Lipidomics reveals new lipid-based lung adenocarcinoma early diagnosis model

Abstract

Lung adenocarcinoma (LUAD) continues to pose a significant mortality risk with a lack of dependable biomarkers for early noninvasive cancer detection. Here, we find that aberrant lipid metabolism is significantly enriched in lung cancer cells. Further, we identified four signature lipids highly associated with LUAD and developed a lipid signature-based scoring model (LSRscore). Evaluation of LSRscore in a discovery cohort reveals a robust predictive capability for LUAD (AUC: 0.972), a result further validated in an independent cohort (AUC: 0.92). We highlight one lipid signature biomarker, PE(18:0/18:1), consistently exhibiting altered levels both in cancer tissue and in plasma of LUAD patients, demonstrating significant predictive power for early-stage LUAD. Transcriptome analysis reveals an association between increased PE(18:0/18:1) levels and dysregulated glycerophospholipid metabolism, which consistently displays strong prognostic value across two LUAD cohorts. The combined utility of LSRscore and PE(18:0/18:1) holds promise for early-stage diagnosis and prognosis of LUAD.

Keywords: Cancer Early Diagnosis Model; LSRscore; LUAD; Lipid Metabolism; Lipidomics.

Figure 1. scRNA-seq analyses of early-stage LUAD and healthy lung tissues identifies aberrant lipid metabolism in cancer cells.

(A) The uniform manifold approximation and projection (UMAP) of 170435 single cells identified by scRNA-seq in tumor and adjacent normal tissues from 17 LUAD patients and healthy lung tissues from eight healthy donors. Seven major cell subpopulations are labeled. Each dot is a single cell and colored based on its cell cypt. (B) Canonical markers of different cell types used to identify cell subpopulations in UMAP plot. (C) The distribution of cell subpopulations in LUAD tumor tissues, adjacent normal tissues and healthy lung tissues. Each color represents a cell subpopulation. (D) Lung tumor cells and normal epithelial and stromal cell subpopulations. Red is malignant cells; blue is nonmalignant cells. (E) The sample of origin for malignant and nonmalignant cells. Red is adjacent normal lung tissues; green is healthy normal lung tissues; blue is LUAD tumor tissues. (F) Enrichment map network of statistically significant GO categories in the malignant cells. Nodes are GO categories and lines their connectivity. Node size is proportional to the number of genes in the GO term and line thickness is the fraction of genes overlapped between GO categories.

Figure 2. The signature lipids for early-stage LUAD detection, n = 60 for LUAD and 30 for HC.

(A, B) Score plot of OPLS-DA model for LUAD and HC on the discovery cohort at positive ion mode (A) and at negative ion mode (B), respectively. Each dot represents a sample. Green is healthy subjects used as control (HC); blue is LUAD patients. (C) Differential lipids identified at positive or negative ion model, respectively. X axis is a base-2 logarithmic scale of fold change for lipid. Y axis is the negative of the base-10 logarithmic scale of FDR value. Triangle represents positive ion mode. Dot represents negative ion mode. Size of each triangle or dot indicates variable importance in the projection (VIP) values. Differential lipids are colored according to lipid class. FA fatty acids, GL glycerolipids, GP glycerophospholipids, SP sphingolipids, ST sterols. (D) The number of differential lipids identified at positive or negative ion mode, respectively. (E) The hierarchical clustering of samples using the 7 signature lipids for LUAD. The clustering method was used ward D2. (F–I) The comparison of lipid concentration between HC and LUAD for FA(20:0) (F), PE(18:0/18:1) (G), LPC(18:1) (H), and PC(18:2/18:2) (I). **** represents P value from the Mann–Whitney U test <0.0001. For box, bottom boundary presented as 25% quantile, upper boundary presented as 75% quantile and line in the middle presented as median value of data. Upper whisker: largest observation less than or equal to 75% quantile + 1.5 *iqr; lower whisker: smallest observation greater than or equal to 25% quantile − 1.5 *iqr;iqr = 75% quantile – 25% quantile. (J) ROC curves of signature lipids on the discovery cohort.

Figure 3. The classification performance of LSRscore on the discovery and validation cohorts.

(A) ROC curves of LSRscore and the four involving signature lipids on the discovery cohort (n = 90). (B) The multivariate logistic analysis of LSRscore and clinical factors including age, gender, and tumorMarker (n = 90). (C) ROC curves of LSRscore and the four involving signature lipids on the prospective validation cohort. (D) The multivariate logistic analysis of 100*LSRscore and clinical factors including age, and gender. (E–H) The comparison of lipid concentration between HC and LUAD for FA(20:0) (E), PE(18:0/18:1) (F), LPC(18:1) (G), and PC(18:2/18:2) (H). **** represents P value from Mann–Whitney U test <0.0001. For box, bottom boundary presented as 25% quantile, upper boundary presented as 75% quantile and line in the middle presented as median value of data. Upper whisker: largest observation less than or equal to 75% quantile + 1.5 *iqr; lower whisker: smallest observation greater than or equal to 25% quantile − 1.5 *iqr;iqr = 75% quantile − 25% quantile. Data information: (C–H) n = 30 samples for each group.

Figure 4. The classification performance of PE(18:0/18:1) on tissue validation cohort, n = 25 samples for each group.

(A, B) The comparison of lipid concentration between LUAD cancer tissue and adjacent lung tissue using Mann–Whitney U test (A) and paired Student t test (B) for PE(18:0/18:1). (C, D) The comparison of PC(18:2/18:2) concentration between LUAD cancer tissue and adjacent lung tissue using Mann–Whitney U test (C) and paired Student t test (D), respectively. (E, F) ROC curves of PC(18:2/18:2) (E) and PE(18:0/18:1) (F). (G) The comparison of PE(18:0/18:1) concentration between LUAD cell lines (A549,NCI-H1975) and control cell lines (HepG2,BEAS-2B,HSF) (each having 3 replicates) using Student t test. Data information: (A, C) For box, bottom boundary presented as 25% quantile, upper boundary presented as 75% quantile and line in the middle presented as median value of data. upper whisker: largest observation less than or equal to 75% quantile + 1.5 *iqr; lower whisker: smallest observation greater than or equal to 25% quantile − 1.5 *iqr;iqr = 75% quantile − 25% quantile.

Figure 5. The prognostic values of KEGG pathways associated with PE(18:0/18:1).

(A) The enrichment status of KEGG pathways involving PE(18:0/18:1) in CPTAC-LUAD. n = 229 LUADs and 120 Healthy lung tissues. (B) Kaplan–Meier curves show the overall survival of LUAD patients with high (> median) or low ($\le \mathrm{median}$) enrichment of glycerophospholipid metabolism in CPTAC-LUAD. P value was calculated by log-rank test. (C, D) Multivariate cox regression analysis reveal that glycerophospholipid metabolism is associated with overall survival independent of other clinical factors on CPTCA-LUAD (C) and TCGA-LUAD cohorts (D), respectively. Line segment represents the 95% confidence interval of hazard ratio (HR), while points represent the HR value calculated by cox regression. (E, F) The level of lipid PE(18:0/18:1) (E) and the enrichment status of glycerophospholipid metabolism pathways (F) between LUAD A549cell line and healthy BEAS-2B cell line with three independent repetitions. Welch’s t test P value less than 0.05 was considered significant. The error bar means the min and max value of data, while the bar is the median value. (G) The Pearson correlation of enrichment scores of glycerophospholipid metabolic pathways with PE(18:0/18:1) levels across A549 and BEAS-2B cell lines. The experiment was independently repeated three times.