The circular RNA circDLG1 promotes gastric cancer progression and anti-PD-1 resistance through the regulation of CXCL12 by sponging miR-141-3p

Fresh-frozen or paraffin-embedded tissues were obtained from advanced-stage gastric cancer patients who underwent gastroscopy and biopsy, fine needle biopsy, or palliative surgery and received anti-PD-1 therapy at Sun Yat-sen University Cancer Center from August 2018 to October 2019. These patients were all included in a real-world study of PD-1 antibody therapy in gastric cancer (No: NCT04086888). A total of 126 gastric cancer patients were enrolled. Eighty-two of these patients had primary tumor tissues available. Seventy-three patients could be evaluated for the efficacy of PD-1 antibody therapy, among which 30 patients had tissues available (including primary tissues, adjacent normal tissues, and distant metastatic tissues). These patients were evaluated every 3-4 cycles of therapy and followed up regularly. The response rate was evaluated based on RECIST 1.1 guidelines. The tumor response included complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD). The objective response rate (ORR) was defined as the percentage of patients who achieved CR and PR. The clinical and pathological parameters, including age, sex, tumor size, tumor cell differentiation, peritoneal metastasis, and Lauren’s classification, were collected from patient records. This study was approved by the ethics committee of the Sun Yat-sen University Cancer Center, and informed consent was obtained from all patients. Overall survival (OS) was calculated from the start of PD-1 antibody treatment to the date of death or last contact, and progression-free survival (PFS) was calculated from the start of PD-1 antibody treatment to the date of progression or death.

Human gastric cancer cell lines (HGC27, BGC823, MKN45, MKN28, SGC7901, AGS), the normal gastric epithelial cell line GES-1, the murine gastric cancer cell line MFC and human embryonic kidney (HEK) 293 T cells were purchased from the Shanghai Institute of Cell Biology, Chinese Academy of Science (Shanghai, China). All cell lines were cultured according to the provider’s instructions.

To identify the potential circRNAs involved in gastric cancer progression and anti-PD-1 resistance, 3 patients who responded to anti-PD-1 (experienced PR) therapy with PFS of more than 10 months and 6 patients who were nonresponsive to anti-PD-1 (2 SD and 4 PD) therapy with PFS of less than 5 months were selected for circRNA microarray analysis.

Total RNA was isolated from human gastric cancer tissues and cell lines using TRIzol reagent (Sigma–Aldrich, St. Louis, MO, USA) according the manufacturer’s instructions. Reverse transcription of mRNA and miRNA was performed using random primers and stem–loop primers, respectively. qRT–PCR was conducted using a TaqMan Universal Master Mix II kit on a Bio–Rad CFX96 qPCR system, and fold changes were determined by using the relative quantification 2^-ΔΔCT method. The nuclear and cytoplasmic fractions of cells were separated using the PARIS Kit (Life Technologies) according to the manufacturer’s instructions. RNA was extracted from both fractions. Then, qRT–PCR was performed to determine the expression ratios of specific RNA molecules between the nuclear and cytoplasmic fractions. GAPDH and U6 served as cytoplasmic and nuclear markers, respectively. The primers used are presented in Additional file 1: Table S1.

Western blotting analysis was performed according to a previously described method [29]. Briefly, proteins were extracted from gastric cancer cell lines, and the protein concentration was calculated using the Pierce BCA protein assay kit (Thermo Scientific, Rockford, IL, USA). The following antibodies were used in this study: anti-GAPDH (CST, #2118 L) and anti-CXCL12 (CST, #3740S).

The IHC analysis was performed according to the method we described previously [30]. Briefly, the paraffin-embedded tissue blocks were cut into 4-μm slides. Immunostaining images were captured using a microscope (Leica, Germany). The immunoreactivity in each tissue section was assessed by two pathologists, and the degree of positivity was evaluated according to the percentage of positive tumor cells. The following antibodies were used: anti-CXCL12 (CST, 97958S; 1:200 dilution), anti-PD-L1 (22C3 pharmDX; Dako, Carpinteria, CA, USA), and anti-CD33 (clone PWS44; 1:200 dilution; Leica Biosystems, Nussloch, Germany). PD-L1 positivity was defined as a membrane staining intensity ≥1% in the tumor cells or tumor infiltrating immune cells. The median value of PD-L1 expression was used as the cutoff value to stratify high and low PD-L1 expression.

