Aerobic Glycolysis Controls Myeloid-Derived Suppressor Cells and Tumor Immunity via a Specific CEBPB Isoform in Triple-Negative Breast Cancer

SUMMARY

Myeloid-derived suppressor cells (MDSCs) inhibit anti-tumor immunity. Aerobic glycolysis is a hallmark of cancer. However, the link between MDSCs and glycolysis is unknown in patients with triple-negative breast cancer (TNBC). Here, we detect abundant glycolytic activities in human TNBC. In two TNBC mouse models, 4T1 and Py8119, glycolysis restriction inhibits tumor granulocyte colony-stimulating factor (G-CSF) and granulocyte macrophage colony-stimulating factor (GM-CSF) expression and reduces MDSCs. These are accompanied with enhanced T cell immunity, reduced tumor growth and metastasis, and prolonged mouse survival. Mechanistically, glycolysis restriction represses the expression of a specific CCAAT/enhancer-binding protein beta (CEBPB) isoform, liver-enriched activator protein (LAP), via the AMP-activated protein kinase (AMPK)-ULK1and autophagypathways, whereas LAP controls G-CSF and GM-CSF expression to support MDSC development. Glycolytic signatures that include lactate dehydrogenase A correlate with high MDSCs and low T cells, and are associated with poor human TNBC outcome. Collectively, tumor glycolysis orchestrates a molecular network of the AMPK-ULK1, auto-phagy, and CEBPB pathways to affect MDSCs and maintain tumor immunosuppression.

In Brief

Tumor-derived myeloid-derived suppressor cells (MDSCs) are critical tumor immunosuppression components. Li et al. show that the high glycolytic rate in triple-negative breast cancer cells is associated with MDSC promotion through an AMPK-ULK1 and autophagy pathway. Glycolysis restriction inhibits tumor G-CSF and GM-CSF and consequently MDSC development.

Graphical Abstract

INTRODUCTION

Tumors reprogram metabolic pathways to meet the bioenergetic, biosynthetic, and redox demands of malignant cells. These reprogrammed activities are recognized as hallmarks of cancer (Hanahan and Weinberg, 2011). Interestingly, recent work has shown that tumors actively reprogram metabolic pathways to evade effective anti-tumor immunity. It has been reported that glycolysis regulates T cell activation and effector function (Chang et al., 2013; Gubser et al., 2013). Given that nutrients, including glucose, are poorly replenished in the tumor, it is assumed that T cell glycolytic metabolism has been altered due to the Warburg effect in the tumor microenvironment (Brand et al., 2016; Chang et al., 2013, 2015; Ho et al., 2015; Zhao et al., 2016). In support of this, tumor glycolysis can alter effector memory (Brand et al., 2016; Chang et al., 2015; Zhao et al., 2016) and naive (Xia et al., 2017) T cell function in the tumor microenvironment and tumor-draining lymph nodes. Furthermore, the oxygen-sensing prolyl-hydroxylase proteins (Clever et al., 2016), necrotic cells releasing potassium ions (Eil et al., 2016), and abnormal zinc metabolism (Singer et al., 2016) can impair effector T cell function in the tumor microenvironment. In addition to T cells, recent studies have shown that natural killer cell function is impaired by tumor glycolysis (Brand et al., 2016) and myeloid dendritic cells (Cubillos-Ruiz et al., 2015), and regulatory T cells (Maj et al., 2017) are functionally altered by oxidative stress in the tumor microenvironment.

Myeloid-derived suppressor cells (MDSCs) are a chief component of immunosuppressive networks (Gabrilovich et al., 2012; Huang et al., 2006; Kusmartsev et al., 2000; Ma et al., 2011; Zou, 2005). Human MDSCs inhibit T cell immunity and promote cancer stem-like properties in the tumor microenvironment in patients with cancer (Cui et al., 2013; Peng et al., 2016). Tumor cells secrete a variety of factors, including granulocyte colony-stimulating factor (G-CSF) and granulocyte macrophage colony-stimulating factor (GM-CSF), to promote MDSC development (Gabrilovich et al., 2012; Morales et al., 2010; Shojaei et al., 2009). However, the potential link between MDSCs and tumor glycolysis is not established in patients with breast cancer, including triple-negative breast cancer (TNBC).

TNBC has been characterized by several aggressive clinical features—including high rates of metastasis, recurrence, and poor survival—compared with those with no-TNBC breast cancers (Bauer et al., 2007; Bianchini et al., 2016; Harris et al., 2016; Schott and Hayes, 2012). In the present work, we have focused our studies on TNBC. We have examined the interactions between glycolytic metabolism and immune system in two mouse TNBC models and extended our research to patients with TNBC. We have found that tumor glycolysis regulates the expression of the secondary isoform of CCAAT/enhancer-binding protein beta (CEBPB), liver-enriched activator protein (LAP), via the AMP-activated protein kinase (AMPK)-ULK1, and auto-phagy-signaling pathways; LAP subsequently controls the expression of G-CSF and GM-CSF in tumor cells and consequently affects MDSC development, anti-tumor immunity, and TNBC outcome.

RESULTS

Glycolysis Regulates Tumor G-CSF and GM-CSF Expression

Aerobic glycolysis affects effector T cell phenotype and function in patients with cancer (Chang et al., 2015; Zhao et al., 2016). Human MDSCs inhibit tumor immunity and endow stem-like properties to cancer cells in patients with cancer, including TNBC (Cui et al., 2013; Peng et al., 2016). TNBC patients have weak anti-tumor immunity and poor survival with limited therapeutic option (Bauer et al., 2007; Harris et al., 2016; Schott and Hayes, 2012). It is unknown whether cancer aerobic glycolysis plays a role in MDSC development, subsequently affecting tumor immunity and outcomes in TNBC patients. To address these questions, we evaluated and compared the metabolism profiles of TNBC and non-TNBC breast cancer in the Cancer Genome Atlas Network database (Koboldt et al., 2012). Gene set enrichment analysis (GSEA) demonstrated that TNBC exhibited an enriched glycolysis profile (Figure S1A) rather than an oxidative phosphor-ylation profile (Figure S1B) as compared with other types of breast cancer (Koboldt et al., 2012). G-CSF and GM-CSF control MDSC development in cancer (Gabrilovich et al., 2012; Morales et al., 2010; Shojaei et al., 2009). To explore the potential relationship between tumor glycolysis and MDSCs in human TNBC, we hypothesized that tumor glycolysis affected expression of G-CSF and GM-CSF to regulate MDSC development. To test this hypothesis, we initially analyzed Oncomine TNBC dataset (Curtis et al., 2012). We found that the expression levels of multiple key glycolytic enzymes, including lactate dehydrogenase A (LDHA), hexokinase-1, glucose-6-phosphate isomerase, phosphofructokinase muscle subunit gene, and pyruvate kinase muscle isozyme 2, correlated with G-CSF (Figures 1A–1E) and GM-CSF (Figures S1C–S1F) expression in TNBC. Thus, glycolytic profile correlates with G-CSF and GM-CSF expression in human TNBC.

Figure 1. Glycolysis Regulates Tumor G-CSF and GM-CSF Expression.

(A–E) Correlations between G-CSF and glycolysis enzymes. Expression of lactate dehydrogenase A (LDHA) (A), hexokinase-1 (HK1) (B), glucose-6-phosphate isomerase (GPI) (C), pyruvate kinase muscle isozyme 2 (PKM2) (D), and phosphofructokinase muscle subunit gene (PFKM) (E) was analyzed based on the dataset containing 250 TNBC patients. Gene expression was normalized to β-actin. Pearson’s correlation was calculated.

(F–I) Effect of 2-DG on tumor G-CSF expression. 4T1 (F and H) and Py8119 (G and I) cells were treated with 5 mM 2-DG for 12 hr. G-CSF transcript was quantified by real-time PCR (F and G) and G-CSF protein was measured in culture supernatant with ELISA (H and I) (n = 3/group, one of three experiments is shown, *p < 0.001).

(J and K) Effect of shLDHAs on tumor LDHA expression. Immunoblot analysis of LDHA knockdown (KD) by shLDHAs (shLDHA1 and shLDHA2) in 4T1 cells (J) and Py8119 cells (K). Scramble shRNA (Scr) was used as control. One of three experiments is shown.

(L–O) Effect of shLDHAs on tumor G-CSF expression. Scramble and shLDHA-expressing 4T1 and Py8119 cells were cultured for 24 and 48 hr, respectively. G-CSF transcript was quantified by real-time PCR (L and M) and G-CSF protein was measured in culture supernatant with ELISA (N and O) (n = 3/group, one of three experiments is shown, *p < 0.003). Data represent mean ± SEM.

(P) Immunofluorescence staining. Scramble and LDHA KD 4T1 cells were cultured for 48 hr and were stained with primary antibody anti-G-CSF and secondary antibody conjugated with Alexa Fluor 594. Scale bar, 20 μm. One of three experiments is shown.

