible binding at P2, P3, or P4 loci of the c-JUN promoter (fig. S6F), supporting a methyltransferase-independent mechanism for c-JUN transcriptional upregulation by pY696-EZH2. Taken together, these data show that both EZH2 and RNA Pol II bind to P3, P4, and nearby promoter regions of c-JUN; thus, pY696-EZH2 may function as a transcriptional co-factor of RNA Pol II to upregulate c-JUN expression independent of EZH2’s methyltransferase function.

To determine whether pY696-EZH2 binds to and functions as a co-factor of RNA Pol II, EZH2 was immunoprecipitated followed by Western blotting of RNA Pol II, and vice versa. These assays detected binding of EZH2 with RNA Pol II (Fig. 6F). Also, EZH2 binding to RNA Pol II was stronger (Fig. 6F), but EZH2 binding to histone H3 was weaker (Fig. 6G), in 231.3C13.YD cells than in 231.3C13.YF cells. Similarly, 293FT.YD cells also showed greatly decreased EZH2 binding to histone H3 compared with 293FT.YF cells, as detected by EZH2 IP followed by Western blotting of histone H3 and vice versa (Fig. 6H). In 231.Src cells with high endogenous pY696-EZH2, EZH2 binding to histone H3 was also reduced, and EZH2 binding to RNA Pol II was increased, compared with 231.pLenti control cells (fig. S6, G and H). Furthermore, ChIP-seq analysis after RNA Pol II IP showed decreased RNA Pol II binding at the promoter regions of 2293 genes, including c-JUN, in 231.3C13.YF cells expressing a non-phosphorylatable EZH2 mutant, compared to that in MDA-MB-231 cells expressing WT EZH2 which is phosphorylated by Src (data file S4). These data indicate that Src-induced pY696-EZH2 switches the binding preference of EZH2 from histone H3 to RNA Pol II, and that pY696-EZH2 cooperates with RNA Pol II as a transcription factor to drive c-JUN transcription.

pY416-Src and pY696-EZH2 are associated with increased neutrophil infiltration in patients’ brain metastases

To determine the clinical relevance of our findings, we detected pY416-Src (activated Src), pY696-EZH2, pS73-c-Jun, S100A8 (neutrophil marker), and Ki-67 (proliferation marker) in breast cancer patients’ brain metastases and matched primary tumors by IHC staining. The signals of pY416-Src and pY696-EZH2 were significantly (P = 0.016 and P = 0.026, respectively) higher in brain metastases than in matched primary breast tumors (fig. S7A, and Fig. 7, A, B, and C), indicating that high pY416-Src and pY696-EZH2 expression in tumor cells is associated with brain metastasis.

Fig. 7. pY416-Src and pY696-EZH2 are associated with increased neutrophil infiltration in patients’ brain metastases.

(A) Representative immunohistochemistry (IHC) staining images (AEC substrate) of pY416-Src and pY696-EZH2 protein expression in 2 sets of matched primary breast tumor and brain metastasis patient samples. Scale bar: 50 μm. (B) Significantly higher IHC staining scores for pY416-Src in brain metastases than in matched primary breast tumors. P = 0.016, Fisher exact test. (C) Significantly higher IHC staining scores for pY696-EZH2 in brain metastases than in matched primary breast tumors. P = 0.026, Fisher exact test. (D) Representative IHC staining images (AEC substrate) of pY416-Src, pY696-EZH2, and pS73-c-Jun expression, S100A8+ cells, and Ki-67 (DAB substrate) in 2 patient brain metastasis samples. Scale bar: 50 μm. (E) pY696-EZH2 staining was associated with S100A8 staining in patients’ brain metastases (P = 0.019, Fisher exact test). (F) Association between S100A8 and Ki-67 staining in patients’ brain metastases (P = 0.0459, Fisher exact test).

In brain metastases, high pY416-Src expression was associated with high pY696-EZH2 expression (P = 0.026), which was significantly associated with high pS73-c-Jun expression (P = 0.037) and increased numbers of brain-infiltrating S100A8+ neutrophils (P = 0.019; Fig. 7, D and E, and fig. S7, B and C), indicating that Src activation-induced pY696-EZH2 in brain metastatic tumor cells contributes to upregulation of pS73-c-Jun and recruitment of neutrophils in brain metastases of breast cancer patients. Both the expression of pY696-EZH2 and the number of S100A8+ neutrophils in brain metastases were associated with positive staining for the Ki-67 proliferation marker (Fig. 7, D and F, and fig. S7D), suggesting that pY696-EZH2-induced brain-infiltrating neutrophils facilitate brain metastasis outgrowth in patients.

