0195168178_0032-0048_Fox-03.

qxd 11/17/05 5:01 PM Page 32

3
Mutation

DAVID HOULE
ALEXEY KONDRASHOV

utation is the ultimate source of all genetic at the DNA level both natural and important. The
M variation, and genetic variation is absolutely
necessary for any sort of evolution to proceed
focus at this level is on the rate at which various sorts
of alterations occur. It is also necessary to understand
(Wayne & Miyamoto, Ch. 2 of this volume). These the effects that these alterations have on the pheno-
two facts should make the study of mutation the type of the organism. This latter aspect of mutation
foundation of evolutionary genetics. Unfortunately, is a key to resolving one of the major controversies
this is very far from the case. Both the processes of in evolutionary biology: whether evolution is limited
mutation and the pattern of effects of those muta- by the supply of genetic and phenotypic variation
tions are relatively little known compared with the (Gould & Lewontin 1979), or by natural selection
properties of the standing genetic variation within (Charlesworth et al. 1982).
populations and among populations and species. The difficulties of the study of mutation stem
However, these properties can only be understood largely from one simple fact: individually, mutations
in light of mutation. are very rare events. This has always made the direct
Two classes of mutation are particularly impor- study of mutation, consisting of recording new
tant to evolution: those with beneficial and those mutations as they arise, exceptionally tedious. For
with deleterious effects on fitness. Evolutionary example, one of the best direct studies of mutation
change depends on the input of beneficial muta- in mammals is that of Russell and Russell (1996),
tions. Unfortunately, such mutations are usually quite who report examinations of over 1,000,000 mice
rare and hard to study. The vast majority of muta- to find just 46 visible mutations.
tions that affect fitness decrease it. Because these The alternative to this tedium is to fit a model to
mutations are common, their effects in total can have genetic variation within a population or to the vari-
very large evolutionary consequences. Such disparate ation found among populations or species. These
phenomena as inbreeding depression, sexual selec- model-based approaches utilize data on contempo-
tion (Moore & Moore, Ch. 22 of this volume), rary variation, which are relatively easy to gather, to
recombination and sexual reproduction, and senes- obtain information about mutations that happened
cence (Promislow & Bronikowski, Ch. 30 of this over a considerable period of time. Consequently,
volume) are all affected by, or even explained by, the much of what we know about mutation comes from
commonness of deleterious mutations. model-fitting. Such efforts have a very important
Genetic information is encoded in nucleic acids; Achilles heel––if the model is not correct, the results
genetic variation is created when the sequence of can be very misleading. A model is necessary because
these is altered. In all cellular organisms DNA is the the variation studied, while certainly due to muta-
genetic material. This makes the study of mutation tion, has also been filtered by natural selection and

32
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 33

Mutation 33

genetic drift. This tension between direct observa- variant is often referred to as a mutant. This is
tion and indirect, assumption-based approaches is justified for deleterious variants, which are usually
very common in evolutionary biology. lost due to natural selection, and thus cannot be
ancestral; in contrast, a low-frequency variant may,
nevertheless, be ancestral. Genetic variants are vari-
CONCEPTS ously called alleles when referring to alternate forms
of a gene or haplotypes when referring to longer
DNA sequences. We will adopt the term “variant”
What is Mutation?
as the most general, and least loaded term.
Most, but not all, changes in DNA sequence are Replication, repair, and recombination of
mutations. The exception is reciprocal recombina- DNA sequences are incredibly complex molecular
tion, where both homologous DNA sequences are processes, each involving interactions of dozens of
broken and rejoined at the same point. This does macromolecules. However, we will ignore biophys-
change the DNA sequence, but its reciprocal nature ical and biochemical aspects of these processes,
conserves information. Small-scale alterations of and adopt a simple transmission genetics approach,
one or a few base pairs are mostly caused by errors concerned with inputs, outputs, and rates.
in DNA replication and repair, and thus are unam-
biguous mutations. Sequences that emerge after
Classification of Mutations
large-scale changes often contain clearly identifiable,
rearranged pieces of old sequences, so that recom- Mutational events may affect anything from 1 base
bination as well as errors per se may be involved. pair to entire genomes. Small-scale mutations that
Many of these changes that can be viewed as either affect only a few nucleotides are categorized in
mutation or nonreciprocal recombination. Such Table 3.1. Small-scale mutation depends on the local
processes will be considered here, because they share sequence context. The most important aspect of
the rarity and irregularity of small-scale mutations. context is whether the sequence consists of repeated
This definitional difficulty carries over to the sequences (for example AGAGAG, known as “peri-
language of variation. A bit of DNA sequence that odic”), or a more typical sequence where the progres-
has been changed from the copy in its parent is sion of base pairs is more or less unpredictable
clearly a “mutation,” but when two sequences differ, (“complex”).
we cannot assume without direct knowledge of their Mutations in short periodic sequences, called
ancestry, that one variant is the “mutation” while micro- and minisatellites, are mostly deletions
the other is “ancestral,” “normal,” or “wild-type,” and insertions, usually of lengths that are multiples
although in practice a deleterious or low-frequency of the period length. They often occur with rates

TABLE 3.1. Classification of mutational events involving small numbers of base pairs
Event l1a l2 Example Description

Nucleotide substitution 1 1 AGC → ATC Any single base pair change

Transition 1 1 AGC → AAC Base pair substitution of a purine (A or G) with
ACC → ATC another purine, or of a pyrimidine (T or C) with
another pyrimidine. More common than transversions
Transversion 1 1 AGC → ATC Base pair substitution of a purine with a pyrimidine,
or a pyrimidine with a purine. Less common than
transitions
Deletion ≥1 0 AGGC → AC One or several successive nucleotides are removed
Insertion 0 ≥1 AC → AGGGC Insertions of one or several successive nucleotides
Complex events — 1 >1 AGC → ATTC Combined indel/substitution very rare. Gene
combine substitution >1 1 AGGC → ATC conversion (see Table 3.2) can cause complex
and deletion or insertion changes
2 2 AGGC →ATTC Simultaneous substitutions do occur at appreciable
rates
a
l1 is the length of the affected sequence before mutation; l2 is the length of the sequence after mutation.
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 34

34 Principles of Evolutionary Genetics

that are orders of magnitude higher than those Large-scale mutations are categorized in
of mutations in complex sequences. Microsatellites Table 3.2. With the exception of long deletions, these
are unlikely to have important functions, but changes usually involve recombination of pre-existing
their frequent mutations provide abundant data for segments of sequences. The reason for this is that
tracing ancestry. Even within complex sequences, de novo origin of a long DNA sequence is a rather
the context can affect mutation rate significan- unlikely occurrence. Thus, sequences added by a
tly (Kondrashov & Rogozin 2004). An important large-scale mutation are usually copies, often modi-
case is the elevated mutation rate in mammals fied, of pre-existing sequences. The probability that
of the dinucleotide CpG. CpG pairs consist, in such a sequence (which can code for a protein
both strands, of a cytosine followed by a guanine, domain, or even an entire protein) would be func-
where p refers to the phosphate that joins them. tional is substantial (Thornton, Ch. 11 of this
The impact of a local context is so important volume). Creation of homologous sequences by
that, say, insertions into complex sequences versus. duplication can catalyze further large-scale events, as
microsatellites, or transitions within versus. outside it creates the opportunity for nonreciprocal recombi-
CpG are often considered as separate types of nation. Transposable elements provide a particu-
mutations. larly important example, as they can also lead to

TABLE 3.2. Mutational events involving large numbers of base pairs

Where does new
Event sequence come from? Example Details

Deletion — A[S1]G → AG Removed sequence can be very long.

