Wild populations can potentially be affected by one-way straying of non-native hatchery fish in three ways. The first is the spread of deleterious alleles into a wild population, and as noted in the previous talk, the ability of non-native alleles to invade a wild population depends critically on selection intensities. The second potential effect of hatchery straying, also previously mentioned, is the eradication of genetic differences between hatchery and wild populations. I do not necessarily mean that some genes have no consequences for fitness, but that the different genes in hatchery and wild fish may represent equally good combinations. If, however, wild populations possess unique attributes, such as unique colors, sizes, or shapes, which do not greatly influence fecundity or other aspects of fitness, but which have some conservation value, these attributes may be eradicated by gene flow from hatcheries. This question has less to do with population fitness than with genic diversity. The third concern is with the demographic effects hatchery strays have on wild populations. Genetics and demography interact to influence wild populations in ways other than the effects brought about by gene flow. If we want to understand the effects of the immigration from hatchery populations, we also have to understand what the fates of wild populations would be in the absence of gene flow from hatcheries. Even in the absence of hatchery gene flow, wild populations may not do well, because of ecological or other non-gene flow effects from the presence of hatchery stocks in the same river. For example, a wild stock may be reduced by shared predation during harvest on both hatchery and wild stocks.

In this talk, I will address only the first issue of what selection intensities are in natural populations and will present information on what selection intensities have been measured in nature for various organisms. Unfortunately, selection has not been measured in natural populations of salmon, so we have to make educated guesses by looking at what has been measured in other organisms, some of which are fishes such as guppies and sticklebacks. I would like briefly to present the results of a survey of the literature on natural selection by Endler (1986). I will then present some caveats in drawing conclusions from these studies, to argue that we may not be able to take these estimates of selection and use them in the equations presented in Joe Felsenstein's talk. Lastly, I would like to talk about the kinds of variability in natural populations and how this variability influences population responses to selection.

By natural selection, we mean the differential survival or reproductive success of various phenotypes. Some individuals succeed and others fail to contribute offspring to the next generation for one reason or another. This should be distinguished from evolution, which is genetic change not only by natural selection, but also by other forces such as genetic drift and migration. Sexual selection is a subset of natural selection and is brought about by differential mating success caused by differences in the phenotypes of the individuals. For example, elk males with large horns may mate with a greater number of females even though the size of the horns may not enhance the fitness of these males in other ways.

Endler (1986) grouped traits under selection into 1) morphological, 2) physiological, and 3) biochemical traits. Morphological traits include such things as external body dimensions, color variation in snails and butterflies, beak size in birds, and so on. Physiological traits include life-history traits such as fecundity, resistance to herbicides or antibiotics, tolerance to heavy metals, and so on. Biochemical traits include allozymes encoded by genetic loci which appear to be affected by selection. Allozymes are alternative states of enzymatic proteins that are encoded by the same locus on a chromosome. Just how allozymes influence an individual's appearance, size, or color is not always understood.

Criteria for Demonstrating Natural Selection

The first criterion that Endler used to include examples of selection in his survey was that the traits were heritable, or were thought to be heritable; that is, the traits had a genetic basis, at least in part. This excluded several cases in the literature that showed that selection was associated with some trait, large body size for example, but that the trait was not heritable. A direct demonstration of selection is made by marking a sample of individuals and measuring them for a trait before and after selection, or by repeatedly measuring a trait in an unmarked cohort of individuals in a single age class.

Using these criteria, Endler estimated that by 1986 natural selection had been directly demonstrated for 314 traits in 141 species (Table 1). Most of these examples were for morphological traits, but some were for physiological traits, especially those involving tolerance or resistance to a stress, and a few were biochemical traits. The immediate importance of these numbers is to show that natural selection is not a rare event in nature. However, it is not correct to say that natural selection occurs on morphological or physiological traits more than on biochemical traits, because at smaller scales of physical organization, it becomes progressively more difficult to measure selection. For example, it is easy to measure the size of a bird's beak, but much more difficult to characterize a bird's physiology or biochemistry, before and after selection.

Table 1. Numbers of species and traits for which natural selection has been directly demonstrated. Endler (1986, p. 156).

