MUTATION AND MIGRATION

We have learned how selection can change the frequencies of alleles and genotypes in
populations. Selection typically eliminates variation from within populations. (The
general exception to this claim is with the class selection models we have called
"balancing" selection where alleles are maintained in the population by
overdominance, habitat-specific selection, or frequency dependent selection). If
selection removes variation, soon there will be no more variation for selection to act
on, and evolution will grind to a halt, right? This would be true if it were not for the
reality of Mutation which will restore genetic variation eliminated by selection. Thus,
mutations are the fundamental raw material of evolution.

We will be gin by considering what mutation will do as an evolutionary force acting

by itself. Simply, mutation will change allele frequencies, and hence, genotype
frequencies. Lets consider a "fight" between forward and backward
mutation. Forward mutation changes the A allele to the a allele at a rate
(u); backward mutation changes a to A at a rate (v). We can express the frequency
(p) of the A allele in the next generation (p t+1) in terms of these opposing forward and
reverse mutations, much like forward and reverse chemical equations: (pt+1) = pt(1-u)
+ qt(v). The first part on the right is accounts for alleles not mutated (1-u), and the
second part accounts for the increase in p due to mutation from a to A (the frequency
of a times the mutation rate to A). We can also describe the change in allele
frequency between generations (p) as: p = (pt+1) - (pt). This is useful because it lets
us calculate a theoretical equilibrium frequency which is defined as the point at
which there is no more change in allele frequencies, i.e. when p = 0 which is
when (pt+1) = (pt); from above: pt(1-u) + (1-p)t(v) = pt [remember, q=(1-p)]. Now
solve for p and convince yourself that the equilibrium frequency = p = v/(u+v).
Similarly the equilibrium frequency of q = u/(u+v).

MUTATION AND SELECTION BALANCE

In the real world we will generally not find specific evolutionary forces acting alone;
there will always be some other force that might counteract a specific force of interest.
Our ability to detect these opposing evolutionary forces depends, of course, on the
relative strengths of the two (or more) forces. However, it is instructive to examine the
conditions where evolutionary forces oppose one another to give us a feel for the
complexity of evolutionary processes. Here we will consider a simple case where
mutation introduces a deleterious allele into the population and selection tries to
eliminate it.

As above we define the mutation rate (u) as the mutation rate to the "a" allele. This
will tend to increase the frequency of a (i.e., q will increase). In fact, q increases at a
rate of u(1-q); remember, (1-q) = p, or the frequency of the A allele. This mutation
pressure will increase the number of alleles which selection can act against.

To select against the a allele, we first will assume complete dominance, i.e., that the
deleterious effects of the a allele are only observed in the aa homozygote. Under these
conditions, the frequency of "a" (q) decreases by selection at a rate of -sq2(1-q), where
s is the selection coefficient. We won't derive this for you, but note that the amount of
change generated by this selection is a function of the frequency of the aa homozygote
(q2) and the frequency of the A allele (1-q). In other words, the amount of change is
proportional to the amount of genetic variation in the population, as we showed last
lecture.

If we put these terms for mutation and selection together, the amount of change in the
a allele is :

q = u(1-q) - sq2(1-q)

Now, if the "fight" between selection and mutation is a "draw", we have a condition
where there is no change in the frequency of the a allele since mutation is increasing q
just as fast as selection is reducing it. Under these conditions, q = 0, and we say that
the equilibrium allele frequency, q-hat, has been reached (in formal notation q-hat is
q with a circumflex over it). We simply rearrange the above formula so that is
becomes : u(1-q) = sq2(1-q). We solve this for q to give the equilibrium allele
frequency , q-hat: q = sqrt(u/s) (sqrt stands for square root).

