GDP-bound Rab37 modulates M2-like tumor-associated macrophage polarization by attenuating STAT1 translocation to downregulate the type I IFN pathway

Abstract

Background

Tumor-associated macrophages (TAMs) in the tumor microenvironment (TME) primarily polarize into the M2-phenotype. Our previous study showed that the small GTPase Rab37 mediates IL-6 trafficking in macrophages for M2 polarization. Here, we uncover an unconventional role of Rab37, independent of vesicle trafficking, in promoting M2 polarization of TAMs.

Methods

The gene profiles in wild-type and Rab37 knockout (KO) bone marrow-derived macrophages (BMDMs) were analyzed using cDNA microarray. The mechanism of Rab37 in regulating the interferon (IFN) pathway was confirmed through in vitro/vivo assays and clinical studies.

Results

Type I IFN signaling was highly enriched in BMDMs from Rab37 KO mice. Moreover, Rab37 induction and decreased type I IFN genes were observed in macrophages treated with lung cancer-conditioned medium and epigenetic drugs, indicating an epigenetic regulation of Rab37 in TAMs. Mechanistically, GDP-bound Rab37 interacted with the nuclear localization sequence of STAT1 to sequest it in the cytosol from its transcription activities, thus leading to the downregulation of IFN genes. Clinically, CD163⁺/Rab37⁺/STAT1^cytosol in TAMs expression signature correlated with advanced tumor stages and poor survival of lung cancer patients.

Conclusions

Our findings highlight the cytosolic interaction of Rab37-STAT1 in M2 TAM polarization, with CD163⁺/Rab37⁺/STAT1^cytosol TAMs as a lung cancer prognosis biomarker.

Introduction

The tumor microenvironment (TME) is a dynamic landscape where cancer cells interact with tumor-infiltrating immune cells, stroma cells, and non-cellular components [1]. Among these cells, tumor-associated macrophages (TAMs) represent the predominant population of tumor-infiltrating myeloid cells within the TME [2–4]. Of note, the high infiltration of TAMs has been proven to correlate with poor outcomes in many malignant tumors [5, 6]. TAMs predominantly polarize into M2-like pro-tumoral macrophages rather than M1-like anti-tumoral macrophages under the stimulation of the TME [7–9]. These M2-like TAMs further suppress the recruitment and function of cytotoxic T cells and dendritic cells by secreting anti-inflammatory cytokines and overexpressing inhibitory ligands, resulting in resistance to cancer immunotherapy [10, 11]. Recently, many studies have shown that M2-like TAMs can repolarize into M1-like TAMs by altering the cell signaling transduction, such as the interferon (IFN) pathway, effectively stimulating the immune system and benefiting inhibitory checkpoint blockade combination therapy [12–14]. Therefore, reprogramming TAMs in the TME may provide an effective cancer immunotherapy strategy.

The IFN family proteins are crucial in orchestrating immune responses, consisting of two main classes of IFNs: type I IFNs (IFN-α and IFN-β) and type II IFN (IFN-γ) [15]. Type I and type II IFNs both trigger the classical Janus kinase (JAK)-signal transducer and activator of transcription 1 (STAT1) pathway, leading to the induction of IFN-stimulated genes (ISGs) expression, which is known to promote M1 polarization of macrophages [16, 17]. Moreover, the upregulation of the IFN-γ-STAT1 responses reprograms TAMs from M2-like to M1-like [18], while the downregulation of the IFN-γ-STAT1 responses promotes M2-like polarization in TAMs [19], indicating that regulating the IFN pathway is crucial for reprogramming TAMs.

Rab GTPase represents the largest family of small GTPase proteins which are responsible for intracellular vesicle trafficking [20]. Rab proteins have an inactive GDP-bound state and an active GTP-bound state. The GTP-bound Rabs are mainly involved in membrane attachment and vesicle trafficking, while the GDP-bound form are distributed in the cytosol [21]. The active GTP-Rab binds with specific effector proteins to facilitate vesicle trafficking, while the function of the cytosolic GDP-Rab remains largely unelucidated [22]. Emerging evidence has suggested that the GDP-bound Rab may be involved in some unconventional mechanisms, such as regulating intracellular signaling pathways or proteasomal degradation [22–24]. For example, Rab1A has been found to interact with mTORC1 and activate mTORC1 signaling, thereby promoting colorectal cancer progression and invasion [25]. Rab25 and Rab35 are involved in PI3K-AKT signaling that regulates cell proliferation [26, 27]. Moreover, Rab2A interacts with p-Erk1/2 to facilitate the nuclear translocation of p-Erk1/2 and downstream β-catenin, thus regulating breast cancer stem-like cells and tumor initiation [28]. Our previous research demonstrated that Rab37 in TAMs mediates interleukin-6 (IL-6) secretion and promotes M2 polarization through the activation of the IL-6/STAT3 pathway [29]. However, whether GDP-bound Rab37 regulates signal transduction in the cytosol to influence macrophage polarization remains unknown. Therefore, we hypothesize that the GDP-bound Rab37 in the cytosol may interact with cytosolic proteins, thereby affecting intracellular signal transduction.

In this study, we unveil the unconventional role of GDP-bound Rab37 in downregulating the type I IFN pathway by sequestering STAT1 nucleus translocation, resulting in promoting the M2 polarization of TAMs. Furthermore, stimulation with lung cancer conditioned medium altered the epigenetic regulation in TAMs, resulting in an upregulation of Rab37 mRNA expression and further promoting the M2 polarization of TAMs. Clinically, the expression of Rab37 in M2-like TAMs increases with the stage of lung cancer patients, suggesting the expression signature of CD163⁺/Rab37⁺/STAT1^cytosol in TAMs can be a biomarker for poor prognosis in lung cancer patients.

Materials and methods

Cell lines and culture conditions

The human lung cancer cell line H460, human embryonic kidney epithelial cell line 293T, mouse macrophage cell line RAW264.7, and mouse Lewis lung carcinoma (LLC) were purchased from the American Type Culture Collection (Rockville, MD, USA) and cultured in DMEM medium (Gibco, Los Angeles, CA, USA). The human monocytic cell line THP-1 was purchased from the Bioresource Collection and Research Center (BCRC, Taiwan) and maintained in RPMI 1640 medium (Gibco). Both DMEM and RPMI 1640 media were supplemented with 10% Fetal Bovine Serum (FBS; Gibco) and 1% penicillin/streptomycin (Gibco), and cultured at 37 °C with 5% CO₂. In some experiments, 5-aza-2’-deoxycytidine (5-aza-dC; Merck Millipore, Burlington, MA, USA) or suberoylanilide hydroxamic acid (SAHA; Merck Millipore) was added to the culture medium at a concentration of 10 μM for 72 h or 1 μM for 24 h.

