Changes in genetic trends in US dairy cattle since the implementation of genomic selection

Genomic selection increases accuracy and decreases generation interval, accelerating genetic changes in populations. Assumptions of genetic improvement must be addressed to quantify the magnitude and direction of change. Genetic trends of US dairy cattle breeds were examined to determine the genetic gain since the implementation of genomic evaluations in 2009. In-breeding levels and generation intervals were also investigated. Breeds included Ayrshire, Brown Swiss, Guernsey, Holstein (HO), and Jersey (JE), which were characterized by the evaluation breed the animal received. Mean genomic predicted breeding values PBV ( ) were analyzed per year to calculate genetic trends for bulls and cows. The data set contained 154,008 bulls and 33,022,242 cows born since 1975. Breakpoints were estimated using linear regression, and nonlinear regression was used to fit the piecewise model for the small sample number in some years. Generation intervals and inbreeding levels were also investigated since 1975.


INTRODUCTION
Official genomic evaluations for dairy cattle in the United States were released in 2009 for Holstein (HO), Jersey (JE), and Brown Swiss (BS), 2013 for Ayrshire (AY), and 2016 for Guernsey (GU; Wiggans et al., 2017). Those evaluations comprise multibreed BLUP followed by single-breed estimation of SNP effects (VanRaden, 2008(VanRaden, , 2009. Genomic PTA are obtained as the combination of direct genomic value based on markers and parent average. Since the implementation of genomic selection (GS) in the United States, it has been documented that genetic trends have changed significantly for the HO breed (García-Ruiz et al., 2016). Genetic trends measure the success or failure of a breeding program and are needed to justify the usefulness of genetic evaluations and, therefore, the rationale behind using new methodologies (Grosu et al., 2014). Milk, fat, and protein yields have increased remarkably alongside the uptake of genotyping in the dairy industry. Since 2009, the genetic improvement in countries such as the United States with advanced dairy industries has predominantly relied on GS (Scott et al., 2021).
There are currently over 6 million genotyped animals in the dairy cattle US evaluation system (https: / / webconnect .uscdcb .com/ #/ summary -stats/ genotype -count/ chip -type), but the uptake of genotyping is not even across breeds. As of May 2022, the number of genotyped AY, BS, GU, HO, and JE is 16. 8K, 69.3K, 8.2K, 5.4M, and 748K, respectively. The increase in reliability of predicting genomic PTA depends on the number of genotyped animals (VanRaden et al., 2009), trait heritability (Hidalgo et al., 2021), phenotype availability (Mulder, 2017), and how the genomic information is used (Scott et al., 2021). Increased reliability of predicting genomic PTA and a reduction in generation interval are known for driving the accelerated rates of genetic gain under GS. García-Ruiz et al. (2016) showed that generation interval for sires and dams of HO bulls in the United States decreased from 7 and 4 yr, respectively, to 2.5 yr after the implementation of GS. As HO is the predominant breed in the United States and made up 81.4% of the national breed composition in 2018 (Guinan et al., 2019), it is no surprise that the improvements after GS are apparent in this breed. Nonetheless, this may not be the reality for all dairy breeds.
The improvement in genetic trends for all breeds has been documented and is available at https: / / webconnect .uscdcb .com/ #/ summary -stats/ genetic -trend. However, there is little evidence and documentation of the improvement and changes in colored breeds (i.e., AY, BS, GU, and JE) in the United States since the introduction of genomic evaluations. Reviewing the assumptions of genetic progress since the introduction of genomics is crucial for advising breed associations and nonacademic stakeholders on how to use this technology moving forward. Therefore, the main purpose of this study was to investigate genetic trends in the 5 breeds receiving genomic evaluations for production, fertility, longevity, and health traits. We further investigated changes in generation intervals and inbreeding levels for those breeds.

Data
No animals were used in this study, and ethical approval for the use of animals was thus deemed unnecessary. The data for this study were extracted from the National Cooperator Database maintained by the Council on Dairy Cattle Breeding (CDCB) in Bowie, Maryland. The data included (genomic) PTA of bulls and cows born since 1975 that incorporated pedigree, phenotypic, and genomic (when available) data from the August 2021 official evaluation with the latest lifetime net merit (NM$) index update (Table 1). Phenotypic data are collected on farms as part of routine herd management by the National DHI.
