Tumor-intrinsic NSUN2 orchestrates immunosuppression in lung adenocarcinoma via the m⁵C-HDAC8-CCL5 axis

Abstract

Background

Insufficient T-cell infiltration is a major barrier to the efficacy of immune checkpoint inhibitors (ICIs) in lung adenocarcinoma (LUAD). We aimed to investigate how the tumor-intrinsic m⁵C methyltransferase NSUN2 shapes the immune landscape of LUAD.

Methods

Nsun2 conditional knockout mice and syngeneic mice models were employed. Single-cell RNA sequencing (scRNA-seq) and m⁵C sequencing were performed to elucidate the downstream pathway regulated by NSUN2. LUAD specimens from patients receiving neoadjuvant immunotherapy were used to assess the correlation between NSUN2 expression and immune infiltration.

Results

Elevated tumoral NSUN2 expression was correlated with a marked lack of CD8⁺ T-cell infiltration and immunotherapy resistance. Mechanistically, NSUN2 enhances the translation of histone deacetylase HDAC8 in an m⁵C-YBX1-dependent manner, which in turn directly represses the transcription of the chemoattractant CCL5, thereby impairing CD8⁺ T-cell recruitment into the tumor. NSUN2 depletion reversed this immunosuppressive axis and converted immunologically “cold” tumors into “hot”. Tumor-targeted liposomes loaded with the NSUN2 inhibitor synergized with anti-PD-1 therapy to induce significant tumor regression.

Conclusions

Our findings identify NSUN2 as a critical orchestrator of T-cell exclusion in LUAD. It serves as both a candidate predictive biomarker for ICI failure and a promising druggable target. Targeting NSUN2 offers a potential strategy to overcome immunotherapy resistance and improve clinical outcomes in LUAD patients.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40164-026-00826-9.

Background

Lung adenocarcinoma (LUAD) is the most prevalent and aggressive subtype of lung cancer, and is typically diagnosed at advanced stages with poor treatment outcomes [1]. Although immune checkpoint inhibitors (ICIs) targeting PD-1/PD-L1 have transformed therapeutic paradigms, their clinical benefits remain limited to approximately 20% of patients with LUAD [2–4]. This treatment resistance primarily stems from an immunosuppressive tumor microenvironment (TME) featuring “cold” tumor phenotypes - characterized by deficient T cell infiltration and abundant immunosuppressive cell populations [5, 6]. In contrast, “hot” tumors with pre-existing T cell infiltration demonstrate significantly better responses to ICIs, underscoring the pivotal role of the immune contexture in determining treatment efficacy. This recruitment of T cells is fundamentally governed by chemokine networks, which normally orchestrate precise immune cell trafficking through coordinated molecular interactions [7]. By dysregulating these secretion patterns, tumors establish barriers that exclude effector T cells [8]. Despite their importance, the upstream epigenetic mechanisms responsible for silencing these essential chemokines in LUAD remain largely unexplored. Consequently, identifying key proteins that drive T-cell infiltration is essential for converting ‘cold’ LUAD tumors into ‘hot’ ones, thereby overcoming immune resistance and broadening the clinical benefits of immunotherapy.

Recent studies have identified RNA methylation as a crucial post-transcriptional regulatory mechanism in cancer biology. However, how RNA methylation regulators orchestrate the tumor immune microenvironment and influence immunotherapy efficacy remains unclear [9]. In this study, based on a screening of high-throughput proteomics and transcriptomic data, we identified the 5-methylcytosine (m5C) methyltransferase, NSUN2, as the most relevant RNA methylation regulator associated with immunotherapy efficacy in LUAD. NSUN2 has recently emerged as a pivotal player in mediating the methylation of mRNA and tRNA to influence RNA metabolism and function [10, 11]. Although the role of NSUN2 in tumor progression is increasingly documented, its specific impact on the LUAD immune microenvironment remains largely unexplored [12–14]. Importantly, the potential of NSUN2-mediated m⁵C modifications to directly regulate chemokine networks and thereby control immune cell trafficking patterns in LUAD represents a critical unanswered question in the field of cancer immunotherapy resistance.

In this study, we integrate clinical specimens from patients receiving neoadjuvant immunotherapy with genetically engineered mouse models, and demonstrate that NSUN2 deposits m⁵C modifications to enhance HDAC8 mRNA translation via the reader protein YBX1, thereby repressing CCL5 secretion and contributing to an immunologically “cold” TME. Genetic inhibition of NSUN2 restores CCL5-dependent CD8⁺ T cell recruitment and synergizes with PD-1 blockade to suppress tumor growth. Pharmacological NSUN2 inhibition delivered via iRGD-liposomes also produced a synergistic response with PD-1 blockade. Our work not only elucidates NSUN2-m⁵C-YBX1-HDAC8-CCL5 axis as a druggable regulator of chemokine networks but also provides a rationale for targeting RNA methylation to overcome ICI resistance in LUAD.

Methods

Animals and tumor models

All mice had a C57BL/6J genetic background. For the spontaneous LUAD model, genetically engineered KrasLSL^−G12D/+,Sftpc-MerCreMer mice combined with either Nsun2^flox/flox or Nsun2^+/+ genotypes were generated by Cyagen Bioscience employing a LoxP-targeting system. At 6 weeks of age, these male and female mice received intraperitoneal injections of tamoxifen (75 mg/kg body weight) for four consecutive days to activate pulmonary-specific Cre recombinase activity, with experimental endpoints set at 7–8 weeks after the completion of tamoxifen induction. For immunotherapy studies in this model, anti‑PD‑1 mAb (BE0146, BioX‑Cell) or IgG isotype control (BE0089, BioX‑Cell) was administered intraperitoneally at 100 µg per injection, starting at week 5 post‑induction and continuing every five days until sacrifice. Mice were randomly assigned to treatment groups, and group sizes were matched for sex.

For the subcutaneous allograft model, 1 × 10⁶ KP or LLC cells in 100 µL PBS were injected subcutaneously into the right flank of 6‑ to 8‑week‑old male and female mice. Tumor dimensions were measured every 2–3 days using a digital caliper, and tumor volumes were calculated as (length × width²)/2. In compliance with institutional guidelines, mice were euthanized when tumors reached a maximum diameter of 15 mm. For immunotherapy, anti‑PD‑1 mAb or IgG isotype control was administered intraperitoneally at 100 µg per injection every three days starting on day 6 post‑implantation until sacrifice. For in vivo pharmacological inhibition of NSUN2, the NSUN2 inhibitor MY-1B (T85335, TargetMol) was encapsulated in iRGD‑functionalized liposomes (iRGD‑MY‑1B liposomes) at a concentration of 3 mg/mL (MY‑1B equivalent) and administered via tail vein injection at 30 mg/kg every three days starting on day 6 post‑implantation until sacrifice. Mice were randomly assigned to treatment groups, and group sizes were matched for sex.

FoxP3^DTR transgenic mice on a C57BL/6 background were used for Treg ablation experiments as previously described [15]. Briefly, 8-week-old FoxP3^DTR mice were subcutaneously implanted with either control or Nsun2-KO KP cells (1 × 10⁶). Mice received three consecutive daily intraperitoneal injections of diphtheria toxin (DT; 30 ng/g body weight; Sigma-Aldrich) or phosphate buffered saline (PBS) starting on day 5 post-implantation, followed by subsequent injections every 3 days at the same dose until the experimental endpoint.

All mice were provided free access to food and water and maintained under controlled environmental conditions with a 12-hour light/dark photoperiod, ambient temperature of 22–26 °C, and ~ 40% humidity. All animals were bred and housed in a specific pathogen-free (SPF) environment at Shanghai Jiao Tong University School of Medicine.

Synthesis and characterization of iRGD-MY-1B liposomes

iRGD-MY-1B liposomes were synthesized by Ruixi Biological Technology (Xi’an, China). Briefly, soybean phosphatidylcholine (SPC), cholesterol, 1,2- distearoyl-sn-glycero-3-phosphoethanolamin(DSPE)- Polyethylene glycol 2000 (PEG2K)- iRGD, and NSUN2 inhibitor MY-1B (T85335, TargetMol) were co-dissolved in 3 mL of chloroform in a sample vial. The organic solvent was removed by rotary evaporation under reduced pressure to form a thin lipid film. Subsequently, 2 mL of deionized water was added to hydrate the lipid film, followed by sonication and extrusion through polycarbonate membranes with a pore size of 200 nm using a liposome extruder.

The resulting liposomal suspension was then dialyzed using a nanodialysis device equipped with a polycarbonate membrane (pore size: 50 nm) to remove unloaded MY-1B. After dialysis, the volume was adjusted to 5 mL with deionized water. Finally, a cryoprotectant was added to the liposome formulation prior to lyophilization.

To prepare the working solution, iRGD-MY-1B liposomal lyophilisate was reconstituted in ultrapure water; complete dissolution was confirmed visually. The dispersion was homogenized by water-bath sonication (2 min, room temperature) to secure uniform nano-sized distribution, then rendered isotonic by addition of an equal volume of 2× physiological saline. The reconstituted nanomaterial suspension was aliquoted under aseptic conditions and stored at − 80 °C until use.

Cell lines

The murine Lewis lung carcinoma (LLC) cell line was obtained from the National Collection of Authenticated Cell Cultures (Cell Bank, Chinese Academy of Sciences, Shanghai, China). The KrasLSL^−G12D/+/P53^fl/fl (KP) cell line, kindly provided by Prof. Hongbin Ji (Chinese Academy of Sciences), was derived as a single-cell clone from a spontaneous KP LUAD mouse model with C57BL/6 genetic background [16]. All cell lines were cultured in DMEM (Gibco, C11965500BT) supplemented with 10% fetal bovine serum (Gibco, 10270-106) and 1% penicillin-streptomycin (Gibco, 15070063) and maintained at 37 °C in a humidified 5% CO₂ incubator.

Generation of CRISPR-edited Nsun2-deficient tumor cell lines

Plasmids of pLenti-U6-sgRNA-EFS-NS-Cas9-2 A-Puro bearing both the U6 promoter and Cas9 coding sequence (CDS) was used for genome editing. The sgRNAs targeting the mouse gene Nsun2 were designed based on corresponding target sequences, as previously validated in the Feng Zhang lab [17]. Mouse Nsun2 sgRNA was designed using the sgRNA designer tool (Obio Technology). All sgRNAs were synthesized and cloned into the pLenti-U6-EFS-NS-Cas9-2 A-Puro vector (Obio Technology). To knock out the target gene, a lentivirus was constructed by co-transfection of the plasmid pLenti-U6-sgRNA-EFS-NS-Cas9-2 A-Puro and packaging plasmids (psPAX2 and pMD2.G) into 293T cells for 48 h. The lentivirus was used to infect the cell line for 48 h, followed by selection with puromycin (2 mg/ml) for consecutive 5 days. The efficacy of knockout (KO) was confirmed by immunoblotting.

In vivo CD8⁺ T-cell, MDSC depletion, and CCL5 neutralization

To determine whether the pro-tumorigenic effect of Nsun2 was immune-mediated, female C57BL/6 mice (6–8 weeks old, n = 25) were administered intraperitoneal (i.p.) injections of either anti-CD8 depleting antibody (10 mg/kg, A2102, Selleck) or IgG isotype control (10 mg/kg, A2145, Selleck). Two days later, mice were subcutaneously implanted with 1 × 10⁶ Nsun2 KO or control KP cells. A second and third dose of antibody (10 mg/kg, i.p.) was administered 5 and 8 days post-implantation. Tumor progression was assessed every three days (n = 5 mice/group), followed by tumor resection and flow cytometry to verify the CD8⁺ T-cell depletion efficacy. To evaluate the role of MDSCs in Nsun2-deficient lung cancer, we depleted MDSCs using an anti-Ly6G monoclonal antibody (A2158, Selleck; 10 mg/kg, i.p.). Treatment was initiated five days after tumor implantation, with antibody administration every three days for four consecutive doses.

