What type of natural selection pressure selects for 1 extreme version of a trait?

Positive Directional Selection: Arms Race Coevolution

Under directional selection, relative fitness increases as the value of a trait increases (positive directional selection) or decreases (negative directional selection). Dawkins and Krebs (1979) argued that reciprocal positive directional selection exerted by coevolving hosts and parasites could lead to a situation where hosts continually become more resistant to parasitism while parasites respond by becoming more virulent or evolving new mechanisms of evading host immunity.

Unlike negative frequency-dependent selection, this so-called ‘Arms Race coevolution’ does not generate a rare advantage per se. Instead, host resistance and parasite ‘virulence’ are inherent properties of the individual genotype and do not depend on the frequency of the other genotypes. In this scenario, repeated selective sweeps favoring resistant hosts and virulent parasites will lead to the evolution of reciprocal host and parasite adaptations that, once overcome by a counter-adaptation, will not again incur resistance/virulence (Figure 1(b); Woolhouse et al., 2002; Burdon et al., 2013). Another important distinction between Red Queen and Arms Race coevolution is that the latter, in favoring traits that counter adaptations in the biological antagonist, has a high potential for evolutionary innovation (e.g., Kerns et al., 2008; reviewed in Daugherty and Malik, 2012).

Directional Selection

Directional selection occurs when individuals homozygous for one allele have a fitness greater than that of individuals with other genotypes and individuals homozygous for the other allele have a fitness less than that of individuals with other genotypes. At equilibrium the population will be composed entirely of individuals that are homozygous for the allele associated with the highest probability of survival. The rate at which the population approaches this equilibrium depends on whether the favored allele is dominant, partially dominant, or recessive with respect to survival probability. An allele is dominant with respect to survival probability if heterozygotes have the same survival probability as homozygotes for the favored allele, and it is recessive if heterozygotes have the same survival probability as homozygotes for the disfavored allele. An allele is partially dominant with respect to survival probability if heterozygotes are intermediate between the two homozygotes in survival probability. This pattern of selection is referred to as directional selection because one of the two alleles is always increasing in frequency and the other is always decreasing in frequency.

When a dominant favored allele is rare most individuals carrying it are heterozygous, and the large fitness difference between heterozygotes and disfavored homozygotes causes rapid changes in allele frequency. When the favored allele becomes common most individuals carrying the disfavored allele are heterozygous, and the small fitness difference between favored homozygotes and heterozygotes causes allele frequencies to change much more slowly (Figure 1). For the same reason changes in allele frequency occur slowly when an allele with recessive fitness effects is rare and much more rapidly when it is common. A deleterious recessive allele may be found in different frequencies in isolated populations even if it has the same fitness effect in every population, because natural selection is relatively inefficient when recessive alleles become rare, allowing the frequency to fluctuate randomly as a result of genetic drift.

Glossary

Directional selection

Natural selection that favors phenotypes that differ from the current mean phenotype of a population in one direction (e.g., those that are larger than the current mean).

Natural selection that favors phenotypes that deviate in either direction from the current population mean.

Phenotypic effect of a particular heterozygous combination of two alleles at a single locus that deviates from the mean of the two homozygous phenotypes.

Phenotypic effects of particular combinations of alleles at two or more loci that differ from the sum of additive effects of these alleles (i.e., the analog of dominance for more than one locus).

The influx of genetic variants into a local gene pool as consequence of immigration of individuals or gametes (i.e., pollen).

Stochastic change in allele frequencies at a given locus from generation t to generation t+1 due to the chance events that affect the number of offspring produced per parent in generation t. The effect of genetic drift on allele frequencies and, by extension, on population mean phenotypes increases with decreasing population size.

For a given taxon, a map of genetic marker loci that are arrayed based on the recombination rates among them. Recombination rates are estimated from crossing-over events in controlled crosses. Given a sufficiently high marker density, the resulting linkage groups correspond to chromosomes.

The co-existence of two or more distinct variants in a population either in the form of alleles for a given locus or as distinct phenotypes.

Natural selection against phenotypes that deviate from the current population mean.

