Lab 6 -- Measuring Selection in Natural Populations


The determination of how natural selection acts in contemporary populations constitutes an important link between the studies of ecology and evolution. While we have referred to natural selection as a "force" that causes populations to evolve, it is perhaps more properly considered as an outcome of an interaction between phenotypic variation in a population and the current environment that population experiences (where the environment is broadly construed to include abiotic and biotic factors). This interaction leads to consistent differences in survivorship and/or reproduction between phenotypic variants, one of the criteria for natural selection to operate (see the introduction to Lab 2). Understanding how biotic interactions and/or the physical environment create selection may provide a clue as to how the current characteristics of a population have been molded through evolution. Through such studies, we come to appreciate the link between ecological interactions and their evolutionary effects.

If we constrain our study of selection to differences in viability (as we will in this lab), we are looking for significant associations of phenotypic variants with the probability of surviving. Phenotypic variation in populations often takes the familiar form of the "bell curve," defined mathematically as the normal distribution. The effects of selection can be seen by comparing the distribution of phenotypes in the population before and after selection acts (i.e., before and after individuals die); specifically, we can compare the means and variances of the distributions and look for three types of effects of biased survivorship associated with different phenotypes:

(1) Directional selection -- The population of survivors can have a higher or lower mean value for the characteristic than the population before selection acted. If individuals with larger values of the trait survived with higher probability, and therefore the mean after selection is greater than the mean before selection, we say that "upward" directional selection has occurred. "Downward" directional selection has occurred when smaller individuals survive with higher probability.

(2) Stabilizing selection -- The population of survivors can have a reduced variance of the characteristic compared to the original population, if individuals with extreme phenotypes have higher rates of mortality than individuals with intermediate phenotypes.

(3) Disruptive selection -- The population of survivors can have a higher variance compared to the original population, if individuals with intermediate phenotypes have higher rates of mortality than individuals with extreme phenotypes.

It’s important to note that selection can affect both the mean and the variance of populations, i.e., both directional and stabilizing(or disruptive) selection can occur.

Comparing selection events -- Our primary goals in studying selection in natural populations is to compare the strength of selection on different traits, different species and between events on the same traits at different times or place. From the first two we can potentially learn why some traits or species evolve and others do not; from the latter, we learn how consistent natural selection is, i.e., whether we should expect traits to evolve in certain directions over long periods of time and whether we should expect different populations to evolve in different directions. Can you think of why this might be important?

But how can we compare selection on two traits or over two events? Let’s just consider directional selection: It might seem logical to compare traits or event for how much the mean changes, but there are two problems with this approach:

(1) Scale bias -- Let’s say you are measuring the effect selection on body mass in your favorite insect and you obtain an answer of 10.2 mg for the strength of selection (i.e., , where refers to the mean of the trait z). But if I repeat your study and measure mass in grams (rather than milligrams), I would get an answer of 0.0102. The problem gets even worse if you wanted to compare selection on body size in elephants vs. flies, for example. Comparing the figures differences in means between species would not be very illuminating.

(2) Variation bias -- Consider the figure below, which illustrates the phenotypic distributions of two populations that have the same mean values for the trait being measured both before and after selection.


Thus, if you use the difference in mean values before and after selection (2 cm) as the "strength" of selection, you would conclude that selection acted in the same way in both populations. However, there is a real sense in which selection is more "intense" in population B than in population A: For selection to change the mean value by 2 cm in population A, the difference in survivorship between those individuals above and below the original mean of the population (10) does not have to be very large; however, for population B to move in mean 2 cm upwards would require virtually ALL individuals smaller than 12 cm to die!

Both these problems can be avoided by standardizing the measurement of change in the mean before and after selection by dividing by a measure of the original amount of variation, the standard deviation (s). This value, called the intensity of selection has the following formula:


Natural History of the Solidago-Eurosta System

Galls are growth deformities induced in certain plants by various insects. These interactions are frequently species-specific, with a particular species of insect inducing galls in a specific tissue of one species of plant. Galls are used by the insects that induce them as sites for larval development and as food. Characteristics of the gall are often under the influence of both the insect that provides the stimulus for gall formation and the plant producing the gall. Therefore, some features of gall morphology may evolve in response to selection on the gall-forming insect. Previous work on this insect-plant system has shown that gall diameter is a heritable character of the insect, as well as the plant. In this laboratory an analysis of gall diameter will be used to determine whether there is selection on the gall-inducing insect for gall size.

