Differences of serum lipid profiles between CA and CRC
TIC chromatography of lipidomic study provided an overview for the general difference of lipid profiles among different groups. Through the direct observation of serum lipid profiles revealed a certain difference between CA and CRC groups by TIC chromatography in both ESI modes (Fig. 1A-D). PCA model was established to analyze the stability of instrument system and detection method as well as the distribution trend of samples in different groups. The three-dimensional score plots of PCA analysis showed that QC samples were clustered closely in both ESI modes, indicating that analysis system and detective method presented good stability and reproducibility during the batch analysis and satisfied the requirements of lipidomic analysis as well as illustrating no obvious drift of lipidomic features (Fig. 2A-B). As a whole, the samples of CA and CRC groups were relative clustering and located on two sides of the score plots, suggesting the obvious difference of serum lipidomic profiles between two groups in both ESI modes and indirectly reflecting their differences in lipid metabolism (Fig. 2A-B). Moreover, with respect to that in ESI- mode, the more obvious trend of separation between sample clusters of CA and CRC groups was observed in ESI+ mode (Fig. 2A-B). Meanwhile, most of the samples were within the 95% confidence interval and just a few samples were beyond the 95% confidence interval owing to a relatively discrete trend, possibly due to the existence of large individual differences (Fig. 2A-B). In addition, the peak areas of each lipid feature in the quality control group were extracted and the RSD values most of them were less than 20.0%, suggesting that analysis system had good stability and reproducibility in the whole analytical process.
To discover the differences in serum lipid profiles with the maximum extent between CA and CRC groups, an OPLS-DA model was constructed by the all-detected lipid features based on PCA analysis to screen potential biomarkers. Compared with PCA analysis, the three-dimensional score plots of OPLS-DA analysis showed that the difference of lipid profiles between the CA and CRC groups was more obvious, demonstrating distinct differences in serum lipid features between the two groups (Fig. 2C-D). The two sample clusters were clearly separated at ESI+ mode with good parameters (R2X (cum) = 0.336, R2Y (cum) = 0.976, Q2 (cum) = 0.893) and ESI- mode (R2X (cum) = 0.535, R2Y (cum) = 0.957, Q2 (cum) = 0.844), which disclosed apparent changes of lipid metabolism between two groups (Fig. 2C-D). These range of parameter values indicated pretty good explanation and predictability ability for this model. In addition, the reliability and suitability of OPLS-DA model were further analyzed and evaluated by 200 permutations test. The results provided a proof that the model was rational and not overfitting for the data analysis with the values of R2 (0.853 and 0.792), Q2 (-0.341 and -0.481) and P (CV-ANOVA) (0.000 and 0.000) in ESI+ and ESI- model, respectively (Fig. 2E-F).
Screening and identification of lipid biomarkers to differentiate CA from CRC
To minimize false positives, according to the multivariate statistical analysis of serum lipid profiles between two groups, the standard of fold change > 1.50 or < 0.67 and P < 0.05 was further used to screen the differential lipid species. Totally, 85 differential lipid species were selected and identified (including 27 in ESI- mode and 58 in ESI+ mode) (Table 2). These differential lipid species mainly included PCs: 35.29%, FAs: 15.29%, TAGs: 12.94%, PEs: 9.41%, PIs: 7.06%, SLs: 7.06%, SMs: 4.71%, LPCs: 3.53%, LPEs: 2.35%, PAs: 1.18%, PGs: 1.18% (Fig. 3). As shown in Fig. 3, PCs was the main component of differential lipid species accounting for more than 35%, followed by FAs and then TAGs. Hence, these three lipid types accounted for 63.52% of the total differential lipid species, indicating that the metabolic perturbation of PCs, FAs, and TAGs could be involved in the mechanisms underlying for CA canceration. Furthermore, compared with the CA group, most of the differential lipid species were significantly down-regulated in the serum of the CRC group, and only 11 of them were significantly up-regulated in the CRC group (Table 2). Additionally, we conducted a clustering heatmap analysis of the screened 85 differential lipid species to further study their level distribution in the individual samples of two groups. The results showed that the samples between two groups were clustered well, indicating that the between-group difference and intra-group similarity of these lipid profiles were relatively obvious (Fig. 4). As seen in Fig. 4, the levels of most differential lipid species mainly including PCs, FAs, and TAGs were significantly up-regulated in the CA group with respect to the CRC group, which is similar to the results shown in Table 2. In summary, PCs, FAs, and TAGs were considered to be the primary influencing factors associated with the carcinogenesis of CA developing into CRC.
