Dog genes and behavior, revisited
As I mentioned in lecture on Tuesday, we are caught-up with our lecture material and will not meet for lecture on Thursday. Instead, I am offering a reading (attached) that I had described earlier, along with some explanation of one of the more important points described in this study.
Last weekend, I sent to our class description (below) of a recent publication examining behavioral-genetic associations in domestic dogs. I hadn't yet seen the original research when I wrote to you last weekend, but forwarded a news report about it that came from the home institution of the senior author on the study. I described in my message to you that some of the behavioral-genetic associations the authors reported were as high as 0.7, near to the limit of those ever reported for narrow-sense heritabilities of behavior.
Over the weekend, I requested a copy of the actual research paper from its senior author, and, upon seeing it, wanted to offer some interpretation.
Early in the term, in chapter 03, we discussed trait variation within species, and we noted (using the canine example) that artificial selection has created an abnormally high amount of trait variation within the single species of domestic dog Canis lupus familiaris.
In our next lecture (Chapter 04), we discussed behavioral genetics and narrow-sense estimates of heritability, describing the upper limit of such associations as around 0.7. We saw in that same chapter (as well as in later chapters, including Chapter 10) some estimates of narrow-sense behavioral-genetic heritability estimates that all were < 0.3, which is typical.
In this new report, the authors report behavioral-genetic associations that are much higher than those typically reported. How can this be? It stems from the artificial (and unusually large) degree of trait variation within this domestic species.
Typically, when one examines associations between traits within a natural (e.g., not artificially-selected) species, we expect some small, defined range of trait values, with correlation (association) between traits of some relatively low magnitude. Here in my Figure 1, I show the trait values and the within-species bivariate trait association for two (hypothetical) different species, such as a fox and a wolf. Within either species, there is some defined range of values for trait X (such as body length) and some defined range of values for trait Y (such as body mass). In my hypothetical example, these traits are correlated somewhat weakly within species A, and uncorrelated within species B. Notice that the two species do not overlap in trait values - a small wolf is always larger than than a large fox.
If domestic dogs were a natural species, chances are good that their trait values would fall somewhere in between these two species, perhaps closer to the 'wolf' end of the spectrum. Nonetheless, they would be expected to occupy only a small potion of the overall trait ranges.
Now, consider what the authors have done in their analysis. They have considered all domestic dog breeds to be of the same species, a fact that is technically true but which ignores the other fact that their range of trait values is anything but normal. They have analyzed behavioral-genetic associations across breeds within this single species, but here the individual breeds represent much more trait variation than natural species might, as size variation across domestic dog breeds is much greater than size variation across canine species in the wild. When associations are evaluated across multiple species (such as in my hypothetical Figure 2), the associations are often of higher magnitude. In the current study, analysis across the very artificially-distributed dog breeds behaves in the same way, resulting in behavioral-genetic associations much higher that those reported within single, natural species. 13/14 of their within-breed estimates (their Figure 1) are <0.3, just as one would expect.
This study is quite interesting, and represent the application of some very modern techniques (canine SNP chip, anyone?) to this interesting question of the heritability of behaviors. It also serves as a very useful reminder of
- the power of artificial selection - modern dog breeds are estimated to have been developed only over the last 300-500 years. For a natural species to evolve as much trait variation in this short time is unheard of.
- the danger of reliance upon secondary news sources - the original news story that I sent to you accurately describes the gist of this research study, and highlights the very strong associations found. But, it also leaves out enough detail that it is not possible to immediately assess why the associations are of such magnitude.
- the importance of proper modeling of evolutionary constraint - as shown in my hypothetical example Figure 2, trait associations across species can be artificially inflated if simple, linear techniques are used instead of methods that account for shared evolutionary history, such as independent contrasts analysis or nested ANOVA. The authors do have a phylogenetic model for their dog breeds; I am not schooled well-enough in the jargon of their analytical models to know if they have fully controlled for relatedness. Whether they have, or have not, these types of broad comparisons should always be examined with an eye for that type of concern.
- the imperfection of any one study - this is a research report describing one body of work on this topic, and I'm certain we could find other, similar/related studies. Is this study perfect? Certainly not. Is it still interesting, and useful? Absolutely. Any one research study can only advance our understanding incrementally. It's too easy, and too common, to dismiss work outright for containing flaws - it's more important to ask, given such flaws, is there anything that we can learn? The latter approach is more fruitful, and provides a much better return on one's investment of time and effort. Here, the traits with the highest across-breed heritabilities are trainability, aggression, and attachment - exactly those traits we might expect to have been key in the artificial selection/shaping of the human-dog relationship. It's a nice confirmation that these are strongly heritable, in ways that have translated into very powerful differences among breeds.
I've spent perhaps too much time dissecting some of these points, but I do so because they put some of our lecture material into sharp relief. Textbook examples are often too carefully culled to represent cutting-edge investigation; it's both fun as well as useful to see where current researchers in these areas actually are working.
Have a great rest of the week - see you on Tuesday for Chapter 11.
Leave a Reply.