Oral/Gut microbiome profiles in pancreatic cancer and their interactions with dietary patterns

Abstract

Background and aims: Pancreatic cancer (PC) has a high case-fatality rate in Western countries, expected to rise in coming years if no immediate actions are taken. Several studies have pointed to an association between the human microbiome and PC risk. Diet is known to play a key role in the microbiota composition in terms of abundance and diversity, and could therefore interact in the microbiome-PC relationship. However, the association between diet and PC is complex. Within the PanGenEU and PanGen-MICROBIOME studies, we aimed to identify a gut and/or oral microbiome signature to detect PC, and to explore interactions with dietary factors. Methods: The MICROBIOME study counts with metagenomic and 16srRNA data of the stool and salivary microbiota of over 50 PC cases and 50 matched controls. In addition, there is 16sRNA data on the salivary microbiota of more than 500 subjects (~250 cases and controls) from the PanGenEU study. After data processing, we compared alpha and beta-diversity measures among cases and controls regarding the oral and gut microbiome. Associations between taxa with case-control status were examined in univariate (Wilcoxon test) and multivariate analyses (via edgeR). Then, a microbiome-based classifier was explored (via LASSO regression) to discriminate between cases and controls. To explore interactions between dietary factors and bacterial taxa, we calculated several dietary scores (the diabetes risk reduction diet score DRRD and the relative Mediterranean Diet score rMED) and microbial risk scores based on the bacterial species conforming the signature. The latter comprised an abundance-based microbial risk score (MRS) and two alpha-diversity measures (richness score, rS, and Shannon score, SS). The association between dietary factors and these scores and with the PC risk was examined using logistic and linear regression models adjusted for potential confounders. Ridge Regression and LASSO as feature selection methods were used to identify foods, food groups and nutrients related to the microbial risk scores. Microbial species related with the dietary scores were also selected using this method. Tuning parameters were optimized on training and test sets. At the individual level, associations (e.g. Spearman correlations) between taxa of both the gut and oral microbiota with foods, food groups and nutrients were explored. Also, hierarchical clustering was applied to identify clusters of taxa and of foods, food groups and nutrients among PC cases and controls. P-values less than 0.05 were deemed significant; p-values were corrected for multiple testing (Benjamini-Hochberg, BH and applying FDR) in association analyses. Results: A stool metagenomic signature of 27 microbial species that discriminated between PC and controls with an accuracy of up to 0.84 area under the curve (AUC). Its performance improved to up to 0.94 AUC when serum levels of CA-19.9, the current diagnostic marker of PC, were incorporated into the signature. The microbial risk signature MRS reflected better PC risk (OR per 1SD increase in MRS=5.5, p= 3.01E-05) than those based on alpha-diversity measures (rS or SS). Regarding its relationship with dietary factors, the most meaningful variables were seafood, some alcoholic beverages, and some polyunsaturated fatty acids (PUFAs). None of the dietary scores, the DRRD and the rMED scores, were significantly associated with PC risk. However, some features in the gut microbiome appeared to be linked to these scores. Also, cluster analyses revealed the existence of gut microbial taxa and diet clusters, with groupings of these taxa with plant-based foods, seafood and PUFAs, mainly. Conclusion: A distinctive gut microbial signature made up of 27 bacterial species allows to discriminate between PC cases and controls. Several dietary factors were related to this microbial signature in terms of a risk score. Future studies with larger sample size are warranted to confirm these findings.