BRAHMAN NEWS SEPTEMBER 2011 Issue #172
By Professor Mike Goddard From Beef Bulletin Quarter 1 - 2011 BEEF CRC PUBLICATION
"When we started Beef CRC 3, the prevailing wisdom at the time was that a handful of genes would explain the variation in the traits of interest to producers, such as meat quality, food conversion efficiency and fertility.
However, when real genomic data became available, it was apparent that the number of genes influencing specific traits was vastly greater than was anticipated.
In fact, there can be thousands of genes scattered across the genome that may affect traits of interest in livestock and for most other complex traits.
This dictated that we needed to change our initial strategy of gene discovery – and also to develop a different model for how we were going to deliver for industry. These outcomes were relevant to genomic researchers across species, including humans, and not just related to the beef industry.
Instead of focusing on finding a small number of markers, the focus switched to genomic selection, which involves using markers scattered randomly over the entire genome. Genomic selection allows us to identify genes associated with a desired trait, for example marbling, no matter where those genes are located, even if there are thousands of them scattered over the genome.
To identify the location within the genome for these genes, we mark where variations (known as single nucleotide polymorphisms, or SNPs), associated with desired traits occur within the genome.
By assaying SNPs of interest across the genome, we can derive a prediction equation from these SNPs that will predict the cumulative effect of all the genes affecting a trait like marbling.
This was a major piece of learning and led to a change in strategy in Beef CRC 3 for both our research focus and commercialisation strategies.
The second thing we realised was that the SNP chips available at the time for cattle – a 10,000 SNP chip and, later, the 50,000 chips – did not contain enough SNP data to be sufficiently accurate across different beef cattle breeds.
The 50k chip works well if you want to predict a breeding value in that same breed. It has, for example, been tremendously successful in dairy cattle because of the heavy focus in that industry on a single breed – the Holstein. There are now datasets of tens of thousands of Holsteins and the industry is using this data to predict breeding values. AI studs are decreasing their progeny programs and basing their selection on genomic predictions.
However, the problem for beef cattle is that we have data spread across 7-8 breeds – and we want to be able to make predictions across these breeds. So for us, the 50k chip is not dense enough. You might have a gene for marbling in both Hereford and Angus but in one breed, SNP number 21 is correlated with this gene and in other breed SNP number 1052 is correlated with this gene. Because the association between SNPs varies from one breed to another, it is almost certainly the case that the variation between SNPs and real genes also varies from one breed to another.
Thus, in the beef industry, we have been a somewhat stymied in delivering prediction equations to industry based on the use of these 50k SNP chips. However, higher density SNP chips of around 650k and 800k have recently been released by Illumina and Affymetrix – and we believe these chips are of a high enough density to develop a prediction equation that will work across breeds. This is what the Beef CRC is currently working towards. We are genotyping animals with these high density SNP chips and we will soon be analysing that data to see if we can develop prediction equations that are useful across breeds.
One new strategy for developing prediction equations across breeds that is showing some promise is to genotype an animal using one of the new highdensity SNP chips, then measure the results against the same genotyping using a smaller 50k chip. Our preliminary data is showing that we can use the data from the 50k SNP chip to impute or make an estimate of the data the high density chip will deliver with a high degree of accuracy (in the order of 90%). If this data proves to be robust, it means we may not have to repeat genotyping using the expensive high-density chip across a whole population of animals, rather focus on genotyping animals in the BIN (Beef Information Nucleus) herds and a smaller number of elite breeding animals.
The Beef CRC has genotyped 6,200 cattle with the 50k SNP chip and is now genotyping 1,750 cattle with the new 800k SNP chip to impute 800k genotypes on all cattle with 50k and generate a discovery population of 7,950 cattle with 800k genotypes and phenotypes for carcase and meat quality traits, NFI or fertility.
Currently this high density genotyping work is being done on samples from all of the breeds where we have lots of data – Angus, Hereford, Shorthorn, Belmont Red, Santa Gertrudis, Brahman and composites from the various pastoral companies.
The difficulty is that the accuracy of these imputed prediction equations depends on the size of the reference database you have – that is, the number of animals that have been measured for the trait and genotyped with the SNP chip. While we would like to have around 30,000 animals within each breed genotyped, this is not possible so our alternative strategy is to build up as many animals as we can, accumulatively, across all the breeds.
This is not the only strategy which can be adopted but, as the Beef CRC represents all breeds, and also because we don't have enough data within any one breed yet to do the job, it is the strategy we are currently working on.
The purpose of prediction equations is to estimate the breeding values of individual animals, so that cattle breeders can select the ones that suit them best. BREEDPLAN already performs this task for phenotypic measurements and pedigree information.
Cattle breeders don't want to have two different systems trying to do the same thing, so it is very important that the validated prediction equation gets incorporated with phenotypic and pedigree information in BREEDPLAN to deliver a single best EBV.
This will enable the cattle breeder to get the overall best estimate of the breeding value of his or her cattle.
Prediction equations will be derived from work we are currently doing with high density chips. We also have 1,300 DNA samples collected by breed associations that will be genotyped by June and used to validate some of the prediction equations derived from CRC cattle.
The CRC has dedicated much research into what determines the accuracy of prediction equations. We have developed prediction equations within a certain referenced dataset, where we have a number of animals that have been measured for the traits and genotypes.
Currently we are trying to demonstrate we can use this method to account for 15% of genetic variation between animals – which may not sound like a lot – but it will mean the accuracy of the EBV will be about 40%, which is at a level that is commercially useful.
Prediction equations are not a product the cattle breeders will "see" – rather they represent complex mathematics that will underpin more accurate EBVs.
Cattle producers can then access thee improved EBVs through BREEDPLAN.
Successfully validated prediction equations for carcase and meat quality traits, NFI and fertility will be transferred to BREEDPLAN and genomics companies in 2012."