For CCL5 neutralization, 8-week-old female C57BL/6 mice were subcutaneously implanted with 1 × 10⁶ control or Nsun2-KO KP cells (n = 5 per group). Starting on day 6 post-implantation, mice received i.p. injections of either anti-mouse CCL5 neutralizing antibody (10 mg/kg; HY-P991753, MCE) or mouse IgG1 kappa isotype control (10 mg/kg; HY-P99977, MCE) every other day until sacrifice.

Plasmid construction and small interfering RNA (siRNA) transfection

To construct the transient-transfection plasmids of Nsun2^WT, the cDNAs were synthesized and subcloned into pcDNA3.1 vector by Genechem Technology.

siRNAs targeting Hdac8, Hdac6, Sirt3, Sirt5, and Kdm5a, Dot1l along a non-targeting scrambled control were obtained from GenePharma (Suzhou, China). The corresponding siRNA sequences are listed in Table 1. Cells were seeded in six-well plates at a density of 2 × 10⁵ cells per well and allowed to adhere for 24 h prior to transfection. Transfection was performed using Lipofectamine 3000 (Life Technologies), with each well containing 20 nM of the respective siRNA.

Table 1.

The target sequences of siRNA and CHIP-qPCR primers used in the study

siRNA	Target sequences
Hdac8	ACTGTGACATCATGGCAGGAA
Hdac6	ATCAACAAGTAGTTTGTGGTT
Sirt5	AAGCAGATGTTGTGTTGACCC
Kdm5a	TTCAATGCTACTTATCAGCCG
Sirt3	AATAGACAGGACAGGTAGGTC
Dot1l	AACATCCTCAGTGTCTCGGCC

CHIP-qPCR	Ccl5
Primer−1-F	GTCCAGCTGAGATGCACTGT
Primer−1-R	GCCCACCAAGGTACAGAGTC
Primer−2-F	GATCCATTCCCACTGTGGCA
Primer−2-R	GGTCAGAGCTGAACGTCACA
Primer−3-F	TGGACTGGAGGGCAGTTAGA
Primer−3-R	CCAGGACTTGGGGAGTTTCC

RNA isolation and RT-qPCR

Total RNA was extracted from the cultured cells using the FastPure Cell/Tissue Total RNA Isolation Kit (Vazyme). For cDNA synthesis, 1 µg of total RNA was converted to complementary DNA using ABScript III Reverse Transcriptase (ABclonal, Wuhan, China), following the supplier’s protocol. Subsequent RT-qPCR analysis was conducted using ChamQ SYBR qPCR Master Mix (Vazyme, Nanjing, China), according to standard procedures. The comparative Ct calculation method was used for data analysis, with β-actin serving as the endogenous reference gene for normalization. The nucleotide sequences of all the PCR primers are provided in Table 1.

Ribosome nascent-chain complex-bound mRNA-qPCR (RNC-qPCR)

The RNC-qPCR assay was performed according to a previously established protocol [18, 19]. To begin, cells were treated with 100 µg/ml cycloheximide for 15 min at 37 °C to stall ribosomes. Subsequently, the cells were lysed on ice for 30 min in buffer containing 1% Triton X-100, 20 mM HEPES-KOH (pH 7.4), 15 mM MgCl₂, 200 mM KCl, 100 µg/ml cycloheximide, and 2 mM dithiothreitol. The crude lysate was clarified by centrifugation at 16,200 × g for 10 min at 4 °C to remove cellular debris. A 10% aliquot of the resulting supernatant was used as an input control. The remaining supernatant was layered over 30% sucrose buffer and ultracentrifuged at 174,900 × g for 5 h at 4 °C. The resulting ribosome-containing pellets were collected, and total RNA was extracted from both RNC samples and input controls using TRIzol reagent for subsequent analysis by RT-qPCR. The translation ratio (TR), defined as the abundance ratio of the translated mRNA to total mRNA for a certain gene, was evaluated using the following formula: TR = RNC-mRNA(2^−ΔΔCT) /Total mRNA(2^−ΔΔCT).

Chromatin immunoprecipitation assays (ChIP)

Chromatin immunoprecipitation (ChIP) was performed using the SimpleChIP Enzymatic Chromatin IP Kit (26157, Thermo Fisher Scientific). For immunoprecipitation, enzymatically sheared chromatin was incubated with an anti-H3K27ac antibody (#8173, CST). A parallel reaction using rabbit IgG isotype antibody (2729 S, CST) was used as the negative control. Subsequent analysis of the purified DNA was performed via quantitative PCR with primers targeting the H3K27ac binding site in the CCL5 promoter, which was pre-validated and synthesized by Saihengbio, Inc.

Dual-luciferase reporter assay

A fragment spanning the 5′ untranslated region (5′-UTR) and the adjacent coding sequence (CDS) of mouse Hdac8 mRNA, which encompasses the primary m⁵C sites identified by MeRIP-seq, was chemically synthesized and cloned into the Renilla luciferase 3′-UTR of the psiCHECK-2 vector (Promega) downstream of the Renilla luciferase stop codon (Genechem Technology, Shanghai, China). The wild-type reporter (Hdac8-WT) contained the native sequences, whereas the mutant reporter (Hdac8-Mut) carried simultaneous substitutions at the identified m⁵C sites: C-to-T mutations in the 5′-UTR (C32T, C71T) and C-to-A silent mutations in the CDS (C118A, C139A) to disrupt methylation without altering the amino acid sequence.

For reporter assays, KP and LLC cells seeded in 24-well plates were co-transfected with 400 ng of psiCHECK2-Hdac8 reporter (wild-type or mutant) and 100 ng of Nsun2 overexpression vector (pcDNA3.1-NSUN2) or empty vector control, with total DNA normalized to 500 ng per well. At 48 h post-transfection, firefly and Renilla luciferase activities were measured sequentially using the Dual-Luciferase Reporter Assay System (Promega). Renilla activity was normalized to firefly for transfection efficiency.

Study participant details

Following approval from the Institutional Review Board of Shanghai Ruijin Hospital and after obtaining informed consent, we retrospectively enrolled two distinct cohorts of LUAD patients, all of whom were managed at the Department of Thoracic Surgery.The first cohort comprised 52 treatment-naïve patients undergoing surgery between 2016 and 2019. Paired LUAD tumor tissues and adjacent normal tissues were collected from these individuals for the construction of a tumor microarray. All resected samples were snap-frozen in liquid nitrogen for preservation. The second cohort consisted of 18 patients who received immune checkpoint blockade (ICB) therapy before surgical resection. Diagnoses were pathologically confirmed by at least two independent pathologists, and clinical characteristics were compiled from their medical records. The efficacy of ICB treatment was evaluated by a minimum of three pathologists. Survival time was defined as the interval from diagnosis to death or the last follow-up, with data censored accordingly for the surviving patients.

Histopathological analysis

For histopathological quantification of tumor burden in the spontaneous LUAD model, whole H&E‑stained lung sections were analyzed using QuPath software (version 0.5.1) [20]. Tumor regions were systematically annotated by a pathologist blinded to the experimental groups based on morphological criteria, including atypical glandular architecture, nuclear atypia, and invasive growth. Inflammatory infiltrates, stromal areas, and hemorrhagic regions were excluded from the tumor regions. The tumor burden was calculated as the percentage of annotated tumor area relative to the total lung section area for each mouse.

Immunohistochemistry (IHC) analysis

IHC staining was performed by Servicebio Inc. commercially available primary antibodies. Following established protocols [21], tissue processing included overnight fixation in 10% neutral-buffered formalin and subsequent paraffin embedding. For IHC preparation, 4-µm sections were dewaxed and treated with 3% H₂O₂ to quench endogenous peroxidase activity. Antigen retrieval was achieved through 5-minute pressurized heating in alkaline buffer (pH 8.0), followed by 1-hour blocking with 3% bovine serum albumin. The primary antibody (20854-1-AP, Proteintech, 1:250) was incubated overnight at 4 °C, with subsequent detection using an HRP-conjugated secondary antibody (Solarbio Detection Kit) and DAB chromogen at room temperature for 1 h. The antibodies used in this study are listed in Table 2. Protein expression levels were independently evaluated by two pathologists using a scoring system described previously [22]. The final IHC score (range 2–8) was calculated as the sum of the proportion score (1, <25%; 2, 25%–50%; 3, 50%–75%; 4, > 75%) and the intensity score (1, negative; 2, weak; 3, moderate; 4, strong).

Table 2.

List of antibodies used for this study

Reagent	Source	Identifier
A2102	Anti-mouse CD8α-InVivo	Selleck
A2158	Anti-mouse Ly6G-InVivo	Selleck
BE0146	InVivoMAb anti-mouse PD-1	BioX-Cell
BE0089	InVivoMAb rat IgG2a isotype	Bio-X-Cell
HY-P991753	Anti-Mouse CCL5 Antibody	MCE
HY-P99977	Mouse IgG1 kappa, Isotype Control	MCE
#8173	Anti-H3K27ac	CST
20854-1-AP	Anti-NSUN2	Proteintech
66009-1-Ig	Anti-ACTIN	Proteintech
A8865	Anti-HDAC8	ABclonal
# 9715	Anti-Histone H3	CST
Ab269456	Anti-CD33	Abcam
ab217344	Anti-CD8	Abcam
#85,336	Anti-CD8	CST
#100,302	Anti-CD3	BioLegend
#317,302	Anti-CD3	BioLegend
#102,102	Anti-CD28	BioLegend
#302,914	Anti-CD28	BioLegend
65-0865-14	Viability Dye	eBioscience
100,753	Anti-CD8	Biolegend
102,008	Anti-Gr-1	Biolegend
109,814	Anti-CD45.2	Biolegend
109,832	Anti-CD45.2	Biolegend
46-1951-80	Anti-CCR5	ThermoFisher
17-7311-82	Anti-IFNγ	eBioscience
372,206	Anti-Granzyme B	Biolegend
101,243	Anti-CD11b	Biolegend
A3534	Anti-YBX1	ABclonal
ab202894	Anti-ALYREF	Abcam
A25318	Anti-YTHDF2	ABclonal
#2729	Normal Rabbit IgG	CST

Immunofluorescence (IF) analysis

Tissue specimens were collected and immediately fixed in 10% formalin for 24 h before paraffin embedding (Servicebio, Wuhan, China) for subsequent immunofluorescence examination. For immunofluorescence staining, 4 μm-thick sections were prepared and processed using standard deparaffinization and rehydration steps. Antigen retrieval was performed using an alkaline buffer (pH 8.0) with a 3-minute pressurized steam treatment. Non-specific binding sites were blocked with 10% fetal bovine serum for 60 min at ambient temperature. Primary antibody incubation was carried out at 4 °C for 16 h using the following reagents: CD8 (ab217344, Abcam, 1:350) for mouse, and NSUN2 (20854-1-AP, Proteintech, 1:1000), CD8 (#85336, CST, 1:250), and CD33 (ab269456, Abcam, 1:50) for human tissue. After washing with phosphate buffered saline (PBS), sections were treated with fluorophore-conjugated secondary antibodies for 60 min at room temperature. Fluorescence imaging was conducted using a ZEISS LSM 880 confocal microscope (Jena, Germany) with the appropriate filter settings. To minimize bias due to tissue heterogeneity, at least five randomly selected, non‑overlapping fields per section were captured at ×200 magnification. Areas with necrosis, hemorrhage, or processing artifacts were excluded. Quantification was performed independently by two experienced pathologists blinded to the experimental groups, and the average values were used for statistical analysis.

