AMPK Alpha-1 Intrinsically Regulates the Function and Differentiation of Tumor Myeloid-Derived Suppressor Cells

“Emergency” myelopoiesis is characterized by an elevated production of myeloid precursors in the bone marrow as a response to acute infection or injury (1). During this process, monocytic and granulocytic cells actively participate in the elimination of agents through direct cytotoxic events and activation of T, B, and NK cells. However, this physiologic process is de-railed in cancer, resulting in a chronic production of myeloid precursors that inhibit the development of protective antitumor immunity and support the formation, growth, and metastasis of tumors (2). These immunoregulatory immature myeloid cell populations are referred as myeloid-derived suppressor cells (MDSC) and are divided into “early stage” (e-MDSC), monocytic (M-MDSC), and granulocytic (PMN-MDSC) subsets (3). MDSC represent a key mechanism for the evasion of protective antitumor T-cell immunity and a significant obstacle for the development of effective therapies against cancer (4). Despite their undeniable relevance, there are no effective clinical therapies to permanently overcome the immunoregulatory effects of MDSC in patients with cancer (5). This can be explained in part by the exaggerated myelopoiesis, the redundancy of the regulatory pathways, and the high plasticity of MDSC in individuals with tumors (5).

The exposure of MDSC to the tumor microenvironment (TME) potentiates their ability to thwart protective antitumor immunity through the induction of multiple pathways, including the expression of arginase I and nitric oxide synthase 2 (Nos2); and the release of reactive oxygen species (ROS) and peroxynitrite (6). Also, the TME drives differentiation of M-MDSC into tumor-associated macrophages (TAM; refs. 7–9). Production of several TME-related factors, such as GM-CSF, G-CSF, and IL6, activate interconnected signaling pathways that control MDSC survival, immunosuppression, and differentiation (4). However, the precise signaling mediators whereby the TME controls MDSC-related immune suppression remain unclear.

AMP-activated protein kinase (AMPK) is a heterotrimeric complex highly conserved from yeast to animals and comprises a catalytic subunit (AMPKα1 or 2) and 2 regulatory subunits (AMPKβ and γ) that function as a metabolic sensor to maintain energy homeostasis in stressed cells (10). Myeloid cells preferentially express AMPKα1 rather than AMPKα2 (11, 12). AMPK activity is regulated by an elevated AMP/ATP ratio and other stress mediators in the TME, and is highly dependent on its expression and phosphorylation (13, 14). Upon activation, AMPK promotes metabolic plasticity through multiple processes including the promotion of fatty acid oxidation and mitochondrial homeostasis (15–17), which have been recently reported as major drivers of the immunoregulatory activity of myeloid cells in tumors (18). Although the role of AMPK is well established in cancer cells, the intrinsic effect of AMPK in the modulation of MDSC in tumors remains controversial. Initial reports showed that pharmacologic activation of AMPK blocked MDSC function in tumors (19–25), whereas additional investigations indicated that AMPK promoted MDSC activity (26–28). Development of conditional AMPK-deficient models will enable precise elucidation of the actions of AMPK signaling in tumor-associated MDSC.

In this study, we sought to dissect the interaction between the expression of AMPKα1 and the immunoregulatory effects triggered by MDSC in tumors. Conditional deletion of the Ampkα1-coding gene, Prkaa1 in myeloid cells, or inhibition of Ampkα in tumor-bearing mice, impaired MDSC suppressive activity, blunted M-MDSC-to-macrophage differentiation, and de-railed M-MDSC into antitumor cytotoxic cells by Nos2-dependent pathways. These results demonstrate the major role of AMPKα1 in the immunosuppressive activity of MDSC in tumors and provide new strategies for the therapeutic inhibition of MDSC-driven T-cell dysfunction in cancer.

Cell lines Lewis lung carcinoma (LLC), EL4 thymoma, B16-F10 melanoma (ATCC), MCA-38 colon carcinoma (Kerafast), B16-GM-CSF melanoma (Dr. Esteban Celis, Augusta University), and ID8-Defb29/Vegfa ovarian carcinoma (Dr. Conejo-Garcia, Moffitt; refs. 29, 30) were cultured in RPMI1640 (Lonza) supplemented with 10% FCS (Gemini), 25 mmol/L Hepes, 4 mmol/L l-glutamine, and 100 U/mL of penicillin–streptomycin (Invitrogen). B16 cells were transduced with lentivirus coding for nontargeting shRNA control or Csf2-targeting shRNA (Dharmacon, RHS6848 and RMM3981-200805373) and selected in 2 μg/mL puromycin-containing medium. Tumor cell lines were authenticated on May 2018 and validated to be mycoplasma-free using an ATCC Detection Kit in October 2018. All studies were conducted with cells within the first 5 passages. C57BL/6J mice (6- to 8-week-old) were from Envigo. Myeloid cell-conditional Prkaa1-deficient (Prkaa1^KO) mice were developed after breeding Prkaa1 loxP/loxP (Prkaa1^flox) mice with those carrying lysozyme promoter-driven Cre recombinase (both from the Jackson Laboratories). Pmel-1 mice and Nos2-deficient mice were from the Jackson Laboratories. Mice were subcutaneously injected with LLC, EL4, MCA-38, B16, or B16-GM-CSF cells, as reported (31). MMTV-PyMT breast tumor cells from transgenic animals (Dr. Ruffell, H. Lee Moffitt Cancer Center & Research Institute) were implanted orthotopically in the mammary fat pads, and ID8-Defb29/Vegfa tumor cells were injected intraperitoneally and mice evaluated until they reached a weight gain greater than 30% (29). Tumor volume was tested using calipers and calculated using the formula [(small diameter)² × (large diameter) × 0.5]. All studies using animals were approved by the Moffitt Institutional Animal Care and Use Committee and followed Moffitt's Comparative Medicine facility guidelines.