Rates increase if flanking sequences are
similar
Tandem duplication Neighboring sequence A[S1]G→ A[S1][S1]G S1 may be a rather long sequence,
sometimes a large proportion of the
chromosome
Nontandem Non-neighboring AG → A[S2]G S2 is a sequence of any length from
duplication LGT sequence elsewhere in the genome. In practice only
detected when S2 is > 20 bp
Transposable TE AG → A[T1]G T1 is DNA derived from a TE. T1 often
element (TE) consists of a fragment of TE sequence.
insertion Rate may be quite high. A common mode
of nontandem duplication. Often insertion
is accompanied by other changes (i.e.,
short duplications)
Lateral gene transfer Other genome AG → A[S3]G Important in prokaryotes
Inversion A[GT … CC]A → Rotates a sequence
A[GG … AC]A
Gene conversion Homologous Sequence of one homolog converted to
sequence that of the other
Transposition Sequence changes Usually through breakage and fusion of
location chromosomes
Chromosome break None Usually deleterious, but may increase
chromosome number
Chromosome fusion None Usually deleterious. May decrease
chromosome number
Aneuploidy Chromosomal Usually lethal, or very deleterious
duplication or loss
Polyploidization Same genome: Fixed more frequently in plants than animals
autopolyploidy
Different genome: Often caused by hybridization. Fixed more
allopolyploidy frequently in plants than animals
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 35

Mutation 35

small duplications through transposition (Petrov & of a variant might depend on the genotype it finds
Wendel, Ch. 10 of this volume). itself in (epistasis).
In diploid organisms, an important additio-
nal consideration is the dominance of a variant.
Classification of Phenotypic
Dominant mutations have their full phenotypic
Effects
effects when present in heterozygous condition, while
Variant DNA sequences may or may not have recessives only affect the phenotype in homozygous
an effect on the phenotype of their carriers. Those condition. If the mutant heterozygote is intermedi-
that do may have a cascading series of effects, from ate between the two homozygotes, the mutant is
the regulation or function of a gene, through the partially dominant. If the heterozygote is exactly
biochemical, developmental, or physiological levels, intermediate, the variant is said to act additively.
and ultimately on the readily observable phenotype The concept of dominance is itself phenotype-
of the organism. It is useful to distinguish loss- specific. For example, Mendel’s wrinkled pea allele
of-function variants, for example a frameshift dele- is recessive when the smoothness of the seed coat is
tion, from those mutations that leave an altered examined, but when the amount of starch in the
functional gene, such as an amino acid substitu- seed is measured, the heterozygote is intermediate
tion. Mutations that alter function usually result in between the two homozygotes. When subjected to
quantitative changes in amount or timing of expres- quantitative analysis, the vast majority of even major
sion. Occasionally such changes may lead to a qual- variants seem to be neither completely recessive or
itatively different effects, such as new substrate dominant. Variants with minor phenotypic effects
specificity of an enzyme. tend to be partially dominant, and are often nearly
The most important phenotype is fitness, additive.
the capacity to produce offspring, and so includes
survival and the ability to breed and reproduce. The
How to Study Mutation
effect on fitness of a variant sequence is an impor-
tant determinant of its evolutionary fate. However, Classification of the types of mutations (Tables 3.1
fitness has surprisingly little effect on the fate of and 3.2) makes the study of mutation sound alto-
any single variant, as all start out rare where stochas- gether straightforward. It is important to realize that
tic effects are very strong (Gillespie, Ch. 5 of this our ability to study these different classes of muta-
volume). tions varies widely depending on their rarity and
Mutations can be conveniently classified the nature of their phenotypic effects. In Table 3.3,
according to their fitness effects into the following we classify the study of mutation according to two
categories: criteria suggested in Kondrashov (1998): the time
scale over which mutation is studied, and the type
1. Lethal mutations kill the individuals that of data that is used to detect mutations.
carry them. Three different classes of characteristics may be
2. Deleterious mutations reduce fitness relative used profitably to detect mutations: DNA sequences,
to alternative states, but not to zero. phenotypes, and fitness. Studies of mutation almost
Considerable evidence suggests that such invariably cover one of three time frames. First
mutations are more common than lethals. there is the direct study of mutation through compar-
3. Neutral mutations do not affect fitness ison of parents and offspring. At a slightly longer
much, either positively or negatively. These time scale, one can set up a mutation-accumulation
too are likely to be common. (MA) experiment. To do so, one maintains the
4. Advantageous mutations increase fitness, population under conditions that minimize the
and therefore will be favored by natural impact of natural selection on the fate of any muta-
selection. These are probably the rarest type tions that may arise (see Case Studies for examples).
of mutation. Finally, the comparative method infers mutation
rate from the rate of divergence between species.
This categorization is context-dependent. An The use of different time frames allows different
advantageous variant in one environment may aspects of mutation to be investigated. The chief
be neutral or deleterious in other circumstances reason for these differences is the degree to which we
(Scheiner, Ch. 21 of this volume). Also, the fitness can assume a realistic model for the interaction
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 36

