SHMT2: a Metabolic and Immune Biomarker of Aggressive Lung Adenocarcinoma

Summary

Integrated metabolomic, transcriptomic, single-cell, and functional analyses identify SHMT2 as a biomarker of SGOC metabolic reprogramming, aggressive behavior, and reduced predicted immunotherapy responsiveness in lung adenocarcinoma.

Abstract

Serine/glycine-one-carbon (SGOC) metabolism is frequently altered in lung adenocarcinoma (LUAD), but its relationship to tumor behavior and predicted immunotherapy responsiveness remains incompletely defined. Metabolomic profiling of 23 paired LUAD and adjacent normal lung tissues was performed using internal extractive electrospray ionization mass spectrometry. Transcriptomic and clinical data from The Cancer Genome Atlas LUAD cohort (TCGA-LUAD) were analyzed to assess SHMT2 expression, prognosis, differentially expressed genes, and immune-related features. Predicted response to immune checkpoint blockade was evaluated using Tumor Immune Dysfunction and Exclusion (TIDE) and The Cancer Immunome Atlas (TCIA), and drug sensitivity was inferred using oncoPredict. Single-cell RNA-seq data were used to examine the cellular distribution of SHMT2. Experimental validation included quantitative reverse-transcription PCR (RT-qPCR), western blotting, Human Protein Atlas (HPA) immunohistochemistry, and short hairpin RNA (shRNA)-mediated SHMT2 knockdown followed by proliferation, wound-healing and colony formation assays. Metabolomic analysis identified glycine, serine, and threonine metabolism as a prominently altered pathway in LUAD. SHMT2 was upregulated in LUAD and associated with worse overall survival and adverse clinicopathological features. SHMT2-high tumors displayed enrichment of cell-cycle and SGOC-related transcriptional programs, lower immune and stromal scores, and reduced predicted responsiveness to immunotherapy. Single-cell analysis showed relative enrichment of SHMT2 expression in B cell populations. In vitro, SHMT2 was overexpressed in LUAD cells, and its knockdown suppressed proliferation, migration, and clonogenic growth. Collectively, SHMT2 is associated with SGOC metabolic reprogramming, aggressive tumor phenotypes, and an immune-disadvantaged state in LUAD, supporting its potential relevance as a biomarker; therapeutic targeting requires additional pharmacologic and in vivo validation.

Introduction

LUAD remains a leading cause of cancer-related mortality worldwide, driven by pronounced molecular heterogeneity, an early propensity for metastasis, and the limited durability of current treatment strategies^1,2. Although immune checkpoint inhibitors (ICIs) have reshaped the therapeutic landscape of LUAD, sustained clinical benefit is restricted to a subset of patients, and both primary and acquired resistance are frequently observed^3,4. This marked inter-patient variability underscores an urgent need to define the biological determinants that sculpt the tumor immune microenvironment and govern immunotherapy responsiveness, while identifying actionable vulnerabilities that can be leveraged to improve patient stratification and treatment outcomes.

Metabolic reprogramming is a central hallmark of malignant progression⁵. Among metabolic programs, SGOC metabolism connects glycolytic inputs to amino-acid metabolism and nucleotide synthesis, supplying one-carbon units for purine/pyrimidine production and methyl-donor generation⁶. Beyond biomass production, SGOC metabolism contributes to redox homeostasis and interfaces with the methionine cycle to influence DNA, RNA, and histone methylation, positioning it at the nexus of proliferation and epigenetic regulation^7,8. Serine hydroxymethyltransferase 2 (SHMT2), the mitochondrial isoform of serine hydroxymethyltransferase, converts serine to glycine while generating 5,10-methylene-tetrahydrofolate and constitutes a major entry point for mitochondrial one-carbon supply within the SGOC network⁹. SHMT2 is aberrantly upregulated in multiple malignancies and has been associated with enhanced proliferation, invasion, therapeutic resistance, and oxidative-stress adaptation^10,11,12,13. SHMT2 also intersects with inflammatory signaling through the BRISC complex, although a direct SHMT2-driven immune mechanism in LUAD remains to be established^14,15. Accordingly, whether SHMT2 marks a clinically meaningful metabolic phenotype in LUAD and how it relates to immune microenvironment features and predicted immunotherapy response require integrated evaluation.

The bidirectional interplay between tumor metabolism and antitumor immunity is central to contemporary cancer biology¹⁶. Tumor cells can compromise immune competence through competitive depletion of nutrients, accumulation of immunomodulatory metabolites, and metabolic-epigenetic coupling that rewires immune-cell differentiation and effector programs^7,16,17. Heightened one-carbon metabolism may influence methylation-dependent programs and metabolic fitness in T cells, but the direction and magnitude of these effects depend on cell type and nutrient context¹⁷. Adaptive immune regulation also extends beyond T cells: B cells, antibody responses, antigen presentation, and tertiary lymphoid structures (TLS) have been associated with immunotherapy efficacy¹⁸. These observations motivate a multi-layered analysis that can place a metabolic candidate within bulk-tumor, single-cell, and functional contexts without assuming that cross-platform concordance alone proves causality.

Unlike a metabolomics-only design, which identifies altered metabolites without establishing their cellular source or phenotypic relevance, and unlike a bulk-transcriptomics-only design, which may infer pathway activity without directly observing metabolite alterations, the present workflow integrates paired-tissue metabolomics, bulk and single-cell transcriptomics, and in vitro functional assays. Concordance across these layers supports candidate prioritization and reduces dependence on any single analytical platform; however, the resulting evidence remains associative for immune regulation unless validated in immune co-culture or in vivo systems.

