Relevant animal experiments were carried out in accordance with the approval of the ethics committee of the First Affiliated Hospital of Soochow University. All experiments are taken in accordance with the standard biosecurity and safety procedures of the First Affiliated Hospital of Soochow University.

Human normal esophageal epithelial Het-1A cell line and human ESCC cell lines (ECA-109 and KYSE450) were obtained from American Type Culture Collection (ATCC, Manassas, VA, USA). The luciferase-labeled ECA-109-luc and KYSE450-luc cells were obtained from Shanghai Chenyishiye Co., Ltd (Shanghai, China). The above cells were cultivated in Dulbecco’s modified Eagle’s medium (DMEM) (MD207-050, Gibco-BRL/Invitrogen, Waltham, MA, USA) supplemented with 10% fetal bovine serum (FBS; 16000-044, Gibco, USA) and 100× penicillin/streptomycin solution (15140122, Gibco-BRL/Invitrogen, USA). Human embryonic lung fibroblast HFL1 cell line was obtained from BeNa Culture Collection (China) and cultured with 90% F-12K plus 10% FBS.

In line with the instruction of TRIzol reagent (T9108, Takara, Japan), the isolation of total RNA samples was extracted in ECA-109 and KYSE450 cells. Synthesis of complementary DNA (cDNA) for miRNAs was carried out using TaqMan™ MicroRNA Reverse Transcription Kit (TaKaRa, Kusatsu, Japan), and the cDNA for mRNAs was synthesized by PrimeScript RT reagent Kit (TaKaRa, Japan). RT-qPCR reaction was achieved with qRT-PCR Kit (QR0100-1KT, Sigma-Aldrich, St. Louis, MO, USA) followed by 2^−ΔΔCt method. In relevant assays, the reference gene was β-actin while the reference gene for miRNA quantification was U6 and miR-16 (in exosome experiments). Related sequence information is included in Supplementary Table S1 .

Specific shRNAs targeting FOXM1 (sh1/2/3-FOXM1) and HER2 (sh1/2/3-HER2), together with their negative control shRNA (sh-NC) were provided by RiboBio Co., Ltd. (Guangzhou, China) to silence FOXM1 expression. For the overexpression of FOXM1 and HER2, the whole sequences were synthesized and subcloned into pcDNA3.1 vector to construct pcDNA3.1-FOXM1 and pcDNA3.1-HER2, with pcDNA3.1 empty vector as the negative control (NC). For the overexpression or silencing of miR-620, miR-620 mimics or miR-620 inhibitor was respectively used, with mimics-NC or inhibitor-NC as internal controls. Transfections were conducted with Lipofectamine 2000 (11668019, Invitrogen, USA) in line with the supplier’s protocols. After incubation for 48 h, cells were collected for subsequent experiments.

Cell viability was evaluated with the application of Cell Counting Kit-8 (CCK-8) according to the manufacturer’s guidelines. Transfected ECA-109 or KYSE450 cells were seeded to 96-well plates and then placed in an incubator with 5% CO₂ at 37°C for 24 h. Next, 10 μl of CCK-8 solution (Dojindo, Kumamoto, Japan) was added into each pore of the plate for 1-h incubation. Finally, the absorbance at 450 nm after 24, 48, or 72 h was measured with the microplate reader to evaluate the viability of indicated cells.

ECA-109 and KYSE450 cells were seeded into the 6-well plates and then cultivated at 37°C with 5% CO₂ for confluence. When the degree of cell confluence was above 90%, a pipet tip was used to scrape the cells as vertically as possible and then the cells were washed with PBS for three times. The scratched cells were removed, the serum-free medium was added, and then the cells were placed into an incubator at 37°C with 5% CO₂. Finally, the scratch was imaged by microscope at 0 and 24 h for analysis.

ESCC cells were planted on the top of 24-well Transwell chambers coated with Matrigel for invasion assay or without Matrigel for migration assay. The lower chambers were loaded with complete medium. Twenty-four hours later, cells in the upper layer were removed by a cotton swab and the bottom of the chamber was fixed in methanol solution for 15 min. Crystal violet was adopted to stain the membranes for 10 min, and the invaded or migrated cells were observed and counted under a microscope (10 × 10).

