In 1898, Hermon Bumpus gathered 136 house sparrows immobilized by an ice storm, noting that the averages of several morphological traits differed between survivors and nonsurvivors. This was one of the first attempts to measure the phenotypic selection component of Charles Darwin’s thesis, that adaptation is driven by heritable traits that affect fitness.

Since then, a vast literature on quantifying associations between trait values and fitness has emerged (1). The quantification of Darwin’s second evolution component—that such traits are heritable—required the development of quantitative genetics by Ronald Fisher in 1918 (2). Although the selection and genetics components can be combined to determine the expected change in any trait, of greater interest is the general adaptive potential of a population. On page 1012 of this issue, Bonnet et al. (3) present a meta-analysis of 19 studies showing the abundance of heritable variations in fitness and the potential for adaptation.

Fisher famously stated that “natural selection is not evolution,” meaning that if a trait is not heritable, no amount of selection will result in a change in the offspring of surviving parents. Fisher’s key to deciphering heritability was noting that parents pass along specific variants of a gene (alleles), rather than entire genotypes, to their offspring. The sum of all the single-allele effects for a given trait carried by an individual is defined as their breeding value (BV) for that trait. BVs are best understood in terms of deviations from the mean, so that a random individual has an expected BV of zero, which implies that its offspring will, on average, be average. The expected deviation of an offspring from the population mean is simply the average of the BVs of its parents. As a result, parents with exceptional BVs have offspring that, on average, deviate substantially from the population mean. Conversely, offspring from parents with modest BVs fall close to the mean.

The spread of individual BVs is a measure of the evolutionary potential of a trait. This is the basis of the additive variance of a trait, defined as the variance among BVs for that trait in a given population. If this variance is small, offspring have very little resemblance to their parents, whereas if it is large, exceptional parents tend to have exceptional offspring. If there is no additive variance for a trait, it will not evolve. More generally, if there is no additive variance in fitness, no trait will genetically respond to selection.

Thus, one of the holy grails in evolutionary genetics is to estimate the additive variation in fitness itself, which gives a general measure of the evolutionary potential of a population and places limits on the maximal response for any trait. This challenging estimation problem was tackled by Bonnet et al. using a collection of 19 long-term vertebrate population studies (covering a total of 561 cohorts and ∼250,000 individuals) from North America, Europe, Africa, and Oceania. The meta-analysis showcases an immense, but doable, effort in estimating this fundamental evolutionary parameter.

Bonnet et al. used total number of offspring, also known as lifetime reproductive success (LRS), as the measure for the fitness of an individual. The LRS parameter is converted to relative fitness simply by dividing LRS for an individual by the average LRS of the population, which allows for quantifiable comparisons across studies. In statistical terms, if a population shows a significant additive variance in fitness among its individuals, this implies that parents with higher LRSs than the population average also have a high BV for LRS, and thus their children also tend to have high LRSs.

Estimating BVs, and thus additive variance, is a common problem in modern animal breeding, built around using pedigree information. A BV exists even when the trait is not displayed, as it is a measure of how exceptional an offspring from that parent would be, if produced. In the case of milk production, information on the BV of a bull is provided by the observed yields of his mother, sisters, and daughters. The same pedigree machinery used by breeders can, in theory, be applied in natural populations to estimate the additive variance of any measured trait. Pedigrees for natural populations can be constructed using molecular markers, and closed populations of vertebrates are well suited for such analyses. Even with perfect pedigrees, the transition of pedigree methods from a large and well-structured domesticated population to a small wild population has been somewhat rocky (4). Domesticated pedigrees tend to be much deeper and denser than those for natural populations, resulting in greater precision in BV estimates. Furthermore, fitness is a problematic trait for standard pedigree methods, which assume trait values are continuous and follow a Gaussian distribution, whereas fitness data are highly discrete—a parent can only have an integer number of offspring, with a large point mass at zero, that is, individuals with zero offspring. Although there have been a few attempts to estimate the additive variance in fitness in wild populations using standard pedigree methods, the failure of the Gaussian assumption suggests that these are likely rather biased.

Bonnet et al. extended these pedigree methods by using a discrete Poisson distribution with an inflated zero value instead of a Gaussian and provided a much better fit for the fitness data. Using the improved fitting, their resulting average estimate of the additive variance in relative fitness, VA(w), was two- to fourfold larger than previous values. To put it in a more tangible context, this means that if the fitness of a population drops by a third, it would take roughly 10 generations to recover back to normal fitness levels. Hence, populations with shorter generation times might have a better chance to somewhat mitigate anthropogenic changes.

In nature, the target of selection is almost certainly a constantly shifting, high-dimensional (i.e., multi-trait) phenotype that may poorly project onto individual traits or even a set of traits. Most studies of adaptation are structured around some assumed edifice of traits that affects fitness. A poor choice of traits can give a misleading impression of population adaptation. Fortunately, an estimate of VA(w) provides an upper bound, and therefore a maximal possible change in any trait independent of selection. For example, a typical trait heritability of 0.3 will mean that 30% of the trait variation is due to variance in BVs, and the maximal possible change in the average value of a trait in the population is about one standard deviation every four generations. A more reliable way to estimate VA(w) can help to better quantify the nature of selection and the robustness of a population to major environmental changes.

SCIENCE

26 May 2022

Vol 376, Issue 6596

FREE JULIAN ASSANGE NOW•••••••••••••••••••••••