Predictive Metabolomic Profiling of Microbial Communities Using

Predictive Metabolomic Profiling of Microbial Communities Using

ARTICLE https://doi.org/10.1038/s41467-019-10927-1 OPEN Predictive metabolomic profiling of microbial communities using amplicon or metagenomic sequences Himel Mallick 1,2, Eric A. Franzosa 1,2, Lauren J. Mclver1,2, Soumya Banerjee 1,2, Alexandra Sirota-Madi1,2, Aleksandar D. Kostic 1,2, Clary B. Clish 1, Hera Vlamakis1, Ramnik J. Xavier 1,3,4,5 & Curtis Huttenhower 1,2 1234567890():,; Microbial community metabolomics, particularly in the human gut, are beginning to provide a new route to identify functions and ecology disrupted in disease. However, these data can be costly and difficult to obtain at scale, while amplicon or shotgun metagenomic sequencing data are readily available for populations of many thousands. Here, we describe a compu- tational approach to predict potentially unobserved metabolites in new microbial commu- nities, given a model trained on paired metabolomes and metagenomes from the environment of interest. Focusing on two independent human gut microbiome datasets, we demonstrate that our framework successfully recovers community metabolic trends for more than 50% of associated metabolites. Similar accuracy is maintained using amplicon profiles of coral-associated, murine gut, and human vaginal microbiomes. We also provide an expected performance score to guide application of the model in new samples. Our results thus demonstrate that this ‘predictive metabolomic’ approach can aid in experimental design and provide useful insights into the thousands of community profiles for which only meta- genomes are currently available. 1 Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA. 2 Department of Biostatistics, Harvard T. H. Chan School of Public Health, Boston, MA 02115, USA. 3 Center for Computational and Integrative Biology, Massachusetts General Hospital and Harvard Medical School, Boston, MA 02114, USA. 4 Gastrointestinal Unit and Center for the Study of Inflammatory Bowel Disease, Massachusetts General Hospital and Harvard Medical School, Boston, MA 02114, USA. 5 Center for Microbiome Informatics and Therapeutics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. Correspondence and requests for materials should be addressed to R.J.X. (email: [email protected]) or to C.H. (email: [email protected]) NATURE COMMUNICATIONS | (2019) 10:3136 | https://doi.org/10.1038/s41467-019-10927-1 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10927-1 dvances in high-throughput metabolomics technology Here we describe MelonnPan (Model-based Genomically have enabled comprehensive coverage of a large number Informed High-dimensional Predictor of Microbial Community A 1 of small-molecule metabolites in microbial communities . Metabolic Profiles), a computational framework to predict com- Analysing metabolic differences between differentially regulated munity metabolomes from microbial community profiles. Mel- biochemical pathways can facilitate the discovery of potential onnPan infers the composite metabolome by enabling (1) data- biomarkers associated with disease and provide insights into the driven identification of an optimal set of predictive microbial underlying pathogenesis2,3. This has been highlighted by an features, and (2) robust quantification of the prediction accuracy increase in studies that rely on multi’omic profiling to simulta- of the well-predicted metabolites. This allows researchers to neously characterize community ecology, metabolic signatures, reproducibly infer metabolites for communities from which only and functional attributes of the human microbiome or other metagenomes are currently available. We applied MelonnPan to environments4–12. For example, among the best-studied bioactive two independent gut metagenome data sets comprising >200 microbial metabolites influencing human health are the short- patients with Crohn’s disease (CD), ulcerative colitis (UC), and chain fatty acids (SCFAs) including propionate, butyrate, and healthy control (HC) participants. This revealed high con- acetate, which have been implicated in the pathogenesis of several cordance between predicted and observed community metabolic diseases, including inflammatory bowel disease (IBD) and col- trends in >50% of metabolites whose identities were confirmed orectal cancer13–15. Other examples include the bile acids16, against laboratory standards, including prediction of metabolic sphingolipids17, and tryptophan derivatives18 all with evidence of shifts associated with bile acids, fatty acids, steroids, prenol lipids, microbial interactions and bioactivity in the gut. and sphingolipids. When using taxonomic features from ampli- Inferring the capacity of a microbial community to produce con sequencing profiles, similar accuracy was maintained for molecules and using