Paired-tissue metabolomics of LUAD and adjacent normal lung was first used to identify glycine-associated metabolic perturbations. TCGA transcriptomic and clinical data were then integrated to evaluate SHMT2, a mitochondrial SGOC enzyme that was supported by pathway relevance, tumor upregulation, prognostic association, and experimental tractability. Immune microenvironment profiling, single-cell mapping, and in vitro functional validation subsequently established a multi-layered, hypothesis-generating framework spanning metabolic alteration, enzyme dysregulation, cellular distribution, and tumor cell phenotype. From a practical perspective, this workflow requires fresh-frozen paired tissues. Metabolite identification by mass spectrometry is semiquantitative; candidate metabolites should be independently validated. Immunotherapy predictions (TIDE/TCIA) and drug sensitivity inferences (oncoPredict) are transcriptome-based computational estimates that do not substitute for clinical validation in real-world ICI-treated cohorts and should be interpreted as hypothesis-generating.

Protocol

This study was approved by the Institutional Review Board of the Second Affiliated Hospital of Nanchang University (CDEFYYLK 3-05). Written informed consent was obtained from all patients prior to sample collection, in accordance with the Declaration of Helsinki. Raw data were exported and analyzed using the software and web resources listed in the Table of Materials.

Clinical tissue collection

Paired tumor and matched distant normal lung tissues were collected from patients who underwent surgical resection between January and June 2024 and were pathologically diagnosed with invasive lung adenocarcinoma. Specimens were excised by the operating surgeon, processed within 5 min of resection, immediately snap-frozen in liquid nitrogen, and stored until analysis. De-identified clinicopathological characteristics are provided in Supplementary Table 1.

Internal extractive electrospray ionization mass spectrometry (iEESI-MS) for discriminating LUAD from normal lung tissues

Tissue specimens (1 mm³) were analyzed by iEESI-MS using a hybrid linear ion trap-electrostatic orbital-trap mass spectrometer in positive-ion mode. Methanol extraction solvent (100%) was delivered at 3.0 µL/min with a +5 kV bias. The iEESI-MS¹ scan range was m/z 50–2,000. iEESI-MS² used collision-induced dissociation with a normalized collision energy of 25–35%, an isolation window of m/z 2.0, and a dynamic exclusion time of 30 s. Daily external calibration according to the instrument-standard procedure maintained a mass error <5 ppm. After median normalization, log₂ transformation, and autoscaling, OPLS-DA was performed with 200 permutation tests. Metabolites were annotated using HMDB, and pathway enrichment was performed using KEGG annotations.

Mass spectrometry data processing and metabolite identification

Raw MS data were normalized by median normalization followed by log₂ transformation and autoscaling (mean centering and division by the standard deviation of each variable). Orthogonal partial least squares discriminant analysis (OPLS-DA) was used to identify discriminatory m/z features. Features were considered differential when VIP > 1.0, |log₂(FC)| > 0.58, and p < 0.05. Differential features were subjected to MS² analysis and annotated using the Human Metabolome Database. KEGG-based pathway enrichment was then used to identify altered pathways and prioritize pathway-relevant enzymes. The annotated differential metabolites are provided in Supplementary Table 2.

TCGA-based SHMT2 expression, survival, and clinicopathological association analyses

RNA-seq expression profiles and corresponding clinical annotations for TCGA-LUAD were downloaded from the Genomic Data Commons on December 18, 2025; tumor (n = 542) and normal (n = 59). Fragments per kilobase of transcript per million mapped reads (FPKM) values were transformed as log₂(FPKM + 1). Only samples with complete survival information were included in prognostic analyses. Tumor and normal SHMT2 expression values were compared using the Wilcoxon rank-sum test. Patients were divided into SHMT2-high and SHMT2-low groups at the median expression value. Overall survival was evaluated by Kaplan-Meier analysis and a log-rank test. Associations with tumor-node-metastasis (TNM) stage, age, and sex were assessed using chi-square or Wilcoxon tests, as appropriate. Time-dependent receiver operating characteristic (ROC) curves were generated for 1-, 3-, and 5-year survival. All statistical tests were two-sided.

Differential expression analysis and functional enrichment (GO/KEGG)

Differentially expressed genes (DEGs) between SHMT2-high and SHMT2-low groups were identified using a false-discovery rate < 0.05 and |log₂(FC)| ≥ 1. Results were visualized as a volcano plot and a heatmap of the top 40 DEGs with row-wise z-score normalization. Gene Ontology (GO) and KEGG enrichment analyses used an adjusted p < 0.05. Human gene annotations were obtained from Ensembl in December 2025.

Immune infiltration, immunotherapy response prediction, and drug sensitivity inference

Tumor-microenvironment scores, including ImmuneScore, StromalScore, and ESTIMATEScore, were calculated using ESTIMATE. Immune cell composition was inferred using CIBERSORT with the LM22 reference set (downloaded December 5, 2025), 1,000 permutations, and retention of samples with deconvolution p < 0.05. Groupwise comparisons were performed between SHMT2-high and SHMT2-low tumors. Spearman correlations between SHMT2 and immune checkpoint-related genes were visualized as a correlation matrix. Immunophenoscores were obtained from TCIA, and TIDE was used for complementary prediction of immunotherapy responsiveness. Drug sensitivity was inferred using oncoPredict trained on the December 2023 GDSC release to estimate IC₅₀ values. Group differences used Wilcoxon tests, and continuous associations used Spearman correlations.

