Trait association and prediction through integrative k-mer analysis.

Genome-wide association study (GWAS) with single nucleotide polymorphisms (SNPs) has been widely used to explore genetic controls of phenotypic traits. Alternatively, GWAS can use counts of substrings of length k from longer sequencing reads, k-mers, as genotyping data. Using maize cob and kernel color traits, we demonstrated that k-mer GWAS can effectively identify associated k-mers. Co-expression analysis of kernel color k-mers and genes directly found k-mers from known causal genes. Analyzing complex traits of kernel oil and leaf angle resulted in k-mers from both known and candidate genes. A gene encoding a MADS transcription factor was functionally validated by showing that ectopic expression of the gene led to less upright leaves. Evolution analysis revealed most k-mers positively correlated with kernel oil were strongly selected against in maize populations, while most k-mers for upright leaf angle were positively selected. In addition, genomic prediction of kernel oil, leaf angle, and flowering time using k-mer data resulted in a similarly high prediction accuracy to the standard SNP-based method. Collectively, we showed k-mer GWAS is a powerful approach for identifying trait-associated genetic elements. Further, our results demonstrated the bridging role of k-mers for data integration and functional gene discovery.

Using recurrent neural networks to detect supernumerary chromosomes in fungal strains causing blast diseases.

The genomes of the fungus Magnaporthe oryzae that causes blast diseases on diverse grass species, including major crops, have indispensable core-chromosomes and may contain supernumerary chromosomes, also known as mini-chromosomes. These mini-chromosomes are speculated to provide effector gene mobility, and may transfer between strains. To understand the biology of mini-chromosomes, it is valuable to be able to detect whether a M. oryzae strain possesses a mini-chromosome. Here, we applied recurrent neural network models for classifying DNA sequences as arising from core- or mini-chromosomes. The models were trained with sequences from available core- and mini-chromosome assemblies, and then used to predict the presence of mini-chromosomes in a global collection of M. oryzae isolates using short-read DNA sequences. The model predicted that mini-chromosomes were prevalent in M. oryzae isolates. Interestingly, at least one mini-chromosome was present in all recent wheat isolates, but no mini-chromosomes were found in early isolates collected before 1991, indicating a preferential selection for strains carrying mini-chromosomes in recent years. The model was also used to identify assembled contigs derived from mini-chromosomes. In summary, our study has developed a reliable method for categorizing DNA sequences and showcases an application of recurrent neural networks in predictive genomics.

Genetic and transcriptomic dissection of host defense to Goss's bacterial wilt and leaf blight of maize.

Goss's wilt, caused by the Gram-positive actinobacterium Clavibacter nebraskensis, is an important bacterial disease of maize. The molecular and genetic mechanisms of resistance to the bacterium, or, in general, Gram-positive bacteria causing plant diseases, remain poorly understood. Here, we examined the genetic basis of Goss's wilt through differential gene expression, standard genome-wide association mapping (GWAS), extreme phenotype (XP) GWAS using highly resistant (R) and highly susceptible (S) lines, and quantitative trait locus (QTL) mapping using 3 bi-parental populations, identifying 11 disease association loci. Three loci were validated using near-isogenic lines or recombinant inbred lines. Our analysis indicates that Goss's wilt resistance is highly complex and major resistance genes are not commonly present. RNA sequencing of samples separately pooled from R and S lines with or without bacterial inoculation was performed, enabling identification of common and differential gene responses in R and S lines. Based on expression, in both R and S lines, the photosynthesis pathway was silenced upon infection, while stress-responsive pathways and phytohormone pathways, namely, abscisic acid, auxin, ethylene, jasmonate, and gibberellin, were markedly activated. In addition, 65 genes showed differential responses (up- or down-regulated) to infection in R and S lines. Combining genetic mapping and transcriptional data, individual candidate genes conferring Goss's wilt resistance were identified. Collectively, aspects of the genetic architecture of Goss's wilt resistance were revealed, providing foundational data for mechanistic studies.

Redox-engineering enhances maize thermotolerance and grain yield in the field.