Genomic PTA were calculated using multistep GB-LUP models as described by VanRaden et al. (2009) and consisted of the following 5 steps: (1) estimation of breeding values using a traditional multibreed BLUP, (2) de-regression of EBV, (3) estimation of SNP effects within each breed, (4) prediction of direct genomic value (DGV), and (5) blending of DGV with parent average/EBV. Given the model, where y is a vector of de-regressed evaluations of genotyped animals, 1 is a vector of ones, µ is the overall mean, Z is a design matrix that allocates records to breeding values, g is a vector of DGV to be estimated, and e is a vector of random residual effects. It was assumed that g ~, ) where G is the genomic relationship matrix based on SNP markers and σ g 2 is the additive genetic variance, and e ~, ) where R is a diagonal matrix containing weights based on the reliability of the breeding values and σ e 2 is the residual variance. To avoid overprediction, DGV were estimated with 10% weight (δ) for the polygenic effect. Thus, the model becomes ) where G δ is a combined relationship matrix, G δ = δA 22 + (1 -δ)G. The blending approach in step 5 is described in VanRaden et al. (2009). Following the CDCB editing (https: / / www .uscdcb .com/ ), for an animal to be included in the analyses, the IDs required a US, Canadian, or 840 identification code to illustrate the progress in North America. Each breed is on an individual base (i.e., within-breed base); therefore, individual plots are not directly comparable. Breed code was based on the one the animals received in the August 2021 evaluation, which reflects the breed base representation. Evaluation breed codes included AY, BS, GU, HO, and JE. Animals with a heterosis value greater than 50% were removed.
We analyzed milk, fat, and protein yield, SCS, productive life (PL), daughter pregnancy rate (DPR), and livability (LIV). These traits were selected based on their weight in the most recent update of the NM$ index . Minimum reliability for milk PTA was 9%, and if milk PTA was missing, the animal was removed from the analyses. For bulls, data were restricted to at least 10 daughters per bull per trait, and a weighted average for bulls was calculated to account for the fact that bulls have different daughter counts. No bulls with less than 10 daughters were included in the calculation of genetic trends as they do not substantially contribute to genetic progress. Cows had to have at least 1 calf (Masuda et al., 2018), but their PTA were not weighted by progeny count. Additionally, the cows must have an evaluation and their milk, fat, and protein records contribute to the sire evaluation. Cows without sire information were removed. After applying the above filters, predicted breeding values (PBV) were calculated by multiplying (G) PTA by 2. Only PBV from bulls born from 1975 up to 2017 and cows up to 2019 were used for plotting genetic trends to highlight the progress made using multiple genetic evaluation methodologies. The genetic base year was 2015 (Norman et al., 2020), meaning that mean PBV PBV ( ) for all milking cows born in 2015 were set to 0.
Pedigree and genomic inbreeding levels were calculated using inbreeding coefficients with the same data. Inbreeding levels were limited to genotyped animals with both genomic and pedigree inbreeding coefficients. The pedigree inbreeding coefficient is defined as the probability that a pair of alleles is identical by descent. Genomic inbreeding coefficients are calculated using the methods used by the USDA Animal Genomics and Improvement Laboratory (AGIL) as described by Van-Raden et al. (2011). The genomic relationship matrix (G) is calculated using the following formula: where p i contains allele frequencies expressed as a difference from 0.5 and multiplied by 2 at locus i, Z is the subtraction of P from M. Variable M is the gene content matrix expressed as −1, 0, and 1 (VanRaden, 2008). Generation intervals were calculated using the same edits and were divided into dams of bulls, sires of bulls, sires of cows, and dams of cows. To stratify by breed, both parents and offspring must have the same breed code. Additionally, pedigree information was determined. Generation interval was limited to where the breed of both parents was known. Generation interval equals animal birth date minus parent birth date divided by 365.25 and dams age must be greater than 1 yr.  (Dickinson et al., 1971); MFP$ = milk-fat-protein dollars (Norman et al., 2010); NM$ = lifetime net merit dollars (VanRaden and Wiggans, 1995

Model and Analysis
In previous studies (Van Tassell and van Vleck, 1991;García-Ruiz et al., 2016;Hagan et al., 2021), the 4-path selection model was used to calculate genetic trends using selection differentials. Dechow and Rogers (2018) found that the 4-path selection model was derived for a closed population; however, through the widespread use of AI commercial herds are not in fact closed. The authors demonstrated that, by tracing the transmission of genetic merit from parents to offspring, the rate of genetic progress in commercial dairy farms is expected to be the same as that in the genetic nucleus. Thus, we generated genetic trends based on bull and cow contributions to genetic improvement.
