Jeju Black Cattle (JBC) is one of the indigenous Korean cattle breeds that has been raised for centuries on Jeju Island [1]. This breed is characterized by its unique black coat, compact body size, and strong adaptability to harsh environmental conditions. It possesses unique genetic and cultural value that sets it apart from the more widely recognized Korean Hanwoo cattle [2]. Historically, JBCs were commonly used for draft and agricultural purposes as well as for their beef. However, their population drastically declined due to their slow growth rate and the national focus on improving Hanwoo for commercial beef production [3]. As a result, this indigenous breed was largely ignored by both the government and farmers for genetic improvement. By the late 20th century, the population of JBC had decreased to critically low levels, placing the breed on the edge of extinction. Conservation efforts began following the identification of a small number of remaining animals in 1992 from isolated farms on Jeju and nearby islands [4]. Since then, structured breeding and conservation programs have been implemented to support population recovery and maintain genetic diversity. The Food and Agriculture Organization (FAO) officially recognized JBC as a distinct Korean cattle breed in 2004. In 2013, the Korean government designated JBC as a Natural Monument (No. 546). It also acknowledges its importance as a genetic resource and as part of the nation’s cultural heritage [5].

Despite these conservation efforts, the practical use of JBC for economically important traits such as carcass yield and meat quality remains limited. In contrast, Hanwoo has undergone continuous genetic improvement since the 1960s through coordinated programs led by government and research institutions, with selection focused on productivity and consumer-driven meat quality traits [6]. On the other hand, JBC has not yet received the same level of genetic improvement effort. Most breeding decisions for JBC are still based on visual traits, limited pedigree information, and small population data, which inhibits accurate and effective selection for desirable characteristics [3,7]. Considering the limited population size and restricted genetic diversity of JBC, more precise and evidence-based approaches are needed. These should aim to improve performance while maintaining genetic resources. Environmental conditions also play a significant role. On Jeju Island, the subtropical maritime climate leads to higher temperatures and humidity, which can negatively affect early growth traits such as birth weight and average daily gain. These effects may extend to carcass weight (CWT) and meat quality in later stages [8]. Such environmental variation makes it difficult to separate genetic potential from external influences. This highlights the need for genomic approaches that can more effectively partition genetic and non-genetic sources of variation in JBC.

Genomic selection (GS) provides an effective framework to accelerate genetic gain while controlling inbreeding, which is particularly relevant for small or conserved populations such as JBC. It uses genome-wide molecular markers, most often single nucleotide polymorphisms (SNPs), to estimate the genetic potential of animals at an early stage. This approach reduces reliance on phenotypic records and complete pedigree information [9]. The genomic best linear unbiased prediction (GBLUP) method has emerged as a widely accepted approach in GS, where the traditional pedigree-based numerator relationship matrix (A matrix) is replaced by a genomic relationship matrix (GRM; G matrix) constructed from dense SNP genotype data [10]. This allows for more accurate estimation of genomic estimated breeding values (GEBVs), especially in breeds with limited population sizes or incomplete pedigree records [11]. While GS has significantly transformed breeding programs in commercial cattle breeds such as Holstein and Hanwoo, its implementation in JBC remains at an early stage.

To date, few studies have been conducted to evaluate the genomic architecture or predict GEBVs for carcass traits in JBC [3,11]. Furthermore, there is a lack of comprehensive studies assessing the accuracy of genomic predictions in this breed, which is essential for developing effective breeding programs. The limited reference population and sparse genomic information available for JBC pose additional challenges to implementing GS.

One well-established solution to improve genomic prediction accuracy in breeds with small populations is to incorporate a larger, genetically related breed into the reference population [12,13]. In the present study, Hanwoo was selected as the reference population for three main reasons. First, JBC and Hanwoo share common ancestral origins as native Korean cattle breeds. This shared ancestry suggests a degree of genetic relatedness, which may allow marker-trait associations established in Hanwoo to remain informative for JBC [2]. Second, Hanwoo has a well-documented cattle population in Korea, supported by a national recording system with extensive phenotypic and genotypic data for the carcass traits examined in this study [14]. Third, previous studies have shown that using a larger, genetically related breed as a reference population can substantially improve GEBV accuracy in small or endangered breeds, especially when the target breed lacks a sufficiently large reference population of its own [12]. Based on these considerations, a multi-breed reference population approach was adopted. This allowed us to evaluate whether including Hanwoo data could improve genomic prediction accuracy for JBC carcass traits, while also comparing it against a reference population composed solely of JBC animals.

