Invasive breast cancer reprograms early myeloid differentiation in the bone marrow to generate immunosuppressive neutrophils

A Coordinated Expansion of CD11b⁺Gr1⁺ Myeloid Cells in Peripheral Tissues Parallels Tumor Development.

In cancer patients, expansion of T cell-suppressive myeloid cells correlates with tumor stage (8). We used the PyMT transgenic mouse model (21), which expresses the polyomavirus middle T antigen driven by the mouse mammary tumor virus (MMTV) promoter, to map expansion of CD11b⁺Gr1⁺ myeloid cells during the multiple stages of breast cancer progression. PyMT mice develop a progressive disease that mimics human pathology (22), in which all mammary glands (MGs) display signs of breast cancer with an initial presentation of hyperplasia (6 wk), which then transitions to adenoma or mammary intraepithelial neoplasia (8 wk), early carcinoma (10 wk), and then late-stage carcinoma (11–15 wk) (Fig. S1 A and B). Similar to human disease, mammary tumor growth is associated with the dissemination of cancer cells in the periphery and notably the lungs. We detected disseminated tumor cells (DTCs) by real-time quantitative PCR (qPCR) in the lungs of three of five PyMT mice as early as 6 wk of age (Fig. S1C). At 14 wk, we detected metastases, and PyMT expression increased dramatically (Fig. S1D).

The expansion of CD11b⁺Gr1⁺ cells occurs within tumors, but also in blood, spleen, and premetastatic lung in murine transplantation models that mimic the late stages of tumor development (9). However, the kinetics of expansion and the location of these myeloid cells during cancer initiation and the early stages of tumor progression remain unknown. To address this question, we analyzed the frequency and cell number of CD11b⁺Gr1⁺ cells in different tissues isolated from PyMT mice at the onset and during the course of disease (from 6 to 15 wk). To discriminate between circulating and tissue resident myeloid cells, we harvested tissues after vascular flushing. Interestingly, the proportion of CD11b⁺Gr1⁺ cells within the developing primary tumors remained constant and below 0.4% of total cells for all tumor stages (Fig. 1A and Fig. S1C). In contrast, we detected a significant systemic increase in proportion and number of these cells in the lung, blood, and spleen at 10 wk, with a continued increase with age (Fig. 1A and Fig. S2 A–C). Thus, we evaluated the proportion of CD11b⁺Gr1⁺ cells within the CD45⁺ immune compartment in the primary tumor and lung. CD11b⁺Gr1⁺ cells represented 8% of CD45⁺ cells in late-stage tumors, but they made up 39% in lungs (Fig. S2D). These data show a coordinated expansion of CD11b⁺Gr1⁺ cells in peripheral tissues during early tumor development and provide a model system to decipher the mechanism for expansion of these cells in cancer.

Activated CD11b⁺Ly6G⁺ Neutrophils Are the Predominant Myeloid Cell Type That Expand in Peripheral Tissues During Tumor Progression.

The Gr1 antibody recognizes two antigens: Ly6G, a specific marker for neutrophils (23), and Ly6C, which is expressed on myeloid and nonmyeloid cells (24). Two distinct subsets of Gr1⁺ myeloid cells, polymorphonuclear (PMN) CD11b⁺Ly6G⁺Ly6C^int (referred to as Ly6G⁺) and monocytic (Mo) CD11b⁺Ly6G⁻Ly6C^hi (referred to as Ly6C^hi) (also called PMN-MDSCs and Mo-MDSCs, respectively) (25) expand in cancer, both with T cell-suppressive activity. We found that Ly6G⁺ cells were the predominant Gr1⁺ cell subset that increased substantially in all tissues from PyMT mice (Fig. 1B and Fig. S3 A and B). Ly6G⁺ cells from PyMT mice showed typical neutrophil morphology (Fig. 1B).

