Targeted metabolomic profiles of serum amino acids and acylcarnitines related to gastric cancer

Abstract

Background: Early diagnosis and treatment are imperative for improving survival in gastric cancer (GC). This work aimed to assess the ability of human serum amino acid and acylcarnitine profiles in distinguishing GC cases from atrophic gastritis (AG) and control superficial gastritis (SG) patients.

Methods: Sixty-nine GC, seventy-four AG and seventy-two SG control patients treated from May 2018 to May 2019 in Gansu Provincial Hospitalwere included. The levels of 42 serum metabolites in the GC, AG and SG groups were detected by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Then, orthogonal partial least squares discriminant analysis (OPLS-DA) and the Kruskal-Wallis H test were used to identify a metabolomic signature among the three groups. Metabolites with highest significance were examined for further validation. Receiver operating characteristic (ROC) curve analysis was carried out for evaluating diagnostic utility.

Results: The metabolomic analysis found adipylcarnitine (C6DC), 3-hydroxy-hexadecanoylcarnitine (C16OH), hexanoylcarnitine (C6), free carnitine (C0) and arginine (ARG) were differentially expressed (all VIP >1) and could distinguish GC patients from AG and SG cases. In comparison with the AG and SG groups, GC cases had significantly higher C6DC, C16OH, C6, C0 and ARG amounts. Jointly quantitating these five metabolites had specificity and sensitivity in GC diagnosis of 98.55% and 99.32%, respectively, with an area under the ROC curve (AUC) of 0.9977.

Conclusion: This study indicates C6DC, C16OH, C6, C0 and ARG could effectively differentiate GC cases from AG and SG patients, and may jointly serve as a valuable circulating multi-marker panel for GC detection.

Keywords: Acylcarnitines; Amino acids; Atrophic Gastritis; Gastric cancer; Metabolomics; Superficial gastritis.

Figure 1. An orthogonal partial least square discriminant analysis (OPLS-DA) model.

(A) GC cases tended to cluster to the right, while the AG and SG groups clustered to the left. The OPLS-DA model revealed fairly separated GC from AG and SG. There was no separation between AG and SG. (B) Score plots for AG and SG cases. The model was unable to distinguish AG and SG. (C) and (E) OPLS-DA achieved a fairly distinct separation between GC and AG and between GC and SG. (D) and (F) The 200-time permutation test revealed no overt model over fitting. The y-axis intercepts were R2 (0.0, 0.136; 0.0, 0.127) and Q2 (0.0, −0.327; 0.0, −0.284).

Figure 2. VIP plot for the OPLS-DA model in preselecting variables.

(A) GC group versus AG group, *VIP >1. (B) GC group versus SG group, *VIP >1.

Figure 3. Box plots for select metabolites (Median (interquartile range)) differentiating GC, AG and SG.

* P < 0.05, ** p < 0.001, and *** p < 0.0001.