large-scale data sets to connect new specific coral-associated, murine gut, and human vaginal microbial genes to metabolites is thus an essential first step towards the goal communities as well. The implementation of MelonnPan, asso- of understanding how and why gut microbiome metabolism ciated documentation, and example datasets are made freely affects human health19. The strength of association between gut available in the MelonnPan software package at http:// microbial and metabolic profiles suggests that it may be possible huttenhower.sph.harvard.edu/melonnpan. to approximately predict the metabolomic activities or features of microbial communities from metagenomes, based on their taxonomic or functional profiles. Easily identifying such asso- Results ciations purely based on enzymatic roles is greatly limited by the The MelonnPan algorithm. We have developed MelonnPan as a currently unsaturated repertoire of gene–metabolite reactions, as computational method to predict metabolite features from well as by the relative (rather than absolute) abundance measures amplicon or metagenomic sequencing data by incorporating provided both by typical sequencing and metabolomic technol- biological knowledge in the form of either taxonomic or func- ogies. Despite these limitations, however, approaches that predict tional profiles. Unlike existing stoichiometry-based methods that metabolite features associated with gut microbial profiles can rely on a limited number of well-characterized taxa, enzymes, and serve as a hypothesis generator that can facilitate population-scale metabolites, functional annotation is not necessary for Melon- discovery of novel associations (e.g. in large metagenomic data nPan, as the tool is designed to capture insights using machine collections) and lead to new sets of testable hypotheses, serving as learning even from uncharacterized microbial features. In this a complementary adjunct to experimental validation studies (e.g. manuscript, we discuss specifically its application to the human as has been the case for predictive functional profiling from gut microbiome, but the methodology is generalizable to any amplicon data20). appropriately profiled microbial environment. Briefly, Melon- Recently, a few studies have taken initial steps to carry out such nPan uses elastic net regularization28 to identify which features predictions in the subset of cases with prior knowledge of the (taxonomic or functional) are predictive for a given metabolite. mechanisms linking microbiome and metabolome (e.g. from Given a new taxonomic profile (from amplicons or a metagen- stoichiometric enzyme reaction matrices derived from databases ome) or metagenomic functional profile (i.e. gene family abun- such as Kyoto Encyclopedia of Genes and Genomes (KEGG)21). dances), it then combines a subset of the sequence features to One set of approaches, collectively referred to as Predicted estimate the associated composite metabolome. The Reactive Metabolic Turnover (PRMT), calculate community- resulting predicted metabolites are each the weighted sum of based metabolite potential (CMP) scores, which represent the relative abundances of predictive features (taxa or gene families), relative capacity of the community in a given sample to generate where the regression coefficients from the trained elastic net or deplete each metabolite22–24. Other methods reconstruct pre- model are used as weights in the prediction algorithm (Fig. 1). dictive metabolic models of community metabolism in either a In the fitting stage, MelonnPan is trained using samples for constraint-based or a network-based modeling framework25–27. which both sequencing data and experimentally measured One common drawback of both these approaches is their inability metabolite abundances are available (Fig. 1a). Both measures to distinguish between failure to predict due to missing annota- are effectively relative abundances—normalized reads or spectral tion or accurate reaction information in the reference database counts, respectively. The training (and, later, inference) meta- and failure due to alternative biological mechanisms, making genomes can be profiled with any system that quantifies relative them difficult to apply or validate in a data-driven manner. In abundances of taxa or functionally related microbial gene addition, these methods depend on accurate characterization and families; here we use previously profiled amplicon data and annotation of species- and even strain-specific metabolites, and metagenomes functionally profiled by HUMAnN229 with Uni- they do not scale well to complex communities with partially Ref90 as the reference catalogue, i.e. clustered sets of sequences referenced taxa or metabolites. All these studies

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    11 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us