Wonder Science spoke with Richard Prum about the mechanisms underlying feather patterning on the Argus Pheasant. Prum is the William Robertson Coe Professor of Ornithology at Yale University, and Curator of Ornithology and Head Curator of Vertebrate Zoology in the Yale Peabody Museum of Natural History.
The identity of a cell at a random place in the vein of a complex feather has to be extremely complicated. It has to be very coordinated. It has to know in some way where it is in the feather and therefore its decisions will affect the shape and size of an object which is much larger than itself. And what that means is that for that to occur you have to have a series of cells in different places in time that make decisions in a coordinated way. The Argus Pheasant secondary feathers are among the most complicated pigmentation patterns found in any bird.
What’s causing the colors in the feathers in the first place, chemically? We know from other species that those browns, blacks, and auburn or golden yellow colors are all created by melanin — various kinds of melanin. It’s actually the same kinds that we see in human hair and human skin. There are eumelanins which make black pigments or their black pigments to make black colors like black hair and there’s also pheomelanin a reddish melanin which is responsible for human red hair and blonde hair.
The pattern in the feather is actually a quite subtle and complicated pattern of the concentration of various kinds of melanin in the cells of the feather.
What’s interesting is that the feather itself, like hair, is dead when it functions. It’s growing and then it’s sticking out of the skin but it’s dead. So all the events in development are occurring right at the moment that the feather’s forming and growing out of the skin.
So the question how do the pigments get into the feather cells? The pigments are made by a class of cells called melanocytes. They are melanin-producing cells. The melanin-producing cells apparently make either eumelanins or pheomelanins.
Apparently the melanocytes don’t make both of them at the same time. So we have one class of cells making one pigment, and one class of cells making the other. It turns out the pigment isn’t soluble in the cell. It’s not just sort of soaking in. The melanocyte actually makes a package of pigment. It’s called a melanosome or a melanin body.
It’s a tiny microscopic, it’s actually bacteria-sized, organelle of the cell. And they package the melanin in there and then hand it over to the keratinocyte, the keratin-producing feather cell. So it’s like a little package of light absorbing chemicals packed together. In order to get a complex coloration pattern in a feather, melanin has to be deposited in some of the keratin-producing cells of the feather but not in others.
It turns out that this is about cell-cell communication decisions during development.
So the melanocytes, the melanin-producing cells, migrate up into the developing feather. And they send out pseudopodia, fine filaments, that interact and touch very close to all of the developing cells of the feather and in some ways that we don’t know, the two cells interact, communicate with each other. And some of the cells say, ‘Hit me, give me some melanin.’ And those melanosomes are transferred from the melanocyte to the keratinocyte. Other cells say, ‘No, I don’t want any.’ And so it doesn’t. These are biological developmental decisions that are being made during development.
When you have multiple kinds of melanin, like pheomelanin and eumelanin, then it’s even finer. It’s not a matter of no and yes, it’s a matter of no or yes a little bit or a little bit more. Or this other kind of melanin versus this first kind. Give me some red and give me some black and just a little bit of a different quantities. And so that’s really complicated.
Over a decade ago, we started getting interested in this question when we were working on how feathers grow. And in that we decided to apply some classic mathematical models that had first been described by Alan Turing back in the 50s. And Turing was asking a fundamental question about how you get order in biology from homogeneity. If you imagine an embryo being a cell that is completely homogeneous.
How do we get differentiated bodies? You know the fact that we’ve got skin and hair and livers and hearts and eyes in the right places. How do you get differentiated bodies from uniformity? And he was really interested in the fundamentals of what we call pattern formation. He had some fantastically interesting ideas that were entirely ignored mostly because the math was complicated and there were no computers that would allow you to solve it or to apply it.
The whole idea was rediscovered in the 1970s and then has been applied across animals and plants to all sorts of questions from zebra stripes to leaf shapes and flower shapes and all sorts of questions in developmental biology. So we decided to do that and to study various kinds of feather patterns. We started way simpler with spots and dots and stripes.
We can think about this in the same way as you might think about weaving a fabric. We weave a fabric from threads that go in one direction and then other threads that go in other directions. And we can automatically see that if we make a material that way certain patterns are really easy. So if you vary the vertical threads in color between two colors and vary the horizontal threads between two colors all of a sudden immediately you have check and plaid. So we’d have to say that plaid is easy to make, relatively easy to make given the nature of fabric. Paisley, really hard to weave. And so you don’t use weaving to make complex patterns. It’s an impossible pattern.
