SAMSN1 restrains NK cell mediated anti-tumor immunity in hepatocellular carcinoma

Abstract

NK cells are critical mediators of anti-tumor immunity whose function is frequently compromised in the tumor microenvironment. Here we identify SAM domain, SH3 domain and nuclear localization signals 1 (SAMSN1) as a previously unrecognized immune checkpoint that predominantly regulates NK cell function in hepatocellular carcinoma (HCC). Single-cell RNA sequencing (scRNA-seq) analysis reveals significant SAMSN1 upregulation in intratumoral NK cells from HCC patients, correlating with reduced granzyme B expression and poor prognosis. In orthotopic Hepa1-6 hepatocellular carcinoma models, global Samsn1 knockout (Samsn1^−/−) mice exhibits 34% tumor burden reduction with enhanced NK cell granzyme B production (P = 0.0002). Critically, NK cell-specific deletion alone (Samsn1^f/f-Ncr1^Cre+) recapitulates this therapeutic effect (41% tumor burden reduction, P = 0.0017), demonstrating that SAMSN1 functions predominantly through intratumoral NK cells rather than other immune populations in the HCC microenvironment. Mechanistically, SAMSN1 suppresses NK cell activation, proliferation, and granzyme B production. These findings indicate SAMSN1 as a targetable NK cell checkpoint with direct therapeutic implications for HCC immunotherapy.

NK cell dysfunction in tumor microenvironment remains elusive. The authors here identify SAMSN1 as an immune checkpoint that mediates NK function. Specifically, global or NK cell-specific deletion of SAMSN1 restores NK function and improves prognosis in hepatocellular carcinoma mouse models.

Introduction

Hepatocellular carcinoma (HCC) is the sixth most common cancer and the third leading cause of cancer-related deaths worldwide¹. Despite advances in diagnostic techniques and treatment modalities, the prognosis for patients with HCC remains poor, with a 5-year survival rate of approximately 18%^2–4. This outcome is often attributed to late-stage diagnosis, high recurrence rates, and the limited efficacy of available treatments^5–7. Current management strategies for HCC include surgical resection, liver transplantation, local ablation, transarterial chemoembolization (TACE), and systemic therapies^6–8. Immune checkpoint inhibitors (ICIs) have demonstrated efficacy in several cancer types by reinvigorating exhausted T cells and restoring anti-tumor immunity^9–11. In HCC, the combination of atezolizumab (anti-PD-L1) and bevacizumab (anti-VEGF) has become the standard of care for first-line treatment of advanced disease, based on the IMbrave150 trial, which showed improved overall survival compared with sorafenib^8,12,13. However, the overall response rate to ICIs in HCC remains low, at 20–30%^14,15. The efficacy of immunotherapy in HCC faces several challenges, including an immunosuppressive tumor microenvironment (TME) characterized by regulatory T cells, myeloid-derived suppressor cells, and tumor-associated macrophages^14,16.

The heterogeneity of HCC, at both the genetic and immunological levels, complicates treatment strategies¹⁵. NK cells, which are components of the innate immune system, can recognize and eliminate transformed cells without prior sensitization^17,18. In HCC, NK cells constitute a significant proportion of resident lymphocytes^19–21. Studies have demonstrated that NK cell infiltration in HCC tumors correlates with patient prognosis²². Tumor-infiltrating NK cells (TINKs) can directly recognize and kill HCC cells through various mechanisms, including the detection of stress-induced ligands and the absence of major histocompatibility complex (MHC) class I molecules^22,23. In HCC specifically, NK cells face unique challenges. The liver harbors the largest population of tissue-resident NK cells in the body, comprising up to 30–50% of intrahepatic lymphocytes. However, within the HCC microenvironment, NK cells often exhibit impaired functionality due to the presence of multiple immunosuppressive factors^19,20,23. This NK cell exhaustion correlates with HCC progression and poor prognosis, yet the molecular checkpoints driving this dysfunction remain incompletely defined. Identifying targetable NK cell checkpoints specific to the HCC microenvironment represents a critical unmet need for improving immunotherapy efficacy in HCC. SAMSN1, also known as HACS1 or SLY2, is an adaptor protein initially identified in leukemia²⁴. Adaptor proteins facilitate protein-protein interactions and coordinate the assembly of signaling complexes^25–27. SAMSN1 has been implicated in various cellular processes, particularly in immune cell regulation^28,29. In B cells, SAMSN1 modulates activation and differentiation processes²⁸. It promotes immunosuppression in sepsis by programming macrophages to express coinhibitory molecules via KEAP1-NRF2 signaling, thereby inducing CD8⁺ T cell exhaustion characterized by impaired IFN-γ and TNF-α production and sustained PD1 and TIM3 expression³⁰. In cancer biology, altered SAMSN1 expression has been reported in various malignancies^30–32. SAMSN1 exhibits context-dependent tumor-suppressive functions in hematopoietic malignancies^31,32. Despite emerging evidence of SAMSN1’s immunoregulatory functions, its role as a potential NK cell checkpoint in solid tumors, particularly HCC, where NK cells comprise up to 30% of tumor-infiltrating lymphocytes, has not been investigated. This knowledge gap is critical given the limited success of current NK cell-based immunotherapies in HCC.

Our study investigates the role of SAMSN1 in NK cell regulation within the HCC tumor microenvironment, filling a gap in understanding innate immune checkpoints for cancer therapy. Our findings reveal that SAMSN1 is specifically upregulated in intratumoral NK cells, where it suppresses cytotoxic activity and correlates with poor patient prognosis. Genetic deletion of SAMSN1 restores NK cell effector functions, resulting in tumor growth inhibition and improved survival in HCC models. These results establish SAMSN1 as a promising target for enhancing NK cell-based immunotherapy, with broader implications for treating malignancies that evade innate immune surveillance.

Results

SAMSN1 expression in tumor microenvironment is associated with a poor prognosis

To validate the correlation between SAMSN1 expression and tumorigenesis, we analyzed Single-cell RNA sequencing (scRNA-seq) data of 18 patients with HCC from published paper³³. scRNA-seq analysis revealed heterogeneous SAMSN1 expression across different cell populations within the tumor microenvironment (Fig. 1a, b). To elucidate the relationship between SAMSN1 expression and HCC development, specimens were classified as intratumor (IT; core tumor region) or para tumor (PT; tissue ≤ 2 cm from tumor margin, pathologically confirmed to contain mixed tumor/stromal cells), respectively (Fig. 1c). We found that the NK cells (1.35-fold) exhibited markedly higher SAMSN1 expression than T cells (1.21-fold), B cells (1.07-fold), myeloid (0.97-fold), parenchymal (1.0-fold), and tumor cells (0.61-fold) in IT compared to PT regions (Fig. 1d). This preferential upregulation in NK cells within the HCC microenvironment suggests a specific role for SAMSN1 in suppressing NK cell-mediated anti-tumor immunity in HCC. Other cell types showed minimal or no SAMSN1 upregulation in IT regions. This suggests that SAMSN1 may be involved in the poor prognosis of HCC.

Fig. 1. SAMSN1 is mainly expressed on lymphocytes and associated with an unfavorable prognosis in patients with HCC.

a–c Single-cell RNA sequencing (scRNA-seq) analysis of 18 HCC patients from China National GeneBank DataBase (CNGBdb). Cells were clustered by their gene expression. NK cells, natural killer cells. a T-distributed Stochastic Neighbor Embedding (t-SNE) plot, showing the annotation and color codes for cell types in the HCC ecosystem with 25,972 cells. b The expression distribution of SAMSN1 across different clusters. c Expression of SAMSN1 in para tumor (PT) and intratumor (IT) samples. d The phenotypic profiles of cells were compared based on the expression of SAMSN1 in PT or IT regions of HCC patients. e, f Kaplan–Meier survival analysis of HCC patients stratified by SAMSN1 expression levels with HCC cohort from The Cancer Genome Atlas (TCGA) database. Patients were divided into SAMSN1-high and SAMSN1-low groups and we identify the best cut-off value for distinguishing between high and low expression using R language. e Overall survival (OS) and f Disease-free survival (DFS) in patients. P values and hazard ratios (the shaded areas correspond to 95% confidence intervals) were calculated using the log-rank test. g XY-plots showing the correlation of SAMSN1 expression with expression of indicated genes in NK cells (KLRC1, TIGIT, TIM3, LAG3) (n = 151). Statistics: significance of Spearman’s rank correlation coefficients for indicated genes was tested using a two-sided t-test. h Schematic representation of the experimental strategy with HCC cohort1 (n = 88). i IHC staining was performed to detect SAMSN1 in PT (left) and IT (right) regions. The lower panels represent high-magnification views of the areas indicated by dashed boxes in the upper panels. Brown staining indicates SAMSN1 expression, and blue staining indicates nuclei counterstained with hematoxylin. Scale bars: 100 μm (top panels) and 20 μm (bottom panels). j Immunohistochemistry score (IOD/Area) of SAMSN1 are statistic in IT compared to PT, as determined by a paired two-sided t-test. k, l Kaplan–Meier survival analysis of HCC patients stratified by SAMSN1 expression levels with HCC cohort1. Patients were divided into SAMSN1-high and SAMSN1-low groups using IHC-based H-scores. k OS and l DFS were presented. P values and hazard ratios (the shaded areas correspond to 95% confidence intervals) were calculated using the log-rank test.