We used SAS (Version 9.4, SAS Institute Inc.) to calculate genetic trends, inbreeding levels, and generation intervals. We calculated PBV, inbreeding, and generation intervals per year using PROC MEANS. The PROC REG was used to create a linear model to estimate the breakpoints (priors) to be included in the nonlinear procedure of SAS to fit the best line between the breakpoints. Breakpoints are the value of x, where the slope of the linear function changes and typically must be estimated. The breakpoints were estimated every 5 yr based on year of birth and were forced to join at the points midway between the ends of the adjacent 5-yr windows. Piecewise (segmented) linear regression is a form of regression that allows multiple linear models to be fitted to the data for different ranges of x. Segmented linear regression was used for this analysis to capture the nonlinear behavior for PBV and inbreeding trends. Also, this aided in plotting the genetic trends of colored breeds when data were limited and based on only a few observations for some years. Finally, PROC GPLOT and PROC SGPLOT were used to plot the data over time.
The yearly unit change was determined by calculating the difference in PBV ( ) 9 yr before and after genomic evaluations were released, and then dividing by the number of years. Percentage change was computed as the difference between annual post genomic evaluations and annual gain pre-genomic evaluations, multiplied by 100.

RESULTS AND DISCUSSION
In this study, we investigated genetic trends for AY, BS, GU, HO, and JE to identify the rates of genetic gain since GS was implemented for these breeds. Genomic evaluations in the United States officially started in January 2009 for HO and JE, in August 2009 for BS, in April 2013 for AY, and in April 2016 for GU (Wiggans et al., 2017). Trends were explored from 1975 up to 2019; therefore, most of the progress achieved for GU is expected to be under traditional selection. To better understand the data and the genotyping uptake in these 5 breeds, descriptive statistics are provided for each breed before examining individual trends.

Descriptive Statistics
Since 1975, there have been 154,008 bulls with at least 10 daughters with milk yield records in the US evaluation system, as shown in Table 2. Of those bulls, AY, BS, and GU make up less than 2%, whereas HO and JE represent 87 and 8%, respectively. Interestingly, BS proportionally has the most genotyped bulls (Table  3). This is a result of the European Brown Swiss Federation Intergenomics project, which established an international database of genotypes for the BS populations. As seen in the frequency distribution in Figure 1a, the number of bulls genotyped has been increasing rapidly per year. Slightly more than half of the bulls born in 2005 were genotyped, and in 2017, all bulls were genomically tested. The BovineSNP50 was manufactured in late 2007. Animals born in 2005 were selected to be genotyped based on their nongenomic PTA to be part of the training data for the GPTA calculation (Weigel et al., 2010). Among all the cows, HO represents the largest proportion and AY is the least represented breed ( Table 2). The number of genotyped cows has increased throughout the years, but still represented only 17.3% of the total cows born in 2017 ( Figure 1b). Proportionally, the JE breed has genotyped the most cows since 1975 (8.6%); however, the proportion of cows genotyped is still a stark contrast to the bulls (overall 3.9% versus 23.8%) since 1975. This is due to the widespread use of bulls and the biological limitation that makes cows have fewer daughters per year.

Genetic Trends
Milk, Fat, and Protein. The most apparent improvement across all traits was observed for production. Because milk historically received so much emphasis in the selection indexes (52% in 1971) and producers were primarily paid for increased production, this is anticipated (Table 1). Breeding objectives of dairy industries around the world have been evolving over the years toward functional and fertility traits and away from production traits (Miglior et al., 2005).
