Energy status dictates PD-L1 protein abundance and anti-tumor immunity to enable checkpoint blockade

SUMMARY

Aberrant energy status contributes to multiple metabolic diseases, including obesity, diabetes and cancer, but the underlying mechanism remains elusive. Here, we report that ketogenic diet-induced changes in energy status enhance the efficacy of anti-CTLA-4 immunotherapy through decreasing PD-L1 protein levels and increasing expression of type I interferon and antigen presentation genes. Mechanistically, energy deprivation activates AMPK, which in turn phosphorylates PD-L1 on Ser283, thereby disrupting its interaction with CMTM4 and subsequently triggering PD-L1 degradation. In addition, AMPK phosphorylates EZH2, which disrupts PRC2 function, leading to enhanced IFNs and antigen presentation gene expression. Through these mechanisms, AMPK agonist or ketogenic diet enhances the efficacy of anti-CTLA-4 immunotherapy and improves the overall survival rate in syngeneic mouse tumor models. Our findings reveal a pivotal role for AMPK in regulating the immune response to immune-checkpoint blockade and advocate for combining ketogenic diet or AMPK agonist with anti-CTLA4 immunotherapy to combat cancer.

Graphical Abstract

eTOC Blurb

Dai et al. show that energy deprivation stimulates anti-tumor immunity and strengthens checkpoint blockade immune-therasspy in part through AMPK-mediated reduction of PD-L1 protein abundance as well as inactivation of the polycomb repressive complex 2 complex, leading to increased expression of IFNs and antigen presentation genes.

INTRODUCTION

Excessive calorie intake-induced diabetes and obesity are potent risk factors contributing to tumorigenesis (Giovannucci et al., 2010; Hursting et al., 2012). On the other hand, calorie restriction has been found to have anti-tumor effects (Collins et al., 2019; Di Biase et al., 2016; Longo and Fontana, 2010; Pietrocola et al., 2016). Furthermore, metabolic ketogenic diet, a high fat, low-carbohydrate/adequate protein nutritional therapy used in the treatment of refractory epilepsy (Freeman et al., 2007; Paoli et al., 2013), has also been implicated as an effective anti-cancer strategy (Freeman et al., 2007; Hopkins et al., 2016; Paoli et al., 2013). For instance, ketogenic diet-mediated suppression of insulin feedback greatly enhances the efficacy of PI3K inhibitors in a pancreatic tumor model (Hopkins et al., 2018). Calorie restriction and ketogenic diet can stimulate T cell functions that contribute to the anti-tumor effects (Collins et al., 2019; Di Biase et al., 2016; Lussier et al., 2016; Pietrocola et al., 2016), but its potential role in response to immune checkpoint blockade therapies, such as anti-PD-1/PD-L1 or anti-CTLA-4 antibody treatment remains largely elusive.

Adenosine monophosphate (AMP)-activated protein kinase (AMPK) is a key sensor for energy stresses, playing crucial roles in adaptive responses to falling energy levels (Hardie et al., 2012; Lin and Hardie, 2018; Mihaylova and Shaw, 2011). AMPK is a heterotrimeric protein, comprised of a catalytic α subunit, a β subunit, and a γ subunit. During energy stress, cellular levels of AMP are increased while ATP levels decline, leading to AMPK activation as a result of allosteric changes (Hardie et al., 2012). AMP binding also activates the upstream tumor suppressor liver kinase B1 (LKB1), in complex with STRAD and MO25, to directly phosphorylate the T172 residue in the α subunit of AMPK, leading to a relatively higher level of activation (Hawley et al., 2003; Shaw et al., 2004; Woods et al., 2003). As a master regulator of metabolism, the functions of AMPK in metabolism, protein synthesis and autophagy have been well studied (Herzig and Shaw, 2018), but the physiological functions of AMPK in tumorigenesis and immune responses remain largely undefined.

While PD-1/PD-L1 or CTLA-4 based immune checkpoint blockade therapy has been approved for treating multiple tumor types with durable clinical benefit (Sharma and Allison, 2015; Sharpe and Pauken, 2018), the efficacy varies in different cancer types, and the response rate to date rarely exceeds 40% (Pitt et al., 2016). Therefore, much effort has been devoted to further improving the efficacy of immune checkpoint blockade. Here, we report that ketogenic diet-induced changes in energy status enhance the efficacy of anti-CTLA-4 immunotherapy through decreasing PD-L1 levels and increasing expression of type I interferon (IFN) and antigen presentation genes. Hence, our findings reveal the pivotal roles of AMPK in regulating cellular response to immune-checkpoint blockade and advocate for combining ketogenic diet or AMPK agonist with anti-CTLA4 immunotherapy to combat cancer.

RESULTS

Energy status dictates PD-L1 protein abundance

Most tumor cells are addicted to high glucose levels (Vander Heiden et al., 2009), which may be an important factor underlying the development and progression of many types of human cancer. However, the underlying molecular mechanism is not fully defined. Given the key role of PD-L1 in tumor immune evasion (Schildberg et al., 2016; Sharpe and Pauken, 2018; Zou et al., 2016), we investigated whether PD-L1 protein abundance was regulated by energy deprivation. Notably, glucose starvation strongly decreased PD-L1 protein abundance in multiple human and mouse cancer cell lines (Figures 1A–1C and S1A–S1E), as well as in mouse macrophage cell lines (Figures S1F and S1G), but other B7 family checkpoint proteins such as PD-L2, B7-H3 and B7-H4 did not change obviously in this experimental setting (Figures 1A and 1B). However, PD-L1 mRNA levels did not decrease upon glucose starvation (Figures S1H and S1I), suggesting that cellular energy stress likely affects PD-L1 protein levels through post-translational regulation. Glucose starvation also activated AMPK, as indicated by increased phosphorylation of AMPK-Thr172 and acetyl-CoA carboxylase (ACC)-Ser79, a well-characterized AMPK substrate (Hardie et al., 2012; Lin and Hardie, 2018; Mihaylova and Shaw, 2011) (Figures 1A, 1B and S1A–S1G).

Figure 1. Energy status regulates PD-L1 protein abundance.

(A and B) Breast cancer cells MDA-MB-231 (A) or BT-549 (B) were starved of glucose (-Glc) for indicated time points before harvesting for immunoblot (IB) analysis.

(C) MDA-MB-231 cells were cultured with (Control, red) or without (-Glc, blue) glucose for 24 h before harvesting for FACS analysis of PD-L1. Mouse IgG2b control antibody, Green.

(D and E) MDA-MB-231 cells (D) or BT-549 cells (E) were treated with 2-DG (5 mM) for indicated time points before harvesting for IB analysis.

(F) Mice were fed with standard diet (SD) or fasted (F) for 24 h, and livers were collected for IB analysis.

(G) Mice were treated with single dose of streptozotocin (STZ) 200 mg/kg. After 2 weeks, livers were collected for IB analysis.

(H) Mice bearing CT26 implanted tumors were fed with standard diet (SD) or ketogenic diet (KD) for 10 days, then tumors were collected for IB analysis.

(I) Mice were fed with SD or KD for 10 days. Small intestines were collected for IB analysis.

See also Figure S1

2-deoxy-D-glucose (2-DG), a non-metabolizable glucose analogue that inhibits normal glucose metabolism and induces cellular energy stress (Inoki et al., 2003), activated AMPK and also resulted in decreased PD-L1 protein abundance in a time- and dose-dependent manner (Figures 1D, 1E, and S1J–S1L). Furthermore, we observed that fasting could dramatically decrease PD-L1 protein abundance in mouse livers and increase AMPK kinase activity (Figure 1F). Interestingly, in streptozotocin (STZ) induced diabetic models (Furman, 2015), we found that PD-L1 protein abundance increased in hyperglycemia condition (Figure 1G). These results together suggest that energy stress could affect PD-L1 protein levels both in cultured cells and in vivo.

