Adipose Progenitor Cell Secretion of GM-CSF and MMP9 Promotes a Stromal and Immunological Microenvironment That Supports Breast Cancer Progression

Postmenopausal breast cancer is one of the leading causes of death in western countries (1). Epidemiologic data indicate that in overweight/obese women, breast cancer incidence is increased (2), prognosis is worsened (3), and the cancers themselves are more resistant to chemotherapy (4). White adipose tissue (WAT) promotes the growth of breast cancer in animal models, is abundant in breast tissue, and many of its cells secrete factors having paracrine and endocrine activity (5, 6). In obese women, adipocytes, inflammatory cells, and the factors they secrete are altered (7, 8) and promote a breast microenvironment that favors cancer development, growth, migration, and angiogenesis (9, 10). Use of autologous adipose tissue in breast reconstruction after mastectomy has been reported to increase the breast cancer relapse rate (11, 12). It is therefore important to identify the molecules that mediate the tumor-promoting activity of WAT, not least because they may be targets for new breast cancer therapies.

WAT includes adult stem cells with progenitor-like phenotype (13, 14). Progenitor cells have been found to support breast cancer growth and metastasis in preclinical models (15). Progenitors isolated from the WAT stromal vascular fraction consist of two subpopulations: mesenchymal progenitor cells or adipose stem cells (ASC), and endothelial cells with a progenitor-like ultrastructure (EC; ref. 16). These have been found to have complementary effects in promoting breast cancer: ASCs support epithelial-to-mesenchymal transition (EMT), whereas ECs support local tumor growth and promote angiogenesis and metastasis (15–17).

However, notwithstanding the abundant evidence that adipose progenitors promote breast cancer, the molecular mechanisms mediating this effect have not been identified. In the present study, we show that two proteins play key mediator roles: granulocyte macrophage colony-stimulating factor (GM-CSF) and matrix metallopeptidase 9 (MMP9). We found that both these proteins are upregulated in WAT progenitors exposed to breast cancer cells (in vitro and preclinical models). We also found that GM-CSF produced by breast cancer cells is an upstream inducer of the aberrant upregulation of GM-CSF and MMP9 in WAT progenitors.

We also investigated these proteins in diet-induced obese syngeneic mouse models of breast cancer, assessing their impact on the tumor microenvironment and tumor progression. Finally, we investigated the effect of metformin on this newly identified interaction. Metformin is a widely used antidiabetic drug recently proposed as a treatment for breast cancer, because its administration is associated with reduced cancer incidence, particularly in obese patients or with metabolic syndrome (18–20). We have shown previously that metformin efficiently reduces tumor growth, metastasis, and angiogenesis in preclinical models of breast cancer and that it targets both breast cancer cells and WAT progenitors (21).

Lipotransfer material was collected from 40- to 65-year-old women undergoing breast reconstruction at the European Institute of Oncology, Milan, after giving informed consent. All patient studies were conducted in accordance with the Declaration of Helsinki and performed after approval by the Institutional Review Board. WAT cells were obtained from lipotransfer material as described elsewhere (15, 16). Cells were labeled magnetically with anti-human CD45 microbeads (Miltenyi Biotec) and separated on a magnetic column (LS columns; Miltenyi Biotec). The CD45^– fraction was collected and its purity assessed by flow cytometry. This fraction was then labeled with anti-human CD31 microbeads (Miltenyi Biotec) to separate ASCs (CD34⁺CD31^–) from ECs (CD34⁺CD31⁺; ref. 16).

All mouse studies were approved by the Italian Ministry of Health and performed in accordance with the Institutional Animal Care and Use Committee. Subcutaneous and intraperitoneal WAT was obtained from 8- to 12-week-old female mice, FVB/N-Tg(MMTV-PyVT)634Mul/J (MMTV-PyMT; Jackson Laboratories), housed in our animal facilities at the IEO-FIRC Institute, Milan. WAT was purified with anti-mouse CD45 microbeads (Miltenyi Biotec).

MDA-MB-436, MDA-MB-231, HCC1937, and 4T1 breast cancer cell lines were purchased in 2014–2017 from the ATCC, tested every 6 months for Mycoplasma by means of the ATCC Universal Mycoplasma Detection Kit 30-1012, expanded, and stored according to the producer's instructions. Cells were tested and authenticated by StemElite ID (Promega) at the European Institute of Oncology, cultured for no more than 2 weeks, and used for no longer than 15 passages. MMTV-ErbB2⁺ breast cancer cells from FVB-NK1Mul/J mice were isolated and cultured as described elsewhere (22).

