ABSTRACT Single-cell transcriptomics has resolved cell-type-specific gene expression in plants, yet maize still lacks an integrated reference and species-specific foundation models. We present scMaize, combining scMaizeAtlas, an integrated atlas of 385,675 cells from 20 projects and 66 samples across seven tissues with hierarchical annotation, with two Transformer-based foundation models pretrained on this atlas. scMaizeExp serves as an expression-only baseline, while scMaizeGO incorporates Gene Ontology (GO) functional embeddings as an inductive bias. Although global expression-prediction accuracy was comparable, the GO prior improved rank-order prediction, strengthened attention toward functionally coherent gene modules, and enhanced embedding topology, with scMaizeGO achieving 86.0% cell-type and 97.1% tissue classification accuracy. Zero-shot evaluation demonstrated the cross-species generalizability of scMaizeGO representations, and few-shot fine-tuning enabled accurate cross-species classification with minimal labeled data. Root perturbation-condition analysis showed that the model encoded treatment-specific cellular states beyond cell-type identity, with the GO prior amplifying perturbation signals approximately threefold. Expression projection identified condition-responsive genes enriched for known stress pathways, and attention analysis revealed predominantly condition-specific changes in gene-gene attention that were weakly associated with expression-projection changes. An online platform ( https://www.scmaize.com ) provides atlas exploration, model access, and zero-code analysis tools. scMaize establishes a framework demonstrating that species-specific pretraining with functional priors enables transferable, perturbation-aware representations for crop single-cell genomics. HIGHLIGHTS scMaizeAtlas integrates 385,675 cells from 20 maize single-cell projects. scMaizeGO incorporates Gene Ontology priors into maize-specific pretraining. GO priors improve rank-order prediction, attention coherence and embeddings. Few-shot tuning enables cross-species cell-type classification with limited labels. Expression projection reveals stress-responsive genes in root cell states.

In recent years, genomic selection (GS) has been widely adopted in plant breeding; however, its practical application is constrained by the high cost of genotyping large segregating populations. To address this issue, this study employed a Quantitative Trait Nucleotide (QTN)-assisted GS strategy to evaluate its efficiency in reducing genotyping costs for soybean seed oil content (OC) and protein content (PC). Based on six multi-parent F4 populations (n = 4404) derived from seven elite soybean cultivars, which were genotyped using a 20K SNP chip, we identified 83 and 110 QTNs that were significantly associated with OC and PC, respectively. Among these loci, 37 and 62 QTNs were specific to OC and PC, respectively. Genomic prediction accuracies were evaluated across different training population (TP) sizes using three marker panels: genome-wide SNPs, all detected QTNs, and trait-specific QTNs. The panel consisting of all detected QTNs exhibited significantly higher prediction accuracy than the other two panels, except for PC when using 90% of the population as the training set. Phenotypic verification of the selected individuals showed that the PC-specific QTN panel yielded higher PC values and increased OC + PC values compared with the other marker panels. These results demonstrate that a small set of QTNs provides a cost-effective approach for genomic selection in practical soybean breeding programs.

Artificial intelligence (AI) is poised to reshape the research paradigm of the life sciences by rapidly advancing the adoption of protein language models and their derivative tools. These technologies are increasingly being applied to protein structure prediction, function analysis, and protein design throughout the life sciences, and have only recently begun to gain attention within the plant science community. Moreover, while the era of AI-driven bio-breeding is on the horizon, it remains largely in the proof-of-concept stage. Therefore, there is a pressing need not only to outline the fundamental principles, models, and tools in this rapidly evolving field, but also to explore their potential applications in plant research and crop breeding. This review begins by introducing general principles and widely used models for protein understanding and generation, supported by illustrative case studies that highlight how these tools are advancing fundamental plant research. For instance, the analyses of 2 maize (Zea mays) genes demonstrate how a structure-aware interpretation of the relationships between mutations and protein function enables more precise hypothesis generation and facilitates experimental validation. Subsequently, the review presents generic AI-enabled protein engineering strategies and pipelines, including rational, semi-rational, refactoring, and de novo design, tailored to diverse protein engineering objectives. These approaches aim to create artificial variants and synthetic proteins with improved or novel functions to foster innovation in crop breeding. Finally, the significant challenges of applying protein design in plants are discussed, particularly in light of the limited availability of experimentally resolved protein structures and the inherent complexity of plant biological systems.

