3.1 Signature construction of lipid metabolism and survival analysis

We downloaded 7 dataset of bulk transcriptome data from GSE37745, GSE19188, GSE30219, GSE31547, GSE41271, GSE42127, GSE72094 in GEO database. Single cell dataset of GSE117570 was obtained from TISCH2 database. LogNormalize was used to standardize the data. To construct a lipid metabolism-related gene signature for predicting prognosis in LUAD, we performed a detailed literature review to identify a set of relevant genes (Table S1). The co-expression pattern of these genes was shown in the form of heatmap (Fig. 1A), and the conservative lipid metabolism subtypes were distinguished using non-negative matrix factorization. The best rank was determined to be 3, which divided 510 LUAD samples in TCGA dataset into three conserved subtypes (Fig. 1B). Survival analysis showed that the grouping based on lipid metabolism-related genes had a prognostic effect. Group 3 showed the best prognosis, and groups 1 and 2 showed poor prognosis (Fig. 1C). We further analyzed the molecular characteristics of lipid metabolism in different groups by showing the expression of lipid metabolism genes in LUAD samples using a box plot (Fig. 1D). We compare them in pairs, and the number above the horizontal line represents P value. GPD1L, PLA2G12B, PLA2G3 and PLA2G1B were markedly up-regulated in the group 3, comparing to the group 1 and 2. On the other hand, MBOAT7 and PTDSS1 showed an enhanced expression in group 1 and 2. In addition, LPCAT1 was uniquely elevated in the group 2.

Fig. 1

Cluster analysis of lipid metabolism-related genes in LUAD. A The co-expression pattern of lipid metabolism-related genes. B 510 lung adenocarcinoma samples in TCGA-LUAD were classified into three conserved subtypes according to the lipid metabolism-related genes. C Prognostic analysis of three conserved subtypes. D The expression of lipid metabolism-related genes in TCGA-LUAD samples

3.2 Prognostic analysis and model construction of lipid-score

We performed univariate cox regression on all lipid metabolism-related genes to evaluate their impacts on the prognosis of LUAD patients. Significant risk factors and protective factors were selected for model construction (Fig. 2A). The risk factors included LPGAT1, PTDSS1 and MBOAT7. The protective factors included ACHE, PLA2G12B, PLA2G1B, MBOAT1, PLA2G3, LPCAT2, CHKB, PLA2G4B, GPAM, LPCAT1, PLA2G15, CRLS1, CHKA, GPD1L and PGS1. The GSVA algorithm was used to calculate the score of the samples in TCGA-LUAD based on the above gene expression, named Lipid-score. In TCGA lung adenocarcinoma samples, Lipid-score showed excellent ability to predict prognosis (Fig. 2B). Subsequently, we found that Lipid-score exhibited excellent and consistent prognostic value across seven independent validation sets (GSE37745, GSE19188, GSE30219, GSE31547, GSE41271, GSE42127 and GSE72094) (Fig. 2C–I). These results demonstrate the robust performance of our Lipid-score in predicting the prognosis of LUAD patients.

Fig. 2

Prognostic analysis and construction of Lipid-score prediction model. A Univariate cox regression model. Significant risk factors (red) and protective factors (green) genes were selected to construct the model. The value of Log (Hazard Ratio) was showed in diaphragm. B Prognostic analysis of low and high Lipid-score groups. C–I The verification of Lipid-score in 7 independent validation sets. The p-value were showed in each diaphragm

3.3 Analysis of immune characteristics of lipid-score

To further elucidate the influences of lipid metabolic signaling on the tumor immune microenvironment, we performed a multidimensional immune cell infiltration analysis. The results showed that M2 macrophages, cDCs (classical DCs), and hematopoietic stem cells (HSCs) were inversely correlated with lipid scores in all groups. We found that Lipid-score exhibited general immunosuppressive efficacy across multiple datasets by using various immune infiltration algorithms (Fig. 3A–D). Specifically, activated CD4 T cells, activated CD8 T cells and CD56 NK cells generally displayed negative correlation with the Lipid-score. Moreover, cancer-associated fibroblasts (CAFs) were also negatively correlated with score and the endothelial cells showed a positive correlation with the score in 5 out of 6 cohorts. The correlation between markers of immunotherapy effect and lipid score showed that high Lipid-scores were mainly negatively related with Tumor Mutation Burden (TMB), gene expression profile (GEP) and PDL1 which implied that the patients with high Lipid-scores may be less likely to benefit from immunotherapy (Fig. 3E–H).

