Through a series of clever experiments, Mendel opened the black box that held the secret of the gene, which he concluded was responsible for the inheritance of traits in plants and animals. Knowing this, breeders could do some simple breeding experiments to reveal whether the genes for particular traits were dominant or recessive, which allowed them to design breedings that produced predictable traits.

While this took some of the mystery out of breeding for many traits, for other traits prediction remained elusive, and breeders realized that in some cases the genetics of a trait must be more complex. The problem, of course, was that some traits are determined by many genes. How could breeders do selective breeding on these so-called polygenic traits to reliably produce particular traits?

The puzzle was finally solved by a university geneticist named Jay Lush. In the 1930s, Lush realized that, for many traits, there could be lots of variation from individual to individual in a trait. That is, some traits were not "binary".

​For instance, there could be a range of size in a population of cows from small to large and everything in between, and not just two (binary) sizes of large and small. He realized that this was the case for many traits and began to think about how genetics could account for this. What Lush had stumbled upon was the key to selective breeding for traits that can vary continuously - genetic variation.

He described this in his groundbreaking book, Animal Breeding Plans.

Here, Lush is making the two observations that transformed animal breeding in the mid-1900s: 1) that genetic variation provides the raw material for selection, and 2) that some of this variation was "additive". Put another way, the effects of multiple genes acting on the same trait can be added. One large dog bred to another large dog could produce offspring larger than either parent.

This is what he is alluding to when he says that it's not necessary for a breeder to find the exact traits desired in a potential breeding animal, because genetic variation can be used to create a new mix of genes that will produce a trait not present in the current population. 

Lush's observation launched a new field of genetics, called "quantitative genetics", that was based in mathematics and statistics and applied specifically to traits that vary continuously. The advances in our understanding of quantitative genetics subsequently revolutionized animal breeding and was the basis of the spectacular improvements in production that became the modern system of commercial breeding of livestock and other animals.

One of the best examples of the gains that could be achieved by understanding the basis of additive genetic variation is for milk production in cattle. At the beginning of the 1900s, average milk yield was about 2,000 kg per year.There were no cows to be found anywhere that could produce 8,000 kg of milk per year. However, with the application of selection based on quantitative genetics in the 1940s, milk production soared and by 2000 it averaged more than 8,000 kg of milk per year - a four-fold increase in milk production over a period of about 50 years. Gains in production were made incrementally, a bit more each generation, and the basis for progress was finding advantageous combinations of genes that, added together, would push a trait in the desired direction. Breeders used the genetic variation in the animal population to create new combinations of genes in the next generation, then selected from among those animals the ones that improved on the trait over their parents.

There was one critical thing that was absolutely necessary for this to work: there needed to be genetic variation in the population. Breeders had to protect the genetic variation in their breeding stock, because without it new combinations of genes would not be possible and there would be nothing to select for.
​

The realization that selective breeding could produce animals that were superior to both parents revolutionized animal breeding. The use of additive genetic variation to produce animals with traits not present in the original breeding stock became the fundamental process of selective breeding in domestic animals.

Of course, dogs provide many examples of traits that have been shaped by selective breeding that takes advantage of additive genetic variation.

Body size is an obvious one. Researchers have identified multiple genes that can be associated with variation in body size in dogs. Some of these have a large effect, accounting for a significant fraction of the variation in size. But there is still much variation that is not yet accounted for genetically, and there might be dozens or even hundreds of other genes that have tiny effects individually but collectively account for differences among individuals.

Let's look at another example.

People probably started racing dogs thousands of years ago, and it's fair to suspect that over this time there has been selection to improve speed. 

One option would be to breed only the best to the best. Identify the sire and bitch with the fastest racing times and breed them together. Will some of the offspring be faster than either parent? Perhaps. Might some be slower? Perhaps. So what do we do from here?

We could mate the fastest male in the litter with the fastest female, and again select the fastest of their progeny. 

