Integrated multi-omics analysis reveals GPD1L-mediated dysregulation of glycerophospholipid metabolism facilitates the progression of lung adenocarcinoma

Abstract

Background

With the widespread of low-dose computed tomography (LDCT), an increasing number of early-stage lung adenocarcinomas (LUAD) presenting as ground-glass opacity (GGO) nodules are being detected. However, the metabolic dysregulation involved in the progression from GGO to solid nodule remains poorly understood. In this study, we investigated the significance of GPD1L-mediated glycerophospholipid metabolism dysregulation during this transition and explored potential therapeutic targets.

Methods

We conducted analysis on multi-omics of lung adenocarcinoma, including bulk RNA-seq, whole-exome sequencing, metabolomics and single-cell RNA seq data. Extensive experiments were conducted to validate the role of GPD1L and the potential therapeutic target GPX4. Rescue assays were performed to further elucidate the mechanism.

Results

Multi-omics analysis revealed the essential role of GPD1L-mediated glycerophospholipid metabolism dysregulation between ground-glass opacity and solid nodule. Knockdown of GPD1L in lung adenocarcinoma cell lines promoted tumor progression and attenuated the degradation of HIF-1α, conferring resistance to hypoxia and apoptosis. Metabolomics, transcriptomics, and in vitro experiments further demonstrated that GPD1L downregulation impairs GPX4 expression by reducing AKT phosphorylation. In addition, GPD1L expression was associated with poor prognosis and may serve as a potential biomarker for immunotherapy.

Conclusion

GPD1L-mediated glycerophospholipid metabolism dysregulation promotes lung adenocarcinoma progress by stabilizing HIF- 1α. Meanwhile, GPX4 were downregulates via AKT/NRF2/GPX4 pathway, suggesting a potential therapeutic target.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12931-026-03558-w.

Introduction

Lung cancer remains the most commonly diagnosed malignancy worldwide and is a leading cause of cancer-related mortality [1], accounting for approximately 125,070 deaths annually in the United States [2]. The widespread adoption of low-dose computed tomography (LDCT) screening has significantly enhanced the early detection of lung adenocarcinoma (LUAD) [3, 4]. During tumor progression, radiological manifestations evolve distinctly. Notably, ground-glass opacity (GGO) nodules are characterized as pathognomonic radiological features by hazy increased lung attenuation on CT imaging, in contrast to solid nodules, while preserving the visualization of underlying vascular and bronchial structures [5–7]. A recent LUAD screening study in China indicated that 84.9% of lung nodule cases contained GGO component [8]. Previous studies have demonstrated that GGO serves as a favorable prognostic marker [9, 10]. However, patients whose nodules progress to solid nodules exhibit significantly poorer clinical outcomes, including a marked decline in overall survival rates and progression-free intervals compared to those with GGO lesion [11, 12].

Metabolic dysregulation, a hallmark of cancer progression, enables tumor cells to reprogram their biochemical networks to support survival and proliferation [13, 14]. This adaptation is driven by the accumulation of genetic mutations and transcriptional alterations that disrupt key metabolic enzymes and pathways [15, 16]. Emerging evidences have highlighted the pivotal role of tumor metabolism in immune evasion, therapeutic resistance, and metastasis [17, 18]. Among various metabolic processes, lipid and glycerophospholipid metabolism have been increasingly recognized as critical regulators of tumor growth and microenvironmental adaptation, yet their role in the radiological and biological transition from GGO to solid nodules remains poorly understood. Building upon our previous comprehensive genomic and transcriptomic analyses across LUAD stages [19], we identified glycerophospholipid metabolism dysregulation as a key event during the progression from GGO to solid nodules. GPD1L (glycerol-3-phosphate dehydrogenase 1-like), a cytosolic enzyme involved in glycerophospholipid metabolism and redox homeostasis, has been implicated in cellular adaptation to hypoxia and oxidative stress. Consistent with these functions, we found that GPD1L downregulation facilitates the transition from GGO to solid nodules by promoting hypoxia resistance and apoptosis evasion via HIF-1α stabilization. Furthermore, we identified glutathione peroxidase 4 (GPX4) as a promising therapeutic target for GPD1L-deficient tumors. Our investigation also explored metabolic correlations and intercellular communication between epithelial and immune cells, highlighting GPD1L’s role in shaping the tumor microenvironment. Collectively, these findings elucidate the function of GPD1L-mediated glycerophospholipid metabolism dysregulation in LUAD progression and shed the light on metabolism-based novel therapeutic strategies for lung adenocarcinomas.

Materials and methods

Study cohort

Our cohort contains 197 lung adenocarcinoma patients underwent surgery between September 2011 and May 2016 at the Department of Thoracic Surgery, Fudan University Shanghai Cancer Center. And none of the patients involved in our cohort received neoadjuvant therapy. The study was approved by the Institutional Review Board(IRB#090977–1) of the Fudan University Shanghai Cancer Center (FUSCC). A total of 121 patients with complete prognostic information were included in this study as training set. We grouped part-solid nodules as GGO group, since nodules containing a ground-glass component are associated with a better prognosis than purely solid nodules [20].

The validation cohort were downloaded from TCGA database (https://portal.gdc.cancer.gov/) and the metabolomic data from an already published paper [21]. Part of the analysis of TCGA-LUAD cohort was generated by the database scCancerExplorer (https://bianlab.cn/scCancerExplorer/home) [22]. The immune therapy cohorts were downloaded from GSE33072 from GEO database.

GSEA and ssGSEA

To discover the different expression pathways between the Solid and ground-glass opacity nodules, we applied GSEA using the Gene set enrichment analysis software(version4.3.2) [23]. TPM form of the bulk RNA-seq data served as the input for GSEA4.3.2, setting the minimal set size to 15 and maximal size to 500. For single-sample gene set enrichment analysis, we used a custom R script to collect 13 lipid metabolism pathways from KEGG database (https://www.genome.jp/kegg/pathway.html#metabolism), consisting 595 genes. To quantify the enrichment of the lipid metabolism pathways, we use R package GSVA (V1.50.0) [24] with the “kcdf = Guassion (using counts), method = “ssGSEA” parameter. Pathway scores between subtypes were then compared using a rank-sum test.

