Using genetic selection and genomics to combat infectious disease - Dr. Jack Dekkers, Iowa State University, from the 2017 North American PRRS/National Swine Improvement Federation Joint Meeting, December 1‐3, 2017, Chicago, Illinois, USA.
More presentations at http://www.swinecast.com/2017-north-american-prrs-nsif-joint-meeting
Dr. Jack Dekkers - Using genetic selection and genomics to combat infectious disease
1. Using genetic selection
and genomics
to combat infectious disease
Jack Dekkers
Animal Breeding & Genetics
Department of Animal Science
Iowa State University
3. The Power of Genetic Improvement
1972: 380 kg feed 100 kg market pig
2007: 325 kg feed 125 kg market pig
Courtesy: Graham Plastow, Univ. Alberta
5. Resistance to some pathogens is controlled by a single gene
F18 E-coli in Pigs
6. Use of Gene Editing to Produce Resistant Pigs
Das et al. J. Virology 2010
Dr. Randy Prather
Univ. Missouri
Rowland Lab,
KSU
7. WUR = 21
18
15
12
9
21
Mutation in
GBP5
explains
15% of
genetic
variance
0 4 7 11 14 21 28 35 42
0
2
4
6
8
Days Post Infection
Viremia(Log)
Viral Load
Area Under Curve
Natural Resistance to PRRS is incomplete and polygenic
“tag SNP”
WUR10000125
GWAS for Viral Load
8. Resistance, Tolerance, or Resilience
0
0.25
0.5
0.75
1
1.25
1.5
1.75
2
0 0.25 0.5 0.75 1
PathogenLoad
Level of Exposure
0.4
0.6
0.8
1
1.2
1.4
0 0.25 0.5 0.75 1 1.25 1.5
Growthrate(lbs/d)
Pathogen Load
0.4
0.6
0.8
1
1.2
1.4
0 0.25 0.5 0.75 1
GrowthRate(lbs/d)
Level of Exposure
B
A B
A
B
A
Resistance
Ability to prevent infection
or limit replication
Tolerance
Ability to maintain
performance at a given
pathogen load
Resilience / Robustness
Ability to maintain
performance upon exposure
to a pathogen
B
9. WUR = 21
18
15
12
9
21
GBP5
region
explains
11% of
genetic
variance
“tag SNP”
WUR10000125
Resilience to PRRS is highly polygenic
0 1 2 3 4 5 6
0
5
10
15
20
25
30
Week
Weight(kg)
GWAS for Weight Gain
10. Challenges for Genetic Improvement of Quantitative Disease
• Difficult to obtain relevant phenotypic records
• No clear easy-to-measure phenotypes
• Inconsistent recording/diagnosis
• Selection candidates are kept
in facilities with high biosecurity
• High-health nucleus farms
• Low heritability
• Selection for resistance to one disease may be
counterproductive for other diseases
11. Approaches for Collection of Health-Related Phenotypes
1) Collect disease indicator traits
on selection candidates
1) Collect health-related phenotypes
(incl. performance) on commercial farms
3) Collect health-related phenotypes in
special experimental facilities
Possible Selection Criteria
• Phenotype-based EBV
• Major genes or genetic markers
• Genomic prediction
Genes
Major genes
Markers
QTL
Black box of
Quantitative genetics
13. 3500 YXLR WEANER PIGS
FROM HIGH-HEALTH HERDS
BATCHES OF 60-75 PIGS/3 WEEKS
650K SNP
genotypes
Trait Genetics
• Parameters
• Architecture
• Mechanisms
Genetic
Prediction
of Resilience
Phenotypic
Prediction
of Resilience
PROJECT
DELIVE-
RABLES
Potential BIOMARKERS
collected on
young healthy pigs
Immune Response
• High Immune Response
• In vitro immune tests
• Phagocytosis tests
Blood Transcriptome
Blood Metabolome
Blood Proteome
Gut Microbiome
Integrated
statistical and
bioinformatic
analyses
NATURAL CHALLENGE FACILITYQUARANTINE NURSERY
SEEDER PIGS
Continuous flow RESILIENCE PHENOTYPES
Mortality/Morbidity
Growth performance
Feed intake
Water intake
Disease diagnostics
Blood transcriptome
Gut Microbiome
Health treatments
14. Vaccine Response as an Indicator of Host Response
ANTIBODY- MEDIATED
IMMUNE RESPONSE
CELL-MEDIATED
IMMUNE RESPONSE
Photos and figures courtesy of Dr. B. Mallard
+
h2 ~ 30%
Disease data from: Wagter, et al. 2000 J. Dairy
Sci. 83:488-498; Thompson-Crispi, et al. 2012.