To address these critical gaps, the present study aims to estimate GEBVs for carcass traits in JBC utilizing high-density whole-genome SNP chip data and to rigorously evaluate the accuracy of genomic predictions across different reference populations using a single-trait animal model. Specifically, this research seeks to develop an optimized reference population by integrating existing phenotypic and genotypic datasets, estimate GEBVs for key carcass traits including CWT, eye muscle area (EMA), backfat thickness (BF), and marbling score (MS), and assess the predictive performance of genomic models for different reference populations. Taken together, these objectives aim to build a practical foundation for implementing GS in JBC. Such a framework could support long-term conservation efforts, improve meat quality and productivity, and strengthen the economic viability of Jeju’s local livestock industry. Ultimately, this work addresses a clear gap in genomic research on this endangered native breed, with direct implications for selection decisions and long-term breeding management. It also contributes to preserving Korea’s agricultural biodiversity by applying genomic tools to a historically neglected population.

The estimates of heritability for carcass traits in JBC differ across reference population composition. When Hanwoo steers alone were used as the reference, the heritability estimates were generally higher compared to those obtained by combining Hanwoo steers and cows or using only JBC animals. This suggests that using a more genetically homogeneous reference population may increase the accuracy of genetic parameter estimation, likely due to reduced environmental heterogeneity [21]. The slight reduction observed in heritability when Hanwoo cows were included in the reference population indicates that adding animals raised under different management or environmental conditions can inflate residual variance and weaken the genetic signal [22]. Since steers and cows are typically raised under different production systems and face distinct physiological demands, their inclusion can introduce additional environmental variation. This added variance may capture the detection of additive genetic variance effects, ultimately lowering the accuracy of heritability estimates. This finding highlights the importance of carefully considering animal classes when expanding reference populations, even within the same breed. The heritability estimates in this study ranged from low to moderate, which aligns with previously reported values for the same breed. For instance, Haque et al. reported estimates of 0.24 to 0.40 for carcass traits in JBC [11]. Another study estimated the heritability values of 0.42 for CWT, 0.27 for EMA, 0.26 for BF, and 0.48 for MS [3].

Although there are limited reports specifically on JBC carcass traits heritability, comparisons could be drawn with Korean Hanwoo and other cattle populations. In Hanwoo, moderate heritability estimates have been observed across various genomic evaluation methods, including GBLUP (0.30–0.39), BayesA (0.33–0.41), BayesB (0.29–0.36), BayesC (0.31–0.38), and BayesCPi (0.32–0.39), as reported by [23]. In contrast, Alam et al. [24] reported higher heritability values using traditional pedigree and phenotype-based models in Hanwoo steers. Similarly, Mehrban et al. [25], analyzing data from 590 Hanwoo young bulls, found heritability estimates of 0.31 for CWT, 0.43 for EMA, 0.49 for BF, and 0.61 for MS. Moreover, Bhuiyan et al. found heritability values of 0.51 for CWT, 0.45 for EMA, 0.29 for BF, and 0.22 for MS in Hanwoo slaughtered at or under 30 months of age [26]. This difference is expected, as heritability estimates based on genomic data, especially those using SNP markers, are often lower than those calculated from pedigree-based models. This observation is consistent with findings by Yang et al. [27]. Notably, the JBC-only reference group maintained relatively stable heritability estimates, particularly for MS and BF. This finding indicates the potential of breed-specific reference populations, even those of smaller size or with limited diversity, to generate reliable genetic parameter estimates for economically important traits. Moreover, the consistently high heritability of MS across all groups underscores the strong genetic control of this trait. This stability suggests that MS can be a key selection target for JBC breeding programs, with the potential to respond well to GS regardless of reference population composition.