A distinguishing characteristic of Ly6G⁺ cells in tumor-bearing mice is their ability to suppress T-cell function (26). To determine whether Ly6G⁺ cells acquired this ability in these mice, we isolated them by flow cytometry from spleens and cocultured them with carboxyfluorescein diacetate succinimidyl ester (CFSE)-labeled splenocytes, in the presence of anti-CD3 and anti-CD28 activating antibodies to stimulate T-cell proliferation. Ly6G⁺ cells from PyMT mice reduced both CD4 and CD8 T-cell proliferation by ∼50%, whereas Ly6G⁺ cells from WT mice did not (Fig. 1C and Fig. S3C). Because T-cell suppression by Ly6G⁺ cells is attributed to enhanced production of ROS (27), we next used a flow cytometry-based assay to assess superoxide production. In the absence of exogenous stimulation, Ly6G⁺ cells in blood and lung of tumor-bearing mice showed increased dihydrorhodamine (DHR) fluorescence, a probe that detects intracellular superoxide, indicating that neutrophils in tumor-bearing mice actively produce more ROS compared with WT mice (Fig. 1D and Fig. S3D). We also observed that total splenocytes and FACS-sorted Ly6G⁺ cells from PyMT mice had lost Rb1 expression, confirming that the loss of Rb1 is a distinct phenotype of Ly6G⁺ cells from tumor-bearing mice (20) (Fig. 1E). Taken together, these data demonstrate that breast cancer development in PyMT mice leads to the generation of T cell-suppressive, activated Ly6G⁺ neutrophils in peripheral tissues.

Breast Cancer Development Results in Profound Remodeling of BM Hematopoiesis.

To determine the origin of Ly6G⁺ cells associated with tumor development, we assessed proliferation of Ly6G⁺ cells by BrdU incorporation (Fig. 2, A–C and Fig. S4 A–C). We did not detect any BrdU⁺ Ly6G⁺ cells in blood, lung, or MG (tumor) from WT or PyMT mice (14–15 wk) after a 2-h BrdU chase, suggesting that these organs do not significantly contribute to their production (Fig. S4 A and B). In contrast, ∼5% of Ly6G⁺ cells in BM were BrdU⁺ in both WT and tumor-bearing mice after a 2-h chase (Fig. 2B), suggesting that Ly6G⁺ cells are generated in BM. After a 24-h chase, 40 ± 5% Ly6G⁺ cells were BrdU⁺ in BM of PyMT mice, which was significantly more than in BM of WT mice in which 25 ± 2% were BrdU⁺ (Fig. 2C). Of note, only Ly6G^low cells incorporated BrdU after a 2-h BrdU chase, whereas Ly6G^med/hi cells showed BrdU incorporation after a 24-h chase. Thus, these data indicate that Ly6G^low cells are immature granulocytes that give rise to mature Ly6G^hi neutrophils with little proliferation potential (28). Consistent with this idea, we found that CD11b⁺Gr1⁺ cells dramatically expand in BM at 10 wk due to a predominant increase in Ly6G⁺ cells (Fig. 2D and Fig. S4 D and E). We also detected BrdU⁺ Ly6G⁺ cells in spleen following a 2-h chase (Fig. 2B), but the frequency was approximately one third in spleen of PyMT mice compared with WT after a 24-h chase (Fig. 2, B and C), and the cell number was less than 10% compared with BM (Fig. S4C). These data demonstrate that the BM is the main production site of Ly6G⁺ cells during breast tumor development.

The pale appearance of the BM from late-stage, tumor-bearing PyMT mice (Fig. 2E and Fig. S4F) and their severe anemia (Fig. S4G) suggested that tumor growth in the breast induces a profound alteration of BM hematopoiesis. Therefore, we examined changes in MPs and more immature HSPCs in BM of late-stage PyMT mice (Fig. 2F). We detected higher frequencies of GMPs within the MP population associated with an overall increase of their absolute numbers in PyMT mice compared with WT controls (Fig. 2, G and H). Although there was no significant change in CMPs, the LSK population expanded approximately twofold, with a significant increase in cell number of HSCs, MPP^F−, and MPP^F+ populations (Fig. 2, G and I). We next performed longitudinal studies to determine the kinetics of HSPC expansion in early tumor-bearing PyMT mice. Strikingly, we found that HSCs expanded as early as 8 wk, whereas both MPP^F+ and MPP^F− expanded at 10 wk in PyMT mice (Fig. 2I and Fig. S4H), concomitant with the increased CD11b⁺Gr1⁺ cells in peripheral tissues (Figs. 2D and 1). Thus, expansion of HSCs occurred during early tumor development, followed by expansion of MPPs and CD11b⁺Gr1⁺ myeloid cells as early as 10 wk, suggesting activation of HSCs leads to increased production of MPPs, which then gives rise to an expanded myeloid compartment.

