The figure above summarizes the experimental design and analysis. The experiment was designed to address the question of whether the microbial community contains sufficient information to predict a biogeochemical state in a dynamic system. The structure of a microbial community is highly sensitive to environmental change. Small changes in the chemical or physical environment will result in a shift in abundance of one or more taxa as mortality and growth rates respond. These shifts in structure are easily observed by amplicon sequencing of taxonomic marker genes. These relative abundance data can be combined with flow cytometry analysis of microbial abundance to yield absolute abundance data.

The trick of course is relating an observed shift in community structure to a specific biogeochemical state. Machine learning provides a number of ways to do this, but all require large training datasets. Fortunately gene sequencing is pretty cheap these days and DNA extractions are much more high-throughput than they were just a few years ago. Because of this it’s possibly to generate community structure data for hundreds of samples in relatively short order. In this study Avishek used over 700 samples from sediment bioreactors and the random forest algorithm to predict the concentration of hydrogen sulfide with a reasonably high degree of accuracy.

Like any statistical model, developing machine learning models takes careful attention to detail. Careful segregation of the data into training and validation sets and engineering of the features used for prediction yield the most honest models that can be best applied for future predictions. Avishek’s paper is an excellent template for developing a predictive machine learning model from microbial community structure data.

Ecuador is ground zero for mangrove deforestation for shrimp aquaculture. Most of Ecuador’s coastline is in fact completely stripped of mangroves. The biogeochemical consequences of this aren’t hard to imagine. Mangrove forests contain a significant amount of carbon in living biomass and in the sediment. Aquaculture ponds, by contrast, contain a large amount of nitrogen as a result of copious additions of nitrogen-rich shrimp feed. The balance of C to N is one of the fundamental stoichiometric relationships in aquatic chemistry. When it shifts all kinds of interesting things start to happen.

The one place in Ecuador where you can find large areas of mangroves is the Cayapas-Mataje Ecological Reserve. CMER is in fact the largest contiguous mangrove forest on the Pacific coast of Latin America. Its status comes from an interesting combination of social and economic factors that left this part of Ecuador relatively undeveloped until recently. There is shrimp aquaculture in the reserve, but it’s nowhere near as expansive as in Muisne and other ex-mangrove sites in Ecuador.

Natalia leveraged the different levels of disturbance present in Cayapas-Mataje, and between Cayapas-Mataje and Muisne, to explore what the impact of all this aquaculture activity is on microbial community structure. After all it’s really the microbial community that responds to and drives the biogeochemistry, so understanding the sensitivity of these communities to the changing conditions gives us insight into how the system is changing as a whole.

By using our paprica pipeline Natalia was able to evaluate changes in microbial community structure, predicted genomic content, and key genome features across the disturbance gradient. A nitrogen excess (relative to phosphorous) was associated with bacteria with larger genomes and more 16S rRNA gene copies, indicative of a more copiotrophic or fast-growing population. This has implications for how carbon is turned over or retained at the higher levels of disturbance.

Different microbial metabolisms are also associated with the level of disturbance. The figure above shows the distribution of predicted metabolic pathways associated with nitrogen metabolism. Nitrogen fixation, a feature of microbial symbionts of many plants, is less abundant at high levels of disturbance, while pathways associated with denitrification are more abundant. The interesting thing about this is that these samples are restricted to the mangroves themselves – the high disturbance samples don’t reflect the actual aquaculture ponds – so these changes reflect altered processes in the remaining stands of mangroves. The loss of beneficial, symbiotic bacteria and elevated abundance of putative shellfish pathogens suggests the impacts of aquaculture are not limited to the physical removal of mangrove trees and associated release of carbon.

Tia filters water just outside the mouth of San Diego Bay.  Coronado Island is in the background.

Why the interest in seagrass?  Unlike kelp, seagrasses are true flowering plants.  They’re found around the world from the tropics to the high latitudes and perform a number of important ecosystem functions.  Considerable attention has been given to their importance as nursery habitat for a number of marine organisms.  More recently we’ve come to appreciate the role they play in mediating sediment transport and pollution.  Recent work in Indonesia (which inspired Tia to carry out this study) even showed that the presence of seagrass meadows between inhabited beaches and coral reefs reduced the load of human and coral pathogens within the reefs.

Seagrass, barely visible on a murky collection day.  Confirming seagrass presence/absence was a considerable challenge during the field effort, and one we hadn’t anticipated.  There’s always something…

There are a number of good papers out on the seagrass microbiome – epibionts and other bacteria that are physically associated with the seagrass (see here and here) – but not so many on water column microbes in the vicinity of seagrass meadows.  In this study we took paired samples inside and outside of seagrass beds within and just outside of San Diego Bay.  I’ll be the first to admit that our experimental design was simple, with a limited sample set, and we look forward to a more comprehensive analysis at some point in the future.  Regardless, it worked well for a factor-type analysis using DESeq2; testing for differentially present microbial taxa while controlling for the different locations.

What we found was that (not surprisingly) the influence of seagrass is pretty minor compared to the influence of sample location (inside vs. outside of the bay).  There were, however, some taxa that were more abundant near seagrass even when we controlled for sample location.  These included some expected copiotrophs including members of the Rhodobacteraceae, Puniceispirillum, and Colwellia, as well as some unexpected genera including Synechococcus and Thioglobus (a sulfur oxidizing gammaproteobacteria).  We spent the requisite amount of time puzzling over some abundant Rickettsiales within San Diego Bay.  We usually take these to mean SAR11 (though our analysis used paprica, which usually picks up Pelagibacter just fine), but didn’t look like SAR11 in this case.  An unusual coastal SAR11 clade?  A parasite or endosymbiont with a whonky GC ratio?  TBD…