Ketogenic diet has been used to treat epilepsy and significantly lowers glucose levels and produces ketone bodies in vivo (Hopkins et al., 2018; Kennedy et al., 2007). To study the effects of ketogenic diet on cells, we used low glucose medium containing β-hydroxybutyrate (βOHB), a ketone body (Cahill, 2006), to mimic the effects of ketogenic diet in cell cultures. Interestingly, we found that βOHB alone did not noticeably decrease PD-L1 protein levels, while low glucose plus βOHB markedly decreased PD-L1 expression (Figures S1M–P). This result indicates that changes of PD-L1 abundance in ketogenic diet setting were mainly due to low glucose status, but not the generation of βOHB that was in part due to βOHB’s inability to activate AMPK (Figures S1O and S1P) in this experimental setting. To test the effect of ketogenic diet on PD-L1 expression in vivo, we fed mice carrying implanted murine CT26 colon cancer cells with ketogenic diet for 10 days. Consistent with previous studies (Hui et al., 2020; Kennedy et al., 2007; Tognini et al., 2017), ketogenic diet, without causing significant changes in caloric intake (Figures S1Q and S1R), induced a significant decrease in blood glucose compared with standard diet (Figure S1S). Furthermore, the AMPK pathway was activated while mTORC1 kinase activity was reduced after ketogenic diet treatment (Figures 1H, 1I, S1U and S1V) (Kennedy et al., 2007; McDaniel et al., 2011). Moreover, we found that PD-L1 protein levels roughly correlated with tumor weight and were decreased in tumors collected from ketogenic diet fed mice compared with standard diet fed mice (Figure 1H and S1T). Ketogenic diet treatment also decreased PD-L1 protein levels in various organs of normal mice (Figures 1I, S1U and S1V) and tumor infiltrating myeloid cells (Figures S1W and S1X). Collectively, our data suggest that ketogenic diet-mediated energy stress inhibits PD-L1 expression both in cells and in vivo.

Ketogenic diet enhances the efficacy of anti-CTLA-4 immune checkpoint blockade

Ketogenic diet has been shown to reduce tumor growth in different cancer types (Paoli et al., 2013). Ketogenic diet also plays a role in alleviating tumor immune suppression in malignant glioma (Lussier et al., 2016), and in enhancing the treatment efficacy of PI3K inhibitors through reducing blood glucose levels (Hopkins et al., 2018). Emerging evidence demonstrates that blocking both PD-L1 and CTLA-4 immune inhibitory checkpoints was more effective than single agent immune blockade cancer therapy (Sharma and Allison, 2015). Given our results demonstrating that energy stress attenuates PD-L1 expression, essentially mimicking PD-1 blockade, we hypothesized that manipulating energy status might synergize with anti-CTLA-4 therapy to more efficiently combat cancers. To this end, we treated syngeneic mice bearing CT26 tumors with ketogenic diet and/or an anti-CTLA-4 antibody (Figure S2A). We found that ketogenic diet alone moderately retarded tumor growth in vivo but no mice survived, while anti-CTLA-4 antibody alone resulted in 5 out of 12 survivors (Figures 2A and 2B). Notably, combining ketogenic diet with anti-CTLA-4 antibody markedly retarded tumor progression and resulted in 8 out of 11 survivors (Figures 2A and 2B).

Figure 2. Ketogenic diet enhances the efficacy of anti-CTLA-4 immune checkpoint blockade.

(A) Mice bearing CT26 implanted tumors were enrolled in different treatment groups as indicated. Tumor volumes were measured every three days and plotted individually. Control (n=12); KD (n=12); CTLA-4 mAb (n=12); Combined (n=11).

(B) Kaplan-Meier survival curves for each group treated in (A).

(C and D) qRT-PCR analyses of relative mRNA levels of type I IFN and interferon-stimulated genes (ISGs) (C) or antigen presentation genes (D) in purified cancer cells derived from SD or KD treated CT26 syngeneic tumors.

(E) Representative dot plots of CD4⁺ and CD8⁺ TILs in CT26 syngeneic tumors.

(F-I) FACS analysis of cell numbers of CD3⁺CD8⁺ TILs (F) or CD3⁺CD8⁺GzmB⁺ TILs (G) or CD3⁺CD4⁺ TILs (H) or CD3⁺CD4⁺Foxp3⁺ TILs (I) from CT26 implanted tumors.Tumor-bearing mice were treated as indicated for 10 days. Control (n=5), KD (n=5), CTLA-4 mAb (n=6), combined (n=6).

*P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant, as determined by Gehan-Breslow-Wilcoxon test (B) or unpaired t test (C, D, F and G). Data are represented as mean ±SD (C and D).

See also Figure S2

The response to cancer immunotherapy also requires the presence of neoantigen and antigen presentation pathways, which are under the regulation of IFN signaling pathways (Dunn et al., 2006; Parker et al., 2016). Consistent with this notion, we found that expression of type I IFNs, interferon-stimulated genes (ISGs) and antigen presentation genes increased in the purified tumor cells (Figures 2C and 2D) and tissues of small intestine following ketogenic diet treatment (Figures S2B and S2C). Furthermore, ketogenic diet treatment significantly increased CD8⁺ and CD8⁺GzmB⁺ tumor-infiltrating T lymphocytes (TILs), but did not significantly affect numbers of CD4⁺ and CD4⁺Foxp3⁺ TILs in tumors of the combined treatment group (Figures 2E–2I, S2D and S2E). Importantly, these CD8⁺ TILs showed enhanced capability to produce cytokines such as interferon gamma (IFNγ), tumor necrosis factor alpha (TNFα) and interleukin 2 (IL-2) (Figures S2F–S2H), indicating a more effective T cell mediated anti-tumor immune response in the combined treatment condition. Moreover, combined treatment of ketogenic diet with CTLA-4 blockade also increased macrophage population and CD86⁺, but not CD80⁺ dendritic cells (Figures S2I–S2L), while monocytic myeloid-derived suppressor cell population (M-MDSC) showed no significant change (Figure S2M). These findings together suggest that ketogenic diet likely enhances the efficacy of anti-CTLA-4 therapy through both decreasing PD-L1 expression and promoting anti-tumor immunity pathways.

Energy stress induced PD-L1 degradation largely depends on AMPK

Since cells sense low energy status mainly through AMPK (Hardie et al., 2012; Lin and Hardie, 2018; Mihaylova and Shaw, 2011), we investigated the potential role of AMPK in regulating PD-L1 protein expression. PD-L1 protein levels increased dramatically in Ampkα1/α2 double-knockout (DKO) mouse embryonic fibroblasts (MEFs) compared with wild-type (WT) MEFs (Figure 3A). Furthermore, knockout of both AMPKα1 and α2, but neither alone could increase PD-L1 protein abundance, but not PD-L1 mRNA levels (Figure 3B and S3A–S3D). In line with this finding, activating AMPK by the AMPK specific agonists A-769662 (Cool et al., 2006) or compound 991 (Xiao et al., 2013) also reduced PD-L1 levels in various cell lines (Figures 3C and S3E–S3H). Consistent with a previous report (Cha et al., 2018), treatment with metformin, another AMPK agonist widely prescribed for type 2 diabetes treatment (Viollet et al., 2012; Zhou et al., 2001), also decreased PD-L1 expression (Figure S3I). On the other hand, treatment with an AMPK inhibitor, Compound C (Zhou et al., 2001), elevated PD-L1 protein abundance (Figures 3D and S3J–S3M).

Figure 3. Energy stress induced PD-L1 degradation largely depends on AMPK.

(A and B) IB analysis of whole-cell lysates (WCL) derived from Ampkα WT and Ampkα1/2^−/− double knockout (DKO) MEFs (A) or two different AMPKα DKO clones of BT-549 cells (B) generated by CRISPR-Cas9 technology.

(C) IB analysis of MDA-MB-231 cells treated with A-769662 (100 μM) for indicated time points.

(D) MDA-MB-231 cells were treated with Vehicle (Control, red) or Compound C (10 μM, blue) for 24 h before harvesting for FACS analysis of PD-L1. Mouse IgG2b control antibody, Green.

(E and F) IB analysis of WT and AMPKα DKO BT-549 cells (E) or CT26 cells (F) glucose starved for the indicated time points.

(G) IB analysis of WT and AMPKα DKO BT-549 cells treated with A-769662 for the indicated time points.

(H) The expression of PD-L1 in LKB1 depleted BT-549 cells.

(I) Blood insulin concentration of mice fed with KD (n = 8) or SD (n = 8) for 7 days. ***P < 0.001, unpaired t test.

(J and K) WT and PTEN KO HCT116 cells were glucose starved (J) or treated with A-769662 (K) for the indicated time points before harvesting for IB analysis.

(L) WT and Ampkα DKO CT26 cells were serum starved for 12 h and stimulated with insulin (0.1 μM) for indicated time points before harvesting for IB analysis.

(M and N) WT and Ampkα DKO CT26 cells were treated with MK2206 (1 μM) or BKM120 (1 μM) for 24h (M) or treated with BKM120 (1 μM) for indicated time points (N) before harvesting for IB analysis.

(O) Mice bearing WT or Ampkα DKO CT26 implanted tumors were fed with SD or KD for 10 days, then tumors were collected for IB analysis. The gels of SD and KD treated samples were transferred to the same membrane, subsequently processed for immunoblot under identical condition, with the same exposure time.

