GTPase RAB31 supports ANXA1-driven macrophage education through clathrin-mediated endocytosis to suppress antitumor immunity Authors: Jingjing Liu, Shanshan Wang, Dengyi Bao, Ge Dong, Yunxi Ma, Guorong Zhang, Xin Liu, Dongli Zhang, Lulu Wang, Shuqian Xu, Zhigang Cai Published: 2026 ABSTRACT AbstractBackgroundTumor-associated macrophages (TAMs) play pivotal roles in shaping the tumor-microenvironment (TME) through functional plasticity, which is regulated by extrinsic and intrinsic signals. However, the role of vesicular trafficking in TAMs remains poorly understood. RAB31, a small GTPase enriched in myeloid cells, was proposed as a potential regulator of TAM polarization through clathrin-mediated endocytosis (CME). We hypothesized that RAB31 modulates TAM education by tumor-derived signals and thereby shapes antitumor immunity.MethodsWe profiled RAB31 expression in human cancers using public single-cell RNA sequencing (scRNA-seq) datasets and clinical sample immunofluorescence staining. Rab31 knockout mice were employed in subcutaneous tumor models. The TME was profiled by scRNA-seq, bulk RNA-seq, and flow cytometry. Bone marrow transplantation, adoptive cell transfer, and antibody-mediated depletion were performed to identify the effector cell populations. Co-immunoprecipitation coupled with mass spectrometry, receptor half-life assays, inhibitor intervention and lysosomal colocalization experiments dissected the molecular mechanism. Functional T-cell chemotaxis, activation, and anti-programmed death-ligand 1 (PD-L1) response assays were performed.ResultsRAB31 was highly expressed in TAMs across multiple cancers and correlated with poor prognosis and immunosuppressive TME. Rab31 deficiency reprogrammed TAMs to M1-like phenotype, enhanced CD8+ T-cell infiltration and activation, and suppressed tumor growth. Mechanistically, Rab31 preserves FPR2 cell surface stability through a CME-dependent mechanism; its loss redirected FPR2 to lysosomal degradation, disrupted tumor-derived ANXA1 signaling, and unleashed NF-κB activity. Rab31 deficiency synergized with anti-PD-L1 therapy in a CD8+ T cell-dependent manner.ConclusionsThese findings establish the CME/RAB31 pathway as an indispensable regulator of TAM polarization, underscoring the pivotal role of vesicular trafficking in the TME. WHAT IS ALREADY KNOWN ON THIS TOPIC Tumor-associated macrophages (TAMs) promote immunosuppression in the tumor microenvironment, but the role of vesicular trafficking in regulating TAM polarization remains poorly understood. RAB31 is a myeloid-enriched small GTPase, yet its function in TAMs and its impact on anti-programmed death-ligand 1 (PD-L1) therapy response have not been characterized. Tumor-associated macrophages (TAMs) promote immunosuppression in the tumor microenvironment, but the role of vesicular trafficking in regulating TAM polarization remains poorly understood. RAB31 is a myeloid-enriched small GTPase, yet its function in TAMs and its impact on anti-programmed death-ligand 1 (PD-L1) therapy response have not been characterized. WHAT THIS STUDY ADDS We demonstrate that RAB31 maintains FPR2 membrane stability via clathrin-mediated endocytosis to sustain tumor-derived ANXA1 signaling, thereby promoting M2-like TAM polarization and immunosuppression. RAB31 deficiency redirects FPR2 to lysosomal degradation, disrupts ANXA1 signaling, triggers NF-κB activity, and reprograms TAMs toward an M1-like antitumor phenotype. This CME/RAB31 axis enhances CD8+ T-cell recruitment and synergizes with anti-PD-L1 therapy. Elevated RAB31 expression correlates with poor prognosis and anti-PD-L1 resistance in patients with breast cancer. We demonstrate that RAB31 maintains FPR2 membrane stability via clathrin-mediated endocytosis to sustain tumor-derived ANXA1 signaling, thereby promoting M2-like TAM polarization and immunosuppression. RAB31 deficiency redirects FPR2 to lysosomal degradation, disrupts ANXA1 signaling, triggers NF-κB activity, and reprograms TAMs toward an M1-like antitumor phenotype. This CME/RAB31 axis enhances CD8+ T-cell recruitment and synergizes with anti-PD-L1 therapy. Elevated RAB31 expression correlates with poor prognosis and anti-PD-L1 resistance in patients with breast cancer. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY This study identifies RAB31 as a novel immunotherapeutic target linking endocytic trafficking to TAM reprogramming and checkpoint response. Targeting the CME/RAB31 pathway may overcome anti-PD-L1 resistance and convert immunologically “cold” tumors to “hot” tumors. Given the favorable safety profile of Rab31-deficient mice under steady-state conditions, RAB31-directed strategies warrant further clinical evaluation. This study identifies RAB31 as a novel immunotherapeutic target linking endocytic trafficking to TAM reprogramming and checkpoint response. Targeting the CME/RAB31 pathway may overcome anti-PD-L1 resistance and convert immunologically “cold” tumors to “hot” tumors. Given the favorable safety profile of Rab31-deficient mice under steady-state conditions, RAB31-directed strategies warrant further clinical evaluation. Introduction Tumor-associated macrophages (TAMs) exhibit functional plasticity within the tumor microenvironment (TME), yet their protumoral functions predominate in most solid malignancies. These pathogenic roles include facilitating tumor initiation, metastasis, angiogenesis, and immunosuppressive niche formation. The mechanisms include trophic support for cancer stem cells, extracellular matrix degradation, and T-cell exhaustion.1–4 Therapeutic targeting of TAMs represents a pivotal option in oncology, particularly given their dominant infiltration in the TME and major contribution to immunotherapeutic resistance and chemoresistance. Current strategies targeting TAMs include: (1) macrophage depletion, such as bisphosphonate and inhibitors targeting CSF-1R (PLX3397 and antibodies)5–7; (2) chemotaxis inhibition, such as the small molecule inhibitor PF-04136309 (targeting the CCL2-CCR2 chemotaxis axis)8–11; (3) functional reprogramming, such as CD47-SIRPα axis antibodies, PI3Kγ inhibitors and CD40 agonists.12–15 However, the molecular mechanisms underlying TAM polarization remain incompletely understood. In particular, the regulatory roles of intracellular endocytic trafficking pathways that mediate crosstalk between tumor cells and immune components have not been studied. Clathrin-mediated endocytosis (CME) is one of the canonical endocytosis processes and plays multifaceted roles in tumor progression.16 17 In general, CME-mediated receptor trafficking (ie, G protein-coupled receptors (GPCRs) and receptor tyrosine kinases) facilitates receptor desensitization, resensitization and signal transduction within the cell.18–20 For example, in tumor cells, the epidermal growth factor receptor (EGFR) phosphorylation triggers receptor internalization via CME and activates downstream signaling pathways such as the RAS/MAPK cascade, thereby regulating cell migration, cell cycle progression (including cell division, quiescence, and replication), and malignant transformation.21 In T cells, inhibiting CME results in impaired T-cell receptor internalization, thereby influencing the development and function of T cells.22 Nano-drugs targeting tumor or immune cell membrane receptors also rely on endocytosis to sustain their long-term efficacy.23 CME regulation of phagocytic cells (ie, macrophages) is also critical for pathogen immunity.24 However, to our knowledge, the specific role of CME in TAM plasticity has not been investigated. Members of the small GTPase RAB family serve as master regulators of phagocytosis and vesicular trafficking.25 26 Membrane internalization is the initial step for intercellular and intracellular cargo delivery, receptor recycling, and signal transduction. Dysregulation of RAB-mediated vesicle trafficking has been directly implicated in hematological diseases and cancers. For example, combined deficiency of Rab32 and Rab38 in mice results in symptoms reminiscent of Hermansky-Pudlak syndrome, which in patients is characterized by gray hair and extended bleeding time, and profoundly impairs hemostasis, suggesting a critical physiological role of RAB proteins in lysosome-related platelet dense granule production.27 Moreover, in tumor cells, inhibition of RAB5-mediated endocytosis and degradation of EGFR promotes the progression of glioma.28 In addition, RAB7 has been reported to regulate the progression of melanoma by regulating the lysosomal pathway.28 29 In macrophages, RAB7 promotes TLR4 degradation under inflammatory conditions, thereby limiting excessive inflammation.30 Under homeostatic conditions, macrophages lacking both Rab32 and Rab38 exhibit defective lysosomal function.31 Although previous studies on RAB family proteins have mainly focused on tumor cell-intrinsic functions or macrophage responses under inflammatory conditions, the function of RAB proteins in shaping TAM plasticity within the TME remains largely unknown. Indeed, only one recent report has implicated RAB37 in TAM-mediated tumor promotion, highlighting a critical knowledge gap.32 Our recent analyses of TCGA (The Cancer Genome Atlas) leukemia datasets suggest that the expression of RAB31 (RAB22B) effectively distinguishes monocytic leukemia from primitive leukemia, reflecting differential expression