Molecular Evolution

Lecture Notes
Anders Gorm Pedersen
[email protected]

February 4, 2005
2
 i

Contents

1 Brief Introduction to Evolutionary Theory 1

1.1 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Darwin and the Theory of Evolution . . . . . . . . . . . . . . 2
1.3 Natural Selection . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 The Modern Synthesis . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Mendelian Genetics . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Brief Introduction to Population Genetics 9

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Population Growth . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Exponential Growth . . . . . . . . . . . . . . . . . . . 10
2.2.2 Logistic Growth . . . . . . . . . . . . . . . . . . . . . 13
2.3 Genotype Frequencies and Growth . . . . . . . . . . . . . . . 13
2.4 Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Mutation-Selection Balance . . . . . . . . . . . . . . . . . . . 21
2.6 Mathematical Modeling: A Few Thoughts . . . . . . . . . . . 23
2.7 Genetic Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.8 Chance of Fixation by Drift . . . . . . . . . . . . . . . . . . . 25
2.9 Drift and Neutral Mutation . . . . . . . . . . . . . . . . . . . 25
2.10 The Neutral Theory . . . . . . . . . . . . . . . . . . . . . . . 28
2.11 The Molecular Clock . . . . . . . . . . . . . . . . . . . . . . . 29
ii
 1

Chapter 1

Brief Introduction to
Evolutionary Theory

c Anders Gorm Pedersen, 2005
2 Molecular Evolution, lecture notes

1.1 Classification
One of the main goals of early biological research was classification, i.e., the
systematic arrangement of living organisms into categories reflecting their
natural relationships. The most successful system was invented by the swede
Carl Linnaeus, and presented in his book ”Systema Naturae” first published
in 1735. The system we use today is essentially the one devised by Linnaeus.
It is a hierarchical system with seven major ranks: kingdom, phylum, class,
order, family, genus, and species.
Carl Linnaeus Specifically, groups of similar species are placed together in a genus,
(1707–1778) groups of related genera are placed together in a family, families are grouped
into orders, orders into classes, classes into phyla, and phyla into kingdoms.
When depicted graphically, the Linnean system can be shown in the form
of a tree with individual species at the tips, and with internal nodes in
the tree representing higher-level categories (Fig. 1.1). Along with this
classification system, Linnaeus also developed the so-called binomial system
The Linnean system: in which all organisms are identified by a two-part Latinized name. The
• Kingdom first name is capitalized and identifies the genus, while the second identifies
the species within that genus. For example the genus Canis includes Canis
• Phylum lupus, the wolf, Canis latrans, the coyote, and Canis familiaris, the domestic
• Class dog. Similarly, the genus Vulpes contains Vulpes vulpes, the red fox, Vulpes
• Order chama the Cape fox, and others. Both genera (Canis and Vulpes) belong
to the family Canidae, which in its turn is part of the order Carnivora, the
• Family
carnivores.
• Genus Note that it is non-trivial to come up with a generally applicable defi-
• Species nition of what exactly a “species” is. According to the so-called biological
species concept, a species is a group of “actually or potentially interbreed-
ing natural populations which are reproductively isolated from other such
groups”. This definition is due to the evolutionary biologist Ernst Mayr
(1904–) and is perhaps what most people intuitively understand by the word
“species”. However, the biological species concept does not address the issue
of how to define species within groups of organisms that do not reproduce
sexually (e.g., bacteria), or when organisms are known only from fossils.
An alternative definition is the morphological species concept which states
that “species are groups of organisms that share certain morphological or
biochemical traits”. This definition is more broadly applicable, but is also
far more subjective than Mayr’s.

1.2 Darwin and the Theory of Evolution

As mentioned, the Linnean system was highly successful. So much so in
fact, that in his publications, Linnaeus provided a survey of all the world’s
plants and animals as then known—about 7,700 species of plants and 4,400

c Anders Gorm Pedersen, 2005
Chapter 1: Brief Introduction to Evolutionary Theory 3

Figure 1.1: Linnean classification depicted in the form of a tree.

species of animals. Linnaeus believed that God was the ordering principle
behind this classification system, and that its structure somehow reflected
the divine master plan.
It was not until after the 1859 publication of Charles Darwin’s “On the
Origin of Species” that an alternative explanation was widely accepted. Ac-
cording to Darwin (and others), the ordering principle behind the Linnean
system was instead a history of “common descent with modification”: all life
was believed to have evolved from one—or a few—common ancestors, and
taxonomic groupings were simply manifestations of the tree-shaped evolu-
tionary history connecting all present-day species (Fig. 1.2).
The theory of common descent did not in itself address the issue of how
evolutionary change takes place, but it was able to explain a great deal
of puzzling observations. For instance, similar species are often found in
adjacent or overlapping geographical regions, and fossils often resemble (but
are different from) present-day species living in the same location. These
phenomena are easily explained as the result of divergence from a common
ancestor, but have no clear cause if one assumes that each species has been
created individually.

1.3 Natural Selection

The mechanism that Darwin proposed for evolutionary change is called nat-
ural selection. This is related to artificial selection—the process of inten-

c Anders Gorm Pedersen, 2005
4 Molecular Evolution, lecture notes

Figure 1.2: The tree of life, Ernst Haeckel, 1866

tional (or unintentional) modification of a species through human actions

which encourage the breeding of certain traits over others. Examples include
crop plants, such as rice and wheat, which have been artificially selected for
protein-rich seeds, and dairy cows which have been artificially selected for
high milk yields. The wide variety of dog breeds is also a result of artifi-
cial selection (for hunting, herding, protection, companionship, and looks)

c Anders Gorm Pedersen, 2005
Chapter 1: Brief Introduction to Evolutionary Theory 5

and illustrates that rather significant changes can be obtained in a limited

amount of time (a couple of thousand years in the case of dogs.) You should
note that for artificial selection to be possible in the first place, there needs
to be naturally occurring and heritable variation in traits of interest: it is
only possible to breed high-protein grass sorts, if there are some grass plants
that produce more seed protein than others, and if that trait is inherited by
their descendants.
Darwin suggested that a similar process occurs naturally: individuals in Charles Darwin
the wild who possess characteristics that enhance their prospects for having (1809-1882)
offspring would undergo a similar process of change over time. Specifically,
Darwin postulated that there are four properties of populations that to-
gether result in natural selection. These are:

1. Each generation more offspring is born than the environment can sup-
port - a fraction of offspring therefore dies before reaching reproductive
age.

