Single-Cell Transcriptome Reveals the Cellular Response to PEG-Induced Stress in Wheat Leaves

Abstract

Drought is a major factor limiting the production and yield of wheat bread (Triticum aestivum). Therefore, investigating the wheat drought-related response mechanism is an urgent priority. Here, the single-cell transcriptome of drought-nonsusceptible and susceptible wheat seedlings subjected to PEG-induced stress was systematically analyzed to study the drought-related response at the cellular level. We identified five major cell types using known marker genes and constructed a wheat leaf cell atlas. On this foundation, several potential specific marker genes for each cell type were identified, which provide a reference for further cell type annotation. Moreover, we identified cellular heterogeneity in the drought-related response mechanisms and regulatory networks among cell types. Specifically, the drought response of mesophyll cells was correlated with the photosynthetic pathway. Pseudotime trajectory analysis revealed the transition of epidermal cells from their normal function to a defense response under stress. Moreover, we also characterized the genes associated with the drought response. Notably, we identified two transcription factors (TraesCS1D02G223600 and TraesCS1D02G072900) as master regulators in most cell types. Our study provides detailed insights into the response heterogeneity of cells under PEG-induced stress. The gene resources obtained in our study could be applied to breed better crop plants with improved drought tolerance.

1. Introduction

Wheat (Triticum aestivum) is one of the three major cereal crops used for human food and livestock feed.¹ However, wheat plants are often subjected to various biotic (fungal, bacterial, viral, and insect pests) and abiotic (drought, salt, and cold) stresses throughout their life cycle. Among these stresses, drought has become a major yield-limiting factor of wheat and threatens international food security.²⁻⁵ Drought stress triggers various morphological, physiological, and biochemical changes, such as root depth and extension, stomata opening and closing, cuticle thickness, osmotic adjustment, hormone fluctuations, photosynthesis inhibition, protein changes, and growth inhibition, ultimately leading to incalculable yield losses and quality reduction.^6,7 Therefore, revealing the underlying regulatory mechanisms of wheat responses to drought stress and breeding drought-resistant cultivars will not only help to avoid yield reduction but also ensure food security and sustainable development of modern agriculture.

Leaves, as the primary place of photosynthesis, serve as the interface between plants and the environment and instinctively respond to environmental stimuli, especially drought stress. Bulk transcriptome sequencing (bulk RNA-seq) provides a routine method for investigating the mechanism by which plants respond to stress.⁸ In recent years, bulk transcriptome profiling of leaves has often been used to research the response mechanisms of plants to drought stress. For instance, transcriptome and anatomical studies of tobacco provided detailed information about the leaf anatomical features and transcriptional changes associated with leaf structure under long-term drought stress.⁹ Zhang et al. conducted the first transcriptome analysis of the Iris germanica cultivar ‘Little Dream’ leaves under drought stress and identified a large number of drought-responsive genes.¹⁰ Leaf transcriptome profiles of Phoebe bournei indicate that drought stress drives large-scale transcriptomic changes in signaling, metabolism, secondary metabolite biosynthesis, and photosynthesis-related pathways.¹¹ A metabolomic and transcriptomic analysis of the leaves of wheat cultivars with different drought resistance revealed different adaptive patterns under drought stress.¹² Notably, plants are composed of various cell types, which function as a whole to secure successful metabolism and growth.¹³ Different cell types play biologically distinct roles in plant development and environmental adaptation. Therefore, studies on the plant transcriptome response to extrinsic stimuli need to be sufficiently heterogeneous.¹⁴ However, such heterogeneity is often masked in traditional bulk transcriptome analyses, since the bulk transcriptome is based on the whole organ and reflects the average response of the different cell types composing the organ.¹⁵ Compared to the limitations of bulk transcriptome sequencing, single-cell transcriptome sequencing (scRNA-seq) facilitates the understanding of gene expression heterogeneity among different cell types and promotes the identification of new cell types.^14,16,17

scRNA-seq could increase the spatiotemporal resolution of transcriptome analyses to the single-cell level.¹⁸ This high resolution provides a previously unattainable view of gene expression dynamics in individual cells and the alteration of these dynamics under various extrinsic conditions; thus, it has been applied to diverse aspects of plant biology in recent years.¹⁹⁻²¹ Since the first report of scRNA-seq in Arabidopsis thaliana,²² many single-cell transcriptomic atlases have been gradually established for various plant organs or tissues, including roots, root calluses, stems, leaves, inflorescences, corollas, ears, shoot apexes, and cotton fibers and glands, and in various species, including Arabidopsis,²³⁻³¹ cotton,³²⁻³⁵ tobacco,³⁶ poplar,³⁷ peanut,³⁸⁻⁴⁰ madagascar periwinkle,⁴¹ tea plant,⁴² rice,^13,43 maize,⁴⁴ alfalfa,⁴⁵ herba schizonepetae,⁴⁶ and wheat.⁴⁷ In addition to revealing the gene expression atlas and developmental and differentiation trajectories, scRNA-seq has also been used to research plant biotic and abiotic stress responses, such as the responses of maize roots to Fusarium verticillioides infection,⁴⁸ rubber tree leaves response to powdery mildew infection,⁴⁹ tomato leaves to tomato chlorosis virus infection,⁵⁰ strawberry (Fragaria vesca) leaves to Botrytis cinerea infection,⁵¹ pea shoots to boron deficiency,⁵² Chinese cabbage leaves to heat stress,⁵³Arabidopsis root tips to osmotic stress,²¹ and cotton roots to salt stress.⁵⁴ Despite dozens of published articles relevant to single-cell sequencing in plants, scRNA-seq research on wheat leaves, especially the drought response of wheat leaves, is unprecedented.