A tissue microarray (TMA; Moffitt Cancer Center) was available for 79 de-identified and pathologically confirmed high-grade advanced serous epithelial ovarian carcinoma tumors and 10 healthy ovary or fallopian tube tissues. Also, peripheral blood from de-identified patients with advanced ovarian carcinoma and healthy donors was obtained from a tissue repository established by Dr. Conejo-Garcia (Moffitt Cancer Center). Moreover, we obtained de-identified mobilized peripheral blood stem cells (PBSC) from healthy donors for hematopoietic stem cell transplantation (HSCT; Georgia Cancer Center Biorepository). Additionally, T-cells were isolated from de-identified buffy coats from healthy blood donors (One-Blood). Studies using de-identified human samples were covered through an exempt-approved Institutional Review Board protocol and were developed following the Regulatory Affairs Committee guidelines at Moffitt Cancer Center. Investigators and biorepository facilities obtained informed written consent forms from the de-identified subjects.

For modulation of AMPK activity, MDSC were treated with 5-aminoimidazole-4-carboxamide 1-β-d-ribofuranoside (Aica-R, 200 μmol/L; Millipore), metformin (10 mmol/L; Millipore), or Dorsomorphin-Compound C (CC, 5 μmol/L; Cayman). Moreover, LLC-bearing mice received CC (15 mg/kg, i.t.), metformin (150 mg/kg, i.p.), or Aica-R (0.5 mg/kg, i.p.) 9 days posttumor injection and continued to be treated daily until tumor endpoint. For in vitro studies inhibiting Nos2, we used L-NG-Monomethylarginine (L-NMMA, 500 μmol/L; Cayman) and Lysine-dihydrochloride (L-NIL, 300 μmol/L; Cayman), whereas for in vivo assays, LLC-bearing mice were treated daily starting at day 0 of tumor injection with 20 mg/kg L-NIL (Cayman, i.p.). Human IL6, mouse granulocyte–monocyte colony-stimulating factor (GM-CSF) and mouse granulocyte-colony-stimulating factor (G-CSF) were from Gemini. Human GM-CSF was from eBioscience. Thioglycolate broth from Sigma-Aldrich was prepared at 4% in water, autoclaved, and stored in dark for 2 weeks before intraperitoneal injection. To test the role of GM-CSF in TES-treated MDSC, we utilized blocking antibodies against mouse-GM-CSF (5 μg/mL, Clone MP1022E9) and/or mouse-GM-CSF receptor α (1 μg/mL, Clone 698423; R&D Systems). Rat IgG2a isotype (Clone 2A3; BioXcell) was used as control.

Surface staining was performed followed labeling with viability Zombie dyes (Biolegend) and purified anti-mouse CD16/CD32 antibodies (Clone 2.4G2; BD Biosciences). The following antibodies were used: CD45-BV785 or BV421 (Clone 30-F11; Biolegend), CD11b-FITC or BV421 (Clone M1/70; Biolegend), Gr1-PE-Cy5 or PE/Dazzle594 (Clone RB6-8C5; Biolegend), F4/80-APC-AF700 (Clone BM8; Biolegend), Ly6G-APC (Clone 1A8; Tonbo), Ly6C-PE or FITC (Clone AL-21; BD Biosciences). For intracellular detection of Nos2 (Clone CXNFT; eBioscience), AMPKα (Clone F6; Cell Signaling Technologies), or phospho-AMPKα (Thr172; Rabbit polyclonal 40H9; Cell Signaling Technologies), tumor cell suspensions or MDSC were cultured for 6 hours in the presence of GolgiStop (0.8 μL/mL; BD Biosciences) plus LPS (1 μg/mL; Sigma-Aldrich) for Nos2; or GolgiStop (0.8 μL/mL), phorbol myristate acetate (PMA, 750 ng/mL; Sigma-Aldrich) and ionomycin (50 μg/mL, Sigma-Aldrich) for total and phospho AMPKα. Intracellular staining was completed using Cytofix/Cytoperm and Perm-Wash buffers (BD Biosciences). T-cell proliferation was detected using carboxyfluorescein succinimidyl ester (CFSE; Molecular Probes). T cells were labeled with 1 μmol/L CFSE at 37°C for 15 minutes. For T-cell:MDSC cocultures (1:0.5 ratio, unless specified), T cells were primed with plate-bound anti-CD3 plus anti-CD28 (1 μg/mL each, Clones 145-2C11 and 37.51 from BD Biosciences) and cultured with MDSC or TAM for 72 hours. For human T-cell:MDSC (1:2 ratio) cocultures, T cells were primed with soluble anti-CD3 (1 μg/mL, clone OKT3; Thermo Fisher Scientific) and anti-CD28 (0.5 μg/mL, clone L293; BD Biosciences) in 96-well plates bound with 10 μg/mL goat anti-mouse (KPL). Results are expressed as percentage of proliferating T cells, determined by the dilution of CFSE fluorescence compared with nonstimulated T cells. Murine MDSC were identified by flow cytometry as: MDSC (CD45⁺ CD11b⁺ Gr1⁺ F4/80^neg), PMN-MDSC (CD45⁺ CD11b⁺ Gr1⁺ Ly6C^low Ly6G⁺), and M-MDSC (3) (CD45⁺ CD11b⁺ Gr1⁺ Ly6C^high Ly6G^neg); whereas TAM were recognized as CD45⁺ CD11b⁺ Gr1^neg F4/80⁺. Cells resembling human MDSC were identified by flow cytometry as e-MDSC (CD45⁺ CD33⁺ HLA-DR^neg/low CD14^neg CD15^neg), PMN-MDSC (CD45⁺ CD33⁺ HLA-DR^neg/low CD15⁺ CD14^neg), and M-MDSC (CD45⁺ CD33⁺ HLA-DR^neg/low CD14⁺ CD15^neg) (3). Cells were collected using a CytoFlex flow cytometer (Beckman Coulter). Analyses were performed using the FlowJo V10 (FlowJo).

