Claudin-12 Deficiency Inhibits Tumor Growth by Impairing Transendothelial Migration of Myeloid-Derived Suppressor Cells

Immunosuppression has been recognized as one of the hallmarks of cancer (1). Myeloid-derived suppressor cells (MDSC) are the main drivers for the establishment of immunosuppressive tumor microenvironment (TME; ref. 2). Notably, the inhibited T cells functions, promoted tumor angiogenesis and enhanced stemness of cancer cells by MDSCs, are characterized by localization dependence, which primarily occur at tumor site in mouse model or patients with cancer (3, 4). The development and expansion of MDSCs have been extensively studied; however, the mechanisms underlying the transendothelial migration of MDSCs into tumor tissue are rarely reported.

Extravasation of immune cells proceeds via several steps, including rolling, adhesion, and transmigration (5). Tight junction proteins, including claudins expressed in endothelial cells (EC) primarily, form barriers to prevent the extravasation of molecules and cells (6, 7). Interestingly, a few studies reported that some immune cells expressed tight junction proteins, which may be actively involved in the transit of immune cells across vascular walls. Expression of occludin on dendritic cells (DC) is sufficient to loosen epithelial tight junctions by establishing tight junction–like structures between DCs and epithelial cells, which helps the DCs to transit across intestinal epithelium without compromising the epithelial barrier function (8). An antibody to the junctional adhesion molecules (JAM) inhibits the recruitment of monocytes and neutrophils into the cerebrospinal fluid of mice with experimental meningitis, suggesting that JAMs also function in transmigration (9). These studies put forwards the possibility that tight junction proteins, such as claudins, might be involved in the transit of immune cells across the blood vessels in diseases.

Claudin-12 (Cldn12) is an unusual member of the claudins, widely expressed on epithelial cells, ECs and pericytes, as it lacks a PDZ-domain–binding motif for interaction with the cytoskeleton (10, 11). Claudin proteins share a common overall structure that includes a short cytoplasmic N-terminal region, two extracellular loops formed by four transmembrane domains, and a cytoplasmic C-terminal tail (12). In the present study, we examined how host Cldn12 deficiency affects tumor growth and immune compositions. Surprisingly, we found that tumor growth was retarded in Cldn12-deficient mice, accompanying with decrease of MDSCs in tumor. Moreover, Cldn12 was expressed by MDSCs, and its expression was required for the transmigration of these cells across vascular walls through formation of Cldn12–Cldn12 homotypic interactions with tumor vascular ECs. And the expression of Cldn12 in MDSCs was modulated by GM-CSF in an AKT-dependent manner. These results suggest a new function for Cldn12 in MDSCs transendothelial migration and shed new light on immunotherapy for tumor by suppressing the immune suppressors.

Cldn12^+/+, Cldn12^+/− and Cldn12^−/− mice in the C57BL/6J background were housed in our laboratory, and detailed information about the generation and characters of these mice has been reported elsewhere (10, 11). All mice were bred under specific pathogen-free conditions, and male mice ages 6–8 weeks were used for the experiments, whereas the female mice exhibited developmental defects. All animal studies were approved by the Animal Care and Use Committee of the Institute of Biophysics, Chinese Academy of Sciences Beijing (No. SYXK2019022).

The melanoma B16F10 and Lewis lung cancer (LLC) cell lines were purchased from the ATCC. MCA205 were generated in Dr. Blankenstein's laboratory from fibrosarcoma in C57BL/6 mice with 3-methylcholanthrene induction as described previously (13). sEND.1 cells were immortalized ECs from blood vessels of the skin and also established in Dr. Blankenstein's laboratory (14). The immortalized MDSCs cell line MSC2 was generously provided by the François Ghiringhelli laboratory (15). Mycoplasma contamination testing was performed on cell lines in culture every 1 month and authenticated by SNP testing. Cells (at passage 10 or below) were routinely cultured in DMEM (or RPMI-1640 for MSC2) supplemented with 10% FBS and 100 U/mL penicillin–streptomycin and subcultured when cells reached 90% confluency.

LLC cells, MSC2 and sEND.1 cell lines with Cldn12-KO (knockout) were constructed with two guide RNAs, respectively, targeted upstream and downstream regions of the Cldn12 gene. GFP-positive single cell was selected and identified by PCR of genomic DNA, qPCR, and western blot. The primer sequences are provided in Supplementary Table S1.

