Background:

Copy number variation -- the duplication of sequences in the genome -- is the fastest mechanism of evolution that we currently know of (see Grauer and Li). When a sequence duplicates, there are far less constraints on it. As long as one of the two dupilcates retains the original function, the other is free to accumulate mutations at a much faster rate. This can result in a sharing of responsibilities ("subfunctionalization") or a brand new function ("neofunctionalization") (see Hurles).

With this in mind, comparing copy number variation (CNV) in the genome between closely related species may be a powerful way to find speciation, a key insight for those of us investigating human origins. That was the logic behind a key paper in 2004 from Jim Sikela's lab in which the authors compared humans to othe primates, looking for changes in sequence copy number between primate species' genomes.

The authors found several interesting insights. For example, the aquaporin 7 gene had duplicated specifically in humans, the only species in the group known to sweat. FGFR3, a gene that can lead to short-limb dwarfism in humans when disrupted, had been duplicated specifically in gorillas, a species in which arm length is significantly longer than height, incidentally. The most interesting finding, however, was a small fragment of a protein, called a domain, that had duplicated hundreds of times in humans -- by far the most duplicated sequence anywhere in the genome. These mysterious domains were so interesting that they made national news.

Nothing was known about these domains (now called Olduvai domains), however. Association and exploratory studies found that higher copy number of Olduvai was associated with larger brains, and that they were expressed at the protein level during human brain development at the right time and place to be influencing brain size (e.g. Keeney et al.), and that they showed associations with head circumference in autism and autism severity, as well as certain IQ test scores. These domains were clearly having a huge impact on human biology, and the hunt was on to figure out what they're actually doing.

Researching the function of these domains turned out to be a lot harder than it is for other genes (which are often already difficult). For one thing, there were so many of them that genome sequencing at the time could not accurately place them in the human genome. The sequences were so similar or even identical, that they could map to multiple locations. They also did not "crystalize," a step used in the process of resolving structure of proteins. Apparently, they seem to have large stretches of highly disorded sequence.

One useful method, however, is RNAseq: an analysis that determines which genes are actually being expressed at a moment in time. For this, a human cell line (DLD1/Tr) was cultured along with an additional cassette that we introduced. The cassette either contained NBPF1 (a gene with 6 of these domains) or an empty cassette ("mock"). The "Tr" promoter meant that we could turn on expression of whatever was inside the cassette at will by treating the cells with doxycycline. With this system, we could compare mock cells with and without doxycycline treatment, and NBPF1+ cells with and without doxycycline treatment, and look for changes in gene expression once NBPF1 was turned on, while controlling for both the addition of the cassette and the doxycycline.

In RNAseq, thousands of genes are detected. After basic quality control, one of the most straightfoward analyses to perform is a Gene Ontology (GO) term analysis. GO annotates genes with categories based on what we know about them (for example, "response to wounding," or "proteasome complex"). As of this writing, there are more than 15 million annotations for genes. To do this analysis, we simply 1. look for genes that have significantly higher or lower expression than control conditions, and 2. evaluate the GO terms evaluated with those top hits, looking for any patterns.

Typically, these are presented as GSEA plots (e.g. see here). GSEA plots are very good at summarizing a vast amount of information, but we wanted a way to summarize multiple plots into an easy to understand figure. The lollipop plots in observablehq are a very good way of doing this. In this figure, now in a preprint, I'm plotting the top up- and down-regulated hits from this analysis, after normalization (a term that applies to hundreds of genes is more likely to be chosen at random than a term that only applies to a few, so normalization accounts for group member size). In the plots below, NBPF1 hits are presented in red, and mock hits in blue. Asterisks represent statistical significance (calculated as a false discovery rate of 0.5 or less), and circle size represents the number of genes found for a given Go term. Plots use the same axis size and direction (absolute value) for ease of comparison. We find terms related to mitochondrial function to be significantly enriched in the data, so these are presented in boldface to highlight the enrichment.