The concept of coherence is implicit in the definition of irreducible complexity in the idea of parts that are “well matched” to a “system.” The standard mechanical mousetrap, with its very well-matched parts, is profoundly coherent. Since it is irreducibly complex, it can’t be built directly by a gradual process that would mimic a Darwinian scenario. But suppose there were some tortuous, indirect route that might lead to the trap. It would be unreasonable to expect the route to be found by a blind process, for the same reason that we wouldn’t expect the blind knight to maneuver through a maze to the summit of the castle—there are too many dead ends and opportunities to go wrong. To trap mice, a deep hole in the ground might do just fine. Yet a hole in the ground isn’t a route to the standard mechanical mousetrap. If the hole then had to be filled in before starting over to build a better mousetrap (pardon the strained analogy), then mice would flourish at least temporarily—ruling out this path. A splotch of glue can catch a mouse, but can’t be turned into a mechanical trap. If the glue trap had to be discarded before starting over to make a mechanical trap, we’d be worse off than before.
The more pieces, and the more intricately they interact, the more opportunities there are to go wrong in building a system. Even with small systems, one can go wrong right off the bat, get trapped in a dead end, and never make it to the top of the castle. If Herman Melville had started writing his novel with “Call me, Ishmael,” Moby Dick would have gotten off on the wrong foot and might never have been written. The benchmarks discussed in this chapter are thus a better guide to the edge of evolution than is irreducible complexity. Now, it’s time to get down to business, using these benchmarks to try to define the ragged edge of Darwinian evolution.
7
THE TWO-BINDING-SITES RULE
A GOOD FIT IS HARD TO FIND
In Darwin’s Black Box I explained how design can be apprehended in the arrangement of parts. It can also be perceived in the fitting of complex parts, even if the reason for the fitting is obscure. Here’s one example. Suppose that you were walking through a junkyard. All sorts of complex parts were lying around: a pipe to your left, springs to your right, bolts, screws, pieces of metal, and much more. Although the yard was filled with many manufactured parts, there would be no reason to think they had anything to do with each other. However, suppose you spotted a compact pile of parts. When you picked up one of the parts, the rest came along—they were attached. What’s more, you saw that the parts matched closely—holes in one piece were aligned to pegs in another; curves fit with indentations, and so on. Even if you didn’t know the function of the aggregate of pieces, you would be pretty sure they had been put together on purpose, unlike the other parts in the junkyard, because they specifically fit each other.
In order to be sure that parts are designed to fit each other, their shapes must be relatively complex and must match each other pretty closely. If the shapes of pieces are comparatively simple, or if the fit is quite loose, their complementarity may just be a matter of luck. For example, it’s hard to decide if, say, a generic book was intended to be mailed in a generic box. The rectangular shapes of both a book and a box are pretty simple, so even if the book fits loosely in the box, it might be a coincidence. On the other hand, it’s easier to conclude that a box with exactly the right-sized compartments was built to ship a particular computer and accessories if all the pieces fit snugly together, with projections and indentations of the computer nicely accommodated by complementary indentations and projections of the box (or, say, Styrofoam packing material). The greater complexity of the second box allows for a firmer decision of purposeful design.
In judging whether two unfamiliar parts were designed to fit each other, one has to be careful. The likelihood of finding two parts that fit by accident has to be weighed against the number of different parts that are on hand. If a warehouse were filled with a million small, rigid, plastic pieces of all different shapes, it wouldn’t be too surprising to find one that, say, fit pretty well inside a heart-shaped locket. The more and more pieces we have on hand, the more and more likely we are to find two that fit each other closely, just by chance. On the other hand, the more and more complex the shapes, the less likely.
PROTEIN PIECES
Proteins have complex shapes, and proteins must fit specifically with other proteins to make the molecular machinery of the cell. Although most proteins were once thought to act individually, in the past few decades science has unexpectedly discovered that most proteins in the cell actually work as teams of a half dozen or more. As former president of the National Academy of Sciences Bruce Alberts remarked:
We can walk and we can talk because the chemistry that makes life possible is much more elaborate and sophisticated than anything we students had ever considered…. [I]nstead of a cell dominated by randomly colliding individual protein molecules, we now know that nearly every major process in a cell is carried out by assemblies of 10 or more protein molecules. And, as it carries out its biological functions, each of these protein assemblies interacts with several other large complexes of proteins. Indeed, the entire cell can be viewed as a factory that contains an elaborate network of interlocking assembly lines, each of which is composed of a set of large protein machines.1
FIGURE 7.1
Cartoon of a smaller protein binding to a larger one. Both the shapes and the chemical properties of the protein surfaces must be complementary to bind. (+ and - stand for positively and negatively charged groups on the protein surfaces. B stands for hydrophobic, oily groups. OH and O stand for polar groups that can “hydrogen bond” to each other.)