Whole cellular lysates were electrophoresed in 8% or 10% Tris-Glycine gels (Novex-Invitrogen), transferred into nitrocellulose membranes (Bio-Rad), and immunoblotted with antibodies against phospho-Ampkα (Thr172; Rabbit polyclonal 40H9; Cell Signaling Technologies), Ampkα (Clone F6; Cell Signaling Technologies), Nos2 (Clone 54/iNOS; BD Biosciences), arginase I (Goat polyclonal N20; Santa Cruz Biotechnologies), phospho-Stat5 (Tyr694; Rabbit polyclonal D47E7, and Clone 14H2; Cell Signaling Technologies), Stat5 (Rabbit polyclonal D2O6Y; Cell Signaling Technologies), Stat5a (Rabbit monoclonal E289; Abcam), Stat5b (Rabbit polyclonal AF1584; R&D Systems), or Vinculin (Clone hVIN-1; Sigma-Aldrich; All at 1:1,000). Horseradish peroxidase linked anti-mouse IgG, anti-rabbit IgG (both from GE Healthcare), or anti-goat IgG (Santa Cruz Biotechnologies) were used as secondary antibodies and used at 1:5,000. Membrane-bound immune complexes were detected using ECL-Western Blot Substrate Reagent (Thermo Fisher Scientific) and images acquired using a Chemidoc Imaging System and analyzed using the Image-Lab software (Bio-Rad).

CD3⁺ T cells were isolated from spleen and lymph nodes of C57BL/6 mice, or from purchased human buffy-coat units (One-Blood) using T-cell negative selection kits (MagniSort; Invitrogen). Purity ranged between 95% and 99% as tested by flow cytometry. MDSC were isolated from cellular suspensions of spleen or tumors digested with DNase I and Liberase (Roche; ref. 31). Also, MDSC, M-MDSC, PMN-MDSC, and TAM were isolated by fluorescence-activated cell sorting. Splenic-MDSC were cultured in the presence of 20 ng/mL GM-CSF or 20% LLC-tumor explants (TES), prepared from filtered supernatants of primary LLC tumors cultured overnight (1 × 10⁷/mL). Also, splenic-MDSC (2 × 10⁶) from CD45.1⁺ mice bearing EL4 tumors (∼2,000 mm³) were transferred into the peritoneal cavity of CD45.2⁺ mice: (i) bearing established intraperitoneal EL4 tumors (CD45.2⁺); (ii) undergoing thioglycolate-induced peritonitis for 12 hours; or (iii) naïve controls. Peritoneal CD45.1⁺ cells were sorted 18 hours later. In vitro developed human MDSC were generated from CD34⁺ CD33⁺ PBSC of healthy donors for HSCT. Precursors were cultured for 7 days with GM-CSF and IL6 (20 ng/mL each) and different AMPKα modulating agents or 30% tumor-conditioned medium from renal cell carcinoma 786-0 cells (hTES; ref. 31).

To eliminate CD8⁺ T cells or MDSC-like cells, LLC-bearing mice were treated with 400 μg anti-CD8 (clone 53.6.72; BioXcell) or 250 μg anti-Gr1 (clone RB6-8C5; BioXcell), respectively. Maintenance intraperitoneal doses of the depleting antibodies were given every third day until tumor endpoint. In MDSC coinjection studies, 1 × 10⁶ tumor-MDSC, PMN-MDSC, or M-MDSC from Prkaa1^flox or Prkaa1^KO mice bearing LLC tumors were coinjected subcutaneously with 1 × 10⁶ LLC cells. For adoptive T-cell therapy (ACT), mice bearing B16 tumors for 6 days, received ACT with 1 × 10⁶ negatively sorted CD8⁺ pmel T-cells preactivated for 48 hours with gp100_25-33 (AnaSpec). Ten days after pmel transfer, lymph nodes were activated for 24 hours with 1 μg/mL gp100_25-33 and tested for IFNγ expression using EliSpot (R&D Systems).

