Consequences of genomic recurrent selection on population diversity, inbreeding, and prediction for CBSD and yield in cassava after three rounds of genomic selection

Cassava breeding for resistance to CBSD in Uganda has advanced to four genomic-enabled recurrent selection (GE-RS) cycles (C_0, C_1, C₂ & C₃). These cycles originated from biparental crosses among 49 progenitors from Tanzania, Uganda, and Colombia, with ~100 superior clones recombined per cycle based on genomic estimated breeding values (GEBVs) from CBSD severity scores. Genotyping used 23K (C_0, C₁ & C₂) and 12K (C₃) DArTseq-derived single-nucleotide polymorphism (SNP) markers. This study evaluated (i) population structure and inbreeding trends across cycles and (ii) genomic prediction model performance for CBSD and yield traits after three selection rounds. Hierarchical clustering and diversity analyses showed a shift from three genetic clusters in C₀ to seven in C_1, followed by a slight collapse to six in C₂ and finally a single cluster in C₃. Observed heterozygosity (Ho) decreased from 47.7% in C₀ to 24.2% in C₃, and inbreeding (F) peaked at 44.5% in C₁. Predictive abilities for foliar (0.13-0.15) and root (0.28-0.29) CBSD severity were low across models (GBLUP, BRR, BayesA/B/C, Bayesian Lasso). Reduced genetic variance after C₃ suggests limited gains from further recombination. Advancing elite clones as variety candidates is recommended, alongside introducing new parental material to maintain genetic diversity for future cycles.