Abstract Grapevines ( Vitis ) belonging to grape family (Vitaceae) are symbolic fruit crops pivotal to human civilization. The evolutionary history of grapevines divergent from other Vitaceae plants remains mysterious, requiring a family-wide whole-genome phylogenomic analysis. Here, we conduct chromosome-level phylogenomics to investigate the origin and evolution of grapevines using 29 genome assemblies of five genera Vitis , Parthenocissus, Ampelopsis, Tetrastigma, and Cissus, 27 of which are newly released in this study. Phylogenomic and macrosynteny analysis unanimously support Ampelopsis as a sister lineage to Parthenocissus, placing both closer to Vitis , with introgression and incomplete lineage sorting contributing to these relationships. Ancestral genome reconstruction delineates the major chromosome rearrangement events in Vitaceae karyotype evolution, highlighting the conserved karyotype in Vitis and the extensive karyotypic reorganization in Tetrastigma and Cissus . Pan-3D genome analysis highlights the contributions of structural variants (SVs) to the variation of A/B compartments and topologically associated domains (TADs), revealing a strong purifying selection of SVs at TAD boundaries. We further demonstrate that Helitron transposons drive the expansion and expression regulation of NLR immune-receptor genes in Vitis . Importantly, we discovered an NLR gene VbRpv35 from wild grapevine V. bellula resistant to downy mildew (DM), whose heterologous expression in V. vinifera confers enhanced DM resistance. Taken together, we provide phylogenomic insight into the origin and evolution of grapevines and valuable resources for grapevine improvement and understanding angiosperm evolution.

Genome-wide association studies (GWAS) play a crucial role in unraveling the genetic foundations of complex traits in plants but are also hampered by the application of heterogeneous tools, incompatible file formats and disparate computational environments. Existing GWAS frameworks are often restricted to a single linear reference genome, limiting the capacity for the analysis of structural variations and presence/absence variations (PAV) within plant populations. These issues pose obstacles to reproducibility, scalability, and comprehensive investigations. Here, we present PlantOmicsGWAS, an open-source Python framework for reproducible plant genome-wide association analysis and genomic prediction. It integrates reference indexing, FASTQ quality control, alignment, variant calling, VCF normalization, PLINK conversion, linkage disequilibrium analysis, population-structure estimation, association testing, marker scoring, genomic prediction, and visualization within a unified Linux and HPC workflow. The framework supports conventional linear-reference analyses and includes an optional pangenome-oriented module for working with multiple assemblies and graph-derived variation. Using a Vitis benchmark dataset containing 120 accessions and 118,247 graph-derived variants, PlantOmicsGWAS reduced manual workflow fragmentation and generated standardized association outputs. This tool provides a modular and extensible platform for plant GWAS and pan-GWAS workflows while retaining compatibility with established command-line tools and common genotype formats. The GWAS workflow described herein is adaptable to a range of sequencing methods and plant genomes, bridging research on crop related issues across various biological levels, from the individual organism to entire populations. PlantOmicsGWAS implements Bayesian sparse linear mixed modeling (BSLMM) through GEMMA for multi-trait association discovery, while also supporting FaST-LMM, regression-based approaches, and machine-learning algorithms (Random Forest, XGBoost) as benchmarking alternatives. The PlantOmicsGWAS, a versatile toolkit is available at GitHub https://github.com/plantomicsgwas1-boop/PlantOmicsGwas_V1 and on Linux and HPC platform (https://pypi.org/project/PlantOmicsGwas/1.0.2/).

