Advances in single-cell and spatial multi-omics for plant development and stress responses

Abstract

Recent advances in single-cell and spatial multi-omics have enabled high-resolution profiling of plant cellular heterogeneity and tissue organization. Single-cell RNA sequencing (scRNA-seq) captures transcriptional variation across cell types and states, while single-nucleus ATAC sequencing (snATAC-seq) provides information on chromatin accessibility and regulatory elements. Spatial transcriptomics (ST) and spatial metabolomics preserve spatial context, enabling in situ mapping of molecular and metabolic features in plant tissues. This review summarizes experimental and computational approaches for these technologies in plants, with emphasis on their applicability and limitations under plant-specific constraints such as cell wall structure, tissue complexity, and genome organization. We further discuss computational integration strategies, including deconvolution, spatial mapping, and cross-modal representation learning, highlighting dependence on reference quality and challenges related to data sparsity and batch effects. We also consider current limitations in applying foundation models to plant data due to limited plant-specific training resources and differences in regulatory architecture across species. Finally, we highlight challenges in extending these approaches to non-model plants and emphasize the need for improved reference resources, standardized analytical frameworks, and robust validation strategies.

Highlights

• •Advances in scRNA-seq, snATAC-seq, spatial transcriptomics, and spatial metabolomics in plants are reviewed.
• •Cell-type-specific and spatial mechanisms underlying plant development and stress responses are summarized.
• •Technical bottlenecks in sample preparation, resolution–throughput trade-offs, and cross-modal integration are evaluated.
• •Future priorities include plant cell atlases, AI-enabled integration, crop improvement, and synthetic biology.

1. Introduction

As sessile multicellular eukaryotes, plants rely on intricate cellular heterogeneity and intercellular communication to sustain growth and respond to environmental cues. The fundamental challenge in plant biology lies in understanding how different specialized cells are coordinated in space and time to generate coherent developmental programs and adaptive responses. Traditional bulk RNA sequencing (bulk RNA-seq) has provided valuable insights into gene expression at the tissue and organ level (Wang et al., 2009). However, by averaging signals across heterogeneous cell populations, it obscures cell-type-specific transcriptional programs and limits mechanistic resolution (Jaitin et al., 2014). The emergence of single-cell RNA sequencing (scRNA-seq) has begun to address this challenge by enabling transcriptomic profiling at single-cell resolution (Fig. 1). Since its initial application in 2009 (Tang et al., 2009), and with the development of high-throughput platforms such as Drop-seq (Macosko et al., 2015) and 10 × Genomics Chromium (Zheng et al., 2017), scRNA-seq has transformed our ability to resolve cellular diversity. Recent surveys indicate that single-cell and spatial omics datasets are currently available for at least 90 plant species (Table S1), encompassing model plants, medicinal plant, food crops, feed crops, horticultural plant, economic crop and forest plant (Liang et al., 2025). Despite this progress, plant taxonomic coverage remains limited compared with the extensive cellular atlases established in animal and human systems.

Fig. 1.

Schematic diagram illustrating the difference between Bulk RNA-seq and scRNA-seq.

Resolving transcriptional states alone does not fully explain how cell identity is established or dynamically regulated. Gene expression programs are constrained by chromatin accessibility and the activity of cis-regulatory elements, necessitating approaches that capture regulatory potential in addition to transcriptional output. Single-nucleus ATAC sequencing (snATAC-seq) provides such a layer by profiling genome-wide chromatin accessibility at single-cell resolution, thereby linking regulatory elements to gene expression and enabling the reconstruction of gene regulatory networks. The integration of scRNA-seq and snATAC-seq thus begins to bridge the gap between regulatory potential and transcriptional states. Because both approaches require tissue dissociation, spatial information essential for understanding cell–cell interactions and tissue organization is lost. Spatial transcriptomics (ST) addresses this limitation by preserving tissue architecture while enabling in situ gene expression profiling (Stahl et al., 2016). Advances from early LCM-seq strategies (Nichterwitz et al., 2020) to high-resolution platforms such as Stereo-seq (Zhao et al., 2025) have substantially improved spatial resolution and throughput. In parallel, spatial metabolomics, driven by mass spectrometry imaging (MSI), extends molecular profiling to the functional level by enabling the in situ mapping of diverse metabolites, thereby providing direct insight into the spatial organization of metabolic processes (Tabeta et al., 2025).

Despite rapid progress, many existing studies and reviews treat single-cell transcriptomics, spatial transcriptomics, and spatial metabolomics as largely independent modalities, limiting a unified understanding of how gene regulation, transcriptional states, and metabolic outputs are coordinated across spatial scales. In addition, current literature is still heavily biased toward a few model species, particularly Arabidopsis, rice, and maize, with insufficient coverage of non-model plants such as ferns, gymnosperms, and woody species, which are essential for understanding the evolutionary diversification of plant cell types. Moreover, a systematic evaluation of experimental and computational differences across platforms remains lacking, including biases arising from cell isolation strategies, spatial transcriptomic technologies, and mass spectrometry imaging methods, particularly in the context of plant-specific constraints such as cell walls and genome complexity.

To address these gaps, this review provides an integrated and critical synthesis of single-cell and spatiotemporal multi-omics in plants by proposing a multi-layer framework that links spatial gene regulation, transcriptional states, and metabolic outputs, by comprehensively covering both model and non-model plant systems, and by critically evaluating current experimental and computational approaches with an emphasis on their limitations and emerging solutions.

2. Advances in single-cell and spatial multi-omics

2.1. Single-cell RNA sequencing technologies

A fundamental prerequisite for plant scRNA-seq is the acquisition of high-quality single-cell suspensions. Plant cell isolation mainly follows protoplast- and nuclei-based strategies. Protoplast isolation relies on enzymatic digestion to remove the cell wall, enabling recovery of intact cells and capturing both cytoplasmic and nuclear mRNAs for high-resolution transcriptomics (Fig. 2) (Birnbaum et al., 2003). However, enzymatic digestion imposes abiotic stress and may induce transcriptional artifacts, while 1–2 h processing further increases technical noise (van den Brink et al., 2017). Recovery efficiency is influenced by protocol conditions, developmental stage, and genotype, and outer or young tissues are generally more efficiently dissociated than internal vascular tissues (Denyer et al., 2019; Jean-Baptiste et al., 2019; Shulse et al., 2019). To mitigate these limitations, fixation-based methods such as FX-Cell, FXcryo-Cell, and cryoFX-Cell have been developed, improving recovery from recalcitrant tissues and enabling cryopreserved sample processing (Ming et al., 2025).

Fig. 2.

Workflow of single-cell and spatial multi-omics approaches.

Single-nucleus RNA sequencing (snRNA-seq) bypasses cell wall digestion, reducing dissociation-induced transcriptional perturbations and better reflecting in situ nuclear transcriptomes (Fig. 2). Nuclei are released via mechanical homogenization or mild lysis and are compatible with frozen tissues, supporting field sampling, multicenter studies, and retrospective analyses, particularly for difficult-to-dissociate tissues (Chiang et al., 2025). However, snRNA-seq primarily captures nuclear RNA and largely misses mature cytoplasmic transcripts, which may limit interpretation of rapid regulatory responses dependent on cytoplasmic mRNA dynamics (Fonseca et al., 2024). To improve applicability in plants, optimized nuclei isolation protocols have been developed, including FlsnRNA-seq, which enables efficient nucleus isolation and facilitates cell-type identification and transcriptomic profiling (Long et al., 2021).

After obtaining single-cell or nuclei suspensions, high-throughput sequencing is commonly performed using droplet-based microfluidic systems. The 10× Genomics Chromium platform co-encapsulates individual cells with barcoded gel beads and unique molecular identifiers (UMIs) in oil droplets, enabling cell lysis, transcript capture, and barcode indexing (Fig. 2) (Zhang et al., 2019). In contrast, plate-based methods such as Smart-seq2 provide full-length transcript coverage at lower throughput, supporting detailed gene structure analyses (Wang et al., 2021). Downstream analysis begins with read alignment and quantification using tools such as Cell Ranger or STARsolo to generate gene-by-cell matrices (Dobin et al., 2013), followed by quality control and doublet removal (Wolock et al., 2019). Data normalization, dimensionality reduction, and graph-based clustering (Louvain or Leiden algorithms) are then used to define cell populations (Jin et al., 2021). Cell-type annotation relies on known marker genes, while pseudotime methods such as Monocle 3 and Slingshot reconstruct developmental trajectories. Gene regulatory network inference is often performed using SCENIC (Aibar et al., 2017); however, its performance in plants is constrained by animal-centric motif databases (RcisTarget). This limitation has been partially addressed through plant-specific resources such as PlantPAN, JASPAR CORE Plants, and CIS-BP (He et al., 2024a), as well as customized Arabidopsis and rice motif databases integrated into SCENIC frameworks (Bravo González-Blas et al., 2023; Wang et al., 2025c). In addition, plant-oriented tools such as scPlant and MINI-EX provide optimized datasets and workflows for regulatory network reconstruction across major crops (Ferrari et al., 2022; Cao et al., 2023).