Breast cancer cells and WAT CD45-CD34⁺ cells were cultured together on collagen-coated 6-well plates (0.8 × 10⁶ cells each) for 72 hours in EBM-2 medium (Lonza), supplemented with 0.5% gentamycin (Thermo Fisher Scientific). Transwell cocultures were obtained using 0.4 μm transwell permeable inserts (Corning). Alternatively, the different cell types were seeded together on collagen coating. After culturing, culture media were collected, briefly centrifuged to remove cell debris, and stored at –20°C. The cells themselves were detached with Accutase (Sigma-Aldrich) and lysed in RLT Buffer (Qiagen).

The following drugs or mAbs were added directly to culture media: metformin hydrochloride (Sigma-Aldrich) 2 to 10 mmol/L; bortezomib (Sigma-Aldrich) 3–50 nmol/L; anti-human IL1β mAb (#MAB201; R&D systems) 0.2 to 0.5 μg/mL; anti-human GM-CSF mAb (#sc-377039; Santa Cruz Biotechnology) 0.5 to 1 μg/mL; SB-3CT (#S1326; Sigma-Aldrich) 0.5 to 1 μmol/L. For each set of experiments, appropriate control solutions were added to the culture medium [IgG mAb or dimethyl sulfoxide (DMSO) for bortezomib and SB-3CT].

Human GM-CSF (hGM-CSF; CSF-2 gene) shRNA lentiviral particles (sc-39391-V) and control particles (sc-108080) were purchased from Santa Cruz Biotechnology. The particles contain 3 target-specific shRNAs (19–25 nucleotides, plus hairpins) that knockdown GM-CSF gene expression. MDA-MB-436 cells were transduced according to the producer's instructions using Polybrene (sc-134220; Santa Cruz Biotechnology). Quantitative reverse transcription PCR (qRT-PCR) and ELISA (#DGM00; R&D) were used to confirm knockdown in breast cancer cells, after puromycin selection.

In vivo experiments were carried out in accordance with Italian legislation and institutional guidelines. Mice were bred and housed in pathogen-free conditions. Xenografts were generated as described elsewhere (15). Six- to 8-week-old NODSCIDIL2Rγnull (NSG) female mice (12 per study arm) were injected (mammary fat pad) with 1 × 10⁶ human breast cancer (hBC) cells (MDA-MB-436, MDA-MB-231, or HCC1937) alone or in combination with 0.2 × 10⁶ human CD45-CD34⁺ WAT cells. Five mice were injected with human CD45-CD34⁺ cells alone as negative controls. In another set of experiments, half the mice were given high-dose metformin (drinking water, 2 mg/mL) 3 days after injection (12 per study arm) until sacrifice. Tumor growth was assessed weekly using digital calipers to determine width (W, mm) and length (L, mm): volume (mm³) was calculated as L x W²/2.

Serum or plasma was prepared from blood collected every 2 weeks from the animals' tail vein; EDTA was added to plasma. Plasma levels of hGM-CSF were determined with high-sensitivity (HS) ELISA (#HSGM0; R&D). Serum levels of cholesterol, triglycerides, and glucose were determined with Architect c8000 (Abbott). When tumors reached 1.2 cm diameter, the animals were killed by CO₂ inhalation.

A syngeneic breast cancer model was generated in diet-induced obese mice. Six-week-old FVB/Hsd or BALB/cOlaHsd females (Envigo) had ad libitum access to a HFD (60% kcal fat, Brogaarden). Weight was assessed weekly. After 30 days, the animals were injected orthotopically with breast cancer cells: 0.5 × 10⁶ MMTV-ErbB2⁺ primary cells (FVB background) or 0.03 × 10⁶ 4T1 cells (BALB/c background). As negative controls, some mice were not injected with cancer cells (n = 3). Three days after injection, inhibition of GM-CSF, MMP9, or both was started by injecting the animals i.p. (n = 10 per study arm) with 50 μg anti–mGM-CSF mAb (#BE0259; BioXcells) dissolved in PBS, or 25 mg/kg SB-3CT (#S1326; Sigma-Aldrich) after dissolution in PBS containing 0.1% Tween20 and 10% DMSO. Injections were administered every other day until sacrifice. In controls, 50 μg mAb rat IgG2a (#BE0089; BioXcell) or PBS-0.1% Tween 20%–10% DMSO, or both, were administered (same schedule).

Five mice were administered high-dose metformin (2 mg/mL in drinking water) 3 days after tumor injection, until sacrifice.

Tumor growth was assessed weekly. Blood was collected monthly from the tail vein, and circulating immune cells were determined by multiparametric flow cytometry (Supplementary Table S1). When tumors reached 1.2 cm, the mice were either sacrificed or the tumor was removed (mastectomy). After mastectomy, the animals were sacrificed 30 days (MMTV-ErbB2⁺ breast cancer) or 15 days (4T1) later. Tumors, subcutaneous WAT, lungs, and spleens were collected from sacrificed animals. Tissue samples were fixed in 4% phosphate-buffered formalin and embedded in paraffin. Other samples were stored for RNA (in RNAlater; Qiagen) or protein analysis (frozen in liquid nitrogen). Five-μm-thick sections of lungs were stained with hematoxylin and eosin (H&E) to assess the presence of metastases. Images were acquired with a ScanScope XT scanner (Leica) and analyzed with Aperio Digital Pathology software.