LLC, MCA205, and B16F10 tumor cells at exponential growth stage were harvested and re-suspended in PBS at concentrations of 5 × 10⁶ cells/mL (LLC), 2.5 × 10⁶ cells/mL (MCA205) and 1.25 × 10⁶ cells/mL (B16F10). 200-μL tumor cell suspensions were injected subcutaneously into mice, and tumor growth was monitored by measuring the long and short diameters every 2 to 3 days starting at day 6 post-inoculation. Tumor volumes (V) were assessed in mm³ using the formula: V = 0.5 × (L × W²), with L being the long and W being the short diameters of the tumor. All animal experiments were repeated at least three times.

Single-cell suspensions prepared from bone marrow, peripheral blood, or whole-tumor tissue suspensions were stained with antibodies, including those specific for CD4 (AB_312719; AB_312713), CD8 (AB_312747), CD11b (AB_2129375), Gr1 (AB_313377; AB_313373), Ly6G (AB_2227348), Ly6C (AB_1186132), TNFα (AB_315428) and IFNγ (AB_315404). Before staining for TNFα or IFNγ, cells were fixed and permeabilized according to the manufacturer's instructions. FACS Calibur device was used to record 20,000–50,000 events for each sample.

Tumor tissue sections or BM-MDSCs seeded on sterile glass coverslips were fixed and incubated with antibodies specific for CD11b (AB_955740), Gr1 (AB_396708), CD31 (AB_940884), Cldn12 (AB_1070543). Fluorescence conjugated secondary antibodies were applied to identify primary antibody binding, and nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI).

Splenocytes of naïve mice were labeled with 5,6-carboxyfluorescein diacetate (CFSE) and cocultured with CD11b⁺Gr1⁺ MDSCs freshly isolated from tumor tissue of Cldn12^+/+ and Cldn12^−/− mice. With concanavalin A stimulation for 72 hours, then cells were collected for CD4 and CD8 staining. The dilution of CFSE on CD4⁺ or CD8⁺ T cells was determined by flow-cytometric analysis.

CD11b⁺Gr1⁺ MDSCs from tumor tissues of Cldn12^+/+ and Cldn12^−/− mice were sorted with FACS Aria device. Two pathways were adopted for transfer of primary MDSCs: Intratumorous and intravenous injections.

Labeled BM-MDSCs resuspended in 2%-serum medium, and then added to Transwell chambers containing the confluent sEND.1 monolayers. In parallel, the medium in the bottom chamber was replaced by 20%-serum medium. After 6 to 8 hours, removed cells on the upper surfaces of the Transwell chambers, and the migrated cells that were attached to the undersides of the chambers were observed with Olympus microscope. Migrated cells were counted from 10 to 20 random microscopic fields using ImageJ (1.52a).

Intravenously injected the fresh bone marrow cells isolated from 6- to 8-week-old donor Cldn12^+/+ or Cldn12^−/− mice into lethally irradiated recipient mice (1 × 10⁷ cells/mouse). The following groups of bone marrow chimeras were established (donor→recipient): WT→WT (Cldn12^+/+→Cldn12^+/+), WT→KO (Cldn12^+/+→Cldn12^−/−), KO→WT (Cldn12^−/−→Cldn12^+/+), and KO→KO (Cldn12^−/−→Cldn12^−/−). After transplantation for 8 weeks, PCR was performed to detect Cldn12 gene expression in peripheral blood leukocytes. The primer sequences are provided in Supplementary Table S1. Constructed chimeric mice were subcutaneously grafted with LLC cells as described above.

Cocultured MSC2 and sEND.1 cells with mCherry–Cldn12 and GFP-Cldn12 overexpression for 48 hours. Cells were lysed with lysis buffer (50 mmol/L Tris-HCl, pH 8.0; 150 mmol/L NaCl; 1 mmol/L EDTA and 0.8% Triton X-100) containing 0.5 mmol/L DTT and protease-inhibitor cocktail for 30 minutes on ice, following by centrifugation at 13,000 rpm for 15 minutes at 4°C. Supernatants were immunoprecipitated with anti-GFP or anti-RFP affinity beads. immunoprecipitates were separated by SDS-PAGE and analyzed with immunoblotting.

