Trans-omic analyses identified novel prognostic biomarkers for colorectal cancer survival.

Dear Editor, Colorectal cancer (CRC) is third most frequently diagnosed malignancy with the second highest mortality rate among all cancers.1-4 As CRC survival is highly dependent upon early diagnosis and treatment, novel sensitive and non-invasive detective approaches f are urgently required.5 CRC is characterized by metabolic reprogramming.6, 7 In present study, using plasma metabolome analysis, a large scale trans-omic network was constructed by connecting polar metabolite data, lipid data, clinical phenoms and survival outcomes of CRC patients. The study revealed that the excavation of trans-omics based on serum metabolome could potentially predict CRC survival. We believe that this trans-omics approach is vital for optimizing personalized treatment and the development of precision medicine for CRC. The plasma samples were collected and analysed for metabolomic and lipidomic analysis. Our experiment consisted of two cohorts (Table S1). In cohort 1, 78 CRC patients and 78 healthy control (HC) samples were collected for lipidomics analysis. In cohort 2, samples from 31 CRC patients and 31 HC were collected for metabolomics analysis (Figure 1A). We found that the polar metabolites and lipids were significantly different between CRC patients and HC (Figure 1B–F). There were eight main classes of lipid detected in our study, including lysophosphatidylcholines, phosphatidylcholines (PC), phosphatidylethanolamine (PE), diacylglycerol, sphingomyelin (SM), cholesterol ester (CE), ceramide and triacylglyceride (TAG). The occurrence and prevalence of CRC are closely related to patient age.8 We further sub-grouped the CRC patients by age into groups of adult (age less than or equal to 50 years) and senior (older than 50 years), and metabolic analysis was performed (Figure 1G). Volcano plots were generated by comparing the three groups in pairwise (Figure 1H). Multiple polar metabolites in the plasma were down-regulated in both the CRC groups compared with the HC group (Figure 1I,K, L). The expression of 5-aminovaleric acid and d-allose in adult CRC patients was higher than that of HC and senior CRC patients (Figure 1J,L). Pathway enrichment analysis identified signalling pathways associated with key polar metabolites between groups (Figure 1M). Pathways of fatty acid degradation were enriched in both senior and adult CRC groups, compared with HC. The variable influence on projection (VIP) score can summarize the importance of each variable. Differentially expressed eight polar metabolites (VIP > 1, p < .05 and FC > 2 or <.5) had shown a great diagnostic potential for CRC (Figure 1N). Averaged ROC curves revealed that these polar metabolites had great diagnostic potential for CRC (Figure 1O). These findings strongly suggested the potential application of polar metabolites as clinical indicator for the early diagnosis of CRC. OPLS-DA (orthogonal partial least squares discriminant analysis) is a statistical technique used for analysing multivariate data in order to identify the differences among different groups. It is commonly used in metabolomics and lipidomics research to identify biomarkers that are associated with different physiological or pathological conditions. We have used OPLS-DA for the pairwise difference lipids analysis among study groups (Figure 2A). The results of the difference analysis are presented as a volcano plot. In the volcano plot, each metabolite or lipid is plotted as a point, its position on the X-axis presents the magnitude of the change in the expression of the substance between the two groups, and its position on the Y-axis presents the statistical significance of the change. This profile can be used to identify candidate metabolites or lipids that may be involved in specific biological processes or diseases. Volcano plots were generated by comparing the three groups populations in pairwise (Figure 2B). Combining results of Venn diagram (Figure 2C, VIP > 1, p < .05 and FC > 2 or <.5), four common differential lipids were selected for further study. These bar charts found that these four lipids were all increased in CRC patients (Figure 2E). ROC analysis of the four common differentially expressed lipids was performed to determine the diagnostic power between groups. Averaged ROC curves confirmed that four lipids had great diagnostic potential in CRC (Figure 2D). With these volcano plots, TAGs of adult CRC group were significantly increased as compared to HC (Figure 2F). This phenomenon was also observed in the volcano plot of senior CRC compared with HC (Figure 2G). A significant number of lipids were up-regulated in adult CRC patients in comparison with that of senior CRC. Most of these lipids were belong to TAGs (Figure 2H). The correlation between polar metabolites and lipids was analysed by Spearman correlation test (Figure 3A). Network plot was used to illustrate these correlations (Figure 3B). By partitioning phenotypes, the lipid was correlated with laboratory examination (Figure 3C), pathognomonic symptoms (Figure 3D), signs (Figure 3D), symptoms (Figure 3E), underlying diseases (Figure 3E) and imaging results (Figure 3F). These results showed that lipids associated with clinical phenotypes were mostly TAGs, PCs and PEs. We further investigated the correlation of clinical phenoms and lipidomic results (Figure 3G). To evaluate the association between lipidomics profiles and patient gender, tumour markers and prognosis of human CRC, survival analysis using Log-rank (Mantel–Cox) test explores the impact of lipids metabolism on CRC prognosis. CRC patients were followed up for up to 36 months. The effects of gender on CRC patients were determined by survival curve analysis and found that females had higher mortality and lower survival rate as compared to that of males (Figure 4A). Heat map was used to present the top 5 up-regulated and 5 down-regulated lipids in female group (VIP > 1, p < .05). In females, various CEs were higher than that of males, and various TAGs were lower. Grouped by tumour markers of plasma carcinoembryonic antigen (CEA), carcinoma antigen, including CA125, CA199 and CA724, the survival curves of CRC patients were analysed and differential lipids analysis performed as shown in Figure 4B–E. Results indicated that all these four common tumour markers can predict the survival of these patients. However, results of the lipidomics analysis after regrouping CRC patients by various tumour markers were different. For example, most of the up-regulated lipids in CEA positive group were TAGs (mostly 46-carbon), and most of the down-regulated lipids were CEs, as compared to CEA negative group. Down-regulations were seen in multiple SMs in the CA125 positive group, in comparison to negative group. The elevated lipids in CA199 positive group were all PEs, which did not appear in other tumour marker positive groups. Both up- and down-regulated lipids in CA724 positive group were mostly TAGs, but the difference was that most of the up-regulated TAGs were 54-carbon TAGs, and most of the down-regulated TAGs were 50-carbon TAGs. Our results demonstrated that the combination of plasma lipids with tumour markers would not only be an effective biomarker for CRC, but also providing a new strategy for CRC diagnosis and monitoring. In conclusion, using an integrated trans-omics approach, the study found that altered metabolites varied among genders, ages and clinical phenome. Relationships between lipids and polar metabolites, lipids and clinical phenotype, and lipids and multiple common tumour markers were also delineated. Furthermore, the study demonstrated the magnitude and complexity of phenome-metabolic networks of CRC, which could facilitate the development of precision treatments that specifically target metabolites. The study underscored the great value of trans-omics to uncover the heterogeneity of metabolism among CRC patients. This study was conducted at the Institute of Clinical Science of Zhongshan Hospital, Fudan University. We sincerely thank all the staff and participants for their important contributions. This work was supported by grants from Shanghai Engineering Research Center of Tumor Multi-Target Gene Diagnosis (20DZ2254300), and Key Subject Construction Program of Shanghai Health Administrative Authority (ZK2019B30), the National Natural Science Foundation of China (82200043), Science and Technology Commission of Shanghai Municipality (21ZR1459000) and Quanzhou City Science & Technology Program of China (2018N008S). All authors obtain permission to acknowledge supports from all those mentioned in this section. The authors declare that they have no conflict of interests. The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.