When the cell confluence was above 80%, the protein extracts were collected from ESCC cell lines using protein extraction kit (PROTTOT-1KT, Sigma-Aldrich, USA) and RIPA buffer (KGP701, KeyGEN BioTECH, Nanjing, China). After being separated through sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE; P0670-250ml, Beyotime Biotech, Shanghai, China), proteins were transferred to polyvinylidene fluoride (PVDF) membranes and cultured in 5% skim milk. The membranes were cultivated with primary antibodies over night at 4°C, followed by being cultivated with secondary antibody for 1 h. After washing in TBST, the secondary antibodies were added and finally assayed by ECL substrate.

The primary antibodies were obtained from Abcam (UK) and listed as follows: anti-E-cadherin (ab40772), anti-Vimentin (ab92547), anti-CD63 (ab1318), anti-CD81, anti-HSP70 and anti-FOXM1 (ab245309), anti-GM130, with anti-β-Actin (ab8227) as the internal control. Experiment was conducted three times.

With the application of Pierce™ Streptavidin Magnetic Beads (Thermo Fisher, USA), RNA pull-down assay was carried out in ECA-109 cells. Experimental groups were divided into Input Antisense (control), HER2 Promoter Sense, and HER2 Promoter (mut) Sense groups. Biotinylated HER2 probes were incubated with cell extracts and streptavidin magnetic beads and then RNAs were purified with TRIzol reagent. The enrichment of FOXM1 was analyzed by Western blot. Likewise, biotinylated FOXM1 probes were incubated with cell extracts and streptavidin magnetic beads, and, finally, the enrichment of miR-620 was examined by qRT-PCR. The experiment was independently conducted in triplicate.

For luciferase reporter assay, FOXM1 3′UTR possessing wild-type and mutant miR-620 binding sites were subcloned into pmirGLO luciferase vectors to obtain pmirGLO-FOXM1 (3′UTR) and pmirGLO-FOXM1 (3′UTR-Mut). MiR-620 mimics or mimics NC were cotransfected with pmirGLO-FOXM1 (3′UTR) and pmirGLO-FOXM1 (3′UTR-Mut) into NSCLC cells. After 48-h transfection, cells were extracted and the luciferase activities were analyzed utilizing the luciferase reporter assay system (Promega).

As for FOXM1 and HER2 promoter luciferase reporter assay, the HER2 promoter was constructed into pGL3 luciferase vectors and then cotransfected with pcDNA3.1 or pcDNA3.1-FOXM1 into 293T cells. The luciferase activities were analyzed utilizing the luciferase reporter assay system (Promega).

By referring to related document, we conducted the exosome uptake experiment (22). A total of 20 ml of culture media (1 × 10 (7) cells) were collected on ice, centrifuged at 800×g for 10 min, and then centrifuged at 12,000×g for 30 min to remove the cellular debris. Exosomes were centrifuged at 100,000×g for 2 h in a SW32 rotor (Beckman Coulter) for separating from the supernatant. The exosome pellet was washed once in a large volume of PBS and then resuspended in 100 μl of PBS. Finally, exosomes were then identified by transmission electron microscope (TEM) to observe the morphologies of exosomes secreted from miR-620 (labeled as miR-620-Exo).

In addition, the exosomes were marked by PKH67 Fluorescent Cell linker kits according to previous guidance (23). Briefly, ECA-109 and KYSE450 cells were seeded into 24-well plates and incubated at 37°C, with 5% CO₂. The exosomes marked by PKH67 as well as ESCC cells were cocultured without light for 12 h and washed with PBS for three times and then fixed by paraformaldehyde for 20–30 min, rinsed by PBS for three times; the nuclei were stained by DAPI for 5 min, rinsed by PBS for three times, and fixed. Finally, the distribution of fluorescence was observed by a laser scanning microscope.

