An animal's parents mate and produce a number of offspring, each with slightly different Instruction Books. After a while developing as an egg or fetus, the animal is born. Some kinds of animals are fed by their parents for a while, but eventually all animals have to get their own food. So, the animal goes out and tries to find plants and/or animals that it can eat. While doing that, it attempts to avoid predators, locate water, and cope with poisons in its domain. After eating and growing for a while, the animal reaches maturity and tries to find a mate. If it gets a mate, it tries to produce children to begin the cycle again.

At any point that this life cycle sketch says "try", an animal has traits that determine how well it will do in that attempt. Similarly, when the sketch says "a while", the length of time that process lasts is a trait. In fact, nearly every word in the sketch can be affected by an animal's traits: an animal can "avoid" predators by blending in with its surroundings, running (or flying or swimming) faster than they do, or having defenses, like spines, a shell, or a bad taste.

Since natural selection is a kind of evolution, it has the two basic steps that all evolutionary processes do: mutation and selection. While animals can't avoid point mutations, they most commonly develop new traits by recombining their Instruction Books and by shuffling exons. This means that they tend to develop new traits by mixing existing Instructions or making new Instructions out of existing exons much more often than by making entirely new Instructions.

Selection, evolution's second step, goes on continuously during an animal's life. As the sketch describes, there are many times when selective pressures act on an animal. In fact, they start working on it even before it's born: if the animal inherited Instruction Books with particularly bad mutations, it may be stillborn. Other mutations may give it traits so bad that it dies soon after birth. Additionally, some kinds of selective pressures, like drought, may simply kill it off regardless of its traits. Accidents, such as being hit by a falling tree, also kill some animals. In short, there are lots of ways for wild animals to die. Fortunately, all animals try to produce many offspring during their lives. Some, like insects, produce many at once. Others, like elephants or people, usually produce only one at a time, but since they have long lives they might have several offspring during their lifetimes.

Let's look at an example of natural selection: a group of animals moves into a new environment and faces a novel selective pressure there. In real life, animals are generally facing multiple selective pressures all the time, but we've simplified the example to make it clear what's happening. In the example, we'll always talk about populations of animals. A population is just a group whose members change over time. Older ones will die off, and new ones will be born into the population.

Natural Selection in Action

While adapting to poisons is best known among bacteria -- they become resistant to antibiotics -- animals evolve to cope with poisons as well. Koalas eat eucalyptus leaves, which contain a poison that sickens most other animals. Humans, too, tolerate some substances that poison other animals. We make workers that detoxify capsaicin and our alcohol dehydrogenases allow us to drink ethyl alcohol, which is poisonous to many living things.

Although the mice all look pretty much alike:

they have a variety of traits and alleles for tolerating capsaicin, ranging from no tolerance whatsoever to a moderate degree of tolerance. None are particularly good at it. The mice have a two-step pathway for breaking down capsaicin, much like the way we break down alcohol. (In real life, the two workers that are breaking down capsaicin probably do other jobs in the mice. However, calling them "workers that can break down capsaicin but really do other things better" is clumsy, so we'll just call them "capsaicin-detoxifying workers".)

To provide a simple measure of how well the mice are doing in their new home, let's color-code the mice and their Instruction Books on a red-to-green scale to indicate how well they deal with capsaicin. The mice's Instruction Books are color coded to show how well each of the two workers produced by the Instructions can do this job. The mice themselves are color-coded for their capsaicin-tolerance trait, which is produced by the interactions among the two pairs of Instructions found in their two sets of Instruction Books.

Also, while the mice are shown with a just single pair of Instruction Books, each of their millions of cells has its own copies. We're showing just the Books in the mice's germ line, since those are the ones that might get passed on to offspring. Barring point mutations all the copies are the same.

Most of the mice don't have good traits, and so don't cope well. Some (shown in red) become very sick from eating the food in the warehouse. Some may die, and they are very unlikely to produce offspring. A few mice (shown in yellow) can tolerate capsaicin ok. They won't do especially well, but they're very likely to live long enough to produce children. Most of the mice are in between: coping poorly. They'll probably each have a child or two.

