Quantitative evaluation of bias in barcode markers derivedfrom complex samples

Abstract

PCR products have become a major commodity used to identify organisms based on polymorphism at the DNA level. One problem arising is that unbiased identification of organisms takes as working hypothesis that when DNA is extracted from a sample, a positive signal will be obtained if universal primers are used and DNA quality is suitable for PCR. As this assumption is not always correct we used a system where large differences in PCR success have been described to identify where biases appear and maybe identify solutions. Plants can be identified with at least seven independent plastid‐located loci. These differ in their degree of PCR success and how informative they are in terms of taxonomically useful sequence polymorphisms. Here we used six common plastid loci spanning 48 plant species and performed a quantitative analysis of bias at each step of the identification process. As expected we found important differences in PCR efficiency within a single species, depending on the barcoding sequence being amplified. Quantitative PCR revealed that the Ct threshold for various plastid loci, even within a single species, could exhibit greater than 2000‐fold differences in DNA quantity after amplification. We then performed Next Generation Sequencing experiments in nine species using equal quantities of three plastid‐based primers and equally‐mixed quantities of DNA from multiple species. The result was significantly biased towards species and specific loci even when using adaptor‐specific primers. Our results caution that Next‐Generation Sequencing projects may suffer dramatic bias, arising largely during DNA amplification steps. Moreover, that amplification‐based Next Generation Sequencing technologies exhibit additional bias despite using adaptor‐specific primers, indicating that amplification success depends on the DNA fragment. As such, while qualitative analysis of unknown samples are prone to false negative results if a combination of widely‐successful amplicons are not used, quantitative results should be considered highly suspect, even if all species in the starting sample are known.