2.2. Single-cell epigenome sequencing technologies

Single-cell epigenomic approaches complement transcriptomic profiling by revealing regulatory features that shape gene expression. Among these, single-nucleus Assay for Transposase-Accessible Chromatin sequencing (snATAC-seq) has become a key tool. By leveraging the Tn5 transposase to selectively tag open chromatin regions genome-wide (Fig. 3), snATAC-seq maps cell-type-specific cis-regulatory elements (CREs), including promoters and enhancers, that are masked by cellular heterogeneity in bulk assays (Lu et al., 2024a). An important practical advantage of snATAC-seq is its compatibility with frozen tissues and complex organs, as direct nuclear extraction largely avoids the transcriptional stress responses induced by enzymatic protoplasting. In Arabidopsis roots, droplet-based snATAC-seq identified approximately 8000 regulatory elements sufficient to resolve all major cell types, while an analytical framework linking open chromatin patterns with transcriptional states successfully characterized developmental progression (Dorrity et al., 2021). In rice root tips, scATAC-seq profiling of over 46,000 nuclei revealed cell-type-specific chromatin accessibility variance across most major cell types and delineated chromatin dynamics during root hair differentiation and in response to heat stress (Feng et al., 2022). In maize, a cis-regulatory atlas constructed from snATAC-seq identified approximately one-third of accessible chromatin regions (ACRs) as cell-type-specific, while integrating these data with scRNA-seq uncovered transcription factor binding motifs that drive cell identity (Marand et al., 2021).

Fig. 3.

Schematic overview of the plant snATAC-seq workflow.

The simultaneous profiling of chromatin accessibility and gene expression within the same nuclei, enabled by 10× Genomics single-nucleus Multiome platform, has been applied to investigate stress-responsive regulatory mechanisms in plants. In Arabidopsis seedlings under drought stress, revealing cell-type-specific gene regulatory networks responsive to water deficit (Liu et al., 2024a). Similarly, in Arabidopsis root tips under osmotic stress, single-nucleus Multiome analysis indicated that changes in chromatin accessibility precede gene expression, establishing a framework for subsequent cell differentiation. Stress exposure reduced root hair cell maturation and revealed cell-type-specific regulatory networks with pronounced heterogeneity (Liu et al., 2024b). This multiome strategy has also been applied to Arabidopsis leaves during bacterial pathogen infection, where integrated analyses identified immune cell-associated transcription factor–ACR–gene modules linked to defense responses (Lu et al., 2024a). This joint RNA + ATAC approach directly correlates chromatin state changes with transcriptional output at the single-cell level, providing mechanistic insights into gene regulatory programs that would remain inaccessible through either modality alone.

Despite these successes, snATAC-seq faces several technical challenges that can affect data quality and biological interpretation. These include extreme data sparsity, flawed normalization, Tn5 cleavage bias and difficulties in accurate chromatin state identification (Kwok et al., 2025). To mitigate these limitations, optimized protocols such as scTurboATAC have been developed to improve Tn5 cleavage efficiency, thereby enhancing signal intensity and reducing sparsity (Seufert et al., 2023).

2.3. Spatial transcriptomics (ST) technologies

ST simultaneously captures gene expression profiles while preserving spatial context within tissue architecture. Based on technical principles, ST is broadly divided into sequencing-based and imaging-based approaches (Table 1). Among these, sequencing-based spatial barcoding technologies have developed rapidly in plant research (Xia et al., 2022). Stahl and colleagues pioneered the capture of whole-transcriptome information from intact tissue sections (Stahl et al., 2016). The core of this technique involves immobilizing oligonucleotide probes with unique spatial barcodes on an ordered glass-slide array. After tissue section attachment, the sample undergoes fixation, staining, and permeabilization. Released mRNAs hybridize with adjacent probes, followed by in situ reverse transcription to generate spatially barcoded cDNA. High-throughput sequencing and barcode decoding then enable mapping of each transcript back to its original spatial location (Fig. 2). In 2018, 10× Genomics acquired this technology and commercialized it as the Visium platform. The standard Visium array covers 6.5 mm × 6.5 mm with ∼5000 circular spots (55 μm diameter, 100 μm spacing) and is compatible with H&E staining, although each spot captures multiple cells due to limited resolution. The Visium HD platform replaces these spots with 2 μm × 2 μm squares, increasing capture elements to over 11 million and enabling continuous tissue coverage with substantially improved spatial resolution (Polański et al., 2024). In parallel with the development of array-based platforms, BGI Group introduced Stereo-seq, a DNA nanoball (DNB)-based spatial transcriptomics technology that enables transcript capture at subcellular resolution. Stereo-seq employs patterned arrays containing densely packed DNA nanoballs with center-to-center distances of ∼500 nm, thereby achieving ultra-high spatial resolution over large capture areas (Chen et al., 2022). Compared with Visium, Stereo-seq provides substantially higher spatial resolution and larger field-of-view, enabling spatial mapping across tissues, organs, and even whole organisms.

Table 1.

Comparison of major spatial transcriptomics technologies.

