Health discussions in dog breeding often sound more decisive than the evidence allows. Genetic information is probabilistic, contextual, and population-dependent, but it is repeatedly treated as categorical, individual, and decisive.
This post separates what we can know from what people keep trying to force genetics to deliver. It moves from what “genetic risk” means, to when tests really do predict outcomes, to when results are information rather than verdicts, to why “the science isn’t there yet” is both true and misused, to what tests systematically miss (including COI and population structure), and finally to what responsible risk management looks like once you accept that risk cannot be eliminated.
1. What “genetic risk” actually means
When people talk about “genetic risk” in dog breeding, they often treat it as if it were a property a dog either has or does not have. A dog is described as risky or safe, compromised or clear, as though risk were something you could locate, identify, and then remove. That way of thinking is intuitive, but it does not match how genetic risk actually works.
Genetic risk is probabilistic rather than binary: it describes the likelihood that certain genetic combinations will produce harmful outcomes, not a guarantee that they will. A genetic change can increase the chance of disease without ensuring it, just as the absence of a known change does not guarantee health. Risk, in this sense, is about distributions and tendencies, not certainties.
This is uncomfortable for people, because we experience dogs as individuals, not as probabilities. Each dog has one body and one life, and when that life goes badly wrong, statistical explanations feel beside the point, but genetics operates across generations, through patterns of inheritance that only become visible when you step back far enough to see how often certain combinations occur.
One reason genetic risk is so often misunderstood is that most genetic variation does nothing at all. Dogs differ from one another at millions of points across the genome, and the vast majority of those differences have no measurable effect on health, behaviour, or function. Even within the smaller portion of the genome that does influence how bodies develop and operate, many variations are neutral or context-dependent. Calling a genetic difference a “variant” simply means it is different from some reference sequence. It does not mean it causes disease.
This matters because the language of genetics quietly encourages people to treat difference as danger. Once a change is named, measured, or categorised, it starts to feel like a problem waiting to happen. In reality, carrying potentially harmful genetic variants is normal. Every dog does. Many of those variants are recessive, meaning they have little or no effect when a dog carries only one copy. Some have effects that are mild, late-onset, or highly dependent on other factors. Some may never cause observable disease at all.
This is why it is misleading to think of genetic risk as something that belongs to an individual dog in isolation. Risk emerges when particular combinations become likely enough to occur. That likelihood is shaped by how genes are distributed across a population and how often related genetic material is paired together. A variant can exist in a population for a long time without causing widespread harm, simply because it is rarely paired with itself or with other variants that amplify its effects.
The distinction between presence and expression is crucial. Many dogs carry genetic changes that would be harmful if inherited in two copies, yet never become ill because those copies are not paired. Problems arise when population structure makes those pairings more common. When related dogs are bred repeatedly, when ancestry becomes concentrated in a small number of lines, or when genetic diversity narrows over time, the probability that the same variants meet increases. What was once latent becomes expressed, not because anything new has appeared, but because existing material is being combined differently.
From this perspective, genetic risk is not a defect you find and remove. It is a background condition created by how genes are distributed and paired over time. It accumulates structurally, through patterns of reproduction, rather than appearing suddenly through a single bad decision. This is why genetic harm can build slowly, without any individual mating looking obviously irresponsible when viewed on its own.
Understanding genetic risk in this way does not minimise harm, and it does not excuse past practices that have caused damage. It does, however, change where responsibility actually sits. Risk is not carried by particular dogs, and it is not created by intent alone. It is shaped by population-level choices, repeated over generations, that determine which genetic material is common, which is rare, and which combinations become likely enough to matter.
Once risk is understood as something that lives at this level, it becomes harder to talk about it as if it could be cleanly identified, eliminated, or certified away. It also becomes clearer why managing genetic risk requires thinking beyond individual outcomes, even though those outcomes are the ones people live with most directly.
2. When genetic tests really do predict outcomes
Not all genetic tests make the same kind of claim. Some are designed to detect conditions where the inheritance pattern is well understood, the clinical outcome is clear, and the result reliably predicts what will happen to the dog. These tests are relatively rare, but they matter because they are often treated as the model for how all genetic testing works.
In these cases, a single gene is involved, and the way it is inherited is well established, most often through a recessive pattern. Dogs who inherit two copies of the relevant mutation almost always develop the disease, while dogs with one copy do not. The clinical outcome is serious and clearly detrimental to the dog’s quality of life, and the association between genotype and disease has been validated within the population where the test is being used.