Most mutation rates are fairly small numbers (about 10 -6), so this equation suggests
that deleterious alleles will be maintained in mutation selection balance at fairly low
frequencies. However, this statement is exactly what this mode is intended to
illustrate: we cannot say what that frequency is until we know BOTH mutation rate
and selection coefficient. However, we will refer back to mutation selection balance
several times during the course, so it is essential that you have a feel for dynamics of
this evolutionary interaction.

MIGRATION

In population genetics, the term "migration" is really meant to describe Gene flow,

defined as the movement of alleles from one area (deme, population, region) to
another. Gene flow assumes some form of dispersal or migration (wind pollination,
seed dispersal, birds flying, etc.) but dispersal is not gene flow (genes must be
transferred, not just their carriers)
We can describe gene flow (migration) in a manner similar to mutation. Consider two
populations, x and y with frequencies of the A allele of px and py . Now consider that
some individuals from population y migrate into population x. The proportion of these
y individuals that become parents in population x in the next generation = m. After the
migration event, population x can be considered to consist of migrant individuals
(proportion m) and non-migrant individuals (proportion [1-m]). Thus the frequency of
the A allele in population x in the next generation (p x t+1) is just the frequency in the
non-migrant portion (= px [1-m]) plus the frequency in the migrant portion (p y m).
Thus: px t+1 = px t [1-m] + py m.

The change in allele frequency due to gene flow is p = (px t+1) - px t which is just;

[px t [1-m] + py m] - px t Multiplying through and canceling terms leaves us with:

p = -m(px t - py t). This makes intuitive sense: the change in p depends on
the migration rate and the difference in p between the two populations. If we
considered a grid or array of populations and focus on one of those populations as the
recipient population with all other populations contributing equally to it, then
py would be replaced by the average p for all the other populations. Many scenarios
are possible.

GENE FLOW AND SELECTION

To address the combined effects of gene flow and selection, we will invoke a "fight"
similar to what we described for mutation selection balance above. Consider that
some weak allele is wafting over to the other side of the tracks, so to speak, where
they do not survive (e.g., fish swimming into New York harbor). There is an
evolutionary pressure changing allele frequencies in one direction ( into the harbor),
and an opposing evolutionary force eliminating those alleles (sewage killing off
genetically intolerant fish). Depending on the relative strengths of these two opposing
forces, an equilibrium condition can arise.

Lets consider the movement of the "a" allele, and assume that it is completely
recessive in its phenotype of death-by-sewage. The change in allele frequency from
the migration into the harbor can be defined as above: q = -m(qx t - qy t). (Note that
we have changed p to q since we are considering the a allele; x and y refer to the two
populations). The change in allele frequency due to selection against this allele is -
sq2(1-q) (note that this is the same expression we used in the mutation selection
balance above). Putting these two pieces together, we can write the expression for the
change in allele frequency that is due to BOTH gene flow and selection: q = -m(qx
2
t - qy t) - sqx (1-qx). When the "fight" between gene flow and selection is a "draw", the
system will be in equilibrium and there will be no change in q,
and -m(qx t - qy t) = sq2(1-q).

So, back to the fish. Lets say that aa homozygotes drop dead the minute they enter the
East river (but that AA and Aa fish are fine). Outside New York Harbor, the
frequencies of the two alleles are equal (p = q= 0.5). It is discovered that in the East
river, q = 0.1 over many generations, and the Mayor wants to know what proportion
of the fish in the East river come from the outside each generation. This information
gives us all we need: it's in equilibrium, q x = 0.1 (East River), qy = 0.5 (outside), and s
= 1.0 since the aa's are dead. Plug in the values and you get m = 0.023. This says that
2.3% of the fish in the East river must come from outside each generation to maintain
the allele frequency at q = 0.1

GENETIC DRIFT

Genetic drift refers to random fluctuations in allele frequencies due to chance events