Type	Method	Principle	Resolution	Throughput	Detectable molecules	Tissue compatibility	Advantages and limitations	References
Sequencing-based	10 × Visium	Spatially barcoded capture array coupled with next-generation sequencing (NGS)	55 μm spots (100 μm spacing)	Whole-transcriptome profiling	Whole-transcriptome RNA	FF and FFPE	Mature and standardized workflow with broad applicability, but limited spatial resolution	Stahl et al. (2016)
Sequencing-based	Visium HD	High-density spatial barcode array integrated with CytAssist imaging and NGS	2 μm × 2 μm capture areas	Whole-transcriptome profiling	Whole-transcriptome RNA	FF and FFPE	Near single-cell spatial transcriptomics with whole-transcriptome coverage	Polański et al. (2024)
Sequencing-based	BMKMANU S1000	Microwell array embedded with spatially barcoded beads for transcript capture	4.8 μm	Whole-transcriptome profiling	Whole-transcriptome RNA	FFPE	Near single-cell spatial resolution with compatibility for FFPE tissues	Wang et al. (2023b)
Sequencing-based	Stereo-seq	DNA nanoball-patterned array for spatial RNA capture and sequencing	500 nm feature spacing	Whole-transcriptome profiling	Whole-transcriptome RNA	FF and FFPE	Ultra-high spatial resolution with large field-of-view imaging	Chen et al. (2022)
Sequencing-based	GeoMx DSP	Imaging-guided selection of regions of interest (ROIs) followed by UV-photocleavage and NGS quantification	ROI-level	ROI-based profiling	RNA and proteins (targeted panels)	FF and FFPE	Flexible region-based profiling but lacking true single-cell resolution	Jerby-Arnon et al. (2021)
Sequencing-based	LCM-seq	Laser capture microdissection coupled with Smart-seq2 or related RNA-seq workflows	Single-cell to multicellular ROI	Low throughput	Whole-transcriptome RNA	Primarily FF	High spatial precision but relatively low throughput and labor-intensive workflow	Nichterwitz et al. (2016)
Sequencing-based	Geo-seq	Laser capture microdissection combined with single-cell RNA-seq	Single-cell	Low-to-medium throughput	Whole-transcriptome RNA	FF	Sensitive spatial transcriptome profiling with relatively low throughput	Chen et al. (2017)
Sequencing-based	Slide-seqV2	Randomly distributed spatially barcoded beads for mRNA capture and sequencing	∼10 μm	Whole-transcriptome profiling	Whole-transcriptome RNA	FF	High-resolution whole-transcriptome profiling with improved RNA capture efficiency	Stickels et al. (2021)
Sequencing-based	HDST	High-density bead array containing spatially barcoded beads	2 μm	Whole-transcriptome profiling	Whole-transcriptome RNA	FF	Ultra-high spatial resolution but relatively low transcript capture efficiency	You et al. (2024)
Sequencing-based	DBiT-seq	Orthogonal microfluidic delivery of DNA barcodes across tissue sections	10 μm	Whole-transcriptome and multimodal profiling	RNA and proteins	FF and FFPE	Multimodal spatial profiling of RNA and proteins with high spatial resolution	Liu et al. (2020)
Imaging-based	Molecular cartography	Sequential hybridization-based imaging using highly multiplexed probe sets	Subcellular	Hundreds to > 1000 genes	Targeted RNA panel	FF and FFPE	Automated single-molecule detection with high reproducibility and subcellular resolution	Heiser et al. (2023)
Imaging-based	MERFISH	Combinatorial barcoding and sequential imaging for multiplexed RNA detection	Single-cell to subcellular	∼1000–10,000+ genes	Targeted RNA panel	FF and FFPE	High sensitivity and multiplexing capability but limited to predefined targets	Sarfatis et al. (2025)
Imaging-based	10 × Xenium	Padlock probe hybridization, rolling-circle amplification (RCA), and imaging-based decoding	Subcellular	∼5000 genes	Targeted RNA panel	FF and FFPE	High sensitivity with robust cell segmentation and subcellular localization	Marco Salas et al. (2025)
Imaging-based	CosMx SMI	Cyclic fluorescence imaging of spatially barcoded probes	Single-cell to subcellular	∼1000–6000 genes	Highly multiplexed targeted RNA and protein panels	FF and FFPE	High multiplexing capacity with single-cell and subcellular resolution	Tsourmas et al. (2025)
Imaging-based	smFISH	Fluorescent oligonucleotide probes for single-molecule RNA detection	Subcellular	Tens of genes	Targeted RNA	FF and FFPE	High sensitivity and quantitative accuracy for low-abundance transcripts	Duncan et al. (2016)
Imaging-based	seqFISH+	Sequential rounds of hybridization and imaging for highly multiplexed RNA detection	Single-cell to subcellular	∼10,000+ genes	Targeted RNA panel	FF and FFPE	Highly multiplexed imaging-based transcriptomics with complex experimental workflows	Eng et al. (2019)
Imaging-based	ISS	Padlock probe-based in situ sequencing	Single-cell to subcellular	Hundreds of genes	Targeted RNA	FF and FFPE	High specificity through sequencing-based in situ detection	Ke et al. (2013)
Imaging-based	FISSEQ	In situ reverse transcription and sequencing-by-imaging	Single-cell to subcellular	Whole-transcriptome profiling	Whole-transcriptome RNA	FF and FFPE	Unbiased transcript detection but relatively low transcript capture efficiency and sensitivity	Lee et al. (2014)
Imaging-based	STARmap/STARmap PLUS	Hydrogel-tissue chemistry combined with in situ sequencing	Subcellular	Hundreds to > 1000 genes	Targeted RNA panel	FF	Three-dimensional spatial transcriptomics with high spatial precision	Shi et al. (2023)

Note: FF (fresh frozen tissue), FFPE (formalin-fixed paraffin-embedded tissue).

Imaging-based spatial transcriptomics technologies predominantly rely on multiplexed fluorescence in situ hybridization (FISH) or in situ sequencing (ISS) approaches. Multiplexed error-robust fluorescence in situ hybridization (MERFISH) employs combinatorial binary barcoding schemes and sequential imaging to enable highly multiplexed RNA detection. Through iterative rounds of hybridization, imaging, and barcode decoding, MERFISH enables quantitative detection of thousands of transcripts within intact tissues at subcellular resolution (Chen et al., 2015). Similarly, STARmap performs in situ amplification and sequencing directly within intact tissues, followed by fluorescence imaging-based readout to reconstruct spatially resolved transcriptomic information (Wang and Allen, 2018). These imaging-based approaches provide exceptionally high spatial resolution and molecular sensitivity; however, their transcriptomic coverage is constrained by reliance on predefined probe panels.

At present, the advancement of spatial transcriptomics in plants remains constrained by multiple structural and genomic challenges. Rigid cell walls and complex tissue organization hinder sectioning and permeabilization efficiency, while large central vacuoles can dilute nucleic acid concentrations and reduce detectable signal intensity. In addition, plant secondary metabolites may interfere with enzymatic reactions during library preparation and hybridization steps. Accurate spatial reconstruction also depends on high-quality genome assemblies and gene annotations, which remain incomplete for many plant species, particularly polyploids (Yang et al., 2025b). Overcoming these limitations will require integrated improvements in sample processing, spatial chemistry, imaging resolution, and genome assembly quality, together with plant-optimized bioinformatic frameworks.

2.4. Spatial metabolomics technologies

The core technology driving spatial metabolomics is mass spectrometry imaging (MSI), which primarily includes MALDI-MSI, SIMS-MSI, and DESI-MSI (Table 2; Fig. 2). MALDI-MSI and SIMS-MSI perform ionization under vacuum conditions, whereas DESI-MSI operates under ambient pressure (García-Rojas et al., 2024). MALDI-MSI employs matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) to acquire pixel-resolved mass spectra and reconstruct the two- or three-dimensional spatial distribution of metabolites based on mass-to-charge (m/z) ratios. With a spatial resolution of 10–50 μm, relatively high sensitivity, and low contamination risk, it is widely used for spatial analysis of proteins, peptides, lipids, and secondary metabolites in plants, including studies of lipid metabolism and medicinal compounds (Xie et al., 2025). SIMS-MSI provides submicron spatial resolution and enables subcellular imaging, making it suitable for detecting plant elements and small metabolites. Combined with stable isotope labeling, it can also be applied to nutrient transport tracing, although limited ionization efficiency and tissue damage may occur in some samples (Liu et al., 2026). DESI-MSI is an ambient ionization technique requiring minimal sample preparation, with a typical spatial resolution of 50–200 μm. Owing to its limited disruption of tissue structure, it is suitable for the analysis of intact plant tissues and root exudates, and has been applied in studies of rhizosphere interactions and living plant samples (Liu et al., 2026).

Table 2.

Classification of mass spectrometry imaging techniques.

	MALDI-MSI	SIMS-MSI	DESI-MSI
Operating environment	Vacuum	Vacuum	Ambient conditions
Ionization type	Soft	Hard	Soft
Matrix required	Yes	No	No
Analytes	Small molecules, lipids, biomacromolecules	Elements/ions, small molecules, lipids	Small molecules, lipids, peptides
Mass range	< 100 kDa	< 1000 Da	< 2000 Da
Spatial resolution	10–50 μm	< 1 μm	50–200 μm

Conventional single-omics approaches are widely used in plant research but generally lack tissue- and cell-level spatial resolution, limiting their ability to resolve in situ associations between gene expression and metabolic activity. Recent advances in MSI spatial resolution and detection sensitivity have promoted the integration of spatial metabolomics and transcriptomics into a high-resolution and systematic analytical framework. This strategy enables simultaneous characterization of spatial gene expression and metabolite accumulation within native tissues and establishes in situ links between gene regulation and metabolic products (Lu et al., 2024b). In cotton, AFADESI-Orbitrap MSI-based spatial metabolomics combined with spatial and single-cell transcriptomics revealed coordinated gene–metabolite networks during fiber initiation and elongation, while 10× Visium and AFADESI-MSI further showed co-localization of polyamines (putrescine, spermidine, spermine) and biosynthetic genes during somatic embryogenesis, implicating polyamine metabolism in embryogenic transition (Ge et al., 2025). In medicinal plants, MALDI-MSI integrated with scRNA-seq in Taxus spp. identified taxol-associated cell types and biosynthetic genes, enabling pathway reconstruction (Zhan et al., 2023). Similarly, DESI-MSI with transcriptomics in Salvia spp. and MALDI-MSI with RNA-seq in Rheum tanguticum resolved spatial distributions and regulatory networks of diterpenoids, anthraquinones, and flavonoids (Wang et al., 2025a; Xia et al., 2023). The integration of spatial metabolomics and transcriptomics is shifting plant research from component identification toward spatial functional analysis and provides a framework for investigating natural product biosynthesis, plant developmental regulation, and crop quality formation.