Protein engineering modifies protein molecules to achieve specific biological or technological functions. Protein design forms the core methodology, and, in recent years, artificial intelligence (AI)-driven approaches have enabled more precise trait design in plants. This review highlights the convergence of protein structure prediction, generative sequence modeling, and function optimization to create synthetic proteins with improved specificity, stability, and activity in plant systems. We trace the development of protein design from rational design to semi-rational strategies and AI-driven platforms that integrate structure prediction, sequence generation, and de novo design. We discuss eight application areas relevant to plant physiology and breeding: enhanced disease resistance via engineered immune receptors, insect resistance through optimized insecticidal proteins, abiotic stress tolerance through metabolic enzyme stabilization, improved nutrient use via transporter redesign, variant mining for trait fine-tuning, genome-editing system optimization, environmental sensing with synthetic biosensors, and programmable regulatory circuits for plant factories (including controlled environment agriculture). Across these areas, we summarize design principles, advances, and translational considerations, emphasizing how AI expands sequence space and improves candidate prioritization. We also address current bottlenecks, including domain shift, reliability gaps in generative models, limited portability, the genotype-to-phenotype gap, and design-to-validation workflow constraints. Finally, we propose a staged roadmap for AI-driven plant trait design and outline the milestones and requirements for translation into breeding.

Abstract Codon optimization is widely used in transgenic crop development, plant synthetic biology, and molecular farming to improve heterologous protein expression in plant cells. Increasing availability of plant omics data now enables optimization strategies that account for species-specific sequence features. We developed HalluCodon, a customizable framework that uses multimodal language models to design coding sequences tailored to individual plant species. The framework allows users to fine tune pre-trained protein and RNA language models with their own datasets to build species-specific codon optimization models. The current implementation includes base models trained on coding sequences and proteomes from fifteen plant species. HalluCodon generates coding sequences through a hallucination-based design strategy guided by two predictive modules that evaluate coding sequence naturalness (CodonNAT) and expression potential (CodonEXP). Benchmark tests using representative proteins show that the generated sequences reproduce host-specific codon usage patterns and support high expression levels in plant systems.

Cabernet Sauvignon (CS), a cultivar of winegrape (Vitis vinifera), is among the most renowned winegrape varieties globally. In this study, we released the haplotype-resolved telomere-to-telomere (T2T) CS genome assembled using a combination of PacBio HiFi, ONT ultra-long, and Hi-C sequencing data. The two T2T gap-free haplotype-resolved assemblies CS-T2T.Hap1 and CS-T2T.Hap2 sized 491.11 Mb (contig N50 = 25.09 Mb) and 491.90 Mb (contig N50 = 25.51 Mb), respectively. Genome annotation predicted a total of 36,456 genes in CS-T2T.Hap1 and 35,471 genes in CS-T2T.Hap2. By genome comparison, we discovered and validated megabase inversion events on Chromosome 03,11,18 and 19, which are not prevalent in other haplotype-resolved V. vinifera genomes. In summary, this haplotype-resolved T2T genome represents an essential genomic resource for Cabernet Sauvignon, and lays the foundation for its genetic studies, improvement and utilization.

Plant genomics faces a paradox: genomic, phenotypic and literature resources continue to expand, yet discovery remains limited by the workflow needed to integrate dispersed literature, heterogeneous analytical tools and iterative reasoning into a defensible candidate hypothesis. We argue that this gap reflects a systems-level constraint: in current human-centred practice, literature depth, analytical breadth and time efficiency are difficult to improve simultaneously, producing an empirical trade-off frontier that bounds many existing workflows. Here we frame the frontier as a conceptual and measurable evaluation agenda for assessing computational systems in plant genomics discovery. On this basis, we propose a four-level based plant-genomics capability hierarchy for autonomous agents, defined by functional and falsifiable criteria rather than by model architecture: L1 knowledge integration, L2 tool orchestration, L3 closed-loop computational discovery, and L4 self-evolving systems. Existing plant-focused systems provide evidence mainly for L1 capabilities and early L2 orchestration, whereas L3 remains an open empirical target and L4 is speculative. To make these distinctions testable, we outline PlantAgentBench, a plant-focused benchmark spanning literature-based gene prioritisation, reproducible multi-tool workflow execution, and iterative candidate-gene refinement. Relevant metrics include recall, provenance completeness, autonomous cycles and human-intervention count. We conclude that autonomous agents are most valuable not as replacements for plant scientists, but as auditable infrastructure for extending expert-guided discovery.