To biologically validate this correlation, we treated mouse 4T1 and Py8119 cells (Gibby et al., 2012), two TNBC cells, with 2-deoxy-D-glucose (2-DG). 2-DG is a glycolysis inhibitor and competitively inhibits the production of glucose-6-phosphate. As expected, 2-DG reduced extracellular acidification rates (ECARs) (Figure S1G) and lactate levels (Figure S1H) in 4T1 cells. Interestingly, 2-DG suppressed G-CSF mRNA (Figures 1F and 1G) and protein (Figures 1H and 1I) but had no effect on interleukin-1β (IL-1 β) expression in 4T1 (Figure S1I) and transforming growth factor β (TGF-β) expression in Py8119 cells (Figure S1J). In addition, 2-DG suppressed GM-CSF expression in 4T1 and Py8119 (Figures S1K and S1L). 2-DG induced 4T1 cell apoptosis but had no effect on Py8119 cell apoptosis as shown by Annexin V and 7-aminoactinomycin D (7AAD) staining (Figure S1M). As 2-DG treatment resulted in less than 5% of tumor cell apoptosis (Figure S1M), apoptotic cell loss was an unlikely cause of the inhibitory effect of 2-DG on tumor cell G-CSF expression. LDHA is a rate-limiting enzyme in glycolysis processes. It reduces the replenishment of NAD⁺, indirectly reduces GAPDH function, and catalyzes pyruvate to lactate. Genetic knockdown of LDHA (LDHA KD) with two specific short hairpin RNAs (shRNAs) (shLDHA1 and shLDHA2) (Figures 1J and 1K) inhibited glycolysis as shown by reduced ECAR (Figure S1N) and lactate production (Figure S1O). Consistent with the effect of 2-DG on G-CSF and GM-CSF expression (Figures 1F–1I), our gene array revealed that G-CSF and GMCSF were among the top reduced five to ten genes in shLDHA1-expressing 4T1 cells (Figure S1P). As a confirmation, we found that LDHA KD caused reduced G-CSF mRNA expression (Figures 1L and 1M) and protein levels, as detected with ELISA (Figures 1N and 1O), as well as by in situ intracellular cytokine staining (Figure 1P) in 4T1 and Py8119 cells, but had no effect on TGF-b expression (Figures S1Q and S1R). We observed a similar effect of shLDHAs on GM-CSF expression in 4T1 (Figure S1S) and Py8119 cells (Figure S1T). Similarly, LDHA KD in human TNBC MDA-MB-231 cells (Figures S1U and S1V) caused reduced GMCSF protein expression (Figure S1W). In addition to 4T1, Py8119, and MDA-MB-231 cells, we included five additional non-TNBC tumor cells in the study. These cells expressed minimal levels of G-CSF and/or moderate levels of GM-CSF (Figures S1X and S1Y). LDHA KD (Figure S1Z) had no effect on their production of G-CSF (Figure S1X) and GM-CSF (Figure S1Y). Thus, tumor glycolysis promotes G-CSF and GM-CSF expression in TNBC cells.

Glycolysis Targets CEBPB Isoform LAP to Control G-CSF and GM-CSF Expression

We next dissected the molecular mechanism by which glycolysis regulates G-CSF and GM-CSF expression in breast cancer cells. CEBPB may regulate G-CSF expression in hematopoietic stem cells (Akagi et al., 2008) and myeloid cells (Marigo et al., 2010). We hypothesized that CEBPB controlled G-CSF expression in breast cancer cells. To test this hypothesis, we genetically knocked down CEBPB with two specific shRNAs (shCEBPB1 and shCEBPB2) (Figures S2A and S2B). CEBPB shRNAs reduced G-CSF transcripts (Figure 2A) and protein expression (Figure 2B) in breast cancer cells. Thus, CEBPB controls G-CSF expression in breast cancer cells.

Figure 2. Glycolysis Targets CEBPB Isoform LAP to Control G-CSF Expression.

(A and B) Effect of shCEBPB on tumor G-CSF expression. Scramble and shCEBPB (shCEBPB1 and shCEBPB2)-infected 4T1 cells were cultured for 24 hr. G-CSF transcript was quantified by real-time PCR (A) and G-CSF protein was measured in culture supernatant with ELISA (B) (n = 3/group, one of three experiments is shown, *p < 0.003).

(C) 4T1 and Py8119 cells were treated with 2-DG (10 mM) for different times. Tumor CEBPB isoforms were detected by western blotting. One of three experiments is shown.

(D) CEBPB isoforms were detected by western blotting in whole-cell lysis of scramble and shLDHA-expressing 4T1 and Py8119 cells. One of three experiments is shown.

(E) 4T1 cells were transfected with plasmid vectors expressing CEBPB isoforms LAP*, LAP* mutant, and LAP. Expression of LAP*, LAP* mutant, and LAP was detected by western blotting. One of three experiments is shown.

(F and G) 4T1 cells were transfected with the plasmid vectors expressing LAP*, LAP* mutant, and LAP. G-CSF transcript was quantified by real-time PCR (F) and G-CSF protein was measured in culture supernatant with ELISA (G) (n = 3/group, one of three experiments is shown, *p < 0.001).

(H and I) LDHA KD 4T1 cells were transfected with plasmid vectors expressing LAP. G-CSF transcript was quantified by real-time PCR (H) and G-CSF protein was measured in culture supernatant with ELISA (I) (n = 3/group, one of three experiments is shown, *p < 0.001).

(J) LAP was knocked out in 4T1 cells (LAP KO) and LAP KO cells were infected with LAP overexpressing (LAP OE) lentivirus. LAP protein was detected by western blot. One of three experiments is shown. Data represent mean ± SEM.

(K–M) Role of LAP in G-CSF expression. G-CSF transcript was quantified by real-time PCR (K). G-CSF protein was measured in culture supernatant with ELISA (L) (n = 3/group, *p < 0.001). (M) Intracellular G-CSF was stained with anti-G-CSF, revealed with Alexa Fluor 594-conjugated secondary antibody, and detected by fluorescence microscope. Scale bar, 20 μm. One of three experiments is shown.

We wondered whether glycolysis affected CEBPB expression and in turn regulated G-CSF expression. To test this, we treated breast cancer cells with 2-DG and assessed the expression of CEBPB. CEBPB mRNA produces four N-terminally truncated isoforms: a 38-kDa full-length CEBPB (LAP*), a 34-kDa LAP, a 21-kDa liver-enriched inhibitory protein (LIP), and a 14-kDa iso-form through alternative translation from different AUG (Schrem et al., 2004) (Figure S2C). We found that 2-DG treatment exclusively and efficiently reduced the expression levels of the second isoform of CEBPB, LAP, in both 4T1 and Py8119 cells (Figure 2C). This suggests that glycolysis inhibition may reduce LAP expression. In support of this notion, knockdown of LDHA exclusively resulted in decreased LAP expression in 4T1 and Py8119 cells (Figure 2D). We questioned whether LAP regulated G-CSF expression in breast cancer cells. To examine this, we constructed the vectors expressing LAP*, LAP* mutant, or LAP (Figures 2E and S2D). Forced expression of LAP* and LAP, but not LAP* mutant, promoted G-CSF transcription (Figure 2F) and protein expression (Figure 2G) in the breast cancer cells. CEBPB isoforms had no effect on TGF-b (Figure S2E) and MMP9 (Figure S2F) expression. LAP also promoted GM-CSF expression (Figure S2G). In addition, we conducted two rescue experiments. First, we expressed LAP in shLDHA-expressing tumor cells and found that forced LAP expression recovered G-CSF transcript (Figure 2H) and protein (Figure 2I) expression in shLDHA tumor cells. Second, we expressed LAP in shCEBPB tumor cells and observed that expression of LAP recovered G-CSF expression in shCEBPB tumor cells (Figure S2H). Furthermore, using CRISPR technology, we specifically mutated the second ATG of CEBPB DNA to knock out LAP (Figures S2D and 2J) and detected a dramatic decrease in G-CSF and GM-CSF mRNA and protein expression (Figures 2K–2M, S2I, and S2J). These effects were recovered by LAP overexpression in LAP knockout (KO) cells (Figures 2K–2M, S2I, and S2J). Thus, tumor LDHA targets specific CEBPB isoform LAP and controls G-CSF and GM-CSF expression in breast cancer cells.