We might end up with dogs faster than the first two parents, but over the generations of inbreeding and strong selection (breeding only "best to best"), progress will start to level off. We will have eliminated much of the original genetic diversity in the dogs of the first generation. In fact, the puppies in each litter will share a larger and larger fraction of the same genes every generation. We are not going to get faster and faster dogs forever. Why? Apart from the deleterious effects of inbreeding, we have tossed out the genetic variation necessary to produce improvement in a continuous trait by exploiting additive genetic variation. We have bred ourselves into a dead end.

So how can you produce faster dogs from selective breeding and not run into this problem of a genetic dead end?

Redefine what you mean by "best". We do definitely want to select the best animals to breed, but we need to think of it in terms of the genetics of the population. To improve on a polygenic trait, we need genetic variation to select for. Instead of selecting a single best dog, we can select a subpopulation of dogs that are all faster than average. We will then use the genetic variation in those dogs, and the random inheritance of that variation in the offspring, to push racing performance in the direction of higher speed.

Can we keep producing faster and faster dogs forever? Probably not. If we don't run out of genetic variation, at some point we will run up against the limits of design and physiology. But we can definitely use selective breeding in this way to improve most traits that are polygenic in dogs.

So think about this. What traits do you strive for that are polygenic? Size. Temperament. Working ability. Coat quality. Ear length. Muzzle length. Shoulder angulation. Intelligence. The list is long.

Note that there is another issue here that breeding in this way addresses. The phenotype of polygenic traits is likely to be a reflection not just of genetics, but also of a roster of non-genetic factors, some of which might be known but most will not. When you select only the "best" individual to breed, you are selecting for the combination of genes and environment that produced what you perceive to be best. But you really only want to be selecting based on genes, because only that part of a trait can be inherited. Maybe the fourth "best" dog has great genes for the trait you want but was raised in a less than ideal situation. That dog's genes will get tossed if you only select the single best animal. By breeding a subpopulation of the top performing dogs, you preserve the genetic variation you need for selection and also reduce the influence of non-genetic factors in your choice.

You have no doubt noticed that this method of breeding, where the "best" is defined as a subpopulation of individuals instead of a single dog, is not the way dog breeders usually breed. We usually choose to breed only the dog we perceive to be the "best", ignoring the possible effects of environment on our assessment of the other apparently-inferior dogs. Doing this, we narrow the gene pool and increase inbreeding every generation, both of which will limit our ability to continue to improve traits. This applies to health as well as physical and behavioral traits. We can't "breed around" health issues if the necessary genetic variation no longer exists in the population of breeding animals.  

Can we breed better dogs by taking advantage of additive genetic variation? Yes, by changing the way we define "best" during selection. The improvement we can achieve might be significant - in health, in physical traits of importance, in temperament, and many other traits that we care about. 

Let's breed better, healthier dogs. Learn about the quantitative genetics of continuous traits and how you can use this knowledge to your advantage in your breeding program.
​

Since tests for specific mutations have become available in the last few years, the mantra of breeders is that their puppies are from "health tested" parents. To breed without first doing the available DNA tests is considered the height of irresponsibility in the fancy. But there is a right way to use DNA test results and also a wrong way. 

If you are reading this, you probably consider yourself to be a responsible breeder. Let's do some cyber dog breeding using some information from DNA testing and see how you do! All you need is a pen and paper along with your expertise as a responsible breeder.

Okay, we have a bitch we want to breed, and we have selected seven possible sires (S1 to S7). We have ranked these according to how well we think they fit what we're looking for in a sire in terms of conformation, temperament, health, longevity, working ability, or whatever criteria are important to you. We have given three green starts to the dogs we rank the highest and one green star to the dogs ranked lowest. However, we would be willing to breed to any of them. Assume we have done the relevant phenotype tests (e.g., hips, elbows, eyes, heart, etc.) and the results for all of those were acceptable.
​

The first thing we need to do is run the appropriate DNA tests for the recessive mutations known to be in the breed. In this case there is one, for the "red circle" trait. Our bitch is  carrier, as are two of the sires. I have identified the carriers with a red X on the chart, and the dogs that were clear have blue check marks.
​

Now we have our health-tested dogs from which we are going to select the perfect sire for our bitch. Using your rankings for overall quality, together with your information from health testing, rank the sires with 1 being your first choice and 7 your last choice. 