J. Dairy Sci. 95:3888-3893; Thompson-Crispi, et
al. 2013. Clin Vacc Immuno. 20:106-112.
High Immune Response
Cows have less disease
7.5%
11.3%
14.4%
High Average Low
64 herds
tested
Bonnie Mallard et al.,
University of Guelph
15. Approaches for Collection of Health-Related Phenotypes
1) Collect disease indicator traits
on selection candidates
1) Collect health-related phenotypes
(incl. performance) on commercial farms
3) Collect health-related phenotypes in
special experimental facilities
Possible Selection Criteria
• Phenotype-based EBV
• Major genes or genetic markers
• Genomic prediction
Genes
Major genes
Markers
QTL
Black box of
Quantitative genetics
16. Approaches for Collection of Health-Related Phenotypes
1) Collect disease indicator traits
on selection candidates
1) Collect health-related phenotypes
(incl. performance) on commercial farms
3) Collect health-related phenotypes in
special experimental facilities
Possible Selection Criteria
• Phenotype-based EBV
• Major genes or genetic markers
• Genomic prediction
Genes
Major genes
Markers
QTL
Black box of
Quantitative genetics
17. Collect Health-Related Phenotypes in the Field
X
X
Purebred
selection
Sire
line
Dam
line
Dam
line
Crossbred - reproduction phase
Cross-bred market pigs
Grow-finish phase
Purebred
lines
Nucleus
Field (Multiplication)
(Multiplication)
CBPh
CBPh
CBPhenotypes
CBPhenotypes
CBPhenotypes
Challenges
• Complicated logistics
• No pedigree (?)
• Time delays
• generation intervals
19. Genomic Selection for Performance in the Field
X
X
Purebred
selection
Sire
line
Dam
line
Dam
line
Crossbred - reproduction phase
Cross-bred market pigs
Grow-finish phase
Nucleus
Field (Multiplication)
(Multiplication)
