G C A T

T A C G
G C A T
genes
Article
Assessment of Imputation from Low-Pass Sequencing
to Predict Merit of Beef Steers
Warren M. Snelling 1, *,† , Jesse L. Hoff 2 , Jeremiah H. Li 2 , Larry A. Kuehn 1,† ,
Brittney N. Keel 1,† , Amanda K. Lindholm-Perry 1,† and Joseph K. Pickrell 2
1 U.S. Department of Agriculture, Agricultural Research Service, U.S. Meat Animal Research Center,
Clay Center, NE 68933, USA; [email protected] (L.A.K.); [email protected] (B.N.K.);
[email protected] (A.K.L.-P.)
2 Gencove, Inc., New York, NY 10016, USA; [email protected] (J.L.H.); [email protected] (J.H.L.);
[email protected] (J.K.P.)
* Correspondence: [email protected]
† The USDA is an equal opportunity provider and employer. The mention of trade names or commercial
products in this article is solely for the purpose of providing specific information and does not imply
recommendation or endorsement by the USD.




Received: 29 September 2020; Accepted: 2 November 2020; Published: 5 November 2020


Abstract: Decreasing costs are making low coverage sequencing with imputation to a comprehensive
reference panel an attractive alternative to obtain functional variant genotypes that can increase the
accuracy of genomic prediction. To assess the potential of low-pass sequencing, genomic sequence of
77 steers sequenced to >10X coverage was downsampled to 1X and imputed to a reference of 946 cattle
representing multiple Bos taurus and Bos indicus-influenced breeds. Genotypes for nearly 60 million
variants detected in the reference were imputed from the downsampled sequence. The imputed
genotypes strongly agreed with the SNP array genotypes (r = 0.99) and the genotypes called
from the transcript sequence (r = 0.97). Effects of BovineSNP50 and GGP-F250 variants on birth
weight, postweaning gain, and marbling were solved without the steers’ phenotypes and genotypes,
then applied to their genotypes, to predict the molecular breeding values (MBV). The steers’ MBV were
similar when using imputed and array genotypes. Replacing array variants with functional sequence
variants might allow more robust MBV. Imputation from low coverage sequence offers a viable,
low-cost approach to obtain functional variant genotypes that could improve genomic prediction.

Keywords: sequence; imputation; genomic prediction; beef cattle

1. Introduction
Current genomic evaluations of beef cattle use genotypes from commercial SNP arrays to predict
breeding values with greater accuracy than breeding values predicted using only pedigree and
performance records. Further increases in accuracy, particularly for multi-breed populations, can be
achieved by including functional sequence variants [1–3]. Obtaining the functional variant genotypes
needed to increase accuracy, however, is a challenge. One array to genotype potentially functional
variants is available [4], but it is missing much of the functional variation detected in the sequence
of beef cattle [5], and many alleles probed by that array are too rare to be informative. One intent
of sequencing efforts is to provide a reference for imputation from array genotypes to sequence
variants, but the disparity in allele frequency distributions of array and sequence variants [4,6] limits
imputation accuracy, especially for the rare variants. Low-pass (<1X) sequence is not subject to the
same limitation and is imputed to comprehensive sets of sequence variants with high accuracy [7,8].
Decreasing sequencing costs [9] coupled with highly multiplexed library preparation methods [10]

Genes 2020, 11, 1312; doi:10.3390/genes11111312 www.mdpi.com/journal/genes

Genes 2020, 11, 1312 2 of 16

make low-pass sequencing (LPS) cost-competitive with SNP arrays, and provides a straightforward
approach to impute functional variant genotypes, without complications of variant selection, probe
design, and call training associated with developing SNP arrays [11]. This study was conducted to
evaluate the potential of LPS in beef cattle, using existing sequence data to mimic LPS before submitting
a large number of samples through an LPS and imputation pipeline.

2. Materials and Methods

2.1. Data Source

Data for this study were obtained from the on-going U.S Meat Animal Research Center (USMARC)
Germplasm Evaluation Project (GPE). Animals were raised, and biological samples for genotyping
and sequencing were obtained following the USMARC standard operation procedures and Federation
of Animal Science Societies (FASS) guidelines [12]. Pedigree (n = 120,207), birth weight (n = 78,625),
postweaning gain (n = 68,846), and marbling score (n = 38,850) records from GPE animals were
extracted from the USMARC cattle records database. A 19,420-animal subset of the GPE population
was genotyped with at least one genotyping array (Table 1). Whole-genome sequence (WGS) was
available from 77 of the 80 GPE steers selected for transcriptome sequencing from individual feed
intake and gain records [13].

Table 1. Animals genotyped in the Germplasm Evaluation Project.

SNP Array N
BovineSNP50 a 9930
BovineHD b 1547
GGP c -F250 2339
GGP-50K 3068
GGP d 5083
aBovineSNP50 (Illumina, Inc.) versions 1 and 2; ~54,000 SNP. b BovineHD (Illumina, Inc.); ~780,000 SNP. c GeneSeek
Genomic Profiler (GGP) F250 (Neogen, Inc.); ~220,000 putative functional SNP. d GGP versions 1 to 4; ~20,000 to
75,000 SNP.

Data included records from the eight historic cycles of GPE and the on-going continuous GPE
project. Starting in 1968, the cycles were breed for comparison experiments, with the base cows
artificially inseminated (AI) to industry sires, representing five to seven breeds. Each cycle included
Angus and Hereford industry sires, and USMARC Angus and Hereford base cows; MARC III composite
cows [14] were introduced in later cycles. Cycle VII was a re-evaluation of the seven breeds (Angus,
Charolais, Gelbvieh, Hereford, Limousin, Red Angus, and Simmental) that were the most influential in
the U.S. beef industry [15], and transitioned into the current continuous GPE project [16]. Sires from 18
breeds were periodically sampled, and the female progeny mated to their breed-of-sire to produce
breeding females that are a high percentage (>87.5%) of one of the 18 breeds. The 18 breeds included
the Cycle VII breeds, and 11 others that conduct national cattle evaluations (NCE) for beef production
traits (Beefmaster, Brahman, Brangus, Braunvieh, ChiAngus, Maine-Anjou, Salers, Santa Gertrudis,
Shorthorn, South Devon, and Tarentaise).
According to the recorded pedigree, the 77 steers with WGS had contributions from 20 different
breeds, and were sired by 70 different registered bulls representing 17 breeds (all continuous GPE breeds
except Tarentaise). Other breeds contributing to the steers included Pinzgauer, Red Poll, and Holstein.
Eighteen steers with MARC III ancestors had up to 7% Pinzgauer and Red Poll, and one was 2%
Holstein, tracing to a twinning study at USMARC [17]. Six steers were purebred, three Angus, and three
Hereford. Twenty were crosses of the predominant Cycle VII breeds, 26 had contributions from other
Bos taurus breeds, and 25 had Bos indicus influence from Brahman or one of the indicus-influenced
Genes 2020, 11, 1312 3 of 16

composites, Beefmaster, Brangus, or Santa Gertrudis. Sixteen steers were sired by one of the 14
sequenced bulls included in the cattle haplotype reference (Table S1).