Glycolysis Targets LAP via the AMPK-ULK1 Pathway

Next, we studied how glycolysis controlled LAP expression in tumor cells. Glycolysis is the predominant pathway to generate ATP for many tumor cells. Glycolysis inhibition can lead to increased AMP and ATP ratios in tumor cells (Pradelli et al., 2010). AMPK is a central sensor for energy stress in cells (Kim et al., 2011). In line with this, we observed that treatment with 2-DG (Figure S3A) and LDHA KD with shLDHA (Figure S3B) increased the AMP and ATP ratios in breast cancer cells. Furthermore, 2-DG treatment (Figure 3A) and LDHA KD (Figure 3B) activated the AMPK-ULK1 pathway in 4T1 and Py8119 cells. To investigate if the AMPK-ULK1 pathway affected the expression of LAP, we treated LDHA KD tumor cells with dorsomorphin, a specific AMPK inhibitor. Dorsomorphin inhibited AMPK phosphorylation (Figure 3C) and recovered LAP protein level (Figure 3D) and G-CSF expression (Figure 3E) in a dose-dependent manner. Similarly, ULK1 knockdown with small interfering RNA (siRNA) restored LAP protein level (Figure 3F), and G-CSF mRNA (Figures 3G and 3H) and protein (Figures 3I and 3J) in 4T1 and Py8119 cells, which expressed specific shLDHA. Thus, glycolysis may regulate tumor G-CSF expression by controlling LAP expression via the AMPK and ULK1 pathway.

Figure 3. Glycolysis Targets LAP via the AMPK-ULK1 Pathway.

(A) 4T1 and Py8119 cells were treated with 2-DG (20 mM) for 1 hr. AMPK and ULK1 were detected by western blotting. One of three experiments is shown.

(B) Immunoblot analysis of AMPK, p-AMPK, ULK1, and p-ULK1 in whole-cell lysis of scramble and shLDHA1-expressing 4T1 and Py8119 cells. One of three experiments is shown.

(C) LDHA KD 4T1 cells were treated with dorsomorphin for 1 hr. AMPK and AMPK phosphorylation were detected by western blotting. One of three experiments is shown.

(D) Scramble and shLDHA-expressing 4T1 cells were treated with dorsomorphin for 24 hr. LAP* and LAP were detected by western blotting. One of three experiments is shown.

(E) Scramble and shLDHA-expressing 4T1 cells were treated with dorsomorphin for 24 hr. G-CSF transcripts were quantified by real-time PCR (n = 3/group, one of three experiments is shown, *p < 0.001).

(F) Scramble and shLDHA-expressing 4T1 and Py8119 cells were transfected with siULK and control for 12 hr. Cells were cultured for an additional 24 hr. LAP* and LAP were detected by western blotting. One of three experiments is shown.

(G–J) Scramble and shLDHA-expressing 4T1 and Py8119 cells were transfected with siULK and control for 12–24 hr. Cells were cultured for an additional 48 hr. G-CSF transcripts and proteins were detected by real-time PCR (G and H) and ELISA (I and J), respectively (n = 3/group, one of three experiments is shown, * p < 0.001). Data represent mean ± SEM.

Glycolysis Controls LAP Expression via Autophagy Activation

We further explored the mechanism by which the AMPK pathway controlled the LAP expression. Autophagy can degrade cytoplasmic proteins and organelles during stress conditions (Klionsky and Emr, 2000). ULK1 is an essential component of autophagy initial complex (Ganley et al., 2009), and AMPK activation induces autophagy via phosphorylation of ULK1 (Alers et al., 2012; Kim et al., 2011). Given that glycolysis restriction activated the AMPK-ULK1 pathway (Figures 3A and 3B), we hypothesized that autophagy affected LAP expression. We observed that 2-DG treatment (Figure 4A) and LDHA KD (Figure 4B) induced the autophagy formation in 4T1 and Py8119 cells, as demonstrated by increased LC3b-II expression. Immunofluorescence staining revealed more autophagy puncta in LDHA-deficient tumor cells compared with control cells with or without chloroquine (CQ), an autophagy inhibitor (Figures 4C and 4D). In addition, siULK1 attenuated the auto-phagy formation in the LDHA KD 4T1 and Py8119 cells (Figure S4A). The data suggest that glycolysis orchestrates a molecular network among LDHA, the AMPK-ULK1 signaling, and the autophagy pathways in breast cancer cells, which might control LAP expression. To test this possibility, we treated 4T1 cells with CQ to inhibit the autophagy formation. When auto-phagy was activated in shLDHA tumor cells (Figures 4B–4D), CQ treatment prevented LAP reduction (Figure 4E) and restored G-CSF expression (Figure 4F) in shLDHA tumor cells. We genetically blocked an autophagy component FIP200 with siRNA (Hara et al., 2008; Wei et al., 2009, 2011) and investigated the level of LAP and the expression of G-CSF. Knockdown of FIP200 restored LAP level (Figure 4G), and G-CSF mRNA (Figures 4H and 4I) and protein (Figures 4J and 4K) expression in 4T1 and Py8119 cells, which expressed shLDHA. We observed similar effects on ATG5 and LC3b knockdown tumor cells (Figures S4B–S4G). In addition to autophagy inhibitor CQ, we treated 4T1 tumor cells with rapamycin, an inducer of auto-phagy via inhibiting the Ser/Thr protein kinase mammalian target of rapamycin (mTOR). We found that treatment with rapamycin reduced LAP expression (Figure S4H). As autophagy inhibition did not cause a full recovery of LAP expression (Figures 4E and 4G), we treated shLDHA 4T1 cells with MG132, a protea-some inhibitor. We observed that MG132 and CQ both partially enhanced LAP protein levels (Figure S4I). The data suggest that proteasomes may also be involved in the regulation of LAP expression (Fu et al., 2015). Thus, tumor glycolysis prevents the AMPK-ULK1 signaling activation and its associated auto-phagy formation and reduces autophagy-mediated partial LAP reduction and, in turn, LAP enhances G-CSF expression and supports MDSC development in the tumor.

Figure 4. Glycolysis Controls LAP Expression via Autophagy Activation.

(A) 4T1 and Py8119 cells were treated with 2-DG (5 mM) for 12 hr. LC3b-I and LC3b-II were detected by western blotting. One of three experiments is shown.

(B) Scramble and shLDHA-expressing 4T1 and Py8119 cells were cultured for 24 hr. LC3b-I and LC3b-II were detected by western blotting. One of three experiments is shown.

(C and D) Scramble and shLDHA-expressing 4T1 cells were treated with and without chloroquine (CQ). Autophagy puncta were revealed with anti-LC3b monoclonal antibody staining and were analyzed with fluorescence microscope (C). Results were shown as the percentage of puncta-positive cells ± SEM (D) (n = 6/group, one of three experiments is shown, *p < 0.0001). Scale bar, 25 μm.

(E) Scramble and shLDHA-expressing 4T1 cells were treated with CQ for 24 hr. LAP* and LAP were detected by western blotting. One of three experiments is shown.

(F) Scramble and shLDHA-expressing 4T1 cells were treated with CQ for 24 hr. G-CSF transcript was detected by real-time PCR (n = 3/group, one of three experiments, *p < 0.0001).

(G) Scramble and shLDHA-expressing 4T1 and Py8119 cells were transfected with FIP200 siRNA for 36 hr. CEBPB isoforms were detected by western blotting. One of three experiments is shown.

(H–K) Scramble and shLDHA-expressing 4T1 and Py8119 cells were transfected with FIP200 siRNA for 12–24 hr. Cells were cultured for 48 hr. G-CSF transcripts and proteins were detected by real-time PCR (H and I) and ELISA (J and K), respectively (n = 3/group, one of three experiments is shown, *p < 0.0001). Data represent mean ± SEM.

Tumor LDHA Affects MDSCs to Control Tumor Immunity

To determine whether tumor LDHA affects MDSCs to control tumor immunity in vivo, we inoculated shLDHA-expressing 4T1 and Py8119 cells into wild-type BALB/C and C57/BL6 mice, respectively. LDHA KD 4T1 tumor-bearing mice exhibited slower tumor growth (Figure 5A), less tumor metastasis (Figures 5B and 5C), and enhanced survival (Figure 5D) compared with control mice. Similarly, LDHA KD resulted in reduced PY8119 tumor growth in C57/BL6 wild-type mice (Figure S5A). The potential differential effects of tumor LDHA on cell proliferation may depend on the tumor cell types (Brand et al., 2016; Le et al., 2010; Xian et al., 2015). We observed similar cell cycles in vitro with or without LDHA KD in 4T1 and Py8119 cells (Figures S5B–S5D), comparable tumor growth (Figures S5E and S5F), lung metastasis (Figure S5G), and mouse survival (Figure S5H) in NOD SCID gamma-deficient (NSG) mice bearing 4T1 tumors expressing shLDHA and control. Interestingly, LDHA KD resulted in a reduced amount of MDSCs in tumor tissues and spleen compared with controls in 4T1 tumor-bearing wild-type mice (Figures 5E–5G) and Py8119 tumor-bearing wild-type mice (Figures S5I and S5J). Although shLDHA did not affect 4T1 tumor growth, metastasis, and mouse survival in NSG model (Figures S5E–S5H), we detected reduced tumor MDSCs in shLDHA 4T1 tumor-bearing NSG mice compared with control NSG mice (Figure S5K). We next examined T cell profiles in mice bearing shLDHA tumor and control. We detected increased interferon-γ⁺ (IFN-γ⁺) and tumor necrosis factor alpha⁺ (TNF-α⁺) effector CD8⁺ T cells in tumor tissues (Figure 5H and 5I) and tumor-draining lymph nodes (Figures 5J and 5K) in mice bearing shLDHA 4T1 tumor compared with control. Similar immune profiles were obtained in mice bearing shLDHA Py8119 tumor (Figure S5L). The results suggest that tumor LDHA may regulate T cell immunity in vivo. In further support of this, we demonstrated that CD4⁺ and CD8⁺ T cell depletion (Figure 5L) abolished the immune protective effect of tumor LDHA KD on tumor growth in vivo (Figure 5M). Thus, the data suggest that tumor LDHA may control MDSCs and, in turn, regulate T cell tumor immunity and tumor progression.