2. Individuals in a population vary in their characteristics.

3. Some of this variation is based on genetic differences.

4. Individuals with favorable characteristics have higher rates of survival

and reproduction compared to individuals with less favorable charac-
teristics.

If all four postulates are true (and this is generally the case) then advan-
tageous traits will automatically tend to spread in the population, which
thereby changes gradually through time. This is natural selection. Let us
consider, for instance, a population of butterflies that are preyed upon by
birds. Now imagine that at some point a butterfly is born with a muta-
tion that makes the butterfly more difficult to detect. This butterfly will
obviously have a smaller risk of being eaten, and will consequently have an
increased chance of surviving to produce offspring. A fraction of the fortu-
nate butterfly’s offspring will inherit the advantageous mutation, and in the
next generation there will therefore be several butterflies with an improved
chance of surviving to produce offspring. After a number of generations it
is possible that all butterflies will have the mutation, which is then said to
be “fixed”.
If two sub-populations of a species are somehow separated (for instance
due to a geographical barrier), then it is hypothesized that this process may
lead to the gradual build-up of differences to the point where the populations
are in fact separate species. This process is called speciation.

1.4 The Modern Synthesis

c Anders Gorm Pedersen, 2005
6 Molecular Evolution, lecture notes

One problem with the theory described in “Origin of Species”, was that its
genetic basis—the nature of heritability—was entirely unknown. In later
editions of the book, Darwin proposed a model of inheritance where “hered-
itary substances” from the two parents merge physically in the offspring, so
that the hereditary substance in the offspring will be intermediate in form
(much like blending red and white paint results in pink paint). Such “blend-
ing inheritance” is in fact incompatible with evolution by natural selection,
since the constant blending will quickly result in a completely homogeneous
population from which the original, advantageous trait cannot be recovered
(in the same way it is impossible to extract red paint from pink paint).
Moreover, due to the much higher frequency of the original trait, the result-
ing homogeneous mixture will be very close to the original trait, and very far
from the advantageous one. (In the paint analogy, if one single red butterfly
is born at some point, then it will have to mate with a white butterfly re-
sulting in pink offspring. The offspring will most probably mate with white
butterflies and their offspring will be a lighter shade of pink, etc., etc. In the
long run, the population will end up being a very, very light shade of pink,
instead of all red).
However, as shown by the Austrian monk Gregor Mendel, inheritance
is in fact particulate in nature: parental genes do not merge physically;
instead they are retained in their original form within the offspring, making
it possible for the pure, advantageous trait to be recovered and, eventually,
to be fixed by natural selection. Although Mendel published his work in
Gregor Mendel 1866 it was not widely noticed until around 1900, and not until the 1930’s
(1822–1884) was Mendelian genetics fully integrated into evolutionary theory (the so-
called “Modern Synthesis”). This led to the creation of the new science of
population genetics which now forms the theoretical basis for all evolutionary
biology.

1.5 Mendelian Genetics

I will here briefly summarize some important aspects of Mendelian genetics,
and present a number of definitions that will be used later in the text.
An organism can be either haploid or diploid. Haploid organisms have
one complete set of genetic material (and therefore one copy of each gene),
while diploid organisms have two complete sets of genetic material located
on two complete sets of chromosomes (and therefore two copies of each gene).
A particular gene in a haploid or diploid organism is said to occupy a par-
ticular locus (plural: loci). If different versions of a gene are present at a
particular locus (e.g., in different individuals of a population) then these are
referred to as alleles of that gene. A diploid organism may have different
alleles present on the two individual copies of a chromosome. If a diploid
organism has the same allele on both chromosomal copies, then it is said to

c Anders Gorm Pedersen, 2005
Chapter 1: Brief Introduction to Evolutionary Theory 7

Figure 1.3: Mendelian genetics. Each diploid parent contains two alleles. The s
allele is recessive and results in wrinkled peas when present in two copies.

be homozygous for that allele (it is a homozygote). If it has two differ-

ent alleles present at a locus, then it is said to be heterozygous for that
allele (and is then referred to as a heterozygote). The total complement of
alleles present in an organism is its genotype. If we are interested in one
particular locus where the alleles A and a occur, then a diploid organism
might for instance have the genotype “AA” or “Aa”. A haploid organism
might have the genotype “a” at such a locus. Depending on the molecu-
lar nature of the different alleles present at a locus in a diploid organism,
one allele may not make an impact on the organisms appearance (its phe-
notype). It is then said to be a recessive allele. An allele that is fully
expressed in the organism’s phenotype is called dominant. For instance,
Fig. 1.3 shows two different alleles—the dominant S allele and the recessive
s allele—of a gene controlling wrinkledness in peas. Occasionally the alleles
will be co-dominant, and this will result in an apparent blending of parental
characteristics. In diploid organisms, one allele comes from the mother, one
from the father. When diploid organisms reproduce sexually, it occurs via
an intermediate, haploid sex cell called a gamete (the gamete is an egg
cell if it is produced by a female, and a sperm cell if it is produced by a
male). During gamete formation, genetic material from the two parents is
mixed by the process of recombination. Recombination is one stage of
the special type of cell division termed meiosis which ultimately results in
formation of the haploid gamete. At any one locus, there will (by necessity)
be only one allele present in the gamete. The diploid cell formed by fusion
of two gametes is called a zygote. Sexually reproducing organisms have life
cycles that alter between a haploid stage and a diploid stage. In some organ-
isms most of the life cycle is diploid (e.g., humans, where only the sex cells
are haploid), while the situation is reversed for other organisms (including
some algae where the diploid zygote quickly undergoes meiosis to form new

c Anders Gorm Pedersen, 2005
8 Molecular Evolution, lecture notes

haploid cells). There are also organisms (e.g., ferns) where the life cycle
alternates between a haploid, multicellular generation and a diploid, multi-
cellular generation. Asexual reproduction is seen in both haploid organisms
(e.g., bacteria) and diploid organisms (e.g., yeast and some plants).