(see figure 6.4, pg. 142). The previous lectures have all dealt with deterministic
(predictable) evolutionary forces often referred to as linear pressures. Genetic drift is a
stochastic (random) force that can scramble the predictable effects of selection,
mutation, and gene flow. While it might seem that a random force would be of little
significance to evolutionary "progress" (we'' confront this loaded term later), genetic
drift is an extremely important force in evolution. However, its strength depends on
the size of the population, as a simple exercise in coin tossing will illustrate. In ten
tosses you might easily get seven heads; in 1000 tosses, however, you would never get
700 heads with a "fair" coin. The same sort of random fluctuation in allele
frequencies can occur in small populations: consider a bag full of red and green
marbles each in equal frequency; pull out a small handful and the frequency in your
hand will probably not equal the frequency in the original bag. Let that handful
determine the frequency in a new population that grows back to the original
population size. A second small handful will randomly shift the frequency to yet
another frequency. If you pulled out all the marbles in the bag (= large population)
then the frequency would be maintained exactly in the next generation. Genetic drift is
not a potent evolutionary force in very large randomly mating populations.

To illustrate the consequences of genetic drift we will consider what happens when
drift alone is altering the frequencies of alleles among many small populations. To
illustrate this we need to understand Population structure, which describes how
individuals (or allele frequencies) in breeding populations vary in time and space.
This structure is determined by the combined effect of deterministic and stochastic
forces. We will introduce the idea of population structure by showing how genetic
drift and inbreeding can change the frequencies of genotypes in populations.
Consider a grid of small populations (e.g., ponds in Minnesota), all with the same
small population size and all starting at time t with p = q= 0.5. Through time each
population will experience genetic drift due to random sampling and the frequencies
in each population will diverge. The distribution of frequencies changes over time
from a tight distribution (all 0.5), to a flat distribution (some populations at p = 0.1,
some at 0.9 and all frequencies in between), to fixation (p =1.0) or loss (p = 0.0) of
the alleles in all populations (see figure below). Fixation is when all alleles in the
population are A; this necessarily implies loss of the a allele ("fixation" or "loss"
should only be used with reference to a specific allele). If each population starts at p =
0.5, then at the end, when all populations have lost their variation, 50% of the
populations will be fixed for the A allele and 50% will be fixed for the a allele (latter
= "loss" for the A allele, get it?). If the initial frequency was p = 0.7, then 70% of the
populations would be fixed for the A allele (again, assuming no selection, migration,
mutation).

Main Points: 1) total variation does not change; variation goes from within
populations (no variation between populations) to between populations (no
variation within populations). 2) genetic divergence of populations entirely
by chance! (no selection). This is why genetic drift can be an important force in
evolution.

At the start of this drift process in our array of populations, p = 0.5 and there are 2pq =
0.5 = 50% heterozygotes. When all populations in the array have fixed or lost the
allele, there can be no heterozygotes (i.e., 0%). This shows that the proportion of
heterozygotes decreases as drift proceeds (this also occurs when there is inbreeding
which can also be thought of as a sampling error phenomenon). We can quantify this
process as follows: the proportion of heterozygotes in the "next " generation is a
function of the proportion of heterozygotes in the present generation and the "rate" at
which drift proceeds: Ht+1 = Ht[1 - (1/2N)] where H = the proportion of heterozygotes
in the population (or in the array of populations) and N = population size. This can be
extended over many generations as follows: Ht = H0[1 - (1/2N)]t where t refers to the
number of generations in the future and 0 refers to the present (or starting) generation.
Looking at these equations it is clear that with small population sizes, heterozygosity
will be lost quickly (drift will proceed quickly), whereas in large populations there
will be little loss of heterozygosity.

If we consider our grid of populations again, we note that as drift proceeds and each
deme

becomes a bit different from every other deme, the variation among demes increases.
Like the loss of heterozygosity due to drift, the increase in the variation among demes
depends on the population size. This variation can be described as Vt = p(1-p)[1 - (1 -
1/2N)t]. Note at t=0, Vt = 0 because the term in brackets = 0. As the number of
generations proceeds, the variation among populations (V t) increases rapidly if N is
small, but slowly if N is large.