The detailed flow cytometry procedure was performed as previously described [31]. The following antibodies were used: anti-mouse IFNγ (BD Biosciences), anti-mouse CD45 (BD Biosciences), anti-mouse Gr-1 (TONBO Biosciences), anti-mouse Ly6G (TONBO Biosciences), anti-mouse ly6C (BD Biosciences), anti-mouse CD11b (TONBO Biosciences), anti-mouse F4/80 (BioLegend), anti-human CD11b (TONBO Biosciences), anti-human CD33 (BD Biosciences), and anti-human HLA-DR (BD Biosciences). Detailed information on the antibodies is provided in Additional file 2: Table S2. Briefly, the cells were digested and suspended as single cells, washed with PBS, and then resuspended in cell stabilizing buffer (BioLegend cat. No. 420201); the supernatant was discarded, and the sample was centrifuged at 350 g for 5 min. Then, 5 μl of Human TruStain FcX™ (Fc Receptor Blocking Solution, BioLegend Cat. No. 422301) and the antibodies were added and cultured for 15-20 min at room temperature. Finally, the cells were assessed by flow cytometry. For intracellular staining of IFNγ, fixation buffer (BioLegend Cat. No. 420801) was added and incubated for 20 min at room temperature. Then, the cells were washed with Intracellular Staining Perm Wash Buffer (BioLegend Cat. No. 421002), resuspended and centrifuged. IFNγ antibody was added, and the cells were cultured for 20 min and finally assessed by flow cytometry.

The FISH assay was performed according to previously described methods [25]. Briefly, cells were hybridized with Cy3-labeled circDLG1 probe and Cy5-labeled miR-141-3p probe (GenePharma, China) for 12 h at 37 °C. Then, the nuclei were counterstained using 4′,6-diamidino-2-phenylindole (DAPI) (Yeasen, Shanghai, China). Finally, the cells were viewed and captured under a ZEISS LSM800 confocal microscope (Carl Zeiss AG, Germany).

Cell counting kit-8 (CCK-8) and colony formation assays were performed to assess the cell proliferation ability. Transwell assays were conducted to evaluate cell migration and invasion abilities. The details of these assays were described in our previous study [32].

The RIP assay was conducted using the Magna RIP RNA-bing Protein Immunoprecipitation kit (Millipore, USA) according to the provider’s protocol. In brief, cell lysates were cultured with Dynabeads-coated IgG antibody (Millipore, USA) or AGO2 antibody (Cell Signaling Technology, USA) for 12 h at 4 °C. The purified RNA was subjected to qRT–PCR to detect the enriched circDLG1 and miRNA.

For the luciferase activity assay, potential binding sites were predicted using StarBase v3.0 and TargetScanHuman 7.2. Gastric cells were cotransfected with pGL-luc-circDLG1, pGL-luc-CXCL12 3′-UTR, and miR-141-3p mimics or negative control mimics for 48 h, and luciferase activity was detected using the dual-luciferase reporter assay system (Promega, USA) according to the manufacturer’s instructions.

Small hairpin RNAs (shRNAs) targeting the junction region of the circDLG1 sequence and the circDLG1-overexpressing lentiviral vector were synthesized by Geneseed Biotech Co., Ltd. (Guangzhou, China). Gastric cancer cell lines were transfected with circDLG1 shRNA or the circDLG1-overexpressing lentiviral vector following the manufacturer’s instructions according to a previously described method [25].

The in vivo tumorigenesis and metastasis experiments in BALB/c nude mice and C57BL/6 mice were approved by the Animal Experiment Ethics Committee of Sun Yat-sen University Cancer Center. The mice were purchased from the Shanghai Institute of Material Medicine (Shanghai, China) and were fed according to the provider’s instructions. The experiments were conducted as described in our previous study [33]. Briefly, to evaluate the in vivo tumorigenesis effect of circDLG1, MFC-sh-circDLG1 and MFC-sh-NC cells (1 × 10⁶ cells/mouse) were inoculated subcutaneously into the flanks of two groups of BALB/c nude mice and C57BL/6 mice (ten for each cell group). Tumor size was measured every 4 days, and tumor volume was estimated. After 4 weeks, the mice were sacrificed, and the tumors were excised. To investigate the effect of circDLG1 on tumor metastasis, MFC-sh-circDLG1 and MFC-sh-NC cells (2 × 10⁶ cells/mouse) were inoculated into the tail vein of two groups of BALB/c nude mice and C57BL/6 mice (ten for each cell group). Four weeks later, the mice were sacrificed, and the lungs were excised and paraffin embedded. Consecutive sections (4 μm) were made and stained with hematoxylin-eosin. The micrometastases in the lungs were examined and counted under a dissecting microscope.