levels of RAB31 in naive and mature myeloid cells.33 Two additional stratification markers identified in the same analysis, LRRC25 and NCF2, have also been reported to play important roles in pathogen and tumor immunity,34 35 but the roles of RAB31 in immunity remain unknown. RAB31 in tumor tissue has been recognized as a prognostic marker in several cancer types, such as breast cancer.36 Within the tumor microenvironment, RAB31 has been suggested in regulating the functions of cancer-associated fibroblasts (CAFs).37 By profiling the expression of 64 RAB family members in human bone marrow, we observed that RAB31 is predominantly expressed in myeloid immune cells, particularly macrophages and neutrophils (online supplemental figure S1). Although other RAB family members such as RAB20, RAB32, and RAB34 share a similar myeloid-lineage enrichment pattern in bone marrow tissue, they are also highly expressed in malignant and normal epithelial cells (online supplemental figure S1 and S2). These gene-expression observations promoted us to hypothesize that RAB31 may exert a unique immunological function in TME. In this study, combining results from human clinical samples and murine tumor models, we show that RAB31 deficiency reprograms TAMs from protumor to antitumor activity. Our findings, for the first time, establish a link between RAB31/endocytosis and the education of TAMs by the tumor-cell-derived immunosuppressive signal ANXA1, thus highlighting RAB31 as a novel immunotherapeutic target in tumor treatment. Results Myeloid-enriched RAB31 facilitates tumor progression in the TME In our previous study, we identified the myeloid-enriched immune signaling molecule RAB31, a member of the small RAS GTPase family, as a biomarker capable of stratifying acute monocytic leukemia into primitive myeloid leukemia, with significant clinical implications.33 RAB31 manifests a myeloid cell-specific expression pattern within the immune system (online supplemental figure S3A). Previous studies suggested that RAB31 also has a prognosis value in solid tumors.38–40 By analyzing 33 cancer types from the TCGA database, we also found that RAB31 expression was significantly elevated in tumor tissues compared with adjacent normal tissues in 8 cancer types (online supplemental figure S3B). Moreover, high RAB31 expression was associated with poor survival outcomes in patients with uveal melanoma (UVM), stomach adenocarcinoma (STAD), colon adenocarcinoma (COAD), and pancreatic adenocarcinoma (PAAD) (figure 1A). Interestingly, single-cell RNA sequencing (scRNA-seq) data from cancer patients revealed that RAB31 was highly expressed in TAMs and almost not expressed in tumor cells (figure 1B–E, online supplemental figure S3C). Myeloid-enriched RAB31 facilitates tumor progression in the TME. (A) Prognosis of patients with the indicated cancers stratified by expression of RAB31. Survival curves for patients with high and low expression of RAB31 are shown and compared (RAB31High, dark green line; RAB31Low, orange line). Source: the TCGA pan-cancer database. UVM, STAD, COAD and PAAD are highlighted. (B–E) RAB31 is predominantly expressed in myeloid-lineage cells but also minimally expressed in CAFs. The connecting lines in COAD and STAD denote paired samples of adjacent non-tumor tissues and tumor tissues from the same patient. Source: scRNA-seq datasets from the GSA or GEO databases as indicated. Cell types are annotated according to established markers. Data are presented as mean±SEM. (F) Representative images stained with indicated markers (left panel) and quantitative analysis (right panel). Immunofluorescence staining and immunohistochemistry were performed to visualize RAB31 (red), CD68 (yellow), CD163 (green), and CD8 (DAB staining) in tumor tissues of patients with colorectal cancer. CD68 was used to label macrophages, CD163 for M2-type macrophages, and CD8 for CD8+ T cells. Staining images are divided into two groups for comparison (RAB31high, dark green dots; RAB31low, orange dots). Mean fluorescence intensity of RAB31 (protein expression per cell), CD68-positive or CD163-positive area (proportion of positive cells), and CD8-positive area were quantified using ImageJ. The number of images included in the analysis was n=18; the number of patients with CRC in the cohort was N=6. (G–I) Monitoring growth of tumors implanted in Rab31−/− or WT mice. We conducted subcutaneous tumor models to examine the general role of Rab31 in TME. Models were established in WT and Rab31−/− hosts using cell lines including MC38 (mouse colon cancer cells), B16F10 (mouse melanoma cells), and KPC cells (mouse pancreatic cancer cells). Tumor growth and survival curves are monitored. Significant differences in tumor volume were assessed using two-way ANOVA, while significant differences in survival rates were evaluated using the Breslow-Wilcoxon test. N=6 per group, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; CAF, cancer-associated fibroblast; COAD, colon adenocarcinoma; CRC, colorectal cancer; IHC, immunohistochemistry; PAAD, pancreatic adenocarcinoma; SATD, stomach adenocarcinoma; sc., subcutaneous; scRNA-seq, single-cell RNA sequencing; TCGA, The Cancer Genome Atlas; TME, tumor microenvironment; UVM, uveal melanoma; WT, wild-type. In previous studies of TAMs, it has been well documented that this population generally acts as “accomplices” of tumor cells, supporting the growth and also immune evasion of the malignant cells. Interestingly, across multiple cancer types, we observed that the infiltration extent of TAM positively correlates with the expression level of RAB31 (online supplemental figure S4). This association is likely attributable to the intrinsically dynamic expression of RAB31 within TAMs. To further investigate whether RAB31 participates in perturbing the TME, we analyzed its expression patterns in tumor and normal tissues from patients with PAAD, COAD, STAD, and UVM. With the exception of UVM, for which no healthy controls were available, scRNA-seq data revealed that RAB31 expression in TAMs was markedly elevated compared with that in macrophages from normal tissue (figure 1B–E). Notably, in COAD and STAD, paired-sample analysis demonstrated that RAB31 expression was further induced during the transition of macrophages from adjacent normal tissues to tumor sites (figure 1D,E). Additionally, in pancreatic cancer, RAB31 was upregulated in CAFs relative to fibroblasts from normal tissue (online supplemental figure S5A). We validated this interesting correlation using clinical samples from our institute. Immunofluorescence analysis of human colorectal carcinoma specimens demonstrated that robust RAB31 expression predominantly localized to TAMs, with negligible expression in other cell types. Consistently, the majority of infiltrating macrophages displayed an M2 immunophenotype (CD163+) (left panel of figure 1F). Stratification of patients based on the average expression of RAB31 in macrophages revealed that the RAB31high cohort was characterized by enhanced infiltration of immunosuppressive M2 macrophages and concomitant reduction in cytotoxic CD8+ T lymphocytes, compared with the RAB31low cohort (right panel of figure 1F). These observations collectively indicate that robust RAB31 expression within the TAMs may impair CD8+ T lymphocytes infiltration and facilitate malignant progression, and potentially vice versa. To assess the functional contribution of RAB31 in the TME, we employed subcutaneous tumor models using wild-type (WT) and Rab31 knockout (Rab31−/−) mice. RAB31 is highly conserved between humans and mice (94.4% similarity in protein sequence; online supplemental figure S5B). Generation and validation of the Rab31−/− strain are shown in online supplemental figure S5C-F. We evaluated tumor progression using three murine cancer cell lines: MC38 (colon), KPC (pancreas) and B16F10 (melanoma). As shown in figure 1G–I and online supplemental figure S5G, tumor growth was significantly attenuated and survival of the host mice was improved in Rab31−/− compared with WT. An exception was observed in the MFC model, where spontaneous tumor regression occurred irrespective of host genotype. Consistently, in a metastatic experimental model, the number of B16F10 pulmonary metastatic nodules was significantly diminished in the host of Rab31−/− relative to in that of WT littermates (online supplemental figure S5H). In summary, these results demonstrate that RAB31 is selectively upregulated in TAMs of multiple solid tumor types and functionally promotes tumor progression in tumor-bearing models. scRNA-seq analysis reveals an immunostimulatory TME induced by Rab31 deficiency For unbiased assessment of the impact of Rab31−/− on the TME, we performed scRNA-seq on CD45+ immune cells isolated from three B16F10 tumors per group, which were mixed with unsorted tumor tissue cells at a 5:1 ratio. Four major cell populations were identified: lymphoid cells, myeloid cells, stromal cells, and melanoma cells. TAMs and monocytes together accounted for ~50% of immune infiltrate. Subclustering divided these TAMs/monocytes into two major subsets: TAM&MONO1 (M1-like, antitumor) and TAM&MONO2 (M2-like, protumor) (figure 2A, online supplemental figure S6A and B). Consistent with human tumor data, Rab31 was predominantly expressed in TAMs/monocytes and showed higher enrichment