Importantly, S100A8 RNA expression in brain metastasis negatively correlated with patients’ overall survival (OS, P = 0.0049) and survival after brain metastasis (SPBM, P = 0.0005) in a cohort of 22 patients with matched primary breast tumors and brain metastases (58) (fig. S7E). However, S100A8 expression in the primary breast tumors had no correlation with OS or SPBM (fig. S7F). Together, these findings demonstrate that activation of Src oncogenic kinase (pY416-Src) is associated with increased pY696-EZH2, which corresponds to a high pS73-c-Jun signal and increased numbers of infiltrating neutrophils in brain lesions. Additionally, the pY416-Src/pY696-EZH2/pS73-c-Jun pathway-induced brain infiltration of neutrophils is functionally associated with brain metastasis outgrowth and poor survival in patients.

Src inhibitor and ICB combination therapies effectively inhibit brain-infiltrating neutrophils and brain metastasis in vivo

Our data from both animal models and patient tissues demonstrated that brain metastases with high pY416-Src/pY696-EZH2 can recruit immunosuppressive neutrophils into the brain to facilitate metastasis outgrowth by deterring T cell functions. Because tumor-associated myeloid cells were found to negatively impact the ICB response (59, 60), we sought proof of concept that reducing the recruitment of neutrophils by targeting Src will enhance ICB therapeutic efficacy in inhibiting brain metastasis. We first tested this in the 4T1 mammary tumor cell model, which has high endogenous pY416-Src/pY696-EZH2 (fig. S5G, and Fig. 2H), high G-CSF secretion (fig. S8A), and many brain-infiltrating PD-L1+ neutrophils (fig. S4, B, C, D, and E, and Fig. 4F). We treated mice with an ICB regimen (anti-PD-1 antibody and anti-cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) antibody), alone or combined with a Src inhibitor, saracatinib, that has been shown to pass the blood-brain barrier and inhibit Src activity in brain metastasis lesions (34). To simulate clinical brain metastases, which frequently coexist with primary tumors, we injected 4T1.Luc.GFP cells into BALB/c mice both intracranially (2 × 10³ cells) to induce brain lesions and into the mammary fat pad (MFP, 2 × 10⁴ cells) to induce mammary tumors. Three days after injection, we began treating mice with i) vehicle, ii) saracatinib (40 mg/kg/day, oral gavage), iii) ICB (anti-PD-1 antibody, 200 μg/mouse and anti-CTLA-4 antibody, 100 μg/mouse, i.p.), or iv) saracatinib combined with ICB (Fig. 8A). Brain metastases were monitored by BLI, and primary tumors were measured with calipers. ICB treatment alone robustly inhibited brain metastases, reduced primary tumor size, and prolonged mouse survival in the 4T1 model (Fig. 8, B and C, and fig. S8, B and C). The combination of ICB and saracatinib inhibited both brain metastasis outgrowth and primary tumor growth most effectively among all treatment groups (Fig. 8, B and C, and fig. S8B).

Fig. 8. ICB and Src inhibitor combination therapies effectively inhibit brain-infiltrating neutrophils and brain metastasis in vivo.

(A) The schedule of combination treatment in mice intracranially injected with 2 × 10³ 4T1.Luc.GFP cells and injected with 2 × 10⁴ 4T1.Luc.GFP cells into the mammary fat pad (MFP). The mice were treated with vehicle (n = 11), the Src inhibitor saracatinib (n = 12), ICB (n = 12), or saracatinib combined with ICB (sar.+ ICB) (n = 12), beginning at day 3 after injection. (B, C) Representative bioluminescence images (BLI) (B) and quantification of BLI photon intensity (C) of the brains of mice in the four groups described in (A). (D-F) Quantification of S100A8+ cells (D), granzyme B (GzmB)+ cells (E), and CD8+ T cells (F) in the brain metastases of mice in the 4 groups described in (A). (G) Quantification of Ki-67- and TUNEL-positive cells in the brain metastases of mice in the 4 groups described in (A). (C-G) Data are means ± s.e.m. *P < 0.05, **P < 0.01; ****P < 0.0001, N.S., not significant, t-test. (H) Kaplan-Meier curves showing overall survival of mice intracranially injected with 5 × 10³ EMT6 cells and injected in the MFP with 5 × 10⁴ EMT6 cells and treated with vehicle, saracatinib, ICB, or saracatinib combined with ICB (sar.+ ICB), beginning at day 6 after injection. Log-rank test, *P < 0.05.