Formalin-fixed paraffin-embedded baked TMA sections were transferred to a BOND RX (Leica Biosystems) and staining performed using an automated OPAL-IHC system (PerkinElmer). Briefly, slides were treated with the PerkinElmer blocking buffer for 10 minutes and incubated with the specific primary antibodies (anti-phospho-AMPKa, rabbit polyclonal 40H9, Cell Signaling Technologies; anti-CD14, Clone LPSR/2386, Abcam; anti-CD15, Clone BRA-4F1, Abcam; anti-HLA-DR, Clone CR3/43, Dako); anti-pCK, Clone AE1/AE3, Dako), followed by OPAL-HRP polymer and one OPAL fluorophore. Individual antibody complexes were stripped after each round of detection and DAPI applied as the last staining. Auto-fluorescence slides (negative control) included primary and secondary antibodies, omitting the OPAL fluorophores. Slides were imaged with a Vectra3 Automated Quantitative Pathology Imaging System. Cells resembling MDSC were identified by imaging as M-MDSC (Pan-cytokeratin^neg HLA-DR^neg/low CD14⁺ CD15^neg) or PMN-MDSC (Pan-cytokeratin^neg HLA-DR^neg/low CD15⁺ CD14^neg). Multilayer TIFF images were exported from InForm (PerkinElmer) into HALO (Indica Labs) for quantitative analysis. Each fluorophore was assigned to a dye color and positivity thresholds determined visually per marker based on intensity thresholds normalized for exposure (counts/2 bit depth × exposure time × gain × binning area). Cell segmentation results from each core were analyzed using FCS Express 6 Image Cytometry (De Novo software).

CFSE-labeled EL4 cells were cocultured with M-MDSC or PMN-MDSC (ratio 1:5; ref. 31), and in the presence or the absence of L-NMMA or L-NIL. EL4 cells were also cocultured with M-MDSC from tumor-bearing mice treated with L-NIL, or from Nos2^KO or wildtype mice treated with CC. After 24 hours of coculture, EL4 cells were stained with Violet Zombie dye and then fixed with Cytofix buffer (BD Biosciences). Percentage of cell-dead within the EL4-CFSE⁺ cells was measured by flow cytometry.

Total RNA was extracted using TRIzol (Life Technologies), according to the manufacturer's instructions. RNA was reversely transcribed using the Verso cDNA Synthesis Kit (Invitrogen). RT-PCR analyses were performed with SYBGreen (Bio-Rad) using primers synthetized by Integrated DNA Technologies. The following primers were used: Gapdh: forward 5′-CTGCCCAGAACATCATCCCT-3′, reverse 5′-ACTTGGCAGGTTCTCCAGG-3′; Prkaa1: forward 5′-GTCAAAGCCGACCCAATGATA-3′, reverse 5′-CGTACACGCAAATAATAGGGGTT-3′. Fold-change expression was calculated by comparing the RNA values from experimental samples relative to the endogenous Gapdh control, compared with the results obtained from a pooled sample. Thus, fold change = 2^−Δ(ΔCT), where ΔCT = CT (Prkaa1) – CT (Gapdh); and Δ(ΔCT) = ΔCT (Prkaa1) – ΔCT (Gapdh pool of all samples). RT² profiler cancer inflammation and immunity crosstalk PCR array (Qiagen) was done in tumor-MDSC from LLC-bearing Prkaal^flox and Prkaa1^KO mice, following the vendor's suggestions.

Chromatin immunoprecipitation (ChIP) assays were done using a SimpleChip Kit (Cell Signaling Technologies), following the vendor's recommendations. Briefly, digested and cross-linked chromatin was prepared from 4 × 10⁶ tumor-MDSC or splenic-MDSC treated or not with GM-CSF or TES for 72 hours, followed by immunoprecipitation with antibodies against phospho-Stat5, Histone H3, or rabbit IgG (Cell Signaling Technologies). Eluted and purified DNA was analyzed by qPCR with the following primers targeting the Prkaa1 promoter: forward 5′-AGCTTTCTTCCCCCTGAATACTTT-3′; reverse 5′-CTCCTAGTTCCATATTCTGGCT-3′. Primers against Rpl30 promoter provided in the kit were used as housekeeping control.

Two-tailed unpaired Student t test was used for most of the statistical analyses using the Prism software (Graph-Pad). Also, analyses of survival were assessed through Mantel–Cox test. P value of <0.05 was considered statistically significant. Specific statistical test results are indicated in each figure: *, P < 0.05; **, P < 0.01; ***, P < 0.001.

To understand the potential interaction between AMPKα signaling and the regulatory function of tumor-associated MDSC, we monitored the expression of Ampkα and the immunosuppressive activity of tumor and splenic-MDSC (CD11b⁺ Gr1⁺) from mice bearing LLC tumors; and splenic PMN, monocytes, and immature myeloid cells (CD11b⁺ Gr1⁺; iMC) from mice without tumors. An increased expression of total and phospho-Ampkα and an augmented ability to impair T-cell proliferation were found in tumor-MDSC, compared with splenic-MDSC from LLC-bearing mice or iMC from tumor-free mice (Fig. 1A and B). Also, an elevation of total and phospho-Ampkα was noted in tumor-MDSC from B16, EL4, or MMTV-PyMT-bearing mice, compared with splenic controls (Supplementary Fig. S1A). Notably, the Ampkα signaling activation in tumor-MDSC occurred in PMN-MDSC and M-MDSC (Supplementary Fig. S1B). Next, we studied whether AMPKα signaling was also heightened in human-MDSC infiltrating tumors. Using high-resolution automated multispectral imaging, we assessed the expression of phospho-AMPKα in a TMA made from 79 patients with advanced high-grade serous ovarian tumors and 10 healthy ovary controls. Higher levels of phospho-AMPKα were noticed in cells resembling M-MDSC (Pan-cytokeratin^neg HLA-DR^neg/low CD14⁺ CD15^neg) from ovarian tumors, compared with cellular counterparts from healthy ovaries (Fig. 1C). However, analysis of phospho-AMPKα in tumor PMN-MDSC could not be completed as limited numbers of Pan-cytokeratin^neg HLA-DR^neg/low CD14^neg CD15⁺ cells were found in ovarian tumors or healthy tissues (Fig. 1C; Supplementary Fig. S1C). To further explore the activation of AMPKα in MDSC subsets, we then examined the expression of phospho-AMPKα in peripheral blood from patients with ovarian cancer and healthy controls by flow cytometry. Higher phospho-AMPKα levels were found in cells resembling e-MDSC, PMN-MDSC, and M-MDSC (3) from patients with advanced ovarian cancer, compared with controls (Fig. 1D), indicating the active AMPKα signaling in circulating MDSC from patients with ovarian cancer.