Conditions such as Progressive Retinal Atrophy (PRA) fall into this category: when a dog inherits two copies of the relevant mutation, progressive blindness is a predictable outcome. The penetrance is high, meaning that most dogs with the at-risk genotype do in fact become affected, and the welfare impact is obvious.
In these situations, genetic testing does what people intuitively expect it to do. It identifies a real, avoidable harm, and it allows breeders to prevent producing affected puppies with a high degree of confidence. Importantly, this can usually be done without removing large numbers of dogs from the breeding population, because carriers can still be used responsibly as long as they are not paired together.
This distinction matters because it shows that decisive action based on genetic test results is not inherently over-reactive or irresponsible. When the inheritance pattern is clear, the disease expression is consistent, and the clinical impact is severe, firm breeding decisions are justified. In those cases, genetic testing functions as a genuinely actionable tool rather than as a proxy for broader genetic risk.
However, this category is much smaller than many people assume. The problem arises when tests that do not meet these criteria are treated as if they do. Once the inheritance pattern becomes less clear, the effect of a variant depends on genetic background, or the association has only been demonstrated in limited contexts, the logic that works for high-penetrance Mendelian conditions no longer applies.
Understanding that difference is essential, because it marks the boundary between tests that can support binary decisions and those that require interpretation at the population level. The danger is not genetic testing itself, but treating all test results as if they belonged in this first, narrow category.
3. When a test result is information, not a verdict
Once you move outside the small group of high-penetrance Mendelian conditions, genetic test results stop behaving like predictions and start behaving like signals. They still contain useful information, but that information no longer maps cleanly onto yes-or-no breeding decisions. Treating these results as if they belonged to the first category creates most of the confusion and much of the unintended harm in contemporary breeding practice.
Many commonly tested variants are associated with disease rather than determinative of it. Their effects vary with genetic background, environment, and population structure. Some influence risk slightly rather than decisively. Some were identified in a narrow context and do not behave the same way elsewhere. In these cases, the test result does not answer the question people usually want it to answer, which is whether a dog will become ill or whether a particular mating is “safe.”
To make this distinction clearer, it helps to separate genetic test results by the kind of claim they are making, rather than by the name of the condition or the emotional weight attached to it.
Different kinds of genetic test results require different kinds of reasoning
| Type of genetic finding | What the result actually tells you | What it does not justify on its own |
|---|---|---|
| Single-gene variants with variable expression (e.g. Adult-Onset Neuropathy, PFK deficiency) | A specific variant is associated with disease, but severity, age of onset, or whether disease appears at all varies depending on genetic background and environment. | Treating the result as a diagnosis, assuming all dogs with the genotype will become ill, or excluding all carriers without considering frequency, severity, alternatives, and population impact. |
| Risk-associated variants without a carrier model (e.g. IVDD-associated variants) | The variant increases risk rather than determining outcome. Having one copy already changes probability, and there is no meaningful “silent carrier” state. | Applying recessive logic (carrier × clear), assuming absence equals safety, or expecting breeding to a clear dog to “hide” or neutralise the risk. |
| Markers outside their validated context (e.g. copper toxicosis variants outside breeds where disease is established) | A statistical association exists in some populations. The marker may track disease risk in that context. | Treating the marker as causal, applying binary exclusion across all breeds, or assuming the same result means the same thing in every population. |
What unites these categories is not that the tests are useless, but that they do not support binary decisions in the way high-penetrance Mendelian tests do. The result is information that needs interpretation, not a verdict that can be acted on in isolation.
This is where many breeding discussions quietly go wrong. A test identifies something, and that something is treated as if it were equivalent to a known, deterministic disease. The language of “carrier,” “clear,” or “affected” is applied even when the underlying biology does not support those distinctions. Once that happens, exclusion starts to feel obligatory rather than discretionary, even when the benefit is uncertain and the cost to the population is real.
Why the usual “carrier” logic does not fit IVDD/CDDY
IVDD is a useful example of why genetic test results often need interpretation rather than automatic rules, because the variant involved is widespread, its effects are real, and yet the most severe outcomes remain hard to predict.
In several breeds, including spaniels, a variant associated with IVDD (often referred to as CDDY) is extremely common. In English Cocker Spaniels and Beagles it is present in the vast majority of the population, yet only a small proportion of dogs ever experience the most severe outcomes such as acute paralysis. Nobody currently knows why – there is no evidence that carrying this variant reliably predicts which dogs will become clinically affected, or how severely.
