SEARCH

SEARCH BY CITATION

Keywords:

  • genetic compatability;
  • good genes;
  • mate choice;
  • extrapair paternity;
  • heterozygosity

Abstract

  1. Top of page
  2. Abstract
  3. The simulation model
  4. Simulation results
  5. Implications of this study
  6. Acknowledgements
  7. References

A key question for molecular and behavioural ecologists who study mating systems is to understand why, in many species, females choose to mate with extra-pair males. Recently a possible explanation, ‘genetic compatibility’, has gained increasing empirical support (for a comprehensive review, see Kempenaers 2007). Genetic compatibility hypotheses assume that females seek extra-pair mates with alleles that complement their own. Typically, this will be achieved by mating with a male of a different genotype than her own, in order to maximise the heterozygosity of her offspring. Because numerous studies have indicated positive associations between heterozygosity and fitness (see Coltman & Slate 2003), it follows that mating with ‘compatible’ males will result in heterozygous, and therefore fit, offspring. Most empirical support for genetic compatibility has been obtained with microsatellite markers that have first been used to assess parentage and then to estimate relatedness and/or individual heterozygosity. A problem with this approach is a possible bias that favours the detection of extra-pair paternity when the extra-pair male has a genotype different from that of the female and her social mate. This in turn could lead to the erroneous conclusion that extra-pair males are less related to the female than within-pair males. In this issue of Molecular Ecology, Wetzel & Westneat 2009 (hereafter W&W), use simulation studies to assess the extent of this bias, using parameter estimates obtained from recent empirical data. They identify two forms of bias that may affect tests of the genetic compatibility hypothesis, and provide guidelines on how these biases may be avoided.


The simulation model

  1. Top of page
  2. Abstract
  3. The simulation model
  4. Simulation results
  5. Implications of this study
  6. Acknowledgements
  7. References

The logic underpinning why there may be a bias in favour of genetic compatibility is quite simple. In most studies, genetic markers resolve only a proportion of paternities and some extra-pair males remain unknown. Suppose a female and her social mate both have genotype AA at a marker and an extra-pair mate has genotype aa. Any offspring with genotype Aa must be sired by the extra-pair male, and the true parentage will be easily detectable by exclusion or likelihood. Under a second scenario the female and extra-pair male both have genotype AA and the social mate has genotype Aa. Here, extra-pair offspring will always have genotype AA but the social mate can also produce offspring with this genotype, and thus will be non-excluded. Similar verbal arguments can be presented for different allelic combinations, but overall a picture emerges that the probability of detecting and assigning parentage to extra-pair young is greatest when the mother has a genotype that is dissimilar to the extra-pair male. Therefore, detection of extra-pair young will be biased towards those extra-pair fertilisations where the female and extra-pair male have low relatedness, or put another way, the offspring have high heterozygosity. Additionally, the probability of detecting extra-pair young is reduced when social mates are highly heterozygous.

To assess whether concerns over this bias were real, W&W simulated bird populations — most of the empirical evidence for genetic compatibility comes from studies of passerine birds — where breeding pairs had broods of four offspring. Some of the offspring were sired by extra-pair males. Microsatellite genotypes were simulated and were used for parentage inference and relatedness or heterozygosity estimation. W&W then tested seven hypotheses commonly used to test genetic compatibility. Broadly, the first three hypotheses relate to the idea that females choose extra-pair males with high heterozygosity and hypotheses 4–7 examine whether females choose extra-pair mates that will maximise the heterozygosity of her offspring. For convenience, W&W lump these seven hypotheses together under the label ‘heterozygosity hypothesis of EPP’. No association between EPP and heterozygosity was built into the simulations.

In each simulation, the seven hypotheses were tested in two data sets: the data set where EPP had been detected by the markers (hereafter the ‘detected data set’) and the entire data set which contained both detected and nondetected EPP (hereafter the ‘actual data set’). Bias was measured as the proportion of statistically significant results obtained from the detected data set minus the proportion of statistically significant results obtained from the actual data set. Thus, if 4% of replicates were significant in the detected data set and 2.5% of replicates were significant in the actual data set, the bias would be 0.04–0.025 = 0.015.