Expansion of T cell-suppressive myeloid cells in cancer is associated with enlargement of the spleen (29), which may act as a reservoir for extramedullary hematopoiesis (30). We also observed an enlarged spleen in late-stage PyMT mice (Fig. S5A); thus, we also assessed MP and HSPC populations in the spleen. Unlike BM, the majority (80–90%) of spleen-residing MPs were MEPs (Fig. S5B), which increased in numbers approximately sixfold in PyMT mice (Fig. S5 C and D). No difference in the frequencies of MPP^F+, MPP^F−, or HSCs was observed (Fig. S5B), although the total number of these cells was higher than in WT mice (Fig. S5E) due to the increased size of the spleen in late-stage PyMT mice (Fig. S5 A and C). Increased production of Ly6G⁺ cells combined with an increase in the frequency of HSCs, MPPs, and GMPs in BM, but not spleen, suggests activation of a specific myeloid differentiation pathway in BM of tumor-bearing mice.

G-CSF Is Necessary and Sufficient for the Activated and Rb1^low Phenotype of Immunosuppressive Neutrophils.

We next hypothesized that the factor(s) driving expansion of Ly6G⁺ cells would be present in the serum of PyMT mice. When we assessed the serum concentration of several cytokines and chemokines, only a few factors were increased systemically during early carcinoma formation (10 wk), whereas multiple factors were deregulated during late-stage tumor progression (>12 wk) (Fig. 3A and Table S1). Notably, G-CSF and the neutrophil-attracting chemokine CXCL1 (KC), and to a lesser extent CCL2 (MCP-1), increased early during disease development (Fig. 3A). In contrast, the related myeloid growth factors GM-CSF and M-CSF and chemokines CCL3 and CCL4 did not increase during the early phase of tumor progression (Fig. 3A). We also detected abundant secretion of G-CSF and CXCL1, but not CCL2, in the supernatants of cultured primary PyMT tumor cells (Fig. 3B) and the VO-PyMT cell line (Fig. 3C), pointing to the tumor cells as the source of these systemic factors.

Given that G-CSF was recently shown to promote tumor progression by contributing to the T cell-suppressive activity of neutrophils (31), we asked whether G-CSF contributed to the activated phenotype of Ly6G⁺ cells in PyMT mice. Although Ly6G⁺ cells from control antibody-treated PyMT mice had increased ROS production compared with WT mice, Ly6G⁺ cells from PyMT mice treated with a neutralizing antibody to G-CSF did not (Fig. 4A). Anti–G-CSF treatment also reduced the frequency of circulating Ly6G⁺ cells (Fig. 4A) and the concentration of G-CSF in the serum of PyMT mice (Fig. 4B). Next, we assessed whether G-CSF was sufficient to induce ROS production. We stimulated WT mice for 5 consecutive days with G-CSF (2 µg/mouse), a dose that recapitulates the persistent high levels detected in late-stage PyMT mice (Fig. 4C). Indeed, G-CSF stimulation alone increased ROS production in Ly6G⁺ cells (Fig. 4D). Thus, tumor-derived G-CSF was both necessary and sufficient for enhanced ROS production in Ly6G⁺ cells in PyMT mice.

We then investigated whether tumor-derived G-CSF also regulated the loss of Rb1 in activated Ly6G⁺ cells. Anti–G-CSF treatment of late-stage PyMT mice restored Rb1 protein expression in splenocytes (Fig. 4E). In addition, Rb1 expression decreased in total splenocytes and isolated Ly6G⁺ cells from WT mice stimulated with G-CSF for 3 d, but not 1 d, and decreased in a time-dependent manner thereafter to levels similar to late-stage PyMT mice (Fig. 4F). To assess the kinetics in the loss of Rb1 expression in Ly6G⁺ cells from multiple tissues, we used a more sensitive intracellular flow cytometry assay (Fig. 4G). Rb1 expression was strongly expressed in circulating Ly6G⁺ cells in spleen and blood of WT mice. Surprisingly, Ly6G⁺ cells from BM of WT and PyMT mice did not express Rb1. These results suggest that Rb1 expression is normally acquired in circulating Ly6G⁺ cells after they leave the BM in WT mice, and its acquisition is inhibited by tumor-derived G-CSF. Thus, tumor-derived G-CSF is the key factor responsible for the two distinguishing characteristics of tumor-derived immunosuppressive neutrophils: enhanced ROS production and loss of Rb1 expression.

