Large-scale meta-analytic approaches for systematic and reproducible associations between the human microbiome and host's conditions

Us as humans are colonised by many microbial communities (the human microbiome) that interact with and regulate the host's physiology, and have been linked with several diseases. The high number of interactions that intercurre between the microbiome and the host requires rigorous statistical approaches to link any condition of interest to microbiome data. Many publicly available microbiome datasets are available that allow to study such interactions. However, strong inconsistencies are found among the reported associations when looking at the same condition in different studies. On the road to consistent statistical microbiome analyses that rely on public data, lack of standardisation and availability are barriers to define reproducible and generalisable associations. The main aim of my PhD was the development of meta-analytical approaches to identify microbial signatures as general hallmarks of health versus disease, integrating diverse cohorts and conditions. During my PhD training, I first explored the associations between the oral microbiome and peri-implantitis, an oral disease of dental implants, in which I defined a microbial signature discriminating diseased from control samples. I further developed and applied discriminative models to multiple colorectal cancer (CRC) cohorts, showing that the microbial signature defined on CRC samples is shared across different populations. To be able to further generalised microbial signature with host's conditions through a meta-analysis approach, I collected and analysed 20,533 public metagenomes from 90 cohorts, that are available through the curatedMetagenomicData (cMD) version 3, an R package providing standardised taxonomic and functional profiles and manually curated metadata. The cMD3 resource was used to derive an easy-to-compute oral-to-gut introgression score that I found systematically associated in a large meta-analysis of twelve diseases and with ageing. Finally, I applied the meta-analysis approach to study diet interventions in mice, exploiting a novel approach able to profile the unexplored fraction of microbiomes, and showing associations driven by previously uncharacterised species. Overall, this thesis contributes to strengthening the links between human and animal microbiomes in normal and altered host conditions.