Genomics-driven breeding for local adaptation of durum wheat is enhanced by farmers' traditional knowledge

Abstract

In the smallholder, low-input farming systems widespread in sub-Saharan Africa, farmers select and propagate crop varieties based on their traditional knowledge and experience. A data-driven integration of their knowledge into breeding pipelines may support the sustainable intensification of local farming. Here, we combine genomics with participatory research to tap into traditional knowledge in smallholder farming systems, using durum wheat (Triticum durum Desf.) in Ethiopia as a case study. We developed and genotyped a large multiparental population, called the Ethiopian NAM (EtNAM), that recombines an elite international breeding line with Ethiopian traditional varieties maintained by local farmers. A total of 1,200 EtNAM lines were evaluated for agronomic performance and farmers' appreciation in three locations in Ethiopia, finding that women and men farmers could skillfully identify the worth of wheat genotypes and their potential for local adaptation. We then trained a genomic selection (GS) model using farmer appreciation scores and found that its prediction accuracy over grain yield (GY) was higher than that of a benchmark GS model trained on GY. Finally, we used forward genetics approaches to identify marker-trait associations for agronomic traits and farmer appreciation scores. We produced genetic maps for individual EtNAM families and used them to support the characterization of genomic loci of breeding relevance with pleiotropic effects on phenology, yield, and farmer preference. Our data show that farmers' traditional knowledge can be integrated in genomics-driven breeding to support the selection of best allelic combinations for local adaptation.

Keywords: Triticum durum Desf.; crop breeding; genomic selection; multiparental populations; smallholder farming.

Fig. 1.

Diversity and agronomic performance in the EtNAM by BLUP value distributions. (A) Distribution of OA scores by gender and by location, with colors according to legend. (B) Distribution of men scores (x-axis) and women scores (y-axis), by location. A regression line is fitted to the score distribution for each location with colors according to legend. The model R² is reported on top left with colors matching the distributions. (C) Distribution of GY performance by location. (D) Farmers’ top choice of genetic materials in the multivariate space of genetic diversity and (E) phenotypic diversity. Individual genotypes are marked in gray, while genotypes scoring above the 95th percentile of OA distribution according to men and women are highlighted in colors according to legend.

Fig. 2.

Breakdown of farmer choices on EtNAM genotypes. The x-axis reports the log-worth, the probability that each genotype within EtNAM families (y-axis) to be selected against the other genotypes. EtNAM families are reported with the corresponding code, N followed by a number. The entry RF represents the recurrent founder used to develop the EtNAM, the modern variety Asassa. The worth of RF was set at 0 for reference. Different groups classified by location and gender according to the model represent different choices in selecting genotypes. Drivers of farmers’ choices, based on agronomic metrics, are presented in SI Appendix, Table S1. Intervals are based on quasi-variance estimates. Data analysis was conducted on BLUP values.

Fig. 3.

Accuracy of GS models considering GY and OA values measured in the DP and the EtNAM. (A) Prediction accuracy of a model trained on DP data and tested on GY in the EtNAM. The pink shading highlights combined data, while location-specific prediction accuracies are given separately. The accuracy of the prediction is reported on the y-axis with bars indicating SEM across 100 repetitions. The predictors are color coded according to legend, while predicted OA and GY measures are reported on the x-axis. (B) Prediction accuracy of a model trained on DP data and tested on OA in the EtNAM, plotted as in panel A. (C) Prediction accuracy of a model trained on the EtNAM and tested on DP data. OA values are split by gender (W, women; M, men) and combined across genders. Data analysis was conducted on BLUP values.

Fig. 4.

Genetic targets for participatory wheat improvement. (A) GWAS reporting marker trait associations for OA scored by women, OA scored by men, OA combined across genders, and GY. On the x-axis, SNP markers are arranged by their estimated physical position, with alternating colors corresponding to the 14 chromosomes of durum wheat plus an unmapped linkage group (UN). The y-axis reports the significance of the association, with SNPs surpassing the Bonferroni threshold (green line) marked as significant. The dashed green line, when present, represents a less stringent threshold for FDR-corrected P-values. (B) QTL mapping on individual EtNAM families. Markers included in EtNAM genetic maps are reported as black ticks according to their physical position. QTL are shown in colors according to legend and correspond to phenotypes grouped by phenology (DB, DH, DF, DM), yield components (GY, TGW, SPL, NSPKPS, SPS, PH, NTPP, BM), and farmers’ appreciation (OA). QTL markers are semi-transparent and have deeper shades of color proportionally to the number of EtNAM subfamilies in which they are detected. Data analysis was conducted on BLUP values combined across locations.