CCL5 measurements

Tumor cells (5 × 10⁵ cells) were cultured in 6-well plates for 48 h. Culture supernatants were collected and passed through 0.45-mm filters. For subcutaneous tumors, tumor interstitial fluid (TIF) was isolated from tumor specimens using the following protocol: tumor tissues (200 mg) were first minced into fragments (approximately 1–2 mm³) and rinsed with ice-cold PBS. The tissue fragments were then suspended in 1 mL of PBS and incubated at 37 °C. Subsequently, the suspension was subjected to sequential centrifugation at 400 × g for 3 min to remove tissue debris, followed by a second centrifugation at 2,000 × g for 20 min at 4 °C to eliminate the remaining particulates. The resulting clear supernatant, representing the TIF fraction, was aliquoted and stored at -80 °C until subsequent analysis. CCL5 levels were determined using human and mouse CCL5 ELISA kits (E-EL-H6179, E-EL-M0009; ElabSciences). All procedures were conducted according to standardized protocols provided by the manufacturer.

Western blot analysis

Cellular lysis was performed on ice using RIPA buffer (Servicebio) containing both protease and phosphatase inhibitors (Servicebio) to extract the total protein. The protein concentration was determined using the Pierce Protein Assay kit (Thermo Fisher). Western blot assay was performed as previously reported [23]. Primary antibodies used were anti-NSUN2 (20854-1-AP, Proteintech), anti-actin (66009-1-Ig, Proteintech), anti-HDAC8 (A8865, ABclonal), anti-H3K27ac antibody (#8173, CST), anti-YBX1(A3534, ABclonal), and Anti-Histone H3 (# 9715, CST).

Flow cytometry analysis

Flow cytometry was performed using LSRII (BD Biosciences), and the data were processed using FlowJo software. For surface marker staining, cells were resuspended in PBS supplemented with 2% FBS and 2 mM EDTA and incubated with the following antibodies for 30 min on ice: Viability Dye (65-0865-14, eBioscience, 1:1000), anti-CD8 (100753, Biolegend, 1:300; 563898, BD Pharmingen, 1:300), anti-CD11b (101243, Biolegend, 1:200), anti-Gr-1 (102008, Biolegend, 1:300), anti-CD45.2 (109814, Biolegend, 1:300; 109832, Biolegend, 1:300), and anti-CCR5 (46-1951-80, Thermo Fisher, 1:100). After staining, the cells were washed twice with ice-cold PBS before acquisition. For intracellular staining, cells were fixed and permeabilized prior to antibody labeling, including anti-IFNγ (17-7311-82, eBioscience, 1:200) and anti-Granzyme B (372206, BioLegend, 1:50). To assess cytokine production, cells were stimulated for 4–6 h with phorbol 12-myristate 13-acetate ( P1585, Sigma), ionomycin (I3909, Sigma), and Golgi Stop (554724, BD Biosciences) before fixation, permeabilization, and staining with cytokine-specific antibodies.

CD8 isolation and in vitro migration assay

Mouse CD8⁺ T lymphocytes were isolated from the spleen or MLNs using the EasySep Mouse CD8⁺ T Cell Isolation Kit (#19853; STEMCELL Technologies). The isolated cells were subsequently stimulated for 48 h with anti-CD3 and anti-CD28 monoclonal antibodies (BioLegend, #100302 and #102102, respectively) in the presence of recombinant human IL-2 (50 IU/ml). Human CD8⁺ T lymphocytes were isolated from PBMCs of healthy donors using the EasySep Human CD8⁺ T Cell Isolation Kit (#19053; STEMCELL Technologies). The isolated cells were subsequently stimulated for 48 h with anti-CD3 and anti-CD28 monoclonal antibodies (BioLegend, #317302 and #302914, respectively) in the presence of recombinant human IL-2 (50 IU/mL).

Cell culture supernatants from murine and human tumor cells were collected and centrifuged to remove dead cells. Prior to the assay, Transwell inserts (5.0-µm pore size, Sarstedt) were coated with 0.8 mg/mL Matrigel for 2 h at 37 °C and washed twice with pre-warmed medium. A total of 1 × 10⁵ CFSE-labeled CD8 + T cells suspended in 100 µL low-serum medium (2% FBS) were seeded into the upper chamber. The lower chamber was filled with 500 µL of cell culture supernatant. After a 4-hour migration period, the number of migrated cells from the lower compartments was harvested and quantified using a hemocytometer or flow cytometry. The percentage of migrated cells was determined using the following formula: percentage of migrated cells = (number of migrated cells) / (number of input cells).

RNA immunoprecipitation

To assess RNA-protein interactions, RIP assays were implemented with the Imprint^® RNA Immunoprecipitation Kit (Sigma-Aldrich, USA) under conditions specified by the manufacturer. Prior to lysis, cells were cultured to ~ 90% confluence, followed by cell disruption in complete RIP lysis buffer fortified with RNase Inhibitor and protease inhibitors. Then, 100-µL samples of the cleared lysates were incubated in RIP buffer alongside magnetic beads that had been conjugated with the antibodies of interest. For specificity verification, Normal Rabbit IgG (#2729, CST) was designated as the negative control, and parallel immunoprecipitations were performed using anti-YBX1 (A3534, ABclonal), anti-ALYREF (ab202894, Abcam), and anti-YTHDF2 (A25318, ABclonal).

m⁵C methylated RNA immunoprecipitation sequencing (MeRIP-seq) and bulk RNA-seq analysis

High-throughput RNA methylation profiling of control and shNSUN2 LLC cells was conducted by Cloudseq Biotech Inc. (Shanghai, China) using MeRIP-seq. Briefly, RNA samples were subjected to immunoprecipitation using anti-m⁵C antibodies, followed by preparation of both input RNAs and enriched m⁵C-modified RNAs libraries. The RNA libraries were constructed using the TruSeq Stranded Total RNA Library Prep Kit (Illumina, USA) following the manufacturer’s protocol. Sequencing was carried out on an Illumina NovaSeq platform with 150-bp paired-end reads.

For bulk RNA‑seq, total RNA was extracted from control and shNsun2 LLC cells. Three independent biological replicates were included per condition, with each replicate derived from a separate culture well. No pooling was performed; each replicate was processed individually for library preparation and sequencing. mRNA sequencing was performed on the XuranGene platform (Shanghai, China). Differentially expressed genes (DEGs) were identified with a |log₂ fold change| ≥1 (corresponding to a 2-fold change) and a statistical significance threshold of P < 0.05.

Single-cell dissociation

Following surgical resection, the tissue specimens were immediately preserved in MACS Tissue Storage Solution (Miltenyi Biotec) to maintain cellular integrity prior to processing. The dissociation protocol commenced with thorough PBS washes, after which the tissues were mechanically minced into 1 mm³ fragments on ice. Enzymatic digestion was performed using a combination of 250 U/mL collagenase I, 100 U/mL collagenase IV ( Worthington), and 30 U/mL DNase I (Worthington) at 37 °C with continuous agitation for 30 min. The resulting cell suspension was sequentially filtered through a 40 μm mesh and pelleted by centrifugation at 300 × g for 5 min. Erythrocyte contamination was addressed by treating the pellet with Miltenyi Biotec’s specialized lysis buffer. Subsequent washes employed PBS containing 0.5% BSA, followed by a final filtration step through a 35 μm strainer to ensure single-cell suspension quality. Cell viability was confirmed by dual staining with calcein-AM (Thermo Fisher Scientific) to identify live cells and Draq7 (BD Biosciences) for dead cell exclusion.

Single-cell RNA sequencing and statistical analysis

Transcriptomic profiling of single cells was conducted using the BD Rhapsody platform, which employs a microwell-based capture system. In this approach, a single-cell suspension is distributed at a limiting dilution across more than 200,000 microwells to ensure single-cell occupancy. Oligonucleotide-barcoded magnetic beads were introduced in excess to guarantee that each microwell contained a bead-cell pair. Following cellular lysis within the microwells, polyadenylated mRNA molecules were captured by hybridization to barcoded oligos on the beads. The beads were then pooled for downstream processing, including reverse transcription, to generate cDNA molecules tagged with unique molecular identifiers (UMIs) and cell-specific barcodes at their 5’ ends, corresponding to the 3’ ends of the original transcripts. Whole transcriptome amplification was performed according to the manufacturer’s WTA protocol, involving random priming and extension, followed by amplification and index PCR for library preparation. Library quality control was performed using capillary electrophoresis (Agilent High Sensitivity DNA chip on Bioanalyzer 2200) and fluorometric quantification (Qubit High-Sensitivity DNA assay). Final library sequencing was carried out on the DNBSEQ-T7 platform (MGI, China) using a 150 bp paired-end configuration to ensure comprehensive transcript coverage. scRNA-seq data analysis was performed by NovelBio Co.,Ltd. with the NovelBrain Cloud Analysis Platform. We applied fastp with the default parameter filtering of the adaptor sequence and removed low-quality reads to achieve clean data. To quantify the gene expression of the single-cell data, we used STARsolo (version 2.7.11b) along with the mouse genome mm10 (Ensembl annotation version 100).

Analysis of public datasets

To characterize the expression profile of NSUN2 in the context of immunotherapy for LUAD, we first downloaded the public scRNA-seq dataset GSE207422 [24]. This dataset was generated from patients with LUAD treated with ICIs and comprised 92,330 cells. A standard pre-processing workflow was performed using the Seurat package (v4.3.0). The cell types were subsequently annotated based on canonical marker genes, resulting in the identification of eight major cell populations: epithelial cells, T/NK cells, myeloid cells, mast cells, B cells, plasma cells, fibroblasts, and endothelial cells. The epithelial cell subset was isolated for downstream analysis, wherein the expression of NSUN2 and eight other RNA methylation regulators was compared between patients with partial response (PR) and stable disease (SD).

To validate the potential association between NSUN2 and response to ICB, we analyzed bulk RNA-seq data from two additional ICB-treated cohorts: GSE126044 (NSCLC) and GSE91061 (melanoma) [25, 26]. Patients were stratified into responder (R) and non-responder (NR) groups according to the therapeutic response criteria from original publications. The Wilcoxon rank-sum test was used to assess the differential expression of NSUN2 between the two groups.

Furthermore, we downloaded transcriptome data from The Cancer Genome Atlas Lung Adenocarcinoma (TCGA-LUAD) cohort. Immune cell infiltration scores for each patient were quantified using the single-sample Gene Set Enrichment Analysis (ssGSEA) algorithm based on signature gene sets for 28 immune cell types. We analyzed the correlation between NSUN2 expression levels and the infiltration abundance of various immune cells (e.g., CD8⁺ T cells, dendritic cells, and M1/M2 macrophages) to investigate the potential functional role of NSUN2 in immune modulation.

Quantification and statistical analysis

All statistical computations were performed using the R language (v4.1.0, Missouri, USA) or GraphPad Prism (v8.0, CA, USA). A p-value of less than 0.05 was set to indicate statistical significance. A two-tailed Student’s t-test was used for comparisons between the two groups. In cases where the data did not follow a normal distribution, the non-parametric Wilcoxon rank-sum test was employed. Associations between NSUN2 expression and clinicopathological features were evaluated using the chi-square (χ²) test or Fisher’s exact test, as appropriate. To analyze survival data, we generated Kaplan-Meier curves and used the log-rank test for comparison, and a Cox proportional hazards regression model was fitted to identify independent prognostic factors. The Spearman’s rank correlation coefficient was calculated to assess the linear relationship between the two variables. Unless otherwise specified in the figure legends, all quantitative data are presented as mean ± standard error of the mean (SEM).