36 Principles of Evolutionary Genetics

TABLE3.3. Categorization of approaches to the study of mutation, with reviews or

examples of successful studies
Generations separating samples

1 10 to 103 >105
Direct Mutation accumulation Comparative

DNA Weber and Wong 1993 Denver et al. 2004; Nachman and
Schug et al. 1997 Crowell 2000
Phenotype Kondrashov 2002 Houle et al. 1996 Lynch 1990
Fitness Woodruff et al. 1983 Mukai et al. 1972

between mutation and natural selection. With depends on the effectiveness of natural selection at
direct studies, the need for assumptions about natu- influencing frequencies; this depends on the size of
ral selection is minimal. Almost all types of muta- the population. In a population where N is small,
tions may be observed in the offspring. As the time genetic drift (luck) will be a relatively strong force,
frame of the study lengthens, the necessary assump- swamping out small differences in fitness. However,
tions about natural selection become more stringent. when N is large, even tiny differences in fitness reli-
Only mutation rates to neutral alleles can simply be ably discriminate higher and lower fitness variants.
inferred over long time periods. As a result, the same Mutation-accumulation experiments are therefore
data can lead to very different conclusions about the designed to maximize the impact of drift, either by
overall mutation rate, depending on the assumptions making N as small as possible, or by equalizing
chosen. family sizes (Shabalina et al. 1997). Thus, the influ-
Despite the complications in applying models to ence of natural selection is minimal in a direct study,
divergence data, the essential neutral theory behind somewhat higher in a mutation-accumulation study,
such models is easy to grasp. If we consider a popu- and very large in a comparative study. The result is
lation of genetic variants with no impact on fitness, that the neutral model can be applied to an uncer-
that is neutral variants, whether they are lost from tain and decreasing proportion of variants as the
the population or will become fixed (rise to a time scale of the study increases. Even at the DNA
frequency of 1) depends only on genetic drift, the level, it is difficult to be sure that a particular segment
luck of sampling during reproduction. Lucky vari- really evolves at the neutral rate. For example,
ants will become fixed; the vast majority will be evolutionary biologists have treated pseudogenes,
lost just by chance. The chance that each particular altered sequences derived from functional genes, as
variant will be fixed in the future is proportional to neutral (e.g. Nachman and Crowell in Case Studies,
its frequency right now: rare variants are likely to below). However, there are at least two possible
be lost, common ones likely to be fixed. Now, let us mechanisms for selection on pseudogenes. First,
consider the fate of each new neutral variant. If recombination between pseudogenes and their
there are N diploid individuals in the population, parent gene is deleterious, so deletions of pseudo-
each new variant starts out at a frequency of 1/2N, genes may be favored by natural selection. Second,
and thus has a chance of 1/2N of rising to fixation. the discovery of naturally occurring nonprotein-
On the other hand, with a mutation rate m per coding genes (such as micro-RNAs) that can regu-
gamete, the number of new mutations in each gener- late expression of their homologous genes suggests
ation is 2Nm. Multiplying these two together gives that some apparent pseudogenes may play such a
the surprisingly simple rate of neutral evolution: selected role.
k = 2Nm × 1/2N × m. This rate is the divergence A second major disadvantage of comparative
from the ancestral sequence; species diverge at studies is that the number of generations that sepa-
twice this rate because variants arise along both rate species is usually known only very approxi-
branches to the common ancestor. mately. As the time scale of any comparison becomes
In reality, variants have a range of effects on longer, these uncertainties become very large.
fitness from undetectable to lethality. Their ability The number of generations in a lineage since the
to persist in the population and so be detected also Mesozoic era will hardly be ever known with any
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 37

Mutation 37

confidence at all. As a result, comparative data are responsible for them, as explained below for
usually summarized as a mutation rate per unit time. human genetic diseases, they can be used to gain
While this may be useful in some contexts, such as very detailed information about mutation rates.
calibrating a molecular clock (e.g. Rodríguez - However, mutations with the very largest
Trelles et al. Ch. 8 of this volume), it does not tell phenotypic effects used in direct mutation studies
us what we want to know about mutation rates in are themselves rarely of evolutionary significance,
organismal terms. as they usually reduce fitness. Quantitative trait locus
These considerations would seem to make direct and developmental studies of species differences
studies preferable, were it not for the fact that the suggest that both detectable-if-you-know-what-
longer the time period, the greater the number of to-look-for and small effect variants are the major
mutational events that can be assayed. Because sort of variation that allows evolution. Their cumu-
mutations are individually rare, direct studies can lative effects are usually studied in a mutation-
only be informative when there is an efficient mecha- accumulation experiment. In most such experiments,
nism for screening enormous numbers of individu- an initially inbred genotype is replicated and selection
als for mutations. Such screening is available for on each replicate minimized by lowering N for each
many inherited phenotypes in human societies with replicate, for example by selfing or brother–sister
advanced health care. At the other end of the biolog- mating. The rate at which variation in the pheno-
ical spectrum, microbial populations can be rapidly type accumulates is used to measure the increase
screened for the converse sort of mutations that in phenotypic variance due to a single generation of
restore function at a defective gene (reviewed in mutation, VM. In addition, a change in the mean
Drake 1991). For other species, the direct data are of the accumulation lines indicates that mutations are
limited. Furthermore, because the phenotypic impact biased in their effects. For example, fitness and its
of most mutations is usually small, we need to be key components of viability, fecundity and mating
able to infer from the minority of mutations that ability, are maximized by natural selection, suggest-
are observable the properties of the full spectrum of ing that the mutations that arise will on average
mutations. decrease fitness. Information on VM and mutational
Of the three different classes of characteristics bias can sometimes be combined to give a very crude
that may be used to study mutation, DNA sequences estimate of the overall mutation rate of all genes
are the most conceptually straightforward. The chal- affecting fitness, as in Mutation Accumulation in
lenge with the use of sequence data is that care must Drosophila in the Case Studies section below.
be taken to account for the possibility of errors in
scoring. This has so far limited the use of sequence
Mutation Rates
data in direct or mutation accumulation studies. For
example, to detect a sample of base pair mutations, The most detailed picture that we have of mutation
which typically occur at a rate of 10−8 per genera- rates in eukaryotes is for humans. This is due to
tion, over a 100 generation mutation-accumulation several factors. First, human–chimpanzee is the only
experiment, one needs the ability to sequence many species pair for which the total number of genera-
more than 106 nucleotides with an error rate well tions since their divergence is reasonably well known,
below 10−6. The necessary methods are emerging facilitating comparative studies (see Nachman and
and are starting to be applied (Denver et al. 2004). Crowell in Case Studies, below). Second, data on
Exceptions are provided by sequences with espe- Mendelian diseases provide the only large-scale
cially high mutation rates, such as microsatellites phenotypic screenings for de novo mutations in any
or mitochondrial DNA. eukaryotic species for direct estimates (Kondrashov
The other two categories of data (phenotypes 2002). For our species, the comparative and direct
and fitness) refer to whole-organism characteristics. approaches suggest a very similar mutation rate
Fitness is, in some respects, just another phenotype, of about 2 × 10−8 per nucleotide per generation.
but is by definition under strong natural selection. The fact that the two methods give essentially the
Mutations with large effects on these phenotypes, same estimate is quite encouraging. Substitutions
such as genetic diseases in humans or visible and account for about 95% of this total, with short
lethal mutations in Drosophila, can be counted. insertions and deletions accounting for almost all
This is the basis for the direct studies in Table 3.3. of the remainder. Large-scale mutations are gener-
By connecting such changes to the DNA changes ally rare.
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 38