2.5. Computational integration of single-cell and spatial omics

The integration of single-cell and spatial multi-omics has emerged as a key strategy for resolving spatial organization in biological systems. By combining scRNA-seq-derived cellular atlases, snATAC-seq-derived epigenomic profiles, ST-based tissue architecture, and spatial metabolomics-derived metabolic phenotypes, this framework enables spatial multi-omics maps capturing cell identity, gene expression, chromatin accessibility, and metabolic activity (Fig. 2) (Cheng et al., 2023a). However, the integrated computing method of multi-omics is still a difficult problem. Current computational approaches include deconvolution, spatial mapping, and cross-modal representation learning (Table 3). Deconvolution methods such as SPOTlight (Elosua-Bayes et al., 2021) and RCTD (Cable et al., 2022) estimate cell-type composition in spatial transcriptomic spots using single-cell references, but performance depends on reference completeness and declines for rare or unannotated cell types.

Table 3.

Representative computational methods for single-cell and spatial multi-omics integration.

Category	Method	Input data	Core strategy	Main application	Strengths	Limitations
Deconvolution	SPOTlight	scRNA-seq + ST	Non-negative matrix factorization	Cell-type proportion estimation	Interpretable, efficient	Depends on reference quality
Deconvolution	RCTD	scRNA-seq + ST	Probabilistic modeling	Spatial cell composition inference	High accuracy for mixed spots	Computational cost, reference dependency
Spatial mapping	Tangram	scRNA-seq + ST	Optimization-based alignment	Cell localization	Applicable across platforms	Requires shared transcriptomic signals
Spatial mapping	CellTrek	scRNA-seq + ST	Spatial nearest-neighbor projection	Single-cell spatial reconstruction	Enables cell-level mapping	Sensitive to spatial noise
Spatial integration	GraphST	ST ± scRNA-seq	Graph neural network	Spatial domain identification and integration	Incorporates neighborhood information	Higher computational complexity
Cross-modal integration	SpatialMETA	MSI ± ST	Deep generative model	Joint analysis of heterogeneous modalities	Handles multiple omics layers	Limited to low-resolution spatial data
Cross-modal integration	MultiGATE	Spatial multi-omics	Hierarchical graph attention autoencoder	Joint spatial multi-omics representation learning	Preserves spatial structure; integrates heterogeneous data types	Sensitive to graph construction; high computational cost

Spatial mapping methods include Tangram (Biancalani et al., 2021) and CellTrek (Wei et al., 2022) for cell localization, and GraphST (Long et al., 2023) for spatial domain identification and representation learning. Tangram performs optimization-based alignment between single-cell and spatial data, while CellTrek projects cells under spatial and transcriptional constraints. GraphST captures spatial structure using graph-based learning but is computationally demanding.

Cross-modal representation learning methods, including SpatialMETA (Tian et al., 2025) and MultiGATE (Miao et al., 2025), learn shared latent embeddings from spatial multi-modal data. SpatialMETA is mainly applied to low-resolution datasets, whereas MultiGATE is sensitive to graph construction and computationally intensive. Both approaches may reduce interpretability in favor of representation capacity.

Overall, these methods show trade-offs among spatial resolution, interpretability, computational scalability, and dependence on reference quality. Future work is expected to focus on unified frameworks for integrative modeling of transcriptomic, epigenomic, and metabolic processes across spatial scales.

3. Single-cell and spatial insights into plant development

3.1. Root development: cellular organization and lineage specification

Because roots possess relatively simple tissue organization and well-defined differentiation trajectories, they have served as an important system for investigating plant development in single-cell level (Fig. 4). The application of single-cell transcriptomics has advanced root biology from traditional anatomical descriptions to high-resolution characterization of cell types and developmental lineages. Early studies in Arabidopsis thaliana generated comprehensive root cell atlases using single-cell RNA sequencing, resolving major cell types, including the stele, endodermis, cortex, epidermis, columella, and stem cell niche. These analyses also identified a series of cell layer-specific marker genes, such as XCP1, DRN1, APL, VND4, TTG1, and GL2 (Table S2), reconstructed continuous differentiation trajectories from the stem cell niche to mature tissues, and identified key regulators associated with lineage specification (Denyer et al., 2019; Lyu et al., 2025). As single-cell datasets have accumulated across species, comparative analyses have begun to provide insights into the evolutionary basis of root development. Core transcriptional programs associated with meristem maintenance, vascular differentiation, and fundamental cellular functions appear to be largely conserved between monocots and dicots (Liu et al., 2021; Zhang et al., 2021b). In contrast, outer tissues such as the epidermis and cortex exhibit greater transcriptomic divergence.

Fig. 4.

Illustration of single-cell and spatial multi-omics applications in plant development, metabolism, and stress responses. Root: cell-type mapping across tissues, developmental trajectories during root growth, and nutrient uptake mediated by specialized cell populations. Leaf: reconstruction of developmental gradients, spatial photosynthetic heterogeneity, and secondary metabolic programs linked to tissue differentiation. Stem: secondary growth involving vascular cambium activity, regulatory networks of wood formation, and spatial patterns of cell wall deposition during xylem differentiation. Flower: reproductive cell lineage specification, transcriptional programs of floral development, and regulation of fertility. Abiotic stress: cell type-specific responses to drought, salinity, and temperature, including adaptive regulatory pathways and intercellular coordination. Biotic stress: spatial host–microbe interactions, immune activation across tissues, and coordinated defense responses. Central cellular processes: control of cell totipotency, regeneration trajectories in tissue repair, and metabolic reprogramming underlying cellular plasticity.

Studies in non-model species have further revealed lineage-specific evolutionary innovations. Single-cell atlases of Phyllostachys edulis indicate that, despite both belonging to monocots, the organization of outer root tissues differs substantially from that of rice. In rice, the epidermis and ground tissue originate from a common initial cell population, whereas root tip tissues in P. edulis appear to have undergone lineage restructuring during evolution, resulting in a distinct developmental pattern in which the root cap and epidermis share a stem cell lineage (Cheng et al., 2023b).

Comparisons across broader evolutionary scales further highlight both conservation and diversification in plant cell type evolution. A recently established unified single-cell atlas of vascular plants, integrating major evolutionary lineages including Selaginella moellendorffii, Arabidopsis thaliana, and Ceratopteris richardii, showed that many cell type-specific genes maintain conserved expression patterns across taxa. For example, homologs of xylem-associated genes (VND1 and TMO5), cortex-associated genes (SHR and SCR), pericycle-associated genes (PFA2 and AHP6), and endodermis-associated genes (CASP1 and SGN) display broad conservation across species (Yang et al., 2023). At the same time, developmental regulatory networks may undergo functional reconfiguration during evolution. In non-seed plants, WOX13 homologs promote root cap formation, whereas Arabidopsis WOX5 functions in maintaining stem cell indeterminacy, indicating divergence in cell fate regulatory mechanisms (Yang et al., 2023).

The development of spatial omics technologies has enabled the integration of cell identity information with tissue architecture, overcoming the lack of spatial context inherent to dissociation-based single-cell approaches. Stereo-seq studies in peanut roots and soybean nodules not only validated cell types identified by single-cell analyses but also revealed spatial organization patterns and dynamic changes of distinct cell populations along the root axis and across tissue layers (Liu et al., 2022, 2023c). In parallel, spatial metabolomics studies have identified pronounced metabolic compartmentalization within root tissues. For example, ginsenosides are primarily enriched in epidermal regions, whereas tanshinones are preferentially localized in periderm tissues (Li et al., 2026b; Yang et al., 2025a).

Collectively, evidence from single-cell and spatial multi-omics studies across diverse plant lineages suggests that root development is not solely governed by species-specific mechanisms but is supported by conserved cellular identity programs shared among vascular plants. Conservation and diversification of plant cell types are ultimately manifested through specific spatial tissue organization and metabolic compartmentalization, which together contribute to root development and adaptive traits.