Field phenotypes
and
HD SNP genotypes
HD SNP genotypes selection candidates
SNP effects
for field
phenotypes
Genomic
training
Genomic
prediction
GEBV for
field
phenotypes
Genomic
selection
Genomic
selection
20. Approaches for Collection of Health-Related Phenotypes
1) Collect disease indicator traits
on selection candidates
1) Collect health-related phenotypes
(incl. performance) on commercial farms
3) Collect health-related phenotypes in
special experimental facilities
Possible Selection Criteria
• Phenotype-based EBV
• Major genes or genetic markers
• Genomic prediction
Genes
Major genes
Markers
QTL
Black box of
Quantitative genetics
21. Approaches for Collection of Health-Related Phenotypes
1) Collect disease indicator traits
on selection candidates
1) Collect health-related phenotypes
(incl. performance) on commercial farms
3) Collect health-related phenotypes in
special experimental facilities
Possible Selection Criteria
• Phenotype-based EBV
• Major genes or genetic markers
• Genomic prediction
Genes
Major genes
Markers
QTL
Black box of
Quantitative genetics
24. Can we use Genomic Prediction across isolates?
NVSL Trials
KS06 Trials
Train
Predict
Train
Predict
Training KS06 NVSL
29
Accuracyofgenomieprediction
Prediction
Training KS06 NVSL
25. Summary / Conclusions
• Resistance/host response to PRRS
and other infectious disease has a
sizeable host genetic component
but tends to be controlled by many genes
• Current breeding programs have limited
opportunities to select for improved health
• Genetic improvement of resistance or
susceptibility to quantitative/complex diseases can be achieved by
• Selecting on disease indicator traits collected on selection candidates or close relatives
• Selecting on health-related phenotypes from the field or from experimental facilities
• Requires continuous recording of disease-related traits on relevant populations
• Genomics increases the potential impact of these strategies
26. Funding Industry Partners
NIFA
Thanks to all Partners
Iowa State University
Nick Serao
Jim Reecy Chris Tuggle
Susan Carpenter
Kansas State University
Bob Rowland PRRS group
USDA-ARS
Joan Lunney group
University of Alberta
Graham Plastow group
Univ. Saskatchewan
John Harding group
Roslin Institute
Steve Bishop
Andrea Doeschl-Wilson
Univ. Minnesota
Monserat Torremorrell
Scientific
Collaborators
27.
28. Discussion questions
• Should genetic improvement focus on specific diseases or on general health?
• What additional information is needed for incorporating host genetics in disease
control strategies?
• What disease phenotypes could be collected in the field (in collaboration with
veterinarians?) that could be used for genetic improvement of health?
• Is it worthwhile to consider added phenotype screens (wide proteomics,
metabalomics, etc. or narrow selected gene/protein targets)?
• Should we consider only monitoring at early times points post infection for most
informative data on immune responses?
• What are the opportunities to select pigs for improved response to vaccines?
Vaccine-ready pigs?
• What role can/should genome editing play in genetic improvement of health?
Editor's Notes
The power of genetic improvement is clearly demonstrated in these examples.
In pigs, it now requires 15% less feed to reach a 25% higher market weight than 40 years ago
In dairy cattle, milk production per cow has almost quadrupled over the past 70 years, resulting in a 60% reduction in the number of cows, while still doubling the nation’s milk production.
The most dramatic example is genetic improvement in broiler chickens, here showing broilers at 7 weeks of age using 2005 genetics versus 1957 genetics on the same diet.
This table summarizes the number of Mendelian traits or disorders that have been discovered in livestock and pet species. For the pig, a total of 248 traits have been reported on, of which 67 have been shown to be affected by a single gene. Of those, the causative mutation is known for 27. The rest of this table shows the number of traits that have been discovered for different catagories, including metabolic disorders, coat color, retinal disorders, bloodgroup systems, an platelet disorders.
This table summarizes the number of Mendelian traits or disorders that have been discovered in livestock and pet species. For the pig, a total of 248 traits have been reported on, of which 67 have been shown to be affected by a single gene. Of those, the causative mutation is known for 27. The rest of this table shows the number of traits that have been discovered for different catagories, including metabolic disorders, coat color, retinal disorders, bloodgroup systems, an platelet disorders.
For any of these genes, extensive information is available, here for the halothane gene, including symptoms, mode of inheritance, how it was mapped, and the molecular basis of the trait. So the halothane gene causes porcine stress syndrome in homozygous form, and was found to be caused by a mutation in the Ryanodine Receptor 1 gene, which encodes a calcium release channel in the skeletal muscle sarcoplasmic reticulum.
One obvious strategy to remove alleles that have negative effects through defects or other disorders, is to not use animals that are affected for breeding. This will of course occur automatically for defects that are lethal. A problem with this approach is that most of these Mendelian traits are deleterious recessives, so carriers are normal. This complicates removal of these deleterious alleles from a population, as is illustrated in this figure, which shows the decline in the frequency of the deleterious allele.
The reason why this happens is because most deleterious alleles ‘hide’ in carriers when the frequency gets low. This is illustrated in this graph, which shows that the frequency of affected animals is close to zero when the frequency of the an allele is low, although there’s still a substantial proportion of carriers present.
This graph shows the Manhattan plot for the GWAS of viral load in the NVSL experimental infection trials, using the Nursery Challenge Model Trials. In this graph, each dot represents the –log10 p-value for a given SNP, with SNPs ranked based on position in the genome by chromosome. The horizontal line indicates the significance threshold.