Overall, genetic trends for milk, fat, and protein yields were similar for both bulls and cows across all breeds (Figures 2-6; Supplemental Figures S1-S5, https: / / github .com/ fionaguinan/ JDS _genetic _trends/ blob/ b2e327e557410b19fd0c904c0741f955f73ac5a1/ Supplementary _materials .pdf). This is partially due to the high genetic correlations between the 3 yield traits. Tsuruta et al. (2004) reported milk-fat genetic correlations to range from 0.48 to 0.69 and milk-protein from 0.82 to 0.87. Milk, fat, and protein PBV for AY bulls show a steady increase ( Figure 2). Interestingly, the rate of genetic improvement for production traits accelerated before 2013, even though AY did not receive a genomic evaluation until 2013. This is due to producers genotyping the best (and consequently older) AY bulls once genomic evaluations were released for HO and JE in 2009. This impacted all the traits because the evaluations include all historical data. Ayrshire bulls born in 2016 had a PBV of 988 kg for milk, 46.6 kg fat, and 31.7 kg protein, whereas the ones born in 2012 (i.e., before GS) had PBV of 203 kg, 12.9 kg, and 10.9 kg, in the same order. A similar trend is evident for the AY cows. In 2013, PBV for milk, fat, and protein for AY cows became positive (Supplemental Figure  S1). This again highlights the impact of genotyping bulls that were being used when genomic evaluations were initially released.
Genetic trends for BS bulls and cows are comparable and moving upward, especially since 2009 when BS received genomic evaluations. This was also the point where milk and fat PBV became positive (Figure 3; Supplemental Figure S2). Protein was a couple of years behind in 2011, probably because of the lower emphasis in the NM$ at that time (Table 1). As GU only began receiving genomic evaluations in 2016, and in the data set, we only have bulls with at least 10 daughters up to 2016, the inference to be made from this trend regarding the use of genomics is limited (Figure 4). However, since 1975, there has been a large increase in milk, fat, and protein production for GU bulls. As our GU cow data set extends to 2019, we see an increase in PBV for the yield traits since 2009 (Supplemental Figure S3). This indicates that the best GU were genotyped before receiving a genomic evaluation.
Due to the large number of HO bulls and cows, the trends are more defined. HO bulls have potentially made the most progress in the evaluated time period. In 1975, the PBV for milk yield was −2,030 kg; however, in 2017, this statistic was 687 kg. A sharper increase is evident for fat and protein yields since implementing HO genomic evaluations in 2009 ( Figure 5). Similarly, milk, fat, and protein PBV have increased substantially over time for HO cows (Supplemental Figure S4). In 2017, JE bulls had average milk, fat, and protein PBV of 387, 21.9, and 17.9 kg. Compared with 2009, this is an increase of 556, 21.4, and 19.9 kg, respectively. Despite not receiving any emphasis in the Jersey Performance Index, milk has also increased over time for both bulls and cows ( Figure 6; Supplemental Figure S5).
SCS. Genetic evaluations for SCS were published for the first time by the USDA Animal Improvement Program Laboratory in 1994. This resulted in favorable changes in SCS that had previously been due to selec-   tion for high milk yields; however, it took 8 yr for this to be evident (Norman et al., 2016). In our study, due to the low number of bulls with daughters with SCS records, the trend is almost nonexistent in the AY bulls; therefore, inferences are not relevant (Figure 2). With AY cows (Supplemental Figure S1), a steady increase in SCS is apparent until 1999 (6.0). After this point, a downward trend is visible, which is favorable. More progress in SCS is being made with AY cows than AY bulls. For BS bulls (Figure 3), again, due to the lower number of bulls with at least 10 daughters with SCS records, there is no apparent trend. The scale of the SCS graph is relevant. Although there may be a trend, the graph ranges from 5.4 to 6.4. We see a gradual increase in SCS PBV for cows until 2012, with a substantial decline after that (Supplemental Figure S2). Trends of SCS for GU bulls (Figure 4) Figure S3, SCS has decreased since 2012. In 1994, SCS was implemented into the NM$. It often takes about 6 yr for the trend to start changing in the desired direction after implementing a new trait in a selection index. The SCS trend for HO bulls gradually increased from 1975 to 1996; however, since 1996, it has been on a steady downward trend (Figure 5), thus highlighting the usefulness of incorporating new traits into the NM$. Even though it is a slight decrease, the PBV for SCS decreased by 0.4 in total from 1975 to 2017. The SCS trend for HO cows is a little more delayed than for the HO bulls (Supplemental Figures  S4 and 5, respectively). Since 2005, HO cows have consistently decreased the PBV for SCS. There is an evident trend change observed for this trait after the implementation of GS. In 2010, SCS received its highest weight in the Jersey Performance Index, which is 6%. Since 2011, SCS has remained constant for JE bulls (6.0 in 2017; Figure 6). Conversely, in JE cows, SCS has remained constant since 2009, as seen in Supplemental Figure S5.