Next, to elucidate the mechanisms driving the poor prognosis associated with high SAMSN1 expression in HCC, we analyzed The Cancer Genome Atlas (TCGA) data from Liver Hepatocellular Carcinoma samples available in the cBioPortal database. We stratified the cohort into SAMSN1^high and SAMSN1^low groups based on best cut-off option. We found that SAMSN1 expression was correlated with significantly reduced overall survival (OS) (Fig. 1e) and disease-free survival (DFS) (Fig. 1f). Expression analysis demonstrated strong positive correlations between SAMSN1 and the key immune-related genes, KLRC1 (Spearman r = 0.59), TIM3 (Spearman r = 0.89), TIGIT (Spearman r = 0.77), and LAG3 in NK cell subset (Spearman r = 0.54) (Fig. 1g). Notably, these correlations were weaker or absent in CD8⁺ T cells (SAMSN1-TIM3 correlation in T cells), further supporting NK cell-specific regulatory function in the HCC context.

To resolve the tumor ecosystem in HCC, we collected surgical tumor specimens and paired adjacent non-tumor tissues from primary HCC patients (cohort1) for to perform Immunohistochemistry (IHC) and cytometry by time of flight (CyTOF) (Fig. 1h). Detailed clinical information is provided of cohort1 (Supplementary Table 1). IHC was performed to verify the spatial distribution of SAMSN1 in HCC tissues (Fig. 1i). The results revealed that integrated optical density (IOD) of SAMSN1 staining was significantly higher SAMSN1 expression within IT regions than in PT areas (Fig. 1j). Subsequently, SAMSN1 expression levels were quantified in HCC patient samples via IHC staining. Samples were stratified into SAMSN1^high and SAMSN1^low groups using IHC-based H-scores. Patients were dichotomized by the cohort-wide median H-score (median = 120). OS and disease-free survival (DFS) were then analyzed across these groups. The OS (Fig. 1k) and DFS (Fig. 1l) was further validated. These findings highlight SAMSN1’s differential expression in the HCC microenvironment, prognostic potential, and association with immune-related genes, suggesting a potential role in tumor immunity.

SAMSN1 restrains activity of immune cells in IT

To further elucidate the mechanism connecting SAMSN1 to diminished anti-tumor immunity in HCC, we isolated tumor-infiltrating lymphocytes (TILs) from HCC cohort2 for CyTOF assay. Detailed clinical information is provided of cohort2 (Supplementary Table 2). The t-SNE analysis revealed distinct clustering of immune cell populations in both PT and IT samples (Fig. 2a). Among the 22 samples processed, 8 samples were excluded from downstream analysis due to failure to meet quality control criteria. Quantification of CD45⁺ cells showed significant differences in the populations of NK and CD4⁺ T cell between PT and IT, with NK cells being more abundant in PT (P = 0.031) and CD4⁺ T cells being more abundant in IT (P = 0.0035) (Fig. 2b).

Fig. 2. SAMSN1 expression and functional markers in tumor-infiltrating immune cells from HCC patients.

a–j Mass cytometry (CyTOF) was used to compare the phenotypic profiles of cells isolated from PT and IT regions of HCC patients in cohort2 (n = 22). The analysis included panels for B cells, CD4 T cells, CD8 T cells, NK cells, NKT cells, myeloid cells, and double negative (DN) cells with specific markers assessed to delineate their phenotypic characteristics. a t-SNE plot showing the annotation and color codes for cell types in the HCC ecosystem. NKT cells, natural killer T cells; DN cells, Double negative cells (n = 14). b The indicated immune cells were isolated from lymphocytes (CD45⁺) and the percentage of cells was quantified (n = 14). Statistical significance was determined using a two-sided unpaired t-test. c Differential abundance of immune cells assessed by expression of cytotoxic molecules (GZMB, Perforin) and cytokines (TNFα, IFNγ) (n = 14). The data are presented as means ± SEM. Statistical significance was determined using a two-sided unpaired t-test. d The different subsets of immune cells were quantified in GZMB⁺ cells (n = 14). Statistical significance was determined using a two-sided unpaired t-test. e Flow cytometry plots shown proportion of GZMB⁺ NK cells (n = 14). f t-SNE plot showing the annotation and color codes for SAMSN1 expression in the HCC ecosystem (n = 14). g The percentage of SAMSN1⁺ different subsets of immune cells were quantified (n = 14). The data are presented as means ± SEM and compared using the two-sided unpaired t-test. h The percentage of SAMSN1⁺ CD45⁺ cells were quantified in PT and IT (n = 14). Statistical significance was determined using the two-sided unpaired t-test. Data are presented as mean ± SEM. i Quantification of GZMB⁺ cells derived from SAMSN1⁺ or SAMSN1⁻ NK cells (n = 14). Statistical significance was determined using paired two-sided t-test. j XY-plots showing the correlation of SAMSN1 expression with expression of indicated proteins (GZMB, CXCR3, LAG3, KLRC1, TIM3, TIGIT). Correlation between SAMSN1 and other proteins based on the protein expression in PT and IT samples. Statistics: significance of Spearman’s rank correlation coefficients for indicated genes was tested using a two-sided t-test.

To further assess the effector functions of these cells, we examined the expression levels of indicated effectors in CyTOF. The gating strategy was shown (Supplementary Fig. 1a). Examination of effector production revealed that granzyme B (GZMB) expression was markedly higher in PT compared to IT (Fig. 2c). This was particularly evident in NK and CD8⁺ T cells (Fig. 2d). NK cells in IT area displayed lower GZMB abundance than PT area (Fig. 2e). Further analysis of cell subsets revealed no significant differences in IFN-γ (Supplementary Fig. 1b), Perforin (Supplementary Fig. 1c), TNF-α (Supplementary Fig. 1d), and GM-CSF (Supplementary Fig. 1e). Similarly, the inhibition markers of TIGIT (Supplementary Fig. 1f) and CD96 (Supplementary Fig. 1g) also showed no significant variation between PT and IT samples.

Further, we next explored the SAMSN1 levels in PT and IT samples by flow cytometry. The t-SNE analysis revealed distinct clustering of SAMSN1 in both PT and IT samples (Fig. 2f). SAMSN1 was strongly upregulated in myeloid cells and NK cells in IT samples (Fig. 2g). SAMSN1 expression was significantly elevated in CD45⁺ cells of IT samples (Fig. 2h). Meanwhile, GZMB expression was significantly decreased in SAMSN1⁺ NK cells compared with SAMSN1^- NK cells (Fig. 2i). Correlation analyses demonstrated significant negative correlations between SAMSN1 and effector molecules, such as GZMB (Spearman r = −0.543) and CXCR3 (Spearman r = −0.02), and positive correlations with immunosuppressive markers, such as LAG3 (Spearman r = 0.077), KLRC1 (Spearman r = 0.311), TIM3 (Spearman r = 0.469), TIGIT (Spearman r = 0.457) and at the protein level (Fig. 2j). In conclusion, our analysis of tumor-specific cells from patients showed upregulation of SAMSN1 in IT area, suggesting GZMB⁺ cells exhibited a significant reduction, decreasing from 13.8% to 6.1% in NK cells (P = 0.0240) and from 19.4% to 8.2% in T cells (P = 0.0099) within the IT region of patients. SAMSN1⁺ NK cells showed 6.5% reduction (from 9.48% to 2.98%) in granzyme B (GZMB) expression (P = 0.0055). However, SAMSN1⁺ CD8⁺ cells were no difference in IT when compared with PT samples. GZMB⁺ NK cells showed 5% reduction (from 42% to 37%) in SAMSN1⁺ cells when compared with SAMSN1⁻ cells (P = 0.0340). The correlation between SAMSN1 and inhibitory markers (TIM3, TIGIT) was strongest in NK cells. Taken together, CyTOF analysis identified SAMSN1 upregulated in NK cell populations that contribute to the anti-tumor immune escape. These patient-derived findings provide the clinical rationale for subsequent genetic dissection of SAMSN1’s cell-type-specific functions in preclinical models, with NK cells emerging as the primary therapeutic target based on their predominant expression and functional correlation with SAMSN1.

SAMSN1 mediates NK cell dysfunction

To determine how SAMSN1 regulate the anti-tumor immunology, we enriched NK cells from human peripheral blood mononuclear cells (PBMCs) and performed SAMSN1 knockout using CRISPR-Cas9 technology (Fig. 3a). The GFP⁺ cells were gated after transfected with Cas9 RNPs complexes (Fig. 3b). The knockout efficiency was detection in mRNA level (Fig. 3c) and protein level by flow cytometry (Fig. 3d). The results indicated that we successfully generated SAMSN1 knockout primary human NK cells using CRISPR-Cas9. Moreover, we recruited SAMSN1 knockout NK cells to perform bulk RNA-seq. Bulk RNA-seq analysis revealed that SAMSN1 knockout dramatically increased surface expression of activation markers while decreasing inhibitory receptors on NK cells (Fig. 3e). This phenotype suggests that SAMSN1 constrains NK cell activation and that its ablation potentiates their anti-tumor function. Gene Ontology (GO) analysis highlighted the involvement of SAMSN1 in various immune-related processes including leukocyte activation, cell adhesion, and lymphocyte activation (Fig. 3f). Gene Set Enrichment Analysis (GSEA) of SAMSN1 knockout human NK cells confirmed that SAMSN1 knockout was associated with positive regulation of NK cell activation (Fig. 3g), suggesting a potential role of SAMSN1 in regulating NK cell function.

