The human ESCC cell lines KYSE-450 and TE-7 were purchased from the Type Culture Collection of the Chinese Academy of Sciences (Shanghai, China). All the cell lines were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) and antibiotics (10,000 U/ml penicillin and 10 μg/ml streptomycin). All cells were maintained in a humidified incubator at 37°C with 5% carbon dioxide. Drugs, including metformin, IL-6, Stattic and Compound C, were purchased from MedChemExpress (MCE).

shRNA constructed to target JAK2 was inserted into the GV493 vector provided by GenePharma (Shanghai, China) ( Supplementary Table 1 ). All plasmids were added with pHelper 1.0 and pHelper 2.0 plasmids into the supernatant of revived 293T cell culture ( Supplementary Figure 1 ). Growth medium was removed after 24 hours of incubation, and fresh DMEM-high glucose (Sigma, USA) with 10% FBS (Dakewe, New Zealand) was added. Then, 293T cells were cultured for another 48 hours. The corresponding lentiviruses were obtained after ultracentrifugation of the collected supernatant for 4 hours. KYSE450 cells were transfected with virus cloned with a stable JAK2 knockdown sequence. The above cells were then purified with 1.5 µg/mL puromycin (Solarbio) for 2 weeks.

Cells were harvested and suspended in RIPA lysis buffer (Thermo Scientific™, 89901) containing a protease inhibitor cocktail (Thermo Scientific™, 78437) and the phosphatase inhibitor cocktail PhosSTOP (Cell Signaling Technology, 5872). After incubation on ice for 30 minutes, the cell lysate was centrifuged at 12,000 rpm for 20 minutes at 4°C. The protein content of the supernatant was quantified using Thermo Pierce™ BCA Protein Assay Reagent (Thermo Scientific™, 23228). A total of 30 μg protein per well was separated by 10% sodium dodecyl sulfate–polyacrylamide gel electrophoresis and transferred to polyvinylidene difluoride membranes (Thermo Scientific™, 0.2 μm, LC2002). The membranes were immunoblotted with primary antibodies against PD-L1 (Cell Signaling Technology, 13684), phospho-JAK2 (Tyr1007/1008, Wanlei, WL02997), JAK2 (Protein Group, Wuhan, 17670-1-AP), phospho-STAT3 (Y705, Abcam, 76315), STAT3 (Protein Group, Wuhan, 10253-2-AP), phospho-ERK1/2 (Thr202/Tyr204, Cell Signaling Technology, 4370), ERK1/2 (Abcam, 17942), phospho-AMPK (Thr172, Abcam, 131357), AMPK (Protein Group, Wuhan, 66536-1-Ig), phospho-ACC (Ser473, Cell Signaling Technology, 3661), and ACC (Protein Group, Wuhan, 67373-1-Ig) overnight at 4°C. After washing three times, the membranes were further incubated with horseradish peroxidase-conjugated goat anti-rabbit (Abcam, 150077) or goat anti-mouse (Abcam, 150117) secondary antibodies (1:2000, Santa Cruz, CA) at room temperature for 2 hours. Signals were detected by Supersignal West Dura Luminol/Enhancer solution.

To quantify PD-L1 mRNA expression, total RNA was extracted from cells with TRIzol reagent (Invitrogen), and cDNA was synthesized using TaqMan MultiScribe Reverse Transcriptase (SuperScript™ IV Reverse Transcriptase, Thermo Fish) according to the manufacturer’s instructions. Quantitative real-time PCR (RT-PCR) analysis was performed using an ABI Prism 7900-HT Sequence Detection System (96-well, Applied Biosystems). For RT-PCR, the following primers were used for amplification: PD-L1, forward primer 5′-TGGCATTTGCTGAACGCATTT-3′ and reverse primer 5′-TGCAGCCAGGTCTAATTGTTTT-3′; and GAPDH, 5′-GGAGCGAGATCCCTCCAAAAT-3′ (forward) and 5′-GGCTGTTGTCATACTTCTCATGG-3′ (reverse). The experiments were performed in triplicate. The relative expression of PD-L1 was normalized to GAPDH expression.