The most striking result from this analysis is the high significance of SNPs at the end of chromosome 4, indicated by the red dots with p-values as low as 10 to the power negative 21.. Further analysis showed that this includes SNPs in a region of about 1 Mb, that explained about 15 % of the genetic variance for viral load. So this region contains a gene or mutation that has a large effect on host response to PRRS. Because of the size of the effect, it is referred to as a major QTL.
Well, when we look at the GWAS for host response to PRRS, not many regions stand out, beyond the GBP5 region, which suggest that the rest of the genetic differences in host response to PRRS is highly polygenic, in other words, affected by many genes across the genome, each with a small effect.
In addition, when we look at the GWAS for the KS06 virus infection data, there are no regions that overlap between the two isolates, apart from the GPB5 region . At first, this was surprising, because we had estimated the genetic correlation between host response to these two isolates to be quite high, at 0.86, which suggest that a large proportion of the genetic variation in response to these two isolates is due to QTL that have an effect for both isolates. This result can, however, be explained by the polygenic nature of host response, which results in low power to identify other regions that are associated with host response, even for these fairly large data sets.
That there is a common genetic basis for host response to these two isolates was, however, confirmed by the overlap in the biological pathways that genes that are near regions that show suggestive associations with host response belong to for these two isolates. So, although we could not identify common regions for these two isolates, other than GBP5, there is evidence that the biological pathways that contribute to response to these two isolates overlap.
In the GWAS for weight gain in these same trials, the same region at the end of chromosome 4 showed high significance. For weight gain, this region explained 11% of the genetic variance, so this QTL region not only has a major effect on VL but also on weight gain following PRRS infection.
Further evaluation of this genomic region using the genome data that is available on the pig showed that this region includes a number of important candidate genes that are known to be involved in innate immune response, including several members of the Guanylate Binding Protein family. Based on further functional genomic analyses, we were able to identify what we believe is the mutation that causes this effect, which we found in the GBP5 gene. More specifically, this mutation results in an early stop codon in the gene, which results in the production of an incomplete and likely non-functional protein.
Although the exact function of the GBP5 gene is not known, a recent report on humans demonstrated that the GBP5 protein gene is involved in reducing the infectivity of HIV. This makes it very plausible that it also could have an function in PRRS.
Although this causative mutation is not on the SNP chip, the chip does include a SNP that is in nearly complete linkage disequilibrium with the causative mutation, such that genotype at this SNP very closely tags the genotype at the causative mutation. This SNP will be referred to as the WUR SNP in the remainder of this presentation.
In the GWAS for weight gain in these same trials, the same region at the end of chromosome 4 showed high significance. For weight gain, this region explained 11% of the genetic variance, so this QTL region not only has a major effect on VL but also on weight gain following PRRS infection.
Further evaluation of this genomic region using the genome data that is available on the pig showed that this region includes a number of important candidate genes that are known to be involved in innate immune response, including several members of the Guanylate Binding Protein family. Based on further functional genomic analyses, we were able to identify what we believe is the mutation that causes this effect, which we found in the GBP5 gene. More specifically, this mutation results in an early stop codon in the gene, which results in the production of an incomplete and likely non-functional protein.
Although the exact function of the GBP5 gene is not known, a recent report on humans demonstrated that the GBP5 protein gene is involved in reducing the infectivity of HIV. This makes it very plausible that it also could have an function in PRRS.
Although this causative mutation is not on the SNP chip, the chip does include a SNP that is in nearly complete linkage disequilibrium with the causative mutation, such that genotype at this SNP very closely tags the genotype at the causative mutation. This SNP will be referred to as the WUR SNP in the remainder of this presentation.
One of the big challenges for the incorporation of selection for disease resistance in breeding programs is that it is very difficult to obtain relevant phenotypic data that can be used to estimate breeding values for disease resistance. Reasons for this include that there are often no clear and easy-to-measure phenotypes, that recording or diagnosis of diseases is often inconsistent, especially across farms. In addition, selection candidates are often kept in facilities with high biosecurity, which means that disease data cannot be recorded on selection candidates. Examples are AI bulls, which must be kept disease free to allow marketing and export of their semen, and the high-health nucleus farms that are used in pig breeding programs, again in order to enable sale of germstock from these farms, both domestically and abroad.