3.2. Leaf development and metabolic specialization

Leaves are highly specialized organs that integrate photosynthetic activity with diverse metabolic processes through multiple distinct cell types (Fig. 4). Recent advances in single-cell and spatial multi-omics have enabled the dissection of leaf development and functional specialization at cellular resolution, revealing how developmental programs, spatial organization, and metabolic compartmentalization are tightly coordinated. In Arabidopsis, single-cell transcriptomics reconstructed continuous differentiation trajectories from leaf primordium cells to epidermal, mesophyll, and vascular lineages. It also confirmed that stomatal development is regulated by the transcription factors SPCH, MUTE, and FAMA (Table S2) (Lopez-Anido et al., 2021). In maize, single-cell transcriptomics combined with gene regulatory network analysis resolved the transition from proliferative mesophyll precursors to photosynthetically specialized cells. Key regulators include HSF, COL, WRKY, and ERF transcription factors (Tao et al., 2022). Although these two species represent C3 and C4 photosynthetic systems, respectively, both rely on hierarchical transcriptional programs that coordinate cell proliferation and functional differentiation, suggesting a conserved developmental logic underlying leaf cell fate specification.

Beyond lineage reconstruction, spatial transcriptomics reveals spatial organization of leaf functions. In Arabidopsis, high-resolution Stereo-seq mapping shows that photosynthesis-related genes increase in expression from the midvein toward the leaf margin. Genes related to transport and ion homeostasis are enriched in vascular tissues (Xia et al., 2022). Distinct functional zones are also observed between adaxial and abaxial epidermis and between palisade and spongy mesophyll (Xia et al., 2022). These results indicate that leaf functional specialization is regulated by both developmental lineage and spatial position. Local microenvironment also contributes to this regulation. Integration of single-cell and spatial transcriptomics therefore provides a more complete framework for understanding leaf development and function.

In medicinal plants and specialized crops, single-cell and spatial multi-omics studies show that secondary metabolism is generally spatially organized. However, the degree of cellular division of labor and pathway organization varies across species. One mode involves cooperative metabolism across multiple cell types. In Catharanthus roseus, monoterpene indole alkaloid biosynthesis is distributed across phloem-associated parenchyma cells, epidermal cells, and idioblasts. Complete biosynthesis depends on intercellular transport of intermediates (Sun et al., 2022). Another mode involves strong metabolic specialization in defined structures or cell types. In Artemisia annua, MVA/MEP precursor pathways and key artemisinin biosynthetic genes are mainly expressed in apical and subapical cells of glandular trichomes. Basal cells contribute partially to precursor supply. This indicates that artemisinin biosynthesis is largely confined within glandular trichomes (Zhang et al., 2025). Similarly, hyperforin biosynthesis in Hypericum perforatum is highly enriched in specialized Hyper cells (Wu et al., 2024). In Camellia sinensis, flavonoids mainly accumulate in palisade mesophyll cells (Zhao et al., 2026).

These different patterns may reflect either true biological divergence or methodological limitations. From a biological perspective, metabolic compartmentalization may reduce cytotoxicity of intermediates, minimize interference between competing pathways, and improve metabolic efficiency (Panda et al., 2021). Storage in specialized structures such as glandular trichomes, idioblasts, secretory cavities, or laticifers may further enhance stress resistance (Al-Khayri et al., 2023). Technical factors may also affect interpretation. Protoplast isolation can lead to loss of fragile or specialized cell types. For example, laticifer cells were not detected in Catharanthus roseus studies. Hyper cells in Hypericum perforatum are also rare in leaf tissues, and their recovery is lower than that of common parenchyma cells. In addition, differences in scRNA-seq, snRNA-seq, and spatial omics platforms in capture efficiency, sensitivity, and spatial resolution may lead to inconsistent pathway localization. Therefore, robust conclusions require integration of spatial transcriptomics, spatial metabolomics, isotope tracing, and functional validation.

3.3. Stem secondary growth and wood formation mechanisms

Radial growth of woody stems depends on secondary growth driven by the vascular cambium, which involves sustained cell proliferation, directional differentiation into xylem and phloem, and coordinated secondary cell wall deposition together with programmed cell death (PCD). Recent single-cell and spatial multi-omics studies suggest that wood formation is not a set of discrete stages but a continuous transcriptional transition field, in which cell fate transitions, lineage plasticity, and spatial positional signals are tightly coupled. In Populus spp., single-cell transcriptomic and in situ analyses have resolved major xylem cell types, including vessel elements, fibers, and ray parenchyma cells, together with stage-specific marker genes (Table S2) (Li et al., 2021). However, single-cell platforms show systematic bias in cell-state capture. scRNA-seq tends to enrich early, less lignified cells, whereas snRNA-seq captures more late-stage, highly lignified secondary wall-forming cells, with overlap mainly in transitional states. Thus, observed heterogeneity likely reflects both biological continuity and technical sampling differences. Integration of scRNA-seq, snRNA-seq, and LCM-based spatial transcriptomics enables reconstruction of a continuous trajectory from cambium to PCD. These analyses indicate that secondary wall formation and PCD initiation are not strictly sequential but partially overlapping transcriptional programs. Along this trajectory, early xylem cells show higher transcriptional heterogeneity, while late cells become more uniform, a pattern consistent across datasets and platforms (Wei et al., 2026).

Cross-species comparisons between Arabidopsis thaliana and Populus indicate a conserved core regulatory framework for xylem formation. The VND, XCP, and MYB gene families are involved in secondary wall formation in both species, although differences are observed in timing and cell-type specificity (Table S2). In Arabidopsis, SND1 mainly regulates fiber differentiation, whereas in Populus its transcripts are also detected earlier in progenitor populations contributing to both fibers and ray parenchyma cells, suggesting a broader activity window (Li et al., 2021). ST analyses further identify two spatially separated cambium-like cell populations: rectangular procambium-like (PCL) cells in the phloem domain and spindle-shaped cambial zone (CZ) cells in the cambial domain. PCL cells are derived from primary root procambium and mainly contribute to phloem, whereas CZ cells arise from transitional states and mainly contribute to xylem. Both populations share proliferative and stem-like markers, but phloem identity genes are enriched in PCL, while xylem and cambial identity genes are enriched in CZ. Along the PC-to-CZ transition, cell cycle genes decrease, while WOX4 and PXY increase, indicating a staged fate transition program (Du et al., 2023).

Across vascular plants, integrated single-cell analyses indicate that major cell types are broadly conserved across evolution, including epidermal, xylem, and phloem cell types. However, in ferns and gymnosperms, cell populations with transcriptional profiles similar to companion cells have been detected. These cells express homologs of companion cell marker genes such as EFM, AAP5, and FTIP1, and are located adjacent to sieve elements, although their morphology cannot be clearly distinguished (Xue et al., 2025). This observation should be interpreted as a companion-cell-like transcriptional state rather than direct structural homology, and may reflect either ancestral regulatory programs or convergent evolution. Overall, Single-cell and spatial multi-omics reveal woody plant secondary growth is controlled by a hierarchical system integrating conserved developmental modules, lineage-specific gene family variation and domain spatial arrangement.

3.4. Floral organogenesis and reproductive cell lineages

During floral organogenesis, diverse plant species undergo a conserved transition from meristematic cells to organ primordia, while exhibiting species-specific regulatory networks and spatial organization. In Arabidopsis, snRNA-seq shows that stage-specific transcriptional changes in floral meristem cells are closely linked to cell-cycle progression: the MADS-box transcription factor FUL and KRP2-associated pathways have been implicated in regulating the G2/M transition (Chen et al., 2025). In grasses, regulatory and spatial features differ. In barley inflorescences, MADS-box genes display stage-specific spatiotemporal expression, and mutant analyses support their roles in the transition from meristem to organ identity (Demesa-Arevalo et al., 2026). In maize, single-cell profiling resolved the inflorescence meristem into three subpopulations, including boundary, adaxial periphery, and base, with primordium-associated transcription factors linked to yield-related variation (Xu et al., 2021). Overall, meristem-to-organ transition is broadly conserved, but interspecific differences in cell-cycle and spatial patterning likely reflect evolutionary divergence and methodological variation.

The formation of reproductive cell lineages also displays substantial variation in regulatory programs and analytical strategies across species. Female germline cells are typically embedded deep within reproductive organs and occur at very low abundance, making their capture technically challenging. High-resolution single-cell transcriptomic analyses in Arabidopsis have reconstructed the developmental trajectory from sporogenous cells to the megaspore mother cell and identified transcriptional signatures associated with reproductive and non-reproductive fate specification at developmental branching points (Hou et al., 2021). In contrast, male reproductive development is generally more amenable to comparative analyses across multiple samples. Single-cell comparisons between maize CMS-C lines and fertility-restored lines revealed transcriptional variation following meiosis and provided evidence that Rf4 contributes to the restoration of transcriptional activity in genes containing E-box motifs (Zhang et al., 2021a).

