Changes to evaluation system (December 2013)
Multi-trait, multi-breed conception rate evaluations
By Paul VanRaden, Chuanyu Sun, Jana Hutchison, Jan Wright, and Mel Tooker
Heifer and cow conception rate (HCR and CCR) evaluations are now multi-trait and multi-breed including crossbred cows instead of the previous single-trait, single-breed evaluations. Fertility traits benefit from multi-trait processing because genetic correlations are high and many observations are missing, with daughter pregnancy rate (DPR) records available since 1960 but HCR and CCR records stored only since 2003. Genetic correlations estimates are 0.45 for HCR with CCR, 0.86 for CCR with DPR, and 0.36 for HCR with DPR. Conception rates were previously modeled using multiple binary success records per lactation (such as no, no, yes), whereas DPR is a continuous lactation measure derived from days open. For simpler multi-trait modeling, conception rate records are now pre-adjusted for environmental effects and are combined into lactation records before analysis.
Inbreeding depression per 1% inbreeding was estimated to be -0.21 for HCR, -0.10 for CCR, and -0.13 for DPR. Heterosis was estimated to be 1.3 for HCR, 3.2 for CCR, and 1.4 for DPR. Crossbred cows get the combined effects of heterosis and no inbreeding compared to purebreds that may average 6%. Thus the combined effect for CCR would be 0.10 * 6 + 3.2 = 3.8% which is added onto the average of the 2 breed means which were 39% for JE and 34% for Holstein cows born in 2005, for example. Genetic differences among breeds were fairly consistent with phenotypic differences. Holsteins had the highest phenotypic and genetic averages for HCR, while for CCR, Jerseys and Milking Shorthorns exceeded Holsteins while Brown Swiss and Guernseys were lowest. Evaluations are adjusted to a 2005 base within each breed as is done for other traits.
Evaluations from the new and previous models were correlated by >0.95 for both HCR and CCR for recent Holstein bulls with >50% reliability, but were less correlated in other breeds because of additional crossbred daughters and contemporaries. For Holstein sires with > 90% reliability, correlations between the single-breed and multi-breed models (both single-trait) were 0.986 for HCR and 0.992 for CCR, indicating little change in rank when adding the other breeds. Genetic trend for CCR was more negative with multi-trait processing because of the correlated influence of DPR. Genetic trends were validated for all breeds using Interbull tests 1 and 3 for CCR and test 3 for HCR. Estimated genetic correlations with other countries from Interbull changed little for Holsteins and averaged 0.02 higher for HCR, 0.02 lower for CCR, and 0.04 higher for the interval from calving to first insemination (CFI). Results were more variable for other breeds, with average correlations that were lower for both HCR and CCR but higher for CFI because the multi-trait model improves the consistency of back-calculation for CFI from CCR and DPR. In general, the new model adds information from crossbred cows and uses DPR as a correlated trait to improve HCR and CCR evaluations of older animals where data is missing.
Nov. 29 Update: A current edit includes HCR in the reference population only for cows with usable milk records, but this edit can distort their sire’s PTA if many daughters are genotyped because the most fertile get included before the less fertile. Two high ranking bulls and their progeny were affected, whereas most other animals were not affected. The two affected bulls were HOUSA000062065919 Charlesdale Superstition whose traditional PTA HCR is 0.3 with 97% REL from 8711 daughters, but his distorted GPTA is 5.2. Similarly, HOUSA000065917481 De-Su Observer has a traditional PTA HCR of 2.5 with 88% REL from 1784 daughters vs. GPTA of 5.3. A new edit was tested to not include genotyped daughters in the reference until they reach 3 years old because that edit already has been used for CCR and DPR. The GPTAs for Superstition and Observer in the test data are within 0.5 of their traditional PTAs, indicating that the edit worked. This problem was detected too late to correct in the December evaluation, so the new edit will be implemented in January.
Genomic evaluations using 61,013 markers
By George Wiggans, Tabatha Cooper, Dan Null, and Paul VanRaden
The number of markers used in computing genomic predictions was increased to 61,013 from 45,195 used previously. The 61,013 markers include all of the previously used markers from the Illumina 50K chip, plus 15,818 markers selected from the GeneSeek High Density (GHD) chip where the magnitude of the marker’s effect was among the top 1,000 effects of the additional markers on that chip for at least one trait. The 61,013 includes markers for the 7 mutation tests included within the HH0, HHB, HHC, HHD, HHM, HHR, and HHP haplotypes; the mutation tests are not yet provided directly to CDCB or AIPL with chip genotypes but instead are provided indirectly by breed associations. Numbers of Illumina BovineHD or GHD genotypes available as of October for within breed imputation of the additional 15,818 markers were 16,956 Holsteins, 1,748 Jerseys, 770 Ayrshires, and 377 Brown Swiss. For Holstein, the average gain in reliability across all traits was 0.5 for predictions using the additional markers. Correlations of new with previous predictions were close to 0.99 for all traits. However, use of additional markers had more impact on the carrier status predicted from haplotypes because each haplotype now has a different set of markers included and may be shifted slightly to the left or right compared to the location of the true genetic defect contained within.