In a classic recessive disease, a carrier has one copy of a mutation that has little or no effect on its own, and disease appears primarily when a dog inherits two copies. With IVDD/CDDY there is no silent carrier state in that sense, because one copy already changes risk even though it does not reliably predict clinical disease. A dog either has the variant or it does not, and the outcome is modified by other genes, body structure, and environment in ways we cannot currently read off a DNA result.
So when you breed a one-copy dog to a clear dog, you reduce the genetic load in the next generation, but you do not create “carriers” in the usual breeding sense where the gene is neutral in one copy. Roughly half the puppies will inherit one copy and half will inherit none, and the one-copy puppies are not genetically equivalent to classic recessive carriers, because the variant is not silent. This is why “carrier” becomes misleading language here. It smuggles in a model of “silent in one copy, harmful in two” that does not match how this variant behaves.
Questions that matter more than the label on the result include:
- Is the inheritance pattern clear, or does one copy already change risk?
- Do most dogs with this genotype actually become ill, or do many live normal lives?
- Has this association been shown across multiple populations, or only in a narrow context?
- What is the typical clinical impact on the dog’s quality of life?
- What happens to genetic diversity if every dog with this result is removed from breeding?
These questions shift attention away from individual dogs and back toward population consequences, where genetic risk actually accumulates. In small or closed populations, removing large numbers of dogs on the basis of uncertain signals can increase relatedness, amplify untested risks, and create the illusion that risk has been “managed” when it has only been displaced.
The problem here is not DNA testing. The problem is treating all genetic test results as if they belonged to the first, narrow category where inheritance is clear, penetrance is high, and outcomes are predictable. When uncertain or context-dependent results are handled as if they were definitive, breeders are pushed toward unnecessary exclusion, accelerated loss of genetic diversity, and a false sense of control.
Used appropriately, genetic tests reduce specific, known risks. Used indiscriminately, they can make broader genetic risk harder to see and harder to manage. Understanding which kind of information a test result actually provides is therefore not a technical nicety, but a practical and ethical necessity.
4. Uncertainty and the misuse of “the science isn’t there yet”
At this point, it should be clear why discussions of genetic risk so often end with frustration. The tools are real, the data are growing, and yet definitive answers remain elusive. In that context, the phrase “the science isn’t there yet” is often heard as a deflection, an admission of ignorance, or an excuse for inaction. It is sometimes used that way. But in genetics, uncertainty is not a temporary failure state. It is a structural feature of the problem.
Why uncertainty persists
Much of what matters for genetic risk cannot be observed directly. Many harmful variants are rare, poorly characterised, or entirely unknown. Others have effects that are small on their own and only become relevant when combined with similar changes elsewhere in the genome. Some interact with environment, age, or physiology in ways that are difficult to predict in advance. None of this disappears as sequencing becomes cheaper or faster, because the limitation is not technological access to DNA but interpretive access to outcomes.
What limits genetic prediction is not the ability to read DNA, but the ability to connect genetic information to lived outcomes. Knowing which dogs become ill, when they do so, how severely, and under what conditions requires large, long-term datasets that track real animals across their lives. Those data are expensive, slow to collect, and unevenly distributed across breeds. Until they exist at scale, many genetic associations will remain probabilistic rather than predictive, even when the underlying biology is real.
This is why uncertainty persists even when individual test results are technically accurate. A result can correctly identify a variant and still leave open the question of what that variant will mean in a particular dog, in a particular population, over time. Treating that uncertainty as a failure of science misunderstands what science is actually doing in this space. Genetics can narrow possibilities, identify constraints, and rule out some harms with confidence. What it cannot do is collapse a complex, multi-generational process into a single definitive answer.
A note on epigenetics and early-life environment
Epigenetics refers to changes in gene expression that do not involve changes to the DNA sequence itself. Genes are not only “present or absent”. They are also regulated: switched on and off in different tissues, at different stages of development, and at different intensities, and that regulation is shaped by both genetic background and lived conditions.
Early-life environment matters here because development is the period when regulatory systems are being built and calibrated. Stress, chronic insecurity, poor maternal care, crowding, illness, inconsistent handling, and other forms of early instability can alter how puppies regulate arousal, inflammation, and stress responses later in life, even when their DNA sequence is unchanged. This does not mean that genetics stops mattering. It means that genetic predisposition and early environment can interact, so two dogs with similar genetic risk can diverge in real outcomes because the systems that buffer and regulate that risk were built under different conditions.
