J. Dairy Sci.
101:1–14
https://doi.org/10.3168/jds.2017-12982
© American Dairy Science Association®, 2018.
Association analysis for udder index and milking speed with imputed
whole-genome sequence variants in Nordic Holstein cattle
Júlia Gazzoni Jardim,*† Bernt Guldbrandtsen,* Mogens Sandø Lund,* and Goutam Sahana*1
*Department of Molecular Biology and Genetics, Center for Quantitative Genetics and Genomics, Aarhus University, 8830 Tjele, Denmark
†Laboratory of Reproduction and Animal Breeding, State University of North Fluminense Darcy Ribeiro, Av. Alberto Lamego,
2000 Parque California, Campos dos Goytacazes, RJ, 28013-602, Brazil
ABSTRACT
Genome-wide association testing facilitates the iden-
tification of genetic variants associated with complex
traits. Mapping genes that promote genetic resis-
tance to mastitis could reduce the cost of antibiotic
use and enhance animal welfare and milk production
by improving outcomes of breeding for udder health.
Using imputed whole-genome sequence variants, we
carried out association studies for 2 traits related to
udder health, udder index, and milking speed in Nor-
dic Holstein cattle. A total of 4,921 bulls genotyped
with the BovineSNP50 BeadChip array were imputed
to high-density genotypes (Illumina BovineHD Bead-
Chip, Illumina, San Diego, CA) and, subsequently,
to whole-genome sequence variants. An association
analysis was carried out using a linear mixed model.
Phenotypes used in the association analyses were der-
egressed breeding values. Multitrait meta-analysis was
carried out for these 2 traits. We identified 10 and 8
chromosomes harboring markers that were significantly
associated with udder index and milking speed, respec-
tively. Strongest association signals were observed on
chromosome 20 for udder index and chromosome 19
for milking speed. Multitrait meta-analysis identified
13 chromosomes harboring associated markers for the
combination of udder index and milking speed. The
associated region on chromosome 20 overlapped with
earlier reported quantitative trait loci for similar traits
in other cattle populations. Moreover, this region was
located close to the FYB gene, which is involved in
platelet activation and controls IL-2 expression. FYB is
a strong candidate gene for udder health and worthy of
further investigation.
Key words: udder conformation, milking speed,
genome-wide association study
Received April 4, 2017.
Accepted October 30, 2017.
1
Corresponding author: goutam.sahana@mbg.au.dk
1
INTRODUCTION
Animal welfare and production costs have spurred
interest in breeding for improved functional traits
(Boettcher, 2005). A cow’s milking ability is defined
by the milking speed, average milk flow rate, maximum
milk flow rate, and total milking time. Milking ability
influences the working time required for milking and
is genetically correlated with mastitis incidence. Heri-
tability values for milking ability traits are medium to
high, with values of 0.42 for average milk flow, 0.56 for
maximum milk flow, and 0.38 for milking time in Ger-
man Holsteins (Gade et al., 2007). Genetic correlations
of these 3 traits with SCS were 0.35, 0.38, and 0.24,
respectively (Gade et al., 2007). Higher milk flow and
shorter milking time were associated with increased sus-
ceptibility to mastitis. Sewalem et al. (2011) reported a
heritability of 0.14 for milking speed in Canadian Hol-
stein. The genetic correlation between milking speed
and SCS was 0.25 (0.41 and 0.25 for first and second
lactations, respectively; Boettcher et al., 1998). Taken
together, these results support an association between
faster milking and higher SCS.
Mastitis could lead to udder injury, longer milking
time, incomplete udder draining, and increased SCS,
whereas complete draining will help to prevent clinical
mastitis (Rupp and Boichard, 2003). Udder conforma-
tion (teat placement, length of fore udder, and udder
depth) has an effect on mammary gland health and
has been used as an indicator trait to enhance selec-
tion for mastitis resistance by inclusion in a selection
index (Lund et al., 1994; Carlstrom et al., 2013). Due
to low heritability and lack of data, however, selection
for udder health traits is generally more difficult than
for production traits in dairy cattle.
The SNP marker panels are routinely used for ge-
nomic prediction in dairy cattle and have almost dou-
bled the rate of genetic gain (Hayes et al., 2013). Faster
genetic gain in low heritability traits can be achieved
through genomic selection augmented by QTL informa-
tion (Boichard et al., 2016). Weights can be defined
2 JARDIM ET AL.
for individual SNP based on their known associations
with phenotypes of interest. This process could increase
the accuracy of genomic prediction (Brøndum et al.,
2015; van den Berg et al., 2016). Due to the availability
of SNP arrays, genome-wide association studies iden-
tifying associations between common genetic variants
with phenotypic differences in a trait have been exten-
sively used in dairy cattle (Sahana et al., 2014). Recent
QTL mapping studies have identified QTL associated
with mammary gland-related phenotypes in several
cattle populations (Cole et al., 2011; Kadri et al., 2015;
Pausch et al., 2016). The QTL for milking speed, ud-
der morphometric, mastitis traits, and milk yield traits
were mapped by Gray et al. (2012).
Farmers in Denmark, Finland, and Sweden frequently
use equipment that automatically measures milk yield
and milking duration on every test day or during every
milking. These high-quality data are collected for use in
management and could be used to improve the genetic
evaluation for milking ability traits. In this study, we
performed association analyses for udder index and
milking speed with imputed whole-genome sequence
(WGS) variants in Nordic Holsteins. Subsequent mul-
titrait meta-analyses were used to verify whether de-
tected QTL were simultaneously associated with both
traits. Finally, we compared the association results in
the present study with previously reported association
results for mastitis resistance in the literature.
MATERIALS AND METHODS
Animals and Phenotypes
Bulls from Nordic Holstein cattle with deregressed
estimated breeding values (DRP) for udder index
and milking speed were used in the analyses. Higher
breeding values indicate better udder conformation
and higher milking speed. The udder index describes
the genetic potential for udder conformation and is
a linear combination of subindices for fore udder at-
tachment, rear udder height, rear udder width, udder
cleft/support, udder depth, teat length, teat thick-
ness, teat placement (front), teat placement (back),
and udder balance. For breeding value predictions,
the Nordic Cattle Genetic Evaluation (www​.nordicebv​
.info) traditionally has used farmers’ recordings of
subjective scores of milking speed of individual cows
compared with their herd mates. Since August 2014,
breeding values for milking speed have been based on
both farmers’ scores and automatic measurements of
milking speed from automatic milking systems and
conventional milking parlors (http://​www​.nordicebv​
.info/​wp​-content/​uploads/​2015/​04/​Improved​-breeding​
-value​-for​-milkability​.pdf). For details on the recording
Journal of Dairy Science Vol. 101 No. 3, 2018
Table 1. Descriptive statistics of deregressed estimated breeding value
(DRP) and reliability of udder index and milking speed in Nordic
Holstein cattle
Udder index Milking speed
Summary DRP Reliability   DRP Reliability
Number 4,921 4,921   4,832 4,832
Mean 94.45 0.77   97.32 0.77
SD 14.39 0.08   15.33 0.13
Minimum 54.80 0.40   55.70 0.33
Maximum 133.50 0.99   138.80 0.99
of phenotypes and breeding value estimates, see www​
.nordicebv​.info. The DRP were derived from Interbull
genetic evaluations on the Nordic scale based on the
EBV and effective daughter contributions (Fikse and
Banos, 2001). Descriptive statistics of DRP and reli-
abilities for these 2 indices are listed in Table 1. Histo-
grams of DRP and their reliabilities for the 2 indices
are presented in Figures 1 and 2. The correlations
between Nordic total merit index and udder index and
milking speed are 0.17 and 0.02, respectively (http://​
www​ . nordicebv​ . info/​ w p​ - content/​ u ploads/​ 2 017/​ 0 3/​
NAV​-routine​-genetic​-evaluation​-122016​_FINAL​.pdf).
The correlation between the DRP of udder index and
milking speed for the bulls used in the analysis was
0.13.
SNP Genotypes and Imputation to WGS Level
The association study was carried out by using im-
puted WGS data, as previously described by Iso-Touru
et al. (2016) and Wu et al. (2016). All 4,921 bulls were
genotyped with the Illumina BovineSNP50 BeadChip
(54k) ver. 1 or 2 (Illumina, San Diego, CA). The 54k
genotypes were imputed to WGS variants by using a
2-step approach. First, all animals were imputed to the
high-density (HD) level by using a multibreed refer-
ence of 3,383 animals (1,222 Holsteins, 1,326 Nordic
Red Dairy Cattle, and 835 Danish Jerseys), which had
been genotyped with the Illumina BovineHD Bead-
Chip. Subsequently, these imputed HD genotypes were
imputed to the WGS level by using a multibreed refer-
ence of 1,228 animals from Run4 of the 1,000 Bull Ge-
nomes Project (1,148 cattle, including 288 individuals
from the global Holstein-Friesian population, 56 Nordic
Red Dairy Cattle, 61 Jerseys, and 743 cattle from other
breeds; Daetwyler et al., 2014) and additional data
from Aarhus University (80 individuals, including 23
Holsteins, 30 Nordic Red Dairy Cattle, and 27 Danish
Jerseys).
Different variant calling pipelines were used for the
1000 Bull Genome Project data and the in-house Nordic
data at Aarhus University. The WGS data at Aarhus
 ASSOCIATION STUDY FOR MILKING ABILITY TRAITS 3