Another challenge for genetic improvement of disease traits is that most disease-related phenotypes have low heritability.
And, finally, care must be taken that selection for resistance to one specific disease may make animals more susceptible to other diseases.
There are several approaches that can be used to overcome at least some of these challenges.
One is to develop disease indicator traits that can be collected on selection candidates without exposure to disease and I will give a couple of examples of such traits.
Another approach is to collect phenotypes in commercial farms and utilize that information to estimate the breeding value of selection candidates in the nucleus.
And finally, health-related phenotypes can be collected in special experimental facilities, followed by the use of the information generated in that facility for selection in the nucleus.
Phenotypic data on disease-related traits that is generated with these approaches can be used for selection using one of three approaches, which have been described in previous models:
- To estimate EBV related to disease on selection candidates using the collected phenotypes.
By identifying major genes or genetic markers that are are associated with disease and using these to genotype selection candidates
By using genomic prediction based on high-density SNP genotypes.
This scatter plot compares PRRS viral load following vaccination with PRRS viral load following co-infection on the pigs that were first vaccinated. There’s a positive relationship between the two that shows that pigs that allow high levels of viremia following vaccination also tend to have higher levels of PRRS viremia following co-infection, but the relationship was not strong. This positive relationship was not only present at the phenotypic level but also appeared to hold at the genetic level, with a genetic correlation of +0.41, although the standard error of this estimate was very large. The red numbers on the diagonal of this table show the heritabilities for these traits, which were low for the vaccinated birds, both after vaccination and co-infection, but relatively high for birds that were not vaccinated. Another interesting relationship is the genetic correlation between PRRS VL following vaccination with PRRS VL following co-infection without vaccination, which was almost 1. Again note the high se., however. This suggests that these traits are genetically the same, which may make sense because it reflects viral load after first exposure to the virus, either as a vaccine or through experimental infection. This is in contrast to the correlations of PRRS VL of the vaccinated birds following co-infection with first exposure VL, either as a vaccine (0.41) or through co-infection (0.21).
This table shows genetic parameter estimates for PCV2 VL of the non-vaccinated or the vaccinated birds. Heritabilities were relatively low, reflecting the noisyness of these data, but the genetic correlation was close to 1, suggesting that differences in PCV2 viremia reflect the same genetic trait with or without prior vaccination against PRRS.
Finally, This table gives estimates of genetic parameters for growth rate. We have growth rate before co-infection of the non-vaccinated and the vaccinated birds, as well as growth rate after co-infection of the non-vaccinated and the vaccinated birds. Heritabilities were reasonably high, except for growth rate following vaccination, but genetic correlations tended to be low, except for growth rates of non-vaccinated and vaccinated birds, either before co-infection (0.6) or after co-infection (0.77).
HIR technology involves two tests that can be performed on healthy animals, including bulls, that capture an animal’s ability to resist a wide array of diseases by measuring the animals potential to generate a cell-mediated response and an antibody-mediated response.
To measure cell-mediated response, a skin test is performed by injecting an antigen that drives this type of response into the tail fold. This picture shows measuring the the tail-folds to determine the cell-mediated response. The side measured here is the control side whereas the other side is the test side. The picture below shows the infiltration of cells at the injection site to cause the swelling at that site, with all the purple dots being immune cells.
To measure antibody mediated response, blood samples are taken after exposure to a pool of antigens and the antibody activity is measured using an ELISA test, with more antibody activity indicating a higher antibody response.
Results in this bar graph show that, on average, cows that score high for both these tests have less disease.
HIR technology for dairy cattle is being used by Semex Canada, under an exclusive license from the University of Guelph, and marketed as immunity +.
As I mentioned, similar tests are currently being evaluated for pigs, which would allow immune response to be measured on pigs in a biosecure nucleus environment as an indicator trait for susceptibility to disease in the field.
There are several approaches that can be used to overcome at least some of these challenges.
One is to develop disease indicator traits that can be collected on selection candidates without exposure to disease and I will give a couple of examples of such traits.
