Genomic Prediction of Complex Traits in Perennial Plants: A Case for Forest Trees

Revealing stable SNPs and genomic prediction insights across environments enhance breeding strategies of productivity, defense, and climate-adaptability traits in white spruce

Exploring the relationship between phenotype, genotype, and environment is essential in quantitative genetics. Considering the complex genetic architecture of economically important traits, integrating genotype-by-environment interactions in a genome-wide association (GWAS) and genomic prediction (GP) framework is imperative. This integration is crucial for identifying robust markers with stability across diverse environments and improving the predictive accuracy of individuals' performance within specific target environments. We conducted a multi-environment GWAS and GP analysis for 30 productivity, defense, and climate-adaptability traits on 1540 white spruce trees from Alberta, Canada, genotyped for 467,224 SNPs and growing across three environments. We identified 563 significant associations (p-value < 1.07 ×10-05) across the studied traits and environments, with 105 SNPs showing overlapping associations in two or three environments. Wood density, myrcene, total monoterpenes, α-pinene, and catechin exhibited the highest overlap (>50%) across environments. Gas exchange traits, including intercellular CO2 concentration and intrinsic water use efficiency, showed the highest number of significant associations (>38%) but less stability (<1.2%) across environments. Predictive ability (PA) varied significantly (0.03-0.41) across environments for 20 traits, with stable carbon isotope ratio having the highest average PA (0.36) and gas exchange traits the lowest (0.07). Only two traits showed differences in prediction bias (PB) across environments, with 80% of site-trait PB falling within a narrow range (0.90 to 1.10). Integrating multi-environment GWAS and GP analyses proved useful in identifying site-specific markers, understanding environmental impacts on PA and PB, and ultimately providing indirect insights into the environmental factors that influenced this white spruce breeding program.

Genomic selection in forest trees comes to life: unraveling its potential in an advanced four-generation Eucalyptus grandis population

Genomic Selection (GS) in tree breeding optimizes genetic gains by leveraging genomic data to enable early selection of seedlings without phenotypic data reducing breeding cycle and increasing selection intensity. Traditional assessments of the potential of GS in forest trees have typically focused on model performance using cross-validation within the same generation but evaluating effectively realized predictive ability (RPA) across generations is crucial. This study estimated RPAs for volume growth (VOL), wood density (WD), and pulp yield (PY) across four generations breeding of Eucalyptus grandis. The training set spanned three generations, including 34,461 trees with three-year growth data, 6,014 trees with wood quality trait data, and 1,918 trees with 12,695 SNPs (single nucleotide polymorphisms) data. Employing single-step genomic BLUP, we compared the genomic predictions of breeding values (GEBVs) for 1,153 fourth-generation full-sib seedlings in the greenhouse with their later-collected phenotypic estimated breeding values (EBVs) at age three years. RPAs were estimated using three GS targets (individual trees, trees within families, and families), two selection criteria (single- and multiple-trait), and training populations of either all 1,918 genotyped trees or the 67 direct ancestors of the selection candidates. RPAs were higher for wood quality traits (0.33 to 0.59) compared to VOL (0.14 to 0.19) and improved for wood traits (0.42 to 0.75) but not for VOL when trained only with direct ancestors, highlighting the challenges in accurately predicting growth traits. GS was more effective at excluding bottom-ranked candidates than selecting top-ranked ones. The between-family GS approach outperformed individual-tree selection for VOL (0.11 to 0.16) and PY (0.72 to 0.75), but not for WD (0.43 vs. 0.42). Furthermore, higher levels of relatedness and lower genotype by environment (G × E) interaction between training and testing populations enhanced RPAs for VOL (0.39). In summary, despite limited effectiveness in ranking top VOL individuals, GS effectively identified low-performing individuals and families. These multi-generational findings underscore GS's potential in tree breeding, stressing the importance of considering relatedness and G × E interaction for optimal performance.

Genomic selection in western redcedar: from proof of concept to operational application

Forests face many threats. While traditional breeding may be too slow to deliver well-adapted trees, genomic selection (GS) can accelerate the process. We describe a comprehensive study of GS from proof of concept to operational application in western redcedar (WRC, Thuja plicata). Using genomic data, we developed models on a training population (TrP) of trees to predict breeding values (BVs) in a target seedling population (TaP) for growth, heartwood chemistry, and foliar chemistry traits. We used cross-validation to assess prediction accuracy (PACC) in the TrP; we also validated models for early-expressed foliar traits in the TaP. Prediction accuracy was high across generations, environments, and ages. PACC was not reduced to zero among unrelated individuals in TrP and was only slightly reduced in the TaP, confirming strong linkage disequilibrium and the ability of the model to generate accurate predictions across breeding generations. Genomic BV predictions were correlated with those from pedigree but displayed a wider range of within-family variation due to the ability of GS to capture the Mendelian sampling term. Using predicted TaP BVs in multi-trait selection, we functionally implemented and integrated GS into an operational tree-breeding program.