What’s more, in contrast to the machines in our everyday world, proteins must self-assemble. Although machine parts in our familiar world must have complementary shapes, they are put together by people (or robot surrogates). Protein parts in cellular machines not only have to match their partners, they have to go much further and assemble themselves—a very tricky business indeed. As a recent issue of Nature put it:
The cell’s macromolecular machines contain dozens or even hundreds of components. But unlike man made machines, which are built on assembly lines, these cellular machines assemble spontaneously from their protein and nucleic-acid components. It is as though cars could be manufactured by merely tumbling their parts onto the factory floor.2
To perform that astounding feat, proteins have to pick their correct binding partners out from the many thousands of other proteins in the cell:
A protein generally resides in a crowded environment with many potential binding partners with different surface properties. Most proteins are very specific in their choice of partner, although some are multispecific, having multiple (competing) binding partners on coinciding or overlapping interfaces.3
In order to assemble correctly, the absolute minimum requirement is that proteins must stick specifically to their partners in the right orientation. As shown in Figure 7.1, not only do the shapes of two proteins have to match, but the chemical properties of their surfaces must be complementary as well, to attract each other. If the shapes of two protein surfaces match each other but their chemical properties don’t, the two surfaces won’t stick; they might bump together in the cell, but if so they would quickly drift apart.
Were complex protein-protein interactions designed, like so many complementary automobile parts? Or are they merely accidental, more like the heart-shaped piece from a large warehouse that fits into a locket? Or are there some in each category—some interactions arising by serendipity, others through intent? Over the next several sections we’ll look closely at how proteins choose a partner and consider the large obstacles that the evolution of new protein-protein interactions presents to random mutation. We’ll see that it is reasonable to conclude that, although some interactions are accidental, the great majority of functional protein interactions arose nonrandomly. And that means nonrandomness extends very deeply into the cell.
BINDING EVERYTHING
Much has been learned ab
out how proteins bind to each other by studying the immune system. The complex vertebrate immune system protects against invasion by microscopic predators. One facet of the system generates a prodigious number of different proteins called antibodies that patrol the circulatory system. Although much of their structure is pretty similar, the antibodies differ from one another at one end, called the binding site. If a foreign cell or virus is lurking in the circulatory system, one or more antibodies will likely stick to the invaders via the binding site, marking them for destruction by other immune system components.
In order for an antibody to stick, the binding site must be geometrically and chemically complementary to the foreign surface. In early work on the immune system, before it was realized just how clever the system is, it was natural to think that vertebrate antibodies had been shaped by natural selection to recognize the surfaces of previously encountered pathogens, so that the only antibodies expected to be available would be ones that had binding sites to past or present invaders. However, subsequent experiments showed that when test animals were injected with synthetic chemicals that had likely never before existed on earth, some existing antibodies were able to bind to the manmade materials. That meant that some binding sites were complementary to shapes that the animal or its ancestors had never encountered! How could that be?
A number of ideas were floated that turned out to be wrong. One proposal, advanced by the two-time Nobel Prize winner Linus Pauling, was that antibody binding sites “mold” themselves to the foreign chemical—wrap around it and “freeze” in position.4 Another was that perhaps the vertebrate genome carried genes for a huge number of antibodies, and by luck the shapes of some of their binding sites matched the shapes of molecules that hadn’t even been invented by nature. That was closer to the mark, but it greatly underestimated the elegance of the design of the immune system. It eventually became clear that, while vertebrates do make a huge number of different antibodies (many billions), they have a limited number of antibody genes (a few hundred). Much hard lab work eventually showed that the trick to generating many different antibodies from a small set of genes is the same principle that allows a huge number of different poker hands to be dealt from a deck of just fifty-two cards. In brief, immune cells contain specific molecular machinery that shuffles segments of genes (and performs other tricks, too), allowing very many antibodies to be produced with an extraordinary diversity of binding sites.
SHAPE SPACE
Over the past decades immunologists have injected test animals with a wide range of synthetic chemicals, and almost unfailingly the animals prove to have antibodies that counteract them. To explain the ability of the immune system to do this, the mathematician Alan Perelson invented the concept of “shape space.”5 Shape space envisions a sort of library of physical objects roughly the size of a protein-protein binding site, with indentations and projections, smooth surfaces and rough ones. Perelson calculated that the huge number of different antibody binding sites made by the immune system was enough to contain essentially every possible shape, so that there would be at least one antibody to bind reasonably well to proteins on the surface of an invading bacterium or virus, even if the body had never been exposed to it before. To revert to my analogy of a warehouse full of plastic shapes, this shows that the binding matches between some antibody and some synthetic chemical were not specifically designed (although the clever, special mechanisms to enable the immune system to produce all those antibodies very likely were). The universe of antibodies is for all practical purposes infinite, which precludes an inference of design for these instances of binding. The knowledge gained from studying the immune system is invaluable for assessing the edge of evolution elsewhere. Is evolution full of processes analogous to our immune system, with automatic shufflings that make all things possible?