38 Principles of Evolutionary Genetics

This simple picture conceals substantial complex- mutation-accumulation studies, as it is possible to

ities based on the sequence that surrounds a partic- rapidly screen huge numbers of individuals for
ular site. In mammals, substitution rate at CpG novel phenotypes (Drake 1991). With far more
sites is elevated by a factor of more than 10. The investigator effort, substantial direct phenotypic
reason for this is that the C tends to be methylated, mutation assays have also been performed in
which misleads DNA polymerase. This confusion several model systems (Schalet 1960; Woodruff
results in a transition from C to T, and the destruc- et al. 1983; Russell & Russell 1996; Drake et al.
tion of the CpG context. Oddly, this results in a 1998). In all these studies, investigators first screen
lower mutation rate of noncoding DNA than in for phenotypic mutations at previously identified
coding DNA. In coding regions, natural selection loci. These values are then combined with the size
to maintain a particular sequence can preserve CpG of the locus and an estimate of what proportion of
pairs despite this high mutation rate away from all DNA changes will result in a mutant phenotype
them; in noncoding regions mutation pressure rapidly to arrive at an overall mutation rate. These estimates
destroys them. Duplicated sequences are another may be inaccurate; for example, the direct C. elegans
frequent source of mutational hot spots because mutation rate from sequencing is one order of
nonhomologous recombination between the similar magnitude higher than the estimate that Drake
sequences leads to large-scale mutations. For exam- et al. (1998) obtained using data from direct pheno-
ple, about half of the mutations causing hemophilia typic assays. Finally, there is a large amount of
A are caused by recombination of the gene with its comparative data from which mutation estimates
nearby pseudogene. can be calculated. As an example of this sort of
Another complexity in the human data is that it estimate we have used the calculations of Keightley
appears that mutation rates are substantially higher and Eyre-Walker (2000), who calculated rates based
in males than females (Drake et al. 1998), contrary to on the assumption that synonymous sites evolve at
the impression created by the well-known maternal the neutral rate.
age effect on Down’s syndrome. Part of this effect The data in Figure 3.1 make it clear that the
is probably due to the fact that the spermatogonia mutational properties of organisms are extremely
divide continuously throughout a male’s lifespan, different. To help interpret this variation, the genome
while oocytes essentially stop dividing before birth. sizes and number of replication events per genera-
It has been estimated that the number of divisions tion are shown in Table 3.4. Figure 3.1a shows that
in the germ line of a 30-year-old human is 31 for a mutation rates per base pair per generation vary
female and 400 for a male; this difference increases over three orders of magnitude. Viruses seem to sacri-
with paternal age. fice accuracy for speed, while multicellular organisms
These facts point up the difficulties in generalizing with long generation times accumulate mutations
about mutation rates across species. The methyla- over many cellular replication events. When the
tion that gives rise to the CpG bias is not universal, numbers of mutations over the whole genome are
and other biases undoubtedly arise in other groups. summed, it is clear that mutation rates on average
While human mutations are overwhelmingly single- rise with the genome size of the organism. Drake
base substitutions, mutations in Drosophila tend to (1991) called attention to the fact that viruses and
involve more short insertions and deletions, and unicellular organisms have a fairly constant muta-
more large-scale mutations. The details of gameto- tion rate per genome. It is now clear, however, that
genesis and details of life history, such as the average this relationship does not hold for multicellular
age at reproduction, can have a big effect on muta- organisms (Figure 3.1b; Drake et al. 1998).
tion rates, even when the cellular details of meiosis A second point concerning Figure 3.1 is that the
and replication remain the same. discrepancies between mutation rates obtained
With these difficulties in mind, data on muta- using different approaches can be very substantial.
tion rates in some other well-studied DNA-based The mutation rates obtained from assuming that
systems are shown in Figure 3.1. In the nematode third-base-pair positions evolve at the neutral rate
worm Caenorhabditis elegans mutation rates were are substantially lower than those obtained using
obtained by sequencing random nuclear sequences direct evidence in Drosophila and Mus. In humans,
in a mutation-accumulation experiment (Denver there is no such discrepancy, with synonymous rates
et al. 2004). Mutation rates are relatively well known being within a factor of 2 of the direct estimate. This
in viruses (phages) and bacteria from direct and difference may be caused by the larger population
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 39

Mutation 39

FIGURE 3.1. Estimated mutation rates in 10 well-studied taxa.

(A) Mutation rates per base pair per generation. (B) Haploid mutation
rates per genome per generation. Circles, direct and mutation accumu-
lation estimates; square, estimate assuming that divergence in pseudo-
genes is neutral (Nachman & Crowell 2000); triangle, estimates
assuming that divergence in synonymous base pairs is neutral (Keightley
& Eyre-Walker 2000).

sizes of Drosophila and Mus relative to that of there is about a 1% chance that a duplicate copy of
Homo, allowing a larger proportion of human gene will be fixed per million years (Lynch &
variants with very small deleterious effects to Conery 2000). For Drosophila melanogaster, this
evolve at the neutral rate. suggests that the fixation rate of duplications is
The above estimates do not reflect whole gene about 3 × 10−10 per gene per generation. This rate
duplications, which may be critical for long-term may be higher or lower than the actual duplication
changes in the genome (Petrov & Wendel, Ch. 10 rate, depending on whether duplications are delete-
of this volume; Thornton, Ch. 11 of this volume). rious (for example due to recombination between
Genomic data from several species suggest that duplicates) or advantageous. Bearing in mind that
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 40

40 Principles of Evolutionary Genetics

TABLE 3.4. Genome sizes, generation times and average number of replication events/life
cycle for taxa with mutation data in Figure 1.
Generation
Taxon Common name Genome sizes time (days)a Rep./gen.b

M13 phage 6.4 × 103 0.01 1

Lambda phage 4.9 × 104 0.01 1
T2 and T4 phages 1.7 × 105 0.01 1
Escherichia coli 4.6 × 106 0.01 1
Saccharomyces cerevesiae Brewer’s yeast 1.2 × 107 0.1 1
Caenorhabditis elegans Nematode worm 1 × 108 4 9
Drosophila melanogaster Fruit fly 1.7 × 108 12 30
Mus musculus House mouse 3 × 109 275 43
Homo sapiens Human 3 × 109 7300 215
a
These times are minimum generation times under ideal conditions. Such conditions may not be typical of those in nature.
b
Estimated number of replication events per generation.