Fig. 3. SAMSN1 was associated with NK cell activation and function.

a Schematic diagram of generating experimental workflow. NK cells were enrichment from PBMCs to knockout SAMSN1 with electroporation. GFP⁺ NK cells were sorted and subjected to bulk RNA-seq. b GFP-positive cells were quantified 72 h after transfection by flow cytometry (n = 4). c SAMSN1 mRNA level was shown (n = 3). Statistical significance was determined using unpaired two-sided t-test. d Cells were fixed/permeabilized and stained with anti-SAMSN1 72 h post-transfection (n = 4). Statistical significance was determined using paired two-sided t-test. e Heatmap showing differential expression of NK cell activation and inhibition markers in human primary NK cells 72 h post-transfection (n = 2). Differentially expressed genes were identified using DESeq2 (adjusted P < 0.05). f GO enrichment analysis was performed on differential expression genes (DEGs). The bar chart shows the top 11 significantly enriched GO biological processes. Enrichment analysis was conducted using DESeq2 and clusterProfiler, with a false discovery rate (FDR) of 0.05. g GSEA was performed to identify enriched pathways and key effector molecules. h Schematic of generating co-culture model. NK92 cells (NK92^Control, NK92^KO-SAMSN1, or NK92^KO-SAMSN1, 5 × 10³ cells each) were co-cultured with HepG2 cells (5 × 10³) for 24 h. i Flow cytometry analysis of perforin, IFN-γ, and GZMB expression in NK92 cells from the co-culture (n = 4). Statistical significance was determined by one-way ANOVA with Tukey’s post-hoc test. j Schematic of the experimental workflow. Spleen were collected for flow cytometry analysis from C57BL/6J WT and Samsn1^−/− mice (n = 6). k Frequencies of CD4⁺ T cells, CD8⁺ T cells, and NK cells in splenocytes, assessed by flow cytometry (n = 6). l Percentage of Ki-67-positive splenocytes, detected by intracellular staining with anti-Ki-67 antibody (n = 6). m Frequency of GZMB-positive lymphocytes in splenocytes after stimulation in vitro with PMA (25 ng/ml), ionomycin (1 μg/ml), and brefeldin A (5 μg/ml) for 4 h (n = 6). All data are presented as mean ± SEM. In (j–m), groups were compared using two-sided Mann–Whitney U-tests.

Next, to functionally elucidate SAMSN1’s function in NK cells, we generated the NK92 knockout (NK92^KO-SAMSN1) cell line with CRISPR-Cas9 technology and overexpression (NK92^OE-SAMSN1) cell line using lentivirus infection. Next, we generated a co-culture model of NK92 cells with HepG2 (Fig. 3h). Knockout of SAMSN1 promoted GZMB, Perforin, and IFN-γ secretion in NK92 cells, whereas SAMSN1 overexpression inhibited cell function (Fig. 3i).

To further explore the function of SAMSN1 in NK cells, we generated the Samsn1 knockout (Samsn1^−/−) mice and detected the function of SAMSN1 in splenocytes. The workflow is shown (Fig. 3j). The generation of Samsn1^−/− mice and genotyping was performed as shown (Supplementary Fig. 2a, b). Analysis of splenocytes from wild-type (WT) and Samsn1^−/− mice revealed no significant differences in immune cell populations (Fig. 3k). However, Samsn1^−/− mice exhibited increased percentages of Ki67⁺ NK cells compared with WT mice in spleen (Fig. 3l). In contrast, the percentages of Ki67⁺CD4⁺ T cells, and CD8⁺ T cells showed no differences (Fig. 3l). Additionally, GZMB⁺ NK cells were significantly elevated in the spleen of Samsn1^−/− mice (Fig. 3m). The gating strategy for flow cytometry is shown (Supplementary Fig. 2c, d). To verify the impact of Samsn1 knockout on GZMB and IFN-γ secretion in splenic NK cells under naive conditions, we compared secretion levels under both naive and PMA/ionomycin-stimulated conditions. The results showed that Samsn1^−/− NK cells exhibited increased GZMB and IFN-γ secretion compared to WT NK cells under PMA/Ionomycin stimulation. However, in the naive state, no significant differences were observed due to the inherently low expression levels of GZMB and IFN-γ (Supplementary Fig. 2e, f). These findings suggest that SAMSN1 regulates NK cell cytotoxic activity against tumor cells.

Samsn1 knockout limits HCC tumor growth and enhances NK cell function

To definitively establish SAMSN1’s role in HCC, we employed orthotopic Hepa1-6 liver tumor models, which recapitulate the immunosuppressive hepatic microenvironment characteristic of human HCC. This model was selected over subcutaneous implantation to preserve liver-resident NK cell populations, recreate the fibrotic milieu present in most HCC patients, and enable physiologically relevant immune cell trafficking patterns (Fig. 4a). Samsn1^−/− mice demonstrated robust anti-tumor efficacy in this HCC-relevant model. Tumor burden, quantified as liver-to-body weight ratio, decreased from 8.8% in wild-type compared to 5.8% in Samsn1^−/− mice (P = 0.0024), representing a 34% reduction (Fig. 4b, c). To determine the cellular mechanism underlying this therapeutic effect, we analyzed TILs from liver tumors at day 21 post-implantation. NK cells from Samsn1^−/− mice exhibited enhanced effector function, with GZMB expression increasing from 49.9% to 63.6% (P = 0.0002, Fig. 4d) and IFN-γ production increasing from 30.9% to 46.7% (P = 0.0050, Supplementary Fig. 3a). This translated to significant survival benefit, with median survival increasing from 27.5 days to 36 days (31% extension, P = 0.0072, Fig. 4e). Critically, these functional enhancements were specific to NK cells. CD8⁺ T cells showed no significant changes in GZMB (4.36% vs. 7.27%, P = 0.22, Supplementary Fig. 3b) or IFN-γ (5.34% vs. 6.51%, P = 0.07, Supplementary Fig. 3c) expression in Samsn1^−/− tumors compared to wild-type controls, despite comparable T cell infiltration. NK cell, CD4⁺ T cell, and CD8⁺ T cell frequencies in TILs were similar between genotypes (Supplementary Fig. 3d–f), indicating that therapeutic efficacy resulted from enhanced NK cell function rather than altered immune cell recruitment. To exclude potential confounding effects from direct tumor cell targeting, we verified that Hepa1-6 cells lack detectable SAMSN1 expression by western blot (WB) and real time PCR (RT-PCR) (Supplementary Fig. 3g, h). This confirms that observed anti-tumor effects derive from immune modulation rather than tumor cell-intrinsic mechanisms. Collectively, these data establish SAMSN1 as a bona fide NK cell checkpoint in hepatocellular carcinoma, with NK cell-intrinsic SAMSN1 deletion sufficient to suppress HCC tumor growth and prolong survival.

Fig. 4. Samsn1 knockout suppresses HCC tumor growth through enhanced NK cell function.

a Schematic of the experimental design for Hepa1-6 orthotopic model. Murine hepatoma Hepa1-6 cells (5 × 10⁵) were subcutaneously injected into the livers of WT and Samsn1^−/− mice (n = 4). b After 21 days, mice were euthanized by carbon dioxide inhalation, and livers were carefully removed, photographed, weighed, and representative images are shown in the panel. c The mean liver weight relative body weight was calculated (n = 4). The data are presented as means ± SEM and compared using the two-sided unpaired t-test. d Frequencies of GZMB positive tumor-infiltrating NK cells (TINKs) after stimulation with PMA (25 ng/ml), ionomycin (1 μg/ml), and brefeldin A (5 μg/ml) for 4 h, compared using unpaired two-sided t-test. e Kaplan–Meier survival analysis in tumor bearing mice (n = 4). P value was calculated using the log-rank test. a–e show data from biological replicates (n = 4 mice per group). Each data point represents one mouse, with no technical replicates used for statistical analysis. No data were excluded from the analyses. f Additional validation in MC38 models is shown. g Tumor growth curves over 21 days measured by caliper (volume calculated as length × width² × 0.5 mm³) and assessed by two-way ANOVA with Holm–Šidák multiple-comparison test (n = 5). h Tumor sizes were shown at day 21. i Mean tumor weights at day 21, measured after excision (n = 5). The data are presented as means ± SEM and compared using the two-sided unpaired t-test. j Frequencies of GZMB positive TINKs after stimulation for 4 h (n = 5). The data are presented as means ± SEM and compared using the two-sided unpaired t-test. k Kaplan–Meier survival analysis in tumor bearing mice (n = 5). P value was calculated using the log-rank test. f–k show data from biological replicates. Each data point represents one mouse, with no technical replicates used for statistical analysis. No data were excluded from the analyses. l Additional validation in B16F10 models is shown (n = 4). Tumor progression was monitored by bioluminescence imaging (BLI) using an in vivo imaging system (IVIS). Mice were injected intraperitoneally with D-luciferin (150 mg/kg) 10 min before imaging. m Bioluminescent signals were measured and quantified as total flux (photons/second) at indicated time points to assess tumor growth (n = 4). n Representative bioluminescence images of mice at day 14, 16, 19, and 21 post-implantations. Quantification of bioluminescent signals over time, and assessed by two-way ANOVA with Holm–Šidák multiple-comparison test (n = 4). o Kaplan–Meier survival analysis in tumor bearing mice (n = 4). The data are presented as mean ± SEM. P value was calculated using the log-rank test. Statistical significance indicated as labeled.