The Effects of Sexual Selection on the Heritability of Traits

Strong directional selection usually exhausts additive genetic variance for a trait in three to five generations. (In this context, this means traits governed by polygenic inheritance, or quantitative trait loci; see Chapter 3 on genetics.) This means that the proportion of variation in the phenotype due to genetic variation, or heritability, approaches zero. After that, there can be no further response to selection because the remaining phenotypic variation is from either environmental or nonadditive genetic variation. In theory, sexual selection on a trait such as antler size should rapidly eliminate the additive genetic variance for the trait. In other words, the trait will be genetically fixed. In practice, many traits that seem to be under strong sexual selection still have considerable heritability.21

Key Term

Directional selection causes one form of a trait in a population, over generations, to be favored (see Figure 11.3).

There are a number of possible explanations for why selection does not eliminate all of the additive genetic variance for traits involved in mate choice. They include the following:

Sexual selection is strong only under extreme environmental conditions in which survivorship is low. Variance is maintained during periods of relaxed selection.

Interactions with other traits (e.g., linkage effects, viability effects) limit sexual selection before the additive variation is exhausted.

Mate choice relies on many factors, rather than one trait. When selection acts on multiple traits, they limit each other’s evolution so that variation remains for each of the traits.

Counterbalancing selection for factors like protection from predators maintains additive genetic variance by limiting the elaborateness of a signal.22,23 It is hard to overemphasize the complexity of mate choice and the need to consider multiple factors involved in any mate choice decision.

B Why are TEs so important?

Insecticide resistance results from very strong, persistent directional selection. TE-mediated changes in regulation can lead to massive and rapid changes in expression, responses that are potentially highly adaptive when an organism is faced with a major, pervasive, and novel mortality agent in the environment, like an insecticide. A useful contrast which illustrates this point is the essential absence of TEs involved in natural xenobiotic resistance—if we consider that mutational changes in plant allelochemicals are unlikely to bring about massive changes in mode of action or in toxicity, then mutational change associated with allelochemical resistance may be acquired more slowly as a result of the accumulation of small changes in structural genes (Li et al., 2007).

Application of insecticide tends to favor insecticide resistance, involving single genes of major effect rather than polygenic resistance (ffrench-Constant et al., 2004), and it has been found that most resistant field strains show monogenic resistance (Roush and McKenzie, 1987). Where resistance genes are already involved in essential functions, as is often the case for metabolic enzymes, it is advantageous to maintain the quality of mRNA to allow wild-type function to be retained and instead regulate gene expression. TE insertion within regulatory regions of genes which confer resistance often results in upregulation, that is, increase in the quantity of mRNA. This may be because many TEs have built-in enhancer sequences related to their transposition (Zhang and Saier, 2009) that have been co-opted by the host, but another possibility is that such spacing may move genes further from existing regulatory sequences (Schlenke and Begun, 2004).

Selection intensity

Selection intensity is a measure of the strength of directional selection applied in a selection experiment or breeding program to change a quantitative trait. If selection is practiced on individual performance, the selection applied can be described by the selection differential (S), which is the difference between the mean performance of the selected individuals and that of the population as a whole. The selection differential is measured in units of the trait, for example, grams of body weight or number of offspring born in a litter of mice or pigs. Formally, the selection intensity, usually denoted i, equals the selection differential measured in phenotypic standard deviations (σP = √VP), that is, i = S/σP. Therefore, the magnitude of the selection intensity so defined does not depend on the variability of the trait.

The selection intensity is a useful measure because its value can be predicted in an artificial selection program from knowledge of the proportion of individuals selected and the distribution of the trait. For example, with truncation selection, in which the highest-performing individuals for the trait are selected, i is a simple function of proportion selected (p) for any distribution. For the normal distribution, tables of i are available, or it can be computed as i = z/p, where z is the ordinate of the standardized normal at the truncation point corresponding to p. For example,

p0.80.50.20.10.050.010.001i0.3500.7981.4001.7552.0632.6653.367

Note that, as the proportion selected becomes very small, i equals approximately 1.2 –0.7log(p). Therefore, for example, the total selection intensity from two generations of selection with p = 0.1 can greatly exceed that for one generation with p = 0.01 and, because Mendelian segregation produces new assortment each generation, the consequent selection response from two generations with weak selection would be expected to be greater than one with intense selection.