The genomic DNA of FVB/N-Tg(MMTV-PyVT)634Mul/J animals was analyzed by PCR of tail biopsies, using the Gentra Puregene Mouse Tail Kit (#158267; Qiagen). Primers, internal positive controls, and cycling conditions were as recommended by The Jackson Laboratory.

Culture media were screened for 55 soluble factors involved in human angiogenesis using a Proteome Profiler Antibody Array (#ARY007; R&D). Most factors were further quantified by ELISA (#DGM00, #DMP900, #MGM00, #MMPT90; R&D).

For Western blot analyses, frozen tissue samples were first lysed in RIPA Buffer; culture media were used directly. Total protein concentrations in samples were measured by BCA assay (Thermo Fisher Scientific). Protein concentrations were equalized by appropriate dilution and the samples (10–15 μg protein) run on 7.5% Mini Protean gels (BioRad), followed by transfer to nitrocellulose membranes (Amersham; GE Healthcare) using the Trans-Blot Turbo Transfer System (BioRad).

Membranes were blocked with 5% skimmed milk in PBS-0.1% Tween20 for 1 hour at room temperature. Immunostaining was performed overnight at 4°C with rabbit monoclonal anti-human MMP9 (1:2,000; #ab76003; Abcam) or rabbit polyclonal anti-human MMP9 (1:500; #10375-2-AP; Proteintech) and anti-human beta actin (1:5,000; #A5441; Sigma-Aldrich). Detection used horseradish peroxidase–conjugated secondary antibodies (1:5,000; Thermo Fisher Scientific), SuperSignal West Dura (Thermo Fisher Scientific), and the ChemiDOC (BioRad) imaging system. Bands were quantified with ImageJ software.

Culture media or whole tumor lysates (15–20 μg) were loaded 1:2 with Zymogram sample Buffer (BioRad) and run on 10% Zymogram gels (BioRad). Gels were then incubated for 0.5 hour in Renaturation Buffer (BioRad), followed by overnight incubation in Developmental Buffer (BioRad) at 37°C, and staining with PageBlue (Thermo Fisher Scientific) for 5 hours at room temperature. Images were acquired with ChemiDOC.

RNA was extracted from samples with the Qiamp Mini Blood Kit (Qiagen). RNA quantity and quality were checked with NanoDrop 2000 (Thermo Fisher Scientific). RNA (0.5–1 μg) was retrotranscribed with the Ipsogen RT Kit (Qiagen). qRT-PCR was carried out on the ABI Prism 7000 platform (Thermo Fisher Scientific) using primers and probes from TaqMan Gene Expression Assays (Thermo Fisher Scientific). RT² Profiler PCR array for mouse inflammation and immunity cross-talk (#PAMM-181Z; Qiagen) was performed following the manufacturer's instructions. The collected data were analyzed with web-based software (RT² Profiler PCR Array Data Analysis; SABiosciences; Qiagen).

At least 500,000 cells per sample were acquired using a 3-laser flow cytometer (Navios; Beckman Coulter). Viable cells (negative for 7-aminoactinomycin) were labeled with a panel of antibodies (Beckman Coulter or BD Biosciences) to analyze immune cell populations and WAT-derived progenitors (Supplementary Table S1). Lymphocytes, macrophages, granulocytes, and dendritic cells were characterized using standard markers (23, 24). Myeloid-derived suppressor cells (MDSC) and inflammatory monocytes were identified as a recent classification (25). Tumor-associated macrophages (TAM) were identified according to Su and colleagues (26). We have previously described the complete characterization of WAT-derived progenitors, including ASCs and ECs (15, 16).

Immunofluorescence studies were performed as described previously (21). Samples were incubated with rabbit polyclonal anti-mouse CD31 (1:20, #ab28364; Abcam) and mouse monoclonal anti-alpha smooth muscle actin (αSMA, 1:3,000, #A2547; Sigma-Aldrich). Secondary antibodies were Alexa Fluor 555 goat anti-mouse IgG2a, and Alexa Fluor 488 donkey anti-rabbit IgG2a (1:200; Thermo Fisher Scientific). Nuclei were revealed with 4′,6-diamino-2-phenylindole (DAPI; Sigma-Aldrich). As negative controls, primary antibodies were omitted. Images were captured with a SP5 II confocal microscope (Leica), x20 objective, and optical configuration as described previously (21). Blood vessel density was quantified with ImageJ software.