Let's see what happens to the mice as they live in the warehouse.

After several generations, the warehouse mouse population has grown a little. The slow growth reflects little selective fitness for the warehouse environment: under ideal conditions, mouse populations can more than double in a single generation. Since the mice with the best traits have each had a few more offspring over the generations, there are more mice with their good alleles (and generally good traits). Still, none of the mice do particularly well in this environment, so all of the alleles and traits are still around. Indeed, many mice have bad alleles, and occasionally one is born with both of the worst alleles.

Even though the mice are not doing well at the moment, their future is not entirely bleak. One mouse in the generation shown had a mutation in one of his capsaicin-detoxifying-worker Instructions that lets it work better. In turn, this better worker gives the mouse a slightly better phenotype, so it can tolerate its food better. With luck, its improved health may increase the chances it will find a mate, and pass on this new Instruction.

Indeed, the mouse with the best allele does find a mate: another mouse that is doing pretty well (shown in yellow near the center of the previous picture). They have six offspring, whose traits and Instructions are shown.

Each child receives a set of Instruction Books from its parents, and these Instruction Books have recombined, so the children's alleles aren't usually the same as either parent's. However, since the two Instructions for breaking down capsaicin are close together in the mice's Instruction Books, recombination during meiosis doesn't always shuffle them randomly. In this case, they're so close that the alleles have a 60% chance to stay together, rather than a 50% chance. This means that a child mouse is more likely than not to inherit an Instruction Book like one of its parents. For example, one of the right-hand parent's Instruction Books contains a pretty good allele for the first capsaicin-detoxifying worker (in yellow-green) and a middling-bad one for the second one (in orange). The 60% linkage between the two of them causes them to be passed on together more often than not, so the middling-bad allele tends to stay in circulation more than would be expected from its selective disadvantage alone. This sort of linkage between alleles is called genetic linkage, and it is one reason that bad alleles persist in populations.

Looking closely at the offspring, we see that most of them have reasonably good capsaicin tolerance, since the new, better allele dominates the older, less-capable ones. We also note that the presence of a single good Instruction does not make a good mouse. The child mouse in the center of the picture inherited one of the best available Instructions, but shows only average overall capsaicin tolerance because neither of its Instructions for the other worker are very good.

Let's let many generations pass before taking another look at the warehouse mice.

The combination of the better allele and the passage of time has allowed the mouse population to start to grow rapidly. And not only has the population grown, but the proportion of mice with the better allele has grown as well. Whereas the better allele was originally found in just a single mouse, then in four of its children, now most of the mice have a copy. Still, a number of worse alleles persist, usually hidden by the better dominant alleles. These worse alleles do sometimes occur together, producing mice that don't do well. For instance, the mouse in the upper right-hand corner doesn't tolerate capsaicin very well, despite the presence of the better allele, since both of the other Instructions for the other worker needed to break down capsaicin aren't very good at all.

Second, a new allele for the second capsaicin-detoxification worker has appeared (in the top center mouse). Exon shuffling combined one of these alleles with a piece of another Instruction, and this new Instruction produces a worker that's better than any of the others. Now, the mouse population has good alleles for both workers needed to break down capsaicin. Let's see what effect this has.

Like most mice in this generation, the mouse with the new allele is able to find a mate and reproduce. Its mate happens to have one of the good alleles for the first capsaicin-detoxifying worker Instruction. Since most mice in this generation have at least one good allele for this Instruction, this isn't unusual. These mice produce five offspring, which get an assortment of their parents' Instructions by recombination.

Two of the children inherited both of the good alleles, giving them an interesting trait -- they're better at coping with capsaicin than either of their parents. These mice have very little problem eating capsaicin-spiced food, and so are likely to grow up healthy and have many offspring. Some of their offspring may be similarly healthy, depending on recombination and their other parent's Instructions. Note, however, that one of these mice, despite having the best-available phenotype, still carries one of the worse alleles, so that allele will survive into another generation.

At this point, the population of warehouse mice is likely to expand rapidly until it meets another selective pressure.