MSC2 cells were seeded in 6-wells plates and treated with GM-CSF for 24 hours. MSC2 cells were collected and resuspended in Seahorse XF DMEM medium (pH 7.4, 10 mmol/L glucose, 1 mmol/L sodium pyruvate, and 2 mmol/L glutamine), then seeded into poly-D-lysine–pretreated XF96 Cell Culture Microplate (5,000 cells per well in 80 μL culture medium). For the measurement of glycolytic rates, Rotenone and antimycin A was added into the medium to a final concentration of 0.5 μmol/L for each. The glycolytic proton efflux rate was determined by Seahorse XF96 extracellular flux analyzer. For measurement of mitochondrial respiration, the blockers were added sequentially through the ports of the Seahorse Flux Pak cartridges. Oxygen consumption rates were measured using a Seahorse XF96 extracellular flux analyzer.

RIPA solution [50 mmol/L Tris-HCl (pH 7.4), 150 mmol/L NaCl, 1.0% NP40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mmol/L EDTA] supplemented with 1 mmol/L PMSF and EDTA-free protease-inhibitor tablets was used for cell lysis. Protein extracts were separated by SDS-PAGE. Specific bands were incubated with chemiluminescent substrate for several minutes and detected by Chemiluminescence Imaging System.

Total RNA was extracted from isolated cells or tumor tissue and quantified with an ND-1000 spectrophotometer. RNA (1 μg) was reverse transcribed into cDNA with random primers. cDNA was the template for real-time PCR conducted according to the SYBR Green Mixture Kit manufacturer's instructions with Rotor-Gene TM6000 software. Relative expression was calculated using the 2^−ΔΔC_T method and Gapdh was used as a reference gene. The primer sequences are provided in Supplementary Table S1.

Statistical analysis was performed using GraphPad Prism V.8.0. Data are shown as mean ± SEM. Comparisons between two groups were calculated using the unpaired nonparametric Mann–Whitney test and comparisons between two groups at multiple time points were calculated by two-way ANOVA Sidak multiple comparisons. Comparisons of more than two groups were calculated using one-way ANOVA Tukey multiple comparisons.

To test the effect of host Cldn12 deficiency on tumor growth, we engrafted LLC tumor cells to Cldn12^+/+, Cldn12^+/−, and Cldn12^−/− mice. The resultant tumors in Cldn12-deficient mice were smaller than those in wild-type (WT) control mice (Fig. 1A–C). Tumor weight tended to decrease along with the number of Cldn12 alleles in the host (Fig. 1B). Similar results were obtained with tumors derived from fibrosarcoma MCA205 and melanoma B16F10 cells (Fig. 1D and E).

Because the expression of Cldn12 by tumor cells might induce antitumor immune responses as immunogen in Cldn12^−/− mice, we generated a Cldn12-KO LLC tumor cell with CRISPR–Cas9 genome editing (Fig. 1F; Supplementary Fig. S1A–S1C). Consistent with previously published results (16), KO of Cldn12 arrested LLC tumor cells growth in vitro (Supplementary Fig. S1D). When Cldn12-KO LLC cells were transplanted into Cldn12^+/+, Cldn12^+/−, and Cldn12^−/− mice, tumor growth was also inhibited in Cldn12^−/− mice (Fig. 1G). The similar growth kinetics between Cldn12-WT and Cldn12-KO LLC tumors in Cldn12^+/+, Cldn12^+/− and Cldn12^−/− mice, excluded the possibility that tumor regression in Cldn12^–/– mice results from immunogenicity of Cldn12 expressed by tumor cells. Thus, transplanted tumor growth is inhibited in Cldn12^–/– mice.

We investigated whether Cldn12 deficiency altered tumor immune cells compositions. Percentages of infiltrated CD45⁺ cells, CD11c⁺ DCs, CD11b⁺F4/80⁺ macrophages, NK1.1⁺ natural killer cells, B220⁺ B cells, CD4⁺ or CD8⁺ T cells and T_REG cells in tumor tissues were comparable between WT and Cldn12^−/− mice (Supplementary Fig. S2). Intriguingly, the proportions of CD11b⁺Gr1⁺ MDSCs in whole-tumor tissue suspensions were drastically lower in the tumor tissues of Cldn12^−/− mice than in those of the WT counterparts on day 22 post-inoculation (Fig. 2A). And the normalized percentages of MDSCs by tumor volume and weight also displayed lower in the tumor tissues of Cldn12^–/– mice, compared with those in WT mice (Fig. 2B). In addition, along with the tumors burden, the slope of percentages of MDSCs in tumor tissues of WT and Cldn12^–/– mice were 1.063 and 0.561, respectively, matched with tumor growth rate (Fig. 2C). Correspondingly, immunofluorescence staining in situ confirmed the presence of lower numbers of CD11b⁺ cells and Gr1⁺ cells in tumor sections from Cldn12^−/− mice (Fig. 2D and E). According to previous findings (2), mouse MDSCs can be subdivided into CD11b⁺Ly6C^hiLy6G⁻ monocytic MDSCs (M-MDSC) and CD11b⁺Ly6C^lowLy6G⁺ polymorphonuclear MDSCs (PMN-MDSC). We found that the reductions in the percentages of M-MDSCs and PMN-MDSCs were similar in Cldn12^−/− and WT mice, and the ratio of PMN-MDSCs to M-MDSCs did not differ significantly between the both mice strains (Supplementary Fig. S3A–S3C).

