Selection strategies to improve yield in macadamia using component traits and genomics

Abstract

Macadamias (Macadamia integrifolia, M. tetraphylla, and their hybrids) are native to the east coast of Australia, and are grown commercially around the world for their high quality edible nut kernel. Breeding new cultivars for high nut yield is a lengthy and laborious process that can take over two decades. Evaluations are time-consuming due to the long juvenile period of four or more years, and the low correlation between young and mature tree yield means that at least eight years of evaluations are required. Furthermore, yield has low heritability, and the large tree size necessitates low field planting density, and thus increases land use and evaluation costs. It is hypothesised that genetic gain for yield may be increased, compared to traditional breeding approaches, through the use of strategies including (i) indirect selection using yield component traits, (ii) marker-assisted selection and (iii) genomic selection.

This thesis employed an experimental population of 295 seedling progeny and their 29 parents, at four sites across south-east Queensland, that were genotyped for 4,113 SNPs and 16,171 silicoDArT markers detected using Diversity Arrays Technology (DArT) methods. Population structure, genetic diversity and linkage disequilibrium (LD) between SNP markers was quantified to inform subsequent genomics analysis. LD decay was initially rapid at short distances (r2 = 0.124 for SNPs within 1 kb of each other), but low level LD persisted for long distances. The seedling population was relatively genetically diverse (HE = 0.255), and very similar in diversity to that of the 29 parents (0.250). Furthermore, progeny with M. integrifolia x M. tetraphylla ancestry were more genetically diverse (HE = 0.278) than M. integrifolia seedlings (HE = 0.189). Progeny were moderately differentiated and clustered into three distinct groups, which represented M. integrifolia germplasm and two hybrid groups.

Flowering and nut characteristics, tree growth and yield were measured on each tree. Estimations of trait heritability and genetic correlations between each component trait and yield were used to calculate selection efficiency of indirectly selecting for yield using component traits. Kernel recovery, an economically important trait, had high heritability (h2 = 0.76) but was negatively genetically correlated with yield (rg = -0.27). Trunk circumference correlated strongly with yield (rg = 0.72) and was moderately heritable (h2 = 0.44); however, a breeding aim is to reduce tree size without compromising yield. No component traits were appropriate or effective for indirectly selecting for high yield.

A genome-wide association study (GWAS) was used to identify genetic markers associated with component traits that were genetically correlated with yield or are considered economically important. SNP markers were significantly associated (after correction for false positives) with nut weight (n = 7), percentage of whole kernels (n = 4), and trunk circumference (n = 44). Multiple regression analysis found that some markers were detecting the same QTLs, and, thus, rendered some redundant. These significantly associated markers could be used for marker-assisted selection (MAS) at the seedling stage to identify trees with desirable nut characteristics prior to fruiting.

Genomic selection (GS) used markers across the whole genome to predict estimated breeding values for yield and yield stability of individuals. Predictions with models using four years of yield data were more accurate (r = 0.12 to 0.94) than those using only one or two years of data (-0.29 to 0.39). Predictions in related families were more accurate (r = 0.57) than in unrelated population predictions (r = 0.22) using four years of yield data, confirming previously reported results in other crops that using GS models will be most beneficial when the target population is closely related to the training population. Predicted genetic gain of yield for related family predictions (421–438 g/year, at 2.5% selection intensity) was more than double that for traditional breeding (202 g/year), and was low for unrelated population predictions (12 to 162 g/year). A comparison of selection strategies indicated that assessing thousands of seedlings in the nursery using genomics-assisted breeding is unachievable due to the current costs of genotyping, though this may decrease in the future. A more feasible approach could be to screen thousands of seedlings for precocity (early flowering) and kernel recovery in the field, and reduce costs by only genotyping the elite individuals to predict yield using GS.

Estimates of heritability and correlations will inform the ease and ability to breed and select for key component traits. MAS could predict phenotypes for yield component traits based on allelic states at key markers, whilst GS could be used to predict yield and yield stability, though this will depend on genotyping costs. Future research requires validation in a separate population, with more individuals and SNP markers to increase accuracy and improve certainty of results. Early identification of elite germplasm would reduce time and labour involved in evaluating progeny, and increase genetic gain by decreasing selection cycle time.