Accounting for Variability Across Years, Depths, and Non-Target Reads Improves Quantification of Root Community Composition in Plant Mixtures.

One key challenge in characterizing plant communities is determining the relative abundance of distinct species in root biomass. Metabarcoding is increasingly used to estimate a species' proportion in root mass based on its proportion in sample reads, but results have been mixed in accurately converting sequencing output to species mass in mixture. We created calibration mixtures from four focal plant species to determine how the association between a species' proportion in mass and its proportion in sequencing reads of the rbcL chloroplast gene varied across sample years, soil depth, and non-target proportion of reads. We also determined whether including these explanatory variables in prediction models could improve accuracy. We found the association between a species' proportion in reads and its proportion in mass varied across years and non-target proportion of reads, and to a lesser extent, rooting depth, and incorporating these explanatory factors into prediction models improved prediction accuracy. A secondary objective was to determine the number of calibration levels and sequencing depth required to accurately predict a species' proportion in mass. Through simulations, we estimated that 42 calibration samples (seven proportional levels * three years * two depths) and 700 reads per sample consistently obtained accurate predictions for simple mixtures. The association between a species' proportion in mass and proportion in reads can vary considerably, and our study suggests that measuring and explaining this variation can lead to more accurate predictions of the relative abundance of plants within root biomass using metabarcoding.