To comprehensively profile the lung cancer metabolome, the plasma of lung cancer patients and HC were detected for polar metabolites and lipids (Fig. 1, panel A). Internal standards and quality control samples were used for quality assurance. Multiple metabolites of polar metabolites (such as carbohydrates and amino acids) closely related to lipids, and together with protein metabolism, constitute the human metabolic map (Fig. 1 panel B). A total of 63 polar metabolites and 742 lipids were annotated (Fig. 1, panels C and D).
Fig. 1Overview of polar metabolome and lipidome detection in lung cancer. A schematic illustration showing the combined analysis using multi-omic metabolites data in the study for lung cancer precision medicine. In summary, blood samples were collected and subjected to detection and identification using LC–MS and GC–MS techniques. Subsequently, a multi-omics correlation analysis was conducted to explore the relationships between the identified metabolites. A Omics such as glycomics, metabonomics, lipidomics and proteomics are interconnected B Pie charts showing the numbers and proportions of annotated polar metabolites C and lipids D identified in present study. Abbreviations: GC–MS: gas chromatography mass spectrometry; LC–MS: liquid chromatography-mass spectrometry; PC: phosphatidylcholines; PE: phosphatidylethanolamine; LPC: lysophosphatidylcholine; LPE: lysophosphatidylethanolamine; SM: sphingomyelin; TAG: triacylglycerol; DAG: diacylglycerol; CE: cholesteryl ester
The polar metabolomic landscape of lung cancerThe comparison of average concentration levels of metabolite species between lung cancer patients and HC revealed the detection of 63 polar metabolites, out of which 33 exhibited significant differential expression (14 were higher and 19 were lower, respectively, compared to HC, all p < 0.05), demonstrating significant differences in abundance between lung cancer and controls. The plasma levels of alcohols, amides, and peptide metabolites of lung cancer patients (n = 111) were significantly increased than that of HC (n = 111), whereas carboxylic acids, hydrocarbons, and fatty acids were decreased (Fig. 2 panels A and B). Of those significantly changed polar metabolites in patients with lung cancer, the top 6 increased metabolites (D-lyxose, L-sorboase, galactinol, urea, D-allose, and L-threonine) and top 6 decreased metabolites (maltitol, palatinitol, eicosapentaenoic acid, D-mannitol, glycolic acid, and thymol), compared to HC, were also identified (Fig. 2 panels C and D). To identify differential metabolite species, OPLS-DA models were established and employed. Six metabolites, such as D-lyxose, galactinol, urea, D-allose, D-glucose, and D-mannose, were upregulated (all VIP > 1, FC > 1, and p < 0.05), while 5 metabolites, namely lactic acid, glycolic acid, D-mannitol, palatinitol, and maltitol, were down-regulated (all VIP > 1, FC < 1, and p < 0.05), defined on the basis of VIP score of lung cancer (Fig. 2 panel E). Furthermore, those remarkably changed metabolites were mapped to KEGG pathways using Metabol Analyst's pathway enrichment tool. Six upregulated pathways, including Arginine biosynthesis, Aminoacyl-tRNA biosynthesis, Galactose metabolism, Glycine, serine and threonine metabolism, Nitrogen metabolism, and D-Glutamine and D-glutamate metabolism, and 8 down-regulated pathways, including Glyoxylate and dicarboxylate metabolism, Alanine, aspartate and glutamate metabolism, Arginine and proline metabolism, Arginine biosynthesis, Citrate cycle (TCA cycle), Pyruvate metabolism, Glycolysis/Gluconeogenesis and Glycine, and serine and threonine metabolism, were identified (Fig. 2 panel H).
