The study area for simulating alternative breeding programs (BPs) focused on smallholder farmers owning crossbred cattle in specific zones of the Oromia Region (Arsi, West Shewa, and North Shewa) and the North Shewa zone of the Amhara Region, Ethiopia. These areas are home to a significant cattle population, estimated at 7,958,831 head [2,253,959 in West Shewa, 1,676,748 in north shewa zone of oromia region (NSHORO), 2,545,778 in Arsi, and 1,482,346 in north shewa zone of Amhara region (NSHAMA)] according to CSA (2017). Crossbreeding practices have been widely adopted in these areas over the past 4 to 5 decades, resulting in approximately 377,729 crossbred cattle, accounting for 35.2% of the national crossbred cattle population (CSA, 2017). The geographical proximity of the study areas, within a 150 km radius of the capital city, Addis Ababa, facilitates the supply of inputs and outputs. For further details on the study area’s geographic references and climate information, refer to Direba et al. (2020).

The simulation of alternative breeding programs (BPs) involved considering genetic parameters, economic values of breeding goal traits, biological and technical parameters, and cost parameters. The goal traits included lactation milk yield (MY), age at first calving (AFC), calving interval (CI), and herd life (HL). Five different BPs were designed and compared within the study area while taking into account the available breeding unit at Holeta Research Center, existing crossbreeding practices, and future prospects of dairy research and development in Ethiopia. These breeding programs included conventional on-station progeny testing (SPT), conventional on-farm progeny testing (FPT), conventional on-station and on-farm progeny testing (SFPT), genomic selection (GS), and genomic progeny testing (GPT). The first three conventional BPs were simulated without utilizing genomic information. A total of 143,576 crossbred cows were assumed to model the alternative BPs. The data collection and selection were presumed only in the breeding unit (BU), and no data were collected from cattle in production units (PUs) in all BPs.

In the FPT, SFPT, GS, and GPT breeding programs, smallholder farmers are expected to participate by providing cows for progeny testing, allowing selected cows or male calves for genotyping, selling the selected male calves to the breeding program, and maintaining records. An on-farm survey indicated that smallholder farmers own an average of 2.65 cows (Direba et al., 2020); approximately 1,000 farmers would participate in implementing the progeny testing scheme. Additionally, around 25 to 27 data recorders should be employed to gather data from the cows in a breeding unit. Over time, the responsibility of data recording in the BU can be transferred to the cattle owners through training and creating awareness about the potential advantages of the breeding program.

The level of exotic inheritance for sires was set at 75% for all alternative BPs. As a result, crossbred generations will be stabilized at 75% exotic inheritance, and thus, adaptive and productive synthetic breeds will be developed in the long term. The comparison of these BPs considered predicted genetic gain, generation interval, the accuracy of selection, and discounted cost, return, and profit.

The alternative BPs were modeled using ZPLAN+ software (Täubert et al., 2011; Vit Verden, 2011). The program follows the deterministic approach that considers population mean and variance. The evaluation of the BP was based on the gene flow method and selection index procedure. That means the fraction of the genes of bulls and cows in successive generations and its associated contribution of breeding values will be considered. ZPLAN+ predicts the genetic gain for the breeding goal traits. In addition, from an economic perspective, the program calculates the discounted cost, return, and profit of BP during the investment. The program requires phenotypic and genetic parameters of breeding goal traits, economic value of traits, fixed and variable costs of selection groups, and other biological coefficients of the study population as input.

The genetic parameters and economic values of breeding goal traits were determined for the study. Estimates for economic value, phenotypic standard deviation, correlations, heritabilities, and repeatability were obtained from survey data collected in the study areas and long-term data on dairy cattle at the HARC (Table 9; Table 10). More detailed genetic parameter information can be found in Direba et al. (2022). For genomic information, input parameters were sourced from relevant literature reports due to the absence of estimates for the target population. The accuracy of polygenic breeding values and the number of animals in the reference population for MY were taken from Erbe et al. (2012). Accuracy estimates for AFC, CI, and HL, along with associated numbers of animals in the reference population, were obtained from the studies of Boison et al. (2017) and Haile-Mariam et al. (2013).