See also Figure S3

In support of an important role for AMPK in regulating energy deprivation induced reduction of PD-L1 protein abundance, glucose starvation or A-769662 treatment did not noticeably decrease PD-L1 protein levels in AMPKα DKO cells (Figures 3E–3G and S3N). To exclude possible off-target effects of A-769662 on PD-L1 regulation (Benziane et al., 2009; Moreno et al., 2008), we generated the AMPKβ1-S108A mutant that was previously reported to abolish the activation of AMPK by A-769662 (Sanders et al., 2007). We found that A-769662 reduced PD-L1 protein abundance in cells expressing WT, but not the S108A mutant form of AMPKβ1 (Figure S3O). Furthermore, depletion of AMPK upstream activator LKB1 (Hawley et al., 2003; Shaw et al., 2004; Woods et al., 2003) could also increase PD-L1 protein levels (Figure 3H). Given that energy stress can lead to cell cycle arrest (Jones et al., 2005; Liang et al., 2007), which plays an important role in regulating PD-L1 levels in a CDK4-dependent manner (Zhang et al., 2018a), we investigated whether energy stress-induced PD-L1 expression is mediated by CDK4. Consistent with our previous work (Zhang et al., 2018a), deletion of CDK4 led to an increase in PD-L1 expression. However, both WT and Cdk4 KO MEFs showed a comparable decrease in PD-L1 expression following glucose deprivation, thus refuting a role for CDK4 in glucose deprivation mediated regulation of PD-L1 (Figure S3P).

In keeping with previous reports (Douris et al., 2015; Hopkins et al., 2018; Kennedy et al., 2007), we found that ketogenic diet could decrease plasma insulin level in vivo (Figure 3I). Since the insulin-PI3K signaling could physiologically elevate PD-L1 transcription or protein translation rate in cancer cells (Lastwika et al., 2016; Parsa et al., 2007; Zhang et al., 2018b), and has been reported to interplay with the AMPK signaling (Han et al., 2018; Hawley et al., 2014; Jeon, 2016; Mihaylova and Shaw, 2011; Suzuki et al., 2013), we further dissected whether suppression of the insulin-PI3K-AKT pathway could contribute to the observed PD-L1 changes following ketogenic diet. Consistent with a previous report (Parsa et al., 2007), depletion of PTEN, which resulted in aberrant elevation of PI3K/AKT oncogenic signaling, could lead to elevated PD-L1 levels in both MEFs and HCT116 cells (Figures 3J–K and S3Q–S3T). Notably, in both WT and PTEN-KO cell lines, activation of AMPK by glucose deprivation or A-769662 treatment could dramatically decrease PD-L1 protein levels (Figures 3J–3K and S3Q–S3R), whereas inhibition of AMPK by Compound C could increase PD-L1 protein levels (Figures S3S and S3T), regardless of PTEN genetic status. These results together implicate the pivotal role of AMPK in energy deprivation-mediated regulation of PD-L1 protein abundance.

As ketogenic diet can cause reduction of plasma insulin levels (Figure 3I), we examined how insulin-mediated activation of PI3K/AKT signaling participates in modulating PD-L1 protein abundance. In keeping with previous reports (Guo et al., 2016; Liu et al., 2014), insulin stimulation can lead to a dynamic activation of the PI3K/AKT signaling (Figures 3L and S3U), with a moderate elevation of PD-L1 expression in relatively later time course in AMPK-WT cells (Figures 3L and S3U). However, PD-L1 protein levels were kept at high levels and did not display obvious changes after insulin stimulation in AMPK-KO cells (Figures 3L and S3U). These results indicate that in acute insulin treatment condition, AMPK signaling might play a more important role in governing PD-L1 protein abundance.

On the other hand, pharmacological inhibition of the PI3K-AKT signaling pathway could markedly inhibit PD-L1 protein expression in AMPK-WT cells (Lastwika et al., 2016; Parsa et al., 2007), but only mildly affected PD-L1 levels in AMPK-KO cells (Figures 3M–3N and S3V). These results validated PI3K/AKT signaling as an upstream regulator of PD-L1, and suggest that the PI3K-AKT pathway likely regulates PD-L1 protein levels in part through modulating AMPK kinase activity (Hawley et al., 2014; Ning et al., 2011; Suzuki et al., 2013). Moreover, compared to normal diet fed mice, ketogenic diet fed mice exhibited a marked decrease of PD-L1 in the syngeneic grafted Ampk-WT CT26 tumors, whereas PD-L1 remained at a higher level in Ampk-KO grafted tumor cells (Figure 3O). Taken together, we propose a model that under normal conditions, insulin-PI3K-AKT and AMPK pathways simultaneously govern the PD-L1 levels, while under the ketogenic diet conditions, activated AMPK likely plays a major role in reducing PD-L1 expression, and PI3K likely regulates PD-L1 protein levels in part via regulating AMPK kinase activity (Figure S3W).

AMPK phosphorylates PD-L1 on Ser283 to trigger its degradation

To gain mechanistic insight into AMPK-mediated regulation of PD-L1 protein abundance, we found PD-L1 interacted with AMPKα (Figure 4A). Since AMPK functions as a kinase, we explored whether AMPK could directly phosphorylate PD-L1. An in vitro kinase assay revealed that compared to other domains of PD-L1, the cytoplasmic terminus of PD-L1 (c-Ter) could be heavily phosphorylated by AMPK (Figures 4B, S4A and S4B). Moreover, a potential AMPK substrate consensus motif in PD-L1 was identified (Figure S4C), with Ser283 being a potential AMPK phosphorylation site. Mutating serine 283 to alanine (S283A) largely diminished AMPK-mediated phosphorylation of hPD-L1 in vitro (Figure 4C). Similarly, mutating Ser285 in mouse PD-L1, the hPD-L1 Ser283 comparable site, also abolished AMPK-mediated phosphorylation of mPD-L1 in vitro (Figure S4D).

Figure 4. AMPK phosphorylates PD-L1 on Ser283 to trigger its degradation.

(A) IB analysis of WCLs and PD-L1 immunoprecipitates (IPs) derived from MDA-MB-231 cells treated with chloroquine (CQ) (50 μM) for 6 h and with or without glucose for 4 h before harvesting for IP and IB analyses.

(B) In vitro phosphorylation assays of GST-PD-L1 truncations demonstrate that AMPK primarily phosphorylates the c-terminus (c-Ter) of PD-L1.

(C) In vitro GST-PD-L1 phosphorylation assays to reveal Ser283 as the main AMPK phosphorylation site of hPD-L1.

(D) IB analysis of WCL and IPs derived from PD-L1 KO MDA-MB-231 cells re-introduced with WT PD-L1, treated with CQ for 6 h and with or without glucose for 4 h before harvesting.

(E) IB analysis of WCL and IPs derived from PD-L1 KO MDA-MB-231 cells re-introduced with WT PD-L1, then treated with or without Compound C (10 μM).

(F and G) PD-L1 knockout MDA-MB-231 cells re-constituted with WT or the S283A mutant form of PD-L1 were starved of glucose (F) or treated with Compound C (G) for the indicated time points before harvesting for IB analysis.

See also Figure S4

The phosphorylation of Ser283 of PD-L1 has been reported by multiple independent groups through mass spectrometry (Beli et al., 2012; Bennetzen et al., 2010; Mertins et al., 2016). Hence, we developed and validated a phospho-specific antibody against PD-L1-pS283 (Figures S4E and S4F). The S283A mutation markedly reduced the capacity of the antibody to recognize PD-L1 (Figure S4G). Moreover, phosphorylation of S283 of PD-L1 was markedly induced by glucose starvation (Figure 4D) and A-769662 treatment (Figure S4H), but was repressed by the AMPK kinase inhibitor, Compound C (Zhou et al., 2001) in MDA-MB-231 cells (Figure 4E). Importantly, A-769662 treatment was capable of promoting phosphorylation of PD-L1 at Ser283 in cells expressing WT, but not the S108A mutant form of AMPKβ1, thereby confirming the notion that WT-AMPKβ1 was required for A-769662 induced phosphorylation of AMPK and PD-L1 (Figure S4I). These data together demonstrate that AMPK can directly phosphorylate PD-L1 primarily on Ser283 both in vitro and in cells.

