Tuesday, January 27, 2009

Two Animal Nervous Systems?

Why would anyone ever care about a phylogeny? Why on Earth should we care which species are each others' closest relatives?
ResearchBlogging.org
The reason boils down to traits. If two species share a very similar trait, but they are distantly related phylogenetically, then something interesting has happened in evolution. Perhaps that trait has evolved twice separately, and if that is true, perhaps we could predict that under similar conditions, another species would also evolve that trait. Alternatively, maybe the trait in question was lost multiple times in intervening groups - were the genes also lost? Now this is the interesting stuff, trait evolution!

Unfortunatley, I think that phylogeneticists sometimes short change traits. Phylophiles' papers can tend to get wrapped up in the relationships of a particular group, to the point of forgetting why anyone would ever care, forgetting what those relationships might mean (anyone else attend an Evolution meeting during the heyday of molecular phylogenetics, circa 1995? You know what I mean).

Symptomatic of this focus on the tree and not on the traits are cases where a paper spends an enormous amount of energy (mental and computational) on inferring the best tree; but then makes claims about the evolution of traits without a single statistic.

A paper this week by Shierwater et al, published in Plos-Biology precisely illustrates this point. The authors do an admirable job of resolving the major branches of the animal tree of life. They present a large dataset and compare many approaches and models with a statistical rigor that would make Willi Hennig, RA Fisher and Reverand Bayes proud. The authors end up with a well-supported, and fascinating result: Animals are divided into two epic clades - the bilaterians (e.g. flies, clams, and humans) on one hand and all other animals (sponges, jellies, and the enigmatic Trichoplax) on the other. But why might we care where Trichoplax falls on the tree? It's because of the traits.

It's in the traits that the paper disappoints me. After putting the computational pedal to the metal to estimate the tree and check its sensitivity to different statistical approaches, the paper simply asserts the paradigm-shifting claim that organ, nervous, and sensory systems must have evolved in parallel in the two epic clades. Here is how they stated it:

"We conclude that the higher animals (Bilateria) and lower animals (diploblasts), probably separated very early, at the very beginning of metazoan animal evolution and independently evolved their complex body plans, including body axes, nervous system, sensory organs, and other characteristics."

in the press:

"Nervous systems are found in both groups (among the lower animals, jellyfish have nervous systems), so the new arrangement means that these systems must have evolved twice in the history of animal evolution..."

and also in the paper:
"Most notable of these aspects is the evolution of the nervous system, which in the hypothesis in [the figure showing their tree], can only be explained by convergent evolution of Cnidaria and Bilateria nervous system organization."

So, what about the possibility of loss? Let's examine this possibility with the most rudimentary, and intuitive of phylogenetic methods, parsimony. Let's also assume that the phylogeny presented in the paper is the correct tree.



panel a and b show the phylogenetic tree supported by the recent paper. Panel a shows two separate gains of nervous systems (black vertical bars). Panel b shows an alternative hypothesis, not mentioned in the paper (unless it's in the supplement, which I haven't read yet). In the alternative, nervous systems originated with animals, and were lost (white vertical bars) in the lineages leading to Trichoplax and sponges.

Yes, two changes is fewer than three, and this is the story the paper and the press went with. But is it really more parsimonious (OED definition "the simplest state, process, evolutionary pathway, etc")? Can we really argue that it is simpler to evolve a nervous system twice than it is to gain it once and lose it twice? Maybe we can? This is what we are left with when using parsimony. (Likelihood and Bayesian methods often get us only slightly farther when dealing with characters).

That said, perhaps we can use these two alternative evolutionary pathways - two gains versus one gain and two losses - as alternative hypotheses (of course there are other possible pathways, but I'll ignore those for now). Do these hypotheses make alternative predictions? (Self promotion: please cite Oakley and Cunningham, 2000 if you buy this alternative hypothesis bit. You only have to read the last paragraph).