Moreover, with less MDSCs within the TME of Cldn12^–/– mice, the relative mRNA expression level of arginase-1 (Arg-1) and inducible nitric oxide synthase (iNOS), key mediators for MDSCs immunosuppression (2), were reduced in tumor tissues of Cldn12^−/− mice, compared with those in WT mice (Fig. 2F). Conversely, the key antitumor mediators’ IFNγ and TNFα mRNA levels were higher in tumor tissues from Cldn12^−/− mice than in those from WT mice (Fig. 2G). The percentages of CD4⁺, CD8⁺ T cells able to produce IFNγ, TNFα displayed tendency to increase in tumor tissues from Cldn12^–/– mice at day 22 post-inoculation, compared with those from WT mice (Fig. 2H and I; Supplementary Fig. S4A–S4C). In addition, there was no significant difference between splenic CD4⁺ or CD8⁺ T cells from naive WT and Cldn12^−/− mice in secreting IFNγ under stimulation (Supplementary Fig. S5A–S5D). Angiostasis mediated by IFNγ and TNFα is crucial for T cells mediated tumor rejection (17–19). Consistently, tumorous blood vessel density was lower in tumor tissue of Cldn12^−/− mice, rather than WT mice (Fig. 2J). Therefore, the tumorous accumulation of MDSCs was impaired in Cldn12^–/‒ mice, leading to decreased antitumor immune responses and angiostasis in TME.

Whether Cldn12 deficiency affects MDSCs functions was further studied. The expression of Cldn12 in MDSCs was first determined. Bone marrow–derived myeloid (BMDM) cells were isolated from WT and Cldn12^−/− mice and cultured with GM-CSF stimulation to induce their differentiation into MDSCs (BM-MDSCs; ref. 19). GM-CSF resulted in similar levels of differentiation of BMDM cells from WT and Cldn12^−/− mice (Supplementary Fig. S6A and S6B), indicating that Cldn12 was not involved in MDSCs development. To detect Cldn12 expression in the BM-MDSCs, absolute-quantitation RT-PCR and relative-quantitation RT-PCR were performed (Supplementary Fig. S6C and S6D). Immunofluorescence staining detected expression of Cldn12 protein in WT BM-MDSCs, but not in Cldn12^−/− BM-MDSCs (Fig. 3A). These results demonstrated that the mRNA and protein of Cldn12 were expressed in WT BM-MDSCs and deficient in Cldn12^−/− BM-MDSCs.

We cocultured splenic T cells from naïve WT mice with CD11b⁺Gr1⁺ MDSCs isolated from the tumor tissues of WT and Cldn12^–/– mice and found no difference in suppression of CD4⁺ and CD8⁺ T-cells proliferation stimulated by concanavalin A (Fig. 3B and C), suggesting that Cldn12 did not affect the immunosuppressive activity of MDSCs. In addition, the relative mRNA expression of Arg-1 and iNOS were comparable in isolated equivalent WT and Cldn12^–/– MDSCs (Fig. 3D).

Furthermore, we manually restored the amount of Cldn12^−/− MDSCs to the same as WT MDSCs within the TME to test the immunosuppressive and tumor promotion effect of WT and Cldn12^−/− MDSCs in vivo. Intratumorous injection of equal numbers of WT or Cldn12^−/− MDSCs into LLC tumor on the left or right flank of individual mouse resulted in similar tumor growth rate and tumor weight (Fig. 3E and F). The ratio of LLC to MDSCs was 5 to 1, which has been characterized by a significant improvement of tumor growth (Supplementary Fig. S7A–S7C). Altogether, the retarded tumor growth in Cldn12-deficient mice could be attributed to the impaired accumulation of MDSCs in the tumor, rather than the compromise immunosuppressive function of MDSCs.