Here, we used scRNA-seq to isolate protoplasts from the leaves of drought-nonsusceptible and susceptible wheat seedlings subjected to PEG-induced stress. We generated a single-cell transcriptome atlas of wheat leaves, annotated five major cell types, and identified several potential cell-type-specific marker genes. Based on the expression patterns for cell lineage, we characterized the cellular heterogeneity in the response mechanism and regulatory network under stress. Our study confirmed the potential of scRNA-seq technology for research on nonmodel plant–abiotic stress interactions and provides new insights into the wheat drought-related response mechanism at the cellular level.

2. Materials and Methods

2.1. Plant Material and Experimental Set Up

The common wheat cultivar Jimai 325 bred by the Institute of Grain and Oil Crops, Hebei Academy of Agriculture and Forestry Sciences was treated by 1.2% ethylmethanesulfonate (EMS) for 10 h, then these seeds were washed with running water for 4 h, dried, sown, and harvested to generate EMS-induced mutants.

The Jimai 325 and its EMS-induced mutants were grown in a greenhouse under 23 °C/14 h light and 15 °C/10 h dark conditions. Wheat seedlings with full grain, consistent size, and no pests or diseases were selected for germination on culture dishes containing distilled-water-soaked filter paper. When the third true leaves fully unfolded, wheat seedlings (Jimai 325 and its EMS-induced mutants) with similar growth statuses were selected and subjected to PEG-induced stress. For stress treatment, the wheat seedlings were transferred to a Hoagland nutrient solution containing 20% PEG-6000 solution. The PEG solution was refreshed every 2 days to maintain the desired concentration.

The Jimai 325 and its mutants were subjected to PEG-induced stress for 12 days to screen the drought-susceptible mutants. Then, Jimai 325 (designated as CK) and drought-susceptible mutant (named T6) were subjected to PEG-induced stress for 7 days to harvest fresh wheat leaves for further protoplast isolation and scRNA-seq.

2.2. Protoplast Isolation from Wheat Leaves

The wheat leaves from CK and T6 were harvested and used for protoplast isolation. Briefly, the leaves were cut into 1–2 mm strips and digested for 3 h at 26 °C in RNase-free enzyme solution (2% cellulase R10, 1% macerozyme R10, 1% hemicellulase; 20 mM MES, 0.8 M mannitol, 20 mM KCl, 10 mM CaCl2, and 0.1% BSA) to release the protoplasts. The protoplasts were filtered with a 40-μm cell strainer and centrifuged at 300 × g for 5 min at 4 °C. After the supernatant was carefully removed, the pellet was washed and resuspended with PBS solution. A small amount of single-cell suspension was mixed with an equal volume of 0.4% trypan blue solution to detect cell viability. The cells were counted using a Countess II Automated Cell Counter. For samples to qualify for further processing, the ratio of viable cells to total cells of each sample had to be higher than 90%, and the live cell concentration had to be 1000–2000 cells/μL.

2.3. scRNA-seq Library Construction and Sequencing

The scRNA-seq library construction and sequencing were completed by Gene Denovo Biotechnology (Guangzhou, China). Libraries were generated and sequenced from the cDNAs with the Chromium Next GEM Single Cell 3′ Reagent Kit v3.1. In brief, cellular suspensions were loaded on a 10X Genomics GemCode single-cell instrument to generate single-cell gel beads in emulsions (GEMs). Upon dissolution of the gel bead, capture sequences containing barcode sequences were released, cDNAs were reverse-transcribed from mRNA, and the samples were labeled. After that, PCR amplification was performed by using cDNA as a template to generate sufficient mass for library construction. The cDNA was cut into 200–300-bp fragments, sequencing connectors (P5 and P7) and sequencing primers (R1 and R2) were added, and PCR amplification was performed to obtain the sequencing library. Finally, high-throughput sequencing of the constructed libraries was performed using the PE150 sequencing mode on the Illumina HiSeq 4000 sequencing platform.

2.4. scRNA-seq Data Preprocessing

The scRNA-seq raw data were deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1191129. The 10X Genomics Cell Ranger software (version 3.1.0, http://support.10xgenomics.com/single-cell/software/overview/welcome) was used to convert raw BCL files to FASTQ files and conduct preliminary statistics on the number of measured reads and the sequencing quality. After the initial quality control, reads with low-quality barcodes and unique molecular identifiers (UMIs) were filtered out. Reads were then mapped to the wheat reference genome (Ensembl_release51) and annotated as specific genes. Reads uniquely mapped to the transcriptome and intersecting an exon by at least 50% were considered for UMI counting. For gene expression quantification, UMI correction and effective cell identification were performed. Then, the UMI number was used to quantify gene expression and obtain cell-by-gene matrices for the downstream analysis.

2.5. Cell Clustering, Visualization, and Identification

The cell-by-gene matrices for each sample were individually imported to Seurat version 3.1.1 for subsequent analysis.⁵⁵ Before cell clustering, low-quality cells were filtered out if they met one of the following criteria: (1) less than 425 or more than 5000 genes detected, which could be that a GEM is packed with multiple cell types, (2) unusually high number of UMIs (≥8000), which might be that two or more cells enter the same GEM, and (3) high chloroplast gene percent (≥10%), which meant the cell is not in good condition. For cell clustering, the gene expression for each cell was normalized using the global-scaling normalization method “LogNormalize”. Harmony was used to aggregate all sample data and minimize the batch effect and the effect of behavioral conditions on clustering.⁵⁶ Subsequently, a shared-nearest neighbor (SNN) graph was constructed, and cells were clustered at a resolution of 0.5 for further analysis. Then, t-distributed stochastic neighbor embedding (t-SNE) and uniform manifold approximation and projection (UMAP) were performed to visualize the cell clustering. UMAP depicted the similarities and differences among cell clusters more clearly than t-SNE. Therefore, UMAP was used to explore the cell types.