Selection intensity also depends somewhat on the size of the population. For a given proportion selected, the intensity becomes slightly less as the population size becomes smaller. Values can be computed using order statistics (i.e., expected values of ranked observations) and are also tabulated (Falconer and Mackay). For example, if N individuals are selected from M recorded, and N = M/10

N/M1/102/205/5010/10020/2000.1M(M→∞)i1.5391.6381.7051.7301.742 1.755

In these calculations for finite numbers, observations are assumed to be uncorrelated. The intensity is a bit further reduced, by a predictable but less easily computed amount in a small population in which animals are related because the family members resemble each other and phenotypes are correlated.

For traits that are not normally distributed, the selection intensities may differ quite substantially from those given above. The notable case is where selection is on some all-or-none character: if it is desirable and has an incidence say q, once the proportion selected is less than q, selecting ever smaller proportions do not increase the selection intensity.

The selection intensity can also be computed if, for example, there is no simple truncation selection, including, for example, where individuals near the middle or extremes of the distribution are favored.

Genetic and Genomic Consequences of Selection: Beyond Gene Lists

Recent studies have used the strong genetic differences produced by directional selection as starting material. They have applied what might be thought of as informatics-driven filters to try to focus on individual genes whose manipulation will affect the selected trait. A clear example comes from Zhifeng Zhou and David Goldman's work with P and nonpreferring (NP) rats.36 P rats have been bred for many generations to have high preference for 10% alcohol versus water, and the NP rat line was bred for low preference. They first sequenced the exomes of six individual rats from each selected line, finding >120,000 single nucleotide polymorphisms (SNPs). About 20% of these SNPs were homozygous and consistently different between P and NP rats, suggesting that they represented the signature of response to selection pressure for high versus low drinking. Alternatively, given the relatively small population sizes necessarily involved in producing the selected lines, these SNPs could represent accidental inbreeding. To distinguish between these alternatives, they mapped the differential SNPs and found numerous relatively large haplotype blocks. About 1000 such blocks of SNPs could be mapped to known functional genes.

Clearly, it was necessary to further reduce the number of potential targets. Two of the segregating SNPs revealed stop codons in known genes, and 31 others showed exomic sequence differences predicted to adversely affect protein function. At this point, Zhou and Goldman turned back to behavior. They used an F2 segregating population of P and NP rats obtained by intercrossing inbred variants of P and NP lines. They tested hundreds of F2 rats for alcohol preference and performed a standard QTL linkage analysis to identify four variants in three genomic regions linked to preference variation. Comparing the QTL results to the exome sequencing SNP patterns consistently identified a stop codon in the Grm2 gene (Grm2 ∗407) (see Fig. 11.2): this gene encodes the metabotropic glutamate receptor 2, and the stop codon variant in P rats leads to widely deficient receptor function. With a clear target gene of interest, these investigators amassed a wealth of evidence that consistently implicated the mGluR2 receptor in P rats, including absence of receptor expression and impaired mGluR-mediated synaptic depression. They demonstrated elevated alcohol consumption in mice with a null mutation for Grm2. Overall, this paper shows how the power of directional selection to influence the genome can yield fruitful results. Three additional possible genes of interest were discussed in the paper, but these have not yet been validated.

A similar conceptual approach was also applied to ethanol preference drinking, also in rats. These investigators took advantage of the other rat preference lines selected at Indiana University, the High Alcohol Drinking and Low Alcohol Drinking rats.37 HAD and LAD rats were bred for high versus low preference in the same way P and NP rats were.38 This population offered three major advantages from a genetic perspective. First, the lines were derived from a genetically heterogeneous intercross of eight inbred rat strains and thus presented high genetic diversity at the outset.39 Rats from this segregating population were also available to serve as controls. Second, two independent populations of HAD rats, and two of LAD rats, were developed. Because laboratory populations are always relatively small (from a population genetic perspective), many differences between divergently selected lines emerge during the course of selection due to accidental fixation of gene variants (i.e., loss of genetic variation, or inbreeding). The occurrence of identical gene or chromosomal regional fixation during selection in two completely independent pairs of lines, but not in the nonselected controls, is thus relatively unlikely; this improves the detection of signatures of selection versus noise.1,2,29

Many details of the genomics analyses in this experiment differed from the analysis of P versus NP rats. Signatures of selection in the HAD/LAD populations were found for nearly 1000 genes, and most were located within a single gene. Within those genes, very few (four) were found in exonic regions: most were in promoter or intronic regions (see Fig. 11.3). Functional overrepresentation analyses suggested an important role for genes involved with synaptic transmission, memory, and reward pathways and included those coding for several ion channels and excitatory neurotransmitter receptors.37 Unlike the NIAAA group, these investigators did not pursue individual candidate genes and attempt to provide further confirmatory evidence.