Figure 5. Tumor LDHA Affects MDSCs to Control Tumor Immunity.

(A–D) Effect of shLDHA on tumor growth (A), metastasis (B and C), and mouse survival (D). Scramble and LDHA KD 4T1 cells (2.5 × 10⁴) were inoculated into female BALB/c mice. (A) Tumor growth was monitored and tumor size was measured (n = 10/group, *p < 0.001). Bioluminescence detection showed the distant 4T1 metastasis (B) and metastatic rate (C) on 30 days (n = 7/group, one of two experiments is shown).

(E–G) Gr1⁺CD11b⁺ cells in tumor-bearing mice. (E) Representative flow cytometry dot plots showed spleen and tumor tissue Gr1⁺CD11b⁺CD45⁺ cells in mice bearing scramble and LDHA KD 4T1 tumors. (F and G) Percentages of Gr1⁺CD11b⁺ cells were shown in spleen (F) and tumor tissues (G) (n = 6/group, one of two experiments is shown, *p < 0.05).

(H and I) Representative flow cytometry dot plots (H) and the percentages of TNF-α and IFN-γ (I) were shown in CD8⁺ T cells in 4T1 tumor tissue (n = 5–6/group, one of two experiments is shown, *p < 0.05).

(J and K) Representative flow cytometry dot plots (J) and the percentages of TNF-α and IFN-γ (K) were shown in CD8⁺ T cells in 4T1 tumor draining lymph nodes (TDLNs) (n = 5–6/group, one of two experiments is shown, *p < 0.05).

(L and M) Effect of CD4 and CD8 T cell depletion on 4T1 tumor growth. (L) Representative flow cytometry dot plots showed T cell depletion efficiency. (M) Tumor volume was shown in different groups (n = 5/group, one of two experiments is shown, *p < 0.05). Data represent mean ± SEM.

Tumor LDHA Regulates Tumor Immunity via G-CSF

We next examined whether tumor LDHA affected G-CSF expression in vivo to control MDSCs and tumor immunity. In line with our in vitro data (Figures 1F–1O, S1X, and 1Y), we detected lower levels of G-CSF mRNA (Figure S6A) and protein (Figure S6B) in tumor tissues (Figures S6A and S6B) and peripheral blood (Figure S6C) in mice bearing shLDHA 4T1 tumor compared with controls. To study the role of G-CSF in vivo, we ectopically expressed G-CSF in shLDHA tumor cells and parental tumor cells and then inoculated these tumor cells into wild-type mice. We confirmed that shLDHA tumors grew slower than scramble tumors, as shown by tumor volume (Figure 6A). As expected, forced G-CSF expression abolished this protective effect of shLDHA on tumor growth (Figure 6A). In line with these observations, tumor LDHA KD caused reduced MDSCs in the tumor tissues and spleen, and forced G-CSF expression recovered the amount of MDSCs (Figure 6B). In addition, tumor LDHA KD resulted in increased IFN-γ⁺ and TNF-α⁺ effector CD8⁺ T cells in tumor (Figure 6C) and tumor-draining lymph nodes (Figure 6D); this effect was diminished with forced G-CSF expression (Figures 6C and 6D).

Figure 6. Tumor LDHA Regulates Tumor Immunity via G-CSF.

(A) Effect of forced G-CSF overexpression (G-CSF OE) on tumor growth. 4T1 cells (0.5 × 10⁵/mouse) were inoculated into female BALB/c mice. Tumor size was measured every 2 days (n = 5/group, one of two experiments is shown, *p < 0.05).

(B) Effect of G-CSF OE on MDSCs. Ly6G⁺CD11b⁺ cells were analyzed by flow cytometry in CD45⁺ cells in tumor-bearing mice. The percentages of Ly6G⁺CD11b⁺ cells were shown in spleen and tumor (n = 5/group, one of two experiments is shown, *p < 0.01).

(C and D) Effect of G-CSF OE on CD8⁺ T cell profile. The percentages of IFN-γ⁺ and TNF-α⁺ in CD8⁺ T cells in tumor tissue (C) and TDLN (D) were shown (n = 5/group, one of two experiments is shown, *p < 0.05).

(E) Effect of G-CSF knockout (G-CSF KO) on tumor growth. 4T1 cells (0.5 3 10⁵/mouse) were inoculated into female BALB/c mice. Tumor size was measured every 2 days (n = 6–7/group, one of two experiments is shown, *p < 0.05).

(F) Effect of G-CSF KO on MDSCs. Ly6G⁺CD11b⁺ cells were analyzed by flow cytometry in CD45⁺ cells in tumor-bearing mice. The percentages of Ly6G⁺CD11b⁺ cells were shown in spleen and tumor (n = 6–7/group, one of two experiments is shown, *p < 0.01).

(G and H) Effect of G-CSF KO on CD8⁺ T cell profile. The percentages of IFN-γ⁺ and TNF-α⁺ in CD8⁺ T cells in tumor tissue (G) and TDLN (H) were shown (n = 6–7/group, one of two experiments is shown, *p < 0.05).

(I) Effect of LAP knockout (LAP KO) on tumor growth. 4T1 cells (0.5 3 10⁵/mouse) were inoculated into female BALB/c mice. Tumor size was measured every 2 days (n = 8–9/group, one of two experiments is shown, *p < 0.05).

(J) Effect of forced LAP KO on MDSCs. Ly6G⁺CD11b⁺ cells were analyzed by flow cytometry in CD45⁺ cells in tumor-bearing mice. The percentages of Ly6G⁺CD11b⁺ cells were shown in spleen and tumor tissues (n = 8–9/group, one of two experiments is shown, *p < 0.01).

(K and L) Effect of LAP KO on CD8⁺ T cell profile. The percentages of IFN-γ⁺ and TNF-α⁺ in CD8⁺ T cells in tumor tissues (K) and TDLN (L) were shown (n = 8–9/group, one of two experiments is shown, *p < 0.05). Data represent mean ± SEM.

To link this observation to the role of MDSCs in vivo, we administered anti-Ly6G antibody to deplete MDSCs in mice bearing different 4T1 tumor cells. We showed that in vivo MDSC depletion reduced tumor growth in mice bearing 4T1 tumor compared with isotype control (Figure S6D). Furthermore, when MDSCs were depleted in mice, tumor volumes were similar between LDHA-proficient and -deficient tumor-bearing mice (Figure S6D). The data suggest that tumor LDHA regulates MDSCs, and in turn, MDSCs mediate immunosuppression. In confirmation, we showed that 4T1 tumor-associated (Figures S6E and S6F) and Py8119 tumor-associated (Figures S6G and S6H) MDSCs inhibited T cell activation (Figures S6E and S6G), as shown by reduced T cell CD25 and CD69 expression and suppressed effector T cell IFN-γ and TNF-α expression (Figures S6F and S6H).

In addition to ectopic G-CSF expression, we used CRISPR gene editing technology, created G-CSF KO 4T1 cells, and further tested the role of tumor G-CSF in tumor progression in vivo. As expected, we detected negligible levels of G-CSF in the in vitro cultured G-CSF KO 4T1 tumor cells (Figure S6I) and in sera (Figure S6J) from mice bearing G-CSF KO 4T1 tumor. Furthermore, we detected smaller tumor volume (Figure 6E), decreased MDSCs in spleen and tumor tissues (Figure 6F), and increased tumor-infiltrating IFN-γ⁺ and TNF-α⁺ effector CD8⁺ T cells in tumor tissues (Figure 6G) and tumor-draining lymph nodes (Figure 6H) in mice bearing G-CSF KO tumor compared with wild-type tumor. As expected, G-CSF KO abolished the protective effect of shLDHA on tumor growth (Figure 6E). In line with this observation, G-CSF KO resulted in comparable IFN-γ⁺ and TNF-α⁺ effector CD8⁺ T cells between the shLDHA and Scr groups in tumor (Figure 6G) and tumor-draining lymph nodes (Figure 6H).