Fig. 2The metabolomic landscape of lung cancer. A Changes in metabolic profiles in plasma of lung cancer patients and health controls detected by GC–MS. Scatterplots with boxplots showing the significantly changed metabolite classes in lung cancer. B Volcano plots of significantly up/down-regulated expressed metabolites in lung cancer. The most differentially expressed polar metabolites were mainly concentrated in 3 types of metabolites: carbohydrates, amino acids and carboxylic acids (log2FC > 0.5 or log2FC < -0.5 and p < 0.05, compared with HC). C D-lyxose (FC = 5.17, p < 0.01), L-sorboase (FC = 1.66, p < 0.01), galactinol (FC = 1.65, p < 0.01), urea (FC = 1.54, p < 0.01), D-allose (FC = 1.50, p < 0.01), and L-threonine (FC = 1.46, p < 0.01) were the top 6 significantly increased metabolites in patients with lung cancer. D Maltitol (FC = 0.37, p < 0.01), palatinitol (FC = 0.38, p < 0.01), eicosapentaenoic acid (FC = 0.44, p < 0.01), D-mannitol (FC = 0.55, p < 0.01), glycolic acid (FC = 0.59, p < 0.01), and thymol (FC = 0.60, p < 0.01) were the top 6 significantly decreased metabolites in patients with lung cancer. E OPLS-DA score plot showed the high separating capacity of polar metabolites. F The top 6 up-regulated polar metabolites in VIP chart (VIP > 1, FC > 1, and p < 0.05). The top 6 up-regulated polar metabolites. G The top 5 scored down-regulated polar metabolites in VIP chart (VIP > 1, FC < 1, and p < 0.05). H Enrichment pathway analysis to identify signaling pathways related to changed metabolites. The 3 most significantly activated signaling pathways were arginine biosynthesis, aminoacyl-tRNA biosynthesis, and galactose metabolism. The 3 most significantly inhibited signaling pathways were glyoxylate and dicarboxylate metabolism, alanine, aspartate and glutamate metabolism, and arginine and proline metabolism. *p < 0.05; ** p < 0.01; *** p < 0.001
Lipidomic profile of lung cancerThe intensity levels of CER, PE, SM, and TAG in the plasma of lung cancer patients (n = 204) were significantly higher than those of HC (n = 204), whereas the levels of CE, DAG and PC were remarkably reduced. Comparison of the mean concentration of lipid species between lung cancer patients and HC identified 742 lipids, of them 541 were differentially expressed (287 were higher and 254 were lower, respectively, compared to HC, all p < 0.05), indicating significant differences of abundance between lung cancer and HC (Fig. 3 panels A and B). Of these significantly changed lipids in patients with lung cancer, the top 6 elevated lipids and top 6 decreased lipids were also identified (Fig. 3, panels C and D): levels of CE(20:1), CE(20:0), CE(24:0), PE(14:0/14:0)-H, TAG42:0-FA16:0, and CE(22:2) were significantly increased, compared to that of HC (all p < 0.05); levels of TAG56:1-FA16:0, TAG56:1-FA18:1, PC(20:0/20:2) + AcO, TAG58:2-FA18:1, PE(14:0/20:2), and TAG58:3-FA18:1 were significantly decreased, comparing to that of HC (all p < 0.05). To better identify differential lipid species, OPLS-DA models were established and employed (Fig. 3, panel E). Twenty-three metabolites, including CE(20:0), TAG54:5-FA18:2, and TAG50:3-FA16:1, were upregulated (all VIP > 1, FC > 1, and p < 0.05); 7 metabolites, including PC(18:0/20:4), PC(16:0/22:6), and CE(18:2), were down-regulated (all VIP > 1, FC < 1, and p < 0.05), defined on the basis of VIP score of lung cancer (Fig. 3 panels F and G). Lastly, the numbers of lipids that were significantly up-or down-regulated in multiple lipid subclasses were counted (Fig. 3 panel H). Most of the significantly upregulated lipids exist in two major classes of lipids as TAG and PE, while the downregulated lipids mainly exist in TAG, PC, PE, and DAG.