Results are summarized as mean and SEMs. The Shapiro–Wilk test was used to assess normality. Most data were not normally distributed so all statistical comparisons used the nonparametric Mann–Whitney U test. All P values are two sided. Differences were considered significant for P < 0.05 after Bonferroni correction. The statistical analyses were performed with GraphPad Prism software.

Cocultures of purified WAT progenitors (CD45-CD34⁺) with breast cancer cell lines were analyzed for soluble factors. Two secreted molecules were found upregulated compared with single cultures: GM-CSF and MMP9 (Fig. 1A). Upregulation was confirmed by ELISA (GM-CSF) and Western blot (MMP9) in WAT progenitors from 15 women cultured with various human triple-negative breast cancer (TNBC) cell lines (Fig. 1B–D). TNBC cells expressed considerably higher baseline levels of GM-CSF than WAT cells cultured alone (Fig. 1B). Zymography revealed increased MMP9 activity in direct cocultures, consistent with the greater expression found on Western blot (Fig. 1C). Transcriptional analysis showed that WAT progenitors were the primary source of GM-CSF and MMP9 in cocultures (Fig. 1E and F). Coculture of murine WAT (mWAT) CD45-CD34⁺ progenitors with hBC further supported these findings because only murine transcripts were upregulated (Supplementary Fig. S1A and S1B).

To assess GM-CSF and MMP9-specific induction in WAT progenitors, in additional experiments, we cocultured human WAT-derived hematopoietic cells (CD45⁺CD34^–) and WAT progenitors (CD45-CD34⁺) with hBC cells. The CD45⁺CD34^– population, which constitutes the main part of the stromal vascular fraction, is increased in WAT from obese persons (9). Upregulation of GM-CSF and MMP9 mRNA did not occur in CD45⁺CD34^– cells on exposure to breast cancer cells, but did occur in CD45-CD34⁺ cells (Supplementary Fig. S1C).

As we reported previously (16), two progenitor subpopulations are present in the CD45-CD34⁺ fraction of WAT: ASCs (CD45-CD34⁺CD31^–) and ECs (CD45-CD34⁺CD31⁺). When cultured alone, ASCs and ECs did not express detectable levels of GM-CSF, but on coculturing with breast cancer cells, GM-CSF was significantly upregulated in both ASCs and ECs (Supplementary Fig. S1D).

Although ECs expressed higher baseline levels of MMP9 than ASCs, MMP9 release was significantly increased when both progenitors were cocultured with breast cancer cells (Supplementary Fig. S1E). We also found that transcripts of GM-CSF and MMP9 were upregulated in ECs and ASCs when cocultured with breast cancer cells (Supplementary Fig. S1F). Thus, both ASCs and ECs contribute to greater production of GM-CSF and MMP9 in WAT progenitors when exposed to breast cancer cells.

MAbs or inhibitors affecting putative upstream regulatory pathways, as identified in previous studies (27–29), were investigated. These were added to cocultures to determine whether they inhibited GM-CSF or MMP9 release (Fig. 2). The NF-κB regulatory pathway was inhibited by bortezomib using concentrations that did not affect cell viability (data not shown). IL1β signaling was blocked by anti-human IL1β mAb. SB-3CT (irreversible inhibitor of MMP9) and anti–hGM-CSF mAb were used to exclude reciprocal regulation. Bortezomib was found not to impair GM-CSF or MMP9 expression in cocultures of breast cancer cells and WAT progenitors (Fig. 2A–E). However, IL1β neutralization impaired MMP9 release in direct cocultures (Fig. 2B) and MMP9 transcript expression in WAT progenitors (Fig. 2E). Neither IL1β neutralization (Fig. 2A and D) nor SB-3CT (Fig. 2A) had any effect on GM-CSF release. However, GM-CSF release was impaired by GM-CSF neutralization in cocultures (Fig. 2A). GM-CSF neutralization also resulted in dose-dependent reduction in MMP9 release (Fig. 2C). qRT-PCR provided additional support for this finding because GM-CSF and MMP9 transcripts were reduced in WAT progenitors after GM-CSF neutralization (Fig. 2D and E). These results suggest that GM-CSF, produced by breast cancer cells, may be an upstream regulator of GM-CSF/MMP9 release by WAT progenitors.