in M2-like than in M1-like macrophages (figure 2B, online supplemental figure S6C). In the Rab31−/− hosts, the fraction of the M2-like subset was reduced, accompanied by an expansion of cytotoxic T cells and a reduction of CAFs (figure 2C). Gene-set-based functional scoring revealed increased M1-like and decreased M2-like signatures in Rab31−/− TAMs/monocytes (figure 2D), and elevated effector/cytotoxic scores in T cells (figure 2E). Interestingly, tumor cells from Rab31−/− mice showed reduced copy-number variants (CNV) scores, suggesting lower degree of malignancy (online supplemental figure S6D). The scRNA-seq analysis indicated a shift toward an immunostimulatory state in the Rab31−/− TME, compared with the controls. scRNA-seq analysis reveals an immunostimulatory TME induced by Rab31 deficiency. (A) Flowchart of scRNA-seq experimental procedure for profiling subcutaneous B16F10 tumor tissues by scRNA-seq analysis. Left panel, a scheme showing that tumors from three WT or Rab31−/− host mice were dissociated into single-cell suspensions, pooled, and subjected to magnetic bead-based cell sorting (CD45+). The CD45+ cells were then mixed with unsorted cells in a 5:1 ratio before proceeding to library construction and RNA sequencing. Right panel, the UMAP visualization of all cells from the scRNA-seq datasets (in total, n=19,093 cells; for tumor tissues from the WT host, n=10,356; for tumor tissues from the Rab31−/− host, n=8,787). We annotated 13 clusters, which belong to four main cell compartments: myeloid cells, lymphoid cells, stromal cells (these cells belong to the TME), and melanoma (these cells are the implanted tumor cells). (B) UMAP plots illustrating the expression of the marker genes for TAMs and monocytes in the TME (WT dataset only). Cd68, for monocytes and TAMs; Tnf, for M1-like TAM; Mrc1, for M2-like TAM. Notice that Rab31 is predominantly myeloid-lineage enriched, and especially biased toward M2-like TAM. See also online supplemental figure S6B for the geneset scoring of M1 and M2 polarization states in these two macrophage populations. (C) The stacked bar chart showing the proportion of each cluster in the TME. (D) Loss of Rab31 altered the M1/M2 balance of TAMs in TME. The M1-like and M2-like scores in TAMs and monocytes were calculated by the AddModuleScore function in Seurat (see Methods for details). (E) T cells in the Rab31−/− host appear to have increased effector function and cytotoxicity scores. The scores in T cells were calculated by the AddModuleScore function in Seurat. (F–G) Validation of the initial scRNA-seq results by bulk RNA-seq analysis of CD45+ cells. The CD45+ cells were sorted from B16F10 subcutaneous tumor tissues. (F) Upregulated genes (ie, Cxcl9, marker for M1-like TAM) are highlighted in red while downregulated genes (ie, Cd163, marker for M2-like TAM) are in blue. (G) The bubble plots display the dysregulated pathways. The upregulated genes in F were used for the enrichment analysis. (H–J) Flow cytometry was conducted to verify the proportions of various immune cell compartments suggested by scRNA-seq dataset analysis, n=4–6 per group. See online supplemental figure S6 for gating strategies and representative flow profiles. Data are presented as mean±SEM. P values were determined by paired t-tests. CAF, cancer-associated fibroblast; CTL, cytotoxic T lymphocyte; DC, dendritic cell; MHC, major histocompatibility complex; NK, natural killer; pDC, plasmacytoid dendritic cell; scRNA-seq, single-cell RNA sequencing; TAM, tumor-associated macrophage; TAN, tumor-associated neutrophil; TME, tumor microenvironment; UMAP, Uniform Manifold Approximation and Projection; WT, wild-type. We further validated these initial findings by bulk RNA-seq and flow cytometry using CD45+ cells from the TME of Rab31−/− and WT mice. RNA-seq confirmed upregulated expression of immunostimulatory markers, including T-cell chemokines (Cxcl9), lymphocyte cytotoxic factors (Cd8a, Gzmb, Nkg7, Ifng), M1 markers (Nos2, Gbp4/5), and interferon-stimulated genes (Ifi47), while immunosuppressive M2 markers (Cd163, Lyve1) were downregulated. The upregulated genes were enriched in pathways associated with immune activation, including T-cell activation, antigen presentation, and tumor necrosis factor (TNF) signaling pathways (figure 2F,G). Flow cytometry profiling confirmed that Rab31−/− TME manifested enhanced Cd45+ infiltration (p=0.0008), decreased Cd11b+ cells (p=0.0061), and increased Cd3+ cells (p=0.0407). In addition, the fraction of total monocytes and macrophages in the Cd45+ pool remained unchanged, but M1-like macrophages increased (Cd86+, p<0.0001; iNOS+, p=0.0059), and M2-like macrophages decreased (Cd163+, p=0.0051; Mrc1/Cd206+, p=0.0125) in the Rab31−/− TME. Moreover, neutrophils were reduced (p=0.0207), and both Cd4+ (p=0.0023) and Cd8+ T cells (p=0.0532) increased in Rab31−/− TME (figure 2H–J, online supplemental figure S6E and S7A and B). Major histocompatibility complex (MHC)-I expression was also elevated in Rab31−/− TAMs (online supplemental figure S7C). In similar tumor-bearing experiments using the cold tumor model KPC (Pdx1-Cre-mediated KrasG12D plus p53R172H mutation to induce pancreatic ductal adenocarcinoma (PDAC) tumors in mice), we examined immune cell infiltration. Consistently, we found that Rab31 deficiency shifted the immune microenvironment from an immunosuppressive to an immunostimulatory state in this KPC model (online supplemental figure S8). In conclusion, our single-cell profiling and experimental validations demonstrate that Rab31 deficiency remodels the TME into an immune-active state, characterized by: (1) enhanced cytotoxic T-cell responses, (2) increased M1-like polarization, and (3) reduced CNV in tumor cells. Therefore, Rab31 deficiency establishes an antitumor TME that constrains tumor progression. Rab31-deficient macrophages are the primary antitumor effectors in the Rab31−/− TME As shown in figures 1B–E and 2A,B, although RAB31 is predominantly expressed in myeloid cells, it is expressed in CAFs, another key stromal component in the TME. CAFs are known to promote tumor progression through immunosuppressive mechanisms. To identify the primary cellular source of antitumor immunity in Rab31−/− TMEs, we performed bone marrow transplantation (BMT) experiments in chimeric mice (figure 3A–H), together with adoptive cell transfer (ACT) experiments in which tumor cells were mixed with distinct immune cell subsets (figure 3I–P). Rab31-deficient macrophages are the primary antitumor effectors in the Rab31−/− TME. We first constructed chimeric mice to demonstrate that Rab31−/− BM cells rather than CAFs are critical for the antitumor function (A-H); we then performed adoptive cell transfer experiments to clarify that Rab31−/− macrophages (BMDMs) rather than NEs are critical for the antitumor function (I-P). (A–D) To assess the involvement of stromal cells, we compared the growth of tumor established in WT→WT or WT→Rab31−/−. WT→WT: WT BM cells were transplanted into WT recipients; WT→Rab31−/−: WT BM cells were transplanted into Rab31−/− recipients. WT or Rab31−/− mice were irradiated with a lethal dose of γ-rays to deplete their bone marrow cells and then transplanted with bone marrow cells from WT mice (WT→WT or WT→Rab31−/−). 1 month after transplantation, the B16F10 subcutaneous tumor model was established, and tumor volume changes were monitored. On day 12 post-inoculation with B16F10 cells, tumors were excised and weighed. (E–H) To assess the involvement of BM-derived immune cells, we compared the growth of tumor established in WT→WT or Rab31−/−→WT. Rab31−/−→WT: Rab31−/− BM cells were transplanted to WT recipients. Notably, we observed significant alterations in tumor growth in Rab31−/−→WT chimeric mice compared with that in →. (I–L) B16F10 cells were mixed with WT or Rab31−/− NEs (at a 1:1 ratio) and injected into WT host mice subcutaneously. On day 5 and day 9 post inoculation of B16F10 cells, additional NEs were injected intratumorally to ensure adequate NEs were present in TME for testing. (M–P) B16F10 cells were mixed with WT or Rab31−/− BMDMs (at a 1:1 ratio) and injected into WT host mice subcutaneously. On day 5, additional BMDMs were injected intratumorally to ensure adequate BMDMs were present in TME for testing. (Q–R) B16F10 tumor-bearing mice were treated with anti-Ly6G (Q) or anti-CSF1R (R) to deplete neutrophils or macrophages, respectively. Tumor weights were measured at day 12 post-inoculation. P values of tumor volume were assessed using two-way ANOVA, n=6 per group. Data are presented as mean±SEM. P values of tumor weight were assessed using paired t-tests. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; BM, bone marrow; BMDM, bone marrow-derived macrophage; CAF, cancer-associated fibroblast; NEs, neutrophils; TME, tumor microenvironment; WT, wild-type. As shown in figure 3A–D, no significant differences in tumor growth were observed between WT→WT (WT bone marrow→WT recipient) and WT→Rab31−/− (WT bone marrow→Rab31−/− recipient) chimeras, indicating that loss of Rab31 in non-hematopoietic compartments (eg, CAFs) does not account for the antitumor effect. In contrast, Rab31−−→WT (Rab31−/− bone marrow→WT recipient) chimeras exhibited a significant suppression of tumor growth compared with WT→WT controls, indicating that Rab31-deficient bone marrow-derived cells (ie, immune cells) mediate the antitumor response (figure 3E–H). Furthermore, in ACT assays, we demonstrated that macrophages, rather than neutrophils, serve as the primary effectors of Rab31-deficiency-driven tumor suppression (figure 3I–P). To further test the myeloid cell-type dependency of Rab31 deficiency-mediated antitumor immunity, we employed well-established depletion strategies using anti-CSF1R antibodies (to deplete macrophages) and anti-Ly6G antibodies (to deplete neutrophils) in tumor-bearing mice. The results indicated that depletion of macrophages, but not neutrophils, was capable of abrogating the tumor-suppressive phenotype in the Rab31-deficient TME (figure 3Q,R). These pharmacological results were consistent with the aforementioned ACT results using mixed cell populations. Taken together, the results from BMT, ACT, and antibody-mediated depletion experiments demonstrate that macrophages are the principal mediators of the immune-activated TME induced by Rab31 deficiency. Rab31 deficiency reshapes the diversity of TAMs by reducing the Dab2+ subpopulation and promoting antitumor polarization Given the broadly elevated expression of Rab31 in monocyte/macrophage populations across multiple cancers and its co-localization with protumorigenic M2 macrophage markers (figure 2B), we hypothesized that Rab31 drives tumor-favoring polarization, that is, the M2-like state. From an in vitro assay, we first validated that stimulation with tumor-conditioned media (TCM)—derived from MC38, B16F10 and KPC cell lines—robustly induced expression of Rab31 in bone marrow-derived macrophages (BMDMs) (figure 4A). Similar induction of RAB31 by TCM was observed in human macrophages (online supplemental figure S9A). In the scRNA-seq datasets, pseudotime trajectory analysis revealed two terminal TAM differentiation states: an M1-like branch and an M2-like branch. Notably, WT TAMs were predominantly skewed toward the M2-like trajectory (figure 4B,C), consistent with their tumor-promoting phenotype. In contrast, Rab31−/− TAMs showed a shift toward the M1-like state (figure 4C). In Rab31−/− TAMs, both Cxcl9 and MHC class I (H2) were significantly upregulated (online supplemental figure S9B). Gene Ontology analysis revealed that upregulated genes in Rab31−/− TAMs were enriched in pathways related to type I interferon response, interleukin-1 signaling, antigen presentation, and T-cell chemotaxis, whereas downregulated genes were associated with Rab31-mediated vesicular trafficking and endocytosis pathways (online supplemental figure S9C). CellChat analysis demonstrated significantly enhanced macrophage-to-T-cell communication in the Rab31−/− TME compared with the WT TME, particularly in interactions related to antigen presentation (eg, H2-Cd8) (online supplemental figure S9D). Loss of Rab31 reshapes the diversity of TAMs subsets and promotes the M1-like trajectory. (A) Signals from tumor cells stimulate expression of Rab31. BMDMs were cultured with basal medium or with TCM. qRT-PCR assays were performed. n=3 per group. Data are presented as mean±SD. (B–C) The developmental trajectories of TAMs and monocytes were analyzed using the tool Monocle2 (balanced for equal cell numbers). (B) branches and marker genes for M1-like and M2-like TAMs and monocytes. (C) Branches and cell density for M0, M1 and M2 states. (D) Following TCM stimulation, functional alterations of BMDMs were revealed by qRT-PCR using key marker genes. In the heatmap, Z-score normalization was performed for each gene to highlight differences in the WT and Rab31−/−. N=3 per group. (E–F) Experimental verification of TAM polarization in the Rab31−/− host compared with that in the WT host. Expression of tumor-suppressive markers ROS, iNOS, MHC-I/II and tumor-promoting markers Mrc1/Cd206 and Cd163 determined by flow cytometry assays. N=3 per group. Data are presented as mean±SD. (G) Flow cytometric quantification of SIINFEKL surface expression. WT and Rab31−/− BMDMs were treated with 5, 10 and 20 µg/mL OVA. Data are presented as mean±SD. P values were determined by paired t-tests. (H–I) T-cell responsiveness (H) and proliferation (I) on macrophage stimulation. Cartoon shows the experimental procedure (for details see the Methods). Data are presented as mean±SD. P values were determined by paired t-tests. (J) Following TCM stimulation, functional alterations of THP-1-derived macrophages were revealed by qRT-PCR using key marker genes. In the heatmap, Z-score normalization was performed for each gene to highlight differences in the WT and Rab31−/−. N=3 per group. (K) Subclusters of TAMs and monocytes. The highly variable genes for each subcluster were listed for each subcluster. In total, n=6912 cells; in the WT host, n=4109; in the Rab31−/− host, n=2803. (L) Stacked bar chart of TAM subset proportions. (M) The expression level of Dab2 in TAMs in the scRNA-seq dataset. P values were determined by paired t-tests. (N–O) qRT-PCR assay (N) and western blot assay (O) were performed on the sorted TAMs from the B16F10 tumor tissue (n=3 per group). Data are presented as mean±SD. (P) Immunohistochemistry staining of CME-related regulator Dab2 using paraffin sections of B16F10 tumor tissue. Scale bar: 200 µm; inset scale bar: 50 µm. N=3 per group. Data are presented as mean±SEM. (Q) Patient survival curves were stratified by the infiltration levels of DAB2+TAMs (DAB2+TAMhigh vs DAB2+TAMlow). Based on quartile stratification, patients were ranked by marker gene scores of DAB2+ TAMs, with the highest quartile designated as the high-infiltration group and the lowest quartile as the low-infiltration group. P values were determined by paired t-tests. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. BMDM, bone marrow-derived macrophage; CME, clathrin-mediated endocytosis; iNOS, inducible nitric oxide synthase; MFI, mean fluorescence intensity; MHC, major histocompatibility complex; OVA, ovalbumin; qRT-PCR, quantitative reverse transcription polymerase chain reaction; ROS, reactive oxygen species; scRNA-seq, single-cell RNA sequencing; TAM, tumor-associated macrophage; TCM, tumor-conditioned media; WT, wild-type. To assess the role of Rab31 in macrophage polarization, we performed TCM-based co-culture assays using BMDMs from WT and Rab31−/− mice, followed by flow cytometry and qRT-PCR (quantitative reverse transcription polymerase chain reaction) assays. We analyzed the expression of tumor-supportive and tumor-suppressive genes in TAMs isolated from B16F10 tumors (online supplemental figure S9E) and in TCM-stimulated BMDMs (figure 4D). Z-score normalized qRT-PCR confirmed that loss of Rab31 impairs M2 polarization in both in vivo and in vitro settings. Flow cytometric analysis revealed that on stimulation with TCM, Rab31-deficient BMDMs exhibited markedly increased production of reactive oxygen species and inducible nitric oxide synthase (iNOS), accompanied by enhanced expression of antigen-presenting molecules MHC-I and MHC-II, while the expression of M2 markers (Cd163 and Cd206) was decreased (figure 4E–F, and online supplemental figure S9F and G). To further investigate the reprogramming of macrophages on Rab31 deficiency, we performed in vitro assays to assess antigen presentation, T-cell activation, and T-cell suppression. The results demonstrated that Rab31−/− BMDMs exhibited enhanced antigen presentation of exogenous OVA (ovalbumin) (figure 4G); accordingly, their capacity for T-cell activation was increased (as evidenced by an increased number of interferon-γ+ T cells), whereas their suppressive function was attenuated (as reflected by reduced T-cell proliferation, that is, fewer BrdU+ cells) (figure 4H,I). We performed similar assays in PMA-differentiated THP-1 cells (sgControl vs sgRAB31), followed by qRT-PCR analysis (figure 4J), further supporting the regulatory role of RAB31 in M1/M2 reprogramming. Re-clustering analysis of TAM and monocyte populations in our scRNA-seq dataset identified one monocyte cluster and eight TAM subsets (transcriptome defined): Mki67+ TAM (proliferative), Folr2+ TAM (tissue-resident), Dab2+ TAM (immunosuppressive, anti-inflammatory), Cxcl9+ TAM (T-cell recruiting, antigen-presenting), Isg+ TAM (interferon-responsive), Mmp12+ TAM (tissue repair, immunosuppressive), Spp1+ TAM (dual pro-inflammatory and anti-inflammatory), Cdk8+ TAM (metabolically active) (figure 4K, online supplemental figure S10A). In the Rab31−/− TAMs and monocytes, tumor-promoting M2-like subpopulations (eg, Mmp12+ and Dab2+ TAMs) were notably decreased, while tumor-suppressive M1-subsets (eg, Isg+ and Cxcl9+ TAMs) were enriched (figure 4L). Genes associated with antigen presentation, lymphocyte chemotaxis, and inflammation were broadly upregulated across all TAM subsets (online supplemental figure S10B). Intriguingly, we also noticed that the fraction of Dab2+ TAM subset is dramatically decreased in the TAM clusters of Rab31−/− TME (figure 4L). The Dab2+ TAM subset is characterized by high expression of membrane-trafficking-related genes, including Dab2 (Disabled-2), Clta (Clathrin light chain A), Stab1 (Stabilin-1), and Cd81 (figure 4K). Dab2 is a well-characterized adaptor protein involved in CME. Within the tumor tissue, Dab2 is specifically enriched in TAMs (online supplemental figure S10C). Expression of Dab2 was significantly decreased in Rab31−/− TAM (figure 4M). Furthermore, decreased