The GEBV estimations for carcass quality traits exhibited distinct variability patterns, emphasizing the importance of reference population composition on genetic predictions. However, the breeding values reported on the carcass traits for the JBC are limited compared to other beef cattle like Hanwoo. A recent study reported GEBVs for the CWT, EMA, BFT, and MS traits as 7.348, 1.515, −0.355, and 0.040, respectively using single-step GBLUP [28]. Importantly, another study using ssGBLUP resulted in significantly higher GEBVs in a Hanwoo full-sib family than our study, where the values for the CWT, EMA, BFT, and MS were 21.5±27.4, 5.5±5, −0.9±1.4, and 0.9±0.8, respectively [29]. A very recent study revealed a wide range of GEBVs for the CWT (−71.8 to 106), EMA ( −19.2 to 22.5), BF (−9.08 to 7.33), and MS (−3.30 to 3.67) traits in Korean Hanwoo cows using GBLUP [30]. In this investigation, HSJB showed negative value in CWT and EMA, with inconsistent fat and marbling quality, indicating a gradual loss in overall carcass size and muscularity on JBC over time. Similarly, HAJB showed larger declines in CWT and EMA. These values decreased further after adding cows to the reference population. This is likely because cows have lower CWT and EMA. Additionally, steers showed higher phenotypic averages, GEBV values were estimated after adjusting for fixed effects such as farm, year, season, and slaughter age. These adjustments likely reduced the observed advantage of steers, resulting in similar or slightly lower mean GEBV values compared to cows. On the other hand, JBJB exhibited the higher GEBV variability across all four traits compared to HSJB and HAJB. This may be because the population had low genetic diversity. It may also be more affected by environmental factors due to a homogeneous reference population. The yearly GEBV trends shown in Figure 3 reveal clear differences among the three groups, highlighting the strong effect of reference population composition on genetic evaluation. JBJB shows a more consistent and slightly positive trend across most traits, especially for MS, while HSJB and HAJB display relatively stable or declining patterns for CWT and EMA. This suggests that within-breed evaluation captures genetic merit more effectively, whereas using different breeds in the reference and test populations may reduce prediction accuracy [31].

GS is extensively applied in the beef industry, especially for improving carcass and meat quality traits [32]. Numerous research efforts have assessed how accurately genomic prediction models perform across various breeds and traits [33]. In Hanwoo steers, the GBLUP method achieved GEBV accuracies between 0.78 and 0.81 for the CWT, EMA, BFT, and MS [34]. On the other hand, Lee et al. [35] reported comparatively lower GEBV accuracies in the Hanwoo population, ranging from 0.54 to 0.78. Another investigation observed GEBV accuracies of 0.7±0.0 for all carcass traits (CWT, EMA, BF, and MS) with ssGBLUP in a Hanwoo full-sib family [29], which partially corresponds with our results. Utilizing 919 Hanwoo cattle, Kim et al. [36] also found higher GBLUP prediction accuracies for CWT (0.779), EMA (0.758), BF (0.766), and MS (0.791), which are partly consistent with our findings.

Our study found ACC_T lower than a recent investigation using GBLUP in Hanwoo cows with values of 0.73, 0.68, 0.68, and 0.74 for CWT, EMA, BF, and MS, respectively [37]. Another recent study observed ACC_T ranging from 0.75–0.78 for CWT, 0.74–0.76 for EMA, 0.75–0.77 for BF and 0.78–0.81 for MS; and ACC_R ranging from 0.70–0.74 for CWT, 0.64–0.69 for EMA, BF 0.59–0.65 and 0.73–0.78 for MS respectively in Hanwoo cows [30], which is significantly higher than our findings. The possible reason for these differences is that their study used within-breed reference and test populations. In contrast, our HSJB and HAJB used different breeds as reference and test populations. Furthermore, lower genomic prediction accuracies for EMA, BF, and MS ranging from 0.27 to 0.30 were also reported in Korean Hanwoo cattle using a 50K SNP chip [38]. This study found that MS had the highest GEBV accuracy while EMA had the lowest, which is probably due to variations in trait heritability [35]. Several studies reported slightly higher ACC_R in the Hanwoo population than our findings with a range of 0.43–0.54 for the carcass traits [35,39], which might be because of differences in reference population size and genetic relationships [30]. However, in JBJB, where the JBC population alone served as the reference, the accuracy was higher than that reported by Lee et al. [28] in Hanwoo. This aligns with our observation that MS showed the highest accuracy while EMA had the lowest.

The GEBV prediction accuracy exhibited unique patterns across reference populations. HSJB displayed moderate ACC_T due to genetic uniformity, but lower ACC_R for all carcass traits, this might be for data quality, population structure, and genetic architecture. Despite a minor reduction in ACC_T, HAJB exhibited comparatively higher ACC_R, particularly for CWT and EMA, owing to increased genetic diversity. Among all three groups, JBJB showed the higher accuracy in both ACC_T and ACC_R. This suggests that within-breed evaluation provides more reliable prediction accuracy. The higher genetic relatedness within the JBC population likely contributed to this improved performance.