Because of lower numbers of MDSCs in tumor tissues of Cldn12^–/– mice, the mechanisms underlying Cldn12 deficiency blocks the accumulation of MDSCs in tumor were further studied. We found that the percentages of MDSCs in the peripheral blood of tumor-bearing Cldn12^−/− mice became even higher in Cldn12^−/− mice after two weeks, rather than in spleen and bone marrow (Fig. 4A; Supplementary Fig. S8A–S8C). To investigate whether Cldn12 affects MDSCs transit across blood ECs, we isolated MDSCs from tumor tissues implanted in either WT or Cldn12^−/− mice, labeled with PKH26 and CFSE fluorescent dyes, respectively, and mixed in 1:1 ratio for adoptive transfer (Fig. 4B). Lower numbers of infiltrated exogenous Cldn12^−/− MDSCs than exogenous WT MDSCs were found in the tumor tissues of recipient mice, suggesting that Cldn12 deficiency inhibits extravasation of MDSCs across tumor vasculature (Fig. 4C–E). The ratio of labeled Cldn12^+/+ to Cldn12^−/− MDSCs increased from 1 to 1 in the mixture to 1.77 to 1 in the recovered sample after 48 hours (Fig. 4F). In addition, the proliferation (BrdUrd⁺ in CD11b⁺Gr1⁺ cells) or apoptosis (Annexin V⁺ in CD11b⁺Gr1⁺ cells) of MDSCs in bone marrow or peripheral blood, spleen and tumor tissues of tumor-bearing WT and Cldn12^−/− mice display no significant difference (Supplementary Fig. S8B–S8E). These results suggest that Cldn12 is required for trafficking of MDSCs into tumor sites, without affecting their proliferation or apoptosis.

We further deeply explored the role of Cldn12 in MDSCs trafficking, and found that expression of chemokines in tumor tissues and chemokine receptors on MDSCs from peripheral blood did not differ between WT and Cldn12^−/− mice (Supplementary Fig. S9A–S9C), suggesting that Cldn12 participates into MDSCs migration independent on chemotaxis axis. Junctional proteins may be function in paracellular transmigration of leukocytes across endothelial/epithelial monolayers (5, 8). To examine whether Cldn12 has such a role, in vitro Transwell assay was used to mimic the trafficking of MDSCs. When the Transwell chambers were coated with immortalized vascular ECs sEND.1, fewer Cldn12^−/− MDSCs migrated under layers of chambers compared with Cldn12^+/+ MDSCs, but when the Transwell chambers were left uncoated, Cldn12^−/− and Cldn12^+/+ BM-MDSCs displayed similar migration to the under layers of the chambers (Fig. 5A). These results highlighted the role of Cldn12-mediated interactions between MDSCs and ECs during MDSCs transendothelial migration.

Claudins usually participate in homotypic interactions between adjacent cells (20). By immunofluorescence microscopy, we confirmed the expression of Cldn12 on vascular ECs of tumor tissue (Fig. 5B). Transmission electron microscopy results showed that myeloid-lineage cells could form “tight junction–like” structures with blood ECs, rather than lymphoid-lineage cells, in tumor tissue from WT mice (Fig. 5C). To directly observe homotypic interactions between MDSCs and ECs, we constructed MSC2–mCherry–Cldn12 and sEND.1–GFP–Cldn12 cells, Time-lapse microscopic imaging showed that Cldn12 was distributed alongside the junction between cocultured MSC2 and sEND.1 cells (Fig. 5D). GFP–Cldn12 protein expressed in sEND.1 coimmunoprecipitated with mCherry–Cldn12 protein expressed in MSC2, and vice versa (Fig. 5E), suggesting the existence of Cldn12–Cldn12 homotypic interactions between MDSCs and ECs. In addition, we observed that Cldn12 preferred to aggregate along the junction between adjacent 293T cells, compared with GFP–Lyn11 (a fusion of GFP with an 11-residue plasma-membrane-targeting N-terminal sequence of tyrosine-protein kinase Lyn). The fold changes of junctional to non-junctional signal intensity of mCherry–Cldn12 were about three times, larger than fold changes of GFP–Lyn11, which evenly distributed on cell membranes (Fig. 5F), implying that the aggregation of Cldn12 proteins toward to cell junctions maybe in favor of Cldn12–Cldn12 homotypic interactions formation.

