As part of the Scripps Ecological Observatory effort we team up with the Southern California Coastal Ocean Observing System (SCCOOS) team for twice-weekly sampling of surface water for microbial community structure via 16S and 18S rRNA gene sequencing and microbial abundance via flow cytometry. As you can see from the SCCOOS and flow cytometry data below it’s a pretty dynamic system! This is why the site is so advantageous for ecological studies; more dynamic means more opportunities to identify co-variants in the environment that signal possible interactions.

At the core of Jesse’s paper is the 16S rRNA gene sequence dataset. What these data provide is a high resolution view of the taxonomy of the bacterial and archaeal community at each sample point. These data are so high resolution – after proper denoising and quality control they represent hundreds to thousands of unique taxa – that it’s often difficult to make inferences from them. Techniques are applied to reduce the complexity of the data and make it easier to see patterns.

Jesse approached the problem from perspectives of both the observations (sampling days) and variables (microbial taxa). For microbial time-series data it is much more common to aggregate variables. A widely used approach involves a technique known as weighted gene correlation network analysis (WGCNA), originally developed for gene expression studies. WGCNA uses network analysis to combine taxa into subnetworks or modules that have like co-occurrence patterns. One advantage to this approach is that the subnetworks are easily correlated to external variables that either drive the pattern (e.g., physical processes) or are influenced by it (e.g., ecophysiology). A disadvantage is that these correlations aren’t predictive. You can’t readily classify new data into the existing subnetworks, and the co-occurrence patterns of the subnetworks themselves contain additional information that isn’t readily captured by this approach.

In a 2017 paper we demonstrated how self-organizing maps (SOMs) can be used to more explicitly link environmental parameters with microbial community structure. SOMs are a form of neural network and collapse complex, multi-dimensional data into a 2D representation that retains the major relationships present in the original data. The end result of the SOM training process is a 2D model of the data that can be further subdivided into distinct classes. Applied to community structure data (i.e. in microbial community segmentation) the SOM flips the aggregation problem, aggregating samples instead of taxa. That means that each unique sample point can be described by the model as a single discrete variable that nonetheless captures much of the key information present. A major advantage to this approach is that the model is reusable: new data can be very efficiently assigned to existing classes, which is a key advantage for an ongoing ecological monitoring effort.

This paper is an exciting but very early effort to track microbial processes at the Scripps Ecological Observatory. The time-series presented here ends in June of 2019 – the date our original (and terrible) flow cytometer terminally failed – but twice-weekly data collection have continued. We now have three years of 16S and 18S rRNA gene sequence and flow cytometry data and this collection will continue as long as we’re able to support it! Students and potential postdocs interested in microbial time-series analysis should take note…

Many thanks to the Simons Foundation Early Career Investigator in Marine Microbial Ecology and Evolution program for supporting this work, and to all the SCCOOS technicians and Bowman Lab personnel for bringing us water and processing samples!