To further test whether glucose deprivation down-regulated PD-L1 expression through AMPK-mediated phosphorylation of PD-L1, we re-introduced PD-L1 WT or the S283A mutant into PD-L1 knockout cells, and observed that glucose starvation or A-769662 treatment decreased the protein abundance of WT, but not the S283A mutant form of PD-L1 (Figure 4F and S4J). Moreover, we found that the S283 site played a pivotal role in conferring sensitivity to glucose starvation-induced PD-L1 degradation, although mutating the S195 site also led to a moderate resistance (Figure S4K). Consistent with these findings, treatment with the AMPK inhibitor Compound C only increased PD-L1 expression in cells expressing WT, but not the S283A mutant form of PD-L1 (Figure 4G). These data suggest that AMPK-mediated phosphorylation of PD-L1 on S283 plays an important role in regulating PD-L1 protein abundance.

Phosphorylation of PD-L1 by AMPK promotes its degradation through inhibiting its interaction with CMTM4

Due to the critical role of PD-L1 in PD-1/PD-L1 blockade based immunotherapy, the regulation of PD-L1 expression has been extensively studied recently. PD-L1 can undergo degradation either in a proteasome-dependent manner, such as via β-TRCP (Li et al., 2016) or SPOP-mediated ubiquitination (Zhang et al., 2018a) or in a lysosome-dependent manner, with CMTM4/6 protecting of PD-L1 from ubiquitination and lysosome-mediated degradation (Burr et al., 2017; Mezzadra et al., 2017). Notably, we found that chloroquine, a lysosome inhibitor, could block PD-L1 degradation induced by glucose starvation (Figure 5A). Furthermore, the half-life of PD-L1 was significantly extended in Ampkα KO cells compared with WT cells (Figures 5B and S5A). Consistent with this finding, phosphorylation deficient PD-L1-S283A had a dramatically prolonged half-life compared with WT-PD-L1 (Figure 5C). However, the interaction of PD-L1 with its established E3 ubiquitin ligases including SPOP or β-TRCP remained unchanged under the condition of glucose deprivation (Figures S5B and S5C), thereby refuting the involvement of SPOP or β-TRCP in glucose deprivation-mediated PD-L1 protein degradation.

Figure 5. Phosphorylation of PD-L1 by AMPK promotes its degradation through inhibiting its interaction with CMTM4.

(A) MDA-MB-231 cells were incubated with media with or without glucose for 24 h and treated with MG132 (10 μM) or CQ (50 μM) for 10 h before harvesting.

(B) IB analysis of PD-L1 in WT versus AMPKα KO BT-549 cells. After addition of 100 μg/ml cycloheximide (CHX), cells were harvested at the indicated time points for IB analysis.

(C) IB analysis of PD-L1 KO MDA-MB-231 cells re-introduced with WT or the S283A mutant form of PD-L1. Cells were treated as in (B).

(D) IB analysis of WCL and IPs derived from 293T cells transfected with the indicated constructs. Cells were treated with CQ for 6 h and with glucose starvation or Compound C for 4 h.

(E) IB analysis of WCL and IPs derived from MDA-MB-231 cells. Cells were treated with CQ for 6 h and with or without glucose for 4 h.

(F) IB analysis of WCL and IPs derived from 293T cells transfected with the indicated constructs. Cells were treated with CQ for 6 h and with or without A-769662 for 2 h.

(G and H) IB analysis of WT and CMTM4 KO BT-549 cells glucose starved (G) or treated with Compound C (H) for indicated time points.

(I) IB analysis of WCL and IPs derived from 293T cells transfected with the indicated constructs.

See also Figure S5

Recently, CMTM4/6 were identified as critical positive regulators of PD-L1 protein stability (Burr et al., 2017; Mezzadra et al., 2017). Interestingly, the interaction of PD-L1 with CMTM4, but not CMTM6, could be attenuated by glucose deprivation (Figures 5D and 5E), and enhanced after treatment with the AMPK inhibitor, Compound C (Figures 5D and S5D). In further support of the notion that this disruption was dependent on AMPK-mediated phosphorylation, treatment with the AMPK specific agonist A-769662 markedly interfered with the interaction between PD-L1 and CMTM4, but not with CMTM6 in cells (Figures 5F and S5E). Furthermore, neither energy stress nor Compound C treatment could modulate PD-L1 protein levels in CMTM4 knockout cells (Figures 5G, 5H and S5F), thereby advocating for a pivotal role of CMTM4 in dictating PD-L1 protein abundance under glucose deprivation condition. Notably, the PD-L1-S283A mutant increased, whereas the PD-L1-S283D mutant decreased the interaction between PD-L1 and CMTM4 (Figure 5I), but not with CMTM6, SPOP or β-TRCP (Figures S5G–S5I). These results suggest that glucose starvation or AMPK activation promotes PD-L1 degradation through S283 phosphorylation, which in turn inhibits PD-L1 interaction with CMTM4 in cells.

AMPK regulates IFN and antigen presentation through phosphorylating EZH2

Given that ketogenic diet activates AMPK and ketogenic diet fed mice exhibit elevated type I IFN and antigen presentation gene expression (Figures 2C, 2D, S2B and S2C), we investigated the underlying mechanism of this elevation and a possible role of AMPK in this process. To this end, we analyzed RNA-seq data to explore whether and how AMPK regulates IFN activation. Gene ontology (GO) enrichment analysis clearly revealed that the IFN response-associated gene set was significantly enriched in the gene sets down-regulated in Ampkα knockout MEFs compared with WT MEFs (Figures 6A and 6B), which is consistent with the previous report that in LKB1 KO cells, type I IFN signaling is significantly down-regulated (Kitajima et al., 2019). These findings were further verified by analyzing selected IFN and ISGs by quantitative PCR (Figures 6C and S6A). Consistently, IFN and ISGs were elevated after metformin treatment (Figure S6B). More interestingly, the antigen presentation genes were also observed to decrease in AMPKα KO cells (Figure 6D). In keeping with a critical role of AMPK in this process, antigen presentation genes could be elevated by glucose starvation or A-769662 treatment (Figures 6E and S6C).

Figure 6. AMPK regulates IFN and antigen presentation through phosphorylating EZH2.

(A) A scatter plot of the top 10 enriched and depleted hallmarks from GSEA analysis by comparing Ampkα KO and WT mouse embryonic fibroblast cells. The gene sets were ordered by their normalized enrichment scores (NES).

(B) Gene set enrichment plots of the interferon alpha and interferon gamma response hallmarks.

(C) qRT-PCR analyses of relative mRNA levels of IFNs and ISGs from samples derived from WT or Ampkα KO MEFs.

(D and E) qRT-PCR analyses of relative mRNA levels of antigen presentation genes from samples derived from WT or AMPKα KO BT-549 cells (D) or glucose starved MDA-MB-231 cells (E).

(F) qRT-PCR analyses of relative mRNA levels of antigen presentation genes from EZH2-depleted BT-549 cells starved of glucose for 24 h.

(G) qRT-PCR analyses of relative mRNA levels of antigen presentation genes from various indicated EZH2 re-introduced BT-549 cells.

*P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant, as determined by unpaired t test. Data are represented as mean ±SD (C, D, E, F and G).

See also Figure S6

Recent studies have shown that EZH2 suppresses tumor immunity through inhibiting IFN signaling and antigen presentation (Burr et al., 2019; Canadas et al., 2018; Kitajima et al., 2019). Our previous work has shown that AMPK could phosphorylate EZH2 at Thr311 to suppress polycomb repressive complex 2 (PRC2) methyltransferase activity (Wan et al., 2018). Thus, we further examined whether AMPK regulates IFN and antigen presentation genes through phosphorylating EZH2. We found that glucose starvation failed to further increase the expression of antigen presentation genes in EZH2-depleted cells (Figures 6F and S6D). Notably, reintroduction of EZH2 WT or the phosphorylation deficient T311A (Wan et al., 2018) in EZH2-depleted cells could strongly suppress the expression of antigen presentation genes, whereas the phospho-mimetic T311E mutant (Wan et al., 2018) could only partially repress these genes (Figures 6G and S6E). These data suggest that AMPK promotes antigen presentation gene expression mainly through phosphorylating EZH2 on the T311 site, as we reported previously (Wan et al., 2018), to dampen PRC2-mediated H3K27me3, thereby modulating the transcription of these critical immune regulatory genes.

AMPK agonist enhances the efficacy of anti-CTLA-4 immune checkpoint blockade through increasing IFN and antigen presentation

Our studies show that energy stress could decrease PD-L1 protein expression, and simultaneously stimulate IFN and antigen presentation gene expression. To further test whether these findings are due to AMPK activation in vivo, we treated mice with A-769662 for two weeks, and found a significant decrease in PD-L1 protein levels in various organs of A-769662 treated mice (Figures 7A, S7A and S7B), accompanied by an increased expression of IFNs as well as the antigen presentation genes in small intestines (Figures S7C and S7D). Moreover, we treated mice carrying murine CT26 tumors and observed that A-769662 treatment significantly decreased PD-L1 protein levels (Figure 7B) and increased IFNs and ISGs, as well as the antigen presentation genes (Figures 7C and 7D) in tumors. Because both elevated IFNs and antigen presentation genes as well as decreased PD-L1 can contribute to enhancing the efficacy of anti-tumor immunotherapy (Zhang et al., 2018b), we hypothesized that combining A-769662 and anti-CTLA-4 antibody could improve the efficacy of tumor therapies in syngeneic mouse models. To this end, we treated mice bearing CT26 tumors with A-769662 or anti-CTLA-4 antibody separately or in combination (Figure S7E). A-769662 treatment alone reduced tumor growth (Figures 7E and 7F) and resulted in 2 long-term survivors out of 15 treated mice while CTLA-4 antibody alone resulted in 6 long-term survivors out of 15 treated mice (Figures 7E and 7F). Notably, A-769662 and anti-CTLA-4 antibody combination treatment resulted in a significant improvement in overall survival compared with either treatment alone (Figures 7E and 7F) in this immune-proficient mouse model, but not in an immunocompromised, T cell deficient mouse model (Figure S7F). Furthermore, the combined treatment caused a significant increase of CD8⁺ and CD8⁺GzmB⁺ TILs (Figures S7G–S7J), and elevated production of cytokines in the tumors (Figures S7K and S7L), but there was no significant change of CD4⁺ or CD4⁺Foxp3⁺ TILs (Figures S7M and S7N).

Figure 7. AMPK agonist enhances the efficacy of anti-CTLA-4 immune checkpoint blockade through increasing IFNs and antigen presentation.

(A) Mice were treated with or without A-769662 (20 mg kg⁻¹ body weight) for 2 weeks. Small intestines were collected for IB analysis.

(B) Mice bearing CT26 implanted tumors were treated with or without A-769662 for 2 weeks. Tumors were collected for IB analysis.

(C and D) qRT-PCR analyses of relative mRNA levels of IFNs and ISGs (C) or antigen presentation genes (D) from samples derived from Vehicle or A-769662 treated CT26 implanted tumors.

(E) Mice bearing CT26 implanted tumors were enrolled in different treatment groups as indicated. Tumor volumes were measured every three days and plotted individually. IgG+Vehicle (n=14), IgG+A-769662 (n=15), CTLA-4 mAb+Vehicle (n=15), or Combined (n=15).

(F) Kaplan-Meier survival curves for each group treated in (E).

(G) Kaplan-Meier survival curves for WT and Ampkα null CT26 cells treated as indicated.

(H) A working model to show how energy status regulates PD-L1 expression and the EZH2 pathway through the AMPK signaling pathway.

*P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant, as determined by unpaired t test (C and D), or the Gehan-Breslow-Wilcoxon test (F and G). Data are represented as mean ±SD (C and D).

See also Figure S7

Moreover, we did not observe the enhanced efficacy in the combined treatment group with CT26 Pd-l1 null cells compared with CT26 wild-type cells (Figures S7O and S7P), suggesting the importance of PD-L1 regulation by AMPK. To further prove the vital roles of AMPK in this process, we demonstrated that Ampkα null cells showed worse survival both in ketogenic diet treatment and in combined treatment conditions compared with wild-type cells (Figures 7G and S7Q). Thus, these results together suggest that combining AMPK agonist or ketogenic diet with anti-CTLA-4 therapy results in a significant improvement in overall survival compared with single agent treatment, which likely relies on decreasing PD-L1 expression and increasing antigen presentation

DISCUSSION

Although immune checkpoint blockade by anti-CTLA-4 or anti-PD-1/PD-L1 has provided substantial benefits to cancer patients (Sharma and Allison, 2015; Sharpe and Pauken, 2018), the response rate remains moderate (Pitt et al., 2016). Currently, much effort has been devoted towards further improvement of the efficacy of immune checkpoint blockade (Sharma and Allison, 2015). Our results show that ketogenic diet significantly improves the efficacy of anti-CTLA-4 immune therapy (Figure 2B). Ketogenic diet as epilepsy therapy has been implicated as an effective anti-cancer strategy (Hopkins et al., 2018; Kennedy et al., 2007). Recent evidence suggests that calorie restriction or ketogenic diet can modulate T cell functions, thereby contributing to the anti-tumor effects (Collins et al., 2019; Di Biase et al., 2016; Lussier et al., 2016; Pietrocola et al., 2016). Our findings further reinforced the possibility that ketogenic diet induced energy stress could boost immune system to suppress tumor progression through decreasing PD-L1 protein levels and promoting IFNs expression, which suggests that manipulating energy status in human body through ketogenic diet or calorie restriction will synergize with immune checkpoint blockade therapy to eradicate tumors.

Notably, combined immunotherapy targeting both PD-1/PD-L1 and CTLA-4 has shown improved efficacy in several cancer types (Amin et al., 2018; Hellmann et al., 2019; Larkin et al., 2015). Our study suggests that dampening PD-L1 expression is an essential aspect of the ketogenic diet enhanced cancer therapeutic effect, thus, it will be interesting to study whether the addition of PD-1/PD-L1 blockade could further improve survival in the ketogenic diet plus CTLA-4 blockade context. Furthermore, as cancer metastasis is responsible for mortality in the majority of cancer types (Chaffer and Weinberg, 2011), it will also be interesting to study whether ketogenic diet could suppress cancer cell metastasis by using lung or peritoneal metastasis mouse models in future studies (Taibi et al., 2019).

AMPK has been shown to suppress tumorigenesis through inhibiting the main anabolic processes (Zadra et al., 2015), but its role in immune checkpoint regulation remains elusive. We found that activated AMPK induced by energy stress phosphorylates PD-L1 on Ser283, thereby disrupting its interaction with CMTM4 and triggering PD-L1 degradation through autophagy (Figure 7H). Although a previous report showed metformin could promote anti-tumor immunity through AMPK phosphorylating PD-L1 on Ser195 site (Cha et al., 2018), our results show the Ser283 site on the PD-L1 cytoplasmic domain, plays a more important role in mediating the immune changes following glucose starvation in our experimental setting (Figure S4J). We also found AMPK could phosphorylate EZH2, which disrupts PRC2 function, leading to enhanced IFNs and antigen presentation gene expression, which facilitates T cell infiltration to kill tumor cells. Taken together, our study not only provides a molecular mechanism for AMPK directly regulating PD-L1 protein levels, but also reveals the mechanisms of AMPK-mediated IFNs and antigen presentation genes expression, highlighting the multiple strategies that combining ketogenic diet or AMPK agonist treatment with anti-CTLA-4 immunotherapy can work together to eradicate tumors (Figure S7F).

LIMITATIONS

Our data provided experimental evidence to support the notion that AMPK activation by ketogenic diet or AMPK agonist is important for PD-L1 protein homeostasis and anti-cancer immune response. However, as ketogenic diet modulates broad biological activities including but not limited to the insulin/PI3K/AKT pathway and the AMPK pathway (Hopkins et al., 2018; Kennedy et al., 2007; McDaniel et al., 2011), the detailed molecular mechanisms underlying regulation of PD-L1 by PI3K/AKT pathway and AMPK pathway as well as the potential crosstalk between these two pathways (Hawley et al., 2014; Mihaylova and Shaw, 2011; Ning et al., 2011), especially under ketogenic diet condition, warrant further investigation. Notably, our study suggests PD-L1 can be degraded through the lysosome pathway under energy stress conditions, whereas PD-L1 has been reported to be degraded through the ERAD pathway when treated with metformin (Cha et al., 2018). How these two processes are coordinated and whether an additional regulatory factor is involved in the AMPK-PD-L1 interaction remains elusive. To this end, a PD-L1 mutant mouse model carrying the mutation(s) in the respective AMPK phosphorylation site, individually or in combination, would help us thoroughly dissect the underlying molecular mechanisms. Furthermore, while our study mainly focuses on energy stress induced suppression of PD-L1 in tumor cells, how ketogenic diet modulates PD-L1 in immune cells and the potential metabolic effects of ketogenic diet in the immune checkpoint block-mediated immunotherapy also require further in-depth investigation.

STAR ★ METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Wenyi Wei (wwei2@bidmc.harvard.edu)

Materials availability

The materials will be available upon request.