To address the effects of Cldn12 deficiency on either myeloid or non-myeloid cells in vivo, bone marrow–chimeric mice were constructed (Supplementary Fig. S10A). LLC tumor cells were subcutaneously inoculated into the bone marrow chimeras, and tumor growth in WT→WT mice was significantly greater than those in WT→KO, KO→WT and KO→KO mice (Fig. 5G and H). MDSCs infiltration reflected tumor volume, with the highest percentage of MDSCs in the tumors of the WT→WT mice, followed by the WT→KO and KO→WT mice, and the lowest percentage in the KO→KO mice, when normalized by tumor volume and weight (Fig. 5I). At the same time, to determine the importance of Cldn12–Cldn12 homotypic interactions in MDSCs transmigration in vitro, we constructed sEND.1 and MSC2 with Cldn12-KO cell lines and rescued Cldn12 expression in KO cells by transfection (Supplementary Fig. S10B–S10D). Transwell assay was applied and displayed that both cell lines with Cldn12 expression or restoration rather than either one or none promoted MSC2 transit across sEND.1 monolayers (Supplementary Fig. S10E), implying that Cldn12–Cldn12 homotypic interactions between MDSCs and ECs were important for MDSCs transendothelial migration. Therefore, we concluded that Cldn12–Cldn12 homotypic interaction between MDSCs and ECs promoted transmigration of MDSCs across tumor vascular walls, resulted in enhanced accumulation of MDSCs in the tumor tissues.

Having demonstrated that Cldn12-mediated homotypic interactions were important for MDSCs tumorous trafficking, we further explored the upstream regulators for Cldn12 expression in MDSCs. We found that the mRNA level of Cldn12 in MDSCs isolated from peripheral blood was higher in the LLC tumor-bearing mice than that in naïve control mice, implying that tumor burden could induce the expression of Cldn12 in MDSCs (Fig. 6A). GM-CSF is often overproduced in solid tumors to reshape TME and is a powerful inducer for MDSCs, involving both signaling activation and metabolic reprogramming (21–23). And we also found that concentration of GM-CSF was significantly higher in the serum of LLC-tumor-bearing mice than that in naïve controls (Fig. 6B), and that the expression of Cldn12 was upregulated in MSC2 cells on both protein and mRNA levels after GM-CSF treatment (Fig. 6C and D; ref. 15). Moreover, expression of Cldn12 in BMDM cells was also induced by GM-CSF (Fig. 6E).

We further profiled the downstream signaling pathway of GM-CSF, and found that the activator of transcription 5 (STAT5), AKT and AMP-activated protein kinase (AMPK) pathway were activated by GM-CSF in MSC2 cells (Fig. 6F and G), which also consistent with previous findings (24). Furthermore, compared with untreated controls, GM-CSF significantly enhanced both glycolysis and mitochondrial respiration of MSC2 cells, resulted in elevated production of ATP (Fig. 6H and I). To further explore the requirements of signaling pathways in the upregulation of Cldn12 in MSC2 by GM-CSF, the kinase inhibitors were used and found that addition of the glucose analog 2-Deoxy-D-glucose (2-DG), an inhibitor of hexokinase-1, resulted in attenuation of the GM-CSF-induced upregulation of Cldn12 and AKT activation (Fig. 6J and K). In addition, inhibition of AKT, but not of STAT5 or AMPK, blocked GM-CSF–induced upregulation of Cldn12 (Fig. 6J, L, and M). These results suggested that the upregulation of Cldn12 by GM-CSF in AKT-dependent manner, and that glucose metabolism was involved in AKT activation.

To further investigate the regulation by GM-CSF of Cldn12-mediated homotypic interactions between MDSCs and ECs. The GM-CSF stimulation caused a large increase in the adhesion of MSC2 cells to sEND.1 cells, and which was blocked by cotreatment with either 2-DG or AKT inhibitor (Fig. 6N). These results suggested that GM-CSF promoted Cldn12 expression in MDSCs and their homotypic-binding with Cldn12 in ECs. In vivo, the LLC tumor cells with GM-CSF overexpression or control were subcutaneously burdened on the left or right flank of each mouse. Compared with LLC-Ctrl tumors, the LLC–GM-CSF tumors were lager with more MDSCs infiltrated, indicating the critical role of GM-CSF in MDSCs tumorous accumulation and tumor growth (Supplementary Fig. S11A–S11C).