Some epigenetic patterns can persist for long periods and, in limited cases, may influence the next generation, but they do not offer a reliable way to bypass genetic constraints in breeding. For the purposes of decision-making, the main point is simpler: outcomes depend on regulation as well as sequence, and regulation depends on early development as well as ancestry, which is one more reason test results and tidy categories cannot substitute for judgement about risk.
How uncertainty gets misused
The more serious problem arises when uncertainty is misread in either direction. On one side, it is used to justify paralysis: if outcomes cannot be predicted with certainty, then no judgement can be made and no responsibility can be assigned. On the other, it is ignored in favour of premature confidence: results are treated as verdicts, numbers as guarantees, and tools as solutions. Both responses distort decision-making, not because they disagree, but because they each try to escape the same discomfort.
Responsible breeding does not require certainty. It requires recognising which uncertainties matter, which risks are being actively managed, and which are being passed forward. Decisions still have consequences even when knowledge is incomplete. In practice, uncertainty is often where population-level harm accumulates, because repeated small assumptions can compound into large structural effects over time.
Understanding uncertainty in this way reframes what it means to act responsibly. The question is not whether science has delivered perfect answers, but whether decisions are being made in a way that respects what is known, acknowledges what is not, and remains attentive to second-order effects. Pretending certainty where it does not exist creates false reassurance. Pretending uncertainty absolves responsibility creates drift.
This is the uncomfortable space in which breeding decisions actually sit: between incomplete knowledge and unavoidable consequence. Genetics does not tell us what to value, but it does constrain what is possible and what is likely. Once those constraints are understood, disagreement about breeding is no longer primarily about facts. It is about how responsibility should be understood when certainty is unavailable but action is unavoidable.
Intervertebral disc disease (IVDD / CDDY) as a case study
IVDD is a useful example of why genetic test results often need interpretation rather than automatic rules, because the information involved is real, widespread, and yet does not behave like a deterministic disease prediction.
IVDD is not like a high-penetrance recessive condition where two copies of a mutation reliably predict illness and the concept of a “carrier” maps cleanly onto practice. In several breeds, including spaniels, variants associated with IVDD are extremely common, in some cases present in the vast majority of the population. Yet only a small proportion of dogs ever experience the most severe outcomes, such as acute paralysis, and there is currently no reliable way to predict which dogs those will be.
This means there is no realistic way to “remove” IVDD-associated variants from some populations without excluding most dogs, and exclusion based on test results alone does not reliably target the dogs who would have become severely affected anyway. That does not make the information useless. It makes it actionable in a different way.
What IVDD-related DNA tests actually measure: Most current DNA tests related to IVDD are not testing “IVDD” itself. They are testing genetic markers associated with increased risk, most commonly FGF4 retrogene insertions linked to chondrodystrophy (short-limbed body type), and occasionally polygenic risk scores built on limited datasets. These markers increase statistical risk, sometimes substantially, but they are neither necessary nor sufficient for IVDD to occur.
IVDD is not a single disease with a single genetic cause. It is a clinical outcome that can arise through multiple pathways, including differences in cartilage biology, spine biomechanics, growth rate, body mass distribution, activity patterns, and age-related degeneration. This is why dogs can develop IVDD without carrying the flagged variants, and why many dogs who do carry those variants never develop clinical disease. From a genetic perspective, this is exactly what is expected for a complex trait.
In some breeds, overall genetic background also matters. Where many risk-shaping variants are already fixed or near-fixed, a specific test result may no longer discriminate well between “higher-risk” and “lower-risk” individuals, because most dogs are genetically similar in the relevant respects. Once a risk factor is widespread, it can stop being informative at the individual level even though it still matters at the population level.
What this means for breeding decisions: Because there is no silent “carrier” state in the usual sense, the logic that applies to recessive diseases does not fit IVDD. Having one copy already changes risk, even though it does not reliably predict disease, and breeding a one-copy dog to a clear dog reduces genetic load in the next generation without creating unaffected carriers in the classical sense.
In practice, this means IVDD needs to be treated as one factor among many rather than as a single veto. For individual dogs, test results support management rather than false reassurance: attention to body condition, core strength, conditioning, and monitoring over time. For populations, the information can still be used to slow further spread in breeds where the variants are not yet effectively fixed, and to avoid compounding genetic load by doubling up unnecessarily.