As the literature on the construction of orthotopic esophageal xenograft tumor has not been retrieved, we referred to the construction method of orthotopic breast xenograft tumor and conducted the following experiment 18. NSG mice was injected intracardiac with 2×105 luciferase-labelled KYSE450-luc cells combined with Matrigel (BD Biosciences; San Jose, CA) in a 1:1 ratio. Starting from day 3 after cancer cell transplantation, oligos (25 mg/kg) were intraperitoneally (i.p.) injected daily for 5 days and then twice weekly until the end of experiment. The mice were sacrificed after four weeks and tumour volume (mm3) was assessed by calliper measurements using the formula (width2 × length)/2.

To verify the effect of exosomal miR-620 on the aerobic glycolysis of HFL1 cells, in vivo experiments were conducted as previously described (20).

The female mice (6–8-week-old) were acquired commercially from the Institute of Zoology, Nanjing University. All the relevant animal experiments were carried out in accordance with the Animal Care and Use Committee guidelines of the First Affiliated Hospital of Soochow University. We intravenously injected exosomes containing low or high levels of miR-620 into mice and measured the aerobic glycolysis change in MLF. Female NOD/SCID/IL2Rγ-null (NSG) mice aged 6–8 weeks were used in this study. Exosomes were isolated from ECA-109 and KYSE450 cells, resuspended in PBS, and followed by centrifugation at 16,000×g for 10 min at 4°C. The supernatant was then transferred to a new tube for mouse injection biweekly for 3.5 weeks (~6 μg/injection).

All data from experiments including three biological replications were exhibited as the mean ± standard deviation (SD). Data analysis was achieved by Student’s t-test (comparison for two groups) and one-way/two-way ANOVA (comparison for more than two groups), applying SPSS 19.0 software (IBM SPSS, Armonk, NY, USA). The statistical significance in differences were confirmed when p < 0.05.

With the application of the GEO database (GSE122497), serum microRNA profiles of 5,531 samples which consist of 566 esophageal squamous cell carcinoma and 4,965 of noncancer controls were exhibited. To make further selection, we searched for relevant literature to find the miRNAs associated with aerobic glycolysis. Finally, a serum miRNA named miR-620 was sifted out ( Supplementary Figure S1A ). It has been proven to promote cancer development, but how it may function in ESCC has not been identified. The role of miR-620 has been revealed in various cancers, but how it may function in ESCC has not been identified. It was observed through CCK-8 assay that the viability of ECA-109 and KYSE450 cells was suppressed by miR-620 overexpression ( Figure 1A ). Next, it was shown from wound-healing assay that the wound width was increased upon miR-620 overexpression, which indicated that miR-620 overexpression inhibited the proliferation of ESCC cells ( Figure 1B ). Results of Transwell assay manifested that the enhanced miR-620 expression inhibited the migration and invasion of ECA-109 and KYSE450 cells ( Figure 1C ), and fewer spindle-shaped ESCC cells were observed through microscope upon miR-620 overexpression ( Figure 1D ). Finally, it was seen from Western blot that after miR-620 was upregulated in ECA-109 and KYSE450 cells, the protein expression of E-cadherin enhanced while that of Vimentin declined, indicating that miR-620 overexpression suppressed the EMT process in ESCC ( Figure 1E ). The above data demonstrated that miR-620 suppressed the malignant progression of ESCC cells. Furthermore, we designed the interference sequence of miR-620 and conducted related assays to verify the influence of miR-620 silencing on ESCC cell progression. As shown in Supplementary Figures S1B–F , decreased miR-620 expression accelerated the proliferation, migration, invasion, and EMT process of ESCC cells, which further manifested the oncogenic property of miR-620 in ESCC.