Assessment of discriminant capability of serum differential lipid species between CA and CRC groups
Based on the MetaboAnalyst 5.0 software, ROC analysis is a useful tool to explore the biomarkers of diseases through evaluating the diagnostic and discriminating ability of metabolites or proteins according to the maximal area under the curve (AUC) values. To calculate the AUC values, the software used the trapezoidal algorithm to select the most appropriate cut-off point for each metabolite with good sensitivity and specificity. Prior to ROC analysis, the sum normalization and auto-scaling of lipidomic data were performed to effectively decrease the influence of individual differences and systematic errors (Fig. 5A-B). In the end, we identified 7 differential lipid species with good distinguishing performance (AUC > 0.8, ranging from 0.80 to 0.93) of high sensitivity (0.76 to 1.00) and specificity (0.72 to 0.98) (Fig. 6), and their identification results were achieved by matching high resolution MS, MS/MS fragments, and RT with Thermo mzCloud and mzVault with Lipidblast database (Fig. 7). Among them, docosanamide had the largest AUC values (AUC = 0.93, 95%CI (0.884-0.974)), indicating its excellent discriminate performance for CA and CRC groups, while SM d36:1 and SM d36:0 had the relatively low AUC values (AUC = 0.80) (Fig. 6). Then, a multivariate ROC curves analysis based on the SVM algorithm was performed in the automatic selection of the best lipid combinations to further screening the hypothetical lipid markers with good performance in distinguishing CA from CRC. The docosanamide, PC 36:1e, triheptanoin, SM d36:0, PC 37:7, SM d36:1, and PC 32:3 were found to be the most vital and frequently selected variables in the panel exploratory analysis of lipid markers, disclosing that these differential lipid species possessed a good discriminative performance for CA and CRC (Fig. 5C). According to the classification model established by the top 2, 3, 4, 5, 6 and 7 of differential lipid species, it could be seen that combination composed of the top 4 differential lipid species (docosanamide, PC 36:1e, triheptanoin, SM d36:0,) was the best potential biomarker panel owing to the AUC values in multivariate ROC curves analysis (AUC = 0.971, 95%CI = 0.921–1.000) (Fig. 5D). Therefore, biomarker-panel of these 4 differential lipid species could act as the potential biomarkers for malignant transformation of CA to develop into CRC.
Trend of differential lipid species with high performance for the distinction between CA and CRC
As shown in lipid profiles analysis between CA and CRC groups, significant difference was observed and totally 85 differential lipid species including 30 PCs, 13 FAs, 11 TAGs, 8 PEs, 6 PIs, 6 SLs, 4 SMs, 3 LPCs, 2 LPEs, 1 PAs, and 1 PGs were discovered (Table 2). Based on the above ROC analysis, the change trend for these 7 potential lipid biomarkers (including 3 PCs, 2 FAs, and SMs) with good distinguish efficacy between two groups was further explored. The results revealed that four lipid species, such as docosanamide, PC 37:7, PC 32:3, and triheptanoin, were significantly up-regulated in the CA group, while the other three lipid species including PC 36:1e, SM d36:1, and SM d36:0 were remarkably down-regulated in the CA group compared to CRC group (Fig. 8). Among them, docosanamide, PC 37:7, and triheptanoin exhibited the most obvious change trend with the fold change > 5 (Table 2), and which was consistent with the clustering heatmap of differential lipid species between groups (Fig. 4). These differential lipid species are mainly involved in the metabolic pathways of FA, PC, and TAG metabolism. Taken together, FAs and PCs were the main dysregulated lipid biomarkers to distinguish between CA and CRC groups. The perturbation of PC, FA, and TAG metabolism was associated closely with CA canceration.