Simulation results

  1. Top of page
  2. Abstract
  3. The simulation model
  4. Simulation results
  5. Implications of this study
  6. Acknowledgements
  7. References

W&W revealed two types of bias — some unexpected statistical biases and the anticipated bias due to non-independence of the loci used for paternity inference and heterozygosity estimation. First, the statistical biases. A priori, 5% of replicates of the actual data set are expected to produce statistically significant results; 2.5% in either tail of the test statistic distribution. However, this was not always the case. For example, when most offspring were sired by the social mate, more than 2.5% of tests support the genetic compatibility hypotheses. When most offspring were sired by the extra-pair male, the opposite was true. In other words, the statistical tests used to examine heterozygosity hypotheses do not always have the expected null distribution, and thus can give misleading results. Furthermore, W&W showed that tests of some hypotheses (particularly those that compare extra-pair males with within-pair males) are downwardly biased. This is because extra-pair and within-pair males are drawn from the same population (violating the assumption they are two different groups), and thus, the power to detect differences between the two groups is compromised if some birds are both extra-pair and within-pair males.

The anticipated bias concerns the paternity assignment process. Under a wide range of simulated parameters, W&W show that using the same set of loci to assign parentage and measure heterozygosity produces biased results in favour of heterozygosity hypotheses. These biases are most pronounced when a modest number (< 5) of loci are used, when markers have low heterozygosity (< 0.60), when a high proportion (> 0.30) of offspring are extra-pair young, and when there are a large number of breeding pairs (gt; 200).

Next, W&W extracted parameter values (for the number and variability of loci, sample sizes, proportion of extra-pair young, etc.) from 25 recent studies that have tested heterozygosity hypotheses, and ran the simulations using the actual values from each study. Although there was some bias in favour of heterozygosity hypotheses, for the majority of studies, the bias was small. Not all of the seven hypotheses were equally biased. The test with greatest bias was one where the null hypothesis is that homozygosity of within-pair males is not a predictor of the proportion of offspring that are sired by an extra-pair male. The bias is towards the alternative hypothesis that homozygous within-pair males are cuckolded more than other males.

Implications of this study

  1. Top of page
  2. Abstract
  3. The simulation model
  4. Simulation results
  5. Implications of this study
  6. Acknowledgements
  7. References

How will this study impact future studies of heterozygosity hypotheses? Most importantly, the biases that researchers in the field suspected were present were confirmed, and guidelines on how to avoid these biases were presented. Two key steps have to be taken to avoid bias. First, some (but not all) of the statistical biases can be eliminated by using randomisation-based tests to derive robust null distributions of test statistics. The bias that remains will probably make tests conservative. The second bias, caused by non-independence of markers used for parentage and heterozygosity estimation is more difficult to deal with. The most robust approach is to use separate markers for parentage and for estimation of heterozygosity. However, even this precaution has a caveat; the bias will only be removed if heterozygosity at the two sets of loci is independent. Given that it is now relatively straightforward to develop and type large numbers of markers (microsatellites or SNPs) in almost any species, we should now expect to see studies where the marker non-independence problem has been remedied. Indeed, this precaution may even be adopted as prerequisite for publication by some journals. Such a move may not be popular, but it would certainly make evidence in favour of genetic compatibility more robust.

Finally, where does this study leave the evidence for the heterozygosity theories of extra-pair paternity? It seems likely that researchers will re-examine the evidence for genetic compatibility hypotheses, both at the level of individual studies and for each specific hypothesis. Encouragingly, among the 25 studies that W&W focused on, there is more support for the idea that females seek extra-pair males who will maximise the heterozygosity of their offspring than can be attributed to bias alone. However, the evidence that females choose extra-pair males with high heterozygosity is much weaker, and probably could be explained by bias.

Acknowledgements

  1. Top of page
  2. Abstract
  3. The simulation model
  4. Simulation results
  5. Implications of this study
  6. Acknowledgements
  7. References

I thank Daniel Wetzel, David Westneat and Bart Kempenaers for insightful comments on this article. Kaspar Delhey provided the blue tit photograph.

References

  1. Top of page
  2. Abstract
  3. The simulation model
  4. Simulation results
  5. Implications of this study
  6. Acknowledgements
  7. References

Jon Slate's research group (http://www.jon-slate.staff.shef.ac.uk) are interested in microevolution and genetic architecture of fitness-related traits in wild vertebrate populations. They combine genomics tools, gene mapping theory and life history data from longitudinal, individual-based field studies to identify genes under selection in the wild.