Multi-Omics profiling identify NNMT in tumor endothelium as a key regulator of CD8⁺ T cell exhaustion via the TGF signaling pathway

Highlights

• •Multi-omics profiling identifies NNMT as a high-confidence therapeutic target overexpressed in tumor-associated endothelial cells in ccRCC.
• •NNMT+ endothelial cells form an immunosuppressive niche in direct contact with exhausted CD8+ T cells via active TGF-β signaling.
• •NNMT knockdown restores CD8+ T cell effector function, reduces pro-inflammatory cytokines, and shifts the BAX/Bcl-2 ratio toward apoptosis.
• •Multi-omics screening identifies I-BET-762, I-BET-151, PFI-1, and BMS-387032 as potent inhibitors of NNMT to reverse T cell exhaustion.
• •NNMT serves as a central metabolic-immune hub orchestrating TGF-β-mediated immune dysfunction and endothelial reprogramming in ccRCC.

Abstract

Background

CD8⁺ T cell exhaustion is a defining feature of the immunosuppressive TME in ccRCC.

Methods

We employed a multi-omics driven pipeline to nominate Nicotinamide N-methyltransferase (NNMT) as a high-confidence therapeutic target in ccRCC. This computational prediction was validated through bulk RNA-seq, single-cell RNA sequencing, and spatial transcriptomics to delineate NNMT-associated molecular and cellular programs. While the discovery phase highlighted endothelial-specific NNMT overexpression, we further validated the functional consequences of NNMT modulation using Caki-1 and A498 cell lines to model the downstream signaling cascades. Functional assays assessed impacts on proliferation, apoptosis, cytokine secretion (IL-6, IL-1β, TNF-α), and TGF-β pathway activity. Immune infiltration and T cell exhaustion signatures were evaluated across TCGA cohorts.

Results

Multi-omics profiling revealed that NNMT is specifically overexpressed in tumor-associated endothelial cells enriched for active TGF-β signaling and inflammatory cues. High NNMT expression strongly correlated with CD8⁺ T cell exhaustion, elevated apoptotic signaling, and immunosuppressive cytokine production. In functional validation, NNMT knockdown suppressed TGF-β activity, reduced pro-inflammatory cytokines, and restored CD8⁺ T cell infiltration and effector function. Mechanistically, NNMT loss shifted the BAX/Bcl-2 ratio toward apoptosis and increased cleaved caspase-3. Spatial transcriptomics confirmed that NNMT⁺ endothelial cells form an immunosuppressive niche in direct contact with exhausted T cells. We also found that I-BET-762, I-BET-151, PFI-1, and BMS-387032 can target and inhibit NNMT to reduce CD8⁺ T cell exhaustion.

Conclusion

We establish NNMT as a central metabolic-immune hub that orchestrates TGF-β-mediated CD8⁺ T cell dysfunction and endothelial reprogramming in ccRCC.

Graphical abstract

Introduction

The intricate interplay that exists among tumor aggressiveness, metabolic reprogramming, and an overall immunosuppressive landscape of tumor microenvironment (TME) places a significant burden on patient long-term survival outcomes for patients with clear cell renal cell carcinoma (ccRCC) [1,31]. Such an important implicated role is also attributed to the functionally exhausted phenotype of T cells, particularly CD8⁺ cytotoxic T cells in the TME, as the dominant immune evasion and resistance mechanisms to anticancer therapy [2]. This exhaustion exceeds functional suppressor, and it involves an in-depth, and multi-layered condition of adaptive failure. Its features are the long-term downregulation of hallmark effector molecules TNF-α and IFN-γ, durable overexpression of numerous co-inhibitory receptors, dysfunction of the mitochondrial machinery, and reprogramming of mitogen-associated epigenetic changes [3]. The acknowledgement and reversal of T cell exhaustion are critical to bridge the treatment obstacles in kidney cancer.

The ccRCC TME is not only a simple immune cell habitat, but also it manifests as an immunosuppressive and metabolically disruptive dynamic ecosystem. Some of the characteristics involve severe regions of hypoxia as a result of aberrant angiogenesis, and substantial infiltration of M2 macrophages, Tregs, and MDSCs [4]. The dominant function of metabolic reprogramming in reconstructing the immunosuppressive TME and triggering T cell exhaustion has become apparent recently [5]. Such unfavorable conditions are characterized by competitive scarce of nutrient glucose, tryptophan, and arginine, and the accumulation of toxic byproducts of tumor metabolism such as lactate, kynurenine, and ROS [[6], [7], [8]]. The effector T cells are also adversely affected by metabolic stress, and they rely on efficient metabolic pathways of oxidative phosphorylation (OXPHOS) in assisting the execution of functional roles. Additionally, the dysregulation of nicotinamide adenine dinucleotide (NAD⁺) metabolic pathway in cellular metabolism and signal transduction, especially the dysfunction of sirtuin family deacetylases [[9], [10], [11]], is becoming a crucial driver linking tumor metabolism, TME immunosuppression, and T cell dysfunction.