c Anders Gorm Pedersen, 2005
 9

Chapter 2

Brief Introduction to
Population Genetics

c Anders Gorm Pedersen, 2005
10 Molecular Evolution, lecture notes

2.1 Introduction
The science of population genetics deals with genetic variation within pop-
ulations, and with the forces that change this variation. I will now give you
a very brief introduction to a few important results in the field.
My goal with this section is mostly to make you aware of some of the
ways in which evolutionary theory can be approached in a stringent, quanti-
tative manner. Specifically, we will discuss the effects that growth, selection,
mutation, and genetic drift have on the genetic composition of a population.
Most of the concepts will be introduced in the context of haploid, asexually
reproducing organisms since that makes the analysis easier.
The material covered here does not directly relate to reconstruction of
phylogenetic trees. However, any evolutionary history is necessarily the
result of processes that resemble the ones described in this section, and it
is therefore relevant to have at least passing knowledge of the underlying
theory.

2.2 Population Growth

2.2.1 Exponential Growth

We will start by analyzing the characteristics of a growing population. Imag-

ine that we are examining a population of asexually reproducing organisms
where each individual produces 200 offspring per generation and then dies.
The number 200 is called the fecundity of the organism and is usually de-
noted m. For various reasons only 2% of the offspring survive sufficiently
long to produce offspring of their own. This is the survival rate and is usu-
ally denoted L. We can easily see that each individual organism will have
a net life-time production of m × L = 200 × 2% = 4 descendants. This
number is the so-called per capita reproductive rate (R) of the population.
It is also clear that since each individual leaves more than one descendant,
then the population will grow. But how exactly will that growth proceed?
Let us first assume that the numbers L and m remain constant in sub-
sequent generations, and that generations are discrete and non-overlapping.
This means that for every organism that was present at some point in time,
there will be R individuals present after one generation (4 in the example
above). Thus, the population size after one generation (N1 ) can be com-
puted from the initial population size (N0 ) as follows:

N1 = N0 × R

The population size after two generations can be found by multiplying this

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 11

900000

800000

700000

600000
Population size, N

500000

400000

300000

200000

100000

0
0 1 2 3 4 5 6 7 8
Generation no., t

Figure 2.1: Exponential growth with discrete, non-overlapping generations. The

plot shows the growth of a population with initial size N0 = 50 and per capita
reproductive rate R = 4.

number by R one additional time, and we therefore have:

N2 = N1 × R
= N0 × R × R
= N0 × R 2
More generally, after t generations we have that
Nt = N0 R t (2.1)
This type of relationship is called exponential growth. Figure 2.1 shows
how the population size will increase for a population with R = 4 and
N0 = 50. After only 7 generations the population size has increased to
more than 800,000. Note that R gives the rate of increase per generation,
and t therefore has to be measured in generations. Moreover, t can only
be changed in discrete steps of one full generation, giving the discontinuous
curve seen above.
The exponential growth model derived here assumes that generations
are discrete and non-overlapping. Typically, however, individuals do not
breed synchronously. Furthermore, it is often the case that offspring is pro-
duced not only once but several times during the life-span of an organism.

c Anders Gorm Pedersen, 2005
12 Molecular Evolution, lecture notes

900000

800000

700000

600000
Population size, N

500000

400000

300000

200000

100000

0
0 1 2 3 4 5 6 7 8
Generation no., t

Figure 2.2: Exponential growth with continuous reproduction and/or overlapping

generations. The plot shows the growth of a population with initial size N0 = 50
and instantaneous rate of increase r = 1.39 per generation (corresponding to a per
capita reproductive rate R = 4).

Accounting for the phenomena of continuous reproduction and overlapping

generations, requires slightly more complicated derivations but leads to mod-
els that are very similar to the one above. Without going into the details,
we may note that the growth in such situations can be described by the
following expression:
Nt = N0 ert (2.2)
Since births can now take place at any given time, the variable t can here
take on any real value (not just whole numbers). Furthermore, t can now
be expressed in any unit of time (hours, days, years, generations, etc.). An
example of continuous exponential growth is shown in figure 2.2.
The constant r is called the “instantaneous rate of increase” and has to
be expressed in units that match those of t (if t is measured in minutes,
then r has to be measured in “per minute”.) If r is expressed in units of
“per generation”, then the per capita life-time rate of increase R can be
found by: R = er .

• Exercise 2–1: The bacterium Escherichia coli has a generation time of

about 20 minutes when growing in rich medium (i.e., R = 2, one generation

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 13

time corresponds to 20 minutes). The weight of a single E. coli cell is

approximately 1 × 10−12 g . The weight of planet earth is approximately
6 × 1024 kg = 6 × 1027 g. Calculate how long it will take for a population
of 100 E. coli cells to grow to the point where the combined weight of the
bacteria is the same as the weight of the earth. Use the fact that equation
2.1 can be rearranged to give:
Nt
log N 0
t=
log R

2.2.2 Logistic Growth

As illustrated by exercise 2–1 the exponential growth model is over-simplified
in that growth will normally be limited by the finite resources available
(food, space, etc.). Exponential growth is therefore only seen for limited
amounts of time and under special circumstances. Examples include the
initial growth of bacterial cells in test tubes and the growth of larger or-
ganisms after entering an unoccupied ecological niche. The so-called logistic
growth model attempts to capture these limits to growth by having the rate
of increase depend on the population size (the rate drops as the population
size increases). Under this model, the population size will eventually reach
a plateau referred to as the “carrying capacity” and usually denoted K.
Logistic growth may be described by the following expression:
K
Nt =   (2.3)
K
1+ N0 − 1 e−rt

An example of logistic growth with carrying capacity K = 10, 000, rate of

increase r = 1.1, and initial population size N0 = 100, is shown in Fig.
2.3. Note how the population size initially seems to grow exponentially,
but subsequently levels off, finally converging on the carrying capacity. The
logistic growth model may be modified further to account for situations
where the effect of population size on the growth rate is not instantaneous
but has a time lag.