In my own breeding decisions, Grace has two copies of the IVDD-associated variant. That mattered, but excluding her would have meant discarding a dog that was otherwise exceptionally sound and genetically valuable. Instead, testing allowed me to reduce genetic load deliberately by choosing a mate who tested clear, so the offspring could inherit one copy rather than two.
This did not eliminate risk completely, but it did allow me to reduce it to a level that was acceptable to my conscience, and to be explicit about the trade-off. I disclosed the result and my reasoning to potential puppy families, and I used the information to guide how Grace herself was managed, including regular physiotherapy checks and monitoring for early signs of spinal wear. At five years old, spinal imaging showed no significant degeneration.
IVDD illustrates a broader point that runs through this series: genetic tests do not divide dogs cleanly into “safe” and “unsafe.” Some results predict a welfare outcome with high confidence. Others describe risk factors that are widespread, variably expressed, and impossible to eliminate without causing greater harm elsewhere. Treating those categories as if they were the same leads to bad decisions.
BONUS: One possible reason severe IVDD outcomes do not track genotype cleanly
What follows is hypothesis rather than settled evidence, but it helps illustrate why genotype does not translate directly into lived outcome for IVDD.
Structure and mechanics likely matter: The most severe forms of IVDD are seen disproportionately in breeds with extreme body proportions, particularly very long backs relative to leg length, as in Dachshunds. A longer spine creates greater mechanical load on intervertebral discs, because increased length increases leverage and cumulative shear forces over time. A more moderate, square structure likely places less mechanical stress—relatively speaking—on each disc segment even when the same variant is present.
Another plausible factor could be the length and strength of the loin: the lumbar region of the spine is largely stabilised by muscle rather than bone, so differences in muscle mass, conditioning, and overall balance are likely to influence how much strain discs experience. Again, this is not something a DNA test can currently quantify.
The point of including this is not to claim a complete explanation, but to show why it is biologically naïve to treat IVDD as a yes-or-no genetic outcome. The same variant can plausibly lead to very different lived outcomes depending on structure, conditioning, and use, which is one reason why removing dogs purely on the basis of this mutation can be both ineffective and damaging to the population.
5. What genetic tests systematically miss
Genetic tests are designed to detect specific, identifiable variants. They answer narrow questions about whether known genetic changes are present. What they cannot do is describe the background risk created when genetic diversity has been reduced across many generations, because that risk is not one variant that can be named and removed.
Why inbreeding can increase immune-mediated disease
The immune system isn’t an on–off switch. Immune cells have to recognise what belongs to the body and what does not, mount a response that is strong enough to clear a real threat, and then stop that response again once the job is done. That sequence depends on many interacting genes and signalling pathways, especially the ones that control timing, intensity, and tolerance, rather than a single “disease gene”.
When genetic diversity is higher, dogs are more likely to carry different versions of key immune-related genes. That variation matters because it broadens the range of immune responses and gives regulation more redundancy. When diversity is lost across generations, more of those immune-related systems become genetically uniform, and immune control becomes less robust. Responses that should be limited in duration or intensity are more likely to overshoot, persist, or activate inappropriately.
The outcome is usually not one neat autoimmune disease with a single causal mutation. What tends to increase instead is general susceptibility: more immune-mediated and inflammatory problems overall, with variable expression between dogs, because what is shifting is the stability of regulation rather than one broken switch.
A dog can be clear on every test you can buy and still be high risk in the only sense that matters here: too much of its genome is identical on both sides, so problems that depend on loss of diversity—unknown recessives and immune dysregulation among them—become more likely.
What panels cannot see
Even leaving immune disease aside, genetic tests miss large categories of risk by design:
- Unknown recessives: every dog carries recessive mutations that have not been identified. They are invisible until two copies happen to meet.
- Polygenic effects: many outcomes come from the combined influence of many small genetic differences, not one mutation.
- Regulatory variation: some changes affect when and how strongly genes are expressed rather than changing the gene itself.
- Background interactions: the effect of one variant can depend on the rest of the genome and on environment.
- Latent risk revealed by inbreeding: when the genome becomes more homozygous, more of the hidden variation becomes expressed.
This is why “clear” results can be true and still incomplete: they can describe what was tested without describing the risk created by how the genome has been structured over generations.
COI and the kind of risk DNA tests cannot certify away
The coefficient of inbreeding (COI) is a probability statement about homozygosity. For recessive disorders, the risk of producing an affected puppy depends on whether a puppy inherits two copies of the same mutation, and COI estimates that risk across the whole genome, including for recessive mutations that have not been identified and cannot be tested for.