Single cell RNA-seq data analysis

We utilized the R package Seurat (V5.0.0) [25] and Harmony (V1.2.3) [26] to work with lung adenocarcinoma samples from GSE117570 [27], GSE189357 [28]. GSE189357 was used as training set while GSE117570 as the validation set. All the lung adenocarcinoma samples from two databases were selected for further analysis. After quality control step and principal component analysis, the uniform manifold approximation and projection (UMAP) method was used to nonlinearly reduce the dimension. Then the R package SingleR (V2.6.0) [29] was performed to annotate the cell clusters.

Cell culture and treatment

RPMI1640 (Meilunbio, China) containing 0.5% Penicillin/Streptomycin,sterile (100X) (Meilunbio, China) and 10% fetal bovine serum (Meilunbio, China) was used for the cultivation of PC9 and A549 cells. All the culturing was carried out in an environment with 5% CO₂ at 37 ℃.

ML162, Sc79, Capivasertib (MedChemExpress, USA) were dissolved in DMSO, while the Cisplatin (MedChemExpress, USA) was dissolved in ddH₂O, yielding a 10 mM masterbatch. The masterbatch was then diluted with culture medium to the indicated concentration for experiment. AnaeroPack (MGC, Japan) with RPMI1640 without fetal bovine serum were used to hypoxia treatment.

Small interfering RNAs (Genepharma, China) targeting GPD1L and HIF1A were delivered into cells via RNAi Max (Thermo Fisher Scientific, USA) for 48 h, and then the cells were harvested for experiments. The siRNA sequences are as follows:

• si-GPD1L-1:

• sense: GCAGCAAAGUAAUGGAGAATT;

• antisense: UCUCCAUUACUUUGCUGCCG.

• si-GPD1L-2:

• sense: GGACAAGUUUCCAUUGUUUTT;

• antisense: AAACAAUGGAAACUUGUCCAG.

• si-HIF1A:

• sense: CCGCUGGAGACACAAUCAUAU;

• antisense: AUAUGAAUGUGUCUCAGCG.

• shGPD1L:

• pLVshRNA-puro combined with sequence of si-GPD1L-2.

Cell–cell communication and trajectory analysis

We used R package CellChat (V1.6.1) [30] for cell communication analysis. Gene expression data was analyzed to identify highly expressed ligand-receptor, followed by the construction of intercellular communication network. Subsequently, network centrality scores were calculated to identify key contributors, and CellChat's pattern recognition was employed to explore signaling patterns. The R package Monocle (V2.32.0) [31] was utilized for trajectory analysis. DDR-Tree algorithm was used to reduce dimension. We then visualized the trajectories by cellular state plots for displaying the change of specific gene’s expression.

LC–MS/MS

Cells after treated by siRNA mentioned above in the logarithmic growth phase were trypsinized after removing the culture medium. Following washes with phosphate-buffered saline (PBS) twice, the harvested cell pellets were snap-frozen in liquid nitrogen and subsequently stored at −80 °C until further processing. The frozen cell samples were then subjected to LC–MS/MS analysis (APExBIO, USA). And the results are analyzed by database MetaboAnalyst (https://www.metaboanalyst.ca/) [32].

Quantitative RT-PCR analysis

RNA was isolated from cells by using total RNA kit I (omega bio-tech, USA) and reverse transcribed by the HiScript IV 1 st Strand cDNA Synthesis Kit(+ gDNA wiper) (Vazyme, China) under the instruction of manufacturer. We then use SYBR Green PCR Master Mix (Abcolnol, China) and qPCR machine product by BAIYUAN GENE-TECH (China) to perform the PCR. The primers used are listed below:

• GAPDH:

• F: GGAGCGAGATCCCTCCAAAAT

• R: GGCTGTTGTCATACTTCTCATGG

• GPD1L:

• F: CCCCTGAAAGTGTGCATCGT

• R: GGCAGCTTGTGTCCAGGAA

Western blot

Cell lysates extracted by the RIPA (Epizyme Biotech, China) and 100 × cocktail (APExBIO, USA) were harvested for Western blot. After separation through SDS-PAGE, the proteins were transferred onto PVDF membrane. Then the membrane was blocked by 5% skim milk. And the membrane was incubated with the primary antibodies, including GPD1L (17,263–1-AP, proteintech, USA), HIF1A (20,960–1-AP, proteintech, USA), Caspase-3/P17/P19 (19,677–1-AP, proteintech, USA), GPX4 (67,763–1-Ig, proteintech, USA), SLC7A11 (26,864–1-AP, proteintech, USA), SLC3A2 (66,883–1-Ig, proteintech, USA), pan-AKT (4691 T, Cell Signaling Technology, USA), p-AKT (4060S, Cell Signaling Technology, USA), NRF2 (12,721, Cell Signaling Technology, USA), HIF1α (ab179483, abcam, USA), Hydroxy-HIF-1 alpha (3434 T, Cell Signaling Technology, USA), β-Actin (ab6276, abcam, USA), overnight at 4 ℃. And the following day, we incubated the second antibody (abcam, UK) for 1–2 h at ambient temperature. The ChemiDoc XRS + system (Bio-Rad, USA) was used to image.

CCK-8 assay and colony formation assay

The Cell Counting Kit-8 test kits (Vazyme, China) was used to measure the proliferation of the PC9 cells at the indicated time points. The PC9 cells were seeded in 96-well plates at 1500 cells per well after interfered by siGPD1L. For colony formation experiment, PC9 and A549 cells were seeded in six-well plates at 1000 cells per well. After culturing for 10–14 days, the cells were fixed with 4% paraformaldehyde and then stained with 0.5% crystal violet staining. Visible colonies were counted by image J (V1.8.0_172).