If the bilaterian and jelly nervous systems are independently evolved, we might expect them to be quite different, and use mostly different genes. This is not the case. We know of many similarities of the nervous systems of jellies and bilaterians.

Alternatively, if nervous systems were lost, we might expect to find remnants of nervous systems littering the genomes of sponges and Trichoplax. In fact, this is exactly what is found. Sponges and Trichoplax have many genes homologous to genes used in bilaterian nervous systems. (Shameless self promotion: please cite Sakarya et al 2007 which shows many synaptic genes are present in the sponge Amphimedon. We mostly assumed sponges as sister to other animals for that paper. But we still mentioned the possibility that the sponge lineage could've lost synapses).


Of course, in the end a "nervous system" is not one thing. Therefore, the question of homology becomes a bit of a red herring. Everything, except protein domain homology is partial homology! Instead, it becomes much more interesting and instructive to break down multi-part systems like nervous systems into components and to trace the evolutionary history of those components. Do these components evolve in a concerted way? Do components come and go? What processes drive divergences of these genes?

I know Shierwater et al are aware of these issues, but I'd have appreciated a more balanced discussion of alternative possibilities (what if Trichoplax is reduced, and not the ancestral animal? See Ryan, Chuck, and Bart on this one.) In the end, the traits are the interesting bit, so we should analyze them with the same statistical rigor as the tree, and treat interpretaions with the same caution and balance!!


Refer

Bernd Schierwater, Michael Eitel, Wolfgang Jakob, Hans-Jürgen Osigus, Heike Hadrys, Stephen L. Dellaporta, Sergios-Orestis Kolokotronis, Rob DeSalle (2009). Concatenated Analysis Sheds Light on Early Metazoan Evolution and Fuels a Modern “Urmetazoon” Hypothesis PLoS Biology, 7 (1) DOI: 10.1371/journal.pbio.1000020

16 comments:

T Ryan Gregory said...

"anyone else attend an Evolution meeting during the heyday of molecular phylogenetics, circa 1995"

In 1998. It was mostly trees with nothing mapped on. I remember thinking, this is essentially worthless without some application of the phylogenies.

Anonymous said...

See a discussion of this article in a recent "news" from Nature. It seems this phylogeny is not that brilliantly constructed after all. Here's a quote from famous molecular phylogenist Hervé Philippe :"I am tired of these molecular papers that don't make sufficient controls to check the reliability of the phylogenetic inferences".
From recent phylogenies (see Dunn et al., Nature, 2008), with a broad taxon sampling, it seems that ctenophores (possessing a nervous system) might be placed as a sister group of all metazoans. Publishing an article after this phylogeny but with a restricted taxon sample is very peculiar (but didn't seem to bother the reviewers appointed by PloS).
Quick question, Schierwater is in the list of authors from Sakarya, et al. What was his reactions concerning the conclusions in the article at odds with his own studies?

Anonymous said...

Sorry, I thought Schierwater was in the list of authors, but he was listed in another article (Trichoplax genome, Nature, 2008) where the phylogeny published was different from his results published in PNAS in 2006. That will teach me to post comments without checking my sources.

Anonymous said...

Whether one gain and two losses is more parsimonious than two gains would depend on what the probabilities are for gain and loss (which might even be different on each lineage, depending on what information one has).

I recall we talked about this some years ago, thinking about the trait of not having, say, eyes. On planet A it might say something highly significant about relatedness that two lineages don't have eyes, if everyone else do, but on planet B it may mean nothing much if very few species have eyes.

Has there been attempts at trait reconstruction based on detailed estimation of probabilities of loss and gain?

Todd Oakley said...

Bjorn -
There is pretty much research on using the parameters of a likelihood model that has gain and loss. My 2002 PNAS paper compared a range of assumptions about rates of gain and loss (gain:loss ratio), to examine under what models ostracod eyes are "independently" evolved.