Nicotinamide N-methyltransferase (NNMT), a central regulatory enzyme in the NAD⁺ metabolic pathway, has garnered extensive focus due to its aberrant overexpression in a broad spectrum of solid tumors, including liver, breast and colon tumour and lung cancers, where it is frequently associated with tumor progression and poor prognosis [[12], [13], [14]]. In these malignancies, NNMT exerts its oncogenic functions primarily through modulating cellular methylation potential and metabolic reprogramming. Specifically, NNMT catalyzes the methylation of nicotinamide (NAM)—the core precursor in the NAD⁺ salvage pathway—utilizing S-adenosylmethionine (SAM) as the methyl donor to produce 1-methylnicotinamide (1-MNAM) [15]. This enzymatic activity constitutes a sophisticated "dual-pronged assault" within the TME. On one hand, NNMT hyperactivity depletes NAM, directly constraining the capacity of tumor cells to regenerate NAD⁺ via the salvage pathway. On the other hand, the accumulation of 1-MNA acts as an immunomodulatory molecule, inhibiting the deacetylase SIRT1. Reduced SIRT1 activity enhances the stability of hypoxia-inducible factor 1-alpha (HIF-1α), thereby amplifying the Warburg effect, promoting the release of pro-angiogenic factors, and stimulating the production of immunosuppressive cytokines such as TGF-β [16,17]. we also found that the TGF signaling pathways can regulate the progression of renal cancer. Despite these established roles in tumor cells and the general TME, the specific contribution of NNMT in the context of ccRCC remains poorly defined. Furthermore, while the impact of NNMT on immune cell function (such as suppression of T-cell stimulation and promotion of Treg activity) has been noted, its potential role in regulating endothelial cell biology within the tumor vasculature has not been elucidated. This gap highlights the unique significance of investigating NNMT-mediated metabolic crosstalk in ccRCC progression.

To reduce the profound heterogeneity and multi-dimensional regulatory complexity of the TME in ccRCC, we found an integrated computational-experimental pipeline combining single-cell and spatial multi-omics profiling with artificial intelligence (AI)–driven target discovery. We leveraged deep learning models to interrogate large-scale drug–target–phenotype databases, enabling unbiased identification of actionable regulatory nodes. This AI-powered approach not only pinpointed nicotinamide N-methyltransferase (NNMT) as a central driver of T cell exhaustion but also revealed that diverse immunotherapeutic agents can suppress NNMT expression—thereby restoring immune function. Our work thus establishes NNMT as a therapeutically tractable target and demonstrates how AI-driven discovery can effectively bridge multi-omics insights with clinical translation, offering a rational foundation for combining NNMT inhibition with immunotherapy in ccRCC.

Thus, we provide an exceptionally promising strategy to NNMT to remodel immunometabolism in the TME of RCC. Important new insights into the metabolic basis of immune evasion in kidney cancer were made for further understanding the specific mechanisms by which regulate T cells and their downstream effector pathways via NNMT. This research will not only assist to identify novel biomarkers to predict response to immunotherapy, but also create a solid theoretical basis for developing innovative NNMT-targeted therapeutic modalities.

Methods

Data collection

The transcriptome profiles of ccRCC were retrieved from TCGA database, as well as GEO database.The bulk transcriptome high throughput sequence was converted to Transcripts per Kilobase Million (TPM).subsequently, duplicates removal and normalization for all the expression profile were performed. The scRNA-seq data was obtained from GSE159115 database, then cluster analysis was conducted,dimensionality reduction; and cell subtype annotation.We used spatial transcriptome tiles from GSM5924036, GSM5924040, GSM5924051 and GSM5924052, which we normalized together so that they can be analyzed in a joint manner.

Single-cell multi-omics progression analysis

The Seurat package was used to construct Seurat objects [[18], [19], [20]]. Then, we assessed the GSE159115 data set using the GEO database (https://www.ncbi.nlm.nih.gov/) and according to the single cell analysis,six samples were prepared using the 10 × Genomics scRNA-seq approach and subjected to further analysis. All samples were integrated in one Seurat object by running Seurat 4.0 R package .QC steps requiring that the number of genes per cell is between 200-4000 and the percent of mitochondrial genes are below 10%. Finally,the data log was normalized and then moved to the ‘LogNormalize’ in which the PCA was performed on the top 1000 most variant genes. The UMAP were used to create a cellular map of the data for visualizing the transcriptome heterogeneity.

For spatial transcriptome data, we applied the following standardized workflow using the Seurat package (v4.1.0): Log-normalization: Raw counts were normalized with a scale factor of 10,000 (scale.factor = 10000) to correct for sequencing depth. n.cells.use = 2000 to identify 3,000 highly variable genes.

Multi-omics cell–cell communication

Single-cell transcriptome: Cell–cell communication was analyzed using the CellChat R package with single-cell and predefined cell annotation data. It used its internal human dataset CellChatDB.human to predict cellular interaction among 32 signaling axes . In addition,we used the Nichenet R package to combine PPI and other interaction information on regulator factors.This integrated analysis allowed us to identify EC-expressed ligands and their target cells with downstream signaling pathways that we validated by CellChat .

Single-cell spatial transcriptome: COMMOT analysis is based on commot, scanpy, pandas, and numpy library implements the collective optimal transport algorithm,is optimized to run on large spatial transcriptomic datasets, written entirely in Python. The runtime as well as the RAM footprint scales linearly in the number of spatial loci.This enables local signal hotspot detection as well as association of predicted intercellular communication with known downstream gene expression profiles.

Non-negative matrix factorization (NMF)

NMF serves as a tool for inferring gene expression patterns from single-cell data. It treats a count matrix (N cells × G genes) as input and generates a gene expression program matrix of size (K × G) and a cell-program assignment matrix of size (N × K), suggesting the program application for each cell. Here, NMF was conducted for clustering tumor cells in RC and identifying the most tumor-associated cell modules.