Transwell migration assay

24-well Transwell inserts (Corning, USA) was used to assess the ability of cells’ migration. 5 × 10⁴ cells were seeded in a serum-free culture media (200μL) in the upper chamber while the lower chamber was filled with culture media (700μL) with 10% fetal bovine serum. After 24 h, the migrated cells were fixed by 4% paraformaldehyde and stained by the 1% crystal violet. Visible cells were counted by image J.

Subcutaneous tumor model

Male C57 mice (6 weeks old) were maintained under SPF conditions. KP cells (2 × 10⁵ in 100 μl PBS) were subcutaneously injected into the right flank. Tumor size was measured every 2 days with calipers, and volume was calculated as 0.5 × length × width². Mice were euthanized at the study endpoint, and tumors were collected for further analysis.

Flow cytometric and fluorescence imaging

Harvested cells (log-phase growth) were washed twice with ice-cold PBS and resuspended in 1 × binding buffer at a density of 5 × 10⁵ cells/mL‌. Cells were then incubated with APC-conjugated Annexin V‌ (5 µL) and Propidium iodide (PI‌, 10 µL) (Vazyme, China) for apoptosis evaluation in the dark at room temperature for 15 min. Stained cells were analyzed immediately using a flow cytometer. Same for the L/D, CD3, CD8 (Biolegend, USA), ROS staining (meilunbio, China). Stained cells were also captured by fluorescence imaging‌‌.

IC50 determination

1500 cells were seeded per well in a 96-well plate with 100 µl 1640 culture medium. The medium was replaced into medium with drugs after culturing for 24 h. And the CCK-8 kit is used to obtain the result, after 48 h incubation period.

IHC staining and evaluation

We employ immunohistochemistry (IHC) to detect GPD1L protein expression in tissue samples. Specimens were fixed in 4% neutral-buffered formaldehyde, embedded in paraffin, and sectioned into 4-μm-thick serial slices using a microtome. After deparaffinization in xylene (3–5 min) and rehydration through a graded ethanol series (100% to 80%), antigen retrieval was performed via microwave heating in citrate buffer (pH 6.0). The staining protocol included sequential steps: endogenous peroxidase blocking with 3% hydrogen peroxide (15 min), nonspecific binding inhibition using 10% normal serum (1 h at room temperature), overnight incubation with anti-GPD1L primary antibody (1:200 dilution at 4 °C), followed by 1-h incubation with horseradish peroxidase (HRP)-conjugated secondary antibody at room temperature, and final chromogenic development with 3,3'-diaminobenzidine (DAB). The staining and scan service is provided by the YangMing MED company (China).

Two pathologists independently evaluated the stained sections using an immunoreactive score (IRS) system. Final results represented the mean of double-blind evaluations.

Evaluation of immune status and tumor mutation

R package EPIC (V1.1.7) [33] was used to evaluate the immune status. For the analysis, TPM expression data of FUSCC cohort was used. Samples with P < 0.05 were considered as usable data. The immunotherapy response was conducted by TIDE database (http://tide.dfci.harvard.edu/). The pathway mutation analysis and top mutation genes were identified by the R package maftools (V2.20.0) [34] in different groups.

Tumor cell–T cell co-culture assay

Human peripheral blood mononuclear cells (PBMCs) were isolated from healthy donors by Ficoll density gradient centrifugation. T cells within PBMCs were activated using anti-CD3/CD28–coated beads in RPMI 1640 medium supplemented with 10% fetal bovine serum and recombinant human IL-2 (50 U/mL) for 48 h.

PC9 cells were seeded in 24-well plates and allowed to adhere overnight. Activated PBMCs were then added to tumor cells at an effector-to-target (E:T) ratio of 2:1 and co-cultured for 48 h. For PD-L1 blockade experiments, a PD-L1–neutralizing antibody (Selleck, USA) or isotype control was added at the start of co-culture.

Tumor cell viability was assessed at the indicated time points, and culture supernatants were collected for IFN-γ measurement by ELISA (WEIAOBIO, China).

Statistical analysis

All data processing and analysis was performed by R version 4.3.3. All the normalization of value for heatmap was completed by z-score. We utilized the survminer package (V0.5.0) for survival analysis and the survival package (V3.6–4) for Cox proportional hazards modeling. Cox analysis was carried out with 95% confidence intervals (CIs) to evaluate overall survival (OS). Additionally, Kaplan–Meier survival analysis, along with the log-rank test, was employed to assess OS. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression. Linear regression was applied to assess associations between continuous variables, while Cox proportional hazards regression was used to evaluate the relationship between variables and survival outcomes. To minimize overfitting, the FUSCC cohort was strictly used as the training cohort and no model parameters, expression cutoffs, or pathway definitions were optimized or adjusted based on the validation cohorts. All validation analyses were performed using predefined hypotheses derived from the discovery cohort. Comparisons between independent two groups used Wilcoxon signed-rank test. One-way ANOVA and post hoc multiple comparisons tests were utilized for experiments with complex conditions. A two-tailed P value < 0.05 was considered statistically significant unless stated otherwise.

Results

Glycerophospholipid metabolism is significantly downregulated in solid nodules

The study design and workflow are illustrated in Fig. 1A. LUAD exhibits distinct radiological manifestations, including GGO and solid nodule. Consistent with previous reports, [10, 12, 35] patients with GGO nodules exhibited more favorable clinical outcomes, a trend also observed in our cohort (Fig. 1B). To explore metabolic pathway alterations during the transition from GGO to solid nodules, we performed GSEA analysis, which revealed significant enrichment of lipid metabolism-related pathways in the GGO group (Fig. S1A). Additionally, we conducted single-sample enrichment analysis on 13 lipid metabolism-related pathways from KEGG database. Among the 13 pathways displayed in Fig. 1C, arachidonic acid metabolism, glycerophospholipid metabolism, fatty acid biosynthesis, and linoleic acid metabolism were significantly enriched in GGO patients. (Fig. 1C). Univariate Cox analysis was employed to identify the prognosis-associated genes (Fig. S1B), which were further restricted to eight glycerophospholipid metabolism–related candidates via LASSO regression, indicating a significant association between glycerophospholipid metabolism and LUAD progression (Fig. S1C, D). An independent validation metabolomics dataset indicated the same result, showing that the key precursor in glycerophospholipid biosynthesis, glycerol 3-phosphate, was downregulated during LUAD progression (Fig. 1D, E), and significant abundance of glycerol 3-phosphate in pre-invasive LUAD with favorable prognosis. To further investigate the cellular origin of this dysregulation, we analyzed single-cell RNA-seq datasets GSE189357 and GSE117570, clustering and annotating the major cell populations (Fig. 1F-G, S1E). Using the irGSEA R package, gene sets consistently identified as differential by multiple algorithms were integrated using robust rank aggregation. We found that glycerophospholipid metabolism was predominantly enriched in epithelial cells (Figure S1F). Pathway activity at the single-cell level visualized by DimPlot further confirmed that metabolic activity was largely localized to epithelial cells (Fig. 1H, S1G). These findings suggest that the dysregulation of glycerophospholipid metabolism observed during the LUAD progression originates primarily from cancer epithelial cells.

