Genomic Selection for Haploid Induction Rate in Tropical Maize ( Zea mays L.) Haploid Inducer Populations: Insights From Heterotic Pools and Recurre

ABSTRACT

Doubled‐haploid (DH) technology based on in vivo haploid induction has become an effective strategy for accelerating maize breeding. However, improving haploid inducer populations remains challenging because accurate estimation of haploid induction rate (HIR) requires labour‐intensive phenotypic validation, while putative HIR estimates are often compromised by incomplete expression of the R1‐nj marker. This study evaluated the predictive ability of genomic selection (GS) for putative (HIRp) and validated (HIRr) HIR in a tropical maize haploid inducer population undergoing recurrent selection. A total of 190 inducer progenies from three recurrent selection cycles (C0, C1 and C2) were evaluated using testers representing the Flint and Dent heterotic pools. Phenotypic data were analysed using linear mixed models, and genomic prediction was performed using GBLUP under random k ‐fold and leave‐one‐cycle‐out (LOCO) cross‐validation schemes. Broad‐sense heritability was substantially higher for HIRr (0.47) than for HIRp (0.16), accompanied by stronger genetic correlations between heterotic pools (0.56 vs. 0.11, respectively). Predictive ability ranged from 0.03 to 0.38 for HIRp and from 0.29 to 0.79 for HIRr, with the LOCO framework consistently providing the most informative prediction scenario and achieving the highest predictive abilities. Marker suppression differed markedly between heterotic pools, reaching 46.8% in Flint but only 0.7% in Dent, which was associated with the reduced heritability and predictive performance observed for HIRp. Relative efficiency estimates further demonstrated the potential of genomic selection to increase selection response, particularly when based on validated HIRr. Overall, the results demonstrate that genomic selection can be effectively integrated into recurrent breeding programmes for tropical maize haploid inducers, reducing dependence on labour‐intensive phenotypic validation while improving selection efficiency and supporting the genetic improvement of HIR.