Fighting Colds with Natural Selection

As a concept, natural selection has come a long way this century. In 1900 many scientists wondered if it was significant, or even real. By 2000, it was possible to witness natural selection reshaping life and helping to carve out new species. The resurgence of natural selection this century has even led scientists to discover it at work in some unexpected places. Anything that fulfills Darwin’s three basic requirements–replication, variation, and rewards through competition–may feel natural selection’s power.

Our bodies, for example, fight off diseases with an immunological version of natural selection. When a virus or some other parasite invades, your immune system tries to mount an attack. But in order to fight off the invader, the immune system has to be able to recognize its opponent. Otherwise it will attack anything it encounters, including the body itself. The immune system harnesses the power of evolution to fine‑tune its assault.

When foreign substances enter the body, they encounter a class of immune cells called B cells. B cells have receptors that can snag foreign substances–a toxin made by bacteria, for instance, or a fragment of a virus’s protein coat. When a B cell snags these substances (known as antigens), a signal travels into its interior, causing it to multiply into millions of new cells.

The new cells start spewing out antibodies, which are free‑floating versions of the receptor that snagged the antigen in the first place. The antibodies course through the body, and when they encounter their antigen, they can lock on to it. While one end of an antibody grabs an antigen, the other end gets rid of it. It may neutralize a toxin, drill holes in the walls of bacteria, or catch the attention of the immune system’s assassin cells, which can swallow a parasite.

B cells can produce antibodies that precisely match any one of billions of antigens, produced by parasites ranging from viruses to fungi to hookworms. Their precision allows the immune system to recognize and destroy only these invaders, without also latching on to the cells of our own bodies and destroying them as well. Yet our DNA does not contain instructions for building an antibody for every antigen our B cells might encounter. There are billions of antigens, but only about 25,000 genes in human DNA. Our immune system uses a different, far more efficient way to create antibodies: our B cells evolve.

This evolution begins when B cells first form deep inside our bone marrow. As they divide, the genes that build their antigen receptors mutate rapidly, randomly creating billions of receptors with different shapes. They thus take the first step of any evolutionary process: the generation of variation.

The young B cells creep from the bone marrow to the lymph nodes, where many antigens are drifting by. Most of the B cells will be unable to latch on to the antigens, but on rare occasions, a B cell will have the right receptor to grab one. The fit doesn’t have to be perfect; if a B cell manages to grab anything at all, it is stimulated to start multiplying madly. You can tell when a B cell has gotten lucky this way: as it proliferates, the lymph node swells into a lump.

Some of the successful B cell’s descendants immediately start releasing antibodies with the same structure as their antigen‑grabbing receptors. But others continue dividing without making antibodies. These B cells begin to multiply, mutating more than a million times faster than normal human cells. The mutations alter only the genes they use to build their antigen receptors and their antibodies. In order to survive, these hypermutating B cells must grab an antigen. If a cell fails, it dies. After successive rounds of mutations and competition, evolution produces B cells that can lock on to antigens more and more precisely. Less‑adapted cells fail to grab antigens and die. Within a matter of days, this evolutionary process can improve the antigen‑snagging ability of B cells by 10 to 50 times.

Imagine if Paley had been introduced to antibodies, so well designed for fighting particular diseases. He might have said that antibodies must be the work of a designer, that nothing so well crafted, so well suited to its own antigen, could have been formed on its own. And yet every time we get sick, he is proved wrong.

Evolution in silico

The powers of natural selection can be seen not only in our own bodies but in computers. Life as we commonly know it is written in only one language: the bases of DNA and RNA. But some scientists have been creating what they claim are artificial life‑forms in computers that have no need for biochemistry. And like DNA‑based life, they can evolve. While critics may question just how alive these creations really are, they nevertheless show how mutations combined with natural selection can turn randomness into complexity. They are even showing how natural selection can create new kinds of technology.

One of the most complex forms of artificial life can be found dwelling in the computers at the California Institute of Technology. There, Christoph Adami, Charles Ofria, and other scientists have created a wildlife refuge they call Avida (A stands for artificial, and vida is the Spanish word for life). The organisms that live in Avida each consist of a program made up of a series of commands. During the organism’s life, an “instruction pointer” moves line by line through the program, carrying out each command that it reads, until it gets to the end, whereupon it automatically returns to the beginning and the program repeats itself.

A digital organism’s program can create a copy of itself that becomes a self‑sustaining organism of its own, multiplying until it takes up all the computer space that Avida can spare. And by allowing the programs of these digital organisms to mutate as they reproduce, Adami can make them evolve. On rare occasions, a line in a digital organism’s program may spontaneously switch to another command; sometimes when an organism tries to copy itself, it may accidentally misread a command and put in the wrong one; it may insert an extra random command or erase one. Just as mutations are usually harmful to biological creatures, most changes to the programs in Avida are harmful bugs that slow down an organism or kill it. But sometimes a mutation can make a digital organism reproduce faster.

