Every gain in knowledge in animal genetics rests on careful collaboration: between scientists, breeding organisations and – above all – the individual breeders. The laboratory contributes the methods and the computing power. Whether a project succeeds or fails, however, is decided long before that: in how the animals are described, and in what material is available from them.
This page sets out how a robust study comes about – from precise phenotyping through the form of organisation and the sampling strategy to the association study itself.
An association study compares animals that show a trait with animals that do not. Everything the statistics can achieve later is decided by how reliably those two groups are formed. An animal placed in the wrong group does not simply add noise – it removes power. The phenotype, not the genotype, is the part of a study that breeders and breeding organisations control.
Before the first sample is drawn, it must be written down what counts as affected: which finding, established by which method, at what age. A diagnosis that rests on an impression cannot be reproduced by a second examiner – and therefore cannot be analysed.
Controls must be examined by the same protocol as the cases. For traits that appear late, the control must have reached an age at which the trait would have become visible. A young, unexamined animal is not a control – it is an unknown.
Where a graded assessment already exists – the FCI grades A to E for hip dysplasia, for example – it should be carried into the analysis as a grade. Collapsing it into affected/unaffected throws away information that has already been paid for.
Standardised assessment also makes data comparable across borders. Wang et al. (2021) combined 1,357 German Shepherd dogs from the United Kingdom, Sweden and Finland into one hip dysplasia study although two different scoring schemes were in use. The same work provides the empirical counterpart to the argument above: using the graded FCI classification produced more significant and suggestive SNPs than the binary case/control split on the very same animals.
One breeder's observation is the beginning of every study – and at the same time the most fleeting part of it. For it to become usable data, something has to bring findings and animals together over years, secure the identity of the animals and govern who may see what. That is exactly what DNA programmes and the Animal Trust Center (ATC) are built for.
They see what nobody else sees: the puppy that develops differently, the bitch that fails to conceive a second time, the foal with the conspicuous gait. What matters is to report the unremarkable animals as well. Passing on only the problem cases delivers a study without controls – and therefore no study at all.
The organisation is the only place with a complete view of a population: studbook, pedigrees, litter registrations and mandatory examinations all come together there. It determines which examination is compulsory for which animal – and in doing so creates a uniform phenotype in the first place.
A breeding organisation stores its own DNA programme in the ATC: which tests apply to which animals, and which findings are shared within the organisation. Many individual orders thereby become one continuous, uniform data collection.
In the ATC, animals and findings are held in a digital online animal record, every finding carrying a secure electronic signature. The owner decides what stays private and what is shared with the organisation – a study is therefore never built behind the breeders' backs, but with their explicit release. See the account types
Whether a question can be answered later depends on whether the material for it exists. There are two fundamentally different routes to sampling. Both are legitimate; they do not differ in quality but in how many questions they keep open.
Only those animals are sampled that have been identified as affected by a particular trait, together with their matching controls. The collection begins with the question and ends with it.
Advantage
Disadvantage
A sample is taken and stored from every breeding animal – whether or not a project is currently running. The archive grows with the population and is not committed to any one trait.
Advantage
Disadvantage
An association study compares marker genotypes between cases and controls across the whole genome and asks where the two groups differ systematically. It becomes possible at the moment when enough informative samples have accumulated. How many that is depends on the trait, not on the budget – and the difference between a monogenic and a complex trait is not a matter of degree but of orders of magnitude.
Karlsson et al. (2007) mapped mendelian traits within a breed using an array of roughly 27,000 SNPs and about 20 dogs, reaching regions below 1 Mb. Fine mapping across two breeds then narrowed the S locus to approximately 100 kb within MITF. The strategy that follows from this is: map within the breed, fine-map across breeds.
Hayward et al. (2016) genotyped more than 4,200 dogs at 180,000 markers and found significant loci for hip dysplasia, elbow dysplasia, idiopathic epilepsy, lymphoma, mast cell tumour and granulomatous colitis. Anyone expecting an answer on hip dysplasia from 30 samples is working three orders of magnitude below what the trait requires.
Sutter et al. (2004) measured linkage disequilibrium within breeds reaching from 400–700 kb in Golden and Labrador Retrievers to 3–3.2 Mb in Bernese Mountain Dogs and Pekingese – up to a hundred times further than in humans, so correspondingly fewer markers are needed. The range is breed-specific, however: between those breeds it differs roughly fivefold.
SNP arrays are sufficient for finding associations within a breed. Across several breeds, and in mixed-breed animals, linkage disequilibrium decays between the markers and the array runs out of power. Two developments have changed this.
Hayward et al. (2019) imputed array genotypes against 365 whole-genome-sequenced dogs and wolves and raised the number of usable markers from 173,000 to 24.0 million, at an overall accuracy of 88.4 per cent (purebred dogs 89.7 per cent, village dogs 84.2 per cent). The gain was not theoretical: a new significant locus for cranial cruciate ligament disease appeared on CFA20.
The Dog10K consortium sequenced 1,987 canids at twentyfold coverage and described more than 48 million variants (Meadows et al. 2023). In the horse the resources are markedly weaker: Reich et al. (2022) reached an imputation accuracy of only 0.66 from a 60K array against 317 horse genomes, but around 0.90 from a 670K array. For horses the shortcut via a sparse array does not carry.
Association studies are carried out together with university research groups, and the result belongs in the scientific literature. The following selection shows papers in which Generatio scientists appear as authors or co-authors – chosen for their relevance to the methods described on this page.
Founder and Managing Director, Generatio GmbH – Center for Animal Genetics
Earlier work at Hannover Medical School dealt with Y-chromosomal gene families in humans and cattle and with DNA fingerprinting for parentage verification – among others Nürnberg P et al. (1998), Paternity assessment in rhesus macaques (Macaca mulatta): multilocus DNA fingerprinting. American Journal of Primatology 44(1):1–18.
Bioinformatics and quantitative genetics, Generatio GmbH – Center for Animal Genetics
Before joining Generatio, Arnav Mehrotra worked in the animal genomics group at ETH Zurich and at the Indian Veterinary Research Institute. His focus – association studies, signatures of selection, sequence-based fine mapping – is exactly the set of methods described on this page.
A selection, not a complete bibliography. Search performed in Europe PMC and Crossref, September 2026. Each paper remains permanently retrievable through the DOI given.