A general result as drift proceeds in small populations is a deficiency of

heterozygotes, and reciprocally, an excess of homozygotes. This is also a common
result when there has been inbreeding (= mating between relatives). In fact genetic
drift and inbreeding are related phenomena. The relation between the frequencies
of expected versus observed heterozygotes allows us to determine the inbreeding
coefficient, F = (He - Ho)/He (subscripts e and o mean expected and observed,
respectively).

One effect of inbreeding is to increase the frequency of homozygotes (and thus,

necessarily, decrease the frequency of heterozygotes). Note: while the frequency
of genotypes change with inbreeding, the frequencies of alleles remains the same
(assuming no selection, migration, mutation). Refer back to the data table presented
on page 2 of Lecture 6 to convince yourself that those data could be a result of
inbreeding: F=(0.343 - 0.14)/0.343 = 0.59. When the allele frequency is not zero, but
there is a complete absence of heterozygotes , F = 1. As an exercise, work through the
data in table 5.2, pg.98. Does this illustrate high or low inbreeding?

Genetic variation is generally "lost" by the action of genetic drift. This is true if we
follow the fate of one deme over time. Note, however, that in our array of populations,
variation is "lost" within demes, but the variation in the total system is preserved, i.e.,
the allele frequency in the entire metapopulation does not change, only the genotype
frequencies and allele frequencies within individual demes).

Inbreeding also has the effect of increasing the variance among the individual demes
of a larger population. As such, drift and inbreeding are closely related evolutionary
forces. Recall that the variance = 1/N_(X-xi)2. In a random mating population with p
= 0.4, f(AA) = 0.16, f(Aa) = 0.48, f(aa) = 0.36. If the alleles act in
an additive manner, the heterozygotes will be intermediate and close to the mean in
phenotype and will contribute little to the variance; with inbreeding most individuals
are homozygous, and thus would deviate from the mean and the variance would be
greater. In effect, inbreeding makes the distribution of phenotypes more "bimodal" by
essentially redistributing the alleles from all three genotypes into the two
homozygotes (see figure 6.5, pg. 143).

Population structure is usually quantified by a simple statistic known as Fst. This

stands for the "fixation" index resulting from comparing sub populations to the total
population, and is used to quantify the proportion of genetic variation that
lies between subpopulations within the total population. An important way of thinking
of this problem is to compare the mean heterozygosity averaged across all demes to
the heterozygosity that would result if all demes were pooled into one big population.
Heterozygosity is the proportion of heterozygotes in the population and is defined as
H = 2 p q. Note that heterozygosity is zero at "fixation", the case where only one
allele exists (p = 0 or 1), and that heterozygosity is at a maximum when alleles are
equally frequent (e.g., p = q = 0.5). [For completeness, H = 1- (p 2 + q2) which follows
from Hardy-Weinberg above. In the case of more than two alleles, we can't just use p
and q, so the following expression for heterozygosity works more generally: H = 1 -
_xi2 where xi is the frequency of the "ith" allele, and summation is across all i alleles.
The expression 1 - (p2 + q2) is identical to 1 - _xi2 when there are only two alleles].

In our metapopulation example above, on the left all demes have p = 0.5 and the allele
frequency for the entire array is also p = 0.5. Right away you can se that there in no
differentiation among demes. To quantify this we use Fst which is calculated as Fst =
(Ht - meanHs) / Ht, where Ht = 2 (pooled p) (pooled q) [note that pooled q = (1-pooled
p)], and meanHs = the average of H values for each of the individual demes (i.e., 2 p q
for each deme, averaged across all demes). For the left-hand metapopulation Ht =
2(0.5)(0.5) = 0.5, and meanHs = 0.5 also. Thus , Fst = (0.5 - 0.5) / 0.5 = 0.0. What does
this mean? In English this says that none of the variation in this system of demes
lies between demes; the population structure is zero. Since there is variation (i.e., p _
0) all of the variation lies within demes.