Fig. 3The lipidomics landscape of lung cancer. A Changes in lipid profiles in plasma of lung cancer patients and normal controls detected by LC–MS. Volcano plots of significantly up-/down-regulated expressed lipids in lung cancer group. TAGs were the most variable lipids, with multiple TAG contained in both elevated and reduced lipids (log2FC > 1.5 or log2FC < -1.5, and p < 0.05, compared with HC). B Scatterplots with boxplots showing the significantly changed lipid classes in lung cancer. C CE(20:1) (FC = 10.22, p < 0.01), CE(20:0) (FC = 6.70, p < 0.01), CE(24:0) (FC = 5.85, p < 0.01), PE(14:0/14:0)-H (FC = 5.00, p < 0.01), TAG42:0-FA16:0 (FC = 4.96, p < 0.01), and CE(22:2) (FC = 4.94, p < 0.01) were the top 6 significantly increased lipids in patients with lung cancer. D TAG56:1-FA16:0 (FC = 0.08, p < 0.01), TAG56:1-FA18:1 (FC = 0.12, p < 0.01), PC(20:0/20:2) + AcO (FC = 0.12, p < 0.01), TAG58:2-FA18:1 (FC = 0.13, p < 0.01), PE(14:0/20:2) (FC = 0.15, p < 0.01) and TAG58:3-FA18:1 (FC = 0.20, p < 0.01) were the top 6 significantly decreased lipids in patients with lung cancer. E The score plot generated by OPLS-DA demonstrated the strong discriminatory ability of lipids. F The top 23 up-regulated lipids in VIP chart (VIP > 1, FC > 1, and p < 0.05). G The top 7 down-regulated lipids in VIP chart (VIP > 1, FC < 1, and p < 0.05). H Among the up-/down-regulated lipids, the cumulative number of significantly changed lipids in each lipid class. The 3 lipid types that increased the most were TAG, PE and CE. The 3 most reduced lipid types were TAG, PC and PE. *p < 0.05; ** p < 0.01; *** p < 0.001
Correlation of polar metabolites and lipidsPearson correlation analysis examined the relationship between polar metabolomics (p < 0.05, compared to HC) and lipids (p < 0.05, logFC > 0.5 or logFC < -0.5, compared to HC), explored the complexity of metabolic networks in the plasma of lung cancer patients (Fig. 4 panel A), and revealed the significant correlations between a few metabolites and lipids. The top 5 significant lipids that were positively and negatively correlated with the expression of these up/down-regulated metabolites were listed in Table 1 with their functional associations and roles in cancers annotated. There was a significant difference in TAG levels between lung cancer and HC (p < 0.001). Up-regulated lipids with VIP > 1 and p < 0.05 were mainly concentrated in TAG. Although lipids in the lung cancer group generally had weak negative associations with the polar metabolites, a variety of metabolites were found strongly positively correlated with TAG. In lung cancer patients, the significant upregulated amino acids generally had a positive correlation with TAG44:0-FA16:0 and TAG48:2-FA18:0, and upregulated carbohydrates had a significant positive correlation with TAG51:1-FA18:0. A few carbohydrate metabolites that were significantly downregulated in the lung cancer group were positively associated with TAG44:1-FA14:0, TAG44:1-FA18:1, and TAG44:2-FA14:0. Down-regulated fatty acids that showed a general positive association with TAG44:1-FA14:0, TAG46:1-FA18:1, and TAG46:2-FA18:2.CE(22:0) was negatively correlated with the downregulated metabolite D-mannitol and pyruvic acid. The discriminative power of significant differenced metabolites (p < 0.05, Fig. 4, panel B) and lipids (p < 0.05, logFC > 0.5 or < -0.5, Fig. 4 panel C) were tested using mean ROC curve analysis generated by tenfold cross-validation, and found that both polar metabolites and lipids profiles could distinguish lung cancer from HC, in particular, lipids showed promising discriminative power with superior sensitivity and specificity.