Fig. 1.

Identification of glycerophospholipid metabolism dysregulation in lung adenocarcinoma progress by multi-omics. A Radiological manifestations of ground-glass opacity and solid nodule and the study design (AIS: adenocarcinoma in situ; MIA: minimally invasive adenocarcinoma; IAC: invasive adenocarcinoma cancer). B Kaplan–Meier survival curves of ground-glass opacity and solid nodule patients showing ground-glass opacity has better prognosis. C Heat map of enrichment scores of 13 lipid pathways showing four lipid pathways enriching in patients with ground-glass opacity. Color bar showing Normalized Enrichment Score of ssgsea result. D Kaplan–Meier survival curves of ground-glass patients with different Glycerol 3-phosphate concentration (G3P) showing patients with higher G3P concentration have better prognosis. E Differences in Glycerol 3-phosphate concentration of AAH/AIS/MIA and IAC. F UMAP clustering plot of the single cell data GSE189357. G UMAP of the marker gene of GSE189357 cohort. H. Enrichment of glycerophospholipid metabolism in different cells in GSE189357 cohort. *P < 0.05, **P < 0.01, and ***P < 0.001

GPD1L regulates glycerophospholipid metabolism and is downregulated during LUAD progression

GPD1L encodes a key initiating enzyme involved in glycerophospholipid metabolism and is a member of the NAD-dependent glycerol-3-phosphate dehydrogenase family. It catalyzes the conversion of dihydroxyacetone phosphate and reduced nicotine adenine dinucleotide to glycerol-3-phosphate (G3P) and NAD⁺ [36] (Fig. 2A). Correlation analysis revealed a significant positive association between GPD1L expression and glycerophospholipid metabolism (Fig. 2B). Survival analysis based on both our cohort and TCGA dataset demonstrated that low GPD1L expression was associated with poorer prognosis in LUAD patients (Fig. 2C, S2A-C). In line with these findings, GSEA showed that the high GPD1L expression group was significantly enriched in glycerophospholipid metabolism pathways (Fig. 2D). GPD1L expression is markedly higher in the GGO group compared to the solid nodule group (Fig. 2E), these results suggested that enhanced glycerophospholipid metabolism in GGO is partly drivenby elevated GPD1L expression. Single-cell RNA-seq analysis revealed that GPD1L is predominantly expressed in epithelial cells, with approximately 20% of epithelial cells expressing detectable levels of GPD1L (Fig. 2F, S2E). To further investigate the dynamic changes in GPD1L expression during tumor progression, we performed trajectory analysis on scRNA-seq data. All epithelial cells were selected for pseudotime analysis, enabling reconstruction of a continuous developmental trajectory (Fig. 2G, S2F). As LUAD progressed, particularly during the transition from AIS or MIA to IAC (Fig. 2H), GPD1L expression displayed a marked downward trend (Fig. 2I, S2G). This trend was consistent with differential expression analyses showing reduced GPD1L levels in LUAD tumor samples compared to adjacent normal tissue, as well as across several other cancer types (Fig. S2D, H). To account for the potential influence of tumor purity on the observed decrease in GPD1L expression in solid nodules, we conducted a linear regression analysis with tumor purity as a covariate. The analysis revealed that the regression coefficient for the comparison between solid nodules and GGOs was −4.518 (SE = 0.806, t = −5.606, P = 1.38 × 10⁻⁷), demonstrating that GPD1L expression remained significantly lower in solid nodules compared with GGOs after adjustment for tumor purity (Fig. 2J). Moreover, clinical correlation analysis within our cohort demonstrated that lower GPD1L expression is associated with higher T and N stages (8th Edition TNM Staging System for Lung Cancer), as well as poorer tumor differentiation (Fig. 2K-M), supporting the hypothesis that GPD1L downregulation contributes to LUAD progression.

Fig. 2.

GPD1L mediates glycerophospholipid metabolism and downregulates during tumor progression. A GPD1L playing an important role in glycerophospholipid metabolism as the initiate enzyme. B Significant correlation between GPD1L expression and enrichment score of glycerophospholipid metabolism of FUSCC cohort with r = 0.53 and p < 0.001. C Kaplan–Meier survival curve showing patients with high GPD1L expression have a better prognosis. D GSEA analysis showing glycerophospholipid metabolism enriching in GPD1L high expression group. Color bar showing p.value while dot size showing the gene numbers of the pathways. E GPD1L expression in patients with ground-glass opacity and Solid nodule. F GPD1L expression and fraction in different cells of GSE189357. color bar showing expression after normalization while dot size showing the percentage of cells expressing GPD1L. G. Differentiation trajectory plot of epithelial cell subtypes showing the transition from AIS/MIA to IAC. H Pesudotime plot of epithelial cell subtypes showing the transition from AIS/MIA to IAC. I GPD1L expression in pseudo-time of epithelial cell subtypes verifies the downregulation of GPD1L during progression. J GPD1L expression in patients with ground-glass opacity and Solid nodule after being adjusted for tumor purity. Boxplot of GPD1L expression in different stage and subgroup showing GPD1L downregulating during LUAD progress (K, L, M). *P < 0.05, **P < 0.01, and ***P < 0.001