Tumor-Derived G-CSF Targets Hematopoietic Stem and Early Progenitors to Drive Myeloid Cell Production in BM.

To test the involvement of G-CSF in activating myeloid cell production in tumor-bearing mice, we treated WT and PyMT mice with the anti-G-CSF antibody. As expected, blocking G-CSF reduced systemic levels of CD11b⁺Gr1⁺ cells in all tissues in WT and PyMT mice (Fig. S6 A–C). Importantly, blocking G-CSF not only reduced Ly6G⁺ neutrophils, but also returned Ly6C^hi, MPP^F+, MPP^F−, and HSC numbers to WT levels in BM of tumor-bearing mice while having no significant effect on MPP and HSC numbers in WT mice (Fig. 5A and Fig. S6D). In addition, the altered phenotypes of the BM and spleen that are characteristic of myeloid dysregulation were restored to WT conditions after anti–G-CSF treatment (Fig. 5B).

Next, we assessed whether G-CSF stimulation alone was sufficient to expand mature myeloid cells, as well as early progenitors in WT mice. After 1 d of G-CSF stimulation, Gr1⁺ cells and specifically Ly6G⁺ cells increased in blood, but not in BM or spleen (Fig. 5, C and D and Fig. S7A), supporting mobilization of Ly6G⁺ cells from BM by G-CSF. Of note, an increase in Ly6G^low cells was observed in BM after 1 d and was more evident after 3 d of G-CSF stimulation (Fig. 5D), consistent with production of immature neutrophils (32, 33). We also observed a similar increase in Ly6G^low cells in blood and spleen, but not until 3 d of G-CSF stimulation (Fig. 5D), which suggests that this increase was due to increased production of Ly6G⁺ cells from the BM. In addition, an increase in GMPs and the loss of red pigment in the BM was also observed after 3 d of G-CSF stimulation (Fig. 5, E and F and Fig. S7B). Importantly, 12 h and 1 d of G-CSF stimulation was sufficient to increase MPPs, which continued to increase after 3 and 5 d of stimulation (Fig. 5G and Fig. S7B). During most of the treatment, the number of HSCs did not significantly change despite a minor and transient increase after 12 h of G-CSF stimulation (Fig. 5G). In contrast to BM, 12-h G-CSF stimulation did not increase HSCs or MPPs in the spleen (Fig. 5H). The kinetics of expansion of these BM populations suggests a rapid activation of a myeloid differentiation program in the early hematopoietic compartment by prolonged G-CSF stimulation (>1 d). These data show that G-CSF is necessary and sufficient for tumor reprogramming of hematopoietic differentiation in the BM in cancer.

G-CSF Acts in a Cell Intrinsic Manner to Drive Altered Myeloid Differentiation in Cancer.

