​One measure of genetic diversity is the fraction of loci with two alleles that are different. There are two expressions of heterozygosity: expected heterozygosity (He) is what we would expect to see if the population is breeding randomly, and observed heterozygosity (Ho) is the actual fraction of loci that are heterozygous.
 
Because heterozygosity depends on actual allele frequencies, two populations with the same COI can have different levels of observed heterozygosity. I won’t go into the math involved, but suffice it to say that you can’t assume that Ho is necessarily 1-COI; that is, you can’t predict one from the other, especially in small, bottle-necked populations (think purebred dogs).
 
One tricky thing to warn you about is that inbreeding can vary from 0 to (theoretically) 100 percent. But heterozygosity is measured using SNPs, which are “biallelic”, meaning that for a given locus, there are only two possible states. Say you have two alleles, A and B; some dogs will be AA, some will be BB, and some will be AB. Because AB and BA are the same, at most only half of the dogs can be heterozygous. As a consequence, graphs of heterozygosity scale from 0 to 0.5, not 0 to 1.0. 

I have made use of published data for heterozygosity of various breeds of dogs (from Bannasch et al. 2021, and the massive study by Donner et al. 2018; ) to produce histograms that make it easy to compare genetic diversity within and across breeds. Within breeds, the range of values reveals the amount of variation in the sampled population. The Donner study supplied data for individual dogs, so I could display them as a histogram. In the Bannasch et al. study, the information provided was the range (blue line) and median (closed circle). (You can browse through the graphs here - https://www.instituteofcaninebiology.org/heterozygosity_donner.html )

For comparison, the distribution of Ho for a mixed breed population of dogs shows that most mixed breed dogs have Ho > 0.4, and in many Ho is > 0.45. As you scan through the graphs for breeds, you can see that there is much variation in Ho, with some showing many dogs with Ho > 0.4, while in other breeds the most dogs are < 0.3 (Basenji). (Note that for some breeds where the sample size is very small, you might see just a few bars that don’t say much about to expect for the breed.) 

These data are very interesting to browse, and can reveal a lot about the genetic health of a breed. Breeds with less genetic diversity in the population are shifted to the left. This tells you that if a breed is trying to breed away from an undesirable gene (or towards a favorable one), the low heterozygosity (read “diversity) could make this difficult. ​​For breeders faced with developing a cross-breeding program to introduce new genetic diversity into the gene pool, the best candidates for those crosses will be the ones with high heterozygosity and low COI; the worst choices will be breeds with low diversity, and you want to avoid high levels of inbreeding because all puppies will inherit the same allele from a parent for homozygous loci.

www.instituteofcaninebiology.org/hi_low_heterozygosity.html


I made two quick graphs to show this. One graph displays the heterozygosity and inbreeding data for the top breeds for genetic diversity, and the other is the breeds with the highest inbreeding and lowest heterozygosity. If you’re looking for genetic diversity for a genetic rescue program, you might steer away from the breeds on the low diversity (although they might be useful for other reasons). I am also working on some analyses that show how diversity in a breed varies among populations in different countries (coming soon…).

Information about heterozygosity can help us make better breeding choices, and Ho is one of the statistics that will be available for all breeds when the updating of the DogsArk Breeding Tool is finished (soon...).

​(Access the graphs for each breed here - )
https://www.instituteofcaninebiology.org/heterozygosity_donner.html

REFERENCES

Bannasch, D, and others. 2021. The effect of inbreeding, body size and morphology on health in dog breeds. Canine Medicine and Genetics 8:12. https://doi.org/10.1186/s40575-021-00111-4

Donner, J and others. 2018. Genetic prevalence and clinical relevance of canine Mendelian disease variants in over one million dogs. PLoS Genetics 19:31010651. ​https://doi. org/10.1371/journal.pgen.1010651

When a genetic disorder or trait is caused by a single gene, the usual protocol is to identify it and create a test to detect it. We can now test for hundreds of single mutations, which allows us to prevent producing any affected animals by screening potential parents before mating.

But for the many disorders thought to be polygenic, we have no simple solution. Identifying multiple or even dozens of genes or loci (a region on the chromosome where a gene is located) that seem to be associated with a particular disorder is of little use to the breeder. Should dogs carrying risk alleles be removed from the gene pool? Do you breed dogs together that have some of the risk alleles? How many overlaps is safe? There are no guidelines and no way to estimate the risk. Unfortunately, many serious and common problems appear to be polygenic, and we have been able to do little to control them. So things like cancer, epilepsy, cardiac and kidney disease, immune system disorders, and more sit like subterranian explosive devices scattered across the genomic landscape of your breed undetected.

DNA tests are not useful for these complex disorders, but we do have techniques that can provide more information about risk for polygenic disorders using just a pedigree database. The analysis uses information about the genetic relationships among the dogs in a population and the identity of animals that are affected with the disorder to determine the genetic risk of a disorder based on the genetic contributions from affected ancestors. This strategy uses a statistical technique called cluster analysis, and requires no information about the genes that might be involved or mode of inheritance.

What does cluster analysis do?

Cluster analysis using pedigree data does not test marker-trait associations. Instead, it tests whether family structure based on ancestry correlates with the distribution of affected animals.
​
In plain terms, it asks:

“Do dogs with this trait disproportionately descend from the same ancestors?”
​
The analysis does this regardless of the complexity of the pedigree, so it can identify risk that would be undetectable by studying individual pedigrees. It uses information about the affected animals, but also their relatives, to estimate the accumulation of genetic risk in lines and individuals.