Productive life, reproductive cycle, age at first reproduction, and survival rate for both zebu and crossbred cows with different levels of exotic inheritance were also derived from long-term data collected at HARC (Direba et al., 2022). Notably, zebu cows exhibited a longer productive life (7.44 years) than crossbred cows, which could be attributed to the adaptive qualities of indigenous breeds and the specific objectives of the farming systems. Differences in reproductive cycles were observed among the selection groups, with the F1 generation displaying a shorter cycle (1.26 years) than other groups. This indicates performance variations within the different selection groups (Table 11). Biological parameters based on 75% exotic inheritance were applied to the CBU and CPU, as admixture analysis revealed a similar level of exotic inheritance (78%–79%) for cows managed by smallholder farmers (Strucken et al., 2017; Netsanet et al., 2021).

Table 12 provides a detailed breakdown of the annual variable and fixed costs per animal for the simulated alternative BPs. The study assumed a 25-year investment period, with a 9.5% interest rate applied to both costs and returns. It is important to note that all the costs, returns, and profits estimated in this study were specifically related to the additional expenses incurred and profits obtained as a direct consequence of implementing the breeding program (Nitter et al., 1994; Tadele et al., 2011; Kahsa et al., 2012). Fixed costs were defined as constant expenses related to salaries, overhead costs, and computers. These costs encompassed the purchase of computers, overhead expenses (including training and monitoring costs), and salaries for labor, veterinarians, recorders, artificial insemination (AI) technicians, and animal breeders. Variable costs represented the total expenses incurred for selecting, testing, and choosing animals for the breeding program (Täubert et al., 2011; Vit Verden, 2011). Additional costs for items such as feed, genotyping, bull purchases, ear tags, veterinary services, and other administrative expenses that arose due to the BP were considered variable costs.

The costs associated with the BPs varied depending on the specific scenarios. For example, in the case of conventional on-SPT and conventional on-SFPT, the cost of feed for cows was included because these cows were entirely used for the breeding program. However, in other BPs, where cows were part of normal dairy business processes, no additional feed costs were considered. The cost of feed for bulls was included in all BPs because bulls were recruited from the on-station breeding unit or purchased from on-farm breeding units and managed at the station until sufficient semen was collected.

In terms of genotyping, the cost for genotyping cows and bulls was 1,916.35 ETB per animal (1 USD = 39.55696 ETB) for GS and GPT. Human resources required for data collection, animal health management, supervision, and data analysis were also factored into the costs. For BPs with on-farm data gathering components (FPT, GS, and GPT), the cost included 25 data recorders (6,500 ETB per person per month) and three animal breeders (16,437 ETB per person per month). In the simulation of SPT, the cost included 20 daily laborers (1,500 ETB per person per month), two data recorders (1,890 ETB per person per month), and three animal breeders (16,437 ETB per person per month). The cost of veterinary services per animal was 94.04 ETB per year for all BPs. Additionally, the salaries of two veterinarians (16,437 ETB per person per month) and two AI technicians (6,500 ETB per person per month) were assumed for the analysis of SPT and SFPT, as these programs required full-time animal health management and AI services.

Approximately 500,000 ETB per year were assumed for training, monitoring, and other administration costs for SPT. This cost was escalated to 1,000,000 ETB per year for each of the remaining BPs, as more supervision was required to mobilize farmers and provide training for data recorders and farmers participating in on-farm breeding units. However, only the cost of AI service was included for CPU because the genetic transfer was through AI, and all farmers should use AI for the practical implementation of BP.

Costs related to the price of animals, animal health services, and AI services were obtained from the survey conducted with smallholder farmers (Direba et al., 2020). The price to purchase bulls/male calves was enhanced by 25% for non-genotyped and 50% for genotyped bulls/male calves to provide better market value for genetically merited bulls and encourage farmers to sell selected young bull/male calves to the BP. Estimates of daily labor and salaries were taken from the HARC payment standard. Feed requirements and associated costs were derived from the HARC feeding standards and market costs of 2021.

