Hip dysplasia is a complex disease. When a trait or disease is described as 'complex,' it is usually meant that the trait is influenced by both genetic and nongenetic, or environmental, effects. This makes it very difficult to determine the mode of inheritance, since the phenotype (the observable manifestation of the trait) is not necessarily an accurate indicator of the genetics; the genetics is only a part of the picture and is 'overlaid' by environmental influences ('good' genetics may be masked by detrimental environment and vice versa). Furthermore, the trait is often under the control of more than one (and usually many) genes, meaning that we can no longer categorise individuals as clear, carriers or affected. In this lecture I will attempt to demonstrate the presumed genetic architecture of complex traits and show how we can achieve more effective selection in spite of the problems bequeathed by this complexity.

You are probably all familiar with Mendelian inheritance, which Gregor Mendel demonstrated with a 3:1 ratio of yellow to green peas. This ratio allowed him to infer that the trait of pea colour was determined by 2 variants (alleles) at a single gene; the yellow allele (A) being 'dominant' and the green allele (a) being 'recessive.' This meant that the two possible phenotypes of pea colour (yellow and green) were in fact produced by three possible genotypes. Homozygotes (so called as both alleles are the same variety) with two yellow alleles (AA) produced yellow peas, and those with two green alleles (aa) produced green peas. Heterozygotes having one green and one yellow allele (Aa) were yellow in appearance (phenotype), the dominant yellow allele masking the recessive green. This is important since it allows phenotypic variations - in this example, the green pea colour - to apparently disappear for a number of generations before suddenly reappearing.

Heterozygotes produce half their gametes (sex cells) with the A allele and half with the aallele. Therefore, progeny of two heterozygotes will have the genotypes AA:Aa:aa in the ratio 1:2:1, but because the yellow allele A is dominant, the phenotypic ratio is 3:1 (see Figure 1). However, this 1:2:1 ratio is very important, because the 'dominance' we have encountered up to this point is not universal or complete across all traits or diseases.

Figure 1. Punnett square showing the genotypes and phenotypes from crossing two heterozygote parents

Consider for a moment (hypothetically) that gene A (with 2 alleles A and a) determines the quantity of peas rather than their colour. So AA might yield 9 peas in each pod, while aa yields only 3. If the A allele is completely dominant, then we expect the heterozygote to show the same phenotype as the dominant homozygote; so in this hypothetical example Aa yields 9 peas per pod. However, as mentioned above, dominance is not universal or always complete. For example, imagine that instead the heterozygote yielded 6 peas per pod - halfway between the two homozygotes. We can begin to look at things more quantitatively, plotting the number of peas per pod against the number of Aalleles:

Figure 2. Hypothetical examples of complete dominance (L) and completely additive (R) A allele

These are two important examples; the chart on the left in Figure 2 shows complete dominance, i.e., the heterozygote (Aa) is the same phenotypically as the AA homozygote. The chart on the right shows no dominance, or complete additivity (i.e., the heterozygote is the intermediate of the two homozygotes, and each A allele adds 3 peas per pod). Additivity is an important concept as we move on to consider 'genetic variation.'

Complete additivity at a single gene will give us a 1:2:1 ratio of phenotypes (reflecting the genotype ratio). However, as stated earlier, many quantitative or complex traits are influenced by multiple genes. As we consider perfect additivity over an increasing number of genes (Figure 3), we can see the phenotypic distribution (discounting nongenetic effects for a moment) approaching a 'normal distribution' (also known as the 'bell curve,' and very important in statistics). Figure 3 shows (from left to right) the genetic distributions of traits controlled by 1, 3 and 6 genes, respectively, followed by a normal distribution on the far right. Hopefully you can see that increasing the number of genes increases the number of phenotypic categories and begins to produce continuous genetic variation for the trait or disease in question. Thus, we have moved from thinking in terms of 'clear,' 'carrier,' and 'affected' to thinking in terms of a continuous scale of liability or risk.

Figure 3. (L to R) Genotype frequency distributions for 1, 3 and 6 completely additive genes, and a normal distribution (far right)

This is probably not as novel a concept as it may appear; think about when you hear news reports about scientists having found a gene for cancer, heart disease, diabetes, Alzheimer's, etc. - it's always a gene, not the gene. There isn't a single gene for any of these diseases, just as there isn't a single gene for height or weight. So, for complex diseases like hip dysplasia, we will have to deal with the concept of genetic variation and risk.

But the complexity doesn't end here. As mentioned at the outset, complex traits are influenced by both genetic and environmental factors. While the genes (and so the genetic risk) are determined at conception, this risk is subsequently modified by the effects of numerous known and unknown nongenetic or environmental influences. Think of heart disease; I may have a moderate genetic risk, but if I smoke, eat a poor diet, eat too much, take no exercise, and have a stressful lifestyle, my actual risk creeps up. My actual risk when I'm 50 may be higher than a 50-year-old with a higher genetic risk, but who watches their weight, eats healthily, has never smoked, and has a low-stress lifestyle. The same is true for hip dysplasia, where known environmental effects include diet and early-life exercise regime.