Single-cell RNA-seq analysis of LUAD

Single-cell transcriptomic data¹⁹ were downloaded from the Gene Expression Omnibus on December 20, 2025, and analyzed. Cells with 200–6,000 detected features and < 10% mitochondrial transcripts were retained. Counts were normalized using LogNormalize with a scale factor of 10,000; the top 2,000 variable genes were selected, and all genes were scaled. Principal-component analysis used the first 20 components, UMAP used dimensions 1–20, and clustering used a resolution of 0.5. Cell types were annotated using the Human Primary Cell Atlas reference. Cell-type proportions were compared between tumor and normal samples, and Spearman correlation matrices were generated. SHMT2 distributions were examined across annotated cell types. Cell-cycle states were inferred using cyclone and visualized with ridge plots.

Cell culture

BEAS-2B, H1299, and A549 cells were cultured at 37 °C in a humidified incubator set to 5% CO₂ and approximately 95% relative humidity, using the media and supplements specified in the Table of Materials.

Quantitative real-time reverse transcription PCR (RT-qPCR)

Total RNA was extracted using a phenol-based RNA extraction reagent. Extracted RNA was reverse-transcribed into cDNA using a reverse-transcription master mix. Reverse transcription was performed in a 20 µL system with genomic DNA removal at 42 °C for 2 min, reverse transcription at 50 °C for 15 min, and termination at 85 °C for 5 s. Quantitative PCR was performed with a DNA-binding dye-based qPCR master mix on a real-time PCR system using 95 °C for 3 min, followed by 40 cycles of 95 °C for 10 s and 60 °C for 30 s, and a melt-curve step of 95 °C for 15 s, 60 °C for 60 s, and 95 °C for 15 s. All reactions were run in triplicate. Relative expression was calculated using the 2^−ΔΔCt method with GAPDH as the housekeeping gene. Primer stock solutions were prepared at 10 µM in double-distilled water and stored at -20 °C in the dark. Primer sequences are provided in Supplementary Table 3.

Western blotting

Total cellular protein was extracted using radioimmunoprecipitation assay buffer on ice for 30 min, and centrifuged at 12,000 x g for 15 min at 4 °C. Proteins were separated by 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred to 0.22 µm polyvinylidene fluoride membranes. Membranes were blocked with 5% (w/v) non-fat milk in Tris-buffered saline with 0.1% Tween 20 (TBST) for 2 h at room temperature and incubated overnight at 4 °C with anti-SHMT2 (1:1,000) and anti-GAPDH (1:2,000) primary antibodies diluted in 5% non-fat milk‑TBST. After three washes of 8 min each with TBST (20 mM Tris-HCl, 150 mM NaCl, 0.1% Tween-20), membranes were incubated with horseradish-peroxidase-conjugated secondary antibodies (1:5,000) for 1 h at room temperature. Signals were developed using an enhanced chemiluminescence substrate prepared by mixing equal volumes of solutions A and B, incubated for 2 min at room temperature in the dark, and captured with a chemiluminescence imaging system.

HPA immunohistochemistry validation

SHMT2 protein expression patterns in normal lung tissue and LUAD were evaluated using the HPA. “SHMT2” was queried in HPA, and representative immunohistochemistry (IHC) images and associated pathology annotations were downloaded for comparison. To minimize bias due to antibody specificity, antibodies with higher validation levels in HPA were prioritized. Where available, IHC results from multiple antibodies were cross-referenced (HPA020543 and HPA020549).

SHMT2 knockdown in LUAD cells

To investigate the tumor cell-intrinsic role of SHMT2 in LUAD, SHMT2 knockdown was performed in H1299 and A549 cells using shRNA. The shRNA was carried by an H1 promoter-driven, puromycin-selectable lentiviral vector; packaging plasmids were co-transfected during virus production. sh-SHMT2 targeted the human SHMT2 coding region, whereas sh-NC was a scrambled negative control. Prepared viral stocks had a titer of ≥ 1 x 10⁸ transducing units/mL and were stored at -80 °C. Cells were infected with the prepared sh-SHMT2 or sh-NC lentiviral particles described in the Table of Materials. After 48 h, knockdown efficiency was assessed by RT-qPCR and western blotting. Only cells with confirmed SHMT2 silencing were used for proliferation, migration, and colony-formation assays.

Cell proliferation assay

Cell proliferation was evaluated using the CCK-8 assay. Transfected H1299 and A549 cells were seeded into 96-well plates at 5 x 10³ cells per well in sextuplicate. On days 1, 2, 3, 4, and 5, 10 µL of CCK-8 reagent was added to each well (10% of the culture-medium volume) and incubated for 2 h at 37 °C. Absorbance was measured at 450 nm and used as the cell-viability readout.

Wound-healing assay

Cell migration was assessed using a wound-healing assay. Transfected H1299 and A549 cells were seeded into 6-well plates and cultured to 90–95% confluence. A linear wound was generated using a sterile 200 µL pipette tip, and detached cells were removed by gentle washing with phosphate-buffered saline. Cells were then maintained in serum-free or low-serum medium, and the wound area was imaged at 0 h and 24 h under an inverted microscope. Wound closure was quantified using the image-analysis software listed in the Table of Materials and calculated as the percentage reduction relative to the initial wound area.

Colony formation assay

For colony-formation analysis, transfected H1299 and A549 cells were seeded into 6-well plates at 800 cells per well and cultured for approximately 10–14 days until visible colonies formed, with microscopic monitoring during culture. Colonies were fixed with 4% paraformaldehyde for 30 min at room temperature and stained with 0.1% crystal violet for 15 min at room temperature. Colonies containing more than 50 cells were counted under a light microscope.

Statistical analysis for in vitro experiments

All in vitro experiments were performed independently at least three times. Data are presented as the mean ± standard deviation. Two-group comparisons used Student's t-test, and p < 0.05 was considered statistically significant.