2.2. SNP Array Genotypes

Genotyped animals represented Cycle VII and continuous GPE. Most sires (AI and natural
service) were genotyped with both the BovineHD (HD; ~770K SNP) and GGP-F250 (F250; ~200K
SNP including ~170K putative functional variant) assays. Additional animals genotyped with those
assays included dams whose sires were not genotyped, and some non-parents with phenotypes for
difficult-to-measure traits. Other dams were genotyped with the BovineSNP50 (50K), and non-parents
with lower density GGP assays. Genotyped DNA was extracted from AI sires’ semen, blood from
USMARC-born single-birth animals, and an ear notch from recorded twins. The sequenced steers were
genotyped with different arrays, 41 with the 50 K, 28 with a lower density GGP, 6 with the F250 and
HD arrays, and 2 with the F250 and a GGP array.
Prior to pedigree imputation with findhap version 3 [18], genotypes were filtered for call rate
(>0.95 by animal and variant) and minor allele frequency (>0.005). Variants were ordered by position
on the ARS-UCD1.2 genome assembly [19], using the coordinates provided in the National Animal
Genome Research Program (NAGRP) Community Data Repository [20]. All animals with genotypes
from any assay were imputed to the combined BovineHD and GGP-F250 variant set. Functional
variant genotypes of 300 2013-born nonparents with F250 genotypes were discarded from the first
round of imputation, as a test of imputation accuracy. Functional variants with correlations <0.95
between the imputed and assayed genotypes of the 300 test animals were removed for the final round
of imputation. The final round imputed genotypes of the 748,804 variants located on autosomes and
the pseudoautosomal region of the X chromosome (paX) for 18,327 animals.

2.3. Genetic Prediction

Breeding values were predicted for birth weight (BW), postweaning average daily gain (PWG),
and marbling score (MARB). The model for each trait was y = Xβ + Zu + e, with var(u) = Aσ2g ,
var(e) = Iσ2e , and cov(u, e) = 0; y is a vector of observations, X is an incidence matrix relating observations
to the vector of fixed effects in β, Z is an incidence matrix relating observations to random additive
genetic effects in u, and e is a vector of random residual effects; A is the numerator relationship
matrix describing pedigree relationships among animals, I is an identity matrix, σ2g is the variance of
additive genetic effects, and σ2e is the variance of residuals. Models to predict genomic breeding values
were identical, except
 that
 A was replaced by a genomic relationship matrix G*. G was computed as
0
G = MM /2Σpi 1 − pi [21], where M is a matrix of variant genotypes (0, 1, or 2 copies of allele B)
and pi is the B allele frequency for the ith variant. G* was scaled as 0.99G + 0.01I to avoid singularity.
Birth weight fixed effects included the age of dam category (2 through 4.5 in half-year increments,
5 to 10 and > 10 years) and the contemporary group (CG) defined by calf sex, year, season, and location
on the research center. Postweaning gain fixed effects included sex at weaning (distinguishing bulls
from steers), and CG defined by year, season, and management group (ration, implant, weigh dates, etc.)
from weaning through yearling (for females retained for breeding) or slaughter. The PWG observations
were computed from all weights observed from weaning through 550 days of age. Following [22],
a quadratic regression on age (days) was fitted for each individual, and weight 160 days post-weaning
was projected to determine the average daily gain. The CG for MARB included PWG CG and slaughter
date. Numeric MARB scores were assigned to USDA Degree of Marbling, with possible scores ranging
from 0 (Devoid00) to 999 (Abundant99).
Best linear unbiased prediction (BLUP) of breeding values for each trait assumed additive genetic
(genomic) and residual variances, estimated with restricted maximum likelihood (REML) algorithms
implemented in WOMBAT [23]. All animals, including the sequenced GPE steers, were predicted with
pedigree relationships, and with the genomic relationships computed with all imputed genotypes.
Phenotypes and genotypes of the sequenced steers were eliminated from the analyses to train variant
Genes 2020, 11, 1312 4 of 16

effects, then to predict molecular breeding values (MBV) of the steers by applying the variant effects to
their genotypes. For each trait, the effects were trained for three sets of variants—(1) variants probed
by the 50K assay, (2) putative functional content of the F250, and (3) the most significant functional
variants selected from 5000 permutations of F250-based breeding values [24]. Variant effects were
solved by α̂ = M0 [MM0 ]−1 û [25], where α̂ is a vector of variant effects and û is a vector of additive
genomic effects predicted with the G for a set of variants. For comparison to the breeding values
predicted with pedigree and genomic relationships with all variants, steers’ MBV were then predicted
by MBV = Ms α̂, where Ms is a matrix of steers’ genotypes. The MBV were predicted with the
genotypes obtained from the SNP arrays and the genotypes imputed from the downsampled WGS.

2.4. Low-Pass Sequence and Imputation

Ten million read pairs (~1X) per steer were randomly sampled from the >10X WGS available
on the 77 sequenced GPE steers, using seqtk [26]. The downsampled sequence was submitted to the
Gencove pipeline for imputation to the cattle haplotype reference panel with loimpute [27].
To build the reference panel for imputation of low-pass sequencing data, sequencing data from
946 animals from two sources were compiled (Table S1): publicly available sequence data available on
the NCBI Short Read Archive, and sequenced samples in the GPE. These animals cover a range of
dairy and beef cattle breeds. For each sample, FASTQ files were obtained and then:

1. Aligned the reads to the ARS-UCD1.2 genome using bwa mem v0.7.17 [28]
2. Sorted the reads using samtools v1.10 [29]
3. Marked duplicate reads using GATK version 4 [30] (MarkDuplicates)
4. Recalibrated base quality scores using GATK version 4 (BaseRecalibrator)
5. Called GVCF in 10Mb windows using GATK version 4 (HaplotypeCaller -ERC GVCF)

Variant calls were then generated and phased:

1. Called variants in the same 10Mb windows as above using the GATK version 4 (GenotypeGVCFs)
2. Filtered single nucleotide polymorphism calls using GATK version 4 (VariantFiltration) with the
filter string ‘QD < 2.0 || FS > 60.0 || MQ < 40.0 || MQRankSum <−12.5 || ReadPosRankSum < −8.00
3. Filtered indel calls using GATK version 4 (VariantFiltration) with the filter string ‘QD < 2.0 || FS >
200.0 || ReadPosRankSum < -20.0 || SOR > 10.00
4. Refined variant calls using BEAGLE v4 [31]
5. Phased variant calls using BEAGLE v5 [32]
6. Filtered indels and multi-allelic sites.
7. Principal components were generated using plink 1.9 [33] restricted to 150,000 randomly-chosen
bi-allelic SNPs with minor allele frequency (across the entire panel) above 5%.