Now consider the metapopulation system on the right above. The pooled allele

frequency is 0.5 because half of the demes are fixed for the A allele (p = 1.0) and half
are fixed for the a allele (p = 0.0). Hence the Ht value is 2 (pooled p) (pooled q) =
2(0.5)(0.5) = 0.5. The meanHs value is very different. Each deme has a heterozygosity
of zero (either p or q is zero in all demes), so the average Hs value is 0.0. Hence, for
the metapopulation on the right, Fst = (0.5 - 0.0) / 0.5 = 1.0. In English this means
that all (100%) of the variation in the system lies between the demes. By definition
then, none of the variation lies within demes, which we know because the little circles
are either filled or empty. By now it should be apparent that in Fst values can range
between the two extremes of zero and one that we have just illustrated. As genetic
drift proceeds, Fst values will increase, but the balance between drift and "linear
pressures" will determine what equilibrium Fst reaches (later lecture).

A further clarification: The "speed" or "intensity" of genetic drift is actually

determined by the effective population size (Ne). This may differ from the total
population size (N) if some individuals do not breed. Two examples where Ne differs
from N are cases of different mating system and temporal fluctuations in population
size. When the number of breeding males (Nm) _ number of breeding females (Nf),
the effective population size can be quite different from the actual population size.
The relation is: Ne = 4NmNf/(Nm + Nf). If Nm = Nf = N/2, then Ne = N. But if a
single male does all the mating (approximated in elephant seals) then Ne = 4 (because
Nf/(1+Nf) is approx. = 1). So the population genetics of elephant seals and sage
grouse will be very different from that of large populations of insects.

With population bottlenecks where the population size drops to a small number in

one generation, the effective population size is not just the average of N's for each
generation. To estimate the Ne, one calculates the harmonic mean population size as
follows:

1/Ne = 1/t_1/Nt where t = the number of generations and Nt = the population size at

each generation. Thus with population sizes of 100, 100, 20, 100 the arithmetic mean
= 80 but 1/Ne = 1/4(1/100 + 1/100 + 1/20 + 1/100) = 1/4(0.08) = 0.02, so Ne = 1/0.02
= 50. Thus the smaller population size has a disproportionate effect on the effective
population size. This is a very important issue in conservation efforts concerning
endangered species.

DRIFT AND MUTATION

Drift will tend to reduce heterozygosity (for our purposes this equals the proportion of
heterozygotes), mutation will introduce new alleles which will serve to increase
heterozygosity. This provides yet another example of a "fight" between opposing
evolutionary forces. When the mutation rate is close to the reciprocal of the
population size, heterozygostity will be high (i.e., a considerable amount of variation
will exist in the population). The "balance" of this equilibrium can be described by an
equation for the equilibrium frequency of heterozygotes:

H Å Plug in Ne's and u's to determine balance: when Ne = 1/u, H = 0.8; when
u>1/Ne then heterozygosity will be higher; when u<1/Ne heterozygosity will be
lower.

DRIFT AND GENE FLOW

Gene flow can counteract the loss of heterozygosity due to drift as well as counteract
the random divergence of allele frequencies among populations. The balance between
these two opposing forces can be described by an equation for the equilibrium
variance among populations

V(among pops.) Å . As m increases, the variance decreases (faster homogenization);

as Ne increases, the variance decreases (drift acts more slowly with larger N e). Major
conclusion is that it takes very little gene flow to keep two "populations"
homogeneous, as little as one reproductive migrant between populations per
generation! See figure 5.13, pg. 128.

A number of different methods have been developed to estimate migration rates from
standing patterns of allele frequency variation. This rests on the fact that Fst values
are a result of a "balance" between gene flow and drift. One can calculate Fst from
molecular markers, and if an estimate of effective population size is available,
migration can be estimated from the following relation: Fst Å 1/(4Nem + 1). By
rearranging the formula and plugging in trial values you can see that low values for m
result in high values for Fst, and high values of m give low values of Fst. This follows
from the homogenizing effect of gene flow on allele frequency variation among
populations.

