Choline metabolism reprogramming mediates an immunosuppressive microenvironment in non-small cell lung cancer (NSCLC) by promoting tumor-associated macrophage functional polarization and endothelial cell proliferation

298 advanced and treatment-naïve NSCLC patients treated with immunotherapy at Sun Yat-sen University Cancer Center between August 2018 and April 2023 were screened in this retrospective, observational study. Exclusion criteria included: receiving prior anticancer therapy; lacking baseline or the alteration level of ChE after 2 cycles of immunotherapy; lacking efficacy and survival data; loss to follow-up. 21 patients who met the exclusion criteria were excluded and finally 277 eligible patients were enrolled in our study. The flow chart of patients screening was summarized in Additional file 1: Figure S1. Ethics approval for this study, including a waiver of informed consent, was achieved from Sun Yat-sen University Cancer Center Institutional Review Board (SL-B2022-680-02).

Patients baseline characteristics including age, gender, Eastern Cooperative Oncology Group performance status (ECOG-PS), smoking status, distant metastasis and treatment regimens were documented. In addition, baseline laboratory parameters (within 7 days prior to the treatment initiation) including cholinesterase, lactate dehydrogenase (LDH), serum albumin (ALB), C-reactive protein (CRP), serum amyloid A (SAA) and neutrophil–lymphocyte ratio (NLR = absolute neutrophil count/absolute lymphocyte count) were collected and analyzed. Meanwhile, ChE level after 8 [± 2] weeks after the initiation of immunotherapy was also collected and all the patients were then categorized into ChE-increased group and ChE-decreased group based on the alteration level of ChE.

Tumor response and progression were regularly performed according to Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST 1.1) per Response Evaluation Criteria. The primary endpoint was overall survival (OS) defined as the time from initiation of immunotherapy to death from any causes. The secondary endpoint was progression-free survival (PFS) calculated from initiation of immunotherapy to tumor-progression or death from any causes. Patients who had not progressed or are not deceased were censored at the time of the last follow up.

RNA sequencing (RNA-seq) data of 600 LUAD samples, including 513 cancer samples and 87 adjacent normal samples were downloaded from The Cancer Genome Atlas (TCGA) database (https://​portal.​gdc.​cancer.​gov/​projects/​TCGA-LUAD). The clinical information including age, gender, TNM stage, survival time and survival status were also obtained.

In addition, the ORIENT-11 study, served as a validation dataset, is a multi-center, randomized, double-blind, phase 3 study which enrolled 398 advanced or metastatic NSCLC patients in China [15], receiving sintilimab (anti-PD1) plus pemetrexed and platinum (experiment group) or placebo plus pemetrexed and platinum (control group). 227 patients lacking RNA-seq data and 58 patients in control group (did not receive anti-PD1 therapy) were excluded in our study. Finally, the RNA-seq data as well as the clinical information of the 113 patients from ORIENT-11 study was collected for further analysis.

Choline metabolism-related genes were obtained from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, AmiGO2 website and Reactome Pathway Databases (Additional file 2: Table S1). Univariate and multivariate Cox analysis were performed to identify significant and independent OS-associated choline metabolism-related genes in TCGA-LUAD database. The risk scores of each patient both in TCGA database and the orient-11 study (the validation cohort) calculated using the following formula.

Patients were then classified into high-risk and low-risk group based on the median level of risk score. R package of “survminer” were used to plot the risk curves and survival curves of two risk groups and “survivalROC” package were conducted to draw the ROC curves.

The "limma" R package was utilized to identify differentially expressed genes (DEGs) between the high-risk and low-risk groups in TCGA-LUAD. Using the R package “clusterProfiler” [16], GO function enrichment analysis and KEGG pathway enrichment analysis of the DEGs were performed. Additionally, the "msigdbr" R package was used to perform Gene Set Enrichment Analysis (GSEA) on the DEGs to analyze the signaling pathways enrichment in different risk-groups. "IOBR" R package [17] was used to calculate the immune infiltration score and metabolism score for high-risk and low-risk patients. The Tumor Immune Dysfunction and Exclusion (TIDE) score [18], a computational method to model two primary mechanisms of tumor immune evasion, were downloaded from the TIDE database (http://​tide.​dfci.​harvard.​edu/​) after following the instructions on the website to uploaded input data, and the TIDE scores in two risk groups were analyzed. Besides, the scoring data of immune checkpoint inhibitors (ICIs) treatment were downloaded from the Cancer Immunome Database (TCIA) (https://​tcia.​at/​) [19].

The single-cell dataset GSE207422 of resectable non-small cell lung cancer (NSCLC) before and after PD-1 blockade combined with chemotherapy was downloaded from Gene Expression Omnibus (GEO) database, and quality control was performed on the raw gene expression matrix using the Seurat R package. The following criteria were applied for filtering cells: exclusion of cells with less than 200 or more than 6000 expressed genes, those with the proportion of mitochondrial gene expression in UMI count was more than 20%, and filtering out genes expressed in fewer than 3 cells. After performing quality control on the raw gene expression matrix, we processed and analyzed the data using the following steps: Firstly, the transcript counts were log-transformed using the NormalizeData function. Subsequently, the FindVariableFeatures function was employed to select the top 2,000 genes with the highest variability within cells. Next, the ScaleData function was applied to normalize the gene expression data, ensuring comparability among genes. To reduce the dimensionality of the data and capture the major sources of variation, we performed linear Principal Component Analysis (PCA) using the RunPCA function from the Seurat package. To correct for batch effects, we employed the Harmony algorithm [20] based on the samples. For clustering analysis, we chose a clustering resolution of 0.6 to better distinguish distinct cell populations. Visualization of cell clusters was performed using the Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction technique. Finally, all cells were annotated based on the expression patterns of cell-specific marker genes.

To define the functional states of myeloid cell clusters, the Ucell was used to calculate functional scores for each cluster based on known functional genes specific to myeloid cells.

Monocle 3 [21] was applied to infer the cell differentiation trajectories of myeloid cell clusters. The functions "cluster_cells" and "learn_graph" were used to partition the myeloid cells and fit the principal graph, while the UMAP visualization was employed to visualize the differentiation trajectories. We selected CD14⁺ monocytes as the starting point to infer the cell differentiation trajectories of myeloid cells.

For the analysis of cell–cell interactions, the CellChat [22] was utilized to predict the communication patterns of cell–cell receptors and ligands from the single-cell transcriptomic data. Special attention was given to the communication strength between myeloid cells and other cells in the tumor microenvironment.

A total of 277 treatment-naïve patients with advanced non-small cell lung cancer, who were treated with first-line anti-PD1 therapy in Sun Yat-sen University Cancer Center between August 3, 2018, and April 19, 2023 were enrolled in the study. Patients ranged in age from 32 to 84 years old, with a median age of 62. Among all the included patients, 222 (80.1%) were males, 135 (48.7%) had an ECOG PS of 0, 183 (66.1%) were current or former smoker, 38 (13.7%) had liver metastasis and 95 (34.3%) had brain metastasis at baseline. In total, 233 (84.1%) patients received chemotherapy plus immunotherapy, 38 (13.7%) received chemotherapy plus immunotherapy combined with antiangiogenic therapy, 3 (1.1%) received single-agent immunotherapy, and the other 3 received immunotherapy combined with antiangiogenic therapy. The baseline clinical characteristics of the patients were summarized in Table 1. Based on the median value of baseline ChE (7611 units/liter, U/L), all the patients were the stratified into two groups. Clinicopathological baseline characteristics were comparable between the two groups (Table 2). Results showed that age, smoking status, high ALB, low NLR, low CRP and low SAA were significantly associated with high baseline ChE level (p < 0.05). At the last follow-up (December 4, 2023), 123 (44.4%) patients died, the median progression-free survival time was 10.44 months (95% CI 8.37–12.50 months) and the median overall survival was 34.17 months (95% CI 30.47–37.86 months).

According to ChE levels at baseline (with the median value as the cut-off), all the patients were divided into two groups (low group: ChE < 7611 U/L; high group: ChE ≥ 7611 U/L). Based on the alteration levels of ChE after the treatment of immunotherapy, patients were also stratified into ChE reduction group and ChE elevation group. First, we explored a prognostic role for baseline ChE levels. Kaplan–Meier survival analysis and log-rank test showed that patients with high ChE at baseline had both significantly prolonged PFS [median PFS, 12.40 vs. 9.37 months; HR = 0.72, (95% CI 0.54–0.96), p = 0.027; Fig. 1A] and OS [median OS, 37.73 vs. 27.20 months; HR = 0.59, (95% CI 0.41–0.85), p = 0.004; Fig. 1B] than those patients with low ChE at baseline. Second, the prognostic value of alteration levels of ChE was evaluated. We noted the similar trend that patients with an elevation of ChE had better PFS [median PFS, 12.80 vs. 6.97 months; HR = 0.56, (95% CI 0.42–0.75), p < 0.0001; Fig. 1C] and OS [median OS, 37.73 vs. 20.13 months; HR = 0.52, (95% CI 0.37–0.74), p < 0.0001; Fig. 1D] than those with a reduction of ChE.

For additional verification, patients were stratified into four groups by both baseline and the alteration of ChE after immunotherapy (baseline ChE low/high and the alteration of ChE reduction/elevation). Results indicated that the combination of both two prognostic factors improved risk stratification and prognostication for survival outcomes in advanced NSCLC (p < 0.0001) (Fig. 2). As expected, those with both high ChE at baseline as well as an elevation of ChE had the best survival outcomes of benefit (n = 83, median PFS = 14.00 months, median OS not reached). In addition, patients with either adverse prognostic feature (baseline low ChE or early reduction of ChE) had intermediate survival. While, Patients who had the low ChE at baseline and an early reduction of ChE after immunotherapy (n = 37) had the particular high risk of death and the worst survival outcomes, with a median PFS of 5.56 months and a median OS of 12.43 months (Fig. 2A, B). Therefore, both baseline high ChE and an elevation of ChE were independently displayed a better clinical benefit from immunotherapy, and the combination of these two prognostic factors can further improve risk stratification.

Based on the univariate analysis, liver metastasis (p < 0.001), bone metastasis (p = 0.006), baseline ChE (p = 0.027) and the alteration of ChE (p < 0.001) were significantly associated with PFS (Additional file 2: Table S2). The multivariate Cox regression analysis revealed that liver metastasis (p = 0.001), bone metastasis (p = 0.009), baseline ChE (p = 0.004) and the alteration of ChE (p < 0.001) were the independent indicators of PFS. Furthermore, univariate and multivariate Cox regression analyses were also performed for identifying OS prognostic factors. Univariate analysis revealed that factors associated with inferior OS included male, age ≥ 62 years old, current or former smoker, liver metastasis, bone metastasis, low ChE at baseline (p = 0.004), a reduction level of ChE (p < 0.001), low ALB, high CRP and high SAA. We further included those significant clinicopathological parameters in univariate analysis into the multivariate model. Multivariate analysis revealed that baseline ChE (high vs. low, HR, 0.65; 95% CI 0.43–0.99; p = 0.047) and ChE alteration (elevation vs. reduction, HR, 0.51; 95% CI 0.35–0.74; p < 0.001) were independent indicators for OS, others including age (< 62 vs. ≥ 62 years old, HR, 0.67; 95% CI 0.46–0.98; p = 0.037), liver metastasis (Yes vs. No, HR, 1.73; 95% CI 1.09–2.75; p = 0.019) and bone metastasis (Yes vs. No, HR, 2.23; 95% CI 1.55–3.22; p < 0.001) at baseline (Table 3).

The significant variables from the multivariate Cox analysis, including age, liver metastasis, bone metastasis, baseline ChE and alteration of ChE were used to establish a prognostic nomogram for OS. Each variable corresponds to a specific point by drawing a line straight up to the score axis, and the total points were calculated by adding up the individual score of each of the 5 variables included in the nomogram. The probability of survival was demonstrated by making a vertical line from the total score axis to intersect the survival probability axis of 1, 2, and 3 years (Fig. 3A). Prognostic accuracy and predictive value of nomogram were evaluated using the time-dependent ROC curves and AUC. The AUC values in the ROC curve analysis showed a good accuracy, with the AUC values of 1-, 2-, and 3-year survival probability of 0.705, 0.743 and 0.691 (Fig. 3B). The calibration plots demonstrated satisfactory consistency between the nomogram-predicted OS and actual observed probability of OS at 1-, 2- and 3-year (Additional file 3: Fig. S2A).

Based on the constructed nomogram, we calculated the total risk score for the enrolled NSCLC patients, and all of them were then divided into high-risk and low-risk groups, according to the median value of total point. As showed in the Kaplan Meir curve, a significant difference in OS were observed that patients in the high-risk group had remarkably worse survival outcomes than patients in the low-risk group (p < 0.0001) (Fig. 3C). Finally, the plotted decision-curve analysis (DCA) curve demonstrated the clinical validity and usefulness of the model (Additional file 3: Fig.S2B).

To further explore the gene changes and specific mechanisms related to choline metabolism, we obtained bulk RNA-seq data and clinical information of LUAD patients from TCGA database (https://​cancergenome.​nih.​gov/​). “Limma” package in R was used to identify DEGs between LUAD and normal tissues (Fig. 4A). The Venn diagrams were plotted to show the overlap between DEGs in LUAD and choline metabolism-related genes obtained from KEGG, Reactome Path and AmiGO2 databases (Fig. 4B). We finally screened 36 choline metabolism-related DEGs, including 14 upregulated genes and 22 downregulated genes. Subsequently, univariate and multivariate Cox regression analysis of the choline metabolism-related DEGs were performed and we finally obtain four choline metabolism-related prognostic genes (MTHFD1, PDGFB, PIK3R3, CHKB) (Fig. 4C).

According to the formula Risk score = 0.311 × MTHFD1 + (0.204) × PDGFB + (−0.277) × PIK3R3 + (−0.207) × CHKB, the risk score of each patient in TCGA-LUAD (training cohort) and Orient-11 study (validation cohort) was calculated. All the patients were divided into high-risk group and low-risk group based on the median value of risk score. Kaplan–Meier survival analysis demonstrated a significant difference in overall survival between two risk group, with the patients in high-risk group having poorer survival and higher risk of death for both training (p = 0.00023) (Fig. 4D) and validation cohort (p = 0.00039) (Fig. 4E). Additionally, we performed ROC curves and calculated AUCs of both training and validation cohorts. Results showed that the ROC curve had acceptable sensitivity and specificity, indicating the satisfactory prognostic and predictive power of the constructed choline metabolism-related signature (Additional file 3: Fig. S2C-D).

We performed the GSEA analysis used those upregulated and statistically significant DEGs in the high-risk group of the choline metabolism signature and results showed that multiple cancer-related pathways, including cell cycle, extracellular matrix (ECM) receptor interaction, cell motility, P53 and PI3k-Akt signaling were significantly enriched (Fig. 5A, Additional file 4: Fig. S3A). These results suggested that the high-risk of choline metabolism signature may have stronger ability of tumor cell growth, proliferation, invasion and metastasis of lung cancer. To further investigate the association between the choline metabolism signature and the immune microenvironment, we used the “IOBR” R package to calculate the scores of tumor microenvironment-related gene sets in the high-risk and low-risk groups of patients in TCGA-LUAD. We found that the gene sets related to lymphocytes, B cells and Mast cells were down-regulated in high-risk patients, whereas the gene sets related to macrophages, CAF cells and endothelial cells were significantly up-regulated (Fig. 5B–D). By comparing the score of metabolism-related gene sets, significant differences in multiple metabolism-related pathways were found between high-risk and low-risk group, suggesting the phenomenon of metabolic reprogramming in patients with high-risk (Fig. 5E). Furthermore, TIDE scores in high-risk group of choline metabolism signature were considerably greater than those in the low-risk group (Fig. 5F). Consistently, patients in the high-risk group of choline metabolism possessed a worse response rate to ICIs (ctla4_pos_pd1_neg, ctla4_neg_pd1_neg, p < 0.001, Fig. 5G), indicating that higher risk of choline metabolism may have stronger ability of immune evasion and immunotherapy resistance. Collectively, the above results suggested that patients in high-risk group were characterized by immunosuppressive tumor microenvironment.

To further explore how choline metabolism mediates the immunosuppressive tumor microenvironment in NSCLC patients, single-cell sequencing data (GSE207422) were downloaded from GEO database, which contained fresh tumor samples from 3 pre-treatment and 12 post-treatment patients with NSCLC who received anti-PD-1 therapy combined with chemotherapy. After performing quality control on the samples, removing batch effects by Harmony analysis, we finally obtained 11 cell types by clustering in reduced dimensions and manual annotation (Fig. 6A). We found that MTHFD1 was mainly expressed in myeloid cells (monocytes & macrophages) and mast cells, while PDGFB was mainly expressed in endothelial cells (Fig. 6B, C). Our previous results demonstrated that multiple macrophage pathway scores were significantly up-regulated in high-risk patients, we further explored the expression of MTHFD1 in myeloid cells. After further grouping and annotation of myeloid cells, 6 subgroups of myeloid cells can be obtained (Fig. 6D). We found that MTHFD1 expression gradually increased as monocytes differentiate into macrophages and it was mainly expressed in CCL18⁺ macrophages (Fig. 6E), and meanwhile CCL18 expression was up-regulated in TCGA-LUAD patients with high-risk (Additional file 3: Fig. S2E), suggesting that MTHFD1 may be implicated in immunosuppressive macrophage differentiation of NSCLC. Ucell score was then used to calculate the functional characteristics of 6 myeloid cell subsets, and results showed that CCL18⁺ macrophage had both a higher anti-inflammatory related score and a lower pro-inflammatory related score (Fig. 6F), suggesting that CCL18⁺ macrophage belonged to immunosuppressive subsets. To infer the developmental trajectory of myeloid cells, monocle3 (v0.2.1) was used to perform pseudotime analysis and results indicated that the subset of CCL18⁺ macrophages belonged to the endpoint of differentiation (Fig. 6G).

To deepen the understanding of the connection between MTHFD1 and macrophages polarization, we stratified all TCGA-LUAD patients into low and high expression groups of MTHFD1 based on the median level. We observed that high expression of MTHFD1 was correlated with elevated level of M2-like macrophages, consistent with previous findings associating high-risk group of choline metabolism with increased M2-like macrophages (Additional file 4: Fig. S3B). Furthermore, GSEA showed the downregulation of M1-like macrophages-related gene sets in MTHFD1-high patient tumors compared with MTHFD1-low patient tumors (Additional file 4: Fig. S3C). Correlation analysis verified a positive association between MTHFD1 expression and CCL18⁺ immunosuppressive macrophages, collectively indicating that increased level of MTHFD1 may be associated with the polarization and immunosuppressive functions of macrophage in the tumor microenvironment (Additional file 4: Fig. S3D).

Those post-treatment samples of single-cell sequencing data were further divided into two groups based on Response Evaluation Criteria in Solid Tumors Version (RECIST): partial-response group (PR; n = 8) and stable-disease group (SD; n = 4). We found that patients in the SD group had a higher expression of MTHFD1 (Fig. 7A), suggesting that higher expression of MTHFD1 may be associated with poor efficacy of immunotherapy. At the same time, by comparing the proportion of 6 myeloid cell subsets in PR and SD group, a significant difference was observed in the proportion of CCL18⁺ macrophages between two groups, with SD group exhibiting a higher proportion (Fig. 7B). These results above suggested that the expression of MTHFD1 was associated with the immunosuppressive functions of CCL18⁺ macrophages and may account for poorer immunotherapeutic effect.

PDGFB, another risk gene of choline metabolism in this study, was mainly expressed in endothelial cells as mentioned above (Fig. 6B, C). It was reported in previous research that higher expression of PDGFB in gastric cancer promoted angiogenesis in metastases via activation of the MAPK/ERK signaling pathway in endothelial cells [23]. Thus, it was reasonable to speculate that the immunosuppressive functions of macrophages in abnormal choline metabolism may be associated with the proliferation of endothelial cells to some extent. CellChat was next performed to identify the intercellular communication and results showed that endothelial cells had strong communication with myeloid cells (Fig. 7C), which was consistent with our previous findings that endothelial cell gene set scores were up-regulated in patients of high-risk. Furthermore, we next compared the differences in intercellular communication between endothelial cells and myeloid subsets of PR and SD group. Results showed that FN1- (ITGA5 + ITGB1) signaling pathway mediating the strongest communication probability from CCL18⁺ macrophages to endothelial cells in SD group (Fig. 7D), while this phenomenon was not significant in PR group. The hierarchy plot showed that CCL18⁺ macrophages were the most important sender in FN1 signaling pathway, endothelial cells were mainly the receiver, and both of them acquired high influencer scores (Fig. 7E), indicating a high capacity of the influencing information flow. By comparing the expression distribution of signaling genes involved in the inferred FN1 signaling network, a higher expression level of FN1 was observed in CCL18 macrophages of SD group compared with that of PR group (Fig. 7G). Finally, further correlation analysis indicated that there was a strong correlation between FN1 and MHTFD1 expression in TCGA-LUAD cohort (Fig. 7F). Previous studies have found that FN1 as well as ITGA5 could facilitate tumor angiogenesis and progression in cervical cancer [24, 25] and may be a prognostic risk factor in many types of cancer including Gliomas, gastric cancer, head and neck squamous cell carcinoma and NSCLC [26‐30]. Above all, our results suggested that overexpression of MTHFD1 in abnormal choline metabolism contributes to the immune suppressive polarization in macrophages, further promoting the proliferation of endothelial cells. Together, these processes jointly shape the immunosuppressive microenvironment of NSCLC, ultimately mediating the occurrence of immunotherapy resistance.