Energy metabolism reprogramming has become an emerging hotspot in cancer research. In most of the cancers, even with abundant oxygen, cancer cells will promote the uptake and absorption of nutrients needed by newborn cells through aerobic glycolysis and reduce oxidative phosphorylation of mitochondria (16). Combining with the information above, we decided to explore whether miR-620 may regulate the aerobic glycolysis of ESCC cells. With the application of glucose determination kit, lactic acid determination kit and pyruvate kinase activity assay kit, it was verified that the glucose consumption, total lactate protein, and relative PK catalytic activity were reduced by miR-620 overexpression ( Figures 2A –C ). It was then measured through Seahorse XF Extracellular Flux Analyzers that the ECAR/OCR ratio was declined in ESCC cells transfected with miR-620 mimics ( Figure 2D ). By comparison, we analyzed the effect of miR-620 knockdown on the aerobic glycolysis of ESCC cells, and opposite results were exhibited ( Supplementary Figures S2A–D ). The above findings suggested that miR-620 inhibited the aerobic glycolysis of ESCC cells. For further verification, we treated ESCC cells with 50 or 100 μM 2-DG (the glycolytic inhibitor) for 48 h (24), and it was then discovered that the glucose consumption, total lactate protein, relative PK catalytic activity, and the ECAR/OCR ratio were all reduced upon 2-DG treatment ( Figures 2E –H ), which demonstrated the effective inhibition of 2-DG on ESCC aerobic glycolysis. Moreover, Supplementary Figures S2E–H shows that adding 2-DG could reverse the promoted aerobic glycolysis caused by miR-620 inhibitor, which further manifested that miR-620 regulated the aerobic glycolysis of ESCC cells.

After we confirmed that miR-620 could regulate the aerobic glycolysis of ESCC cells, we continued to probe into the specific molecular mechanism. Through the GEO database (GSE75241), we selected the top 30 genes upregulated in ESCC, and then potential mRNAs which could bind to miR-620 were selected through starBase database. As a result, 8 candidates were sifted out, which were ITM2C, FNDC38, CD276, LAMC2, LHFPL2, LAMC1, PARVB, and FOXM1 ( Supplementary Figures S3A, B ). Through qRT-PCR detection, it was found that only FOXM1 was significantly reduced by miR-620 overexpression in ECA-109 cells ( Supplementary Figure S3C ). Therefore, we chose FOXM1 for further investigations. The result of RNA pulldown assay with qRT-PCR showed that the enrichment of miR-620 was enhanced in the wild-type Bio-FOXM1 (3′UTR) group rather than in the mutant group ( Figure 3A ). Subsequently, it was observed through luciferase reporter assay that the luciferase activity of HEK293T cells transfected with miR-620 mimics was declined in the pmirGLO-FOXM1 (3′UTR) group while on obvious change was seen in the corresponding mutant group ( Figure 3B ). We utilized 50 mM α-amanitin to treat ESCC cells, and then it was discovered that the upregulation of miR-620 inhibited the stability of FOXM1 mRNA in ECA-109 cells ( Figure 3C ). The above findings manifested that miR-620 combined with FOXM1 3′UTR region to decline FOXM1 mRNA stability in ESCC cells. In order to figure out the downstream target gene of FOXM1, KEGG (https://www.genome.jp/kegg/) was applied and potential 6 genes related to aerobic glycolysis were detected ( Supplementary Figure S3D ), and then it was verified through hTFtarget database (http://bioinfo.life.hust.edu.cn/hTFtarget#!/) that only HER2 was forecast to be the target gene of FOXM1 ( Supplementary Figure S3E ). Therefore, HER2 was chosen for the next-step researches. It was shown from qRT-PCR assay that HER2 expression was suppressed upon sh1/2-FOXM1 transfection in ECA-109 cells, as evidenced by qRT-PCR assay ( Figure 3D ). In addition, FOXM1 expression was enhanced by transfecting pcDNA3.1-FOXM1 in ESCC cells ( Supplementary Figure S4A ), and then the result of luciferase reporter assay manifested that HEK293T cells displayed an elevated luciferase activity in pGL3-HER2 promoter cotransfected with pcDNA3.1-FOXM1, while the corresponding mutant group exhibited no obvious change ( Figure 3E ). Through RNA pulldown assay with Western blot, it was shown that FOXM1 was pulled down by the HER2 promoter sense ( Figure 3F ). Finally, qRT-PCR data showed that the addition of pcDNA3.1-FOXM1 could normalize the reduced HER2 expression caused by miR-620 overexpression in ECA-109 cells ( Figure 3G ). To conclude, FOXM1 combined with the HER2 promoter so as to transcriptionally activate HER2 expression in ESCC cells.

