An overview of my doctoral dissertation.

There is an estimated 5 nonillion prokaryotic cells on Earth. That is a 5 followed by 30 zeroes, a number so large that “astronomic” falls short. For example, there are about 7 billion prokaryotic cells on Earth for every single planet in the observable universe (estimated 700 quintillion, a 7 followed by 20 zeroes). However, we know remarkably little about the distribution of prokaryotic species around the globe, or the ecologic principles that dictate how these species get to live together in the same environment: how are communities assembled.

In his seminal 1934 work ‘Geobiology or Introduction to the Science of the Environment’, professor L.G.M. Baas Becking wrote about microorganisms:

Everything is everywhere, but the environment selects¹

In this elegantly concise dictum (and throughout his work), Baas Becking provided an early version of the ecologic niche concept as the driver of community assemblies, while predicting that all microorganisms are cosmopolitan. Importantly, he understood from the onset of microbial ecology, that the limit of detection and the selectiveness of techniques in microbiology hide deeper truths about the real distribution of microorganisms. Just because we don’t see a microbe, it doesn’t mean it’s not there. Since then, there have been enormous advancements in both microbial ecology theory and microbiology techniques, but the limitations recognized by Baas Becking continue to hunt the analysis of the growing observation data. During the past decade, a novel set of techniques collectively known as metagenomics has allowed the study of genetic material directly extracted from environmental samples. By allowing the direct observation of genomic fragments from microorganisms, bypassing cultivation and targeted amplification, metagenomics offers unique opportunities to close the gaps that exist between theory and empirical testing in microbial ecology. However, realizing the full potential of metagenomics requires the advancement of computational and statistical techniques for data analyses accounting for the detection limits inherent to the sampling methods.

So how much can we see in a metagenomic sample? Not a trivial question that could be tackled in different ways. For example, one could simply estimate the fraction of successfully sequenced fragments of DNA from all the prokaryotic cells in an ecosystem. One of the microbial communities we have studied is that of Lake Lanier, located in the Southeast of the USA. In Lake Lanier, we estimate the sampling fraction in each one of our metagenomes at about 2 in 1,019. This is akin to attempting to describe all insects on Earth from a random sample of two, likely a couple of ants. This exercise helps visualizing the magnitude of the problem, but could be misleading because not all cells in a community are uniformly different. Rather, they reflect their evolutionary history, and tend to cluster in tractable and genetically distinct groups (species) that can be detected and characterized using metagenomics². Therefore, the fraction of species in the community normalized by abundance is much more informative, as it reflects the descriptive power of a sample as well as the accuracy in sample comparisons³. This problem, measuring metagenomic coverage, has been previously explored from a theoretical standpoint, typically by modeling or assuming parameters describing the community and the sample taken (reviewed by Luo et al⁴). However, no scalable implementations that could be applied to the large numbers of sequences in metagenomes were available. We proposed an alternative approach that didn’t attempt to directly model the community or the sample, but instead leveraged a more general statistical result spotted by Allan Turing and formalized by his assistant Irving J. Good: a random sample from large categories tends to be more redundant when it’s more comprehensive. I.e., coverage can be estimated from redundancy⁵. This principle has been previously applied to other related problems, such as the estimation of richness or other metrics of diversity^6,7, but not for the estimation of metagenomic coverage. Our approach, Nonpareil, does exactly that^8,9. And the results are much more encouraging than the bleak image of two ants representing all insects: our metagenomes from Lake Lanier have over 60% abundance-weighted average coverage.

Being able to directly address the coverage problem, we next embarked on the use of metagenomics to describe the influence of different factors on community assembly. First, we expected that the most conspicuous signs could be detected after a large disturbance (cf. Shade et al¹⁰). For this, we used the unintended experiment provided by the 2010 BP oil spill in the Gulf of Mexico, to describe the succession dynamics in the microbial community of a Florida beach. We were able to obtain samples before the oil reached the shoreline, as well as a time series from the oiling extending for a year. We demonstrated that this secondary succession only had minor lasting effects in community composition after recovery, and was mainly driven by degradation of hydrocarbons and nutrient scavenging in concert with the degree of specialization of community members¹¹. We saw a strong environmental selection, but we also saw that many of the organisms responding to the disturbance were either present in the undisturbed community or typically found in the waters of the Gulf. Perhaps some microorganisms are local, and not everywhere. With this lingering question in mind we returned to Lake Lanier, a relatively stable environment offering the perfect natural laboratory: seasonal cycles and well defined connections to four downstream lakes through the Chattahoochee River until the estuaries in the Gulf of Mexico. After monitoring this system for a span of six years, we found strong effects of both season and geographic distance on the communities, only minor landscape effects, and documented cases of local and regional endemism, indicating that both historic and contemporary factors affect community assembly in this system¹². In other words: the environment does select, but there are such things as local microbes.

References