High-density genetic linkage mapping in Sitka spruce advances the integration of genomic resources in conifers

In species with large and complex genomes such as conifers, dense linkage maps are a useful resource for supporting genome assembly and laying the genomic groundwork at the structural, populational, and functional levels. However, most of the 600+ extant conifer species still lack extensive genotyping resources, which hampers the development of high-density linkage maps. In this study, we developed a linkage map relying on 21,570 single nucleotide polymorphism (SNP) markers in Sitka spruce (Picea sitchensis [Bong.] Carr.), a long-lived conifer from western North America that is widely planted for productive forestry in the British Isles. We used a single-step mapping approach to efficiently combine RAD-seq and genotyping array SNP data for 528 individuals from 2 full-sib families. As expected for spruce taxa, the saturated map contained 12 linkages groups with a total length of 2,142 cM. The positioning of 5,414 unique gene coding sequences allowed us to compare our map with that of other Pinaceae species, which provided evidence for high levels of synteny and gene order conservation in this family. We then developed an integrated map for P. sitchensis and Picea glauca based on 27,052 markers and 11,609 gene sequences. Altogether, these 2 linkage maps, the accompanying catalog of 286,159 SNPs and the genotyping chip developed, herein, open new perspectives for a variety of fundamental and more applied research objectives, such as for the improvement of spruce genome assemblies, or for marker-assisted sustainable management of genetic resources in Sitka spruce and related species.

Comparison of ddRADseq and EUChip60K SNP genotyping systems for population genetics and genomic selection in Eucalyptus dunnii (Maiden)

Eucalyptus dunnii is one of the most important Eucalyptus species for short-fiber pulp production in regions where other species of the genus are affected by poor soil and climatic conditions. In this context, E. dunnii holds promise as a resource to address and adapt to the challenges of climate change. Despite its rapid growth and favorable wood properties for solid wood products, the advancement of its improvement remains in its early stages. In this work, we evaluated the performance of two single nucleotide polymorphism, (SNP), genotyping methods for population genetics analysis and Genomic Selection in E. dunnii. Double digest restriction-site associated DNA sequencing (ddRADseq) was compared with the EUChip60K array in 308 individuals from a provenance-progeny trial. The compared SNP set included 8,011 and 19,008 informative SNPs distributed along the 11 chromosomes, respectively. Although the two datasets differed in the percentage of missing data, genome coverage, minor allele frequency and estimated genetic diversity parameters, they revealed a similar genetic structure, showing two subpopulations with little differentiation between them, and low linkage disequilibrium. GS analyses were performed for eleven traits using Genomic Best Linear Unbiased Prediction (GBLUP) and a conventional pedigree-based model (ABLUP). Regardless of the SNP dataset, the predictive ability (PA) of GBLUP was better than that of ABLUP for six traits (Cellulose content, Total and Ethanolic extractives, Total and Klason lignin content and Syringyl and Guaiacyl lignin monomer ratio). When contrasting the SNP datasets used to estimate PAs, the GBLUP-EUChip60K model gave higher and significant PA values for six traits, meanwhile, the values estimated using ddRADseq gave higher values for three other traits. The PAs correlated positively with narrow sense heritabilities, with the highest correlations shown by the ABLUP and GBLUP-EUChip60K. The two genotyping methods, ddRADseq and EUChip60K, are generally comparable for population genetics and genomic prediction, demonstrating the utility of the former when subjected to rigorous SNP filtering. The results of this study provide a basis for future whole-genome studies using ddRADseq in non-model forest species for which SNP arrays have not yet been developed.

Genomic evaluation for breeding and genetic management in Cordia africana, a multipurpose tropical tree species

BACKGROUND

Planting tested forest reproductive material is crucial to ensure the increased resilience of intensively managed productive stands for timber and wood product markets under climate change scenarios. Single-step Genomic Best Linear Unbiased Prediction (ssGBLUP) analysis is a cost-effective option for using genomic tools to enhance the accuracy of predicted breeding values and genetic parameter estimation in forest tree species. Here, we tested the efficiency of ssGBLUP in a tropical multipurpose tree species, Cordia africana, by partial population genotyping. A total of 8070 trees from three breeding seedling orchards (BSOs) were phenotyped for height. We genotyped 6.1% of the phenotyped individuals with 4373 single nucleotide polymorphisms. The results of ssGBLUP were compared with pedigree-based best linear unbiased prediction (ABLUP) and genomic best linear unbiased prediction (GBLUP), based on genetic parameters, theoretical accuracy of breeding values, selection candidate ranking, genetic gain, and predictive accuracy and prediction bias.