Construction and validation of models

To construct a robust prognostic signature, we developed a Consensus Inverse Regularized Least Squares (CIRLS) model by integrating the results of 10 distinct machine learning algorithms and 101 algorithmic combinations. The workflow involved the following steps: First, Differentially Expressed Genes (DEGs) were identified and subjected to univariate Cox analysis using a bootstrap kernel to extract candidate prognostic signatures. Second, these DEGs were used to build classifiers via a Leave-One-Out Cross-Validation (LOOCV) scheme within the TCGA-KIRC training cohort; this step systematically evaluated the 101 algorithm combinations, which included Decision Trees (DT), k-Nearest Neighbors (kNN), Stepwise Cox, Lasso, Supervised Principal Components (Supervised PC), Cox Partial Least Squares regression (plsRcox), CoxBoost, Gradient Boosting Machines (GBM), and Support Vector Machines (SVM). Finally, the generalizability and stability of the resulting consensus model were rigorously validated using independent datasets, including the TCGA-KIRC test set, ICGC, and GSE22541 [32].

Cell culture and sh-NNMT transfection

The human Caki-1 and A-498 cell lines were acquired from the Institute of Cell Biology, Chinese Academy of Sciences (Shanghai). Cell culturing was performed in DMEM supplemented with 10% fetal bovine serum (FBS, Mema, # YSN1121). All cells were cultivated at 37°C in a 5% CO₂ environment.

Gene silencing was employed via the HiPerFect Transfection Reagent (Qiagen, Germany). Transfecting Caki-1 and A-498 cell lines was processed with 5 nM NNMT-specific shRNA in accordance with to the established protocol, with non-targeting shRNA and sham transfections as negative controls. After a 24-hour incubation period post-transfection, the efficiency of knockdown was assessed by reverse transcription qRT-PCR analysis.

Subcutaneous tumor formation

A-498 cells were divided into groups as:①sh-NC, sh-NNMT. Cells at the logarithmic growth phase (80-90% confluency) were cultured and subjected to trypsinization to produce a single-cell suspension. It was centrifuged to get rid of the supernatant, and resuspended in PBS or serum-free medium for cell counts. Cell concentration was changed into 1-5 × 10⁷ cells/ml to ensure that each mouse situated at a subcutaneous injection of 0.1 ml (containing 1-5 × 10⁶ cells). C57BL/6 mice, aging from 6 to 8 weeks, were utilized for the study. Mice were manually restrained, and 0.1-0.2 ml of the cell suspension was subcutaneously injected into the axillary region. Tumor growth was monitored weekly using caliper measurements. Finally, the mice were put into euthanasia to harvest tumors for subsequent analysis.

Immunoflow cytometry

To quantitatively analyze the infiltration of CD8+ T cells in tumor tissues, we conducted multicolor flow cytometry on mouse tumor samples from tumor-bearing mice. First, fresh tumor tissues were mechanically minced and digested with collagenase to prepare single-cell suspensions. After red blood cell lysis and washing, the cells were mixed with fluorescently labeled antibodies and incubated at 4°C in the dark for 30 minutes. The antibodies used included anti-mouse CD3-PerCP-Cy5.5 and anti-mouse CD8-APC. After staining, the cells were washed twice with PBS and resuspended in PBS containing 0.5% paraformaldehyde for fixation. Data were collected using the BD LSRFortessa flow cytometer and analyzed using FlowJo v10 software. During the analysis, cell debris was first excluded by forward scatter light (FSC-A) and side scatter light (SSC-A), and then single-cell populations were distinguished using FSC-H and FSC-A. Finally, through CD3+ cell gating, the proportion of CD8+ T cells was further analyzed to evaluate the difference in CD8+ T cell infiltration between the Sh-NC and Sh-NNMT groups.

qRT-PCR

Researchers extracted total RNA from Caki-1 and A-498 cell lines via an RNA extraction kit (Aibotek Biotechnology, Wuhan). Then, they quantified mRNA expression levels under primer-specific amplification.

Cell viability assay

Caki-1 and A-498 cells were cultured with 3000 cells per well, 10 μL of CCK8 (P0010, Beyotime, China). Researchers measured the OD value by the microplate reader (mμlISKANMK3, Thermo, USA).

Migration and invasion assays

Caki-1 and A-498 cell were detected via the 6-well migration and invasion inserts (Corning, USA). In the migration and invasion assay, the upper wells were filled with 8 × 104 cells in 200 μL of DMEM (0.5% FBS) and 600 μL DMEM (10% FBS) was supplement to the lower chamber. Researchers fixed the chambers with methanol following 48-hour incubation. Then the upper chamber was gently cleaned with the cotton swabs and stained using 0.5% crystal violet.

ELISA

The concentrations of IL-6 and IL-1βwere quantified via ELISA kits (Biotech, Beijing, China) as the instructions of manufacturers. Caki-1 and A-498 cell cells, along with their corresponding supernatants, were acquired from cell-conditioned medium and cell lysates.

Cell scratch

Caki-1 and A-498 cells were divided into two groups:① sh-NC, sh-NNMT.The cells were seeded on the bottom of 6-well plates and scratch lines are drawn in order to induce wounds.The plate is cultured under conditions of 37°C , 5% CO2 .At regular time points, especially at 0 h and 48 h, the cells were visualized using microscope for measuring the width of the scratch at constant positions,and pictures were taken.The percentage of migrated cells, as well as the area of the scratch were measured using ImageJ software.