To demonstrate the role of G-CSF in activating myeloid differentiation in the early hematopoietic compartments in tumor-bearing mice directly, we developed an orthotopic transplantation model (Fig. 6A). First we established mixed BM chimeras by transplanting lethally irradiated WT (CD45.1) mice with a 1:1 ratio of WT (CD45.1) and G-CSF-R^−/− (CD45.2) BM cells. Then, following BM reconstitution, we orthotopically transplanted primary C57BL/6 late-stage PyMT tumor cells or vehicle (sham-treated) into chimeric mice. Similar to PyMT transgenic mice, we found expansion of Ly6G⁺ cells in lung, blood, spleen, and BM (Fig. 6B), consistent with elevated G-CSF production by primary C57BL/6 late-stage PyMT tumor cells (Fig. 3B). We observed a highly significant expansion of GMPs, MPPs, and HSCs, but no expansion of CMPs in orthotopically transplanted mice (Fig. 6C), similar to changes in FVB/n PyMT transgenic mice (Fig. 2, H and I). Chimerism analyses showed a defect in the ability of G-CSF-R^−/− cells to contribute to the expanded GMP compartment in tumor-bearing mice (Fig. 6D), which was expected because G-CSF is known to regulate GMPs directly (34). However, G-CSF-R^−/− cells did not contribute to the expanded MPP compartment, and there was an overrepresentation of these cells in the expanded HSC compartment in tumor-bearing mice (Fig. 6E). These results confirm the engraftment of G-CSF-R^−/− HSCs and demonstrate their inability to contribute to the activated myeloid differentiation in the BM of tumor-bearing mice. The specific expansion of G-CSF-R^−/− HSCs may result from their inability to differentiate into MPPs or their inadequate mobilization compared with WT HSCs in the G-CSF–rich milieu of tumor-bearing mice, although we cannot rule out that other tumor-derived factors may increase HSC numbers in these animals. Taken together, these data directly demonstrate that G-CSF produced by tumor cells activates a myeloid differentiation program in the early hematopoietic compartment, before the more committed CMPs, to skew hematopoiesis toward the myeloid lineage in the BM, resulting in increased production of a specific neutrophil subset, which on circulation in the periphery, acquires an activated, T cell-suppressive, and Rb1^low phenotype (Fig. 6F).

Mice.

All mouse experiments were approved by the Institutional Animal Care and Use Committee of the University of California San Francisco, in accordance with the guidelines of the National Institutes of Health. Transgenic PyMT (MMTV-PyMT^634Mul/J) mice (21, 22) were maintained on FVB/n and C57BL/6 backgrounds; littermate controls were used for all experiments. WT FVB/n female mice were purchased from Charles River and WT C57BL/6 female (CD45.1 or CD45.2) mice were purchased from the Jackson (JAX) Laboratories for cross-breeding, in vivo G-CSF stimulation, or BM transplants. Homozygous C57BL/6 mice lacking expression of the receptor for G-CSF (G-CSF-R^−/−) (63) were provided by Daniel Link, Washington University, St. Louis, and then later acquired from JAX [B6.129 × 1(Cg)-Csf3r^tm1Link/J], with age-matched WT C57BL/6 (JAX) mice used as controls.

Flow Cytometry.

All antibodies are listed in Table S2. Cells were blocked with rat IgG (10 μg/mL; Sigma) for at least 20 min on ice, washed with staining media [2% (vol/vol) HI, FBS in HBSS (BSS) without Ca²⁺ or Mg²⁺, denoted SM], and then stained with fluorescently conjugated antibodies in SM for 30 min on ice, unless stated otherwise. For evaluation of MPs and HSPCs in BM and spleen, cells were stained as previously described (64). In brief, a six-step staining procedure was used that included the following combinations of antibodies: (i) mixture of unconjugated rat antibodies to lineage specific markers (B220, CD3, CD4, CD5, CD8, CD11b, Ter119, Gr1); (ii) fluorescently-tagged (PE-Cy5 or PE-Cy5.5) anti-rat secondary antibody; (iii) rat IgG (as described above) to block nonspecific binding; (iv) mixture of biotin- or fluorophore-conjugated antibodies to c-Kit, Sca-1, FcγR, and CD34 (for staining MPs) or c-Kit, Sca-1, Flk2, CD48, and CD150 (for staining HSPCs); (v) streptavidin conjugated to a fluorophore; and (vi) propidium iodide to identify live and dead cells. In some experiments evaluating mature myeloid cells (i.e., Gr1⁺, Ly6G⁺, or Ly6C^hi), stained cells were fixed with 2% (vol/vol) PFA at 4 °C overnight and analyzed 24–48 h later. For evaluation of intracellular Rb1 protein expression using flow cytometry, cell were stained for extracellular cell surface markers, fixed in 4% (vol/vol) PFA, permeabilized briefly (5 min) in 0.1% Triton X-100, blocked for 20 min at 4 °C with the 2.4G2 Fc-blocking antibody, and then stained with an anti-Rb1 antibody for 30 min on ice. The data were acquired with the FACSCalibur, LSRII, or FACSAria instruments (BD Biosciences) and analyzed using the FlowJo software (Tree Star, Inc.).

