A comprehensive metabolomics urine study has been done for the first time with both volatile and liquid fractions of urine for CRC. We have analyzed by mass spectrometry (MNS) either with gas chromatography (GC-MS, GCxGC-MS), and liquid chromatography (LC-MS) over 200 urine samples from a CRC screening program. Big part of the work has been the learning of new analytical techniques by the researcher fellow at UC Davis, in US. Even though the project has been affected by the pandemic, we have finalized the samples analysis during the outgoing phase. During the lock-down periods in which we could not do laboratory work, we performed a systematic review and meta-analysis of CRC markers in urine reported so far for volatiles (volatilomics) and metabolites (metabolomics).
Comprehensive gas chromatography is very well suited for the measurement of complex matrices, such those found in metabolomics. However, the high dimensionality of the raw data obtained makes the analysis and processing difficult to use in an automated way. A new algorithm for GCxGC has been developed, in an open-source code.
The repeatability and reproducibility of the GC-MS method was assessed, and it was found that samples kept at 4ºC before analysis exhibited an increase in the number of linear and reproducible peaks compared to room temperature (RT). Furthermore, repeatability and reproducibility of the method also improved when samples were stored at 4ºC. This issue was only observed in automatized sampling, meaning the samples remained longer hours sitting at RT. To overcome this issue, a Peltier cooled drawer was acquired for the samples while waiting for their processing (stirring, heating, solid phase microextraction (SPME) exposure and finally, chromatographic analysis).
From the univariate and multivariate statistical analysis, 10 metabolites were found relevant, however they were reduced to 2 with a LASSO selection. Noteworthy, one of the biomarkers is specific of CRC staging. The area under the curve (AUC) of the CRC progression model is 0.890 meaning we have a very good model to explain the differences between early and metastatic CRC groups. The model ROC curve was created with a random forest algorithm. The prediction test was further evaluated with a validation set of metastatic cancer patients, obtaining a significant model with a similar AUC of 0.869.