The results in Table 13 show estimates of genetic gain (GG) of breeding goal traits per year and per generation. The GGs predicted for lactation MY per year were 34.52 kg, 49.63 kg, 29.35 kg, 76.16 kg, and 77.51 kg for SPT, FPT, SFPT, GS, and GPT, respectively. The corresponding GGs per generation were 209.27 kg for SPT, 273.22 kg for FPT, 174.30 kg for SFPT, 362.33 for GS, and 426.71 for GPT. The genomic BPs attain at least 26.53 kg more GG per year than the conventional BPs.

The GGs calculated for age at first calving (AFC) were 1.19 and 7.22 days for SPT, 0.97 and 5.37 days for FPT, 0.69 and 4.12 days for SFPT, 1.11 and 5.26 days for GS, and 0.90 and 4.94 days for GPT per year and generation, respectively. A lower GG of calving interval (CI) was recorded in SFPT (1.20 days per year and 7.13 days per generation) than other BPs (varied from 1.57 to 2.35 days per year). The value of GG obtained for CI was almost similar among FPT, GS, and GPT. The estimates of GG for herd life (HL) were very low (0.06–0.12 days per year) and almost similar among all alternative BPs.

Generation interval (GI) refers to the age of parents when their replacement offspring is born. When the GI is low, the transfer of genetic gain from generation to generation becomes rapid. Hence, there will be higher return and genetic progress in breeding goal traits for the target population. The values of predicted GI for conventional BPs were close to each other and ranged from 5.51 years to 6.06 years (Table 14). However, GI was reduced by 21% in GS compared to SPT and reduced by 19% compared to SFPT. Table 14 summarizes the accuracy of selection for test and progeny-tested bull in different BPs. The accuracy of selection calculated for test bulls was 0.33 for SPT, 0.34 for FPT, 0.37 for SFPT, and 0.43 for both genomic BPs. Furthermore, the accuracies of the selection of progeny-tested bull estimated here were 0.85 for SPT and 0.99 for other BPs.

The ultimate goal of the BP is to ensure the profitability and sustainability of the dairy business for the target population. As indicated in Table 12, the total costs calculated for breeding units varied among simulated BPs. The annual cost for BUs in ETB was approximately 15.15 million for SPT, 5.00 million for FPT, 22.07 million for SFPT, and 6.67 million for each genomic BP. ZLAN+ distributes costs and returns for the entire cow population included in the BP as the genetic gain obtained at the breeding unit is transmitted to the whole population (Täubert et al., 2011; Vit Verden, 2011). Furthermore, the program discounted the cost of BP considering the interest rate, generation interval, and number of animals in different selection groups.

Table 15 contains discounted cost, return, and profit per animal per generation and year. The discounted costs of BPs per animal per generation in ETB were 45.85, 52.71, 71.51, 190.28, and 172.57 for SPT, FPT, SFPT, GS, and GPT, respectively. The corresponding discounted returns were ETB 1,559.07 for SPT, 2,475.83 for FPT, 1,765.54 for SFPT, 5,741.69 for GS, and 3,969.88 for GPT. The discounted profit of the BPs ranged from ETB 1,513.22 in SPT to 5,551.40 in GS per cow per generation.

Among the trait goals, milk yield had the highest GG in all BPs. This could be attributed to the availability of higher genetic variance of this trait in the study population and better heritability than other traits. Furthermore, the higher GG of MY recorded in this study could create a good opportunity for the future sustainability of BPs as MY accounts for 69% of the relative economic value of the dairy business in the study area. The result revealed that GG was positive for all BPs. However, the positive GG obtained for AFC and CI is undesirable. The overall results indicated that the application of any of the five BP could bring genetic progress in the crossbred population of the study area.

Genetic gain for MY estimated in the present study concurred with the report of other studies on different dairy breeds (Börner and Reinsch, 2012; García et al., 2016; Fedorovych et al., 2021). Opoola et al. (2020) and Tobias et al. (2010) predicted 245 kg–734 kg and 366 kg–410 kg MY GG per generation by modeling different BPs for dairy cattle in South Africa, Zimbabwe, and Kenya, respectively. Kudinov et al. (2018) calculated 56 kg–59 kg MY GG per year for black and white dairy cattle in Russia. However, a higher estimate (512 kg per generation) was reported by Mulder et al. (2005) using genetic evaluation data collected from different breeds.