Results

NSUN2 is highly expressed in LUAD and is associated with poor prognosis

To systematically identify RNA methylation regulators that shape the immune microenvironment and influence immunotherapy outcomes in lung adenocarcinoma, we adopted a multi-stage screening strategy. First, to prioritize candidates with direct clinical relevance to ICB, we analyzed publicly available single-cell RNA-seq data (GSE207422) from NSCLC patients who had received anti-PD-1 therapy (Fig. 1 A and B) [24]. By comparing the transcriptomes of malignant cells between patients with stable disease (SD) and those with partial response (PR), we identified 16 RNA methylation regulatory genes that were significantly upregulated in the SD group (P < 0.05) (Fig. 1C). We then cross-referenced this ICB-associated gene set with our integrative proteomic analysis of two independent LUAD cohorts (cohort 1, n = 110; cohort 2, n = 103), which identified proteins overexpressed in tumors versus adjacent normal tissues [27, 28]. This cross-validation yielded three overlapping candidates (NSUN2, YTHDC1, ALKBH5) that were both upregulated in LUAD tumors and associated with ICB resistance (Fig. 1D). Among these, NSUN2 showed the most significant difference, with malignant cells from SD patients expressing 2.4-fold the levels of those from PR patients (Fig. 1, E to G, and fig. S1, A and B), prompting us to select it for further investigation. To further validate the prognostic value of NSUN2 in independent cohorts receiving immunotherapy, analysis of two additional public bulk RNA-seq datasets from ICB-treated cohorts showed that NSUN2 expression was consistently higher in non-responders than in responders (fig. S2A) [24, 25].

Fig. 1.

NSUN2 is highly expressed in LUAD and correlates with poor prognosis. A UMAP plot of all cells from the GSE207422 scRNA-seq dataset, color-coded according to annotated cell types. B UMAP plot of malignant cells, color-coded by patient response group (PR vs. SD) from the GSE207422 scRNA-seq dataset. C Ranking of RNA methylation regulators according to their association with immunotherapy response (PR vs. SD) in NSCLC, derived from the GSE207422 scRNA-seq dataset. NSUN2 was highlighted as on of the top candidates. D Venn diagram showing the intersection of 16 immunotherapy-non-response-associated RNA methylation regulators (from GSE207422 scRNA-seq) with proteins upregulated in LUAD tumors across two independent proteomic cohorts. NSUN2 was highlighted the top candidate. E Expression of NSUN2 projected onto the UMAP of malignant cells from the GSE207422 scRNA-seq dataset. F. Violin plots comparing the expression levels of NSUN2 in malignant cells of patients with PR and SD from the GSE207422 scRNA-seq data. G. The bar plot illustrates the proportion of patients with PR or SD stratified by NSUN2 expression levels (high vs. low) derived from the GSE207422 scRNA-seq data. H. Analysis of TCGA-LUAD cohort demonstrated that NSUN2 expression was significantly upregulated in tumor tissues compared to their matched adjacent non-tumorous tissues. I. Kaplan-Meier analysis of the TCGA-LUAD cohort indicated that high NSUN2 expression was associated with significantly poorer overall survival. J. A meta-analysis visualized by a forest plot across four independent NSCLC datasets confirmed that high NSUN2 expression is a robust unfavorable prognostic factor. K. Representative IHC images of NSUN2 staining in a tissue microarray containing 52 paired LUAD and adjacent nontumorous tissues. The lower panels illustrate examples of low, medium, and high NSUN2 expression in the LUAD cores. L. Quantification of NSUN2 expression from the TMA cohort confirmed its significant upregulation in LUAD tissues compared with paired paracancerous tissues. M. Stratification of NSUN2 expression across LUAD and normal lung tissues. N. Kaplan-Meier survival analysis of the TMA cohort demonstrated that high NSUN2 expression was significantly associated with poorer overall survival. Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

To clarify the source of NSUN2 in lung cancer, several NSCLC and LUAD scRNA-seq datasets confirmed that the overexpression of NSUN2 was predominantly confined to tumor cells, rather than other cell types within the LUAD microenvironment (fig. S2B) [29–33]. Building on these observations, we confirmed the consistent upregulation of NSUN2 in large-scale public repositories. Both TCGA and Clinical Proteomic Tumor Analysis Consortium (CPTAC) datasets revealed a significant increase in NSUN2 at both the transcriptional and protein levels in LUAD specimens when compared to normal controls (Fig. 1H and fig. S2C). Importantly, this upregulation carried adverse prognostic implications, as high NSUN2 expression demonstrated a strong correlation with shorter overall survival in the TCGA-LUAD cohort and across multiple lung cancer cohorts (Fig. 1 I and J) [34–38].

To independently validate these bioinformatic findings at the protein level within our clinical samples, we performed IHC staining for NSUN2 in a cohort of 52 pairs of surgically resected primary LUAD and adjacent non-tumorous tissues (Fig. 1K). While the intensity of NSUN2 staining exhibited a degree of inter-patient heterogeneity, a clear pattern of upregulation was observed in LUAD tissues compared to their corresponding paracancerous counterparts (Fig. 1, L and M). To evaluate the prognostic significance of NSUN2, the LUAD patient cohort was stratified into NSUN2-high and NSUN2-low expression groups based on IHC scores, with a cut-off value of 5.5 (scores > 5.5 defined as high, and ≤ 5.5 as low). Subsequent Kaplan-Meier survival analysis revealed that patients with high NSUN2 expression exhibited significantly poorer overall survival than those with low expression (Fig. 1N). The detailed clinicopathological characteristics of this patient cohort and their statistical correlation with the NSUN2 expression status are summarized in Table 3.

Table 3.

Clinicopathological characteristics of IHC cohort

Characteristics		NSUN2 expression
	Total (n = 52)	Low (n = 23)	High (n = 29)	P value
Gender
Male	16	7 (30.4%)	9 (31.0%)	0.963
Female	36	16 (69.6%)	20 (69.0%)
Age at surgery (years)
≥60	25	10 (43.5%)	15 (51.7%)	0.555
<60	27	13 (56.5%)	14 (48.3%)
Clinical disease stage
I-II	27	12 (52.2%)	15 (51.7%)	0.974
III-IV	25	11 (47.8%)	14 (48.3%)
TNM stage: T
1–2	44	19 (82.6%)	25 (86.2%)	0.721
3–4	8	4 (17.4%)	4 (13.8%)
TNM stage: N
N0	27	10 (43.5%)	17 (58.6%)	0.278
N1-2	25	13 (56.5%)	12 (41.4%)
TNM stage: M
M0	46	22 (95.7%)	24 (82.8%)	0.148
M1	6	1 (4.3%)	5 (17.2%)
Grade
Low (G1-2)	20	7 (30.4%)	13 (44.8%)	0.289
High (G3-4)	32	16 (69.6%)	16 (55.2%)

The convergence of these findings from proteomic, transcriptomic, and clinical outcome datasets establishes NSUN2 as a critical mediator of LUAD pathogenesis and highlights its potential utility as a candidate predictive biomarker for immunotherapy resistance, thus warranting further mechanistic investigation into its role in modulating anti-tumor immunity.

NSUN2 promotes LUAD progression in an immunity-dependent manner

Although the RNA methyltransferase NSUN2 has been implicated in the pathogenesis of various malignancies, its specific role in shaping the TME in LUAD remains poorly understood. To investigate this, we first determined the functional impact of Nsun2 loss on LUAD cell proliferation both in vitro and in vivo. We established two murine Nsun2-deficient cell lines: LLC cells with stable Nsun2 knockdown mediated by lentiviral shRNA and KP cells with complete Nsun2 knockout achieved through CRISPR-Cas9 gene editing (Fig. 2A). In vitro CCK-8 assays revealed that Nsun2-deficient LLC and KP cells had significantly reduced viability (Fig. 2B). A similar effect was observed upon the pharmacological inhibition of NSUN2 using MY-1B, a small-molecule inhibitor of NSUN2 (Fig. 2B) [39, 40]. The above observation was further validated by colony formation assays, which demonstrated a marked impairment in their long-term clonogenic potential (fig. S3, A and B). Consistently, NSUN2 knockdown in human A549 and H1975 cells also significantly impaired cell viability and colony formation (fig. S3, C and D). These in vitro data indicate that NSUN2 supports cell-intrinsic proliferation in LUAD cells. In contrast to the in vitro findings, when these cells were transplanted subcutaneously into immunocompetent C57BL/6 hosts, both Nsun2 knockdown and knockout led to profound suppression of tumor growth in vivo (Fig. 2, C and D). However, a different result emerged when we performed the same tumor implantation experiment in Rag1^−/− immunodeficient mice. Interestingly, the potent tumor growth inhibition observed in immunocompetent mice was largely abrogated in the Rag1^−/− background (Fig. 2, E and F). These findings indicate that the in vivo pro-tumorigenic function of NSUN2 is primarily immune-dependent rather than cell-autonomous.

Fig. 2.

NSUN2 promotes LUAD progression in an immunity-dependent manner. A Immunoblot analysis confirmed efficient knockdown of Nsun2 in LLC cells and its complete knockout in KP cells. B CCK-8 analysis of the proliferation of LLC and KP cells following Nsun2 depletion or pharmacological inhibition of NSUN2 with MY-1B (20µM) (n = 3 per group). C Tumor growth curve (left), final tumor weight (right upper), and representative tumor images (right lower) for control versus shNsun2 LLC tumors in immunocompetent C57BL/6 mice (n = 5 per group). D Tumor growth curve (left), final tumor weight (right upper), and representative tumor images (right lower) for control versus Nsun2 KO KP tumors in immunocompetent C57BL/6 mice (control, n = 5; NSUN2 KO, n = 6). E Tumor growth curve (left), final tumor weight (right upper), and representative tumor images (right lower) for control versus shNsun2 LLC tumors in immunodeficient Rag1^⁻/⁻ mice (n = 5 per group). F Tumor growth curve (left), final tumor weight (right upper), and representative tumor images (right lower) for control versus Nsun2 KO KP tumors in immunodeficient Rag1^-/- mice (n = 5 per group). G Schematic diagram illustrating the induction of LUAD in KrasLSL^−G12D/+, Sftpc-MerCreMer genetically engineered mice with or without conditional knockout of Nsun2. H Representative H&E‑stained lung images (left) and the corresponding tumor area annotations (right, in red) from mice with spontaneous LUAD (Nsun2^+/+ versus Nsun2^flox/flox). Scale bar, 200 μm. I Quantification of tumor burden from the spontaneous LUAD model in Nsun2^+/+ and Nsun2^flox/flox mice (n = 6 per group). J Kaplan-Meier survival analysis of Nsun2^+/+ versus Nsun2^flox/flox mice in the spontaneous LUAD model (n = 7 per group). Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

To further validate these findings in a more clinically relevant model that recapitulates spontaneous tumorigenesis, we engineered a conditional knockout mouse model: KrasLSL^(−G12D/+),Nsun2^flox/flox, and Sftpc-MerCreMer. This model permits the tamoxifen-inducible deletion of Nsun2 specifically within alveolar type II epithelial cells, the known cell of origin for this LUAD subtype (Fig. 2G and fig. S3, E and F). Consistent with our findings from the transplant models, Nsun2^flox/flox mice exhibited a significantly attenuated lung tumor burden when compared to Nsun2^+/+ controls (Fig. 2, H and I). This reduction in tumorigenesis translated directly into a significant increase of overall survival (Fig. 2J).