Another approach is to collect phenotypes in commercial farms and utilize that information to estimate the breeding value of selection candidates in the nucleus.
And finally, health-related phenotypes can be collected in special experimental facilities, followed by the use of the information generated in that facility for selection in the nucleus.
Phenotypic data on disease-related traits that is generated with these approaches can be used for selection using one of three approaches, which have been described in previous models:
- To estimate EBV related to disease on selection candidates using the collected phenotypes.
By identifying major genes or genetic markers that are are associated with disease and using these to genotype selection candidates
By using genomic prediction based on high-density SNP genotypes.
There are several approaches that can be used to overcome at least some of these challenges.
One is to develop disease indicator traits that can be collected on selection candidates without exposure to disease and I will give a couple of examples of such traits.
Another approach is to collect phenotypes in commercial farms and utilize that information to estimate the breeding value of selection candidates in the nucleus.
And finally, health-related phenotypes can be collected in special experimental facilities, followed by the use of the information generated in that facility for selection in the nucleus.
Phenotypic data on disease-related traits that is generated with these approaches can be used for selection using one of three approaches, which have been described in previous models:
- To estimate EBV related to disease on selection candidates using the collected phenotypes.
By identifying major genes or genetic markers that are are associated with disease and using these to genotype selection candidates
By using genomic prediction based on high-density SNP genotypes.
Another strategy to obtain information on disease-related traits or on performance under disease is to record disease-related phenotypes on commercial animals in the field. So for reproduction-related traits, phenotypes would be recorded on sow herds and for grow-finish related diseases, phenotypes would need to be recorded on finisher pigs. These phenotypes would then be used for genetic evaluation of animals in the nucleus herds. A major challenge with this approach, however, is that, apart from the logistics involved in collecting this data, it would require pedigrees to be tracked through the system because in order to compute an EBV of a nucleus animal for a trait that is recorded in the field, you need to have phenotypes on relatives. Recording pedigrees on commercial animals, however, is a big challenge, especially with the use of mixed semen.
In addition, there will be a time delay in getting information on relatives from the field, especially if you wait for phenotypes on commercial progeny. This would increase generation intervals and reduce response per year. Nevertheless, this approach has been implemented in a number of breeding programs, allowing to select more directly for performance in the field, rather than performance in a high-health environment.
Another strategy for utilizing phenotypes collected in the field for selection of breeding stock in the nucleus is genomic prediction
As discussed in a previous module, the general idea behind genomic prediction and genomic selection is to genotype a substantial number of animals that have recorded phenotypes for traits of importance, and to genotype them using a high-density SNP chip. Using these phenotypes and genotypes, the prediction model is trained by estimating the effect of each SNP and using the resulting estimates to compute a genomic EBV of selection candidates.
In this case, the training population would consist of genotypes and phenotypes collected on crossbred animals in the field and the selection candidates would be animals in the nucleus.
So data and high-density SNP genotypes would be collected on crossbred pigs in the field. On crossbred market pigs for the grow-finish phase and on crossbred sows for the reproduction phase. To focus on disease-related traits or performance under disease, the farms that these data are collected on should have some history of disease pressure.
The resulting genotypes and phenotypes would then by used to estimate SNP effects for disease and performance in the field through genomic training based on the HD SNP genotypes of selection candidates, which would then be used to predict the breeding value of nucleus animals for field phenotypes. This would allow genomic selection of nucleus populations for disease and performance in the field.
One important difference of this genomic prediction approach compared to the strategy where phenotypes collected in the field are directly used for genetic evaluation in the nucleus is that the genomic prediction approach does not require pedigree to be tracked through the system. In addition, the individuals that are phenotyped do not need to be close relatives of the selection candidates, although accuracies of genomic prediction will be higher if they are.
There are several approaches that can be used to overcome at least some of these challenges.
One is to develop disease indicator traits that can be collected on selection candidates without exposure to disease and I will give a couple of examples of such traits.
Another approach is to collect phenotypes in commercial farms and utilize that information to estimate the breeding value of selection candidates in the nucleus.
