Variation is common, and often lingers in places where it is unexpected. The drawing to the left is from West-Eberhard’s Developmental Plasticity and Evolution(amzn/b&n/abe/pwll), and illustrates six common variations in the branching pattern of the aortic arch in humans. These are differences that have no known significance to our lives, and aren’t even visible except in the hopefully rare situations in which a surgeon opens our chests.

This is the kind of phenomenon in which I’ve become increasingly interested. I work with a model system, the zebrafish, and supposedly one of the things we model systems people pursue is the ideal of a consistent organism, in which the variables are reduced to a minimum. Variation is noise that interferes with our perception of common underlying mechanisms. I’ve been thinking more and more that variation is actually a significant phenomenon that tells us something about where the real constraints in the system are. It is also, of course, the raw material for evolution.

Unfortunately, variation is also relatively difficult to study.


There is some good work in the scientific literature that emphasizes variation. The drawing to the right is from a well-known paper on the morphology of amphibian limbs. There was a fortuitous (for us, not them) mass kill of newts in a small pond in California — a sudden freeze catastrophically killed the entire population. Shubin et al. (1995) collected more than 500 dead rough-skinned newts (Taricha granulosa), and counted and characterized the pattern of bones in their limbs. The drawing is of the canonical pattern. Only 70% of the population actually had limbs like this: the rest had missing bones, additional bones, and fused bones; interestingly, some of the patterns were atavisms, in that they resembled the limb structure of related but more primitive species.

Variation is common and important, and it’s all around us. The questions are about the source of that variation, and as I’m finding, how to study it. It’s hard to capture and isolate noise in a way that allows one to study it with the same consistency and convenience as regularities.


West-Eberhard discusses another particularly instructive paper, some observations made by Slijper in the 1940′s on a bipedal goat. This poor beast was born with missing forelimbs. Surprisingly, it lived for a year before it died in an accident (presumably unrelated to its deformities), and adapted in unexpected ways. It could hop around on its two hind legs, and developed additional peculiarities: enlarged hind limbs, a curved spine, and large neck.

At left are the bones of the hind limb of a normal goat (A) and one that had been born with congenitally absent forelimbs (B). The differences are obvious, and are related to the changes in posture the goat had to maintain.


Pelvic musculature of a normal (A) and two-legged (B) goat. The gluteus muscle (“gt”) has a long anterior extension, which has been reinforced anteriorly by novel tendons (“t”). These are new anatomical structures that were generated in the absence of any direct genetic specification.


The pelvic bones responded to the unusual stresses imposed on them with changes in shape, as well. To the left is the pelvic skeleton of a normal (A) and two-legged (B) goat (“i”=ischium). Slijper noted that the dorsoventral flattening and elongation of the ischium resembled the forms seen in kangaroos, a naturally bipedal animal.

These are not genetic changes — the goat was ordinary domestic stock, and presumably had perfectly ordinary genes that, under normal circumstances, would have generated more typically goatish morphology. These are instead the indirect consequences of a plastic phenotype, responding to a radical change in its environment.

The point of the two-legged goat example

…is not to argue that these handicapped individuals might have given rise to durable novelties, for it is unlikely in the extreme that such individuals would outperform normal individuals in nature. Rather, the point is to dramatize how a change in one aspect of the phenotype — in this case the front legs — can lead to correlated changes that show a degree of complexity and functional interaction that we usually assume to require generations of natural selection and genetic change at many loci. Clearly, the phenotype can be elaborately restructured, due largely to adaptive plasticity, without a proportional restructuring of the genome.

I think there are several important messages here. One is that genes are not “for” some feature; the absence of forelimbs did not conjure a gene “for” gluteal tendons into existence. Rather, cellular patterns of gene expression are regulated in response to the environment, and in turn modulate the environment of other cells and tissues. We have been conditioned by years of good (but selective) results in genetics and molecular biology to view gene expression as an end result. It is not. In development, gene expression is part of a process that produces an end result in collaboration with multiple other factors.