Estimates of the strength of selection in nature are crucial for using models of gene flow and selection. Let us first look at polymorphic traits, or those traits occurring in more than one easily discernible state. These include such things as eye or hair color, or other discretely varying traits such as allozyme genotypes, which can be scored for each individual. For example, a researcher measures the survival of individuals during a drought and is able to show that survival is better in individuals with particular character states. One particular genotype might be associated with 80% survival, on average, whereas individuals with another genotype may survive only 60% of the time, on average (Fig. 1). This then is natural selection, the differential survival of different genotypes or phenotypes. The selection coefficient, s, is the difference in fitness between the genotype with the highest level of survival and the genotype of interest. The genotype with the greatest fitness (w) is generally given the value 1.0, and the fitnesses of other genotypes are measured relative to this value. The range of s is then between 0.0 and 1.0. The maximal value of s is 1.0, and such a level of selection is very strong. On the other hand, a value of 0.0 means that selection is not occurring on a particular genotype or phenotype. As noted previously, the effects of gene flow depend on the amount of gene flow and the strength of selection. If the migration rate is 50%, then a selection coefficient of 50% is needed to keep adaptive alleles in the natural populations. So just what are the levels of selection in nature?

Discrete characters

Endler classified the cases of selection into two groups, varying in the types of environments: A, selection in undisturbed environments, and B, selection in disturbed environments, including selection in environments with introduced organisms and selection in enclosures (Fig. 2A, 2B). Perhaps category B is most relevant to hatchery straying. Endler also classified examples into two groups according to the fitness components investigated: C, mortality selection, and D, data from fecundity, fertility, and sexual selection (Fig. 2C, 2D). Shaded bars represent values that are significantly different from 0.0 (P < 0.05). Unshaded values indicate the values were not distinguishable from chance fluctuation. The lack of significance for many of the small values may have been due to the inability of the experiment, because of small sample sizes for example, to detect small but real changes. The basic finding is that selection intensities vary considerably; some are strong, but most are weak. One problem with these distributions is that they represent a sample of values in the literature and not a random sample from nature itself. The lack of detectable selection is usually not reported, so there is a bias toward publishing large values of selection. Cases of weak, but real, selection are underestimated in the literature because the problems of measuring small intensities of selection are greater than those of measuring large selection intensities. The median of selection in undisturbed and disturbed environments is about 0.3, a high level of selection intensity.

Continuously varying characters

Next, we will look at traits, such as body mass, height, and gill raker length, that vary continuously in a population and not in discrete, countable units. Measurements of one of these traits usually show some distribution, with an average value (X) and a standard deviation () as in Figure 3. The unshaded distribution represents a hypothetical population before selection, and the shaded distribution represents the same population after some kind of selection. Directional selection occurs when the mean value of the trait shifts after a period of selection. The selection illustrated in this figure is quite strong, because no one under a certain value survives. Only individuals with extremely large values survive. The standard deviation is about one-fourth of the width of the distribution and is used to scale the coefficient of directional selection. The intensity of selection (i) is the difference between the means X₁ and X₂ divided by the standard deviation , and represents the amount of change in terms of standard deviation units. In this hypothetical example, the shift was about one standard deviation, which is quite large. A shift of two standard deviations is enormous. Endler used the index, i, in his survey.

The results of Endler's survey of selection, i, on continuously varying traits in natural populations appears in Figure 4. As before, shading indicates cases in which the shift was statistically significant, and did not result from chance alone. The median is about i = 0.3, but a few extreme values exceed 1.0. Keep in mind, however, that these are published values, and because of the tendency not to publish small values of selection, they probably under represent examples of low selection intensity.

As a specific example, Figure 5 shows the results for a population of song sparrows on Mandarte Island in Juan de Fuca Strait, which has been studied for several years by Jamie Smith at the University of British Columbia (Schluter and Smith 1986). Tarsus (a bone in the lower part of the leg) length in millimeters is shown versus the probability of survival after the first winter of life in female birds. Symbols (+) represent individual measurements of tarsus length before winter when the birds were banded for later identification. In spring, the presence (upper row) or absence (lower row) of a bird was used as a measure of survival: if present the bird was given a value of one, if absent a value of zero. The curve, then, shows the relationship between probability of survival and variation among females in their first year of life. A difference of only 3 mm, from 18 mm to about 21 mm, produced a reduction in survival from 90% to 20-30%. This shows that large changes in the probability of survival are produced by only small changes in the length of the tarsus. Such intensities of selection are quite common and are not limited to one life-history stage. For this population, about one selection event is detectable each year for such things as juvenile and adult survival, reproductive success, and so on. This number, however, probably underestimates the number of selection events because of our limited power to detect them.