THP-1 macrophage differentiation and polarization

THP-1 monocytes were differentiated into macrophages by treating with 20 ng/ml phorbol 12-myristate 13-acetate (PMA; Sigma Aldrich, Burlington, MA, USA) for 48 h, then polarized into M1 phenotype with 20 ng/ml IFN-γ and 1 μg/ml LPS for 24 h, or into M2 phenotype with 20 ng/ml IL-4 and IL-13 for 48 h. Cells were maintained in 5% CO₂ at 37 °C during these processes.

Bone-marrow-derived macrophages (BMDMs) isolation and culture conditions

Bone marrow cells were sterilely isolated from the tibia and femur of 6–8-week-old WT and Rab37 knockout (KO) mice. After removing the epiphyses from the tibia and femur, the bone marrow cells were eluted with complete DMEM and then seeded at a concentration of 3 × 10⁶ cells in a 10 cm dish with complete DMEM containing 10 ng/ml macrophage colony-stimulating factor (M-CSF) (Peprotech, Cranbury, NJ, USA). BMDMs were differentiated for 7 days at 37 °C with 5% CO₂, with medium changes every 2 days during the ex vivo culture.

LLC and H460 lung carcinoma-derived conditional medium (CM) collection

LLC and H460 cells were seeded (5 × 10⁵ cells in 10 cm dishes) and incubated overnight. After refreshing the culture medium, cells were further incubated for 48 h. Subsequently, the LLC and H460 CM were collected and centrifuged at 1400 r.p.m. for 5 min at 4 °C. The supernatant was stored at −80 °C for up to two weeks. The cell viability (of >98%) was monitored using the trypan blue dye exclusion assay.

RNA extraction and quantitative reverse transcriptase polymerase chain reaction (RT-qPCR) assay

Total RNA was extracted using Trizol reagent (Invitrogen, Waltham, MA, USA) and reverse transcribed into cDNA with the PrimeScript RT reagent kit (TaKaRa, San Jose, CA, USA). RT-qPCR was performed to analyze the mRNA using SYBR Green Master Mix (Invitrogen). The results were normalized to the housekeeping gene β-actin. The primers used for RT-qPCR analysis are listed in Supplementary Table S1.

cDNA microarray and pathway analysis

We obtained gene expression profiles from BMDMs of WT and Rab37 KO mice treated with or without LLC CM for 24 h (n = 2 technical replicates per group). Total RNA was isolated and then analyzed using the Clariom D array Mouse Transcriptome Array (Thermo Fisher Scientific, Waltham, MA, USA). Primary data analysis was performed using Affymetrix Transcriptome Analysis Console software (version 3.1.0.5). Gene expression data were log-transformed, and when the ANOVA P-value met the criterion P < 0.05 at fold changes >|2| were selected as significantly differentially expressed genes. To identify differential transcriptional changes in BMDMs from Rab37 WT or KO mice, we subtracted the differences in average gene-expression values between WT BMDMs after 0 or 24 h of treatment (i.e., ΔWT, representing WT-24h minus WT-0h) from the differences in average gene-expression values between BMDMs from Rab37 KO after 0 or 24 h treatment (i.e., ΔKO, representing KO-24h minus KO-0h). The resulting values (ΔKO − ΔWT) were utilized for further analysis. Data are available at the Gene Expression Omnibus (GSE254823).

We performed gene set enrichment analysis (GSEA) on the differential gene-expression changes induced between Rab37 KO and WT BMDMs (ΔKO − ΔWT) using gene sets of the canonical pathway (C2) from the Molecular Signature Database (MSigDB; version 5.0) [30]. Transcriptome Analysis Console (TAC) (Affymetrix Transcriptome Analysis Console software) was used to further prioritize the top 10 mapped pathways on the differential gene-expression changes induced between Rab37 KO and WT BMDMs (ΔKO − ΔWT).

Bisulfite conversion and methylation-specific PCR (MSP)

Genomic DNA from macrophages was purified using the Quick-DNA^TM Microprep Plus kit (Zymo Research) and unmethylated cytosine was converted to uracil with sodium bisulfite using the EZ DNA Methylation-Gold^TM kit (Zymo Research). The methylation status of Rab37 gene promoters was determined through MSP analysis using EpiTaq^TM HS (for bisulfite-treated DNA) (TaKaRa)with methylated and unmethylated specific primers. The primers used for MSP are listed in Supplementary Table S1.

Protein extraction, Western blot (WB) and immunoprecipitation (IP) assays

Cells were harvested and lysed in RIPA buffer with a protease inhibitor cocktail (Sigma Aldrich). Protein lysates were solubilized in sample dye and 50 μg of protein lysates were loaded into 8% sodium dodecyl sulfate/ polyacrylamide gel electrophoresis and then transferred to a polyvinyl difluoride (PVDF) membrane. The protein was identified by incubating the PVDF membrane with primary antibodies at 4 °C for 18 h, then with horse radish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature.

For the IP assay, cells were harvested and lysed in 1× IP buffer and 500 μg of protein lysates were incubated with 5 μg of the indicated antibodies at 4 °C for 3 h. Next, 25 μl Protein G PLUS/Protein A mixture agarose beads (Merck Millipore) were added to the protein-antibody mixture and incubated at 4 °C overnight. After washing with 1× IP buffer, proteins were eluted with 2× sample dye. For the GFP or Flag-tagged IP, the reaction was performed using the anti-GFP nanobody immunomagnetic beads (Elabscience, Houston, TX, USA) or the Pierce^TM anti-DYKDDDDK magnetic agarose (Thermo Fisher Scientific). The antibodies and conditions are described in Supplementary Table S2.