(5.8). For GU cows in Supplemental
PL. Productive life measures the ability of a cow to stay alive on the farm while producing milk and is measured in months. Genetic evaluations for PL were first published by Animal Improvement Program Laboratory in 1994. Although cows have been directly selected for longer productive lives, management decisions have shortened the actual tenure of cows in the milking herd (Norman, 2019). The PBV for PL has increased for AY bulls over the past 41 yr, from −5.6 mo in 1975 to 1.1 mo in 2016 (Figure 2). The AY cows have made similar progress in PL, with an PBV for PL in 1975 of −10.9 to 1.9 in 2019, again becoming a positive PBV in 2013 (Supplemental Figure S1).
In 2011, a positive PL PBV was observed for BS bulls. Since 2011, the genetic trend has been positive and with a steep slope. Productive life has increased 9.7 mo since 1975 for BS bulls (Figure 3). Similarly, with BS cows, PL has been gradually increasing over time (Supplemental Figure S2). Since 1999, there has been a consistent increase in PBV per year for PL. This trait has also been increasing over time for GU bulls, from a low of −11.4 mo in 1975 to 2.9 mo PBV in 2015 ( Figure  4). For GU cows, PL has increased to 2.5 mo of extra PL in 2019 (Supplemental Figure S3). About 6 yr after PL was introduced to the genetic evaluation, we started to see an increase in PBV for HO bulls and cows ( DPR. Genetic evaluations for DPR were introduced in the United States in February of 2003 to improve reproductive performance (VanRaden et al., 2004). Historically, DPR had not been a major focus due to its low heritability (0.04), despite research showing unfavorable genetic correlations between yield and fertility (VanRaden et al., 2004). The same authors showed that from 1960 to 2000, PTA for DPR decreased for bulls in all breeds, including Milking Shorthorn.
The DPR PBV for AY bulls has been reducing since 1975 and, over this period, declined from 10.3 to −2.9% in 2016 (Figure 2). Conversely, DPR PBV for AY cows began increasing in 2005 through 2010, and a downward trend is evident thereafter (Supplemental Figure  S1). Although the PBV for DPR for AY cows trended downwards, in 2019 it is not a negative PBV (0.6%), as we see with the bulls. For BS bulls, PBV for this trait decreased from 1975 to 2011 (Figure 3). From 1975 to 2011, DPR dropped 14 percentage points to −3.7% in 2011. Since 2011, DPR PBV has increased 3.7 percentage points to 0% in 2016. Similarly, the fertility of BS cows declined to 0% in 2019 (Supplemental Figure S2). The DPR for GU bulls decreased from 10.6% in 1975 to −4% in 2001 and has not changed notably since ( Figure  4). For cows, DPR started at 16.8% in 1975 and declined to 0.7% in 2010, remaining approximately at this level ever since (Supplemental Figure S3).
Although there has been a steep decline of 12.1 percentage points in DPR for HO bulls from 1975 to 2001, minor improvements have been made since then ( Figure  5). The DPR is trending upward since receiving increased weight in NM$ in 2006. For HO cows, we see a slightly different trend (Supplemental Figure S4). Between 1975 and 2010, DPR decreased from 15% to 0.2%. Holstein DPR PBV have remained centered around 0 since 2010. For JE bulls and cows, DPR has decreased steadily since 1975. There was a 15.8-and 17.7-percentage-point reduction in DPR for JE cattle, respectively, for this range of years ( Figure 6; Supplemental Figure S5). Since 2006, a shift in the genetic trends for DPR is visible. This can be attributed to the increased selection pressure for reproductive efficiency and the widespread use of assisted reproductive technologies around this time, which led to a reduction in calving interval and days open.