Immunofluorescence

ccRCC tissue sections and adjacent normal tissue sections were retrieved from storage at -80°C. Researcher deparaffinized them in xylene (twice for 10 min each), and put them into rehydration through a graded ethanol series. Heat-induced antigen retrieval was deployed using the citrate buffer. Sections were subsquently washed with PBST, permeabilized with 0.5% Triton X-100 (three times for 5 min each), blocked for endogenous peroxidase activity with 3% H₂O₂, and washed again with PBST. Following blocking with goat serum for 1 hour and a subsequent PBST wash, sections were maintained at 4°C with primary antibodies (rabbit anti-NNMT and rabbit anti-CD3, diluted 1:100) overnight. After further PBST washes, they were maintained with a secondary antibody (anti-rabbit) for 2 hours, washed, and counterstained with DAPI for 10 min. After a final PBST wash, sections were mounted through anti-fade mounting medium and imaged via a laser scanning confocal microscope.

Statistical analysis

All data were calculated using R software and GraphPad Prism. Current analyses performed student's t-test to determine differences between groups, treating p<0.05 as statistical significance.

Result

NNMT passes through the tumor region of inflammation in the complex TME

In order to examine the specific characteristics of the TME in ccRCC, we compared spatial transcriptomic results of ccRCC samples. The spatial patterns of various cell subtypes within the TME were systematically mapped. Supplementary Fig. 4A illustrates an annotated spatial localization of all immune cell subsets, such as plasma cells, macrophages, and T/NK cells. The spatial annotation and distribution of parenchymal/stromal cells within the areas of tumor regions were presented in Supplementary Fig. 4B, encompassing epithelial cells, endothelial cells, and CAFs. Results reveal the spatial localization of specific cell subtypes (Supplementary Fig.s 4C and 4D). It is interesting to note that clearly different localization of T/NK cell subpopulations and endothelial cell subtypes within the tumor regions was observed. Overall, the spatial expression patterns of these diverse cell populations are crucial for key processes as immune infiltration of tumor tissue and stromal cell-mediated remodeling of the microenvironment, which offer molecular mechanistic insights of spatial-level tumor immunotherapy and targeted therapy.

Subsequently, we discovered the overexpression of NNMT was detected in the shape of red foci, because this gene creates metabolically active tumor regions Fig. 1A). The CXCL14 implicates on its role for delineation of immune cell recruitment sectors and inflammation chemotactic signalization area (Fig. 1B). ANGPTL4 has localised enrichment,possibly suggesting a link to the vascular and/or stromal compartment (Fig. 1C). ESM1 was highlighted in red and represented thick vascular vessels (Fig. 1D).The ROIs are marked by black dash lines. The space partitioning on the ROI of NNMT and ANGPTL4 obtained different results, as shown in Fig. 1E that we can clearly see the separation between high- and low-expressing NNMT area,that reflects the local metabolic heterogeneity. The ANGPTL’s expression profile and its regulation near vessels/stroma are shown in Fig. 1F.These are the patterns that are intrinsically linked to tissue functional zonation. Collectively, these results show an important enrichment of T/NK cell populations at the regions of NNMT expression,suggesting that NNMT expression might mediate impairment of T/NK immunity via cellular inflammation as a mediator. We demonstrate that the contributions from NNMT,CXCL14, ESM1, and ANGPTL4 for metabolic, immune, and vascular processes, respectively.These local specific molecular signature derives locale specific molecular proofs to explore mechanism of tumor progresssion,immune evasion; angiogenesis.

Fig. 1.

Spatial Transcriptomics Identifies Functional Zonation of Key Genes in ccRCC. (A-F) Spatial localization of NNMT, CXCL14, ESM1, and ANGPTL4.

T/NK is depleted in the progression of renal cancer

Here we integrated spatial transcriptomics with pseudotime to completely decipher the cellular heterogeneity,spatial localization, and differentiation trajectories in TME. We visualize the spatial clustering using UMAP in Fig. 2A,which groups 12 different cell subpopulations (Cluster 0–11) that define spatial compartments spanning from the tumor core to stromal matrix,and immune-infiltrated areas.Fig. 2B integrates the UMAP clustering and pseudo-time analysis to illustrate global cellular structure heterogeneity and development trajectory from immature cell to mature cell,wherein they have a colour gradient based on their pseudotime trajectory.Fig. 2C and D combine space with time: spatial pseudotime map illustrates spatial abundance and trajectory (blue arrows) of major cell types, e.g., stromal/structural (endothelial,epithelial/tumor, CAFs), and immune cells (macrophages, T/NK, plasma).Concomitantly, for dynamic lineage plots, we observe lower differentiation potentials for endothelial cells in contrast with higher activities found for T/NK cells or macrophages at later stage of tumour progression when increasing pseudo time. Together these results suggest that there are spatially resolved differentiation hierarchies and functionally transitions within the TME.

Fig. 2.

TME Cell Fate Spatially Resolved by Pseudotime Trajectories.(A) Spatial localization of all clusters;(B) UMAP shows each cluster to express the quasi-temporal change expression;(C-D) Expression of temporal changes between various cell types during the development of renal cancer.

Spatially resolved cell–cell communication networks in the ccRCC TME

To describe the spatial arrangement and connection of cell-cell interactions in the TME, we performed integrative analysis. H&E staining determines the pathological structure,showing clear delineation between the tumor nests and the stroma (Fig. 3A).Spatial NMF further partitioned the tumor core into nine functional sub-regions (NMF1-NMF9) that mapped onto their spatial locations (Fig. 3B,D), showing the distribution of immune and other cell subsets. Strikingly, NMF signatures were strongly enriched at tumor-stroma boundaries,suggesting a spatial crosstalk of endothelial cells and CAFs.

Fig. 3.