TurboID-based proximity labeling assay

To analyze Rab37 interacting proteins in macrophages. Cells were transfected with pcDNA3.1-TurboID, pcDNA3.1-Rab37^WT-TurboID, pcDNA3.1-Rab37^Q89L-TurboID, or pcDNA3.1-Rab37^T43N-TurboID. After 8 h, 50 μM biotin was added to the culture medium for 16 h, allowing the proteins in proximity (10 nm) to TurboID-tagged Rab37 to be biotinylated. After biotinylation, cells were lysed in RIPA buffer, and the biotinylated proteins were pulled down by using streptavidin beads (Merck Millipore). The beads were washed twice with RIPA buffer and once with 1 M KCl, each for 2 min, followed by a brief 10 sec wash with 0.1 M Na₂CO₃ and 2 M urea in 10 mM Tris-HCl (pH 8.0). Finally, the beads were washed with RIPA buffer twice and protein samples were eluted by 25 μl of 2× sample dye with 2 mM biotin and 20 mM DTT at 95 °C for 10 min.

Nuclear and cytoplasmic fractionation assay

Cells were harvested in cytoplasmic buffer (1 M HEPES, 1 M MgCl2, 1 M KCl, and 1 M DTT), kept on ice for 10 min, then lysed by adding 1% IGEPAL. The samples were mixed gently before centrifugation at 13,000 r.p.m. at 4 °C for 2 min. The supernatant (cytoplasmic protein) was then collected, and the pellet was washed twice with sterile PBS and lysed with nuclear buffer (1 M HEPES, 1 M MgCl₂, 25% glycerol, 1 M NaCl, 0.5 M EDTA, and 1 M DTT). After centrifugation at 13,000 r.p.m. for 10 min at 4 °C, the supernatant (nuclear protein) was collected.

Immunofluorescence (IF) assay and confocal microscopy

Cells were fixed with 4% paraformaldehyde for 30 min, washed three times with PBS, and permeabilized with 0.1% Triton-100 for 5 min, followed by three times PBS washes. After blocking for 30 min, samples were covered with the primary antibody at 4 °C overnight. The next day, IF assay was conducted following the manufacturer’s instructions (Opal stain kit #NEL810001KT, Akoya Biosciences, Marlborough, MA, USA). The antibodies and conditions are described in Supplementary Table S2. The confocal images were captured using the Olympus FV3000 confocal microscope and analyzed using the FV31S-SW software (Olympus, Tokyo, Japan).

Plasmids and transfection

GFP-tagged Rab37^WT plasmid was purchased from OriGene (Rockville). Other plasmids used in this study are listed in Supplementary Table S3. Plasmids transfection was carried out with TurboFect transfection reagent (Thermo Fisher Scientific) or Lipofectamine^® 2000 (Invitrogen) according to the manufacturer instructions. Overexpression efficiency was checked by RT-qPCR and WB assay.

Protein stability assay

To analyze the protein stability of p-STAT1 in macrophages, cells were treated with 50 μg/ml protein synthesis inhibitor cycloheximide or 10 μM proteasome inhibitor MG132. Protein lysates were harvested in RIPA buffer at the indicated time points.

Flow cytometry

To analyze surface markers on macrophages, cells were washed twice with PBS, and suspended in staining buffer (2% FBS, 0.1% sodium azide in PBS). After centrifugation at 1400 r.p.m. for 5 min, cells were then stained with F4/80, CD86, or CD206 (BD Biosciences, San Jose, CA, USA) antibodies formulated with staining buffer (1:200). After incubating for 30 min, the samples were centrifuged at 400 × g for 5 min and washed once with staining buffer. Subsequently, the samples were re-suspended in 300 μl staining buffer, and analyzed using CytoFLEX (Beckman, Coulter, CA, USA). The antibody conditions are listed in Supplementary Table S2.

Multiplex fluorescence immunohistochemistry (IF-IHC)

Multiplex IF-IHC was performed to detect STAT1, IFN-α, IRF7, and Rab37 in TAMs from LLC lung orthotopic tumors of WT and Rab37 KO mice. The process was performed according to the manufacturer’s instructions (Opal stain kit #NEL810001KT). After antigen retrieval and blocking, slides were incubated with primary antibody at 4 °C overnight. The slides were then incubated with polymer HRP Ms+Rb for 10 min, followed by incubation with Opal fluorophore for 10 min at RT. For staining of another primary antibody, antigen retrieval steps were repeated. DAPI was then applied for nucleus staining. IF-IHC was quantified by an average of immunoreactive positive cells per five regions of interest (ROIs, 200 ×200 μm) using the ImageJ software. The antibodies and the experimental conditions are listed in Supplementary Table S2.

Patient samples and clinical information

A total of 48 lung cancer patients from NCKU Hospital were enrolled with institutional review board permission approval. Overall survival was calculated as the time from the day of surgery to the date of death or the last follow-up. Disease-free survival was calculated as the time from the day of surgery to either the date of disease recurrence or the date of death. Tumor type and disease staging were performed according to the World Health Organization classification and the TNM classification system, respectively.

Statistical analysis

Cell studies were conducted in three independent experiments unless indicated otherwise. Data represented mean ± s.d. using the two-tailed Student’s t test. Pearson χ2 for protein expression correlation in patients, Kaplan–Meier for survival curves with log-rank tests, and Cox regression for patient outcome risk analysis. The level of statistical significance was taken as P value, *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001.

Results

Type I IFN pathway enrichment in BMDMs from Rab37 KO mice suppresses macrophage M2 polarization

Our previous study revealed that Rab37 in TAMs promotes tumor growth in lung cancer TME by exocytosing IL-6, and driving M2 polarization via the Rab37/IL-6/STAT3 axis [29]. However, the mechanism of Rab37 in regulating macrophage polarization remains elusive. We first analyzed the gene profiles of WT and Rab37 KO BMDMs treated with mouse Lewis lung carcinoma cell-derived conditioned media (LLC CM) using cDNA microarray (GSE254823). The results showed that 70 and 166 genes exhibited differential alterations in WT and Rab37 KO BMDMs, respectively (Fig. 1a). Next, the 166 genes from Rab37 KO BMDMs were analyzed using Gene Set Enrichment Analysis (GSEA) and Transcriptome Analysis Console (TAC) software. The results demonstrated that these genes correlated with both type I and type II IFN signaling, which are associated with M1 phenotype [31] (Fig. 1b, c). The volcano plot identified six interferon-stimulated genes (ISGs) that were upregulated in LLC CM-treated Rab37 KO BMDMs, including ubiquitin-specific peptidase 18 (Usp18), XIAP-associated factor 1 (Xaf1), interferon-induced GTP-binding protein Mx1, interferon regulatory factor 7 (Irf7), IFN-induced protein with tetratricopeptide repeats 3 (Ifit3), and 2’-5’-oligoadenylate synthetase 2 (Oas2) (Fig. 1d). These results suggest that type I IFN pathway is enriched in BMDMs from Rab37 KO mice.