In addition, to confirm the enhanced glucose metabolism and adhesion ability in MSC2 by GM-CSF stimulation, we performed quantitative proteomics analysis on primary MDSCs. And we found that glycolytic and cell–cell adhesion process ranked at enriched pathways (Supplementary Fig. S12A–S12B, Supplementary Table S2).

Finally, we identify the features of CLDN12⁺ MDSCs in human lung cancer, as well as in an alternative mouse lung-cancer model, based on the published single-cell RNA sequencing data (GSE127465) from human non–small cell lung carcinoma (NSCLC) and murine KP1.9 lung adenocarcinoma (25). The expression of CLDN12 was identified in human CD11b⁺CD33⁺ MDSCs and mouse CD11b⁺Gr1⁺ MDSCs (Supplementary Fig. S13A and S13B; ref. 26). Functional-pathway enrichment with Gene Ontology and the Kyoto Encyclopedia of Genes and Genomes were performed to compare the differentially expressed genes in CLDN12⁺ and CLDN12⁻ MDSCs (Supplementary Fig. S13C–S13E, Supplementary Tables S3–S4). The “leukocyte transendothelial migration” and “tight junction” terms were enriched in both mouse and human samples, supporting our hypothesis that CLDN12 is involved in the trafficking of MDSCs into tumors. In mouse samples, Cldn12⁺ MDSCs were also active in “oxidative phosphorylation” and the “PI3K–AKT signaling pathway,” which was consistent with our observation that GM-CSF stimulated metabolic and AKT pathways to regulate Cldn12 expression. Altogether, these results implied that Cldn12 facilitated transendothelial migration of MDSCs in both human and mouse lung cancers.

MDSCs mediated in situ immunosuppression impedes acute and chronic responses to immunotherapy (2, 27). Therefore, transmigration of MDSCs across vascular walls into tumor site is an immune suppressive concern, and a better understanding of the underlying mechanisms may help to develop novel combined immunotherapeutic strategies. We have now identified a new junctional molecule–mediated mechanism for MDSCs trafficking, in which homotypic interactions between Cldn12 in MDSCs and tumor blood vascular ECs facilitate tumorous infiltration of MDSCs. This was proven by the findings that systemic Cldn12 deficiency in vivo attenuated tumor growth via impaired MDSCs tumorous infiltration and further validated the important role of Cldn12-mediated homotypic interactions in transendothelial migration of MDSCs with bone marrow chimeric mice and Cldn12-KO MSC2 and sEND.1 cell line. To specifically and intrinsically strengthen the role of Cldn12 in the transendothelial migration of MDSCs, the mice with specific Cldn12 KO in myeloid cells and endothelial cells are needed in the future work.

Chemokine–chemokine receptor interactions are mainly responsible for the recruitment of MDSCs into tumors. The chemokine receptor CXCR2 antagonists and CCR5 neutralizing antibodies could inhibit the aggregation of MDSCs at tumor sites, and can be function as a combination strategy for tumor immunotherapy to enhance therapeutic effects (28, 29). In addition, blocking CSF1/CSF1R affects the infiltration of MDSCs into the TME and promotes antitumor effects of immunotherapy (30). GM-CSF, IL6, TNFα, and protein S100 family have also been reported to involve in MDSCs tumorous accumulation by participating into the development and expansion process of MDSCs (31–33). As presented by our data that expression of Cldn12 in MDSCs was upregulated by GM-CSF, which highlighted a new Cldn12-dependent mechanism for GM-CSF–driving tumorous MDSCs accumulation and leading to immunosuppressive tumor microenvironment. The roles of Cldn12 plays in the development, immunosuppression, proliferation, apoptosis or chemotaxis of MDSCs were not observed in this study, but more evidence maybe still needed.

Tight junction proteins, including Claudins, are primarily expressed on ECs and epithelial cells and form barriers to defend against the entry of bacteria and harmful inflammatory cells (7, 34). Whereas, the roles of Claudins play in ECs proliferation are rarely reported. We found that tumor vasculature establishment was defect in Cldn12^–/– mice compared with WT mice, which may be related with less MDSCs infiltrated into tumor tissue of Cldn12^–/– mice or subsequently weakened T cells derived IFNγ and TNFα stimulations based on our observation in this study. However, the detailed mechanisms are still needed to be further exploration.