Another message is that the flexibility of an individual’s genome is great. It’s easy to slip into the trap of thinking that the phenotype is a predictable consequence of the genotype, that development is a process of translating genes into morphologies, in a way that is just more prolonged and elaborate than the way transcriptional machinery translates gene sequences into proteins. This isn’t the case. Developmental plasticity means that a single genome contains the potential to generate multiple morphologies.

Shubin N, DB Wake, and AJ Crawford (1995) Morphological variation in the limbs of Taricha granulosa (Caudata: Salamandridae): evolutionary and phylogenetic considerations. Evolution 49(5): 874-884.

West-Eberhard MJ (2003) Developmental Plasticity and Evolution. Oxford University Press.


  1. #1 Mrs Tilton
    January 19, 2006

    Fascinating! I may have to look into West-Eberhard, but I have a nagging fear she will not prove as laity-friendly as Sean Carroll.

    Of course it makes perfect sense that the very different mechanical forces at play in quadra- and bipedal goats would mould very different bodies. (After all, if anybody doubted that unusual mechanical stresses can produce highly unusual musculo-skeletal systems, they need only look at the governor of California.) But if I may play devil’s advocate, how do we know that a variation in genotype was not (also) a cause of the bipedal goat’s unorthodox build? Was Slijper able to determine that the missing forelimbs had a purely environmental cause? If not, why could not a mutation have helped form the unusual pelvis and hindlimbs, just as a mutation might have suppressed the forelimbs? (Sorry, my mind is all abuzz with appendage-suppression these days. Did I mention that I am reading Carroll?)

  2. #2 Kagehi
    January 19, 2006

    Well, if you think about this, its hardly a surprise that the plasticity that exists on the gene level would extend to a morphological level from environmental effects. I essence, if RNA is plastic enough to adapt to 50 different sets of codes that all produce insulin, why wouldn’t the body as a whole show similar plasticity? As I told someone yesterday, if you take the 25% successful birth rate and add in the estimate of eight out of nine births producing twins, but with only about 1% of the twins surviving, the number of non-viable offspring is roughly 9 out of every 200 “possible” births. And this is *with* plasticity in how many versions of a sequence still produce a working (or even close enough) version of a protein.

    Imagine the number of tries you would need without that plasticity, where only *one* sequence for every protein actually worked… Something like 1 in every 100 billion tries? 100 trillion? Without knowing how many total proteins are produced by genes, and how many possible variations could exist in each sequence, its impossible to even guess what the odds would be for every one of thousands to have the ‘exact’ pattern needed. Plasticity of morphology on the environmental scale seems to me to be a logical consequence of a system that is so adaptive to differences that a 1 in near-infinity possibility of success is expanded to be more like a 9 in 200 chance. Of course, this all assumes I didn’t screw up my math again, like I realized I did yesterday. I am not that good at statistics. lol

  3. #3 Mnemosyne
    January 19, 2006

    If I live past the point where my organs would be useful for donation (and since the women in my family live well into their 90s, it’s definitely possible), I would seriously consider donating my body to science for dissection by medical students. It’s not like I would be using it any more, and my religious beliefs (such as they are) don’t demand bodily integrity after death, so I’ve been pondering it.

    Of course, at this point, I’ve got about 60 more years to consider it, almost twice as long as I’ve already lived. So I suppose I might change my mind.

    My only regret would be never knowing what strange anomolies might lie within me.

  4. #4 Todd
    January 19, 2006

    My question is rather similar to Mrs. Tilton. Unsurprisingly, I too just finished Carroll.

  5. #5 John Wilkins
    January 19, 2006

    I’m so pleased to see a decent review of Slijper’s goat, but saddened, because it means I have to do actual work, that you didn’t give a reference to that study directly…

  6. #6 Troy Britain
    January 19, 2006

    According to Armand Leroi, 1 in 10 people have an extra set of ribs!

    And some are a set short of average. This appears to be not too uncommon amongst vertebrates and it makes hash out of the common antievolutionist objection to horse evolution based on the fact that the fossil “horses” vary slightly in rib count. If vertebrates can have more or less sets of ribs within a species, then more or less sets between fossil species is hardly evidence against a descent relationship.

