168 CHAPTER 7Figure 7.2 illustrates five fundamental features of drift. The first is that drift is
unbiased: an allele frequency is as likely to go up as to go down. Natural selection
can favor one allele over another, but genetic drift does not.
Second, the figure shows that the random fluctuations in allele frequency are larger
in smaller populations. That results from a basic law of probability. When you flip a
coin, you expect to get heads half the time and tails half the time. If you flip a coin
only twice, there is a probability of 1/2 that you will get all heads or all tails, rather
than half and half. But if you flip the coin 1000 times, the probability that you will
get all heads or all tails is less than 10–300. It is much more likely that about half
of the flips will come up heads and half of them tails. Outcomes become more
predictable when averaging over a larger number of random events. This is why
random genetic drift is stronger in small populations and weaker in large ones.
A third basic feature is that drift causes genetic variation to be lost. An allele frequency
that fluctuates randomly up and down will eventually reach either p = 0 or p = 1.
(Picture a New Year’s Eve partyer staggering along a long train platform with railroad
tracks on each side. Sooner or later, he will fall off the platform onto one track or the
other.) One of the alleles is then fixed. While the allele that is lost can be reintroduced
by mutation, drift by itself causes genetic variation to be lost. The loss is faster in
smaller populations because they have larger allele frequency fluctuations.
A fourth basic feature seen in Figure 7.2 is that drift causes populations that are
initially identical to become different. A useful way to think about this point is to con-
sider the variance in allele frequencies among the populations in our experiment. At
generation 0, all populations have an allele frequency of p = 0.5, and so the variance
among them is zero. After one generation, however, variation among the popula-
tions is generated by drift, and the variation grows with time. In quantitative terms,
starting with an allele frequency p, the average of the allele frequencies across the
replicates in the next generation is also expected to be p, and the variance of allele
frequencies across the replicates will be p(1 – p)/2N. (This quantity is based on the
binomial distribution, which describes how allele frequencies change by genetic
drift.) This confirms our earlier conclusion that the variation among populations
generated by drift grows more slowly in large populations than in small ones.
Fifth, the figure shows that an allele can become fixed without the benefit of natu-
ral selection. If we wait long enough, it is certain that one of the two alleles will
become fixed and the other lost. A simple rule tells us the probabilities of those two
outcomes: if an allele’s current frequency is p, then the probability that it becomes
fixed is also p. This result implies that a new mutation that has no effect on fitness
has a probability of 1/2N of ultimately becoming fixed, since it is initially present as
single copy among 2N copies of the gene in a diploid population.
We can see several key features of drift in an experiment that used the fruit fly
Drosophila melanogaster [5]. Replicate populations were established, each with eight
females and eight males. Each replicate was begun with equal frequencies of two
alleles that do not measurably affect survival or reproduction (at least in the lab).
The replicate populations were propagated by allowing the parents to reproduce,
then choosing eight female and eight male offspring at random to start the next
generation. FIGURE 7.3 shows the results. After one generation, the allele frequen-
cies varied substantially among the replicates. By ten generations, the frequencies
were distributed evenly between 0 percent and 100 percent. After 19 generations,
more than half of the replicates had lost one allele or the other.
Figure 7.3 shows one more general feature of genetic drift. Across the populations
taken as a whole, there are fewer heterozygotes than predicted by the Hardy-Wein-
berg ratios. This deficit is cause by allele frequency differences among populations.
If the Drosophila experiment were continued, one or the other allele would become
fixed in every population. At that point, roughly half the populations would be07_EVOL4E_CH07.indd 168 3/23/17 9:09 AM