Immune-cell infiltration into brain metastases under different treatments was detected by IHC staining. Saracatinib alone resulted in a reduction of S100A8+ neutrophils and an increase of granzyme B-positive (GzmB+) cytotoxic T lymphocytes in brain metastases, indicating that Src inhibition can partially relieve the immunosuppressive brain TME by blocking brain-infiltrating neutrophils (Fig. 8, D and E, fig. S8D). ICB treatment alone did not markedly change neutrophil infiltration but increased GzmB+ lymphocyte and CD8+ T cell infiltration, showing that ICB treatment rekindled cytotoxic T cell function in brain metastases (Fig. 8, E and E, fig. S8D). The combination of ICB plus saracatinib robustly blocked S100A8+ neutrophil infiltration (as did saracatinib alone) and activated cytotoxic T cell function (as did ICB alone) in brain metastases (Fig. 8, D, E, and E, fig. S8D). ICB plus saracatinib combination therapy significantly decreased the proliferation (P < 0.0001) and increased apoptosis (P = 0.0288) of tumor cells in brain metastatic lesions (Fig. 8G, and fig. S8E).

We further compared the therapeutic response to the combination treatment versus the response to ICB alone in the EMT6 mammary tumor model, which induces ICB-resistant brain metastases. BALB/c mice were injected with EMT6 mammary tumor cells intracranially (5 × 10³ cells) and into the MFP (50 × 10³ cells) and were divided into 4 treatment groups as in the 4T1 model above, except the treatments were started 6 days after injection. Compared to treatment with vehicle, treatment with ICB or saracatinib alone did not prolong mouse survival (Fig. 8H). However, the combination of ICB plus saracatinib significantly (P = 0.0088) prolonged the survival of mice (Fig. 8H), despite all treatment groups having similar MFP tumor volumes (fig. S8F). Together, data from these preclinical models demonstrate that inhibiting pY696-EZH2 with a Src inhibitor combined with ICB enhances the therapeutic efficacy of ICB in deterring brain metastases by relieving immunosuppression.

Discussion

In this study, we found that brain metastatic cells, in which Src is highly activated (34), overexpress EZH2, which directly binds to Src in the nucleus and is phosphorylated by Src at the Y696 site. This Y696 phosphorylation switches the binding preference of EZH2 from histone H3 to RNA Pol II; pY696-EZH2 interacts and cooperates with RNA Pol II to transcriptionally upregulate c-JUN, independently of the methyltransferase function of EZH2. c-Jun induces the expression and secretion of inflammatory cytokine G-CSF, which recruits immunosuppressive Arg1+/PD-L1+ neutrophils into the brain to drive lethal brain metastasis outgrowth (fig. S9). The discovery of this immunosuppressive function by neutrophils in brain metastases highlights that dysfunction of the innate immune response can contribute to suppression of the adaptive immune response in the brain. These findings point to targeting neutrophils as a therapeutic strategy for brain metastasis.

EZH2 is well known to contribute to the development and progression of various human malignancies and is deemed an attractive cancer therapeutic target (61). However, targeting EZH2 has not always been clinically beneficial. Several EZH2 inhibitors showed efficacy in clinical trials for treating non-Hodgkin lymphoma by inhibiting the increased methyltransferase activity of a gain-of-function EZH2 mutant (Y641F/N/S), but they failed in patients with WT EZH2-overexpressing solid tumors (30, 61–63). Counterintuitively, high expression of EZH2 was associated with low H3K27me3 expression in breast cancer patients (64, 65), and low H3K27me3 expression correlated with poor survival in patients with breast, ovarian, and pancreatic cancers (66). These clinical dilemmas pose a challenge in clinically targeting the EZH2 methyltransferase. Here, we found that WT EZH2 is phosphorylated at Y696 by the oncogenic kinase Src. pY696-EZH2 has reduced H3K27 methyltransferase activity, and this activity could not be blocked by an EZH2-methyltransferase inhibitor. Because solid tumors frequently have Src activation that phosphorylates WT EZH2 to pY696-EZH2, our findings clarify why WT EZH2-high-expressing solid tumors generally do not respond to EZH2 methyltransferase inhibitors in clinical trials and explain the observation of low H3K27me3 expression in WT EZH2-high-expressing breast cancers (64, 65). Our data suggest that blocking Src-induced phosphorylation of WT EZH2, targeting pY696-EZH2 downstream components such as G-CSF, or reversing neutrophil-induced immunosuppression would be beneficial.