2.3 Genotype Frequencies and Growth

Above, we have started developing an understanding for how populations
grow. We will now move on to investigate what happens when a growing
population consists of a number of distinct genotypes. Let us first consider
the fate of two different alleles—A and a—that are present in a population
of haploid, asexually reproducing, exponentially growing organisms with
discrete, non-overlapping generations (i.e., a population whose growth is

c Anders Gorm Pedersen, 2005
14 Molecular Evolution, lecture notes

10000

8000
Population size, N

6000

4000

2000

0
0 2 4 6 8 10 12
Generation no., t

Figure 2.3: Logistic growth. The plot shows the growth of a population with
initial size N0 = 100, rate of increase r = 1.1, and carrying capacity K = 10, 000.

described by the equation Nt = N0 ∗ Rt ). The allele A is present in a

fraction p of all individuals at the time we start our examination. The other
allele, a, is present in the remaining fraction q. Note that p + q = 1. If we
denote the initial (total) population size by N0 , then the initial number of
individuals with alleles A and a are:

NA,0 = pN0
Na,0 = qN0

We again assume that the average, per capita reproductive rate (R) of the
entire population remains constant in subsequent generations. We further-
more assume that the two genotypes have the same growth rate (Thus,
RA = Ra = R). The average, per capita life-time reproductive rate of a
genotype is also referred to as that genotype’s “absolute fitness”. From
equation 2.1 we have the following expressions for the number of individuals
with genotypes A and a after one generation:

NA,1 = NA,0 R = pN0 R

Na,1 = Na,0 R = qN0 R

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 15

The total population size after one generation will therefore be

N1 = NA,1 + Na,1
= pN0 R + qN0 R
= N0 R × (p + q)
= N0 R

(since p + q = 1). We can now compute the frequency of individuals with

genotype A after one generation:
NA,1 pN0 R
p1 = = =p
N1 N0 R
The frequency of individuals with genotype A after one generation of
growth (p1 ) is therefore the same as the frequency of A at the outset (p),
and we can conclude that if different genotypes have the same reproductive
rate, then their proportions are not changed by asexual reproduction. (The
same is obviously true for allele a).
Note that this result holds regardless of whether the population size is
increasing (corresponding to R > 1) or decreasing (corresponding to R < 1),
as long as the different genotypes have the same absolute fitness, R. (If the
two genotypes do not have the same absolute fitness then there is natural
selection for the genotype with the higher value. We will return to that situ-
ation below). Importantly, constancy of genotype frequencies during growth
will be true for any type of organism with asexual reproduction, regardless
of whether the organism is haploid or diploid. In the latter case p would
refer to the frequency of a given diploid genotype. If all diploid genotypes
(AA, aa, and Aa) retain their initial frequencies, then the frequency of any
single allele (A and a) will again remain unchanged.
In this section, we have ignored the statistical uncertainty that will play
a role for small populations. We will return to chance effects and the phe-
nomenon of “genetic drift” in section 2.7.

2.4 Selection
Let us now consider the more interesting case where two haploid genotypes
do not have the same absolute fitness. We will again investigate an example
where the alleles A and a are present at a locus in a haploid organism that
reproduces as described above. Let us imagine that the absolute fitness of
genotype A is RA = 4, while that of genotype a is Ra = 2. Recall that
for organisms such as the one we are examining here, R is the product of
fecundity and survival rate. It is therefore possible that the difference in
fitness between the two genotypes is caused by differential fecundity, dif-
ferential survival rate, or both. Let us say, for instance, that genotype A

c Anders Gorm Pedersen, 2005
16 Molecular Evolution, lecture notes

20000
Genotype A
Genotype a

15000
Population size, N

10000

5000

0
0 1 2 3 4 5 6 7 8
Generation no., t

Figure 2.4: Natural selection in haploid organisms. Genotype A (initial number

NA,0 = 1) has the fitness RA = 4, while genotype a (Na,0 = 99) has a lower
fitness (Ra = 2). Differential exponential growth of two genotypes is one instance
of natural selection

has a fecundity of 200 offspring per generation, and a survival rate of 2%,
giving RA = 200 × 2% = 4. We may further imagine that genotype a has
a higher fecundity (400 offspring per generation) but a much lower survival
rate (0.5%) resulting in an overall fitness that is half that of genotype A
(Ra = 400 × 0.5% = 2).
The number of individuals with genotype A therefore grows faster than
the number of individuals with genotype a. An example of this is shown
in Figure 2.4. Table 2.1 gives the corresponding genotype numbers and
frequencies, and includes a few extra generations compared to the figure.
In this example, the initial population consists of one single individual with
genotype A (perhaps a newly created mutation), and 99 individuals with
genotype a. It can be seen how the proportion of individuals with genotype
A rapidly increases from the initial 1%, and after 7 generations A is the
predominant genotype. After only 10 generations genotype A makes up more
than 90% of the population, and intuitively it seems to be approaching 100%
asymptotically (Table 2.1). But instead of guessing, we should of course
develop a mathematical model that we can use to predict the genotype
frequencies at any time t.

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 17

Table 2.1: Differential growth. p: freq. of genotype A, q: freq. of genotype a.

t NA Na Ntot p q
0 1 99 100 0.01 0.99
1 4 198 202 0.02 0.98
2 16 396 412 0.04 0.96
3 64 792 856 0.07 0.93
4 256 1584 1841 0.14 0.86
5 1024 3168 4192 0.24 0.76
6 4096 6336 10432 0.39 0.61
7 16384 12672 29056 0.56 0.44
8 65536 25344 90880 0.72 0.28
9 262144 50688 312832 0.84 0.16
10 1048576 101376 1149952 0.91 0.09

Let us again define the initial frequencies of individuals with genotypes

A and a to be p and q respectively (p + q = 1), and denote the initial total
population size N0 . We therefore have the following numbers of individuals
with A and a at the time we start:

NA,0 = pN0
Na,0 = qN0

After t generations we have the following numbers:

NA,t = NA,0 (RA )t = pN0 (RA )t

Na,t = Na,0 (Ra )t = qN0 (Ra )t

Therefore the total population size at time t will be:

Ntot,t = NA,t + Na,t = pN0 (RA )t + qN0 (Ra )t

The frequency of individuals with genotype A after t generations is therefore:

NA,t pN0 (RA )t p(RA )t

pt = = =
Ntot,t pN0 (RA )t + qN0 (Ra )t p(RA )t + q(Ra )t

Where the last simplification was obtained by eliminating N0 . The ex-

pression can be further simplified by dividing both the numerator and the
denominator with (RA )t :
1
p(RA )t × (RA )t p p
pt = 1 = (Ra )t
= Ra t
(2.4)
)t
(p(RA + q(Ra )t ) × (RA )t p+ q (R t
p + q( R )
A) A

Ra
The term R A
is the so-called “relative fitness” of allele a. The relative fitness
of a genotype (usually denoted W ) is the fitness of that genotype relative

c Anders Gorm Pedersen, 2005
18 Molecular Evolution, lecture notes

Genotype A
Genotype a
1

0.8
Genotype frequency

0.6

0.4

0.2

0
0 2 4 6 8 10 12 14 16 18
Generation no., t

Figure 2.5: Change in genotype frequencies as a result of natural selection. Geno-

type A (initial number NA,0 = 1) has the fitness RA = 4, while genotype a
(Na,0 = 99) has a lower fitness (Ra = 2).

to a reference genotype (typically the genotype with the highest fitness). In

our example we therefore have that Wa = R Ra
A
= 42 = 0.5, and WA = R RA = 1.
A

Ra
Substituting Wa for RA in equation 2.4, we get:
p
pt = (2.5)
p + qWat

And here, finally, is our result: equation 2.5 enables us to compute how
natural selection changes the frequencies of genotypes A and a over time
(The frequency of genotype a can, for instance, be found by qt = 1 − pt .) An
important conclusion from equation 2.5 is that the effect of natural selection
only depends on the relative fitness. This means that we would get the same
change in frequency regardless of whether the absolute fitnesses of A and a
were, for instance, 10 and 5, or 6 and 3, or even 0.8 and 0.4 respectively.
The so-called selection coefficient (s) is often used instead of the relative
fitness W . The selection coefficient against the least fit allele is defined as
s = 1 − W where W is the relative fitness of the least fit allele. This means
that W = 1 − s and equation 2.5 can of course be rewritten by substituting
(1 − s) for Wa , if one is interested in expressing the frequencies in terms of
selection coefficients instead of relative fitness.

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 19

Figure 2.5 shows how the genotype frequencies change over time in our
example (i.e., when Wa = 0.5). A relative fitness of 0.5 corresponds to a
selection coefficient of s = 1 − 0.5 = 0.5. It can be seen that even though
genotype A has an initial frequency of only 1%, it has essentially reached
fixation after just 16 generations. It should be noted that a selection coeffi-
cient of 0.5 is quite high, but it is not unrealistic. For instance it has been
estimated that natural selection acting on the so-called melanic peppered
moths, that spread in industrial Britain during the 1800’s, involved a selec-
tion coefficient of approximately 0.3. It is believed that this selection was
driven by the dark, melanic forms being harder to detect on soot-covered
tree bark compared to the lighter, more easily spotted form of the moth.
Selection for drug resistance in HIV and pesticide resistance in mosquitoes
has also been reported to be of this magnitude.
Figure 2.6 shows another example where the a allele has a relative fitness
of 0.99 (corresponding to a selection coefficient s = 0.01). In this example
genotype A has essentially reached fixation after 1000 generations. It is
important to note that there are situations where natural selection will not
lead to fixation of one allele, but will instead act to maintain a certain level
of diversity at a locus. One example of this is when a diploid organism
that is heterozygous at some locus has higher fitness then either of the two
homozygotes.
There is one final issue we can investigate using this simple model of
selection. You may have noted from figures 2.5 and 2.6 that the frequency of
A appears to change rapidly when both genotypes are fairly common, while
it changes more slowly when one genotype predominates. Let us derive an
expression that sheds light on this problem. The frequency of genotype A
is initially p. Using equation 2.5, we can see that after one generation, the
frequency of A will be:
p
p1 =
p + qWa
We can rewrite this expression using that p = 1 − q and that Wa = (1 − s):
p p p
p1 = = = (2.6)
(1 − q) + q(1 − s) 1 − q + q − sq 1 − sq

This expression tells us the frequency after one generation has passed. The
amount of change during one generation can now be seen to be:
p
∆p = p1 − p0 = −p
1 − sq

Multiplying p by 1−sq
1−sq (i.e., multiplying by 1) allows us to collect all the
terms in one single fraction:

p p(1 − sq) p − p(1 − sq) p − p + spq

∆p = − = =
1 − sq 1 − sq 1 − sq 1 − sq

c Anders Gorm Pedersen, 2005
20 Molecular Evolution, lecture notes

Genotype A
Genotype a
1

0.8
Genotype frequency

0.6

0.4

0.2

0
0 200 400 600 800 1000 1200
Generation no., t

Figure 2.6: Change in genotype frequencies as a result of natural selection. Rela-

tive fitness of genotype a is 0.99 corresponding to a selection coefficient of s = 0.01

which can finally be reduced to:

spq
∆p = (2.7)
1 − sq

Since 1 > s > 0 (genotype a has a lower fitness than genotype A), it can be
seen from equation 2.7 that ∆p must be positive. This is consistent with the
fact that genotype A is more fit than genotype a and that selection should
therefore act to increase its frequency. The change in frequency of the less
fit genotype is simply:
−spq
∆q = −∆p = (2.8)
1 − sq

Plotting ∆p as a function of p for a given value of s, results in a single-

humped curve with a maximum in the middle (for an example see Fig. 2.7).
The exact location of the maximum depends on the value of s, but for
small s, ∆p is essentially driven by the numerator of this equation and will
be highest when p ≈ q. We will use equation 2.7 below when considering
the interplay of mutation and selection.