a gene duplication will affect many nucleotides, the the remainder evolves slower than the neutral rate,
overall rate of duplications may affect as many or and is therefore functionally important (estimates
more base pairs as do base pair mutations. range from 15% to 3%; Shabalina et al. 2001;
Dermitzakis et al. 2002). Mutations affecting the
remainder are phenotypically silent and selecti-
Mutations and Fitness
vely neutral. Quantitatively, the fraction of neutral
Mutations that affect fitness are more likely to be sequences is smaller in compact genomes of bacteria
deleterious than advantageous. As a result, the rate (where over 80% of sequences code for proteins)
of evolution of DNA sequences is on average and much larger in eukaryotic genomes. From these
inversely related to their functional importance. figures, together with the data on per nucleotide
Within protein-coding genes, synonymous sites mutation rates referred to above, the 100 muta-
evolve faster than nonsynonymous sites, and homol- tions expected per human genome per generation
ogous exons are much more similar than homolo- can translate into from 4 to 14 deleterious muta-
gous introns. Direct estimates of the fitness effects tions (assuming that 50% of mutations in coding
of new mutations generally corroborate this by sequences are deleterious).
showing a decline in fitness during mutation accu- Direct evidence makes it clear that most mutations
mulation experiments (see Mutation Accumulation that affect fitness have relatively small deleterious
in Drosophila in Case Studies, below). This is hardly effects (see Mutation Accmulation in Drosophila in
surprising: spoiling the product of 3.5 billion of the Case Studies section below). A striking confir-
years of evolution is easier than improving it. mation of this fact is that systematic knockouts of
However, there are parts of some genes where genes in eukaryotes reveal that less than 30% of all
nonsynonymous substitutions occur faster than genes are essential to viability. For example, the
synonymous, suggesting that many replacements of fitness effects of knocking out nearly every gene in
amino acids were advantageous. For example, the the yeast genome have been measured. Under typi-
anitgen binding region of the HLA gene in the major cal laboratory conditions, only 18.7% of the genes
histocompatibility complex in humans and mice are essential, while quantitative decreases in fitness
has a higher rate of amino acid substitutions than are detectable in another 15% of the knockouts
synonymous substitutions. This suggests that positive (Giaever et al. 2002). Since any gene that is not
selection for diversity or at least change in antigen- capable of affecting fitness will rapidly be destroyed
binding makes advantageous amino acid changes by mutation, the remaining genes must either have
quite frequent (Hughes & Nei 1989). effects on fitness that are too small to be detectable,
Another key property of DNA sequences in or be advantageous under conditions not found in
eukaryotes is that large portions are irrelevant these experiments. Most spontaneous mutations
to fitness. Only approximately 2% of the human must have smaller effects on fitness than the whole-
genome codes for proteins, and only a minority of gene knockouts used in this experiment.
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 41

Mutation 41

Unfortunately, our ignorance regarding the muta- these mutations that are most difficult to detect and
tion rate to beneficial mutations is quite profound. study. These questions have so far been addressed
Recent work holds out hope that this ignorance is by relatively crude mutation-accumulation experi-
curable. For example, microbiologists have long been ments in which the aggregate properties of unknown
able to exploit the lack of recombination to detect numbers of mutations are studied.
fixation of beneficial mutations. When a novel bene- These studies make it clear that the effects
ficial variant arises, the genetic background in which of mutation differ with phenotype (Houle et al.
it occurs increases along with it. Thus, perturbations 1996). Figure 3.2 summarizes estimates of the
of neutral allele frequencies are good markers of mutational variance, VM, in seven species expressed
fixation. A recent study exploited this fact to esti- as coefficients of variation, CVM. A CVM of 1%
mate that the genomic beneficial mutation rate in a means that after one generation of mutation, the
set of E. coli populations was 4 × 10−9/replication standard deviation among initially identical lines is
(Imhof & Schlotterer 2001), or only one millionth of 1% of the mean. The traits are classified by their
the total genomic mutation rate shown in Figure 3.1. presumed relationship to fitness. Life history traits
It is not easy to generalize from this, as presumably are measures of viability, fecundity, or mating abil-
the evolutionary history of a population and the ity, and are expected to be closely related to fitness.
constancy of the environment will have an impact Morphological traits are features such as bristle
on this rate. Comparative data can also be used to number or leaf size that are probably under stabi-
estimate beneficial mutation rates if one is prepared lizing selection. Growth traits reflect the size of the
to accept a fairly simple set of assumptions about the organism during growth, which may or may not be
distribution of effects. For example, fixation events closely related to fitness.
are slightly more likely to involve amino acid Two facts are apparent from Figure 3.2. First,
substitutions than predicted from the pattern of the variation in CVM is large, ranging from 0.1% to
within-species polymorphism in two Drosophila over 4%. This reflects variation both within and
species. Under a simple model this suggests that among species. Species-level CVMs are correlated
about 45% of the amino acid substitutions are due with generation time and genome size, suggesting
to positive selection (Smith & Eyre-Walker 2002). that, as for molecular mutation rates, large genomes
and/or large numbers of cell divisions increase muta-
tional variance (Lynch et al. 1999). Second, it is clear
Mutations and Phenotypes
that morphological traits accumulate variance less
The importance of mutation for evolution also fast than life history traits (median CVM 0.24% vs.
depends on the precise pattern of effects on the 1.47%). At least part of the explanation for this
phenotype. For example, the degree to which the difference seems to be that larger numbers of loci
evolution of two parts of the body may be decoupled affect life history traits, because they summarize
depends on whether and how often mutations variation in the overall function of the organism
that affect the parts in different ways arise (Wagner (Houle 1998). Thus the concept of mutational target
& Altenberg 1996). While our knowledge of size––the number of base pairs which, when
the molecular and fitness effects of mutation are far mutated, affect a trait––can help to explain both
from comprehensive, we are more ignorant of these among- and within-species variation in the impact
important phenotypic properties. We know a bit of mutations on phenotypes.
about the amount of phenotypic variation produced
by mutation. The study of the correlated effects
Why Are Mutation Rates What
of mutations is just beginning (see Mutation
They Are?
Accmulation in Drosophila in the Case Studies
section below). Mutation rates themselves may evolve. Since the
The basic challenge is that discrete mutations of 1930s it has been clear that three major factors
large effect such as lethals, visibles, or human genetic potentially determine the outcome of this process:
diseases, can be readily observed, but these are irrel- the inevitability of some mutation, the costs of
evant to long-term evolution as they have extremely making replication as accurate as possible, and the
large deleterious fitness effects. Mutations with possible advantages of beneficial mutations. Thus,
small positive or negative effects on fitness are the there are three possible sorts of equilibrium muta-
ones that we need to understand, and it is precisely tion rates: the minimum possible, an optimal rate
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 42

42 Principles of Evolutionary Genetics

FIGURE 3.2. Mutational coefficients of variation from a review of studies

prior to 1995 (Houle et al. 1996). See text for explanation.