Additional validation in other solid tumor models (MC38 colon cancer and B16F10 melanoma) confirmed enhanced NK cell function upon Samsn1^−/− mice (Fig. 4 and Supplementary Fig. 4). In these models, Samsn1^−/− mice exhibited significantly reduced tumor size, lower liver-to-body weight ratio, and increased overall survival compared to WT mice (Fig. 4i–o and Supplementary Fig. 4a–c). NK cells from Samsn1^−/− mice showed elevated GZMB (Fig. 4j), IFN-γ (Supplementary Fig. 4d), and Perforin secretion (Supplementary Fig. 4e), with no significant changes in CD8⁺ T cells (Supplementary Fig. 4f–h). These findings establish that Samsn1 knockout enhances NK cell-mediated anti-tumor immunity in HCC.

NK cell-specific Samsn1 knockout alone is sufficient to suppress HCC tumor growth

While global Samsn1^−/− mice demonstrate anti-HCC efficacy, critical questions remain: Is NK cell-intrinsic SAMSN1 deletion sufficient, or do other immune cells contribute? Could developmental defects in germline knockout mice confound interpretation?

To definitively address these concerns, we generated NK cell-specific conditional knockout mice with Samsn1 floxed alleles (Samsn1^f/f) crossed with Ncr1-2A-Cre mice, restricting Cre recombinase activity exclusively to NK cells (Fig. 5a and Supplementary Fig. 5a, b). Critically, NK cell frequencies were identical between Samsn1^f/f-Ncr1^Cre+ and control littermates in blood (7.1% vs. 7.3%, P = 0.85) and spleen (5.4% vs. 5.5%, P = 0.79) (Fig. 5b–d), excluding developmental artifacts that might confound tumor studies.

Fig. 5. Samsn1 knockout promotes NK cell-mediated anti-tumor immunity.

a Schematic of generating NK cell-specific Samsn1 conditional knockout (Samsn1^f/f-Ncr1^Cre+) and control (Samsn1^f/f-Ncr1^Cre−) mice. b Representative staining of NK and T cells were shown in Samsn1 conditional knockout and control mice. c, d The frequencies of NK cells in blood and spleen of mice were shown (n = 3). The data are presented as means ± SEM and compared using the two-sided unpaired t-test. e Schematic illustration of the experimental design for Hepa1-6 orthotopic model in Samsn1^f/f-Ncr1^cre+ and Samsn1^f/f-Ncr1^Cre− mice. After 21 days, mice were euthanized by carbon dioxide inhalation, and livers were carefully removed, photographed, weighed, and representative images are shown in the panel. f Tumor sizes were shown at day 21 in liver (n = 4). g The mean liver weight relative body weight was calculated and compared using unpaired two-sided t-test (n = 4). The data are presented as means ± SEM. h Percentage of NK cells in liver tumor of Samsn1^f/f-Ncr1^Cre+ and Samsn1^f/f-Ncr1^Cre− mice were measured and compared using unpaired two-sided t-test (n = 4). The data are presented as means ± SEM. i, j Expression frequency of GZMB⁺ and IFN-γ⁺ NK cells from tumor tissue after stimulation with PMA (25 ng/ml), ionomycin (1 μg/ml), and brefeldin A (5 μg/ml) for 4 h, compared using unpaired two-sided t-test. The data are presented as mean ± SEM. k Kaplan–Meier survival analysis in tumor bearing mice (n = 4). P value was calculated using the log-rank test. e–k show data from biological replicates. Each data point represents one mouse, with no technical replicates used for statistical analysis. No data were excluded from the analyses. l Schematic illustration of the experimental design for the B16-F10-luc melanoma orthotopic model in Samsn1^f/f-Ncr1^Cre+ and Samsn1^f/f-Ncr1^Cre− mice. m Bioluminescent signals were measured and quantified as total flux (photons/second) at indicated time points to assess tumor growth (n = 3 for WT and n = 4 for conditional knockout mice). n Representative bioluminescence images of mice at days 10, 14, 17, and 21 post-implantations (n = 3 for WT and n = 4 for conditional knockout mice). Quantification of bioluminescent signals over time, and assessed by two-way ANOVA with Holm–Šidák multiple-comparison test (n = 4). o Kaplan–Meier survival analysis in tumor bearing mice (n = 4). P value was calculated using the log-rank test.

To further confirm the function of SAMSN1 in HCC model, we generated the orthotopic mouse model to validate the function of Samsn1 in vivo (Fig. 5e). Samsn1^f/f-Ncr1^Cre+ mice exhibited markedly reduced tumor size (Fig. 5f), with the liver-to-body weight ratio showing a 41% reduction in tumor burden compared to Samsn1^f/f-Ncr1^Cre− mice (Fig. 5g, P = 0.0017). Analysis of NK cells (Fig. 5h) and CD8⁺ T cells (Supplementary Fig. 5c) in TILs of liver from Hepa1-6 orthotopic model revealed no significant differences in immune cell populations. NK cells from Samsn1^f/f-Ncr1^Cre+ tumors also displayed increased secretion of GZMB (Fig. 5i) and IFN-γ (Fig. 5j). In accordance with prevailing predictions, no alterations were observed in the levels of GZMB (Supplementary Fig. 5d) and IFN-γ (Supplementary Fig. 5e) in CD8⁺ T cells. Samsn1^f/f-Ncr1^Cre+ mice was also associated with increased OS compared with Samsn1^f/f-Ncr1^Cre− mice, with median survival increasing from 27 days to 36 days (33% extension, P = 0.0101, Fig. 5k). To further validate the NK cell-dependent nature of this phenotype, we generated NK-specific Samsn1 knockout metastasis mice model (Fig. 5l). However, one mouse in WT group died within 24 h after tumor cell inoculation and was excluded from statistical analysis. There was no significant different in radiance intensity (Fig. 5m, n). However, Samsn1^f/f-Ncr1^Cre+ resulted in significantly longer OS (Fig. 5o). In summary, the phenotypic concordance between global Samsn1^−/− and Samsn1^f/f-Ncr1^Cre+ mice establishes that Samsn1 knockout in NK cells alone is both necessary and sufficient for the observed anti-tumor effects. The comparable tumor reduction (35% in Samsn1^−/− vs. 41% in Samsn1^f/f-Ncr1^Cre+) and survival benefit (median survival 36 days in Samsn1^−/− vs. 36 days in Samsn1^f/f-Ncr1^Cre+) demonstrate that other immune cell types contribute minimally to the therapeutic efficacy, definitively positioning NK cells as the primary functional target of SAMSN1 in HCC.

Enforced expression of SAMSN1 in NK92 cells induces cell exhaustion

To determine whether SAMSN1 is associated with poor prognosis in human, we detected the phonotypes of NK92^KO-SAMSN1 cell line and NK92^OE-SAMSN1 cell line (Fig. 6a). The SAMSN1 knockout efficiency was showed (Supplementary Fig. 6a–c). Given the high baseline and transduction efficiencies, the SAMSN1 expression level was showed by Mean fluorescence intensity (MFI) values in flow cytometry (Fig. 6b). The SAMSN1 expression level was also showed in WB (Supplementary Fig. 6d) and RT-PCR (Supplementary Fig. 6e). The proliferation of cells was found to be significantly inhibited in NK92^OE-SAMSN1 cells (Fig. 6c). In contrast, the knockout of SAMSN1 in NK92 cells was found to be a stimulant of cell proliferation (Fig. 6d). Meanwhile, SAMSN1 exhibited a marked propensity to elicit apoptosis (Supplementary Fig. 6f). NK92^KO-SAMSN1 cells promoted GZMB (Fig. 6e), and IFN-γ (Fig. 6f) secretion when compared with control cell line (NK92^Control), whereas SAMSN1 overexpression in NK92 cells hindered the secretion of GZMB (Fig. 6e). And the secretion of IFN-γ (Fig. 6f), Perforin (Fig. 6g) was no significantly different, when compared with NK92^Control with PMA/Ionomycin stimulation. In contrast, the inhibition marker of TIGIT was significantly increased in NK92^OE-SAMSN1 cells, when compared with NK92^Control with PMA/Ionomycin stimulation (Fig. 6h). Importantly, NK92^OE-SAMSN1 cells showed reduced lower oxygen consumption rate (OCR) (Supplementary Fig. 6g, h) and increased extracellular acidification rate (ECAR) (Supplementary Fig. 6i, j), indicating metabolic alteration.

Fig. 6. SAMSN1 overexpression and knockout effects in NK92 cells and xenograft models.

a Schematic of generating NK92 cell lines with lentivirus. b The mean fluorescence intensity (MFI) of SAMSN1 in NK92 cell lines was detected by flow cytometry (n = 3). Statistical significance was determined by one-way ANOVA. The data are presented as means ± SEM. c The Annexin V and 7-AAD were used for the combined detection of early-stage cell apoptosis (using Annexin V) and late-stage cell apoptosis or necrosis (using both Annexin V and 7-AAD) by flow cytometry (n = 3). Statistical significance was determined by one-way ANOVA. The data are presented as means ± SEM. d Cell proliferation rates were assessed over time using the Cell Counting Kit-8 (CCK-8) assay (n = 5). The statistical method used was two-way ANOVA. The data are presented as means ± SEM. e–h Expression frequency of GZMB⁺(e), IFN-γ⁺(f), Perforin (g), and TIGIT (h) in NK92 cell lines following PMA (25 ng/ml), ionomycin (1 μg/ml), and brefeldin A (5 μg/ml) for 4 h (n = 3). In all panels, groups were compared using one-way ANOVA. Data are presented as means ± SEM. i Schematic illustration of the experimental design for the humanized mouse model. NOD-SCID-IL2Rγ^null (NSG) mice were injected with NK92^OE-SAMSN1 or NK92^OE-control cells (5 × 10⁶ cells/mouse), and HepG2 cells (5 × 10⁶ cells/mouse) were injected subcutaneously. recombinant human IL2 was intraperitoneally injected every 2 days to support NK cell survival. j–m Tumor size was measured every two days starting on day 6 after tumor cell injection (n = 4). j Tumor size was compared using two-way ANOVA with Holm–Šidák multiple-comparison test. The data are presented as mean ± SEM. k–m Tumor growth curves of individual mice are shown for the three different treatment groups (n = 4). In this model, Complete Response (CR) is calculated as 100% reduction in tumor volume from baseline, measured via caliper (volume = length × width² × 0.5 mm³). Partial Response (PR) is typically ≥30% (but <100%) reduction in tumor volume from baseline. n Graphical abstract for the research.