Since the functions of most genes in wheat are largely unknown, the cell types of each cell cluster were annotated using two methods: (1) Cell types were identified using marker genes previously reported in several plants, such as Arabidopsis thaliana and Oryza sativa. Briefly, the marker genes were obtained from previous literature and plant single-cell marker gene databases, including the Plant Cell Marker Database (PCMDB, http://www.tobaccodb.org/pcmdb/)⁵⁷ and plant single-cell transcriptome database (PlantscRNAdb, http://ibi.zju.edu.cn/plantscrnadb)/.⁵⁸ Then, the sequences of marker genes were used to search for homologous genes of wheat, and the corresponding cell type was annotated. A bubble diagram was utilized for the visualization of marker gene expression. (2) Cell types were identified based on correlations between cell clusters. In short, the Pearson’s correlation coefficients between cell clusters were determined and used to construct a heatmap. If two highly correlated cell clusters (P > 0.95) had relatively similar gene expression patterns, this indicated that they might be of the same cell type.

After cell type identification, the up-regulated differentially expressed genes (DEGs) for each cell type were identified according to three criteria: overexpressed by at least 1.28-fold in the target cluster compared with in all other clusters (|log₂fold-change| > 0.36), expressed in more than 25% of the cells belonging to the target cluster, and a P-value less than 0.01. To explore and analyze the marker genes, the top 10 DEGs with the highest log₂fold-change value in target clusters were selected, and their expression distribution was visualized using a heatmap. The specific expression of representative marker genes in each cell type was visualized using UMAP and a violin diagram.

2.6. DEG Analysis between Groups and Regulatory Network Construction

To find the key genes responding to stress, the DEGs of CK versus T6 in different cell types were analyzed. Genes with an expression greater than 1.28-fold (|log₂fold-change| > 0.36) and P-value less than 0.05 were considered DEGs in CK vs T6 comparison. Additionally, differentially expressed transcription factors (TFs) were screened and analyzed. The coexpression relationship tables of these TFs were obtained from WheatOmics 1.0 (http://202.194.139.32/). According to the plant TF database PlantTFDB (https://planttfdb.gao-lab.org/index.php), differentially expressed target genes of these TFs were obtained. The TF coexpression network and TF-target gene regulatory network were constructed using Cytoscape_v3.8.2.

2.7. DEG Functional Enrichment Analysis

To understand the biological function of DEGs identified in this study, a functional enrichment analysis was performed using the Gene Ontology (GO, http://geneontology.org/) and Kyoto Encyclopedia of Genes and Genomes (KEGG, https://www.kegg.jp/) databases in corresponding R packages.^59,60 The DEGs were annotated, classified, and calculated by using the GO database to find significantly enriched GO terms. DEGs were analyzed by using the KEGG database to identify enriched metabolic or signal transduction pathways. P-values for the GO and KEGG analyses were calculated and corrected using the FDR correction. GO terms and KEGG pathways meeting the condition FDR ≤ 0.05 were defined as significantly enriched in DEGs.

2.8. Pseudotime Trajectory Analysis

The gene expression matrix of CK and T6 epidermal cells was imported into Monocle version 2.0. Based on key gene expression patterns of epidermal cells generated from CK and T6, Monocle sequenced individual cells at a pseudotime to simulate the dynamic changes of the temporal development process and visualized the developmental trajectory. Based on the pseudotime values of each cell, we screened DEGs with FDR < 0.05 along the timeline to identify key genes related to developmental differentiation processes. Subsequently, the DEGs with similar expression trends were clustered, and KEGG/GO enrichment analysis was performed to uncover the potential biological functions.

2.9. RT-qPCR Analysis

For the validation of scRNA-seq data, easily isolated tissues of wheat leaves were collected. The epidermal layers were isolated using tweezers. The leaves without epidermis were dissected to collect the vasculature (mainly big leaf vein), and the remaining leaves were collected as mesophyll tissue. The isolated epidermis, mesophyll, and vascular regions were further used for RT-qPCR analysis.

For the functional verification of candidate transcription factors, Jimai325 was cultivated and subjected to PEG-induced stress, as Section 2.1 describes. The leaves were harvested 0, 3, and 7 days after stress and used for further RT-qPCR analysis.

For RT-qPCR analysis, Total RNA was isolated from each sample using the plant total RNA extraction kit (TIANGEN, Beijing, China); there were three replicates for each sample. Subsequently, cDNA synthesis was performed using a Fastking RT Kit (TIANGEN, Beijing, China). The RT-qPCR analysis was performed using SYBR Green qPCR Mix (TIANGEN, Beijing, China) on the iQ5Multicolor Real-Time PCR detection system (Bio-Rad Laboratories, Hercules, CA, USA). The specific primers used for RT-qPCR are listed in Table S9. Gene expression levels were normalized using the wheat Actin gene. The relative gene expression levels were calculated by using the 2^–ΔΔCt method. Data are represented by the mean values (±SD) of three replicates.

3. Results

3.1. Generation of a Wheat Leaf Single-Cell Transcriptome Atlas

According to stress screening for 12 days, we identified T6 as the most drought-susceptible EMS mutant derived from Jimai 325 (Figure S1A). To investigate the response mechanism and response differences of Jimai 325(CK) and its drought-susceptible mutant(T6) at the cellular level, we first performed high-throughput single-cell transcriptome sequencing (scRNA-seq) to generate single-cell profiles of wheat leaves. However, T6 failed to be used for protoplast isolation and scRNA-seq due to the lack of fresh leaves under stress treatment for 12 days. Thus, the wheat seedlings of CK and T6 were subjected to PEG-induced stress for 7 days and then harvested and cut into strips for protoplast isolation. The protoplasts were subjected to GEM generation, library construction, and sequencing to generate wheat leaf scRNA-seq data (Figure S1B). After removing the low-quality reads, more than eight hundred million reads were generated from two samples (395,166,461 reads in CK and 498,682,911 reads in T6). The reads were mapped to the wheat reference genome (Ensembl_release51) at a rate of more than 77% for each sample. In total, we detected 52,971 genes from 3954 cells in CK samples, with a median number of 850 genes and 1079 UMIs per cell. We captured 4258 cells in T6 samples containing 47,299 detected genes, with a median of 424 genes per cell and 520 UMIs per cell (Figure S1C). A summary of the quality control for the single-cell transcriptome is given in Table S1. The results indicated that high-quality wheat leaf scRNA-seq data were obtained for further analysis.

