Combined analysis of transposable elements and structural variation in maize genomes reveals genome contraction outpaces expansion.

Structural differences between genomes are a major source of genetic variation that contributes to phenotypic differences. Transposable elements, mobile genetic sequences capable of increasing their copy number and propagating themselves within genomes, can generate structural variation. However, their repetitive nature makes it difficult to characterize fine-scale differences in their presence at specific positions, limiting our understanding of their impact on genome variation. Domesticated maize is a particularly good system for exploring the impact of transposable element proliferation as over 70% of the genome is annotated as transposable elements. High-quality transposable element annotations were recently generated for de novo genome assemblies of 26 diverse inbred maize lines. We generated base-pair resolved pairwise alignments between the B73 maize reference genome and the remaining 25 inbred maize line assemblies. From this data, we classified transposable elements as either shared or polymorphic in a given pairwise comparison. Our analysis uncovered substantial structural variation between lines, representing both simple and complex connections between TEs and structural variants. Putative insertions in SNP depleted regions, which represent recently diverged identity by state blocks, suggest some TE families may still be active. However, our analysis reveals that within these recently diverged genomic regions, deletions of transposable elements likely account for more structural variation events and base pairs than insertions. These deletions are often large structural variants containing multiple transposable elements. Combined, our results highlight how transposable elements contribute to structural variation and demonstrate that deletion events are a major contributor to genomic differences.

Cross-species modeling of plant genomes at single nucleotide resolution using a pre-trained DNA language model.

Understanding the function and fitness effects of diverse plant genomes requires transferable models. Language models (LMs) pre-trained on large-scale biological sequences can learn evolutionary conservation, thus expected to offer better cross-species prediction through fine-tuning on limited labeled data compared to supervised deep learning models. We introduce PlantCaduceus, a plant DNA LM based on the Caduceus and Mamba architectures, pre-trained on a carefully curated dataset consisting of 16 diverse Angiosperm genomes. Fine-tuning PlantCaduceus on limited labeled Arabidopsis data for four tasks involving transcription and translation modeling demonstrated high transferability to maize that diverged 160 million years ago, outperforming the best baseline model by 1.45-fold to 7.23-fold. PlantCaduceus also enables genome-wide deleterious mutation identification without multiple sequence alignment (MSA). PlantCaduceus demonstrated a threefold enrichment of rare alleles in prioritized deleterious mutations compared to MSA-based methods and matched state-of-the-art protein LMs. PlantCaduceus is a versatile pre-trained DNA LM expected to accelerate plant genomics and crop breeding applications.