Results

Tissue metabolomics reveals extensive metabolic remodeling in LUAD with prominent enrichment of glycine-associated pathways

Paired tumor and adjacent normal lung tissues from 23 patients with LUAD were profiled by direct iEESI-MS; patient characteristics, and differential metabolite annotations are provided in Supplementary Table 1 and Supplementary Table 2. After log₂ normalization, OPLS-DA showed clear separation between tumor and normal metabolic profiles (R²X = 0.824, R²Y = 0.968, Q² = 0.757) (Figure 1A). Model reliability was further supported by 200-permutation testing, with intercepts of R² = 0.789 and Q² = −0.451 (Figure 1B). Using prespecified criteria—variable importance in projection (VIP) > 1, |log₂(FC)| > 0.58, and p < 0.05; 45 differential m/z features were identified (Figure 1C). MS/MS acquisition and HMDB annotation yielded six altered endogenous metabolites, including glycine, serine, and butanone (Supplementary Table 2). KEGG enrichment identified 11 pathways, with glycine, serine, and threonine metabolism among the most prominent (Figure 1D), indicating glycine-related metabolic remodeling in LUAD.

TCGA-LUAD confirms SHMT2 upregulation in tumors and associates high expression with poor prognosis

SHMT2 was prioritized through convergence of four criteria: its direct position at the mitochondrial entry point of SGOC metabolism, significant tumor-associated upregulation in TCGA-LUAD, association with overall survival, and suitability for experimental perturbation. TCGA-LUAD RNA-seq data showed significantly higher log2(FPKM + 1) SHMT2 expression in tumors than in normal lung tissues (Figure 2A). Kaplan–Meier analysis using the median cutoff showed that high SHMT2 expression was associated with worse overall survival (Figure 2B). Exploratory clinicopathological analyses further suggested associations with tumor stage and lymph-node status (Figure 2C), while time-dependent ROC analysis yielded AUC values > 0.70 at 1, 3, and 5 years (Figure 2D).

SHMT2-high tumors display transcriptional programs enriched for cell-cycle progression and the SGOC metabolic axis

To delineate biological programs associated with SHMT2, TCGA-LUAD tumors were divided at the median SHMT2 expression value into SHMT2-high and SHMT2-low groups before differential-expression analysis (Figure 3A). A heatmap of the top 40 differentially expressed genes showed distinct groupwise expression patterns (Figure 3B). GO enrichment identified cell-cycle progression, chromosome segregation, and mitotic division (Figure 3C), providing the analytical basis for describing an SHMT2-high proliferative transcriptional phenotype. KEGG enrichment also highlighted glycine, serine, and threonine metabolism (Figure 3D), indicating coordinated SGOC-associated transcription rather than an isolated change in SHMT2.

High SHMT2 expression associates with an immune-disadvantaged microenvironment, reduced predicted immunotherapy benefit, and distinct drug sensitivity patterns

Tumor microenvironment profiling using ESTIMATE revealed lower ImmuneScore and StromalScore values in the SHMT2-high group (p < 0.001) (Figure 4A). CIBERSORT deconvolution suggested lower proportions of memory B cells, resting memory CD4 T cells, monocytes, and resting mast cells, alongside higher fractions of activated memory CD4 T cells and follicular helper T cells (Figure 4B). Correlation analysis also identified positive associations of SHMT2 with BTLA, TNFSF15, and CD28 and negative associations with CD160 and TNFSF14 (Figure 4C). Complementary transcriptome-based prediction using TIDE and TCIA-derived immunophenoscores indicated lower predicted immunotherapy benefit in SHMT2-high tumors (p < 0.05) (Figure 4D). These estimates were not derived from a clinically treated LUAD cohort and therefore do not demonstrate actual treatment resistance.

Drug sensitivity was computationally inferred rather than experimentally measured. SHMT2-high tumors had lower predicted IC₅₀ values for a glycogen synthase kinase-3 inhibitor and selumetinib but lower predicted sensitivity to talazoparib and a PLK1 inhibitor (Figure 4E–H). Reduced predicted sensitivity to nine additional agents, including cytarabine, irinotecan, and palbociclib, is shown in Supplementary Figure 1A–I. These findings are candidate hypotheses and do not establish clinical or pharmacologic efficacy.

Single-cell analysis reveals distinct cellular composition in LUAD and identifies B cells as the predominant SHMT2-expressing population

To examine the cellular distribution of SHMT2 within the tumor microenvironment, the LUAD single-cell RNA-seq data were analyzed. UMAP clustering identified B cells, endothelial cells, epithelial cells, T cells, and other major compartments (Figure 5A). Cell-type proportions differed between tumor and adjacent normal tissues (Figure 5B–C), and B- and T-cell proportions were positively correlated (p < 0.001) (Figure 5D). SHMT2 expression was relatively enriched in the B cell compartment but was also detectable in epithelial cells (Figure 5E). Because transcript detection alone does not establish cell function or malignant status, these results are interpreted as a localization hypothesis. The tumor cell-intrinsic role of SHMT2 was examined separately in LUAD cell lines.

Single-cell cell-cycle analysis was performed by calculating normalized G1, S, and G2/M scores across cell types. G1 scores were concentrated near zero, whereas S and G2/M scores showed broader distributions (Supplementary Figure 2) and marked cell-type heterogeneity (Figure 5F). These findings indicate proliferative-state heterogeneity within the LUAD microenvironment and are concordant with the cell-cycle and mitotic signatures identified in bulk GO/KEGG analyses.