We revealed that nuclear Src phosphorylates EZH2 at Y696 in the C-terminal SET domain of EZH2, and pY696-EZH2 binds to RNA Pol II, not histone H3, and activates a c-Jun/GCSF/neutrophil axis to foster brain metastasis. This defines a “phospho-switch” that alters EZH2’s binding preference from its canonical substrate histone H3 to RNA Pol II and thereby changes EZH2’s function from an epigenetic transcription repressor to a transcription co-factor. Although EZH2 has been implicated in transcriptional activation in breast cancer and castration-resistant prostate cancer (33, 67), here, we report the evidence that Src-induced pY696-EZH2 interacts with RNA Pol II and functions as a transcription co-factor in solid tumors. Because chromatin reorganization occurs with activating transcription factors, pY696-EZH2 may impact three-dimensional chromatin structure by several mechanisms, including modulating H3K27me3 and RNA Pol II.

The immune-cell landscape of brain metastasis is unique and complex. Normalizing the immune response in the brain TME could be an effective therapeutic strategy for inhibiting brain metastasis. Although neutrophils are important effector cells in the innate arm of the immune system, their role in brain metastasis development has been largely ignored, partly because neutrophils are phenotypically and functionally heterogeneous and have several subpopulations with different functions (68). Here, we identified a special population of brain-infiltrating neutrophils that are PD-L1+ and Arg1+ with immunosuppressive functions in brain metastasis. These neutrophils affect tumor cell growth through both T cell dependent and independent mechanisms. The number of brain-infiltrating neutrophils was positively associated with brain metastasis tumor cell proliferation in breast cancer patients, and systemic G-CSF blockade thwarted brain metastasis outgrowth in 2 preclinical models. These findings provide a scientific basis for G-CSF targeting and/or blocking brain-infiltrating neutrophils as therapeutic options for treating brain metastases. The G-CSF receptor-blocking antibody CSL324 is currently in clinical trial (NCT03972280) for the treatment of inflammatory and immune disorders; our data warrant further clinical testing of the potential of blocking G-CSF/G-CSF receptor for treating brain metastasis.

Recently, a clinical trial for treating melanoma-derived brain metastases with nivolumab (anti-PD-1) combined with ipilimumab (anti-CTLA-4) showed marked clinical benefit in 57% of patients (69). However, a clinical trial of ICB for lung cancer brain metastasis showed no efficacy (70). These clinical observations raise the question of how to predict and enhance the response of brain metastases to ICB. Response to anti-PD-L1 antibody treatment was found to be strongly associated more with PD-L1 expression on tumor-infiltrating immune cells than on tumor cells (71). Here, we found that ICB can activate cytotoxic T cells and impede 4T1-induced brain metastasis outgrowth. Moreover, targeting Src, the EZH2 upstream kinase, combined with ICB treatment, further improved the therapeutic efficacy in inhibiting brain metastasis outgrowth by relieving the immune suppression induced by brain-infiltrating neutrophils. Since Src inhibitors and ICB therapies (blocking antibodies to CTLA-4, PD-1, and/or PD-L1) are clinically applicable treatments, our findings could be quickly translated into the clinic to treat patients with refractory brain metastases, especially those with excessive brain infiltration of immunosuppressive neutrophils.

Although our results open a promising translational direction in treatment of brain metastasis, we need to consider the limitations of our study. Brain metastasis experimental models have been modestly successful in predicting patient responses to therapy (72, 73). Due to the lack of spontaneous brain metastasis models, we used experimental brain metastasis models (ICA or intracranial injection) of 4T1 or EMT6 cells to induce brain metastasis to test the effects of combination treatment. The 4T1 and the EMT6 brain metastasis models are very aggressive, and they might not represent the relatively longer time of brain metastasis outgrowth in patients. Additionally, although we have observed markedly improved efficacy using the combination of ICB and Src inhibition, the concentrations of drugs and the treatment schedule still need to be optimized to achieve maximum therapeutic efficacy for the clinical management of patients with brain metastasis. Also, we have focused on pro-inflammatory cytokines regulated by pY696-EZH2; thus, other mechanisms that drive brain metastasis are yet to be investigated.