Fig. 1. cDNA microarray and pathway analyses revealed upregulation of type I IFN signaling pathway in Rab37 KO BMDMs.

a Affymetrix Transcriptome Analysis showed that WT BMDMs mock versus LCM treatment altered the expression of 585 genes in the WT group and 681 genes in the Rab37 KO group (n = 2 technical replicates for each group). Gene expression data were log-transformed, and differentially expressed genes were selected based on the ANOVA P-value < 0.05 at fold changes >|2| (GSE254823). b, c GSEA analysis was performed in 166 Rab37 KO BMDMs-specific genes, which are highly enriched with type I IFN (IFN-α/β) and type II IFN (IFN-γ) pathways. d Volcano plot from TAC software identified six candidate genes with significant change (fold change >2) in the type I IFN pathway. e RT-qPCR and western blot analysis confirmed Rab37 KO in BMDMs. f RT-qPCR analysis was performed to examine the mRNA expression of type I IFNs (Ifn-α and Ifn-β) and ISGs (Irf7) in WT and Rab37 KO BMDMs with or without LLC CM treatment (24 h). g Western blot analysis of IFN-β, IRF7, and Rab37 in WT and Rab37 KO THP-1 cells with or without H460 CM treatment (48 h). h RT-qPCR and western blot analysis confirmed the efficiency of Rab37^WT-GFP overexpression in RAW 264.7 cells. i RT-qPCR analysis of the mRNA expression of Ifn-β and Irf7 in EV or Rab37^WT overexpressed RAW 264.7 cells with or without LLC CM treatment. j Western blot analysis of IFN-β, IRF7, and Rab37 in EV or Rab37^WT overexpressed RAW 264.7 cells with or without LLC CM treatment. The band intensities in (g) and (j) were quantified, and the normalized fold changes are indicated below the blots. k M2-type macrophage surface marker (CD206) on WT and Rab37 KO THP-1 cells treated with or without M2-inducer (20 ng/ml IL-4 and IL-13, 48 h). The data are shown as the mean ± s.d. (n = 3 per group). P values were determined by one-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Furthermore, we collected WT and Rab37 KO BMDMs to analyze the expression of Rab37 and ISGs (Fig. 1e, f). The results showed that type I IFNs and ISGs were upregulated in Rab37 KO BMDMs, especially after LLC CM treatment (Fig. 1f; Supplementary Fig. S1A). The protein levels of type I IFNs and ISGs were analyzed in THP-1 human macrophage cells with Rab37 WT or Rab37 KO via CRISPR/Cas9. The result showed that IFN-β and IRF7 expression were upregulated in Rab37 KO THP-1 cells, particularly after H460 CM treatment (lanes 4 vs. 2) (Fig. 1g). In contrast, we overexpressed Rab37^WT in RAW 264.7 mouse macrophage cells (Fig. 1h), confirming significant downregulation of type I IFNs and ISGs expression, which further decreased following LLC CM treatment. (Fig. 1i; Supplementary Fig. S1B). Consistently, IFN-β and IRF7 protein levels were downregulated in Rab37^WT overexpression group (lanes 2 and 4) (Fig. 1j).

We hypothesized that Rab37 KO macrophages disrupt M2 polarization by elevating the type I IFN pathway. Therefore, we treated WT and Rab37 KO THP-1 cells with M2 inducers (IL-4 and IL-13) and measured the M2 surface marker CD206. Indeed, KO Rab37 in THP-1 cells significantly inhibited M2 polarization (Fig. 1k; Supplementary Fig. S1C). Moreover, we utilized GEO database (GSE14759, GSE35825, and GSE69607) to compare the gene profiles of Rab37 KO BMDMs with M1 or M2-type BMDMs in a total of 11 conditions (Supplementary Table S4). The results demonstrated that Rab37 KO phenocopied M1 BMDMs treated with LPS, IFN-α, or IFN-γ (Supplementary Fig. S1D–F), but negatively correlated with M2 BMDMs stimulated with IL-4 (Supplementary Fig. S1G). Interestingly, Rab37 was upregulated in M2-type THP-1 cells (Supplementary Fig. S1H). Collectively, these results suggest that Rab37 promotes macrophage M2 polarization by downregulating the type I IFN signaling pathway.

Rab37 expression is increased in TAMs due to epigenetic regulation

We found that Rab37 was upregulated in cancer cell CM-treated macrophages (Fig. 1h). This prompted us to investigate the transcriptional regulation of Rab37 in macrophages. Our previous study indicated a highly methylated state in Rab37 promoter of lung cancer cells, leading to gene silencing [32]. Therefore, we examined whether Rab37 in TAMs was also epigenetically regulated. Methylation-specific PCR (MSP) was performed to analyze the methylation status of the Rab37 promoter using the methylated and unmethylated-specific primers for the CpG island (from -2184 to -2322 bp) on the murine Rab37 promoter (Fig. 2a). MSP results revealed that Rab37 promoter was highly methylated in untreated RAW 264.7 cells (lanes 1 vs 2, Fig. 2b). Notably, a significant decrease in the methylation status was observed in cells treated with DNA methyltransferase inhibitor 5-Aza-2′-deoxycytidine (5-aza-dC) or LLC CM (lanes 4 and 6 vs 2, Fig. 2B). Moreover, Rab37 protein levels were upregulated in 5-aza-dC and LLC CM-treated RAW 264.7 cells (Fig. 2c), indicating that the upregulation of Rab37 in TAMs is influenced by CpG demethylation.