GPD1L-mediated glycerophospholipid metabolism dysregulation promotes LUAD Progression

To further validate the functional of GPD1L, we performed GPD1L knockdown in PC9 cells using siRNAs and confirmed knockdown efficiency by qPCR and Western blot (Fig. 3A, B). As demonstrated by CCK-8 assays and colony formation assay, knockdown significantly enhanced tumor cell proliferation (Fig. 3C-F). Additionally, transwell migration assays revealed the increased migratory capacity in knockdown cells (Fig. 3G-H), which is consistent with clinical observations showing a correlation between low GPD1L expression and advanced N stage (Fig. 2L). Untargeted metabolomic profiling of PC9 cells identified glycerophospholipid metabolism as one of the top 10 significantly enriched pathways (impact value > 25%, p-value < 0.05), supporting the notion that GPD1L downregulation contributes to the glycerophospholipid metabolism dysregulation during LUAD progression (Fig. 3I). GPD1L knockdown was performed in KP cell line by shRNA (Figure S3A, B) and GPD1L-deficient tumors exhibited enhanced growth compared to controls both in vivo and in vitro (Fig. 3J-K, S3C, D). Collectively, both in vitro and in vivo data demonstrated that GPD1L downregulation promotes LUAD progression.

Fig. 3.

Validation of downregulation of GPD1L promotes lung adenocarcinoma progress. A Knockdown efficiency of GPD1L in PC9 cells confirmed by qPCR with GAPDH as control. B Knockdown efficiency of GPD1L in PC9 cells confirmed by Western blot with β-Actin as a loading control. Knockdown of GPD1L in PC9 and A549 cells promoting the proliferation, verified by CCK-8 assay (n = 3) (C, D), colony formation (n = 3) (E, F). Transwell assay showing knockdown of GPD1L promotes the migration of PC9 and cells (n = 3) (G, H). I Pathway enrichment of metabolomics result between NC and si-GPD1L PC9 cell line, verifying GPD1L’s impact on glycerophospholipid. Subcutaneous tumors of KP cell line showing knockdown GPD1L promoting tumor progress in vivo (J, K). *P < 0.05, **P < 0.01, and ***P < 0.001

GPD1L downregulation promotes tumor progression via stabilization of HIF-1α

As LUAD progresses from GGO to solid nodules, increased tumor cellularity and reduced vascular infiltration contribute to a hypoxic tumor microenvironment. Hypoxia is one of the significantly enriched pathways in epithelial cells of invasive LUAD (Fig. 4A). Previous studies have suggested that GPD1L can modulate the activity of the prolyl hydroxylase domain enzymes (PHDs) [37]. Under normoxic conditions, PHDs hydroxylates specific proline residues on HIF-1α, targeting it for degradation [38]. Based on this, we hypothesized that GPD1L downregulation may enhance the hypoxia tolerance and anti-apoptotic capacity by indirectly mediating HIF-1α degradation. GSEA analysis comparing low- and high-GPD1L expression groups revealed significant enrichment of hypoxia and glycolysis hallmarks, both HIF-1α-regulated pathways, in the low-GPD1L group (Fig. 4B, S4A). Consistently, HIF1A mRNA level was elevated in the low-GPD1L group and increased during LUAD progression (Fig. 4C, D). We then examined HIF- 1α protein level by western blot after 8 h hypoxia treatment. GPD1L knockdown resulted in reduced HIF-1α degradation and lower levels of cleaved caspase-3 (p17 fragment), indicating diminished apoptosis in GPD1L-deficient PC9 cells (Fig. 4E). To validate these findings, we assessed apoptotic cell proportion under hypoxic conditions. Flow cytometry after hypoxia treatment demonstrated significantly reduction in apoptotic cells in si-GPD1L group under hypoxia condition (Fig. 4F-H).

Fig. 4.

Downregulation of GPD1L reduces the degradation of HIF-1α and influences GPX4. A Enrichment pathways of IAC epithelial cells of single cell data compared to the AIS/MIA epithelial cells. B GSEA enrichment plots showing that hallmark of hypoxia enrich in low GPD1L group of FUSCC cohort. C HIF1A expression in low and high GPD1L expression group. D HIF1A expression during pesudotime of the single cell data. E GPD1L knockdown promoting the expression of HIF-1α and reduced Caspase-3 P17 in PC9 cell line verified by the western blot, with β-Actin as a loading control. Flow cytometry showing blocking GPD1L giving PC9 cells resistance to hypoxia (24 h), making less apoptosis compared to the control group (F, G). H.PC9 cells after 36 h hypoxia, making less apoptosis compared to the control group. I GPD1L knockdown and DMOG promoting the expression of HIF-1α but combined use having no synergistic effect. J GPD1L knockdown increasing the stability of HIF-1α protein after exposure to oxygen. K GPD1L restoration effectively attenuating HIF-1α hydroxylation when PHD activity was intact, but this regulatory effect was lost following DMOG-mediated PHD inhibition. L Difference in reduced L-Glutathione (Normalized) of NC and si-GPD1L PC9 cells. M GPD1L knockdown reducing GPX4 but having no influence on SLC3A2 in PC9 cell line, showed in the western blot with β-Actin as a loading control. *P < 0.05, **P < 0.01, and ***P < 0.001

To further elucidate the mechanism by which GPD1L regulates HIF-1α, we employed DMOG, a pharmacological inhibitor of PHDs. Knockdown of GPD1L alone led to an increase in HIF-1α protein levels; however, co-treatment with DMOG did not further enhance HIF-1α accumulation (F ig. 4I), suggesting that GPD1L and PHD inhibition act on a shared regulatory axis. Silencing of GPD1L did not significantly alter HIF1A mRNA expression (Fig. S4E), indicating that GPD1L regulates HIF-1α predominantly at the post-transcriptional level. Consistent with this, GPD1L knockdown markedly increased HIF-1α protein stability after exposure to oxygen (Fig. 4J). These findings imply that loss of GPD1L enhances HIF-1α stabilization through inhibition of PHD-dependent hydroxylation. To further validate this mechanism, rescue experiments were performed in shGPD1L cells. Restoration of GPD1L expression effectively promoted HIF-1α hydroxylation in the absence of PHD inhibition, whereas this effect was abolished upon DMOG treatment (Fig. 4K). Together, these results demonstrate that GPD1L regulates HIF-1α stability primarily through a PHD-dependent hydroxylation pathway.