Isolation of Epithelial-Enriched Cells from Tumors.

Tumors were digested in collagenase as described above, with several modifications. Tumors were digested for 30 min in collagenase media, centrifuged, DNase I treated, and then resuspended in DMEM complete. Single cells (which consisted primarily of stromal cells) were then removed from the cell suspension by a series of brief centrifugations (500 × g for 1 s, four times), which resulted in epithelial-enriched organoids (65). Organoids were then further digested in DMEM complete with trypsin (0.05% in EDTA) and DNase I (20 U/mL) for 20 min at 37 °C. Digestion was stopped by the addition of 10% (vol/vol) heat inactivated (HI) FBS in DMEM complete. Resulting epithelial-enriched cells were washed, filtered (as described above) and then cultured for in vitro assays or frozen in DMEM/F12 with 30% (vol/vol) HI FBS, 100 U/mL penicillin and 100 μg/mL streptomycin (P/S), and 10% (vol/vol) DMSO for in vivo transplant experiments.

Cell Lines.

An MMTV-PyMT (FVB/n) cell line (66), which we refer to as VO-PyMT, was stably transduced with GFP and luciferase, as previously described (67).

Conditioned Media from Tumor Cells.

Primary epithelial-enriched tumor cells (C57BL/6 PyMT) were cultured in 24-well plates at 4 × 10⁵ cells/mL for 48 h in DMEM/F-12 medium with 1% FBS and P/S. VO-PyMT cells were cultured for 48 h in DMEM/F-12 medium and P/S (no FBS) with a final concentration of ∼1 × 10⁶ cells/mL. Tumor cell-conditioned media were centrifuged at 800 × g for 5 min, and then supernatant was collected and stored at −80 °C.

In Vivo Studies and Reagents.

Murine G-CSF (Peprotech) and a blocking antibody to murine G-CSF (R&D Diagnostics), along with the isotype control antibody, were prepared per manufacturer’s instructions. For in vivo stimulation, 2 μg (in 200 μL) of G-CSF were injected s.c. daily, unless indicated otherwise. Fifty micrograms (in 100 μL) anti-G-CSF or isotype control antibody was injected i.p. every 48 h (total of four injections) for ∼1 wk, unless stated otherwise.

In Vivo BrdU Proliferation.

Mice were injected i.p. with 100 μL (10 mg/mL) BrdU (BD Pharmingen) 2 or 24 h before death. Cells, isolated from multiple tissues, were first stained for extracellular surface markers, fixed, and permeabilized using the BD Cytofix/Cytoperm buffer (15 min at room temperature), washed, and then stored at 4 °C overnight. The next day, intracellular BrdU was stained according to the FITC BrdU Flow Kit Staining Protocol (BD Pharmingen). Following BrdU staining, cells were stored at 4 °C overnight, and data were then acquired on a BD LSR II flow cytometer and analyzed using FlowJo software.

BM Transplantation.

Total BM cells from donor mice (CD45.1 WT and CD45.2 G-CSF-R^−/− C57BL/6) (63) were collected from femurs by flushing. Cells were passed through a 70-μm filter, RBCs were lysed, and then cells were resuspended at 1:1 with a total of 3 × 10⁶ cells/100 μL in SM. Congenic recipient mice (CD45.1 WT C57BL/6) were irradiated using a cesium source irradiator with a lethal dose (9 Gy) delivered in two doses (4.5 Gy) 3 h apart and were given antibiotic-containing water (1.l g/L neomycin sulfate and 10 U/L of polymyxin B sulfate) for 4 wk after irradiation as described previously (64).

Statistics.

All of the data are expressed as mean ± SD or SEM as indicated. GraphPad Prism6 was used for all statistical analysis. P values were generated using the unpaired Student t test and considered significant when <0.05. In analysis of some experiments, normalization was performed by dividing individual raw data values (such as median fluorescent intensity values acquired by flow cytometry) by the mean of WT samples in that experiment.