Fig. 2. Treatment with epigenetic inhibitor or cancer cell media upregulated Rab37 expression in macrophages, leading to the downregulation of type I IFN pathway.

a Map of CpG island (-2184~-2322 bp) and designed primers of murine Rab37 promoter. b Methylation-specific PCR analysis of the methylation status of Rab37 promoter in RAW 264.7 cells treated with 5-aza-dC (10 μM, 48 h) or LLC CM (48 h). c Western blot analysis of Rab37 protein expression in RAW 264.7 cells treated with 5-aza-dC (10 μM, 72 h) or LLC CM (72 h). d, e RT-qPCR analysis of the mRNA expression of Rab37, Ifn-α, Ifn-β, Ifit3, Irf7, Mx1, and Usp18 in DMSO or 5-aza-dC (10 μM, 72 h) treated RAW 264.7 cells. f Western blot analysis of IFN-β, IRF7, and Rab37 in DMSO or 5-aza-dC (10 μM, 72 h) treated RAW 264.7 cells. g Flow cytometry analysis of CD206 expression on RAW 264.7 cells treated with 5-aza-dC (10 μM, 96 h). h, i RT-qPCR analysis of the mRNA expression of Rab37, Ifn-α, Ifn-β, Irf7, and Oas2 in DMSO or SAHA (1 μM, 24 h) treated RAW 264.7 cells. j Western blot analysis of IFN-β, IRF7, and Rab37 in DMSO or SAHA (1 μM, 24 h) treated RAW 264.7 cells. k Flow cytometry analysis of CD206 expression on RAW 264.7 cells treated with SAHA (1 μM, 24 h). The data are shown as the mean ± s.d. (n = 3 per group). P values were determined by Student’s t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

We further confirmed the impact of two epigenetic drugs, 5-aza-dC and SAHA, a histone deacetylase inhibitor, on the expression of Rab37 and type I IFN pathway in macrophages. The results showed a significant upregulation of Rab37 mRNA expression in RAW 264.7 cells with 5-aza-dC treatment (Fig. 2d), while a downregulation of type I IFNs and ISGs mRNA (Fig. 2e) and protein expression (Fig. 2f). Notably, CD206 was significantly upregulated in macrophages treated with 5-aza-dC (Fig. 2g; Supplementary Fig. S2A). Similarly, Rab37 mRNA was upregulated after SAHA treatment (Fig. 2h), accompanied by a downregulation of type I IFNs and ISGs mRNA (Fig. 2i) and protein expression (Fig. 2j), and increased CD206 expression in macrophages (Fig. 2k; Supplementary Fig. S2B). These findings demonstrated that the expression of Rab37 in macrophages is regulated by both promoter methylation and chromatin condensation.

Moreover, we treated Rab37 KO BMDMs with two epigenetic drugs and analyzed the expression of type I IFNs and ISGs. The results showed that the mRNA and protein expression of type I IFNs and ISGs were not significantly altered in 5-aza-dC-treated (Supplementary Fig. S2C, D) or in SAHA-treated BMDMs in the absence of Rab37 (Supplementary Fig. S2E, F). Together, these findings confirm that the two epigenetic drugs upregulate Rab37 expression in macrophages, leading to the downregulation of the type I IFN pathway and promoting M2 polarization.

Rab37 attenuates p-STAT1 cytosol-nucleus translocation

Our data indicate that Rab37 regulates type I IFN pathway at the gene level. We therefore hypothesize that Rab37 may influence the transcription factors in the type I IFN pathway. STAT1 plays a crucial role in regulating the IFN pathway and the expression of ISGs [15]. Therefore, we investigated the role of Rab37 in STAT1-mediated IFN pathway. Interestingly, confocal images revealed colocalization of p-STAT1 with Rab37 in the cytosol of WT THP-1 cells, while p-STAT1 displayed predominantly nuclear translocation of Rab37 KO THP-1 cells (Fig. 3a). Consistently, the immunofluorescence (IF) images showed the same results in BMDMs (Supplementary Fig. S3A).

Fig. 3. Rab37 downregulated type I IFN pathway through attenuation of STAT1 nuclear translocation.

a Confocal images of Rab37 (green) and p-STAT1 (red) in WT and Rab37 KO THP-1 cells. Blue signals represent DAPI (nucleus staining). Z-score images of lines indicated in the merge panel are shown. Scale bars: 10 μm. b IP of STAT1 and immunoblotting with Rab37 in WT and Rab37 KO THP-1 cells. c IP of biotinylated protein and immunoblotting of STAT1 in EV and Rab37^WT-TurboID overexpressed RAW 264.7 cells. d Nucleus-cytoplasmic fractionation immunoblotting of p-STAT1, STAT1, and Rab37 in WT and Rab37 KO BMDMs. Histone H3 is a nuclear protein marker. e Predicted interaction residues between human RAB37 and STAT1 DBD domain using AlphaFold2 and Chimera. f Schematic figure of STAT1 NTD deletion (ΔNTD) and DBD deletion (ΔDBD) constructs. g IP of Rab37 and immunoblotting with GFP-tagged STAT1 in EV, WT, ΔNTD, or ΔDBD of Rab37 overexpressed 293T cells. h Protein stability analysis of p-STAT1 in WT and Rab37 KO THP-1 cells treated with CHX (50 μg/ml). i Quantification results of p-STAT1 protein half-life in WT and Rab37 KO THP-1 cells treated with CHX (50 μg/ml). j IP of p-STAT1 and immunoblotting with ubiquitin in WT and Rab37 KO THP-1 cells treated with MG132 (10 μM). The data are shown as the mean ± s.d. (n = 3 per group). P values were determined by two-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001.

These results prompted us to hypothesize that Rab37 may interact with p-STAT1 to sequester it from nuclear localization. The immunoprecipitation (IP) and reverse IP-WB demonstrated protein-protein interactions between Rab37 and STAT1 (Fig. 3b, c). Moreover, nuclear and cytoplasmic fractionation assays confirmed that more p-STAT1 translocated into the nucleus of Rab37 KO BMDMs (Fig. 3d). To examine whether the interaction between Rab37 is specific to STAT1 but not other STAT members, we performed IP-WB and reverse IP-WB analyses for STAT3, which plays a role opposite to STAT1 and primarily regulates anti-inflammatory pathways [33]. The results showed that Rab37 did not interact with STAT3 in macrophages (Supplementary Fig. S3B, C), supporting the specific interaction between Rab37 and STAT1.