Further comparison among BPs indicated that genomic systems noticeably increase the GG of MY. A better GG was obtained by GPT and GS than by conventional BPs. The results imply that the genetic progress of MY can be enhanced by 53%–164% by supporting the BP with genomic information. Scott et al. (2021) estimated that the rate of genetic gain increased by about 160% in Holsteins and 100% in Jersey cattle when genomic information was included in the progeny testing scheme in Australia. Consistent with the present study, the substantial contribution of genomic BPs was reported by several authors (Panigrahi and Parida, 2012; Thomasen et al., 2014; Boisonn et al., 2017).

Values calculated for AFC and CI are lower than the report of Opoola et al. (2020), who found 41–65 days for AFC and 20–42 days for CI GG per generation using data from Zimbabwe and South Africa. Similarly, Tobias et al. (2010) noted higher genetic response (90–101 days per generation) for AFC in Kenya than our estimate. Even though it was low, the GG obtained for AFC and CI is undesirable, as the economic value of these traits was negative when their GG became positive. Furthermore, selection only for MY may adversely prolong CI and AFC as the genetic correlation of MY with these two traits is positive. Consistent with our findings, García et al. (2016) calculated a GG of 0.03–0.88 days for HL per year for Holstein cows in the United States. It can be noted that the variation observed among BPs for AFC, CI, and HL looks insignificant. Although the GGs of AFC and CI were low, care should be taken in the selection index to further reduce the GG of these two traits.

Genomic selection (GS) reduces the generation interval (GI) by up to 21% compared to conventional BPs. This is mainly because bulls in GS were selected and transferred genetic merit at an early age. Estimates of GI in the present study were higher than those of other studies in developed countries. Täubert et al. (2011) calculated 5.02, 3.44, and 4.64 years for conventional, genomic, and combined conventional-genomic breeding programs, respectively. Similarly, Garcia et al. (2016) discussed a reduction of GI from 7 years to 2.5 years by implementing genomic selection in dairy cattle in the United States. The difference with the current result might be due to the lack of application of appropriate genetic improvement programs in Ethiopia and farming practices, as the herd life of dairy cows was longer in the Ethiopian condition.

The accuracy of selection calculated for test bulls is comparable to the report of Brown et al. (2016), who found 0.28 to 0.41 prediction accuracy for the selection of crossbred dairy cattle in East Africa. However, Täubert et al. (2011) found better accuracy (0.54) for test bulls in Germany than the present study in conventional BPs. The improvement of accuracy from 0.3 to the 0.43 obtained in the present study is attributed to the increase in number of cows (2500) in the BU or the reference population. The result also indicated that genotyping bulls could enhance the accuracy of the selection of test bulls by 16%–30%. Similar conclusions are reported by other authors (Haile-Mariam et al., 2013; Garcia et al., 2016; Aliloo et al., 2018). The values of accuracy of selection of progeny-tested bulls in this study (0.85–0.99) are close to the 0.89 estimated for Holstein cattle in Germany (Täubert et al., 2011). Hayes et al. (2007) also reported a 0.84 accuracy using marker-assisted (haplotype) selection with phenotype records for cattle in Australia.

The discounted cost in genomic BPs looks higher than that of conventional systems, but it was offset by the high return in these systems. Among conventional progeny testing schemes, FPT generated 53% more profit than SPT and 37% more than SFPT. The genomic systems enhanced the profit by at least 63% compared to the conventional counterparts. The GS outperformed the SPT, SFPT, and FPT by 266%, 227%, and 138%, respectively. Similarly, the profit obtained from GPT was better than that of all conventional BPs. On the other hand, the profitability of GS was reduced by 31% when it was supported by progeny testing. The substantial difference in profit observed among BPs is attributed to the BP scenario. The costs of SPT and SFPT were inflated mainly due to the cost of animal feed and salary, as these BP maintained many cows at the station. The other BPs (FPT, GS, and GPT) kept about 20 sires at the station, which significantly reduced the cost required for animal feed and labor. However, considering the contribution of the BPs to the entire population, the calculated cost looks reasonable.