levels of Dab2 in tumors from Rab31−/− mice were validated by qRT-PCR, western blotting, and immunohistochemistry (figure 4N–P). To assess the clinical relevance of this finding, we examined the association between infiltration levels of DAB2+ and patient prognosis across four publicly available cancer cohorts. Patients with high infiltration of DAB2+ exhibited poorer overall survival, regardless of whether the cohort was stratified by median or quartile stratification (figure 4Q, online supplemental figure S10D). To experimentally validate the role of Dab2+ TAMs, we transfected BMDMs with Dab2-targeting siRNA. Under TCM stimulation, Dab2 knockdown skewed macrophages toward a protumorigenic phenotype (online supplemental figure S10E and F). Collectively, Dab2+ TAMs can be regarded as a protumorigenic macrophage subset. Of note, Dab2 was reported to suppress the NF-κB signaling pathway in macrophages.41 Here we noticed that an expression score based on several NF-κB target genes was maintained at a higher level than WT controls throughout the different stages of TAM differentiation (online supplemental figure S10G). Moreover, pySCENIC analysis revealed that the activity of the downstream transcription factor NF-κB was enhanced in Rab31−/− TAMs (online supplemental figure S10H-J), accompanied by higher signature scores (downstream target gene-based) compared with WT counterparts (online supplemental figure S10K). In summary, the loss of Rab31 reduced the proportion of Dab2+, a subset with protumorigenic properties. The CME pathway is required for Rab31-mediated polarization of TAMs By what mechanism does Rab31-deficiency reshape TAM diversity, specifically the pronounced depletion of the Dab2+ subset? To further dissect the Rab31 interactome, we established a macrophage model stably overexpressing FLAG-tagged Rab31 and conducted co-immunoprecipitation (co-IP) followed by mass spectrometry (MS) (figure 5A). As shown in figure 5B, Rab31 interacted with Phosphatidylinositol-binding clathrin assembly protein (Picalm) and Dynamin 2 (Dnm2), and these interactions were further enhanced on stimulation with TCM. Picalm and Dnm2 are both key components of the CME machinery. PICALM functions as an adaptor that facilitates the recruitment of Clathrin and AP-2 to the plasma membrane, regulates vesicle formation, and coordinates the trafficking of specific cargoes through CME.42 DNM2, a large GTPase, mediates the final scission of Clathrin-coated vesicles from the plasma membrane, thereby completing vesicle internalization.43 These interactions were further validated by Western blotting assays, which demonstrated an interaction between Rab31 and Picalm that was enhanced on TCM stimulation (figure 5C). Interestingly, analysis of human single-cell transcriptomic datasets revealed strong positive correlations between RAB31 and the CME-related genes DNM2 and PICALM, indicating that these endocytosis-related molecules may function in a symphonic manner within TAMs (figure 5D). We examined the members involved in the CME pathway (KEGG (Kyoto Encyclopedia of Genes and Genomes) database) and found that several components were downregulated in Rab31−/− TAMs (online supplemental figure S11A), red stars for those downregulated). These results are consistent with the observation that the Dab2+ TAM subset—characterized by the expression of CME-related markers—was markedly decreased in Rab31−/− TAMs (+). A schematic illustration is provided to depict that Rab31, Dnm2, Picalm, and Dab2 cooperatively participate in the process of CME (online supplemental figure S11B), based on current literature.44–46 The CME pathway is required for Rab31-mediated polarization of TAMs. (A) Co-IP experimental workflow for screening Rab31-interacting proteins. Rab31 overexpression (with a FLAG tag) was first established in mouse macrophage cell line RAW264.7, which was then cultured in either normal medium or MC38 TCM for 24 hours. FLAG-tagged proteins were then immunoprecipitated using anti-FLAG magnetic beads and the precipitated products were subjected to mass spectrometry analysis. (B) The interaction between Rab31 and endocytic pathway-related proteins was investigated using the Co-IP/MS dataset. Interaction levels (range: 0–1) were normalized to levels observed with TCM stimulation for each protein (n=3 per group). The proteins Picalm and Dnm2 (highlighted in red) are known to be mediators of CME. (C) Interaction between Rab31 and Picalm was verified by Co-IP assays. (D) The correlation between RAB31 and DNM2 or between RAB31 and PICALM in TAMs was examined by the human pancreatic cancer scRNA-seq dataset CRA001160. Correlation was assessed using the Pearson correlation coefficient (rpearson). (E–G) Intratumoral injection of the small molecule drug CPZ (an inhibitor of CME) reversed the effect of Rab31−/− in suppressing the growth of subcutaneous B16F10 tumors. P values of tumor volume were assessed using two-way ANOVA. P values of tumor weight were assessed using paired t-tests. N=6 per group. Data are presented as mean±SEM. (H) Flow cytometry analysis of Mrc1/Cd206 (MFI) of BMDMs following TCM stimulation. Data are presented as mean±SD. (I) Relative mRNA expression of M1 (Cd86, Il1b, Cxcl9, Cxcl10) and M2 (Cd206, Cd163) markers of BMDMs following TCM stimulation as determined by qRT-PCR, N=3 per group. Data are presented as mean±SD. (J) Schematic diagram of the experimental procedure. B16F10 cells were mixed with BMDMs (WT and Rab31−/−) transiently transfected with siRNA for 48 hours, then combined with 50% matrigel and inoculated subcutaneously into WT mice. Additional BMDMs were injected on day 5 and day 8. (K) Tumor weight was measured at day 12 as the experimental endpoint. Data are presented as mean±SEM. N=6 per group. (L–M) qRT-PCR assays were conducted to examine the effects of different genetic perturbations on the pro-inflammatory, antigen presentation, T-cell chemotaxis, anti-inflammatory, and tissue repair functions of BMDMs (L). Flow cytometry was performed to analyze the effects of different genetic perturbations on the expression of M2 marker Mrc1/CD206, M1 marker iNOS, and antigen-presenting molecule MHC-I in BMDMs following TCM stimulation (M). Both siCtrl and siCltc refer to siRNA transfection following TCM stimulation at 48 hours after transfection. Pitstop2 was used at a concentration of 30 µM. Heatmaps were generated using GraphPad software, with each colored block representing the mean expression value of three independent replicates. N=3 per group. Data are presented as mean±SD. P values were determined by paired t-tests. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; BMDM, bone marrow-derived macrophage; CME, clathrin-mediated endocytosis; Co-IP, co-immunoprecipitation; CPZ, chlorpromazine; iNOS, inducible nitric oxide synthase; MFI, mean fluorescence intensity; MHC, major histocompatibility complex; mRNA, messenger RNA; MS, mass spectrometry; PDAC, pancreatic ductal adenocarcinoma; qRT-PCR, quantitative reverse transcription polymerase chain reaction; sc., subcutaneous; scRNA-seq, single-cell RNA sequencing; TAM, tumor-associated macrophage; TCM, tumor-conditioned media; WT, wild-type. To evaluate the functional involvement of CME in Rab31-mediated immune modulation, we treated tumor-bearing Rab31−/− mice with chlorpromazine (CPZ), an inhibitor of CME. Strikingly, on treatment with CPZ, Rab31−/− mice no longer exhibited the tumor-suppressive phenotype (figure 5E–G). Considering that CPZ may affect the viability of tumor cells themselves, we isolated BMDMs for in vitro assays. Flow cytometry and qRT-PCR assays showed that, in the presence of CPZ, Rab31−/− BMDMs did not exhibit a further reduction of M2 polarization compared with WT, indicating that CME is a key pathway through which Rab31 regulates TAM behavior (figure 5H,I). We also assessed the effects of CPZ on tumor cells. At a low concentration (<20 µM), CPZ did not alter programmed death-ligand 1 (PD-L1) expression on B16F10 and MC38 cells (online supplemental figure S11C). CCK-8 (cell counting kit-8) assays showed that the IC₅₀ (half maximal inhibitory concentration) values of CPZ at 24 hours/48 hours are 95.46/55.79 µM (B16F10) and 61.18/33.49 µM (MC38), respectively (online supplemental figure S11D). These results indicate that although CPZ may not induce immunotherapy resistance at non-cytotoxic concentrations, it is capable of inducing cytotoxicity against tumor cells at a high concentration. To further clarify the role of the CME pathway in Rab31−/− macrophage immunity function, we therefore designed experiments to knock down Cltc, a key mediator of the CME pathway, or to treat with a chemical Pitstop2, an alternative pharmacological inhibitor of the CME pathway.47 The siRNA experimental workflow is illustrated in figure 5J: siCtrl (control) or siCltc was transiently transfected into WT or Rab31−/− BMDMs, which were subsequently mixed with B16F10 cells and subcutaneously inoculated into mice. As shown in online supplemental figure S11E, Cltc knockdown with siCltc-1 or siCltc-2 led to a 60–80% reduction of Cltc in BMDMs. As shown in figure 5K, Cltc knockdown in WT BMDMs alone did not affect tumor growth, but reversed the inhibitory effect of Rab31−/− BMDMs on tumor growth. These results are consistent with the above conclusion: when the CME pathway was blocked, Rab31 