Previous studies have shown that STAT1 contains two nuclear localization signals (NLS) in its N-terminal domain (NTD) and DNA binding domain (DBD), respectively [34, 35]. Through AlphaFold2 prediction and Chimera software analysis, we found that human RAB37 may interact with STAT1 DBD region (Fig. 3e; Supplementary Fig. S3D–F). Therefore, we designed an NTD deletion construct targeting NLS-1 (ΔNTD) and a DBD deletion construct targeting NLS-2 in STAT1 (ΔDBD) (Fig. 3f). IP-WB showed that only deleting the NLS-2 region (ΔDBD) disrupted the interaction between Rab37 and STAT1 (Fig. 3g). Moreover, we designed a CCD deletion (ΔCCD) construct (Supplementary Fig. S3G), and our result demonstrated that Rab37 could still interact with STAT1^ΔCCD mutant (Supplementary Fig. S3H). We also confirmed that the interaction between Rab37 and STAT1 primarily occurred in the cytoplasm rather than in the nucleus (Supplementary Fig. S3I). These results indicated that Rab37 can shield the NLS sequence of STAT1 by interacting with its DBD region, thereby inhibiting nuclear import of STAT1.

As Rab37 could sequester p-STAT1 in the cytosol, we hypothesized whether Rab37 influences the protein half-life of p-STAT1, further downregulating the type I IFN pathway. The cycloheximide (CHX) chase assay revealed that the protein stability of p-STAT1 was increased in Rab37 KO THP-1 cells compared to WT cells (Fig. 3h, i). Moreover, pull-down p-STAT1 and immunoblotting of ubiquitin revealed a higher ubiquitination level of p-STAT1 in WT THP-1 cells compared with Rab37 KO group (Fig. 3j). In addition, the protein stability of p-STAT1 was increased in WT THP-1 cells treated with proteasomal inhibitor, MG132 (Supplementary Fig S3J, K). These data suggest that Rab37 sequesters p-STAT1 in the cytosol, leading to a reduction of p-STAT1 through proteasomal degradation in macrophages.

The GDP-bound Rab37 strongly interacts with p-STAT1 to attenuate its cytosol-nucleus translocation

Rab proteins cycle between GTP-bound and GDP-bound states. GTP-bound Rabs anchor to the vesicle membrane and participate in vesicle trafficking. While GDP-bound Rabs are freely present in the cytosol, and may have an unconventional role in regulating signaling pathways [36, 37]. Therefore, we transfected empty vector control (EV), Rab37^WT, constitutive active mutant Rab37^Q89L, or trafficking-inactive mutant Rab37^T43N into RAW 264.7 cells. The confocal-IF images indicated that both Rab37^WT and Rab37^T43N sequestered STAT1 in the cytosol, with the Rab37^T43N group showing a more significant effect. In contrast, the EV group and the Rab37^Q89L group showed increased STAT1 nuclear localization (Fig. 4a, b). Furthermore, the IP-WB results demonstrated a stronger interaction between STAT1 and trafficking-inactive GDP-bound Rab37^T43N than Rab37^WT and Rab37^Q89L (Fig. 4c). These findings suggest that the GDP-bound Rab37 may have an unconventional role in sequestering and interacting with p-STAT1. In addition, we observed the colocalization of IRF7, another transcription factor in type I IFN pathway, with Rab37 in the cytosol of WT THP-1 cells (Supplementary Fig. S4A). Moreover, nuclear and cytoplasmic fractionation assays also confirmed that more IRF7 translocated into the nucleus of Rab37 KO BMDMs (Supplementary Fig. S4B). Together, these results indicate that GDP-bound Rab37 interacts with p-STAT1 to attenuate its nucleus translocation and IRF7/IFN pathway.

Fig. 4. Rab37 sequestered STAT1 in the cytosol of macrophages in a GDP-dependent manner.

Confocal images (a) and quantitative results (b) for STAT1 (red) and Rab37 (green) in EV, Rab37^WT, Rab37^Q89L, or Rab37^T43N overexpressed RAW 264.7 cells. Z-score images of lines indicated in the merge panel are shown. Blue signals represent DAPI (nucleus staining). Scale bars: 10 μm. c IP biotinylated protein and immunoblotting of STAT1 and Rab37-TurboID in EV, Rab37^WT, Rab37^Q89L, or Rab37^T43N overexpressed RAW 264.7 cells, pull-down with streptavidin beads. The data are shown as the mean ± s.d. (n = 3 per group). P values were determined by one-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Increased STAT1 and IRF7 nucleus translocation and IFN-α expression are observed in Rab37 KO TAMs compared to WT TAMs

To verify the mechanism of Rab37 on type I IFN pathway in vivo, we collected endpoint LLC lung orthotopic tumors from WT and Rab37 KO mice and performed immunofluorescence-immunohistochemistry (IF-IHC). In line with our in vitro findings, the results showed that STAT1 was retained in the cytosol of WT TAMs (Fig. 5a), while STAT1 was translocated to the nucleus in Rab37 KO TAMs (Fig. 5b, c). Consistently, the results showed that IRF7 was retained in the cytosol of WT M2 (CD163⁺)-TAMs (Supplementary Fig. S5A), while it was translocated to the nucleus in Rab37 KO M2-TAMs (Supplementary Fig. S5B, C). Additionally, the expression of IFN-α was significantly lower in WT TAMs compared with Rab37 KO TAMs (Fig. 5d–f). These in vivo results further support our hypothesis that Rab37 in TAMs downregulates the type I IFN pathway by attenuating STAT1 nuclear translocation.

Fig. 5. Rab37 attenuated STAT1 and downregulated IFN-α expression in TAMs in vivo.