This is also why COI can be misread when it is treated as a property of one mating rather than as the output of a multi-generation history. A “COI of zero” often reflects limited pedigree depth rather than true genetic independence, because it can mean only that the system cannot see shared ancestors far enough back to calculate overlap. A low COI in one mating also cannot restore diversity lost through repeated inbreeding in earlier generations; it only avoids adding more shared ancestry at that step.
You can see the practical consequence if you imagine pairing a dog whose own COI is moderate to high, and whose pedigree contains many generations of heavier inbreeding, to a dog that is genuinely unrelated in recorded terms. The mating COI can come out very low or even zero, but the puppies still inherit half their genome from the more inbred parent, including long stretches of identical-by-descent DNA created by that parent’s history. The number is therefore describing the absence of new overlap between the parents in the available records, not the absence of accumulated homozygosity in the genetic material being passed on.
For that reason, COI is most useful when seen as a population pattern over time rather than as a single reassuring number attached to a pairing, because what matters for long-term risk is not only whether a particular mating adds overlap, but whether the population as a whole is continuing to concentrate ancestry in ways that increase the probability of pairing hidden mutations.
This is the core point for anyone relying on tick-box testing: panel testing can reduce the risk of specific known disorders, but it does not reduce the underlying probability of pairing hidden mutations across the genome if ancestry continues to be concentrated.
Pedigree COI vs. genetic COI
Up to this point, I have been talking about pedigree-calculated COI, which is an estimate derived from recorded ancestors. That estimate is useful, but it inherits the limits of the record. If the pedigree is shallow, incomplete, or simply cannot see far enough back, it will systematically underestimate overlap, and a “COI of zero” can mean only that the database cannot detect shared ancestors within the depth available.
Some testing companies now offer a genetic COI, where the estimate is derived from the dog’s DNA rather than from the pedigree. This kind of estimate does not depend on whether ancestors were recorded, because it looks directly at how much of the genome is homozygous across many locations, and it can therefore detect accumulated concentration that a pedigree calculation misses. When a pedigree COI comes out low because the paper trail ends, a genetic COI can still come out high because the dog’s genome contains long stretches that are identical on both sides.
This does not make pedigree COI useless, and it does not make genetic COI a definitive verdict. It means they answer slightly different questions. Pedigree COI estimates the probability of pairing identical material based on known ancestry. Genetic COI describes how much identical material is already present in the dog’s genome, regardless of how it got there. Used together, they make it harder to hide behind reassuring numbers, because one measures what the paperwork can see, and the other measures what the dog actually carries.
When people present “clear DNA panels” as proof of safety while ignoring genome-wide homozygosity, they are treating named variants as if they exhaust genetic health, even though both pedigree COI and genetic COI are pointing at the same background risk: how often hidden recessives and regulatory fragility are being doubled up.
Why you need pedigree COI and DNA-based COI for complete information
Pedigree COI is an expected probability computed from the relationship structure in the pedigree. DNA-based measures of inbreeding estimate how much homozygosity a particular dog actually has across its genome. Those two measures are related because they are both tracking shared ancestry, but they are not identical because inheritance does not distribute genetic material as a smooth average in each individual.
Even when two parents have a pedigree relationship that implies a particular expected COI, a puppy does not inherit a perfectly average fraction of each ancestor’s DNA in neat proportions. The puppy inherits segments of DNA, and the exact segments it receives depend on recombination, which means that siblings from the same parents can inherit different patterns of homozygous and heterozygous regions even though their pedigree COI is the same.
This is why two littermates can share the same pedigree COI while differing in the amount and distribution of long homozygous stretches in their genomes. It is also why a dog can have a high pedigree COI because its parents are closely related on paper, while its genome-wide homozygosity measured from DNA comes out somewhat lower than the pedigree expectation, and why the reverse can happen when a pedigree is shallow or incomplete, because ancestry outside the recorded depth can still contribute substantial shared genetic material.
For practical purposes, pedigree COI is most useful for describing the expected effect of a particular mating given the recorded history, while DNA-based COI is most useful for describing what a dog actually carries regardless of what the records can see. When the two disagree, the disagreement is not a paradox. It is information about the difference between an expectation derived from a pedigree and an outcome realised in an individual genome.