deficiency no longer exerted antitumor effects in macrophages. Flow cytometry results further demonstrated that siCltc or Pitstop2 reversed the effects of Rab31−/− on pro-inflammatory, antigen presentation, T-cell chemotaxis, anti-inflammatory, and tissue repair functions in BMDMs (figure 5L), as well as on the expression of M2 marker Mrc1/Cd206, M1 marker iNOS, and antigen-presenting molecule MHC-I (figure 5M). Notably, neither siCltc nor Pitstop2 alone affected these functions. We speculate that CME inhibition completely blocks receptor internalization, trapping receptors at the plasma membrane—a state distinct from the receptor degradation phenotype induced by RAB31 deficiency, and that CME inhibition non-selectively targets both M1 and M2 receptor systems, whereas Rab31 preferentially regulates M2-associated trafficking. Thus, inhibition of CME is not equivalent to loss of Rab31 in TAM. Taken together, these results indicate that CME is an indispensable pathway by which Rab31 modulates TAM polarization. Loss of Rab31 disrupts Anxa1/Fpr2-mediated polarization of TAMs through CME As described above, macrophages undergo tumor-induced education (known as dynamic cancer immunoediting), which leads to upregulated expression of RAB31. To identify potential tumor-derived signals mediating this process, we performed cell–cell communication analysis based on the scRNA-seq dataset described above where all major cell types are included. As shown in figure 6A, Anxa1 and Hmgb1 ligands’ activities were significantly downregulated in the Rab31−/− TME compared with the WT TME, suggesting that these cell−cell signaling events are functionally impaired in the absence of Rab31. Compared with Hmgb1, Anxa1 expression appeared to be generally elevated in malignant cells compared with normal epithelial cells (figure 6B and online supplemental figure S12A). Interestingly, expression of RAB31 in TAMs was positively correlated with that of ANXA1 in tumor cells across multiple cancer types, including PDAC, UVM, COAD, and STAD (Pearson correlation coefficients were 0.77, 0.93, 0.63, and 0.66, respectively) (figure 6C). However, only a weak correlation between HMGB1 levels in tumor cells and RAB31 expression in TAMs was observed (online supplemental figure S12B). ANXA1 is an endogenous anti-inflammatory mediator involved in the progression of inflammation and cancer. It is also reported that ANXA1 acts as an upstream inhibitor of NF-κB signaling.48 The preliminary clues of RAB31/ANXA1 led us to hypothesize that RAB31 modulates the tumor cell-derived ANXA1 signaling in TAMs. Rab31 deficiency disrupts Anxa1/Fpr2-mediated regulation of TAMs. (A) The signals sent to TAMs and monocytes in the WT and Rab31−/− TME were analyzed using NicheNet. Upper: Tumor cells and TAMs are senders of the signal while TAMs are receivers; only these groups of cells are included for simplicity. Lower: The first column lists the top 15 ligands ranked by activity; the second column presents the fold change of signaling flow between the two host mice (Rab31−/− TME vs WT TME). Heatmap, red blocks for upregulation in the Rab31−/− TME while blue blocks for downregulation. (B) ANXA1 expression levels in normal epithelial cells and paired cancer cells (connecting lines denote samples from shared patients) in three cancer types. Source: scRNA-seq datasets from GSA and GEO. P values were determined by paired t-tests. PDAC (n=5 vs 10), COAD (n=9 vs 22), STAD (n=10 vs 10). Data are presented as mean±SEM. (C) Expression of ANXA1 in tumor cells is associated with expression of RAB31 in TAMs. Four cancer types were analyzed as indicated: PDAC, UVM, COAD and STAD. Source: scRNA-seq datasets from GSA and GEO. (D–F) Control B16F10 (sgCtrl_B16F10) cells or Anxa1-knockout B16F10 (sgAnxa1_B16F10) cells were inoculated subcutaneously into WT or Rab31−/− mice. Tumor growth was monitored by tumor volume (mm³) over time (D), tumor images (E) and final tumor weight (mg) (F). P values of tumor volume were assessed using two-way ANOVA. P values of tumor weight were assessed using paired t-tests. Data are presented as mean±SEM. (G) BMDMs from WT or Rab31−/− mice were stimulated with TCM from sgCtrl_MC38 or sgAnxa1_MC38 (left); TAMs in the sgCtrl_MC38 or sgAnxa1_MC38 TME (right). M1-like (iNOS+) and M2-like (Mrc1/CD206+) macrophage subsets were quantified by flow cytometry. Data are presented as mean±SD for TCM experiments and mean±SEM for TME. (H) BMDMs were stimulated with MC38 TCM in the presence or absence of 1 µM Ac2-26, followed by flow cytometric analysis of Mrc1/Cd206 MFI. (I) BMDMs were stimulated with sgAnxa1_MC38 TCM in the presence or absence of 1 µM Ac2-26, followed by flow cytometric analysis of Mrc1/Cd206 MFI. (J) Schematic representation of the intracellular trafficking fate of receptors following clathrin-mediated endocytosis. On internalization, clathrin-coated vesicles uncoating and fuse with early EEs characterized by RAB5 GTPase activity. From this sorting station, receptors are directed into two alternative pathways: (1) non-ubiquitinated cargo enters the fast-recycling route involving RAB4-positive and RAB11-positive REs for return to the PM; (2) ubiquitinated receptors are sequestered into RAB7-positive late endosomes (ILVs of MVBs), ultimately leading to lysosomal degradation. (K–L) THP-1 cells were treated with 50 µg/mL CHX for 2 hours, followed by stimulation with 1 µM Ac2-26. Samples were collected at 0, 2, 4, and 10 hours after stimulation for western blot analysis. Band intensities were quantified using ImageJ, and statistical analysis was performed with GraphPad software using two-way ANOVA to assess significant differences in protein expression. (M) Representative immunofluorescence images showing the co-localization of Fpr2 (red) and Rab31 (green) in BMDMs on TCM treatment. Fluorescence intensity and the spatial distance between Fpr2 and Rab31 signals were quantified using ImageJ. Three representative images are shown. (N) Representative immunofluorescence images showing Fpr2 (red) and the lysosomal marker Lamp1 (green) in BMDMs on TCM treatment. Arrows denote regions of Fpr2-Lamp1 colocalization. Three representative images are shown. P values were determined by paired t-tests. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; BMDM, bone marrow-derived macrophage; CHX, cycloheximide; COAD, colon adenocarcinoma; EEs, endosomes; FC, fold change; ILVs, intraluminal vesicles; iNOS, inducible nitric oxide synthase; LFC, log2 fold change; MFI, mean fluorescence intensity; MVBs, multivesicular bodies; REs, recycling endosomes; PDAC, pancreatic ductal adenocarcinoma; PM, plasma membrane; scRNA-seq, single-cell RNA sequencing; STAD, stomach adenocarcinoma; TAM, tumor-associated macrophage; TCM, tumor-conditioned media; TME, tumor microenvironment; UVM, uveal melanoma; WT, wild-type. To assess the involvement of tumor-derived Anxa1 in Rab31−/− TME, we generated Anxa1-knockout MC38 cells and implanted them subcutaneously into WT and Rab31−/− mice. As shown in online supplemental figure S12C-E, Anxa1 deficiency significantly suppressed tumor growth in WT mice [WT (sgAnxa1_MC38) vs WT (sgCtrl_MC38)]. However, there was no significant difference [WT (sgAnxa1_MC38) vs Rab31−/−(sgAnxa1_MC38)]. Two-way analysis of variance revealed a significant interaction between host genotype and tumor ANXA1 status (p=0.0049), indicating that Rab31-mediated tumor suppression is, at least in part, dependent on Anxa1 signaling from tumor cells. Notably, Anxa1-deficient tumors exhibited signs of resorption, resulting in constrained tumors at endpoint. To further validate the involvement of tumor-derived Anxa1 in Rab31−/− TME, we performed a similar experiment using Anxa1-deficient B16F10 cells. As shown in figure 6D–F, the growth difference between Anxa1-deficient B16F10 tumors in Rab31−/− versus WT hosts was substantially diminished compared with that observed with WT tumor cells (p=0.0008). We also generated Hmgb1-deficient tumor cells and implanted them into WT and Rab31−/− hosts. Hmgb1 knockout alone significantly impaired tumor growth in WT mice. However, the tumor-suppressive difference between WT and Rab31−/− groups remained nearly unchanged regardless of Hmgb1 status (2.57 vs 2.49), indicating that the antitumor effect of Rab31 deficiency is largely independent of Hmgb1 (online supplemental figure S12F and G). Together, the data suggest that tumor-derived Anxa1 is a key contributor to the immunosuppressive phenotype in Rab31−/− TME. We also assessed the effect of Anxa1-deficient TCM on Rab31−/− TAM polarization. Depletion of tumor cell-derived Anxa1 attenuated M2 polarization in both in vitro and in vivo assays and substantially abrogated the difference in TAM polarization between WT and Rab31−/− groups (figure 6G, online supplemental figure S12H and I). Interestingly, supplementation with the Anxa1-derived peptide Ac2-26 restored the M2 polarization impaired by Rab31 deficiency (figure 6H). Moreover, when the peptide Ac2-26 was added to Anxa1-deficient TCM, the difference in M2 polarization between WT and Rab31−/− groups was rebuilt (figure 6I). These results support the hypothesis that Rab31 maintains a protumor TME by facilitating Anxa1-driven TAM education. The protein FPR2 is the primary receptor for ANXA1 and is a member of the G protein-coupled receptor family. On sustained