To evaluate the effect of CXCL12 on immune evasion and the efficacy of anti-PD-1 therapy in vivo, an in vivo xenograft experiment was conducted in C57BL/6 mice. In brief, four groups of mice were implanted subcutaneously in the left flanks with MFC-sh-NC, and two groups of mice were implanted subcutaneously in the left flanks with MFC-sh-CXCL12 cells (ten mice for each group, 1 × 10⁶ cells/mouse). The MFC-sh-NC implanted mice were treated with 0.9% normal saline (NS), anti-PD-1 antibody, AMD3100 (a CXCR4 inhibitor), and the combination of AMD3100 and anti-PD-1 antibody. MFC-sh-CXCL12-implanted mice were treated with 0.9% normal saline (NS) and an anti-PD-1 antibody. A mouse anti-PD-1 antibody was purchased from Bio X Cell (West Lebanon, NH, USA). AMD3100 was purchased from Selleck (Shanghai, China) and dissolved in phosphate-buffered saline (PBS). The mice were intraperitoneally injected with 0.9% NS, AMD3100, or anti-PD-1 antibody according to a previously described method [34]. The mice were sacrificed at 4 weeks or before becoming moribund. The survival time was defined as the date of first therapy to the date of sacrifice.

The methods for ISH, sphere formation assay and RNA-seq are described in Additional file 3: Supplementary materials and methods.

To identify the circRNA expression profiles involved in gastric cancer progression and immune evasion, we performed a circRNA microarray analysis on 9 paired gastric cancer tissues (primary tissues and paired distant metastatic lesions) from patients who received anti-PD-1 treatment. The circus plot shows the distribution and expression profiles of the detected and differentially expressed circRNAs on human chromosomes, as well as potential miRNAs interacting with differentially expressed circRNAs (Fig. 1a). The heatmap presents the significantly expressed circRNAs in gastric cancer and paired distant metastatic tissues (Fig. 1b). To identify the potential circRNAs involved in anti-PD-1 resistance, we compared circRNA profiling in the primary tissues of 3 patients who were responsive (experienced PR) and 4 patients who were nonresponsive (experienced PD) to anti-PD-1 therapy (Fig. 1c). To select a circRNA that plays a critical role in gastric cancer progression and immune evasion, circRNAs (hsa_circ_0008583, has_circ_0002387) that were upregulated in both distant metastatic lesions and tissues resistant to anti-PD-1 therapy were identified. However, only hsa_circ_0008583 (derived from exons 13, 14, 15 and 16 of DLG1, thus designated circDLG1) was confirmed to be upregulated in distant metastatic lesions and primary gastric cancer tissues resistant to anti-PD-1 therapy in a larger cohort of patients (Fig. 1d and e). Moreover, circDLG1 expression was higher in primary tumor tissues than in adjacent normal tissues, although the difference was not significant (Additional file 4: Fig. S1). Kaplan–Meier analysis showed that circDLG1 expression was significantly associated with PFS in gastric cancer patients treated with anti-PD-1 therapy, and patients with low circDLG1 expression presented with significantly better prognosis than those with high circDLG1 expression (Fig. 1f). The patients were divided into two groups (high expression and low expression) based on the mean circDLG1 expression level from qRT–PCR. circDLG1 expression was significantly associated with tumor size and peritoneal metastasis. However, no association was observed between circDLG1 expression and age, sex, tumor cell differentiation, or Lauren’s classification (Additional file 5: Table S3). Several previous reports indicated that PD-L1 expression and tumor mutation burden (TMB) were associated with the anti-PD-1 response in gastric cancer [8, 9]. We questioned whether circDLG1 expression is associated with PD-L1 expression and TMB. ISH analysis was also used to confirm circDLG1 expression in gastric cancer tissues (blue staining indicates positive expression; red staining indicates negative), and IHC was performed to detect PD-L1 expression in gastric cancer tissues (Fig. 1g). However, no significant correlation was found between circDLG1 expression and PD-L1 expression or TMB (Fig. 1h and i).

Taken together, these results demonstrate that circDLG1 might play an important role in gastric cancer progression and immune evasion.

