Broadscale Ecological Patterns Are Robust to Use of Exact Sequence Variants versus Operational Taxonomic Units

1Characterization of microbial communities by amplicon sequencing introduces biases and errors at every step. Hence, choices concerning all aspects of molecular processing from DNA extraction method to sequencing platform are debated. Further downstream, the choices for computational processing of amplicon sequences are similarly deliberated (e.g., see references to ). Yet despite these ongoing debates, microbial ecology has made great strides toward characterizing and testing hypotheses in environmental and host-associated microbiomes (e.g., see references and ).

2Within microbiome studies, operational taxonomic units (OTUs) have been used to delineate microbial taxa, as the majority of microbial diversity remains unrepresented in global databases. While any degree of sequence similarity could be used to denote individual taxa, a 97% sequence similarity cutoff became standard within microbial community analyses. This cutoff attempted to balance previous standards for defining microbial species and recognition of spurious diversity accumulated through PCR and sequencing errors.

3Recently, it has been suggested that taxa should be defined based on exact nucleotide sequences of marker genes. Delineation of taxa by exact sequence variants (ESVs), also termed amplicon sequence variants (ASVs) or zero-radius OTUs (zOTUs), is not only expected to increase taxonomic resolution, but could also simplify comparisons across studies by eliminating the need for rebinning taxa when data sets are merged. Due to these advantages, there has been a surge in bioinformatic pipelines that seek to utilize ESVs and minimize specious sequence diversity. Moreover, some proponents have stated that ESVs should replace OTUs altogether. However, as with the adoption of any new approach, there remains a need to quantify how this new method compares to a large body of previous research. Furthermore, OTU classifications remain biologically useful for comparing diversity across large data sets or identifying clades that share traits.

4Here, we tested if use of ESVs versus 97% OTUs affected the ecological conclusions, including treatment effects and α and β diversity patterns, from a large field study of leaf litter communities. This study included a “site” and “inoculum” treatment, in which all microbial communities were reciprocally transplanted into all five sites (see Text S1 in the supplemental material) along an elevation gradient. We sequenced both bacteria (16S rRNA) and fungi (internal transcribed spacer 2 [ITS2]) from litterbags collected at three time points (6, 12, and 18 months after deployment) in separate sequencing runs. While we expected that the binning approach would alter observed richness, we hypothesized that it might not alter trends in α and β diversity, but that these results might differ based on the amplicon sequenced.

5In total, we analyzed >15 million bacterial and >20 million fungal sequences using UPARSE v10 (see Table S1 in the supplemental material), which allowed for a direct comparison of ESV versus 97% OTU approaches by keeping all other aspects of quality filtering and merging consistent. We selected a direct comparison with 97% OTUs as it is the most standard threshold and the clustering algorithms appear to be most effective at this level (R. Edgar, personal communication). A recent study also found that clustering thresholds from 87% to 99% yield highly stable results.

6ESV and OTU α diversity was strongly correlated across samples using four metrics for both bacteria and fungi (mean Pearson’s r = 0.95 ± 0.02; all P values are <0.001). For three metrics (Berger-Parker, Shannon, and Simpson), the ESV and OTU approaches were not only highly correlated (mean Pearson’s r = 0.95 ± 0.02), but nearly equivalent in their values (mean slope = 0.97) (see Table S2 in the supplemental material). For observed richness, ESV versus OTU was also highly correlated across all time points/sequencing runs (Pearson’s r > 0.92) (Fig. 1A and B). However, bacterial OTU richness was approximately half of ESV richness for the same sample (mean slope = 0.46), and fungal OTU richness was approximately three-quarters of ESV richness (mean slope = 0.79). We speculate that this difference between bacteria and fungi is due to the coarser phylogenetic breadth of the 16S versus ITS genetic regions.

7β diversity metrics were also strongly correlated across samples for ESVs and OTUs (Bray-Curtis average Mantel’s r = 0.96 for bacteria and 0.98 for fungi; all P values are <0.01 [Fig. 1C and D]), whether assessed by abundance-based (Bray-Curtis) or presence-absence (Jaccard) metrics (Table S2). Moreover, the values of the β diversity metrics were nearly identical regardless of binning approach (slopes of ~1).

8The highly correlated α and β diversity metrics indicated that results based on these metrics should yield similar ecological conclusions. Indeed, the patterns of bacterial and fungal richness and community composition across the elevation gradient were nearly indistinguishable (Fig. 2; see Fig. S1 in the supplemental material), as were the statistical tests for both richness (see Table S3 in the supplemental material) and community composition (see Tables S4 and S5 in the supplemental material). Moreover, family- and genus-level compositions at each site along the gradient were virtually identical for bacteria (see Fig. S2 in the supplemental material) and highly similar for fungi (see Fig. S3 in the supplemental material), with no taxa being over- or underrepresented in the ESV versus OTU approaches for bacteria (Fig. S2C) and only one for fungi (Fig. S3C). We also included a mock community of eight distinct bacterial species in our PCR and sequencing runs. Both approaches resulted in highly similar mock community composition (see Fig. S4 in the supplemental material). Thus, we found no evidence that ESVs yield better taxonomic resolution or are more sensitive to detecting treatment effects. If anything, the ESV method appeared to be slightly less sensitive to detecting treatment effects on richness than the OTU method, especially for fungi in which fewer significant treatment effects were detected using ESVs (Table S3).

9Despite quantitative differences in microbial richness, ecological interpretation of our large bacterial and fungal community data set was robust to the use of ESVs versus 97% OTUs. Thus, even though there are good reasons to take an ESV approach, we need not question the validity of ecological results based on OTUs. Indeed, while previous studies have found that ESVs can help explain additional variation among samples, the α and β diversity patterns of ESVs and OTUs in these studies were also quite similar. In general, we suspect that the robustness of such comparisons will vary depending on the breadth of the microbial community targeted. For instance, here we characterized all bacteria and fungi in a diverse environmental community, as opposed to a narrower subset of taxa or a less diverse, host-associated community.

10Finally, both 97% OTUs and ESVs mask ecologically important trait variation of individual taxa. In our study, ESVs only slightly increased the detection of fungal and bacterial diversity (1.3 and 2.1 times OTU richness, respectively), highlighting that ribosomal marker genes at any resolution are generally poor targets for improving genetic resolution within a microbial community. For example, it is widely known that many taxa can share the same 16S rRNA or ITS. Thus, if strain identification is critical, then a full genome or amplicon of a less conservative marker gene is required. However, for broadscale community α and β diversity patterns, although the vagaries of molecular and bioinformatics processing inevitably add noise to microbial sequencing data, strong community-level signals will likely emerge with suitable study designs and statistics regardless of binning approach.

11Sequences were submitted to the National Center for Biotechnology Information Sequence Read Archive under accession no. SRP150375 and BioProject no. PRJNA474008. All data and scripts to recreate all figures and statistics from this article can be found on github at https://github.com/sydneyg/OTUvESV.