Measurements of spontaneous rates of mutations in the recent past and the near future

Abstract

The rate of spontaneous mutation in natural populations is a fundamental parameter for many evolutionary phenomena. Because the rate of mutation is generally low, most of what is currently known about mutation has been obtained through indirect, complex and imprecise methodological approaches. However, in the past few years genome-wide sequencing of closely related individuals has made it possible to estimate the rates of mutation directly at the level of the DNA, avoiding most of the problems associated with using indirect methods. Here, we review the methods used in the past with an emphasis on next generation sequencing, which may soon make the accurate measurement of spontaneous mutation rates a matter of routine.

1. Introduction

Mutation, as the ultimate source of all genetic variations, is arguably among the most fundamental processes in biology, and the mutation rate is one of the key quantities that describe life. Estimating the rates of different kinds of mutations and characterizing the mechanisms of their origin are important ingredients for understanding many other biological phenomena, such as the evolution of sex or genetic disease. However, aside from these practical implications, an accurate knowledge of the mutation rate has an inherent value in expanding our fundamental understanding of the natural world.

Mutations have been the subject of quantitative research for quite some time (Danforth 1923; Muller 1927; Haldane 1935; Timofeeff-Ressovsky 1935; Mukai 1964; Mukai et al. 1972; Crow & Abrahamson 1997) and there have been many attempts at estimating the rates of spontaneous mutations (for a review, see Drake 1993; Drake et al. 1998; Lynch et al. 1999, 2006; Lynch 2006; Baer et al. 2007). The efforts of the past century have produced a fragmented picture. On the one hand, our level of awareness of the importance of mutations is phenomenal, embracing various branches of biology from the mechanisms of mutation to its impacts on fitness and disease. On the other hand, study of spontaneous mutation has barely reached a level of accuracy where it can be considered a quantitative science. Why are we still struggling to make accurate measurements of this fundamental parameter, 100 years after the mass of the electron and the speed of light were determined?

There are six reasons why measuring the rate of mutation has proven to be so difficult. First and foremost, it is because the per-nucleotide rate of mutation is extremely low, especially in eukaryotes (Drake et al. 1998; Lynch 2006; Lynch et al. 2006), but can be high in viruses (Drake et al. 1998; Gago et al. 2009). To measure such low rates, extremely careful and laborious methods are required. Second, mutation is a random process and there is no way to know where and when a mutation will occur. Third, there are several kinds of mutations, including point mutations (single-nucleotide substitutions, insertions, deletions, and complex events) and large-scale mutations (long insertions and deletions, duplications and polyploidizations, inversions, results of gene conversion, transpositions and long complex events; Petrov et al. 2000; Yang et al. 2001a,b; Ball et al. 2005; Chen et al. 2005; Hastings et al. 2009) and their rates need to be measured separately. Fourth, the rate of mutation can depend on the genetic background, especially on the context surrounding the site in question or its methylation (Holliday & Grigg 1993; Rogozin & Pavlov 2003). Thus, simply determining the average rate even of a particular kind of mutation is not enough: ideally an accurate measurement would also take into account the frequency of different contexts in the genome. Fifth, the mutation rate can be influenced dramatically by many mutagenic environmental and physiological factors, including temperature (Muller 1932), chemicals (compilation of hazardous chemicals (HAZARD DB 2009) lists approx. 700 known mutagens), radiation, age, sex, etc. Finally, the mutation rate itself is subject to selection and evolution, with some authors suggesting that some organisms increase the rate of mutation in times of stress (Sniegowski et al. 1997; Loewe et al. 2003; Avila et al. 2006; Lynch 2008).

2. Methods for determining mutation rates

Several authors have classified conceptual approaches of measuring spontaneous mutation rates (Kondrashov 1998; Fu & Huai 2003; Baer et al. 2007) while others have collected qualitative estimates of mutation rates (Drake et al. 1998; Keightley & Eyre-Walker 1999; Lynch 2006; Lynch et al. 2006; Keightley & Halligan 2009). However, these approaches and estimates, in our opinion, are rapidly becoming obsolete, at least in the context of measuring spontaneous mutation rates at the sequence level. Therefore, we present only a brief overview of these estimates, not for the purpose of guidance in selecting one of these approaches for further usage, but as a tribute to the enormous efforts that generations of scientists have invested in studying spontaneous mutation.