On your piece of paper, make columns for S1 through S7, and in the first row below that write your ranks for each of the potential sires.

​

Let's say you went to a little more trouble before making your breeding decision and you were able to get the  predicted average inbreeding coefficient of the litters that would be produced from our bitch and each of the seven sires. (You can get this information from the kinship coefficient, which is an index of the relatedness of a pair of dogs and can be determined either from pedigree or DNA information. The kinship coefficient of a pair of dogs is equal to the inbreeding coefficient of their potential offspring.)

Let's add the predicted litter COI to our table. With this information to consider as well, again rank the sires from 1 to 7 and add these data to the next row of your table. (Did this alter your first set of ranks?)

This is usually as much information as a breeder has about the dogs they want to breed. 

You did the testing for the "red circle" mutation in our potential parents and discovered that two of the sires were carriers. For both of them, there would be a 25% risk of producing puppies that are homozygous for the red circle mutation. You probably chose to eliminate those two sires from breeding consideration, which reduced the risk of producing affected pups in this litter from 25% to 0%.

You breed using your top-ranked sire and your bitch produces a beautiful litter of puppies. You advertise your pups as being from health-tested parents, keep a lovely pup for your breeding program, and send the rest off to pet homes. Sometime later one of the pet owners contacts you with the report that their puppy has a serious inherited blood disorder, and they ask how this could be if you claimed your pups were from health tested parents?

Indeed. How can this be???

We only know what we can know. We can know if a dog possesses a mutation if we have a test for it. We did our tests so we knew about the status of the red circle mutation in all of the potential parents.

But most dogs have many recessive mutations lurking in their genome for which there is no DNA test, and we have no way to know they are there if they are not expressed. Clearly, there was a mutation in your parents that you didn't know about. You feel horrible, offer to replace the puppy, and bury yourself in pedigrees trying to figure out how you could have produced a puppy with a genetic disease after being so careful.

If there is no DNA test for a recessive mutation, a disease can suddenly appear out of the blue that you have never seen in your lines if you happen to pair two carriers. This is why, if somebody claims that they only breed within their own kennel because they "know what's in their lines", you know that either they don't understand genetics, or they are being dishonest. It is not possible to know what recessive mutations are in your lines if you have no way to detect them.

So this leaves you in a pickle. All dogs have recessive mutations, and there are always going to be some that you don't know about. How do you avoid them? This is like tip-toeing across a mine field in which there are 3 mines that you can't see but are flagged, but you know that there are 28 more hiding out there somewhere. One wrong step and you will meet your maker. If these are mutations hiding in your dogs, this is a pretty hectic way to breed.

What if there were two mine-fields, but you can choose which one to cross. The first minefield has 3 mines you know about and can avoid and 28 that are hidden. The second mine field also has 3 that are flagged, but only 6 more that are hidden. If you're not ready to start your after-life, you would logically decide to cross the second mine field, where the risk of getting blown up is only 25% of the risk in the other one.

​Now we can see the problem. Our bitch has not one but three mutations that are also found in our "health tested" sires. Have a look at your last set of sire rankings. Now that you can see these other mutations, are you still happy with those ranks? If you would like to change your rankings, add another row to your table with these new assessments.

Great, you say. But in the real world, if we don't have tests for those unknown mutations, there is no way for us to avoid them, right?

Actually, there is. You just need one piece of information that in fact you already have - the predicted inbreeding coefficient of the litter.

Remember that inbreeding is defined as the probability of inheriting two copies of the same allele from an ancestor. By definition, then, inbreeding is homozygosity. If the inbreeding coefficient of a dog is 25%, then the probability of inheriting two copies of the same allele from an ancestor is 25%. If that allele happens to be a recessive mutation, then the risk of producing a genetic disorder caused by that recessive mutation is also 25%. Likewise, if the inbreeding coefficient is 10%, then the risk of a genetic disorder produced by a recessive mutation is also 10%.