Nevertheless, genetics makes an important contribution to the overall risk. The heritability of a trait tells us how important genetics is relative to nongenetic effects - strictly it is the proportion of phenotypic variation that is due to genetic variation. For hip dysplasia, about 40% of the overall observable variation is due to genetic variation. This may not seem much - it is less than half after all - but it is by far the biggest single component.

However, when it comes to breeding, it is only the genetic risk we are concerned with, as it is only genetics that is passed across generations. This presents us with a problem - we are using phenotypes (hip scores) to guide our selections, but we know that they are not necessarily the best guide to genetics. We may unwittingly choose a dog with a good hip score, not knowing that this is actually more to do with a beneficial environment and that the genetic risk, which is passed to the progeny, is actually fairly high. But to date, hip scores are all that breeders have had to guide them.

This is where estimated breeding values, or EBVs, come in. EBVs are a quantitative estimate of the true genetic risk, or breeding value. We make the estimate using trait information, in the case of hip dysplasia using the hip score, on an individual and all its relatives. We are able to do this thanks to the availability of pedigree information, which allows us to quantify the relationship between all the dogs therein. Information on relatives, who share genes to a quantifiable degree, will allow us to make a better judgment on an individual's genetics. For example, we may feel very differently about using a stud dog with a poor hip score if we knew that he had over 50 progeny scored with a very good average hip score. The performance of the progeny tells us about the genetics of the parent. In fact, this aspect has been key to the success of EBVs, which have been extensively used in the dairy industry for over 20 years. Here we are concerned with milk production traits - traits that are only expressed in females. Yet we have very accurate EBVs for dairy bulls based on the milking performance of thousands of their daughters. Somewhat paradoxically, we know more about a bull's genetics with respect to milking traits than we do for any cow!

As with all estimates, it is useful to know how good an estimate the EBV really is. It is intuitive that we will have more confidence in the genetics of the stud dog mentioned above, with his own hip score and scores of 50 progeny known, than a stud dog with no information on itself or its progeny. Just as EBVs are a quantitatively formal way of taking account of relatives' information in the assessment of an individual's genetic liability, so we can formally calculate the accuracy of our estimate of true genetic liability (the EBV).

So, EBVs are a more accurate indicator of genetics than an individual phenotype - and are more abundant. Because we calculate the accuracy of each EBV, we are able to quantify how much more accurate selection using EBVs will be than selection using phenotype, with more accurate selection delivering greater genetic progress. Results from research show that selection using EBVs is an average of 1.16 times more accurate than using hip scores, across 16 breeds (Lewis et al. 2013). Furthermore, EBVs are available for all animals in the pedigree. Selection using EBVs of dogs too young to have their own hip score is on average 1.30 times more accurate than selection using the parental phenotypes (Lewis et al. 2013). We also showed that a far greater proportion of animals had an EBV more accurate than knowing both parental hip scores, than actually had both parental hip scores known, demonstrating that EBVs are an effective way of providing more reliable information on a far greater proportion of the breed or population.

Further improvements in the accuracy of selection have been shown to be available from the way we use the phenotypes. For example, we have shown with selection index methodology that for Labradors EBVs for elbow dysplasia score are up to 10% more accurate when computed from a bivariate analysis of elbow and hip scores than from a univariate analysis of elbow scores alone. A positive genetic correlation between hip and elbow score means that hip score acts as a much more abundant, if slightly less accurate, indicator of elbow dysplasia (Lewis et al. 2011). This method was also used to determine a more effective combination of the nine component traits of the UK hip score than a simple aggregate total (Lewis et al. 2010b), again delivering more accurate selection.

Finally, it is important to remember that EBVs are simply a more effective way of using the hip score data in selection - they are not a direct replacement! Quality data is critical for the calculation of accurate EBVs. Furthermore, hip scores themselves have significant prognostic value for individual dogs.

References

1.  Lewis TW, Blott SC, Woolliams JAW. Genetic evaluation of hip score in UK Labrador retrievers. PLoS One. 2010a;5(10):e12797. www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0012797

2.  Lewis TW, Woolliams JAW, Blott SC. Genetic evaluation of the nine component features of hip score in UK Labrador retrievers. PLoS One. 2010b;5(10):e13610. www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0013610

3.  Lewis TW, Ilska JJ, Blott SC, Woolliams JAW. Genetic evaluation of elbow scores and the relationship with hip scores in UK Labrador retrievers. Vet J. 2011;189:227–233. www.sciencedirect.com/science/article/pii/S1090023311002383

4.  Lewis TW, Blott SC, Woolliams JAW. Comparative analysis of genetic trends and prospect of selection against hip and elbow dysplasia in 15 UK dog breeds. BMC Genet. 2013;14:16. www.biomedcentral.com/1471-2156/14/16