Additional chips introduced
By George Wiggans, Lillian Bacheller, and Paul VanRaden
Two new chips from Zoetis and one from Europe were introduced in October, November, and December evaluations. The Zoetis low density chip will be referred to as ZLD in xml and csv output files and has 11,404 markers of which 10,922 were used in 50K genomic evaluations. The Zoetis medium density chip will be referred to as ZMD and has 56,955 markers of which 39,784 were used in 50K evaluations. The added 17,171 markers on the ZMD that were not from the 50K chip were selected from the Illimuna HD chip and have only a small overlap of 750 markers with the 48,212 additional markers on the GeneSeek high density (GHD) chip. The European low density chip will be referred to as ELD and has 9,072 markers of which 8,348 were used in the 50K evaluation. The ZLD, ZMD, and ELD chips each contain all 6,909 Illumina low density (LD) markers (Boichard et al., 2012) and thus have good overlap with GeneSeek Genomic Profiler versions 1 and 2 (GGP and GP2) that also contain the LD markers. Overlap of marker sets is very useful for quality control and for imputation when different generations or population subsets are genotyped with different chips. In evaluations using 61,013 markers, numbers of markers used from each chip are 2,683 3K, 6,836 LD, 8,032 GGP, 8,465 GP2, 10,932 ZLD, 39,969 ZMD, 8,381 ELD, 45,195 50K, 43,550 GHD, and 55,178 HD. Other factors such as marker spacing, allele frequency, and error rate also affect imputation success, but the chip with highest number of usable markers is reported for animals genotyped with multiple chips. In the format 38 and 105 chip fields, chip numbers will be 10 for ZLD, 11 for ZMD, and 12 for ELD.
Genomic weighting and deregression improved
By Paul VanRaden, Mel Tooker, and Tabatha Cooper
The weighting factors to include cow records in genomic evaluations and the deregression methods to obtain pseudo records from traditional predicted transmitting abilities (PTA) were both revised to improve the accuracy of predictions. Since 2010, cow PTAs for yield traits were adjusted to better match the mean and standard deviation of bull PTAs, and now the weights on cow PTAs are also reduced in all traits because cows contribute smaller gains to prediction accuracy than previously assumed from theory. Weights were reduced to 70% of the theoretical daughter equivalents derived from the cow's traditional reliability and reliability of parent average. This adjustment improved the observed reliabilty of genomic predictions by about 0.5% for young animals. The second change was to deregress the traditional PTAs jointly across animals instead of 1 animal at a time. Each animal gets credit for its own records and for records of its non-genotyped progeny but not for its genotyped progeny. This prevents double counting of traditional information when parents and progeny are both genotyped. This new deregression method was programmed and tested a few years ago but did not show an advantage at that time because fewer cows contributed to the reference population, multiple generations of genotyped females were not yet present, and some weight was shifted from genotyped sons to non-genotyped daughters. However, joint deregression now helps to remove biases that may be created when only the best genotyped progeny receive phenotypes, called genomic pre-selection. The improved deregression increased the observed reliability of prediction by another 1.0% compared to the simple deregression used previously. Neither of the changes affected the means, standard deviations, or slopes of the regressions for predicting future data. Correlations were about 0.996 between genomic predictions before and after these changes to the weights and the deregression. Largest changes were for older dams that had many proven, genotyped sons or genotyped daughters with records, and reliabilities of these dams also decreased slightly because information from their genotyped progeny were no longer double counted.
Jersey haplotype for Polled
By Paul VanRaden and Dan Null
A haplotype test for Polled was introduced for Jerseys (JHP) in September using the same methods as for Holsteins (HHP) in August. In both cases, laboratory tests for Polled are included as data for genotyped animals, together with an assumed status of homozygous normal for USA and CAN bulls with at least 100 daughters and not designated as Polled. Using this data, Polled status is imputed for all other genotyped animals. Laboratory test results are provided by breed associations. A test for Brown Swiss (BHP) could be added in the near future, but may require genotypes from a larger number of Polled animals.
The dominant inheritance of polled / horned was first documented more than a century ago by USDA (Spillman, 1905). At least 2 different mutations in the same genetic region around 1.6-1.9 Mbases (UMD3.1 map) on chromosome 1 cause the polled trait (Medugorac, 2012). The first mutation is called Celtic, where 10 base pairs at 1,706,051–1,706,060 are deleted and 212 base pairs of DNA from 1,705,834–1,706,045 are duplicated and inserted instead. The second mutation is called Friesian and has been fine mapped to a small haplotype region, but the exact causative QTL is not yet known. Medugorac (2012) reported that the Friesian mutation was present only in Holsteins, but by examining haplotype similarity we found that the Friesian mutation actually has slightly higher frequency than the Celtic mutation in Jerseys. The Friesian mutation was introduced to Jerseys primarily via Fair Weather Bernard-P (USA634104) and Fair Weather Case-PP (USA633214) both born in 1978, whereas the earliest genotyped carrier of the Celtic mutation is Normsland Belle Boy-P (USA592073) born in 1960. Jersey pedigree files include Polled ancestors earlier than this (Herron, 2011), but those animals are not genotyped. Jerseys and Holsteins share fairly long sections of both the Friesian and Celtic haplotypes, but very little of either haplotype is preserved in the Brown Swiss breed, which may have a different mutation as noted by Medugorac (2012).References for Polled:
Medugorac et al., 2012. http://www.plosone.org/article/info:doi/10.1371/journal.pone.0039477
Herron, J.E. 2011. History of the polled Jersey in the USA. http://www.polledjerseys.com/history.htm
Specht, L.W. 2008. Polled Holstein history. http://extension.psu.edu/animals/dairy/documents/polled-holsteins-history
Spillman, W. J. 1905. Mendel’s law in relation to animal breeding. Journal of Heredity. http://jhered.oxfordjournals.org/content/os-1/2/171.full.pdf