To confirm the effect of LAP on the regulation of G-CSF and tumor immune response, we used CRISPR/CAS9-mediated point mutation to mutate the second ATG in the CEBPB gene, specifically knocked out LAP in 4T1 cells, and examined the role of LAP in tumor progression in vivo. LAP KO dramatically decreased tumor volume (Figure 6I), MDSCs in spleen and tumor tissues (Figure 6J), and increased tumor-infiltrating IFN-γ⁺ and TNF-α⁺ effector CD8⁺ T cells in tumor tissues (Figure 6K) and tumor-draining lymph nodes (Figure 6L) in mice bearing LAP KO tumor compared with wild-type tumor. This phenotype was reversed by ectopic expression of LAP (Figures 6I–6L). Thus, tumor LDHA regulates MDSCs via LAP, controlling G-CSF expression, and affects tumor immunity and tumor progression.

Glycolysis, MDSC, and T Cell Response Correlate in TNBC Patients

Finally, we evaluated the potential correlations among the glycolytic pathway, MDSC levels, immune signatures, and their associations with patient survival. We performed GSEA in a dataset from the European Genome-phenome Archive (EGAS00000000083) (Curtis et al., 2012). Based on the median values of LDHA expression in 250 TNBC patients, we found enriched glycolytic gene signature (Figure 7A) and MDSC gene signature (Figure 7B) in patients with high LDHA expression. In contrast, the T cell metagenes (Figure 7C) and T cell receptor (TCR) complex signature (Figure 7D) were enriched in patients with low LDHA expression. We observed similar correlations in the enrichment network analysis (Figure 7E). We confirmed these observations in an additional dataset from the GEO (GEO: GSE58812) (Jezequel et al., 2015) with 107 TNBC patients (Figures S7A–S7D). The data suggest that glycolysis and MDSC may be important negative modulators in anti-tumor immunity in patients with TNBC.

Figure 7. Glycolysis, MDSCs, and T Cell Response Correlate in TNBC Patients.

(A–D) Correlation between LDHA and protective immune signature in human TNBC. Glycolytic (A) and immune cell (B–D) gene signatures were analyzed and compared between high and low LDHA-expressing TNBCs (EGAS00000000083, n = 250). The normalized enrichment score (NES) (green line) reflects the degree to which the gene set is over-represented at the top or bottom of the ranked list of genes. A positive value indicates more correlation with “high LDHA-expressing tumor,” and a negative value indicates more correlation with “low LDHA-expressing tumor.”

(E) GSEA comparison between patients with high LDHA expression (red) and low LDHA expression (blue). Median split, n = 250 (EGAS00000000083). Cytoscape and Enrichment map were used for visualization of the GSEA results (p value cutoff: 0.05). Enrichment results were mapped as a network of gene sets. Nodes represent enriched gene sets, which were grouped by their similarities according to the related gene sets. Node size was proportional to the total number of genes within each gene set. Proportion of shared genes between gene sets was presented as the thickness of the green lines between nodes.

(F–I) Estimates of signature correlation among MDSC, glycolysis, and T cell immune response. Signature scores were calculated as reported previously (Welte et al., 2016). The correlations were evaluated by Person’ correlation test controlled by cohort (EGAS00000000083 and GEO: GSE58812, n = 357).

(J–M) Kaplan-Meier estimates of overall survival in TNBC patients. Patients were divided into low and high groups based on the median of LDHA expression (J), glycolysis signature score (K), MDSC signature score (L), and T cell metagene signature score (M); p values were calculated using the log-rank test controlled by cohort (EGAS00000000083 and GEO: GSE58812, n = 357).

We next evaluated the relationship between the glycolytic pathway and MDSCs in the TNBC tumor microenvironment. A recent report has identified a specific MDSC-associated gene signature in patients with cancer (Welte et al., 2016). Based on this (Welte et al., 2016) and the availability of clinical information, we combined and analyzed the two breast cancer data-sets, EGAS00000000083 (Curtis et al., 2012) and GEO: GSE58812 (Jezequel et al., 2015). We found that glycolysis signature score correlated with that of MDSCs in 357 patients with TNBC (Figures 7F and 7G; Table S1). In contrast, MDSC signature score negatively correlated with that of T cell meta-genes (Figures 7F and 7H), TCR complex (Figures 7F and S7E), and adaptive immune response signatures (Figures 7F and S7F). In addition, glycolysis signature negatively correlated with T cell metagenes (Figure 7I). We extended our studies to the third breast cancer dataset (Koboldt et al., 2012) and obtained similar correlations among these factors (Figures S7G–S7J). Furthermore, we found that in the combined datasets (EGAS00000000083 and GEO: GSE58812), LDHA expression, the glycolysis-associated signature score, and the MDSC-associated signature score were negatively associated with overall survival (Figures 7J–7L) and metastasis-free survival (Figures S7K and S7L) in patients with TNBC, whereas the T cell metagene score favorably predicted the TNBC patient outcome (Figure 7M; Table S1). Furthermore, the hazard function of MDSC signature, LDHA, glycolysis signature, and T cell response signature are risk factors for TNBC survival (Table S1). Thus, glycolysis, MDSC, and T cell response functionally correlate in TNBC patients and are clinically associated with the patient outcome.

DISCUSSION

Tumor cells reprogram metabolic pathways to meet their bioenergetics and biosynthetic demands. Preferential aerobic glycolysis is observed across many types of cancer cells and is considered a hallmark of cancer (Hanahan and Weinberg, 2011). However, whether and how aerobic glycolysis affects MDSCs and, in turn, regulates tumor immunity, are not well defined in patients with cancer. In this work, we focus on patients with TNBC and demonstrate that aerobic glycolysis controls tumor G-CSF and GM-CSF expression, regulates MDSC development via distinct molecular mechanisms, and consequently affects tumor progression and outcome. We, along with our peers, have shown that tumor glycolysis can modulate effector memory T cell phenotype and function in the tumor microenvironment in cancer patients (Chang et al., 2013, 2015; Ho et al., 2015; Zhao et al., 2016) and in tumor-bearing mouse models (Brand et al., 2016; Chang et al., 2013, 2015; Ho et al., 2015; Zhao et al., 2016). Thus, our current work has established a potential causal link between tumor metabolic reprogramming and MDSC-mediated immunosuppression.

In line with previous reports (Gabrilovich et al., 2012; Morales et al., 2010; Shojaei et al., 2009), we have found that tumor-derived G-CSF and GM-CSF are important cytokines promoting MDSC development in TNBC both in vitro and in vivo. Recent studies have elucidated the role of tumor glycolysis in T cells (Brand et al., 2016; Chang et al., 2015; Zhao et al., 2016). Thus, we have focused our studies on whether tumor glycolysis can regulate MDSCs via targeting G-CSF and GMCSF expression. We have shown that biochemical and genetic inhibition of tumor glycolysis results in an increase of the AMP and ATP ratio (Pradelli et al., 2010), which subsequently activates an energy sensor, AMPK (Mihaylova and Shaw, 2011); stimulates ULK1 phosphorylation; and activates and induces autophagy formation (Kim et al., 2011). Interestingly, we have made a molecular link between G-CSF and GM-CSF expression, autophagy function, and CEBPB expression. There are three isoforms of CEBPB including LIP, LAP, and LAP* (Qiu et al., 2008), and different CEBPB isoforms can be induced during granulocyte differentiation (Marigo et al., 2010; Rosenbauer and Tenen, 2007). It is thought that the transcription factor CEBPB in myeloid cells regulates MDSCs in tumor (Marigo et al., 2010). However, it is unknown whether CEBPB in tumor cells and/or which CEBPB isoform(s) in tumor cells affects the production of key cytokines critical for tumor MDSC development, and how CEBPB expression is regulated in the context of tumor glycolysis. Unexpectedly, we have observed that the AMPK-ULK1-activated autophagy signaling pathway is involved in the regulation of the secondary isoform of CEBPB, LAP. Furthermore, we have defined that LAP is a relatively specific isoform of CEBPB controlling tumor G-CSF and GM-CSF expression through glycolysis. Our work shows that (1) CEBPB in tumor cells regulates the expression of G-CSF and GM-CSF; (2) this regulation is controlled by the second isoform of CEBPB, LAP; and (3) the levels of LAP protein are modulated by the autophagy-associated pathway. The latter may be under the control of tumor glycolysis via the AMPK-ULK1 pathway (Kim et al., 2011). Notably, CEBPB protein stability can be regulated by the 26S proteasome in myoblasts (Fu et al., 2015). Furthermore, CEBPB isoforms may be kinetically controlled by endoplasmic reticulum stress in fibroblasts (Li et al., 2008) and by mTOR signaling in osteoclasts (Smink et al., 2009). We have studied the impact of tumor glycolysis on CEBPB isoform expression in TNBC epithelial cells. We suggest that an auto-phagy-mediated regulation may be a previously unappreciated mechanism controlling the second isoform of CEBPB isoform LAP in the context of tumor glycolysis. The potential molecular details (including specific CEBPB isoform synthesis, translation, and stability) remain to be examined in the context of tumor glycolysis. Nonetheless, we provide important insights into CEBPB biology in the regulation of tumor MDSC development and tumor immunity. We suggest that tumor glycolytic metabolism may potentially orchestrate a molecular network of AMPK-ULK1, autophagy, and CEBPB (LAP) to efficiently promote tumor G-CSF and GM-CSF expression and to establish and maintain MDSC development and to evade tumor immunity.