LIV. Genetic evaluations for LIV were first computed in 2016 to improve the ability of a cow to stay alive while in the milking herd. Livability is an indirect measure of mortality. Cow LIV instead of mortality is reported so that positive PTA are favorable (0 = died; 100 = lived for each lactation), and as a per lifetime Guinan et al.: CHANGES IN GENETIC TRENDS IN US DAIRY CATTLE instead of per lactation basis to express LIV differences as a percentage of all cows exiting the herd . Predicted transmitting ability LIV is expressed as a probability value of a lactation not ending in death or on-farm euthanasia. Currently, LIV receives 4.4% relative emphasis in the NM$ selection index (Table 1). Due to the low number of AY bulls with daughters with greater than 10 records for LIV, there is not a clear trend since 1975 (Figure 2). For AY cows (Supplemental Figure S1), we see a slightly downward trend in LIV since the beginning of genomic evaluations in 2013, perhaps because LIV was introduced into the evaluation system in August 2016. Livability in BS bulls has not changed over time (Figure 3). The PBV for LIV in BS bulls in 2015 was −1.3. We see a similar downward trend for BS cows and in 2019 the PBV for LIV was 0 (Supplemental Figure S2). Livability has been increasing since 2011 for GU bulls (Figure 4), despite a considerable reduction between 1975 and 1996 of 7.3 percentage points. Guernsey cows have also decreased LIV since 1975, from a starting point of 12.8. However, since 205 this trend has remained positive and is slowly increasing over time (Supplemental Figure S3).
Livability was incorporated into the NM$ in 2017; however, since 1996, there has been a positive increase in LIV PBV of HO bulls after a steady downward trend since 1980 ( Figure 5). This could be due to the inclusion of PL into the index in 1994, as LIV is one of the traits that make up PL. A much clearer trend is evident in the cows due to the higher number of observations per year (Supplemental Figure S4). Livability has remained constant in JE bulls ( Figure 6) and JE cows (Supplemental Figure S5). In 2016, the PBV LIV for JE bulls was −1.9 and for in 2019 cows was 0.56. Although LIV was added to the Jersey Performance Index in 2017, the trend started to improve slightly a decade earlier, probably due to the inclusion of PL in 2002. Although LIV is a lowly heritable trait (0.013), it has a high impact on profit as it leads to a reduction in involuntary culling and therefore retention of cows with desirable traits.
Few studies have documented changes in genetic trends for multiple traits in different dairy cattle breeds in the United States, most of which were based on the animal model (Hintz and Van Vleck, 1978;Blanchard et al., 1983;Nizamani and Berger, 1996;Gengler et al., 1999). At this time, selection goals, evaluation methodologies, and production systems were vastly different to current systems. García-Ruiz et al. (2016) has shown that based on the 4-path selection model, rates of genetic gain per year increased from ~50 to 100% for yield traits and from 3-fold to 4-fold for lowly heritable traits for HO cattle. Percentage changes in our results were slightly different, where HO and JE bulls saw up to a 192% increase in production traits and 150% increase in fertility 9 yr before and after 2009 (Table  4). Differences in rates of genetic improvement between both studies could be due to the alternative methods used to estimate genetic gain as highlighted by Dechow and Rogers (2018). In that paper, the authors showed that the 4-path selection model predicts a higher rate of genetic progress in elite herds that provide AI sires than in commercial herds that use such sires.
Pedigree and Genomic Inbreeding. Inbreeding is defined as the probability that 2 random alleles at the same locus from 2 uniting gametes are identical by descent from a common ancestor (Malécot, 1948).  As the level of pedigree errors and missingness can be considerable in dairy populations, genomic information can more precisely estimate the inbreeding levels (Howard et al., 2017). Pedigree and genomic inbreeding coefficients were analyzed for bulls and cows in this data set. The same animals were used in both the genomic inbreeding trends and the pedigree inbreeding trends. This was to ensure equal comparisons among trends on the same graph. Genomic selection was expected to reduce the rate of inbreeding and increase genetic gain per generation. García-Ruiz et al. (2016) noted an increase in pedigree and genomic inbreeding in HO.
Rates of inbreeding in recent years correspond closely to rates of genetic progress and selection intensity, which tend to be highest in the JE and HO breeds. Scott et al. (2021) found that GS has also increased inbreeding rates of HO artificially inseminated bulls in Australia.
The use of young animals as sires and dams increased the rate of inbreeding per generation, which is caused by the lower accuracy of selection when young animals are selected as parents, leading to more weight on parent average and therefore more co-selection of sibs (De Roos et al., 2011). Inbreeding is currently controlled in the US dairy cattle population by adjusting the EBV for inbreeding of future progeny. The rate of inbreeding per year is much higher with genomics because of the shorter generation interval (de Roos et al., 2011). González-Recio et al. (2007) showed that cows with higher inbreeding coefficients had impaired fertility. Similarly, a study from Ireland (McParland et al., 2007) highlighted that higher rates of inbreeding had a deleterious effect on fertility. Pryce and Daetwyler (2012) predicted that the rate of inbreeding would be lower per generation in genomic breeding programs than in progeny testing and that shorter generation intervals would be responsible for higher rates of inbreeding per year.