Muti-plex IF-IHC images of F4/80 (blue), Rab37 (green), and STAT1, IFN-α (red) of LLC lung orthotopic tumors from WT (a, d) and Rab37 KO mice (b, e). Staining was performed using the opal stain method. Scale bars: 10 μm. Quantitative results of macrophages with STAT1 nucleus translocation (c) or IFN-α expression (f). The data are shown as the mean ± s.d. (n = 3 per group). P values were determined by Student’s t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

The expression signature of CD163⁺/Rab37⁺/STAT1^cytosol in TAMs is a biomarker for poor prognosis in lung cancer patients

We further investigated the impact of Rab37 in TAMs on lung cancer patients. We analyzed samples from a total of 48 lung cancer patients. The multi-plex IF-IHC images showed that tumor-infiltrating M2 macrophages exhibit lower Rab37 expression and more nuclear translocation of STAT1 in early-stage patients (Fig. 6a). In contrast, TAMs have higher Rab37 expression and less nuclear translocation of STAT1 in late-stage patients (Fig. 6b). Quantitative results of IF-IHC images further confirmed that CD163⁺/Rab37⁺/STAT1^cytosol of TAMs was significantly increased in late-stage lung patients (Fig. 6c). Moreover, Kaplan–Meier analysis showed that patients with higher CD163⁺/Rab37⁺/STAT1^cytosol TAMs were associated with poorer overall survival (Fig. 6d) and recurrence-free survival (Fig. 6e). We further analyzed the correlation between CD163⁺/Rab37⁺/STAT1^cytosol TAMs and clinicopathological parameters, with CD163⁺ TAMs included as a control. In univariate analysis, CD163⁺/Rab37⁺/STAT1^cytosol TAMs (P = 0.004), CD163⁺ TAMs (P = 0.030), and tumor metastasis (P = 0.048) were significantly associated with mortality. However, in the multivariate Cox regression analysis, only CD163⁺/Rab37⁺/STAT1^cytosol TAMs (P = 0.040) and tumor metastasis (P = 0.010) remained significantly associated with mortality in lung cancer patients, while CD163⁺ TAMs (P = 0.832) did not (Table 1). In addition, high expression of CD163⁺/Rab37⁺/STAT1^cytosol TAMs in lung cancer patients was associated with male (P = 0.034), advanced tumor stage (P < 0.001), lymph node metastasis (P < 0.001), differentiation grade (P = 0.049), and recurrence (P = 0.002) (Supplementary Table S5). These data suggest that the expression signature of CD163⁺/Rab37⁺/STAT1^cytosol in TAMs serves as a biomarker for poor lung cancer prognosis.

Fig. 6. The expression signature of CD163⁺/Rab37⁺/STAT1^cytosol in TAMs was a biomarker for poor prognosis in lung cancer patients.

Multi-plex IF-IHC images of CD163 (blue), Rab37 (green), and STAT1 (red) of tissue microarray from early-stage (a) and late-stage (b) lung cancer patients. c Quantitative results of CD163⁺/Rab37⁺/STAT1^cytosol TAMs expression in early and late stage lung cancer patients. d Kaplan–Meier analysis of the overall survival in CD163⁺/Rab37⁺/STAT1^cytosol high and low lung cancer patients. e Kaplan–Meier analysis of the recurrence-free survival in CD163⁺/Rab37⁺/STAT1^cytosol high and low lung cancer patients. The data are shown as the mean ± s.d. (n = 3 per group). P values were determined by Student’s t test. *P < 0.05, **P < 0.01, ***P < 0.001, **** P < 0.0001.

Table 1.

Cox regression analysis of risk factors for cancer-related death in lung cancer patients.

Characteristics	Univariate analysis		Multivariate analysis
	HRa (95% CIb)	P-valuec	HR (95% CI)	P-value
CD163+/Rab37+/STAT1cytosol cells (%)d
<4.26%	1		1
≥4.26%	3.225 (1.450–7.171)	0.004	3.369 (1.060–10.709)	0.040
CD163+ cells (%)e
<8.05%	1		1
≥8.05%	2.364 (1.087–5.141)	0.030	1.129 (0.370–3.446)	0.832
Age
<65	1
≥65	0.699 (0.327–1.495)	0.356	–g
Gender
Female	1
Male	0.651 (0.307–1.377)	0.261	–g
Stage
Stage I-II	1
Stage III-IV	4.005 (1.809–8.869)	0.001	–g
T stagef
T1	1
T2	1.734 (0.711–4.230)	0.226	–g
T3	5.280 (1.274–21.877)	0.022	–g
T4	0.946 (0.196–4.572)	0.945	–g
N stagef
N0	1
N1	2.724 (0.585–12.685)	0.202	–g
N2	4.906 (2.106–11.431)	<0.001	–g
N3	22.873 (2.427–215.539)	0.006	–g
M stagef
M0	1		1
≥M0	4.464 (1.017–19.601)	0.048	7.781 (1.631–37.105)	0.010
Differentiation grade
Well	1
Moderate	2.273 (0.677–7.635)	0.184	–g
Poor	7.055 (1.406–35.397)	0.018	–g
Recurrence
No	1
Yes	4.003 (1.792–8.942)	0.001	–g

^aHR Hazard ratio.

^bCI Confidence interval.

^cBold values indicate statistical significance (P < 0.05).

^dPatient with CD163⁺/Rab37⁺/STAT1^cytosol cells (%) ≥4.26% of ROI (200 × 200 μm) was defined as high expression.

^ePatient with CD163⁺ cells (%) ≥8.05% of ROI (200 × 200 μm) was defined as high expression.

^fT status: primary tumor; N status: lymph node metastasis; M status: distant metastasis.

^gThe variables without significant HR in the univariate analysis were not included in the multivariate analysis.

Discussion

In this study, we discover that the trafficking-inactive GDP-bound Rab37 interacts with the NLS sequences of the transcription factor STAT1. This interaction inhibits STAT1 nuclear translocation, thus downregulating the type I IFN pathway and ultimately promoting M2 polarization of macrophages. Furthermore, stimulation with lung cancer CM can trigger epigenetic changes, including promoter de-methylation and chromatin de-condensation. This leads to an upregulation of Rab37 expression and further promotes M2 polarization in macrophages. Interestingly, the presence of Rab37 significantly reduces the protein half-life of p-STAT1 compared to the Rab37 KO group, suggesting that Rab37 interaction with STAT1 contributes to the protein instability of STAT1. The tumor-infiltrating TAMs predominantly have CD163⁺/Rab37^-/STAT1^nucleus signature in Rab37 KO group, consistent with the in vitro results. Clinically, CD163⁺/Rab37⁺/STAT1^cytosol in TAMs serves as a poor prognostic marker of lung cancer patients.