Liquid-liquid phase separation regulates biological processes through dynamic condensates. Despite its significance, experimentally validated phase-separating proteins in plants remain limited, complicating predictions. We overcome this gap by applying positive-unlabeled learning, a semi-supervised approach optimized for imbalanced datasets. Leveraging 6,559 reported plant phase-separating proteins from eight species, we train a model integrating sequence-structural features, enabling prediction of 174,656 high-confidence candidates across 14 species. Experimental validation confirms liquid-liquid phase separation in 67.9% of the candidate proteins from Arabidopsis, rice, and maize. This positive-unlabeled framework demonstrates robust predictive power while providing open resources to advance plant phase separation research.

Longan (Dimocarpus longan) is a tropical tree in Sapindaceae family with economic importance known for its nutritous fruits. Genetic improvement of longan requires knowledge of reliable molecular markers and functional genes associated with key traits. However, it is largely impeded by lacking a complete genome and pangenome reference encompassing the broad genetic diversity in diverse longan germplasms. Here, we present a telomere-to-telomere (T2T) gap-free genome of longan cultivar 'Shixia' and a graph-based pangenome constructed using newly assembled chromosome-level genomes of 101 accessions. We completely assemble the longan centromere regions primarily composed of Gypsy-CRM retrotransposons. The pan-genome analysis reveals 58,978 non-redundant structural variants (SVs) that exhibit signs of genomic selective sweep during longan domestication and breeding. Additionally, haplotype-resolved genome analysis suggests allele-specific gene expression associated with SV-driven changes in 3D genome architecture. Importantly, the graph-based pan-genome empowers population-scale SV genotyping and genomewide association with various longan traits. SV-GWAS revealed 12 QTLs significantly linked with longan maturity period. Among them, a 758bp insertion is located downstream of the DlDAZ gene encoding a C2H2 zinc-finger transcription factor, and DIDAZ-transgenic tomatoes showed delayed maturity. Together, our T2T genome and pan-genome provide valuable resources to facilitate longan genetic research and precise improvement. ### Competing Interest Statement The authors have declared no competing interest. Shandong Provincial Natural Science Foundation

Endosperm is a primary source of food and feed globally. Early developmental events including proliferation and cell differentiation in this dedicated seed sink structure are expected be responsive to the availability of photoassimilates and in turn to affect seed size and quality. However, the gene regulatory programs underlying early development of cereal endosperm remain largely unknown. We profiled the maize (Zea mays) endosperm transcriptome during the first four days after pollination using laser-capture microdissection and identified multiple temporal co-expression modules including a subset activated upon fertilization. Our analysis of the associated cis-regulatory elements and co-expressed transcription factor (TF) families enabled us to construct a gene network regulating basal endosperm transfer layer (BETL) differentiation through MYB-related (MYBR) transcription factors and a network of E2F TFs for early endosperm proliferation. We found a significant association of the BETL network with kernel size variation and a contribution of the Glucose-target-of-rapamycin (TOR)-dependent processes to the E2F-mediated regulation of endosperm proliferation. Using the available data, we propose early proliferative development and transfer cell differentiation in endosperm are coordinated via sugar-sensing inputs to ultimately control seed size. Our analysis can guide the development of strategies for improvement of seed composition and yield. ### Competing Interest Statement The authors have declared no competing interest. U.S. National Science Foundation, https://ror.org/021nxhr62, IOS-0923880, IOS-1444568