Adami can set up experiments with Avida that mimic the evolution of biological organisms. In one early experiment, Adami created a digital organism that could replicate, but which also carried several useless (but harmless) commands in its program. This progenitor gave rise to millions of descendants, which diverged into mutant strains. Within a few thousand generations some of the strains became far more common and successful than the others. What the successful digital organisms shared in common was a short program. In each case, mutations had stripped them down to the simplest program that could still replicate–about 11 steps long.

Evolution drives the digital organisms toward simple genomes in this experiment because they live in a simple environment. In more recent experiments, Adami has made Avida more like the real world, by requiring his digital organisms to eat. In Avida, numbers are food–an infinite string of 1s and 0s that digital organisms digest and turn into new forms. Just as bacteria can eat sugar and transform it into the proteins they need to survive, a digital organism with the proper commands in its program can read the numbers Adami supplies it with and transform them into various forms.

In the natural world, evolution favors organisms that can turn their food into proteins that help them reproduce more successfully. Adami created a similar system of rewards for his digital organisms in Avida. He set up a list of tasks for the organisms to perform, such as reading a number and transforming it into its opposite, so that 10101 becomes 01010. If an organism evolves the ability to do this, he rewards it by speeding up the rate at which its program runs. With a faster‑running program, a digital organism can replicate faster. And the rewards for carrying out more complex operations are bigger than for carrying out simpler ones. This reward system radically changes the direction of evolution in Avida. Instead of becoming stripped‑down virus‑like organisms, the organisms evolve into sophisticated data processors.

Avida is essentially evolving new pieces of software, although they are unlike any program written by a human being. The alien structure of Avida’s programs has attracted the attention of Microsoft, which has funded some of Adami’s research. They recognize that our DNA is in some ways like an extraordinary computer program, but it can keep a trillion‑cell human body going for 70 years without crashing. There seems to be something about the way that evolution produces information processing that makes it more robust than human creations. Microsoft would like to know whether they might someday be able to evolve software rather than write it. The programs that evolve inside Avida today are as simple compared to a spreadsheet as bacteria are to a blue whale. Yet evolution has produced blue whales, and it’s conceivable that in an artificial world like Avida, it might be able to produce spreadsheets as well. The challenge will be to landscape the evolutionary hills and valleys of artificial life in the right way, so that a spreadsheet design represents the highest peak of fitness.

Avida is part of an infant science, known as evolutionary computing. Its disciples are discovering that natural selection can shape not only software but hardware as well. A computer can be challenged to come up with thousands of different designs for a device, which it can then test in a simulation. The ones that run best are saved, and then randomly altered in small ways to create a new generation of designs. Without any more guidance than this, computers can evolve some extraordinary inventions.

In 1995, for instance, the engineer John Koza used evolutionary computing to design a low‑pass filter, a device that can cut off sounds above a certain frequency. Koza chose 2,000 cycles a second as his cutoff. After 10 generations, his computer produced a circuit that muffled frequencies above about 500 cycles and only completely extinguished them above about 10,000. After 49 generations, it had created a circuit that produced a sharp drop‑off at 2,000 cycles. Natural selection had created a design for a seven‑rung ladder made out of inductors and capacitors. The same design had been invented in 1917 by George Campbell of AT&T. The computer, without any direction from Koza, had infringed on a patent.

Since then, Koza and others have evolved thermometers, amplifiers complete with woofers and tweeters, circuits that control robots, and dozens of other devices, many of which replicate the work of great inventors. It won’t be long, they predict, before evolutionary computing will create devices that warrant patents of their own.

For the moment, this kind of evolution remains trapped inside computers, dependent on human programmers and engineers for its existence. But within a few decades, independent robots may be able to evolve on their own, transforming themselves into new forms that humans could never imagine. In a sign of things to come, Hod Lipson and Jordan Pollack, two engineers at Brandeis University in Massachusetts, announced in August 2000 that they had programmed a computer to use evolution to design a walking robot.

Lipson and Pollack’s computer evolved 200 robot designs, each starting completely from scratch. Using a simulation program, Lipson and Pollack scored the robots by how fast they could move across the floor, replaced low‑fitness robot designs with ones of higher fitness, and mutated all the remaining robots again. After several hundred generations, the computer then built some of the most successful robots out of molded plastic. These evolved robots walk like inchworms, crabs, and other real animals, yet they look unlike any real animal (or, for that matter, the animal‑like robots that humans have built).

The dawn of artificial evolution is a triumph Darwin could not have imagined. Four billion years ago, a new form of matter emerged on this planet: a substance that could store information and replicate itself, that could survive as that information gradually changed. We humans are made of that mutable stuff, but we may now be carrying its laws into new forms, into silicon and plastic, into binary streams of energy.

PART TWO