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 21

0.40

0.35

0.30
Frequency change (delta-p)

0.25

0.20

0.15

0.10

0.05

0.00
0 0.2 0.4 0.6 0.8 1
p

Figure 2.7: Change in p (i.e., the frequency of genotype A) during one generation,
as a function of the current value of p. In this example the selection coefficient
against allele a is s = 0.8

2.5 Mutation-Selection Balance

The model of selection developed above, shows how an advantageous allele
(A) will spread through a population and eventually reach fixation. Sim-
ilarly, selection will act to eliminate the disadvantageous allele (a) from
a population. But what if the disadvantageous allele is constantly being
produced at some rate by mutation? In that case selection cannot remove
the allele entirely and instead its frequency will eventually reach a constant
equilibrium value. (If the frequency of the disadvantageous allele falls be-
neath the equilibrium value then mutation will act to increase it again. If
the frequency of the disadvantageous allele becomes too high, then natu-
ral selection will lower it). We are interested in computing the equilibrium
frequency.
Let us again consider a haploid, asexual organism where we are exam-
ining the two alleles A and a, and where genotype a has a lower fitness
than genotype A. For instance, genotype a may have a relative fitness of
Wa = 0.95 corresponding to a selection coefficient s = 0.05. Now, let us
also imagine that allele A mutates to allele a with mutation rate m (we
can imagine, for instance, that A corresponds to a functional enzyme while

c Anders Gorm Pedersen, 2005
22 Molecular Evolution, lecture notes

a refers to all non-functional forms of the enzyme). Let us set m = 10−7

mutations per A-gene per generation. This is a fairly typical mutation rate
1
for a gene, and means that in any one generation 10,000,000 of all A-alleles
will mutate to an a-allele.
We will now derive an expression that gives us the frequency of the
disadvantageous allele a at equilibrium. We will do this by first considering
how the forces of mutation and selection combine to change the frequency
of A during one single generation. To simplify matters, we will assume that
the effects of selection and mutation act sequentially1 :

G1 (p) −−−−−→ G2 (p∗ ) −−−−−→ G2 (p1 ) −−−−→ ...

selection mutation

Here, p is the frequency of A in generation 1 organisms just prior to repro-

duction; p∗ is the increased frequency of A in generation 2 after selection
has acted (via fecundity and survival rate); p1 is the final frequency of A
in adult generation 2 organisms after one full life-cycle has elapsed, and
mutation has decreased the frequency of A (thereby producing more a).
To find the equilibrium value, you should first recall that equation 2.6
(page 19) tells us that the natural selection step will change the frequency
of genotype A as follows:
p
p∗ = (2.9)
1 − sq
We now need to figure out how the mutation step will change this frequency.
Note that the mutation rate m gives the fraction of A alleles that will change
into a in the course of one generation. The fraction of A individuals that
remain A must therefore be 1 − m and we consequently have that:

p1 = p∗ (1 − m)

Substituting the expression in equation 2.9 for p∗ gives us:

p(1 − m)
p1 = (2.10)
1 − sq

This expression tells us how the forces of selection and mutation combine
to change the frequency of A (from p to p1 ) in one generation. We can now
find the equilibrium frequency by using the following insight: at equilibrium
the frequency of A is constant. From this it follows that the frequency of
A after one generation (p1 ) must equal the frequency of A in the previous
life-cycle (p). Substituting p for p1 in equation 2.10 gives us:

p(1 − m)
p=
1 − sq
1
This derivation follows Felsenstein. See http://evolution.genetics.washington.
edu/pgbook/pgbook.html

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 23

Multiplying either side of this equation by (1 − sq) gives us

p(1 − sq) = p(1 − m)

We can now eliminate p from this equation and isolate the equilibrium fre-
quency of a:

1 − sq = 1 − m

sq = m
m
q= (2.11)
s
Equation 2.11 shows us that the equilibrium frequency of the disadvanta-
geous allele depends on only the mutation rate and the selection coefficient
in a very simple way. In our example we find that the equilibrium frequency
of a is:
m 10−7
q= = = 2 × 10−6
s 0.05
The conclusion from the analysis presented here, is that when both se-
lection and mutation are acting at the same time, then a constant and
predictable level of genetic diversity will be maintained in the population.

2.6 Mathematical Modeling: A Few Thoughts

The science of population genetics deals with understanding the genetic
variation within populations, and the forces that change this variation. As
you should know by now the use of mathematical models is important in
this field. Mathematical modeling will also play an important role later in
this course when we discuss phylogenetic reconstruction and it is therefore
relevant at this point to briefly consider the subject. First, I would like to
make two important points about the nature of models:

1. A mathematical model is an explicitly stated hypothesis

A mathematical model that describes a biological system can be thought
of as a very explicitly and stringently phrased hypothesis about how
that system works. The explicitness of a hypothesis stated this way,
means that it is possible to make very detailed predictions about how
the investigated system will behave under different conditions. As is
the case for predictions based on qualitative hypotheses, such quan-
titative predictions can be checked against real world data, and the
model modified or abandoned for a more suitable model if necessary.

c Anders Gorm Pedersen, 2005
24 Molecular Evolution, lecture notes

2. A mathematical model does not represent full reality

It is typically not possible to represent full reality in a mathematical
model. For instance, even the intuitively reasonable growth models
considered above, assumed that there was a well defined (and con-
stant) average fitness for a given genotype. In reality, both fecundity
and survival rate depend on a large number of interacting biological
and non-biological, internal and external factors for each individual in
the population. Some of these factors will be stochastic (falling trees,
heart attacks, infection, etc.). To approach full reality in the model we
would therefore need to model fecundity and survival rate of all indi-
viduals, and for each individual these would be complicated functions
of huge numbers of different terms. It should be clear that for most
biological systems, representing full reality in a mathematical model is
impossible. Fortunately, that’s not something we are interested in do-
ing in the first place! For a typical biological system, we will instead be
more interested in finding an approximating model that captures the
most important features, and allowing us to understand the dynamics
of the system.