Like something from a science fiction story, the objects in Perelson’s protein shape space are multidimensional—they have more than the usual three dimensions found in our ordinary world. They also have “dimensions” that take into account their varying chemical properties—positive charges, oily areas, and more. Imagine a huge library of complexly shaped objects that also had a half dozen or so weak bar magnets inserted at irregular spots, with just the tip of each magnet sticking out on the surface. The magnets could be oriented with their north poles facing either in or out, so that the magnet tips on the surface could either attract or repel each other. The extra dimensions of shape space can be thought of as accounting for the placement and orientation of the magnets. So shape space accounts not only for the physical shapes of objects, but also for their ability to stick to each other.
Suppose that thousands of objects from the shape space library are placed in a well-stirred swimming pool, and a prime goal for each is to find its one ideal mate. Reaching that goal faces a “Goldilocks” problem. First, consider objects that bind indiscriminately. Suppose some shape-space objects weren’t very rigid—they had flexible octopus arms lined with magnets that allowed them to stick to many other objects in the swimming pool. Although they’d bind strongly, those stick-to-everything objects would gum up the works. The cell cannot tolerate objects that bind haphazardly.6 They must be eliminated. Next, think about objects that are rigid, but don’t match. Suppose a pair of objects do have complementary shapes, but they don’t have magnets lined up in the right positions. Although, by rotating around, one or two pairs of magnets might be brought close to each other, that’s not nearly enough to hold them together in the swirling waters of the pool, so they immediately fall apart. The two lessons are: 1) Nonspecific, octopuslike objects that bind strongly to many other objects are hazardous and must be removed; and 2) in order to stick specifically and well, two rigid objects have to fit each other in both shape and magnet pattern.
Binding doesn’t have to be all or nothing. A pair of rigid objects might have shapes that are pretty complementary, and magnets that are fairly well aligned. They might stick pretty well, but every so often a strong current in the pool or unusually big collision knocks them apart. After stirring for a while they reconnect. On average they might spend about half their time together and half apart. On the other hand, another pair of objects might fit together like a hand in a glove, with all magnets aligned perfectly. They might stick so tightly that they rarely fall apart and spend about 99 percent of their time together.
FIGURE 7.2
Cartoon of objects from a shape-space library. N and S stand for the north and south poles of magnets. (A) The floppy, octopuslike shape would bind nonspecifically to many objects. (B) These rigid objects are complementary in magnet pattern, but not in shape, so they would not bind well to each other. (C) These rigid objects are complementary in shape, but not in magnet pattern, so they would not bind well to each other either.
HOW BIG IS SHAPE SPACE?
We can adapt the lessons of the immune system and shape space to help understand the problems random mutation would face in making new protein-protein binding sites in the cell. The immune system is set up to get around the problem. But what happens when you remove the manufacturer who threw all these shuffled pieces of plastic into the pool? What if new pieces had to be made by rare dents or scrapes to old pieces? The immune system is capable of producing an essentially infinite warehouse of little plastic shapes. Experiments with similar systems in laboratories have enabled scientists to test finite subsets of that warehouse, to see how big it must be in order to handle particular challenges. These experiments yield generalizable parameters for the limits of binding proteins in any context, not just immune systems.
A huge hurdle confronting Darwinian evolution is the following: Most proteins in the cell operate as specific complexes of a half dozen or more chains.7 Hemoglobin comprises a complex of two kinds of amino acid chains (alpha and beta) stuck together, but hemoglobin is relatively simple. Most cellular proteins have six or more kinds of amino acid chains. So, unless those complexes were all together from the start, then at some time in the past
separate cellular proteins had to develop the ability to bind to each other. But that would be a very tricky business indeed. On the one hand if, like the octopus objects above, a protein developed a surface that stuck indiscriminately to a lot of other proteins, it would gum up the workings of the cell. It would have to be eliminated. On the other hand, most protein pairs wouldn’t bind to each other at all, or bind very weakly, because their surfaces don’t match closely enough. Only when a Goldilocks match randomly developed between their multidimensional surfaces would two proteins bind to each other tightly and specifically enough to make an effective pair. So we can ask, how difficult would it be for two proteins that initially did not bind to each other to develop a strong, specific interaction by random mutation and natural selection?
To start to answer that question, let’s take the measure of shape space. Is shape space small, medium, or huge? How many protein binding sites do we need in our shape-space library to find a decent match for another given protein? In the 1990s Greg Winter and coworkers at the Medical Research Council in Cambridge, England, were interested in developing artificial antibodies to use as tools in medical research and treatment. In a series of papers they focused on the question of how large a shape-space library would be needed to have a good chance of containing at least one antibody with a binding site that would stick pretty specifically to an arbitrary test protein. Using clever laboratory methods, they made antibodies in which the amino acids of the binding site had been randomized, and generated a very large number of different combinations—a hundred million, in fact.
The Edge of Evolution Page 14