A foundation model (FM) is a neural network trained on large-scale data using unsupervised or self-supervised learning, capable of adapting to a wide range of downstream tasks. This review provides a comprehensive overview of FMs in plant molecular biology, emphasizing recent advances and future directions. It begins by tracing the evolution of biological FMs across the DNA, RNA, protein, and single-cell levels, from tools inspired by natural language processing (NLP) to transformative models for decoding complex biological sequences. The review then focuses on plant-specific FMs such as GPN, AgroNT, PDLLMs, PlantCaduceus, and PlantRNA-FM, which address challenges that are widespread among plant genomes, including polyploidy, high repetitive sequence content, and environment-responsive regulatory elements, alongside universal FMs like GENERator and Evo 2, which leverage extensive cross-species training data for sequence design and prediction of mutation effects. Key opportunities and challenges in plant molecular biology FM development are further outlined, such as data heterogeneity, biologically informed architectures, cross-species generalization, and computational efficiency. Future research should prioritize improvements in model generalization, multi-modal data integration, and computational optimization to overcome existing limitations and unlock the potential of FMs in plant science. This review serves as an essential resource for plant molecular biologists and offers a clear snapshot of the current state and future potential of FMs in the field.

Watermelon (Citrullus lanatus L.) is a globally important fruit crop, yet it is susceptible to devastating diseases such as vascular wilt caused by Fusarium oxysporum f. sp. niveum (Fon), with limited control options. Fon rapidly evolves to overcome host resistance, constantly threatening production through new pathogenic races. High-quality genomic resources are key to understanding the molecular mechanisms underlying Fon virulence evolution for disease management. Here, we de novo assembled and annotated gapless genomes of three isolates affiliated with different physiological races of Fon (race 1, 2, and 3), and dissected the mechanisms behind their distinctive virulence through comparative genomics and transcriptomics. Core and accessory chromosomes in Fon were identified, where each race-affiliated isolate carried a unique set of accessory chromosomes or regions. Comparative transcriptomics of Fon infection revealed distinctive temporal patterns of gene expression even among core gene families, particularly those related to cell wall degradation enzymes. Effectoromic prediction and comparative analysis in three gap-free genomes identified 13 FonR3-specific effectors (FonR3SEs), one (FonR3SE1) of which was a critical virulence factor of FonR3 on watermelon as demonstrated via functional experiments. These gap-free genome assemblies and FonR3SEs provide valuable resources for studying Fon pathobiology and evolution and improving development of disease control strategies.

Maize (Zea mays L.) is not only an important cereal crop, but also a model plant species for genetic, cytologic, genomic, and molecular studies. Maize possesses tremendous phenotypic and genetic diversity. During the past few decades, researchers have made significant advances in multiple areas, including the genomic compositions and variations of maize and its ancestors, the genetic and genomic bases of maize domestication and evolution, the genetic architecture of various agronomic traits (yield, quality, biotic and abiotic stress responses, nutrient use efficiency, fertility and heterosis), and the development of novel molecular breeding technologies. In this review, we summarize these research achievements and provide a perspective for future maize research and breeding.

Grapevines are among the most economically important fruit crops, and the microbiome profoundly influences their health, yield, and quality. However, mechanistic insights into microbiome-orchestrated grapevine biology remain limited. Here, we conduct large-scale pan-metagenomic and pan-metatranscriptomic analyses of the phyllosphere microbiome from 107 grapevine accessions spanning 34 Vitis species. We show that the grapevine core microbiome is dominated by phyla Bacillota and Pseudomonadota. Leveraging PacBio sequencing, we assembled 19 high-quality metagenome-assembled genomes (MAGs) from the grapevine pan-microbiome, representing the first MAG reconstruction in plant-associated microbial communities using PacBio reads. These MAGs encode genes associated with antibiotic resistance, secondary metabolism, and carbohydrate-active enzymes (CAZymes), which could potentially influence grapevine biology. During downy mildew (DM) infection, DM-resistant grapevines exhibit significantly higher microbial network complexity than susceptible counterparts. Among the key taxa contributing to this complexity, Bacillota emerged as the dominant phylum, displaying strong abundance correlations with phylum Euglenozoa and Cyanobacteriota, and an isolated Bacillota species from the grapevine leaves, Bacillus cereus, demonstrated potent biocontrol activity against DM infection. Pan-metatranscriptomic analysis further revealed significant upregulation of eukaryotic microbial genes involved in primary and secondary metabolism. This pan-metagenomic study offers unprecedented insights into the complex structure, diversity, and functional roles of the grapevine phyllosphere microbiome and presents valuable genomic and microbial resources for microbiome research and engineering to enhance viticulture productivity and quality.