To generate the cell atlas of wheat leaves, we merged the CK and T6 samples for cell clustering and identification. After filtering and removal of low-quality cells, 5451 cells (3681 from CK and 1770 from T6) were grouped into eight clusters (named cluster 0–7). The number of cells distributed in each cluster ranged from 32 to 2364 (Table S2). Uniform manifold approximation and projection (UMAP) and t-distributed stochastic neighbor embedding (t-SNE) were performed to visualize the results of cell clustering (Figures 1A and S2A). Furthermore, the known cell-type-specific marker genes were collected and used to annotate corresponding cell clusters (Table S3). As shown in Figure 1B, the genes related to photosynthesis, such as the genes encoding chlorophyll a-b binding protein (CAB), light-harvesting chlorophyll b binding protein (LHCB), and rubisco bisphosphate carboxylase small subunit (RBCS), were used as marker genes to identify mesophyll cells (MC) (cluster 3). Vascular tissue is composed of highly heterogeneous cells; therefore, multiple reported markers were selected, including the phloem-related genes PP2A1 and SWEET4, the xylem-related gene CCOAOMT1, the procambium-related gene HB1, and the vascular system-expressed genes XTH8 and EXPB6. These genes were predominantly expressed in cluster 2 and cluster 7. Hydathode cells (HC) were assigned to cluster 6 because the hydathode-specific gene encoding thaumatin-like protein (TLP) was expressed in this cluster. Furthermore, the genes encoding early nodulin-like protein (ENODL14) and cell division control protein 48 homologue B (CDC48), which are associated with cell division, were notably expressed in cluster 1, suggesting this cluster contained proliferating cells (PC). The known epidermal cell (EC) marker genes DUF538, glutathione s-transferase F subunit (GSTF1), lipid transfer protein 3 (LTP3), and GPAT5 were observed in clusters 4 and cluster 5. Therefore, cluster 4 and cluster 5 were considered epidermal cells. To verify the reliability of scRNA-seq data, some representative marker genes were analyzed for RT-qPCR analysis. We found that GPAT5, CAB7, and SWEET4 had high expression levels in epidermis, mesophyll, and vascular tissue, respectively (Figure 1D). The expression patterns of marker genes were consistent with UMAP and bubble plot, which indicates the accuracy of scRNA-seq analysis.

Figure 1.

Single-cell transcriptome atlas of wheat leaves. (A) Visualization of the eight cell clusters by UMAP. Each dot represents a cell and is colored according to the cell cluster. Eight colors represent the eight distinct cell clusters. (B) Bubble plot showing the expression pattern of known cell type marker genes. The dot diameter represents the proportion of cells expressing the gene. The color indicates the average expression level. (C) UMAP illustrated the final cell type annotation of wheat leaves. (D) Expression of epidermal, mesophyll, and vascular cell-specific marker genes.

Based on the expression analysis of known marker genes, cluster 0 could not be identified. To further annotate cluster 0, correlation analysis between clusters was performed, which showed a high correlation (P > 0.95) between clusters 0 and 2 (Figure S2B). In addition, we also found that cluster 0 was adjacent to cluster 2 in UMAP (Figure 2A). This evidence suggested that cluster 0 contained the same cell types as cluster 2 (vascular cells, VC). Taken together, eight cell clusters from wheat leaves were identified into the following five populations: mesophyll cells (cluster 3), epidermal cells (cluster 4 and 5), vascular cells (cluster 0, 2 and 7), proliferating cells (cluster 1), and hydathode cells (cluster 6) (Figure 1C). These results indicate that the wheat leaf is a highly heterogeneous tissue. Additionally, the three major leaf tissues (mesophyll cells, epidermal cells, and vascular cells) were annotated, and their transcripts were captured, indicating that our approach was reliable for single-cell transcriptome analyses.

Figure 2.

GO and KEGG enrichment analysis of up-regulated DEGs for each cell type. Pie graph showing the number of DEGs in each cell type. GO and KEGG enrichment analysis of DEGs for each cell type (red boxes represent GO analysis, black boxes represent KEGG analysis).

3.2. Identification of Potential Marker Genes

Based on single-cell studies in Arabidopsis, rice, and other species, several marker genes were reported for cell type identification in different organs. However, the homologous genes of some known marker genes in wheat were not expressed, especially in single cell clusters in this study, due to the hexaploid property of wheat. Therefore, it was necessary to identify potential marker genes with cell-type-specific expression in wheat. Before the characterization of potential marker genes, we first validated the reliability of the cell type identification. To do this, we selected the DEGs (up-regulated expression in the target cell type compared with that in other cell types) and performed GO and KEGG enrichment analysis. In total, 328 genes in EC, 255 genes in HC, 249 genes in VC, 168 genes in MC, and 212 genes in PC were identified (Figure 2). GO and KEGG analyses revealed that transmembrane transporter activity, ATPase activity, and metabolite-related pathways were enriched in VC, consistent with their core role in the conduction of substances. Mesophyll cells, which are involved in light capture, were specifically enriched with genes that participate in the photosynthesis pathway. In PC, the up-regulated genes were mainly associated with the perinuclear region of the cytoplasm pathway, as well as protein folding and processing, revealing the primary action of cell division. As expected, the molecular processes matched the specialized biological functions of the corresponding cell types, demonstrating the accuracy of the cell type annotation in this study.

