Presentation made to the staff of Keygene, NV, in Wageningen, The Netherlands. (I don't know what the problem is with the template here. It looks fine if you use a dark background.)

1. J. B. Cole
Animal Improvement Programs Laboratory
Agricultural Research Service, USDA
Beltsville, MD 20705-2350, USA
john.cole@ars.usda.gov
Genomic
selec+on
and
systems
biology
–
lessons
from
dairy
ca5le
breeding

2. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(2)
Cole
Dairy Cattle
 9 million cows in US
 Attempt to have a calf born every year
 Replaced after 2 or 3 years of milking
 Bred using artificial insemination
 Popular bulls have 10,000+ progeny
 Cows can have many progeny though
superovulation and embryo transfer

3. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(3)
Cole
Embryo transferred to
recipient"
Parents selected"
Dam inseminated"
Bull born"
Semen collected (1 y)"
Daughters born (9 m later) "
Daughters have calves (2 y later)"
Bull receives
progeny test"
(5 y)"
Genomic Test"
Lifecycle of bull

4. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(4)
Cole
Phenotypes recorded
 Monthly recording
 Milk, fat, and protein yields
 Somatic cell count (udder health)
 Visual appraisal for type traits
 Breed associations record pedigree
 Calving difficulty and stillbirth

5. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(5)
Cole
Available data
Type of Data Number of Records
Cows with lactation data 28,394,976
Lactations 68,373,863
Individual test days 508,574,532
Dystocia records 20,770,758
Animals in pedigree file 58,893,009
Genotyped bulls 105,654
Genotyped cows 276,173

6. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(6)
Cole
0
50000
100000
150000
200000
250000
300000
1004 1008 1012 1104 1108 1112 1204 1208 1212 1304
Bulls Cows
Cole"
Many animals have been genotyped
Evaluation Date (YYMM)"
Genotypes"
381,827 genotyped animals"
Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(6)

7. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(7)
Cole
How does genetic selection work?
 ΔG = genetic gain each year
 reliability = how certain we are about our estimate of
an animal’s genetic merit (genomics can é)
 selection intensity = how “picky” we are when making
mating decisions (management can é)
 genetic variance = variation in the population due to
genetics (we can’t really change this)
 generation interval = time between generations
(genomics can ê)

8. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(8)
Cole
8"
Calculation of genomic evaluations
 Deregressed PTA derived from traditional
evaluations of predictor animals
 Allele substitution effects estimated for
45,188 SNP
 Polygenic effect estimated for genetic
variation not captured by SNP
 Selection index combination of genomic
and traditional not included in genomic

9. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(9)
Cole
Many chips are available
HD"
50KV2 "
LD "
GGP HD!
 BovineSNP50
 Version 1 54,001 SNP
 Version 2 54,609 SNP
 45,188 used in evaluations
 High-density (HD)
 777,962 SNP
 Only 50K SNP used,
 Low-density (LD)
 6,909 SNP
 Geneseek Genomic Profiler & GGP-HD

10. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(10)
Cole
What is a SNP genotype worth?
For the protein yield
(h2=0.30), the SNP
genotype provides
information
equivalent to an
additional 34
daughters"
Pedigree is equivalent to information on about 7 daughters "

11. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(11)
Cole
And for daughter pregnancy rate (h2=0.04), SNP = 131 daughters"
What is a SNP genotype worth?"

12. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(12)
Cole
High density SNP chip
 Currently only 50K subset of SNP used
 Some increase in accuracy from better
tracking of QTL possible
 Realized gains have been small
 Potential for across-breed evaluations
 Requires few new HD genotypes once
adequate base for imputation developed

13. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(13)
Cole
Low density SNP chip
 6909 SNP mostly from SNP50 chip
 Evenly spaced across 30 chromosomes
 Addresses performance issues with 3K
while providing low-cost genotyping
 Provides over 98% accuracy imputing
50K genotypes

14. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(14)
Cole
Parentage validation and discovery
 Parent-progeny conflicts detected
 Animal checked against all other genotypes
 Reported to breeds and requesters
 Correct sire usually detected
 Maternal grandsire checking
 SNP at a time checking
 Haplotype checking more accurate
 Breeds moving to accept SNP in place of
microsatellites

15. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(15)
Cole
Imputation
 Based on splitting the genotype into
individual chromosomes
 Missing SNP assigned by tracking
inheritance from ancestors and
descendants
 Imputed dams increase predictor
population
 3K, LD, & 50K genotypes merged by
imputing SNP not on LD or 3K

16. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(16)
Cole
Genotypes and haplotypes
 Genotypes indicate how many copies of
each allele were inherited
 Haplotypes indicate which alleles are on
which chromosome
 Observed genotypes partitioned into the
two unknown haplotypes
 Pedigree haplotyping uses relatives
 Population haplotyping finds matching
allele patterns

17. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(17)
Cole
Haplotyping program – findhap.f90
 Begin with population haplotyping
 Divide chromosomes into segments,
~250 to 75 SNP / segment
 List haplotypes by genotype match
 Similar to fastPhase, IMPUTE
 End with pedigree haplotyping
 Detect crossover, fix noninheritance
 Impute nongenotyped ancestors

18. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(18)
Cole
O-Style Haplotypes Chromosome 15

19. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(19)
Cole
We’re working on new tools

20. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(20)
Cole
Recessive defect discovery
 Check for homozygous haplotypes
 7 to 90 expected but none observed
 5 of top 11 are potentially lethal
 936 to 52,449 carrier sire-by-carrier
MGS fertility records
 3.1% to 3.7% lower conception rates
 Some slightly higher stillbirth rates
 Confirmed Brachyspina same way

21. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(21)
Cole
Impact on producers
 Young-bull evaluations with accuracy of
early 1st­crop evaluations
 AI organizations marketing genomically
evaluated 2-year-olds
 Rate of genetic improvement may
increase by up to 50%
 Studs reducing progeny-test programs

22. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(22)
Cole
Why genomics works in dairy
 Extensive historical data available
 Well-developed genetic evaluation program
 Widespread use of AI sires
 Progeny test programs
 High-valued animals, worth the cost of
genotyping
 Long generation interval which can be reduced
substantially by genomics

23. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(23)
Cole
Where do we go from here?
 We found a few QTL
 Most traits show infinitessimal
inheritance
 Dominance effects also are small
 What about epistasis?
 Systems biology – gene/protein/
transcription factor networks

24. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(24)
Cole
24"
We confirmed known QTL
Cole, J.B. et al. 2009. Distribution and location of genetic effects for dairy traits. ICAR Tech Ser. 13:355–360."

25. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(25)
Cole
Gene set enrichment analysis-SNP
Gene
pathways (G)"
GWAS results"
Score increase is proportional to SNP test
statistic"
Nominal p-value corrected for multiple
testing"
Pathways with
moderate
effects"
Holden et al., 2008 (Bioinformatics 89:1669-1683. doi:10.2527/jas.2010-3681)"
SNP ranked by
significance
(L)"
SNP in
pathway genes
(S)"
Score
increases for
each Li in S"
Permutation
test and FDR"
Includes all SNP, S, that are included in L"
The more SNP in S
that appear near the
top of L, the higher the
Enrichment Score"

26. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(26)
Cole
Association weight matrix
 Find gene coexpression networks (Fortes et al., 2010)
 Select SNP by significance, correlation, dist’n, etc.
− Favor intragenic SNP significant across traits
 Construct weight matrix
− Rows are SNP, columns are traits cols
− Cells are normalized z-score of the additive
effect of ith SNP on jth trait
 Significant correlations are identified using PCIT
(Reverter and Chan, 2008) and visualized
− Cells randomly permuted as control

27. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(27)
Cole
Can we identify regulatory networks?
Fortes et al., 2011 (J. Animal Sci. 89:1669-1683. doi:10.2527/jas.2010-3681)"
Candidate
genes and
pathways that
affect age at
puberty
common to
both breeds"

28. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(28)
Cole
Network analysis
Fortes et al., 2011 (J. Animal Sci. 89:1669-1683. doi:10.2527/jas.2010-3681)"
Gene network – the
red center identifies
highly connected nodes."
Subnetwork of interacting
transcription factors from
the puberty network."
Subnetwork of
interacting
transcription factors
from a collection of
mouse and human
data. (Validation
step.)"

29. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(29)
Cole
Enriched pathways
Fortes et al., 2011 (J. Animal Sci. 89:1669-1683. doi:10.2527/jas.2010-3681)"

30. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(30)
Cole
Transcription factor network
Fortes et al., 2011 (J. Animal Sci. 89:1669-1683. doi:10.2527/jas.2010-3681)"
Yellow genes were
submitted to
database.
Other nodes were
mined from
FunCoup.
Red: protein-
protein interaction
Blue: mRNA
coexpression"

31. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(31)
Cole
How do we rank allele effects?
 GSEA and AWM require that we order
SNP on some criterion
 p-values (actual or nominal)
 q-values (false discovery rate)
 Not all models provide p-values
 Allele substitution effects (not so good)
 Scaled substitution effects (better)
 It’s not clear (to me) which is best

32. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(32)
Cole
Aren’t P-values easy?
 Single SNP, fixed-effects model
 Inflation of error variances
 Spurious associations
 e.g., Plink
 Multiple SNP, mixed-effects model
 Accounts for population structure
 e.g., TASSEL, GoldenHelix SVS

33. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(33)
Cole
A recent example from dairy
 Extreme birth weights are associated
with increased risk of stillbirth and
calving difficulty
 Birth weights are not measured on most
dairy farms in the US
 With German colleagues, we developed a
predictor based on traits we do measure

34. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(34)
Cole
GWAS for birth weight PTA
h"
Cole et al.(2013), unpublished data"

35. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(35)
Cole
KEGG pathways for birth weight
What does
regulation of the
actin
cytoskeleton
have to do with
birth weight in
cattle?
That is, do
these results
make sense?"
Maybe…these
pathways may
be involved in
establishment &
maintenance of
pregnancy, as
well as
coordination of
growth and
development.
"
Cole et al.(2013), unpublished data"

36. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(36)
Cole
A new project
 The Brown Swiss, Holstein, and Jersey
breeds experience dystocia at different
rates
 We are applying the AWM method of
Fortes et al. to these data
 The goal is to identify gene networks…
 Common to all breeds
 Different by breed

37. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(37)
Cole
We have divergent populations
Cole et al., 2005 (J. Dairy Sci. 88(4):1529–1539)"

38. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(38)
Cole
Challenges
 Annotation
 This is a mess in the cow
 The reference assembly may not be
representative of all taurine cows
 Validation
 Doing functional genomics with large
mammals is expensive – who pays?
 When have we proven something?

39. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(39)
Cole
Conclusions
 We’re not going to find big QTL for most
traits
 We may identify gene networks affecting
complex phenotypes
 We’re learning how much we don’t know
about functional genomics in the cow
 Validation remains a problem

40. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(40)
Cole
Partners
 Illumina
 Marylinn Munson
 Cindy Lawley
 Christian Haudenschild
 BARC
 Curt Van Tassell
 Lakshmi Matukumalli
 Tad Sonstegard
 Missouri
 Jerry Taylor
 Bob Schnabel
 Stephanie McKay
 Alberta
 Steve Moore
 USMARC – Clay Center
 Tim Smith
 Mark Allan
iBMAC Consortium" Funding Agencies"
 USDA/NRI/CSREES
 2006-35616-16697
 2006-35205-16888
 2006-35205-16701
 USDA/ARS
 1265-31000-081D
 1265-31000-090D
 5438-31000-073D
 Merial
 Stewart Bauck
 NAAB
 Godon Doak
 ABS Global
 Accelerated Genetics
 Alta Genetics
 CRI/Genex
 Select Sires
 Semex Alliance
 Taurus Service

41. Keygene
N.V.,
Wageningen,
The
Netherlands,
28
May
2013
(41)
Cole
Questions?
http://gigaom.com/2012/05/31/t-mobile-pits-its-math-against-verizons-the-loser-common-sense/shutterstock_76826245/"