The type-set version of my 1st chapter is out now in @AmNat!

With @ronbassar, @CGhalambor, @GuppyProject and folk not on Twitter.

https://doi.org/10.1086/711874 

If you like guppies and/or detecting rapid evolution in the wild, please take a look!

Here is a summary thread (with pics)

In a nutshell:

Male guppies evolved to mature at larger sizes, but we couldn’t see this in the wild because food levels decreased at the same time - despite selection for increased size, you can’t be big if you can’t eat enough food!

This is an example of cryptic evolution.

But there was a bit more to it than that.

Q: How did we know they evolved larger size at maturity, if we couldn’t see it in the wild population?

Was it through:

A: Common garden experiments?

or

B: Quantitative genetics, using a pedigree, mark-recapture data, mixed-effects models, and a whole lot of MCMC?

Answer: both!

Common garden experiments on ancestral and derived pops are the standard for demonstrating evolutionary change.

We then tested a quant gen approach - the animal model - using field data, and compared to CG results.

Before we go any further, I should introduce the stars of the show: the guppies!

Guppies are endemic to Trinidad and Venezuela, but are now found all over the . Males are smaller and more colourful than females.

In Trinidad, guppies downstream of barrier waterfalls live in communities with lots of predators. Some guppies escaped upstream, beyond the reach of large predators. These low-predation (LP) guppies live at higher population densities than their high-predation (HP) ancestors.

(This has happened several times - ask @jimwhiting_sci - and the distinction between HP and LP is often a bit more nuanced than that - ask @amy_e_deacon - but we’ll stick with this dichotomy here)

LP guppies have evolved a whole host of traits that are distinct from their HP ancestors.

Here, we focus on one trait - size at maturity. Male LP guppies (right) mature at larger sizes than HP (left).

PC: Helen Rodd

OK - time to introduce our study population.

Our focal population live in a stream called the Lower La Laja (go on, say it out loud - such a treat to say!), in the Northern Range Mountains in Trinidad.

I it there, even when I look like a steamed ham (see pic).

This population was founded from HP ancestors in 2008, introduced into a closed section of LP-type stream as part of a long-term experimental mark-recapture study.

You can find out more about The Guppy Project here:
https://theguppyproject.weebly.com/ 

Since then, the stream has been sampled every month.

Each fish is ID'd, measured and photographed, and if it's the 1st time we have caught them, we sample DNA and give them a tattoo.

This is a mammoth task - huge thanks to the amazing field teams over the years!

From the DNA samples, we built a pedigree. Here it is, in all it’s monstrous glory. I plotted this in #Rstats using the kinship2 package.

The pedigree structures our mixed-effects models.

It also tells us the lifetime reproductive success of each fish - our measure of fitness.

The animal model (ME-model with pedigree) lets us partition the variance in traits (z) and fitness (w) into additive genetic (A) and environmental (E) components , i.e. z = A + E.

This allows us to parametrise quant gen models of evolutionary change.

Ok - so how exactly did we quantify evolutionary change in our guppies? Here are the four methods we used.

1-3 are the quant gen approaches, parameterised by the animal model.
(See the paper for details)

4 is the common garden approach

For the CG, each year we collected juvenile fish from the intro and ancestral populations, and reared them for 2 gens in the lab to control for maternal E differences.

We then compared traits between ancestral and intro F2s reared under identical conditions - this controls for E

CG work in 4 intro streams show that the intro guppies evolve to be more like the natural LP type - and we can see this in just a few years!

Evolution is delayed until around the time when pop densities max out.

For details see: https://www.journals.uchicago.edu/doi/10.1086/705380?mobileUi=0

So - we knew that our focal population had evolved to mature at larger sizes after 3 years: z increased.

But what did the mark-recapture data look like over 36 months? (hint: see pic)

And what did the quant gen methods think?

This plot show the 3 yr change in size at maturity in the field (top) and the predictions of change from the 3 quant gen methods we used.

This looks like a classic case of “The Paradox of Stasis” - h^2 and S (in the Breeder’s Eqn), but no response seen in the field data.

This seems to be confirmed by the other two approaches, which have been used to show that the BE, when applied to wild populations, can predict evolution when none is happening (aka the paradox of stasis)…

…But wait!

We know that our population HAS evolved - so why are QG predictions 2 and 3 so wrong?

Let’s think a bit about what the animal model is actually doing.

The animal model calculates additive genetic variances and covariances, based on phenotypic similarity among relatives.

But what if relatives experience very different environments? They will be less similar than their relatedness alone would predict.

In our three year study period, population density increased more than 20 fold (!) - this meant that offspring almost always experienced higher population densities than their parents did.

Fig shows pop density on y-axis, and lines connect offspring (l) to their dad (r).

At high pop densities, there is less food, and guppies mature smaller and have lower fitness: parents and offspring were less similar than their relatedness would predict.

P.S. more about food-dependent plasticity of guppy traits soon in exciting new work from @AnjaFelmy!

When we included pop density in the animal model, predictions generally improved, more closely matching the common garden results (top).

In fact, the STS was almost bob on (if you excuse the large uncertainty).

I think the main point of the paper is that environmental drivers of selection (in this case, population density) can inhibit our ability to detect the evolutionary change they cause.

This seems like a problem.

When the environment changes over time, offspring and parents will be less similar than their relatedness would predict.

I think this is probably reasonably common in wild systems - climate change, anyone?

But - a simple solution is to include environmental factors as covariates in your animal model!

This corrects the genetic var/cov for dissimilarity among relatives that have experienced different environmental conditions.

If you’ve made it this far, you deserve a medal.

I’m afraid I’ve only got this awful pun, which you will know if you have ever seen me give a talk.

Sorry. I promise this is it’s last outing.

Thanks for reading!