NSUN2 elicits an immunosuppressive microenvironment by inhibiting infiltration of CD8⁺ T cell

Building upon the above findings, we formulated the hypothesis that NSUN2 orchestrates an immunosuppressive tumor microenvironment, thereby driving a state of “immune desertification”. To test this hypothesis, we employed an integrated bulk and single-cell RNA sequencing approach in both subcutaneous and spontaneous LUAD models to comprehensively map the immunomodulatory consequences of NSUN2 deficiency.

Our initial transcriptomic characterization of bulk subcutaneous tumors provided evidence that Nsun2 knockdown triggered reprogramming of the TME, leading to a significant enhancement of gene signatures associated with T cell-mediated anti-tumor immunity (Fig. 3, A and B, and fig.S4A). This was particularly evidenced by a marked upregulation of key cytotoxic effector molecules (for example, GZMA, GZMB, and PRF1; fig. S4B). Analysis of public data from the TCGA-LUAD cohort also corroborated these findings (fig.S4, C, and D). To dissect the cellular drivers of this enhanced immune signature at a higher resolution, we performed a comprehensive scRNA-seq analysis on spontaneous tumor models. This approach not only substantiated our bulk sequencing data but also revealed the sheer magnitude of immune remodeling. We observed a notable 32% increase in total T/NK cell infiltration within Nsun2^flox/flox tumors compared with Nsun2^+/+ controls, indicating a fundamental shift in immune cell recruitment and retention within the tumor niche (Fig. 3C). Detailed examination of the immune subpopulations through unsupervised clustering analysis revealed a landscape-level shift. CD8⁺ T cells increased from 35.8% in Nsun2^+/+ tumors to 62.3% in Nsun2^flox/flox tumors (Fig. 3D). Concurrently, MDSCs showed a sharp decline in the myeloid compartment, decreasing from 64.7% to 35.3% (Fig. 3D). A closer look at CD8⁺ T cell expansion revealed a complex picture encompassing multiple functionally distinct subsets, including significant increases in effector and naïve populations (Fig. 3E). Notably, we also observed an increase in exhausted T cells, suggesting that Nsun2 ablation may promote both the recruitment and subsequent exhaustion of CD8⁺ T cells (Fig. 3E). Concurrently with the expansion of the CD8⁺ T-cell population, we observed a marked decline in other lymphocyte subsets such as regulatory T cells (Tregs) and Th2 cells (Fig. 3E). This concurrent shift indicates a profound rebalancing of the immune landscape toward a more potent anti-tumor phenotype.

Fig. 3.

NSUN2 shapes an immunosuppressive tumor microenvironment by impeding CD8⁺ T cell infiltration. A Volcano plot of bulk RNA-seq data displaying differentially expressed genes in shNsun2 versus control cells. B GSVA revealed alterations in immune-related pathways from the bulk RNA-seq data of shNsun2 versus control cells. C Quantification of the proportions of six major cell types in spontaneous LUAD from Nsun2^+/+ and Nsun2^flox/flox mice as determined by scRNA-seq analysis. D UMAP visualization of T cell and myeloid cell subclusters identified from scRNA-seq data of spontaneous LUAD from Nsun2^+/+ and Nsun2^flox/flox mice. E Quantification of T cell subpopulation proportions in Nsun2^+/+ versus Nsun2^flox/flox tumors based on scRNA-seq data. F Proportions of tumor-infiltrating CD8⁺ T cells within KP, LLC, and spontaneous LUAD tumors as determined by flow cytometry (n = 5/6 for KP; n = 5 for LLC; n = 6 for spontaneous LUAD). G Proportions of tumor-infiltrating IFNγ⁺CD8⁺T cells within KP, LLC, and spontaneous LUAD tumors as determined by flow cytometry (n = 5 for KP; n = 5 for LLC; n = 6 for spontaneous LUAD). H Proportions of tumor-infiltrating GZMB⁺CD8⁺T cells within KP, LLC, and spontaneous LUAD tumors as determined by flow cytometry (n = 5 for KP; n = 5 for LLC; n = 6 for spontaneous LUAD). I Proportions of tumor-infiltrating MDSCs within KP, LLC, and spontaneous LUAD tumors as determined by flow cytometry (n = 5/6 for KP; n = 5 for LLC; n = 6 for spontaneous LUAD). J Representative flow cytometry plots showing the proportion of tumor-infiltrating CD8⁺ T cells, GZMB⁺CD8⁺ T cells, IFN-γ⁺CD8⁺ T cells, and MDSCs in spontaneous LUAD tumors. K Representative IF images of CD8⁺ T cells in control and Nsun2-KO tumors. L Corresponding quantification of CD8⁺ T cell abundance from IF analysis. Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

To confirm that these transcriptomic shifts translated to corresponding changes at the protein and cellular levels, we conducted validation using multiparameter flow cytometry in both subcutaneous and spontaneous LUAD models. Consistent with our sequencing data, Nsun2-deficient tumors exhibited a significant increase in the frequency of tumor-infiltrating CD8⁺ T-cells (Fig. 3F). Crucially, this expansion was not merely quantitative; we observed an enhancement of functionally active subsets, including IFN-γ⁺ and GZMB⁺ CD8⁺ T cells (Fig. 3G and H, respectively), confirming a heightened state of cytotoxic potential within the tumor following Nsun2 depletion. Interestingly, the elevation of CD8⁺ T cells extended systemically, as they were also elevated in the spleens of mice bearing Nsun2-deficient subcutaneous tumors (fig. S6A). Despite the enhanced effector function of CD8⁺T cells, analysis of exhaustion markers revealed that, while the overall PD-1⁺ CD8⁺ T cell frequency remained unaltered, the subset of terminally exhausted PD-1⁺TIM-3⁺ CD8⁺ T cells was significantly increased in Nsun2 KO and Nsun2^flox/flox tumors (fig. S6, B and C). In parallel with reinvigoration of the lymphoid compartment, we investigated its impact on immunosuppressive myeloid populations. MDSC populations showed a dramatic reduction in both the subcutaneous and spontaneous LUAD models (Fig. 3I and fig. S6, D, and E), suggesting that the immunomodulatory role of Nsun2 extends beyond the lymphoid lineage to actively regulate accumulation of the suppressive myeloid compartment. Representative flow cytometry plots of the tumor-infiltrating immune cells are shown in Fig. 3J. Finally, to provide a spatial context for these cellular changes, we employed complementary immunofluorescence analysis. These imaging studies visually confirmed our flow cytometry and sequencing findings, demonstrating a significantly increased infiltration of CD8⁺ T cells throughout the tumor parenchyma in Nsun2-deficient tumors (Fig. 3, K and L). In conclusion, ablating Nsun2 shifts the LUAD TME into a T cell-inflamed milieu by boosting CD8⁺ T cell function and curbing immunosuppression, thereby enabling effective anti-tumor immunity.

NSUN2 impairs CD8⁺ T cell recruitment through suppression of the CCL5/CCR5 axis

Building on our initial findings that established NSUN2 as a critical regulator of immune evasion in LUAD, we sought to elucidate the molecular mechanisms underlying its immunosuppressive function. To this end, we performed a subsequent pathway enrichment analysis of the bulk RNA-seq data, comparing the transcriptomes of Nsun2-deficient tumors with those of the control. The results revealed that pathways related to chemotaxis were the most significantly and robustly enriched biological processes in the absence of Nsun2 (Fig. 4A), with a particular emphasis on those governing immune cell migration. This primary observation provides a direct mechanistic link to our phenotypic findings, aligning with emerging evidence that implicates chemokine networks as pivotal players in shaping the immune landscape of tumors responsive to therapy [41, 42]. Crucially, the specific epigenetic regulators that control these critical chemokine networks have remained largely uncharacterized. Upon interrogating the differentially expressed genes within these chemotaxis pathways, chemokine Ccl5 emerged as a significantly upregulated factor following Nsun2 depletion, exhibiting a 1.2-fold increase in expression (Fig. 4B).

Fig. 4.

NSUN2 suppresses CD8⁺ T cell recruitment via the CCL5/CCR5 axis. A Gene Ontology (GO) analysis of differentially expressed genes in shNsun2 versus control LLC tumors based on bulk RNA-seq. B Heatmap depicting mRNA expression of chemokine-related genes in control versus shNsun2 tumors determined by bulk RNA-seq. C UMAP plot illustrating Ccl5 mRNA expression across malignant cells of spontaneous LUAD derived from single-cell RNA-seq data. D Violin plot comparing Ccl5 mRNA expression in malignant cells from Nsun2^+/+ versus Nsun2^flox/flox spontaneous LUAD tumors based on single-cell RNA-seq data. E RT-qPCR analysis of Ccl5 mRNA expression in control and Nsun2-depleted LLC and KP cells. F ELISA quantification of secreted CCL5 in the supernatant of control and Nsun2-depleted LLC and KP cells. G ELISA quantification of CCL5 in tumor interstitial fluid (TIF) from LLC and KP tumors. H Schematic diagram of the in vitro CD8⁺ T cell transwell migration assay. I Transwell assay showing migration of splenic or MLN-derived CD8⁺T cells toward conditioned media from LLC or KP cells. Data are representative of three independent experiments. J Representative flow cytometry plots showing CCR5⁺CD8⁺ T cells in subcutaneous and spontaneous LUAD tumors. K Proportions of CCR5⁺CD8⁺ T cells in subcutaneous tumors and spontaneous LUAD by flow cytometry. Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

Consequently, we prioritized CCL5 for further investigation. This selection was strategically based on its extensively documented role as a primary mediator of anti-tumor immunity across a diverse array of cancer types [8, 43, 44]. A deficiency in CCL5 signaling has been consistently associated with the formation of immunologically “cold” tumors [45], which are characterized by lymphocyte exclusion and a correspondingly poor response to immunotherapy.

To validate our bulk RNA-seq findings and to pinpoint the specific cellular source of this chemokine within a more physiologically relevant context, we analyzed the scRNA-seq data from the spontaneous LUAD tumor model. This high-resolution analysis revealed that malignant epithelial cells, rather than immune or stromal populations, serve as the predominant source of Ccl5 transcripts within the complex tumor microenvironment. UMAP visualization and subsequent quantification demonstrated that Nsun2-knockout tumor cells exhibited 3.7-fold higher Ccl5 expression than their wild-type counterparts (Fig. 4, C and D).

This key finding firmly established that tumor-intrinsic NSUN2 acts as a negative regulator of CCL5 production. We conducted a series of in vitro experiments to corroborate this cell-intrinsic regulatory axis. We observed highly concordant and significant increases in both Ccl5 transcript levels and the quantity of secreted Ccl5 protein following genetic ablation of Nsun2 in LLC and KP cells (Fig. 4, E and F). Consistently, NSUN2 knockdown in human A549 and H1975 cells also led to significant upregulation of CCL5 secretion (fig. S5A). Finally, to confirm the physiological relevance of these findings in vivo, we quantified Ccl5 protein levels in the tumor interstitial fluid (TIF). ELISA measurements revealed that Nsun2-deficient subcutaneous tumors contained significantly higher concentrations of Ccl5 than controls (Fig. 4G).