Think about this. You did mutation testing to eliminate the 25% risk of producing a recessive genetic disorder from two carrier parents. But we just explained that the inbreeding coefficient tells us the risk of producing a genetic disorder from ANY recessive mutation, the ones we know about as well as the ones we don't.

So, we can be clever about this. We can control the risk of genetic disease from all recessive mutations by controlling the inbreeding coefficient of the litter. If we test for red circle and eliminated the carrier sires, it would be dumb to then breed to sire #1. We have paid to eliminate the 25% risk from the known mutation then done a breeding with a 31% risk of producing a disorder from some other mutation we don't know about.

Let's read that again.

We tested, eliminated the carrier sires so there is no chance of producing affected puppies from the known mutation, then we bred to a "clear" dog with a 31% risk of producing puppies with a disorder from some other mutation we don't know about.

You might be breeding "health tested" dogs, but you are still producing puppies with a significant risk of genetic diseases from recessive mutations.

Go back to what we know about the inbreeding coefficient. If our litter has a predicted COI of 10%, then the risk of ANY mutation causing a problem is only 10%, including the one we paid to test for. This is because parents with a kinship coefficient of 10% are less similar genetically than parents with a kinship coefficient of 25%. If the parents have fewer genes in common, there is a lower chance of a puppy getting two of the same mutation from both.

So it's simple. Reduce the risk of genetic disorders in your puppies by reducing the predicted inbreeding coefficient of the litter. 

If you've never thought about this before, this should be your light-bulb moment. All this "health testing" is a waste of time and money if we choose parents that produce a litter with an inbreeding level of 38%.

This is the WRONG WAY to use DNA testing. The breeder believes they are doing everything they can to produce puppies that will live long, healthy lives. But in fact they mark the known land mines with flags then assume the coast is clear without considering the risk produced by the many unseen explosives.

To produce healthy dogs, we need to get the risk of problems caused by ALL recessive mutations as low as possible, the ones we know about as well as the ones we don't. You can estimate this risk from the kinship coefficients of the potential parents. These numbers can be computed from a pedigree database, or (even better) they can be determined from the genotype data produced by DNA testing.

Don't waste your money on DNA tests then breed in a way that has a high risk of producing disease. Use those tests to eliminate all risk from mutations we can test for, then use kinship coefficients to make sure we have reduced as much as possible the risk from the unknown mutations hiding out in the genome of every dog.

To learn more about the genetics of dogs, check out
ICB's online courses

***************************************

Visit our Facebook Groups

ICB Institute of Canine Biology
...the latest canine news and research

ICB Breeding for the Future
...the science of animal breeding

There are some fun tools we can use to explore the effects on genetics of changes to some of the properties of a population. This is one that we use in my population genetics courses to help students understand how genetics can be affected by changes in various properties of a population such as size or addition of new individuals.

One of these is an online population simulator called Red Lynx. It's easy to use, and if you have access to the internet you can take it for a test drive.

Go the the Red Lynx website at ​https://cartwrig.ht/apps/redlynx/. 

On the landing page, click on the button to start the Red Lynx simulator.

This will take you to the simulator and you should see the screen below.

Red Lynx simulates the "behavior" of a single allele, "A1", in populations with various properties. There are two alleles at every locus, so actually A1 has a pair (lets call it A2). But we only need to simulate one allele, because if the frequency of allele A1 in a population is going up, the frequency of A2 must go down. If we model the behavior of A1, we know that A2 will be responding as a mirror image. So, we will run our simulations using only a single allele at a locus.

On the simulation page, you will see a graph, and below that a series of parameters that you can change using some simple sliders. We are only going to play with the first three: number of generations, population size, and initial allele frequency (green arrows). You can change these either by using the slider or just type the number you want in the box. You can ignore the rest of the sliders on the page.

​This line is the predicted behavior of the allele A1 in this population over time, assuming a 50/50 chance of inheriting the allele at each generation in a population under "ideal" conditions - no selection, no migration of A1 alleles in or out of the population, and no mutation. You can see that the initial frequency is 50%, and from there it change in frequency in the population with each generation.