Immunohistochemical analysis of PD-L1 in 4-μm paraffin-embedded archived ESCC tissue biopsy specimens was performed in accordance with a standard protocol using commercial antibodies. After deparaffinization and dehydration, sections were placed in 10 mM sodium citrate buffer (pH = 6.0), autoclaved at 121°C for 10 minutes, and then incubated in normal goat serum (KL-D1418; KALANG, Zhengzhou, China) for 30 minutes to block nonspecific antibody binding sites. Each section was then incubated for 10 minutes at room temperature with an anti-PD-L1 rabbit monoclonal antibody (Abcam, 228415) diluted 1:500 in phosphate-buffered saline (PBS) containing 1% bovine serum albumin. PBS was used as a negative control.

The Human IL-6 ELISA Kit and Human IL-2 ELISA Kit were purchased from Protein Group, Wuhan. The levels of IL-6 and IL-2 in tissue lysates were measured according to the manufacturers’ protocols. Absorbance at 450 nm was measured using a microplate reader (Bio-Rad, Hercules, CA, USA).

First, human peripheral blood mononuclear cells (PBMCs) were isolated from peripheral blood donated by healthy volunteers by Ficoll-Paque density centrifugation. KYSE-450 cells were treated with metformin, IL-6 or PBS for 24 hours. PBMCs were cultured in RPMI medium containing 10% FBS and phytohemagglutinin (PHA) (Sigma) at 0.1 μg/mL as indicated for 3 days, followed by stimulation with anti-CD3/CD28 antibodies (100 ng/mL) to activate T cells (PBMCs). Then, the activated PBMCs were added to the coculture system containing KYSE-450 cells at a ratio of 2:1. After 24 hours of coculture, the wells were washed with PBS 3 times to remove T cells. The T cells were then collected and lysed. ERK phosphorylation was evaluated by Western blotting. The cell-free supernatant from the coculture system was collected by centrifugation at 12,000 rpm for IL-2 analysis by ELISA.

T cells (PBMCs) were activated with ImmunoCult™ Human CD3/CD28 T Cell Activator (100 ng/mL; STEMCELL Technologies, 19071) and IL-2 (10 ng/mL; Abcam, 174444). After 4 days of coculture of tumor cells and T cells with metformin, IL-6 or PBS in 12-well plates, the wells were washed with PBS 3 times to remove the T cells. The surviving tumor cells were fixed and stained with a crystal violet solution (Sigma, V5265). The dried plates were scanned, stained, solubilized in 100 μL 33% glacial acetic acid, and shaken until the color was uniform. The absorbance at 450 nm was measured using a spectrometer.

Male NPIdKO™ mice (NOD-Prkdc^scid-Il2^rgem1IDMO) (6 weeks old) were purchased from Beijing IDMO Company, and all animal experimental procedures were conducted with the approval of the Animal Ethics Committee of The First Affiliated Hospital of Zhengzhou University. NPIdKO™ mice are immunodeficient with NPI™ as the genetic background, resulting from the knockout of the MHC I molecule β2-microglobulin (β2M) gene and MHC II molecule IAβ gene. These mice lack mature T, B and natural killer (NK) cells, produce no immunoglobulin; and have dendritic cells (DCs) with abnormal functions.

PBMCs from healthy donors were diluted to a ratio of 1:2 with PBS, and mononuclear cells were isolated by using density-gradient centrifugation with Ficoll medium. Approximately 1 × 10⁵ freshly isolated PBMCs were injected intravenously into NPIdKO™ mice. Then, the mice were randomized into four treatment groups in all of the experiments (each group contained 5 animals).