Solid tumors with low GPD1L expression exhibited increased HIF-1α accumulation and suppressed apoptosis-related pathways than GGO nodules, conferring resistance to hypoxia and apoptosis during LUAD progression. This mechanism may also underlie the radiological transition from GGO to solid nodules, as increased cellular density during tumor growth requires adaptation to hypoxic stress. Our data suggest that GPD1L downregulation facilitates this transition by promoting hypoxia tolerance through HIF-1α stabilization.

GPX4 is a potential therapeutic target in low-GPD1L LUAD patients

Beyond hypoxia and apoptosis resistance, GPD1L downregulation may also modulate other forms of cell death. To explore this possibility, we next examined whether ferroptosis, another regulated cell death pathway, was affected by GPD1L expression. The MDA assay demonstrated that GPD1L knockdown markedly decreased cellular resistance to ferroptosis (Fig. S4B, C). Metabolomic profiling of PC9 cells also revealed a decrease in reduced glutathione (GSH) levels, a key ferroptosis resistance mediator [39, 40] (Fig. 4L). Given this finding, we focused on the ferroptosis related gene expression. Notably, GPX4—a central regulator of ferroptosis resistance—was significantly downregulated in the low-GPD1L expression group and showed a similar decreasing trend during LUAD progression in bulk and single-cell transcriptomic data (Fig. 4M, S4D, F). To further investigate the effect of GPD1L on the ferroptosis resistance pathway, we analyze the expression of SLC3A2, another key component involved in ferroptosis regulation. However, SLC3A2 expression was not significantly affected by GPD1L knockdown (Fig. 4M, S4G).

Collectively, these findings indicate that GPD1L downregulation impairs ferroptosis resistance by suppressing GPX4 expression, rendering low-GPD1L LUAD tumors potentially vulnerable to ferroptosis-inducing therapies.

GPD1L positively regulates GPX4 via the AKT/NRF2/GPX4 pathway

To elucidate the mechanism by which GPD1L regulates GPX4 expression, we first performed transcription factors analysis by pySCENIC analysis on epithelial cells stratified by GPD1L expression levels in the single-cell dataset. This analysis revealed reduced transcriptional activity of NRF2 in the GPD1L-low subpopulation (Fig. 5A). Differentially expressed genes in the bulk RNA-seq data further supported this finding, showing significantly lower NRF2 expression in the low-GPD1L group, while AKT1, AKT2, and AKT3 expression remained unchanged (Fig. 5B). Given that GPD1L is involved in the glycerol phosphate shuttle, we hypothesized that knockdown of GPD1L may influence the ATP generation [41], thereby attenuating AKT phosphorylation. To rule out upstream regulatory effects, we examined the PI3K/AKT pathway. The expression of PIK3CA, a catalytic subunit of PI3K, did not show significant difference between groups (Fig. 5C). Moreover, there were no notable differences in the mutational status of PI3K or NRF2 pathway genes, despite broader genomic variation across samples (Fig. 5D, S5A, B). We next performed western blot to verify the bioinformatic result and it revealed that blocking GPD1L had no impact on the pan-AKT, while significantly reduced the phosphorylation of AKT and NRF2 protein levels (Fig. 5E).

Fig. 5.

GPD1L regulates GPX4 via AKT/NRF2/GPX4 pathway. A Transcription factor analysis (TF) with significant functional differences in low and high GPD1L expression epithelial cells of GSE189357 single cell data (“ + ” presents active TFs; Regulon Specificity Score, RSS). B Difference in NFE2L2 in high and low GPD1L group, but AKT1, AKT2 and AKT 3 showing no difference in our cohort. C PIK3CA expression showing no difference in high and low GPD1L group. D Mutation pathways status showing no difference in high and low GPD1L group of FUSCC cohort. E Western blot showing that knockdown of GPD1L reduced the phosphorylation of AKT and then affected AKT/NRF2 pathway, with β-Actin as a loading control. F Dose response curve showing the resistance of A549 to ML162 after blocking the GPD1L (n = 3). G Western blot showing that AKT agonist Sc79 rescued the downregulation of NRF2 by knockdown of GPD1L, with β-Actin as a loading control. H AKT agonist Sc79 rescued the downregulation of GPX4 caused by knockdown of GPD1L, giving the resistance to the ML162, while AKT inhibitor Capivasertib enhancing the downregulation (n = 3). Combined use of Capivasertib promoted ML162 lethality to A549 cells with the knockdown of GPD1L (n = 3) (I, J). K IHC staining representing low and high GPD1L expression. L Kaplan–Meier survival curve showing low expression of GPD1L, evaluated by IHC staining, is a poor prognostic marker. M Plasma glycerophospholipid metabolites G3PC concentration of metabolomic data decreasing with the lung adenocarcinoma progress. *P < 0.05, **P < 0.01, and ***P < 0.001

To further confirm the functional relationship between GPD1L and GPX4, we employed ML162, a covalent inhibitor of GPX4 [42] known for selective lethality in A549 cells. We performed knockdown of GPD1L in A549 cells using the same siRNAs described above, which resulted in significantly increased sensitivity to ML162 compared with the negative controls. The IC₅₀ of ML162 in the GPD1L knockdown group dropped to 0.2396 and 0.6692 μM, compared to 2.041 μM in the control, demonstrating an approximate ten-fold difference (Fig. 5F). These findings confirmed enhanced ferroptosis susceptibility and GPX4 as a potential target in GPD1L-deficient cells. We then used AKT inhibitor Capivasertib to enhance the influence and AKT agonist Sc79 to rescue the influence induced by blocking GPD1L. In ML162 sensitivity assays, Capivasertib-treated cells (1 μM/L) exhibited an IC50 of 0.05504 μM/L, while Sc79 treatment (5 μM/L) increased the IC50 to 1.115 μM/L, similar to control levels (Fig. 5G), indicating the rescue of GPX4 protein level. Western blot results confirmed that Sc79 also restored NRF2 expression in GPD1L-deficient cells (Fig. 5H), further validating the role of the AKT/NRF2 axis in regulating GPX4.