The expression of human GM-CSF (hGM-CSF) and human MMP9 (hMMP9) was assessed in xenograft models of TNBC. NSG mice were injected with TNBC cells alone or with hWAT CD45-CD34⁺ progenitors (Supplementary Fig. S2A–S2C). Circulating hGM-CSF was quantified by ELISA and found to be significantly higher in coinjected (hWAT+hBC) than single-injected (hBC) mice (Fig. 3A; Supplementary Fig. S2D). Mice injected with hWAT progenitors were used as negative controls to support the specificity of the assay.

hMMP9 expression was also investigated in tumors and was found at higher levels in coinjected (hBC+hWAT) than single-injected tumors (Fig. 3B; Supplementary Fig. S2E). The full-length precursor (pro-MMP9, 92 kDa) and cleaved (active) MMP9 forms (82 or 67 kDa) were upregulated in whole tumor lysates from coinjected mice, with the biologically active 67 kDa form expressed most prominently. Zymography showed that MMP9 had greater enzymatic activity in coinjected tumors (Fig. 3B).

qRT-PCR of breast cancer cells isolated from mouse tumors did not indicate significant induction of GM-CSF or MMP9 transcripts, suggesting that WAT progenitors were secreting these proteins (Fig. 3C). To provide support for this supposition, mWAT CD45-CD34⁺ progenitors were isolated from transgenic tumor-bearing mice (MMTV-PyMT). These cells were compared with mWAT progenitors collected from wild-type (WT) mice of the same age and sex. It was found the GM-CSF and MMP9 transcripts were significantly upregulated in WAT from transgenic mice compared with WAT from WT mice (Supplementary Fig. S3A). Protein release was also significantly greater in WAT from transgenic mice (Supplementary Fig. S3B), indicating that in vivo, upregulation of these proteins in adipose progenitors depends on the presence of breast cancer.

GM-CSF was identified in vitro as a potential upstream regulator of both GM-CSF and MMP9 production in WAT progenitors. To provide further evidence on this, lentiviral vectors were used to knockdown GM-CSF in the MDA-MB-436 TNBC cell line. shGM-CSF hBC cells were injected into NSG mice alone or together with hWAT progenitors. GM-CSF knockdown in tumors was demonstrated by qRT-PCR (Fig. 3D). Tumor growth in coinjected (hBC+hWAT) mice was significantly impaired when GM-CSF knockdown cells were used, compared with scramble, and did not differ from that in single-injected (hBC) mice (Fig. 3E). There were also fewer metastatic foci in lungs in GM-CSF knockdown hBC cells (Fig. 3F). These data indicate that breast cancer–derived GM-CSF supports local tumor growth and metastatic progression in preclinical models. Plasma hGM-CSF was significantly reduced after GM-CSF knockdown in both single- and coinjected mice (Fig. 3G). Thus, GM-CSF released from breast cancer cells is required to support GM-CSF production in WAT progenitors. Our data also indicate that GM-CSF is an upstream regulator of MMP9 because we found that hMMP9 protein levels in tumor lysate were significantly lower in GM-CSF knockdown tumors compared with scramble in coinjected mice (Fig. 3H).

The preclinical effects of inhibiting GM-CSF and MMP9 were investigated in immunocompetent obese mice injected with TNBC (4T1 cells, in BALB/c background) or MMTV-ErbB2⁺ (in FVB/Hsd background). Because WAT progenitors are significantly increased (per unit weight of WAT) in obese mice (15, 17), before injection of breast cancer cells, the mice were fed an HFD and rapidly gained body weight (Supplementary Fig. S4A). BALB/c mice have been reported as more resistant to obesity than FVB mice, although they have similar adiposity (30). After tumor injection, the animals were administered anti–mGM-CSF mAb, SB-3CT (MMP9 inhibitor), or both. Mouse weight was unaffected by these treatments (data not shown). Preliminary experiments (using IgG2a mAb or vehicle as controls) showed that both anti–mGM-CSF mAb and SB-3CT reduced tumor volume and metastatic spread in obese mice (Supplementary Fig. S4B and S4C). Administration of GM-CSF and MMP9 inhibitors together revealed they acted synergistically to reduce local breast tumor growth in MMTV-ErbB2⁺ breast cancer (Fig. 4A). Metastatic spread to lungs was investigated 30 days after mastectomy. Administration of both inhibitors together had the best efficacy in reducing metastatic spread (P < 0.001; Fig. 4B).

In the other obese model of breast cancer (4T1-injected Balb/c), tumor growth was significantly reduced by GM-CSF neutralization and MMP9 inhibition (P < 0.001; Fig. 4C). SB-3CT was the most effective in reducing lung metastases in this model (P < 0.01; Fig. 4D). Spleens from these mice were less enlarged (lower weight) than those from IgG2a⁺ vehicle controls (P < 0.01, Fig. 4E). These findings indicate that GM-CSF and MMP9 are involved in the local and metastatic growth of triple-negative and ErbB2 overexpressing breast cancers in diet-induced obese mice.