The LD results further support these findings. Although both Hanwoo and JBC showed similar LD decay at short distances, JBC maintained higher LD at longer distances and showed slower decay at the (r²<0.1) threshold. The overall mean r² was also higher in JBC. This suggests more persistent LD in JBC, which may be due to a smaller effective population size or stronger genetic relatedness within the breed. However, even with higher LD, the multibreed prediction accuracy remained low. This may be because LD phase and allele frequencies differ between Hanwoo and JBC. As a result, marker effects estimated in one breed may not be directly applicable to another breed [31].

The genetic relationship within the JBC population was notably high, with a value of 0.1743447 compared to the Hanwoo population (0.0004212), suggesting that JBC are more genetically homogeneous compared to Hanwoo, due to a more closed breeding structure, possibly reflecting a smaller effective population size or strong selection within the JBC line. Furthermore, compared to JBC, the lower value implies that Hanwoo individuals have more genetic diversity and are less linked, which could be the result of different sire contributions or selection procedures that refuse to put a high priority on similarities in genetic makeup in this group [39]. Meanwhile, the genetic relationship between Hanwoo and JBC individuals was negative (−0.0089897), indicating genetic divergence and very limited genetic overlap between the two populations [40]. This low genetic relationship likely contributed to the reduced prediction accuracy observed in HSJB and HAJB. One study [30] found the higher genetic relatedness among cows (0.001277) compared to steers (0.0000919) in the Hanwoo population, which is comparatively lower than our estimations. However, genomic prediction accuracy relies on the genetic closeness between the reference and testing populations. Since the JBC disclosed higher relatedness within the group, using a reference population composed of genetically similar individuals (JBC) can substantially enhance prediction accuracy. This finding reinforces the established principle that GS achieves the highest accuracy when the reference population closely mirrors the genetic makeup of the target population [41]. In contrast, a reference population consisting of genetically distant individuals provides limited benefit to prediction accuracy.

Our results show how the composition of the reference population affects the accuracy of genomic predictions for key economic traits such as CWT, EMA, BF, and MS in the JBC population. However, research on JBC remains limited, and there is a lack of established breeding strategies incorporating GS for this population. The small population size of JBC could reduce the accuracy and robustness of genomic predictions. Additionally, insufficient genetic variation within reference populations may limit the prediction of diverse genetic backgrounds. Despite these challenges, incorporating breed-specific reference data could facilitate the identification of unique genetic variants and trait associations relevant to JBC, thereby improving genomic evaluations in JBC breeding programs.

In this study, a single-trait animal model was applied to each carcass trait separately. Multi-trait models can improve GEBV accuracy by using genetic correlations among traits [42]. The single-trait model provided a more practical and computationally feasible framework. It also improved transparency, as prediction accuracy could be assessed separately for each trait. This avoided potential complications associated with estimating genetic correlations across breeds, which can be unstable when population structures differ considerably. Future studies with larger and more uniform datasets may benefit from adopting a multi-trait framework to further improve prediction accuracy.

From a practical breeding perspective, the effectiveness of a GS program depends not only on prediction accuracy but also on genetic gain per unit time, which integrates selection intensity, generation interval, and accuracy. Although a formal genetic gain analysis was beyond the scope of this study due to the absence of generation interval data specific to JBC, the substantially higher ACC_R achieved in JBJB suggests greater potential for genetic gain through within-breed GS. Future studies incorporating generation interval estimates and selection intensity parameters for JBC would enable a more comprehensive comparison of breeding program efficiency across the reference population configurations evaluated here.

This study highlights the critical role of reference population composition in the application of GS for JBC. The outcomes of this study indicate that using the same breed for both the reference and test population, as in JBJB, yields substantially higher prediction both in ACC_T and ACC_R, compared to other reference scenarios. This improvement was particularly evident for EMA and MS, where the ACC_T and ACC_R were closely aligned, enhancing the reliability of predictions for practical breeding applications. Notably, this breed-specific approach proved more accurate and consistent, reflecting the advantage of capturing genetic variation unique to the target breed. In conclusion, employing a JBC-specific reference population offers a robust and reliable framework for genomic prediction of carcass quality traits in JBC. By leveraging breed-specific genetic information, this strategy can enhance the precision of GS, particularly for traits influenced by unique genetic variability, and ultimately strengthen breeding programs aimed at improving meat quality in JBC.