Experimental validation confirms SHMT2 overexpression and demonstrates that SHMT2 knockdown suppresses malignant phenotypes in LUAD cells

To validate the bioinformatic findings and to evaluate the tumor cell-intrinsic role of SHMT2, expression was first examined in normal bronchial epithelial cells (BEAS-2B) and LUAD cell lines (H1299 and A549), consistent with its detection in bulk TCGA data and epithelial cells in the single-cell analysis (Figure 5E). RT-qPCR showed higher SHMT2 mRNA levels in H1299 and A549 cells than in BEAS-2B cells (Figure 6A), and western blotting confirmed higher protein expression (Figure 6B–C). Representative HPA immunohistochemistry images also showed stronger SHMT2 staining in LUAD than in normal lung tissues (Figure 6D). To further investigate the biological function of SHMT2 in LUAD, SHMT2 was silenced in H1299 and A549 cells using shRNA. RT-qPCR analysis confirmed that SHMT2 mRNA expression was markedly reduced in the sh-SHMT2 groups compared with the sh-NC groups in both cell lines (Figure 6E). Consistently, Western blotting demonstrated effective downregulation of SHMT2 protein expression after knockdown (Figure 6F–G).

Functional assays showed that SHMT2 knockdown significantly impaired the malignant phenotype of LUAD cells. CCK-8 assays revealed that silencing SHMT2 markedly inhibited the proliferative capacity of both H1299 and A549 cells, with the difference becoming more pronounced over time (Figure 6H). Wound-healing assays further showed that SHMT2 depletion significantly reduced the migratory ability of both LUAD cell lines compared with control cells (Figure 6I). Similarly, colony formation assays demonstrated that SHMT2 knockdown significantly decreased the clonogenic potential of H1299 and A549 cells (Figure 6J). Collectively, these findings indicate that SHMT2 promotes LUAD cell proliferation, migration, and colony-forming ability, supporting its pro-tumorigenic role in LUAD.

DATA AVAILABILITY:

All raw data supporting the findings of this study, including the data underlying the figures, are publicly available in Zenodo at https://doi.org/10.5281/zenodo.20676134.

Figure 1: Tissue metabolomics reveals glycine-related metabolic reprogramming in LUAD. (A) OPLS-DA score plot showing separation between LUAD tissues (T, green) and paired adjacent normal tissues (N, blue). (B) Permutation test evaluating the stability and predictive performance of the OPLS-DA model based on R² and Q² values. (C) Venn diagram showing the overlap of differential features screened by VIP score > 1.0, |log₂(FC)| > 0.58, and p < 0.05. Forty-five features were retained. (D) KEGG-based metabolite set enrichment analysis of differential metabolites. Bar length indicates the enrichment ratio and color intensity represents the p value. Abbreviations: VIP = variable importance in projection; FC = fold change. Please click here to view a larger version of this figure.

Figure 2: SHMT2 is upregulated in LUAD and predicts an unfavorable prognosis in TCGA-LUAD. (A) Violin plot comparing SHMT2 expression between normal lung tissues (n = 59) and LUAD tumors (n = 542) using TCGA-LUAD RNA-seq data. (B) Kaplan-Meier overall-survival curves for patients divided at the median SHMT2 expression value. SHMT2-high is shown in pink, and SHMT2-low is shown in blue. (C) Heatmap showing the associations between SHMT2 expression and clinicopathological characteristics, including age, sex, overall pathological stage (Stages I–IV), primary tumor stage (T1–T4), regional lymph node stage (N0–N2), and distant metastasis stage (M0–M1). * indicates statistically significant differences between the SHMT2-high and SHMT2-low groups (*p < 0.05; **p < 0.01; Chi-square test). (D) Time-dependent ROC curves evaluating the prognostic performance of SHMT2 expression for predicting 1-, 3-, and 5-year overall survival in patients with LUAD. The corresponding AUCs are shown in the panel. Abbreviations: SHMT2 = serine hydroxymethyltransferase 2; FPKM = fragments per kilobase of transcript per million mapped reads; ROC = receiver operating characteristic; AUC = area under the curve. Please click here to view a larger version of this figure.

Figure 3: Transcriptomic profiling of SHMT2-high vs SHMT2-low tumors highlights SGOC/one-carbon metabolism and proliferative programs. (A) Volcano plot showing differentially expressed genes between SHMT2-high and SHMT2-low LUAD tumors. Upregulated genes are shown in red, downregulated genes in blue, and nonsignificant genes in gray. (B) Heatmap of representative DEGs with unsupervised hierarchical clustering. (C) Gene Ontology enrichment analysis of differentially expressed genes. Bubble size indicates gene count, color represents the adjusted p value, and the x-axis indicates the gene ratio. (D) KEGG pathway enrichment analysis of differentially expressed genes. Dot size represents gene count, and the x-axis indicates −log₁₀(p value). Abbreviations: SHMT2 = serine hydroxymethyltransferase 2; FDR = false discovery rate; FC = fold change. Please click here to view a larger version of this figure.