To further characterize the function of SAMSN1 in NK cells in anti-tumor immunity, we generated a humanized mouse model (Fig. 6i) and investigated whether the exhaustion of NK92^OE-SAMSN1 correlated with the anti-tumor capacity of NK92 cells. We adoptively transferred NK92^OE-SAMSN1 cells or NK92^Control into immunodeficient mice bearing HepG2 xenografts. NK92^OE-SAMSN1 cells exhibited significantly reduced tumor growth suppression compared to NK92^Control cells (Fig. 6j–m). Specifically, tumor volumes in mice treated with NK92^OE-SAMSN1 cells were significantly larger than those in the control group by day 16 post-transfer, indicating a diminished anti-tumor effect. This finding is consistent with our in vitro cytotoxicity assays, where NK92^OE-SAMSN1 cells showed reduced specific lysis of tumor targets compared to NK92^KO-SAMSN1 and NK92^Control cells (Supplementary Fig. 6k). Taken together, the present study investigates the regulatory role of SAMSN1 expression levels in modulating NK cell activity and its implications in HCC (Fig. 6n). Collectively, these data implicate SAMSN1 as a potential biomarker of dysfunctional NK cell states and a contributor to poor clinical outcomes, possibly through its dual effects on immune effector function and metabolic fitness. The targeting of SAMSN1 or its downstream pathways may represent a novel therapeutic strategy to restore NK cell-mediated anti-tumor immunity in patients with SAMSN1-driven immunosuppression.

Discussion

NK cells play a pivotal role in immunological surveillance against malignancies, particularly in HCC, where they constitute up to 30% of tumor-infiltrating lymphocytes³⁴. However, their function is frequently impaired within the tumor microenvironment, limiting the efficacy of NK cell-based immunotherapies³⁵. In this study, we establish SAMSN1 as an immune checkpoint that predominantly suppresses NK cell function in HCC. NK cell-specific knockout alone is sufficient for therapeutic efficacy, as demonstrated by our conditional knockout studies where Samsn1^f/f-Ncr1^Cre+ mice recapitulated the tumor-suppressive phenotype of global Samsn1^−/− mice. Preliminary analyses indicate reduced GZMB⁺ NK and T cells in the IT region of HCC patients. However, SAMSN1⁺ T cells frequencies remain comparable between PT and IT regions. Moreover, T cells from Samsn1 knockout mice exhibit markedly weaker secretion of cytotoxic factors relative to NK cells. Notably, Samsn1 knockout significantly suppresses tumor progression across primary HCC models and extends to other solid tumor models, suggesting its broad therapeutic potential. This genetic evidence definitively positions NK cells as the primary cellular target of SAMSN1’s immunosuppressive function.

Our study addresses a gap in understanding NK cell dysfunction in HCC. Through convergent evidence from patient samples and preclinical models, we demonstrate that SAMSN1 upregulation in intratumoral NK cells correlates with reduced cytotoxic function and poor prognosis. Mechanistically, SAMSN1 suppresses NK cell cytotoxicity by downregulating granzyme B expression, as evidenced by a significant reduction in GZMB⁺ NK cells within the intratumoral region (from 42% to 37% in SAMSN1⁺ vs SAMSN1⁻ NK cells). This functional impairment is accompanied by upregulation of inhibitory receptors, including TIM3 (r = 0.469) and TIGIT (r = 0.457), suggesting SAMSN1 promotes an exhausted NK cell phenotype. Importantly, our NK92 cell studies demonstrate that SAMSN1 overexpression not only reduces cytotoxicity but also induces metabolic dysfunction, characterized by reduced oxygen consumption and increased glycolysis, indicating that SAMSN1’s suppressive effects extend beyond transcriptional regulation to metabolic reprogramming.

The quantitative concordance between NK cell-specific and global knockout phenotypes provides compelling evidence for NK cell-intrinsic mechanisms. In Hepa1-6 orthotopic models, NK-specific deletion achieved 41% tumor burden reduction compared to 34% in global knockout mice, with comparable survival benefits (median survival 36 days vs 33 days for Samsn1^−/−, both P < 0.05 vs WT). Importantly, CD8⁺ T cells showed no significant functional changes in either knockout model, excluding major contributions from adaptive immunity in our experimental context. This specificity contrasts with previous studies showing SAMSN1’s role in T cell regulation during sepsis, highlighting tissue- and disease-specific immunoregulatory functions.

Our findings build upon and extend previous characterizations of SAMSN1 in other disease contexts. In multiple myeloma, SAMSN1 functions as a tumor suppressor, with Samsn1^−/− mice showing increased susceptibility to malignant transformation³¹. This contrasts sharply with its tumor-promoting role in HCC, where SAMSN1 expression facilitates immune evasion. These context-dependent effects likely reflect differences in cellular expression patterns and microenvironmental cues. In HCC, SAMSN1 is predominantly expressed in infiltrating immune cells rather than tumor cells themselves, positioning it as an immune checkpoint rather than a tumor cell-intrinsic regulator. This distinction is critical for therapeutic development, as it suggests SAMSN1 inhibitors would primarily modulate immune function rather than directly target malignant cells.

The identification of SAMSN1 as a NK cell checkpoint has significant therapeutic implications. Current immune checkpoint inhibitors targeting PD-1/PD-L1 and CTLA-4 have shown limited efficacy in HCC, with overall response rates of only 20–30%^36,37. This modest benefit likely reflects the predominance of innate immune dysfunction in HCC, where NK cells rather than T cells dominate the immune infiltrate. Our findings suggest that SAMSN1 inhibition could complement existing checkpoint blockade by specifically enhancing NK cell-mediated immunity. The prognostic value of SAMSN1 expression (HR > 2.0 for both OS and DFS in our cohorts) further supports its potential as a biomarker for patient stratification. Future studies should investigate whether high SAMSN1 expression predicts resistance to current immunotherapies and whether SAMSN1 targeting can sensitize resistant tumors.

While our study provides definitive evidence for SAMSN1’s role in HCC, several limitations warrant discussion. First, although NK-specific deletion is sufficient for therapeutic efficacy in our models, we observed modest SAMSN1 expression in other immune populations. Whether SAMSN1 contributes to dysfunction in specific NK cell subsets (CD56^bright vs CD56^dim in humans; Ly49⁺ populations in mice) remains unexplored. Second, the molecular mechanisms linking SAMSN1 to granzyme B suppression require further elucidation. Our metabolic data suggest involvement of metabolic reprogramming, but dedicated mechanistic studies, including ChIP-seq and metabolomics profiling are needed to define the complete signaling pathway. Third, while our orthotopic models recapitulate key features of human HCC, patient-derived xenograft models would strengthen clinical relevance. Finally, exploratory studies in other solid tumor models suggest broader applicability, but disease-specific validation is required before generalizing these findings beyond HCC.

Methods

Study design

The primary objective of this study was to elucidate the role of SAMSN1 in HCC and the underlying mechanisms involved. To achieve this goal, we investigated the expression and distribution of SAMSN1 in HCC patient specimens and Samsn1 knockout mice, and explored the mechanism by which SAMSN1 contributes to tumor growth.

Human samples

Human samples were obtained under the ethical approvals detailed in the “Ethics statement” above. For the retrospective HCC cohort (Cohort 1), we collected a total of 90 samples from patients diagnosed with HCC who underwent liver resection surgery without neoadjuvant chemotherapy. These samples comprised: (1) 80 retrospective paraffin-embedded tissues obtained under approval KY2021204 (consent waiver); and (2) 10 additional retrospective samples obtained under approval 2023-A-(H)-048 (written informed consent). All 90 samples were used for IHC staining experiments (Fig. 1i–l). However, 2 of the 10 additional samples lacked complete survival follow-up data and were therefore excluded from the survival analysis, resulting in n = 88 for the prognostic analyses shown in Fig. 1i–l. All samples were anonymized, and data were handled in compliance with privacy regulations.

Cell lines

Mouse MC38 colorectal cancer cell line was purchased from Cell bank of Chinese Academy of Sciences. Human HepG2, Hepa1-6 murine hepatoma, and B16-F10 mouse melanoma cell lines were purchased from the American Type Culture Collection (ATCC). A B16-F10 luciferase cell line was established by transducing B16-F10 cells with a lentiviral vector encoding the luciferase gene. B16-F10 and B16-F10-luc cell lines were maintained in DMEM containing 10% FBS. NK92 cells were purchased from ATCC (Cat: CRL-2407) and maintained according to the protocol provided by ATCC. NK92 overexpression and knockout cell lines were established by transducing NK92 cells with lentiviruses. K562 cell line was also purchased from ATCC (Cat: CCL-243) and maintained in IMEM medium containing 10% FBS. All cells were cultured in a humidified, 5% CO₂ incubator at 37 °C, and grown in respective medium and 100 U/ml penicillin–streptomycin (Gibco, Cat: 15070063). All cell lines were tested and were negative for mycoplasma contamination.