  7. #7 miko
    January 19, 2006

    I think this is fascinating stuff…and mysterious because it is relatively unstudied (because, as you mention, most scientific techniques and reasoning rely on working with supposedly uniform populations). Since it just came up before, it seems like developmental systems theory (DST) is trying to tackle some of this, but with no data. And it just doesn’t fit into current experimental methods.

    So…how do you study developmental noise? What is the adaptive advantage (if any) of developmental processes evolving to respond to or compensate for a deleterious mutation, like no forelimbs? Is it selection pressure for variety?

    One perhaps related example is the mind-blowing (to me, anyway) work by Susan Lindquist (and here) proposing that Hsp90 buffers genetic variation in developmental genes until there are environmental changes, then unleashes the “suppressed” phenotypes to be subject to new selection conditions. Then, if you want your mind blown again, read her stuff on prions, particularly the work on prions and memory with Eric Kandel.

  8. #8 rubberband
    January 19, 2006

    Todd and Tilton make good points (At least I think so, because they are exactly what my first thoughts were!), but wouldn’t any “unintended” genetic-developmental cascade produce random, and therefore most likely harmful/inappropriate changes? It seems far too fortuitous that the alterations that occurred were precisely those most useful in the birthed offspring. This seems like simple plasticity–environmentally imposed changes, like developing callouses after hard abrasion, big muscles after weightlifting, scar tissue after injury, or tiny feet after binding, than a genetic-developmental cascade. Those leg-parts were physically restrained from maturing normally.
    Unless somehow the alternative (no forelimb) variation allows expression of latent atavistic genes (which had already been subject to selection as being useful for some kind of no-forelimb predecessor), but this requires formidible mutation resistence in inactive sequences, and begs for a two-limbed ancestor (which I do not think exists).
    On the other hand, at my level of understanding there is very little difference between what I’ve written above and just making shit up.

  9. #9 Ricardo Azevedo
    January 20, 2006

    Interest in phenotypic plasticity, canalization (now called robustness) and developmental stability aren’t new in evolutionary biology — Schmalhausen and Waddington, two pioneers of evo-devo, devoted large parts of their careers to their study. What West-Eberhard is talking about (she calls it genetic accommodation), is not too far from the so called Baldwin-Waddington effect (by her own admission). However, the evolutionary importance of these mechanisms is still controversial. For example, Williams has pointed out that the arguments against the evolutionary significance of “hopeful monsters” do not depend on whether the monstrosities have a genetic or an epigenetic basis. Another example of a genuine, meaningful “controversy” in modern evolutionary biology…

  10. #10 Anonymous
    January 20, 2006

    Interest in phenotypic plasticity, canalization (now called robustness) and developmental stability aren’t new in evolutionary biology — Schmalhausen and Waddington, two pioneers of evo-devo, devoted large parts of their careers to their study.

    Yes, but before molecular genetics (and until development is better understood in evolutionary terms and vice versa) it’s pretty much theoretical conjecture…not to minimize the coolness and brilliance of people like Waddington). These issues have not been subjected to very much rigorous experimental investigation, partly because they lie predominantly outside the current methodological practice of developemental genetics.

  11. #11 CCP
    January 20, 2006

    Very cool…I’ll be showing that aorta figure in my Physiology course this semester.
    I study an interestingly relevant system: phenotypic variation induced by incubation temperature in turtles. Gonadal sex is determined by incubation temperature in most, and we find a bunch of other traits that are also affected, many of them likely components of fitness, including hatchling size, post-hatchling growth rate, temperature preference, locomotor performance, metabolic rate…if I wasn’t so averse to killing the little guys I’d look for morphological variation too.
    What’s interesting to me is that some of this environmentally induced variation is temporary and thus reversible (“acclimation”) while some is seemingly permanent and irreversible (like sex)…and we’re starting to get hints that which is which differs among species.
    There’s a hell of a lot more to phenotype than the Central Dogma, and natural selection has to sort phenotypes in order to affect heritable, genetic variation…much complexity.

  12. #12 Dave Harmon
    January 20, 2006

    “These are instead the indirect consequences of a plastic phenotype, responding to a radical change in its environment.”

    I’d also call it a demonstration of at least some “front-loading”.

    Rubberband: “wouldn’t any “unintended” genetic-developmental cascade produce random, and therefore most likely harmful/inappropriate changes?”