Genotypes of SNP array variants were extracted from the variant call format (VCF) files written
by the imputation pipeline. Identity of the sequencing libraries was confirmed by comparing imputed
genotypes to array genotypes and the genotypes of the variants expressed in muscle transcriptome
of each steer [34]. Additionally, a phred-scaled call confidence (CC) score was assigned to each steer
as a measure of imputation quality. Genotype probabilities (GP) for each array variant listed in the
VCF were extracted, and CC was computed as the mean 10 × log10 (1−GPmax ) of each uncertain call
(GPmax < 1), where GPmax is GP of the most probable of the three possible genotypes at a variant site.
Functional impact of each variant was predicted with snpEff v4.3 [35], using ensemble annotation
(release 96) of the ARS-UCD1.2 assembly [24]. Figure S1 depicts the general flow from the GPE project
data and steer sequence through the MBV of the steers.
Genes 2020, 11, 1312 5 of 16

3. Results

3.1. Cattle Haplotype Reference Panel

Sequence contributing to the imputation reference was generated in different projects, using SOLiD
and Illumina instruments.
Genes 2020, 11, x FOR PEERPrincipal
REVIEW component analysis (PCA) of the sequenced individuals showed
5 of 16

considerable overlap among projects (Figure 1a), suggesting that sequence from the different projects
showed considerable overlap among projects (Figure 1a), suggesting that sequence from the different
and platforms could be combined to construct a haplotype reference panel. The main differences
projects and platforms could be combined to construct a haplotype reference panel. The main
betweendifferences
projects between
where whether or not
projects where they or
whether included or Bosorindicus-influenced
HolsteinHolstein
not they included Bos indicus-influenced animals
(Figure 1b). The(Figure
animals first principal
1b). The component componentBos
first principalseparated taurus from
separated Bos indicus,
Bos taurus from Bosand indicated
indicus, and some
variationindicated some variation
in the individual in the individual
separation of Brahman separation
from Bosof Brahman fromsecond
taurus. The Bos taurus. The second
principal component
separatedprincipal
Holstein component separated
from Angus, Holstein
with otherfrom
Bos Angus,
taurus with other
breeds Bos taurus breeds
intermediate betweenintermediate
Holstein and
between Holstein and Angus. Continental European breeds, such as Simmental and Gelbvieh,
Angus. Continental European breeds, such as Simmental and Gelbvieh, appeared closer to Holstein,
appeared closer to Holstein, and Hereford was closer to Angus. Various Bos taurus crossbreds in the
and Hereford was closer to Angus. Various Bos taurus crossbreds in the reference were along the
reference were along the continuum between Continental breeds and Angus, and Bos indicus
continuum between Continental
influenced crossbreds breeds and
and composite Angus,
breeds and
in the Bosbetween
space indicus Bos
influenced
taurus andcrossbreds
Brahman. and composite
breeds in the space between Bos taurus and Brahman.

(a)

Figure 1. Cont.
Genes 2020, 11, x FOR PEER REVIEW 6 of 16
Genes 2020, 11, 1312 6 of 16

(b)
Figure 1. Principal
Figurecomponent (PC) analysis
1. Principal component of the
(PC) analysis haplotype
of the reference
haplotype reference panel.
panel. (a) Overlap
(a) Overlap among among projects
sequenced withprojects
different platforms;
sequenced (b) depicts
with different PC1
platforms; separating
(b) depicts the Bostheindicus
PC1 separating from
Bos indicus fromBos
Bos taurus
taurus breeds, and PC2
breeds, and PC2 separating Holstein from Angus, with other Bos taurus breeds intermediate
separating Holstein from Angus, with other Bos taurus breeds intermediate between Holstein and Angus. The between
Holstein and Angus. The first two PC explained 11% of genomic relationships among the reference, 7%
first two PC explained 11% of genomic relationships among the reference, 7% by PC1, and 4% by PC2.
by PC1, and 4% by PC2.

3.2. Variants Imputed from Low-Pass Sequence

3.2. Variants Imputed from Low-Pass Sequence
After filtering, 59,198,025 variants with a mean spacing of 44 bp were detected in the haplotype
After reference
filtering, 59,198,025
panel variants
and imputed with
with the a mean
low-pass spacing
pipeline (Table of
S2).44There
bp were detected
were 332,714 in the haplotype
variants,
which were expected to alter the proteins coded by 21,066 of the 21,861 annotated
reference panel and imputed with the low-pass pipeline (Table S2). There were 332,714 variants, protein-coding
genes, and another 327,367 that might affect regulation of those genes (Table 2). Genotypes for
which were expected to alter the proteins coded by 21,066 of the 21,861 annotated protein-coding
715,402 of the 748,804 usable autosomal and paX variants on the SNP arrays were imputed from the
genes, anddownsampled
another 327,367 that might affect regulation of those genes (Table 2). Genotypes for 715,402
sequence.
of the 748,804Onusable
average,autosomal and paX
98.9% of all genotypes variants
called from theon the SNP exceeded
downsampled arrays the
were > 0.9
imputed
GPmax from the
threshold to pass imputation, and 92.6% of variants had pass rates greater than 95%. Low pass rates
downsampled sequence.
were most prevalent for BTA 23, which was the only chromosome with more than 10% of variants
having pass rates less than 95%. Across the genome, every one-megabase (MB) interval contained
Table 2. Functional
variants classification
with pass rates less than 95%of variants
(Table detected
S3). More in of
than 38% thethecattle haplotype
sites within reference
the interval panel.
around
BTA 23:26 MB, and within three consecutive intervals on BTA 10 around 23, 24, and 25 MB had pass
.
rates less than 95%. The BTA 23:26 MB interval was Reference
the most variant-denseSNP
b
singleArray
c
MB interval of
Classification
the genome, with the 77,545 variants
a Variants
separated by a mean of Genes Variants
12.9 bp between variants.Genes
This region

Protein-changing 332,714 21,066 29,519 10,673

High impact 14,773 9084 545 509
Non-synonymous SNP 318,269 20,978 29,011 10,576
Genes 2020, 11, 1312 7 of 16

contained part of the bovine major histocompatibility complex, containing highly polymorphic loci
associated with immunity [36]. The BTA 10:23 and 10:25 MB intervals were relatively dense (18.3 to
24.3 bp mean separation) but with 58.5 bp between variants, the BTA 10:24 MB interval was less dense
than the mean 47.3 ± 26.3 bp separation between variants.

Table 2. Functional classification of variants detected in the cattle haplotype reference panel.

Reference b SNP Array c

Classification a Variants Genes Variants Genes
Protein-changing 332,714 21,066 29,519 10,673
High impact 14,773 9084 545 509
Non-synonymous SNP 318,269 20,978 29,011 10,576
Potentially regulatory 327,357 18,110 13,072 8076
Untranslated region (UTR) 318,495 15,288 12,447 7557
Non-coding RNA 8940 2822 627 519
Intergenic 38,694,029 396,306
Intronic 19,533,912 272,510
Total 59,198,026 21,334 715,402 10,683
aVariants classified with snpEff v4.3 using ensembl ARS-UCD1.2.96 annotation. b Variants detected in the cattle
haplotype reference panel and imputed from the low-pass sequence. c Autosomal and pseudo-autosomal variants
detected in the reference panel and with usable SNP array genotypes in Germplasm Evaluation Project cattle.

None of the downsampled libraries had pass rates less than 95%. While the pass rate and CC
scores rank libraries were similar (Spearman r = 0.90), the phred-scaled CC scores provided clearer
separation between libraries. The CC scores were indicative of the agreement between the genotypes
imputed from the downsampled sequence and called from SNP arrays. The libraries with noticeably
lower CC also had a lower agreement between the sequence and array genotypes. Correlations between
the sequence and array genotypes (rsa ) were < 0.90 for libraries with CC < 36.6, and rsa was > 0.95 for
all but one library with CC > 37.6 (Figure 2).
There was complete agreement between genotypes, which passed imputation from sequence and
called from SNP arrays for 70% of the variants called for at least 35 steers (Figure 3a). The lowest mean
rsa within 0.01 minor allele frequency (MAF) increments was 0.93 at MAF = 0.02, and > 0.98 for all
MAF increments > 0.08. Concordance between sequence and array calls was consistently > 0.98 for all
MAF increments. Agreement between genotypes imputed from downsampled sequence and called
from transcript sequence was somewhat less, but followed a similar pattern (Figure 3b). There was
perfect agreement between the transcript and downsampled calls for about half the transcript variants.
The lowest mean correlation between the downsampled sequence and transcript genotypes (rst ) was in
the MAF = 0.03 increment, with rst = 0.90, and MAF increments > 0.08 had rst > 0.95.
 scores rank libraries were similar (Spearman r = 0.90), the phred-scaled CC scores provided clearer
separation between libraries. The CC scores were indicative of the agreement between the genotypes
imputed from the downsampled sequence and called from SNP arrays. The libraries with noticeably
lower CC also had a lower agreement between the sequence and array genotypes. Correlations
between the sequence and array genotypes (rsa) were < 0.90 for libraries with CC < 36.6, and rsa was >
Genes 2020, 11,0.95 for all but one library with CC > 37.6 (Figure 2).
1312 8 of 16