Figure 4: SHMT2 expression is associated with tumor immune contexture, predicted immunotherapy response, and drug sensitivity. (A) Comparison of TME score, ESTIMATE score, stromal score, immune score, and tumor purity between the SHMT2-high and SHMT2-low expression groups. (B) Comparison of immune cell infiltration estimated by CIBERSORT between SHMT2-low and SHMT2-high tumors. (C) Spearman correlation matrix between SHMT2 and immune checkpoint-related genes. Circle size reflects correlation strength and color represents Spearman’s ρ (positive to negative). (D) Immunophenoscore comparisons under four immune checkpoint blockade settings (CTLA4-/PD-1-, CTLA4-/PD-1+, CTLA4+/PD-1-, CTLA4+/PD-1+), indicating reduced predicted benefit in SHMT2-high tumors. (E) Predicted sensitivity to Talazoparib in the SHMT2-high and SHMT2-low expression groups, shown as estimated IC₅₀ values (top) and the correlation between SHMT2 expression and predicted drug sensitivity (bottom). (F) Predicted sensitivity to BI-2536 in the SHMT2-high and SHMT2-low expression groups, shown as estimated IC₅₀ values (top) and the correlation between SHMT2 expression and predicted drug sensitivity (bottom). (G) Predicted sensitivity to S63845 in the SHMT2-high and SHMT2-low expression groups, shown as estimated IC₅₀ values (top) and the correlation between SHMT2 expression and predicted drug sensitivity (bottom). (H) Predicted sensitivity to Selumetinib in the SHMT2-high and SHMT2-low expression groups, shown as estimated IC₅₀ values (top) and the correlation between SHMT2 expression and predicted drug sensitivity (bottom). Abbreviations: SHMT2 = serine hydroxymethyltransferase 2; TME = tumor microenvironment; ESTIMATE = Estimation of STromal and Immune cells in MAlignant Tumor tissues using Expression data; IC₅₀ = half-maximal inhibitory concentration. Please click here to view a larger version of this figure.

Figure 5: Single-cell landscape reveals altered cellular composition, SHMT2 expression bias, and cell-cycle heterogeneity in LUAD. (A) UMAP visualization of publicly available single-cell transcriptomes from LUAD and adjacent normal lung tissues, colored by annotated cell type. (B) Relative cell type proportions in normal and tumor samples shown as horizontal stacked bars. (C) Overall cellular composition shown as vertical stacked bars. (D) Spearman correlation heatmap of cell-type proportions; red indicates positive and blue indicates negative correlation (*p < 0.05, **p < 0.01, ***p < 0.001). (E) Violin plots of SHMT2 expression across major cell types. (F) Dot plot showing G1-, S-, and G2/M-phase cell-cycle scores across the identified cell types; dot size indicates the percentage of cells, and color indicates the average cell-cycle score. Abbreviations: UMAP = Uniform Manifold Approximation and Projection. Please click here to view a larger version of this figure.

Figure 6: Experimental and histological validation of SHMT2 expression and functional effects of SHMT2 knockdown in LUAD cells. (A) RT-qPCR analysis of SHMT2 mRNA expression in normal bronchial epithelial cells (BEAS-2B) and LUAD cell lines (H1299 and A549). (B) Western blot analysis of SHMT2 protein expression in BEAS-2B, H1299, and A549 cells, with GAPDH as the loading control. (C) Densitometric quantification of SHMT2 protein expression normalized to GAPDH in H1299 and A549 cells relative to BEAS-2B. (D) Representative immunohistochemistry (IHC, x20) images of SHMT2 in normal lung and LUAD tissues from the HPA using antibodies HPA020543 and HPA020549. (E) RT-qPCR validation of SHMT2 knockdown efficiency in H1299 and A549 cells transfected with sh-SHMT2 or sh-NC. (F) Western blot analysis of SHMT2 protein expression after SHMT2 knockdown in H1299 and A549 cells. (G) Densitometric quantification of SHMT2 protein levels normalized to GAPDH after SHMT2 knockdown. (H) Cell Counting Kit-8 (CCK-8) assays showing the effects of SHMT2 knockdown on cell proliferation in H1299 and A549 cells. (I) Wound-healing assays (x200) showing reduced migratory ability after SHMT2 knockdown in H1299 and A549 cells; the right panel shows quantification of relative migration. (J) Colony formation assays showing decreased clonogenic ability after SHMT2 knockdown in H1299 and A549 cells; the right panel shows quantification of colony numbers. Data are presented as the mean ± SD from at least three independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Abbreviations: sh-NC = short hairpin RNA negative control; sh-SHMT2 = short hairpin RNA targeting SHMT2. Please click here to view a larger version of this figure.

Supplementary Figure 1: SHMT2 expression is inversely associated with predicted drug sensitivity in TCGA-LUAD. Predicted drug response was inferred for TCGA-LUAD samples and compared between SHMT2-low and SHMT2-high tumors defined by the median cutoff. For each compound, the upper panel shows groupwise differences in an IC₅₀ (half-maximal inhibitory concentration) based response score (lower values indicate greater predicted sensitivity), and the lower panel shows the association between continuous SHMT2 expression and predicted sensitivity with a fitted linear-regression line and 95% confidence band. Group comparisons used a two-sided Wilcoxon rank-sum test, and correlations are reported as Spearman's r with corresponding p values (* p < 0.05, ** p < 0.01, *** p < 0.001). (A) Cytarabine. (B) GDC0810. (C) P22077. (D) Irinotecan. (E) AZD4547. (F) Palbociclib. (G) VE821. (H) I-BRD9. (I) Oxaliplatin.Please click here to download this file.

Supplementary Figure 2: Ridge plot visualization of cell-cycle phase score distributions across cell types at single-cell resolution. Ridge plots show normalized cell-cycle score distributions across major cell types, including B cells, T cells, epithelial cells, macrophages, monocytes, endothelial cells, and smooth muscle cells. (A) G1-phase scores. (B) S-phase scores. (C) G2/M-phase scores. Differences in distribution shape and peak position indicate heterogeneity in proliferative states among cell populations within the LUAD microenvironment.Please click here to download this file.

