Get to Know Genomic Testing

by John Genho, Livestock Genetic Services

Over the past several years, genomic tests have become more available to breeders for genetic improvement. These have come in several different forms and have been used in different ways. The following is a review of these forms and methods.

Initially, our collective thought was that we could find certain genes that coded for certain traits and select for the positive forms of those genes. A lot of research, time, and money was spent searching for these genes. This resulted in a few markers for tenderness and a single marker for marbling. The GeneStar markers are an example of this. However, as we looked at the impact that these genes had on these relatively easy traits, and the cost that went into searching for them, we soon realized that this model was not going to be used long term.

Following this realization, a new idea developed. We would look at a very high number of markers, spread evenly throughout the genome, and determine how these markers were associated with certain traits. This would result in a set of prediction equations that could be applied to other animals. The idea was that a very limited number of animals could be tested and then the results of this could be applied to a larger population. The double-step methods of breed associations and the commercial panels of genotyping companies relied on this idea. However, we soon realized that once we left the initial population that we studied, the predictions were not very accurate. In addition, breeders began to see that they could simply turn in a DNA test, in place of actually gathering phenotypes and performance data on their animals. Long-term DNA tests cannot, and should not, take the place of good phenotypes but instead should enhance the phenotypes that are collected.

Today, the industry, in general, has arrived at the idea that a single-step model is the best method for genomic prediction. The basic idea is that the values of genomic markers are determined at the same time that the expected progeny differences (EPDs) are calculated to prevent any problems between populations or double counting of effects. In the near future, all breed associations running genomic-enhanced EPDs (GE-EPDs) in the United States will be switching to single-step models. The International Brangus Breeders Association (IBBA) was the second association to run EPDs based on a single-step model with the change coming two years ago.

There are two single-step models that are used currently. The first is GBLUP, which is the model that IBBA uses. The idea behind this model is that we use genomic tests to help firm up the relationships between animals. The American Angus Association is in the process of switching to this model for their EPDs. The second single-step model is the super hybrid model. Since the BOLT platform is built on this model, I’ll simply refer to this model as BOLT going forward. The idea behind this model is that we run successive attempts to simultaneously estimate the EPDs and genomic values until an acceptable version is reached. International Genetic Solutions, the group which runs genetic evaluations for Simmental, Red Angus, and many other associations, is in the process of switching their evaluation to the BOLT platform.

These two models each possess pros and cons. GBLUP is easier to implement with a limited number of genotyped animals and fits nicely with our traditional genetic evaluation software. However, as the number of genotyped animals goes up, it becomes more difficult to run. Also, the basic form of GBLUP assumes all markers have the same effect on each trait, a simplifying assumption we know is not true. The model which BOLT uses on the other hand is more difficult to work into existing evaluations. However, once BOLT is implemented, it handles a high number of genotyped animals more easily. In addition, it does not assume each marker has the same effect on each trait. While these comparisons hold true in general, the American Angus Association and University of Georgia are developing GBLUP to overcome some of the negatives making GBLUP a viable option going forward.

An additional consideration when comparing the models is how many markers are included in genomic prediction. Many of the current genotypes for all associations have been done at 50K densities. We’re just getting into ultrahigh density genotypes in the industry. At these lower densities, the models will likely pick very similar animals making the model decision less relevant.

IBBA is currently using the GBLUP model, which has proven to be an effective model. Figure 1 shows a portion of the 2015-born bulls that were both genotyped and ultrasounded. For this test, all scan records for these bulls were removed. The EPDs were then run with genomic tests included and without genomic tests included. The correlation between these EPDs and the actual scan records were then measured (displayed as r in the figure). When genomic tests were used, the corre lation between the EPDs and the scan records jumped from 40 percent to 45 percent. Similar results can be found for different traits and breeds. The bottom line is that GBLUP works in the IBBA dataset. We are better able to predict performance when breeders send in a DNA sample than when breeders do not. Going forward, we can continue to evaluate further models, but for the time being, the genomic model is working.

The majority of the problems with all models are the imbalance between the number of animals registered and the number of animals with DNA tests. Looking further into the future, the absolute best situation is going to be when all animals that are registered are also genotyped. For the time being, this is cost prohibitive, but down the road it is likely to become more affordable.

We’ve reviewed several models in this article and weighed out the costs and benefits to them. While there are many exciting options open to associations, all of these depend on the continued collection and reporting of performance data. Without phenotypes, all of this is just sounding brass and tinkling cymbals. The EPDs must be built on sound data being collected and reported to the association. DNA tests do not remove the need to collect phenotypes. Instead, they put the phenotypes that you collect to better use in building selection tools.