    Not if the cascade was a reactivated atavism; then it might be incomplete, but not randomized.

    “Unless somehow the alternative (no forelimb) variation allows expression of latent atavistic genes (which had already been subject to selection as being useful for some kind of no-forelimb predecessor)”

    Almost right — you don’t need a “no-forelimb” predecessor, just one that had some adaptations towards standing on hind legs, e.g. to crop tree leaves or tall bushes. (It wouldn’t necessarily have been as big as a modern goat!).

    Developmental cascades are complex weaves; on one side you have the genetic “imperatives”, but the actual results depend heavily on “expectations” about the local environment. Between the two stands developmental “state”, referring to the current results of that interaction. E.g., if you raise a kitten blindfolded, the visual systems won’t develop properly — the cat will be blind. Even if you uncover the adult cat’s eyes, the developmental state needed to develop them has been overwritten by the developmental process itself, which “couldn’t wait forever” for the “expected” stimuli.

    I’m quite sure that if you tried to raise mammals in zero-G, it would play holy hell with their skeletal development, if not kill them in utero with biochemical disruptions. I’m not too surprised that a two-legged goat could “dig up” a few alternate developmental pathways triggered by the muscle and bone stresses of walking on hind legs. If you think about it, it’s hardly more remarkable than the point that this subsentient quadruped was able to learn to walk on two legs in the first place!

  13. #13 Anonymous
    January 20, 2006

    Hah, here’s the photo I was thinking of. Gerenuk is another name for Dik-Diks, Pudus are adorable little deer about the size of a Chilhuahua. (sp?)

  14. #14 Kagehi
    January 21, 2006

    I don’t get it. Out of 200 conceptions, you say you had (8×2+1)/9×200 ~ 378 fetuses. About 1/4×200 ~ 50 births happened, of which ~ 0 was twin. So the nonviability of fetuses ought to be (378-50)/378 or 87 %, much higher than 4.5 %.

    Its **impossible** for you to have more viablity than the already determined fail rate of 75%. If only 1 in 4 pregnancies produce offspring, its impossible for you to end up with an 87% viability rate. You only get that if you devide by the 50 you already know are successes, not the original 200 fertilizations, of which only 25% actually implant.

    I did say hoever that I might have gotten the math wrong, so I rechecked my figures. Actually (rounding), since 44 cases might have been a twin, while 6 wouldn’t have (according the the estimate), that is 21 failed case where the twin wasn’t born, plus one that was, plus the 6 that never where twins, or 28 in 200, which is 14%. Yes, that is a lot higher than my original figure, but yours is still just completely nuts, since I think you forgot to account for the 150 that never survived to be born in the first place. Or in other words:

    X = Number of successful pregnancies = (200 * 25%) = 50
    Y = Number of possible twins = (X * 8 / 9)
    Z = Number of non-twins = X – (X * 8 / 9)
    98% = Number of failed twins.
    Total births = Y/2 + (Y – (Y * .98)) + Z = 28.66667

    This time I think I have the math right. If the number of successful twin births they quoted was 1%, not 2%, then this is 14.1% successes, if it was something I got it wrong the other direction, like 4%, then its still only 14.7% viability. Even if I really screwed up and it was 40% of twins, that is still *not* your figure, its more like 22.7778%, which still falls below the 25% total success rate assumed from merely pregnancies vs. actual births. If however *all* births start as twins and some are just not detectable as such, i.e. 9-9, not 8-9, then the math gets much easier, just X/2 + X * ‘% of living twins’. For 1% that is 25.25/200 or 12.625%. You can do the math for 2%, 4% or 40%. It is obviously going to be lower than if some births never started that way in the first place. While this is close to 3 times better than I miscalculated initially, it is still hardly impressive.

  15. #15 Rubberband
    January 21, 2006

    Dave Harmon-
    I really don’t know much about atavistic genes–(do you have any recommended reading?)–but what keeps them from being effectively destroyed? I mean, mutations happen, but with the expressed genes doesn’t natural selection keep them “healthy?” Lacking expression, and thus selection, it seems sequences would degrade over time. I suppose the same logic might apply to recessive genes, but I’m assuming atavistic genes are expressed MUCH less frequently, because they are only triggered when development is significantly altered.

