Do we need to understand microbial community structure to predict function?

Background/Question/Methods  

Ecological modeling has been useful in predictions of shifts in major biogeochemical cycles. Yet, despite the central role of microbes in ecosystem processes, such models rarely account for variation in microbial community composition. Terrestrial models typically assume that microbial function is wholly determined by climatic and edaphic factors. However, several studies show that microbial community structure can be decoupled from environmental parameters. Here, we use statistical models to address a fundamental, but unanswered question in ecology: do we need to understand microbial community structure to accurately predict function?

We compared the power of models constructed with and without community data for predicting soil N cycling processes. Nitrification, denitrification, and DNRA rates were quantified in a tropical rainforest soil using ¹⁵NO₃ stable isotope tracers, and gene abundances of napA, narG, nirK, nirS, nosZ, nifH, and amoAwere determined using qPCR. We then statistically compared three types of predictors using stepwise linear regressions: edaphic factors, functional gene abundances and both edaphic factors and functional gene abundances.

Results/Conclusions

We found that no gene abundance models explained more variation in process rates than edaphic factor models. As well, models that contained both edaphic factors and community structure data did not explain more variation in process than models containing data on edaphic factors alone. Functional gene abundances did not explain any variation in the residuals from edaphic factors, suggesting that they explained the same fraction of variance as edaphic factors. Our results support that edaphic factors structure microbial communities in this tropical rainforest soil, and that edaphic factor models may implicitly consider community structure. How our results apply to other ecosystems remains unknown, but the simple statistical approaches presented here can accommodate a variety of spatial and temporal scales, as well as multiple types of microbial community composition data.  If future cross-system comparisons reveal greater explanatory power of community data in other ecosystems or for other types of communities, the approach we present should help prioritize what forms of data may be most useful in ecosystem model development.