GM-CSF and MMP9 are known modulators of angiogenesis. GM-CSF influences angiogenesis by regulating the coordinated expression of VEGF (31); MMP9 triggers the angiogenic switch during carcinogenesis (32). To investigate whether angiogenesis is affected by GM-CSF neutralization or MMP9 inhibition, we performed double immunofluorescence analysis, staining endothelial cells (CD31⁺ cells) and pericytes (αSMA⁺ cells) in tumors from obese mice. In both models, GM-CSF neutralization and MMP9 inhibition resulted in strong inhibition of intratumoral angiogenesis, with significantly (P < 0.01) lower microvessel density (MVD; Fig. 5A–C). Use of both inhibitors together further impaired tumor angiogenesis, suggesting that both factors are crucially involved in breast cancer neovascularization. Use of both inhibitors did not affect the ratio of αSMA⁺ to CD31⁺ blood vessels, but did preferentially target αSMA⁺ cells (Fig. 5D).

Multiparametric flow cytometry was used to investigate immune cells (for gating strategy, see Supplementary Fig. S4D). Peripheral blood, tumors, and peritumoral WAT were collected from obese tumor-bearing mice. A month after starting GM-CSF neutralization, the number of circulating monocytes was found to be significantly lowered in GM-CSF–neutralized mice, compared with controls (CTR)-administered IgG2a mAb (Fig. 5E). Immunosuppressive cells were also significantly reduced by GM-CSF neutralization in tumors and peritumoral WAT: T-regulatory (Treg) cells, and granulocytic (G)-MDSCs were reduced in tumors and WAT (Fig. 5F and G), whereas monocytic (M)-MDSCs were reduced in tumors only (Fig. 5F). Macrophage and TAM infiltration was also significantly reduced by GM-CSF neutralization. Taken together, these findings suggest that GM-CSF is able to promote the recruitment of immunosuppressive cells to the tumor microenvironment, to thereby promote immune escape by tumor cells. By contrast, MMP9 inhibition (SB-3CT) had no effect on immune cells in peripheral blood, WAT, or tumors (data not shown).

To further investigate immunosuppression, WAT and tumor cells from GM-CSF–neutralized mice were analyzed by qRT-PCR. In comparison with IgG2a controls, transcripts of several immunosuppressive factors were downregulated, including IL10, IL5, CXCL5, CCL22, CCL4, CXCR5, and CD274 (PD-L1; Supplementary Fig. S4E). GM-CSF neutralization more profoundly reduced the expression of these genes in WAT than tumor cells. It is noteworthy that GM-CSF and IL1β gene expression was also downregulated in WAT and tumors, providing in vivo support to the GM-CSF regulation observed in vitro.

The antidiabetic drug metformin has been reported to inhibit the onset of several breast cancer subtypes in diabetic patients, especially those who are obese (18, 19). The drug targets neoplastic and microenvironment cells, including endothelial cells and WAT-derived progenitors (17, 21). Metformin was added directly to in vitro cocultures to determine whether it affects GM-CSF and MMP9 release. Metformin was added at the concentration of 5 mmol/L—a level at which cells remained viable at 72 hours (Supplementary Fig. S5A and S5B). GM-CSF and MMP9 release was reduced in coculture media to which metformin had been added, but not in single-cell cultures (Fig. 6A and B). As expected, zymography showed reduced MMP9 enzymatic activity in the presence of metformin (Fig. 6C). GM-CSF and MMP9 transcript levels were unaffected by metformin (data not shown), suggesting that metformin does not affect the expression of these proteins at the transcriptional level.

Xenograft NSG mice, injected with hBC cells alone or with CD45-CD34⁺ hWAT progenitors, were given metformin in drinking water (2 mg/mL) at a higher dosage than that used in diabetic patients, but found to be effective and nontoxic in preclinical models of breast cancer (21). Neither mouse weight, nor serum levels of glucose, cholesterol, or triglycerides were significantly affected by the drug (data not shown). As expected in this model (21), tumor growth was significantly reduced (Supplementary Fig. S5C). Plasma levels of hGM-CSF were reduced in metformin-administered mice, compared with controls (Fig. 6D). hMMP9 expression in whole tumor lysates, previously shown to be upregulated in hBC-hWAT coinjected mice, was reduced in metformin-treated mice (Fig. 6E). By contrast, metformin did not reduce hMMP9 expression in tumor lysates from single-injected (hBC) mice, in fact pro-MMP9 levels were higher in metformin-administered mice (Fig. 6E).

Because metformin reduces GM-CSF and MMP9 release in vitro and in vivo, we investigated the effect of the drug in obese syngeneic models of breast cancer. Mice were administered either metformin, anti–mGM-CSF mAb, or SB-3CT (Fig. 7). These treatments significantly reduced local breast cancer growth (tumor volume) compared with control, with GM-CSF-neutralization achieving a greater (not significant) reduction (Fig. 7A). Metastases in lungs were significantly reduced in treated mice (P < 0.01; Fig. 7B).