Supplementary Table 1: De-identified clinicopathological characteristics of the 23 patients with LUAD included in paired-tissue metabolomics. Variables include patient number, sex, age, primary tumor stage (T), regional lymph node stage (N), and overall pathological stage. Please click here to download this file.

Supplementary Table 2: Differential metabolites identified in LUAD versus paired adjacent normal tissues. The table lists the HMDB accession number, metabolite name, molecular formula, VIP score, log₂(FC), and p value for each differential metabolite. Abbreviations: HMDB = Human Metabolome Database; VIP = variable importance in projection; FC = fold change. Please click here to download this file.

Supplementary Table 3: Primer sequences used for quantitative reverse transcription PCR (RT-qPCR). The table lists the target gene, primer direction (forward or reverse), and primer sequence (5′–3′) used for gene expression analysis.Please click here to download this file.

Discussion

This study used a sequential framework comprising metabolomic discovery, enzyme prioritization, transcriptomic interpretation, immune-context assessment, single-cell localization, and tumor cell functional validation. The analyses identify glycine-related and SGOC metabolic remodeling in LUAD and associate SHMT2 with adverse outcomes and lower computationally predicted immunotherapy benefit. This prioritization is stronger than reliance on a single omics layer but does not establish an immune-causal mechanism.

SHMT2 constitutes a major mitochondrial entry point for conversion of serine to glycine and generation of one-carbon units. The TCGA-LUAD analysis showed tumor-associated SHMT2 upregulation and linked high expression to poorer survival and adverse clinicopathological features. These findings agree with studies associating SHMT2 with cancer-cell survival, proliferation, invasion, and therapy resistance^20,21,22,23. Additional work has connected SHMT2 to epigenetic regulation and has motivated development of selective one-carbon enzyme inhibitors^24,25. Differential-expression and enrichment analyses further placed SHMT2 within a coordinated SGOC and cell-cycle program rather than an isolated gene-level change. Concordance between metabolite-pathway enrichment and SHMT2-associated transcription therefore strengthens candidate prioritization, while remaining correlative across the omics layers.

The association between SHMT2-high tumors and an immune-disadvantaged context can be framed through three nonexclusive routes. First, increased tumor cell demand for serine, glycine, folate-linked one-carbon units, and methionine-cycle substrates may alter nutrient partitioning and reduce the metabolic fitness of neighboring lymphocytes. Second, SHMT2-supported NADPH, glutathione, and nucleotide production may improve tumor cell survival under oxidative and nutrient stress, thereby changing stress-associated signals released into the microenvironment. Third, one-carbon flux and S-adenosylmethionine availability can influence methylation-dependent programs in both malignant and immune cells^26,27,28,29. These mechanisms are biologically plausible, but the present bulk and single-cell analyses did not measure local metabolites, cytokine secretion, or immune cell function directly. Consequently, lower TIDE/TCIA-predicted benefit in SHMT2-high tumors should be interpreted as a hypothesis linking SGOC activity to immune disadvantage, not as proof that SHMT2 causes clinical immunotherapy resistance.

Single-cell analyses provide additional structural resolution. Analysis of publicly available single-cell RNA-seq data showed altered cellular composition in tumor versus normal lung tissues and a positive association between B- and T-cell proportions, consistent with coordinated adaptive immunity³⁰. SHMT2 expression was relatively enriched in B cells but remained detectable in epithelial cells. This observation is relevant because B cells and tertiary lymphoid structures can support antigen presentation, local antibody responses, and response to immune checkpoint blockade^18,31,32,33; however, it does not identify the responsible B cell subtype or establish that B cell SHMT2 is beneficial or detrimental. Bulk and single-cell results also converged on cell-cycle activity: mitotic and chromosome-segregation terms were enriched in SHMT2-high tumors, and S- and G2/M-phase scores varied across cell types. Because mitochondrial one-carbon metabolism supplies nucleotide precursors, these findings support a plausible link between SHMT2-associated SGOC activity and proliferative demand.

Xi et al. identified a TP63-RAC2 program that enhanced macrophage efferocytosis, promoted M2-like polarization, and remodeled esophageal cancer toward immunosuppression³⁴, providing a direct example of tumor-macrophage communication. Jiang et al. linked SATB2 in pancreatic cancer to tumor cell proliferation and migration as well as altered T-cell cytotoxicity³⁵, illustrating simultaneous tumor-intrinsic and immune effects. By contrast, Liu et al. showed that TRIM29 drove glioblastoma through NEFL degradation and PI3K/AKT activation³⁶; this primarily tumor-intrinsic mechanism cautions against assigning every adverse immune association to cytokine signaling. Zhai et al. identified WDR54-mediated amplification of NF-κB signaling in hepatocellular carcinoma³⁷, highlighting a cytokine-responsive inflammatory node that can couple malignant behavior to microenvironmental signaling. Yin et al. used co-culture and in vivo experiments to show that PLAU cooperated with nerve-growth-factor-associated perineural interactions in head and neck cancer³⁸, demonstrating that soluble and extracellular factors can organize non-immune stromal crosstalk. Finally, Liu et al. summarized the dual, context-dependent roles of cytokine-driven JAK/STAT signaling: persistent IL-6/STAT3 and IFN-related signaling can promote PD-L1 expression, suppressive myeloid states, and T-cell exhaustion, whereas appropriately timed signaling is also required for antigen presentation and antitumor immunity³⁹. Collectively, these studies provide a biological rationale for testing whether SGOC reprogramming modifies cytokine, macrophage, stromal, or lymphocyte states, but none establishes an SHMT2-specific cytokine circuit in LUAD. The present TIDE/TCIA findings therefore define a testable direction for mechanistic work.