Finally, co-treatment with Capivasertib and ML162 produced a synergistic cytotoxic effect in vitro (Fig. 5I, J), suggesting that targeting the AKT/NRF2/GPX4 pathway may offer a promising therapeutic strategy for GPD1L-deficient LUAD.

Low GPD1L expression serves as a poor prognostic marker

To evaluate the clinical significance of GPD1L, we assessed its expression in lung adenocarcinoma tissues using immunohistochemical (IHC) staining on a tissue microarray (Fig. 5K). After excluding samples with incomplete IHC data or missing clinical information, a total of 149 cases were included. Patients were stratified into high- and low-GPD1L expression groups based on the median IHC score. Kaplan–Meier analysis revealed that the patients with low GPD1L expression had a significantly worse prognosis (Fig. 5L). These results indicate that reduced GPD1L expression is associated with poor clinical outcomes and support its potential as a prognostic biomarker in LUAD. Additionally, the plasma level of Glycero-3-phosphocholine also exhibited a decreasing trend during the tumor progression (Fig. 5M), suggesting that the plasma glycerophospholipid metabolites may possess similar prognostic utility and could be explored as non-invasive biomarkers for monitoring lung adenocarcinoma progression.

GPD1L as a potential biomarker for immunotherapy response

Flow cytometry analysis of subcutaneous tumors revealed a significant reduction in CD8⁺ T cell infiltration in the shGPD1L group, suggesting that GPD1L downregulation may facilitate tumor immune evasion and contribute to disease progression (Fig. 6A, S6A). This aligns with growing evidence that tumor metabolic reprogramming influences the tumor microenvironment and alters intercellular interactions [17, 18]. Cell communication analysis by Cell chat conducted in single cell data also revealed that epithelial cells of IAC have communications with T cells, while T cells exhibited limited suppressive activity (Fig. 6B, S6B, C). EPIC analysis stratified by GPD1L expression demonstrated the same results that the infiltration level of cancer associated fibroblasts (CAFs) and CD8 + T cells differed between high and low- GPD1l expression groups (Figure S6D). Correlation analysis further confirmed that GPD1L expression was significantly associated with CAF and CD8⁺ T cell infiltration (Fig. 6C, S6E). Further clustering of T and NK cells in the single-cell dataset showed a marked decrease in CD8⁺ T cells in IAC (Fig. 6D-F). Given that HIF-1α have been indicated to regulate PD-L1 expression [43, 44], we analyzed CD274 (encoding PD-L1) expression (Fig. 6G). Both low-GPD1L expression and reduced glycerophospholipid metabolism were associated with elevated CD274 expression (Fig. 6H). Western blot data confirmed the GPD1L downregulation may have an impact on PD-L1 via HIF-1α (Fig. 6I). To further assess the potential involvement of GPD1L in tumor–immune interactions, PC9 cells were co-cultured with activated T cells. Knockdown of GPD1L in tumor cells was associated with increased tumor cell viability compared with control cells, suggesting reduced sensitivity to T cell–mediated cytotoxicity (Figure S6F). In parallel, IFN-γ levels in the culture supernatant were decreased in the GPD1L knockdown group (Figure S6G). Treatment with a PD-L1–blocking antibody partially reversed the increased tumor cell viability and restored IFN-γ production. These findings suggest that downregulation of GPD1L is associated with altered tumor–T cell interactions and may contribute to immune modulation in the tumor microenvironment, potentially involving PD-L1–related pathways.

Fig. 6.

Potential use as immune therapy marker of GPD1L. A Flow cytometry of subcutaneous tumors showing CD8 + T cells filtrating less in shGPD1L tumors. B Communication of metabolism pathway in different cell groups of GSE189357. C Correlation analysis showing GPD1L significant associated with CD8 + T cells infiltration. D UMAP clustering plot of T cells of the single cell data GSE189357. E UMAP of the marker gene of T cells of GSE189357 cohort. F Filtration percentage of different T cells of GSE189357 cohort. G Correlation analysis showing HIF1A expression significantly associated with CD274 expression. H CD274 upregulating in both low GPD1L expression group and low glycerophospholipid metabolism group. I GPD1L knockdown promoting the expression of PD-L1 in PC9 cell line verified by the western blot, with β-Actin as a loading control. J Boxplot of GPD1L expression of responder and non_responder. K AUC value of using GPD1L as biomarker for predicting response of immunotherapy. *P < 0.05, **P < 0.01, and ***P < 0.001

Furthermore, the increased CAF infiltration associated with low GPD1L expression may impede CD8⁺ T cell infiltration, while PD-L1 upregulation could contribute to CD8⁺ T cell exhaustion, orchestrating an immunosuppressive microenvironment. To evaluate the clinical relevance of these findings, we examined the independent immunotherapy dataset GSE33072 demonstrating low GPD1L expression was associated with poorer immunotherapy outcomes (Fig. 6J, K). TIDE prediction further indicated these results, that patients with low GPD1L expression showing higher TIDE scores, indicating a worse outcome with limited response to immunotherapy (Figure S6H, I). Together, these results suggest that GPD1L expression may serve not only as a prognostic biomarker but also as a predictive marker of immunotherapy response in lung adenocarcinoma. Low GPD1L expression may identify patients less likely to benefit from immune checkpoint inhibitors, underscoring the importance of incorporating metabolic markers into immunotherapeutic decision-making.