Figure 1. Distribution of deregressed breeding values (DRP) and reliabilities for udder index in Nordic Holstein cattle.

University were analyzed as described by Brøndum et al.
(2014), whereas the same for 1000 Bull Genome Project
was described by Daetwyler et al. (2014) and detailed
guidelines are available at http://​www​.1000bullgenomes​
.com. Data from both sources were available as VCF
files. The data from 2 sources were combined using
Picard MergeVCF (http://​broadinstitute​.github​.io/​
picard/​). As the 1000 Bull Genomes Project shares
data after variant calling, some markers were not called
for all animals in the combined data set. To avoid large
gaps of missing markers in the data set, only markers
that were called in both the Nordic and the 1000 Bull
Genomes Project data sets were kept. For positions
containing both a SNP and an INDEL, the INDEL was
deleted as the imputation methods rely on unambigu-
ous sequences of variants. Positions with disagreements
between alleles for sequence and HD data were also
deleted. Reference genotype probability data was run
through BEAGLE (Browning and Browning, 2007)
and all markers with an R2 value (imputation quality

Journal of Dairy Science Vol. 101 No. 3, 2018

4 JARDIM ET AL.

at imputed marker) below 0.9 were removed from the
original sequence data. This was done to remove uncer-
tain marker genotypes that might have adverse effects
on the imputation procedures.
Imputation from 54k to HD genotypes and imputa-
tion to the WGS level were undertaken with IMPUTE2
v2.3.1 (Howie et al., 2011) and Minimac2 (Fuchsberger
et al., 2015), respectively. The imputation to WGS
was done in chunks of 5 Mb with the length of buffer
region of 0.25 Mb on either side. A total of 22,751,039
biallelic variants were present in the imputed sequence
data. After excluding SNP with minor allele frequency
below 1% or with large deviation from Hardy-Weinberg
proportions (P < 1.0−6), we used 15,355,402 SNP on
29 autosomes in Nordic Holstein cattle for associa-
tion analyses. The average accuracy (R2-values from
Minimac2) was 0.85 for across breed imputation and
the distribution of imputation accuracy with respect
to minor allele frequency was published earlier (Wu et
al., 2016).

Figure 2. Distribution of deregressed breeding values (DRP) and reliabilities for milking speed in Nordic Holstein cattle.