Fig. 3

Immune characteristics of Lipid-score. A–D Enrichment analysis of immune infiltration in validation sets. E–H Spearman correlation between immunotherapy indicators and Lipid-score

3.4 Drug sensitivity analysis and the predictive effects of lipid-score in immunotherapy

We identified the distribution of Lipid-score in different cancer cell lines in the GDSC cancer cell line database (Fig. 4A). We performed drug sensitivity analysis based on Lipid-score, and the results revealed a variety of drugs positively correlated with Lipid-score, suggesting the possibility of drug combination (Fig. 4B, C). The drugs and corresponding targets and pathways are shown in Fig. 4D. To validate the predictive performance of Lipid-score on curative effect of immunotherapy, we calculated the Lipid-scores in PMID: 26359337, PMID: 32472114, PRJEB23709, Phs000452 data sets, which all contain data of immunotherapy efficacy. The results demonstrated the ability of Lipid-scores to distinguish the efficacy of immunotherapy (chi-square test, Fig. 5A, B). Based on Lipid-score grouping, the high Lipid-score group showed a lower progression-free survival (Fig. 5C–F).

Fig. 4

Drug sensitivity analysis based on Lipid-score. A The distribution of Lipid-score in different cancer cell lines in the GDSC cancer cell line database. B, C Drug sensitivity analysis identified a variety of drugs positively correlated with the Lipid-score. D Sankey diagram showing drugs and corresponding targets and pathways

Fig. 5

The predictive performances of Lipid-score in immunotherapy. A, B The distinguishing ability of Lipid-score groups in the response of immunotherapy: Van Allen CTLA4 Science study (A) and phs000452 study (B). C–F Prognostic analysis of low and high Lipid-score groups in different immunotherapy cohorts

3.5 Single-cell transcriptome analysis reveals diverse involvement of lipid metabolism pathway signals in LUAD microenvironment

By utilizing UMAP dimensionality reduction, we annotated malignant cells, stromal cells, and immune cells including B, Th2, CD8 T effector cells, pDCs (plasmacytoid dendritic cells), endothelial cells, epithelial cells, malignant cells, M1 macrophages, M2 macrophages, Monocytes, NK cells, and plasma cells (Fig. 6A). GSVA was used to calculate the distribution of Lipid-score. It was found that the Lipid-score was distributed in various cell types (Fig. 6B), suggesting that lipid metabolism exists in various cells in the tumor microenvironment. The dotplot diagram demonstrated a relatively high expression of Lipid-score in monocytes, epithelial cells and pDCs (Fig. 6C). We further characterized the functions of two groups based on the median of Lipid-score in the single-cell transcriptome dataset. GSEA results revealed the enrichment of Lipid-score in antigen presentation, immune activation, interferon signaling, and other immune pathways (Fig. 6D). The distribution of Lipid-score genes with negative prognostic effects, which is LPGAT1, PTDSS1 and MBOAT7, was particularly evident in malignant cells and macrophages (Fig. 6E).

Fig. 6

Single-cell transcriptome analysis of Lipid-score in LUAD. A The landscape of the cell types in LUAD. B, C Distribution of Lipid-score in different LUAD cell types. D The results of GSEA enrichment analysis. E Expressions of marker genes in cell subsets

3.6 Analysis of lipid-score associated intercellular communication processes

We utilized NicheNet to analyze intercellular communication differences between two groups based on the Lipid-score. We identified ligand molecules with high correlation, among which APOE showed the highest correlation and is closely linked to lipid metabolism (Fig. 7A). These ligands were predominantly expressed in myeloid cells including M2 macrophages and monocytes, suggesting their potential roles as functional carriers of immunosuppression (Fig. 7B, C). Downstream target analysis of these molecules revealed functional information related to Lipid-score mediated immune tolerance and ICAM-related immune cell adhesion (Fig. 7D).

Fig. 7

Analysis of intercellular communication processes in different Lipid-score groups. A Highly correlated ligand molecules between Lipid-score groups. B, C Heatmap of the expressions of ligand molecules. D The regulatory potential of the downstream predicted genes of ligand molecules

3.7 PTDSS1-knockdown tampered the proliferation and facilitated the apoptosis of LUAD

The phosphatidylserine synthase 1 (PTDSS1) is a key gene in risk score model. In order to further reveal the potential roles and mechanisms of PTDSS1 in the development of LUAD, we further conducted in vitro experiments. We used A549 and NCI-H1975 cell lines for in vitro experiments. We transfected PTDSS1 siRNA into these cell lines and verified the knockdown effects by applying western blot. We found that PTDSS1 was significantly reduced in siRNA groups (Fig. 8A, B). Edu assays revealed that LUAD cell proliferation was significantly inhibited after reduction of PTDSS1 expression (Fig. 8C, D). Colony formation assays also showed that knockdown of PTDSS1 could inhibit the proliferation of LUAD cells (Fig. 8E, F). Furthermore, we investigated whether PTDSS1 affects apoptosis in LUAD cells. LUAD cells showed a significant increase in apoptotic cells after the knockdown of PTDSS1 (Fig. 8G, H).

Fig. 8

PTDSS1 affected proliferation and apoptosis in LUAD. A, B Western blot showed that PTDSS1 was successfully knockdown by siRNA. C, D Edu assays showed that proliferation of A549 and NCI-H1975 cells were inhibited by si-PTDSS1. E, F Clonal formation assays showed that proliferation of A549 and NCI-H1975 cells were inhibited by si-PTDSS1. G, H Flow cytometry assessing the LUAD cell apoptosis