Overall differential metabolite and lipid profiles between normal colon cells and CRC cells
PCA model was first built to evaluate the natural clustering of samples and to identify outliers. PCA score plots showed that QC samples were clustered closely in both ESI modes in metabolomic and lipidomic analyses, indicating that the analysis system and detection methods presented good robustness and reproducibility during the batch analysis of samples. Also, clear separation between the control cell group and CRC cell groups was observed in PCA analysis, confirming their evident differences in metabolite and lipid profiles (Fig. 1A1, B1, C1, and D1).
Next, OPLS-DA model was created to further describe inter-group differences. Results showed that all the five groups were well discriminated. For metabolomics, good model parameters in ESI + mode [R2X(cum) = 0.787, R2Y (cum) = 0.988, Q2 (cum) = 0.975] and in ESI- mode [R2X (cum) = 0.776, R2Y (cum) = 0.989, Q2 (cum) = 0.974] were obtained, respectively, which indicated remarkable differences in metabolite features among groups in both ESI modes (Fig. 1A2, B2). Similarly, for lipidomics, the five sample clusters were clearly separated with good model parameters in ESI + mode [R2X (cum) = 0.930, R2Y (cum) = 0.999, Q2 (cum) = 0.998] and ESI − mode [R2X (cum) = 0.970, R2Y (cum) = 0.999, Q2 (cum) = 0.999], respectively, which also revealed obvious differences of lipid features among groups (Fig. 1C2, D2). Meanwhile, 200 times permutation tests were conducted to further assess the reliability and applicability of OPLS-DA model. Results of metabolomic analysis exhibited that the values of R2Y and Q2 were 0.297 and − 0.457 in ESI + mode, R2Y and Q2 were 0.241 and − 0.479 in ESI- mode, respectively. And values of P (CV-ANOVA) were less than 0.001 in both modes (Fig. 1A3, B3). Permutation tests resulted in lipidomics were presented with the values of R2Y (0.195 and 0.152), Q2 (-0.424 and − 0.428), and P (CV-ANOVA) (all less than 0.001) in ESI + and ESI-modes, respectively (Fig. 1C3, D3). All parameters of the permutation test provided evidence that the OPLS-DA model was rational and not overfitted for the data analyses of metabolomics and lipidomics. By this point, UHPLC-HRMS-based multiomics provided relatively comprehensive metabolites and lipids profiles to differentiate the control cell group from CRC cell groups.
Exploration of metabolites and lipids to distinguish CRC cells from normal colon cells
On the basis of the metabolite and lipid profiles, combined with FC > 2 or < 0.5, P < 0.05, a total of 82, 94, 80, and 90 differential metabolites were selected and identified in the between-group analyses, including Control vs. Stage A, Control vs. Stage B, Control vs. Stage C, and Control vs. Stage D, respectively. As shown in Fig. 2A, these differential metabolites mainly included amino acid (AA), fatty acid (FA), glycerophospholipid (GPL), nucleotide (NL), organic acid (OA), polyamine (PAM), sphingomyelin (SM), steroid (ST), and sugar (SUG). Among them, lipids were the main component in CRC cells at different stages with the percentage of composition more than 50% (Fig. 2A). It is also worth noting that GPL accounted for the most proportion with more than 25% in the total differential metabolites of four stages of CRC cells, followed by FA and AA (Fig. 2A). Additionally, GPL was composed of lysophosphatidylcholine (LPC), lysophosphatidylethanolamine (LPE), lysophosphatidylglycerol (LPG), phosphatidylcholine (PC), and phosphatidylethanolamine (PE) in all differential metabolites (Fig. 2B). Furthermore, the percentage of PC is significantly higher in CRC cells (Fig. 2B).
To further investigate the distribution and trend of relative levels of differential metabolites between the two groups, a clustering heatmap analysis was utilized. The findings revealed that the levels of most differential metabolites dominated by GPL were significantly up-regulated in CRC cells at different stages compared to the control cell group (P < 0.05, Figures. S1-S4) (see Additional file 1).
Owing to the limitation of the sample preparation method in metabolomics to extract lipids, we further systematically and extensively investigated the dysregulation of lipid metabolism in CRC cells. Similarly, differential lipids in the between-group analysis were screened according to the above criteria (FC > 2 or < 0.5, P < 0.05) based on the differences in lipid composition found in the multivariate analysis. 104, 92, 102, and 122 differential lipids were selected by comparison of Control vs. Stage A, Control vs. Stage B, Control vs. Stage C, and Control vs. Stage D, respectively. The lipids discriminating CRC cells from normal colon cells were grouped into classes including FA, GPL, SM, ST, sphingosine (SPH), and ceramide (CER) (Fig. 2A). Interestingly, similar to metabolomics results, GPL disorders were the most evident in these four stages of CRC cells with a composition ratio of more than 75% (Fig. 2A). Moreover, PC took up the highest ratio of differential GPL (the proportion is more than 65%), implying that the metabolism of PC was most obviously dysregulated in CRC (Fig. 2C). Clustering of the differential lipids declared that most PC were notably up-regulated in CRC cell groups compared to the control cell group (P < 0.05, Figures. S5-S8) (see Additional file 1).