Journal of Dairy Science Vol. 101 No. 3, 2018

 ASSOCIATION STUDY FOR MILKING ABILITY TRAITS
Association Analysis with EMMAX
Association analyses were carried out for each im-
puted sequence variant by using a 2-step variance
component-based approach accounting for population
stratification, as implemented in the EMMAX software
tool (Kang et al., 2010). Details about the model are
given by (Kang et al., 2008, 2010).
In the first step, polygenic and error variances were
estimated by using the model:
y = 1µ + Za + e, puted sequence variants, following Bolormaa et al.
where y is a vector of phenotypes (DRP), 1 is a vector
of ones, μ is the overall mean, a is a vector of random
polygenic effects, which are multivariate normally dis-
tributed a ~ N (0, Gσa2 ), G is the genomic relationship
matrix (GRM) built from SNP genotypes on the HD
panel using EMMAX software, σa2 is the additive ge-
netic variance, Z is an incidence matrix relating pheno-
types to corresponding random polygenic effects, e is a
vector of random individual error terms, assumed to
have a multivariate normal distribution e ~ N (0, Iσe2 ), I
is an identity matrix, and σe2 is the error variance. Ele-
ments in the GRM were twice the estimated coefficients
of coancestry between each pair of individuals, where
the kinship matrix was inferred by an identical-by-state
allele-sharing matrix. If a model fits a candidate SNP
as fixed effect for testing association and at the same
time includes it in the GRM, this may lead to loss of
power due to double fitting of the candidate SNP both
as a fixed regression and a random effect as part of the
GRM (Listgarten et al., 2012). Therefore, we followed
the leave-one-chromosome-out approach of Yang et al.
(2014) to build a kinship matrix specific to each chro-
mosome by leaving the markers on that chromosome
out.
In the second step, each SNP effect was obtained by
using a linear regression model:
y = 1µ + xg + η, QTL region. If the QTL region was larger than 1 Mb,
where y, 1, and µ are defined as before, x is a vector of
imputed allele dosages (expected number of copies of a
specified allele, ranging from 0 to 2), g is the fixed SNP
effect, and η is a vector of random multivariate normal
residual deviates with variance Gσa2 + Iσe2 . We applied
genomic control to adjust the for genomic inflation by
diving the chi-squared test statistics by the genomic
inflation factor, λ (Devlin and Roeder, 1999). Bonfer-
roni multiple testing correction was applied to control
for false-positive associations. We declared a SNP sig-
5
nificant if the P-value was less than 0.05 divided by the
number of SNP. The significance threshold value for −
log10(P) was 8.48 with 15,355,402 SNP. The Bonferroni
multiple testing is conservative as SNP are not inde-
pendent due to linkage disequilibrium (LD) among
them; therefore, we consider a suggestive significant
threshold at P < 1.0 × 10−6.
Multitrait Meta-Analysis
We carried out multitrait meta-analyses for the im-
(2014). The test statistic was T 2 = ti′ V−1 ti , where ti is a
 b 
2 × 1 vector of signed t values t =  , b is the allele
 se (b)
substitution effect, se(b) is the standard error of the
allele substitution effect estimated by single-variant
analysis using EMMAX software, as described above,
and V is the correlation matrix between traits esti-
mated from the correlation of summary SNP statistics
(signed t-value from single-trait GWAS) over all the
SNP in the analysis. Under the null hypothesis of no
effect, T 2 follows a χ2 distribution with 2 df. The test
statistics were adjusted for genomic inflation (Devlin
and Roeder, 1999). The significance threshold was kept
at 8.48 as applied above for single-trait analysis.
Defining a QTL Region
The QTL intervals were defined as continuous re-
gions including SNP with −log10(P) ≥ 8.48. The SNP
with the highest –log10(P) within an associated region
was taken as the lead SNP. The start and end of the
QTL region were determined by using the following
calculation. First, a value of 3 was subtracted from
the −log10(P) value of the lead SNP. Second, from the
remaining SNP, the most upstream and downstream
SNP were chosen with −log10(P) values not lower than
3 from the −log10(P) for the lead SNP of the region.
Positions of these SNP were taken as boundaries of the
then the 0.5 Mb upstream and 0.5 Mb downstream
from the position of the lead SNP were demarcated as
the QTL region. Subsequent QTL regions on a chromo-
some, if distinctly separated by visual inspection of the
Manhattan plot, were considered as separate QTL.
Functional Annotations of Associated SNP
For each QTL peak, the SNP with the lowest P-value
was designated as the lead SNP. Information on the
lead SNP was obtained from the National Center for
Biotechnology Information database of genetic varia-
Journal of Dairy Science Vol. 101 No. 3, 2018
6 JARDIM ET AL.

tions (https://​www​.ncbi​.nlm​.nih​.gov/​snp/​), functional tal Table S1; https://​doi​.org/​10​.3168/​jds​.2017​-12982).

annotation of genes from BioMart at the Ensembl
Genome Browser (http://​www​.ensembl​.org/​biomart;
Kinsella et al., 2011) and Animal QTLdb (http://​www​
.animalgenome​.org/​cgi​-bin/​QTLdb/​index; Sherry et
al., 2001; Baker, 2012; Hu et al., 2016). The position
of each SNP was defined according to the Bos taurus
genome assembly UMD3.1 (Zimin et al., 2009).
RESULTS AND DISCUSSION
Associated Genomic Regions for Udder Index
Genome scans for udder index revealed significantly
associated markers in Nordic Holstein cattle (Figure
3). A total of 3,859 SNP on 2 chromosomes, BTA6 (771
SNP) and BTA20 (3088 SNP), showed genome-wide
significant associations with udder index. A total of
18,194 SNP on 11 chromosomes were significant at the
suggestive threshold level (P < 1.0 × 10−6; Supplemen-
The associated genomic region and its lead SNP for ud-
der index for each chromosome are presented in Table 2.
The variance for udder index explained by individual
QTL varied from <1 to 13.58% (Table 2). However,
upward bias could be present in the estimate of QTL
effect (Weller et al., 2005). This is due to the winner’s
curse. Introducing a threshold on statistical significance
means that the selected set of signals will preferentially
contain variants whose effects are overestimated in a
particular study sample due to chance noise. Besides,
the standard error of the estimate of SNP effect, espe-
cially with a large allele substitution effect (for example
BTA6), was relatively high (Table 2).
Lead SNP for associated regions on BTA1, 3, 4, 5, 6,
11, 12, 17, and 20 were located within genes (Table 2).
The strongest association signal for udder index was ob-
served on BTA20. The lead SNP (rs111003349; P-value
= 2.31 × 10−12) in this region was located at 35,310,279
bp, which is an intron variant of the FYB1 gene. The

Figure 3. Manhattan plots of whole-genome association analysis for udder index in Nordic Holstein cattle. The x-axis represents chromo-
somes. The y-axis represents −log10(P-value). The horizontal line indicates a genome-wide significance threshold at P < 3.25 × 10−9. Color
version available online.

Journal of Dairy Science Vol. 101 No. 3, 2018

 Table 2. Lead SNP from significantly associated genomic regions for udder index in Nordic Holstein cattle

QTL interval1 Lead SNP

Effect Effect SE of
BTA Start (bp) End (bp)   Position2 (bp) MAF3   allele size4 effect size P-value VA%5 Refsnp id6   Gene7   Consequence type
1 133,380,781 134,423,336   133,863,441 0.12 C −6.01 0.79 4.69e-8 3.70 rs521423052 STAG1 Intron variant
3 25425805 25,833,258   25,433,311 0.44 G −1.81 0.25 2.84e-7 0.78 rs109151138 GDAP2 Intron variant
4 77,833,074 78,803,311   78,424,609 0.06 C −8.26 1.01 5.51e-9 3.73 rs210496253 HECW1 Intron variant
5 88,511,252 89,143,099   88,824,857 0.47 G 1.68 0.24 6.55e-7 0.68 rs209893772 ABCC9 Intron variant
6 16,438,959 17,412,771   16,938,766 0.05 T −12.52 1.37 5.22e-11 7.19 rs481885372 MCUB Intron variant
6 88,423,063 89,398,459   88,899,845 0.08 T 13.82 1.46 1.30e-11 13.58 rs468282389 NA Intergenic variant
11 38,003,904 38,891,799   38,392,647 0.42 G −1.85 0.24 4.81e-8 0.81 rs209257618 EFEMP1 Intron variant
12 20,306,904 20,967,325   20,775,545 0.45 C −1.88 0.25 1.24e-7 0.85 rs385076875 FAM124A Intron variant
16 44,854,031 45,150,866   44,990,424 0.46 A −1.89 0.27 5.43e-7 0.86 rs41807582 SPSB1 Downstream gene variant
17 13,683,310 14,674,260   13,635,276 0.38 G −1.93 0.26 8.50e-8 0.85 rs42382190 HHIP Intron variant
20 9,492,696 10,480,244   9,988,243 0.07 A −6.07 0.74 3.61e-9 2.31 NA NA NA
20 34,810,491 35,904,566   35,310,279 0.26 C −2.92 0.30 2.31e-12 1.37 rs111003349 FYB1 Intron variant
23 35,291,213 35,919,749   35,604,326 0.30 A 1.78 0.26 1.87e-7 0.64 rs133280722 HIST1H1E Intergenic variant
1
QTL interval = QTL interval between the start and the end of the region in one BTA.
2
Position = base pair position in BTA.
3
MAF = minor allele frequency.
4
Effect size = allele substitution effect of the SNP.
ASSOCIATION STUDY FOR MILKING ABILITY TRAITS

5
VA% = percentage of genetic variation explained by SNP (2pqβ2)/VA, where p and q are allele frequencies, β is the marker effect estimate, and VA is the genetic variance of the trait.
6
Refsnp id = reference SNP ID.
7
Gene = lead SNP located in the gene. NA = not available.