In this part, we tried to verify the regulatory mechanism of FOXM1 and HER2 on the aerobic glycolysis of ESCC cells. As shown in Figure 4A–D , the aerobic glycolysis of ESCC cells was suppressed after FOXM1 was knocked down, as the glucose consumption, total lactate protein, relative PK catalytic activity, and the ECAR/OCR ratio were all declined. Same results were observed in the sh1/2-HER2 transfection group in ESCC cells ( Figures 4E –I ). By comparison, the effect of FOXM1 and HER2 overexpression on the aerobic glycolysis of ESCC cells was respectively verified in Supplementary Figures S4B–J , which further manifested that both FOXM1 and HER2 could effectively promote the aerobic glycolysis of ESCC cells. After that, experimental groups were divided into sh-NC, sh1-FOXM1, sh1-FOXM1+pcDNA3.1, and sh1-FOXM1+pcDNA3.1-HER2 for rescue assays. As shown in Figure 4J , the glucose consumption of ESCC cells was inhibited by FOXM1 silencing, while such effect was offset by the cotransfection of pcDNA3.1-HER2. Same opposite results were observed in the total lactate protein, relative PK catalytic activity, and the ECAR/OCR ratio of ESCC cells in different transfection groups ( Figures 4K –M ). In conclusion, FOXM1 promoted the aerobic glycolysis of ESCC cells via upregulating HER2 expression.

According to the qRT-PCR data in Fig. S5A , the expression level of miR-620 in the supernatant of ESCC cells was lower than that of normal Het-1A cell, whereas the intracellular expression of miR-620 in ESCC cells was higher than that that in normal Het-1A cell. Previous study has reported that miRNAs can be secreted to the extracellular environment by membrane-enclosed vesicles (such as exosomes) (6). Therefore, we tried to explore whether miR-620 may play its regulatory role through exosomes. At first, we observed through electron microscope that the exosomes secreted from miR-620-overexpressed ESCC cells (named miR-620-Exo) was 40–100 nm in diameter with oval-shaped vesicles, which conformed to the basic features of exosomes ( Figure 5A ; Supplementary Figure S5B ). Meanwhile, the expression of exosome-specific marker proteins (CD63, CD81, and HSP70) was detected in ESCC cell-derived exosomes instead of the exosome-negative marker GM130 ( Figure 5B ), and qRT-PCR data showed that miR-620 was highly expressed in the miR-620-Exo group ( Figure 5C ). In order to examine whether the receptor cells could accept the exosomes secreted from the donor cells as well as the exosomal miR-620, we constructed Cy3-miR-620 vectors in the exosomes secreted from ESCC cells (ECA-109/NC-Exo and KYSE450/NC-Exo) and the exosomes secreted from miR-620-overexpresed ESCC cells (ECA-109/miR-620-Exo and KYSE450/miR-620-Exo), as previously described (23). During Transwell assays, the donor ESCC cells containing Cy3-miR-620 were cultured for 24 h and then collected and seeded into the apical chamber of Transwell, and the receptor lung fibroblast cells (HFL1) cells were placed into the basolateral chamber. Through fluorescence microscope, it was observed that most of the receptor cells displayed red fluorescence, indicating that miR-620 could be transmitted through exosomes ( Figure 5D ). With the application of PKH67 Green Fluorescent Cell Linker Kit, it was further manifested that PKH67-marked exosomes were absorbed and internalized by ESCC cells ( Supplementary Figure S5C ), and miR-620-Exo could enhance the expression of miR-620 in the receptor HFL1 cells ( Figure 5E ). Moreover, it was shown that miR-620 expression remained basically unchanged compared with the control group after we treated exosomes with RNase A, while the addition of Triton X-100 declined miR-620 expression in the exosomes, which proved that miR-620 was located in the membrane ( Supplementary Figure S5D ).