The nature of population differentiation can depend on population structure. Typical

scenarios include a Continent-Island model (gene flow from continent to island),
an Island model (equal probability of gene flow among several/many
populations), Stepping Stone model (gene flow is sequential among
populations), Continuous Model (e.g., carpet of individuals). In all cases (except
Island model) there can be Isolation by Distance simply reflecting incomplete
homogenization of populations due to incomplete flow of genes among all
populations.

INTEGRATION OF EVOLUTIONARY FORCES

Selection leads to an increase in the average fitness of the population. Illustrated by

the following example. Recall that the average fitness ("w bar") = w AA p2 + wAa 2pq +
waa q2. Now consider the case where p=.2, q= .8 so f(AA)=0.04, f(Aa)=0.32 and
f(aa)=0.64. If their respective fitnesses are 1.0, 0.2 and 0.1, then w bar = .
04(1)+.32(.2)+.64(.1) = .132. If selection were to continue for a number of
generations q would decrease and p would increase. Lets say that we recalculated the
average fitness when p=0.8 and q=0.2: thus f(AA)=.64, f(Aa)=.32 and f(aa)=.04 so w
bar = .64(1)+.32(.2)+.04(.1)=.708. When will the average fitness reach a maximum?
when the deleterious allele (a) is selected out of the population so the only genotype is
AA, the most fit genotype. (How would this differ if there was dominance?)

The sickle cell story with a new twist. There are actually more than just two alleles
relevant to the sickle cell story. The normal allele (A) and the sickle allele (S) produce
the genotypes traditionally considered: AA is normal but susceptible to malaria, AS
individuals are not severely debilitated, but they are resistant to malaria, SS
individuals are severely affected and rarely survive to reproduce. A third allele (C)
results in the following genotypes and phenotypes: AC is normal but susceptible, SC
has mild anemia and CC is normal and resistant to malaria. The following
approximate fitnesses have been assigned to these genotypes:

Genotype w phenotype
AA 0.9 malarial susceptibility
AS 1.0 malarial resistance
SS 0.2 anemia
AC 0.9 malarial susceptibility
SC 0.7 anemia
CC 1.3 malarial resistance

Bantu speaking people moved into central and western Africa. A slash and burn
agriculture opened up habitat for mosquitoes and along with them
came Plasmodium the malarial agent. Consider a large population with the above
fitness exposed to malaria for the first time. f(A) ~ 1.0 f(S) and f(C) very low, thus all
S and C alleles will be in heterozygous states as AC or SC. Virtually No SS or
CC genotypes will be present (the product of two very small numbers). Selection will
lead to an increase of the S allele due to its higher fitness as a AS heterozygote.
Expect no change in the C allele because heterozygotes have equal or lower fitness
than other common genotypes. Can't "pass through" the heterozygote stage.

Selection is "short-sighted" cannot "see" the best solution, i.e., the highest fitness
state where C goes to fixation. How do we get to the "best" condition?

Consider the effects of population structure: Bantu people establish local breeding
groups of small effective population size. This necessarily will bring on
some inbreeding which serves to increase homozygotes and decrease heterozygotes.
Some CC genotypes will be produced by the chance effects of inbreeding and now the
"best" genotype is present in the population and selection can "see" it so the frequency
of the C allele increases, and the population would go to fixation for C.

The highest fitness state could not be reached by selection alone; Drift can affect the
outcome of selection. This has been conceptualized as an Adaptive landscape:

Pick two allele frequencies, one for the A locus and one for the B locus. This defines a
"population" (obviously this would be done in many dimensions for all loci in the
genome; only two here for illustration). The population will evolve by selection to the
top of the nearest peak. If a population starts with f(A) ~ f(B) ~ 0.2, this population
would evolve to the top of the peak at the lower right of the diagram. This is not the
highest peak, but selection acts to increase average fitness and can only "see" the
nearest peak. If drift due to low effective population size rapidly shifted both f(A) and
f(B) to higher frequencies, then the population might be in the "domain of
attraction" of the highest peak in the upper right corner.