To confirm that the elevated secretion of CCL5 was directly responsible for driving CD8⁺ T cell influx, we performed Transwell migration assays (Fig. 4H). As hypothesized, supernatants from Nsun2-deficient tumor cells triggered a marked migratory response, recruiting a significantly greater number of primary CD8⁺ T cells than those exposed to the control media (Fig. 4I). Similarly, conditioned media from NSUN2-knockdown A549 and H1975 cells also exhibited enhanced chemotactic activity toward human CD8⁺ T cells (fig. S5B). We next sought to determine whether this mechanism translated into a specific impact on CD8⁺T cell populations within the complex in vivo tumor microenvironment. Since CCL5 exerts its chemotactic function primarily through its cognate receptor CCR5, we specifically investigated the CCR5⁺ CD8⁺ T cell subset, which represents the direct cellular target of this signaling axis. Flow cytometric analysis revealed a consistent expansion of this population following Nsun2 depletion (Fig. 4J). We observed a significant increase in CCR5⁺ CD8⁺ T cells in the subcutaneous tumor model, and a comparable increase in the spontaneous LUAD model (Fig. 4K).

Collectively, our study delineates an epigenetic checkpoint where tumor-intrinsic NSUN2 suppresses the CCL5-CCR5 signaling axis, thereby limiting the infiltration of effector CD8⁺ T cells and maintaining an immunologically “cold” tumor microenvironment.

NSUN2 suppresses CCL5 expression through m⁵C modification of HDAC8 mRNA

To dissect the molecular cascade linking NSUN2 to CCL5 expression, we delved into the epitranscriptomic landscape by performing comprehensive m⁵C-MeRIP-seq in LLC cells following Nsun2 knockdown versus control (Fig. 5A). However, we found no change in the m⁵C status of the Ccl5 transcripts upon Nsun2 depletion. This finding prompted us to hypothesize that Nsun2 governs Ccl5 expression through an indirect intermediary pathway. Accordingly, our subsequent bioinformatics analysis of differentially methylated transcripts provided a critical clue: we observed a highly significant enrichment for genes integral to histone modification pathways (Fig. 5B), strongly suggesting this as the missing link. A closer examination revealed a specific cohort of 12 key histone modification regulators, whose transcripts underwent significant m⁵C hypomethylation upon Nsun2 depletion (Fig. 5, C and D). This finding provides a crucial mechanistic link, as these regulators are enzymes pivotal to orchestrating the chromatin landscape, thereby controlling gene expression [46, 47]. Histone acetyltransferases (HATs) foster a transcriptionally permissive state by neutralizing histone charges, whereas histone deacetylases (HDACs) induce chromatin condensation and transcriptional repression [48]. The regulatory role of histone methylation is more complex, with the antagonistic activities of methyltransferases (HMTs) and demethylases (KDMs) dynamically shaping gene accessibility [49].

Fig. 5.

NSUN2 mediates m⁵C modification of HDAC8 mRNA and thereby inhibits CCL5 expression. A Schematic of the m⁵C MeRIP-seq experimental workflow B Top 10 enriched Gene Ontology (GO) terms for genes with reduced m⁵C methylation upon Nsun2 knockdown. C Venn diagram showing the overlap between histone-modifying enzymes (66 genes) and potential NSUN2 target genes (1750 genes with reduced m⁵C methylation). D Analysis of m⁵C methylation levels in the overlapping genes identified from the Venn diagram. E Effect of siRNA-mediated knockdown of the candidate NSUN2 target genes on Ccl5 mRNA expression. F m⁵C MeRIP-seq reads along Hdac8 mRNA. Ranges of reads are indicated. G Immunoblot analysis of HDAC8 protein levels following Nsun2 depletion. H RT-qPCR analysis of Hdac8 mRNA levels in LLC and KP cells after Nsun2 depletion. I The translation ratio of Hdac8 mRNA in ribosome-nascent chain (RNC) complex was analyzed using RT-qPCR. J Left: Schematic of the Renilla luciferase reporters. The Hdac8 fragment containing 5‘UTR and adjacent CDS was inserted downstream of Renilla luciferase. Mutations were introduced at m⁵C sites in the 5’UTR (C32T, C71T) and CDS (C118A, C139A; silent mutations). Right: Relative luciferase activity (Renilla/Firefly ratio) in KP cells co-transfected with the indicated reporters and either Nsun2 overexpression or empty vector. K RIP-qPCR analysis of Hdac8 mRNA enrichment with anti-YBX1 antibody versus IgG control in control and Nsun2-KO KP cells. L RT-qPCR analysis of Hdac8 and Ccl5 mRNA levels following Ybx1 knockdown in KP cells. M Immunoblot analysis of YBX1 and HDAC8 protein levels following Ybx1 knockdown in KP cells. N RT-qPCR analysis of Hdac8 and Ccl5 mRNA levels following Ybx1 knockdown in KP cells. O RT-qPCR analysis of Ccl5 mRNA expression in KP cells treated with or without the HDAC8 inhibitor PCI-34,051. P ELISA of secreted CCL5 protein in the culture supernatants of KP cells treated with or without the HDAC8 inhibitor PCI-34,051. Q Immunoblot analysis of HDAC8 and H3K27ac in KP cells treated with or without HDAC8 inhibitor PCI-34,051. β-actin and total H3 were used as loading controls. R ChIP-qPCR analysis of H3K27ac occupancy at the Ccl5 promoter in cells treated with or without the HDAC8 inhibitor PCI-34,051. S Immunoblot analysis of NSUN2 and HDAC8 levels in control and NSUN2-overexpressing KP cells treated with or without the HDAC8 inhibitor PCI-34,051. T ELISA of secreted CCL5 protein in the culture supernatant of control or Nsun2-overexpressing KP cells treated with or without the HDAC8 inhibitor PCI-34,051. U CD8⁺ T cell migration assay using culture supernatant from control or Nsun2-overexpressing KP cells treated with or without the HDAC8 inhibitor PCI-34,051. Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

Crucially, RNA-seq results revealed that the transcript levels of all 12 identified enzymes remained unchanged after Nsun2 knockdown, suggesting that Nsun2’s influence might not be transcriptional but rather post-transcriptional. Given that Nsun2 loss upregulates Ccl5, we reasoned that the relevant intermediary must be a transcriptional repressor whose function is impaired. Therefore, we selected six identified enzymes known to function as repressors (Hdac8, Hdac6, Kdm5a, Sirt3, Sirt5, and Dot1l) for a targeted functional screen (Fig. 5D).

Individual knockdown using siRNAs yielded a clear outcome; only the depletion of Hdac8 robustly recapitulated the Ccl5 upregulation observed in Nsun2-deficient cells (a 1.5-fold increase, P < 0.001), with Sirt3 depletion showing a more modest effect (Fig. 5E). This functional validation, combined with the observation that Hdac8 mRNA exhibited the most marked loss of m⁵C methylation (a 1,440-fold reduction), solidified its status as the primary downstream mediator of Nsun2’s effect on Ccl5 (Fig. 5F).

Consistent with this, immunoblotting confirmed a marked reduction in HDAC8 protein levels following NSUN2 depletion (Fig. 5G and fig. S5C). This presented a key mechanistic puzzle: since Hdac8 transcript levels remained unchanged (Fig. 5H), we hypothesized that NSUN2 regulates HDAC8 at the translational level. To test this directly, we employed RNC-qPCR assay. This technique, which quantified the engagement of mRNA with active ribosomes, revealed that the translational efficiency of Hdac8 mRNA was significantly downregulated upon Nsun2 depletion (Fig. 5I), providing strong evidence that NSUN2-mediated m⁵C modification enhances HDAC8 translation.

To establish a causal link between NSUN2-mediated m⁵C modification and HDAC8 translational regulation, we performed a dual-luciferase reporter assay using psiCHECK-2 vectors carrying either wild-type or mutant Hdac8 fragments in which the primary m⁵C sites were simultaneously mutated (5’UTR C32T/C71T; CDS C118A/C139A silent mutations) (Fig. 5J). The results showed that Nsun2 overexpression significantly increased luciferase activity in WT reporter-transfected cells. In contrast, this Nsun2-driven enhancement was completely abolished in cells carrying the Mut reporter, with no significant difference between Mut+Vector and Mut+Nsun2 groups (Fig. 5J and fig. S5, D and E). These results provide direct evidence that the specific m⁵C motifs on Hdac8 mRNA are required for Nsun2-driven translational enhancement.

YBX1 recognizes m⁵C-modified HDAC8 mRNA to enhance its translation

To complete the mechanistic cascade from NSUN2-mediated m⁵C modification to HDAC8 translational enhancement, we next sought to identify the m⁵C reader protein responsible for recognizing the methylation mark on HDAC8 mRNA. Among the well-characterized m⁵C readers, YBX1 and ALYREF have been previously implicated in regulating mRNA stability and translation, whereas YTHDF2 is known to function as an m⁶A reader but has also been reported to interact with m⁵C-modified transcripts [10, 50]. We performed RNA immunoprecipitation assays followed by RT-qPCR to detect Hdac8 mRNA enrichment. Results showed that Hdac8 mRNA was robustly pulled down by anti-YBX1 in control cells, whereas this enrichment was diminished upon Nsun2 depletion (Fig. 5K). In contrast, neither anti-ALYREF nor anti-YTHDF2 exhibited significant enrichment of Hdac8 mRNA over the IgG control, regardless of Nsun2 status (fig. S5, F and G), indicating that YBX1, rather than ALYREF or YTHDF2, specifically recognizes the m⁵C-modified Hdac8 transcript.

To functionally validate this interaction, we knocked down Ybx1 in KP cells using Ybx1-targeting shRNA. Consistent with our hypothesis, Ybx1 knockdown resulted in a significant reduction of Hdac8 protein levels without affecting its mRNA abundance (Fig. 5, L and M), phenocopying the effect of Nsun2 loss. Importantly, this translational repression of Hdac8 upon Ybx1 knockdown led to a consequent upregulation of Ccl5 at both the mRNA and protein levels (Fig. 5, L and N), further confirming that the NSUN2-m⁵C-YBX1 axis operates upstream of the HDAC8-CCL5 cascade.

HDAC8 silences CCL5 expression through H3K27 deacetylation

HDAC8 is a well-defined enzyme that specifically catalyzes the removal of acetyl groups from histone H3 lysine 27 (H3K27ac), an epigenetic marker that is strongly associated with transcriptional repression [51]. Its clinical relevance has been underscored by several studies demonstrating that pharmacological inhibition enhances CD8⁺ T cell infiltration and improves immunotherapy responses in several malignancies [52, 53]. To directly verify if HDAC8 activity governs CCL5 expression, we treated LUAD cells with PCI-34,051, a highly selective HDAC8 inhibitor. This intervention resulted in significant upregulation of Ccl5 at both the transcript and protein levels, as evidenced by a more than 2-fold increase in its mRNA quantified by qPCR (Fig. 5O and fig. S5H) and a more than 90% increase in secreted protein quantified by ELISA (Fig. 5P and fig. S5I). To confirm the on-target epigenetic mechanism, we verified that the pharmacological inhibition of HDAC8 led to a global increase in H3K27 acetylation in KP cells (Fig. 5Q). More critically, to determine whether this regulation occurred directly at the Ccl5 locus, we performed ChIP-qPCR. This analysis revealed that HDAC8 inhibition markedly enhanced H3K27ac enrichment at the Ccl5 promoter, providing evidence that HDAC8 directly silenced its transcriptional activation (Fig. 5R).

The pivotal question then became whether the repressive effect of NSUN2 on CCL5 was mechanistically dependent on HDAC8. To address this, we conducted a rescue experiment (Fig. 5S). As expected, Nsun2-overexpressing KP cells exhibited a strong suppression of CCL5. Notably, HDAC8 inhibitor treatment effectively reversed this suppression (Fig. 5T), restoring both CCL5 expression and consequently the migration of CD8⁺ T cells (Fig. 5U).