Integrated Spatial Dissection of TME Architecture and Signaling.(A) HE staining of renal cell carcinoma (B, D) COMMOT algorithm to reveal the signal transduction direction of spatial pathway; (C) Spatial cell communication analysis of the expression of interaction among all cell subpopulations; (E) Spatial conduction status analysis of ANGPTL, CXCL, and TGFβ signaling pathways in renal cell carcinoma; (F) TRAC serves as a key downstream marker for ANGPTL, CXCL, and TGFβ signaling.

Spatial transcriptome map of ligand receptor interaction identified strong interacting network around the CAFs,T/NK cells, and macrophage (Mac) (Fig. 3C). Further application of COMMOT decoded spatial signals of gradients in pivotal tumor-associated pathways - including ANGPTL,CXCL, and TGFβ–to demonstrate a predominant activity of theses three axes (Fig. 3E). The pseudotime analysis demonstrated discrete temporal activation patterns.For example, the hub gene TRAC was significantly upregulated for the ANGPTL axis during mid- to late-stages but showed an early- to mid-stage activation for the CXCL and TGFβ axes (Fig. 3E).

NNMT promotes apoptosis-associated immune evasion in ccRCC

We conducted various systematic analyses in order to realize a comprehensive realization of the TME at single-cell resolution. Initially, t-SNE was plotted to the single-cell transcriptomes to identify major cell subtypes within the TME, encompassing endothelial cells, CAFs, and macrophages(Fig. 4A). The comparison of all the subpopulations revealed that NNMT was a key endothelial cell marker (Supplementary Fig. 1A). DEGs were graphically represented using volcano plots (Supplementary Fig. 1D), while functional enrichment analysis provided an overview of highly significant association of endothelial cells with apoptosis pathways (Supplementary Fig. 1B). Proportional heterogeneity was done further via cellular composition analysis among renal cancer subclones (Supplementary Fig. 1C).

Fig. 4.

Activation of PI3K-Akt signaling pathway accelerates the progression of renal cancer. (A) Different cell types were found in renal cell carcinoma, including endothelial cells, cancer-associated fibroblasts, and macrophages; (B) Analysis of cell population differentiation; (C) Volcanic plot analysis of the differences in endothelial cell groups; (D) Expression of GO enrichment analysis in endothelial cell groups; (E-F) NMF analysis was divided into different groups from MP1 to MP10, and KEGG enrichment analysis was conducted through the characteristic genes of each group.

The inference of developmental trajectory placed endothelial cells at the precursor state with multiple lineages (Fig. 4B). Hypoxia signatures projection indicated heterogeneous responses to oxygen homeostasis in cell subtypes (Fig. 4B), and t-SNE mapping highlighted variations in hypoxia-related gene expression in the endothelial compartment (Fig. 4C). Subsequently, we constructed a co-expression network of endothelial hypoxia-responsive genes in endothelium with six functional modules involved in oxygen sensing, hypoxic response, and processes (Fig. 4D). Similarity heatmaps were used to define interactions among gene modules (MP1-MP8; Fig. 5E), and functional enrichment analysis implied their biological functions. For example, MP1 regulates apoptotic signaling cascade (Fig. 4F). NNMT expression was specifically localized to MP1-enriched cell subpopulations to regulate apoptosis-mediated immune evasion mechanisms and advanced renal carcinoma progression.

Fig. 5.

High expression of NNMT is a key factor inducing the progression of ccRCC. (A) (B) Multiple machine learning algorithms revealed the AUC expression of MP1 group genes, and all showed benign diagnostic efficiency when it was greater than 0.6. (C) TCGA-LIHC analysis of the expression of NNMT; (D) Analysis of the expression prognosis of NNMT combined with KI67 gene; (E) Spatial density and spatial expression of NNMT; (F) NNMT was divided into high-expression and low-expression groups, and then expression analysis was conducted on four idle samples of renal cancer.

High expression of NNMT promotes the progression of ccRCC

Additionally, we separated cells of the endothelial MP1 cell subpopulation.Using the Boruta algorithm, we were able to identify significant feature genes including NNMT which showed a significantly high score,which suggests that it is a key contributor to progression of disease (Fig. 5A). Using multi-approach coefficient heatmaps,the core genes of NNMT was removed. All the scores of NNMT are the largest ones that reflects the high predictability of the tumor’s phenotype.The Ridge AUC matrix showed that the model with NNMT was usually inclined towards higher AUC value and thereby confirmed the effectiveness of NNMT for tumor prediction (Fig. 5B,supplemented with Fig. 2B). We further plotted the prognostic importance of genes in OS and DSS analyses during all forest plotting analysis, where the dots indicate the Hazard ratio of evey gene,while the horizontal lines represent the 95% CI (95%-CI) and the P-value represents the level of statistic significance. In OS analysis, NNMT (P = 0.018, HR = 1.1973) demonstrated different risk associations, and the DSS analysis demonstrated disparate risk hazard ratios and significance level(P <0.001, HR = 1.3554).This versatile dataset could explain the gene-tumor survival prognosis relationship as a foundation to explore the tumor prognostic biomarkers and molecular mechanism (Supplementary Fig. 3A), we also illustrated the multi-cohort HR analysis using a forest plot. The pooled HR is more than 2 under the random-effect model,showing that this is a sufficient condition for the existence of an invariant density, andfects and overall heterogeneity of genetic prognostic risk across different cohorts (Supplementary Fig. 2A).On the difference of survival between the risk group, The difference in survival were evaluated by using the log-rank test based on Kaplan-Meier survival analysis,which based on the prognostic model and several data sets such as TCGA-KIRC, ICGC, GSE22541. The two datasets of GSE167573 and E_MTAB_1880 both confirmed that the high risk group had poor survival outcomes,thus offering systematic support for developing and clinically applying models. Different expression analysis results revealed that NNMT expression was signifi cantly elevated in tumor tissues than normal tissues (p < 0.001) (Fig. 5C), which we validated in several cohorts. In prognostic value analyses, after combining NNMT with the proliferation marker MKI67,Patients were divided in to the following four categories. It demonstrated that patients, whose NNMT and MKI67 were both positive (NNMT+ & MKI67+), had the poorest survival outcomes, whereas the ones with NNMT and MKI67 both negative were the best (NNMT- & MKI67-). This suggests that patients with high NNMT expression and active tumor proliferation have a significantly worse prognosis (P<0.001) (Fig. 5D). Through single-cell distribution studies, tSNE analysis revealed tumor cells works as a superpopulation with the high expression of NNMT(Fig. 5E). The data of spatial transcriptome were used to categorize all samples into the NNMT high- and low-expression group. It validated the higher expression level of NNMT in the high-expression group.