The development of reproductive-derived structures further illustrates how similar cellular morphologies may arise through distinct developmental routes. Cotton fiber cells originate from ovule epidermal cells and enter a predominantly elongation-driven developmental program after initiation, accompanied by dynamic expression of hormone-related and circadian-associated genes (Wang et al., 2023a). In contrast, fiber cells of Bombax ceiba originate from epidermal cells of the inner ovary wall and undergo an additional round of cell division following differentiation, resulting in multicellular fiber clusters (Ding et al., 2023). Conventional bulk RNA-seq approaches often have limited ability to disentangle transcriptional differences arising from heterogeneous tissue origins. Single-cell analyses reduce interference from non-fiber cell populations and facilitate direct comparisons of fiber-development programs, revealing both shared fiber-related gene expression patterns and differences in cell-cycle regulation between the two species.

Overall, two concurrent patterns can be observed during flower development and reproductive lineage formation. One involves regulatory modules that recur across angiosperms, including MADS-box-associated pathways and cell-cycle regulation during the meristem-to-organ transition. The other comprises species-specific expression patterns associated with differences in developmental trajectories and structural organization, such as variation in reproductive lineage progression and the formation of reproductive-derived structures. The integration of single-cell and spatial multi-omics not only provides a common framework for comparing cellular states and regulatory networks across plant species, but also offers an important basis for evaluating evolutionary interpretations within their spatial context.

3.5. Cellular totipotency and plant regeneration mechanisms

Plant cells universally possess remarkable totipotency and regenerative capacity. Analysis of Arabidopsis hypocotyl-derived calli utilizing scRNA-seq has demonstrated that their internal architecture exhibits a three-layered structure analogous to root primordia or root apical meristems, with the middle-layer cells possess defining characteristics of the root quiescent center (QC) (Zhai and Xu, 2021). By activating transcriptional networks mediated by key regulatory genes including WOX5 and PLT, and synergistically modulating hormone signaling pathways, these cells acquire pluripotency and ultimately, they serve as the primary cellular source for regenerated shoots or roots. During the somatic embryogenesis of Dimocarpus longan, the application of single-cell technologies has delineated the developmental trajectories progressing from embryogenic calli to early somatic embryos, directly identifying critical transitional states including proliferating cells and meristematic cells, alongside their corresponding marker genes (Zhang et al., 2023). Building upon this foundation, ST has resolved the spatial dynamics underlying organ regeneration. Within the organogenesis model of tomato calli, this technology has precisely localized the specific spatial coordinates of shoot primordium initiation and comprehensively reconstructed the spatiotemporal evolutionary processes governing vascular tissue reestablishment (Song et al., 2023).

At a finer mechanistic level, recent single-cell studies indicate that reprogramming toward totipotency can also proceed through direct lineage conversion at single-cell resolution. In Arabidopsis, the transcription factor LEC2 reprograms somatic epidermal cells into totipotent states by activating local auxin biosynthesis via TAA1 and YUC4 in coordination with SPEECHLESS (SPCH)-expressing meristemoid mother cells, thereby establishing a direct reprogramming route that bypasses an intervening callus phase and highlighting an alternative regeneration pathway governed by lineage plasticity and localized hormonal regulation (Tang et al., 2025). Cell fate reprogramming is not solely dependent on transcriptional network reconstruction but is also accompanied by systemic reorganization of metabolic pathways. Spatial metabolomic analysis of somatic embryogenesis in cotton reveals that metabolites such as linoleic acid, proline, polyamines, and zeatin exhibit pronounced spatial distribution disparities across heterogeneous embryonic regions, particularly the cortex, the cotyledons, and the vascular tissues (Ge et al., 2025). By becoming integrated into core biological processes including lipid metabolism, hormone signal transduction, antioxidant responses, and nucleotide synthesis, these metabolites synergistically dictate the capacity for cell fate reprogramming and the trajectory of orderly differentiation at the chemical molecular level.

4. Multi-omics dissection of plant stress responses

4.1. Abiotic stress responses at single-cell and spatial resolution

The application of single-cell and spatial multi-omics technologies has profoundly revealed the high degree of cell type specificity in plant abiotic stress responses, overturning the traditional perspective based on uniform tissue or organ level responses (Fig. 4). Recent studies have increasingly shifted the focus from static cell-type-specific responses to dynamic cell-state transitions, in which cells may adopt stress-responsive, adaptive, or recovery states in response to environmental changes. Distinct cell types often deploy different molecular strategies under the same stress condition, resulting in substantial response heterogeneity. However, plant adaptation to environmental stress is not solely determined by individual cell types but also depends on coordinated regulation among different cellular populations. Accordingly, the inference of cell–cell communication networks based on ligand–receptor expression patterns has emerged as an important direction in single-cell data analysis. Computational frameworks such as CellChat, CellPhoneDB, and NicheNet have been widely applied in animal systems and provide useful references for plant studies (Jin et al., 2023). Given the unique features of plant signaling systems, including receptor-like kinases, peptide-mediated signaling, and hormone-regulated pathways, plant-specific resources such as PlantPhoneDB and ScTensor have been developed to facilitate the construction of putative ligand–receptor interaction networks and the identification of signal-sending and signal-receiving cell populations (Xu et al., 2022; Tsuyuzaki et al., 2023). Integration with spatial transcriptomic data may further improve the characterization of signal transmission and functional coordination at the tissue level. Although current analyses remain limited by the availability of curated ligand–receptor annotations and standardized validation frameworks, the continued expansion of plant single-cell atlases and spatial omics datasets is expected to broaden the application of cell–cell communication analysis in studies of plant stress adaptation.

Emerging evidence suggests that stress adaptation is coordinated through signaling interactions among distinct cell populations. During phosphate starvation in Arabidopsis, vascular cells promote cytokinin biosynthesis through activation of the TMO5-LHW transcriptional network. The resulting mobile signals diffuse outward and act on epidermal cells, regulating their elongation and differentiation, thereby increasing root hair density and enhancing phosphorus acquisition (Wendrich et al., 2020). The response of the maize root apex to nitrate nitrogen further highlights this spatial division of labor, epidermal and root hair cells exclusively execute the absorption and primary assimilation of nitrate nitrogen through the upregulation of ZmNAR2.1 and ZmGS2 genes, while meristematic zone cells dominate nitrate nitrogen reduction via the high expression of ZmNIR1 (Li et al., 2022).

Confronting osmotic stress (such as salinity and drought) and oxidative stress, diverse plant cell types similarly construct a complementary defense network. Transcriptomic dynamics of the cotton root apex under salt stress reveal that epidermal, root hair, and root cap cells are enriched in stress responses, glutathione metabolism, and defense reactions establishing primary physical and biochemical barriers. In contrast, cortical, xylem, and endodermal cells are predominantly enriched in metabolic processes including amide biosynthesis, translation, and peptide synthesis driving osmotic regulation and maintaining cellular homeostasis and water transport functions (Li et al., 2024). Under drought stress conditions, Arabidopsis mesophyll cells spatially differentiate into two functionally complementary populations: one subpopulation activates the classical abscisic acid dependent pathway to cope with water deficit, whereas the other specifically initiates iron starvation response genes. This division of labor at the cellular level synergistically maintains ion homeostasis and reactive oxygen species balance, achieving adaptive defense of the entire tissue within a complex environment characterized by intertwined osmotic and oxidative stresses (Tenorio Berrío et al., 2025).

Spatial metabolomics provides intuitive chemical evidence for these transcriptional networks, profoundly elucidating the metabolic mechanisms underlying cellular specialization during stress responses. In the halophyte Kandelia obovata, flavonoid precursors involved in the synthesis of condensed tannins are specifically enriched within mesophyll cells, and this active metabolic zone highly coincides with the spatial locations where sodium ions are sequestered into vacuoles, directly corroborating the cell-type-specific mechanisms of salt tolerance (Huang et al., 2024). Additionally, heterogeneous alterations in the spatial distribution of amino acids and organic acids within the embryo and endosperm of wheat grains, induced by a prolonged high carbon dioxide environment, intuitively demonstrate the cellular metabolic foundation through which climate change impacts crop nutritional quality (Li et al., 2023).

