Parsing ecological signal from noise in next generation amplicon sequencing

Abstract

It is clear that the use of next generation sequencing (NGS) applied to environmental DNA is changing the way researchers conduct experiments and significantly deepening our understanding of microbial communities around the globe (Amend et al., 2010; Caporaso et al., 2011; Bik et al., 2012; Bates et al., 2013). The lower per unit cost and sheer number of sequences relative to traditional methods provide tremendous advantages in characterizing the richness and composition of highly diverse microbial systems (Bokulich et al., 2013). In a recent volume of New Phytologist, Lindahl et al. (2013) presented an excellent introduction into high-throughput sequencing of amplified gene markers for fungi, and broadly discussed field sampling and handling, DNA extraction, markers, primers, amplicon library construction, sequencing platform, bioinformatic analyses and data interpretation. We applaud their overview as an important general guide, but we have found that there are significant additional issues regarding NGS that have not been well articulated in the literature, especially when applied to fungi. Below we highlight a series of platform-independent recommendations based on our recent experiences with NGS, which we think are critical for maximizing the signal:noise ratio in molecular ecological analyses. The inclusion of both negative and positive controls is indispensable in NGS-based studies due to the greater detection level than traditional sequencing (i.e. sequences can be readily detected in controls with NGS methods even in the absence of positive PCR bands). These controls are essential at multiple steps during the experimental process, for example, in the field, in laboratory settings where samples are processed, during DNA extraction, as well as before and after PCR. To be useful, the controls must be treated identically to other samples from initial processing through library preparation. As an example, we recently conducted a field-based study using the Illumina MiSeq platform to amplify the ITS1 region of soil fungi, which included a series of negative controls. Of the total sequence pool generated, we detected 0.01% from soil sieve controls (3.17% total OTUs (operational taxonomic unit(s))), 0.0001% (0.2% total OTUs) from DNA extraction controls, and 0.001% sequences (0.67% total OTUs) from PCR controls. Together, these controls accounted for 0.01% of total sequences (3.8% of total OTUs). While detection of fungal taxa in negative controls is key to determining which fungal taxa should be included in subsequent ecological analyses, there is currently no consensus on how to handle these sequences. One approach would be to simply delete any OTUs that appeared in negative controls across all samples (e.g. Vik et al., 2013). However, in our study, this would have deleted many of the most abundant OTUs in the experimental samples. It seems highly likely that those abundant OTUs were in fact present in the field because (1) many had been previously encountered in soil and (2) their abundance in the controls was multiple orders of magnitude lower. To avoid eliminating OTUs that appeared to be ecologically valid, we addressed this issue by subtracting the number of sequences of each OTU present in the negative controls from the sequence abundance of that OTU in the experimental samples (essentially, after subtraction, the negative control samples will contain zero sequences, and other samples will have reduced abundances). In our dataset, this approach eliminated only two low abundance OTUs (each had < 40 total sequences) instead of 56 OTUs had we used the deletion approach. While we recognize that the inclusion of two low abundance OTUs would have negligible effects on the large conclusions drawn, the alternative deletion approach would have most likely created ecologically spurious results. Although this subtraction approach may not be best for all studies, we strongly recommend that researchers sequence all negative controls and explicitly report those results as well as how sequences in controls are processed before ecological analyses. Further, if this subtraction method is applied, we advocate adding as equivalent volumes of the negative control and experimental samples as possible to avoid sequence inflation biases. We also suggest that researchers consider additional types of negative controls in experimental designs. In the aforementioned study, we were particularly interested in characterizing the richness of active ectomycorrhizal (ECM) fungal communities, so we sampled soils from monodominant plots containing either ECM or arbuscular mycorrhizal (AM) hosts. Because ECM hosts are absent in the AM host plots, we considered those samples to be a negative field control for ECM fungal taxa that were present as spores but not mycelium. We found that while ECM fungal taxa represented only 1.6% of the sequences from the AM host plots, they included nearly 60% of all the ECM fungal OTUs present in the E