The average pedigree inbreeding per year in AY bulls has increased since 1976 from 0.02 to 4.8% in 2016. For BS bulls, the pedigree inbreeding level was 10.4% in 2017. Guernsey bulls have increased pedigree inbreeding from 1.6% in 1975 to 5.5% in 2016 (Figure 7). The breed with the highest level of pedigree inbreeding in 2017 was HO at 10.9% (Figure 7). Although inbreeding levels are now increasing at a higher rate than before GS, genetic gains currently outweigh the effects of inbreeding depression for most traits. However, this is not true for fertility and disease resistance traits. Makanjuola et al. (2021) determined that unique runs of homozygosity regions with unfavorable effects within and across traits exist in the genome. Further research is required into alternative inbreeding measurements to disentangle and manage the negative effects on phenotypic traits at harmful regions of the genome. Pedigree inbreed-ing for JE bulls was higher than genomic inbreeding in 2017 (8.8 vs. 7.4%; Figure 7). This indicates that the American Jersey Cattle Association has successfully implemented GS as it was intended to be used. Due to the low number of genotyped bulls per year, genomic inbreeding trends cannot be inferred for AY and GU bulls (Figure 7). The level of genomic inbreeding for BS bulls in 2016 was extremely high (16.1%); however, because of data edits, this number reflects the information on a single BS bull born in 2016. For HO bulls, the level of genomic and pedigree inbreeding was 12.7 and 11.1% in 2017, respectively. Because SNP can more accurately calculate the proportion of the genome that is homozygous, it would be proper to update genomic PTA with the genomic inbreeding coefficient instead of the pedigree inbreeding coefficient.
Similarly, genomic inbreeding trends for AY and GU cows cannot be inferred due to the small number of observations per year. Although bulls born before receiving genomic evaluations were genotyped, this was not the case with cows. This is obviously due to the lower number of offspring per year and the higher cost of genotyping at that time. For BS, pedigree inbreeding increased from 1.4% in 1982 to 7.7% in 2019 ( Figure  7). Guernsey cows have shown a slow increase in pedigree inbreeding levels, reaching 7.7% in 2019 ( Figure  7). Holstein cows averaged 8.8% in 2019 and JE cows 8.86% (Figure 7). Genomic inbreeding has increased steadily since 2009 for BS cows (5.0-6.6%). The average genomic inbreeding for HO and JE cows in 2019 was 9.4 and 6.7%, respectively (Figure 7). Fortunately, the genomic inbreeding levels can be controlled by incorporating this information into the genomic PTA calculation.
Generation Interval. In 1975, the average age of US HO sires of bulls when offspring were born was 9.4 yr, whereas, in 2017, it was 2.2 yr (Figure 8). For JE, there are similar patterns, although not as low as for HO (9.8-3.2 yr). Brown Swiss bulls have also made progress in sire generation intervals from 2009 to 2016 (−3.9 yr); however, very little progress has been made for AY and GU since 1975, probably because of the later implementation of genomics and the smaller number of genotyped selected candidates compared with the other breeds. For cows, the reduction in generation interval has been somewhat slower. For sires of HO cows, generation interval has decreased from approximately 6.8 yr in 2009 to 3.9 yr in 2019. Jersey and BS also saw a reduction in the sire generation interval of 2.3 and 2.4 yr since 2009. Although it could be interpreted that longevity of cows is decreasing, this is not the case. This reduction indicates that age of first calving is decreasing (2020: HO, 24 mo; JE, 22.8 mo). Longevity receives emphasis in selection indexes through PL and LIV; however, management decisions are driving generation intervals. Additionally, one of the benefits of genomics is to decrease the generation interval, which is highlighted in the generation interval graphs. The advancement and increased adoption of reproductive technologies have also contributed to genetic gain. Generation intervals for the parents of AY and GU cows have not decreased since 1975 (Figure 8). Perhaps this is an area AY and GU breeds could benefit significantly from genomic technologies. Incentives could encourage data collection, genotyping, and the widespread use of young genomic bulls and heifers. García-Ruiz et al. (2016) reported generation intervals of 2.4, 5.0, 2.6, and 3.6 yr for sires of bulls, sires of cows, dams of bulls, and dams of cows, respectively, for HO offspring born in 2015. In contrast, our results show these averages as 2.2, 3.9, 2.2, and 3.1 yr for HO bulls born in 2017 and cows born in 2019. Given the additional data, generation intervals have continued to decline for HO. These findings confirm the findings of García-Ruiz et al. (2016) and highlight the continued improvement in genetic gain since 2009, which is ultimately driven by producer decisions regarding selection of parents of the next generation. Schaeffer (2006) projected the total generation interval to be 9.75 yr due to GS, and Hagan and Cue (2020) showed that in the Canadian HO population, the total generation interval was 13.5 yr in 2016. Although Schaeffer (2006) predicted the generation intervals of dams of cows to have slower improvements, recently, there has been a rapid  increase in the genotyping of dams of cows for management purposes along with a reduction in genotyping costs. This is reflected in the reduction in generation interval in dams of cows for most breeds in our study.
