The compositional nature of microbial sequencing datasets hides significant community changes in wetlands

Wetland research today is concerned about the biochemical responses to climate and land-use change. These changes are substantially mediated by microbial activity. But often enough, these changes are not reflected in microbial community compositions.
In the SMARTX experiment in a coastal wetland on the Chesapeake Bay (USA) biogeochemical responses to several years of simulated elevated temperatures and atmospheric CO₂ were observed. However, similar to comparable studies on microbial compositional changes, no significant change in microbial community composition was observed. We propose that the discrepancy between measurable biogeochemical changes and the lack of an apparent microbial response stems to a large part from our data analysis approach, which does not account for the compositional nature of microbial sequencing datasets.

There are two reasons why sequencing datasets should be analysed as compositional. Firstly, to determine mean microbial community compositions, the overall count quantity, so the sequencing depth, is not relevant. However, it is relevant when comparing compositions. Additionally, the statistical error strongly depends on the total number of counts. This also matters when investigating the variability of community compositions. Secondly, the bias induced through nucleic acid extraction, amplification and sequencing cannot be expected to cancel out when comparing community compositions, even if the bias is constant for each amplicon sequence variant (ASV, a proxy for species).Approaching our data as compositional solves both of the above described problems. Firstly, the total sum of counts becomes irrelevant for our statistical analysis (Subcompositional Coherence). Secondly, the analysis is invariant to the constant extraction bias (Perturbation Invariance). By adapting our data analysis approach, we expect to see changes of other measurable parameters reflected in microbial community compositions. In a first attempt to reanalyse ASV reads from the SMARTX experiment, a temperature effect on community composition, which was not apparent when using rarefaction and Bray-Curtis distance measures, could be identified.