The population would stop evolving when it reached the top of the highest peak
because there is no higher peak to shift to unless the environment changes at which
point we would have to redraw the adaptive landscape.

Sewell Wright conceived of this view of evolution and believed that this was a more
accurate description of how "real" populations evolved sine most species do have
some structure to their populations and experience drift. So there will be a shifting
balance between allele frequencies and a shifting balance between drift and selection
as the causative agents in evolution. This is the so called shifting balance theory.

Wright envisioned different stages of evolution by shifting balance: 1) Drift in local

populations would shift allele frequencies to new values and the demes may evolve
up a local peak because the allele frequency in such a deme drifted near the 'domain of
attraction' of a peak. The assumption is that without drift, a population on a 'flat'
section of the adaptive landscape would not evolve by natural selection, because there
would be no fitness variation in a flat region of the landscape. 2) intrademic
selection where selection within local populations (demes) would drive the various
demes to the top of their nearest peak. Even if several populations were at different
"locations" on the adaptive landscape, the highest peak may not be reached. One can
invoke stage 2.5 by saying that drift in local populations might move such a
population's allele frequency off one peak and into the 'domain of attraction' of an
adjacent peak with a different maximum fitness. This peak may be lower or higher
than the old one, but after several rounds of drift at least one population may evolve to
the top of the highest peak. 3) This (these) high-fitness populations will produce
many emigrants and tend to change the other demes' allele frequencies closer to their
own as a result of gene flow; this third stage is called interdemic
selection (selection among demes); emigration rates are proportional to the extent that
the fitness of a given population is greater or less than the average fitness of all
populations. When all populations are homogenized to the allele frequencies of
maximal fitness a new balance will be achieved and the allele frequencies will
be maintained by selection until the environment changes the adaptive landscape.
(For empirical support of this theory see Wade and Goodnight, 1991 Science vol. 253
pg. 1015-1018 and a commentary on page 973 by Crow).

Alternative way to view the shifting balance theory: consider a surface with troughs
and pits in it. Put several marbles on the surface. If marble is near pit it falls
in selection ~ gravity. Shake surface and balls will roll up out of pits against
gravity and make their way to new pit. Shaking ~ drift. See figures 8.8-8.11, pgs.
215-219.

Before discussing the shifting balance view of evolution we considered selection as if

it were acting on a single locus. This is a gross oversimplification because many loci
are linked along the chromosome. Who's to say that selection is acting the same way
on both loci? Things get much more interesting (but more complicated) when we face
the reality of linked loci. Consider the cross between the two two-locus genotypes:

The offspring can be AB/AB, AB/ab or ab/ab. Other two locus

genotypes are possible:  or  . But these can only be produced in the cross if
there is recombination between the two loci. We can thus refer to four two-locus
gametes AB and ab are the coupling gametes and aB and Ab are the repulsion
gametes (another way to think about gametes is to just refer to them as a
"chromosome" since this will reflect the linear array of whatever alleles are linked
together). The frequency of these four gametes will be determined by two
things 1) the frequencies of the respective alleles (p and q for the A locus and a
different p and q for the B locus) and 2) the degree of linkage disequilibrium which
describes whether recombination has broken up any association between the two
linked loci.
When allele frequencies all = 0.5 and all gametes are in equal frequency then f(AB) =
f(ab) = f(Ab) = f(aB). But if A alleles tend to be associated (linked) to B alleles then
AB gametes will be in higher frequency than expected at random. We can quantify
the disequilibrium as follows:

D = [f(AB) f(ab)] - [f(Ab) f(aB)]. (Note frequencies are multiplied) When all gametes
are in equal frequency D = 0 i.e., linkage equilibrium. When only the coupling
gametes are present D = 0.25; when only the repulsion gametes are present D = -0.25.
If the frequencies of the alleles are less than 0.5, then the maximum value for D will
be less than 0.25.

