GenBio2 - Lesson 1 - Introduction to Genetics.pptx
Heat tolerance, real-life genomics and GxE issues
1. Heat tolerance, real-life genomics
and GxE issues
Ignacy Misztal
University of Georgia
ILRI – LiveGene Seminar, Addis Ababa, 3
February 2016
2. Research in Misztal’s lab at UGA
• 8-15 people (Postdocs + grad students+ visitors)
• Focus on practical results
• Only provider of genetic (genomic) evaluation software in US
– Holsteins Assoc
– Angus Assoc +_
– Major pig companies
– Cobb (broiler chicken)
• High scientific output: 10 papers/year over 10 years
3. Current Projects
• Genomic selection (single step methodology)
• Genetics of:
– heat stress
– mortality
– competition (social interaction)
• Issues in genetic evaluation in dairy, beef, pigs and
chicken (also sheep and fish)
5. Selection and environment
Almost all dairy bulls selected in mild or cold environments
Assume genetic relationship between mild and hot
performance antagonistic
Is selection indirectly against heat tolerance?
Can one select for heat tolerance?
6. How to proceed with genetic studies in heat
tolerance?
What data to record on heat tolerance?
Recording of rectal temperatures or respiration rate
expensive
What models to use?
8. Assumptions
Production
f(heat index)
cow 2
cow 3
cow 1
Breeding value: BV = a + f(THI)*v
a – regular breeding value v – heat-tolerance breeding value
f(THI) – function of temperature humidity index
9. Studies
Ravagnolo et al., 2000ab
and 2002ab
Milk test days from
Georgia or FL
NR45-90 from FL
THI from weather stations
13. Genetic component for heat stress present
Genetic correlations between regular and heat
stress effects -0.40
14. National genetic evaluation of
Holsteins for heat tolerance
Can one identify heat-tolerant sires in Holsteins?
What are they?
Bohmanova et al. (2005 and 2006)
National US data
15. Differences between most 100 and least 100
heat tolerant sires
Milk -1100kg
Fat% +0.2%
Prot% +0.1%
Dairy Form -1.4
Udder +0.7
Longevity +0.90
Fertility +1.6
Total Index +36
16. Comments
Currently selection against heat tolerance in fluid markets
Heat tolerant cows may also be stress tolerant in general
Welfare perspective (comfort, mortality, susceptibility to diseases)
17. Does heat stress vary by parity?
Ignacio AGUILAR*1,2, Ignacy MISZTAL1 and Shogo TSURUTA1
1 Animal and Dairy Science Department, University of Georgia
2 Instituto Nacional de Investigación Agropecuaria, Las Brujas, Uruguay
18. Genetic trends of daily milk yield for 3
parities – regular effect
First Second Third
19. Genetic trends for heat stress effect at 5.5o
C over the threshold
First Second Third
20. ssGBLUP for Heat Stress in Holsteins
(Aguilar, 2011)
• Multiple-Trait Test-Day model, heat stress as
random regression
• ~ 90 millions records, ~ 9 millions pedigrees
• ~ 3,800 genotyped bulls
• Computing time
• Complete evaluation ~ 16 hRegular effect -first parity Heat stress effect – first parity
21. Heat stress in Days Open (Oseni et al, 2003)
Seasonal trends for DO in Georgia
24. Heat stress-summary
• Can evaluate Holsteins for heat stress
– Negative correlations with production
• Largest effect in later parities
– Poor survival in hot climates
• AI companies not much interested in heat
tolerance – small market
• Hot areas - revolution with sexed semen
25. Genetics of growth in pigs under different
heat loads (Zumbach et al., 2007)
• Pigs in NC or TX exposed to heat
stress
• Heat stress affect growth
• How to model heat stress for
growth?
26. pigs with wings - 27
Heat stress and reproductive capacity of
sows in Spain (Bloemhof et al.)
Two lines: Adapted (I) and Performance (D)
10.5
10.7
10.9
11.1
11.3
11.5
11.7
11.9
12.1
12.3
12.5
1 2 3 4 5 6 7 8 9 10 11 12
Month of insemination
Totalnumberpigletsborn
I
D
27. Focus on research in animal breeding
• Quantitative genetics and BLUP (up to 1990)
• QTL and marker assisted selection (up to 2007)
• Genomic selection (now)
28. SNP chips and genetic evaluation
• Meuwissen et al. paper
• Early claims of prediction for life with 1000 genotypes
• Initial Experiences
– Encouraging results if > 2k genotypes and 50k chips
– No long-range prediction under selection
• Multistep methods to use phenotypes of ungenotyped
animals
29. Multistep genomic evaluation in dairy
(VanRaden, 2008)
BLUP
DD or deregressed
proofs
BayesX or
GBLUP &
Index hard to calculate
Index &
y
GEBV
DGV*w1
PA *w2
PI *w3
Deregression good if only if high accuracy animals
Complex
30. National Swine Improvement Federation Symposium, Dec. 2008 (31) Paul VanRaden
2008
Value of Genotyping More Bulls
Bulls R2 for Net Merit
Predictor Predicted PA Genomic Gain
1151 251 8 12 4
2130 261 8 17 9
2609 510 8 21 13
3576 1759 11 28 17
31. Typical result of assuming different SNP
distributions
Verbyla et al, 2009
59
58
60
60
59
59
59
Similar results with genomic relationships and SNP effects
32. Single-step evaluation (Misztal et al., 2009)
Merge pedigree relationship matrix (A) and genomic
relationships (G) into a joint matrix (H) and use in BLUP
H=A+”modifications due to genomic information”
Single-step GBLUP = ssGBLUP
33. Inverse of matrix that combines pedigree
and genomic relationships
-1 -1
-1 -1
22
0 0
H = A +
0 G - A
Aguilar et al., 2010
Christensen and Lund, 2010
Boemcke et al., 2010
GEBVyoung
=w1
PA + w2
DGV-w3
PI
34. Single-Step / Unified Method
• Simple and fast
• Any model
• Usually more accurate than other methods
• Now industry standard
• Lots of research at UGA
– Extensive quality control
– Approximation of accuracies
– Origin of convergence problems (except in broilers)
– GWAS
– …
35.