In summary, our study discovered a site of phosphorylation by Src on EZH2 at Y696 that changes EZH2’s function from an epigenetic transcription repressor to a transcription co-factor. We also demonstrated that pY696-EZH2 promotes brain metastasis by recruiting immunosuppressive neutrophils and that this function is independent of EZH2’s methyltransferase function. On the basis of these findings, we present an effective strategy to target immunosuppressive neutrophils for treating brain metastasis, which may be readily translated into the clinic for treating patients suffering from brain metastasis.

Materials and Methods

Study design

The objective of this research was to determine the function of p-Y696 EZH2 in brain metastases and to test blockade of immunosuppressive neutrophils as a strategy for treating brain metastasis. The number of mice used in each experimental group was determined by power analysis and on the basis of prior experience with animal models of metastasis. ICA injection of 4T1 and MDA-MB-231 cells induces symptomatic brain metastases in 100% of BALB/c mice or nude mice, so the sample size of 15 female mice/group can achieve 81% power at 0.05 significance level. For other in vivo experiments, a sample size of 7 to 10 mice per group was typical. All sample sizes are listed in the corresponding figure legends or on the figures. Animals were randomly assigned to treatment groups. The quantitative experiments were repeated using at least 3 independent biological replicates, except RPPA experiments, which used 2 biological replicates. The number of biological replicates for each experiment is indicated in the figure legends. The investigators were not blinded during data collection and analysis. Pathologists were blinded to group allocation during analysis of staining and when assessing outcomes.

In vivo experiments in mice

All animal experiments were carried out in accordance with protocols approved by the Institutional Animal Care and Use Committee of The University of Texas MD Anderson Cancer Center. C57BL/6J and BALB/cJ mice were purchased from The Jackson Laboratory, and athymic NCR nu/nu mice were obtained from MD Anderson. The mice were exposed to a 12-h light/dark cycle, bred as specific-pathogen-free and given free access to food and water. The number of mice used in each experimental group was determined by power analysis, and mice were grouped randomly for each experiment. Female mice were used for breast cancer and mammary tumor experiments; both female and male mice were used for melanoma experiments.

Brain metastases were induced by intracarotid, intracardiac, or intracranial injections. Six-to-eight-week-old C57BL/6J and BALB/cJ mice, or 8-to-10-week-old athymic NCR nu/nu mice were used in these experiments. For intracarotid injections, tumor cells (100,000 cells in 0.1 mL Hanks’ balanced salt solution/mouse for MDA-MB-231, HCC1954, A375, and B16BL6 cell lines and 50,000 cells for the 4T1 cell line) were injected into the right common carotid artery as described previously (34). For intracardiac injection, 4T1.Ctrl or 4T1.EZH2.KO cells (200,000 cells/mouse), 4T1.C50.pLenti, 4T1.C50.EZH2.WT, or 4T1.C50.EZH2.H694A cells (100,000 cells/mouse), or 231.3C13 cells and 4 related sublines (100,000 cells/mouse) were injected into the left ventricle of anesthetized female mice as described previously (74). For intracranial injection, 4T1 cells (2000 cells/mouse) were injected into the right hemisphere as described previously (15). Tumor cells for most in vivo brain metastasis experiments were labelled with luciferase by transducing a pLOVE.Luc2GFP plasmid before injection. To generate lung metastasis, 500,000 A375 cells in 0.2 mL Hanks’ balanced salt solution/mouse were injected via the lateral tail vein. For subcutaneous xenograft experiments, 500,000 A375 cells in 0.05 mL Hanks’ balanced salt solution/mouse were injected in the right hind footpad. MFP tumors were established by injection of 20,000 4T1 tumor cells in 50 μL of PBS orthotopically into the MFP of 6-to-8-week-old BALB/c mice. MFP tumor size was measured with a digital caliper, and tumor volume was calculated using the formula: L × W²/2, where L is the longest diameter and W is the shortest diameter.

Brain metastases generated by injection of luciferase-labeled tumor cells were monitored by BLI every 3 days (for 4T1 and sublines) or every week (for MDA-MB-231 and sublines) using an IVIS Lumina XR system (Perkin Elmer). The endpoint brain metastasis-free survival was defined as the first day on which the bioluminescence signal was detected in the mouse brain. For the analysis of overall survival of brain metastasis-bearing mice, the endpoints were based on bioluminescence signals and clinical signs of brain metastasis including (but not limited to) behavioral abnormalities resulting from primary central nervous system disturbances and weight loss. Animals were euthanized upon showing signs of brain metastasis or 1 to 4 weeks after injection for detection of neutrophil recruitment and brain metastasis outgrowth. Brain metastasis lesions were counted as the experimental readout. Overt metastases or whole brains were dissected, and brain-seeking cells were disaggregated in DMEM/F-12 medium using a Tenbroeck homogenizer. Bone metastasis-free survival is defined as the time before the luciferase signal was detectable in mouse legs.

