Genomic selection in crops and livestock
How genomic selection works: training populations, GEBVs, accuracy, why it doubled dairy genetic gain, and its use in crops and Indian cattle.
On this page
For most of the twentieth century, a dairy bull's genetic merit for milk could only be judged from his daughters. That meant waiting about five years while they were born, grew up and completed a lactation. Crop breeders faced a similar wait: years of multi-location yield trials before a line's real value was known.
Genomic selection (GS) removes much of that waiting. Breeders genotype a young animal or seedling with thousands of DNA markers across the whole genome, and a statistical model predicts its breeding value immediately. Since its widespread adoption in dairy cattle around 2009, GS has roughly doubled the annual rate of genetic gain in many dairy populations. It is now spreading through crop breeding, from CIMMYT's maize and wheat to rice and pulses in India.
The idea
Marker-assisted selection works well when a few genes with large effects control a trait. Most economic traits, such as yield, milk production, fertility and drought tolerance, are controlled by hundreds or thousands of genes with tiny effects. Trying to find each one with significance tests misses most of them.
Meuwissen, Hayes and Goddard proposed a different approach in 2001: estimate the effects of all markers at once and sum them. With markers dense enough that every gene is in linkage disequilibrium with at least one marker, the sum captures most of the genetic variance, without anyone needing to know where the genes are.
The process has three parts:
- A training population is genotyped and phenotyped. In crops these are lines with field trial data; in dairy, bulls with daughter records.
- A model estimates marker effects or genomic relationships. Common methods are GBLUP, which uses a genomic relationship matrix, ridge-regression BLUP, and Bayesian methods (BayesA, BayesB, BayesCπ), which allow some markers to have larger effects. Machine-learning models are also being tested.
- Selection candidates are genotyped only. Their genomic estimated breeding values (GEBVs) are predicted, and the best are selected as parents.
Why it speeds up gain
The response to selection per year is:
Genomic selection may not improve accuracy (r) much. Its main effect is to cut the generation interval (L) sharply, because candidates can be selected before they are phenotyped. It can also raise i, because many more candidates can be screened cheaply.
The expected accuracy depends on how much information the training population provides:
Phenotypic versus genomic selection in wheat
For a yield trait with h2 = 0.4, selection on replicated trial means might reach accuracy r = 0.63 with a five-year cycle. A genomic model trained on 1,000 lines with Me = 1,000 gives r ≈ √(400 ⁄ 1,400) ≈ 0.53, but candidates can be selected within two years. Gain per year is proportional to r ⁄ L: 0.63 ⁄ 5 = 0.13 for phenotypic selection, against 0.53 ⁄ 2 = 0.27 for genomic selection. That is about twice the gain per year, even with lower accuracy.
Livestock: where it took off
Dairy cattle were ideal for GS: progeny testing was slow and costly, artificial insemination spreads each bull's genes widely, and large training populations of progeny-tested bulls already existed.
- The USA began official genomic evaluations in 2009, and other major dairy countries followed.
- The average generation interval for sires of bulls in the US fell from about 7 years to about 2.5 years. Annual gains in milk yield, fat and protein roughly doubled, and fertility and health traits improved for the first time.
- Pigs, poultry, beef cattle, sheep and aquaculture species now use GS widely.
India's large indigenous breeds lacked tools until recently. The IndiGau SNP chip, developed for Indian cattle such as Gir, Sahiwal and Tharparkar and launched in 2021, and similar chips for buffalo, are the first step toward genomic evaluation of Indian bulls.
Crops: rapid spread
Crops have different challenges. Genotype × environment interaction is strong, breeding programmes are smaller, and lines are cheap to produce but expensive to test. GS is used to:
- predict line performance before multi-location trials, so that only the most promising lines are tested;
- choose parents and crosses early in rapid-cycle recurrent selection, combined with speed breeding or off-season nurseries;
- predict hybrid performance in maize, sorghum and rice from parental genotypes, avoiding the need to test every cross.
CIMMYT uses GS routinely in its maize and wheat programmes. Its maize work in sub-Saharan Africa has shown faster gains for drought tolerance. Indian institutes, including IARI, IRRI's South Asia hub and ICRISAT, are applying GS in rice, wheat, chickpea and pearl millet. Falling genotyping costs, now a few US dollars per sample for low-density SNP panels, have made GS affordable for public programmes.
Challenges
- Training population design: accuracy falls when candidates are distantly related to the training set, so models must be retrained regularly with new data.
- Genotype × environment interaction: multi-environment models and weather or soil covariates are being developed to handle it.
- Loss of diversity: fast selection of closely related elite parents can narrow the genetic base. Optimal contribution methods help balance gain and diversity.
- Data and skills: GS needs high-quality phenotypes, data management and statistical expertise, which are often bottlenecks in smaller programmes.
Key points
- Genomic selection predicts breeding values from genome-wide markers using a model trained on genotyped and phenotyped individuals (Meuwissen, Hayes and Goddard, 2001).
- It captures many small-effect genes that marker-assisted selection misses.
- Gain per year = i r σA ⁄ L. GS mainly shortens L, and can raise i.
- Accuracy rises with training size, heritability and relatedness: r ≈ √[N h2 ⁄ (N h2 + Me)].
- Dairy cattle adopted GS around 2009, and the rate of genetic gain roughly doubled.
- In crops, GS is used for early-generation testing, rapid-cycle selection and hybrid prediction. India's IndiGau chip supports indigenous cattle.
What the exams ask
- Define genomic selection and GEBV, and distinguish GS from MAS.
- Explain the role of training and test populations.
- Use the breeder's equation per year to show why GS speeds up gain.
- List factors affecting prediction accuracy.
- Give examples of GS use in livestock and crops.
Described in general terms: the kind of examination, not a past paper. Laisha never reproduces exam questions.
Questions people ask
Does genomic selection replace field trials?
No. Models depend on good phenotype data from the training population, and they must be updated as programmes advance and climates change. GS lets breeders test fewer, better candidates in the field and recycle parents faster. Field trials are still essential before release.
Is genomic selection the same as genetic engineering?
No. GS only uses DNA information to choose parents; the DNA itself is not altered. Animals and varieties bred with GS are conventional.
Why does accuracy drop in a new population?
The model learns associations between markers and genes that hold in the training population. In a distantly related population, these linkage patterns and allele frequencies differ, and gene effects may differ too. Predictions then rely less on true linkage and more on noise.
Sources
- Meuwissen, T. H. E., Hayes, B. J. and Goddard, M. E. "Prediction of total genetic value using genome-wide dense marker maps." Genetics 157, 1819–1829 (2001).
- García-Ruiz, A. et al. "Changes in genetic selection differentials and generation intervals in US Holstein dairy cattle as a result of genomic selection." PNAS 113, E3995–E4004 (2016).
- Daetwyler, H. D., Villanueva, B. and Woolliams, J. A. "Accuracy of predicting the genetic risk of disease using a genome-wide approach." PLoS ONE 3, e3395 (2008).
- Crossa, J. et al. "Genomic selection in plant breeding: methods, models, and perspectives." Trends in Plant Science 22, 961–975 (2017).
- Varshney, R. K. et al. "Designing future crops: genomics-assisted breeding comes of age." Trends in Plant Science 26, 631–649 (2021).
More in genetics and plant breeding
Discussion
Questions, corrections and notes from readers on this article. Every comment is checked before it appears: keep it on the topic, and do not share phone numbers, links or contact details.
Last reviewed