Mice

The primary tumor models used for mechanistic validation were HCC-relevant orthotopic Hepa1-6 mice models. Additional exploratory validation was performed in MC38 and B16F10 models to assess generalizability. All mice used in this study were of the species Mus musculus. C57BL/6 (C57BL/6Smoc) (Cat. NO. SM-001), M-NSG (NOD-Prkdc^scidIl2rg^em1/Smoc) (Cat. NO. SM-001), and Ncr1-2A-Cre (C57BL/6Smoc-Ncr1^em2(2A-^iCre)Smoc) (Cat: NM-KI-190027) mice were purchased from Shanghai Model Organisms Center, Inc. Samsn1^f/f and Samsn1 knockout (Samsn1^−/−) mice were also generated on a C57BL/6 background in Shanghai Model Organisms Center, Inc. Mice used in the cancer model (Samsn1^+/+ and Samsn1^−/−) were crosses of Samsn1^+/− bred in-house carrying the following genotype: Samsn1^+/+, Samsn1^+/−, and Samsn1^−/−, and mice used in the cancer model (Samsn1^f/f Ncr1^cre− and Samsn1^f/f Ncr1^cre+) were crosses of Samsn1^f/f and Ncr1^cre+ bred in-house carrying the following genotype: Samsn1^f/f Ncr1^cre+, Samsn1^f/+ Ncr1^cre+, and Samsn1^f/f Ncr1^cre−, where the plus (+) and cre indicates presence of the mutant/transgenic allele and a minus (−) indicates allele absence. Samsn1^f/f Ncr1^cre− and Ncr1^cre+ mice had a similar tumor growth profile (not shown), and we preferentially used the Samsn1^f/f Ncr1^cre− mice as controls. All the mice were maintained in a specific pathogen-free (SPF) environment with controlled temperature (22 ± 2 °C) and humidity (50 ± 15%) under 12 h light/dark cycle at the University of Science and Technology of China (USTC) Animal Facility. Mice aged 6–8 weeks were used for experiments. Group sizes are specified per experiment in figure legends and Source Data. Breeding mice were housed separately by sex prior to grouping. After grouping, experimental and control animals were not co-housed; they were placed in separate cages within the facility and bred separately. The maximum tumor burden permitted by our institutional ethical board is 1500 mm³. We strictly adhered to these size limits, and no tumor exceeded this volume in our experiments. PCR primers used in this study can be found in Supplementary Table 3.

Validation of Ncr1-Cre conditional knockout system

Ncr1-2A-Cre mice were validated for NK cell-specific recombination through multiple approaches. Ncr1-2A-Cre mice were crossed with Rosa26-tdTomato reporter mice. Flow cytometric analysis confirmed tdTomato expression exclusively in NKp46⁺ NK cells in spleen (2.97%) and blood (2.51%)³⁸, with negligible recombination in CD3⁺ T cells ( < 0.5%), CD11b⁺ myeloid cells ( < 0.1%), or CD19⁺ B cells ( < 0.3%). Ncr1-2A-Cre+ mice (without floxed Samsn1 alleles) showed no differences in tumor growth, survival, or NK cell function compared to Cre-negative littermates in Hepa1-6 models (data available upon request), excluding Cre toxicity artifacts. In addition, SAMSN1 protein levels were reduced specifically in NK cells (78% reduction by flow cytometry) but unchanged in CD4⁺ or CD8⁺ T cells from Samsn1^f/f-Ncr1^Cre+ mice. These validations confirm that observed anti-tumor effects result from NK cell-intrinsic SAMSN1 deletion rather than off-target recombination or Cre expression artifacts.

In vivo mouse models and analysis

To evaluate tumor growth in Samsn1 knockout mice, Samsn1^−/− or Samsn1^f/f-Ncr1^cre+ mice (C57BL/6 background, 6- to 8-week-old females) and littermate control mice (C57BL/6; 8-week-old females) were anesthetized via intraperitoneal injection of ketamine (100/10 mg/kg), and were inoculated subcutaneously into the right flank at a final volume of 100 μL with MC38 cells (2 × 10⁵) or B16-F10 cells. Tumor dimensions were measured every 2–3 days using digital calipers, and tumor volume was calculated as 0.5 × length × width × width. Mice were monitored daily for health status. Criteria for early termination (humane endpoints) included tumor volume reaching 1500 mm³, tumor ulceration, weight loss >20%, or signs of severe distress. Mice were euthanized by carbon dioxide inhalation when these endpoints were reached.

For the metastatic B16-F10-luc model, B16F10-luc cells (1 × 10⁶) were injected into the left hepatic lobe of anesthetized control and Samsn1^−/− or Samsn1^f/f-Ncr1^cre+ mice (6- to 8-week-old females). Tumor growth was monitored by bioluminescence imaging using an IVIS system. For internal tumor models where tumor size could not be externally measured, the criteria for early endpoints included weight loss >20%, lethargy, hunched posture, abdominal distension, or when bioluminescence signal exceeded 1 × 10⁸ photons/s/cm²/sr. Euthanasia was performed by carbon dioxide inhalation immediately upon reaching these endpoints.

For orthotopic Hepa1-6 mouse models, 6- to 8-week-old female control and Samsn1^−/− or Samsn1^f/f-Ncr1^cre+ mice were anesthetized. Aseptic midline laparotomy ( ~ 1.5 cm) exposed the left hepatic lobe. Hepa1-6 cells (1 × 10⁶) suspended in 1:1 (v/v) PBS/matrigel (Corning, Cat:356231) were injected into the subcapsular parenchyma (50 μL total) using a 30-gauge needle. The abdomen was closed in layers (5-0 Vicryl sutures for muscle; staples for skin). For tumor burden assessment, mice (n = 8/group) were euthanized at 21 days by carbon dioxide inhalation; livers were weighed, imaged, and liver index calculated as (liver weight/body weight) × 100. A survival cohort (n = 8/group) was monitored daily. Humane endpoints for this non-measurable tumor model included weight loss >20%, lethargy, or palpable abdominal masses >1 cm.

For subcutaneous tumor model in M-NSG mice, 6- to 8-week-old female M-NSG mice received subcutaneous injections of HepG2 cells (5 × 10⁶) and IL-2-primed NK92 cells (2 × 10⁶) suspended in PBS/matrigel (1:1 v/v; total volume 100 μL) into the right flank. Control groups included HepG2-only injections. Tumor dimensions were measured three times weekly by blinded operators using digital calipers; volume was calculated as 0.5 × length × width × width. Humane endpoints (tumor volume >1500 mm³, weight loss >20%, or ulceration) were strictly enforced, and mice were euthanized by carbon dioxide inhalation.

Single-cell RNA sequencing and data analysis

scRNA-seq data from datasets from published paper³³ were used for analysis of SAMSN1 expression in lymphocytes of PT and IT samples. Seurat package (version: 4.0.3, https://satijalab.org/seurat/) was used for cell normalization and regression based on the expression table according to the UMI counts of each sample and percent of mitochondria rate to obtain the scaled data. PCA was constructed based on the scaled data with top 2000 high variable genes and top 10 principals were used for tSNE construction. Utilizing graphbased cluster method, the unsupervised cell cluster result based the PCA top 10 principal was acquired and the marker genes were calculated by FindAllMarkers function with Wilcox rank sum test algorithm under following criteria: 1. lnFC > 0.25; 2. p-value < 0.05; 3. min.pct > 0.1. In order to identify the cell type detailed, the clusters of same cell type were selected for re-tSNE analysis, graph-based clustering, and marker analysis. For pseudotime analysis, the Single-Cell Trajectories analysis utilizing Monocle2 (http://cole-trapnell-lab.github.io/monocle-release) using DDR-Tree and default parameter. Based on the pseudo-time analysis, branch expression analysis modeling (BEAM Analysis) was applied for branch fate determined gene analysis.

TCGA data analysis

We analyzed the Cancer Genome Atlas (TCGA) data from Liver Hepatocellular Carcinoma samples available in the cBioPortal database and used to analyze the gene expression correlations between SAMSN1 and other immune checkpoints as well as the overall survival of HCC patients based on their SAMSN1 gene expression. Correlations were analyzed by UCSC Xena (https://xena.ucsc.edu). The Kaplan–Meier survival of patients with HCC was analyzed using OncoLnc (http://www.oncolnc.org).

Bulk RNA sequencing

Total RNA was isolated using RNeasy Mini Kit (QIAGEN, Germany) from purified human NK cells derived from PBMCs, following SAMSN1 knockout via the CRISPR-Cas9 system. Total RNAs (2 μg) were used for stranded RNA sequencing library preparation by means of a Stranded mRNA Library Prep Kit from DR08502 (Bioyigene) according to the manufacturer’s instructions. The library products corresponding to 200–500 bp were enriched, quantified, and finally sequenced on DNBSEQ-T7. The gene expression profiles of NK cells from WT and SAMSN1 knockout were determined by RNA-Seq data analysis (Bioyigene). In brief, raw sequencing data were first filtered by FastQC; low-quality reads were discarded, and adaptor sequences were trimmed. RNA-seq reads were aligned to the human GRCh38 reference genome (GENCODE v43 annotation) using HISAT2 v2.2.1. Only samples with >90% uniquely mapped reads were retained. Significantly differentially expressed transcripts were screened by applying the criteria FC ≥ 2 or ≤ −2 and p-value < 0.05.