Now, without clearing the graph, do a few more simulations. Each of these lines will be different because the role of chance in inheritance at each generation. If you do enough of these, you might see one of the lines go either all the way to 0% and flat-line (red arrow), or to 100% and stay there (green line). In the first case, all copies of the red allele have been lost from the population; for the green arrow, the alternative allele (A2) has been lost completely, leaving only A1, a situation we call "fixed" for that allele.
​

So what?

I hear you say "That's cool, but so what?".

Let's say the A1 allele was for something like the ability to smell a particular scent. If this isn't something you would be selecting for, this simulation suggests that there is a chance it could be lost from the population just by chance, a phenomenon called "genetic drift". Maybe this scent wasn't important to you, but if it was to the dog (e.g., a pheromone), it would have consequences that might affect some aspect of the physiology or behavior of the dog. Perhaps it is an allele that prevents fearfulness; losing it from the breed would definitely have consequences for the welfare of the dog. Or maybe it is an allele at a locus that is most beneficial when heterozygous - that is, the genotype A1A2 is more beneficial to the dog than either A1A1 or A2A2, something called "over-dominance". If one of the alleles is inadvertently lost from the population, all individuals will be homozygous for the other allele, and beneficial effects of heterozygosity will be unavailable.

So now let's do the fun stuff.

Let's make the number of generations in the simulation 100 so it will be more relevant to dog breeding. Change the population size to 50, and leave the allele frequency at 50%. Use the button to clear the graph, and run some simulations. 

You should find that the frequency of A1 goes to one extreme or the other (0% or 100%) in many of these simulations. 

Now run some simulations with different population sizes, perhaps 100 and 25. How does this affect the graph?

This simulator makes it easy to explore the effect of population size on the behavior of alleles in the population that you might not be paying any attention to. Clearly, the genetic stability of a population is sensitive to the number of individuals.

Why does this matter to your breed?

How big is your breed? What fraction of the puppies produced are used for breeding? Remember, only reproducing animals count when we're considering genetics. What are the effects of breeding contracts or spay/neuter policies on the population genetics of your breed? How stable is it genetically to the chance loss of alleles by genetic drift?

These changes in allele frequency occur in all populations. If a population is divided into closed subpopulations, the same phenomenon will occur in each of them. This will cause them to drift apart genetically over time. Breeders can take advantage of the genetic differences in these subpopulations to use them as outcrosses that can restore alleles that have been lost in another subpopulation. You can simulate this using the slider for "migration". If breeders can work together to monitor the genetic status of these subpopulations in a breed, they can reduce the loss of genetic diversity, slow the increase in inbreeding, and benefit from the hybrid vigor produced by an outcross.

​Is the population of your breed stable, or is it one of the many with registrations dropping off and fewer active breeders than there used to be? The simplest thing you can do to reduce the chance of losing important genetic diversity from your breed is to use more of the dogs produced in the breeding program. If the usual fraction is 20%, increase it to 40% or more. Don't send so many puppies off with contracts that prohibit breeding, or breed once before spaying or neutering. How much of a difference will it make? You can check it out using the Red Lynx simulator. Just remember that the simulator assumes a population with no selection, no migration, and no mutation (unless you stipulate conditions for these using the sliders). Certainly a purebred population will be under selection for the traits for type, but you can assume there is no selection for the "neutral" alleles that have no effect on type.

This simulator can also have a very practical use. Let's say your breed is discussing separating various colors in the breed into separate populations that would not interbreed. Or perhaps it is different sizes, or some feature of phenotype. If the groups are prevented from interbreeding, what will the consequences be for the genetic stability of the resulting smaller populations? There are many examples of breeds being split in the past - Norwich and Norfolk were separated based on ears, Toy and Standard Manchester were separated by size, the Belgian Shepherds are separated in the US based on color (but not in Europe), and perhaps you know of others. Splitting a breed WILL have consequences to the genetics of the separate populations, as you can see here effects of genetic drift on allele frequency. There will also be changes in other genetic properties of the populations such as the rate of inbreeding, and breeders should carefully consider these before deciding to make any changes in the breed.