Collectively, these studies highlight that abiotic stress responses at cellular resolution are inherently shaped by the spatial organization of tissues and the functional specialization of cell types. Distinct cellular populations exhibit coordinated yet non-uniform transcriptional reprogramming, often forming spatially structured patterns associated with their physiological roles. At the same time, stress adaptation relies on the integration of intercellular signaling and dynamic shifts in cellular states, enabling plants to maintain metabolic balance and developmental plasticity under fluctuating environmental conditions.

4.2. Biotic stress responses revealed by multi-omics

Single-cell and spatial multi-omics systems have systematically revealed the dynamic networks governing interactions between plants and microbes, spanning from microscopic cellular responses to macroscopic tissue coordination (Fig. 4). Rather than being uniformly activated, plant immune responses are spatially confined into localized defensive microdomains centered around pathogen invasion sites. During the initial stages of interaction, scRNA-seq precisely captures the heterogeneous responses of the host. Under pathogenic stress, host cells exhibit a continuous spectrum of states ranging from highly activated to completely suppressed. The pathosystem comprising Arabidopsis and Pseudomonas syringae has comprehensively mapped the dynamic transitions among immune, susceptible, and transitional cell states (Zhu et al., 2023). This process often forms a spatially layered activation pattern, where pathogen-contacting cells initiate primary defense responses that subsequently propagate to adjacent and distal cells through intercellular signaling. Upon the infection of wheat by Fusarium graminearum, this cellular differentiation evolves into a defensive division of labor at the tissue level: phloem and bundle sheath cells synergistically confer disease resistance, whereas the pathogen induces transcriptional repression within chlorenchyma cells to facilitate infection (Wei et al., 2025).

Spatial multi-omics further anchors these heterogeneous cells to the physical tissue architecture, elucidating the spatial deployment strategies of defense and symbiotic modules. The pathosystem of rice and Magnaporthe oryzae demonstrates a defense gradient strictly dependent upon tissue context, where immune receptors are enriched within vascular tissues and exhibit a progressive enhancement extending from the leaf base toward the apex (Wang et al., 2025b). The pathosystem involving potato and late blight reveals a strict spatial division of labor: cells targeted by pathogens execute localized defense mechanisms, while pericycle cells dominate systemic immunity (Li et al., 2026a). At the level of metabolic execution, spatial metabolomics directly reconstructs the in situ biosynthetic pathways of defense compounds. During the graft healing period in grapevine, the site of xylem injury exhibits a spatiotemporal progression, characterized by the sequential conversion of stilbene monomers into oligomers (Loupit et al., 2023). When maize defends against aflatoxins, the accumulation of toxins within the roots and the concurrent synthesis of benzoxazinoids in the aerial parts constitute a spatial complementary defense mechanism spanning across organs (Righetti et al., 2021).

The integration of single-cell and spatial multi-omics weaves isolated cellular differentiation and spatial localization into complete developmental trajectories and global synergistic networks. Within the symbiotic system of soybean and rhizobia, the combination of single-nucleus RNA sequencing and Stereo-seq technology not only maps the in situ distribution of cell types but also reconstructs the developmental timeline through RNA velocity analysis, successfully validating the functions of key regulatory factors including GmbHLH93 (Liu et al., 2023c; Sun et al., 2023). High precision spatiotemporal cellular atlases indicate that nodule maturation represents a dynamic specialization process. Uninfected cells within the central infection zone undergo transformation into heterogeneous functional subpopulations, and core genes such as leghemoglobin alongside essential nitrogen fixation metabolic pathways are precisely allocated to specific cellular spaces (Liu et al., 2023c).

Taken together, these findings redefine plant–microbe interactions as spatially organized and dynamically coordinated processes, in which distinct cell populations adopt specialized and context-dependent roles. Rather than a uniform immune activation, plant defense emerges from the integration of localized signaling, cell-type-specific regulatory programs, and intercellular communication networks. This spatially compartmentalized and state-driven framework provides a conceptual basis for understanding how plants balance defense and growth under complex biotic environments.

5. Challenges and future perspectives

5.1. Technical bottlenecks in plant single-cell and spatial multi-omics

Despite continuous methodological advancements, plant single-cell and spatial multi-omics remain fundamentally constrained by critical bottlenecks encompassing sample preparation, technical performance, data interpretation, and functional validation. First, sample preparation remains a primary technical challenge in plant single-cell studies. Plant cell walls are composed of cellulose, polysaccharides, lignin, and other components, and their structural complexity substantially increases the difficulty of enzymatic protoplast isolation, particularly in tissues with thickened secondary walls or highly specialized wall composition. Protoplasting itself may also induce stress-related transcriptional responses and introduce cell capture bias due to differences in cell wall properties among cell types (Grones et al., 2024). Single-nucleus transcriptomics can avoid part of the perturbation associated with enzymatic digestion, but it mainly captures nuclear RNA and often shows a bias toward nuclear transcripts, together with relatively low transcript recovery. It therefore remains systematically different from whole-cell transcriptomics (Sunaga-Franze et al., 2021). Recently developed methods such as FX-Cell use tissue fixation, RNase removal, and RNA protection strategies to reduce mRNA degradation during digestion, thereby improving RNA integrity and gene detection efficiency (Ming et al., 2025). However, these approaches mainly represent workflow optimization, and standardized low-bias and reproducible preparation protocols are still lacking across species and tissue types.

At the data analysis level, plant single-cell and single-nucleus omics remain limited by signal sparsity, insufficient annotation resources, and incomplete adaptation of analytical pipelines. In plant scATAC-seq/snATAC-seq, successful analysis requires efficient nucleus isolation, which is often hindered by cell wall properties and endogenous metabolites, necessitating species- and tissue-specific protocol optimization (Lu et al., 2024a). Data sparsity and the lack of plant-specific analytical tools further complicate cell-type annotation, particularly in non-model species where marker genes remain scarce. Although databases such as PCMDB, scPlantDB, PscOA, and PlantscRNAdb 4.0 provide valuable references, they are largely based on Arabidopsis and rice (Jin et al., 2022; He et al., 2024b; Xu et al., 2026; Zheng et al., 2026). Current approaches integrate gene activity scores, motif enrichment, marker peaks, and scRNA-seq-guided annotation, supported by tools including SEMITONES, COSG, Cellcano, and EpiAnno (Lu et al., 2024a). However, annotation accuracy still depends heavily on data quality, reference completeness, genome assembly quality, and the transferability of marker genes across species, highlighting the need for plant-specific analytical frameworks and expanded reference resources.

Spatial omics is constrained by inherent trade-offs between spatial resolution, molecular coverage, and throughput. Sequencing-based spatial transcriptomics typically operates at subcellular-to-multicellular resolution, leading to partial signal mixing across neighboring cells and limiting precise delineation of cellular boundaries and microenvironments (Fan et al., 2023). While higher-resolution platforms improve spatial precision, they reduce capture efficiency and increase experimental cost and data complexity (Chen et al., 2022). Imaging-based approaches provide near in situ resolution but remain restricted by probe design capacity and multiplexing limits, as well as optical crowding and signal deconvolution challenges (Liu et al., 2023a). Similar constraints extend to spatial metabolomics, where achieving high spatial fidelity requires balancing sensitivity, metabolite identification confidence, and tissue-induced matrix effects. In plant tissues, additional complications arise from ion suppression, metabolite instability, and sectioning-induced redistribution (Li et al., 2025). Collectively, these limitations highlight that a central challenge is not only spatial mapping itself, but the construction of quantitatively reliable, multi-omic spatial regulatory networks linking metabolites with transcriptional programs.

As plant multi-omics studies move from single-modality profiling toward joint analysis of scRNA-seq, scATAC-seq, spatial transcriptomics, and spatial metabolomics, data integration has emerged as another major computational bottleneck. Differences among platforms in resolution, noise structure, and missing-data patterns make high-dimensional alignment and integration strongly dependent on algorithmic assumptions and reference quality. Existing methods can support cell deconvolution, spatial mapping, and cross-modal label transfer when suitable reference atlases are available, but their performance may become less stable in tissues with complex organization, stress-induced cell states, or large numbers of previously unseen cell types (Biancalani et al., 2021). Batch effects introduced by different experimental platforms and sample preparation procedures also remain insufficiently quantified in plant systems, which further limits cross-study integration and comparability. One large-scale Arabidopsis study benchmarked seven integration methods using ten complementary metrics for batch-effect removal and biological conservation, with RPCA achieving the best overall balance across 63 scRNA-seq datasets (Cao et al., 2024). Nevertheless, comparable benchmarking across species, tissues, platforms, and experimental conditions remains scarce. Establishing unified evaluation criteria and benchmark frameworks for plant multi-omics data therefore remains an important task. More fundamentally, most current integrative analyses still operate largely at the level of correlation. Dynamic modeling frameworks that can connect regulatory networks, cell-state transitions, and metabolic phenotypes in a unified manner remain underdeveloped, and this limitation is particularly evident in plant studies aimed at resolving metabolic mechanisms.

