Integrative holo-omic data analysis predicts interactions across the host-microbiome axis

Abstract

ABSTRACT Understanding the interplay between host organisms and their microbiomes is central to the development of sustainable food systems. However, high dimensionality and spurious associations remain major obstacles to extracting meaningful biological insight from multi-omic host-associated microbiome data; a challenge further exacerbated when “holo-omic” analyses across the host-microbiome boundary are considered. Here, we show that a computational method designed for multi-omic analysis in eukaryotes can be leveraged to integrate and analyze five layers of holo-omic data from porcine hosts and their gut microbiomes. We collected caecal tissue and digesta samples during a feeding trial that tested the impact of microbiota-directed fibers (acetylated galactoglucomannan) at critical developmental stages. From 800,000 features including microbial and host genes, metagenome-assembled genomes, and metabolites from caecal tissue and digesta, we used multiset correlation and factor analysis to select the most relevant features for capturing coordinated patterns across omic layers. From these features, we predicted over 2,000 putative host-microbiome interactions based on co-occurrence. Some interactions reflected previously known relationships between animal and microbiome features, such as microbial genes for carbohydrate metabolism being linked to glycoside abundances in host tissue. Other predicted co-occurrences included features that were not detected in single-omic analysis and offer new hypotheses of host-microbiome interactions that warrant future investigation. Hence, we showcase an application of holo-omic analysis that avoids common pitfalls in high-dimensional data analysis, identifies known interactions as a form of validation, and most importantly, predicts new leads for understanding host-microbiome symbiosis. IMPORTANCE While study systems involving mammalian hosts and their microbiomes are inherently complex, multi- and holo-omic analyses promise to provide interpretable results with translational value for the animal production industry. Unfortunately, computational methods capable of this kind of integration are currently scarce, as most existing multi-omics approaches have been developed for analysis of data layers within a single multicellular organism. We propose to adapt existing multi-omic methods for holo-omics by combining feature selection and interaction inference. This two-step analysis approach addresses common challenges in data-driven studies and can be implemented with a variety of tools for feature selection and interaction modeling. Through this holistic approach, we show that both known and novel relationships across the holobiont axis can be identified in a data-driven manner, offering new targets for the continued study and experimental validation of host-microbiome interactions and the effect of dietary interventions on production animals. While study systems involving mammalian hosts and their microbiomes are inherently complex, multi- and holo-omic analyses promise to provide interpretable results with translational value for the animal production industry. Unfortunately, computational methods capable of this kind of integration are currently scarce, as most existing multi-omics approaches have been developed for analysis of data layers within a single multicellular organism. We propose to adapt existing multi-omic methods for holo-omics by combining feature selection and interaction inference. This two-step analysis approach addresses common challenges in data-driven studies and can be implemented with a variety of tools for feature selection and interaction modeling. Through this holistic approach, we show that both known and novel relationships across the holobiont axis can be identified in a data-driven manner, offering new targets for the continued study and experimental validation of host-microbiome interactions and the effect of dietary interventions on production animals.