(a). Phenotypic mutation screening

A straightforward approach to estimate the rate of mutation is to screen for phenotypically evident mutants (for a review, see Drake et al. 1998; Fu & Huai 2003). Then, to estimate the per nucleotide rate of mutation it is necessary to know the mutational target, the length of the sequence which, when disrupted, leads exclusively to the screened phenotype. The application of this method has been hindered in the past by a poor knowledge of the fraction of functional sites per locus (Kondrashov 1998), but the availability of complete genomes has eliminated this problem. This approach has been applied to humans (Sommer 1995; Sommer & Ketterling 1996; Kondrashov 2003; Nachman 2004), Drosophila (Drake et al. 1998) and other species (Schrag et al. 1999; Mackwan et al. 2008), but has never been very popular for eukaryotes, with only one large-scale study performed in recent years (Yang et al. 2001a). In general, phenotype screening almost certainly underestimates mutation rates, either because not all nonsense mutations produce a phenotype (Nachman 2004) and available corrections (Drake et al. 1998) are not perfect, or because of the difficulties arising from clusters of mutants (Yang et al. 2001a; Fu & Huai 2003; Gong et al. 2005a,b), or because a mutant can be easily overlooked in the course of screening. The main problem, however, with this type of approach is the sheer volume of work involved in phenotype screens owing to the low spontaneous mutation rates. For example, Yang et al. (2001a) screened approximately 900 000 flies and found only 16 mutation events at eight loci.

(b). Measuring the fitness impact of accumulating mutations

Measuring the fitness effect of mutations accumulating over time in a laboratory population under relaxed or absent selection has been a popular choice for estimating the rate of deleterious mutation (Mukai 1964; Mukai et al. 1972; Charlesworth et al. 1990, 1994; Kondrashov & Houle 1994; Fernández & López-Fanjul 1996; Kibota & Lynch 1996; Keightley & Caballero 1997; Schultz et al. 1999; Vassilieva & Lynch 1999; Loewe et al. 2003; Charlesworth et al. 2004; Houle & Nuzhdin 2004; Gong et al. 2005a,b; Roles & Conner 2008). Starting from an inbred population, it is possible to measure the impact on fitness of many mutations that have accumulated over dozens of generations. Thus, this approach measures the rate of deleterious mutations and, if the deleterious fraction of the total mutation rate can be estimated, then one can deduce the total rate. This approach, however, has a variety of problems and its application remains controversial (Keightley & Eyre-Walker 1999; Fry 2001). An inbred population may not reflect mutation patterns expected in a wild population. Also, a high rate of homozygosis can largely obscure selection and makes it difficult to maintain the population in the laboratory, although the latter issue is not as important in selfing organisms. One way to avoid these problems is to maintain a small, randomly mating population that is not inbred, the so-called ‘middle class neighbourhood’ population (Shabalina et al. 1997). Nevertheless, other problems are unavoidable, including the principal impossibility of eliminating selection completely even under laboratory conditions, difficulties in detecting very mildly deleterious mutations, and the possibility of mutations affecting the rate of mutation itself in the course of the many generations under relaxed selection. Therefore, it is not surprising that different authors obtained very different estimates of the mutation rate using this approach. Of course, this approach is also very time-consuming, requiring constant vigilance to maintain the inbred population, to prevent contamination and to repeatedly assay fitness.

(c). Measuring the rate of evolution at neutral sites

The rate of neutral evolution is equal to the mutation rate (Kimura 1968). This simple relationship has been used several times to estimate the spontaneous rate of mutation (Crow 1993; Drake et al. 1998; Nachman & Crowell 2000; Kumar & Subramanian 2002; Silva & Kondrashov 2002; Mitchell & Graur 2005), with varying degrees of success. In particular, it has been suggested that contrasting the rate of evolution of neutral sequences with that of sequences under selection can be used to estimate the rate of deleterious mutation (Kondrashov & Crow 1993). Several such estimates have since appeared (Eyre-Walker & Keightley 1999; Keightley & Eyre-Walker 2000; Cutter & Payseur 2003), yet the accuracy of the underlying rate of spontaneous mutations in these estimates has been questioned, most notably on account of the usage of synonymous sites as neutral markers (Kondrashov 2001). Two problems are apparent with this method. Firstly, as exemplified by the use of synonymous sites, there must be a good reason to assume neutrality of particular sequences. Secondly, the rate of neutral evolution should not be measured by comparing distant species owing to the difficulty of estimating the number of hidden multiple substitutions per site (Li 1997) and a lack of knowledge of the true number of generations separating the two species. The study of Nachman & Crowell (2000) avoided both of these issues by selecting orthologous pseudogenes of humans and chimps and, as a result, their estimate remains the best application of this approach.