The present study on Rab37 focuses on its unconventional role apart from vesicle trafficking. Many Rab family proteins can regulate signaling pathways and affect tumor progression. For example, Rab1a regulates mTORC1 signaling to promote tumor invasion; and Rab2a influences Erk1/2 signaling to facilitate tumor progression [25–28]. Our study uncovers a previously unidentified non-canonical role of Rab37 in regulating the IFN pathway. Rab37 interacts with and sequesters STAT1 in the cytosol, inhibiting its nuclear translocation. Further investigations reveal that Rab37 exerts stronger inhibition of STAT1 nuclear translocation in its trafficking-inactive state compared to its active state, suggesting that GDP-bound Rab37, mainly present in the cytosol, effectively regulates the IFN pathway. Notably, the IP-WB demonstrated that Rab37 did not interact with IRF7 (Supplementary Fig. S4C). Furthermore, the protein half-life of IRF7 remained unchanged in both WT and Rab37 KO THP-1 cells treated with CHX (Supplementary Fig. S4D, E) or MG132 (Supplementary Fig. S4F, G), suggesting that IRF7 localization is a downstream effect of Rab37-mediated p-STAT1 inactivation. These results indicate that trafficking-inactive Rab37 sequesters p-STAT1, but not IRF7, to prevent its nuclear translocation.

Our study elucidated that the upregulation of Rab37 expression in lung cancer cell CM-educated macrophages is caused by epigenetic regulation. However, which factors within the cancer cell-CM affect the epigenetic changes in macrophages is still unknown. Previous studies have indicated that IL-6 released by tumor myoepithelial cells promotes the degradation of DNA methyltransferase 1 in endothelial cells through the proteasomal degradation pathway in breast cancer TME. This leads to hypomethylation of the vascular endothelial growth factor receptor 2 (VEGFR2) promoter and increases VEGFR2 transcription through activating the STAT3 transcription factor [38]. Another study on glioma stem cells found that activated STAT3 in cancer cells upregulates the expression of DNA dioxygenase, ten-eleven translocation enzyme 3 (TET3), promoting gene demethylation [39]. Whether cancer cell-CM influences the epigenetic regulation of macrophages through the IL-6/DNMT1 or IL-6/STAT3/TET3 axis warrants further investigation. Moreover, whether co-treatment with 5-Aza and SAHA synergistically enhances Rab37 expression merits additional study.

Nuclear translocation of STAT1 requires initial phosphorylation at the Y701 site and SH2-mediated dimerization, followed by recognition of the NLS sequence on the DBD domain by importin-α5 [34]. Additionally, one study has suggested that the NTD domain of STAT1 also influences STAT1 nuclear translocation, but the detailed mechanisms remain unclear [33]. Our IP-WB results demonstrate a decreased interaction between Rab37 and STAT1 DBD domain deletion mutant, suggesting that Rab37 may compete with importin-α5 for the NLS sequence on the STAT1 DBD domain, leading to reduced STAT1 nuclear translocation. Interestingly, we found that Rab37 potentially binds to the DBD region of STAT1, and the binding site was very close to where importin-α5 binds to STAT1 using AlphaFold2 prediction tool. Further structural analyses will confirm this predicted 3D structure on human Rab37 and STAT1 interaction. Additionally, whether Rab37 interacts with other STAT members requires further examination.

In previous studies, it is known that STAT1 undergoes ubiquitination mediated by the E3 ligase suppressor of cytokine signaling 5 (SOCS5), leading to its proteasomal degradation [40], or undergoes de-ubiquitination through ubiquitin-specific peptidase 22 (USP22) or USP2a, enhancing STAT1 protein stability [41, 42]. We observed a significant increase in the half-life of p-STAT1 in Rab37 KO cells compared to the WT group. Moreover, treatment with the proteasome inhibitor MG132 significantly prolonged the half-life of p-STAT1 in cells. These findings suggest that the interaction between Rab37 and STAT1 may reduce STAT1 nuclear translocation and promote its proteasomal degradation.

Our clinical analysis confirms that TAMs with high Rab37 expression sequester STAT1 in the cytoplasm, while those with low Rab37 expression show enhanced nuclear translocation of STAT1 in tumor specimens from lung cancer patients. Particularly, fewer M2-type TAMs with low Rab37 expression exhibit increased nuclear translocation of STAT1 in early-stage lung cancer. In contrast, more M2-type TAMs with high Rab37 expression show cytoplasmic accumulation of STAT1 in late-stage lung cancer. These clinical results show for the first time that the expression signature of CD163⁺/Rab37⁺/STAT1^cytosol in TAMs can serve as a potential prognostic biomarker in lung cancer.

In conclusion, we found that GDP-bound Rab37 in macrophages inhibits STAT1 nuclear translocation and promotes its degradation. GDP-bound Rab37 also downregulates the IFN pathway and induces M2 macrophage polarization. The lung TME alters the epigenetic profile of TAMs, thereby enhancing Rab37 expression. Clinically, Rab37 expression in TAMs increases with advancing stages, further promoting M2 polarization, and is associated with lower overall survival rates in lung cancer patients. Our findings suggest that the CD163⁺/Rab37⁺/STAT1^cytosol signature in TAMs predicts a poor prognosis for lung cancer patients. Targeting Rab37 specifically within macrophages to reduce its expression may serve as a new strategy for cancer therapy.

Author contributions

Chen-Tai Hong: Writing – original draft, Writing – review & editing, Conceptualization, Methodology, Visualization, Formal analysis, Data curation. You-En Yang: Writing – original draft, Writing – revised draft, Writing – review & editing, Conceptualization, Methodology, Data curation, Investigation, Visualization. Hsueh-Fen Juan: Software, Data curation, Investigation. Chih-Peng Chang: Methodology, Validation, Investigation. Yi-Ching Wang: Supervision, Project administration, Conceptualization, Validation, Investigation, Funding acquisition, Data curation, Writing - original draft, Writing - review & editing.

Funding

This work was supported by Taiwan Ministry and Science Technology grant MOST 109-2327-B-006-004 and Taiwan National Science and Technology Council grant NSTC 112-2311-B-006-004-MY3.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

All experiments using mice were approved by the animal ethics committee of National Cheng Kung University and complied with all relevant ethical guidelines (#112062). All lung cancer samples were conducted in accordance with the requirements of Research Center of Clinical Medicine, National Cheng Kung University Hospital. The use of clinical samples was approved by the institutional review board of the hospital with the ethical number #A-ER-111-517. All patients provided informed consent for the use of the tumor tissues for research.

Consent for publication

All authors have reviewed the final version of the manuscript and are in agreement its content and submission.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-025-02955-0.