The TE-7-luc cell line (TE-7 cells labeled with luciferase) was kindly provided by Dr. Ruizhe Li (Department of Pathology, The First Affiliated Hospital of Zhengzhou University, China). Fifteen days after PBMCs were injected, TE-7-luc cells (1 × 10⁶ cells) in 100 μl medium were subcutaneously administered into the right flank of NPIdKO™ mice.

Treatments began when the tumor size was 100 mm³. The mice in different groups were administered metformin (35 mg/kg/d, diluted in 0.1 mL normal saline), anti-PD-1 (camrelizumab, Hengrui, China) monoclonal antibodies (200 μg/3 days/mouse, diluted in 0.1 mL normal saline), combination treatment with metformin and the anti-PD-1 antibodies, or control treatment (0.1 mL normal saline/day).

Tumor size was measured every 3 days for 4 weeks with a caliper, and tumor volume was calculated using the following formula: width × width × length/2. Tumor volume was also measured by quantifying bioluminescence intensity with a small-animal in vivo imaging system (IVIS 200; Caliper Life Sciences). At the endpoint of experiments, defined as whichever came first among a tumor volume greater than 1,000 mm³, tumor ulceration, or study end, the mice were euthanized, and the tumors were harvested and weighed quickly. Then, the tumors were fixed in neutral buffered formalin and embedded in paraffin for immunohistochemical analyses.

The Cancer Genome Atlas- Esophageal Cancer (TCGA-EC) RNAseqV2 gene expression data and clinical data were obtained from the TCGA. Data Portal Transcriptional values were Log2-transformed from the normalized fragments per kilobase transcript per million mapped reads values using R package “limma” in R 3.6.0. We divided patients into different groups by the medians of IL-6 genes expression. Kaplan-Meier analysis was performed for survival curves and the significance was determined by log-rank, which was completed by R software. The associations between IL-6 levels and clinicopathologic characteristics of patients were analyzed using one-way analysis of variance (ANOVA). The correlation between IL-6 and PD-L1 levels was analyzed using the Pearson correlation coefficient.

Representative results from three independent experiments are shown in the present study. Numerical data are presented as the mean ± standard deviation of the mean (SD). The p-values between two experimental groups were determined by a two-tailed Student’s t test, and p-values less than 0.05 were considered significant.

We downloaded IL-6 and PD-L1 gene expression information and the corresponding clinical data for esophageal cancer from TCGA database (https://cancergenome.nih.gov). Altogether, 160 esophageal cancers from TCGA with normalized gene expression and specific clinical status were collected and analyzed. For the transcriptional data for IL-6 in esophageal cancer tissue, the median value was selected as the cutoff point, and the patients were divided into high and low IL-6 expression groups. The results showed that high expression of IL-6 was associated with a poor prognosis (P=0.022) ( Figure 1A ). Combined with tumor stage data, the results suggested that IL-6 expression was significantly decreased in low-stage tissues compared with higher-stage tissues ( Figure 1B ). Furthermore, the correlation between IL-6 and PD-L1 gene expression was evaluated, and a positive correlation was found between PD-L1 and IL-6 expression in esophageal cancer (R=0.34, P<0.01) ( Figure 1C ).

To investigate the role of metformin in esophageal cancer, we compared the antitumor effects of metformin on ESCC cell lines, including KYSE-450 and TE-7. Interestingly, we found that metformin treatment significantly decreased the expression of IL-6 ( Figure 2A ). Based on the related results for the TCGA data, we next examined the effect of metformin on PD-L1 and found that metformin could significantly inhibit the expression of PD-L1 in a dose-dependent manner. The effect could be reversed by adding exogenous IL-6, which suggests that the downregulation of PD-L1 induced by metformin is associated with the IL-6 signaling pathway ( Figures 2B, C and Supplementary Figure 2A ). The involvement of the JAK2/STAT3 signaling pathway in IL-6 signaling has been shown by many results (7, 21, 22). In our study, metformin alone effectively downregulated JAK2 and STAT3 activation and the PD-L1 expression level in the two cell lines. To expand on this result, we also examined the role of IL-6 in PD-L1 expression and found that IL-6 significantly activated the JAK2/STAT3 signaling pathway and increased the level of PD-L1 expression ( Figure 2D ). Therefore, we speculated that the reduction in PD-L1 expression induced by metformin was related to the IL-6/JAK2/STAT3 signaling pathway.