Quantification of intratumoral CD31⁺αSMA⁺ blood vessels indicated that metformin was more effective (P < 0.001) in reducing angiogenesis than GM-CSF neutralization or MMP9 inhibition (P < 0.01; Fig. 7C). Flow cytometry of tumors and peritumoral WAT indicated that metformin had similar effects to GM-CSF neutralization (Fig. 7D): several immunosuppressive populations, including Tregs and TAMs, were downregulated in peritumoral WAT, whereas G-MDSCs were strongly downregulated in tumors. Metformin significantly reduced the presence of inflammatory monocytes in both tumors and peritumoral WAT, providing further evidence of its anti-inflammatory activity. These findings suggest that GM-CSF and MMP9 might be previously unrecognized targets of metformin, which would explain some of its antitumor effects in breast cancer preclinical models.

We investigated interactions between breast cancer cells and CD45-CD34⁺ WAT progenitors because the latter promote tumor growth, metastasis, and angiogenesis in preclinical breast cancer models (15, 16). We found that the proteins GM-CSF and MMP9 were significantly upregulated in murine and human WAT progenitors on exposure to a variety of breast cancer cell types in vitro and in vivo. These proteins were not upregulated in other WAT cells (e.g., hematopoietic cells). Both ASCs and ECs—constituents of the CD45-CD34⁺ WAT fraction—produced GM-CSF and MMP9 when cocultured with breast cancer cells, consistent with previous findings that ASCs and ECs cooperate to support breast cancer growth, angiogenesis, and metastatic spread (16).

GM-CSF is a growth factor for hematopoietic and immune cells, which mobilizes stem cells and induces macrophage/granulocyte differentiation (33). Other roles include the regulation of inflammation and autoimmunity (34). MMP9 is a type IV collagenase whose increased expression has been reported associated with higher grade, metastasis, and angiogenesis in several cancers (29, 35).

We investigated various pathways to clarify breast cancer–dependent GM-CSF/MMP9 upregulation. Our findings indicate that GM-CSF produced by breast cancer cells induces GM-CSF and MMP9 release from WAT progenitors; IL1β might be also induced by GM-CSF, as our and previous data suggest (28), possibly implicating it in MMP9 release by WAT progenitors.

In mice coinjected with breast cancer cells and WAT progenitors, we found greater tumor volume and more metastases than in mice injected with breast cancer cells alone. However, when GM-CSF was knocked down in breast cancer cells, there was no increase in tumor growth. Thus, tumor-derived GM-CSF seems to be essential for triggering the protumorigenic actions of WAT progenitors. Other studies support the existence of such regulation. Thus, GM-CSF has been reported to upregulate MMP9 in squamous cell carcinoma (29), and GM-CSF feedback regulation has been reported in myeloid cells (36). Considering that CD116, the GM-CSF receptor, is expressed on myeloid cells (36), we are now investigating why WAT progenitors do not produce GM-CSF and MMP9 in isolation but only when cocultured, and the role of CD116 expression in obesity and WAT-embedded tumors.

To further explore the roles of GM-CSF and MMP9 in promoting breast cancer, we examined the effects of inhibiting these proteins in a syngeneic breast cancer model generated in immune-competent obese mice. Obese mice were used because WAT progenitors are markedly increased in adipose tissue in obesity (17). Use of immune-competent animals allowed us to investigate the effect of GM-CSF on immune cells. We found that both GM-CSF neutralization and MMP9 inhibition significantly reduced tumor volume and number of metastases. However, these antitumor effects may not be due (entirely) to direct interaction with tumor cells, but to microenvironment modulation, because GM-CSF neutralization reduced angiogenesis and immunosuppressive cells within the tumor, and MMP9 inhibition reduced angiogenesis and metastasis. MMP9 may exert its cancer-promoting effects by multiple mechanisms including degrading basement membranes (35), thus MMP9 inhibition could reduce tumor invasion into surrounding tissues.

Some recent data suggest that although angiogenesis inhibitors may suppress growth of primary tumors, they might also push the tumor into a more aggressive metastatic state (37). Accordingly, more studies in different models would be useful to better investigate the role of GM-CSF and MMP9 blockade in cancer local and metastatic progression.

Myeloid cells are known to be targeted by GM-CSF, which has been identified as a macrophage chemoattractant in the presence of WAT inflammation (28). In our models, GM-CSF neutralization was associated with reduced numbers of monocytes in PB, and reduced macrophages and MDSCs in tumors and peritumoral WAT, whereas TAMs were reduced in tumors (TNBC model) and also peritumoral WAT—supporting the finding of Su and colleagues (26) that GM-CSF as a TAM inducer in breast cancer. Macrophage accumulation has been found to promote tumor growth in obesity by increasing chronic inflammation, immune escape, and angiogenesis (38). Thus, it is possible that some of the antitumor effects of GM-CSF neutralization found in our study might be due to reduced macrophage activation. We note also that GM-CSF can expand MDSCs in breast cancer (39).