The problem is, we never know what the true rates of gain and loss are, and if we estimate from the data, we can be misled.

People often think only about the mutational bias of gain versus loss of complex traits. It's easy to imagine mutations that disrupt complex traits, but originating them could require many many mutations.

But it's not only about mutation - those traits also have to be fixed in populations. So it really comes down to macroevolutionary/ecological processes. In the case of eyes, we can imagine that losing eyes would occur when shallow water taxa invade a dark niche. How easy might that be for a shallow water adapted species to invade the deep, versus a deep eyeless species colonizing the shallows?

Hard to know.

Anonymous said...

Is it practically possible in some cases to estimate the probabilities of gain or loss by looking at the selective advantages that they confer? For example, loss of vision in cavefish makes sense not only because they actually have eyes that just don't confer vision, but also because they live where there is no light. Thus that environmental factor increases the probability that a loss is more parsimonious in that case.

Anonymous said...

Thanks for posting this -- I ostensibly still work on metazoan phylogeny, but I just can't keep up with it any more. Too much happens too fast.

I agree with your interpretation, but it's hard for me to accept that tree. It smells like another rooting problem. If the Schierwater et al. tree is rooted on the branch leading to sponges, you get (sponges ((cnidarians + ctenophores) (placozoans (bilaterians)))) (if my notation is correct there), which makes a bit more sense to me. I'm sure they tested that rooting and rejected it, though. Guess I'd better read the paper!

Of course, I said basically the same thing about Dunn et al. (where ctenophores were sister to all other animals) because I simply cannot accept anything that doesn't put sponges as sister to all other animals. I guess I've turned into one of those old guys who is convinced something is wrong with all these newfangled trees unless they exactly match my notion of how evolution worked!

Unknown said...

As an absolute naive and ignorant bystander (and journalist, go figure), I didn't get it this way. They say Trichoplax (better, the placula) has some hint of nervous system in the genome - as you say. And they conclude that "the genome of the placozoan [...] ideed delivers some notable evidence that the genetic inventory may precede morphological manifestation of organs". It looks much as your hypothesis - one gain, two losses. Where am I mistaken?

Todd Oakley said...

Marco wrote "delivers some notable evidence that the genetic inventory may precede morphological manifestation of organs". It looks much as your hypothesis - one gain, two losses. Where am I mistaken?"


By stating the genetic inventory PRECEDES the manifestation of organs, they are assuming panel a of my figure - that "organs" (e.g. nervous system, but also other traits) evolved separately in the jellies and bilaterians. They are saying that the genes for nervous systems were there in the ancestral animal, but not the morphological traits, not the organs.

My point is that they didn't discuss the alternative of one gain *of morphology* and two losses.

One has to decompose the genes and the morphology.

(I wrote this very quickly, if it makes no sense, let me know)...


Andy wrote "It smells like another rooting problem".

I agree that rooting the animal tree is a difficult problem. But I don't know what they could've done differently. Choanos are pretty clearly the best outgroup.

Also, to Pierre's suggestion of limited taxon sampling. Taxon sampling can often be improved, but this seems pretty reasonable to me, relative to other studies.

Todd Oakley said...

Bjorn wrote "Is it practically possible in some cases to estimate the probabilities of gain or loss by looking at the selective advantages that they confer? ....."

This is only part of the equation. The other important part is knowing biogeographically, how often do lineages move between places with very different selective regimes. In your example, it'd be about knowing how often fish moved into caves and stayed long enough to lose eyes; relative to knowing how often cavefish moved out of caves and regained eyes.

Rates of gain v loss are an amalgam of mutation, fixation (selection), AND biogeography/ecology. (And as you say, this could all be different in different parts of the tree. ouch).

Unknown said...

Now your thoughts make perfect sense. You wanna see a brain in Trichoplax 8-)

Jokes aside, don't you think having genes for nerve structures (or some such) and NOT the nerve structures themselves is a case of exaptation?