General Observations. Remarkable improvements have been made since the implementation of genomic evaluations in the United States in 2009. Bulls are making more progress than cows as bulls have more progeny, and a greater proportion of bulls (24%) are genotyped versus cows (3.9%) since 1975. Genetic gain for protein yield increased 192% for HO bulls born between 2009 and 2017, whereas genetic gain for PL increased 50% (Table 4). Because of the increased use of young, genotyped bulls and heifers, a decrease in generation interval followed the implementation of GS for all breeds, except AY and GU. The generation interval before and after GS changed from 7.1 to 2.2 yr in HO, 7.3 to 3.2 in JE, 8.3 to 4.4 in BS, and 7 to 5.1 in AY for sires of bulls (Figure 8). For dams of bulls, the generation interval before and after GS changed from 4 to 2.2 for HO, 5.2 to 2.3 yr for JE, 5.8 to 3.6 for BS, and 4.1 to 4.3 yr for AY (Figure 8). The generation interval before and after GS changed from 6.8 to 3.9 yr in HO, 6.4 to 4.0 in JE, 7.2 to 4.9 in BS, and 6.6 to 6.3 in AY for sires of cows (Figure 8). For dams of cows, the generation interval before and after GS changed from 3.8 to 3.1 for HO, 3.7 to 3 yr for JE, 4.3 to 3.7 for BS, and 4.5 to 4 yr for AY ( Figure 8). As genomic evaluations were only released for GU in 2016, it is too early to see the benefits as clearly as in other breeds, and this should be reviewed in the next 5 yr for both GU and AY. Applying a negative weight to inbreeding in the NM$ (Table 1) could result in an increased ranking of colored breed bulls in North America. More focus on communication between the industry and producers on how to use genomic PTA can help to control inbreeding levels. Progress for smaller breeds will need to be continually monitored to assess the level of genetic gain post genomic evaluations.

CONCLUSIONS
This research provides benchmarks for breed associations to evaluate the updates in genetic evaluations methodology in the genomics era. Genomic information has mainly benefited HO and JE because the implementation of GS for these 2 breeds happened as early as 2009. Another reason is that HO and JE have larger reference populations and more refined selection indexes. However, it does not mean the other breeds are not benefiting from GS. In the short time period investigated, there is success in most traits for the colored breeds. Colored breeds in the United States could significantly benefit from genotyping and phenotyping more of the population and increasing the adoption of genomic methods. This study should be updated in another 5 yr to provide these breeds further opportunity to develop.

ACKNOWLEDGMENTS
This study was partially funded by the Council on Dairy Cattle Breeding (CDCB, Bowie, MD) and Agriculture and Food Research Initiative Competitive Grant no. 2020-67015-31030 from the US Department of Agriculture's National Institute of Food and Agriculture (Washington, DC). The authors thank the CDCB for providing access to the data. The contribution of dairy producers who supplied data through their participation in the Dairy Herd Improvement program and the Dairy Records Processing Centers that edited and relayed information on to the CDCB are also acknowledged. Comments from Shogo Tsuruta, Yutaka Masuda, Jorge Hidalgo, Felipe Ruiz-Lopez, and Adriana Garcia-Ruiz are gratefully acknowledged. We also acknowledge the two anonymous reviewers for thoughtful suggestions and improvements. The authors have not stated any conflicts of interest.