After deciphering the molecular mechanism by which tumor glycolysis regulates MDSC development, we investigated the immunological and clinical relevance of this regulation in TNBC-bearing animal models and patients. We have shown that genetic silence of tumor LDHA and manipulation of tumor G-CSF results in alterations in MDSCs and effector T cell profile in vivo. Analogously, tumor growth, metastasis, and mouse survival are accordingly changed. In line with mouse data, the scores of glycolytic gene signature including LDHA, MDSC gene signature, and T cell gene signature correlate in TNBC patients. Furthermore, these three signature scores are associated with TNBC patient survival. Human TNBC is pathologically and clinically aggressive, with high rates of metastasis and recurrence and poor patient survival. TNBC patients are poor responders to hormonal or trastuzumab-based standard therapies (Bauer et al., 2007; Bianchini et al., 2016; Harris et al., 2016; Schott and Hayes, 2012) and checkpoint blockade treatment (Zou et al., 2016). Given that autophagy-associated innate immune signaling (Yu et al., 2017) and effector T cells (Wang et al., 2016) regulate cancer chemotherapy resistance, it will be interesting to explore whether this LAP and autophagy network additionally contributes to TNBC therapeutic response.

Our work demonstrates that tumor glycolytic metabolism biologically, immunologically, and clinically controls MDSCs and affects anti-tumor immunity in TNBC patients. Thus, reprogramming tumor glycolysis may be a biologic and immuno-logic strategy to treat patients with TNBC.

Limitations of Study

Crosstalk between AMPK and mTOR signaling pathways affects protein translation. We have found that the autophagy pathway regulates LAP expression in the context of tumor glycolysis. Thus, the next step is to examine whether and how autophagy, AMPK, and mTOR signaling pathways may be coordinated in the regulation of LAP expression. Moreover, it is important to understand the genetic and molecular basis of potent glycolytic metabolic pattern in TNBC. Given that myeloid cells including macrophages, myeloid dendritic cells, and myeloid suppressor cells express PD-L1 and mediate immunosuppression in the human tumor microenvironment and draining lymph nodes (Curiel et al., 2003; Lin et al., 2018), it is clinically relevant to explore therapeutic approaches by targeting MDSCs and tumor glycolysis in combination with PD-L1/PD-1 blockade (Zou et al., 2016) to treat patients with TNBC.

STAR⋆METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-human/mouse LDHA	Cell Signaling Technology	Cat# 2012S; RRID: AB_2137173
Anti-human/mouse AMPKα	Cell Signaling Technology	Cat# 2532S; RRID: AB_330331
Anti-human/mouse p-AMPKα	Cell Signaling Technology	Cat# 2535S; RRID: AB_331250
Anti-human/mouse ULK1	Cell Signaling Technology	Cat# 8054S; RRID: AB_11178618
Anti-human/mouse p-ULK1 (Ser317)(D2B6Y)	Cell Signaling Technology	Cat# 12753S; RRID: AB_2687883
Anti-human/mouse LC3B (D11)	Cell Signaling Technology	Cat# 3868; RRID: AB_2137707
Anti-human/mouse β-actin (8H10D10)	Cell Signaling Technology	Cat# 3700; RRID: AB_2242334
Anti-human/mouse CEBPB (C-19)	Santa Cruz Biotechnology	Cat# sc-150; RRID: AB_2260363
Anti-mouse CD3e (clone 500A2)	BD Biosciences	Cat# 557984; RRID: AB_396972
Anti-mouse CD4 (clone RM4–5)	BD Biosciences	Cat# 558107; RRID: ABJ397030
Anti-mouse IFN-γ (clone XMG1.2)	BD Biosciences	Cat# 560660; RRID: AB_1727533
Anti-mouse CD8a (clone 3–6.7)	BD Biosciences	Cat# 557654; RRID: AB_396769
Anti-mouse TNF-α (clone MP6-XT22)	BD Biosciences	Cat# 557644; RRID: AB_396761
Anti-mouse Ly-6G/Ly-6C (clone RB6–8C5)	BD Biosciences	Cat# 565033
Anti-mouse Ly-6G (clone 1A8)	BD Biosciences	Cat# 560601; RRID: AB_1727562
Anti-mouse CD45 (clone 30-F11)	Thermo Fisher Scientific	Cat# MCD4530; RRID: AB_2539700
Anti-mouse Ly-6C (clone HK1.4)	Thermo Fisher Scientific	Cat# 47-5932-82; RRID: AB_2573992
Anti-mouse CD11 b (clone M1/70)	Thermo Fisher Scientific	Cat# 17-0112-83; RRID: AB_469344
Anti-Rabbit IgG (H+L), Alexa Fluor 594	Thermo Fisher Scientific	Cat# R37117; RRID: AB_2556545
Anti-mouse G-CSF (clone EPR3203 (N) (B))	Abeam	Cat# ab181053
Anti-mouse Ly6G (clone 1A8)	Bioxcell	Cat# BE0075–1; RRID: ABJ 107721
Anti-mouse CD4 (clone GK1.5)	Bioxcell	Cat# BE0003–1; RRID: ABJ 107636
Anti-mouse CD8α (clone YTS169.4)	Bioxcell	Cat# BE0117; RRID: ABJ 0950145
Anti-Rat lgG2b isotype control (clone LTF-2)	Bioxcell	Cat# BE0090; RRID: ABJ 107780
Bacterial and Virus Strains
Stbl3 Chemically Competent E. coli	Thermo Fisher Scientific	Cat# C737303
DH5α Competent Cells	Thermo Fisher Scientific	Cat# 18258012
Chemicals, Peptides, and Recombinant Proteins
Chloroquine	Sigma-Aldrich	Cat# C6628
2-Deoxy-D-glucose	Sigma-Aldrich	Cat# D6134
Dorsomorphin	Sigma-Aldrich	Cat# P5499
AMP	Sigma-Aldrich	Cat# A2252
ATP 100 mM Solution	Sigma-Aldrich	Cat# GE27-2056-01
Ovalbumin (323–339)	Sigma-Aldrich	Cat# 01641
Ovalbumin (257–264)	Sigma-Aldrich	Cat# S7951
Critical Commercial Assays
Mouse G-CSF DuoSet ELISA	R&D systems	Cat# DY414
Mouse GM-CSF DuoSet ELISA	R&D systems	Cat# DY415
Human GM-CSF DuoSet ELISA	R&D systems	Cat# DY215
Lactate Assay Kit	Sigma-Aldrich	Cat# MAK064–1KT
Direct-zol RNA MiniPrep Kit	Zymo Research	Cat# R2051
Myeloid-Derived Suppressor Cell Isolation Kit	Miltenyi Biotec	Cat# 130-094-538
Deposited Data
Raw data for gene expression arrays	This study	GEO: GSE103925
Experimental Models: Cell Lines
4T1 cells	Fred R Miller (Karmanos Cancer Institute)	N/A
Py8119 cells	ATCC	Cat# CRL-3278
MDA-MB-231 cells	ATCC	Cat# HTB-26
NIH/3T3 cells	ATCC	Cat# CRL-1658
LLC1 cells	ATCC	Cat# CRL-1642
B16-F10 cells	ATCC	Cat# CRL-6475
HEK293T cells	ATCC	Cat# CRL-3216
MC38 cells	Walter Storkus	N/A
ID8 cells	Roby et al., 2000	N/A
Experimental Models: Organisms/Strains
Mouse: C57BIV6J	The Jackson Laboratory	Stock No: 000664
Mouse: BALB/cJ	The Jackson Laboratory	Stock No: 000651
Mouse: NSG	The Jackson Laboratory	Stock No: 005557
Mouse: C57BU6-Tg(TcraTcrb)110OMjb/J	The Jackson Laboratory	Stock No: 003831
Mouse: C.Cg-Tg(DO11.10)10Dlo/J	The Jackson Laboratory	Stock No: 003303
Oligonucleotides
Primers for colony	This paper	See Table S2
siRNA for mouse ULK1	Qiagen	Cat# GS22241
siRNA for mouse FIP200	Qiagen	Cat# GS12421
siRNA for mouse ATG5	Qiagen	Cat# GS11793
siRNA for mouse LC3b	Qiagen	Cat# GS67443
Mouse G-CSF primer	Origene	Cat# MP202847
Mouse GM-CSF primer	Origene	Cat# MP202843
Mouse LDHA primer	Origene	Cat# MP207257
Mouse ACTb primer	Origene	Cat# MP200232
Mouse IL-1β primer	Origene	Cat# MP206724
Mouse TGF-β primer	Origene	Cat# MP217184
Mouse CEBPB primer	Origene	Cat# MP201856
Mouse MMP9 primer	Origene	Cat# MP207904
Mouse LC3B primer	Origene	Cat# MP208170
Human LDHA primer	Origene	Cat# HP208683
Human GM-CSF primer	Origene	Cat# HP200702
Human ACTB primer	Origene	Cat# HP204660
Recombinant DNA
Mouse LDHA shRNA#1	Dharmacon	Cat# V2LMM_80290
Mouse LDHA shRNA#2	Dharmacon	Cat# V2LMM_165731
Mouse CEBPB shRNA#1	Dharmacon	Cat# V3LMM_504726
Mouse CEBPB shRNA#2	Dharmacon	Cat# V3LMM_504727
Human LDHA shRNA#1	Dharmacon	Cat# V3LHS_388269
Human LDHA shRNA#2	Dharmacon	Cat# V3LHS_388270
Scramble shRNA	This paper	N/A
psPAX2	This paper	N/A
pMD2.G	This paper	N/A
pLLEV-Luc	Biomedical Research Core Facilities of University of Michigan	N/A
Csf3 Mouse Tagged ORF Clone	Origene	Cat# MR225697
Cebpb Mouse Tagged ORF Clone	Origene	Cat# MR227563
pcDNA 3.1 (−) mouse C/EBP beta (LAP)	Addgene	Cat# 12557
pSpCas9(BB)-2A-Puro (PX459) V2.0	Addgene	Cat# 62988
G-CSF Double Nickase Plasmid	Santa Cruz	Cat# sc-419844-NIC
Software and Algorithms
GSEA	Subramanian et al., 2005	http://software.broadinstitute.org/gsea/index.jsp/
Cytoscape 3.4	Shannon et al., 2003	http://www.cytoscape.org/
ImageJ	NIH	N/A
SAS 9.4	SAS Institute	N/A