Increasing populations and temperatures are expected to escalate food demands beyond production capacities, and the development of maize lines with better performance under heat stress is desirable. Here, we report that constitutive ectopic expression of a heterologous glutaredoxin S17 from Arabidopsis thaliana (AtGRXS17) can provide thermotolerance in maize through enhanced chaperone activity and modulation of heat stress-associated gene expression. The thermotolerant maize lines had increased protection against protein damage and yielded a sixfold increase in grain production in comparison to the non-transgenic counterparts under heat stress field conditions. The maize lines also displayed thermotolerance in the reproductive stages, resulting in improved pollen germination and the higher fidelity of fertilized ovules under heat stress conditions. Our results present a robust and simple strategy for meeting rising yield demands in maize and, possibly, other crop species in a warming global environment.

Development of a Model for Genomic Prediction of Multiple Traits in Common Bean Germplasm, Based on Population Structure.

Due to insufficient identification and in-depth investigation of existing common bean germplasm resources, it is difficult for breeders to utilize these valuable genetic resources. This situation limits the breeding and industrial development of the common bean (Phaseolus vulgaris L.) in China. Genomic prediction (GP) is a breeding method that uses whole-genome molecular markers to calculate the genomic estimated breeding value (GEBV) of candidate materials and select breeding materials. This study aimed to use genomic prediction to evaluate 15 traits in a collection of 628 common bean lines (including 484 landraces and 144 breeding lines) to determine a common bean GP model. The GP model constructed by landraces showed a moderate to high predictive ability (ranging from 0.59-0.88). Using all landraces as a training set, the predictive ability of the GP model for most traits was higher than that using the landraces from each of two subgene pools, respectively. Randomly selecting breeding lines as additional training sets together with landrace training sets to predict the remaining breeding lines resulted in a higher predictive ability based on principal components analysis. This study constructed a widely applicable GP model of the common bean based on the population structure, and encouraged the development of GP models to quickly aggregate excellent traits and accelerate utilization of germplasm resources.

Chromosome-level genome assembly of a regenerable maize inbred line A188.

BACKGROUND: The maize inbred line A188 is an attractive model for elucidation of gene function and improvement due to its high embryogenic capacity and many contrasting traits to the first maize reference genome, B73, and other elite lines. The lack of a genome assembly of A188 limits its use as a model for functional studies. RESULTS: Here, we present a chromosome-level genome assembly of A188 using long reads and optical maps. Comparison of A188 with B73 using both whole-genome alignments and read depths from sequencing reads identify approximately 1.1 Gb of syntenic sequences as well as extensive structural variation, including a 1.8-Mb duplication containing the Gametophyte factor1 locus for unilateral cross-incompatibility, and six inversions of 0.7 Mb or greater. Increased copy number of carotenoid cleavage dioxygenase 1 (ccd1) in A188 is associated with elevated expression during seed development. High ccd1 expression in seeds together with low expression of yellow endosperm 1 (y1) reduces carotenoid accumulation, accounting for the white seed phenotype of A188. Furthermore, transcriptome and epigenome analyses reveal enhanced expression of defense pathways and altered DNA methylation patterns of the embryonic callus. CONCLUSIONS: The A188 genome assembly provides a high-resolution sequence for a complex genome species and a foundational resource for analyses of genome variation and gene function in maize. The genome, in comparison to B73, contains extensive intra-species structural variations and other genetic differences. Expression and network analyses identify discrete profiles for embryonic callus and other tissues.

Genomic Prediction using Existing Historical Data Contributing to Selection in Biparental Populations: A Study of Kernel Oil in Maize.

Maize ( L.) kernel oil provides high-quality nutrition for animal feed and human health. A certain number of maize breeding programs seek to enhance oil concentration and composition. Genomic selection (GS), which entails selection based on genomic estimated breeding values (GEBVs), has proven to be efficient in breeding programs. Here, we estimate the robustness of predictions for the oil traits of maize kernels in biparental recombination inbred lines (RILs) using a GS model built based on an association population. Most statistical models, including ridge regression-best linear unbiased prediction (RR-BLUP), showed high prediction accuracy in the training population through a cross validation procedure. The training population size was more important than marker density and a statistical model for prediction performance. Using the optimized GS model, prediction of the biparental RIL population showed medium-high prediction accuracy (0.68) compared with prediction using only oil associated markers ( = 0.43). The potential to apply the GS model to another RIL population that is genetically less related to the training population was also examined, showing promising prediction accuracy in the top selected lines. Our results proved that genomic prediction using existing data is robust for the prediction of polygenic traits with moderate to high heritability.