ligand activation, GPCRs are internalized via CME and fuse with RAB5+ early endosomes, undergoing two fates: (1) non-ubiquitinated receptors recycle to the plasma membrane through RAB4/11-driven rapid recycling; (2) ubiquitinated receptors will be sorted by the endosomal sorting complex required for transport (ESCRT) machine into RAB7+ late endosomes for lysosomal degradation. FPR2 in general undergoes the first process via CME to achieve receptor desensitization and resensitization49–53 (figure 6J). Immunofluorescence analysis showed that Rab31 deficiency reduces the expression of Fpr2 in BMDMs in response to TCM stimulation, thereby disrupting Anxa1 signaling. Given that CPZ reversed this cascade, we concluded that Rab31 deficiency inhibits Anxa1 signaling through the CME pathway (online supplemental figure S13A and B). To assess plasma membrane stability of FPR2, CHX chase assays were performed in THP-1 to inhibit new protein synthesis, followed by plasma membrane separation (online supplemental figure S13C). In siCtrl-treated THP-1 (control), membrane FPR2 remained stable. However, sgRAB31-treated THP-1 showed time-dependent FPR2 decay on Ac2-26 stimulation; the FPR2 loss was reversed by siCLTC co-treatment (figure 6K–L, online supplemental figure S13D). On blockade of protein synthesis, Rab31 exhibited prominent co-localization with Fpr2 in TCM-stimulated BMDMs (figure 6M). Fpr2 showed minimal lysosomal co-localization in WT BMDMs, whereas Rab31 deficiency markedly increased Fpr2-lysosome co-localization (figure 6N). Consistently, following bafilomycin A1 treatment, surface FPR2 in RAB31−/− THP-1 cells no longer decreased with increasing ligand stimulation time, indicating that lysosomal inhibition rescued the FPR2 surface loss caused by RAB31 deficiency (online supplemental figure S13E and F). Collectively, these results demonstrate that Rab31 maintains Fpr2 membrane stability through CME-dependent trafficking in macrophages, while Rab31 deficiency redirects Fpr2 toward lysosome, accelerating its degradation. Considering that FPR2 activation results in inhibition of the NF-κB pathway,54 55 we also accessed the NF-κB pathway in Rab31−/− TAM. Screening the Harmonizome transcription factor dataset identified five targets of macrophage-related NF-κB: Cd40, Ccl20, Itga1, Nlrp3, Tnf. qRT-PCR assays showed that Rab31 deficiency enhanced expression of these NF-κB targets under TCM stimulation. However, such enhanced expression was blocked by CME inhibition (online supplemental figure S13G). Western blotting analysis demonstrated enhanced activation of the canonical NF-κB signaling pathway in Rab31-deficient macrophages, as evidenced by increased phosphorylation of p65, IKKβ, and IκBα (online supplemental figure S13H and I). The NF-κB inhibitor markedly suppressed M2 polarization in BMDMs and partially rescued the M1-like polarization induced by Rab31 deficiency (online supplemental figure S13J). Loss of Rab31 in TAMs enhances recruitment and activation of T cells and increases sensitivity to anti-PD-L1 therapy The antitumor functions of TAMs can be mediated through a direct mechanism (eg, macrophage-mediated phagocytosis) or an indirect mechanism (eg, primarily through CD8+ T-cell activation). In most tumor contexts, T cell-mediated adaptive immunity plays a central role in orchestrating effective antitumor responses. As shown in the above scRNA-seq results figure 2J and online supplemental figure S9D, compared with tumor-bearing WT mice, Rab31−/− mice exhibited markedly increased T-cell infiltration in subcutaneous tumor tissues, along with enhanced antigen-presenting interactions between TAMs and T cells. To further investigate the involvement of T cells in the Rab31−/− TME, we first analyzed CD8+ T cell-related signals in the scRNA-seq dataset. As shown in figure 7A, tumor-killing cytokines (Tnf, Ifng), T-cell chemoattractants (Cxcl9), and activation markers (Ox40) were all significantly upregulated in Rab31−/− hosts, supporting that T cells were functionally activated in this context. Importantly, Rab31 is not expressed in T cells, suggesting that the observed T-cell activation is mediated via cell–cell interactions. Consistent with this, Cxcl9 signaling from TAMs to T cells was markedly increased in Rab31−/− mice, while Tnf and Ifng signals from T cells to melanoma cells were also significantly elevated (figure 7B–C, online supplemental figure S14A-C). Taken together, these results demonstrate that enhanced bidirectional communication between TAMs and T cells took place in the Rab31−/− TME, thereby promoting T cell-mediated tumor cytotoxicity. Rab31 deficiency in TAMs enhances recruitment and activation of T cells and increases sensitivity to anti-PD-L1 therapy. (A) Differential information flow patterns in the WT and Rab31−/− TME. The overall information flow was computed by CellChat. Information flows that are dominant in Rab31−/− TME are highlighted in bold green (TNF, CCL, IFN-II, IL-1 and OX40). (B) The communication from TAMs and monocytes to T cells in the two groups was analyzed using NicheNet. The first column lists the top 20 ligands ranked by activity; the second column presents the fold change of signaling flow between the two groups (with red and blue indicating upregulation and downregulation, respectively, in the Rab31−/− group). (C) The communication from T cells to melanoma in the two groups was analyzed using NicheNet. (D–E) Cxcl9 protein levels were measured by ELISA. (D) Cxcl9 levels in tumor homogenates (n=5 per group), and data are presented as mean±SEM; E, Cxcl9 levels in the culture medium of BMDMs co-cultured with TCM (n=6 per group), and data are presented as mean±SD. (F–G) The transwell assay was used to assess T-cell migration in response to macrophage-derived chemotactic signals following TCM treatment. (F) mouse cell models. (G) human cell models. After 4 hours co-culture, the number of T cells that migrated to the lower chamber was assessed (n=6 per group). Data are presented as mean±SD. (H) Functional changes in CD8+ T cells were assessed by expression of cytotoxicity and activation marker genes (Gzma, Gzmb, Nkg7, Itga1, Klrg1, Cd40lg and Prf1). qRT-PCR assays were performed on the sorted CD8+ T cells from the B16F10 tumor tissue (n=3 per group). Data are presented as mean±SD. (I) T cells exhibit increased exhaustion in the WT TME compared with those in the Rab31−/− TME. The proportion of exhausted T cells (marked by Lag3 staining or by Pd1 staining) in tumor tissue was detected by flow cytometry (n=5 per group). Data are presented as mean±SEM. (J–K) Blocking the CXCL9-CXCR3 signaling axis results in loss of Rab31-deficiency-mediated antitumor effect. The inhibitor SCH546738 was used to inhibit T-cell chemotaxis, and its effect on the progression of B16F10 subcutaneous tumors in Rab31−/− mice was assessed (n=6 per group). (J) photograph of tumors; K, quantification of tumor weights. (L–M) On αPD-L1 treatment, the disease progression was associated with RAB31 expression in TAMs. Source: scRNA-seq dataset GSE169246. (L) Pearson correlation analysis. (M) comparing the expression of RAB31 in TAMs in patients with SD (no response) and with PR. Data are presented as mean±SEM. (N–O) Tumor growth in WT and Rab31−/− host mice treated with anti-CD8α and anti-PD-L1 antibodies. (N) Photograph of the tumors; (O) Quantification of tumor weights. B16F10 subcutaneous tumor-bearing models were established in WT and Rab31−/− mice, followed by intervention with CD8+ T cell-depleting antibody (anti-CD8) or immune checkpoint inhibitor (anti-PD-L1), administered either as monotherapy or in combination. Tumor weights were recorded at day 12. Data are presented as mean±SEM. P values were determined by paired t-tests. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. BMDM, bone marrow-derived macrophage; CCL, C-C motif chemokine ligand; IFN, interferon; IL, interleukin; LFC, log2 fold change; PD-L1, programmed death-ligand 1; PR, partial response; qRT-PCR, quantitative reverse transcription polymerase chain reaction; scRNA-seq, single-cell RNA sequencing; SD, stable disease; TAM, tumor-associated macrophage; TCM, tumor-conditioned media; TME, tumor microenvironment; TNF, tumor necrosis factor; WT, wild-type. To experimentally validate these observations, we performed a series of functional assays. ELISA was used to quantify the protein levels of Cxcl9 and Il1b in tumor lysates and in BMDM supernatants following TCM stimulation (figure 7D,E online supplemental figure S14D and E). In the presence of TCM, BMDMs or THP-1 cells were co-cultured with T cells in transwell assays to assess T-cell chemotaxis (figure 7F–G, online supplemental figure S14F). qRT-PCR assays were used to measure activation markers in T cells (figure 7H, online supplemental figure S14G-I), and T-cell exhaustion was also evaluated (figure 7I). Collectively, these results indicate that Rab31−/− TAMs secrete higher levels of Cxcl9 and Il1b, thereby promoting T-cell recruitment and enhancing T cell-mediated cytotoxicity, ultimately driving a potent antitumor immune response. Moreover, inhibition of T-cell recruitment with the CXCR3 antagonist SCH546738 abolished the Rab31-deficiency-induced antitumor effect, demonstrating that the antitumor phenotype in Rab31−/− TME is T cell-dependent (figure 7J,K, online supplemental figure S14J). For further