Abstract
Background: Pancreatic cancer (PC) is projected to become the second most mortal type of cancer by 2030. Several factors increase the PC risk. However, these conditions are not enough to define high-risk populations towards preventive interventions. Oral and gut microbiota have been associated with PC risk, and a gut microbiota signature has shown to be a good predictor of PC risk. Preliminary studies point out that the metabolomic profile is associated with PC risk and can be used for further stratification. Our purpose was to build an integrative predictive model based on microbiome, jointly with serum, stool, and urine metabolome
features, to identify PC high-risk population
Methods: The PanGenMic population was selected on the basis of oral, blood, urine, and stool samples availability with a total of 44 PC cases and 38 hospital controls. All subjects had metabolomics data generated with NMR assays quantify metabolite’s concentration in stool (25 metabolites) and urine (25 metabolites), and with liquid/gas chromatography technology to quantify serum’s metabolites (470-2000 metabolites). Oral and faecal samples were processed to extract and sequence the bacterial 16S (250 and 290 ASVs, respectively). Zero inflation was managed through pseudocounts, missing values were imputed following bayesian strategy, and Z-score normalization was implemented. A Kernel-based Bayesian regression model was fitted to obtain both the univariate layer risk score and the integrated risk score. All analyses were adjusted for sex, age, centre, analytical batch, and diabetes mellitus. The predictive ability of the models were evaluated with the area under the ROC curve using a crossvalidation approach.
Results: The combined 7 layers of data showed an AUC of 0.80 (SD 0.11). The best-case combination included, serum gas (i.e. best univariate predictor) plus faecal microbiota and metabolome (i.e. lowest correlation to serum gas), with an AUC of 0.83 (SD 0.09).
Conclusion: This is the first study that compares and combines metabolome and microbiome performance into PDAC risk prediction. Regarding predictability, metabolome outperformed microbiome. The combination of faecal metabolomic þ faecal microbiomic þ serum gas metabolomic’s risk score provided the greatest AUC, pointing to the translational potential of this assay for the sake of applicability.