(d). Inferring parameters of mutation from data on within-population variation

Many properties of equilibrium populations depend on the mutation rates and, thus, can be used to estimate them. In particular, nucleotide diversity at the sequence level is 4N_eμ in diploids, and thus can be used to estimate μ (Deng & Lynch 1996; Kondrashov 1998; Li & Deng 2005; Deng et al. 2006; Messer 2009). However, because N_e is not known independently, this relationship is more useful for estimating N_e and not μ (Kondrashov 1998; Caballero 2006). A more promising approach is to estimate the rate of deleterious mutation from the data on inbreeding depression in highly selfed populations, as this is mainly owing to recent mutations and thus does not depend on the intensity of genetic drift (Charlesworth et al. 1990).

(e). Direct sequence-level screening of mutations accumulated in the course of 10–1000 generations

In recent years, several studies have been published in which the rate of spontaneous mutation was measured directly at the DNA level (Denver et al. 2000, 2004; Malpica et al. 2002; Howell et al. 2003; Haag-Liautard et al. 2007, 2008; Lynch et al. 2008; Keightley et al. 2009). Three of them (Lynch et al. 2008; Keightley et al. 2009; Xue et al. 2009) used a direct sequencing approach, demonstrating that direct sequencing is now accurate enough to be used to estimate even low rates of mutations (Kondrashov 2008). Most of these studies have sequenced a mutation accumulation line, with the exception of those of Howell et al. (2003) and Xue et al. (2009), who sequenced a section of the mitochondrial genome and Y chromosome, respectively, traced through known pedigrees of approximately 10 generations. Pedigree data essentially replace mutation accumulation lines in the human population, and although the use of non-recombining loci simplifies their use by eliminating the possibility of contamination from unrelated individuals, data obtained for the human mitochondria (Howell et al. 2003) remain controversial (Henn et al. 2009).

(f). Applying mutagens

One of the easiest approaches is to apply artificial mutagens. If the ratio of the spontaneous to mutagen-induced mutation rates is known a priori, then measuring the mutagen-induced rate of mutation can be used to estimate the rate of spontaneous mutation. The mutagenic treatment can greatly increase the rate of mutation, thereby reducing the amount of work. While artificial mutagenesis has long been used to obtain mutants and estimate the impact of mutation on quantitative traits (Mukai 1970; Ohnishi 1977), only two studies on estimating the spontaneous rate of mutations have been performed (Keightley & Ohnishi 1998; Yang et al. 2001b). A potentially resolvable issue with this approach is the difficulty of quantifying the impact of the mutagen. A more fatal issue is that, given the nature of the mutagen, it is likely to have profoundly different effects on different kinds of mutations, such as point mutations and indels, and, therefore, cannot be used to obtain good estimates of all spontaneous mutations.

3. A snapshot of current knowledge

What do we know about spontaneous rates of mutation? It seems that the problems with measuring rate of mutations have up until very recently thwarted direct quantitative measurements. Measurements of more qualitative factors of mutation have been much more successful. We know that transitions are usually (Keller et al. 2007) more frequent than transversions (Kondrashov 2003; Denver et al. 2004; Haag-Liautard et al. 2007; Lynch et al. 2008). The rate of mutation is dependent on the number of germline cell divisions per generation (Drost & Lee 1995; Drake et al. 1998), which implies an accumulation of mutations in older individuals and more mutations in males owing to more pre-meiotic cell divisions (Crow 2000; Makova & Li 2002; but see Crow 2006) and that intra-spermatogonial selection can alter the rate of germline mutation (Crow 2006; Arnheim & Calabrese 2009). We have a good understanding of the mechanisms involved in spontaneous mutations, including recombination (Hastings et al. 2009), various repair mechanisms (Wyman & Kanaar 2006; Lewin 2008, ch. 20) and of context dependence, the influence of the nucleotide state of neighbouring sites (Gordenin & Resnick 1998; Kondrashov & Rogozin 2004; Kondrashov et al. 2006; Hodgkinson et al. 2009). Different polymerases can lead to different patterns of mutation (Pavlov et al. 2006; Arana et al. 2008) and their function also causes the mutagenic effects of methylation (Holliday & Grigg 1993) and mutational hotpots (Rogozin & Pavlov 2003). We also have some estimates of the relative rates of indels (Petrov et al. 2000; Yang et al. 2001a,b; Denver et al. 2004; Haag-Liautard et al. 2007), transposon insertions (Nuzhdin & Mackay 1994) and strand biases in the mitochondrial (Tanaka & Ozawa 1994), nuclear and prokaryotic genomes (Green et al. 2003; Morton & Morton 2007; Unniraman & Schatz 2007). Finally, the deleterious mutation rate is likely to be U > 1 for many eukaryotic species with measured mutation rates (Kondrashov 2001, 2003; Denver et al. 2004; Haag-Liautard et al. 2007).