Application of these values to models

It would be tempting to use these observed values of selection in the models for migration, genetic drift, and natural selection. If we did, these large values of selection would suggest that hatchery straying will not have a large effect on the genetics of wild salmon populations. However, let me mention several caveats. The first is that in virtually all cases of selection in Endler's survey, the actual cause of selection was unknown. Although differential survival or reproductive success may be associated with a particular heritable trait, we still have no idea of the mechanism of selection in most cases. Selection may actually be occurring on another unmeasured trait associated with the measured trait. Understanding the mechanism of selection is important so we can judge whether the selection is the result of factors in the local environment, or food supply, or whatever.

A second warning is that just because a trait does not show a change, it does not necessarily indicate that no selection is occurring on that trait. Strong selection may be holding the trait in place, the direction of selection may be different in the various life-history stages, or selection may be acting on another trait strongly correlated with the first in an opposing direction (Lande and Arnold 1983). A third problem is that selection may be occurring on a non-heritable trait or series of traits that are correlated with a heritable trait, so that selection is not actually acting on the heritable component of the variability. Traits that depend on the nutritional condition of an individual, such as fecundity or body size, are particularly susceptible to this kind of artifact. It is therefore difficult to extrapolate from the measurement of apparent selection to the kind of selection that may be operating.

Another problem is that only a single life-history stage was examined in most of the cases summarized by Endler. These estimates of selection, therefore, reflect only one component of fitness, and another view of selection may emerge if all life-history stages of an organism were studied. In some of the cases where selection was observed for more than one life-history stage, the directions of selection changed at the different stages (Schluter et al. 1991). For example, in the song sparrow population on Mandarte Island, selection on tarsus length operated in one direction in females in their first year, but in the opposite direction later in life (Fig. 5B). If you look over the whole life span, the two forces cancel each other out, but you would not see this if you observed only a single episode of selection.

Yet another problem is that the direction of selection may oscillate within or between generations. A well-known study of a species of Darwin's finches in the Galapagos Islands (Gibbs and Grant 1987), showed that, during a prolonged drought, 85% of the birds died, and those that survived were considerably larger for every trait (weight, wing length, beak length, etc.) measured. A subsequent, milder drought also produced changes in the same direction but on a smaller scale. When the rains associated with the El Nino returned, food abundance on the island increased and brought about selection for smaller individuals. In general, long-term studies show that the directions and intensities of selection are constantly oscillating, and the magnitudes of the selected characters tend to wobble not just within generations, but between generations. If selection is measured at only one life-history stage, it may appear that selection is powerful and operates consistently in one direction. Short-term measurements of selection tend to be large, but long-term average measurements probably tend to be smaller.

Finally, when we attempt to look at the relationship between variability for a trait and fitness, we would like to know about the intensities of selection in different populations. It is of great interest to know what kinds of selection produce differences between populations or even between hatcheries. Long-term studies of selection in a range of populations are indeed rare, but one such study does exist for the peppered moth (Biston betularia) in Britain. Dark pigmentation increases crypsis on tree trunks denuded of light-colored lichens by sulfur dioxide and darkened by industrial soot, and confers resistance to visual predation. The more ancestral light form was displaced by the dark form in industrial areas which produce high levels of air pollution (Kettlewell 1973). A cline developed near Liverpool, England in which most moths near the industrial area were dark and the moths farther away, 50 km or so, were more lightly pigmented. Experimental translocations of the dark and light forms allowed the researchers to measure changing natural selection pressures over the whole environmental gradient. Mortality of the light form was high near Liverpool and low farther away from Liverpool, whereas the opposite trend was seen in the dark form. Even this textbook example of selection may not be as simple as first thought; gene flow and some component of non-visual selection are necessary to explain, for example, the somewhat higher-than-expected frequency of the dark form in non-industrial areas (Brakefield 1987).

An important problem is to understand the extent that continuously varying traits are heritable. Many researchers assume that most of these traits are heritable, and that variation in only a few traits is caused by something other than genetic differences among individuals. Figure 6 shows the cumulative frequency distribution of heritability values in the literature for morphological traits such as beak size. This distribution suggests that the median heritability for these kinds of traits is about 40%; that is, about 40% of the variation in a trait is due to the additive effects of genes. This is quite high. Life-history traits tend to have heritabilities that are about 30%, and heritabilities for behavioral and physiological traits lie between these values.