2.7 Genetic Drift

In the discussion presented above we have been using models that are “deter-
ministic”. Deterministic models are characterized by always giving a specific
result when starting from a specific set of conditions. For instance, when we
investigated exponential growth above, it was assumed that if a population
has growth rate R = 4 and initially consists of N0 individuals, then there
will be exactly 4 × N0 individuals in the next generation. Based on this as-
sumption we showed that the proportions of different genotypes will remain
constant provided that the genotypes have the same fitness (section 2.3).
This is, however, a simplification. Different individuals will not leave
exactly the same number of offspring, and death will also not remove exactly
the same fraction of different broods. This means that for every generation
there is some chance that the frequency of a genotype will change, even
though it has the same fitness as all other alleles at the locus (alleles with
the same fitness are said to be “neutral”).
The frequency of an allele has an equal chance of changing to a higher
or a lower value. This is true for every consecutive generation, regardless
of what the frequency has been at any previous time. Imagine for instance
that an allele has changed from its initial value of p = 0.3 to p = 0.26. In
the next generation the chance of ending with a frequency that is higher
than 0.26 will be the same as the chance of ending with a frequency that
is lower than 0.26. If, after 100 generations, the frequency has changed to

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 25

0.0001, then there will still be the same chance for the frequency to either
increase or decrease in the next generation. If this fluctuation continues for
sufficiently long then p will eventually wander to either 0 or 1. Once that has
happened the allele frequency can no longer change (at least if we assume
that there is no mutation and no migration from other populations).
The process of random change in genotype frequencies is called genetic
drift. From the discussion above, it can be seen that genetic drift (on its own)
tends to reduce the level of genetic variation in a population. This is similar
to the effects of selection described in section 2.4, but in the case of fixation
by drift, the fixed allele will not be advantageous compared to the lost alleles.
In fact, fixation will be the result of entirely random processes, and different
alleles will be fixed in different populations. The DNA sequences in isolated
populations will therefore tend to drift apart over time.

2.8 Chance of Fixation by Drift

We will consider the fate of individual genes in a population of haploid,
asexually reproducing organisms subject to only genetic drift. Let us as-
sume that the number of individuals (N ) is constant. For instance, we can
1
imagine that the average fecundity is 200 and the average survival rate 200 ,
meaning that each individual leaves one offspring per generation. This is
only the average however. Some individuals will leave no offspring, some will
leave one, and some will leave two or more, resulting in a gradual change in
genotype frequencies by genetic drift. Let us further imagine that every indi-
vidual in the population initially has a distinct genotype (A1 , A2 , · · · , AN ).
The frequency of each genotype is therefore initially N1 . According to the
argument above, one of the individual genes in the population will eventu-
ally reach fixation. Since this process is entirely random the different alleles
must all have the same chance of fixation, and we can conclude that the
chance that any particular allele (A5 for instance) is fixed, must be N1 . Sim-
ilarly, if an allele is initially present in x copies, then it has the probability
x 1
N of eventually being fixed (since each of the x copies have probability N
of being fixed). Generally, the probability that an allele Ai will be fixed is
the same as the frequency pi of that allele .

2.9 Drift and Neutral Mutation

Genetic drift ultimately results in the loss of genetic diversity, but since this
loss is the result of random fluctuations, drift is not a particularly strong
or fast-acting evolutionary force. It can be shown that, on average, it takes
2N generations for a new mutation to reach fixation in a haploid, asexually
reproducing population of N individuals. The process takes 4N generations
for diploid, sexually reproducing organisms. We will not prove this but

c Anders Gorm Pedersen, 2005
26 Molecular Evolution, lecture notes

Genotype
frequency

A2

A4

A3
A1

Time
t1 t2 t3 t4

Figure 2.8: Genetic drift is counterbalanced by mutation resulting in the

dynamic maintenance of genetic variation at a locus.

only note that, for some organisms, 2N (or 4N ) generations is a very long
time indeed. (This obviously depends on both the population size and the
generation time). You should compare these time spans to the speed with
which natural selection leads to fixation of an advantageous allele (section
2.4).
The loss of genetic diversity caused by genetic drift is, however, coun-
terbalanced by the constant production of new mutations. The net result is
a dynamic equilibrium where the population maintains a certain amount of
variation, but the specific alleles making up this variation are changing over
time. This process is illustrated in figure 2.8 where we follow the frequencies
of alleles at a specific locus. Initially, alleles A1 and A2 are present. At time
t1 , allele A3 is produced by mutation. Allele A1 is lost by drift at time t2 ,
and a new allele (A4 ) subsequently arises by mutation at time t3 . Later,
allele A3 is again lost by drift. Note that the average level of variability
at the locus remains roughly constant over time, but that the actual alleles
(and their frequencies) accounting for this variability changes over time.
We will now consider some aspects of how genetic drift interacts with
selectively neutral mutations that are generated at a constant rate. Let us
again assume that we are examining a haploid, asexual population with a
constant size of N individuals. Mutations are constantly being produced at
a rate µ. This rate is fairly constant and is perhaps mostly controlled by the

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 27

Genotype
frequency

2N 1/u
Time

Figure 2.9: The average time it takes for a neutral mutation to reach fixation
(in a population of haploid organisms) is 2N . The average time between the
fixation of different alleles at a locus is 1/u.

interplay between the error rate of the DNA polymerase during replication
and the activity of DNA repair systems. A certain fraction f0 of mutations
are neutral. These are produced at the rate u = f0 µ. Note that u only refers
to the neutral mutation rate and that this is lower than the total mutation
rate. Most newly arisen neutral mutations are immediately lost due to
genetic drift, but some eventually become fixed. We are now interested in
determining the over-all rate at which neutral mutations become fixed in the
population. This “rate of fixation” tells us how quickly the DNA sequences
of two isolated populations drift apart.
The number of mutations produced per generation (at the locus we are
examining) is N u. For instance, if u = 2 × 10−7 mutations per generation
for this locus, and if N = 106 then an average of 2 × 10−7 × 106 = 0.2
new mutations will be produced at this locus per generation in the entire
population (corresponding to one new mutation every five generations). In
the previous section we found that a single gene in a population of N haploid
individuals has the probability N1 of being fixed by genetic drift. This must
therefore also be the probability a newly arisen neutral mutation has of
becoming fixed, since the mutant allele will initially be present as a single
copy among a total of N genes.
Recall that the rate of fixation is the number of new mutations that
become fixed in a given population per generation (or any other unit of
time). We can now determine this rate simply by multiplying the number of
mutations produced per generation (N u) by the probability that a mutation
eventually becomes fixed ( N1 ). Denoting the rate of fixation by k we therefore

c Anders Gorm Pedersen, 2005
28 Molecular Evolution, lecture notes

have:
1
k= uN = u (2.12)
N
This simple but slightly surprising result shows that the rate, at which neu-
tral mutations become fixed, is independent of the population size. More-
over, the rate of fixation is simply equal to u - the neutral mutation rate.
Note that the average time between fixation of different alleles at a locus
is t = k1 = u1 . In our example from before, we have that k = u = 2 × 10−7
fixed mutations per generation. The average time between fixations is there-
fore t = k1 = 2×10 1
−7 = 5, 000, 000 generations per fixed mutation (at this
locus). You should distinguish between the average time it takes for a mu-
tation to reach fixation (2N generations in a haploid) and the average time
between fixation of different mutants ( k1 ; figure 2.9).