Solidago gigantea, or Late Goldenrod, is a common perennial of the eastern and midwestern United States that is frequently parasitized by the gall fly, Eurosta solidaginis. In the spring, adult female gall flies lay a single egg in each of many terminal buds of developing goldenrod shoots. The fly larva tunnels into the stem just below the apical meristem, where it secretes compounds believed to be similar to normal plant growth substances. As a result the plant undergoes abnormally high rates of cell division in the area occupied by the larva, resulting in the formation of a spherical gall. Gall fly larvae feed off the plant tissue, growing to full size by early Fall, overwintering in the gall, and pupating in the Spring. After metamorphosis is completed in May, the adult emerges from the gall to seek a mate. [Note that a related species, Solidago altissima, is also attacked by a separate, reproductively isolated, host race of this fly species. The natural history of this interaction is almost identical to that between the fly and S. gigantea, and has received a greater amount of study.]

Sources of Eurosta Mortality - Mortality of fly larvae within galls may result from a number of different causes, including interactions with predators or other herbivores of the goldenrod. The following lists the major, diagnosable causes of mortality in this system:

(1) Parasitoid wasps --- Parisitoids are insects that lay their eggs on or in a host, but whose effect is to kill the host (unlike a true parasite). The wasp Eurytoma gigantea is such a species -- a female wasp inserts its eggs into the central chamber of goldenrod galls. The resulting wasp larva eats the fly larva, and then switches to a vegetarian diet, eating gall tissue the rest of the growing season. Flies in smaller galls may be more susceptible to attack by this parasitoid wasp, since wasps can attack only those fly larvae that are within reach of the wasp's ovipositor. If this is the case, then attack by wasps may be a factor causing directional selection on the size of galls induced by the flies.

(2) Bird predators -- During the winter, downy woodpeckers (Picoides pubescens) and black-capped chickadees (Parus atricapillus) also prey on the gall fly larvae. These birds peck through the tissue of the gall and extract the soft-bodied fly larva. These birds are visual predators and thus larvae living in galls more easily seen by birds may suffer higher rates of mortality. If gall size is a determinant of which fly larvae are attacked by birds, then predation by birds will cause directional selection on the size of galls induced by flies.

(3) Other herbivores -- The stems and galls of the goldenrods are attacked by a large number of herbivorous insects. One common herbivore often found in the galls is Mordellistena unicolor, a beetle species that lays its eggs on the surface of the gall early in the summer. When many larvae burrow into the gall tissues they often cause the death of the fly larva and may consume it. If gall size is a determinant of which fly larvae are killed by this herbivore, beetle attack will be a cause of selection on the size of galls induced by the flies.

(4) Plant interactions -- Plants may have mechanisms to resist herbivory, in some cases causing the death of the herbivore. This may be an explanation of the phenomenon of Early Larval Death for Eurosta flies -- the gall continues to form although the fly larva has died early during gall formation. This is a common cause of mortality for flies on S. gigantea, and often leads to smaller than average gall sizes due to the early death of the gall inducer. Since the gall size is not indicative of a fly phenotype in this case (the fly hasn’t been present to stimulate gall growth throughout the entire growing period), we will remove these galls from the analysis before estimating selection on gall size.

By collecting galls, measuring the phenotypic distribution of gall sizes and determining how the distributions change after mortality agents act, you will be able to estimate the strength and form of selection on this phenotype for these populations. You will also be able to look at the rates of mortality due to different agents, which will help you interpret why selection has acted in the way it has. Finally, you’ll be able to compare your data to the data from last year’s class and the literature to determine how consistent selection has been over time or place. This may be important in understanding why populations have the phenotypic traits they do.



Sampling Galls (Week 1) -- When collecting a sample of individuals from a population, it is important to consider carefully how the methods used to choose measured individuals may bias the results. Sampling is a complicated area of ecology, with different techniques used in different situations. Ideally, we want to choose individuals from a population randomly, although sometimes this is not practical. The technique described below does not produce a truly random sample of the population of galls, but should guard against systematic biases in the sample (can you think of biases that might be introduced by other methods of sampling?):

  1. Lay out a 30m measuring tape along one edge of the population and determine the end-points of belt transects along this tape using a random number table.
  2. Run a belt into each population perpendicular to the end-point line using the long measuring tapes.
  3. Collect all galls within 0.5 meter of the measuring tape. Make sure you do not miss small galls!

Data Collection (Week 2):

1. Measure the diameter at the widest point of each gall by fitting it into the metric template -- find the smallest hole that the gall will fit through.