The limited development of functional validation systems further constrains the transition from descriptive atlases to mechanistic studies. Although single-cell and spatial multi-omics can identify candidate cell populations, regulatory factors, and spatial features at high resolution, the corresponding validation strategies remain relatively underdeveloped. In addition, non-model and medicinal plants often lack mature cell-type-specific promoters, stable transformation systems, and in situ perturbation tools, which restricts systematic functional testing (Moskal et al., 2025). Taken together, future progress from atlas construction to mechanistic dissection will require not only continued optimization of sample preparation, analytical workflows, and multi-omics integration, but also the establishment of validation frameworks that connect molecular measurements with cell-type-specific perturbation and spatially resolved functional evidence.

5.2. Emerging directions and technological convergence in plant multi-omics

To address the limitations outlined above, future development of plant single-cell and spatial multi-omics will likely depend on progress in three interconnected areas: improving the robustness of sample preparation and spatial profiling, developing more interpretable strategies for multi-omics integration, and extending these approaches to non-model and agronomically important plant systems. Progress in these areas will also require continued expansion of reference atlases, cell marker resources, metadata standards, and cross-study benchmarking frameworks, so that datasets generated across species, tissues, and experimental conditions can be compared and reused more reliably.

A first priority is to improve the robustness and adaptability of experimental workflows across diverse plant tissues. Sample preparation in plants is constrained by cell wall composition, tissue heterogeneity, and transcriptional perturbation induced during enzymatic dissociation (Grones et al., 2024). Future methodological development may therefore rely less on a single universal workflow and more on modular, tissue-adapted strategies. These may include optimized nucleus-based protocols for lignified or mature tissues, improved fixation and RNA preservation procedures, and dissociation or sectioning workflows tailored to different species and organs. In spatial omics, gains in resolution and sensitivity depend not only on instrumentation, but also on sample preservation, spatial registration accuracy, and the chemically complex background of plant tissues. Reducing experimental perturbation while improving reproducibility across laboratories and plant lineages therefore remains a key methodological goal.

A second major direction is the development of multi-omics integration strategies that improve the interpretation of cell states (Fig. 5). Transcriptomic data are central to cell identity analysis, but they do not fully capture the regulatory complexity contributed by chromatin state, epigenetic regulation, protein abundance, and metabolic specialization. In plants, this limitation is particularly relevant during developmental transitions and environmental responses, where transcriptional regulation is closely linked to hormone signaling, chromatin remodeling, and specialized metabolism. Current studies often combine sc/snRNA-seq with chromatin accessibility data or jointly analyze spatial transcriptomic and spatial metabolomic datasets. At present, however, the main challenge is less the generation of multimodal data than the development of analytical frameworks that can accommodate sparse signals, partially overlapping feature spaces, and modality-specific noise structures. Existing studies suggest that integration of chromatin accessibility, developmental trajectory, and spatial context can improve the interpretation of lineage progression and regulatory network inference, but these analyses remain strongly dependent on reference quality, and many models still show limited interpretability and limited generalization across biological conditions. Future methodological work will therefore need to place greater emphasis on interpretable cross-modal alignment, uncertainty-aware modeling, and evaluation frameworks based on biologically matched perturbation datasets rather than computational alignment metrics alone.

Fig. 5.

Types of Single-Cell Multi-omics: This figure illustrates various methods in single-cell multi-omics and shows how different levels of biological data—including transcriptomics (scRNA-seq), epigenomics (snATAC-seq), proteomics, and metabolomics—are integrated at the single cell level.

Foundation models are emerging in plant single-cell research, but their application remains limited by small datasets, inconsistent annotations, and weak cross-species transferability. Although models such as scGPT demonstrate the potential of self-supervised learning, benchmark studies show limited advantages over simpler methods in zero-shot settings (Kedzierska et al., 2025). These challenges are amplified in plants because of limited species diversity, inconsistent annotation standards, widespread gene duplication, polyploidy, and complex developmental–environmental interactions (Guillotin et al., 2023). Plant-specific foundation models, including scPlantLLM and scPlantFormer, represent initial progress but remain largely trained on model species and require broader validation (Zhang et al., 2024; Cao et al., 2025). In the near term, foundation models are likely to be most useful for workflow optimization, cross-dataset annotation, candidate regulator prioritization, and standardized data analysis rather than end-to-end prediction.

A third important direction is the extension of single-cell and spatial multi-omics to non-model and crop species. This shift is important for understanding the evolutionary diversification of plant cell types and for linking cell states to agronomic traits and environmental adaptation. However, non-model plants present additional challenges in sample preparation, annotation, and functional validation. At the experimental level, protocols often need to be adapted to lineage-specific tissue properties, including lignified cell walls, high levels of polysaccharides or phenolic compounds, and large vacuoles. In many species, nucleus-based approaches may be more feasible than protoplast-based workflows, but further optimization of nucleus isolation and quality control is still required (Cervantes-Pérez et al., 2022).

Reference resources will be particularly important for cell-type annotation in non-model plants. This includes the development of cross-species marker frameworks informed by orthology, expansion of cell atlases across representative organs and developmental stages, and incorporation of spatial information to improve annotation consistency. Marker genes are not always strictly conserved across plant lineages, and one-to-one orthology alone is often insufficient for polyploid crops or taxa with extensive gene family expansion. Recent cross-species annotation strategies have therefore moved beyond simple ortholog mapping toward approaches that combine homology, expression similarity, and anatomical correspondence (Haghani et al., 2026). Future annotation frameworks will likely need to integrate orthology, co-expression patterns, chromatin information, and spatial context, while continuously incorporating new species and tissue datasets into reference atlases to improve robustness and scalability.

Finally, annotation validation remains a major limitation in non-model plants that lack mature genetic manipulation systems. In such systems, cell-type annotation often needs to rely on convergent evidence from multiple sources, including spatial transcriptomics or in situ hybridization, histological and cytological features, consistency with tissue anatomy and developmental trajectories, and physiological or metabolic measurements. When stable transformation systems are unavailable, transient expression assays, laser capture microdissection, and targeted transcript detection may serve as alternative validation strategies. Whenever possible, it should be supported by mutant comparisons, validation in homologous tissues across species, chromatin accessibility data, and spatial localization information, so that candidate regulatory relationships can be evaluated more rigorously (Shahan et al., 2022; Liu et al., 2023b; Nolan et al., 2023). Future studies should also report annotation confidence explicitly and distinguish well-supported cell types from candidate cell populations that remain insufficiently validated. This will be particularly important for separating conserved cross-species cell states from candidate states that may be lineage-specific.

In conclusion, progress in plant single-cell and spatial multi-omics is unlikely to depend on any single technical advance. It will more likely require coordinated development of experimental workflows, multi-omics integration methods, plant-oriented computational models, and validation frameworks. These efforts should strengthen the connection between molecular measurements, cell states, and plant phenotypes, and support the extension of single-cell and spatial multi-omics from model plants to a broader range of non-model systems. In the next, single-cell spatiotemporal multi-omics research across multiple plant species and even the entire plant kingdom will become a key focus, and play an important role in investigating the origin, evolution, adaptation, genetic and developmental processes of plants and their specific organs.

CRediT authorship contribution statement

Ping Xu: Conceptualization, Software, Visualization, Writing - original draft, Writing - review & editing. Baozheng Wang: Investigation, Software, Writing - review & editing. Yanting Hu: Conceptualization, Investigation, Writing - review & editing. La Qiong: Funding acquisition, Methodology, Supervision, Writing - review & editing. Ticao Zhang: Funding acquisition, Methodology, Supervision, Writing - review & editing.

Data availability

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is/are the supplementary data to this article: