Establishment of single-cell transcriptional states during seed germination

Abstract

Germination involves highly dynamic transcriptional programs as the cells of seeds reactivate and express the functions necessary for establishment in the environment. Individual cell types have distinct roles within the embryo, so must therefore have cell type-specific gene expression and gene regulatory networks. We can better understand how the functions of different cell types are established and contribute to the embryo by determining how cell type-specific transcription begins and changes through germination. Here we describe a temporal analysis of the germinating Arabidopsis thaliana embryo at single-cell resolution. We define the highly dynamic cell type-specific patterns of gene expression and how these relate to changing cellular function as germination progresses. Underlying these are unique gene regulatory networks and transcription factor activity. We unexpectedly discover that most embryo cells transition through the same initial transcriptional state early in germination, even though cell identity has already been established during embryogenesis. Cells later transition to cell type-specific gene expression patterns. Furthermore, our analyses support previous findings that the earliest events leading to the induction of seed germination take place in the vasculature. Overall, our study constitutes a general framework with which to characterize Arabidopsis cell transcriptional states through seed germination, allowing investigation of different genotypes and other plant species whose seed strategies may differ.

Fig. 1. Germinating embryo scRNA-seq experimental design and impact of protoplast isolation on the transcriptome.

a, Arabidopsis seed germination analysis showing the percentage rupture of different stages during seed germination by measuring testa rupture (red line) and endosperm rupture (green line). Representative images for each time point are shown below. b, Germination procedure and sampling for transcriptomic analyses: (1) RNA-seq of whole isolated embryos without protoplast isolation; (2) RNA-seq of embryo protoplasts; and (3) scRNA-seq of individual embryo protoplasts. c,d, Expression changes of 1,202 genes in response to protoplast isolation (log₂[fold change]). The transcriptional response to protoplast isolation was consistent across time points (heatmap (c)) and included large fold-changes for well expressed genes (MA-plot (d)). e, Venn diagram of the limited overlap in upregulated genes upon protoplast isolation between this study and a previous study by Birnbaum and colleagues. CPM, counts per million; np, non-protoplast; p, protoplast. Source data

Fig. 2. Annotation of germinating embryo cell types using literature-curated marker transcripts.

a, Uniform manifold approximation and projection (UMAP) dimensional reduction and visualization of 12,798 cells in 15 clusters. b, UMAP dimensional reduction and visualization of cells across three time points—12, 24 and 48 h after placement to light—showing the temporal changes in cell and cluster detection. c, Bubble plots showing the enrichment of expression of representative cell type-specific marker transcripts in 15 clusters and the percentage of cells from the two biological replicates (replicate 1, R1; replicate 2, R2) within each cluster at each time point. Marker transcripts were identified from published studies. d, Spatial distribution of cell clusters in the Arabidopsis embryo. SAM, shoot apical meristem; QC, quiescent centre. Source data

Fig. 3. Validation of cell type annotation using RNA in situ hybridization.

a,b, Expression domains of independent marker transcripts of cluster 9 (a) and cluster 14 (b) confirmed the physical location of the cells in these clusters. UMAP dimensional reduction graph visualized the expression of marker transcript at 48 h. The hybridization signals were detected at 48 but not 24 h, also confirming the temporal detection of these clusters in scRNA-seq data. For each time point, the results of hybridization with antisense probes (that is, the test) and sense probes (negative controls) are shown. Representative images are shown (more than ten embryos were examined). Scale bars, 200 μm.

Fig. 4. Clusters 1, 5 and 7 define a trajectory of hypocotyl cortex cell states.

a, UMAP dimensional reduction and visualization of clusters 1, 5 and 7 cells across three timepoints, 12 h, 24 h and 48 h. b, Confirmation of the locations of clusters 1, 5 and 7 using RNA in situ hybridization of a marker transcript specific to these clusters at 12 and 24 h. Left, UMAP dimensional reduction plots of the expression of the marker transcript of AT4G16410 across all cells at each time point. Right, locations of the hypocotyl cortex cell type. Representative images are shown (more than ten embryos were examined). Scale bars, 200 μm. c, Cells of clusters 1, 5 and 7 form a contiguous group together. They sit upon a temporal trajectory from 12 h to 24 h to 48 h, which corresponds to the transition from cluster 5 cells, through cluster 7 to cluster 1 cells (top graph). Reconstruction of pseudotime (bottom graph) follows a trajectory that corresponds to the real time of germination (middle graph). On the trajectory, branch points are denoted by black circles and leaves (outcome cell states) are denoted by grey circles, while numbers are for reference purposes only. d, Co-expressed gene modules across the pseudotime trajectory of cluster 1, 5 and 7 cells. Early pseudotime is equivalent to early germination. The module annotations are major representative Gene Ontology terms associated with the modules, assessed using Gene Ontology term reduction. Source data

Fig. 5. Initial transcriptional stages are established early in germination.

a, Top, UMAP dimensional reduction and visualization of all cells across three time points (that is, 12, 24 and 48 h), showing the temporal changes in cell and cluster detection. Bottom, UMAP dimensional reduction and visualization of cluster 8 and 11 cells across three time points. b, Comparison of transcriptomes of each cell, grouped by cluster, against whole (bulk) seed transcriptomes from early time points during germination. The transcriptomes of cluster 8 cells are strongly similar to the transcriptomes of bulk seeds at 1 and 6 h of germination. c,d, RNA in situ hybridization to confirm the locations of clusters 8 (c) and 11 (d) at 1, 6 and 12 h after 48 h of stratification using two cluster-specific marker transcripts for each. Illustrated sketches indicate the area of expression. The expression of each marker is shown in an adjacent UMAP dimensional reduction plot that displays all of the cells detected at 12 h, some of which belong to other clusters. Representative images are shown (more than ten embryos were examined). Scale bars, 200 μm. The arrows indicate regions where signals were detected. Source data

Fig. 6. Transcription factors of clusters 8 and 11 are predicted and validated to affect seed germination.

a, Pseudotime models of cell developmental trajectories for clusters 8 and 11, constructed using CSHMM-TF. P indicates paths and D indicates nodes. Nodes demarcate the start and end of each path. Mutants with mutation of the genes encoding the transcription factors highlighted in red were selected for germination assay. b, Germination assay of T-DNA mutant lines predicted in CSHMM-TF of clusters 8 and 11. The percentage of germination of seeds (endosperm rupture) at 24 h is shown. Each data point represents one biological replicate of 50 scored seeds. Seeds from two individual parent plants (_1 and _2) of each mutant line were included. Four biological replicates are included for each mutant line. Asterisks denote significant differences compared with wild-type Col-0. Statistical significance was determined by two-tailed paired Student’s t-test (P < 1 × 10⁻⁵). Mutants with mutation of the genes highlighted in red were selected for RNA-seq. c, Number of DEGs (FDR < 0.05) in mutants at 24 h compared with wild-type Col-0. Positive values represent upregulation in the mutant and negative values represent downregulation in the mutant. d, Active transcription factors (TFs) in every CSHMM-TF model of the 15 clusters identified, comprising a total of 81 unique transcription factors, 39 of which are active in one cell cluster only. Source data

Extended Data Fig. 1. Quality assessment and integration of scRNA-seq data.

a, Numbers of cells captured for each time point (12, 24, 48 h) and biological replicate (R1, R2) in Col-0. b, Distribution of number of detected genes per cell. c, MDS plot of pseudo-bulks for each scRNA-seq sample. d, UMAP plot depicting relative similarity of all cells post batch correction and data integration. e, Total Unique Molecular Identifier (UMI) count per cell for all cells. f, Proportional distribution of cells between clusters in each sample.

Extended Data Fig. 2. RNA in situ hybridisation of marker transcripts specific to cluster 9 (a) and cluster 14 (b).

Representative images are shown with more than 10 embryos examined. Scale bars indicate 200 μm.

Extended Data Fig. 3. Identification of marker transcripts for clusters 1, 5 and 7.

a, UMAP dimensional reduction and visualisation of cells at three individual time points, 12 h, 24 h, 48 h, and all time points together. b, Most highly specific marker transcripts for each of clusters 1, 5 and 7 individually. The plots illustrate that marker transcripts highly specific to each of these clusters individually could not be identified, likely due to high similarity in transcriptomes between the three clusters. These most highly individual cluster-specific marker transcripts were still expressed in clusters other than 1, 5 and 7. Source data

Extended Data Fig. 4. Identification of combined marker transcripts for clusters 1, 5 and 7.

a, UMAP dimensional reduction and visualisation of cells at three individual time points, 12 h, 24 h, 48 h, and at all time points together. b, Most highly specific marker transcripts for clusters 1, 5, 7 combined. Marker transcripts identified were more highly specific to the clusters when these clusters were analysed as a group. Source data

Extended Data Fig. 5. RNA in situ hybridisation of marker transcripts.

a–c, Results of RNA in situ hybridisation of marker transcripts specific to clusters 1, 5 and 7 combined (a), cluster 8 (b) and cluster 11 (c). The icon on the top summarizes the location of the cell type where the marker transcripts expressed. UMAP dimensional reduction plot of expression of the marker transcript across all cells at the corresponding time point. Representative images are shown with more than 10 embryos examined. Scale bars indicate 200 μm.

Extended Data Fig. 6. Each cell cluster is defined by a unique gene regulatory program, which reflects dynamic and differing function.

a, Co-expressed gene modules across the pseudotime trajectory of cluster 13 cotyledon mesophyll cells. b, Enriched gene ontology terms of cluster 13. c, Co-expressed gene modules across the pseudotime trajectory of cluster 15 protophloem cells. d, Enriched gene ontology terms of cluster 15. Gene ontology enrichment p1552 values were determined with one-sided Fisher’s exact test, 1553 adjusted for multiple testing with a Bonferroni correction. Source data

Extended Data Fig. 7. CSHMM-TF models for clusters 5, 7, 1, and 6.

CSHMM-TF models for clusters 5 (hypocotyl cortex, early), 7 (hypocotyl cortex, mid), 1 (hypocotyl cortex, late) and 6 (hypocotyl epidermis).

Extended Data Fig. 8. CSHMM-TF models for clusters 2, 4, 10, 14.

CSHMM-TF models for clusters 2 (hypocotyl/radicle endodermis), 4 (radicle quiescent centre - QC, shoot apical meristem - SAM, columella), 10 (hypocotyl/radicle epidermis, and 14 (radicle apical meristem region).

Extended Data Fig. 9. CSHMM-TF models for clusters 3, 12, 9, 8, and 11.

CSHMM-TF models for clusters 3 (cotyledon mesophyll), 12 (stele), 9 (protoxylem), 8 and 11.

Extended Data Fig. 10. The qRT-PCR and bulk RNA-seq of T-DNA mutant lines.

a, Validation of target gene expression in t-DNA mutants by qRT-PCR. Three biological replicates are included and data are presented as mean value +/- SEM. Asterisks (*) indicate statistical significance using two-tailed paired Student’s t-test (P < 0.001), b,c, Overlap of differentially expressed genes (upregulated (b) and downregulated (c)) between t-DNA mutants. Source data