Computational protocol: Exercise Is More Effective at Altering Gut Microbial Composition and Producing Stable Changes in Lean Mass in Juvenile versus Adult Male F344 Rats

Similar protocols

Protocol publication

[…] Samples were prepared for sequencing using established protocols [,]. After sample preparation, variable region 4 (V4) of 16S rRNA genes present in each sample was PCR-amplified with forward and reverse primers (F515/R806). The reverse primer is barcoded with an error-correcting 12-base Golay code to facilitate demultiplexing of up to 1,500 samples []. Following purification and precipitation to remove PCR artifacts, samples were subjected to multiple sequencing on an Illumina Genome Analyzer IIx. Operational taxonomic units (OTUs) were picked using a ‘closed reference’ approach []. In brief, this approach takes sequenced reads and compares them to a reference database. A sequence is considered a ‘hit’ if it matches something in the reference database at greater than 97% sequence identity. If an experimental sequence failed to match any member of the reference collection, it was discarded. Closed-reference picking is preferable to ‘de-novo’ or ‘open-reference’ picking in well characterized rat gut communities because the curated reference database acts as a filter; low quality or noisy sequences which get past the quality control steps, but do not actually represent novel OTUs, are eliminated. GreenGenes May 2013 version was the reference database used [], and all sequence processing was done with QIIME v 1.8.0 [] using the UCLUST algorithm []. Taxonomy and phylogeny were taken from the GreenGenes reference collection. The current experiment generated 5,787,335 sequences, of which 1,132,569 were discarded because of uncorrectable barcode errors, low quality, or for being too short (using the default parameters in the QIIME script ‘split_libraries_fastq.py’). The remaining 4,654,766 sequences of median length 151 nucleotides were clustered. The resulting OTU table was rarefied at 8468 sequences/sample to correct for uneven sequencing depth due to amplification differences between samples. Rarefaction is a conservative approach that normalizes library size to prevent type I errors in a variety of techniques applied by QIIME. Recent literature has questioned the ‘statistical admissibility’ of rarefaction [] in the context of differential abundance testing (e.g. ANOVA), but provide a superior method for only the basic two-way comparison. PCoA, supervised learning, and other methods perform poorly without rarefaction when sequencing depth differs between samples. To check that our selected rarefaction depth was not responsible for erroneous conclusions, these data were also rarefied at higher levels to check that patterns were not artifacts of low sequence coverage. PCoA visualizations were done using the Emperor software package [].N = 9 samples/group were submitted for sequencing at the first two time points, and n = 6 samples/group were submitted for the final time point. Rats were excluded due insufficient fecal samples. Final group sizes used in all microbial analyses are as follows: Adult sed at three days (n = 9), six weeks (n = 7; two samples excluded due to yield less than 8468 sequences; see rarefaction description above), and 25 days post (n = 5; one sample excluded due to yield less than 8468 sequences); Adult run at three days (n = 9), six weeks (n = 9), and 25 days post (n = 6); Juvenile sed at three days (n = 9), six weeks (n = 9), and 25 days post (n = 6); Juvenile run at three days (n = 9), six weeks (n = 9), and 25 days post (n = 6). [...] Principal coordinates analysis (PCoA) was performed using unweighted UniFrac distances. Briefly, UniFrac is an algorithm that determines differences between microbial communities between samples based upon their shared branch length on a phylogenetic tree []. [...] Statistical analyses were conducted using the SPSS software package V.21 (SPSS, Chicago, IL). Running distance was compared using a 2 (age) by 6 (weeks of exercise) mixed design ANOVA, and body weight was compared using a 2 (age) by 2 (exercise status) by 9 (weeks) mixed design ANOVA. Chemical carcass body composition and body weight were analyzed using a 2 (age) by 2 (exercise status) by time point (6 wks vs. 25d post). Alpha diversity measures (Shannon entropy, species richness, and phylogenic diversity) as well as relative abundance of microbial taxa at the phylum and the genus levels were subjected to normality tests (Shapiro-Wilk), and all non-normal data were subsequently rank transformed. A 2 (age) by 2 (exercise status) by 3 (time point of fecal sample collection; 3d vs. 6 wk vs. 25d post) mixed design ANOVA was then used to investigate measures of alpha diversity and relative abundance at the phylum and genus levels. At the genus level, taxa without an order classifier were excluded from analyses. Correction for multiple comparisons was conducted using the Benjamini-Hochberg step down method [] implemented in the QIIME 1.8.0. Significant interactions were further investigated with Fisher’s PLSD, with alpha set to p<.05. Supervised machine learning using random forests as implemented in QIIME were employed to classify and differentiate sample classes. […]

Pipeline specifications

Software tools QIIME, UCLUST, EMPeror, UniFrac, SPSS
Databases Greengenes
Applications Miscellaneous, Phylogenetics, Metagenomic sequencing analysis, 16S rRNA-seq analysis
Organisms Rattus norvegicus, Homo sapiens