There are three things to note about the example as a whole:

First, not all selective pressures are lethal. While many of the mice with the worst traits did die, most of the bad alleles diminished because mice that had them didn't have as many children as mice that had better ones.

Second, the traits produced by natural selection are not perfect. The capsaicin-detoxifying workers the mice have are pretty good, but not the best possible. While better ones could evolve by future rounds of mutation, they probably won't provide much selective advantage over the current ones, so their Instructions won't necessarily be inherited by a large proportion of future mice. (This happens with human traits, too: despite millennia of walking upright, people still develop back problems.) Good enough suffices for natural selection.

Third, natural selection can't eliminate all the bad alleles from a population. Both genetic linkage of bad alleles to good ones and the masking of bad recessive alleles by better dominant ones tends to keep a few copies of bad alleles circulating in a population.

Another point to keep in mind is that the mutations that produce better traits are random. It's certainly possible that if we were to take ten populations of mice and put them in ten similar new environments, none of them would acquire advantageous mutations -- indeed, the very Instructions they need to adapt might get worse through mutation -- and they'd all die out. This doesn't make an interesting example, though.

The initial population of mice probably had pretty good alleles for their fat storage Instruction. Here's our initial population of mice again, with color added to show their fat-regulator Instructions (in the lower left corner of each of their Instruction Books). The mice are still color-coded for capsaicin tolerance, as are their capsaicin-detoxifier Instructions, but their fat-regulator Instructions are color coded for how good they are in the wild.

Now, let's change the mice's color-code so that it shows how good the mice are at storing fat:

So the initial population generally had good Instructions and good phenotypes for storing fat. This isn't surprising, since mice that didn't make workers to store a lot of fat were likely to be killed off or weakened by famine, and thus not have a lot of children. A few worse fat-regulator alleles are still around, but the better alleles are so common that the mice generally don't have problems storing fat.

Now, we'll skip ahead to the last generation of mice we looked at. Here they are, color-coded to show that they've been selected for good capsaicin tolerance.

However, when we look at them color-coded for fat regulation, we see a different story:

We see that the fraction of good traits is smaller, and the proportion of the good alleles for fat regulation has declined. There are two reasons for this. First and most importantly, the mice's capsaicin tolerance determined which ones would reproduce most, not their fat regulation. So, the mice that had better capsaicin tolerance passed on their fat-regulator Instructions regardless of how much fat they stored. Second, mutations continue in all the mice's Instructions. Regardless of whether a mutation made a fat-regulator Instruction better or worse, it was likely to be passed on as long as the mouse it was in had good capsaicin tolerance. The lack of selective pressure for storing fat let the distribution of the mice's alleles for fat regulation become uniform (just as many bad alleles as good ones) since nothing was removing bad alleles, or encouraging the spread of better ones.

If our population of warehouse mice should return to their original environment where fat-regulation matters, they'll once again be subject to selective pressure eliminating mice that don't cope well with scarce food. How the mice will rise to that challenge isn't certain: mice with the best remaining fat-regulator alleles may do well; mice that store seeds instead of fat may do well; or they all may perish. Time will tell.

Natural Selection in Miniature

Natural selection works just the same in plants and bacteria as it does in animals. Bacteria act like very small animals, with three exceptions. First, they can acquire new Instructions (and therefore new traits) by conjugation. Second, they don't make new Instructions by exon shuffling. Third, since they're prokaryotes, they have only a single copy of their Instruction Book, so they're more vulnerable to mutation changing an Instruction badly. On balance, though, bacteria seem to be more robust than animals at evolving.

Make Shade, not War

Due to the random nature of natural selection, it's really possible to see how living things evolved only in retrospect. The mice in the warehouse have no idea why they survive, or whether they'll be exposed to capsaicin long enough for mutation to produce new traits to cope with it. Only people, looking back through history, can say, "These mice evolved different workers to cope with capsaicin." Since nobody can see the future, and what selective pressures it might apply, it's not possible to say that a particular organism is evolving by natural selection in a particular direction. This limitation is in direct contrast to the goals of artificial selection.