Discussion

Glycerophospholipid metabolism dysregulation in the progress of cancer has been implicated in previous studies [45, 46]. These studies mainly focus on its intrinsic metabolic roles and therapeutic targeting strategies [47]. However, due to the essential role of glycerophospholipid in maintain cell membrane integrity and mitochondrial respiration, direct therapeutic interventions targeting this pathway face considerable limitations [48]. In contrast to those studies, our study identifies the role of GPD1L-mediated glycerophospholipid metabolism in lung cancer progression, particularly during the transition from ground-glass opacity to solid nodule. Through multi-omics analysis, including transcriptomics, metabolomics, and single-cell sequencing, we demonstrated that glycerophospholipid metabolism is significantly downregulated in solid nodules compared to ground-glass opacity. Furthermore, we identified GPD1L as a key regulatory gene in this metabolic pathway and demonstrated its crucial role in tumor progression, hypoxia response, and therapeutic targeting in patients with low GPD1L expression.

Notably, we identified four significantly altered lipid metabolism pathways, suggesting a fundamental metabolic shift in tumor evolution, consistent with the previous studies [49]. Metabolomics analysis further supports these findings, demonstrating a significant reduction in key intermediates such as glycerol-3-phosphate and glycero-3-phosphocholine in progress cancers. Among glycerophospholipid metabolism-related genes, GPD1L emerged as a critical regulator. As an enzyme catalyzing the conversion of dihydroxyacetone phosphate to glycerol-3-phosphate, GPD1L plays a crucial role in lipid homeostasis. Our cohort analysis and TCGA validation collectively showed that low GPD1L expression is associated with poor prognosis, while trajectory analysis of single-cell data revealed its progressive downregulation along tumor evolution, especially in epithelial subpopulations. Functional assays confirmed the tumor-suppressive role of GPD1L, as its knockdown significantly enhanced LUAD cell proliferation, migration, and colony-forming ability.

Beyond its metabolic role, GPD1L downregulation contributes to lung adenocarcinoma progression by stabilizing HIF-1α. Mechanistically, GPD1L suppression inhibits HIF-1α degradation via impaired prolyl hydroxylation, thereby enhancing hypoxia tolerance and suppressing apoptosis [50, 51]. This adaptation suits tumor transition from ground-glass opacity to solid nodule and other progression, allowing tumor cells to thrive in a hypoxic and dense microenvironment. Importantly, GPD1L-low tumors exhibited ferroptosis vulnerability, as evidenced by reduced levels of glutathione and downregulation of GPX4, the key ferroptosis suppressor. Pharmacological inhibition of GPX4 sensitized GPD1L-deficient cells to ferroptosis, suggesting a potential therapeutic strategy. Experiment and RNA-seq analysis confirmed that GPD1L regulates GPX4 expression via the AKT/NRF2/GPX4 pathway, highlighting a potential therapeutic strategy for targeting GPD1L-low tumors.

Clinically, low GPD1L expression was associated with higher T stage, N stage, and poor differentiation, reinforcing its prognostic significance. Plasma glycerophospholipid metabolites also showed a downward trend during tumor progression, highlighting their potential as non-invasive biomarkers. Furthermore, single-cell and bulk transcriptomic analysis indicated that GPD1L may modulate immune cell infiltration, particularly CD8 + T cells, and potentially modulate response to immunotherapy, as supported by immune therapy cohorts. Given that tumor metabolism can influence immune cells, GPD1L-mediated glycerophospholipid metabolism dysregulation may also impact tumor microenvironment. These findings suggest that targeting GPD1L-associated vulnerabilities could offer novel therapeutic and diagnostic opportunities in LUAD.

Compared to the published studies, our work elucidates the mechanism of GPD1L tumor suppress role in lung adenocarcinoma. Furthermore, we verified the ferroptosis vulnerability and its mechanism in GPD1L-Low patients. But there still remains problems to solve. Since GPD1L is functionally connected to the glycerol phosphate shuttle and oxidative phosphorylation [52, 53], its role in modulating mitochondrial function and reactive oxygen species (ROS) production warrants further exploration [39, 54, 55]. The precise mechanisms by which GPD1L influences the immune landscape remain to be elucidated. In addition, recent evidence has identified FOXO1 as a key transcription factor regulating metabolic and redox-related gene programs in lung adenocarcinoma, with its activity tightly controlled by the PI3K–AKT signaling pathway. Given the close association between FOXO1 signaling, lipid metabolism, and hypoxia-related pathways, dysregulated FOXO1 activity may contribute to the reduced GPD1L expression observed during LUAD progression [56]. Our ongoing work will focus on how GPD1L affects oxidative phosphorylation and immune modulation within the tumor microenvironment.

In conclusion, our study revealed the glycerophospholipid metabolism dysregulation during lung adenocarcinoma progression, especially from ground-glass opacity to solid nodules. GPD1L acts as a tumor suppressor, regulating HIF-1α degradation and ferroptosis susceptibility via the AKT/NRF2/GPX4 pathway. Its downregulation not only contributes to tumor progression and hypoxia resistance but also presents a targetable vulnerability in LUAD. Furthermore, GPD1L expression may serve as a valuable prognostic and immunotherapeutic biomarker, offering new insights into metabolic-immune crosstalk and therapeutic stratification in lung cancer.

Authors’ contributions

YS and YR designed the research and wrote the paper. YS and CL analyzed the data. YS and YR performed experiments. YZ and YC participated in the validation. YB helped with the visualization. YZ, HH and HC revised the manuscript and guided the work. All authors read and approved the manuscript.

Funding

The study was supported by National Key R&D Program of China (2022YFA1103900), National Natural Science Foundation of China (82430099 and 82403923), the Clinical Research Special Project of Shanghai Municipal Health Commission (202440070), Shanghai Pujiang Program (24PJD017), China Postdoctoral Science Fundation (2025T180572) and the Medical Research Special Project for Shanghai Science and Technology Innovation Action Plan (24Y12800400).

Data availability

All data generated or analyzed during this study were included in this article’s methods section.

Declarations

Ethics approval and consent to participate

This study has been approved by the research ethics review committee of Fudan University Shanghai Cancer Center (FUSCC) Institutional Review Board (No. 090977–1).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.