determined by the costs of fidelity, and an optimal possible (regardless of why exactly sex is good). As
rate determined by selection for evolvability indicated above, the rate of mutation depends on
(Kondrashov 1995). the DNA sequence, and the chromosomal context
It is clear that the first two processes must it is in. There is also substantial evidence for envi-
influence mutation rates. Mutation is unavoidable ronmental effects on mutation rates. Nutritional
because recovery of destroyed information is gen- deficiencies, the presence of mutagens, and tempera-
erally not possible. For example, when bases on ture can change both rates and patterns of mutation.
complementary DNA strands do not match, it is All this suggests that the variation necessary to alter
difficult to see how a repair mechanism could always mutation rates is abundant.
determine which of the two bases in the mutant In a few cases, organisms seem to have evolved
that should be repaired. Similarly, physical consid- portions of their genomes to be susceptible to muta-
erations show that, if one attempts to reduce the tion. For example, a region of the gene specifying
error rate in DNA replication and repair to zero, host recognition in the Bordetella bacterium
the cost of these processes, in terms of both time mutates at a very high rate because the bacterium
and energy, could be large. Thus, the first hypothe- has harnessed a retrotransposition-like process that
sis, that mutation rates are minimal, is not tenable. targets that region (Doulatov et al. 2004). The attrac-
This is also suggested by the inverse relationship tiveness of the notion that mutation rates are adap-
between genome size and mutation rates among tive must be tempered by the evidence that the vast
microorganisms (Figure 3.1). What is controversial majority of mutations are deleterious, and there-
is the effect of selection for evolvability on muta- fore costly to the individual in whose genome they
tion rate. occur (Johnson 1999). In sexual organisms, this
The fact that mutation is essential for evolution creates a typical conflict between group-level evolu-
has made the notion that its rate is tuned to allow tion and individual selection over the fate of a variant
evolution attractive to many. Furthermore, there is that increases the mutation rate. The high mutation
plenty of evidence for variation in mutation rate rate variant can increase in frequency when it causes
within species that could be exploited by natural a beneficial variant; however, this advantage benefits
selection. The fact that most eukaryotes employ the high mutation variant only as long as it and the
meiotic recombination to produce genetic variation beneficial variant remain together in the same geno-
shows that variation-generating adaptaions are type. This may not be long at all if the loci involved
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 43

Mutation 43

are unlinked. On the other hand, most copies of the It is much more difficult to imagine adaptive
high mutation variant will produce a steady stream increase of mutation rates in multicellular organ-
of deleterious variants that will reliably decrease the isms. In fact, the study of mutation in cancerous cells
transmission of the high mutation rate variants. of multicellular eukaryotes suggests that similar
From the point of view of a population, more exper- hypermutability may occur in tumors, where it facil-
imentation is ultimately beneficial if it can speed itates the evolution of high tumor growth rate and
adaptation. However, from the viewpoint of the resistance to chemotherapy, resulting in death of
individual, it would be much better if another indi- the organism. The fact that similar phenomena occur
vidual took the risk of experimentation. It takes whether or not they can be adaptive favors break-
special conditions for the advantages of mutation down in normal replication, recombination, and
to the population to outweigh its costs to the indi- repair machinery. The isolation of the germline from
vidual (Sniegowski et al. 2000). the soma in most multicellular animals is a powerful
Closely related to the idea that the overall muta- argument against the generality of adaptive mutation.
tion rate is adjusted to promote evolvability is the
idea that organisms can increase their mutation
What Limits the Rate of Evolution,
rate in relation to their need for variation, so-called
Mutation or Selection?
adaptive mutation. The converse idea that muta-
tions occur at random with respect to their useful- One of the central paradoxes of evolutionary biol-
ness has been a cornerstone of evolutionary thinking ogy is that most of the time, organisms do not
since the late nineteenth century, so evolutionary evolve at all (reviewed in Gould & Eldredge 1993).
biologists reacted with outrage when adaptive muta- There are two sorts of explanation for this stasis
tion was revived again by Cairns et al. (1988). These (reviewed by Hansen & Houle 2004). First, many
researchers observed mutation rates that restored believe that this is due to stabilizing selection which
growth in nondividing “stationary phase” cells in is somehow maintained over very long time periods.
E. coli and other single-celled organisms are higher The weakness in this hypothesis is simply that it is
than the rates of the same mutations when the cells difficult to see why selection should be constant over
are growing. This basic observation that has now periods of tens of millions of years. The other alter-
been made in many microorganisms (Foster 2000). native is that the kinds of variation necessary for
There are three potential explanations for the populations to evolve in response to whatever novel
increase in mutations observed in stationary phase. selection pressures come up are often not produced
The increase can be adaptive in two senses: The (Gould & Lewontin 1979). Such limitations on
strong version is that the rate of beneficial mutations variation are usually referred to as constraints, but
can be increased at need, which we can call “directed this suggests that the necessary variation is never
mutation.” The weak version is that the overall produced by mutation.
rate of mutation may be increased when variation The common argument against the constraint
is needed, or “hypermutability.” Finally increased hypothesis is that nearly every trait studied does
mutation may arise because of an unavoidable break- display genetic variation. This is insufficient to
down in normal repair and replication. resolve the issue because all aspects of the pheno-
Foster (2000) and others have pinned down the type will be selected simultaneously. It is not enough
mechanisms which underlie several cases of high to produce variation in each trait, the variation must
mutation in stationary phase. In each case the effects also be relatively free of entangling effects on other
are not confined to genes where mutations might selected traits. The capacity of the genome to produce
be adaptive, ruling out the directed mutation hypoth- appropriate sorts of phenotypic variation may deter-
esis. However, the specific mechanisms by which mine which of the many pressures that natural selec-
the mutations arise, including DNA synthesis initi- tion places on an organism it will be capable of
ated by recombination and activation of transpos- responding to. If mutations tend to affect a limited
able elements, do not necessarily suggest a general number of phenotypes, or phenotypes that tend to
breakdown of fidelity. Thus hypermutability is real, have similar selection pressures on them, the struc-
but its adaptive signficance is still not clear. The ture of variation is said to be modular (Mezey, Box
deleterious consequences of increases in mutation 19.5 of this volume; Wagner & Altenberg 1996).
are not readily avoided, unless death of the cell is Thus not only the rate but also the nature of muta-
certain in the absence of mutation. tion may itself be shaped by natural selection.
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 44