This analysis has obvious advantages over trying to identify the specific genes involved. Polygenic disorders or traits might involve dozens or even hundreds or genes, each with a small effect, but which collectively affect the expression of a trait or disease. Teasing apart the effects and interactions of many genes is difficult, and even if you could, you would still not know how to use this information to make breeding decisions. Great knowledge, but not useful to the breeder.

​With a pedigree database, however, we can determine risk of producing affected dogs, putting a valuable tool in the hands of breeders that can use this information to take some of the uncertainty out of mate selection. With the right analysis pipeline, breeders can use basic information they already have about pedigree relationships and health records to, for the first time, put these data to good use in mate selection. 

These analysis are particularly useful for:

• Polygenic traits
• Founder effects
• Rare risk alleles
• Line-specific risk accumulation

​
The analysis does not require:

• Knowing the causal variants
• Assuming additivity
• Detecting large effect sizes

The beauty of this technique is that you don't need DNA data, your pedigree database doesn't have to be perfect or complete, and it will produce information that is actionable for beeders immediately. It produces a map of risk superimposed on the pedigree that identifies where breeders need to avoid selecting parents that are both from within the risk population. They can, however, breed dogs with risk to dogs outside the risk population to retain genetic diversity in the breed. 

I recently analyzed a pedigree database for Yorkshire Terriers to identify dogs at risk of producing offspring with liver shunts. The database was relatively small, only about 30,000 dogs (only a fracion of the complete breed database), which included about affected 40 animals. From this, the dogs with geneticc risk in the current population were clearly identified, giving breeders the opportunity to produce litters with low risk of liver shunt while protecting the genetic diversity of the current breeding population.

If your breed is struggling to manage complex health problems for which we have no suitable tests, this analysis might provide the information you need to avoid risk and the incidence of some of the most serious health problems in dogs.

Commercial DNA analysis has changed the way we breed dogs, giving us the ability to prevent disorders caused by recessive mutations with just a cheek swab. It has been less successful at addressing polygenic disorders. Identifying 12 genes "associated" with a health issue, with none identified as directly causal, poses a selection nightmare for breeders. What if your bitch has 4 of them and a potential mate has numerous others? What is the risk of producing puppies with problems? Can you discard all dogs with any of the genes, or even some of them, from the breeding program?

This is becoming a huge problems. Many breeds are wrestling with disorders that seem to be genetic because they run in breeds or families, but breeders can only guess at levels or risk involved in breeding a pair of dogs. Without some way to shed light on the underlying genetics of a probem, breeders have to weigh personal responsibility for health outcomes without useful information about risk. 

Several years ago, I wrote a blog post that described a way for breeders to deal with genetic disorders when nothing was known about the genes involved or mode or inheritance (Cool tricks with Kinship Coefficients, part 3: "How can I manage a disease without a DNA test?" ). In the past, the technique has been used to address several othewise intractable genetic problems in dogs, but it seems to be forgotton now that we look to DNA analysis to address genetic problems.

Recently, I was asked by some breeders of Yorkshire Terriers in Norway to see if I could help with the problem of liver shunts in this breed, and I decided to see if I could shed some light on the genetic basis of the problem using this technique.

The analysis uses a statistic computed from population genetics called the kinship coefficient, which quantifies the degree of relatedness between two dogs, assessed either from pedigree data or DNA. For a male and female, you determine the kinship coefficient for the pair, and that value is equal to the expected average coefficient of inbreeding in their puppies. This works because the kinship coefficient compares the genetic makeup of the sire and dam, then uses their genetic similarity (i.e., their relatedness) to estimate the degree of inbreeding if their genes were combined to produce puppies. In fact, this is how predicted litter COI is estimated in software that does test matings. 

So we have a statistic, the kinship coefficient, that allows us to assess the genetic similarity of two dogs. If you have a dog with a health issue (or any trait for that matter) that is suspected to be genetic, you would expect the dogs that are genetically similar to your dog would be more likely to carry the genes for the trait than dogs that were less similar. If you wanted to choose a mate that is less likely to carry the genes responsible for the health problem in your dog, you would look for individuals that are less similar genetically.

This seems logical enough, but how does this actually identify the dogs you could breed with that would have a lower risk of producing the disease? First, we need to identify as many dogs as we can that have been identified with the disorder, and we need the pedigree ancestry of these individuals. For a dog that has been diagnosed with the disorder, we know that it carries the risk genes that were inherited from its parents. This allows us to identify carriers of the risk genes, even though they are not affected and we don't know which specific genes are involved. 

Armed with the pedigree information for affected dogs in Norway, I performed a "cluster analysis" that identifies groups of closely related dogs using a "family tree" called a dendrogram, which is a visualization of the genetic reationships among groups of dogs. When I superimposed the information about affected Yorkies on the dendrogram, they all clustered in one group of relatives and there were none anywhere else among less related dogs.
​

This almost seems magical, but it's a very powerful technique. We have no easy way to breed for or against polygenic disorders or traits using DNA data, and staring at a pile of pedigrees doesn't tell a breeder much about risks in a quantitative way. But we know that the degree of genetic similarity of a pair of dogs will depend on how closely related they are. Just as we can use kinship coefficients to predict the average inbreeding of a litter of puppies, we can use them to evaluate risk of a genetic disorder.

The clever thing about this type of analysis is that it can be used for any trait with a genetic basis. It will work better when the heritability of the trait or disorder is high; that is, when most of the variation in the trait is due to genetics, and relatively less is a consequence of environmental (non-genetic) factors. In either case, it is a way of addressing genetic issues that are complex, involving both genes and environment. Most breeds have some issue that they are trying to deal with by selective breeding, but without any useful genetic information except knowing individuals that are actually affected. This is a way to address those problems.