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Weiping Zou (wzou@med.umich.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animals

Six to eight week old female BALB/c, C57/BL6 and NOD SCID gamma deficient (NSG) mice were used for this study (Jackson Laboratory). The work was approved by the committee on Use and Care of Animals at the University of Michigan. LDHA knockdown and scramble 4T1 cells with luciferase expression were implanted into the secondary right mammary fat pad of BALB/c mice. Tumor size was measured every three days using calipers fitted with Vernier scale. Tumor volume was calculated as described previously (Tanikawa et al., 2012). Py8119 cells with LDHA knockdown were implanted into the mammary fat pad of C57/BL6 mice. The distant metastatic foci were quantified by Bioluminescence imaging system. In some cases, single cell suspensions were made from spleen and tumor tissues for immune phenotype and functional analysis

Cell Culture Studies

All cells were cultured at 37°C in a humidified atmosphere containing 5% CO2. Murine 4T1 breast cancer cells (sex: female) were cultured in PRMI 1640 medium supplemented with 10% fetal bovine serum (FBS). Py8119 cells (sex: female) were cultured in F-12K medium (ATCC, 30–2004) supplemented with 5% FBS. MDA-MB-231 breast cancer cells (sex: female) were cultured in Leibovitz’s L-15 medium (Thermo, 11415064) supplemented with 10% FBS. NIH/3T3 (sex: male), LLC1, MC38 (sex: female), ID8 (sex: female), B16-F10 (sex: male), and HEK293T (sex: female) cells were cultured in DMEM (Thermo, 11965167) medium supplemented with 10% FBS.

METHOD DETAILS

Short Pairpin RNAs (shRNA), Small Interfering RNA (siRNA) and Genetic Knockout

Plasmids expressing short hairpin RNA targeting LDHA, CEBPB or scramble sequences were purchased from GE Dharmacon. ShRNA sequences were packed into a lentivirus packaging construct and transfected into HEK293T cells with lipofectamine 2000 (Invitrogen). 4T1 and Py8119 cells were infected with shRNA expressing lentiviruses and selected with 10 mg/ml puromycin. siRNAs targeting ULK1, FIP200, ATG5 and LC3b (Qiagen) were transfected into tumor cells with HiPerFect transfection reagent (Qiagen).

G-CSF and shLDHA Co-Expression Vectors

G-CSF fragment with CMV promoter was cloned from G-CSF expression plasmid (MR225697, Origene) using KOD Xtreme hot start DNA polymerase (EMD Millipore). The GFP with CMV promoter in pGIPZ plasmid was cut off from shRNA (shLDHA and Scramble) expression plasmids (GE Dharmacon) with Xbaǀ and NotI (New England Biolabs) and replaced by G-CSF fragment with CMV promoter using T4 DNA Ligase (New England Biolabs). The following primers were used to clone G-CSF fragment: CTAGtctagatagttat taatagtaatcaattacggggtc and TTTTCCTTTTgcggccgcCTAGGCCAAGTGGTGCAGAGCA. G-CSF-pGIPZ plasmid was packaged with psPAX2 and pMD2.G plasmids in HEK293T cells to produce G-CSF expressing lentivurs.

G-CSF Knockout Cells

4T1 cells were transfected with G-CSF Double Nickase Plasmid (sc-419844-NIC, Santa Cruz) according to the manufacturer’s instructions. Two days later, the transfected cells were cultured in RPMI 1640 complete medium with 5 μg/ml puromycin for 3 days. Living cells were seeded into 96 well plates with unlimited dilution to reach one cell per well. G-CSF knockout clones were validated with ELISA.

MDSC and CD4/CD8 T cell Depletion

MDSCs were depleted with anti-Ly6G antibody (1A8, BioXCell). Anti-Ly6G antibody and isotype controls (200μg per mouse) were administered intraperitoneally (i.p.) into tumor bearing mice every three days starting from day 3. CD4⁺ and CD8⁺ T cells were depleted with anti-CD4 (GK1.5, BioXCell) and anti-CD8 (YTS 169.4, BioXCell) antibodies. Anti-CD4 (100μg per mouse) and anti-CD8 (100μg per mouse) antibodies were injected i.p. at the beginning of tumor inoculation and continuously administered every three days.

LAP*, LAP* Mutant, and LAP Expressing Vectors

Full length CEBPB LAP* expressing plasmid was purchased from Origene company. Site-directed mutation of second translation initiation site in LAP* mRNA ATG/GCG (Uematsu et al., 2007) was introduced using KOD Xtreme hot start DNA polymerase (EMD Millipore) according to the manufacturer’s instructions. The following primers were used to yield LAP* mutant expressing vector: gcctttagacccGCggaagtggccaac and gttggccacttccGCgggtctaaaggc. LAP expressing plasmid was gifted from Addgene (Plamid #12557).

LAP Knock Out Cells

Guide RNA sequences for CRISPR/CAS9 were developed at CRISPR design web interface (http://crispr.mit.edu/). Three gRNAs around second ATG (initial translation site for LAP) were selected.

gRNA#1

5’-CACCGGCCCGCCGCCTTTAGACCCA-3’/5’-AAACTGGGTCTAAAGGCGGCGGGCC-3’.

gRNA#2

5’-CACCGGTCGGGCTCGTAGTAGAAGT-3’/5’-AAACACTTCTACTACGAGCCCGACC-3’.

gRNA#3

5’-CACCGGTAGAAGTTGGCCACTTCCA-3’/5’-AAACTGGAAGTGGCCAACTTCTACC-3’.

The complementary oligonucleotides for guide RNAs were annealed and cloned into pSpCas9(BB)-2A-Puro (PX459) V2.0 (Addgene Plasmid #62988). The homologous recombination template with ATG mutation (ssDNA Ultramer from IDT) is ssDNA #1 CGCGTTCATGCACCGCCTGCTGGCCTGGGACGCAGCATGCCTCCCGCCGC

CGCCCGCCGCCTTTAGACCCgcaGAAGTGGCaAACTTCTACTACGAGCCC

GACTGCCTGGCCTACGGGGCCAAGGCGGCCCGCGCCGCGCCGCGCGCCCC and ssDNA #2 GGGGCGCGCGGCGCGGCGCGGGCCGCCTTGGCCCCGTAGGCCAGGCAGTCGGGCTCGTAGTAGAAGTTtGCCACcTCtgcGGGTCTAAAGGCGGCGGGCGGCG GCGGGAGGCATGCTGCGTCCCAGGCCAGCAGGCGGTGCATGAACGCG

4T1 cells were transfected with either pX459/gRNA#1 with ssDNA#1 or pX459/gRNA#2 and gRNA#3 with ssDNA#2 using Lipofectamine 2000 (Thermo Fisher Scientific). Two days later, the transfected cells were cultured in RPMI 1640 complete medium with 5 μg/ml puromycin for another two days. Living cells were seeded into 96 well plates with unlimited dilution to reach one cell per well. LAP knockout clones were confirmed by Western blot.