Redox regulation provides a particularly relevant mechanistic context. DeNicola et al. demonstrated in non-small cell lung cancer that NRF2 regulates PHGDH, PSAT1, and SHMT2 through ATF4, supporting glutathione and nucleotide production and linking this program to poor prognosis⁴⁰. More recently, Zhang et al. showed that esophageal adenocarcinoma cells surviving HER2 inhibition accumulated NRF2; NRF2 knockdown increased lapatinib cytotoxicity, whereas sustained NRF2 expression reduced sensitivity and created NRF2 dependence⁴¹. The latter study did not directly test SHMT2 inhibition, but together these findings suggest that SHMT2 may function as part of an NRF2-supported redox and biosynthetic program. In LUAD, this hypothesis should be tested by stratifying KEAP1/NFE2L2 status and determining whether SHMT2 perturbation selectively increases reactive oxygen species or restores treatment sensitivity in NRF2-active models.

Compared with metabolomics-only or transcriptomics-only analyses, the integrated workflow links a tissue-level metabolite signal to pathway-relevant gene expression, cell-type distribution, and a perturbation phenotype. This design improves candidate prioritization and can be scaled by substituting public transcriptomic or single-cell cohorts, but each layer introduces distinct sources of variability. Rapid, uniform tissue freezing is critical because post-excision delay can alter small-molecule abundance. Mass spectrometric drift, unstable total-ion signal, or weak permutation performance should trigger recalibration, signal-quality review, and reanalysis before metabolite interpretation. Bulk analyses require prespecified grouping and filtering thresholds, while single-cell conclusions should be tested across reasonable quality-control and clustering-resolution settings. In vitro validation requires matched cell passage, comparable confluence, a uniform wound width, and confirmed SHMT2 knockdown before phenotype assays. These checkpoints improve reproducibility but do not eliminate cohort, platform, or model-specific effects.

Several limitations constrain interpretation. The tissue-metabolomics cohort was modest, and the identified metabolites require targeted quantitative validation. Immunotherapy benefit and drug sensitivity were inferred from transcriptome-based models rather than measured in an ICI-treated cohort or pharmacologic experiment. The single-cell dataset suggests B cell-biased SHMT2 expression but does not resolve B cell subtype, spatial organization, or SHMT2-dependent immune function. The in vitro experiments establish a tumor cell growth and migration phenotype after genetic knockdown but do not establish a cytokine-mediated immune mechanism or therapeutic safety. Future studies should stratify tumors by KEAP1/NFE2L2 status, quantify SGOC flux and cytokine secretion, and combine SHMT2 perturbation with immune co-culture, spatial profiling, pharmacologic selectivity studies, and in vivo validation.

In summary, SHMT2 is associated with SGOC metabolic remodeling, unfavorable prognosis, proliferative transcriptional programs, an immune-disadvantaged tumor context, and lower computationally predicted immunotherapy responsiveness in LUAD. Single-cell analysis suggested relative enrichment of SHMT2 expression in B cell populations, while in vitro knockdown reduced proliferation, migration, and clonogenic growth in LUAD cells. These findings support SHMT2 as a candidate biomarker for further validation and provide a testable framework for studying metabolic-immune interactions.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
A549 cells	CCL-185	ATCC
anti-GAPDH antibody	M1310-2	Huaan Biotechnology
anti-SHMT2 antibody	D197021	Sangon Biotech
BEAS-2B cells	CRL-9609	ATCC
Cell Counting Kit-8 (CCK-8)	K1018	APExBIO
Database: TCGA / GDC	Accessed 2025-12-18	National Cancer Institute (NCI)
Database: GEO	Accessed 2025-12-20	NCBI
Database: HMDB	Accessed 2025-12-12	Human Metabolome Database
Database: KEGG	Accessed 2025-12-11	Kanehisa Laboratories
Database: Human Protein Atlas		Protein Atlas
Database: GDSC	2023-12 release	Sanger Institute
Database: Ensembl	2025-12	EMBL-EBI
Database: TCIA	Accessed 2025-12-10	The Cancer Immunome Atlas
H1299 cells (NCI-H1299)	CRL-5803	ATCC
HRP-conjugated secondary antibody	D110087	Sangon Biotech
HiScript II Q Select RT SuperMix	KR116	Tiangen Biotech
LightCycler 480 System	5015278001	Roche
Non-fat milk	Common laboratory reagent	BD (or as used)
PVDF membrane, 0.22 µm	GVHP00010 (or common laboratory reagent)	MilliporeSigma (or as used)
R package: pRoloc	1.40.0	Bioconductor
R package: TCGAbiolinks	2.26.0	Bioconductor
R package: survival	3.4-0	CRAN
R package: survminer	0.4.9	CRAN
R package: timeROC	1.0.4	CRAN
R package: DESeq2	1.38.1	Bioconductor
R package: ggplot2	3.4.4	CRAN
R package: pheatmap	1.0.12	CRAN
R package: clusterProfiler	4.6.2	Bioconductor
R package: org.Hs.eg.db	3.16.0	Bioconductor
R package: estimate	1.0.13	Bioconductor / GitHub
R package: CIBERSORT	1	(standalone R script)
R package: corrplot	0.92	CRAN
R package: TIDE	0.3.0	R package
R package: oncoPredict	1.1.1	CRAN / GitHub
R package: GEOquery	2.66.0	Bioconductor
R package: Seurat	4.3.0	CRAN
R package: SingleR	1.10.0	Bioconductor
R package: scran	1.26.0	Bioconductor
SYBR Green qPCR Master Mix (Universal)	HY-K0501A	MedChemExpress (MCE)
TRIzol Universal Reagent	DP424	Tiangen Biotech