Note that when allele frequencies are different from p = 0.5 = q, the maximum value
of D (absolute value)will be less than 0.25. For example if p=0.8, q=0.2 and if only
the coupling gametes were in the population then D = 0.16

A worked example: gamete frequencies in 1000 observations: 580 AB's, 140 Ab's, 60
aB's and 280 ab's. Thus f(A) allele = (520+140)/1000 = 0.66 so f(a) = .34. f(B) =
(520+60)/1000 = 0.58 so f(b) = 0.42. At random we expect the following gamete
frequencies: f(AB) should be .66(.58)1000 = 383. f(Ab) should be .66(.42)1000 =
277. f(aB) should be .34(.58)1000 = 197 and f(ab) should be .34(.42)1000 = 143.
These numbers of expected gametes are clearly different from the observed gametes.
We can thus calculate the linkage disequilibrium as d = [.52(.28)] - [.14(.06)] =
0.1372. This tells us that the A and B alleles are in linkage disequilibrium.

This disequilibrium will be broken up by recombination and the rate of breakup will
be determined by the rate of recombination (see figure 8.2, pg. 202).

Now let's say that the A locus was under selection with A alleles favored. If the A and
B loci were in linkage disequilibrium in the coupling state what would happen to the
B alleles? They too would be selected for, but not because they were under selection.
This is a very important phenomenon in population and evolutionary genetics
called hitchhiking. It demonstrates a very important distinction we must make about
selection and phenotype: we need to distinguish between selection "of" and
selection "for" If the A allele is favored, there is selection for the A allele and
selection of the B allele due to its linkage to the A allele (i.e., linkage between the A
and B loci).

Now consider the situation where the nose length is the result of interactions between
the two loci. In the first case the interaction is additive, in the second case there
is epistasis
 AA Aa aa AA Aa aa
BB 1 2 3 BB 1 2 3
Bb 2 4 6 Bb 4 6 4
bb 3 6 9 bb 9 6 3

In the first case if we selected for long noses, we would tend to drive the a and b
alleles to high frequency. If we selected for long noses in the second case we would
tend to drive the A and b alleles to high frequency. The important distinction between
these two tables is that in the simple two-locus additive case on the left,
heterozygotes at one locus are intermediate between the two
homozygotes regardless of the genotype at the other locus. In contrast, the table on the
right shows that the relationship between genotype and phenotype at one
locus depends on the genotype at the other, interacting locus. In a sense, one locus
is modifying the expression of the other locus. If selection were to act in favor of nose
length in the right-hand epistatic system, the way alleles "marched to fixation" would
be very different.

Now consider how linkage and epistasis can affect the response to selection. In the
second case above if there was high linkage disequilibrium so that all we had we AB
and ab chromosomes in the population (= AB or ab gametes in the gamete pool), there
would be less variation to select on (sizes 1, 6 and 3). Now if there was recombination
such that Ab and aB chromosomes were produced, then the full range of phenotypic
variation would be exposed (up to 9) and selection would rapidly shift the mean
phenotype to longer noses and to high frequency of A and b alleles.

The general point is that loci do not act independently and their response to selection
depends critically on their linkage relationships and their interaction with other
loci. For the ecologically minded, there are some interesting parallels between
community ecology and population genetics: there are an uncountable number of
ways that the interacting participants can interact. In community ecology one
considers species in a community; in population genetics one considers genes in the
genome. The fate of each player depends on the degree to which it is "connected" to
the other players in the system. Darwin referred to the complexity of nature as a
"tangled bank"; this is very true of the genes within genomes within populations.