RESULTS

Genotyping a subset of the study population provided insights into the level of relatedness in BSOs, allowing better genetic management. Due to the inbreeding detected within the genotyped provenances, we estimated genetic parameters both with and without accounting for inbreeding. The ssGBLUP model showed improved performance in terms of additive genetic variance and theoretical breeding value accuracy. Similarly, ssGBLUP showed improved predictive accuracy and lower bias than the pedigree-based relationship matrix (ABLUP).

CONCLUSIONS

This study of C. africana, a species in decline due to deforestation and selective logging, revealed inbreeding depression. The provenance exhibiting the highest level of inbreeding had the poorest overall performance. The use of different relationship matrices and accounting for inbreeding did not substantially affect the ranking of candidate individuals. This is the first study of this approach in a tropical multipurpose tree species, and the analysed BSOs represent the primary effort to breed C. africana.

Realized genomic selection across generations in a reciprocal recurrent selection breeding program of Eucalyptus hybrids

Introduction

Genomic selection (GS) experiments in forest trees have largely reported estimates of predictive abilities from cross-validation among individuals in the same breeding generation. In such conditions, no effects of recombination, selection, drift, and environmental changes are accounted for. Here, we assessed the effectively realized predictive ability (RPA) for volume growth at harvest age by GS across generations in an operational reciprocal recurrent selection (RRS) program of hybrid Eucalyptus.

Methods

Genomic best linear unbiased prediction with additive (GBLUP_G), additive plus dominance (GBLUP_G+D), and additive single-step (HBLUP) models were trained with different combinations of growth data of hybrids and pure species individuals (N = 17,462) of the G1 generation, 1,944 of which were genotyped with ~16,000 SNPs from SNP arrays. The hybrid G2 progeny trial (HPT267) was the GS target, with 1,400 selection candidates, 197 of which were genotyped still at the seedling stage, and genomically predicted for their breeding and genotypic values at the operational harvest age (6 years). Seedlings were then grown to harvest and measured, and their pedigree-based breeding and genotypic values were compared to their originally predicted genomic counterparts.

Results

Genomic RPAs ≥0.80 were obtained as the genetic relatedness between G1 and G2 increased, especially when the direct parents of selection candidates were used in training. GBLUP_G+D reached RPAs ≥0.70 only when hybrid or pure species data of G1 were included in training. HBLUP was only marginally better than GBLUP. Correlations ≥0.80 were obtained between pedigree and genomic individual ranks. Rank coincidence of the top 2.5% selections was the highest for GBLUP_G (45% to 60%) compared to GBLUP_G+D. To advance the pure species RRS populations, GS models were best when trained on pure species than hybrid data, and HBLUP yielded ~20% higher predictive abilities than GBLUP, but was not better than ABLUP for ungenotyped trees.

Discussion

We demonstrate that genomic data effectively enable accurate ranking of eucalypt hybrid seedlings for their yet-to-be observed volume growth at harvest age. Our results support a two-stage GS approach involving family selection by average genomic breeding value, followed by within-top-families individual GS, significantly increasing selection intensity, optimizing genotyping costs, and accelerating RRS breeding.

A simulation study comparing advanced marker-assisted selection with genomic selection in tree breeding programs

Advances in DNA sequencing technologies allow the sequencing of whole genomes of thousands of individuals and provide several million single nucleotide polymorphisms (SNPs) per individual. These data combined with precise and high-throughput phenotyping enable genome-wide association studies (GWAS) and the identification of SNPs underlying traits with complex genetic architectures. The identified causal SNPs and estimated allelic effects could then be used for advanced marker-assisted selection (MAS) in breeding programs. But could such MAS compete with the broadly used genomic selection (GS)? This question is of particular interest for the lengthy tree breeding strategies. Here, with our new software "SNPscan breeder," we simulated a simple tree breeding program and compared the impact of different selection criteria on genetic gain and inbreeding. Further, we assessed different genetic architectures and different levels of kinship among individuals of the breeding population. Interestingly, apart from progeny testing, GS using gBLUP performed best under almost all simulated scenarios. MAS based on GWAS results outperformed GS only if the allelic effects were estimated in large populations (ca. 10,000 individuals) of unrelated individuals. Notably, GWAS using 3,000 extreme phenotypes performed as good as the use of 10,000 phenotypes. GS increased inbreeding and thus reduced genetic diversity more strongly compared to progeny testing and GWAS-based selection. We discuss the practical implications for tree breeding programs. In conclusion, our analyses further support the potential of GS for forest tree breeding and improvement, although MAS may gain relevance with decreasing sequencing costs in the future.