Quantifying microbial memory and its influence on ecosystem time lags using Bayesian modeling
Ecosystems do not respond instantaneously to environmental change, but rather possess a "memory" of past conditions. This memory alters the magnitude and timescale of an ecosystem's response to climate stress. Therefore, a lack of understanding of what controls ecosystem memory could contribute to inaccurate predictions of climate patterns and biogeochemical dynamics. Microbial communities control carbon and nitrogen (biogeochemical) cycling in the soil. Contrary to previous assumptions, recent evidence shows that microbial communities themselves possess a memory that could contribute to ecosystem responses. However, microbial memory has not been explicitly quantified, or assessed as a possible mechanism contributing to ecosystem memory. The aim of this project is: to develop a model framework to examine the magnitude and timescale of microbial memory; to determine the importance of this microbial mechanism in predicting lags in soil gas flux; and to identify the microbial traits (physiology or community composition) that are best linked to the larger scale functions they mediate.
This research couples innovative statistical tools (hierarchical Bayesian modeling) and new biological concepts (microbial memory), bridging conceptual and collaborative gaps between ecology and mathematics. The project will improve the postdoctoral fellow?s statistical skills, broaden the scope of future work, and give the fellow novel tools to analyze large ecological datasets. Results from this project will be widely disseminated through peer-reviewed publications and university-mediated public seminars, and contribute to ecological workshops that introduce life scientists to Bayesian methods.
Microorganisms are the most abundant life forms on earth, and are responsible for a myriad of essential functions, from food digestion and immunity to decomposition of dead plant material. These organisms live in immensely diverse communities: over 50,000 species can fit in the cap of a pen. Molecular tools like high-throughput sequencing have helped to characterize this diversity. However, our quantitative methods and theory have lagged behind this deluge of genomic descriptions. Although often the species in a given microbial community relates to environmental conditions (e.g. pH in soil, or diet in the gut), in most studies, a substantial portion of the variation in species composition remains unexplained. One reason it is difficult to predict microbial communities using only their habitat is that other "stochastic" processes also determine the species present in a community. These random processes include changes in populations that naturally occur from random birth and death events, and random dispersal (e.g. by wind) of a subset of microbes into a community. However, if we are going to better predict how microbial communities change, and manipulate communities to perform certain functions, it is essential we understand how these processes influence their composition and function. Microbial "traits" - such as their pH tolerance or how fast they degrade substrates - determines how each species survives, which in turn, determines how the community functions. Only about 1% of bacterial species can be cultivated, so it is difficult to study these traits in whole communities. To better track these traits, and fully control all processes that influence microbial community composition, we conducted "experiments" using a theoretical model. The model creates hypothetical microbial communities, each with a different set of traits. After placing microbes on a simulated "dead leaf" to degrade, they compete with one another to survive. Taxa grow and die over time, depending on the adaptiveness of their traits. Throughout this process, certain taxa are selected based on environmental conditions in the model, like the chemical properties of the leaf. In addition to selection, stochastic or random processes occur: microbes randomly are born and die, and new taxa are disperse into the community from the outside each year. Using this model as our experiment, we tested the effect of dispersal rate on the predictability of microbial communities subject to different leaf chemistries. In an experiment in nature, we would expect different types of leaves to host communities that adapted to degrade those leaves. However, since stochastic effects are at play, we might also find that even identical leaves have quite distinct communities. This difference – the variation among communities under identical environmental conditions – can be used as an indicator of predictability. When communities under similar conditions are more similar, predictability is high, and vice versa. We expected that it would be harder to predict community composition at high dispersal because a large proportion of the community would consist of randomly dispersed, immigrant microbes. We also expected that when there was no dispersal, communities that started as distinct would not have many opportunities to become more similar; highly-adapted taxa destined to be abundant on all leaves would not get a chance to dominate if they are not present. However, our results did not support this hypothesis. We found that at high dispersal, selection of those microbes that specialized on certain types of leaves was extremely high, making it easy to distinguish groups that were characteristic for each leaf type, even when 60% of the community was exchanged each year. When dispersal was zero, selection still occurred, but the signal was dampened. At intermediate dispersal rates (<25%), a combination of dispersal limitation and random immigrants made differences in microbial communities on different leaf types undetectable. We also analyzed the same simulations using the traits of the microbes instead of their identity; here, two species with similar traits were treated as the same group. Using these groups, it was much easier to predict the communities that would emerge under different conditions (although dispersal still played a role). The greater predictability is most likely because several species that filled the same function were equally as likely to be abundant, but the success of one or the other resulted in differences in species composition, when the communities’ basic functions were similar. In this work, we have demonstrated how hypothetical experiments model can be used to study changes in microbial communities. Managing microbes to optimize health and ecosystem services requires that we understand what factors determine the species that are present, including selection, dispersal, and random drift. We identified relationships that describe how these factors are likely to influence microbial diversity in nature. These patterns can be tested in nature to develop much-needed theory for enhancing predictability in microbial community ecology.