Gene set and pathway enrichment analyses

GSEA was performed using gsea v4.2.3, by comparing single cells annotated as sgSAMSN1 vs. sgControl; all genes with log2FC threshold ≥0.1 were included. Pathway enrichment analysis was performed using the top 20 most upregulated genes by log2FC for both single cells annotated as sgSAMSN1 vs. sgControl for single cells within the leukemia cluster, with CytoTRACE >0.95 vs <0.95. These gene sets were compared against the ChEA and ENCODE transcription factor targets databases via the enrichrplatform.

Isolation of human PBMCs and NK cells

Peripheral blood samples from healthy controls were collected from The Affiliated Hospital of University of Science and Technology of China (Cohort 2, n = 4). And PBMCs were isolated by FicollPaque (GE Healthcare, Cat: 17-1440-02) density gradient centrifugation according to the manufacturer’s instructions. NK cells were purified by negative selection using human NK Cell Isolation Kit (Miltenyi Biotec, Cat: 130-092-657) following the manufacturer’s protocol. The purity of NK cells was > 90% as determined by flow cytometry. Human PBMCs and purified NK cells were cultured in RPMI-1640 medium with 10% FBS and 100 IU/ml recombinant human IL2 (Proteintech, Cat: Hz-1015).

In vitro knockout of SAMSN1 with CRISPR/Cas9

NK cells magnetically enriched (Mitenyi) from PBMCs were cultured in RPMI-1640 supplemented with recombinant human IL2 (Proteintech, Cat: Hz-1015) for expansion. To disrupt the target gene, we employed the type II CRISPR-Cas9 system from S. pyogenes. Specifically, cells (1 × 10⁶) were resuspended in 20 μl P4 solution with NLS-Cas9-EGFP Nuclease (Novoprotein, Cat: E379-02A) and guide RNA (gRNA) (0.1 nmol, GenScript) to make RNP complex, followed by nucleofection using Lonza 4D Nucleofector with program CM137 (Lonza). The single-guide RNA (sgRNA) designed using the CRISPR Design Tool (Zhang Lab, MIT) to target exon 6 of SAMSN1. This is a widely used, well-characterized version of CRISPR-Cas9 with established protocols for in vitro applications. The cells were cultured post-electroporation for 72 h. Flow cytometry was used to gate GFP⁺ cells and verify the knockout efficiency by CHANGE-seq, and mRNA was purified to perform bulk RNA-seq. Sanger sequence was used for indel efficiency analysis. gRNA sequences used in this study can be found in Supplementary Table 3.

Generation of SAMSN1 overexpression and knockout cell lines

The following lentiviral plasmids were used for this experiment: pLVX-EGFP-IRES-puro was a gift from Qingming Fang (Addgene, Cat: 128652) and modified by replacing the puro sequence with EGFP and inserting SAMSN1 in the original EGFP position to generated the pLVX-SAMSN1-IRES- EGFP and pLVX-SAMSN1-IRES- puro plasmids. lentiCRISPR v2 (Addgene, Cat: 52961) and lentiCRISPR v2GFP (Addgene, Cat: 82416) were inserted with sgRNA sequences to knockout SAMSN1 in NK92 cells. HEK293T cells were used to produce lentiviral particles using psPAX2 (Addgene, Cat: 12260) and pMD2.G (Addgene, Cat: 12259) plasmids. Once the presence of lentivirus was confirmed, the supernatant was divided and stored in −80 °C until use. PCR primers used in this study can be found in Supplementary Table 3.

Transduction of NK92 cells with concentrated lentiviral particles

NK92 cells (3 × 10⁵) were resuspended with 2 ml lentivirus (6 × 10⁶ PFU/ml) and supplemented with 8 μg/ml polybrene. The mixture was subjected to centrifugation at 30 °C for 90 min in 6-well plate, after which it was transferred to a 37 °C incubator for 12 h. Thereafter, centrifugation was performed once more with the objective of replacing the liquid. The cells were validated after 72 h.

CHANGE-seq

The CHANGE-seq was performed on genomic DNA isolated from human NK cells using the Gentra Puregene Kit (Qiagen, Cat: 158767). The DNA was tagmented with a custom Tn5-transposome to an average length of 400 bp, followed by gap repair using Kapa HiFi HotStart Uracil+ DNA Polymerase (Roche, Cat: KK2601) and Taq DNA ligase (NEB, Cat: M0208V). The tagmented DNA was then treated with USER enzyme (NEB, Cat: M5505S) and T4 polynucleotide kinase (NEB, Cat: M5505S), circularized intramolecularly with T4 DNA ligase (NEB, Cat: M0202V), and residual linear DNA was degraded using a cocktail of exonucleases (Plasmid-Safe ATP-dependent DNase [Lucigen], Lambda exonuclease [NEB], and Exonuclease I [NEB]). In vitro cleavage was conducted with 125 ng circularized DNA, 90 nM SpCas9 protein, NEB buffer 3.1, and 270 nM gRNA in a 50 μL reaction. Cleaved products were A-tailed, ligated with a hairpin adaptor, treated with USER enzyme, and PCR-amplified using barcoded primers (NEBNext Multiplex Oligos for Illumina, Cat: E6444S) and Kapa HiFi Polymerase. Libraries were quantified by qPCR and sequenced with 151 bp paired-end reads on an Illumina NextSeq. Data analysis used open-source CHANGE-seq software. For CHANGE-seq-BE analysis, adapters were trimmed with cutadapt (v1.18) using parameters. to remove Tn5 adapters. Paired-end reads were mapped via bwa mem, filtered for outward orientation and self-overlap (default min_overlap = 0 and max_overlap = 15), and start positions were tabulated to identify enriched intervals in nuclease-treated samples. Flanking sequences ( ± 30 bp) were searched for off-target sites with edit distance ≤6 (allowing gaps). Percentage was calculated as the ratio of deamination-containing reads to total valid reads per off-target.

Cytotoxicity assay

The tumor cell lines were labeled with CFSE (ThermoFisher, Cat: C34554) according to the manufacturer’s instructions. Labeled tumor cells were co-cultured with effector cells in 96-well plates for 4 h at different effector/target ratios. For the spontaneous death control, CFSE-labeled target cells were cultured alone, followed by the addition of 7AAD (Biolegend, Cat: 420403), and lysed cells (CFSE⁺ 7AAD⁺) were identified using flow cytometry. Details information of antibody used in this study can be found in Supplementary Table 4.

Immunohistochemistry

Formalin-fixed paraffin-embedded HCC tissue sections (4 μm) were deparaffinized and rehydrated. Antigen retrieval was performed using citrate buffer (pH 6.0) at 95 °C for 20 min. Endogenous peroxidase was blocked with 3% H2O2 for 10 min. Sections were incubated with anti-SAMSN1 antibody (1:200, Atlas Antibodies, Cat: HPA059729) overnight at 4 °C. After washing, the sections were incubated with an HRP-conjugated secondary antibody (1:500, Abcam, Cat: ab205718) for 1 h at room temperature. 3,3′-Diaminobenzidine (DAB) was used for color development. Lymphocytes were identified based on their characteristic morphological features, including small, round cell morphology, hyperchromatic nuclei, scant cytoplasm, and a high nuclear-to-cytoplasmic ratio, with cell diameters typically less than 10 μm. In the IHC images, blue staining (hematoxylin) denotes nuclei, while brown staining (DAB) indicates SAMSN1 expression. The IHC score was calculated as the integrated optical density (IOD) divided by the area (IOD/area) to quantify SAMSN1 expression. SAMSN1 expression was quantified using ImageJ software (NIH, USA) in five random high-power fields (400×) per region. The staining intensity was scored as 0 (negative), 1 (weak), 2 (moderate), or 3 (strong). The percentage of positive cells was scored as 0 (0%), 1 (1–25%), 2 (26–50%), 3 (51–75%), or 4 (76–100%). The average IOD/area was calculated for each sample. Wilcoxon rank-sum test was used in the statistical analysis. The final score was calculated by multiplying intensity and percentage scores.

Metabolism assays

Extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) were measured with the Seahorse XF96 Flux Analyser (Agilent) and Agilent Wave 2.6.1 software according to the manufacturer’s instructions. 1 h before testing, the media was replaced with assay media, and cells were seeded on plates. For ECAR test, 1 mmol/L oligomycin (O), 0.5 mmol/L carbonylcyanide p-trifluoromethoxyphenylhydrazone (FCCP), and 0.1 mmol/L antimycin A plus 0.5 mmol/L rotenone (A & R) were injected to the wells. For OCR test, 10 mmol/L glucose, 1 mol/L oligomycin (O), and 50 mmol/L 2-deoxyglucose (2-DG) were added to the wells.

Preparation of cell suspensions

Single-cell suspensions were prepared from mouse spleens or tumors as previously described³⁹. In brief, tumors were dissociated mechanically and digested with 1 mg/ml collagenase A and 0.1 mg/ml DNase I for 20 min at 37 °C. Spleens were mechanically dissociated, digested with 0.1 mg/ml collagenase A and 0.01 mg/ml DNase I for 20 min at 37 °C, and passed through a 40 μm cell strainer and lysed of red blood cells (using ACK buffer), then washed with cold PBS and centrifuged.