LAP Over Expression Lentivirus

LAP fragment with CMV promoter was cloned from LAP expression plasmid (plasmid #12557, Addgene) using KOD Xtreme hot start DNA polymerase (EMD Millipore). The GFP with CMV promoter in pGIPZ plasmid was cut off from Scramble plasmids (GE Dharmacon) with Xbaǀ and NotI (New England Biolabs) and replaced by LAP fragment with CMV promoter using T4 DNA Ligase (New England Biolabs). The following primers were used to clone LAP: 5’-CTAGtctagatagttattaatagtaatcaattacggggtc-3’ and 5’-TTTTCCTTTTgcggccgcCTAGCAGTGGCCCGCCGAGGCC-3’. LAP-pGIPZ plasmid was packaged with psPAX2 and pMD2.G plasmids in HEK293T cells to produce LAP expressing lentivurs.

Cytokine Detection

Tumor cells were cultured in different conditions. The cells were subject to real-time PCR assay. The culture supernatants were subject to cytokine measurement with ELISA kits (R&D System).

Western Blot

Cell lysates were prepared in RIPA buffer (Thermo Fisher Scientific). The protein concentrations of cell lysates were determined by BCA protein assay kit (Thermo Scientific). Equivalent amounts of total cellular protein were separated by SDS PAGE and transferred to nitrocellulose membranes. The proteins were detected with specific antibodies to CEBPB (SantaCruz Biotechnology), LDHA, p-AMPK, AMPK, p-ULK1 (317), ULK1, LC-3b, and β-actin (Cell Signaling Technology). The protein expression levels were quantified with ImageJ software and were normalized to 1 in specific control groups (Liu et al., 2010).

RNA Extraction, Reverse Transcription, and Real-Time Polymerase Chain Reaction

RNA extraction, reverse-transcription, and real-time polymerase chain reaction (PCR) were described previously (Curiel et al., 2004; Wang et al., 2016). Primers were purchased from OriGene Technologies.

Flow Cytometry Analysis (FACS)

Single cell suspensions were prepared from different organs and tumor tissues. Single cells were stained with fluorescence-conjugated anti-CD45, anti-CD3, anti-CD4, anti-CD8, anti-TNF-α, anti-IFN-γ, anti-IL-2, anti-Gr-1, anti-Ly6G, and anti-CD11b (BD, USA), and were analyzed with LSR II flow cytometry (BD Biosciences, San Jose).

Immunofluorescence Staining

For G-CSF staining, tumor cells were cultured and fixed with Foxp3 fixation transcription factor and permeabilization buffer and washed with Foxp3 transcription factor wash buffer (Thermo Fisher Scientific). Then, the cells were blocked with 2.5% normal goat serum blocking solution (Vector laboratories) and were stained with anti-G-CSF antibody (abcam) and goat anti-rabbit Alexa fluor 594 secondary antibody (Thermo Fisher Scientific). For autophagy staining, tumor cells were seeded in 24 well plates and treated with 50 μM chloroquine (CQ) (Sigma-Aldrich) for 6 hours. Cells were washed and stained with primary antibody, anti-mouse LC-3b antibody (Cell Signaling Technology) and secondary antibody, goat anti-rabbit Alexa fluor 594 secondary antibody (Thermo Fisher Scientific), and counter-stained with DAPI. Cells were analyzed with fluorescence microscope (Leica Biosystems).

Metabolic Assay

Tumor glucose metabolism was measured in a Seahorse XFe24 analyzer on V7-PET XF24 cell culture microplates (Seahorse Bioscience). One day before the assay, Tumor 4T1 cells were seeded into the wells and cultured in complete RPMI 1640 medium. Before testing, the cells were washed three times with Seahorse XF assay medium (Seahorse Bioscience) and were added 225 μl Seahorse XF assay medium. Different metabolic modifiers were added as indicated.

Lactate Detection

Tumor cells were cultured in six well plates with complete RPMI 1640 medium for 24 hours. Culture supernatant was collected and filtered with protein concentrators PES 10 K MWCO (Thermo Fisher Scientific). Lactate was measured with lactate assay kit (Sigma-Aldrich) according to the instruction.

ATP and AMP Measurement

Tumor cells (10⁷) were lysed with 1 ml lysis buffer and filtered with protein concentrators PES 10 K MWCO (Thermo Fisher Scientific). The filtered lysates were detected by Liquid chromatography–mass spectrometry (LC-MS) (Angilent). ATP and AMP standards were purchased from Sigma-Aldrich.

Microarray

Scramble and LDHA KD 4T1 cells were cultured for 24 hours and the total RNA of Scramble and LDHA KD 4T1 cells was extract with Direct-zol RNA MiniPrep Kit (Zymo Research). Gene expression microarray was carried out at the University of Michigan Microarray Core Facility using Affymetrix Mouse Gene ST 2.1 Chip according to the standard protocol. Two biological replicates of each sample were prepared independently. Data was analyzed by the University of Michigan Bioinformatics Core Facility. RMA was used to fit log2 expression values to the data using the oligo bioconductor package in R version 3.4.0. Volcano plots were generated by R. The results were uploaded into GEO as GEO: GSE103925.

MDSC Suppression Assay

MDSC suppressive assay was performed as previously described (Hamilton et al., 2012; Srivastava et al., 2010). Ly6G+ MDSCs cells were isolated from tumor tissues of 4T1 and Py8119 tumor models with Myeloid-Derived Suppressor Cell Isolation Kit (Miltenyi Biotec). Splenocytes were isolated from OT-II and OT-I transgenic mice, respectively. OT-II and OT-I cells were stimulated with 1 μg/ml OVA peptide (323–339) and 10 μg/ml OVA peptide (257–264), respectively, and co-coultured with MDSCs. T cell activation markers and cytokines were detected by flow cytometry (FACS).

Bioinformatics Analysis

TCGA (n=547) data set was obtained from the Cancer Genome Atlas Network reported by Kobodt (Koboldt et al., 2012). METABRIC data set was obtained from the European Genome-phenome Archive (EGAS00000000083) reported by Curtis (Curtis et al., 2012) and the clinical information of this data set was download from Oncomine reported by Curtis (Curtis et al., 2012). The clinical information was matched with the gene expression data on the basis of patient identifiers. GEO: GSE58812 data set was downloaded from Gene Expression Omnibus (GEO) (Jezequel et al., 2015). GSEA analysis was performed as we previously reported (Subramanian et al., 2005; Sun et al., 2016). The gene sets used for the enrichment analysis including h.all.v5.2.symbols, c5.all.v5.2.symbols and c2.cp.kegg.v5.2.symbols were from the Molecular Signatures Database-MsigDB and immune response metagene clusters (Rody et al., 2009). MDSC signature gene set and gene signature score calculation were previously defined (Welte et al., 2016). High specific signature score indicates high specific pathway activation. The network of enrichment results was mapped with Cytoscape 3.4 (Shannon et al., 2003). Data sets were normalized by standard deviation by R. Pathway signature scores were calculated and re-scaled by R.

QUANTIFICATION AND STATISTICAL ANALYSIS

Comparisons of measurement data between two groups were performed using two-sided Student’s t-test. Tumor growths in different groups were tested by repeated measures ANOVA. The Pearson’s correlation coefficient was calculated to evaluate the association between markers and a p-value < 0.05 indicates that the correlation is significantly different from zero. To control for cohort (dataset), the Pearson’s partial correlation was used. Overall patient survival was defined as the time from date of diagnosis to disease-related death. Metastasis free survival was defined as the interval from date of diagnosis to first metastasis. Survival data was censored at the last date the patient was known to be alive or free of metastasis. Survival functions were estimated by the Kaplan-Meier method and compared using the log-rank test. To control for cohort, the log-rank test was stratified according to cohort. Furthermore, the Cox proportional hazards model was used to assess the effect of each maker to survival. In the model, continuous maker data was used and cohort was controlled by allowing a different baseline hazard function for each cohort. The assumptions of proportional hazard and linear form of covariates were assessed by martingale residuals plots and the Kolmogorov-type supremum test. All analyses for survival data were done using SAS 9.4 software. P < 0.05 was considered significant.

DATA AND SOFTWARE AVAILABILITY

The accession number for the raw data for gene expression arrays reported in this paper is GEO: GSE103925.

Highlights.

• Aerobic glycolysis affects G-CSF and GM-CSF expression in TNBC

• Aerobic glycolysis regulates CEBPB isoform, LAP, via AMPKULK1-autophagy pathway

• LAP controls G-CSF and GM-CSF expression and MDSC development

• Aerobic glycolysis impacts tumor immunity and patient outcome through MDSCs