Re-examination of reported biological patterns through a shallow shotgun metagenomic lens
In the second half of our analyses, we sought to determine whether shallow shotgun sequencing data (rarefied to 0.38 million bacterial read pairs) could replicate the biological patterns reported in a recently published 16S amplicon dataset We used mantel tests to test for correlations between spatial and Bray-Curtis dissimilarity matrices, PERMANOVAs to test for the effect of environmental variables on microbiome beta-diversity, and ANCOM-BC tests to identify taxa which differ in abundance between horses with access to sandwort versus horses without access to sandwort (with false discovery rate adjustments and a conservative variance estimate; To allow for comparisons between 16S amplicon and shotgun metagenomic results, ANCOM-BC tests for taxon differential abundance were performed on datasets binned to bacterial family. However, to prevent large imbalances in false discovery rate penalties between datasets, we further filtered the shallow shotgun dataset to only families which were present at a percent abundance of 0.1% in at least 1 sample. This resulted in the retention of 144 families, which was comparable to the 147 families observed in the 16S amplicon dataset. Notably, although Kaiju accurately estimates abundance weighted patterns in the microbiome , it overestimates richness and can result in a large number of spurious, low abundance taxa. Abundance filtering of count tables is therefore prudent. Similar analyses as described above were used to analyze differential abundances in MetaCyc reactions and metabolic pathways. For all functional analyses, reads which could not be classified or those which could not be integrated into pathways (in the case of pathway abundance estimates) were removed. Datasets were normalized via rarefaction for all analyses other than ANCOM-BC tests, which possess a built-in normalization process . Unless otherwise stated, all analyses were completed in R (v. 4.1.2) using the package phyloseq v. 1.38.0 and vegan v. 2.5–7