- If you are choosing a mating: What you need most is the relatedness between the two parents, because that is what determines whether the puppies become autozygous (inherit identical-by-descent material from both sides). Pedigree COI is one proxy for that, and DNA-based relatedness is another proxy for that, but the key object is the pairing, not the individual dog. In practice, if you have deep pedigrees you trust, pedigree-based calculations are very good for comparing options, because they tell you what overlap you are introducing by choosing A×B instead of A×C. If pedigrees are shallow or you have reasons to distrust them, DNA-based relatedness measures become more informative, because they are not dependent on what the records can see.
- If you are trying to understand a particular dog’s genetic background: DNA-based measures often become more informative, because they describe what the dog actually carries. Pedigree COI tells you what you would expect on average given recorded ancestry, but the genome is what was actually inherited, and individual dogs can land above or below expectation.
What “COI 0%” does and does not mean
A pedigree COI of 0% for a planned litter usually means only that, within the depth of the pedigree used for the calculation, the two parents do not share recorded ancestors. If the parents are genuinely unrelated, then the offspring’s inbreeding coefficient in the strict sense is effectively zero, because inbreeding (autozygosity) is about the same ancestral DNA segment arriving through both parents. When there is no shared ancestry between the parents, the puppy does not inherit identical-by-descent material from both sides, so you do not expect the long runs of homozygosity that come from close parental relatedness. That does not mean the puppy is “genetically diverse” in a broader sense, because the puppy still inherits half its genome from each parent.
If one parent comes from a line where ancestry has been repeatedly concentrated, that parent may carry large stretches of genetic material that are already narrow and repetitive, and the puppy inherits half of that genetic background regardless of whether the mating itself introduces new overlap. A low mating COI therefore tells you that the pairing is not adding further shared ancestry in that generation, but it does not describe how homozygous either parent already is, and it does not restore genetic variation that has been lost in earlier generations.
This is also why DNA-based measures can behave differently depending on what they are measuring. If a company’s “genetic COI” is primarily estimating autozygosity (for example by measuring long homozygous runs), it may not spike in a puppy from an unrelated pairing, because the mating itself did not create autozygosity. If the measure is closer to overall heterozygosity or genome-wide diversity, you may still see that the puppy is less diverse than it would have been if both parents came from broad genetic backgrounds, even though the pedigree mating COI is 0%.
In practice, the cleanest way to see the accumulated narrowing is to DNA-test the more inbred parent, because that parent’s genome carries the signature directly. Testing only the puppy can understate it, because the puppy carries only one copy of that parent’s segments.
As a one-generation intervention, pairing to an unrelated dog is one of the most effective ways to avoid creating new inbreeding in the offspring. As a population strategy, it only helps if it is repeated and supported by restraint, because the benefit disappears if the descendants are then repeatedly bred back into the same concentrated lines.
This is why “COI 0%” in an puppy announcement can be technically correct and still incomplete. It answers a narrow question about whether the parents overlap in the visible pedigree. Many readers assume it answers a broader question about overall genetic risk and genetic diversity, even though the number does not contain that information.
Questions to ask a breeder about COI and genetic concentration
COI is one of the few tools that points at a kind of risk that DNA panels cannot certify away, because it tells you about shared ancestry and the probability of doubling up genetic material across the genome rather than about a short list of named mutations. You do not need to be an expert to ask useful questions here, and you do not need a perfect answer to every question, because the point is to find out whether the breeder understands what COI can and cannot tell you and whether they have thought about relatedness as something that accumulates across generations.
Start with these three questions, because they are hard to dodge and easy to answer if someone actually knows their own breeding choices.
- How was the COI calculated, and how far back does the calculation go? If the breeder gives you a number, the follow-up that keeps it concrete is: “How many generations is that based on, and which database or tool did you use?”
- What are the COIs of the sire and dam themselves, not just the planned litter? This matters because a litter number describes the relationship between the two parents, while the parents’ own COIs tell you what level of concentration each dog is already carrying forward.
- Do the same dogs show up repeatedly behind either parent? If you want a plain-language version that does not require you to read pedigrees, you can ask: “Are there a few names that appear over and over in the last few generations, or is there a broad mix of different lines?”
You can also ask to see the pedigrees of each parent separately, or ask if there is a database for the breed – example pedigree here from Crufts Agility Champion Blink (note that the colours are for dogs that repeat!) Or you can practice looking for the answers to these questions from his pedigree!
FINAL NOTE: If a breeder cannot at least some of these questions, or feels uncomfortable about you asking them, choose another breeeder.
6. What responsible risk management looks like
If the earlier sections are right, then “responsible breeding” cannot mean eliminating risk, because too much of the risk that matters is either unknown, distributed across many genes, or expressed only when population structure makes certain combinations likely. What changes, once you take that seriously, is that responsibility shifts away from trying to certify safety and toward managing probability under constraints, which means you make decisions that reduce the chances of harm without pretending you can guarantee outcomes.
Elimination thinking treats risk as a contaminant: you find the bad thing, remove the dogs that carry it, and assume the remaining population is safer. That approach works in a narrow set of cases, especially when a variant is clearly causal, penetrance is high, and the welfare outcome is severe and predictable. Outside those cases, elimination thinking tends to create a second problem while it solves the first, because every time you exclude dogs you shrink the breeding population, and shrinking the breeding population increases relatedness, which increases the probability of pairing up the recessives and background vulnerabilities you do not have tests for.
Management thinking keeps the same moral aim (avoid preventable harm) but it works with the actual structure of the problem. Instead of asking “how do I remove every risk factor,” it asks “how do I avoid producing affected puppies when I can, without narrowing the genome so much that I increase background risk everywhere else.” That is why the practical goal is usually not “never use carriers,” but “do not produce affected puppies while keeping enough diversity in play that the population can still absorb what we do not know.”
Most breeding decisions feel local, because you are choosing a mate for a specific dog and you can justify that choice in terms of temperament, performance, type, or health information you can see. The population-level damage rarely comes from one obviously reckless decision. It comes from many individually defensible decisions that all point in the same direction, because the same admired dogs are used repeatedly, the same lines dominate reproduction, and selection pressure keeps narrowing the pool in the name of predictability.
This is why restraint matters more than cleverness: you can run every test, pick mates that look sensible, and still contribute to long-term decline if you keep feeding the same genetic bottlenecks. The simplest high-impact behaviours are often the least glamorous ones: limiting the use of any one dog, being willing to breed from dogs that are healthy and functional even when they are not fashionable, and spreading reproduction across more families so that no single line becomes structurally unavoidable.
Restraint also shows up in how people talk: if the story you tell puppy buyers is “this pairing is safe,” you set yourself up to either mislead or retreat into vagueness later. If the story you tell is “this pairing manages the risks we can see, and it does not pretend to solve the risks we cannot,” you stay honest about what breeding can and cannot deliver.
A single litter is not where most genetic damage becomes visible – the damage shows up when today’s successful dogs become tomorrow’s bottlenecks, when high representation propagates for several generations, and when the costs of increasing homozygosity appear as diffuse fragility rather than as one named disease that can be blamed on a single decision.
Thinking in population time horizons changes what you pay attention to: you stop asking only whether a mating “looks good” and you start asking what it contributes to the distribution of ancestry over time, because the question is not just what this litter will be like, but what this litter will make more likely three generations from now if its descendants are used widely. That is also where modest choices matter. A breeder who prevents one popular sire from dominating a local population may never be celebrated for it, but the effect compounds quietly in the only place that matters: in the genetic options the population still has later.Preview of ethical responsibility
Once you accept that risk cannot be eliminated, responsibility stops being a claim you make about being careful and becomes a pattern you can be held to over time. It becomes about what you choose to compound and what you choose to limit, about how much uncertainty you pass on without disclosure, and about whether you treat puppy buyers as consumers to reassure or as participants in a system where informed demand is one of the few forces that can change norms.
That is where the argument moves out of genetics and into ethics: genetics can tell you what tends to happen when you concentrate ancestry and when you narrow diversity. It cannot tell you what level of risk is acceptable, what trade-offs are justified, or what anyone owes to the dogs they produce and the people who will live with them. The next post takes that step directly, because once the mechanisms are understood, the remaining disagreements are not mainly about facts but about responsibility under uncertainty.
Next post: The ethics of dog breeding (coming tomorrow)
Please note: I have not made detailed in-line references in this article simply because of the time it takes – my blog does not usually have a huge volume of readers, so I balance the time and effort it takes to write these longer educational pieces with the number of people who will actually benefit from the additional time of being academically precise with referencing. If there is a particular claim that you’d like me to point you to, please contact me and I’d be happy to do so.
For further reading, I recommend this open access source that has been hugely helpful to me: Institute of Canine Biology (Blog) They also offer free and paid courses. Much of the detail in this blog post is thanks to what I have originally learned from this site.
BusyDoggie 