Our group has also taken advantage of the existence of replicated selectively bred lines.40 A binge has been defined by the National Institute on Alcohol Abuse and Alcoholism of the US National Institutes of Health as a period of temporally focused drinking that leads to a blood ethanol concentration (BEC) > 80 mg%, or 0.8 mg/mL.41 Binge drinking is a risk for development of an alcohol use disorder, and most alcohol abusers binge drink. Binge drinking is also a strong predictor of medical diagnosis and has deleterious health consequences.42 Prevalence of binge drinking is increasing in the United States,43,44 and it is highly prevalent in both veterans and active military duty personnel. Alcohol use disorder is comorbid with many other psychiatric conditions: in these populations, posttraumatic stress disorder is also a frequent diagnosis.45 To develop an animal model of binge-like drinking, we explored several alternatives with the goal of achieving a simple behavioral assay for binge-like drinking in the mouse. Following earlier work in the area,46 we developed the drinking in the dark (DID) assay, where mice consume enough alcohol in 2–4 h to reach intoxicating BECs.47 The basic paradigm we used was to substitute 20% ethanol for water for a limited period each day, during the early hours of the circadian dark cycle, as this is when rodents consume much of their daily food and fluid. We have determined the optimal time after “lights off” to start access,47 and the optimal duration of access to result in elevated blood alcohol levels.48 Consumption of the ethanol solutions remains relatively consistent across 12 days. When we examined panels of multiple inbred strains in the DID procedure, we found that the trait was reliable upon retest and significantly heritable.49,50

We subsequently selectively bred high DID (High Drinking in the Dark [HDID]-1 and HDID-2) mouse lines for high BECs after a 4-h DID session; these mice drink to the point of behavioral intoxication and reach blood levels that average about 200 mg/mL [Refs. 48,51; see Fig. 11.4]. Behavioral characterization of HDID mice has revealed that HDID mice exhibit behavioral impairment after drinking, withdrawal after a single binge drinking session, and escalate their intake in response to induction of successive cycles of dependence.48,50 Notably, HDID mice do not exhibit altered tastant preference or alcohol clearance rates.52,53 One clear limitation of the DID model is that ethanol is not offered as a choice versus water, and when it is, ethanol intake and blood alcohol levels are somewhat lower.48,53 This selection has one or two unusual features. The first is that we bred for a pharmacological endpoint (blood alcohol level after drinking) rather than for increased intake. As Fig. 11.4 shows, animals did nonetheless show elevated drinking across generations, which was expected. However, they achieve higher blood levels by patterning their drinking differently. HDID-1 mice show larger (longer) bouts of sustained drinking, while HDID-2 mice show more frequent, smaller bouts.54 Second, given that the HS/Npt foundation population shows very low blood levels (and intake: see data at Generations S0 and S14 in Fig. 11.4), we elected not to develop parallel lines for low blood levels after DID.

In an initial attempt to explore the genomic structure of response to intense selection, we compared patterns of gene expression in ventral striatum tissue from 48 naïve, male mice from all three genotypes.40 We compared HDID-1 mice from the 22nd selected generation, HDID-2 mice from the 15th selected generation, and HS/Npt unselected controls. Using Illumina WG 8.2 arrays, we analyzed SNP variation in 3683 markers: analyses were carried out marker by marker and with Weighted Gene Coexpression Network Analysis [WGCNA: Refs. 55,56]. For both QTL analyses and network analyses, we predicted that genetic variability across animals would be greatest in the unselected HS mice, which we found to be true. We predicted that the HDID-2 mice would show differences from HS, and that the HDID-1 mice would show even larger changes given their greater response to selection at that point. Of the more than 9000 transcripts (more than 7000 unique genes) surveyed, there were more genes differentially expressed between HDID-1 and HS than HDID-2 and HS, and 94 transcripts differed from HS in both lines, with the same directionality.