After verifying the accuracy of cell type identification, the top 10 genes with the highest fold-change in the five known cell types were collected for further potential marker gene characterization, and their expression profiles were exhibited in a heatmap (Figure 3A; Table S4). Among these DEGs, several known marker genes had expression profiles consistent with previous results, such as marker genes LTP3 (TraesCS5B02G145900) in EC, CAB1B (TraesCS7D02G276300) in MC, HB1 (TraesCS4B02G237300) in VC, and Thaumatin-like protein genes (TraesCS7A02G558500 and TraesCS4A02G296300) in HC (Figures 3A and S3). In addition to these characterized marker genes, we also found that BCA2 (TraesCS7B02G354800), a gene encoding beta-carbonic anhydrase, which mediates carbon dioxide control of stomatal movements, was highly expressed in EC (Figure 3A, B).⁶¹ The gene encoding PETE (TraesCS4A02G499500), which is an electron carrier in photosynthesis, and other photosynthesis-related genes, namely PSAK (TraesCS2B02G272300), PSBW (TraesCS3B02G344200), and CAB8 (TraesCS6A02G159800), were highly abundant in MC (Figures 3A, B and S3).⁶² In addition, RSCA (TraesCS3B02G293200), HSP18 (TraesCS3B02G131300), and CYP714C2 (TraesCS6A02G404500) had the highest up-regulated and specific expression in HC, PC, and VC, respectively (Figure 3B; Table S4). The identification of cell-type-specific potential marker genes provided an essential reference for cell cluster annotation of wheat and other plants in scRNA-seq studies.

Figure 3.

Identification of potential marker genes in each cell type. (A) Heatmap of the top 10 marker genes with the highest expression levels in each cell type. The color bar indicates the gene expression level in each cell type. The known marker genes were marked with black asterisks and potential marker genes were marked with red asterisks. (B) Specific expression patterns of five potential marker genes distributed in UMAP and violin plots. Each dot represents a cell and is colored according to the gene expression level.

3.3. Single-Cell Heterogeneity Analysis of Wheat Leaves under PEG-Induced Stress

To explore the differential stress response, we first analyzed the cellular heterogeneity of drought-nonsusceptible (CK) and susceptible (T6) wheat under PEG-induced stress. By comparing the UMAP of CK and T6, we found that both samples contained all five cell types, suggesting that the cell types were not affected by the stress (Figure 4A). However, the total number of captured cells in T6 was notably lower than that in CK. The proportion of each cell type was also different (Figure 4B). For example, the proportion of mesophyll cells in CK (12.63%) was significantly higher than that in T6 (7.46%). The proportions of proliferating cells and epidermal cells were lower in CK than in T6 (21.24 and 15.42% in CK; 14.94 and 14.42% in T6, respectively). The cell proportions of vascular and hydathode cells also varied between CK and T6 (Figure 4B, Table S2).

Figure 4.

Single cell profiles in CK vs T6 comparison under PEG-induced stress. (A) UMAP plot of CK+T6 and separated single cells of CK and T6. (B) Cell number and relative proportion of each cell type in CK and T6. (C) Number of DEGs in CK vs T6 comparison under PEG-induced stress. (D) GO enrichment analysis of DEGs in CK vs T6 comparison under PEG-induced stress. The red represent the top five biological process GO terms enriched in up-regulated DEGs for each cell type. (B) Green represent the top five biological process GO terms enriched in down-regulated DEGs for each cell type. The color bars represent the −log₂(P-value).

The UMI reflects the number of transcripts captured by scRNA-seq. The total UMI of mesophyll cells, epidermal cells, proliferating cells, and hydathode cells was significantly lower in T6 than in CK, but there were no differences in the vascular cell types (Figure S4A). These results indicate distinct transcriptional characteristics between CK and T6 in different cell types. To further analyze the transcriptional heterogeneity under PEG-induced stress, the DEGs for each cell type in the CK vs T6 comparison were detected. We identified 1811 DEGs in EC (655 up, 1156 down), 1798 DEGs in PC (618 up, 1180 down), 1455 DEGs in VC (637 up, 818 down), 969 DEGs in MC (231 up, 738 down), and 202 DEGs in HC (39 up, 163 down) (Figure 4C). The DEGs with annotations identified from each cell type are listed in Table S5. Among them, only 21 DEGs overlapped across all five cell types, and a majority of DEGs were specific to a single cell type (Figure S4B). These results suggested cellular and transcriptional differences between the CK and T6 samples under PEG-induced stress.

3.4. GO and KEGG Enrichment Analysis of DEGs

To gain insights into the response pattern of wheat leaves to stress, we performed GO and KEGG pathway analyses on the DEGs in each cell type. The enrichment analysis revealed that drought-response-related GO terms were shared in all cell types, such as the responses to water deprivation, osmotic stress, reactive oxygen species, and salt stress (Figure S4C). The stress-related KEGG pathways, such as the MAPK signaling pathway, were also enriched in DEGs of all cell types (Table S6). GO enrichment analysis of the up- and down-regulated DEGs in each cell type was further conducted, and the top five biological process GO terms were displayed in a heatmap (Figure 4D). Some GO terms were enriched in DEGs from specific cell types; for example, up-regulated DEGs from HC were enriched in MAPK cascade and signal transduction by protein phosphorylation. Although PC and VC DEGs had many GO terms in common, the preferences of enriched GO terms were different. Unlike VC, the up-regulated DEGs in PC were more enriched in protein folding and response to heat, and the down-regulated DEGs in PC were more enriched in response to inorganic substances and photosynthesis. Interestingly, the up-regulated DEGs in EC and MC showed similar enrichment patterns, except for the specific enrichment in MC of photosynthesis and light harvesting in photosystem I. However, the GO terms enriched in down-regulated DEGs in these cell types were distinct, showing significant enrichment of the cell wall macromolecule catabolic process, aminoglycan metabolic process, and chitin metabolic process in EC, as well as translation, peptide biosynthetic process, cellular amide metabolic process, and organonitrogen compound metabolic process in MC. These results suggested that there were distinct response mechanisms to stress in each cell type, although with certain common ground among cell types.