36. • Large research interest in GWAS
• Limitations of classical and Bayesian methods
Can ssGBLUP be used for GWAS?
ssGBLUP for GWAS
2/9/2016 PAG 2012 Meeting
37. Correlations between QTLs and clusters of
SNP effects - simulation
45
50
55
60
65
70
75
80
85
1 2 4 8 16 32
SNP cluster size
BayesBSingle SNP
ssGWAS/1
Wang et al., 2012
38. Comparison of Three Methods (broilers)
ssGBLUP
Iterations on SNP (it3)
Classical GWAS
BayesB
0.8%
2.5%
23%
Wang et al., 2012
39. Plots and accuracies in Zhang et al. (2010)
Therefore, the proposed mutation rate gave an expected heterozy- between QTL. The simulated additive genetic variance of each
simulation
BayesB
Acc 0.83
Weighted RRGBLUP
Acc 0.75
40. GWAS findings
• Better estimates of QTL effects with a cluster of adjacent SNP –
why and what size?
• BayesB gives inflating readings but misses a lot
– Multiple SNP solutions for same GEBV? Singularity problem?
– Only informal reporting. Why?
• Little or no improvement in accuracy of GEBV with large
number of genotypes
– Are many large SNP readings artifacts?
– Impacts on papers in high-impact journals
2/9/2016 PAG 2012 Meeting
41. Result with GWA
• Best correlations with QTL with cluster of SNP
• Manhattan plots: Few or no common peaks between
breeds or lines
• Peaks smaller with larger number of genotypes
• Weird behavior of BayesB
– “con-artist method”
42. Size problem
• Number of genotyped animals
– About 1 million Holsteins
– Close to 200,000 Angus
• Big computations – what to do
• APY methodology – based on large haplotype sizes
in farm populations
43. APY with Holsteins (Fragomeni et al., 2015)
G needed G-1
APY inverse
Regular inverse
Correlations of GEBV
with regular inverse
23k bulls
as proven
17k cows as
proven
> 0.99
> 0.99
20k random
animals as
proven
> 0.99
44. Costs with 570k genotyped animals
(Masuda et al., 2016)
• 10 M US Holsteins for type
• Computing time for APY inverse 2h
– Would be one month with direct inverse
• Genomic single-step sGBLUP evaluation ~ 2 times
more expensive than BLUP
45. Why APY works?
• Limited dimensionality of genomic information
– Limited number of independent SNP clusters
– Limited number of independent chromosomal segments (Me)
• Dimensionality ≈10000 for cattle, 4000 for swine, ≈3000
for broilers (Pocrnic et al., 2016)
46. Impact of reduced dimensionality
• Lmited resolution of GWAS ≈1/(2Ne) Morgans
– About 0.5 to 5 Mbases farm animals
– About 2-5 Kbases in humans (Li et al., 2012)
• Seems impossible to find causative SNP with GWAS in farm
animals (e.g., Veerkamp et al., 2015)
• Can use APY with sequence data if causative SNP identified
and qualified by other means
– (Brøndum et al., 2014)
– Do causative SNP with large effect exist?
47.
48. Genetic selection in chicken
• 60+ generations of selection (Eitan and Soller, 2002 & 2012)
• Great progress for production traits
• Problems with reproduction and disease resistance
• Problems solved by management (environmental changes),
• not genetics
• No reduction of genetic variance
• Important genes become fixed and selection follows with new genes
49. Selection as optimization
• Some traits improve, some traits deteriorate
• Hard to detect deterioration if low heritability and small data sets
• Need improved management to compensate
• Why efficiency differs by commercial species?
• Pigs and chicken - environment strongly controlled
• Dairy - environment controlled
• Beef – environment little controlled
50. Can large QTL exist despite selection?
• Genetics and genomics of mortality in US
Holsteins
• (Tokuhisa et al, 2014; Tsuruta et al., 2014)
• 6M records, SNP50k genotypes of 35k bulls
53. African connections at UGA
Graduate students
Saudi Oseni – Dept head in Nigeria
Emannuel Lutaaya –UgandaNamibia
Dr. Boly – Burkina Faso
ILRI (Dempfle, McClintock, Gibson)
South Africa – universities
UNDP-FAO
54.
55. Renumbering
RENUMF90
BLUP in memory
BLUPF90
Variance component estimation
REMLF90 AIREMLF90
GIBBS2F90 THRGIBBS2F90
BLUP – iteration on data
BLUP90IODF
CBLUP90IOD
Approximate accuracies
ACCF90
Sample analysis
POSTGIBBSF90
Computing of extra matrices
PreGSF90
GEBV to SNP conversions
GWAS
PostGSF90Predictions via SNP
PredGSF90
56. Summary
• Long time from first idea to practical results
• Commercial twists
• Genomic selection now mature
• Lots of exaggeration and misinformation
• Need large data sets across species
• Selection as optimization - winners known but what are the losers
• UGA has tools and focuses on deliverables
60. Modeling issues in genomic selection
• Many quantitative issues
– Nucleus - commercial performance
– Mortality
– Heat stress
– New problems (e.g., survival of low-weight piglets)
• Need to have:
– Good selection index
– Good model
– Troubleshooting skills