According to the circBase database and UCSC, circDLG1 (circRNA ID: hsa_circ_0008583) is derived from exons 13, 14, 15 and 16 of DLG1, which is located on chromosome 3 and has a length of 496 nucleotides (Fig. 2a). CircDLG1 was amplified by using divergent primers and confirmed by Sanger sequencing in AGS and SGC7901 gastric cancer cells (Fig. 2b), and PCR analysis of reverse-transcribed RNA (cDNA) and genomic DNA (gDNA) was performed. The results showed that the divergent primers could amplify products from cDNA but not from gDNA (Fig. 2b). To further confirm the characteristics of circDLG1, we used random hexamer or oligo (dT)₁₈ primers for reverse transcription of RNA from gastric cancer cells. The relative expression of circDLG1 was significantly decreased, while the expression of DLG1 mRNA was not decreased when oligo (dT)₁₈ was used (Fig. 2c). This finding indicates that circDLG1 has no poly A tail. Moreover, circDLG1 was resistant to RNase R (a highly processive 3′ to 5′ exoribonuclease that degrades linear RNAs but not circRNAs), while DLG1 mRNA was sensitive to RNase R (Fig. 2d). In addition, SGC7901 cells were treated with actinomycin D (an inhibitor of transcription), and RNA was extracted at the indicated times for the detection of circDLG1. The results demonstrated that the half-life of circDLG1 was significantly longer than that of DLG1 mRNA (Fig. 2e). Furthermore, the qRT–PCR and FISH results indicated that circDLG1 is mainly located in the cytoplasm (Fig. 2f and g). These results demonstrate that circDLG1 is a stable and abundant circRNA transcript.

To explore the biological function of circDLG1 in gastric cancer, we first measured circDLG1 expression in gastric cancer cell lines and the gastric normal epithelial cell line GES-1. The results showed that circDLG1 expression was upregulated in gastric cancer cell lines compared with GES-1 gastric normal epithelial cells (Fig. 3a). We designed three shRNA plasmids to target the unique back-splice junction. Back-splice junction-specific shRNAs (sh-1, sh-2, sh-3) effectively knocked down circDLG1 expression in HGC27 and SGC7901 cells (Fig. 3b). In vitro CCK-8 and colony formation assays indicated that knockdown of circDLG1 significantly inhibited the proliferation and colony formation ability of HGC27 and SGC7901 cells (Fig. 3c and d). In addition, transwell assays indicated that knockdown of circDLG1 significantly inhibited the invasion ability of HGC27 and SGC7901 cells (Fig. 3e). Moreover, considering that epithelial-mesenchymal transition (EMT) plays an important role in the process of cancer progression and invasion, we explored the effects of circDLG1 on the EMT and stemness of gastric cancer cells. Immunofluorescence analysis showed that knockdown of circDLG1 significantly increased the expression of E-cadherin while decreasing the expression of N-cadherin (Fig. 3f). qRT–PCR analysis indicated that knockdown of circDLG1 markedly increased the expression of epithelial markers (E-cadherin, α-catenin, β-catenin) while decreasing the expression of mesenchymal markers (N-cadherin, vimentin, Snail, Slug) (Fig. 3g). IHC showed that the expression of E-cadherin decreased, whereas vimentin expression increased in tissues with high expression of circDLG1 (Additional file 6: Fig. S2). In addition, sphere formation assays showed that knockdown of circDLG1 markedly inhibited sphere formation (Fig. 3h), and qRT–PCR analysis indicated that knockdown of circDLG1 significantly inhibited the expression of stemness-associated markers (CD44, CD133, Oct4, CD24, CD155, CD166, Nanog, Sox2) (Fig. 3i).

We generated MFC-sh-circDLG1 cell lines through the stable transduction of circDLG1-targeting shRNA vectors into the MFC murine gastric cancer cell line. qRT–PCR analysis confirmed that knockdown of circDLG1 significantly decreased circDLG1 expression in MFC cells (Additional file 7: Fig. S3). To elucidate the function of circDLG1 in vivo, MFC-sh-circDLG1 cells were injected into immunocompetent mice subcutaneously or via the tail vein. The results showed that knockdown of circDLG1 significantly inhibited tumor progression and metastasis compared with the control treatment (Fig. 4a). In contrast, no significant difference was observed in tumor progression or metastasis between the MFC-sh-circDLG1 and MFC-control groups of immunodeficient mice (Fig. 4b). These results demonstrated that circDLG1 promotes gastric cancer progression and metastasis and might be related to tumor immune evasion. IHC analysis showed that MFC-sh-circDLG1-derived tumors were associated with increased CD8⁺ tumor-infiltrating lymphocytes (TILs) and decreased infiltration of Gr-1- and CD11b-positive MDSCs (Gr-1 and CD11b are mouse MDSC markers) (Fig. 4c, Additional file 8: Fig. S4). Moreover, flow cytometric analysis of immune cells in subcutaneous tumors showed that knocking down circDLG1 significantly increased the number of CD8⁺ T cells and IFNγ⁺ cells but significantly decreased the number of MDSCs and granulocytic (G)-MDSCs (Fig. 4d). We further evaluated the relationship between circDLG1 and MDSCs in gastric cancer tissues. The expression of CD33 (a marker of human MDSCs) was detected by IHC, and the results showed a significant correlation between circDLG1 and CD33⁺ cells (Fig. 4e and f). These results demonstrated that circDLG1 promotes gastric cancer progression by inducing MDSC infiltration.