Marco

Anonymous said...

Todd wrote: I agree that rooting the animal tree is a difficult problem. But I don't know what they could've done differently. Choanos are pretty clearly the best outgroup.

As you know, sometimes you're just stuck. You probably can't do better than choanos, but it's quite possible that even heavily sampling choanos and other possible relatively close outgroups won't allow us to nail down a root for Metazoa with high confidence.

I think this is an underreported issue that shows up in lots of groups we're really interested in. Closest extant outgroup for Aves? Crocs -- a long branch that probably isn't going to be broken without a time machine. Ditto for coleoid cephalopods and Nautilus. Anywhere you have a deep divergence and the closest outgroup is depauperate you have a potential problem. In those cases, throwing ever more distant outgroups into the analysis probably won't help, and can make things worse.

The position of the root just may not be resolvable in some of those cases. I hope Metazoa isn't one of them, but given the conflicting hypotheses that have been floated just in the past couple of years, I am starting to worry.

Todd Oakley said...

Andy - Yes, adding more choanos might help. I don't know off hand how disparate they are (ie if they are all very closely related it won't help much).

One other idea would be to examine genes that duplicated between choanos and animals. Each paralog could be used to root the other paralog. I did this for my MS degree on salmonid fishes, which are a tetraploid family. Also, Donoghue's lab had a big paper using the approach to root Angiosperms, perhaps about 10 years back.

This would make a cool bioinformatics project!

Todd Oakley said...

Marko wrote "Now your thoughts make perfect sense. You wanna see a brain in Trichoplax 8-)"

Actually, this makes me think you might not be understanding my point. Or maybe I am just not getting the joke. My main point is that Trichoplax is not an ancestor - it could be that it changed a lot since the ancestor of animals, and that its "brain" (or synapses, or neurons) were lost during evolution, but that its common ancestor with other animals had neurons. Losing neurons doesn't mean it has to lose all the genes that make a neuron. Those genes could be - and are - used in many other things



Marko wrote "Jokes aside, don't you think having genes for nerve structures (or some such) and NOT the nerve structures themselves is a case of exaptation?"

Yes, absolutely. In fact, I wrote most of this paragraph in our Plos-1 paper on sponge synapse genes:

"A fundamental question in biology is how complex, integrated traits like nervous systems originated. One approach to this question involves analyzing individually the numerous components of a defining structure, such as the post-synaptic proteins of nervous systems, to determine which – if any – predate an integrated nervous system. These exapted [1] components, defined as a biologic unit originating with a function other than that for which it was later selected, can be revealed by comparative genomics of animals with nervous systems and their closest relatives that lack them."

See these posts in pharyngula, Newsweek. The open access paper and other news sites (including radio interview) are here.

Anonymous said...

Todd wrote: Yes, adding more choanos might help. I don't know off hand how disparate they are (ie if they are all very closely related it won't help much).

Right. And that's often the problem -- there are two species of Nautilus that are quite divergent from each other, but relative to the distance from them to the other living cephalopods, I think they are quite closely related. I'm pretty sure it's the same deal with crocs and birds.

One other idea would be to examine genes that duplicated between choanos and animals. Each paralog could be used to root the other paralog. I did this for my MS degree on salmonid fishes, which are a tetraploid family. Also, Donoghue's lab had a big paper using the approach to root Angiosperms, perhaps about 10 years back.

And it's been used to attempt to root the entire Tree of Life. In fact, it's probably the only way you could root life, unless we find Martian relatives (and if life is even rootable, of course).

This would make a cool bioinformatics project!

Yes, it would! And we now have genomes for at least one representative of several of the key taxa...hmm. Could be a good excuse to finally learn a bit more about bioinformatics/comparative genomics...?

John Cowan said...

"Never say 'higher' or 'lower'." --Darwin's notebooks

(I realize it's just a quotation, but I can't post a comment on a scientific paper. Yet.)