2. In order to determine the fate of the fly that induced the gall, carefully cut open the gall with the pruners and examine the contents.

a. If a cream-colored, fat larva or a tan-colored fly pupal case is present in the gall, that fly has clearly survived all the mortality agents discussed above and will almost certainly survive to emerge later in the spring. All other galls were induced by a fly that did not survive.

b. After examining each gall, categorize the fate of the fly that induced the gall. Examples of galls in each class will be available in lab. For each gall you will record the diameter and the fate on the data sheet and in the class computer file.

Data Analysis -- All members of the class will analyze the same data set, which will be posted as an EXCEL file on the Macintosh common servers. [If you prefer to work with a Windows machine, bring a IBM-formatted disk to a Macintosh and copy the file onto it. EXCEL for windows should be able to read the data file.] While you may consult with others about the analyses, each of you should produce your own results, including figures and/or tables for your paper.

The data are organized into two columns, "Fate" and Gall size". First, organize the data so you can use EXCEL to do all the number crunching for you:

  1. Sort the data by gall fate by highlighting all the cells in both columns, choosing Sort from the Data menu, and indicating that you want to sort the cells by the "Fate" column.
  2. Divide the data into separate columns on Sheet 2. Start by transferring the list of gall sizes for all galls EXCEPT those that died due to Early Larval Death into the first column, "Pre-selection." Then highlight all the gall sizes associated with "Survivors", copy them, and paste them into the "Survivors" column on Sheet 2. Do the same with gall sizes associated with mortality due to parasitoid wasps, birds, beetles, and unknown insects.
  3. Calculate the rates of mortality for different agents (% killed by each agent), and the rate of survivorship for flies (% survived), using the number of "pre-selection" galls as the numerator for these rates. [You may also wish to calculate the rate of early larval death, as it may prove interesting in interpreting year-to-year and/or site-to-site variation in overall fly survivorship.]
  4. Calculate averages, variances, and standard errors for each column. [In a cell at the bottom of the column type "=Average(", then highlight the column of data, type ")" and press <Return>. Type "=Var(" to get the variance. The S.E. can be calculated from the variance and the sample size.]
  5. Determine whether the mean size of galls attacked by the each of the mortality agents is significantly different from the mean size of "Pre-selection" galls, using the criterion of overlapping 95% confidence intervals. Note that if a mean size of gall for a mortality agent is LESS THAN that of pre-selection galls, that agent imposes directional selection favoring LARGER galls.
  6. Determine whether the mean gall size for Survivors is significantly different from the mean for pre-selection galls, using the 95% C.I. test. This will tell you whether the total effect of all mortality agents has resulted in directional selection.
  7. Calculate the intensity of directional selection for gall size in this population.


Are changes in variance significant?

How would you determine whether stabilizing or disruptive selection is occurring? Remember that we want to know whether the phenotypic variance has increased or decreased after selection has acted. But just as it is legitimate to ask "how much different does it have to be to be REALLY different" when we compare two means, we need a statistical test to determine how much of a difference in variances is significant (i.e., not likely due to chance). This test is called the F-test.

Here’s the problem: Imagine you have two populations that have the same variance, but you estimate that variance by taking a sample of each. If you take a small number of individuals from each population, it’s quite likely by chance that your estimates of the variances would be different. The more you sample, the more likely the two variances should be equal. The statistical test uses this principle by asking you to express the difference between the variances as a ratio (larger variance divided by smaller variance) called the F-ratio. The null hypothesis is that the variances are equal, i.e., that the ratio is equal to 1. The F-test asks whether ratio you got could happen easily by chance (given how many individuals you sample) when the true ratio is 1; if the probability of getting this ratio by chance is less than 5%, then we reject the null hypothesis of equal variances and say that the variances are significantly different.

To test whether there is significant stabilizing or disruptive selection , calculate the ratio of pre- and post-selection variances (larger variance over smaller variance) and compare that ratio to the appropriate critical value in the table below. [Remember that d.f. = n-1 and use the largest value in the table below that is less than your actual d.f.] If the ratio is higher than the critical value, you can reject the null hypothesis of equal variances with 95% confidence.

Critical values of F (p = 0.05)

d.f. numerator












d.f. denominator


















9. The class server has data sets from the last two years available on it (note that the class samples two different sites at CERA, "Plantation" and "Lake"). Choose at least one other data set and compare your results to it. Repeat your calculations for the other data set(s) and compare the results.

Questions for consideration in your papers:

1. What do your analyses suggest about the evolution of gall size? Can you think of possible constraints on adaptation by the gallmaker to such selection?

2. Is natural selection consistent from year-to-year or place-to-place? Why or why not? What are the ramifications of this for understanding the evolution of the flies?

On to Lab 7 - Field Community Ecology

Back to Lab 5 - Animal Behavior

Back the BIO 136 Homepage