An increasing number of evidence has proven that miRNA can regulate cell function by transferring exosomes to adjacent or distant cells (7, 8). For example, it has been reported that breast cancer-secreted miR-122 reprograms glucose metabolism to promote metastasis (20); meanwhile, it has been illustrated that cancer cell-secreted IGF2 instigates fibroblasts and bone marrow-derived vascular progenitor cells to promote cancer progression (21). Therefore, we wanted to explore whether exosomal miR-620 may regulate the aerobic glycolysis of lung fibroblast cells (HFL1). HFL1 cells absorbing the exosomes were renamed HFL1/Exo, and then it was discovered from qRT-PCR that the expression of FOXM1 and HER2 was decreased in HFL1 cells receiving ECA-109/miR-620-Exo and KYSE450/miR-620-Exo ( Figures 6A, B ). In addition, it was observed in Figures 6C –F that the glucose consumption, total lactate protein, relative PK catalytic activity, and the ECAR/OCR ratio were all reduced in HFL1 cells receiving miR-620-Exo, which indicated that exosomal miR-620 inhibited the aerobic glycolysis of HFL1 cells.

In order to verify whether ESCC cell-secreted miR-620 may regulate the aerobic glycolysis of premetastatic mouse lung fibroblasts (MLF), related in vivo were conducted. qRT-PCR results show that FOXM1 and HER2 expression was reduced in MLF cells receiving ECA-109/miR-620-Exo and KYSE450/miR-620-Exo ( Figure 7A ). Moreover, it was discovered that after MLF received exosomal miR-620, the glucose consumption, total lactate protein, relative PK catalytic activity, as well as the ECAR/OCR ratio were all decreased ( Figures 7B –E ), indicating that exosomal miR-620 could inhibit the aerobic glycolysis of MLF.

Furthermore, we investigated the effect of exosomal miR-620 on ESCC metastasis. First of all, the mice were pretreated with exosomes secreted from different ESCC cells for 6 weeks. Luciferase-labeled ESCC cells (ECA-109-luc and KYSE450-luc) were then injected intracardiac, with exosome treatment for another 3 weeks. As shown on Figure 7F , qRT-PCR data manifested that the expression level of miR-620 in mouse serum was higher in ESCC cells accepting exosomal miR-620 (ECA-109/miR-620-Exo and KYSE450/miR-620-Exo) than in tumor-transplanted groups (ECA-109-luc tumor and KYSE450-luc tumor). Subsequently, it was analyzed that the expression of FOXM1 and HER2 was reduced in MLF receiving ECA-109/miR-620-Exo and KYSE450/miR-620-Exo ( Figure 7G ). Moreover, it was found that the relative MLF luciferase level was higher in mice treated with miR-620-Exo, which indicated that ESCC metastasis was more significant in mice with higher miR-620 expression in serum ( Figure 7H ). The number of pulmonary metastatic nodules was further tested to be enhanced in MLF receiving ECA-109/miR-620-Exo and KYSE450/miR-620-Exo ( Figure 7I ). To conclude, exosomal miR-620 inhibited the aerobic glycolysis of MLF and thus promoted ESCC metastasis.

To determine if the primary tumor-secreted miR-620 enhances metastasis, orthotopic mammary xenografts were constructed. It was shown that after miR-620 was upregulated in KYSE450-luc cells, the xenograft tumors displayed a slower growth tendency as well as a lighter tumor weight at the 28th day ( Figures 8A, B ). It was then observed through qRT-PCR that the expression of FOXM1 and HER2 in KYSE450-luc cells was reduced upon miR-620 overexpression ( Figure 8C ). In addition, the aerobic glycolysis of KYSE450-luc cells was discovered to be suppressed by miR-620 overexpression, as the glucose consumption, total lactate protein, relative PK catalytic activity, and the ECAR/OCR ratio of KYSE450-luc cells were all declined ( Figures 8D –G ). To identify the effect of miR-620 on the ESCC metastasis, miR-620 was overexpressed in KYSE450-luc cells, which turned out that miR-620 expression was enhanced in mouse serum ( Figure 8H ). It was then observed that the expression of FOXM1 and HER2 in KYSE450-luc cells was reduced in MLF after miR-620 was overexpressed ( Figures 8I, J ), and the relative MLF luciferase level was higher in the xenograft tumors treated with enhanced miR-620 expression ( Figure 8K ). The above findings showed that miR-620 inhibited the aerobic glycolysis of primary tumors while promoting ESCC metastasis.