Since life-history traits have lower heritabilities than morphological traits, you might think that the smaller the heritability, the more resistant the population is to selection; however, this is not exactly true. The unshaded curve in Figure 7 indicates the probability of survival or some other measure of fitness as a function of variation in that trait. The shaded curve indicates the distribution of a trait in a population experiencing selection. This relationship can be used to predict the magnitude of the response to selection in the next generation (i.e., the amount of evolution). The response is not actually determined by heritability, but by the absolute amount of additive genetic variation in the population experiencing a fitness function. The actual levels of additive genetic variation in life-history and morphological traits are about the same even though heritabilities differ (e.g., Price and Schluter 1991). Life-history traits have lower heritabilities than morphological traits because they are influenced by environmental variation to a greater degree. More environmental noise is associated with them. The message is that nearly all traits are heritable and that they do respond to natural selection.

My first conclusion is that natural selection is pervasive in nature, and my second is that the intensity of selection is quite strong. We have seen the patterns of selection for several kinds of single traits. However, numerous problems arise in attempting to use the estimates of selection from these limited studies, because the direction of selection may vary from one life-history stage to another or because of several other factors. The third conclusion is that since most traits are heritable to some extent, local selection on life-history, morphological, physiological, and biochemical variability confers adaptation to local conditions. However, none of these conclusions implies that selection is sufficiently strong or consistent in direction to overcome the effects of migration from non-adapted genes. These conclusions also do not imply that evolution is repeatable. Just because most traits are heritable does not mean a genetically altered population can revert to its original genetic state.

Brakefield, P. M. 1987. Industrial melanism: Do we have the answers? Trends in Ecology and Evolution 2(5):117-122.

Endler, J. A. 1986. Natural selection in the wild. Princeton University Press, Princeton, NJ, 336 p.

Gibbs, H. L., and P. R. Grant. 1987. Oscillating selection on Darwin's finches. Nature 327:511-513.

Kettlewell, B. 1973. The evolution of melanism: The study of a recurring necessity. Clarendon Press, Oxford, 423 p.

Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, applied to brain:body size allometry. Evolution 33:402-416.

Lande, R., and S. J. Arnold. 1983. The measurement of selection on correlated characters. Evolution 37:1210-1226.

Mousseau, T. A., and D. A. Roff. 1987. Natural selection and the heritability of fitness components. Heredity 59:181-191.

Price, T. D., and D. Schluter. 1991. On the low heritability of life history traits. Evolution 45:853-861.

Schluter, D. 1988. Estimating the form of natural selection on a quantitative trait. Evolution 42:849-861.

Schluter, D., T. D. Price, and L. Rowe. 1991. Conflicting selection pressures and life history trade-offs. Proceedings of the Royal Society of London 246:11-17.

Schluter, D., and J. N. M. Smith. 1986. Natural selection on beak and body size in the song sparrow. Evolution 40:221-231.

Question: Richard Carmichael: In the distributions of selection intensities you presented for discrete and continuously varying traits, it looked as though these studies reported more statistically insignificant than significant values for selection. In calculating the average values of selection intensity, were the non-significant values treated as zero?

Answer: Dolph Schluter: The median values I reported were estimated by eye and included all the values as they were reported, significant or not. I should point out that selection has been observed in 141 species, but that several different components of selection were observed in some of the same species, so the total number of estimated selection coefficients is greater than the number of species. The problem is that some of the observations may not be independent of other observations.

Question: Nils Ryman: Do you have any idea about how strong selection must be to be observed at all?

Answer: Dolph Schluter: The most important factor in detecting selection is the interaction between the strength of selection and the sample sizes used to measure selection. Small samples sizes decrease the power of an experiment to detect small selection coefficients.

Question: Audience: You said that evolution is not necessarily repeatable, but it is striking to me how similar odd- and even-year pink salmon are to each other even though they are reproductively isolated from each other. They use the same streams in very much the same way, but in different years.

Answer: Dolph Schluter: This is an indication that similar selection pressures can produce similar phenotypes, even though biochemical data indicate odd- and even-year populations at the same locality are reproductively isolated from each other.

Question: Richard Carmichael: Do estimates of heritability suffer from the same bias that estimates of selection coefficients do, in that estimates of zero heritability tend not to be published?

Answer: Dolph Schluter: Yes, the distribution of heritability estimates also suffers from the sample size problem. Undoubtedly many researchers have dropped efforts to measure heritability when it appeared they were not going to find heritable traits.

Table of Contents