The higher profitability of genomic systems probably resulted from lower feed and labor costs for animal management and a lower generation interval, which leads to rapid genetic gain. The contribution of progeny testing BPs and further enhancement by genomic BPs for better profitability of the dairy business was inferred in several reports (Börner and Reinsch, 2012; Weller et al., 2017; Mrode et al., 2019; Newton and Berry, 2020; Scott et al., 2021). For instance, discounted profits of 18–26 USD per generation in Kenya (Kahi and Nitter, 2002), 29.92 euros per year for Danish Jersey dairy cattle (Thomasen et al., 2014), and 238 to 532 euros per generation for Holstein dairy cattle in Germany (Täubert et al., 2011) were reported for different BPs. The difference in profit and values of other parameters of the present result with other reports could be due to variations in economic values and the number of traits included in BPs, the size of the population included in the BPs, and parameters fitted in the model.

All suggested BPs showed a positive profit. However, each has its own advantages and limitations. The SPT is highly suitable for record keeping and selection of animals as the BU is maintained at the station. However, the number of cows in the BU was very low to produce and recruit test bulls. This highlighted that sires were evaluated with a very low number of daughters (3–4 daughters per sire). As a result, the selection accuracy of test bulls and selected bulls was lower than other BPs. On the other hand, the annual total cost of SPT was also higher than FPT and genomic BPs.

The result showed that the GG and profit obtained by FPT were better than SPT and SFPT. Likewise, the accuracy of sire selection was better than SPT and comparable with SFPT. This BP was superior to SPT and SFPT by 53% and 37% discounted profit, respectively. Furthermore, farmers practicing FPT participate in a community that can enhance the success of a BP. The practical limitations of this BP are 1) lack of a record-keeping system at the farmer level and establishing pedigree record may take longer; 2) in the absence of a pedigree record, selection based on phenotypic performance may not bring genetic progress; and 3) it was difficult to estimate the level of exotic inheritance for sires as there were no pedigree records at the farmer level.

The combined on-SFPT could solve the shortcomings of SPT and FPT. The on-station part supports the BP by a pedigree record, and the on-farm part avails a sufficient number of cows for progeny testing. Community participation in this BP could also create a smooth environment (enhance cooperation of farmers) for practical implementation of this BP as farmers provide cows for progeny testing and support recording and benefit from the BP. Increasing the number of cows for progeny testing by participating dairy farmers has improved the accuracy of test bull selection from 0.33 to 0.37 and selected bulls from 0.85 to 0.99. The profit was increased by 11% compared to SPT. However, it was decreased by 37% compared to FPT. The GIs of SPT and SFPT were greater than all BPs. In addition, the annual cost of SFPT was also much higher than all other BPs.

As expected, the potential advantages of genomic systems were enormous. This system overtakes the conventional counterpart by at least 63% profit in GPT and 138% when GS was compared. The annual cost of genomic BPs was decreased by about 56%–69% compared to SPT and SFPT. The GI was reduced by up to 21% by using genomic BPs. Furthermore, the accuracy of test bull selection was improved to 0.43 in genomic BPs. The practical challenge of genomic BPs could be the establishment, phenotype data collection, and genotyping of the reference population.

In conclusion, all alternate BPs produced a profit and contributed to MY’s genetic gain. These imply that implementing any of the alternative BPs can bring genetic progress. Genomic BPs overtake the conventional BPs in terms of genetic gain, generation interval, accuracy of sire selection, and discounted profit. Hence, establishing GS and gradually supporting it with progeny testing (GPT) are recommended in that order as the main way forward to attain better genetic progress in dairy farms in Ethiopia and similar scenarios in other tropical countries. However, until compulsory conditions are in place for genome selection, SFPT can be considered more practical. For the success of the breeding program, relevant governmental and non-governmental institutes should be engaged, and their roles and responsibilities in the implementation of the breeding program should be defined. It is also crucial to enhance community participation through training and supervision, as well as build the capacity of a biotechnology laboratory to facilitate the genotyping of the reference population and test bulls.