As far as quantitative estimates of mutation rate are concerned, a large fraction of them have been obtained in recent studies that use a multigeneration direct sequencing approach (table 1). Other methods have also produced quantitative estimates (for a review, see Drake et al. 1998; Lynch 2006; Lynch et al. 2006), but, owing to the inherent caveats associated with all approaches other than direct DNA sequencing, these estimates are uncertain and must be repeated. A symptom of the uncertainty of indirect estimates is the large variability of estimates obtained by different authors, especially for Drosophila melanogaster. The direct DNA sequencing methods have yet to be independently replicated, and perhaps the estimates reported in table 1 should also be treated with caution. However, direct sequencing appears to be associated with low error rates; in particular Lynch et al. (2008) and Keightley et al. (2009) have not found a single false-positive mutation, indicating that the false positive rate of direct sequencing is lower than the limit of accuracy. In addition, it is heartening to see that the studies that have obtained these estimates have placed the accurate measurement of the spontaneous mutation rate, not its implications, as the primary goal.

Table 1.

Direct sequencing estimates of mutation rates. Two probably accurate but indirect estimates (Nachman & Crowell 2000; Kondrashov 2003) are included for Homo sapiens.

species	rate of mutation per site per generation	comments	references
Homo sapiens	2.5 × 10−8 total (from species divergence)	indels are rare	Nachman & Crowell (2000), Kondrashov (2003), Xue et al. (2009)
	1.8 × 10−8 total (phenotypic mutations analysis)
	3.0 × 10−8 total (direct sequencing of Y chromosome)
Drosophila melanogaster	8.4 × 10−9 total	some variation of rates between different lines; supports higher rate estimates	Haag-Liautard et al. (2007), Keightley et al. (2009)
	3.5 × 10−9 to 5.8 × 10−9 point mutations
Drosophila melanogaster mitochondrion	7.2 × 10−8 total	high rate of G→A mutations on the major strand; higher than estimates from species divergence	Haag-Liautard et al. (2008)
	6.2 × 10−8 point mutations
Caenorhabditis elegans	2.1 × 10−8 total	insertions very common; higher than previous estimates	Denver et al. (2004)
	9.1 × 10−9 point mutations
Caenorhabditis elegans mitochondrion	1.6 × 10−7 total	higher than previous estimates	Denver et al. (2000)
	9.7 × 10−8 point mutations
Saccharomyces cerevisiae	0.33 × 10−9	similar to previous estimates	Lynch et al. (2008)
S. cerevisiae mitochondrion	1.2 × 10−8 point mutations (average of two methods)		Lynch et al. (2008)
	7.5 × 10−9 indels (average of two methods)

4. Perspectives

(a). The dawn of a new era in measuring spontaneous mutation rates

The use of mutation accumulation over the course of many generations, which may lead to an underestimation of the rate of mutation owing to the action of selection in the laboratory (Kondrashov & Houle 1994), is the last bastion that is yet to fall before the advance of sequencing technology. The 1000-dollar human genome (Mardis 2006), if realized, will make sequencing cheap and accurate enough to eliminate the need to accumulate mutation for many generations, which is obviously impossible for any organism with a not-too-short generation time. Even if the cost of genome sequencing is currently not as cheap as advertised for the future it is doubtful that at present new mutation accumulation lines will be started for estimating spontaneous mutation rates; by the time enough generations pass, genome sequencing may already be cheap enough to sequence individuals separated by very few generations.

In the near future, researchers may sequence randomly selected individuals from parental (P), and two generations of the offspring generations, F₁ and F₂. The sequencing of individuals from the F₂ is required to eliminate somatic mutations, and several such triplet generations could be sequenced to obtain accurate measurements. If the per nucleotide mutation rate is approximately 10⁻⁸, to obtain approximately 100 mutations it is necessary to sequence approximately 10¹¹ nucleotides (estimated from two P genotypes, one F₁ and seven F₂ genotypes for one generation triplet). A key fact is the apparent accuracy of even the current versions of the next generation sequencing methods, with Lynch et al. (2008) and Keightley et al. (2009) reporting no false positives among approximately 5 × 10⁶ nucleotides sequenced with 3× or greater coverage and approximately 7.2 × 10⁷ nucleotides sequenced with 5× or greater coverage, respectively.