Date and code availability

This study did not generate/analyze datasets/code.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

HEK293, HEK293T, MDA-MB-231, MC38, B16-F10, J774A.1, Raw264.7, wild-type and Ampk α1^−/−Ampk α2^−/− Mouse Embryonic Fibroblasts (MEFs) (a gift from Dr. Benoit Viollet, Institut Cochin), wild-type Cdk4^−/− (a gift from N. Mitsiadesa) were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) containing 10% fetal bovine serum (Gibco), 100 unit/ml penicillin and 100μg/ml streptomycin (Gibco). BT-549 and CT26 cells were cultured in PRMI-1640 medium. BT-549 PD-L1 KO cells were kindly gifted form Dr. Mien-Chie Hung.

Mice and ethics statement

C57BL/6J and BALB/cJ mice were obtained from The Jackson Laboratory. All mouse experiments were conducted under a protocol approved by the Beth Israel Deaconess Medical Center (BIDMC) institutional animal care and use committee (IACUC) (Protocol #043–2015) and Dana-Farber Cancer Institute Institutional Animal Care and Use Committee (IACUC; protocol number 04–047). For fasting and STZ induced diabetic model, 6–8 weeks old male C57BL/6J mice were used. For CT26 syngeneic mice tumor models, 6 weeks old female BALB/cJ mice were used. All mice were housed in a pathogen-free environment at BIDMC animal facility and were handled in strict accordance with the “Guide for the Care and Use of Laboratory Animals”.

METHOD DETAILS

Transfection, viral infection and cell treatment

Transfection with Lipofectamine (Invitrogen) or PEI (Polysciences) was performed according to the manufacturer’s instructions.

To generate stable cells, 293T cells were used for packaging of lentiviral and retroviral cDNA-expressing viruses. After transfection 48 h and 72 h, supernatant was harvested and filtered through a 0.45 μm syringe filter and used to infect cells in the presence of 4 μg/ml Polybrene. Infected cells were selected using hygromycin B (200 μg/ml) or puromycin (1 μg/ml) for three days.

For glucose starvation, cells were washed twice with PBS and then incubated in DMEM without glucose and pyruvate (Invitrogen) supplemented with 10% FBS, dialyzed (Invitrogen) and 100 unit/ml penicillin and 100μg/ml streptomycin (Gibco).

Plasmid construction

PGEX-GST-6P2-PD-L1 truncations were constructed by cloning the corresponding cDNAs into PGEX-GST-6P2 vector. PD-L1 cDNA was amplified and sub-cloned into pLenti-CMV-Hygro vector to generate lentiviral vector encoding PD-L1. The plasmids of Flag-CMTM4, CMTM6 were purchased from Genescript. HA-Ins-PD-L1 and Flag-SPOP constructs have been described previously (Zhang et al., 2018a). pLenti-CMV-hygro-HA-EZH2, point mutants and PLKO-shEZH2 have been described previously (Wan et al., 2018).

For generating CRISPR-mediated knockout cell lines, sgRNAs were sub-cloned into pLenti-CRISPRV2 GFP vector (Addgene). The sgRNA sequences, for human AMPKα1: sg1-GAAGATCGGCCACTACATTC, sg2-TACATTCTGGGTGACACGCT; for human AMPKα2: sg1-TCAGCCATCTTCGGCGCGCG, sg2-GAAGATCGGACACTACGTGC; for mouse Ampk α1, sg1-ACGCTTGGTGTCGGCACCTT; sg2-GCATCAAGCAGGACGTTCTC; for mouse Ampkα2, sg1-TTGAAGAGGTGGAAGCGCGC; sg2-TCGACTACATCTGCAAACAT; for human CMTM4 sg1-AAAGTAGAGGCCTTCACACG; sg2-CCGCCTCAAGGTCGCCCAAG.

PD-L1 mutants were generated using QuickChange XL Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer’s instructions.

Immunoblot and immunoprecipitation

Cells were lysed in EBC buffer (50 mM Tris pH 7.5, 120 mM NaCl, 0.5% NP-40) supplemented with protease inhibitors (Complete Mini, Roche) and phosphatase inhibitors (phosphatase inhibitor cocktail set I and II, Calbiochem). The cell lysates were centrifuged at 12,000 g for 15 min at 4 °C. Protein concentrations were measured using a Beckman Coulter DU-800 spectrophotometer using the Bio-Rad protein assay reagent. Equal amounts of protein were resolved by SDS–PAGE using a standard protocol.

For immunoprecipitation analysis, 1,000 μg total cell lysates were incubated with the primary antibody-conjugated beads or the PD-L1 antibody and protein A/G conjugated beads for 3–4 h at 4 °C. The recovered immunocomplexes were washed four times with NETN buffer (20 mM Tris, pH 8.0, 100 mM NaCl, 1 mM EDTA and 0.5% NP-40) before being resolved by SDS-PAGE.

In vitro kinase assays

Detection by anti-thiophosphate-ester antibody: For AMPK in vitro kinase assays, PD-L1 proteins purified from E. coli were subjected to kinase assays in the presence of 500 μM ATP-γ-S with recombinant AMPK α1/β1/γ2 proteins (SignalChem). The reaction mixtures were incubated for 30 min at 30°C. The reaction mixtures were further supplemented with 2.5 mM PNBM. Alkylating reactions were allowed to proceed for 1 h at room temperature. The reactions were terminated with SDS sample buffer and boiled before analysis by SDS-PAGE (Dai et al., 2013).

Detection by phosphorus-32 (³²P): PD-L1 protein purified from E. coli was subjected to kinase assays in the presence of 10 μM ATP and 5 μCi of [γ-³²P] ATP with recombinant AMPK α1/β1/γ2 proteins. The reaction mixtures were incubated for 30 min at 30°C, then terminated with SDS sample buffer and boiled before analysis by SDS-PAGE.

Preparation of recombinant proteins

E. coli BL-21 was transformed with bacterial expression constructs (pGEX-KG) containing the indicated genes. The expression of the recombinant GST-fusion proteins within the transformed bacteria was induced by using 0.1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) at 18°C overnight. Cells were re-suspended in PBS containing protease inhibitor cocktail followed by ultrasonication. The proteins were purified by a single step using glutathione beads according to the manufacturer’s protocol (GE healthcare). Purified proteins were dialyzed against 50 mM Tris at pH 8.0. Purified GST-proteins (0.5 μg) were used as substrates for AMPK in vitro kinase assay.

Protein half-life assays

Cells were transfected or treated under indicated conditions. For half-life studies, cycloheximide (100 μg/ml, Sigma) was added to the medium. Cells were collected at indicated time points and protein concentrations were measured, then analyzed by immunoblot analysis.

Tumor-infiltrating lymphocytes isolation

Tumors were resected from each of the tumor bearing mice. The tumor tissue was minced into about 3 mm pieces and transfer into a 15 ml tube. Add tumor dissociation buffer (RPMI1640, 5% FBS, 1mg/ml Collagenase IV (Sigma), 200U/ml DNase I (Roche)) in each tube containing tumor tissue. Rotating incubation at 37 °C for 20 minutes. Pass through a 70 μM strainer on the top of a 50 ml tube for each sample. Centrifuge 1900 RPM for 5 minutes, and aspirate the supernatant. Add 2 ml of RBC (red blood cell) lysis buffer, gently mix and incubate on ice for 2 minutes. Wash once with MACS buffer (HBSS without Ca2⁺, without Mg2⁺, 1% FBS, 2mM EDTA) by centrifuge 2000 RPM for 5 minutes. After Percoll (GE Healthcare) gradient centrifugation (2000 RPM, 30 minutes, room temperature, no brake), the layer of mononuclear cells was harvested. Add MACS buffer to wash the pellet. After counting the cells, the cells were seeded in 96 well plates (0.05×10⁶ cells/well), then stimulated in cell culture media (RPMI1640, 10% FBS, 1% Penicillin/Streptomycin of 10000 U each, 1.5 mg/ml Gentamycin, add 550 μl of 2-mercaptoethanol solution (55 mM in DPBS) with cell stimulation cocktail (eBioscience) plus protein transport inhibitor (Brefeldin A and Menesin) for 4 h. After stimulation, the cells were proceeded to Flow cytometry staining procedure.

Mouse tumor cell isolation

Tumors tissues resected from mice are dissociated into single-cell suspensions as described above. Then mixed cells were processed immediately for magnetic bead mediated negative selection to separate tumor cells by using a mouse tumor cell isolation kit (Miltenyi Biotac, #130–110-187).