A mutant KYSE450 (JAK2 KD) cell line was prepared by knocking out the JAK2 gene in KYSE-450 cells ( Supplementary Figure 2B, C ). The expression of p-STAT3 and PD-L1 in the KYSE450 (JAK2 KD) cell line was significantly reduced compared with that in the KYSE450 (JAK2-NC) cell line. The regulatory effects of metformin and IL-6 on PD-L1 disappeared when the JAK2 gene was knocked down ( Figure 3A ).

To further verify that p-STAT3 can regulate the expression of PD-L1, we pretreated KYSE450 cells with Stattic, which could inhibit STAT3 phosphorylation. The regulatory effects of metformin and IL-6 on PD-L1 disappeared when STAT3 phosphorylation was inhibited ( Figure 3B ). In brief, metformin downregulated PD-L1 expression through the JAK2/STAT3 signaling pathway.

Metformin is a well-known AMP-activated protein kinase (AMPK) activator. In our study, we investigated whether the AMPK signaling pathway is involved in the downregulation of PD-L1 expression induced by metformin. The levels of phosphorylated AMPK and acetyl-CoA carboxylase (ACC), an AMPK target, were both decreased in the AMPK inhibitor Compound C pretreatment group ( Figure 4A ) and AMPKα 1/2-specific siRNA-transfected group ( Figure 4B ), demonstrating the loss of AMPK activity. Notably, metformin still decreased PD-L1 expression by inhibiting JAK2/STAT3 signaling. This result indicated that metformin downregulated PD-L1 expression in a manner independent of the AMPK pathway.

It is well known that PD-L1 in tumor cells can lead to inactivation of T cells. First, human PBMCs were isolated from peripheral blood donated by healthy volunteers by Ficoll-Paque density centrifugation. Then, PBMCs were added to a coculture system containing KYSE-450 cells at a ratio of 2:1. To evaluate whether the downregulation of PD-L1 mediated by metformin could promote the CD3/CD28-triggered T cell activation signaling pathway, we generated T cell blasts by treating PBMCs with PHA to induce PD-1 expression. The results showed that metformin markedly promoted CD3/CD28-induced ERK phosphorylation in a dose-dependent manner ( Figure 5A ). The coculture supernatant was collected for IL-2 measurement. We found that IL-2 was markedly increased in the supernatant of the KYSE450/PBMC coculture system after metformin treatment ( Figure 5B ). Moreover, IL-6 inhibited CD3/CD28-induced ERK phosphorylation and PHA-induced IL-2 secretion, both of which are indicators of T cell activation. Next, we performed a T cell-mediated tumor cell-killing assay to examine the influence on T cell function. The results showed that metformin-induced PD-L1 downregulation protected the cytotoxic function of T cells, while this effect was reversed by IL-6 ( Figure 5C ).

NPIdKO™ mice were subcutaneously implanted with TE-7 cells, followed by administration of metformin and an anti-PD-1 antibody as a monotherapy or combination therapy ( Figure 6A ). The anti-PD-1 antibody or metformin alone exhibited anticancer activity, and metformin significantly enhanced the efficacy of combination therapy in TE-7 cell-harboring mice, which was reflected by a tumor size assay and small-animal in vivo imaging system ( Figures 6B–E ). Consistent with the aforementioned mechanistic findings, immunohistochemical staining of tumor tissues showed that metformin reduced PD-L1 levels. In addition, metformin and anti-PD-1 combination therapy produced a significant reduction in the tumor expression of PD-L1 ( Figure 6F ). In this study, we did not observe any abnormalities in mouse body weight or behavior.