Taken together, these findings delineate a novel regulatory axis in which NSUN2 acts upstream of HDAC8, leveraging m⁵C modification of its transcript to orchestrate the suppression of CCL5.

NSUN2 is a candidate biomarker associated with immune infiltration status and immunotherapy efficacy

To translate our mechanistic findings into a clinically relevant context, we first sought to delineate the significance of NSUN2 in the human LUAD tumor microenvironment. To this end, we performed multiplex immunohistochemistry (mIHC) to simultaneously assess NSUN2 expression and the density of tumor-infiltrating CD8⁺ T cells in a well-annotated cohort of 43 primary LUAD specimens (Fig. 6A). Stratification of the cohort based on NSUN2 expression revealed a stark dichotomy in the immune landscape. Tumors with high NSUN2 levels were consistently characterized by a markedly sparse infiltration of CD8⁺ T cells, whereas NSUN2-low tumors displayed a markedly denser population of these lymphocytes (Fig. 6B). This culminated in the establishment of a statistically significant inverse correlation between NSUN2 protein levels and CD8⁺ T cell density (Fig. 6C), providing clinical evidence that high NSUN2 expression is linked to an immune-excluded or “cold” tumor phenotype, thereby impairing anti-tumor immunity.

Fig. 6.

NSUN2 negatively stratifies immune infiltration and immunotherapy response in LUAD. A. Representative immunofluorescence images of NSUN2 expression and infiltrating CD8⁺ T cells in human LUAD tissue microarray(n = 43). B Comparison of CD8⁺ T-cell infiltration density between NSUN2-high and NSUN2-low human LUAD tissues. C Correlation analysis between NSUN2 expression levels and CD8⁺ T cell infiltration density in human LUAD tissues. D Representative immunofluorescence images showing NSUN2 expression, CD8⁺ T-cell infiltration, and MDSC (CD33⁺) infiltration in LUAD tissues from patients who received neoadjuvant immunotherapy. E Quantification of NSUN2 expression levels in LUAD tissues from patients receiving neoadjuvant immunotherapy stratified by pathological response (pCR, MPR, pPR, and pNR). F Quantification of CD8⁺ T-cell infiltration in LUAD tissues from patients receiving neoadjuvant immunotherapy stratified by pathological response. G Correlation between NSUN2 expression and CD8⁺ T cell infiltration percentage in patients who underwent neoadjuvant immunotherapy. H Quantification of MDSC (CD33⁺) infiltration in LUAD tissues from patients receiving neoadjuvant immunotherapy, stratified by pathological response. I. Correlation analysis between NSUN2 expression and MDSC (CD33⁺) infiltration percentage in patients underwent neoadjuvant immunotherapy. J Tumor growth curve (left), final tumor weight (middle) and representative tumor images (right) of KP allografts (control vs. Nsun2 KO) from mice treated with CD8⁺ T-cell depletion antibody (anti-CD8α), MDSC depletion antibody (anti-Ly6G), or IgG control (n = 5 per group). K Tumor growth curve (left), final tumor weight (middle), and representative tumor images (right) of KP allografts (control vs. Nsun2 KO) in FoxP3^DTR mice treated with DT or PBS (n = 5 per group). Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

Building on this correlative evidence, we next investigated whether the expression level of NSUN2 could serve as a candidate predictive biomarker for patient response to ICB. For this purpose, we analyzed a distinct cohort of post-surgical specimens from 18 patients with LUAD who had received neoadjuvant treatment with either anti-PD-1 or anti-PD-L1 antibodies (Fig. 6D, fig. S7A, and Table 4). These patients were stratified according to their treatment outcomes into pathological complete response (pCR; n = 6), major/partial pathological response (MPR/pPR; n = 6), and pathological non-response (pNR; n = 6) groups. The results showed that patients who achieved a favorable pathological response (pCR and MPR/pPR groups) consistently displayed significantly lower intratumoral NSUN2 expression (Fig. 6E). Concurrently, these responders showed significantly higher CD8⁺ T-cell infiltration than the pNR group (Fig. 6F). These findings also established a significant inverse correlation between NSUN2 levels and CD8⁺ T cell infiltration ( Fig. 6G). Importantly, these expression patterns for NSUN2 and CD8 held true when patients were grouped into pCR/MPR versus non-pCR/MPR (fig. S7, B and C). As a secondary observation, we noted that the expression of the human MDSC marker CD33 was significantly elevated in pCR patients relative to that in MPR/pPR patients (Fig. 5H and fig. S7D) and a positive, albeit not statistically significant, trend was observed between NSUN2 and MDSC infiltration (R = 0.1626; Fig. 7I), which suggests a complex interplay that warrants further investigation.

Table 4.

Clinicopathological characteristics of the LUAD neoadjuvant cohort

No.	Pathological response	Sex	Age		Neoadjuvant therapy	Neoadjuvant therapy cycles	Baseline T stage	Baseline N stage
1	pCR	Male	63	anti-PD-L1 + chemo		2	T1	N0
2	pNR	Male	62	anti-PD-L1 + chemo		3	T2	N0
3	pCR	Male	67	anti-PD1 + chemo		1	T2	N0
4	pPR	Male	75	anti-PD1 + chemo		2	T2	N0
5	pCR	Male	68	anti-PD1 + chemo		1	T2	N0
6	pCR	Male	74	anti-PD1 + chemo		2	T2	N1
7	pNR	Male	56	anti-PD1 + chemo		2	T2	N2
8	MPR	Male	61	anti-PD1 + chemo		1	T4	N2
9	MPR	Male	74	anti-PD1 + chemo		2	T1	N2
10	pNR	Male	69	anti-PD-L1 + chemo		3	T3	N2
11	pNR	Male	66	anti-PD1 + chemo		3	T3	N2
12	MPR	Male	70	anti-PD-L1 + chemo		2	T3	N0
13	pCR	Female	69	anti-PD1 + chemo		2	T3	N2
14	pPR	Female	75	anti-PD1 + chemo		2	T3	N2
15	pNR	Female	67	anti-PD1 + chemo		2	T1	N0
16	MPR	Male	60	anti-PD1 + chemo		2	T2	N2
17	pNR	Male	73	anti-PD1 + chemo		4	T2	N0
18	pCR	Male	71	anti-PD1 + chemo		2	T1	N0

Fig. 7.

Targeting NSUN2 enhances immunotherapy efficacy in LUAD. A Tumor growth curve (left), final tumor weight (middle), and representative tumor images (right) of KP allografts (control vs. Nsun2 KO) in mice treated with anti-PD-1 or IgG control antibody (n = 5 per group). B Tumor growth curve (left), final tumor weight (middle), and representative tumor images (right) of LLC allografts (control vs. shNsun2) in mice treated with anti-PD-1 or IgG control antibody (n = 5 per group). C Representative H&E‑stained lung images (left) and the corresponding tumor area annotations (right, in red) from mice with spontaneous LUAD (Nsun2^+/+ versus Nsun2^flox/flox) treated with anti-PD-1 or IgG control antibody. D Quantification of tumor burden in a spontaneous LUAD model (Nsun2^+/+ vs. Nsun2^flox/flox) following treatment with anti-PD-1 or IgG control antibody (n = 4 per group). Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

To establish a correlation between the pro-tumor effect of NSUN2 and CD8⁺ T cell infiltration, we conducted a key CD8⁺ T cell depletion experiment. The efficacy of CD8⁺ T cell depletion was validated using flow cytometry (Fig. 6I). Critically, we found that administering a CD8-depleting antibody to mice bearing Nsun2-knockout tumors significantly abrogated the anti-tumor benefits conferred by Nsun2 loss, leading to the rescue of tumor growth (Fig. 6J). To distinguish whether the observed alterations in MDSCs and Tregs represented active mediators or secondary consequences of Nsun2 deficiency, we performed sequential depletion studies. Depletion of MDSCs using anti-Ly6G antibody failed to reverse the loss of tumor control caused by CD8⁺ T cell depletion in mice bearing Nsun2 KO tumors (fig. S8, A to C), indicating that MDSCs are not primary mediators of the Nsun2-driven antitumor effect. Furthermore, we performed Treg ablation experiments using FoxP3^DTR mice. As shown in Fig. 6K, while DT-mediated Treg depletion significantly suppressed the growth of control tumors, confirming the functional activity of Tregs in this model, it failed to further suppress tumor progression in Nsun2-KO tumors (fig. S8D). These results indicate that Tregs are secondary consequences rather than active mediators of the NSUN2-driven antitumor effect.

Collectively, these convergent findings strongly support that elevated NSUN2 expression fosters a profoundly immunosuppressive tumor microenvironment in LUAD by suppressing CD8⁺ T cell-mediated immunity, which likely leads to resistance against ICB.

Targeting NSUN2 is a promising strategy to enhance ICIs efficacy in LUAD

Building on these findings, we sought to therapeutically validate our hypothesis that targeting NSUN2 could serve as a powerful strategy to sensitize LUAD to ICB. To rigorously test this, we first established subcutaneous allograft tumor models by injecting WT mice with either control or Nsun2-depleted LUAD cells, followed by treatment with anti-PD-1 antibodies. These results are encouraging. While either genetic depletion of Nsun2 or immunotherapy conferred modest therapeutic benefits, resulting in a reduction in tumor size and weight, the combination group resulted in a synergistic anti-tumor response (Fig. 7A and B). This was evidenced by a tumor remission rate of more than 80% in the combination group. To further corroborate these results in a model that more faithfully recapitulates human disease progression, we employed immunotherapy in the spontaneous LUAD mouse model. Consistent with the allograft data, the dual strategy of Nsun2 ablation combined with anti-PD-1 treatment markedly suppressed tumor progression and significantly diminished the overall tumor burden, providing direct validation of therapeutic synergy (Fig. 7 C and D).

Mechanistically underpinning this powerful synergy, immunophenotyping analyses revealed substantial remodeling of the tumor immune microenvironment. We observed a significant increase in tumor-infiltrating CD8⁺ T cells in Nsun2-deficient tumors from both the allograft and Nsun2^flox/flox spontaneous LUAD, particularly following immunotherapy (fig. S8, E and F). The proportion of MDSCs and exhausted CD8⁺ T cells was also analyzed (fig. S8, G to I).

NSUN2 inhibitor MY-1B delivered by iRGD-liposomes synergizes with anti-PD-1 to suppress tumor growth

To establish the causal requirement of CCL5 in mediating the antitumor effects of Nsun2 depletion, we neutralized CCL5 in control and Nsun2-KO subcutaneous allografts (Fig. 8A). Anti-CCL5 treatment significantly promoted tumor growth in control tumors, and more importantly, reversed the growth inhibition conferred by Nsun2 depletion, with tumor volumes in the Nsun2 KO+anti-CCL5 group becoming comparable to those in the control + IgG group (Fig. 8A). Flow cytometric analysis further revealed that the increased infiltration of CD8⁺ T cells in Nsun2-KO tumors was also abrogated upon CCL5 neutralization (fig. S8J), establishing that NSUN2 suppresses antitumor immunity by impairing CCL5-dependent CD8⁺ T-cell recruitment.