Human studies

The retrospective evaluation of pY416-Src, pY696-EZH2, pS73-c-Jun, calgranulin A (S100A8), and Ki-67 expression was performed in TMAs of matched primary breast tumor and brain metastasis samples from 41 patients, of whom 14 had human epidermal growth factor receptor 2-positive breast tumors, 16 had triple-negative breast tumors, 9 had estrogen receptor-positive breast tumors, and 2 had uncategorized breast tumors according to clinical diagnostic criteria (75). Several TMA dots were detached from slides during the process of TMA slide construction or staining. The TMAs of formalin-fixed, paraffin-embedded matched primary breast tumor and metastatic brain tumor samples were obtained from The University of Queensland Centre for Clinical Research (Herston, Queensland, Australia). Tissues were collected with the approval of the human research ethics committees at the Royal Brisbane and Women’s Hospital (2005/022) and The University of Queensland (2005000785). For TMA construction, tumor-rich regions (guided by histological review) from each specimen were sampled using 1-mm cores. All of the archival paraffin-embedded tumor samples were coded with no patient identifiers. All patient samples in the second patient cohort (IRB protocol No. PA16–1122) were collected with the approval of the Institutional Review Board of The University of Texas MD Anderson Cancer Center. The tumor contents of frozen tissues were reviewed by a pathologist (J.T.H.). Tumor material was available from 24 patients with paired primary tumor tissue and brain metastasis tissue (breast cancer: n = 15, lung cancer: n = 7, melanoma: n = 2). All patients gave informed consent to the use of their tissue for research purposes.

Bioinformatic and statistical analyses

Univariate analyses with Student t tests were performed to identify differentially expressed transcripts in our cDNA microarray analyses of A375 cells and their corresponding tumors and metastases. The P values obtained from multiple t tests were corrected for the false discovery rate (FDR) using the beta-uniform mixture method. With an FDR threshold of 0.15, 590 genes were found to be upregulated, and 336 genes were found to be downregulated in brain metastases compared with other samples. Supervised cluster analysis was performed using Cluster 3.0, and heatmaps were generated using Java Treeview software. GSE14020 microarray data were processed to select the best probe for each gene, the probe showing the highest geometric median of intensity and median absolute deviation across all samples for the 20,307 genes on GPL570. Differential gene expression between brain metastases and other sites was analyzed using the limma package for R software with a P-value threshold of 0.05 and a 1.25-fold change between groups. RNA sequencing data were converted to Fastq format using Illumina bcl2fastq software. Fastq reads were aligned to the human reference genome (hg19) using the STAR software (76) and further processed by using the Picard software (v1.112) for marking duplicated reads. The HTSeq (v0.6.1) (77) was used to generate raw read counts for each gene. Reads per kilobase of transcript, per million mapped reads (RPKM) were normalized to the raw count by sequencing depth and gene length. RPKM was calculated by using the R (v3.2) statistical language. ChIP-Sequencing reads were mapped to human genome (hg19) by bwa software (78) and putative Pol II binding peaks were called using HOMER software (79) using a fold change of >3 in peak area versus control. The peaks are further narrowed down to be within 2 kbp of any transcription starting sites for all RefSeq gene transcripts.

The differences in EZH2 mRNA expression between primary tumors and brain metastases were analyzed by Wilcoxon test. For power analysis, we calculated sample size using the Pwr package in the R environment. For staining of patient samples, we calculated the association between pY696-EZH2 and S100A8, pY416-Src, pS73-c-Jun or Ki-67, and the association between S100A8 and Ki-67 by Fisher exact test. The quantitative experiments were repeated using at least 3 independent biological replicates (2 biological replicates for RPPA), and data are presented as means ± standard errors of the mean. Quantitative data were analyzed using either 1-way analysis of variance (for multiple groups) or t-tests (for 2 groups) by Graphpad Prism 8 software. Survival was analyzed by Kaplan-Meier curves and log-rank tests. Two-sided P values of less than 0.05 were considered statistically significant. All statistically significant values shown in the figures are indicated as: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, N.S. not significant.