44 Principles of Evolutionary Genetics

CASE STUDIES variants. The insertion-deletion variants were all of

4 nucleotides or shorter. These data are biased
against detection of large insertions and deletions as
Nachman and Crowell
this would tend to preclude recognition of a pseudo-
Nachman and Crowell (2000) studied the gene in the first place. CpG contexts accounted for
rate of mutation in humans using the comparative about 25% of all the substitutions, with such sites
approach. We are unusually confident of both the having a 10-fold higher rate of substitution than
time since we diverged from chimpanzee lineage, and non-CpG contexts. The estimated rate of substitu-
of the number of generations that this represents. tion and mutation differed by a factor of 6 among
Nachman and Crowell took advantage of this by different pseudogenes. These differences were
studying divergence of DNA sequences that are statistically significant, suggesting that the region
among the most likely to be neutral and thus evolve in which the pseudogenes inserted influenced their
at the mutation rate: processed pseudogenes. mutation rates.
A processed pseudogene is a bit of DNA that The simple neutral divergence model outlined
has been reverse transcribed from a messenger RNA above assumes that a single copy of DNA is split
back into DNA and incorporated into the genome. into two lineages at the time of species divergence.
Differences of a pseudogene from the homologous In reality the ancestral species was very likely to
gene, such as frameshift deletions or insertions, already have DNA sequence variation at the time
make it clear that the pseudogene cannot encode a of speciation. This means that some of the differ-
protein, hence the name. Processed pseudogenes ences fixed in each lineage after speciation are actu-
can be recognized because they have had their ally variants that arose before the time of speciation.
introns edited out, and often have a poly-A sequence To compensate for this an assumption about the
attached. Therefore, pseudogenes are expected to effective size, N, of the ancestral species must also
have no function, and consequently to evolve at the be made. When the number of generations since
neutral rate, although this may not always be so divergence is not much greater than N, which is likely
(see above). to be the case for humans and chimps, the effect of
A potential complication with the use of pseudo- this adjustment can be substantial.
genes is that there may be many pseudogenes Thus, there are three unknown factors that
derived at different times from the ancestral gene. still must be taken into account to convert the
If an older pseudogene in one species were to be 1.2% divergence into an estimate of the mutation
compared with a younger one in the other species rate per generation: N before speciation, the time
then a very misleading picture of the rate of diver- since divergence, and the average generation time
gence would be obtained. Nachman and Crowell since divergence. Nachman and Crowell consid-
were able to guard against this possibility by a care- ered N values up to 105, divergence times between
ful choice of methods. They used polymerase chain 4.5 and 6 million years ago, and generation times
reaction (PCR) amplification to obtain material of 20 and 25 years. This gives an estimate of
directly from genomic DNA. For each pseudogene, the number of generations separating chimps and
they chose one of their PCR primers to lie in the humans of between 360,000 and 600,000. The range
genomic DNA outside the pseudogene itself. Thus, of possibilities suggests mutation rates per base
only sequences that possessed the same flanking per generation pair between 1.3 and 3.4 × 10−8.
sequence in both humans and chimpanzees would This agrees very well with direct estimates of
amplify. This ensures that each pair of compared human mutation rates (Kondrashov 2002), and more
human and chimpanzee pseudogenes was ortholo- recent analyses of a much larger human–chimp
gous, that is, it originated from the pseudogene data set.
already present in the last common ancestor.
In total Nachman and Crowell sequenced 18
Mutation Accumulation in
different pseudogenes in a chimpanzee and two
Drosophila
humans. In each individual, 16 kb was sequenced.
Overall, they found 199 differences between the Studying mutations with small effects on the
human and chimpanzee sequences, for a divergence phenotype and fitness is both important for under-
of 1.2%. These differences consisted of 131 transi- standing evolution, and difficult experimentally.
tions, 52 transversions, and 16 insertion-deletion Much of the data on such mutations come from
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 45

Mutation 45

mutation–accumulation experiments on Drosophila 1994) a second chromosome balancer was used to

melanogaster, due to its 2 week generation time, extract and replicate a single test chromosome, then
readily observed morphology and life history, and to preserve independent copies in heterozygous
the genetic tools available. condition (Figure 3.3). The second chromosome in
The most important such tools are balancer chro- D. melanogaster contains about 40% of the
mosomes, so called because they allow the preserva- genome. These copies start out genetically identical,
tion of a sampled chromosome from the disruptive but diverge over time as each copy independently
effects of recombination. Balancer chromosomes accumulates spontaneous mutations. If the heterozy-
have three key properties: a multiply inverted gene gous fitness effects of mutations are small, then
order, a dominant morphological marker, and a they will accumulate at very close to the mutation
recessive lethal mutation. When an individual is rate. To detect the effects of mutation, inversion
heterozygous for chromosomes with very different heterozygotes are crossed, and the ratio of test chro-
gene orders, chromosomes can pair at meiosis, but mosome homozygotes to inversion heterozygotes
recombination between them results in duplica- observed in the offspring, as shown in the last two
tions and deficiencies in the products, and inviable rows of Figure 3.3.
gametes. Therefore, a fly heterozygous for a balancer Mukai utilized this basic design several times to
chromosome will only give rise to gametes that study the effects of mutation on egg-to-adult viabil-
carry the unrecombined balancer or wild-type chro- ity, the probability that an egg survives to become
mosomes. As shown in Figure 3.3, this fact can be an adult. In his 1972 paper, three test chromosomes
exploited to capture (or “extract”) and replicate were each replicated 50 times to make sublines. Each
single chromosomes from any population, allowing subline was then subjected to the accumulation
their properties to be studied. process. Every 10 generations, the relative homo-
In the mutation-accumulation experiments we zygous viability of each subline was measured. From
want to discuss (Mukai et al. 1972; Houle et al. this, Mukai estimated the rate at which viability went

FIGURE 3.3. The use of balancer chromosomes to extract intact

chromosomes and to serve as standards for the measurement of
viability. The thick line denotes a multiply inverted chromosome
(the balancer), while thin lines are chromosomes with the usual
gene order. The circle denotes the dominant mutation Curly (Cy),
which causes the adults to have wings that curl upwards. The
square denotes the mutation brown-dominant (bwD), which causes
its carriers to have brown rather than red eyes.
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 46

46 Principles of Evolutionary Genetics

down over generations (M), and the rate at which the large effects must have occurred. If all mutations had
variation among sublines went up (VM). exactly the same effect on homozygous viability
The distribution of homozygous viabilities this relationship would be mathematically precise:
showed a minority of chromosomes acquired lethal M = Us, where U is the total mutation rate on the
recessive mutations, while most chromosomes chromosome to alleles affecting viability by amount
showed modest declines in fitness, with very few s, and VM > Us2. In reality, s does vary from variant
chromosomes having reductions of viability between to variant, and this increases VM, so Mukai et al.
50% and 100%. The rate at which chromosomes could set a lower limit to the mutation rate U <
acquired lethal mutations was 0.006 per chromo- M2/VM and an upper limit on the average s, s < VM/M.
some per generation. Even with the lethals excluded, For this experiment U > 0.06 when all the nonlethal
M was negative, as shown in Figure 3.4, as expected chromosomes were considered, or even > 0.17
due to a preponderance of deleterious mutations, when the few chromosomes with homozygous viabil-
while VM was positive as expected since each line ities near 0 were excluded. Thus the mutation rate to
accumulates its own unique set of individually rare variants with small homozygous effects on viability
mutations. is at least 10 times greater than the recessive lethal
These two quantities, M and VM, together give mutation rate. When extrapolated to the whole
some indication of the mutation rate and the effects genome, this suggests a deleterious mutation rate
of the nonlethal mutations. To see this, imagine greater than 0.4 per genome per generation.
two cases with the same decline in the mean: if the The high genomic deleterious mutation rate esti-
variance among sublines had not increased at all, mates of Mukai et al. (1972) have proved to be
this would indicate the presence of a very large quite controversial. Many subsequent studies have
number of mutations that each had very small effects. undertaken similar estimates in Drosophila and
Conversely, if the variation among sublines was other organisms; some broadly support Mukai’s
high, this would indicate that a few mutations with results, while others do not (Lynch et al. 1999).