One interesting finding from this study is shown in Fig. 11.5. The WGCNA identified 21 modules, each representing a number of coexpressed genes. Of these, four modules were strikingly and consistently affected by selection. In some instances, the coherence of the module was increased by increased selection (i.e., HDID-1 > HDID-2 > HS), while for others, the opposite was true (see Fig. 11.5). The overall signature of selection indicated that intramodule coherence was more meaningfully responsive to selection (i.e., consistent across the two replicates) than the specific differences in expression of individual genes.40

Abstract

Heritable phenotypic variation determines how a population’s mean phenotype evolves under directional selection. The additive genetic covariance matrix, G, summarizes the genetic variation among individuals (due to individual’s carrying different alleles), and can be used to predict (or possibly reconstruct) phenotypic responses to directional selection. Moreover, G itself is subject to evolution, and by comparing G among naturally or experimentally evolving taxa we can understand how selection and drift drive changes in frequencies of alleles that determine phenotypes. Here, we present some commonly used tools for comparing matrices in order to understand the evolution of this important evolutionary parameter.

Selection Based on Progeny Performance

For quantitative traits, the expected response (R) per generation of directional selection is given by

R=ρ×i× σA2

where ρ is the accuracy of selection, which is defined as the correlation between the true breeding values (TBVs) and the estimated breeding values (EBVs) of the selection candidates; i is the selection intensity, which is equal to the mean superiority of the selected individuals expressed in terms of phenotypic standard deviations; and σA2 is the additive genetic variance of the trait being directly selected for.

If selection is based on a single phenotypic record on own performance of the selection candidates, the selection accuracy is ρOP=h2, where h2 is the heritability of the trait of interest. Alternatively, if selection is based on the average performance of n offspring of each selection candidate, then the selection accuracy is shown to be ρPT=n×h2/[4+(n−1)h2] .

The ratio between the ‘progeny test’ (ρPT) and the ‘own performance’ (ρ OP) accuracies is equal to v=ρPT/ρOP=n/[4+(n−1)h2]. Selection based on progeny testing will be more effective than phenotypic selection if v > 1, which occurs only if n>(4−h2)/(1−h2). Hence, it is seen that progeny testing is more effective for lowly heritable traits, but that even in such cases at least five progeny are necessary for progeny test to be more efficient than phenotypic selection.

For reasons of economic profitability, progeny testing protocols are usually applied to selection of males in animal breeding programs. First, males can be mated with a large number of females to produce a large number of offspring needed for analysis, especially with the use of artificial insemination. Second, in many species, generation intervals for males are shorter than those for females.

The principal drawback of progeny testing is a substantial increase in time and associated cost needed for animal evaluation. To be evaluated for most traits of economic importance, the progeny has to reach maturity, thus adding at least one generation to the time required for a round of selection (up to 8 years in some species). Genetic progress per unit of time can be improved with progeny test only if the resulting increase in generation interval is compensated with satisfactory increments in prediction accuracy. To obtain such high accuracy of selection, large populations of offspring have to be produced and maintained, thus making this approach feasible mostly to large-scale breeders.

Modifier Theory in Small Populations – The Role of Genetic Drift

In small populations, genetic drift acts as a stochastic force in generating nonrandom associations. Directional selection acting on beneficial mutations in such populations generates, on average, negative linkage disequilibrium. Allelic associations in positive linkage disequilibrium (either beneficial with beneficial, or deleterious with deleterious) are rapidly either fixed or lost when they arise due to chance – it is combinations that include both beneficial and deleterious mutations that, on average, persist the longest (Barton and Otto, 2005). This leads to increases in frequency for recombination modifier alleles, with the strongest effects occurring in small populations (Otto and Barton, 2001), in large populations where spatial structure increases the effect of genetic drift (Martin et al., 2006), and in population where directional selection is acting simultaneously on multiple loci (Iles et al., 2003). However, this combined effect of genetic drift and directional selection on recombination requires a relatively high rate of beneficial selective sweeps to fix or remove allelic combinations in positive disequilibrium.