The low-level expression of several tight junction proteins with unknown functions has been detected in myeloid cells (35). JAM-A has been reported to facilitate leukocyte diapedesis in various inflammatory models (36). Under inflammatory conditions, JAM-A redistribution in ECs from the intercellular area to the apical surface facilitates interactions in trans with JAM-A on leukocytes or with the beta 2-integrin lymphocyte function-associated antigen 1 (37). Our results suggested that the Cldn12 expressed in MDSCs had a similar role to JAM-A, in the formation of homotypic interactions with Cldn12 in ECs. However, it is still unknown that whether Cldn12 could form heterotypic interactions with other molecules or affect distribution of other junctional molecules are also unclear. In addition, the low-level expression of CLDN12 in lymphoid cells also has been reported (38–41), we believe that it will be very interesting to figure out the effect of Cldn12 deficiency on T cells functions. So far, we have not observed the significant difference in ability to secret IFNγ and TNFα or proliferation between CD4, CD8 T cells isolated from naïve WT and Cldn12^–/– mice after stimulation. And we proposed that Cldn12 is more likely to influence T cells antitumor responses by interfering MDSCs infiltration into tumor tissues.

Clinically, elevated numbers of MDSCs are positively correlated with cancer stages and tumor burdens in cancers, including NSCLC, melanoma, and breast cancer (2). And the amount of tumor-infiltrated MDSCs is predictive of responses to various cancer therapies, such as chemotherapy, or immunotherapy with ipilimumab (a monoclonal antibody against cytotoxic T-lymphocyte protein 4) or nivolumab (a monoclonal antibody targeting programmed cell death protein 1; refs. 2, 42). In comparison, tumor types that are more responsive to immune checkpoint blockade, such as B16F10 melanoma, exhibit less MDSCs tumorous infiltration (26, 43). Notably, results with mouse tumor models have shown that inhibition of MDSCs tumor-accumulation increases the therapeutic effect of immunotherapy (23, 28, 44). Strategies to block accumulation of MDSCs should, therefore, be urgently explored to enhance tumor therapeutic efficacy (27). Structure analysis of Cldn12 protein suggests that the extracellular loop 1 mainly has a barrier function and the extracellular loop 2 can mediate interaction between adjacent cells (12, 45), which strongly pointed to a therapeutic potential of targeting the extracellular loop 2 of Cldn12 by the peptiomimetic designed, bacterial toxins, or small-molecule inhibitor (46). Importantly, CLDN12 expression is relevant to transendothelial migration of MDSCs in human lung cancer, implying that CLDN12 may represent a promising target for immunotherapy of human lung cancers.

In summary, our results demonstrate an unidentified mechanism influencing the entry of MDSCs from peripheral blood into tumor sites, in which homotypic interactions between Cldn12 molecules expressed in MDSCs and ECs enable the transmigration of MDSCs across the endothelial monolayer. And the expression of Cldn12 in MDSCs was regulated by GM-CSF in an AKT-dependent manner. These findings suggest a promising strategy to stimulate antitumor immune responses and to enhance the efficacy of immunotherapy by interfering the entry of MDSCs into the tumor.

C. Ni reports grants from National Natural Science Foundation of China and Key Project of Medical Science and Technology of Henan Province during the conduct of the study. No disclosures were reported by the other authors.

H. Cao: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. C. Ni: Data curation, software, formal analysis, validation, investigation, visualization, writing–review and editing. L. Han: Data curation, investigation, methodology. R. Wang: Conceptualization, formal analysis, investigation, methodology, writing–review and editing. R. Blasig: Resources, methodology. R. Haseloff: Resources, formal analysis, validation, visualization. Y. Qin: Visualization, methodology. J. Lan: Data curation, software, methodology. X. Lou: Methodology. P. Ma: Data curation, methodology. X. Yao: Methodology. L. Wang: Methodology. F. Wang: Validation. L. Zhu: Visualization. N. Lei: Visualization. I.E. Blasig: Resources, validation, visualization. Z. Qin: Conceptualization, resources, supervision, funding acquisition, validation, project administration, writing–review and editing.

This work was supported by the National Science and Technology Major Project (grant number 2021YFA1201102); the National Natural Science Foundation of China (grant number 82073231 and 81902336); the Key Project of Medical Science and Technology of Henan Province (grant number SB201902019). Editorial assistance was provided by Dr. R. Phillips of Insight Editing London.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.