FACS analyses

For cultured MDA-MB-231 tumor cell lines, the cells were detached by using 2mM EDTA in PBS and harvested in a 15 ml tube. Resulting cells were washed once with FACS buffer (PBS, 2% FBS, 0.1mM EDTA, 0.02% Azide) and filtered with 70 μm cell strainer, then adjusted the concentration to 1×10⁶ cells/ml. Cells were incubated with human Fc block at room temperature for 10 min. After that, 1μl of Live/Dead NIR was added into 1×10⁶ cells in a volume of 100 μl, and cells were washed once with FACS buffer, and resuspended in 50 μl of FACS buffer. In another tube, appropriate antibodies (listed in key resource table) were added into 50 μl of FACS buffer in which 50 μl of cells and 50 μl of antibody solution were mixed together, and incubated for 30 minutes at 4°C in dark. Cells were washed twice with FACS buffer and resuspended in 0.5 ml of 2% formaldehyde in PBS. The tubes were stored at 4°C in dark until analyzed by FACS.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-AMPKα	Cell Signaling Technology	Cat#: 5831
anti-AMPKα1	Cell Signaling Technology	Cat#: 2795
anti-pAMPKα T172	Cell Signaling Technology	Cat#: 2535
anti-pACC Ser79	Cell Signaling Technology	Cat#: 11818
anti-ACC	Cell Signaling Technology	Cat#: 3676
anti-human PD-L1 (E1L3N)	Cell Signaling Technology	Cat#: 13684
anti-PD-L2	Cell Signaling Technology	Cat#: 82723
anti-B7-H3	Cell Signaling Technology	Cat#: 14058
anti-B7-H4	Cell Signaling Technology	Cat#: 14572
anti-GST	Cell Signaling Technology	Cat#: 2625
anti-H3K27me3	Cell Signaling Technology	Cat#: 9733
anti-phsopho-Thr389-S6K	Cell Signaling Technology	Cat#: 9234
anti-p70 S6K	Cell Signaling Technology	Cat#: 9202
anti-phospho-S6 (Ser235/236)	Cell Signaling Technology	Cat#: 4858
anti-S6	Cell Signaling Technology	Cat#: 2217
Anti-phospho-Ser473-AKT	Cell Signaling Technology	Cat#: 4060
anti-phospho-Thr308-AKT	Cell Signaling Technology	Cat#: 2965
anti-AKT total antibody	Cell Signaling Technology	Cat#: 4691
Anti-vinculin	Sigma-Aldrich	Cat#: V9131
anti-Flag	Sigma-Aldrich	Cat#: F-3165
HRP-conjugated anti-mouse secondary antibody	Sigma-Aldrich	Cat#: A-4416
HRP-conjugated anti-rabbit secondary antibody	Sigma-Aldrich	Cat#: A-4914
Anti-AMPKα2	Bethyl	Cat#: A300-508A
anti-mouse PD-L1, clone EPR20529	Abcam	Cat#: ab213480
Anti-Thiophosphate ester antibody	Abcam	Cat#: ab92570
anti-HA	Biolegend	Cat#: MMS-101P
PE anti-human PD-L1	Biolegend	Cat#: 329706
Brilliant Violet 785™ anti-mouse CD3ε	Biolegend	Cat#: 100355
Brilliant Violet 510™ anti-mouse IL-2	Biolegend	Cat#: 503833
FITC anti-mouse IFN-γ	Biolegend	Cat#: 505806
APC anti-mouse TNF-α	Biolegend	Cat#: 506308
Brilliant Violet 711™ anti-mouse CD8a	Biolegend	Cat#: 100748
Brilliant Violet 785™ Armenian Hamster IgG Isotype Ctrl	Biolegend	Cat#: 400947
Brilliant Violet 510™ Rat IgG2b, κ Isotype Ctrl	Biolegend	Cat#: 400646
FITC Rat IgG1, κ isotype Ctrl	Biolegend	Cat#: 400406
APC Rat IgG1, κ isotype Ctrl	Biolegend	Cat#: 400412
Brilliant Violet 711™ Rat IgG2a, κ Isotype Ctrl	Biolegend	Cat#: 400551
Brilliant Violet 605™ anti-mouse CD45	Biolegend	Cat#: 103140
Brilliant Violet 605™ Rat IgG2b, κ Isotype Ctrl	Biolegend	Cat#: 400657
Brilliant Violet 650™ anti-mouse CD4	Biolegend	Cat#: 100469
Brilliant Violet 650™ Rat IgG2b, κ Isotype Ctrl	Biolegend	Cat#: 400651
Pacific Blue™ anti-human/mouse Granzyme B	Biolegend	Cat#: 515408
Pacific Blue™ Mouse IgG1, κ isotype Ctrl	Biolegend	Cat#: 400151
Brilliant Violet 650™ anti-mouse/human CD11b	Biolegend	Cat#: 101259
Brilliant Violet 510™ anti-mouse CD11c	Biolegend	Cat#: 117353
Brilliant Violet 510™ Armenian Hamster IgG Isotype Ctrl	Biolegend	Cat#: 400942
Brilliant Violet 785™ anti-mouse F4/80	Biolegend	Cat#: 123141
Brilliant Violet 785™ Rat IgG2a, κ Isotype Ctrl	Biolegend	Cat#: 400546
PE anti-mouse CD274 (B7-H1, PD-L1)	Biolegend	Cat#: 124308
PE Rat IgG2b, κ Isotype Ctrl	Biolegend	Cat#: 400608
APC anti-mouse CD86	Biolegend	Cat#: 105012
APC Rat IgG2a, κ Isotype Ctrl	Biolegend	Cat#: 400512
PE/Cy7 anti-mouse Ly-6C	Biolegend	Cat#: 128018
PE/Cy7 Rat IgG2c, κ Isotype Ctrl	Biolegend	Cat#: 400721
PerCP/Cyanine5.5 anti-mouse Ly-6G	Biolegend	Cat#: 127616
PerCP/Cyanine5.5 Rat IgG2a, κ Isotype Ctrl	Biolegend	Cat#: 400532
Pacific Blue™ anti-mouse CD80	Biolegend	Cat#: 104724
Pacific Blue™ Armenian Hamster IgG Isotype Ctrl	Biolegend	Cat#: 400925
Brilliant Violet 711™ anti-mouse I-A/I-E Antibody	Biolegend	Cat#: 107643
Brilliant Violet 711™ Rat IgG2b	Biolegend	Cat#: 400653
PerCP-Cy™5.5 Rat Anti-Mouse Foxp3	BD Pharmingen	Cat#: 563902
Anti-CDK4	Thermo Fisher Scientific	MS-616-P1
InVivoPlus anti-mouse CTLA-4 antibody, clone 9D9	Bio X cell	BP0614
InVivoPlus mouse IgG2b isotype control antibody, clone MPC11	Bio X Cell	BP0086
Anti-human PD-L1 for immunoprecipitation, clone 29E. 12B1	(Chaudhri et al., 2018)	N/A
Phos-PD-L1-S283	Abclonal Technology	N/A
Bacterial and Virus Strains
XL10 Gold Escherichia coli	Agilent	Cat #200314
BL21(DE3) Escherichia coli	Dr. William G. Kaelin, Jr., Dana-Farber Cancer Institute	N/A
Chemicals, Peptides, and Recombinant Proteins
A-769662	Selleck Chemicals	Cat#: S2697
Dorsomorphin (Compound C) 2HCl	Selleck Chemicals	Cat#: S7306
Metformin HCl	Selleck Chemicals	Cat#: S1950
BKM120	Selleck Chemicals	Cat#: S2247
LY294002	Selleck Chemicals	Cat#: S1105
MK2206	Selleck Chemicals	Cat#: S1078
insulin	Invitrogen	Cat#: 41400–045
2-deoxy-D-glucose	Sigma-Aldrich	Cat#: D8375
Cycloheximide	Sigma-Aldrich	Cat#: C7698
Polyethylene glycol	Sigma-Aldrich	Cat#: 202371
Tween 80	Sigma-Aldrich	Cat#: P4780
Streptozotocin	Sigma-Aldrich	Cat#: S0130
AMPK (A1/B1/G2) Kinase	SignalChem	Cat#: P55–10H-10
MG-132	Enzo Life Sciences	Cat#: BML-PI102
Ketogenic diet	Bio-Serv	Cat#: S3666
p-Nitrobenzyl mesylate	Abcam	Cat#: ab138910
ATP-γ-S	Abcam	Cat#: ab138911
Ultra Sensitive Mouse Insulin ELISA Kit	Crystal Chem	Cat#: 90080
Experimental Models: Cell Lines
HEK293	Dr. William G. Kaelin, Jr., Dana-Farber Cancer Institute	N/A
HEK293T	Dr. William G. Kaelin, Jr., Dana-Farber Cancer Institute	N/A
MDA-MB-231	ATCC	CRM-HTB-26
CT26	Dr. Gordon Freeman	N/A
BT-549 PD-L1 KO cells	Dr. Mien-Chie Hung	N/A
WTand Ampk α1−/−Ampk α2−/− MEFs	Dr. Benoit Viollet, Institut Cochin	N/A
wild-type and Cdk4−/− MEFs	Dr. Mitsiadesa	N/A
DLD1	Dr. Bert Vogelstein	N/A
HCT116 PTEN+/+ and PTEN−/− cells	Dr. Todd Waldman	N/A
MEFs Pten+/+ and Pten−/− cells	Dr. Pier Paolo Pandolfi	N/A
B16-F10	Dr. Gordon Freeman	N/A
MC38	Dr. Arlene H. Sharpe	N/A
A375	In our lab	N/A
RAW264.7	Dr. Gordon Freeman	N/A
J774A.1	Dr. Gordon Freeman	N/A
Experimental Models: Organisms/Strains
C57BL/6J	Jackson Laboratory	000664
BALB/c	Jackson Laboratory	000651
NCr nude mice	Taconic	NCRNU-F
Recombinant DNA
PGEX-GST-6P2-PD-L1 truncations	This paper	N/A
pLenti-CMV-Hygro-hPD-L1 WT and mutations	This paper	N/A
Flag-CMTM4	Genescript	OHu07533
Flag-CMTM6	Genescript	OHu24864
HA-Ins-PD-L1	(Zhang et al., 2018a)	N/A
Flag-SPOP	(Zhang et al., 2018a)	N/A
pLenti-CMV-hygro-HA-EZH2 WT and mutants	(Wan et al., 2018)	N/A
PLKO-shEZH2	(Wan et al., 2018)	N/A
pLenti-puro-AMPKα1	This paper	N/A
pcDNA-HA-AMPKβ1	This paper	N/A
pcDNA-HA-AMPKγ1	This paper	N/A
hPD-L1-C-Flag	(Gao et al., 2020)	N/A
pLenti-hPD-L1	(Gao et al., 2020)	N/A
Oligonucleotides
qPCR primers	This paper (Table S1)	N/A
Software and Algorithms
Prism V7	Graphpad	N/A
FlowJo X software V10.0.07	FlowJo	N/A