Searching for differential metabolites and lipids common to different stages of cancer cells to characterize CRC
Molecular characterization of CRC cells is crucial to diagnosis for CRC. The association of differential metabolites and lipids of the above four groups was confirmed by Venn diagram analysis. These common differential metabolites and lipids may be potential biomarkers for the diagnosis of CRC. Therefore, their diagnostic performance was evaluated and their area under the curve (AUC) values were listed in Tables S1-S2 (see Additional file 1). In metabolomics, the results showed a high similarity of the differential metabolites among the groups. 46 differential metabolites common to CRC cells at different stages were found, including 34 in ESI + mode and 12 in ESI- mode, respectively (Fig. 3A and Table S1) (see Additional file 1). The composition of these 46 differential metabolites included 10 AA, 10 FA, 6 ST, 5 NL, 3 LPE, 3 LPG, 3 PC, 2 OA, 2 PAM, 1 LPC, and 1 SUG (Fig. 3B). Among them, lipids made up a high percentage of more than 50% (26/46), meaning that lipids as the major components showed obvious metabolic disorders in CRC (Fig. 3B), similar to the results in Fig. 2. Particularly, AA and FA occupied the first place in equal numbers in 46 shared metabolites in different stages of CRC cells, which is noteworthy. This suggests that the metabolism of AA and FA may get involved in the occurrence and development of CRC.
To discover the pivotal dysfunctional lipids in the different four stages of CRC cells, Venn diagram analysis was also applied to lipidomics. In the lipidomics, there were 29 shared differential lipids in CRC cells at different stages, with 12 in ESI + mode, and 17 in ESI- mode (Fig. 3D and Table S2) (see Additional file 1). These shared differential lipids included 15 PC, 6 LPE, 5 PE, 1 SM, 1 SPH, and 1 ST (Fig. 3E). Similarly, PC was also the key disorder category, which was consistent with the composition of differential lipids shown in Fig. 2. Cluster analysis showed that differential metabolites and lipids common to CRC cells at different stages were mostly up-regulated compared to normal colon cells (P < 0.05, Fig. 3C and 3F). Consequently, the results of lipidomics further confirmed that disruption of PC metabolism may be the key determinant in the malignant process of CRC.
Performance evaluation of potential biomarkers for the diagnosis of CRC
After selecting the common differential metabolites and lipids in CRC cells at different stages, ROC analysis was applied to evaluate their diagnosis performance for CRC. As shown in Tables S1-S2 (see Additional file 1), in general, the AUC values of most differential metabolites and lipids were relatively high. Then, the change trend of these common differential metabolites and lipids was observed, those that have the same change trend as differential metabolites and lipids in the serum of CRC patients deserve further research. Ultimately, eight differential compounds including palmitoylcarnitine, oleoyl ethanolamide, ACar 16:1, tryptophan, LPC 16:0, LPE 16:0, PC 22:5e and sphingosine were picked out (Fig. 4). Among them, LPE 16:0 and PC 22:5e were significantly up-regulated in CRC compared to the control groups, while the remaining 6 differential metabolites were significantly down-regulated (P < 0.05, Fig. 4).
ROC analysis was performed on these eight differential metabolites in the serum of CRC patients, their ROC curves were showed in Fig. 4. Among them, palmitoylcarnitine and sphingosine were identified with good discrimination performance for CRC (AUC > 0.80), with relative high sensitivity and specificity (more than 0.70). And, the AUC values of these two compounds in CRC cells at different stages were mostly 1.00 (except for palmitoylcarnitine in CRC cells at stage C with AUC value less than 1.00).
To further confirm these two differential compounds, the precursor and their fragments of metabolite standards were matched with mass spectrometry (MS) information of database under the same data acquisition conditions. The results showed that the two metabolite standards matched well with the MS and MS/MS spectra of the database, and the high-resolution mass error was less than 10 ppm (Fig. 5A).
What’s more, to seek an optimized diagnostic model, multivariate ROC analysis was employed on biomarkers panel composed of palmitoylcarnitine and sphingosine. As a result, the panel showed excellent performance in discriminating CRC cells from normal colon cells, with AUC values of 1.00 in all diagnostic models (Fig. 5B). This panel may have an important application value in the diagnosis of CRC, and this further prove that palmitoylcarnitine and sphingosine have great potential as diagnostic biomarkers for CRC on the basis of results from human serum.