So we wanted to know what are the possible and impossible patterns that we could make with feathers? What are the ones that are easy? And we were successfully able to show that Turing pattern, Turing mathematics, would simulate feather patterns that were exactly like lots of living bird patterns. We were able to do stripes and chevrons and spots and dots and arrays of dots. The Turing models basically implied that within this homogeneous embryo or homogeneous field that there were two kinds of chemicals. One he called the “activator” and the other he called the “inhibitor”.
And the activator has two consequences it upregulates itself. That is it makes more of itself or promotes the production of more activator. And it also promotes the production of an inhibitor of another molecule. The inhibitor on the other hand sends negative feedback or inhibits the production of the activator. So we have both autocatalysis, meaning the production of more of the activator, and the activator producing another thing which inhibits the activator. And because the activator and inhibitor work at different size scales, that is, the activator is local and the inhibitor is broad scale, you can end up with localized peaks of the activator where it runs away and produce much much much of itself but that’s going to be restricted to a small portion, a small point, because the inhibitor is broadly suppressing the production of the activator over a broader scale. Eventually that inhibition will wear off and at a certain distance you have an opportunity for another activator peak to arise.
This is the way you can produce stable spots or stable stripes if it’s propagating in one direction. And at the heart of this observation is that pattern formation almost always requires regardless of the math requires some kind of activation and inhibition some kind of positive and negative feedback.
The way in which we would get spots and dots and stripes over time depends on what’s going on at that growing edge. Let’s imagine that we create a periodic pattern like horizontal stripes. One way to do that is to have a wave of change saying, we’re accepting pigment now, but then later on you inhibit later cells, subsequent cells, cells that are younger than you are suddenly say no no no we’re not going to take it.
And then suddenly cells that are younger than they are when they grow up they say, oh no no we’re going to take it. So you end up with a wave going down the feather as it grows. A wave in time. Then you can if you superimpose on that a wave in space that is there a certain horizontal positions or rotary positions, radial positions in the tube where you turn it on, you end up with an array of dots with the interaction of a temporal wave and a spatial wave. Those two things together can give you an array of dots.
Those patterns in and of themselves are not sufficient to give you the complexity of an Argus Pheasant because you have really many more dimensions of complexity.
If you read Darwin’s description of the Argus Pheasant feathers in “Descent of Man,” it’s really amazing because he talks about the variation across the feather. Not just the golden balls in their socket but all the other, and they’re amazing. And he really actually starts to use the languages of waves. He talks about the patterns as a wave and then the spots are kind of like start as an eddy in the wave and then the eddies become even more localized and rotating and then they become discrete spots.
He’s really trying to come up with a scenario for how you would get the peacock eye through a series of increasingly more complicated wave like patterns of pigmentation pattern on a feather. It was a super creative and really interesting passage and indeed the differential equations that are part of the Turing model are expressions of waves. They really do have this property of waves so in some way, Darwin understood that the math or that the pattern formation mechanism must include a regard for waves in time and waves in space.
And that’s exactly what Turing models are about.
And so there’s a class of models, the models we use, imagines that the pattern is being developed like as if it were being woven at the edge. Or as if it’s a forest fire proceeding through a field. That means it has a growing edge. There’s another class of models where you can imagine a fixed volume or fixed area. Imagine a square or a football field and then you imagine that chemical reactions are taking place within this domain in two dimensions and then ultimately they reach some stable state or some predictable state.
And these are two classes of the models.
But I can tell you that we failed to be able to produce any patterns that are as complicated as the Argus Pheasant. And I think I have some mathematical ideas about why we failed having to do with how the math is solvable. Obviously the Argus Pheasant is solving this. It’s not worried about dividing by zero or any other kind of mathematical challenges. It just does it. But currently we don’t actually have an appropriate theory to describe the development of the patterns of the Argus Pheasant.
It seems to be in a mathematically more complicated class than the kind of Turing patterns that we’ve developed for feathers or for zebra stripes or other similar kinds of patterns. The problem is that with the growing edge models, the woven models, you cannot produce the kind of complicated patterns that Argus Pheasant has. You need to have a 2D model. But the 2D models also have constraints. So that where we need to go I think mathematically to produce a model of the Argus Pheasant is a 2D model that has a growing edge and the senescing or aging edge.
And this produces problems with the mathematics that I haven’t yet solved. I’ve recruited a couple of students to say, hey — math students who are really into this — and say, ‘hey, we want to try this, and no no solutions. And I haven’t recruited other math talent with the math chops to be able to produce these new kinds of models. So I think we need a new class of models and mathematical methods to solve the challenge of simulating an Argus Pheasant.