(b). Unanswered questions concerned with spontaneous mutation rates

As powerful as direct sequencing may be, it has not yet reached a level of economy to be able to answer all fundamental questions pertinent to mutation rates. The four recent studies that involved direct sequencing have yielded very few mutations: four in the human pedigree (Xue et al. 2009), 35 in Saccharomyces cerevisiae (Lynch et al. 2008), 37 (Haag-Liautard et al. 2007) and 174 (Keightley et al. 2009) in D. melanogaster and 30 in Caenorhabditis elegans (Denver et al. 2004). Although some qualitative observations on different kinds of mutations have been made in some of these studies (table 1), obviously, to be of use for refining this knowledge on a quantitative level, the amount of data must increase by more than an order of magnitude. Nevertheless, it may be useful to define a set of unanswered fundamental questions pertaining to the study of mutation rates that may yet be tackled with the advance of cheaper sequencing.

At the heart of all questions, secondary to the average genomic mutation rate is the issue of its natural variability. The first set of questions deals with the intraspecies and interspecies variation of mutation rates. How different are the rates of mutations between species? To some degree a crack at this question has been attempted with the study of selfing versus obligatory sexual organisms (Charlesworth et al. 1994; Kondrashov 2001; Denver et al. 2004), but it must be set broader than just a question of life cycle. A study of the mutation rate of closely related species will give an idea of how quickly the rate of mutation evolves. Similarly, the distribution of the rates of mutation within a population is intriguing and if such mutational variability is high, then the second type of questions can be of fundamental importance: what is the impact of environmental factors on the rate of mutation? Is the discharge of anthropogenic mutagens into the environment having an effect on the mutation rate of various species? Is there a difference in the mutation rate between populations of the same species living in different environments, for example, hot versus cold? Finally, the question of intragenomic mutation rate variability must be addressed. Methylation and transition/transversion bias surely have a large impact on such variability, but the more detailed impact of context, chromosomal location or other uncharacterized factors is uncertain (Kondrashov et al. 2006; Hodgkinson et al. 2009). Similarly, precise measurements of the rate of different kinds of mutations, such as indels and substitutions, need to be performed.

(c). Impact of mutation at the phenotype and fitness levels

While new-generation sequencing will produce good quantitative description of the mutation process at the sequence level in the not-too-distant future, this will not lead automatically to understanding of the impact of mutation at the higher level of organization. What fractions of de novo mutations are deleterious, neutral and advantageous (Keightley & Eyre-Walker 2010; Orr 2010; Trindade et al. 2010)? How rapidly do fresh mutations supply genetic variation of quantitative traits, i.e. how large is the mutational heritability of various traits (Lynch et al. 1999)? What is the magnitude of mutational covariation of traits and fitness? Even a complete knowledge of mutation at the sequence level will not be enough to address these questions, because the answers to them depend on the nature of genotype > phenotype (including genotype > fitness) map. Thus, accumulation of mutations in the course of many generations may still be used to study the phenotypic impacts of mutations. Still, indirect estimates of the genomic deleterious mutation rate based on the fraction of non-neutral sites in the genome may be more promising, because selection at many sites is too weak to be detected in the course of a mutation accumulation study.

(d). Conclusion

In the recent past, two circumstances have inhibited a more in-depth study of spontaneous rates of mutation. First, it was the natural difficulties that arise in measuring such infrequent random events. Even a decade ago, direct sequencing of entire genomes of closely related individuals to measure the mutation rate was inconceivable owing to the high cost and low accuracy, although acknowledged as the most problem-free approach of all possibilities (Kondrashov 1998). Two such attempts have already been made (Lynch et al. 2008; Keightley et al. 2009) and we have every reason to believe that many more estimates obtained in a similar manner will follow.

The second hindering circumstance is that many scientists were more focused on the implications of spontaneous mutations, such as the evolution of sex, and for this purpose have been estimating the rate of mutation indirectly (Drake et al. 1998; Kondrashov 2008). At the same time, few studies were devoted exclusively to improving the accuracy and cost effectiveness of such measurements. Improvement in the cost and accuracy of next generation sequencing technologies in the near future may allow the accurate estimation of spontaneous mutation rates in many species through direct sequencing of closely related individuals. Such estimates can be accelerated by a conscious community-driven effort, similar in spirit to that of the Human Genome Project or the CERN Large Hadron Collider, aimed at the scrupulous measurements of one of the most fundamental parameters in evolution—the rate of spontaneous mutation.