Monday, June 26, 2006

Redundantia ad absurdum

[Update: I've fixed and added some links. An argument has been added, in red.]

You might have thought that this discussion was over (see also here and here). However, Salvador has now decided to reply (also here) to parts of my last piece. My immediate responses can be found under the comments cited, but I thought I should bring some of the more interesting points over here for detailed consideration.

Full or partial redundancy?

Salvador has finally clarified what he and Denton were talking about in their repeated claims that the C. elegans vulva "is generated by means of two quite different developmental mechanisms, either of which is sufficient by itself to generate a perfect vulva". It turns out that they meant the graded, action at a distance by the EGF-like ligand LIN-3 vs the sequential, non-graded lateral signal mediated by LIN-12. (Note: The earlier ambiguity was a serious failure of communication on their part. When you make a scientific assertion you must be very precise on the details. Mentioning "two mechanisms" in the context of vulval development will be interpreted as referring to "vulval induction and lateral signaling" by any C. elegans biologist. My second guess was the synMuv genes, because Salvador was discussing knockout experiments at the time. How was I supposed to know what Salvador meant?)

They based their arguments on a mini review by Cynthia Kenyon. Here's a crucial passage:
"[...] A simple and attractive model is that the two pathways both operate and are partially or fully redundant. This set-up would enable the vulva to form perfectly in every animal, which it does."
Salvador then decided to go on the offensive:
"I will be interested to see if he affirms or retracts his comments based on Cynthia Kenyon's article which way pits Sternberg (not richard) against others in the debate over nematode vulva development. Ricardo cited sternberg, but not the competing opinion it seems. However, I am willing to stand corrected."
First of all, let me introduce the figures involved. Cynthia Kenyon and Paul Sternberg are both giants of C. elegans developmental biology. Apart from doing classic work on several systems, such as vulval and embryonic development, and Hox genes, Kenyon turned to the genetics of aging and completely revolutionized the field (for example, with her work on the insulin receptor daf-2 and the transcription factor daf-16, or the signal from the germline that inhibits life-span). While I occasionally disagree with some of her evolutionary thinking, I have enormous respect for her work. Paul Sternberg began by resolving the gonadal and nongonadal post-embryonic cell lineages of the nematode Panagrellus redivivus, a classic evo-devo study from before the time it was called that. He then became famous for discovering the EGF receptor / RAS / RAF / MAPK vulval induction signaling pathway. He has also spearheaded the development and evolution of WormBase in recent years (that's the tool that allows me to gene-name-drop so effectively). Sternberg (definitely not Richard) is perhaps the most knowledgeable scientist in the world when it comes to vulval development.

I hate to disappoint Salvador, but it isn't helpful to talk about "competing" opinions here, for three reasons. First, the mechanisms raised by Kenyon are explicitly discussed in Sternberg's review. Second, Kenyon's piece is a small review written over 10 years ago, and it's hardly appropriate to compare it to the latest comprehensive survey of the field. Third, Kenyon was clear about the speculative nature of her proposal. She admitted that the papers she was reviewing did "not indicate which of the two signaling mechanisms plays the predominant role during normal development" (p. 172). Her proposal was described as "a simple and attractive model", not the final answer to the problem.

So what's the answer? My answer is that the two mechanisms are only partially redundant, not fully redundant (you'll notice that Kenyon was open to either possibility), for two reasons. First, the signal from the anchor cell (encoded by lin-3) is necessary and sufficient for vulval development. If the signal from the anchor cell, graded or not, is abolished (for example, by killing the anchor cell), all
vulval precursor cells (VPCs) acquire tertiary fates and no vulva is formed at all. Second, lin-12 (lateral signaling) is necessary and sufficient for specification of the secondary fate VPCs. Mutant individuals with null alleles of lin-12 do not generate secondary fate VPCs and, thus, cannot form a vulva. However, lateral signaling cannot form a vulva on its own, in the absence of a signal from the anchor cell. All this means that Denton's contention that the vulva "is generated by means of two quite different developmental mechanisms, either of which is sufficient by itself to generate a perfect vulva" is incorrect. And so is Salvador's original contention that got this discussion started:
"One can do a knockout experiment on one of the develomental pathways of a nematode vulva and then an alternative develomental pathway kicks in to create the vulva. There are two independently successful redundant developmental pathways in the vulva."
No. The graded signal is not sufficient to generate a vulva (it cannot generate secondary fate VPCs in the absence of lin-12 activity). I hope Salvador remains "willing to stand corrected".