Journal of Dairy Science Vol. 101 No. 3, 2018

7
8 JARDIM ET AL.
second strongest association signal for udder index was
observed on BTA6. The lead SNP (rs468282389; P-
value = 1.30 × 10−11) was located at 88,899,845 bp, in-
tergenic between GC and NPFFR2 genes. Pausch et al.
(2016) identified QTL for mammary gland morphology
on BTA6 at 88.72 and 90.37 Mbp in cattle. Wu et al.
(2015) detected QTL for mastitis on BTA6 and BTA20
in Nordic Holstein. They suggested candidate genes in
these 2 regions (i.e., DCK, SLC4A4, and NPFFR2 on
BTA6, and LIFR and EDN3 on BTA20) based on the
location of peak association signals and the biological
function of these genes.
Another strong association signal for udder index
was detected on BTA4. The lead SNP (rs210496253; P-
value = 5.51 × 10−9), an intron variant, was located at
78,424,609 bp on BTA4 in the gene E3 ubiquitin-protein
ligase HECW1-like (HECW1), whose general function
is protein modification (Kinsella et al., 2011). This gene
was differentially expressed when primary cultures of
bovine epithelial with stromal endometrial cells were
exposed to bacterial lipopolysaccharide (Oguejiofor et
al., 2015). Gram-negative Escherichia coli is a common
mastitis-causing bacteria (Wellnitz and Bruckmaier,
2012). Lipopolysaccharides released from E. coli induce
acute inflammatory responses (Jiang et al., 2008).
The lead SNP (rs41807582; P-value = 5.43 × 10−7)
on BTA16 was a downstream variant of SPSB1, which
is involved in mammary gland function and immune
regulation (Ramey et al., 2013). The major QTL or as-
sociation regions reported for udder composite index in
Animal QTLdb are distributed on BTA 12, 14, and 18
(Hu et al., 2016). Chromosomal regions on BTA2, 10,
11, 16, 20, 22, 25, and X associated with udder traits
were reported by Cole et al. (2011).
Associated Genomic Regions for Milking Speed
Genome-wide significant associations (P < 1.0 ×
10−6) with sequence variants were identified with milk-
ing speed on 8 chromosomes (Figure 4). A total of 1198
SNP from 5 chromosomes [BTA4 (7), BTA6 (56), BTA9
(49), BTA19 (233) and BTA20 (853)] showed genome-
wide significant associations with milking speed. 12,363
SNP were significant at the suggestive threshold of P <
1.0 × 10−6 (Supplemental Table S1; https://​doi​.org/​10​
.3168/​jds​.2017​-12982). The associated genomic regions
and lead SNP are presented in Table 3. The variance
for DRP of milking speed explained by individual QTL
varied from <1 to 1.9% (Table 3). These estimates of
QTL variance could be upward bias for the detected
QTL (Weller et al., 2005).
The strongest QTL signal for milking speed was found
on BTA19. The lead SNP (rs110663209; P-value = 4.96
× 10−13) was an intergenic variant located at 7,565,171
Journal of Dairy Science Vol. 101 No. 3, 2018
bp. The lead SNP on BTA20 was located at 39,474,131
bp (rs137401174; P-value = 1.62 × 10−11), an intronic
variant of the retinoic acid induced 14 (RAI14) gene.
The lead SNP (rs43485634; P-value = 3.25 × 10−10) on
BTA6 was located at 102,847,978 bp within the mito-
gen activated protein kinase 10 (MAPK10) gene, which
previously was reported to be expressed in human
mammary glands. Chockalingam et al. (2005) mapped
this gene and found it to be related to the intramam-
mary gland defense mechanism against gram-negative
bacteria associated with bovine mastitis. Guo et al.
(2012) mapped significant SNP on BTA6 associated
with milking speed. Hiendleder et al. (2003) detected
QTL affecting the general udder quality, which could
influence clinical mastitis, on BTA6.
Associations Detected by Multitrait Meta-Analysis
The multitrait meta-analyses results for udder index
and milking speed are shown in Figure 5. A total of
11,444 SNP from 4 chromosomes [BTA4 (6), BTA6
(875), BTA19 (208), and BTA20 (10,355)] showed
genome-wide significant associations in these multitrait
meta-analyses. A total of 36,829 SNP were significant at
the suggestive threshold (P < 1.0 × 10−6; Supplemen-
tal Table S1; https://​doi​.org/​10​.3168/​jds​.2017​-12982).
The highest associated regions on each chromosome
and lead SNP are presented in Table 4. On a number
of chromosomes, the P-value for markers from meta-
analysis decreased substantially compared with single-
trait analyses, for example BTA5, 6, 7, and 20 (Table
4). This could be an indication of pleiotropic QTL af-
fecting both traits. In such scenarios, SNP associated
with both traits are expected to show lower P-values
from meta-analysis compare with single-trait analysis.
Therefore, multi-trait meta-analysis might help in nar-
rowing down the location of the causal variant. We also
observed that some SNP are highly significant for one
trait but not significant for other traits, for example
BTA19 (Table 4). These indicate that the QTL is only
affecting one of the traits.
The most significantly associated region was located
on BTA20, and the lead SNP (rs111003349; P-value
= 2.61 × 10−14) was located in the FYB gene. FYB
encodes a protein participating in the T-cell signaling
cascade and modulation of IL-2A expression (Kin-
sella et al., 2011). This gene has 1 transcript (ENS-
BTAT00000001953.4) and is expressed in the mammary
gland (Nayeri and Stothard, 2016). Our study reports
a QTL near the QTL previously reported for SCC and
mastitis susceptibility (Tiezzi et al., 2015). This udder
health QTL overlaps with QTL associated with heat
stress/tolerance, production, and health traits, and an-
alyzed for thermotolerance in Holstein cattle (Dikmen
 ASSOCIATION STUDY FOR MILKING ABILITY TRAITS 9

et al., 2015). Other lead SNP on BTA19 (rs110663209;
P-value = 3.57 × 10−12), BTA6 (rs470757828; P-value
= 8.32 × 10−13), and BTA4 (rs378411878; P-value =
3.18 × 10−10) were not located within genes.
The lead SNP (rs208392119; P-value = 7.58 × 10−7)
on BTA10 was located in RAD51B, which has DNA-
dependent ATPase activity (Cammack et al., 2008).
The lead SNP (rs209257618; P-value = 2.92 × 10−7) on
BTA11 was located in EFEMP1, which encodes a cal-
cium ion-binding protein and has been associated with
udder index and milking speed in dairy cattle (Ehler-
mann et al., 2003; Matika et al., 2016). The lead SNP
(rs520845091; P-value = 2.91 × 10−7) on BTA12 was
located within FAM124A, which is strongly expressed
in the mammary gland myoepithelial cells (Uhlen et
al., 2010). The lead SNP (rs41704885; P-value = 9.68
× 10−7) on BTA13 was located within ANGPT4, which
is responsible for initiation of the activation of trans-
membrane receptor protein tyrosine kinase (Cammack
et al., 2008). Two-trait meta-analysis identified novel
associated regions (on BTA7 and 10) compared with
single-trait analyses, and 9 out of 13 lead SNP were
located within genes (Table 4).
Multiple QTL on a Chromosome
The associated regions from single-trait analyses
were quite broad (Tables 2 and 3) and many times were
not distinctly separated. The LD in Holstein cattle is
spread over large genomic region. For example, aver-
age LD (r2) was 0.22 at 100 kb distance in Holstein
(de Roos et al., 2008). Therefore, it is possible that
a broad genomic associated region on a chromosome
represents one causal variant. It is also possible that
multiple closely located QTL are segregating for a
trait. However, association peaks on some of the chro-
mosomes were clearly separated from each other (e.g.,
BTA6 and 20 for udder index and BTA19 and 20 for

Figure 4. Manhattan plots of whole-genome association analysis for milking speed in Nordic Holstein cattle. The x-axis represents chro-
mosomes. The y-axis represents −log10(P-value). The horizontal line indicates a genome-wide significance threshold at P < 3.25 × 10−9. Color
version available online.

Journal of Dairy Science Vol. 101 No. 3, 2018

10 JARDIM ET AL.

milking speed), indicating multiple causal factors on

VA% = percentage of genetic variation explained by SNP (2pqβ2)/VA, where p and q are allele frequencies, β is the marker effect estimate, and VA is the genetic variance of the trait.
Downstream gene variant
those chromosomes. A proper test of multiple causal
variants on a chromosome could be done by fitting the
Consequence type
Intergenic variant
Intergenic variant

Intergenic variant

Intergenic variant
lead SNP as a cofactor and repeating the association
Intron variant analyses for that chromosome.

Intron variant

Intron variant
Genomic Inflation

NA In this study, the GRM was used to control the infla-

 

tion caused by the population stratification and familial

5S_rRNA

relatedness (Yu et al., 2006). The same was also ob-

ANGPT4
MAPK10

RAI14

served in cattle population with large half-sib families

  Gene7
NA
NA

NA

NA
NA

(Kadri et al., 2014). However, we observed high genomic

inflation factors (lambda) of 2.25 and 2.20 for udder
index and milking speed, respectively. These lambda
rRs110357799

values are quite high compared with what is generally

rs378411878
rs384905882

rs208012539
rs134787577

rs110663209

rs137401174
rs43485634

rs41704885
VA%5 Refsnp id6

recommended for GWAS study and generally reported

in other dairy cattle studies, for example (Pausch et al.,
2016; Butty et al., 2017). Guo et al. (2012) reported the
lambda value from ranged from 1.3 to 1.5 for various
Lead SNP

1.42
1.07
1.41
1.47
0.81
0.94
1.88
1.39
1.38

production traits in Brown Swiss cattle. Pausch et al.

(2016) reported no genomic inflation (inflation factors
Table 3. Lead SNP from significantly associated genomic regions for milking speed in Nordic Holstein cattle

1.96e-10

3.25e-10
7.70e-10

ranged between 1.04 to 1.08) for traits related to mam-

4.96e-13

1.62e-11
P-value

2.29e-7

8.56e-7
9.45e-8

4.22e-9

mary gland morphology. Beside population structure,

the inflation factor of GWAS can be influenced by level
of polygenic nature of the trait, heritability, sample
SE of
effect
0.32
0.31
0.31
0.36
0.36
0.30
0.30
0.40
0.31

size, and LD pattern (Yang et al., 2011). For highly

polygenic trait, the GWAS inflation might remain high
QTL interval = QTL interval between the start and the end of the region in one BTA.

in the absence of population structure (Yang et al.,

2.85
2.26
2.70
−3.12
2.48
2.20
2.99
−3.27
2.94
Effect
size4

2011). Both the traits analyzed in this study are highly

polygenic in nature. The reliability (heritability) of
pseudo phenotypes for bulls (DRP) used in this study
Effect
MAF3   allele

was rather high. The LD is extensive because of the

G
G
A
A
T
C

T
C
T

small effective population size for Holsteins (de Roos et

al., 2008). Many SNP reached significant level possibly
0.29
0.43
0.35
0.23
0.19
0.35
0.45
0.16
0.41

due to high LD with the same mutations. However, to

Gene = lead SNP located in the gene. NA = not available.

examine whether there was still inflation in single-trait

GWAS test statistics after correcting the population
70,634,510
37,921,163
102,847,978
68,678,693
22,434,962
60,762,237
7,565,171
60,044,004
39,474,131
Position2

structure using genomic data, 10 top eigenvectors with

(bp)

Effect size = allele substitution effect of the SNP.

the highest eigenvalues were further incorporated into

the association model as covariates. The lambda value
for udder index was 1.94 (details not shown) compared
 
 
 
 
 
 
 
 
 
 

with the value (2.25) without fitting any eigenvector.

Position = base pair position in BTA.
71,132,127
38,393,106
103,307,041
69,171,560
22,469,289
61,230,783
8,038,634
60,541,209
39,974,329

Therefore, the high inflation factor observed in our

End (bp)

study could be partly due to unaccounted population

MAF = minor allele frequency.

Refsnp id = reference SNP ID.

QTL interval1

structure due to the leave-one-chromosome (LOCO)

approach (Yang et al., 2014) to build the genomic rela-
tionship matrix.
70,136,308
37,436,199
102,377,851
68,495,197
22,139,742
60,277,395
7,212,536
59,544,275
38,974,932
Start (bp)

High inflation factor in dairy cattle could be caused

by the effect of strong artificial selection influencing
a smaller subset of loci, which then will be different
from the general effect of subpopulation divergence on
the whole genome (Guo et al., 2012). Such inflation
BTA

will differ between different types of traits. However, to

12
13
19
19
20
4
5
6
9

Journal of Dairy Science Vol. 101 No. 3, 2018

 ASSOCIATION STUDY FOR MILKING ABILITY TRAITS 11

Figure 5. Manhattan plot for 2-trait meta-analyses for udder index and milking speed in Nordic Holstein cattle. x-axis represents chromo-
somes. The y-axis represents −log10(P-value). The horizontal line indicates a genome-wide significance threshold at P < 3.25 × 10−9. Color
version available online.

avoid the risk of false positive detection, we adjusted CONCLUSIONS

the test statistics using genomic control (Devlin and
Roeder, 1999). The Q-Q plots of the genome-wide as-
sociation study for milking speed after genomic control
are presented in Supplemental Figure S1 (https://​doi​
.org/​10​.3168/​jds​.2017​-12982) and the lambda value
after genomic control was 1.08.
We observed an increase in number of associ-
ated SNP in meta-analysis of udder index and milking
speed. This demonstrates that multitrait meta-analysis
for genetically correlated traits can increase power to
detect association compared with single-trait analyses.
Meta-analysis results helped narrow down to candidate
genes as many lead SNP were located within genes and
thereby may further facilitate identifying causal muta-
tions. Further, this associated SNP information could
be included in the genomic selection model for better
prediction of breeding values for these traits (Brøndum
et al., 2015; van den Berg et al., 2016).
Association analyses using over 15 million sequence
variations across the whole genome imputed for almost
5,000 progeny-tested bulls indicated several variations
that likely influence udder index and milking speed in
Nordic Holstein Cattle. We identified several candidate
genes for these 2 traits. Finding quantitative trait nu-
cleotides is challenging because there are many regions
of LD and small-effect QTL. The lead SNP found here
can be used in addition to the 54k SNP genotypes for
better prediction accuracy in routine genomic selection.
ACKNOWLEDGMENTS
We are grateful to the Nordic Cattle Genetic Evalu-
ation (Aarhus, Denmark) for providing the phenotypic
data used in this study and Viking Genetics (Rand-
ers, Denmark) for providing samples for genotyping.

Journal of Dairy Science Vol. 101 No. 3, 2018

12 JARDIM ET AL.

This work was supported by research projects funded

by the Center for Genomic Selection in Animals and
Plants (GenSAP) funded by Innovation Fund Denmark
  Consequence type

Intergenic variant
Intergenic variant
Intergenic variant
Intergenic variant
(grant 0603-00519B), the Grønt Udviklings- og Dem-

Intron variant
Intron variant
Intron variant
Intron variant
Intron variant
Intron variant
Intron variant
Intron variant
Intron variant
onstrationsprogram (GUDP) project from the Ministry
of Environment and Food of Denmark, the Milk Levy
Fund (Mælkeafgiftsfonden, Skejby, Aarhus), Viking
Genetics, and Nordic Cattle Genetic Evaluation. The
1,000 Bull Genomes Project is kindly acknowledged for
sharing WGS data. The Brazilian government agency
LOC107132625
LOC101904601

Coordination for the Improvement of Higher Education

FAM124A

Personnel (CAPES) is acknowledged for awarding a

ANGPT4
EFEMP1
RAD51B
STAG1

scholarship grant (process no. 99999.007355/2015-07)

HHIP
  Gene4

FYB
NA
NA
NA
NA

to Júlia Gazzoni Jardim. Anonymous reviewers pro-

vided very helpful comments on this manuscript.
The authors declare no competing interest.
rs110663209
rs111003349
rs208392119
rs209257618
rs520845091
rs135809877
rs209670381
rs133605053
rs470757828
rs521423052
rs378411878

GS, JGJ, BG, and MSL conceived and designed the

rs41704885
rs42382190
(meta-analysis)   Refsnp id3

study. GS and JGJ analyzed the data. JGJ wrote the

manuscript. MSL and BG contributed materials and
analysis tools and participated in the discussion. All
Lead SNP

authors read, revised, and approved the final manu-

script.
3.57e-12
2.61e-14
8.32e-13
3.18e-10
P-value

9.68e-7
8.51e-7
2.92e-7
2.91e-7
9.49e-8
5.66e-9
7.58e-7
5.40e-7
3.09e-7

REFERENCES
Baker, M. 2012. Quantitative data: Learning to share. Nat. Methods
9:39–41.
Table 4. Boundaries of detected QTL and most significant markers in 2-trait meta-analyses

Boettcher, P. 2005. Breeding for improvement of functional traits in

(milking speed)

dairy cattle. Ital. J. Anim. Sci. 4:7–16. http://​dx​.doi​.org/​https://​

QTL interval = QTL interval between the start and the end of the region in one BTA.
4.96e-13
1.96e-10
P-value

doi​.org/​10​.4081/​ijas​.2005​.3s​.7.
4.71e-4
9.45e-8
7.29e-5
2.57e-9
1.04e-5
1.72e-9

0.576
0.030
0.011

Boettcher, P. J., J. C. Dekkers, and B. W. Kolstad. 1998. Develop-

0.43
0.48

ment of an udder health index for sire selection based on somatic

cell score, udder conformation, and milking speed. J. Dairy Sci.
81:1157–1168.
Boichard, D., V. Ducrocq, P. Croiseau, and S. Fritz. 2016. Genomic
(udder index)

selection in domestic animals: Principles, applications and per-

2.31e-12

spectives. C. R. Biol. 339:274–277.

P-value

8.50e-8
4.26e-6
4.81e-8
1.62e-7
3.15e-3
2.26e-5
7.46e-5
4.69e-8

Bolormaa, S., J. E. Pryce, A. Reverter, Y. Zhang, W. Barendse, K.

0.069
0.026

0.68
0.08

Kemper, B. Tier, K. Savin, B. J. Hayes, and M. E. Goddard.

2014. A multi-trait, meta-analysis for detecting pleiotropic poly-
Gene = lead SNP located in the gene. NA = not available.

morphisms for stature, fatness and reproduction in beef cattle.

PLoS Genet. 10:e1004198.
7,565,171
35,310,279
60,762,237
13,635,276
38,392,647
20,767,647
40,545,358
70,902,375
80,567,467
81,561,956
98,759,865
133,863,441
70,634,510

Brøndum, R. F., B. Guldbrandtsen, G. Sahana, M. S. Lund, and G.

Position2

Su. 2014. Strategies for imputation to whole genome sequence us-

(bp)

ing a single or multi-breed reference population in cattle. BMC

Genomics 15:728.
Brøndum, R. F., G. Su, L. Janss, G. Sahana, B. Guldbrandtsen, D.
Boichard, and M. S. Lund. 2015. Quantitative trait loci markers
derived from whole genome sequence data increases the reliability
Position = base pair position in BTA.
 

 
 
 
 
 
 
 
 
 
 
 
 
 

of genomic prediction. J. Dairy Sci. 98:4107–4116.

14,133,474
8,064,523
35,807,496
21,212,112
61,230,783
80,860,563
38,891,799
41,023,990
71,401,053
82,053,693
99,257,277
134,336,137
71,134,368
End (bp)

Browning, S. R., and B. L. Browning. 2007. Rapid and accurate haplo-

Refsnp id = reference SNP ID.

type phasing and missing-data inference for whole-genome associa-

QTL interval1

tion studies by use of localized haplotype clustering. Am. J. Hum.

Genet. 81:1084–1097.
Butty, A. M., M. Frischknecht, B. Gredler, S. Neuenschwander, J.
Moll, A. Bieber, C. F. Baes, and F. R. Seefried. 2017. Genetic and
34,810,294
13,149,930
7,065,518
37,893,450
20,268,239
60,301,896
70,430,316
80,191,919
98,260,455
40,056,015
133,380,781
70,136,308
81,537,874
Start (bp)

genomic analysis of hyperthelia in Brown Swiss cattle. J. Dairy

Sci. 100:402–411.
Cammack, R., T. K. Attwood, P. N. Campbell, J. H. Parish, A. D.
Smith, J. L. Stirling, and F. Vella. 2008. Oxford Dictionary of
Biochemistry and Molecular Biology. 2nd ed. Oxford University
Press, Oxford, UK.
BTA

19
20
12
13
17
10
11
7
9
1
4
5
6

Journal of Dairy Science Vol. 101 No. 3, 2018

 ASSOCIATION STUDY FOR MILKING ABILITY TRAITS
Carlstrom, C., G. Pettersson, K. Johansson, E. Strandberg, H. Stal-
hammar, and J. Philipsson. 2013. Feasibility of using automatic
milking system data from commercial herds for genetic analysis of
milkability. J. Dairy Sci. 96:5324–5332.
Chockalingam, A., M. J. Paape, and D. D. Bannerman. 2005. In-
creased milk levels of transforming growth factor-alpha, beta 1,
and beta 2 during Escherichia coli-induced mastitis. J. Dairy Sci.
88:1986–1993.
Cole, J. B., G. R. Wiggans, L. Ma, T. S. Sonstegard, T. J. Lawlor,
B. A. Crooker, C. P. Van Tassell, J. Yang, S. W. Wang, L. K.
Matukumalli, and Y. Da. 2011. Genome-wide association analysis
of thirty one production, health, reproduction and body conforma-
tion traits in contemporary US Holstein cows. BMC Genomics
12:408.
Daetwyler, H. D., A. Capitan, H. Pausch, P. Stothard, R. van Binsber-
gen, R. F. Brondum, X. Liao, A. Djari, S. C. Rodriguez, C. Grohs,
D. Esquerre, O. Bouchez, M. N. Rossignol, C. Klopp, D. Rocha, S.
Fritz, A. Eggen, P. J. Bowman, D. Coote, A. J. Chamberlain, C.
Anderson, C. P. VanTassell, I. Hulsegge, M. E. Goddard, B. Guld-
brandtsen, M. S. Lund, R. F. Veerkamp, D. A. Boichard, R. Fries,
and B. J. Hayes. 2014. Whole-genome sequencing of 234 bulls fa-
cilitates mapping of monogenic and complex traits in cattle. Nat.
Genet. 46:858–865.
de Roos, A. P., B. J. Hayes, R. J. Spelman, and M. E. Goddard. 2008.
Linkage disequilibrium and persistence of phase in Holstein-Frie-
sian, Jersey and Angus cattle. Genetics 179:1503–1512.
Devlin, B., and K. Roeder. 1999. Genomic control for association stud-
ies. Biometrics 55:997–1004.
Dikmen, S., X. Z. Wang, M. S. Ortega, J. B. Cole, D. J. Null, and P.
J. Hansen. 2015. Single nucleotide polymorphisms associated with
thermoregulation in lactating dairy cows exposed to heat stress. J.
Anim. Breed. Genet. 132:409–419.
Ehlermann, J., S. Weber, P. Pfisterer, and H. Schorle. 2003. Clon-
ing, expression and characterization of the murine Efemp1, a gene
mutated in Doyne-Honeycomb retinal dystrophy. Gene Expr. Pat-
terns 3:441–447.
Fikse, W. F., and G. Banos. 2001. Weighting factors of sire daugh-
ter information in international genetic evaluations. J. Dairy Sci.
84:1759–1767.
Fuchsberger, C., G. R. Abecasis, and D. A. Hinds. 2015. minimac2:
Faster genotype imputation. Bioinformatics 31:782–784.
Gade, S., E. Stamer, J. Bennewitz, W. Junge, and E. Kalm. 2007. Ge-
netic parameters for serial, automatically recorded milkability and
its relationship to udder, health in dairy cattle. Animal 1:787–796.
Gray, K. A., C. Maltecca, A. Bagnato, M. Dolezal, A. Rossoni, A.
B. Samore, and J. P. Cassady. 2012. Estimates of marker effects
for measures of milk flow in the Italian brown Swiss dairy cattle
population. BMC Vet. Res. 8:199.
Guo, J., H. Jorjani, and O. Carlborg. 2012. A genome-wide association
study using international breeding-evaluation data identifies major
loci affecting production traits and stature in the Brown Swiss
cattle breed. BMC Genet. 13:82.
Hayes, B. J., H. A. Lewin, and M. E. Goddard. 2013. The future of
livestock breeding: Genomic selection for efficiency, reduced emis-
sions intensity, and adaptation. Trends in genetics. TIG 29:206–
214.
Hiendleder, S., H. Thomsen, N. Reinsch, J. Bennewitz, B. Leyhe-Horn,
C. Looft, N. Xu, I. Medjugorac, I. Russ, C. Kuhn, G. A. Brock-
mann, J. Blumel, B. Brenig, F. Reinhardt, R. Reents, G. Aver-
dunk, M. Schwerin, M. Forster, E. Kalm, and G. Erhardt. 2003.
Mapping of QTL for body conformation and behavior in cattle. J.
Hered. 94:496–506.
Howie, B., J. Marchini, and M. Stephens. 2011. Genotype imputation
with thousands of genomes. G3 (Bethesda) 1:457–470.
Hu, Z., C. A. Park, J. M. Reecy, and Z. L. Hu. 2016. Developmental
progress and current status of the animal QTLdb. Nucleic Acids
Res. 44(D1):D827–D833.
Iso-Touru, T., G. Sahana, B. Guldbrandtsen, M. S. Lund, and J. Vilk-
ki. 2016. Genome-wide association analysis of milk yield traits in
Nordic Red Cattle using imputed whole genome sequence variants.
BMC Genet. 17:55.
13
Jiang, L., P. Sorensen, C. Rontved, L. Vels, and K. L. Ingvartsen.
2008. Gene expression profiling of liver from dairy cows treated
intra-mammary with lipopolysaccharide. BMC Genomics 9:443.
Kadri, N. K., B. Guldbrandtsen, M. S. Lund, and G. Sahana. 2015.
Genetic dissection of milk yield traits and mastitis resistance quan-
titative trait loci on chromosome 20 in dairy cattle. J. Dairy Sci.
98:9015–9025.
Kadri, N. K., B. Guldbrandtsen, P. Sorensen, and G. Sahana. 2014.
Comparison of genome-wide association methods in analyses of
admixed populations with complex familial relationships. PLoS
One 9:e88926.
Kang, H. M., J. H. Sul, S. K. Service, N. A. Zaitlen, S. Y. Kong, N.
B. Freimer, C. Sabatti, and E. Eskin. 2010. Variance component
model to account for sample structure in genome-wide association
studies. Nat. Genet. 42:348–354.
Kang, H. M., N. A. Zaitlen, C. M. Wade, A. Kirby, D. Heckerman, M.
J. Daly, and E. Eskin. 2008. Efficient control of population struc-
ture in model organism association mapping. Genetics 178:1709–
1723.
Kinsella, R. J., A. Kahari, S. Haider, J. Zamora, G. Proctor, G. Spu-
dich, J. Almeida-King, D. Staines, P. Derwent, A. Kerhornou, P.
Kersey, and P. Flicek. 2011. Ensembl BioMarts: A hub for data
retrieval across taxonomic space. Database (Oxford) 2011:bar030.
Listgarten, J., C. Lippert, C. M. Kadie, R. I. Davidson, E. Eskin, and
D. Heckerman. 2012. Improved linear mixed models for genome-
wide association studies. Nat. Methods 9:525–526.
Lund, T., F. Miglior, J. C. M. Dekkers, and E. B. Burnside. 1994.
Genetic relationships between clinical mastitis, somatic cell count,
and udder conformation in Danish Holsteins. Livest. Prod. Sci.
39:243–251.
Matika, O., V. Riggio, M. Anselme-Moizan, A. S. Law, R. Pong-Wong,
A. L. Archibald, and S. C. Bishop. 2016. Genome-wide association
reveals QTL for growth, bone and in vivo carcass traits as assessed
by computed tomography in Scottish Blackface lambs. Genet. Sel.
Evol. 48:11.
Nayeri, S., and P. Stothard. 2016. Tissues, metabolic pathways and
genes of key importance in lactating dairy cattle. Springer Sci.
Rev. https://​doi​.org/​10​.1007/​s40362​-016​-0040​-3.
Oguejiofor, C. F., Z. Cheng, A. Abudureyimu, O. L. Anstaett, J.
Brownlie, A. A. Fouladi-Nashta, and D. C. Wathes. 2015. Global
transcriptomic profiling of bovine endometrial immune response in
vitro. II. Effect of bovine viral diarrhea virus on the endometrial
response to lipopolysaccharide. Biol. Reprod. 93:101.
Pausch, H., R. Emmerling, H. Schwarzenbacher, and R. Fries. 2016. A
multi-trait meta-analysis with imputed sequence variants reveals
twelve QTL for mammary gland morphology in Fleckvieh cattle.
Genet. Sel. Evol. 48:14.
Ramey, H. R., J. E. Decker, S. D. McKay, M. M. Rolf, R. D. Schnabel,
and J. F. Taylor. 2013. Detection of selective sweeps in cattle using
genome-wide SNP data. BMC Genomics 14:382.
Rupp, R., and D. Boichard. 2003. Genetics of resistance to mastitis in
dairy cattle. Vet. Res. 34:671–688.
Sahana, G., B. Guldbrandtsen, B. Thomsen, L. E. Holm, F. Panitz,
R. F. Brondum, C. Bendixen, and M. S. Lund. 2014. Genome-
wide association study using high-density single nucleotide poly-
morphism arrays and whole-genome sequences for clinical mastitis
traits in dairy cattle. J. Dairy Sci. 97:7258–7275.
Sewalem, A., F. Miglior, and G. J. Kistemaker. 2011. Genetic pa-
rameters of milking temperament and milking speed in Canadian
Holsteins. J. Dairy Sci. 94:512–516.
Sherry, S. T., M. H. Ward, M. Kholodov, J. Baker, L. Phan, E. M.
Smigielski, and K. Sirotkin. 2001. dbSNP: the NCBI database of
genetic variation. Nucleic Acids Res. 29:308–311.
Tiezzi, F., K. L. Parker-Gaddis, J. B. Cole, J. S. Clay, and C. Maltec-
ca. 2015. A genome-wide association study for clinical mastitis
in first parity US Holstein cows using single-step approach and
genomic matrix re-weighting procedure. PLoS One 10:e0114919.
Uhlen, M., P. Oksvold, L. Fagerberg, E. Lundberg, K. Jonasson, M.
Forsberg, M. Zwahlen, C. Kampf, K. Wester, S. Hober, H. Werner-
us, L. Bjorling, and F. Ponten. 2010. Towards a knowledge-based
Human Protein Atlas. Nat. Biotechnol. 28:1248–1250.
Journal of Dairy Science Vol. 101 No. 3, 2018
14 JARDIM ET AL.
van den Berg, I., D. Boichard, B. Guldbrandtsen, and M. S. Lund.
2016. Using sequence variants in linkage disequilibrium with caus-
ative mutations to improve across-breed prediction in dairy cattle:
A Simulation Study. G3 (Bethesda) 6:2553–2561.
Weller, J. I., M. Shlezinger, and M. Ron. 2005. Correcting for bias
in estimation of quantitative trait loci effects. Genet. Sel. Evol.
37:501.
Wellnitz, O., and R. M. Bruckmaier. 2012. The innate immune re-
sponse of the bovine mammary gland to bacterial infection. Vet.
J. 192:148–152.
Wu, X., B. Guldbrandtsen, M. S. Lund, and G. Sahana. 2016. As-
sociation analysis for feet and legs disorders with whole-genome
sequence variants in 3 dairy cattle breeds. J. Dairy Sci. 99:7221–
7231.
Wu, X., M. S. Lund, G. Sahana, B. Guldbrandtsen, D. Sun, Q. Zhang,
and G. Su. 2015. Association analysis for udder health based on
SNP-panel and sequence data in Danish Holsteins. Genet. Sel.
Evol. 47:50.
Journal of Dairy Science Vol. 101 No. 3, 2018
Yang, J., M. N. Weedon, S. Purcell, G. Lettre, K. Estrada, C. J.
Willer, A. V. Smith, E. Ingelsson, J. R. O’Connell, M. Mangino,
R. Magi, P. A. Madden, A. C. Heath, D. R. Nyholt, N. G. Mar-
tin, G. W. Montgomery, T. M. Frayling, J. N. Hirschhorn, M. I.
McCarthy, M. E. Goddard, P. M. Visscher, and G. Consortium.
2011. Genomic inflation factors under polygenic inheritance. Eur.
J. Hum. Genet. 19:807–812.
Yang, J., N. A. Zaitlen, M. E. Goddard, P. M. Visscher, and A. L.
Price. 2014. Advantages and pitfalls in the application of mixed-
model association methods. Nat. Genet. 46:100–106.
Yu, J., G. Pressoir, W. H. Briggs, I. Vroh Bi, M. Yamasaki, J. F.
Doebley, M. D. McMullen, B. S. Gaut, D. M. Nielsen, J. B. Hol-
land, S. Kresovich, and E. S. Buckler. 2006. A unified mixed-model
method for association mapping that accounts for multiple levels
of relatedness. Nat. Genet. 38:203–208.
Zimin, A. V., A. L. Delcher, L. Florea, D. R. Kelley, M. C. Schatz, and
D. Puiu. 2009. A whole-genome assembly of the domestic cow, Bos
taurus. Genome Biol. 10:R42.