Lymphoid cells were not affected by GM-CSF neutralization in our experiments, consistent with reported lack of expression of GM-CSF receptors on T cells, B cells, and natural killer cells (40). Nevertheless, Tregs were downregulated in GM-CSF–neutralized mice, so GM-CSF may modulate these cells, as suggested elsewhere (41). Several immunosuppressive interleukins and cytokines have also been reported downregulated in WAT and tumors from GM-CSF–neutralized mice, including IL10, CCL22, and PD-L1 (42–43).

The prominent protumorigenic effect of GM-CSF found in our obese syngeneic models is consistent with recent findings that GM-CSF ablation in in vitro primary pancreatic cancer cells resulted in defeated immune escape (44). Nevertheless, the tumor-promoting effects of GM-CSF are surprising when considered alongside current therapeutic applications of GM-CSF such as hematopoietic stem cell mobilization (45) and immunotherapy (46). However, many GM-CSF effects are dose- and context-dependent (47), Eubank and colleagues (48) have reported opposite effects of GM-CSF in a different breast cancer model, thus the quantity of GM-CSF present in the tumor microenvironment may determine its biological effects on WAT progenitors.

As regard MMP9 inhibition, achieved by administration of SB-3CT, we found that this was associated with significant reduction in tumor progression in obese preclinical models. However, in general, MMP inhibitors have failed to produce beneficial effects in cancer patients (49). This may be due to enrollment of patients with late-stage disease, whereas MMP9 inhibition might be more efficient in early stage disease or in the presence of concomitant obesity.

We used metformin to disrupt the interaction between WAT progenitors and breast cancer cells. We and others have shown that metformin delays tumor progression by effects on both breast cancer cells and WAT progenitors (17, 21). Metformin also reduces hyperglycemia and hyperinsulinemia, which are frequently associated with inflammation of breast WAT and breast cancer in humans (50). We found that metformin significantly reduced GM-CSF and MMP9 release in cocultures, but not in single-cell cultures, without affecting transcription. In vivo, in xenograft mice, high-dose metformin reduced levels of hGM-CSF and hMMP9 proteins. Metformin probably reduced the translation of these proteins in association with mTOR inhibition and decreased phosphorylation of S6 kinase (51). When we administered high-dose metformin to obese breast cancer models, it significantly reduced local tumor growth and metastasis, in a similar way to GM-CSF/MMP9 inhibition. The drug was also effective in reducing intratumoral angiogenesis and intratumoral immunosuppressive cells (TAMs, G-MDSCs, and Tregs). Metformin blocks GM-CSF and MMP9 production only in cocultured cells. As there is an evidence that—depending upon models and drug dosage—metformin can target a variety of pathways other than mTOR (17, 21, 51), we are currently investigating multiple potential effects of metformin in the presence or in the absence of concomitant GM-CSF or MMP9 blockade.

Overall, these findings suggest that GM-CSF and MMP9 may be new targets of metformin in breast cancer and add support to use of the drug in clinical studies on breast cancer, particularly in women who are obese or have insulin resistance. Nevertheless, further studies are required to clarify the mechanisms by which metformin prevents GM-CSF/MMP9 upregulation in WAT progenitors.

Although in our model GM-CSF/MMP9 blockade had an effect also in WAT-poor organs such as the lungs, it would be crucial to further investigate how metformin and GM-CSF/MMP9 blockade can be effective in metastases arising in organs where there is little or no WAT, like—for instance—the brain or the bone, and how metformin and GM-CSF/MMP9 blockade can be effective in inhibiting primary breast cancer cells collected from these metastatic sites. Another relevant field for investigation will involve the study of the protumorigenic role of WAT cells from mice of different ages and different strains.

No potential conflicts of interest were disclosed.

Conception and design: F. Reggiani, F. Bertolini

Development of methodology: F. Reggiani, C. Rabascio, F. Bertolini

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Reggiani, P. Mancuso, G. Talarico, S. Orecchioni, A. Manconi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Reggiani, V. Labanca, P. Mancuso, C. Rabascio, F. Bertolini

Writing, review, and/or revision of the manuscript: F. Reggiani, F. Bertolini

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Bertolini

Study supervision: V. Labanca, F. Bertolini

F. Bertolini is a scholar of the U.S. National Blood Foundation. The authors thank Stefano Freddi and Giovanna Jodice for technical assistance, and Donald Ward for help with the English text.

This study was supported in part by the Italian Association for Cancer Research (AIRC grant IG14188 to F. Bertolini), the Umberto Veronesi Foundation (to F. Reggiani and F. Bertolini), and the Italian Ministry of Health (to F. Bertolini).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.