RNA isolation and real-time RT-PCR analyses

Total RNA from cells was extracted using an RNeasy kit (Qiagen) according to the manufacturer’s instruction. One microgram of RNA was used for reverse transcription with iScript Kit (Bio-Rad). cDNA was then diluted and used for real-time PCR with gene-specific primers using SYBR Select Master Mix (Thermo Fisher Scientific). Relative abundance of mRNA was calculated by normalization to GAPDH mRNA. Primer sequences were listed in key resource table.

RNA-seq data analysis and Gene Set Enrichment Analysis (GSEA)

Our previously published RNA-seq dataset (GSE97735) comparing Ampkα knockout with wild type in mouse embryonic fibroblast cells were used (Wan et al., 2018). DESeq2 1.14.1 was used to call differentially expressed genes (Love et al.). The log₂ fold change values of all the genes were then used to perform a pre-ranked GSEA analysis by using JAVA GSEA 2.2.0 program by searching against the hallmark gene set from the Broad Molecular Signatures Database (MsigDB) (Liberzon et al., 2011; Subramanian et al., 2005). The gene symbols in hallmark gene set were converted to mouse orthologues using BioMart (Durinck et al., 2009).

Mouse fasting experiment

6 weeks old C57BL/6J male mice were randomly assigned to standard diet and fasting groups. All animal studies were approved by the Beth Israel Deaconess Medical Center (BIDMC) Institutional Animal Care and Use Committee (IACUC: Protocol #043-2015). Mice were fasted for 24 h or fed with standard diet. Livers from these animals were collected and subjected to the indicated experiments.

STZ induced diabetic model

8 weeks old C57BL/6J male mice were randomly divided into 2 main groups: diabetic and non-diabetic. The diabetic group received a single dose of streptozotocin (1% w/v solution in fresh cold sodium citrate buffer, pH 4.5) injection into the abdominal cavity, with the dosage ranging from 180–200 mg/kg of body weight.

Treatment of wild-type mice with ketogenic diet

The animal experiments were performed in accordance with protocol approved by the Beth Israel Deaconess Medical Center (BIDMC) Institutional Animal Care and Use Committee (IACUC: Protocol #043-2015). Mice were maintained in temperature- and humidity-controlled specific pathogen-free conditions on a 12 h light/dark cycle, and performed in accordance with guidelines established by NIH Guide for the care and use of laboratory animals. 6-week old BALB/cJ female mice (Jackson Laboratory) received a normal chow diet (Formulab Diet 5008* Formulab Diet, Irradiated 5008C33) or ketogenic diet (Bio-Serv (S3666)) with free access to drinking water. Serum insulin levels were quantified by ELISA Kit (Crystal Chem Cat#: 90080).

In vivo experimental therapy in CT26 mice tumor models

Animal studies were approved by the Beth Israel Deaconess Medical Center (BIDMC) Institutional Animal Care and Use Committee (IACUC: Protocol #043-2015) or Dana-Farber Cancer Institute Institutional Animal Care and Use Committee (IACUC; protocol number 04-047). All mice were housed in a pathogen-free environment at the animal facility in BIDMC or Dana-Farber Cancer Institute, and performed in accordance with guidelines established by NIH Guide for the care and use of laboratory animals. CT26 WT, Pd-l1 KO or Ampkα KO tumors were established by subcutaneously injecting 1×10⁵ tumor cells in 100 μl HBSS into the right flank of 6-week old BALB/cJ female mice (Jackson Laboratory) or NCRNU-F nude mice (TACONIC). Tumor sizes were measured every three days by caliper after implantation and tumor volume was calculated by length × width² × 0.5. Seven days after tumor cells were injected, animals were pooled and randomly divided into designated experimental groups with comparable average tumor size. Laboratory members who measured the mice were blinded to the treatment groups.

For ketogenic/anti-CTLA-4 combination treatment, mice were grouped into control antibody treatment, anti-CTLA-4 antibody treatment (clone 9D9), ketogenic treatment, and anti-CTLA-4 antibody plus ketogenic treatment. As illustrated in Figure S2A, control and anti-CTLA-4 antibody treatments were conducted by intraperitoneal injection (200 μg per mouse in 200 μl HBSS saline buffer) every three days for a total of 3 injections. Ketogenic diet was administered until end point.

For A-769662/anti-CTLA-4 combination treatment, mice were grouped into control antibody treatment, anti-CTLA-4 antibody treatment (clone 9D9), A-769662 treatment, and anti-CTLA-4 antibody plus A-769662 treatment. As illustrated in Figure S7E, control and anti-CTLA-4 antibody treatments were conducted by intraperitoneal injection (200 μg per mouse in 200 μl HBSS saline buffer) every three days for a total of 3 injections. The A-769662 treatments were conducted by intraperitoneal injection (400 μg per mouse in 200 μl 5% DMSO + 10% polyethylene glycol + 1% Tween 80 in PBS buffer) every day for two weeks.

For survival studies, BALB/c mice were monitored for tumor volumes every three days after initial treatment and NCRNU-F nude mice were monitored every two days, until tumor volume exceeded 1,500 mm³, or until tumor had an ulcer with a diameter of 1 cm. Statistical analysis was conducted using the GraphPad Prism software (GraphPad Software). Kaplan–Meier curves and corresponding Gehan-Breslow-Wilcoxon tests were used to evaluate the statistical differences between groups in survival studies. P < 0.05 was considered significant. For tumor-infiltrating lymphocytes isolation, tumors were harvested on day 16 post tumor cell injection.

QUANTIFICATION AND STATISTICAL ANALYSIS

The majority of experiments were repeated two or three times. All quantitative data were presented as mean ±SD, as indicated of at least three independent experiments or biological replicates by Student’s t test among group differences. Values of P < 0.05 was considered statistically significant.

Highlights.

• Energy stress or ketogenic diet treatment decreases PD-L1 protein abundance

• AMPK phosphorylates PD-L1 at Ser283 to disrupt its interaction with CMTM4

• AMPK enhances IFNs and antigen presentation genes expression via repressing PRC2

• AMPK agonist or ketogenic diet enhances the efficacy of anti-CTLA-4 immunotherapy