Next, we investigated the contribution of the TME in the induction of Ampkα in MDSC by treating splenic-MDSC with LLC tumor explants (TES). Upregulation of total and phospho-Ampkα, higher levels of the MDSC-inhibitory factors arginase I and Nos2, and elevated T-cell immunoregulatory activity, were noticed in TES-treated MDSC, compared with controls (Fig. 1E and F). To elucidate the specific effect of the TME versus overall inflammation in the induction of Ampkα in MDSC, we transferred CD45.1⁺ splenic-MDSC into the peritoneum of CD45.2⁺ mice previously injected with peritoneal EL4 tumors or the inflammatory agent thioglycolate. CD45.1⁺ cells were sorted 18 hours later and tested for the expression of Ampkα and the ability to impair T-cell proliferation. Higher levels of total and phospho-Ampkα and an augmented capacity to block T-cell proliferation were noted in MDSC previously transferred into EL4 tumors, compared with those from chemically-induced peritonitis or untreated mice (Fig. 1G and H). Thus, results indicate the driving effect of the TME in the induction of AMPKα in MDSC and the potential interaction between AMPKα and MDSC activity.

We aimed to identify the TME factors driving Ampkα expression in tumor-MDSC. We focused on the role of GM-CSF as its production in the TME plays a primary role in MDSC activity (32, 33). Treatment of splenic-MDSC with GM-CSF induced the expression of total and phospho-Ampkα and enhanced their ability to blunt T-cell proliferation (Fig. 2A and B; Supplementary Fig. S2A-C). Moreover, blockade of GM-CSF and/or GM-CSF receptor α partially blunted the upregulation of Ampkα and the immunosuppressive activity of TES-treated MDSC (Fig. 2C and D; Supplementary Fig. S2D), suggesting the effect of the TME-associated GM-CSF in the induction of Ampkα in MDSC. Next, we tested the role of the cancer cell-derived GM-CSF in the induction of Ampkα in MDSC. Lower expression of total and phospho-Ampkα and reduced immunosuppressive activity were found in tumor-MDSC from mice bearing GM-CSF-silenced (shCsf2) B16 tumors compared with controls (Fig. 2E and F; Supplementary Fig. S2E-F). In agreement, higher levels of total and phospho-Ampkα and augmented ability to impair T-cell proliferation were found in splenic-MDSC from mice carrying B16 cells overexpressing GM-CSF (B16-GM-CSF; Fig. 2G and H; Supplementary Fig. S2G), indicating the role of cancer cell–derived GM-CSF in the induction of AMPKα in MDSC.

Next, we sought to identify the intracellular mediators by which GM-CSF promoted Prkaa1 transcription. Our recent report showed the interaction between GM-CSF production in the TME and the phosphorylation of Stat5 in tumor-MDSC (18). Elevated levels of total Stat5, Stat5a, Stat5b; and augmented expression and endogenous binding of phospho-Stat5 to the Prkaa1 promoter were found in tumor-infiltrating MDSC compared with splenic-MDSC or iMC (Fig. 2I and J; Supplementary Fig. S2H), and in splenic-MDSC treated with TES or GM-CSF (Fig. 2K and L; Supplementary Fig. S2I). Notably, the inhibition of Stat5 through Pimozide impaired the induction of Prkaa1 mRNA and Ampkα protein in GM-CSF or TES-treated splenic-MDSC (Fig. 2M). Thus, our results suggest the primary role of TME-associated GM-CSF and intrinsic signaling through Stat5 in the induction of Ampkα in tumor-MDSC.

We tested the therapeutic effects of the inhibition of AMPKα in tumor-bearing mice. Delayed tumor growth and alterations in tumor-MDSC, including impaired immunosuppressive activity, diminished expression of the immune inhibitory factor arginase I, and increased Nos2 levels, were found LLC-bearing mice treated with the AMPK inhibitor, CC, compared with controls (Fig. 3A–C). Because MDSC activity depends on tumor burden, we further tested the intrinsic effects of the modulation of AMPKα in the activity of human MDSC developed from myeloid precursors (Supplementary Fig. S3; refs. 18, 34), after exposure to effective doses of the AMPK agonists, Aica-R and metformin, or the AMPK inhibitor, CC (Fig. 3D). In agreement with the immune suppressive role of AMPK in MDSC, we observed a higher immunoregulatory activity in Aica-R pretreated MDSC and reduced immunoinhibitory potential in CC-conditioned MDSC, compared with controls, which correlated with corresponding changes in phospho-AMPKα levels (Fig. 3D and E). Also, despite the elevation of phospho-AMPKα in metformin-treated MDSC, we found a decrease in their immunosuppressive potential (Fig. 3D and E), suggesting opposite effects of the AMPK agonists, Aica-R and metformin, in human MDSC. To further evaluate possible off-target effects of the AMPK agonists, we repeated the treatments in splenic-MDSC from myeloid cell-conditional Ampkα1-null (Prkaa1^KO) mice, created after crossing Prkaal^flox mice with those carrying lysozyme-driven Cre recombinase. Prkaa1 deletion antagonized the increased immune regulatory activity observed in TES or Aica-R–exposed MDSC, without impacting metformin-treated MDSC (Fig. 3F), suggesting that TES and Aica-R, but not metformin, modulate MDSC regulatory activity in an Ampkα1-dependent manner. Furthermore, we tested the role of myeloid cell-Ampkα1 in the tumor effects induced by Aica-R or metformin treatments in mice. Prkaa1 deletion overcame the slight potentiation of tumor growth induced by Aica-R treatment in mice (Fig. 3G). Conversely, the delayed tumor growth noticed in LLC-bearing mice treated with metformin was not altered by deletion of Prkaa1 in myeloid cells (Fig. 3H), showing that Aica-R, but not metformin, modulated tumor growth in mice in a myeloid cell Ampkα-related manner.