Fig. 4Correlation between polar metabolome and lipidome. Pearson correlation analysis examined the relationship between polar metabolomics and lipids. Heatmaps were used to show the relationship between these two omics results. The color gradients of the heatmap indicates the Pearson correlation coefficients. The horizontal line in the middle of heatmap is the dividing line between up and down regulated metabolites, and above the horizontal line is increased metabolites, below the horizontal line is decreased metabolites. A variety of polar metabolites were found to have a clear correlation with lipids (A). ROC curves of polar metabolites (B) and lipids (C) were constructed using tenfold cross-validated evaluation based on the random forests algorithm. The averaged ROC curves (represented in blue) were generated by calculating the mean of all ROC curves for polar metabolites or lipids. Our AUC was calculated using the mean ROC curve. AUC stands for the area under the curve, the shaded part is 1 standard deviation above and below. ROC curves demonstrated that the expression of polar metabolites and lipids could distinguish lung cancer patients from HC
Table 1 The top 5 significant lipids that were positively and negatively correlated with the expression of these up/down-regulated metabolitesTrans-omic profiles integrating lipidomes with clinical phenomesBased on the eQTL model, we modified the lipid quantitative trait loci model to investigate lipid-clinical phenomes correlation. In addition, MatrixlQTL R package was used to obtain the significant phenome-lipid pairs and their p values. Results showed that CE(C = 22, 24) were significantly associated with lower limb edema, smoking, cough, and hypertension stratification (Fig. 5, panel A). The close correlation between DAG and clinical phenomes focused on C18 (Fig. 5, panel B). Metastasis was related to the level of LPC(C = 20) (Fig. 5, panel C). Marasmus and diarrhea were associated with LPC(C = 18) and SM(C = 18) (Fig. 5, panels C and G). Pulmonary-related phenotypes including wheeze, chronic lung disease history, Velcro, pack-years of smoking, and nutritional state were significantly associated with LPE(C = 20) (Fig. 5 panel D). Pathological examination results, ALK, 34βE12, Cam5.2, CK8/18, and lymphatic pleural metastasis were related to PC(C = 18) (Fig. 5 panel E). PE(C = 18) was associated with a variety of clinical phenotypes, including schistosome and pleural metastasis (Fig. 5, panel F). TAG(C = 52, 54, and 56) mainly affect clinical phenotype (Fig. 5, panel H).
Fig. 5Trans-omic nodules cross-clinical phenomes and each lipid class measured by simulating the expression quantitative trait locus (eQTL) model. Heatmap showing the associations between the clinical phenomes and CE (A), DAG (B), LPC (C), LPE (D), PC (E), PE (F), SM (G) and TAG (H). The bluer the color of the heatmap, the weaker the correlation between the clinical phenomes and lipids; and the redder the color of the heatmap, the stronger the correlation between clinical phenomes and lipids
The phenotype-related lipid classes were classified by different clinical phenotypes. TAG was found to be the most abundant lipid class in plasma, and most of the lipids that associated with pathological results, physical examination, symptoms, primary disease, and metastasis belonged to TAG. The results revealed that patients with positive expression of Napsin A had a strong correlation with lipids (Fig. 6, panel A). The lipids most relevant to physical examination were DAG, PC, and TAG. Among these phenotypes, lower limb edema, voice transmission, thoracocyllosis, and intercostal changes were associated with lipids (Fig. 6, panel B). Most of the symptoms associated with lung cancer were significantly related to PE in addition to TAG (Fig. 6, panel C). In patients who had a history of drinking or smoking, PC and PE were dysfunctional (Fig. 6, panel D). In patients with underlying disease, PE was prone to be abnormal (Fig. 6, panel E). PC, PE, and TAG were associated with tumor metastasis, especially pleural and osseous metastasis. Lung cancer was frequently accompanied by these two types of metastasis (Fig. 6, panel F).