To further explore the underlying molecular mechanism by which circDLG1 promotes gastric cancer progression and immune evasion, gene expression profiling was performed in gastric cancer cells with or without knockdown of circDLG1. SGC7901 cells were transfected with circDLG1 shRNA, and the expression levels of mRNAs were detected by RNA sequencing in these cells (Additional file 9: Fig. S5a). Gene annotation enrichment analysis showed that differentially expressed genes involved in T cell migration, EMT and immune responses were significantly enriched based on Gene Ontology (GO) analysis, and TNF signaling and the PI3K-Akt signaling pathway were significantly enriched based on Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis (Additional file 9: Fig. S5b). The circRNA/mRNA microarray dataset was analyzed, and the correlation between circDLG1 and mRNAs was determined. The results showed that CXCL12 had the highest correlation coefficient. In addition, CXCL12 was one of the most significantly decreased genes by knockdown of circDLG1 in gastric cancer cells based on the RNA sequencing assay. Thus, we speculated that there might be an association between circDLG1 and CXCL12. qRT–PCR analysis showed that knockdown of circDLG1 significantly decreased the expression of CXCL12, whereas ectopic expression of circDLG1 significantly increased CXCL12 expression in gastric cancer cell lines (Additional file 9: Fig. S5c and d). Western blot analysis confirmed that knockdown of circDLG1 significantly decreased the protein level of CXCL12 in SGC7901 and HGC27 gastric cancer cells (Additional file 9: Fig. S5e). Moreover, the decreased level of CXCL12 caused by CXCL12 knockdown could be restored by circDLG1 ectopic expression (Additional file 9: Fig. S5f). Accordingly, the suppressed cell migration ability caused by CXCL12 knockdown could be restored by circDLG1 ectopic expression in gastric cancer cells (Additional file 9: Fig. S5g). Furthermore, circDLG1 expression was significantly associated with CXCL12 in gastric cancer tissues (Additional file 9: Fig. S5h). These data indicated that circDLG1 exerts its effect at least partly by regulating CXCL12 expression.

Previous studies have reported that circRNAs can act as miRNA sponges to regulate the expression of protein-coding genes. In the present study, we found that circDLG1 was mainly located in the cytoplasm, which indicated that it might function as a miRNA sponge. Interestingly, by using the online bioinformatics tools miRanda and TargetScan, 46 miRNAs were predicted to be potential targets of circDLG1 (Additional file 10: Table S4). We first explored the occupancy of AGO2 in the region of circDLG1. The results of RNA immunoprecipitation for AGO2 in SGC7901 cells stably expressing Flag-AGO2 and Flag-GFP showed that endogenous circDLG1 pulled down by Flag-AGO2 was significantly enriched, suggesting that circDLG1 is incorporated into the RNA-induced silencing complex (RISC) (Fig. 5a). To identify the miRNAs that can interact with circDLG1, a circDLG1 fragment was amplified and inserted downstream of a luciferase reporter gene. Transfection of circDLG1 shRNA significantly decreased the luciferase activity of the circDLG1 reporter (Fig. 5b). To further identify the key miRNAs that can bind to circDLG1, a luciferase screen with a miRNA mimic library was performed. A total of 46 miRNA mimics were cotransfected with the circDLG1 luciferase reporter into SGC7901 cells (Fig. 5c). The luciferase activity of the circDLG1 reporter was significantly reduced by four miRNAs (miR-141-3p, miR-138-5p, miR-651-3p, miR-155-3p) (Fig. 5d). In addition, RNA precipitation (RIP) analysis with a biotin-labeled circDLG1 probe was conducted to evaluate the key miRNAs that can interact with circDLG1. Compared with the negative control, miR-141-3p, but not the other three miRNAs, was enriched by the biotin-labeled circDLG1 probe (Fig. 5e), and the enrichment of miR-141-3p by circDLG1 was reduced when circDLG1 was knocked down in gastric cancer cells (Fig. 5f). In addition, an inverse association was observed between the expression of circDLG1 and miR-141-3p in gastric cancer tissues (Fig. 5g). Furthermore, the RNA FISH assay demonstrated that miR-141-3p colocalized with both endogenous and exogenous circDLG1 in gastric cancer cells (Fig. 5h). These data demonstrated that circDLG1 could function as a miR-141-3p sponge in gastric cancer cells.

