Bacterial functional prediction tools detect but underestimate metabolic diversity compared to shotgun metagenomics in southwest Florida soils

Abstract

Profiling the 16S-rRNA gene is a common method in microbial ecology to provide estimates of microbial diversity and abundance, but it cannot directly produce information about the functional potential of these taxa. While shotgun sequencing can provide functional information, the cost can be prohibitive for large scale comparative studies. Functional prediction tools have been developed to generate expected gene counts from 16S-rRNA amplicon data and allow for initial exploration of hypotheses prior to shotgun sequencing. PICRUSt2 and Tax4Fun2 are two such prediction tools that are becoming increasingly popular in microbial ecology studies. However, there are only a few benchmarking studies that have tested their performance, and none that focused on soils. The goal of the present study was to compare the performance of both prediction tools with each other and shotgun sequencing. We assessed prediction output from both tools used on the same samples from two sites in Southwest Florida and compared the predicted functional profiles with MG-RAST functional profiles on a shotgun-sequenced subset of samples. Shotgun and amplicon sequencing datasets showed comparable 16S rRNA gene profiles with observed differences likely stemming from PCR bias and taxa-dependent variability of 16S rRNA gene density in shotgun sequences. PICRUSt2 functional profiles represented greater relative abundances of genes in the pathway for oxidative phosphorylation, while Tax4Fun2 detected more genes from specialized metabolic pathways, such as for methane metabolism. PICRUSt2 functional profiles showed somewhat stronger correlations with shotgun-sequenced functional profiles than Tax4Fun2. However, both prediction tools significantly underestimated gene frequencies in many KEGG categories, including genes with biogeochemical significance for soil carbon- and nitrogen cycling.