Abstract:

Objective To analyze the diagnostic value of metabolic makers in cerebrospinal fluid in advanced lung adenocarcinoma patients with leptomeningeal metastases （LM）. Methods A total of 46 cerebrospinal fluid samples （LM group） from lung adenocarcinoma patients with LM admitted to Nanjing Drum Tower Hospital Affiliated to Nanjing University Medical School from December 2019 to December 2021 were collected, and 48 cerebrospinal fluid samples （control group） from patients with benign neurological diseases during the same period were collected. Metabolomics analysis of cerebrospinal fluid was carried out by high-performance liquid chromatography-mass spectrometry. Principle component analysis （PCA） and orthogonal to partial least squares discriminant analysis （OPLS-DA） were used for modeling. Multi-criteria assessment was used to identify the different metabolites between the two groups. Receiver operating characteristic （ROC） curve, pathway enrichment analysis and other methods were used to screen metabolites and pathways related to LM from lung adenocarcinoma. Results There were no statistically significant differences in the proportions of age （Z=-0.41, P=0.210）, gender （χ²=1.19, P=0.275）, history of smoking （χ²=2.86, P=0.091）, Karnofsky performance status score （χ²=0.65, P=0.419） and increased intracranial pressure （χ²=0.65, P=0.419） between the LM group and control group. The models of PCA （R2X was 0.608 and 0.583, Q2 was 0.462 and 0.513 in electrospray ion positive and negative modes, respectively） and OPLS-DA （R2Y was 0.967 and 0.889, Q2 was 0.959 and 0.852 in electrospray ion positive and negative modes, respectively） showed that the overall data quality was good. Meanwhile, the model interpretation rate and prediction rate were effective. The permutation tests duplicated for 200 times and showed no over-fitting of the established model. The metabolic profiles of the two groups were significantly different. A total of 30 endogenously differential metabolites were screened by using multi-criteria assessment. Six potential biomarkers with larger area under the curve （AUC） were identified through ROC curve analysis, including tyrosine （AUC=0.967, 95%CI： 0.906-1.000）, phenylalanine （AUC=0.992, 95%CI： 0.973-1.000）, pyruvate （AUC=0.976, 95%CI： 0.935-1.000）, tryptophan （AUC=0.935, 95%CI： 0.880-0.973）, glucose （AUC=0.932, 95%CI： 0.880-0.975） and adenosine monophosphate （AUC=0.993, 95%CI： 0.987-1.000）. The 30 selected differential metabolites were enriched and analyzed for metabolic pathways, and 20 relevant metabolic pathways were matched. Among them, the four metabolic pathways most likely to cause changes in metabolites were glycolysis and glucose metabolic synthesis, pyruvate metabolism, phenylalanine metabolism, phenylalanine, tyrosine and tryptophan biosynthesis. Conclusion Untargeted metabolomics analysis can effectively screen specific cerebrospinal fluid metabolites in lung adenocarcinoma patients with LM. Six potential metabolites such as tyrosine, phenylalanine, pyruvate, tryptophan, adenosine monophosphate, glucose and their metabolic pathways may be involved in the pathogenesis of LM from lung adenocarcinoma.

Key words: Metabolomics, Lung adenocarcinoma, Leptomeningeal metastases, Cerebrospinal fluid