validating the functional cooperation between RAB31-regulated TAMs and CD8+ T cells, we analyzed a cohort of patients with breast cancer receiving anti-PD-L1 therapy. We observed a positive correlation between expression of RAB31 in TAMs and degree of disease progression in the patients (figure 7L). Patients who achieved partial remission exhibited lower expression of RAB31 in TAMs compared with those with only disease stabilization (figure 7M). These results imply a potential link between expression level of RAB31 in TAMs and degree of resistance to anti-PD-L1 therapy in the patient cohort. Using our tumor-bearing mouse model, we validated the potential involvement of RAB31 in anti-PD-L1 resistance. As shown in figure 7N,O, compared with anti-PD-L1 monotherapy, the combination of Rab31 deficiency with anti-PD-L1 further enhanced tumor inhibition, indicating that Rab31 loss sensitizes tumors to checkpoint blockade. CD8+ T-cell depletion completely abolished the tumor-suppressive effect of Rab31 deficiency, demonstrating that this phenotype is dependent on CD8+ T cells. Furthermore, under CD8+ T cell-depleted conditions, Rab31 deficiency failed to enhance anti-PD-L1 efficacy, confirming that the synergistic antitumor effect is CD8+ T cell-mediated. Collectively, these experiments establish that Rab31 deficiency in TAM promotes antitumor immunity through a CD8+ T cell-dependent mechanism and potentiates anti-PD-L1 therapy in a CD8+ T cell-restricted manner. Discussion Here, we demonstrate that RAB31 modulates macrophage sensing of tumor-derived cues via the CME pathway, establishing a previously unrecognized “membrane trafficking-TAM polarization-immunosuppression” axis that underscores the functional interplay between neoplastic signals and TAM endocytic capacity. While a series of prior studies on RAB31 have explored its role in tumor cells,36 40 56–58 the present work elucidates its uncharacterized functions and mechanisms in TME, especially in TAMs. Our study demonstrates that RAB31 regulates macrophage polarization via supporting the tumor-derived ANXA1 signal with the assistance of the CME machine. The cascades mediated by RAB31/CME in TME are summarized in figure 8 and can be divided into three steps: Step 1, Rab31 deficiency disrupts the normal endocytosis-recycling process of the TAM membrane receptor Fpr2 via the CME mechanism, redirecting internalized Fpr2 to lysosomes for accelerated degradation, thereby blocking the educational effect of tumor cell-derived Anxa1 on macrophages; Step 2, Rab31−/− macrophages are deprived of the Anxa1/Fpr2 signal and undergo transformation from the protumoral M2 phenotype to the antitumoral M1 phenotype, characterized by enhanced antigen presentation capacity and T-cell chemotaxis and activation; Step 3, Rab31 deficiency-induced transformation of TAMs cooperates with CD8+ T-mediated immunity, reprograms the TME into an antitumor state, and augments the sensitivity to anti-PD-L1 immunotherapy. Our findings underscore the therapeutic potential of targeting intracellular trafficking pathways, such as RAB31-mediated endocytosis, to reprogram the immunosuppressive landscapes of solid tumors. Targeting RAB31 may thus represent a promising strategy to enhance checkpoint inhibitor responses, particularly in tumors characterized by TAM-driven immune suppression. Graphical abstract and summary of Rab31-mediated regulation of TAM plasticity in TME. See the Discussion section of the main text for the description about the proposed mechanisms including the Step 1, Step 2 and Step 3 cascade induced by the loss of Rab31. MHC, major histocompatibility complex; TAM, tumor-associated macrophage; TCR, T-cell receptor; TME, tumor microenvironment; WT, wild-type. It has been well recognized that the plasticity of TAMs is orchestrated by a complex interplay of extrinsic cues (eg, cytokines and metabolites) and intrinsic regulators (eg, transcription factors and epigenetic modifiers).55 59 However, the role of vesicular trafficking in regulating TAM polarization and plasticity had remained largely unexplored until the present study. This unique expression of RAB31 in myeloid cells prompted us to propose its function in the TME. We therefore speculate that RAB31 represents a selective target for reprogramming TAMs. As proposed in figure 8, RAB31 involves critical and intriguing communications between intercellular and intracellular compartments within the TME. These findings are in line with emerging evidence that endosome-related signaling hubs. For example, the trafficking of the cargos including toll-like receptors and programmed cell death protein 1, are key modulators of immune cell function.60 Herein, we provided evidence from clinical samples and tumor-bearing models that high RAB31 levels in TAMs correlate with adverse outcomes in diverse cancers and facilitate the formation of an immunosuppressive, tumor-promoting microenvironment. RAB proteins are tightly associated with multiple steps of cargo transport, especially the CME process. CME is the canonical eukaryotic process through which numerous receptors undergo endocytosis. RAB5 regulates early endosome fusion and sorting; receptors may recycle back to the plasma membrane via RAB4/RAB11-positive recycling endosomes, or be sorted by the ESCRT machinery into RAB7+ late endosomes for lysosomal degradation.49–53 Members of the RAB family participate in multiple steps of this process, highlighting the potential importance of RAB31 in endocytic regulation. Here, we identified a distinct subset of TAMs, Dab2+TAMs, that was nearly absent in the Rab31−/− group. Dab2, an adaptor protein associated with the CME pathway, marks a population characterized by high expression of membrane trafficking-related genes. We also demonstrated that in clinical samples a high infiltration of Dab2+TAMs was correlated with a poor prognosis. Co-IP/MS analyses revealed that Rab31 interacts with key CME-associated proteins, including Picalm and Dnm2. These interactions are enhanced under tumorous stimulation. Interestingly, we investigated the contribution of the tumor-derived ligand ANXA1, a well-characterized anti-inflammatory mediator known to promote tumor progression, in Rab31−/− TAMs. We further examined the roles and mechanisms of Anxa1 in our tumor-bearing murine models. It has been reported that a high proportion of ANXA1+ tumor-infiltrating macrophages correlates with poor clinical outcomes.61 62 Activation of FPR2, one of the primary receptors for ANXA1, in TAMs drives M2 polarization and fosters an immunosuppressive TME by inducing T-cell exhaustion.62 In our tumor-bearing mouse models, we demonstrated that Rab31−/− macrophages exhibited a significantly accelerated degradation of surface FPR2 in a CME-dependent manner following exposure to tumor-derived signals. This effect was attributable to enhanced redirection of FPR2 to lysosomes. Although Rab31 is robustly expressed in TAMs and not expressed in T cells, the antitumor immunity in tumor-bearing Rab31−/− mice still requires T cells. Rab31−/− TAMs exhibit increased production of Cxcl9, a chemokine critical for the recruitment of CD8+ T cells into tumors,63 aligning with the observed increase in T-cell infiltration and enhanced antitumor responses in these host mice. More importantly, our tumor-bearing models showed that Rab31 deficiency enhances the therapeutic efficacy of anti-PD-L1 treatment. Currently, strategies targeting TAMs are rapidly emerging, exemplified by antitumor IgE antibodies that activate macrophages, providing a novel therapeutic option for IgG-resistant or immunologically “cold” tumors.64 However, allergic safety concerns and immunosuppression within the tumor microenvironment remain major obstacles to clinical translation. Rab31−/− mice exhibited completely normal physical characteristics and hematopoietic function, including conventional hematological parameters, immune cell composition in bone marrow and spleen, and competitive repopulation capacity of hematopoietic stem cells, indicating a favorable safety profile for targeting RAB31. Limitations of the study Although we demonstrated that FPR2 recycling is dependent on the RAB31-associated endocytosis machinery, we should point out that the cargo selectivity controlled by Rab31 during the process remains unclear. We speculate that this selectivity depends on the stimuli received by macrophages from their microenvironment, where only receptors exposed to sustained or high concentrations of ligands undergo endocytosis and resensitization. Furthermore, the function of RAB31 in dendritic cells, another myeloid lineage population, represents a promising avenue for future investigation, including potential roles in regulating antigen processing, cross-presentation, and costimulatory molecule expression. Finally, the development of an RAB31-specific GTPase assay will facilitate a more detailed understanding of its function in macrophages. In conclusion, through exploring the expression of RAB31 in clinical samples and characterizing tumor growth in Rab31−/− host mice, we identified that the RAB31/CME machinery mediates the communication between TAMs and tumor cells in the TME. Our results suggest that the strategy targeting the RAB31/CME pathway in TAMs represents a promising and novel option in cancer immunotherapy. Supplementary material