Fig. 8.

iRGD‑MY‑1B-liposomes synergize with anti‑PD‑1 to suppress tumor growth. A Tumor growth curve (left), final tumor weight (middle), and representative tumor images (right) of KP allografts (control vs. Nsun KO) in mice treated with IgG or anti-CCL5 neutralizing antibody (n = 5 per group). B Schematic representation of synthetic process of iRGD-MY-1B liposomes and drug treatments in the subcutaneous tumor models. C Tumor growth curve (left), final tumor weight (middle), and representative tumor images (right) of LLC allografts from mice treated with anti-PD-1, iRGD-MY-1B liposomes, iRGD-MY-1B liposomes plus anti-PD-1, or control (n = 5 per group). D Tumor growth curve (left), final tumor weight (middle), and representative tumor images (right) of KP allografts from mice treated with anti-PD-1, iRGD-MY-1B liposomes, iRGD-MY-1B liposomes plus anti-PD-1, or control (n = 4 per group). Data are presented as mean ± SEM. Statistical significance was determined by one-way ANOVA or Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001; ns, not significant

We then sought to determine whether pharmacological inhibition of NSUN2 could recapitulate the anti-tumor effects observed with genetic ablation (Fig. 8B). To achieve tumor-selective delivery and maximize therapeutic potency, we encapsulated the small-molecule NSUN2 inhibitor MY-1B in iRGD-liposomes, a PEGylated nanocarrier whose surface is post-inserted with the cyclic peptide iRGD (CRGDK/RGPD/EC). The peptide sequentially binds αvβ3/αvβ5 integrins displayed on tumor endothelium and is then proteolytically cleaved to expose the CendR motif (CRGDK), which activates neuropilin-1–dependent trans-tissue transport. iRGD-liposomes sequentially engage endothelial integrins and tumor-cell NRP-1, anchoring the particle to the neovasculature and then propelling the drug beyond the vascular wall into the tumor core [54, 55]. We systematically evaluated the safety of iRGD-MY-1B liposomes in tumor-naïve mice and found no significant toxicity (fig. S9, A to F). Consistent with our hypothesis, combined treatment with the iRGD-MY-1B-liposomes and anti-PD-1 exhibited a robust anti-tumor response in KP and LLC allograft models, with significantly enhanced efficacy compared to individual agents (Fig. 8 C and D).

In conclusion, these data provide compelling evidence that targeting NSUN2 appears well-tolerated and potentiates the efficacy of immunotherapy in preclinical models.

Discussion

ICB has revolutionized the treatment of LUAD [56]; however, resistance remains a central challenge, largely driven by tumor-intrinsic mechanisms that create an immunosuppressive microenvironment. Here, we identified the m⁵C RNA methyltransferase NSUN2 as a critical regulator of immune evasion. Our study demonstrated that high NSUN2 expression orchestrates a “cold” tumor phenotype. Conversely, NSUN2 inhibition reprograms the tumor immune microenvironment into an inflamed, “hot” state, sensitizing tumors to anti-PD-1 therapy. This study identifies NSUN2 as both a candidate predictive biomarker for ICB efficacy and a potential therapeutic target to overcome ICB resistance (Fig. 9).

Fig. 9.

Mechanistic model of the study. (Graphical abstract). NSUN2 enhances the translation of histone deacetylase HDAC8 in an m⁵C-YBX1-dependent manner, which in turn represses CCL5 transcription and CCR5⁺CD8⁺ cytotoxic T cell recruitment. Inhibition of NSUN2 converted immunologically “cold” tumors into “hot” ones and synergized with immunotherapy to induce potent tumor regression

Mechanistically, we delineated a novel NSUN2-m⁵C-YBX1-HDAC8-CCL5 regulatory axis that directly links RNA epigenetics to T cell exclusion. We demonstrated that NSUN2, through its m⁵C methyltransferase activity, post-transcriptionally upregulates HDAC8 by enhancing its mRNA translation via the m⁵C reader protein YBX1. The resulting accumulation of HDAC8 targets the promoter of the T-cell-attracting chemokine CCL5, where it catalyzes the removal of the activating H3K27ac mark, leading to transcriptional repression of CCL5. This cascade reduces CCL5 in the TME, impairs CD8⁺ T-cell trafficking, and establishes an immune-excluded phenotype.

Previously, we demonstrated that the N6-methyladenosine (m⁶A) writer METTL3 drives NSCLC metastasis by enhancing aromatase translation in an m⁶A-dependent manner [23]. In the present study, our findings bring m⁵C methylation, a less-studied modification compared to m⁶A, to the forefront of cancer immunology. The emerging role of the m⁵C methyltransferase NSUN2 as a multi-faceted driver of malignancy and therapy resistance in NSCLC is becoming increasingly apparent. Recent studies have implicated NSUN2 in this process, linking it to intrinsic resistance to EGFR-targeted therapies via the QSOX1 axis and to immune evasion through the stabilization of the checkpoint molecule PD-L1 [56–58]. Our work further advances this understanding by uncovering a non-cell-autonomous mechanism driving ICB resistance: we show that NSUN2’s pro-tumorigenic effects are critically dependent on shaping the adaptive immune response, which may represent a conceptual advance for the field. This study has potential implications from a translational perspective. The strong correlation between high NSUN2 expression and poor ICB response establishes NSUN2 as a candidate predictive biomarker to stratify patients with LUAD and guide treatment decisions. The use of iRGD-functionalized liposomes for the targeted delivery of NSUN2 inhibitor MY-1B to tumors markedly enhanced the efficacy of immunotherapy, leading to significant suppression of tumor growth in LUAD models. Therefore, by defining the mechanism of NSUN2-mediated immune reprogramming, we validated NSUN2 as a therapeutic target aimed at reversing immune exclusion and sensitizing tumors to immunotherapy. While NSUN2 depletion augmented CD8⁺ T-cell infiltration and effector function and reduced MDSC accumulation, we also observed a concurrent increase in exhausted CD8⁺ T cells, including the terminally exhausted subset. We interpret this as a consequence of sustained T-cell activation and antigen exposure within the hot tumor microenvironment, rather than an indication of immune suppression [59, 60]. Notably, the preserved synergy with anti-PD-1 therapy suggests that these exhausted cells might remain partially responsive to ICB.

We acknowledge the apparent discrepancy between the in vitro and in vivo growth phenotypes following NSUN2 depletion. In vitro, NSUN2-deficient cells exhibit a clear proliferation defect, consistent with its established role in supporting cell cycle progression and translation. However, in immunocompetent hosts, Nsun2 deletion remodels the tumor immune landscape by enhancing CD8⁺ T cell recruitment, ultimately leading to tumor growth suppression. In immunodeficient Rag1⁻/⁻ hosts, this immune-dependent tumor control is absent, allowing Nsun2-deficient cells to persist. Moreover, the complex tumor microenvironment, particularly stromal cells such as cancer-associated fibroblasts, likely compensates for the cell-intrinsic defects of KO cells through paracrine secretion of essential growth factors or metabolites [61, 62]. This non-autonomous compensation rescues the tumorigenic potential of Nsun2-deficient cells in the absence of adaptive immunity, restoring their growth rate to wild-type levels. Thus, while NSUN2 does support cell-intrinsic proliferation in vitro, its major in vivo function lies in orchestrating immunosuppression rather than cell-autonomous growth control.

Several aspects of this work warrant further discussion and future exploration. While we have outlined the NSUN2-YBX1-HDAC8-CCL5 axis, future studies should delineate the full spectrum of NSUN2-mediated m⁵C targets that contribute to immune modulation. A recent discovery by Khavari et al. revealed that glucose directly binds to and activates NSUN2, thereby promoting its methyltransferase activity [63]. This finding provides a crucial backdrop for our work. Furthermore, because NSUN2 is aberrantly expressed in LUAD, identifying the upstream signaling pathways that drive NSUN2 overexpression in LUAD could reveal additional therapeutic nodes. Additionally, our therapeutic efficacy studies were conducted in subcutaneous allograft models rather than orthotopic models, which may not fully recapitulate the native tumor microenvironment of LUAD. Moreover, the sample size of clinical LUAD specimens treated with immunotherapy was relatively limited; thus, expanding the cohort could yield more robust conclusions. Finally, although we validated NSUN2 as an attractive therapeutic target, developing more inhibitors or degraders against NSUN2 holds considerable clinical promise for expanding the benefits of immunotherapy.

Conclusions

In summary, this study uncovered a critical role of the m⁵C methyltransferase NSUN2 in fostering an immunosuppressive TME and driving immunotherapy resistance in LUAD. These findings highlight the interplay between RNA epigenetics and cancer immunity and support further exploration of NSUN2-targeted strategies to promote anti-tumor immune responses.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ICIs

Immune checkpoint inhibitors

LUAD

Lung adenocarcinoma

scRNA-seq

Single-cell RNA sequencing

TME

Tumor microenvironment

m⁵C

5-methylcytosine

SPF

Specific pathogen-free

SPC

Soybean phosphatidylcholine

DSPE

1,2-distearoyl-sn-glycero-3-phosphoethanolamin

PEG2K

Polyethylene glycol 2000

LLC

Lewis lung carcinoma

CDS

Coding sequence

KO

Knockout

siRNAs

Small interfering RNAs

TR

Translation ratio

RNC-qPCR

Ribosome nascent-chain complex-bound mRNA-qPCR

ChIP

Chromatin immunoprecipitation

ICB

Immune checkpoint blockade

IHC

Immunohistochemistry

IF

Immunofluorescence

DT

Diphtheria toxin

PBS

Phosphate buffered saline

TIF

Tumor interstitial fluid

anti-PD-1

Anti–programmed cell death protein 1

MeRIP-seq

m⁵C methylated RNA immunoprecipitation sequencing

UMIs

Unique molecular identifiers

PR

Partial response

SD

Stable disease

TCGA-LUAD

The cancer genome atlas lung adenocarcinoma

ssGSEA

Single-sample gene set enrichment analysis

SEM

Standard error of the mean

HATs

Histone acetyltransferases

HDACs

Histone deacetylases

HMTs

Histone methyltransferases

KDMs

Histone lysine demethylase

H3K27ac

Histone H3 lysine 27

mIHC

Multiplex immunohistochemistry

pCR

Pathological complete response

MPR

Major pathological response

pPR

Partial pathological response

pNR

Pathological non-response

Author contributions

Wangyang Meng : Writing – review & editing, Writing – original draft, Data curation, Funding acquisition. Tong Lu : Writing – review & editing, Writing – original draft, Data curation. Mingyuan Du : Writing – review & editing, Writing – original draft, Methodology. Xin Dai : Writing – original draft, Methodology. Yichao Han : Writing – original draft, Methodology. Yingyi Li : Writing – review & editing. Haimin Xu : Methodology, Resources. Biaolong Yang : Writing – review & editing. Zeyu Wang : Writing – review & editing. Hailei Du : Methodology, Resources. Keman Liao : Methodology, Resources. Xinnan Liu : Writing – review & editing. Yidan Sun : Writing – review & editing. Cui Yang : Methodology. Dong Dong : Methodology. Yan Yan : Data curation. Wei Guo : Supervision. Bin Li : Supervision, Funding acquisition. Hecheng Li : Supervision, Funding acquisition.

Funding

This study was supported by the National Natural Science Foundation of China (82372855, 32130041, 82241222, 82503628), Novel Interdisciplinary Research Project from Shanghai Municipal Health Commission (2022JC023), and Interdisciplinary Program of Shanghai Jiao Tong University (YG2023ZD04). The funders had no involvement in the design of this study; the collection, analysis, and interpretation of data; the writing of the report; or the decision to submit the article for publication.

Data availability

Data is available upon reasonable request from the corresponding author.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine (2021 − 224). All participants provided written informed consent before their inclusion in the study. All procedures involving animals were conducted in compliance with ethical guidelines and approved by the Institutional Animal Care and Use Committee (IACUC) of the Shanghai Immunology Institute, School of Medicine, Shanghai Jiao Tong University (Approval No. A-2022-031).

Competing interests

The authors declare no competing interests.