Identification of T cell molecular subtypes

Building on previous studies we propose a novel classification scheme that dissects the complexity of ccRCC. We incorporate four key features such as immune microenvironment,stromal microenvironment, genome stability, and malignant potential.

In order to better understand the pathological mechanisms of ccRCC, we constructed a multi-dimensional dataset based on patients with KIRC in The Cancer Genome Atlas (TCGA),completing with transcriptomic profiles, prognostic methylation marker and top 50 somatic mutation loci. The best estimate of cluster silhouette index and gap statistics were estimated to be a k = 2 as an optimal cluster resolution for downstream analysis. With the approach of ssGSEA,we further classified the TCGA-KIRC patients into two subclasses as follows: CS1 and CS2.This grouping was strongly supported by several consensus clustering approaches including SNF,multi-kernel learning (CIMLR) and PINSPlus (Fig. 6A-B).

Fig. 6.

Functional characterization of molecular subtypes in ccRCC (A) Consensus clustering heatmap based on the multi-omics data,best partitioning of the ccRCC samples into two molecular subtypes (CS1, CS2). The brighter a color is, the more similar are those samples.(B) Subtype agreement comparison among different clustering methods (SNF,CIMLR, PINSPlus, NEMO, COCA, MoCluster, LRCluster, ConsensusClustering, IntNMF, iClusterBayes).The high concordance of CS1 and CS2 among different approaches suggests that the subtype classification is reliable.(C) Heat map of representative immune- and apoptosis-related pathways activities between CS1 and CS2. The yellow color represents a higher level of pathway activity,while cooler colors indicate reduced activity.(D) Heatmap of differential activity for the TF MUC regulon, and chromatin remodeling genes between CS1 and CS2. The color scale represents relative activity level.(E) Kaplan–Meier overall survival (OS) analysis of CS1 and CS2 in discovery cohort. The OS is significantly higher for CS1 compared to CS2(P<0.001). (F) OS survival analysis of CS1 and CS2 in validation cohort,also confirmed that the survival rate of CS1 was much higher than that of CS2(P<0.001).

The survival analysis also showed that there is a significant difference in prognosis among different subtypes (P<0.001) and CS2 has the worst prognosis(Fig. 7E-F). The immune characteristics showed that increased T-cell apoptosis pathway enrichment in CS1 compared to dominant immunosuppressive character of CS2. PD-1 blockage treatment had promising therapeutic effects for both subtypes.Interestingly, we found WNT signaling axis to be highly enriched in CS2(Fig. 6C) suggesting its involvement in the progression of RCC.

Fig. 7.

NNMT-Linked Immunomodulation in ccRCC.(A) Expression of immune checkpoint molecules associated with NNMT.(B) Immune expression profiles of CS1 and CS2 subtypes in ccRCC.(C) Comparative analysis of immune infiltration between high NNMT-expressing and low NNMT-expressing groups.

NNMT-related T cell infiltration induces ccRCC immune escape

An analysis of tumor immune checkpoints was performed in a systematic way of characterizing the immune regulation in ccRCC. Heatmaps were generated to visualize the distribution of molecular feature—including mRNA expression, methylation differences, and copy number variation frequencies—annotated by immune checkpoint inhibitor (ICI) functions (stimulatory/inhibitory) across patient samples.

We observed an upregulation of co-stimulatory molecules such as CD80, CD28, ICOSLG whereas the expression of ICIs such as SLAMF7 and CD274 (PD-L1) was downregulated. In addition,inflammatory cytokine receptors CXCL10, CXCL9 and IL1R1 were significantly activated (Fig. 7A).The heatmap of the correlations between immunity score, stromal score, estimated cellularity content, and cluster status (cluster 1 and 2) is shown in Fig. 7B,with examples of annotated immune/stromal scores, MeTIL scores, and the assigned subtypes. Both subtypes showed a heterogenous immune cell infiltration pattern,including B cell, T-cell subtypes, and NK cell. CS2 subtype highly activate of PD-1 pathway receptors(CD80, CD28, CD276), and significant enriched NNMT.According to the stratification of NNMT expression, we found that there was a significantly positive correlation between the high expression level of NNMT and higher infiltrated effector T cell levels including CD8+ effector memory or CD8+T cell.

