Creating a Metagenomic Data Analysis Pipeline Using Simulated Infant Gut Microbiome Data for Genome-Resolved Metagenomics in the Infant Gut Microbiome

Abstract

Background: Studying the infant gut microbiome during the period of solid food introduction may provide valuable insight into gut colonization, microbial evolution, and the ecological role of bacterial metabolic pathways in microbial succession. However, since infant gut microbial communities are made of bacterial genera with high relative abundance, within-genus and within-species diversity, the efficacy of current computational tools in elucidating strain-specific differences is not known.

Methods: 34 infant gut metagenomic samples were simulated with the CAMI-Simulator, using 16S rRNA gene profiles from subjects of the Baby & Mi study as a reference. Raw simulated reads were trimmed, assembled, and binned into metagenome-assembled genomes (MAGs) using mg_workflow, a Snakemake-based pipeline of current metagenomic analysis protocols. Results were compared to gold-standard references in order to benchmark the success of current computational methods in retrieving strain-level MAGs from the gut, and in predicting bacterial carbohydrate active enzymes. Real metagenomic samples from the Baby, Food & Mi cohort were processed through the bfm_mg_flow pipeline to study the taxonomic and metabolic changes in the infant gut microbiome during the solid food introduction period. Post-pipeline analyses were conducted in R.

Results: Misassemblies were significantly impacted by sample community composition, including Shannon diversity, number of strains in the sample, and relative abundance of the most dominant strain. MAG completeness, contamination, quality, and reference coverage were significantly impacted by choice of assembly software, and choice of single- or co-sample assembly. Different assemblies yielded different MAGs from the same samples. Reference coverage of MAGs recovered from co-assemblies were lower than for those from single assemblies and CAZyme predictions were more accurate from MetaSPAdes than from MEGAHIT assemblies at both the assembly-level and the MAG-level. Based on these results, we propose the MetAGenomic PIpelinE (MAGPIE), with recommendations for ensemble methods for assembly, binning, and gene predictions. Using these methods, we identified changes in microbial community composition before and after solid food introduction in real Baby & Mi infant gut samples. These changes included an increase in bacteria that can digest a wide variety of carbohydrates, such as Bacteroides, and a decrease in Bifidobacterium.

Conclusions: In this study, we characterized the current state of tools for genome-resolved metagenomics, and contributed a framework to tailor metagenomic data analysis for the unique composition of the infant gut microbiome. We further used this framework to study bacterial metabolism in the infant gut microbiome before and after the introduction of solid foods.