To evaluate whether circDLG1 could regulate CXCL12 expression by sponging miR-141-3p, we first confirmed the regulation of CXCL12 by miR-141-3p. The results showed that ectopic expression of miR-141-3p could significantly decrease the mRNA level of CXCL12 in gastric cancer cells (Fig. 5i). Moreover, Western blot analysis showed that ectopic expression of miR-141-3p or knockdown of circDLG1 markedly reduced the protein level of CXCL12, inhibition of miR-141-3p significantly increased the protein level of CXCL12, and the increased level of CXCL12 by miR-141-3p inhibition could be partly restored by circDLG1 knockdown (Fig. 5j). Accordingly, the luciferase activity assay showed that ectopic expression of miR-141-3p or knockdown of circDLG1 significantly reduced the luciferase activity of wild-type CXCL12-3′-UTR, inhibition of miR-141-3p significantly increased the luciferase activity of wild-type CXCL12-3′-UTR, and the increased luciferase activity of wild-type CXCL12-3′-UTR by miR-141-3p inhibition could be partly restored by circDLG1 knockdown; however, no significant change in luciferase activity was observed when the CXCL12-3′-UTR was mutated (Fig. 5k). In addition, miR-141-3p expression was inversely associated with CXCL12 expression in gastric cancer tissues (Fig. 5l). These results indicated that circDLG1 could function as a miR-141-3p sponge to regulate CXCL12 expression.

It has been reported that CXCL12 plays an important role in cancer progression and immune evasion [35, 36], and we questioned whether CXCL12 is implicated in gastric cancer progression and anti-PD-1 therapy resistance. The potential role of CXCL12 was further explored in gastric cancer cells. In vitro experiments showed that knockdown of CXCL12 significantly inhibited the invasion ability of SGC7901 and HGC27 cells, whereas ectopic expression of CXCL12 significantly increased the invasion ability of AGS cells (Additional file 11: Fig. S6a). Moreover, knockdown of CXCL12 markedly decreased the expression of N-cadherin but increased the expression of E-cadherin (Additional file 11: Fig. S6b).

To evaluate the effect of CXCL12 on anti-PD-1 efficacy, MFC-sh-NC or MFC-sh-CXCL12 cells were injected into C57BL/6 mice subcutaneously or via the tail vein. Then, the mice were treated with 0.9% normal saline (NS), AMD3100 (CXCR4 inhibitor), or anti-PD-1. The results showed that compared with those in the control group, knockdown of CXCL12 significantly inhibited tumor growth and metastasis, and anti-PD-1 treatment further reduced tumor growth and metastasis in C57BL/6 mice (Fig. 6a and b). Likewise, blocking CXCL12/CXCR4 using AMD3100 significantly inhibited tumor growth and metastasis, and the combination of AMD3100 and anti-PD-1 further reduced tumor growth and metastasis (Fig. 6a and b). Moreover, Kaplan–Meier analysis indicated that knockdown of CXCL12 or AMD3100 treatment could significantly prolong the OS of mice, and anti-PD-1 therapy could further increase the OS in mice (Fig. 6c). In addition, flow cytometric analysis of immune cells in subcutaneous tumors showed that knockdown of CXCL12, AMD3100 treatment and anti-PD-1 therapy significantly increased the number of CD8⁺ T cells and IFNγ⁺ cells but decreased the number of MDSCs (Fig. 6d-f). We then evaluated the association of CXCL12 expression and the anti-PD-1 response in gastric cancer patients. The results showed that high expression of CXCL12 had a strong trend to be associated with resistance to anti-PD-1 therapy, and patients with high CXCL12 expression presented with adverse survival after anti-PD-1 therapy (Fig. 6g-i).