3.5. Transcriptional Regulation of Wheat Leaves under Stress

To adapt to stress, plants have developed complex transcriptional regulatory networks. TFs function as key switches by recognizing specific DNA sequences to regulate their gene expression in these networks. To understand the transcriptional regulation and to search for core TFs that might be responsible for the drought-related response, the differentially expressed TFs in each cell type were identified. There were 129 TFs in EC (44 up, 85 down), 83 TFs in PC (24 up, 59 down), 35 TFs in MC (10 up, 25 down), 57 TFs in VC (16 up, 41 down), and 6 TFs in HC (6 down) were identified (Figure 5A). We mainly focused on the TFs that were annotated as associated with the drought-related response. By constructing a coexpression network of these potential TFs, we found that several TFs were related to each other, indicating that these TFs not only act alone but may work together in response to stress (Figure 5B). In addition, we constructed a heatmap to display the fold change of these TFs in CK versus T6 comparison among each cell type (Figure 5C). Two homeologous genes of WRKY24 (TraesCS1D02G072900 and TraesCS1B02G088900), a bHLH128 gene (TraesCS3A02G395000), and two homeologous genes of dehydration-responsive element binding (DREB) TFs genes DREB2B (TraesCS1D02G223600 and TraesCS1B02G235100) showed significant fold change in most cell types, suggesting these TFs were core transcriptional regulators in the stress response. We further validated this result using RT-qPCR. Although the fold changes of TF expression levels observed in RT-qPCR differed from those in scRNA-seq, the expression trends of the TFs were consistent with those in scRNA-seq data (Figure 5C). In addition, we found some TFs showed fold change in specific cell types, such as ERF4 (TraesCS3A02G328000) in EC, WRKY71-like (TraesCS6D02G136200) in MC, NAC048 (TraesCS3B02G439600) in PC, NAC074 (TraesCS3D02G183900) in VC, and NAC048 (TraesCS3D02G401200) in HC indicating that these TFs might participate in cell type-specific regulatory networks.

Figure 5.

Differentially expressed TF analysis between CK and T6 under PEG-induced stress. (A) Number of differentially expressed TF genes in each cell type. (B) Putative coexpression network of TF genes. Different colors represent different transcription factor families. (C) Heatmap and column chart showing the fold change of TFs in CK vs T6 comparison in each cell type. The color bar represents the absolute value of log₂(fold change). The column chart were drawn according to the results of RT-qPCR. (D) TF-target genes regulatory network in epidermal cells and mesophyll cells under PEG-induced stress. The red circles represent TFs.

The regulatory networks of these TFs were further generated (Figures 5D and S5). From the regulatory network analysis, we could clearly see that WRKY24 participated in the regulatory networks of multiple cell types, as expected. ERF4 in EC and WRKY71 in MC were connected to genes involved in multiple downstream processes and to other transcription factors, suggesting a potential regulatory role in specific cell types for these two TFs. Notably, ERF4 and WRKY71 could be regulated by the same transcription factor DREB2B, indicating that DREB2B might work as a core regulator of stress responses. Although the network of EC and MC was mastered by the same TFs, the downstream regulatory networks were specific in different cell types. Specifically, the downstream genes in MC were clearly related to chloroplastic organization and function, such as TraesCS5B02G111800 (TRXM) and TraesCS2A02G206100 (THF1). However, the downstream genes in EC were involved in phospholipid transport(ALA10, TraesCS1A02G002500), protein ubiquitination (PUB23, TraesCS6D02G367600), the oxidation–reduction process (SPT154, TraesCS2D02G000200 and ME6, TraesCS1B02G141700), and the chitin metabolic process (Cht8, TraesCS5D02G408200) (Figure 5D, Table S7). Taken together, the TFs' regulatory networks indicated a certain cell type specificity, and these networks were also associated with each other to form a sophisticated regulatory network to respond to stress.

To verify the changes in expression of these key TFs in stress response, wheat seedlings were subjected to PEG-induced stress for 7 days. The gene expression patterns of some genes and their homologous genes at different treatment times (0, 3, and 7 days) were detected by RT-qPCR (Figure S6). In general, the expression levels of these genes were up-regulated under PEG-induced stress, indicating they were involved in the stress response. However, we found different expression patterns among the three homologous genes. For example, the expression of WRKY24-A and WRKY24-D was activated at 3d after stress treatment, while at this time, the expression level of WRKY24-B was not significantly different compared to the control. These results indicated that these homologous genes had the same function in response to stress, but may respond to stress in various regulatory ways.

3.6. Pseudotime Trajectory Analysis of Epidermal Cells under PEG-Induced Stress

To analyze the impact of stress on epidermal cell development, pseudotime analysis was performed on epidermal cells generated from CK and T6 using Monocle2. Based on key gene expression patterns, Monocle sequenced individual cells at a pseudotime to simulate the dynamic changes of the temporal development process and visualized the developmental trajectory. The pseudotime analysis showed that each cell was distributed in different positions along the path. There were two branch points in the pseudotime trajectory, which marked the developmental state’s transition, that divided all cells into four states marked as 1, 2, 3, and 4 (Figure 6A–C). The cells shifted gradually from state 1 to other states as the pseudotime value increased (Figure 6A, B). According to the pseudotime order, DEGs were screened, and cluster analysis was performed according to the similarity in expression trends. The DEGs were grouped into four clusters with distinct gene expression patterns, reflecting the changes of gene expression patterns and the developmental process of epidermal cells under drought stress (Figure 6D and Table S8). The genes in clusters 1 and 2 were significantly enriched in the GO terms fatty acid biosynthetic process (GO:0006633), lipid biosynthetic process (GO:0008610), chitin catabolic process (GO:0006032), carbohydrate metabolic process (GO:0005975), glutathione metabolic process (GO:0006749), and protein phosphorylation (GO:0006468), which was consistent with the biological function of epidermal cells (Figure 6D). GO terms related to the abiotic stress response, including the response to water deprivation (GO:0009414), response to osmotic stress (GO:0006970), and response to hydrogen peroxide (GO:0042542) were enriched in cluster 3 and cluster 4, indicating the transition of epidermal cells to a stress response under drought (Figure 6D). Additionally, the expression levels of corresponding genes such as TraesCS6D02G234700 (COR410), TraesCS6B02G273400 (COR410), TraesCS1B02G088900 (WRKY24), and TraesCS3D02G046300 (HSP16.9) showed increasing trends at the middle and end of the pseudotime axis (Figure 6E), implying that these genes play important roles in the stress response.

