Butyrate

Stool samples from 15 different individuals were randomly selected from the HMP Data Analysis and Coordination Center (http://www.hmpdacc.org; parameters defining health can be obtained from the website). Raw nucleotide read sequences were aligned (blastn) against our database, requiring a minimum alignment length of 70 bp and sequence identity of ≥80%. Only the best-scoring alignment (lowest E value) was used for further analysis. The abundance of individual butyrate-producing pathways (Fig. 4) was calculated as follows: (i) (#reads_tot × length_pathway)/4 × 10⁶ bp = th_100%, and (ii) #reads_pathway/th_100% = result (genomes exhibiting pathway [%]), where #reads_tot is the total number of reads for a sample, length_pathway stands for the total length (bp) of all unique pathway genes (calculated from the median length of all entries in the database for a specific gene), 4 × 10⁶ bp corresponds to an average genome size, th_100% is the theoretical number of reads if all genomes exhibit the pathway, and #reads_pathway corresponds to the number of reads matching the pathway (BLAST result). Detailed results are presented in Fig. S7 in the supplemental material.
Prior to diversity analysis, individual genes from the database were subjected to multiple complete linkage clustering (using the Pyrosequencing Pipeline provided by the Ribosomal Database Project; http://rdp.cme.msu.edu) on the nucleotide level, applying a 10% cutoff. All genes of an individual pathway clustered very similarly (clusters for all individual pathway genes were usually associated with the same genomes), allowing us to group individual clusters of all genes of a specific pathway together. Thus, obtained groups contained all genes of a specific pathway. If cluster results varied between genes (e.g., all thl genes from three candidates cluster together, whereas two clusters were generated for the hbd gene), then clusters were manually merged (e.g., merging of all three hbd genes as associated thl genes) to achieve consistency, and the most conservative approach was always applied, i.e., clusters were only merged and never split. Genes of the same strain were always merged. For metagenomic analysis, a specific group (e.g., the group Faecalibacterium prausnitzii for the acetyl-CoA pathway consists of all pathway genes from all five strains of this taxon) was considered present only if all pathway genes could be identified for that group in the BLAST result (thus, BLAST hits did not have to match all genes from the same strain but only from the same group—an example [sample A] is shown in Fig. S5 in the supplemental material). Results presented in Fig. 5 are a median value for all individual pathway genes (see Fig. S5). The degree of explanation was calculated as the percentage of reads matching groups that were included in the diversity analysis (average from individual genes) from the total number of reads matching any gene in the database.

Individual pathways shown in Fig. 1 are based on KEGG with modifications. Most importantly, the entire lysine pathway and certain steps in the 4-aminobutyrate pathway are not present in KEGG and were included based on references 22 (link) and 43 (link). KEGG additionally displays the conversion from butanol to butyrate, which was not included in this study. Furthermore, a possible route from acetoacetate via poly-β-hydroxybutyrate and crotonoyl-CoA to butyrate is suggested in KEGG. However, this pathway contains an unlikely reverse reaction of extracellular poly-β-hydroxybutyrate degradation enzymes that differ considerably from intracellular depolymerases (44 (link)), and this route was hence not considered. The stereospecific separation between R-hydroxybutyrate and S-hydroxybutyrate in the acetyl-CoA pathway was omitted, and the two routes were merged.
Screening of genomes was divided into two main parts, where the first was based on EC number searches (from KEGG) within the Integrated Microbial Genome (IMG) (http://img.jgi.doe.gov) database and the second part used HMM models (both approaches were applied on a protein level). A detailed schematic representation of the work flow and abundance of obtained candidates (and associated genes) at each step is given in Fig. S1 in the supplemental material. First, all genes matching individual EC numbers were obtained, and the data were queried for all candidates exhibiting all genes of a specific pathway. Since several model butyrate producers failed the query, we allowed for one missing gene in each pathway. Candidates were then subjected to synteny analysis (see Fig. S1 and Text S1 in the supplemental material). Since it was proposed that several different gene products are able to catalyze the final step in the acetyl-CoA pathway and their location is often apart from other genes in this pathway, we excluded the terminal enzymes here and treated them in separate analyses. After these first steps, we harvested genes from model butyrate producers and candidate strains displaying all genes of the individual pathway in close synteny (not considering terminal genes) and used the obtained sequences to construct HMM models to screen genomes again. After applying certain cutoffs based on HMM scores (for details, see Fig. S1 and Text S1), candidates were filtered for exhibiting entire pathways (allowing one missing gene), and terminal genes were treated in separate analyses (for details, see Fig. S1 and Text S1). Finally, candidates from both EC number and HMM searches were combined and subjected to additional filtering based on detailed gene analysis considering synteny and phylogenetic trees (for details, see Fig. S1 and Text S1). Protein sequences were aligned in the software program Clustal Omega (http://www.ebi.ac.uk/Tools/msa/clustalo), and neighbor-joining trees were constructed using the program MEGA (http://www.megasoftware.net). Taxonomy is displayed as provided by IMG with some modifications for the phylum Firmicutes based on RDP’s classifications.