Brassica rapa includes a variety of important vegetable and oilseed crops,yet it is significantly challenged by clubroot disease.Notably,the majority of genotypes of B.rapa with published genomes exhibit high susceptibility to clubroot disease.The present study presents a high-quality chromosome-level sequence of the genome of the DH40 clubroot-resistant(CR)line,a doubled haploid line derived from the hybrid progeny of a European turnip(ECD01)and two lines of Chinese cabbage.The assembled genome spans 420.92 Mb,with a contig N50 size of 11.97 Mb.Comparative genomics studies revealed that the DH40 line is more closely related to the Chinese cabbage Chiifu than to the turnip ECD04.The DH40 genome provided direct reference and greatly facilitate the map-based cloning of the clubroot resistance gene Crr5,encoding a nucleotide-binding leucine-rich repeat(NLR)protein.Further functional analysis demonstrated that Crr5 confers clubroot resistance in both Chinese cabbage and transgenic Arabidopsis.It responds to inoculation with Plasmodiophora brassicae and is expressed in both roots and leaves.Subcellular localization shows that Crr5 is present in the nucleus.Notably,the Toll/interleukin-1 receptor(TIR)domain of Crr5 can autoactivate and trigger cell death.In addition,we developed two Crr5-specific Kompetitive allele-specific PCR(KASP)markers and showcased their successful application in breeding CR Chinese cabbage through marker-assisted selection.Overall,our research offers valuable resources for genetic and genomic studies in B.rapa and deepens our understanding of the molecular mechanisms underlying clubroot resistance against P.brassicae.

Grapevine (Vitis) is one of the oldest domesticated fruit crops with great cultural and economic importance. Here we assembled and annotated haplotype-resolved genomes of 72 global Vitis accessions including 25 wild and 47 cultivated grapevines, among which genomes for 60 grapevines are newly released. Haplotype-aware phylogenomics disentangled the mysterious hybridization history of grapevines, revealing the enormous genetic diversity of the Vitis genus. Pangenomic analysis reveals that European cultivars, more susceptible to the destructive disease downy mildew (DM), have a smaller repertoire of resistance genes in the NLR family encoding the TIR-NBARC-LRR domain. Through extensive structural variation (SV) characterization, phenotyping, DM-infection transcriptome profiling of 113 Vitis accessions, and SV-expression quantitative trait loci analysis, we have identified over 63 SVs and their relevant genes significantly associated with DM resistance, exemplified by a lysine histidine transporter, VvLHT8. This haplotype-resolved super pangenome of the Vitis genus will accelerate breeding and enrich our understanding of the evolution and biology of grapevines.

Structural variations (SVs), especially presence-absence variations (PAVs), are crucial in crop domestication and trait improvement. Although pan-genome analysis provides an exhaustive view of PAVs, it is often limited by high costs and restricted sample sizes. In contrast, genome-wide association studies (GWASs) can effectively identify trait-marker associations in large populations but typically overlook PAVs and face challenges in distinguishing causal variants due to linkage disequilibrium. In this study, we performed de novo assembly of eight reference-quality foxtail millet (Setaria italica) genomes and constructed a graph-based pan-genome to systematically explore PAVs. We subsequently performed a GWAS with 344 millet accessions, targeting genomic regions associated with the color of the leaf, leaf sheath, and leaf pulvinus. Using interpretable machine-learning models, we identified large-effect variants in the 26.84-26.94 Mb interval on chromosome 7, including a 5002-bp Copia element insertion and other key variants associated with phenotypic variations in leaf color traits. This integrative approach combines the detailed variant-detection capabilities of pan-genome analysis with the large-scale mapping potential of GWASs and enhances variant prioritization using interpretable machine learning, providing a cost-efficient yet effective framework for studying agronomic traits in crops.