To determine the specific contribution of Ampkα1 in myeloid cells in tumor growth, we used myeloid cell–conditional Prkaa1^KO mice. A delay in the growth of several tumors, including LLC, MCA-38, EL4, B16, and MMTV-PyMT was noticed in Prkaa1^KO mice, compared with Prkaal^flox controls (Fig. 4A). Also, extended survival was found in LLC, EL4, B16, and ID8-Defb29/Vegfa bearing Prkaa1^KO mice, compared with controls (Supplementary Fig. S4A). Furthermore, an evident decrease in the T-cell suppressive potential was noticed in tumor-MDSC from Prkaa1^KO mice, compared with those from control mice (Fig. 4B), which correlated with a lower expression of arginase I and a surprising upregulation of Nos2 (Fig. 4C). Additionally, gene transcript assessment showed increased levels of Nos2, Il12a, Tlr4, Cxcl1, and Myd88; and lower Aicda, Tnfsf10, and Cxcl10 mRNA expression in tumor-MDSC from Prkaa1^KO mice compared with controls (Supplementary Fig. S4B–S4C). Because gene targeting using Lysozyme-Cre recombinase does not specifically affect MDSC, we then distinguished the effect of the deletion of Prkaa1 on the 2 most frequent myeloid subsets in LLC tumors, MDSC and TAM. Increased accumulation of cells resembling MDSC, PMN-MDSC, and M-MDSC; and lower frequency of TAM were found in tumors from Prkaa1^KO mice compared with controls (Fig. 4D). Because M-MDSC represent a primary source for TAM expansion in the TME (7–9), we tested the effect of Ampkα in the M-MDSC-to-TAM differentiation. In accordance with the lower frequency of TAM in tumors from Prkaa1^KO mice, a diminished differentiation of TES-conditioned M-MDSC to macrophages was observed upon Prkaa1 deletion (Fig. 4E). Also, we compared the role of Ampkα1 in the T-cell–suppressive activity of MDSC versus TAM. Prkaa1 deletion impacted the immunoregulatory activity and the levels of arginase I in tumor-MDSC, while promoting Nos2 expression. However, these effects were not observed in TAM (Fig. 4F and G). Interestingly, the low immunosuppressive activity of Prkaa1^KO MDSC from tumors was observed both in PMN-MDSC and M-MDSC (Fig. 4F). Thus, results show the role of Prkaa1 in the differentiation and immunoinhibitory function of tumor-MDSC, but not in TAM suppressive activity.

We next sought to elucidate the role of T-cell immunity in the antitumor effects induced by the conditional deletion of Prkaa1 in myeloid cells. In agreement with the role of Ampkα1 in the promotion of MDSC function, elevated frequency of CD8⁺ T-lymphocytes and CD8⁺ CD44⁺ CD69⁺ antigen-experienced T cells was observed in tumors from Prkaa1^KO mice, compared with controls (Fig. 5A and B). Moreover, depletion of CD8⁺ T cells restored tumor growth in Prkaa1^KO mice (Fig. 5C). Next, we tested the impact of the MDSC-Ampkα1 in tumor-induced T-cell tolerance using an ACT model against the tumor-antigen gp100, in which activated anti-gp100_25–33 transgenic pmel T cells were transferred into mice bearing B16 tumors. A significant delay in tumor growth and a higher frequency of IFNγ-expressing pmel T cells were found in B16-bearing Prkaa1^KO mice undergoing ACT, compared with flox controls receiving the same amount of pmel T cells (Fig. 5D and E). Thus, results suggest the key role of MDSC-related Ampkα1 in tumor-induced T-cell dysfunction.