Flow cytometry

All antibodies used in this study can be found in Supplementary Table 4. For flow cytometry, tumor samples were collected from The First Affiliated Hospital of Anhui Medical University (Hefei, China). Tumor tissues were finely minced and digested at 37 °C in digestion buffer: RPMI-1640 medium with 1 mg/ml collagenase IV (Worthington Biochemical, Cat: 9001-12-1) and 0.1 mg/ml DNase I (Sigma-Aldrich, Cat: AMPD1-1KT). All organs were processed into a single-cell suspension over a 70-μm filter and washed with PBS and 5 mM EDTA. cells were left untreated or stimulated with 30 ng/ml phorbol 12-myristate 13-acetate (PMA) (Sigma-Aldrich, Cat: P1585) and 1 μg/ml ionomycin, followed by 2.5 μg/ml monensin (Sigma-Aldrich, Cat: 22373-78-0). After washing with FACS buffer (PBS + 0.5% FBS + 2 mM EDTA), the cells were stained with a Zombie Aqua Fixable Viability Kit (BioLegend, Cat: 423101) or a Horizon Fixable Viability Stain 700 (BD Biosciences, Cat: 564997). The reaction was terminated by washing with FACS buffer. Subsequently, the cells were stained for 30 min at 4 °C in the dark with fluorophore-conjugated antibodies specific for panels of cell-surface markers. Intracellular staining was performed using the FoxP3/Transcription Factor Fix/Perm Kit (eBioscience, Cat: 00-5523-00). Antibodies used in this study were diluted at 1:200 unless otherwise specified. The anti-SAMSN1 antibody used for FACS was sourced from Boster Biological Technology (Cat: A08977-2, 1:100 dilution) and labeled with PE Donkey anti-rabbit IgG (Biolegend, Cat: 406421, 1:100 dilution). The following cell populations were identified on the basis of cell marker expression: CD4⁺ T cells (CD45⁺ CD3⁺CD4⁺), CD8⁺ T cells (CD45⁺CD3⁺CD8⁺), B cells (CD45⁺CD3⁻CD19⁺), natural killer (NK) cells (CD45⁺NK1.1⁺) for mice and (CD45⁺CD56⁺) for human samples, NKT cells (CD45⁺CD3⁺NK1.1⁺) or (CD45⁺CD3⁺ CD56⁺), monocytes (CD45⁺CD3⁺CD11b⁺). All samples were acquired on an LSRFortessa (BD, Franklin Lakes, USA) and were analyzed using FlowJo software (BD, Franklin Lakes, USA).

Cytometry by time of flight (CyTOF) analysis

For mass cytometry, Single-cell suspensions were prepared from tumors as described above. All antibodies used in this study can also be found in Supplementary Table 4. Antibodies were purchased unlabeled and conjugated to heavy metals in-house. Antibody conjugation to the heavy metals tags was performed using the MaxPar Antibody Conjugation Kit (Fluidigm), according to the manufacturer’s protocol. After labeling, antibodies were diluted to 0.2–0.5 mg/ml in antibody stabilization solution (Candor Bioscience) and stored at 4 °C until use. Before experimental use, conjugated antibodies were titrated against tissue to determine the optimal staining concentration. Before sample staining, all antibodies were pooled into either the surface or intracellular master mix and stained.

Cells were resuspended at a 1:1 ratio in PBS + 5 mM EDTA and 100 μM cisplatin (Sigma-Aldrich; diluted in PBS + 5 mM EDTA) for 60 s before quenching at a 1:1 ratio in FACS buffer to determine viability. Cells were then treated with an Fc receptor blocking reagent to minimize nonspecific binding, followed by incubation with a mixture of surface antibodies for 30 min on ice. Cells were fixed overnight at 4 °C in Fix/Perm Buffer (eBioscience). Samples were washed with permeabilization buffer (eBioscience) prior to intracellular staining. Intracellular antibody cocktails were applied for 30 min on ice, after which cells were washed, resuspended in ultrapure water, and mixed with EQ calibration beads (20%, Fluidigm) for instrument standardization. Data acquisition was performed on a Helios mass cytometer (Fluidigm). Among the 22 samples processed, 8 samples were excluded from downstream analysis due to failure to meet quality control criteria. Raw data were preprocessed by debarcoding using a mass-tag-specific method that incorporated doublet filtering. Bead-mediated normalization was applied to standardize signals between acquisition runs. Viable singlet immune cells were selected through manual gating in FlowJo software, with exclusion of debris, non-viable cells, and multiplets. Cellular subtypes were delineated via the PhenoGraph clustering method, which partitioned cells on the basis of their marker profiles. Subtype annotations were determined from standard marker signatures. Low-dimensional projections of the clustered dataset were produced with the t-SNE technique to aid in assessing subtype abundance, marker levels, and inter-group variations. Variations in subtype proportions across conditions were analyzed with unpaired t-tests.

Western blotting

NK92 cell lines were harvested and washed with cold PBS, then lysed with RIPA lysis buffer (Solarbio) on ice with addition of protease inhibitor cocktail (Sigma) and subsequently heated at 98 °C for 10 min. After total protein normalized, protein samples were separated by 8–10% SDS-PAGE gel electrophoresis and transferred onto 0.22 mm NC membranes. After blocking with 5% BSA diluted in Tris buffer saline plus 0.1% Tween 20 (TBST) for 1 h at room temperature, membranes were incubated overnight with SAMSN1 antibody (1:1000, Boster Biological Technology, Cat: A08977-2), and 2 h with Anti-rabbit IgG, HRP-linked antibody (1:5000, CST, Cat: 7074S). Blots were visualized by the Bio-Rad system.

RT-qPCR assay

To determine the expression levels of SAMSN1, NKp30, CCR3, and CXCR3 in NK92 cells, total RNA was isolated using a MolPure Cell/Tissue Total RNA Kit (YEASEN, Cat: 19221ES50) and converted to cDNAs using the Superscript III First Strand Synthesis System (YEASEN, China). RT-qPCR was performed using a Bio-Rad IQ5 System. PCR reactions were performed in 20 μL reactions using the SYBR Green PCR master mix (YEASEN, Cat: 11201ES) and 0.2 μM specific primers. The relative expression levels of mRNAs were calculated using the comparative CT method normalized to GAPDH. Primers used for RT-qPCR are shown in Supplementary Table 3.

Statistical analyses

Data are presented as mean ± standard deviation. All statistics were carried out using GraphPad Prism 10.0. Student’s t-test was used to compare the differences between two groups. One-way analysis of variance (ANOVA) and Tukey’s multiple comparisons test were used for the three or more groups comparisons. At least three independent biological replicates have been performed for each experiment. The number of independent experiments is indicated. P < 0.05 was considered to be statistically significant. All statistical was labeled as shown in figure unless otherwise indicated.

Ethics statement

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Anhui Provincial Hospital, China (approval numbers KY2021204 and KY2022222) and the Biomedical Ethics Committee of the University of Science and Technology of China (approval numbers 2019-N-(H)-121 and 2023-A-(H)-048).

Written informed consent was obtained for: (1) 4 healthy donor blood samples (Cohort 2, n = 4) for bulk RNA-seq under approval 2019-N-(H)-121, using the approved informed consent form (version dated March 15, 2019); and (2) 10 retrospective HCC samples (Cohort 1, n = 10) for expanded IHC validation under approval 2023-A-(H)-048, using the approved informed consent form (version T2.2023-09, dated September 17, 2023). Informed consent was waived by the ethics committees for: (1) 80 retrospective paraffin-embedded HCC tissues (Cohort 1, n = 80) for IHC staining under approval KY2021204; and (2) 22 fresh HCC samples (Cohort 2, n = 22) for CyTOF under approval KY2022222, as all samples were de-identified and used for observational research.

All mouse experiments were conducted in accordance with animal protocols approved by the Laboratory Animal Centre of the University of Science and Technology of China (USTCACUC192201040, USTCACUC25030123013), and every effort was made to minimize suffering.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C.S. contributed to the conception and design of the study. Study materials, data collection, and assembly were provided by R.F.W. and H.D.C.. R.F.W. and H.D.C. provided important information for the initial experimental exploration. H.Y.L. provided assistance in bioinformatics data processing. Q.L., G.J.Y., F.L.L, P.S., and T.L.D. assisted in obtaining mouse samples. J.B.W. assisted in obtaining human samples. Data analysis and interpretation were conducted by R.F.W. and H.D.C.. Manuscript writing was performed by R.F.W. and C.S.. All authors reviewed and provided final approval of the manuscript.

Peer review

Peer review information

Nature Communications thanks Domenico Mavilio and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The raw bulk RNA-seq data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA016031) that are publicly accessible (https://ngdc.cncb.ac.cn/gsa-human). The scRNA-seq data analyzed in this study were downloaded from the China National GeneBank DataBase (CNGBdb) under accession codes CNP0000650³³. The TCGA data for Liver Hepatocellular Carcinoma samples were downloaded from the cBioPortal database (https://www.cbioportal.org/study/summary?id=lihc_tcga). All data reported in this manuscript are available in the main text and supplementary information. Source data are provided with this paper.

Code availability

The scripts for all bioinformatic analyses of scRNA-seq and bulk RNA-seq are freely available at https://github.com/RuifengWan/SAMSN1-NC-2026.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-68661-4.