So does it matter whether any particular biological system shows partial as opposed to complete redundancy? I believe it does. Most biological systems show some redundancy, but full redundancy is relatively rare. Partial redundancy is exactly what you'd expect to appear under natural selection. In contrast, I suspect that if perfect, full redundancy were all around us in biological systems, Salvador would be claiming it as evidence for design (more about this later).

Here's a point by Sternberg which Salvador might appreciate:
"C. elegans VPC patterning involves multiple sources of information each of which would be capable, in principle, of generating a normal pattern. [...] These pathways might act in combination to generate an exquisitely precise pattern. In C. elegans the graded anchor cell LIN-3 signal and LIN-12-mediated lateral signaling are the major players. In other species, the other pathways might predominate."
Surprise, surprise, this Sternberg is talking about evolution.

[Note: Pharyngula has some nice introductions to vulval development and evolution.]

The Platonic vulva

So why are vulval development mechanisms redundant? Apparently, we are all in agreement about that. In Kenyon's memorable phrase "this set-up would enable the vulva to form perfectly in every animal, which it does". Sternberg makes essentially the same argument. Redundancy is for robustness. Where we differ is over whether this redundancy and robustness have evolved (Kenyon, Sternberg, myself, and just about every practicing biologist in the planet) or designed (Salvador, Denton, and a few others). Before getting to that question, let me just make another point.

I suspect that another reason why Kenyon's piece appeals to Denton and Salvador, is the Platonic undercurrent of the concept of a "perfect vulva". As I've argued in the past, this is an old foible of C. elegans biologists; taking the whole "959 somatic cells" thing far too seriously. Christian thinkers have long been attracted to Platonism, of course. Darwin showed that this was the wrong way to look at biology. A better way is to embrace variation -- what Ernst Mayr called "population thinking". So are worm vulvae as Platonic as Kenyon suggests? No, worms do not challenge population thinking in any way. In one of my favorite C. elegans evolution papers of the last decade (i.e., one I wish I had written), Marie Delattre and Marie-Anne Felix showed that the vulva doesn't actually "form perfectly in every animal": they found a few errors here and there when they looked at thousands of individuals. Incidentally, Felix and her co-workers are doing some of the most exciting work on developmental robustness I know of, much of it on the vulva. It's largely unpublished at the moment, but I'll summarize it here when it comes out.

An argument from redundancy?

So why does Salvador believe redundancy supports ID creationism? This is actually a difficult question to answer because the argument hasn't been laid out in sufficient detail (compared to, say, Behe's "irreducible complexity" argument). Indeed, redundancy is the opposite of "irreducible complexity" (see comments here), so it isn't at all clear what Salvador is talking about. However, going back to the original discussion, Salvador did sketch an argument:
"Regarding the issue of redundant functionality, it does pose a problem for natural selection as a mechanism for it's evolution. Since there are frequently circumstances under which the functions may not visible to selection over several generations, this poses a problem for natural selection. I would not say insurmountable, yet, but substantial."
Substantial? Why? You agree that redundancy is for robustness, right? Now, we know that robustness evolves. For example, C. H. Waddington selected in the laboratory on several different kinds of developmental robustness in Drosophila melanogaster and readily observed selection responses. In the 1950s! Since then, robustness has been characterized in different kinds of organisms (e.g., viruses, bacteria) and systems (e.g., genetic code, development, circadian clocks). We know that robustness responds to selection imposed by high rates of mutation, recombination or environmental perturbation. We know it responds to different kinds of selection, such as, stabilizing, directional or fluctuating. We have simulation models and mathematical theory (also here, here, here, ...). The field has been reviewed, re-reviewed and re-re-reviewed. So what is the problem, exactly? Redundancy can be selected for if it promotes robustness, which it obviously does -- no problem!

Salvador then returns to the vulva:

"In the case of the nematode vulva, without some co-option, the independent path way can not be selectively advantaged unless the other develpmental pathway is knocked out. The independent pathway, would have to be pretty much functional when it appears as it is critical to perpetuation."

No. Partial redundancy does not require such an explanation. An initial sloppy pattern can evolve, and then partial redundancy can be added piecemeal to stabilize it. Even complete redundancy could, in principle, have evolved from partial redundancy, making highly suspect any "arguments from redundancy" for ID. In order for you to claim that redundancy implies design, you must show that there is no way the redundancy could have evolved by natural selection. Good luck establishing that for any system.

I'm afraid that against this background Salvador and Denton and Sanford have to be a lot more specific. It isn't good enough to assert that "problems" exist, especially when you give the impression that you haven't done your homework. If anyone is going to pay any attention you also have to explain where exactly the hundreds of earlier studies went wrong.