And finally, health-related phenotypes can be collected in special experimental facilities, followed by the use of the information generated in that facility for selection in the nucleus.
Phenotypic data on disease-related traits that is generated with these approaches can be used for selection using one of three approaches, which have been described in previous models:
- To estimate EBV related to disease on selection candidates using the collected phenotypes.
By identifying major genes or genetic markers that are are associated with disease and using these to genotype selection candidates
By using genomic prediction based on high-density SNP genotypes.
There are several approaches that can be used to overcome at least some of these challenges.
One is to develop disease indicator traits that can be collected on selection candidates without exposure to disease and I will give a couple of examples of such traits.
Another approach is to collect phenotypes in commercial farms and utilize that information to estimate the breeding value of selection candidates in the nucleus.
And finally, health-related phenotypes can be collected in special experimental facilities, followed by the use of the information generated in that facility for selection in the nucleus.
Phenotypic data on disease-related traits that is generated with these approaches can be used for selection using one of three approaches, which have been described in previous models:
- To estimate EBV related to disease on selection candidates using the collected phenotypes.
By identifying major genes or genetic markers that are are associated with disease and using these to genotype selection candidates
By using genomic prediction based on high-density SNP genotypes.
Another strategy is to obtain disease related phenotypes in experimental facilities. This approach has been used by Hy-Line International to improve survival to Marek’s disease in layer chickens. In their system, each generation, progeny of male selection candidates are submitted to the latest most virulent field isolate of Marek’s disease and the % mortality is recorded and to select roosters with greater survival to Marek’s disease. In this system, phenotypes recorded on relatives are fed back to the nucleus and used to estimate breeding values of selection candidates for disease-related traits.
Another approach, is to genotype individuals that are challenged using a high-density SNP chip and then use the resulting SNP effect estimates for genomic prediction and genomic selection in the nucleus, similar to the approach of genomic selection using commercial crossbred data collected in the field. Again, the advantage of the genomics approach is that the individuals that are challenged do not need to be close relatives of the selection candidates and pedigree is not needed
This is the approach that was used in the PRRS Host Genomics Consortium, which we will discuss in Module 4, in which large numbers of weaner pigs from commercial lines were challenged with the PRRS virus to identify genomic regions and genetic markers that are associated with resistance to PRRS virus infection, which could then be used for marker-assisted or genomic selection in nucleus breeding programs.
One strategy to deal with the polygenic nature of traits in selection using genomics is genomic prediction and genomic selection, which I introduced in an earlier module.
Applied to selection for host response to PRRS, the idea would be to utilize the data that has been collected on host response in the PHGC trials to estimate the effects of large numbers of SNPs across the genome, including SNPs that are not significantly associated with host response in a GWAS, and then use these estimates to predict the genetic value of new individuals that have been genotyped for host response.
To evaluate the potential of this approach, we used the PHGC data. Where we used the NVLS phenotypes and SNP genotypes to estimate the SNP effects on host response and used these estimates to predict the genetic value of animals in the KS06 trials. To determine how good the resulting predictions were, we then correlated the predictions for host response with the phenotypes for host response on these same animals. So, here we trained on host response to one PRRS strain and then evaluated how good predictions were for response to another PRRS strain, which would reflect what you would need to be able to do in practice, with the wide range of PRRS strains that are present in the field.
We then also did this the other way around, ie. Train on the KS06 trial data and validate the accuracy of the predictions based on the NVSL data.
These bars show the accuracy of the genomic predictions when training on KS06 and validating on NVSL. The blue bar shows the accuracy when using all SNPs across the genome for prediction, the green bar the accuracy when the prediction was based only on the WUR genotype, and the purple bar the accuracy obtained based on the rest of the genome, beyond the WUR region. Results show that we obtained reasonable accuracies of prediction when using the whole genome but that most of that accuracy came from the WUR genotype.
When this was repeated for training on the NVSL data and validating on the KS06 data, similar results were obtained, although the accuracies in general were a bit lower.
Nevertheless, this showed some promise to not only use the WUR genotype for genetic selection for host response to PRRS but also the rest of the genome.