To determine whether Prkaa1 deletion transforms MDSC into cells that elicit antitumor actions, we eliminated MDSC using anti-Gr1 antibodies; and coinjected Prkaa1^flox mice with LLC cells and tumor-MDSC from Prkaa1^KO or flox mice (31). MDSC depletion restored tumor growth in Prkaa1^KO mice and delayed tumor progression in control mice (Fig. 6A). Also, in agreement with the antitumor actions of Prkaa1^KO MDSC, delayed tumor growth was found after transfer of Prkaa1^KO-MDSC, compared with mice receiving flox-MDSC (Fig. 6B). Notably, Prkaa1^KO-MDSC triggered similar antitumor effects after transfer into immunodeficient Rag^KO mice (Fig. 6C), indicating that Prkaa1 deletion transforms MDSC into cells that potentially elicit direct tumor cytotoxicity. To further explore this possibility, we assessed the cytotoxic potential of tumor-MDSC subsets from flox and Prkaa1^KO mice on cultured EL4 tumor cells. Prkaa1^KO-M-MDSC showed enhanced ability to eliminate EL4 tumor cells, compared with Prkaa1^KO-PMN-MDSC or tumor-MDSC subsets from Prkaa1^flox mice (Fig. 6D). Moreover, coinjection of LLC cells with Prkaa1^KO M-MDSC, but not with Prkaa1^KO PMN-MDSC, resulted in delayed tumor growth in both wild-type and Rag^KO mice (Fig. 6E; Supplementary Fig. S5), confirming the ability of Prkaa1^KO M-MDSC to induce direct antitumor effects. Interestingly, the increased antitumor cytotoxic activity of Prkaa1^KO-M-MDSC correlated with a higher expression of Nos2 (Fig. 6F) and was completely inhibited after treatment with the Nos2 inhibitors, L-NMMA or L-NIL (Fig. 6G), suggesting the impact of Nos2 in the antitumor cytotoxic actions induced by Prkaa1^KO-M-MDSC. Next, we elucidated whether the inhibition of Nos2 overcame the effects observed in Prkaa1^KO mice. Treatment of LLC-bearing Prkaa1^KO mice with the Nos2 inhibitor, L-NIL, partially restored tumor growth and ablated the ability of tumor-M-MDSC to kill tumor cells (Fig. 6H and I). Accordingly, knockdown of Nos2 in LLC-bearing mice partially prevented the antitumor effects and M-MDSC cytotoxic activity induced after treatment with CC (Fig. 6J and K). Thus, the overall results indicate that knockdown of Prkaa1 induced M-MDSC-like cells with the ability to elicit direct antitumor cytotoxic effects through Nos2-dependent pathways.

Our study reveals a new role of AMPKα as a direct mediator of the immunosuppressive activity and differentiation of MDSC in tumors and suggests the therapeutic potential of inhibiting AMPK signaling as a strategy to restore protective myelopoiesis in cancer.

Upregulation and phosphorylation of AMPKα are key steps in the cellular adaptation to various stress conditions, including nutrient deprivation, elevation of intracellular AMP-ADP/ATP ratio, accumulation of reactive species, and hypoxia (10). Priming of AMPKα1 occurs through its phosphorylation by the liver-kinase-B1 (LKB1) and the calcium/calmodulin-dependent kinase kinase 2 (CAMKK2; refs. 13, 14); and aims to restore energy homeostasis in stressed cells by inhibiting anabolic processes consuming ATP and by activating catabolic signals that generate ATP (15–17). Interestingly, stimulation of AMPK signaling in cancer cells has emerged as a key regulator of antitumor immune responses. Activation of AMPK by aerobic glycolysis in breast tumors promoted MDSC expansion through the production of G-CSF and GM-CSF (35). Also, rapidly proliferating cancer cells imposed nutrient competition and accumulation of multiple reactive metabolites, such as adenosine, that activated AMPK in tumor-associated myeloid cells (36). In contrast to the immunosuppressive effects of cancer cell–expressed AMPK, stimulation of AMPK in breast cancer tumors through the AMPK agonist, metformin, endogenously induced phosphorylation and degradation of PD-L1, thereby promoting protective T-cell immunity (37). Thus, priming of AMPK signaling in malignant cells can activate both antitumor and protumor responses.

Our results show that treatment of tumor-bearing mice with CC or conditional deletion of Prkaa1 in myeloid cells blunted the immunoregulatory function of MDSC and improved protective T-cell immunity. Similarly, inhibition of AMPK by CC thwarted the expression of the immune inhibitory factor arginase I in bone marrow–derived MDSC treated with GM-CSF and IL6 (28). The immunoregulatory effects of AMPK are not restricted to MDSC. Indeed, previous reports described the role of AMPK in the activation of immunosuppressive phenotypes in macrophages and in the restriction of effector T-cell expansion in tumors (12, 38). In contrast to our argument that AMPK drives the tolerogenic actions of MDSC, treatment of tumor-bearing mice with the AMPK agonists metformin, phenformin, or OSU-53 exerted antitumor activities that correlated with a lower immunosuppressive function of MDSC (19–25). Conversely, additional studies showed that treatment of M-MDSC with metformin promoted their immunosuppressive activity in allogeneic skin grafts (26); and enhanced MDSC response to the immunoinhibitory factor prostaglandin E₂ in models of doxorubicin-resistant tumors (27). The paradoxical effects induced by metformin, and other AMPK activators, on the activity of MDSC could be explained by several possibilities, including the potential combined effects of the modulation of AMPK in different cellular populations in the TME, the levels of the nutritional stress and AMPK signaling in different tumor models, and potential off-target effects (39). In agreement with the AMPKα-independent effects of metformin (39–41), our results showed that the intrinsic effects induced by metformin in MDSC were not responsive to the deletion of Prkaa1. Similarly, the antitumor effects induced by metformin in LLC-bearing mice were not impacted by the elimination of Prkaa1 in myeloid cells. Because metformin has been reported to modulate additional targets, including mTORC1, protein kinase A, and mitochondrial glycerophosphate dehydrogenase (39–41), development of therapeutic models in specific deficient mice for these targets will enable to establish the precise mechanism of action of metformin in MDSC.