Figure 6.

Pseudotime analysis of epidermal cells of CK and T6. (A) Pseudotemporal distribution of cell trajectories. Each dot represents a single cell. Color bar represent the pseudotime score. (B) Distribution of cells with different differentiation states in the trajectory. Colors represent different states. (C) Cell distribution of different sample in the trajectory. Colors represent different samples. (D) Pseudotime heatmap and GO analysis of differentially expressed genes. The color bar indicates the relative gene expression level. (E) Gene expression kinetics of representative genes along with the pseudotime order. Colors represent different samples.

4. Discussion

4.1. A Single-Cell Transcriptome Atlas of Wheat Leaves Was Constructed

Drought severely impacts crop growth, yield, and quality, especially for vital food crops like wheat, posing a threat to global food security.^3,6 Additionally, the frequency and severity of droughts are expected to intensify at a global scale due to global warming.⁶³ Therefore, a better understanding of the genes controlling drought resistance and the underlying regulatory mechanisms in wheat is indispensable to avoid yield reduction and help develop drought-tolerant cultivars. Recently, scRNA-seq technology, which has great potential to address complex biological questions with high precision, has been successfully applied in plant biology. High-throughput scRNA-seq improves our ability to identify cell types and essential cellular activities in plant development and stress adaptation at the single-cell level.^18,19,64 Although scRNA-seq studies on wheat roots have been published previously, no scRNA-seq data have been reported for wheat leaves.⁴⁷ In this study, we constructed a single-cell atlas for wheat leaves using scRNA-seq technology. This allowed us to further explore cell-type-specific drought-related responses in wheat leaves.

We successfully isolated high-quality protoplasts from the leaves of wheat seedlings by enzyme-induced protoplast generation and identified eight cellular clusters and five cell types using several known homologous markers and GO analysis of the highly expressed genes in each cell cluster. Mesophyll, epidermal, vascular, hydathode, and proliferating cells were clearly identified according to the marker genes of different cell clusters (Figure 1D). Although the cell numbers and cell types in our study were lower than those in previous leaf scRNA-seq atlases of model plants,^13,29,30 the main leaf tissues (mesophyll, vascular, and epidermal) were identified, revealing the applicability of scRNA-seq to nonmodel plants. It is worth noting that some cell types, such as phloem and xylem cells, were not assigned to any cell clusters in our scRNA-seq data. However, the expression levels of phloem-related marker PP2A1 and xylem-related marker CCOAOMT1 were high in cluster 2, which was assigned to vascular cells, consistent with the notion that the phloem and xylem are components of vasculature.⁶⁵ One possible explanation is that these cells failed to be computationally clustered at the resolution value used (0.5). Generally speaking, the higher the resolution, the more cell clusters can be identified. Kim et al. identified at least 19 clusters and 10 cell types in Arabidopsis leaf using a resolution value of 0.8.²⁵ In addition, the resolution of 0.8 was used to cluster the cells of the wheat root, which identified 22 cell clusters.⁴⁷ As a very first attempt to apply scRNA-seq to the wheat leaves, we used a resolution of 0.5 for cell cluster analysis to avoid increasing the difficulty of cell type identification while also reducing the number of cell types. Another potential reason is that the vascular tissue is embedded by the bundle sheath (BS) with a rigid cell wall, and the vascular-enriched protoplast populations might not be captured fully by the protoplast isolation protocols used in this study. Thus, the phloem and xylem cell populations were relatively small, and their gene expression was not sufficiently specialized to a unique cell cluster. Similarly, in a previous study, guard cells were not assigned to any cell cluster even though their marker gene was expressed in the epidermal cell cluster.¹³ With advancements and continued application of scRNA-seq technology, the annotation of cell types or a subset of cell types will likely become more accurate.

Cell type identification is a core step in the application of single-cell sequencing in plants. Several databases and maker genes are available for annotating cell types in model plants, including Arabidopsis and rice. However, expression patterns of these marker genes may not be conserved between model and nonmodel plants.⁶⁶ For example, the phloem and procambia marker gene SWEET of Arabidopsis was highly expressed in the lower epidermal cells of tea leaves.⁴² Thus, for nonmodel plants, the lack of cell-type-specific marker genes makes cell annotation more challenging. To allow easier cluster annotation in future wheat leaf scRNA-seq studies, we characterized the highly expressed genes in each cell cluster and identified potential marker genes. A beta-carbonic anhydrase gene (BCA2, TraesCS7B02G354800) was identified as a new marker for EC. The carbonic anhydrases in Arabidopsis mediate carbon dioxide control of stomatal movements, implying their specific presence in the epidermis.⁶¹ We also identified four photosynthesis-related genes (TraesCS4A02G499500, TraesCS2B02G272300, TraesCS3B02G344200, and TraesCS6A02G159800) as the marker genes for MC, of which TraesCS4A02G499500 encodes a plastocyanin involved in linear as well as cyclic photosynthetic electron transport.⁶² Interestingly, we found that a large number of heat shock protein (HSPs) genes were significantly enriched in cluster 1, which was consistent with previous observations in strawberry leaves. However, there were not enough marker genes to determine its identity in the previous report.⁵¹ In our study, we identified this cell type as proliferating cells using marker genes ENODL14 and CDC48. We know that HSPs are highly conserved molecular chaperones that participate in protein folding and unfolding, protein assembly, cell-cycle control, and signaling.^67,68 This result suggests the important role of HSPs in the cell division of PC. The marker genes developed in this study could be used for identifying wheat leaf cells in the future and provide valuable knowledge for other single-cell studies of plants with fewer known marker genes.