AI-driven targeting of NNMT emerges as a promising synergistic strategy for ccRCC immunotherapy

While NNMT inhibition restricts ccRCC development, BET inhibitors (BETis) such as I-BET-762 and I-BET151—prime candidates for combination with immunotherapy—downregulate PD-L1 and repolarize immune cells via the BRD4 acetyl-lysine pocket (Fig. 8B). Machine learning-based modeling predicts that these interactions are key to their immune-enhancing activity. Notably, PRISM analysis (Fig. 8A) reveals that these compounds, along with kinase inhibitors BMS-387032 (CDK9i) and Pacritinib (JAK2i), also modulate NNMT, suggesting a broad epigenetic-metabolic-immune crosstalk. Intriguingly, our models predict that BMS-387032 and Pacritinib exhibit unexpected BET-binding modes (Fig. 9B), indicative of polypharmacology. This AI-guided insight suggests that smart combination strategies, leveraging NNMT status alongside BET inhibition, may overcome resistance in "cold" tumors like ccRCC.

Fig. 8.

Computational and structural analysis of NNMT-related molecules having immuno-modulating properties in RCC. (A) Relevance score calculated by PRISM (correlation with AUC) for the selected molecules towards NNMT,highlighting I-BET151, I-BET-762, BMS-387032 and Pacritinib with non-negligible association pointing to possible cross-talk of NNMT with the epigenetic/kinase pathway.(B) Predicted or co-crystal binding poses of the same compound in the acetyl-lysine recognition pocket of BRD4 BD1 (PDB: 3MXF),highlighting the critical contacts to preserved amino acids (e.g., Q133, R136, Y94). The structures of these complexes were optimized by machine learning-based virtual screening and binding free energy calculations. Together,these data provide the rationale of a rational AI-guided approach, which can be used to either re-purpose or optimize existing NNMT modulators as immunotherapeutics against RCC by enhancing anti-tumour immune response.

Fig. 9.

NNMT Mediates Oncogenic Invasion and Renal Dysfunction. (A) Cell viability of A498 and Caki-1 renal cancer cells transfected with sh-NNMT-1 and sh-NNMT-2 was assessed by CCK-8 assay. (B) The effects of sh-NNMT-2 transfection on the proliferation and migration of A498 and Caki-1 cells were examined using Transwell assay. (C) NNMT expression and its correlation with renal glomerular injury were compared between adjacent non-tumor tissues and tumor tissues. (D) NNMT expression levels in adjacent non-tumor tissues versus tumor tissues. (E) Colony formation assay assessing the proliferative capacity of A498 and Caki-1 renal cancer cell lines following transfection with sh-NNMT-2.

Decreased NNMT expression inhibited the progression of ccRCC

Previous evidence identifies NNMT as a critical therapeutic target, specifically upregulated in renal carcinoma later. We developed NNMT-specific shRNA interference fragments (sh-NNMT-1 and sh-NNMT-2). Transfection into renal carcinoma A498 and Caki-1 cell lines demonstrated that optimal knockdown efficiency for sh-NNMT-2, as confirmed by CCK8 assays, thereby establishing this construct for further experiments(Fig. 9A).

Invasion and migration functional analyses showed that the knockdown of NNMT significantly affected the invasion and migration abilities of ccRCC in Transwell and colony formation experiments(Fig. 9B,E). IHC analysis showed that the clinic samples exhibited positive staining for NNMT (Fig. 9D) and,we further observed that the expression of NNMT was also associated with glomerular injury in ccRCC (Fig. 9C).In conclusion, our findings support that NNMT is a pro-progressor factor in ccRCC and its persistent up-regulation could be responsible for kidney failure in late stages.

Reducing NNMT expression restores CD8 T cell function and inhibits immune escape in renal cancer

Elevated NNMT expression also presupposed quicker exhaustion of CD8⁺T-cell. Immunofluorescence co-localization studies revealed significant spatial overlap of elevated NNMT and KI67 expression in ccRCC. Such co-localization was markedly reduced when NNMT was knocked down, thus confirming that NNMT suppression inhibits tumor progression (Fig. 10A,E).

Fig. 10.

NNMT Depletion Reverses T-cell Exhaustion. (A,E) Fluorescence co-localization expression of KI67 and NNMT in renal cancer tissue sections; (B,F) Detection of the co-expression of KI67 and T-cell marker CD3 in renal cancer tissues; (C-D) Immunoflow cytometry was used to detect the expression of CD3 T and CD8 T cells.

Further immunofluorescence co-localization demonstrated decreased CD3⁺T-cell infiltration in tumor regions using CD3⁺T-cell markers. On the one hand, sh-NNMT treatment had a great impact on significantly enhanced CD3⁺T-cell infiltration (Fig. 10 B,F). This was eventually proved by flow cytometry analysis ultimately proving that there was a substantial recovery in proportions of CD8⁺T-cell in the sh-NNMT group(Fig. 10 C-D).

Inhibition of NNMT expression promotes ccRCC cell apoptosis and inflammation

Major amplification of TGF signaling and inflammation were found by our investigation, which we confirmed using qPCR and ELISA in the A498 and Caki-1 cells,which demonstrated that the knockdown of NNMT (sh-NNMT) led to a significant decrease in the expression levels of TGF-β1, KI67, IL-6 and IL-1β.BAX and caspase-3 levels increased; thus indicating that downregulation of NNMT induces apoptosis in addition to enhancing the ability of CD8+ T cells to kill tumour cells thereby preventing their growth (Fig. 11 A-F).

Fig. 11.

NNMT Silencing Suppresses Inflammation and Activates Apoptosis. (A,C) Investigate the expression of the TGF-β signaling pathway transfected with sh-NNMT in cell lines A498 and Caki-1; (B,D) To investigate the expression of IL6 and IL-1β transfected with sh-NNMT in cell lines A498 and Caki-1; (E,F) To investigate the expression of Caspase3,Bax,Ki67 and Bcl-2 transfected with sh-NNMT in cell lines A498 and Caki-1.