2.10 The Neutral Theory

Different ideas about the degree of polymorphism in real populations have
been entertained at various times in the history of evolutionary theory. Dur-
ing the early days of the modern synthesis, it was generally believed that
natural selection very quickly removed any disadvantageous alleles, and that
a single predominant allele (the so-called “wild-type”) was present at most
loci. Occasionally an advantageous mutation would arise, and it would then
very quickly be brought to fixation, replacing the previous wild-type in the
process. This viewpoint is now referred to as the “Classical School”.
In contrast to this view, the so-called “Balance School” believed that an
appreciable amount of polymorphism was present in real populations. It was
believed that polymorphism was being actively maintained by natural selec-
tion. One way in which selection can maintain several alleles in a population
of diploid organisms is if heterozygotes are more fit than homozygotes, but
there are also other selection-based scenarios with this outcome.
According to both schools of thought, essentially all evolutionary change
(meaning change in genotype frequencies) was brought about by natural
selection. The phenomenon of drift was not believed to play a significant
role, since it was assumed that two alleles were very unlikely to have the
same fitness. At the time it was not possible to directly measure molecular
diversity.
When the first electrophoretic studies of protein polymorphism were pub-
lished in the 1960’s, the level of genetic diversity was much higher than an-
ticipated by adherents of either school of thought. The classic hypothesis
was obviously wrong (as there was in fact a great deal of polymorphism at
many loci), but even the balance theory did not seem to be able to account
for the observed levels of polymorphism. This led several authors (among
them most prominently Motoo Kimura) to propose that perhaps most of

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 29

the observed molecular polymorphism was in fact neutral and therefore had
no effect on fitness.
According to this so-called “Neutral Theory” of molecular evolution most
mutations are disadvantageous and are quickly removed by natural selection,
a vanishingly small proportion are advantageous and are quickly brought to
fixation, while the vast majority of fixed (and therefore observed) muta-
tions are selectively neutral. That most mutations are disadvantageous and
rarely observed is in agreement with the previously prevalent views (now
referred to as “selectionist”). Selectionists and neutralists also agree that
adaptation must be the result of advantageous mutations that are brought
to fixation by natural selection. The main point of difference concerns the
fraction of mutations that are advantageous: the extreme selectionist view
is that almost all observed mutations are advantageous, while the neutralist
believes that practically all observed mutations are neutral with respect to
fitness. Today, we have many examples of mutations that appear to have
been fixed by natural selection, but there is also a great deal of evidence for
the importance of neutral mutation and genetic drift. The truth probably
lies somewhere between the two extreme viewpoints.

2.11 The Molecular Clock

In addition to the issue of the surprisingly high level of polymorphism, an-
other observation was also taken as evidence for the neutral theory—the
constancy of the rate of molecular evolution.
If a particular protein is examined in a number of species, then it is—for
each pair of species—possible to determine (1) the approximate time since
the species diverged, and (2) the number of differences between the protein
sequences. Strikingly it was observed that if divergence time was plotted
against genetic distance for many pairs of species, then the points would fall
on a straight line. Figure 2.10 shows an example where the number of differ-
ences have been computed from a collection of 4 different proteins (α-globin,
β−globin, cytochrome c, and fibrinopeptide A) from various mammalian
groups. (Note that here the difference between sequences is expressed as
replacements per site. This simply means that the number of differences
have been divided by the total length of the protein).
Since genetic distance is approximately proportional to divergence time,
it appears that molecular evolution must be proceeding at a roughly constant
rate. In figure 2.10 the distance between pairs of sequences is increasing at
a rate of approximately 0.006 amino acid replacements per site per million
years (i.e., for every 1000 amino acids in the protein, 6 differences accumu-
late every 1 million years). If this is true of most molecular evolution, then
sequences may be used to estimate approximate divergence times for species
that lack an informative fossil record.

c Anders Gorm Pedersen, 2005
30 Molecular Evolution, lecture notes

Figure 2.10: The molecular clock. Genetic distance (number of amino acid
replacements per site) plotted against a geological estimate of divergence
times for various pairs of mammalian groups. Combined sequence data
from α-globin, β−globin, cytochrome c, and fibrinopeptide A. (From Graur
and Li, 2000, Molecular Evolution, 2nd edition, Sinauer Associates Inc.)

An approximately constant rate of molecular evolution is exactly what

would be expected if most mutations are neutral: as was shown in section
2.9 above neutral mutations are fixed at a constant rate k regardless of
population size. (Note that, obviously, neutral mutations are not produced
at a perfectly constant rate—they appear at random intervals. But if they
are observed over sufficiently long periods of time, then the rate of change
will appear to be approximately constant. This has been described as a
“stochastic clock”).
This rate constancy would, however, not be expected if the selectionist
scenario is correct. If most substitutions are the result of natural selection,
then we would assume that the rate of evolution is heavily influenced by
environmental change (where the word “environment” includes the impact
of other living organisms). Intuitively, it seems unlikely that the rate of
environmental change is sufficiently stable to produce the constant rate of
molecular evolution that has been observed in a wide range of organisms,
and over long periods of time.
There are a number of things that should be noted with regard to the
molecular clock. First, molecular clocks do not run at the same speed in
different proteins. Generally, it appears that less constrained proteins evolve

c Anders Gorm Pedersen, 2005
Chapter 2: Brief Introduction to Population Genetics 31

faster. Secondly, things are not quite as tidy as figure 2.10 implies. There
are many examples where evolution does not proceed at a constant rate.
However, it is probably fair to say that all in all, the examples that we
do have of rate constancy are sufficiently striking to require some sort of
explanation. Finally, selectionist explanations for the molecular clock have
also been proposed, although generally these seem to be slightly ad hoc and
unsatisfactory.

c Anders Gorm Pedersen, 2005