FIGURE 3.4. Changes in the mean viability (left scale, filled circles)
and genetic variance (right scale, open circles) in viability over
40 generations of mutation accumulation in Drosophila melanogaster.
Mukai et al. (1972), reprinted with permission of the Genetics Society
of America.
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 47

Mutation 47

More troubling is that estimates of the genomic This substantially undermines the mutation-accu-
mutation rate in Drosophila, such as the direct esti- mulation hypothesis for the evolution of lifespan.
mate of 0.8 shown in Figure 3.1, are not much
larger than this estimate. If the direct estimate is
correct, then half of all mutations in Drosophila FUTURE DIRECTIONS
must be deleterious, an implausibly large propor-
tion. The direct estimate is itself subject to uncer- Knowledge of the rates and properties of mutation
tainty but where the discrepancy lies is not certain. is critical to understanding evolution. The study of
Two other potential gaps in the Mukai et al. mutation is difficult, and so has been relatively
study are less well known. Mukai et al. assumed neglected by evolutionary biologists. The result
that balancer heterozygotes provide a standard for is that our generalizations about mutation rest on
viability to compare with that of the heterozygotes. somewhat shaky ground. Consequently, we expect
This is almost certainly incorrect due to partially that the study of mutation will be among the most
dominant or epistatic effects of the mutations them- productive areas of evolutionary genetics in the
selves. Second, early development in Drosophila is future. Here are a few areas where we hope to see
primarily driven by maternal message, so that the great progress over the next 10 years.
homozygous effects of maternal-acting genes are
not assayed with their technique.
1. Improvements in sequencing technology
Houle et al. (1994) used the same technique as
have now made it possible to directly assay
Mukai et al. to accumulate mutations, but focused
mutations in DNA sequences after a muta-
instead on the pleiotropic effects of mutation on tion-accumulation experiment. Such studies
life history traits. This work was undertaken largely should increase rapidly in number. These data
to test the mutation-accumulation theory for senes- should remove much of our uncertainty about
cence, one of the major theories concerning the molecular aspects of mutation, as it will
the evolution of lifespan (Promislow & Bronikowsi, no longer be necessary to find phenotypic
Ch. 30 of this volume). The mutation-accumulation evidence for a mutation before molecular
theory requires that many mutations occur that analysis.
decrease fitness of old individuals, while leaving 2. Availability of complete genome sequences
early-life fitness unimpaired. Natural selection is less for closely related species pairs will tell us
effective at removing mutations that act late in the what proportion of genomes are subject to
lifespan than those that act earlier, potentially causing natural selection. Combined with direct data
a decrease in lifespan. To test for such mutations, on mutation rates, this will reveal what
Houle et al. studied female fecundity in young flies proportion of mutations affect fitness.
(5 and 6 days old) and old flies (27 and 28 days old), 3. Automated phenotyping is being pursued in
as well as average male and female lifespan. If the several systems, making it possible to
mutation-accumulation hypothesis is correct, there rapidly screen for mutants or increases in
should be variation in late fecundity that does not variation in many traits simultaneously. This
affect early-life fecundity. will be necessary to understand the correla-
tion structure of mutational effects.
After 44 generations of mutation accumulation,
4. After many years of relative eclipse, the
there was significant genetic variation due to muta-
study of advantageous mutations should
tion for all four traits. The mutational correlation
take its rightful place as the most important
of early and late fecundity was not different from a
aspect of evolutionary biology. New theory
perfect correlation of 1, but was significantly higher and large amounts of molecular data should
than 0, suggesting that mutations tended to affect make it much easier to detect the signature
early- and late-life fitness components similarly. of positive selection, for example in bacter-
Consistent with this, the correlations of both fecun- ial cultures, or from population survey data.
dities with lifespan were also positive and not differ- 5. One of the most neglected issues in evolu-
ent from 1. Mutations that affect one aspect of tionary genetics is what limits evolutionary
fitness also seem to impair all aspects of fitness –– progress (Hansen & Houle 2004), with one
a fly with poor fecundity early in life will also of the major possibilities being limits on the
live less long and have lower fecundity late in life. phenotypic effects of mutation. Progress in
0195168178_0032-0048_Fox-03.qxd 11/17/05 5:01 PM Page 48

48 Principles of Evolutionary Genetics

the previous three areas may make it possi- Denver DR, Morris K, Lynch M & WK Thomas
ble to test this idea. 2004 High mutation rate and predominance
of insertions in the Caenorhabditis elegans
nuclear genome. Nature 430:679–682.
SUGGESTIONS FOR FURTHER Drake JW, Charlesworth B, Charlesworth D & JF
READING Crow 1998 Rates of spontaneous mutation.
Genetics 148:1667–1686.
We know of very few general works on the role of Foster PL 2000 Adaptive mutation: implications
mutation in evolution, that is those that cover the for evolution. BioEssays 22:1067–1074.
range of issues raised in this chapter. Molecular Graur D & W-H Li 2000 Fundamentals of
mechanisms of mutation are introduced in Lewin’s Molecular Evolution. Sinauer Assoc.
series of Genes books (e.g., 2004). Graur and Li Kondrashov AS 1995 Modifiers of
(2000) provide an overview of molecular evolution mutation–selection balance: general approach
with a reasonable emphasis on mutation as a driv- and the evolution of mutation rates. Genet.
ing force. For information on molecular mutation Res. 66:53–69.
rates the review of Drake et al. (1998) and the Lewin B 2004 Genes VIII. Oxford Univ. Press.
recent paper by Denver et al. (2004) introduce the Lynch M, Blanchard J, Houle D, Kibota T,
important literature. For phenotypic mutation rates Schultz S, Vassilieva L & J Willis 1999
Drake et al. (1998) and Lynch et al. (1999) provide Perspective: Spontaneous deleterious
reviews. For the evolution of mutation rates generally, mutation. Evolution 53:645–663.
Kondrashov (1995) gives an overview of hypothe- Sniegowski PD, Gerrish PJ, Johnson T & A
ses. Foster (2000) and Sniegowski et al. (2000) Shaver 2000 The evolution of mutation rates:
provide a quick summary of the evidence for and separating causes from consequences.
against the adaptive values of environmental respon- BioEssays 22:1057–1066.
siveness of mutation rate.