    Hmmm, mutation allowed to proceed for long periods without environmental “trimming,” then followed by expression, what might be the evolutionary implications? (Especially if these genes were at one point quite useful) Combine with myriad possible developmental pathways . . . my head hurts. It seems any damn thing might happen.

  16. #16 David Harmon
    January 22, 2006

    Rubberband: you’re getting the idea. Yes, unused, duplicated, or otherwise “mooted” genes do “drift” over time, but that tends to be fairly slow — indeed, slow enough to use as an “evolutionary clock”. Often they *do* get more-or-less destroyed, but if they’ve already been duplicated (likely), the damaged copies can just hang around indefinitely. (These relics are often called “pseudogenes”.)

    Both gene duplication and redundant developmental pathways give developmental/genetic networks a good deal of redundancy, so they can degrade gracefully. Also consider that in evolutionary terms, an early lethal counts as a “graceful exit” — if there’s enough survivors to propagate, the loss of (say) one in four embryos, is not a huge problem.

    In any case, if enough carriers of the mutation survive, the damaged gene can continue to mutate, possibly becoming a new, “differently useful” gene. A stray tendon turns out to enable a new posture; a variant digestive enzyme allows eating something new; a changed pigment regulator produces a new fur pattern….

  17. #17 Anonymous
    January 22, 2006

    Of course internal organs are variable and this variability allows Roman augurs to accurately predict battle outcomes. The form and colour of the liver is most significant. An algoritm as good as any.

  18. #18 Kagehi
    January 23, 2006

    You count twins as a single individual in one place (‘Y’) and as two in another (‘Total births’), which I guess explain why you first say you have 50 succesful pregnancies, ie 50-51 babies, and then end up saying you have 29 births, ie 29-30 babies.

    No, I double the number of “possibles”,, with the expectation that a pecentage of the 50 possible births would be failures, to more easilly do the math. Trying to work it out using 100 and 25 seemed more complicated, since that only include successful pregnacies, not “all births from those pregnancies”. Logically, some of those births will produce twins, so the actually number of possible children is 50 in 200, not 25 in 100. Thinking about it, maybe I needed to adjust the 200 and 50 “first” to account for the 8-9 ratio, then run the calculations, but I think I subconsciously kept thinking about the 1-1 twin ratio, where all pregnancies start as twins, with some non-detectables making it appear 8-9.

    And yeah, I guess I did misread you. Sorry, I misread the sentence. Still, your number didn’t end up that far off from mine in the end. In any case, the point is still the same, without maliability in the geneset, the actual birth rate would be astronomically small, so having different patterns produce the same proteins is all that keeps the non-viability from being more like 1:1,000,000,000 or some similarly insane number.

  19. #19 Greg Downey
    February 22, 2006

    Dear PZ,
    Thanks for posting the essay, which is obviously drawing a lot of comments. I’m similarly interested in phenotypic plasticity and working on a book on how sports and other forms of extreme human performance demonstrate some of the range of human anatomical, perceptual, and neurological plasticity, inspired by a lot of the work in DST (dynamic or developmental systems theory) and folks like Esther Thelen, West-Eberhard, Ann Fausto-Sterling, and others.

    Another case that I can NOT find any interesting information on that might be intriguing was a young woman who was born without any functional hands and had learned to do most everyday activities with her feet. From watching a documentary about her, it seemed pretty clear that her feet and toes were extraordinarily adept and dextrous. I’m wondering if anyone out there knows of any research on this type of unusual development, either compensatory or due to other forms of training, that might drive anatomical unfolding into unusual channels. I’ve been looking for material on anatomical studies of elite athletes, circus performers, and others and finding that it’s very difficult to find any more information.

    My goal, obviously, in this sort of enquiry is not to treat this young differently-abled woman as odd, but actually to demonstrate how she shows us just what humans are capable of, like Olympians or other highly-trained individuals.

    If anyone can get in touch with me, please don’t hesitate to email me at I’ve moved to a new institution (university, not holding facility), but this email address should still be good for a few months.

    Many thanks,
    Greg Downey