Russell's Blog

When the first environmental sequencing projects were conducted, the genetic bredth present within an environmental sample so far outstripped the available sequencing capacity at the time that it was only possible to obtain a tiny slice of the genetic material present. This gave researchers two choices; either target a particular gene, or go fishing. Both approaches have been extremely fruitful. Targeted studies of ribosomal RNA led to the discovery of the archaea, among other important accomplishments. The "fishing" approach (which has a shorter history) has also led to exciting discoveries. If you do a literature search for your favorite enzyme with the word "novel," it's quite likely that most of the recent publications will involve some kind of metagenomic survey.

As the cost of sequencing continues to plummet, a third approach to environmental sequencing has suddenly become possible: Exhaustive sequencing. It should be possible not only to survey the entire genomes of the organisms present (although assembling them is another story), but also to survey the population-level variability of the organisms present. This is a rather unprecedented development. Microbial communities have suddenly gone from the most challenging ecologies, with only a handful of observable characters, to a spectacularly detailed quantitative picture.

Here is an example from one of my datasets :

This is a small region in the genome of Roseiflexus castenholzii. I have mapped reads from an environmental sample to the reference genome, yielding an average coverage of about 190x. If you look closely at the column in the middle (position 12519 in the genome, in case you care), we see some clear evidence of a single nucleotide polymorphism in this population of this organism.

As it happens, this coordinate falls in what appears to be an intergenic region, between a phospholipid/glycerol acyltransferase gene on the forward strand to the left and a glycosyl transferase gene one the reverse strand to the right. The two versions appear with roughly equal frequency in the data. For this organism, I've found single nucleotide polymorphisms at thousands of sites. There are also insertions and deletions, and probably rearrangements.

In this ecosystem, I'm able to get between 50x and 300x coverage for almost every taxon present. This should make it possible to see variants that make up only a percent or two of their respective taxon's population. With data like this, it should be possible to do some really beautiful ecology!

For example, suppose one wanted to see if a community obeys the island biogeography model. One could measure the theory's three parameters, immigration, emigration and extinction, by comparing the arrivals and disappearances of variants between the "mainland" and the "island" over time. The ability to examine variants within taxa should make these measurements very sensitive. Additionally, because these are genomic characters, it should be possible to control for the effects of selection (to some extent) by leveraging our knowledge of their genomic context. The 12519th nucleotide of the R. castenholzii genome is perhaps a good example of a character that is unlikely to be under selection because it happens to sit downstream from both flanking genes.¹

So, here is my question to you : What ecological model or process would you be most excited to see studied in this way?

¹ Well, actually I haven't looked at this site in detail, so I'm not sure if one would or wouldn't reasonably expect it to be under selection. My hunch is that it is less likely to be under stringent selection than most other sites. I'm basing this hunch on eyeballing the distance of this locus from where I think RNA polymerase would be ejected on either side, and that both transcripts terminate into its neighborhood. My point is that it should be possible to have some idea of how selection might operate on a particular locus based on its genomic context. One should take this with the usual grain of salt that accompanies inferences drawn solely from models. A better example would be a polymorphism among synonymous codons, but I wasn't able to find one in a hurry.