Fig. 6Correlations between various clinical phenotypes and lipids. Heatmap showing the associations between lipid classes and pathological results (A), physical examination (B), symptom (C), personal history (D), underlying disease (E), and metastasis (F). Numbers in the heatmap represent the number of related lipids detected per lipid class. The bluer the color of the heatmap, the less lipids are associated with the clinical phenotypes; and the redder the color of the heatmap, the more lipids are associated with the clinical phenotypes
Lipidomic profiles of lung cancer subtypesDifferent lipids are generally interconverted by specific cellular lipid phosphatases and kinases. DAG is derived from phosphatidate and converts into PE, PC, and TAG. PC and PE further convert into LPC and LPE, respectively. In addition, PC converts into CER to generate DAG and SM. And cholesterol produces CE (Fig. 7, panel A). The proportions (%) of the 9 main lipid elements of ADC (n = 20), SCC (n = 4), and SCLC (n = 9) were plotted through the pie charts (Fig. 7 panel B). Notably, SM, CE and LPE increased mainly in SCLC, whereas PC and PE significantly decreased, compared with ADC and SCC. In addition, LPC and TAG were decreased in SCC. Differences exist in the changed lipids among lung cancer subtypes. Among the lipid elements that showed significant changes in lung cancer, the top three lipids from each group were identified. Levels of TAG(49:1-FA16:1, 49:1-FA17:0, and 49:2-FA16:1) were significantly higher (Fig. 7, panel C), and TAG(53:4-FA16:0, 53:6-FA20:4), LPC(20:3) were lower (Fig. 7 panel D) in ADC as compared with that of SCLC. Levels of CE(18:2), SM(18:0, and 18:1) were higher (Fig. 7, panel E), and PC(18:2/20:5) and DAG(18:0/18:1, and 18:1/18:1) were lower (Fig. 7 panel F) in ADC as compared with that of SCC. Levels of PE(O-16:0/22:6, P-16:0/22:6, and P-18:1/22:6) in SCC were up-regulated (Fig. 7 panel G), and levels of LPC(16:0, 18:0, and 20:0) in SCC were down-regulated (Fig. 7 panel H), when compared with that of ADC and SCLC. In SCLC, the levels of SM(22:1, 24:0, and 26:1) were increased compared to the groups of ADC and SCC (Fig. 7, panel I). Moreover, the levels of PE(O-16:0/22:6, O-18:0/22:6, and P-16:0/22:6) were decreased in SCLC when compared with SCC (Fig. 7 panel J). Levels of LPC(16:1 and 22:5), and CE(24:1) were also decreased when compared with ADC (Fig. 7, panel K).
Fig. 7Top 3 significantly changed lipids in patients with different lung cancer subtypes. Schematic of interconversions among different lipids. DAG is derived from phosphatidate and converts into PE, PC and TAG. PC and PE further converts into LPC and LPE, respectively. In addition, PC converts into CER to generate DAG and SM. And cholesterol produces CE (A). Pie charts showing the proportions of multiple lipid classes in patients with different lung cancer subtypes (B). TAG49:1-FA16:1, TAG49:1-FA17:0, and TAG49:2-FA16:1 were the top 3 up-regulated lipids in ADC compared to SCLC (C). CE(18:2), SM(18:0), and SM(18:1) were the top 3 down-regulated lipids in ADC compared to SCLC (D). LPC(20:3), TAG53:4-FA16:0 and TAG53:6-FA20:4 were the top 3 up-regulated lipids in ADC compared to SCC (E). PC(18:2/20:5), DAG(18:0/18:1), and DAG(18:1/18:1) were the top 3 down-regulated lipids in ADC compared to SCC (F). PE(O-16:0/22:6), PE(P-16:0/22:6), and PE(P-18:1/22:6) were the top 3 up-regulated lipids in SCC compared to ADC and SCLC (G). LPC(16:0), LPC(18:0) and LPC(20:0) were the top 3 down-regulated lipids in SCC compared to ADC and SCLC (H). SM(22:1), SM(24:0), and SM(26:1) were the top 3 up-regulated lipids n SCLC compared to ADC and SCC (I). PE(O-16:0/22:6), PE(O-18:0/22:6), and PE(P-16:0/22:6) were the top 3 down-regulated lipids in SCLC compared to SCC (J). LPC(16:1), LPC(22:5), and CE(24:1) were the top 3 down-regulated lipids n SCLC compared to ADC (K). For all comparisons, *p < 0.05,**p < 0.01,***p < 0.001
留言 (0)