Discussion

The more immunosuppressive/metabolically ignored microenvironment of ccRCC represents an important barrier for immune surveillance,especially leading to functional exhaustion and accelerated cell death of CD8+ T lymphocytes. for tumour immune escape [21]. Extensive studies have shown that NNMT is an important hub within this pathogenic loop.High levels of NNMT expression contribute to metabolic dysregulation, immunosuppressive signaling axis (especially the TGF-β pathway),and CD8+ T-cell fate. Understanding the complex network will be key to develop novel immunometabolic therapeutic approaches in renal cancer,which is schematically illustrated on the particular flowchart.

The TME of ccRCC represents a primary catalyst in promoted tumor progression [22]. Research has demonstrated that the stabilization of HIF-alpha as a result of VHL deficiency, hypoxia environments, and build-up of immunosuppressive cytokines (especially IL-10) [23,24] all combine to form a highly hostile environment that causes serious destroys the CD8 + T cell functions.CD8+ T cells enter a typical exhaustion state, further complicating therapeutic intervention. It is important to note that NNMT increases significantly in such an environment, as a key activity that contributes to the ineffective production and apoptosis of CD8+ T cells. The experiments conducted by us validated the NNMT’s role in modulating apoptosis in renal cancer cells and attenuating tumor-associated inflammation. Crucially, the upregulation of NNMT establishes a robust interactive relationship with the hyperactive TGF-β signaling within the RCC TME, which collaboratively builds a potent immunosuppressive and pro-apoptotic axis. On the one hand, TGF-β acts as a strong inducer of CD8+ T cell apoptosis [25] through the classical Smad2/3/4 signaling cascade to downregulate BAX and Caspase-3, but to upregulate Bcl-2. Our experimental findings are another indication that NNMT downregulation is mainly the key to reinvigoration of CD8+ T cell performance and immune responses. NNMT inhibition can regulate TGF-β pathway to prevent tumor invasion and proliferation.

Our experimental findings further demonstrate that exposure to unremitting cellular inflammation within the TME is another basic stimulus of RCC progress, which greatly perturbs the dynamics interplay between vascular endothelial cells with infiltrating T cells,and all together aggravate disease aggressiveness. Many pro-inflammatory cytokines (TNF-α, IL-1β) support this chronic inflammatory condition and trigger key signaling pathways of TGF-β in tumor vascular endothelial cells [26]. We observed that there was obvious expression of NNMT in kidney cancer ECs and it often occurred together with the expression of inflammatory factor; ThereforeDownregulation of NNMT largely dampens levels of IL-6, IL-1b and TNF-a which is possibly regulated through the TGF-b axis. Since we have previously reportedT cells were recruited to the tumor site via inflammatory-activated ECs that upregulated PD-L1 [27]. This docks straight onto PD-1 found on T cells and delivers a potent negative signal Whilst inflammation impairs endothelial integrity:increases vascular permeability (facilitating metastasis), and results in production of pro-angiogenic factor VEGF, creating an abnormal, dysfunctional tumor vascular architecture [28]. Infiltrated T cells not only encounter direct inhibition via PD-L1 but are also chronically exposed to an inflammatory microenvironment enriched with diverse immunosuppressive factor IL-6 and IL-10 [29,30]. All this contributes to fatigue and reduced activation of effector T cells, and sets the stage for expansion of Tregs. Finally,the inflammatory-induced functional disorder of endothelial cell resulting in an immunosuppressive phenotype, barrier dysfunction,and anti-angiogenesis and T cell exhaustion form a self-reinforcing vicious cycle.

Despite the robust findings presented here, our study has several limitations. Firstly, while we demonstrate that NNMT downregulation suppresses inflammation possibly through the TGF-β axis, the specific molecular mechanisms remain to be fully elucidated. Future studies are warranted to identify whether this regulation involves specific transcription factors or epigenetic modifiers. Secondly, our in vivo experimental models have limitations in fully recapitulating the complexity of the human immune microenvironment. While murine models provide valuable insights, there are inherent differences in immune cell composition and function between mice and humans, which may affect the translatability of our findings regarding the NNMT-mediated immunosuppressive niche.

Data availability

The datasets were sourced from GEO (https://www.ncbi.nlm.nih.gov/geo/), Xena (https://xena.ucsc.edu/), which are available from the corresponding authors of this study upon reasonable request.

Ethical statement

All animal experiments and human renal cancer tissue sections were approved by the Ethical Editorial Committee of Ningxia Medical University and completed in Ningxia Medical University, ethics number: KYLL-2022-0117

Funding

This work was supported in Chongqing Natural Science Foundation (No. CSTB2023NSCQ-LZX0099).; Chongqing Talent·Innovation and Entrepreneurship Demonstration Team, CQYC202203091342.

CRediT authorship contribution statement

Lexin Wang: Methodology, Funding acquisition, Formal analysis, Data curation, Conceptualization. Honglin He: Data curation, Conceptualization. Ke Su: Data curation, Conceptualization. Yunjing Gao: Formal analysis, Data curation, Conceptualization. Ying Luo: Formal analysis, Data curation, Conceptualization. Huanhuan Tan: Formal analysis, Data curation, Conceptualization. Ziyang Liu: Funding acquisition, Formal analysis, Data curation, Conceptualization. Ke Xu: Funding acquisition, Formal analysis, Data curation, Conceptualization. Yi Li: Funding acquisition, Formal analysis, Data curation, Conceptualization. Xiaosong Li: Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.

Declaration of competing interest

The authors declare no conflict of interest.