4.2. Wheat Leaf Cells Exhibit Varying Responses to Stress

Based on a pairwise comparison of single-cell profiles from CK and T6 under PEG-induced stress, we found that wheat leaf cells were heterogeneous in their response to the stress. We can clearly notice that stress did not substantially alter cell type identity (all identified cell types appeared in both CK and T6), but led to changes in the relative proportions of cell types (Figure 4). This result is consistent with previous observations reported in many biotic and abiotic stress experiments, such as in tomato leaves infected with tomato chlorosis virus, pea shoots treated under boron deficiency, and Chinese cabbage leaves under heat stress.^50,52,53 Notably, the proportion of MC was significantly decreased under stress in this study. Perhaps directly related to this event was the change in the internal morphological structure of the leaves. Some studies have reported that plants experience an obvious reduction in leaf size, width, and length under abiotic stress.⁶⁹⁻⁷¹ These findings emphasize the plasticity and adaptability of leaves to balance natural growth with the stress response.

In addition to heterogeneity in cell proportions, heterogeneity in the gene expression response to stress was also observed among cell types. Although some genes had similar gene expression patterns, most genes responded to stress in a cell-type-specific manner (Figure S4B). The GO analysis of DEGs indicated that some GO terms were enriched in all cell types, including response to stimulus, response to abiotic stimulus, response to water deprivation, and response to osmotic stress, which have been reported to play significant roles in the drought stress response.^12,72−74 This result indicates that wheat leaf cells initiated drought-related genes in response to stress. Pseudotime trajectory analysis further revealed the transition of epidermal cells from their normal function to defense response, showing the upregulation of drought-related genes during the late stage of pseudotime (Figure 6). Additionally, the specifically enriched GO terms were separately identified in different cell types (Figure 4D). Plants have evolved a variety of adaptive mechanisms to cope with drought stress, such as the alteration of biological processes in photosynthesis, chitin metabolism, and cell wall macromolecule metabolism.^71,75−77 Consistent with this, wheat leaf cells might adapt to stress by regulating photosynthesis and light harvesting in photosystem I (GO:0009768) in MC and by adjusting the chitin metabolic process (GO:0006030) and the cell wall macromolecule catabolic process (GO:0016998) in EC. Thus, our results reveal the cell-type-specific drought-related response mechanism of wheat leaves at a more detailed level.

4.3. Regulatory Network Uncovered the Coordination among Cell Types in the Stress Response

Transcription factors are key controllers in the regulatory networks underlying plant responses to abiotic stresses, especially drought stress.⁷⁸⁻⁸⁰ The dehydration-responsive element binding (DREB) TFs belong to the AP2/ERF family, which specifically interacts with the DRE sequence and controls the expression of a large number of drought stress-responsive genes.^80,81 In addition to DREB, other TF families have been extensively reported to play important roles in the drought response, including NAC,⁸⁰ MYB,⁷⁹ WRKY,⁸² and bHLH.⁸³ This is consistent with our results, as we found that members of these TF families were differentially expressed in our study, implying the essential roles of these TFs in the stress response. We also found that some TFs and their homologous genes were differentially expressed in stress response, but their fold change and expression patterns in cell type and response time were different (Figures 5C and S6); thus, we suggested that they had the same function but different regulatory pathways. However, the biological function and regulatory mechanism of these key TFs require further validation by genetic transformation and genome-editing technology.

Moreover, regulatory network analysis suggested that each cell type actively participates in the stress response through a distinct transcriptional regulation process. As the main leaf photosynthetic tissue, the regulatory network in MC was clearly related to chloroplastic organization and function, indicating that mesophyll cells respond to drought stress by regulating the chloroplast content and photosynthesis.⁸⁴ Not only that, but the regulatory networks in different cell types communicate with each other to build a powerful and synergistic transcriptional regulatory network to resist drought in the whole leaves of wheat seedlings. These genes used to construct regulatory networks might be potential gene resources for drought-resistant breeding.

Data Availability Statement

The scRNA-seq raw data were deposited in the NCBI Sequence Read Archive (SRA, https://www.ncbi.nlm.nih.gov/sra/) with accession number PRJNA1191129.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jafc.4c12749.

• scRNA-seq process and reports (Figure S1), single-cell transcriptome atlas of the wheat leaves (Figure S2), UMAP and violin plots of specific expression of representative marker genes in each cell type (Figure S3), single-cell transcriptome analysis of CK and T6 under PEG-induced stress (Figure S4), regulatory network of TF-target genes identified in PC (Figure S5), and relative expression level of key TFs and their homeologous genes under PEG-induced stress (Figure S6) (PDF)

• Single-cell RNA sequencing quality of CK and T6 (Table S1), cell number and proportion of CK and T6 in each cluster (Table S2), known marker genes used for cell identification (Table S3), the top 10 DEGs and their annotation from each cell type (Table S4), the DEGs with annotation identified from different cell types (Table S5), top 10 KEGG enriched pathways for up and down regulated DEGs in each cell type (Table S6), differential transcription factors and the corresponding target genes (Table S7), DEGs with annotation based on the pseudotime line (Table S8), and primers used for RT-qPCR in this study (Table S9) (XLSX)

This research was supported by funding from the Earmarked Fund for Hebei Wheat InnovationTeam of Modern Agro-industry Technology Research System (21326318D), Youth Innovation Fund of Grain and Oil Crop Research Institute, Hebei Academy of Agriculture and Forestry Sciences (2023LYS01), Scientific and technological innovation personnel construction of Hebei Academy of Agriculture and Forestry Sciences (C22R0313), Research on Digital Breeding of Crops in Hebei Province (2022KJCXZX-NXS-5), National Natural Science Foundation of China (32160353), and the Natural Science Foundation of Gansu Province, China (22JR5RA263).

The authors declare no competing financial interest.