.
2. Relationship
Figure Figure between
2. Relationship betweenimputation accuracy,
imputation accuracy, expressed
expressed as a correlation
as a correlation (r)genotypes
(r) between betweenimputed
genotypes
imputed from sequence and called from SNP arrays, and call confidence—a function of imputed
from sequence and called from SNP arrays, and call confidence—a function of imputed genotype probabilities.
genotype probabilities. Accuracy and call confidence were lowest for the known crossbred (XB) steers,
which were sequenced with DNA extracted from blood, another low-confidence, low-accuracy steer
was suspected to be a twin. The purebred (PB) Bos taurus steer with lowest accuracy had the lowest
call confidence of any Bos taurus and was a known twin. Bos indicus-influenced steers (>0.1 Brahman)
tended to have lower call confidence and accuracy than Bos taurus steers.

Call confidence and agreement between imputed sequence and array genotypes were strongly
influenced by Bos indicus. Ignoring the steers with unusually low CC, Bos indicus-influenced steers had
lower CC (p < 1e−13 ) and lower rsa (p < 1e−11 ) than Bos taurus steers. Within the Bos indicus-influenced
steers, when the pedigree contributions ranged from 12% to 85% Brahman, the amount of Brahman
influence did not affect CC (p = 0.58) or rsa (p = 0.10). Purebred steers and steers whose sire was in
the haplotype reference had somewhat higher CC than crossbred steers (p = 0.03) and steers whose
sire was not in the reference (p = 0.04), but being purebred or having a reference sire did not affect
rsa (p > 0.10). Influence from minor Bos taurus breeds did not appear to affect CC or rsa , which were
similar for steers composed of only Cycle VII breeds and those with some contribution from other
Bos taurus breeds (p > 0.24). Steers sired by any other Bos taurus breed had a CC and rsa similar
to Angus-sired steers. Steers sired by all Bos indicus-influenced breeds had CC and rsa lower than
Angus-sired steers (p < 1e−3 ), but Brangus-sired steers had higher CC and rsa than steers sired by the
other Bos indicus-influenced breeds (p < 0.003). Sire breed differences were less for agreement with
genotypes called from transcript sequence. Correlations between the imputed sequence and transcript
genotype calls were not different for Angus, other Bos taurus, and Brangus-sired steers (p > 0.11).
Correlations for Brahman-sired steers were less different from the Angus-sired steers (p = 0.04) than
Beefmaster- (p = 0.002) or Santa Gertrudis-sired steers (p < 3e−5 ). Sire breed differences in correlations
tested on a log scale (−log(1−r)), however, revealed some differences among Bos taurus breeds (Table 3)
that were not evident when testing differences on the correlation scale.
 all MAF increments > 0.08. Concordance between sequence and array calls was consistently > 0.98 for
all MAF increments. Agreement between genotypes imputed from downsampled sequence and
called from transcript sequence was somewhat less, but followed a similar pattern (Figure 3b). There
was perfect agreement between the transcript and downsampled calls for about half the transcript
variants. The lowest mean correlation between the downsampled sequence and transcript genotypes
(rst1312
Genes 2020, 11, ) was in the MAF = 0.03 increment, with rst = 0.90, and MAF increments > 0.08 had rst > 0.95. 9 of 16

Genes 2020, 11, x FOR PEER REVIEW 9 of 16

Figure 3. Relationship between imputation accuracy, expressed as a correlation (r) between genotypes
imputed from sequence
Figure andbetween
3. Relationship called imputation
from SNPaccuracy,
arraysexpressed
(a) or transcript sequence
as a correlation (b),genotypes
(r) between and minor allele
frequency (MAF).
imputed Mean correlation
from sequence between
and called from imputed
SNP arraysand called
(a) or genotypes
transcript sequencewithin
(b), and0.01
minorMAF increments
allele
is shown byfrequency (MAF).
blue lines, andMean correlation
the green linesbetween imputed
show mean and called genotypes
concordance within the within
0.01 0.01
MAF MAFincrements.
increments is shown by blue lines, and the green lines show mean concordance within the 0.01 MAF
increments.

Call confidence and agreement between imputed sequence and array genotypes were strongly
influenced by Bos indicus. Ignoring the steers with unusually low CC, Bos indicus-influenced steers
had lower CC (p < 1e−13) and lower rsa (p < 1e−11) than Bos taurus steers. Within the Bos indicus-
Genes 2020, 11, 1312 10 of 16

Table 3. Sire-breed differences among correlations between genotypes imputed from downsampled
sequence and called from transcript sequence.

Correlation (r) Scale −log(1−r) Scale

Sire Breed Effect a SE p Value Effect a SE p Value
Red Angus 4.60 × 10−4 4.92 × 10−3 9.26 × 10−1 −0.07 0.21 7.52 × 10−1
Brahman −2.79 × 10−2 3.81 × 10−3 1.59 × 10−9 1.86 0.16 8.17 × 10−16
Beefmaster −2.08 × 10−2 3.36 × 10−3 1.01 × 10−7 1.60 0.14 1.78 × 10−15
Brangus −1.05 × 10−2 3.11 × 10−3 1.37 × 10−3 1.15 0.13 1.25 × 10−11
Charolais −2.02 × 10−3 3.36 × 10−3 5.50 × 10−1 0.36 0.14 1.55 × 10−2
ChiAngus −2.48 × 10−3 4.92 × 10−3 6.16 × 10−1 0.40 0.21 6.38 × 10−2
South Devon −2.54 × 10−3 6.60 × 10−3 7.02 × 10−1 0.44 0.28 1.22 × 10−1
Gelbvieh −1.63 × 10−3 3.81 × 10−3 6.71 × 10−1 0.29 0.16 7.83 × 10−2
Hereford −6.70 × 10−4 3.55 × 10−3 8.51 × 10−1 0.13 0.15 4.10 × 10−1
Limousin −1.80 × 10−3 6.60 × 10−3 7.86 × 10−1 0.34 0.28 2.34 × 10−1
Maine-Anjou −3.20 × 10−4 4.92 × 10−3 9.48 × 10−1 0.09 0.21 6.72 × 10−1
Salers −2.75 × 10−3 3.81 × 10−3 4.74 × 10−1 0.47 0.16 5.95 × 10−3
Braunveih −3.89 × 10−3 4.92 × 10−3 4.33 × 10−1 0.61 0.21 5.66 × 10−3
Simmental −2.57 × 10−4 4.21 × 10−3 9.52 × 10−1 0.07 0.18 6.79 × 10−1
Shorthorn −1.25 × 10−3 3.55 × 10−3 7.62 × 10−1 0.24 0.15 1.22 × 10−1
Santa
−2.21 × 10−2 3.36 × 10−3 2.34 × 10−8 1.66 0.14 5.55 × 10−15
Gertrudis
a Difference from Angus.

3.3. Genomic Prediction

The three traits examined in this study, birth weight, postweaning gain, and marbling score, were
all estimated to be at least moderately heritable. Heritability estimates were always greatest with
pedigree relationships and the complete set of GPE phenotypes, followed by genomic relationships
using the combined HD and F250 with phenotypes of genotyped GPE animals (Table 4). Functional
content from the F250 explained more variation than the 50K marker set, but less than the full set
of variants. Sets with a few hundred variants selected after permutation to eliminate variants with
consistently large, spurious effects [37,38], explained approximately 2/3rds the variation explained by
the full variant set.
The pedigree and genomic BLUP with all variants included the sequenced steers’ data to predict
(genomic) the estimated breeding values (G) (EBV). The steers’ phenotypes and genotypes were
eliminated from analyses with variant subsets in order to compute variant effects that were not directly
influenced by the steers’ data. Molecular breeding values from applying variant effects to steers’
genotypes had stronger correlations to their GEBV than to their pedigree EBV (Table 5). In all cases,
correlations between MBV and (G) EBV were similar (within SE) using steers’ genotypes imputed
with pedigree from their assayed genotypes or imputed from downsampled sequence. Correlations
between MBV using either set of genotypes were > 0.96 (Table S4).
Genes 2020, 11, 1312 11 of 16

Table 4. Restricted maximum likelihood heritability (h2 ) estimates for birth weight, postweaning gain,
and marbling score using pedigree and different genomic relationship matrices.

Birth Weight Postweaning Gain Marbling Score

Relationship a h2 (SE) N h2 n h2 n
Pedigree a 0.595 (0.008) 78,625 0.526 (0.010) 68,846 0.538 (0.018) 33,850
Gall b 0.573 (0.011) 16,512 0.474 (0.013) 16,144 0.508 (0.017) 10,898
GF250 c 0.545 (0.011) 16,440 0.442 (0.012) 16,068 0.471 (0.016) 10,822
GF250s d 0.380 (0.023) 16,440 0.270 (0.019) 16,068 0.342 (0.021) 10,822
GF250r e 0.066 (0.007) 16,440 0.062 (0.007) 16,068 0.105 (0.009) 10,822
G50K f 0.519 (0.011) 16,440 0.437 (0.012) 16,068 0.466 (0.016) 10,822
a Pedigree BLUP including downsampled steers. b
748,804 autosomal and pseudo-autosomal variants from
GGP-F250 and BovineHD arrays, filtered for >0.95 call rate and pedigree imputation accuracy. Genomic BLUP
(GBLUP) included downsampled steers. c 116,472 filtered variants from GGP-F250. GBLUP excluded downsampled
steers. d GGP-F250 subsets selected for trait-specific effects: 551 birth weight; 585 postweaning gain; and 698 marbling
score. GBLUP excluded downsampled steers. e Randomly selected GGP-F250 subsets, same size as trait-specific
subsets. GBLUP excluded downsampled steers. f 51,496 BovineHD variants common with BovineSNP50 array.
GBLUP excluded downsampled steers.

P
Table 5. Correlations (SE) between molecular breeding values ( (marker effect estimates × genotypes))
and predicted breeding values.

Birth Weight Postweaning Gain Marbling Score

Predictions Using Imputed SNP Array Genotypes
Pedigree a b Pedigree Gall Pedigree Gall
Gall
GF250 c 0.738 (0.061) 0.904 (0.037) 0.779 (0.055) 0.881 (0.041) 0.770 (0.057) 0.926 (0.032)
GF250s d 0.555 (0.079) 0.681 (0.067) 0.653 (0.069) 0.714 (0.063) 0.655 (0.069) 0.750 (0.059)
GF250r e 0.379 (0.090) 0.481 (0.083) 0.344 (0.093) 0.385 (0.090) 0.629 (0.070) 0.741 (0.058)
G50K f 0.710 (0.063) 0.888 (0.039) 0.785 (0.055) 0.886 (0.040) 0.794 (0.053) 0.950 (0.026)
Predictions Using Genotypes Imputed from Low-Coverage Sequence
GF250 c 0.680 (0.067) 0.866 (0.044) 0.779 (0.055) 0.887 (0.040) 0.769 (0.057) 0.936 (0.030)
Gf250s d 0.531 (0.081) 0.634 (0.073) 0.635 (0.071) 0.722 (0.063) 0.641 (0.071) 0.738 (0.062)
GF250r e 0.286 (0.100) 0.390 (0.094) 0.332 (0.096) 0.395 (0.094) 0.649 (0.071) 0.776 (0.057)
G50K f 0.676 (0.067) 0.866 (0.044) 0.805 (0.052) 0.903 (0.037) 0.760 (0.058) 0.941 (0.029)
a Breeding values predicted by BLUP with pedigree relationships, including downsampled steers. b Breeding
values predicted by BLUP with genomic relationships computed using genotypes of 748,804 autosomal and
pseudo-autosomal variants from the GGP-F250 and BovineHD arrays, including downsampled steers. c Molecular
breeding values of the downsampled steers predicted with effects of 116,472 GGP-F250 variants solved from GBLUP,
excluding downsampled steer records. d GGP-F250 subsets selected for trait-specific effects: 551 birth weight; 585
postweaning gain; and 698 marbling score. GBLUP excluded downsampled steers. e Randomly selected GGP-F250
subsets, same size as trait-specific subsets. GBLUP excluded downsampled steers. f 51,496 BovineHD variants
common with BovineSNP50 array. GBLUP excluded downsampled steers.

4. Discussion
Existing WGS available from steers produced by the multi-breed, industry-representative
USMARC GPE project was downsampled to mimic low-pass sequencing, and provide an indication of
how imputing low-pass sequence to the variants detected in a comprehensive haplotype reference
panel might perform. For most of the steers sequenced, there was a strong agreement between
genotypes imputed from downsampled sequence and genotypes called from SNP arrays and
transcriptome sequence.
Five steers, however, had noticeably low agreement with the SNP array genotypes. This lack of
agreement was initially indicated by genotype probabilities included in imputation results, which were
Genes 2020, 11, 1312 12 of 16

summarized into a call confidence score for each individual. Extracting more complete records from the
USMARC database revealed that four of the five low CC, low-agreement steers were twins to another
calf. As the sequenced DNA was extracted from blood, the twins’ DNA would have included DNA
from their co-twin, due to blood cell chimerism resulting from twins sharing blood across placental
membranes [39,40]. The fifth low-confidence, low-agreement steer might have been a single-birth
twin, whose co-twin embryo was lost early in pregnancy [39–41]. The CC score summarizing imputed
genotype probabilities at least provides an indication of imputation accuracy, and possible issues
with the sequenced DNA. Reasons for low CC scores included insufficient sequence reads to match
reference haplotypes, missing reference haplotypes to match sequence reads, and contamination
resulting in sequence matching conflicting reference haplotypes. As DNA extracted from twins’ blood
is contaminated, low CC scores might indicate infertile single-birth heifers that were co-twins to a male
embryo [41]. Further confirmation might be the presence of Y-chromosome sequence in DNA from the
heifer’s blood [42], and higher CC with no Y sequence in the DNA extracted from other tissue.
Lower CC for Bos indicus-influenced steers suggests that haplotypes that match their sequence are
missing from the reference panel. Although the reference panel contains more Brahman cattle than
cattle from several Bos taurus breeds, PCA shows separation between Brahman that were influential
in Australia and some Brahman sampled from the U.S. industry for GPE. Additionally, the Brahman
and Bos taurus contributions to the Beefmaster (25% Hereford, 25% Shorthorn, 50% Brahman) and
Santa Gertrudis (62.5% Shorthorn, 37.5% Brahman) breeds might be isolated. Both breeds descended
from narrow bases, Beefmaster from a single closed herd that originated with Brahman bulls mated
to Hereford and Shorthorn cows [43], and Santa Gertrudis from a single bull mated to F1 Brahman x
Shorthorn heifers [44]. Both breeds allow grading up through mating Beefmaster or Santa Gertrudis
bulls to undocumented females, but do not allow re-creating the composites from unrelated cattle
representing the contributing breeds. Brangus policy, however, allows mating registered Angus (black
and red) and Brahman to create the 62.5% Angus, 37.5% Brahman composite, which might maintain
stronger connections to the contributing breeds, and explain Brangus as having somewhat higher
CC and agreement between imputed genotypes and calls from SNP arrays and transcript sequence.
Broader sampling of Bos indicus-influenced breeds for the imputation reference should increase the
imputation accuracy for these cattle; further increases might be realized by reference construction and
imputation strategies that consider the assembled genome of a Brahman cow [45].
The generally strong agreement between genotypes imputed from downsampled GPE steers and
genotypes called from SNP arrays and transcript sequence certainly suggests imputation from low-pass
sequence is a viable approach to genotyping sequence variants. Having a sequence of influential GPE
animals in the haplotype reference, including sires of 20% of these steers, contributes to the quality
of imputation. Further evaluation outside of GPE is needed to determine suitability of the current
reference for imputing sequence genotypes of current seedstock and commercial crossbred cattle.
Existing SNP array genotypes on current commercial and seedstock cattle might be useful to identify
additional animals who would be informative in the haplotype reference panel. Genomic relationships
among commercial calves, seedstock influencing those calves, and animals in the current reference
could reveal influential seedstock lowly related to cattle in the current reference. Following [24,46–48],
a more refined approach might infer haplotypes from array genotypes, prioritize the haplotypes based
on frequency and existing coverage, then prioritize additions to the reference to add sequence to the
highest frequency haplotypes that are lacking coverage.
The strong agreement between imputed and array genotypes allowed predicting steer MBV with
imputed genotypes that agreed with MBV from variant effects, applied to those array genotypes.
Even with the loss of assayed variants that were not imputed, correlations with pedigree EBV and
GEBV using all assayed variants were similar for MBV computed with both array and imputed
genotypes. Agreement was stronger with GEBV, predicted with available phenotypes for genotyped
GPE animals, than with EBV, which used all available GPE phenotypes and pedigree records, but no
genomic information. Agreement was similar for MBV that used either F250 or 50K genotypes, and was
Genes 2020, 11, 1312 13 of 16

lower for small subsets of the F250. The small subsets selected, based on association with BW and
PWG had a better agreement with corresponding (G) EBV than same-size randomly selected subsets,
but agreement for MARB-associated and random subsets with MARB (G) EBV was similar. Previous
work showed that small sets of SNP, selected with different approaches, might not fully explain
variation within a population, but can predict across populations more accurately than larger sets of
whole-genome SNP [1–3]. These subsets should be examined in cattle that are distant from the GPE
population, before drawing conclusions about their effectiveness. Beyond this, including functional
variants imputed from low-pass sequence that are not interrogated by the F250 might be considered.
The smaller panels were proposed for low-cost genotyping arrays. For a similar cost, the genotypes
could be imputed from low-pass sequence, while avoiding complications of array design and
development. Imputing the full set of variants detected in the haplotype reference from low-pass
sequence is relatively straightforward and can capture individual variants within variant-dense regions
where close, interfering SNP preclude designing probes for genotyping arrays. Especially important
for low-frequency variants, imputed genotypes can be called from matches to haplotype reference
sequence, without the need for sufficient data to train clustering algorithms to call array genotypes.
Somewhat similar to selecting variants to probe with an array, a manageable number of variants might
be selected from the full set of imputed genotypes for genomic analysis. Unlike an array, the set of
variants extracted is flexible, without redesigning and manufacturing a different array.
Genotypes for the 50K variants imputed from low-pass sequence could be extracted to include
with existing array genotypes for genome-enhanced national cattle evaluation (NCE). National cattle
evaluation might be extended to traits that are not routinely recorded, and cattle that are not usually
evaluated if the LD-dependent 50K were replaced with causal variants. Current within-breed NCE
rely on consistent LD between 50K and unknown causal variants for genomic predictions of routinely
recorded traits in seedstock cattle. Causal variants, at least functional variants that are likely to
affect phenotype, could reduce reliance on LD and enable genomic predictions that are more robust
across populations [1–3]. This could allow genomic prediction of difficult-to-measure traits, based on
records from intensely measured herds, and predictions for commercial cattle that are not included
in seedstock evaluations. Reliable predictions to guide sorting commercial cattle for management
and marketing could help to justify the expense of low-pass sequencing. Phenotypes and genotypes
imputed from low-pass sequence on commercial cattle could further increase reliability of genomic
prediction for both commercial and seedstock cattle, if data-sharing mechanisms are in place to allow
commercial records to inform NCE. Similarly, reducing per-sample costs of low-pass sequencing to a
point well under current array costs, perhaps through less expensive DNA extraction and sequencing
library preparation, might encourage more complete genotyping of seedstock and commercial calves,
and provide even more data to support accurate genomic prediction.

5. Conclusions
Existing genome sequence from individuals that also had transcriptome sequence and SNP array
genotypes provided an opportunity to assess low-pass sequence and imputation to sequence variants.
Downsampling mimicked low-pass sequencing, and genotypes for nearly 60 million variants detected
in a broad haplotype reference panel were imputed. Agreement between imputed genotypes and
genotypes called from the SNP arrays and transcriptome sequence was generally strong, somewhat
stronger for Bos taurus than Bos indicus-influenced cattle. Expanding the reference panel to include
more Bos indicus-influenced haplotypes might increase agreement for those cattle. Further evaluation
of relationships among current industry cattle and individuals in the reference panel might reveal
additional cattle that might contribute to the reference. Owing to the agreement between SNP array
and imputed genotypes, MBV with array variant effects applied to either array or imputed genotypes
were similar. Molecular breeding values that more completely explained sequence variation that affect
phenotypic variation might be obtained by transitioning genomic prediction from the limited set of
variants interrogated by SNP arrays, to functional variants detected in sequence. These variants could
Genes 2020, 11, 1312 14 of 16

currently be imputed from low-pass sequence at a cost similar to the least expensive SNP arrays.
Further developments that could lower costs of obtaining low-pass sequence and increase accuracy of
imputation and genomic prediction might make genotyping from low-pass sequence more accessible
and worthwhile for seedstock and commercial cattle.

Supplementary Materials: The following are available online at https://doi.org/10.5281/zenodo.4147100. Table S1:
Sup1.tsv, tab separated text list of animals in imputation reference, containing ID, SRR Accessions, Source, Breed,
and number of downsampled progeny examined in this study; Table S2: Sup2.vcf, VCF-formatted text containing
snpEff annotation of variants imputed from low-pass sequence; Table S3: Sup3.tsv, tab separated text summarizing
number of variants, low-confidence variants, and variant spacing in 1 Mb intervals; Figure S1: steerworkflow.pdf,
diagram depicting processes to obtain imputed genotypes, (G) EBV and MBV of the study steers; Table S4; Sup4.tsv,
tab separated text containing correlations between array- and sequence-based MBV. The loimpute software is
available from https://gitlab.com/gencove/loimpute-public.
Author Contributions: Conceptualization, W.M.S., J.L.H., L.A.K., and J.K.P.; Data curation, W.M.S., B.N.K.,
and A.K.L.-P.; Formal analysis, W.M.S., J.L.H., and J.K.P.; Investigation, W.M.S., J.L.H., J.H.L., B.N.K., and A.K.L.-P.;
Methodology, W.M.S., J.H.L., and J.K.P.; Supervision, L.A.K.; Writing—original draft, W.M.S.; Writing—review
and editing, J.L.H., J.H.L., L.A.K., B.N.K., A.K.L.-P., and J.K.P. All authors have read and agreed to the published
version of the manuscript.
Funding: This research received no external funding.
Acknowledgments: The authors thank the U.S. Meat Animal Research Center staff for animal care, data recording
and management, and assistance with tissue collection.
Conflicts of Interest: J.L.H., J.H.L., and J.K.P. are employees of Gencove, Inc.

References
1. Moghaddar, N.; Khansefid, M.; van der Werf, J.H.J.; Bolormaa, S.; Duijvesteijn, N.; Clark, S.A.; Swan, A.A.;
Daetwyler, H.D.; MacLeod, I.M. Genomic prediction based on selected variants from imputed whole-genome
sequence data in Australian sheep populations. Genet. Sel. Evol. 2019, 51, 72. [CrossRef] [PubMed]
2. MacLeod, I.M.; Bowman, P.J.; vander Jagt, C.J.; Haile-Mariam, M.; Kemper, K.E.; Chamberlain, A.J.;
Schrooten, C.; Hayes, B.J.; Goddard, M.E. Exploiting biological priors and sequence variants enhances QTL
discovery and genomic prediction of complex traits. BMC Genom. 2016, 17, 144. [CrossRef] [PubMed]
3. Xiang, R.; Berg, I.v.d.; MacLeod, I.M.; Hayes, B.J.; Prowse-Wilkins, C.P.; Wang, M.; Bolormaa, S.; Liu, Z.;
Rochfort, S.J.; Reich, C.M.; et al. Quantifying the contribution of sequence variants with regulatory and
evolutionary significance to 34 bovine complex traits. PNAS 2019, 116, 19398–19408. [CrossRef] [PubMed]
4. Rowan, T.N.; Hoff, J.L.; Crum, T.E.; Taylor, J.F.; Schnabel, R.D.; Decker, J.E. A multi-breed reference panel and
additional rare variants maximize imputation accuracy in cattle. GSE 2019, 51, 77. [CrossRef] [PubMed]
5. Snelling, W.M.; Bennett, G.L.; Keele, J.W.; Kuehn, L.A.; McDaneld, T.G.; Smith, T.P.; Thallman, R.M.;
Kalbfleisch, T.S.; Pollak, E.J. A survey of polymorphisms detected from sequences of popular beef breeds.
Anim. Sci. J. 2015, 93, 5128–5143. [CrossRef]
6. Snelling, W.M.; Kuehn, L.A.; Keel, B.N.; Thallman, R.M.; Bennett, G.L. Linkage disequilibrium among
commonly genotyped SNP variants detected from bull sequence. Anim. Genet 2017, 48, 516–522. [CrossRef]
7. Wasik, K.; Berisa, T.; Pickrell, J.K.; Li, J.H.; Fraser, D.J.; King, K.; Cox, C. Comparing low-pass sequencing and
genotyping for trait mapping in pharmacogenetics. bioRxiv 2019, 632141. [CrossRef]
8. Davies, R.W.; Flint, J.; Myers, S.; Mott, R. Rapid genotype imputation from sequence without reference
panels. Nat. Genet 2016, 48, 965–969. [CrossRef] [PubMed]
9. DNA Sequencing Costs: Data. Available online: https://www.genome.gov/about-genomics/fact-sheets/DNA-
Sequencing-Costs-Data (accessed on 26 October 2020).
10. Baym, M.; Kryazhimskiy, S.; Lieberman, T.D.; Chung, H.; Desai, M.M.; Kishony, R. Inexpensive Multiplexed
Library Preparation for Megabase-Sized Genomes. PLoS ONE 2015, 10, e0128036. [CrossRef] [PubMed]
11. Matukumalli, L.K.; Lawley, C.T.; Schnabel, R.D.; Taylor, J.F.; Allan, M.F.; Heaton, M.P.; O’Connell, J.;
Moore, S.S.; Smith, T.P.L.; Sonstegard, T.S.; et al. Development and Characterization of a High Density SNP
Genotyping Assay for Cattle. PLoS ONE 2009, 4, e5350. [CrossRef]
12. FASS. Guide for the Care and Use of Agrictultural Animals in Research and Teaching, 3rd ed.; FASS: Champaign,
IL, USA, 2010.
Genes 2020, 11, 1312 15 of 16

13. Keel, B.N.; Zarek, C.M.; Keele, J.W.; A Kuehn, L.; Snelling, W.M.; Oliver, W.T.; Freetly, H.C.;
Lindholm-Perry, A.K. RNA-Seq Meta-analysis identifies genes in skeletal muscle associated with gain
and intake across a multi-season study of crossbred beef steers. BMC Genom. 2018, 19, 430. [CrossRef]
[PubMed]
14. E Gregory, K.; Cundiff, L.V.; Koch, R.M. Breed effects and heterosis in advanced generations of composite
populations for preweaning traits of beef cattle. J. Anim. Sci. 1991, 69, 947–960. [CrossRef] [PubMed]
15. Wheeler, T.L.; Cundiff, L.V.; Shackelford, S.D.; Koohmaraie, M. Characterization of biological types of cattle
(Cycle VI): Carcass, yield, and longissimus palatability traits12. J. Anim. Sci. 2004, 82, 1177–1189. [CrossRef]
16. Ahlberg, C.M.; A Kuehn, L.; Thallman, R.M.; Kachman, S.D.; Snelling, W.M.; Spangler, M.L. Breed effects and
genetic parameter estimates for calving difficulty and birth weight in a multibreed population1. J. Anim. Sci.
2016, 94, 1857–1864. [CrossRef]
17. E Gregory, K.; E Echternkamp, S.; E Dickerson, G.; Cundiff, L.V.; Koch, R.M.; Van Vleck, L.D. Twinning in
cattle: I. Foundation animals and genetic and environmental effects on twinning rate. J. Anim. Sci. 1990,
68, 1867–1876. [CrossRef]
18. VanRaden, P.; Null, D.; Sargolzaei, M.; Wiggans, G.; Tooker, M.; Cole, J.; Sonstegard, T.; Connor, E.; Winters, M.;
Van Kaam, J.; et al. Genomic imputation and evaluation using high-density Holstein genotypes. J. Dairy Sci.
2013, 96, 668–678. [CrossRef]
19. Rosen, B.; Bickhart, D.; Schnabel, R.; Koren, S.; Elsik, C.; Zimin, A.; Dreischer, C.; Schultheiss, S.;
Hall, R.; Schroeder, S.; et al. Modernizing the Bovine Reference Genome Assembly. In Proceedings
of the World Congress on Genetics Applied to Livestock Production 2018, Molecular Genetics 3, 802,
Auckland, New Zealand, 11–16 February 2018.
20. Schnabel, R. NAGRP Community Data Repository. Available online: https://www.animalgenome.org/
repository/cattle/UMC_bovine_coordinates/ (accessed on 11 February 2020).
21. VanRaden, P.M. Efficient Methods to Compute Genomic Predictions. J. Dairy Sci. 2008, 91, 4414–4423.
[CrossRef] [PubMed]
22. Snelling, W.M.; Allan, M.F.; Keele, J.W.; A Kuehn, L.; Thallman, R.M.; Bennett, G.L.; Ferrell, C.L.; Jenkins, T.G.;
Freetly, H.C.; Nielsen, M.K.; et al. Partial-genome evaluation of postweaning feed intake and efficiency of
crossbred beef cattle1,2. J. Anim. Sci. 2011, 89, 1731–1741. [CrossRef]
23. Meyer, K. WOMBAT—A tool for mixed model analyses in quantitative genetics by restricted maximum
likelihood (REML). J. Zhejiang Univ. Sci. B 2007, 8, 815–821. [CrossRef]
24. Snelling, W.M.; Kachman, S.D.; Bennett, G.L.; Spangler, M.L.; A Kuehn, L.; Thallman, R.M. 197 Functional
SNP associated with birth weight in independent populations identified with a permutation step added to
GBLUP-GWAS. J. Anim. Sci. 2017, 95, 97–98. [CrossRef]
25. Strandén, I.; Garrick, D. Technical note: Derivation of equivalent computing algorithms for genomic
predictions and reliabilities of animal merit. J. Dairy Sci. 2009, 92, 2971–2975. [CrossRef] [PubMed]
26. Seqtk. Available online: https://github.com/lh3/seqtk (accessed on 25 October 2020).
27. Loimpute-Public. Available online: https://gitlab.com/gencove/loimpute-public (accessed on 25 October 2020).
28. Li, H.; Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics
2009, 25, 1754–1760. [CrossRef]
29. Li, H.; Handsaker, B.; Wysoker, A.; Fennell, T.; Ruan, J.; Homer, N.; Marth, G.; Abecasis, G.; Durbin, R.
The Sequence Alignment/Map format and SAMtools. Bioinformatics 2009, 25, 2078–2079. [CrossRef] [PubMed]
30. Van Der Auwera, G.A.; Carneiro, M.O.; Hartl, C.; Poplin, R.; Del Angel, G.; Levy-Moonshine, A.; Jordan, T.;
Shakir, K.; Roazen, D.; Thibault, J.; et al. From FastQ Data to High-Confidence Variant Calls: The Genome
Analysis Toolkit Best Practices Pipeline. Curr. Protoc. Bioinform. 2013, 43, 11.10.1–11.10.33. [CrossRef]
31. Browning, S.R.; Browning, B.L. Rapid and Accurate Haplotype Phasing and Missing-Data Inference for
Whole-Genome Association Studies By Use of Localized Haplotype Clustering. Am. J. Hum. Genet. 2007,
81, 1084–1097. [CrossRef]
32. Browning, B.L.; Zhou, Y.; Browning, S.R. A One-Penny Imputed Genome from Next-Generation Reference
Panels. Am. J. Hum. Genet. 2018, 103, 338–348. [CrossRef]
33. Chang, C.C.; Chow, C.C.; Tellier, L.C.A.M.; Vattikuti, S.; Purcell, S.M.; Lee, J.J. Second-generation PLINK:
Rising to the challenge of larger and richer datasets. GigaScience 2015, 4, 7. [CrossRef] [PubMed]
Genes 2020, 11, 1312 16 of 16

34. Cingolani, P.; Platts, A.; Wang, L.; Coon, M.; Nguyen, T.; Wang, L.; Land, S.; Lu, X.; Ruden, D.M. A program
for annotating and predicting the effects of single nucleotide polymorphisms, SnpEff: SNPs in the genome of
Drosophila melanogaster strain w1118; iso-2; iso-3. Fly (Austin) 2012, 6, 80–92. [CrossRef]
35. Ensembl No Title. Available online: ftp://ftp.ensembl.org/pub/release-96/gtf/bos_taurus/Bos_taurus.ARS-
UCD1.2.96.gtf.gz (accessed on 14 April 2019).
36. Zorc, M.; Ogorevc, J.; Dovc, P. The new bovine reference genome assembly provides new insight into genomic
organization of the bovine major histocompatibility complex. J. Central Eur. Agric. 2019, 20, 1111–1115.
[CrossRef]
37. Lillie, F.R. The Theory of the Free-Martin. Science 1916, 43, 611–613. [CrossRef]
38. Owen, R.D.; Miller, N.E.; Bailey, C.J.; Stevenson, J.A.F. Immunogenetic consequences of vascular anastomoses
between bovine twins. Science 1945, 102, 400–401. [CrossRef]
39. López-Gatius, F.; Hunter, R. Spontaneous reduction of advanced twin embryos: Its occurrence and clinical
relevance in dairy cattle. Theriogenology 2005, 63, 118–125. [CrossRef]
40. Padula, A.M. The freemartin syndrome: An update. Anim. Reprod. Sci. 2005, 87, 93–109. [CrossRef]
41. McDaneld, T.G.; Kuehn, L.A.; Thomas, M.G.; Snelling, W.M.; Sonstegard, T.S.; Matukumalli, L.K.; Smith, T.P.L.;
Pollak, E.J.; Keele, J.W. Y are you not pregnant: Identification of Y chromosome segments in female cattle
with decreased reproductive efficiency. J. Anim. Sci. 2012, 90, 2142–2151. [CrossRef]
42. Lasater Beefmasters, Lasater Philosophy, Composite Cattle. Available online: https://isabeefmasters.com/
about-us/beefmasters-history/ (accessed on 1 March 2020).
43. Breed History—Santa Gertrudis Breeders International. Available online: https://santagertrudis.com/sgbi/
santa-gertrudis-breed-history/ (accessed on 1 March 2020).
44. Koren, S.; Rhie, A.; Walenz, B.P.; Dilthey, A.T.; Bickhart, D.M.; Kingan, S.B.; Hiendleder, S.; Williams, J.L.;
Smith, S.R.; Phillippy, A.M. De novo assembly of haplotype-resolved genomes with trio binning.
Nat. Biotechnol. 2018, 36, 1174–1182. [CrossRef] [PubMed]
45. Ros-Freixedes, R.; Gonen, S.; Gorjanc, G.; Hickey, J.M. A method for allocating low-coverage sequencing
resources by targeting haplotypes rather than individuals. Genet. Sel. Evol. 2017, 49, 78. [CrossRef]
46. Snelling, W.M.; Cushman, R.A.; Keele, J.W.; Maltecca, C.; Thomas, M.G.; Fortes, M.R.S.; Reverter, A.
BREEDING AND GENETICS SYMPOSIUM: Networks and pathways to guide genomic selection. J. Anim. Sci.
2013, 91, 537–552. [CrossRef]
47. Saatchi, M.; Schnabel, R.D.; Taylor, J.F.; Garrick, D.J. Large-effect pleiotropic or closely linked QTL segregate
within and across ten US cattle breeds. BMC Genom. 2014, 15, 1–17. [CrossRef]
48. Saatchi, M.; Garrick, D.J. Developing a Reduced SNP Panel for Low-cost Genotyping in Beef Cattle. Anim. Sci.
Pap. Rep. 2014, 660. [CrossRef]

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional
affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).