Previous results demonstrated the driving effect of the TME in the differentiation of M-MDSC into TAM (7–9). Our findings show that deletion of Prkaa1 antagonized M-MDSC-to-TAM differentiation, which agrees with a recent report highlighting the role of Ampkα1 in the differentiation of atherosclerosis-linked monocytes-to-macrophages (42). Notably, although AMPK has been previously shown to control anti-inflammation signals in macrophages (11, 12), we did not observe alterations in the immunosuppressive activity or changes in the levels of arginase 1 or Nos2 in Prkaa1^KO TAM, indicating that Prkaa1 deletion impairs differentiation of M-MDSC-to-TAM, but not TAM immunoregulatory activity. Mechanistic mediators for the M-MDSC-to-TAM differentiation process include the induction of the hypoxia-inducible factor-1 alpha, the decreased transcriptional activity of Stat3, the activation of the RAR-related orphan receptor C, Notch, or Aryl hydrocarbon receptor, and the signaling mediated by M-CSF and GM-CSF receptors (8, 9, 43–45). Although the mechanisms by which AMPKα1 regulates M-MDSC-to-TAM differentiation remain to be elucidated, it is conceivable that the effect of TME-derived GM-CSF in the transcriptional induction of Prkaa1, combined with relevant stress mediators in the TME such as hypoxia or nutrient deprivation, could induce activation of Ampkα1 and promote M-MDSC-to-TAM differentiation. Moreover, accumulation of anti-inflammatory cytokines, IL10 and TGFβ in the TME could modulate AMPK and regulate M-MDSC-to-TAM differentiation (36).

The immunosuppressive activity of MDSC is regulated in part through an increased coexpression of arginase I and Nos2 and the production of peroxynitrite (6). Paradoxically, Nos2-dependent tumoricidal and immunogenic antitumor effects were induced in TIP-DCs following T-cell ACT (46). In addition, treatment of tumor-bearing mice with the synthetic double-stranded RNA analog Poly-I:C transformed MDSC into cells that produced higher levels of nitric oxide and exhibited cytotoxic activity against cancer cells (47). Similarly, our results demonstrate that Prkaa1^KO M-MDSC triggered direct antitumor cytotoxic actions in a Nos2-dependent manner. Also, the upregulation of Nos2 in Prkaa1^KO M-MDSC complements previous reports showing the inhibitory effect of AMPK on the expression of Nos2 in macrophages, myocytes, and adipocytes (48). An important question unaddressed by our results is how the expression of Nos2 in MDSC could elicit paradoxical immunosuppressive or cytotoxic antitumor effects. Although the mechanisms mediating these opposite processes remain unknown, it is conceivable that the availability of the amino acid l-arginine could play a role. In fact, the expression of arginase I in tumor-MDSC decreases the availability of the amino acid l-arginine (49), which uncouples Nos2 to generate peroxynitrite rather than nitric oxide (50). Because we observed a dramatic decrease in the expression of arginase I correlating with higher levels of Nos2 in Prkaa1^KO MDSC, it is possible that the production of immunosuppressive peroxynitrite in M-MDSC is replaced for tumor-cytotoxic nitric oxide. This prediction remains to be tested.

In summary, our results demonstrate the primary role of AMPKα1 in the immunosuppressive activities induced by tumor-MDSC; and provide strategies to overcome MDSC-driven T-cell dysfunction in tumors, which could enhance the effects of different forms of immunotherapy.

B. Ruffell is a consultant at Merck & Co., Inc. and reports receiving a commercial research grant from Tesaro, Inc. J.R. Conejo-Garcia is an external advisory board member of Compass Therapeutics, is an external advisory board member of Anixa Biosciences, and reports receiving a commercial research grant from Anixa Therapeutics, has ownership interest (including patents) in Anixa Bisociences, is an unpaid consultant, and has an advisory board relationship with KSQ Therapeutics. No potential conflicts of interest were disclosed by the other author.

Conception and design: J. Trillo-Tinoco, P.C. Rodriguez

Development of methodology: J. Trillo-Tinoco, E. Mohamed, Y. Cao, Á. de Mingo-Pulido, S. Wei

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Trillo-Tinoco, R.A. Sierra, Á. de Mingo-Pulido, D.L. Gilvary, C.M. Anadon, T.L. Costich, E.R. Flores, B. Ruffell, J.R. Conejo-Garcia

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Trillo-Tinoco, D.L. Gilvary, C.M. Anadon, E.R. Flores, J.R. Conejo-Garcia, P.C. Rodriguez

Writing, review, and/or revision of the manuscript: J. Trillo-Tinoco, E. Mohamed, Á. de Mingo-Pulido, B. Ruffell, J.R. Conejo-Garcia, P.C. Rodriguez

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Trillo-Tinoco, Y. Cao, C.M. Anadon, T.L. Costich, P.C. Rodriguez

Study supervision: J. Trillo-Tinoco, P.C. Rodriguez

Authors would like to thank J. Kroger from the Flow Cytometry core, S. McCarthy from the CLIA Tissue Imaging core, and Janis De la Iglesia, PhD, for their insightful advice. This work was partially supported by R01-CA184185 and R01-CA233512 to P.C. Rodriguez, and R01-CA157664, R01-CA124515, R01-CA178687, R01-CA211913, and U01-CA232758 to J.R. Conejo-Garcia. Support for shared resources was provided by Cancer Center Support Grant (CCSG) CA076292 to H. Lee Moffitt Cancer Center.

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.