Prediction of Cacao (Theobroma cacao) Resistance to Moniliophthora spp. Diseases via Genome-Wide Association Analysis and Genomic Selection

Abstract

Cacao (Theobroma cacao) is a globally important crop, and its yield is severely restricted by disease. Two of the most damaging diseases, witches' broom disease (WBD) and frosty pod rot disease (FPRD), are caused by a pair of related fungi: Moniliophthora perniciosa and Moniliophthora roreri, respectively. Resistant cultivars are the most effective long-term strategy to address Moniliophthora diseases, but efficiently generating resistant and productive new cultivars will require robust methods for screening germplasm before field testing. Marker-assisted selection (MAS) and genomic selection (GS) provide two potential avenues for predicting the performance of new genotypes, potentially increasing the selection gain per unit time. To test the effectiveness of these two approaches, we performed a genome-wide association study (GWAS) and GS on three related populations of cacao in Ecuador genotyped with a 15K single nucleotide polymorphism (SNP) microarray for three measures of WBD infection (vegetative broom, cushion broom, and chirimoya pod), one of FPRD (monilia pod) and two productivity traits (total fresh weight of pods and % healthy pods produced). GWAS yielded several SNPs associated with disease resistance in each population, but none were significantly correlated with the same trait in other populations. Genomic selection, using one population as a training set to estimate the phenotypes of the remaining two (composed of different families), varied among traits, from a mean prediction accuracy of 0.46 (vegetative broom) to 0.15 (monilia pod), and varied between training populations. Simulations demonstrated that selecting seedlings using GWAS markers alone generates no improvement over selecting at random, but that GS improves the selection process significantly. Our results suggest that the GWAS markers discovered here are not sufficiently predictive across diverse germplasm to be useful for MAS, but that using all markers in a GS framework holds substantial promise in accelerating disease-resistance in cacao.

Keywords: GWAS; SNPs; Theobroma cacao; frosty pod rot; genomic selection; witches’ broom disease.

FIGURE 1

Correlation (R-values) between phenotypes in three cacao populations. Phenotypes were log-transformed (Veg. br., Chir. pod, Cus. br., Fr. wt.) or set as proportion of total pods (Mon. pod, Hea. pod), then adjusted using site, year, and plant age to get a mean value per genotype.

FIGURE 2

Ancestry proportions for 1,345 accessions from three cacao populations. Each accession is represented by a vertical line and derives its ancestry from up to 10 ancestral groups which are indicated by the various colors in the legend. Ancestry was estimated using supervised Admixture analysis using a genome-wide panel of 9,640 SNPs.

FIGURE 3

Principal component analysis (PCA) of genetic relatedness of 1,345 cacao individuals in three sites using a genome-wide panel of 9,640 SNPs. Shapes refer to the population of the individual. Colored points are individuals showing >0.5 proportion ancestry of an ancestral group (see Figure 2 for description). Percentage of the variation captured by each component is given on the axis labels.

FIGURE 4

Mean pairwise SNP intra-chromosomal linkage disequilibrium (LD) by inter-SNP distance for three populations of cacao. Lines represent Loess-smoothed averages.

FIGURE 5

Genomic position of SNP markers significantly associated with five phenotypes among three populations of cacao (see Supplementary Table S3 for SNP information).

FIGURE 6

Simulated selection screen of two traits (vegetative broom and chirimoya pod) in three populations of cacao using three genetic prediction methods and one phenotypic method, compared against a random sampling of the populations. The predicted top ∼10% (40 individuals) performers for each phenotype from each population (‘Ganaderia,’ ‘Malvinas,’ ‘Las Tecas’) were selected using predictions from the training population (‘Las Tecas’), using three different methods (‘GWAS’ = ranking by sum of desirable GWAS-derived markers, ‘GS’ = ranking by genomic selection model GEBV, ‘GS + GWAS’ = ranking by genomic selection model GEBV with GWAS markers as fixed effects, ‘Pheno’ = phenotypic selection of seedlings for disease susceptibility (in Las Tecas only). Curve indicates the distribution of means from a 10,000-fold sampling of 40 random accessions from the training population. Lines indicate the position of the mean of the set selected by each method, including the actual top 10% selected by observed phenotypes. Sets outside of the random distribution are significantly different than the population mean at P < 0.0001.

FIGURE 7

Simulated selection screen of two traits (monilia Pod, total fresh weight) in three populations of cacao using five genetic prediction methods, compared against a random sampling of the populations. The predicted top ∼10% (40 individuals) for each phenotype from each population (‘Ganaderia,’ ‘Malvinas,’ ‘Las Tecas’) were selected using predictions from the training population (‘Las Tecas’), using three different methods (‘GWAS’ = ranking by sum of desirable GWAS-derived markers, ‘QTL’ = ranking by sum of desirable biparental population QTL markers, ‘GS’ = ranking by genomic selection model GEBV, ‘GS + GWAS’ = ranking by genomic selection model GEBV with GWAS markers as fixed effects, ‘GS + QTL’ = ranking by genomic selection model GEBV with QTL markers as fixed effects). Curve indicates the distribution of means from a 10,000-fold sampling of 40 random accessions from the training population. Lines indicate the position of the mean of the set selected by each method, including the actual top 10% selected by observed phenotypes. Sets outside of the random distribution are significantly different than the population mean at P < 0.0001.