文献分享 | 浙江大学樊龙江团队绘制首张水稻种胚单细胞时空图谱

近日，浙江大学樊龙江教授课题组联合华大生命科学研究院等研究团队在国际知名期刊Developmental Cell 发表了题为”Spatiotemporal transcriptomic landscape of rice embryonic cells during seed germination” 的研究论文。该论文利用空间和单细胞转录组测序技术，首次构建了水稻种胚单细胞分辨率的空间转录图谱。该研究利用吸胀24小时的种子构建了至今为止细胞类型最为完整的水稻胚细胞时空图谱。基于构建的胚细胞时空图谱，同时开展了水稻种子萌发过程中不同细胞类型的基因表达以及营养代谢、植物激素传导相关的关键基因通路的分析。该研究为水稻胚细胞类型鉴定提供了重要空间图谱参考信息，为萌发过程不同类型细胞功能的分子生物学研究奠定了基础。

该研究首先建立了水稻种胚单细胞水平的空间转录组分析流程。具体来说，基于切片生成的高清细胞壁染色图像，进行了深度学习模型的训练以实现细胞的自动分割。收集浸水吸胀的水稻种子胚样本进行Stereo-seq空间和单细胞转录组测序，并分别使用划块（Bin50）和细胞分割方法对水稻种胚5个主要解剖区域的细胞类型进行了精细的注释。该研究鉴定出14种水稻胚细胞类型，并发现了此前从未报道的盾片细胞类型。

Figure 1. An overview of this study
(A) Workflow of Stereo-seq and scRNA-seq data acquisition from germinating rice embryos. Rice embryos were stripped from 6-, 24-, 36-, and 48-HAI seeds, and half of them were used in Stereoseq. Embryos were paraffin-embedded and sectioned. The sections were stained for cell wall observation. In situ RNA capture was performed on the sections loaded onto the Stereo-seq chip. Finally, cDNA amplification, library construction, and sequencing were conducted. Meanwhile, embryos of the four time points were used in scRNAseq.
(B) Performance of different cell-segmentation models. The models of ‘‘nuclei,’’ ‘‘cyto,’’ CP, and ‘‘CPx’’ used images of nuclei-stained, cytoplasm fluorescent, or diverse cells as the training dataset,15 respectively. The CP_embryo is an updated model by training the CP model, using the rice embryo cell wall images generated by this study. Left panel: the DICE threshold measures similarity between the predicted and the ground-truth masks, with 1.0 being a perfect match. Right panel: predicted cell area distribution relative to the ground truth. The CP_embryo model had the closest cell area distribution relative to the ground truth and the highest average recall of cells predicted from the ground truth.
(C) Comparison of cell-segmentation results from the same image by different models.
(D) An example of cell segmentation by CP_embryo. The right-top panel displays details of cell structure in the plumule (PL) and coleoptile (CO) areas. The right-middle panel displays the predicted cell-segmentation result, and the right-bottom one shows the comparison of predicted cell-segmentation result with the ground truth, respectively. (E) The top panel illustrates examples of lasso sampling in cell gaps (gap) and environment (env) regions, and the bottom panel displays the expression correlation analysis among cell gaps, environment, and various other cell regions. Different colors refer to different cell types.

Figure 2. Annotation of cell types of rice embryo
(A) An example of cell clustering and cell-type annotation based on the 24-HAI spatial transcriptomic data. The middle panel shows the five anatomical regions of rice embryo, including scutellum (SC), coleoptile (CO), plumule (PL), radicle (RA), and epiblast-coleorhiza (EP-CR). The left and right panels show the UMAP clustering results of embryo cells and their spatial locations and cell-type annotation, using bin-based and cell segmentation-based method, respectively. Details about the cell types were described in the main text. Scale bar represent 500 mm. (B) Comparison of the number of cells per sample and the count per cell produced by the Bin50 and single-cell segmentation method. Error bars represent standard deviation (SD) calculated from independent slides of emrbyo samples.
(C) Dotplot showing the expression pattern of the selected putative cell-type marker genes. The circle size indicates the relative percentage of cells expressing the marker genes, and the color represents the relative expression level.
(D) The spatial expression pattern and RNA ISH validation results of two selected genes.
(E) Visualization of the spatial expression pattern of the selected marker genes in different cell types based on Bin50 and single-cell unit. The deeper the color, the higher the value.

Figure 3. Cell types and trajectory inference of scutellum region (24 HAI)
(A) Cell distribution of the five cell types in the scutellum region and the expression pattern of the key marker gene of the corresponding cell type at 24 HAI.
(B) In situ hybridization detection of OsSAG12-1 (Os01g0907600) and OsCYS9 (Os04g0350100).
(C) FISH validation result of OsMFT2 (Os01g0111600).
(D) Germination rate of the OsMFT2 knockout line and the wild type (WT). Error bars represent standard deviation (SD) from three independent experiments.
(E) Cell abundance of SCL2 in the OsMFT2 knockout line and WT based on cell-type deconvolution of their embryonic bulk RNA-seq data. Error bars represent standard deviation (SD) from three independent experiments.
(F) Trajectory inference analysis of SCL1-2, SCL2, and BESCL2. Cells are colored by cell type (left) or developmental pseudo-time (right).
(G) Expression heatmap and clustering of the differentially expressed genes (DEGs) along a pseudo-time progression of scutellum parenchyma cells (left); the top DEGs of the cluster showing on the left panel and the enriched Gene Ontology (GO) terms (middle); and scatterplots showing the pseudo-time dynamics of the expression of the selected top DEGs (right).

Figure 4. Spatiotemporal transcriptomic landscape of rice embryo during seed germination
(A) Spatial visualization of cell types identified in the rice embryo sections by Stereo-seq at the four germination time points (6, 24, 36, and 48 HAI). Each image of the same time point represents one replicate. Scale bar represent 500 mm.
(B) Barplot showing the cell count ratio of each cell type identified by Stereo-seq at different samples. The same cell type in different samples is represented by the same color. * indicates a significant change (p < 0.05) in the proportion of cell types at different germination time points, and ** indicates an extremely significant change (p < 0.01). Line plot depicting standardized spatial expression signal by time-series analysis. Three typical expression patterns, patterns A, B and C,in6to 48-HAI embryos are shown, with the enriched representative GO terms of the genes from each pattern shown on left.
(C) Representative gene expression patterns of Stereo-seq data by time-series analysis (right) and the GO enriched pathways corresponding to the expression pattern (left).

Figure 5. Expression pattern of the genes related to nutrients and plant hormones at the cell-type resolution
(A) Expression trend of the genes related to nutrients and plant hormones in 6- to 48-HAI embryos. The relative expression level was generated based on pseudobulk RNA analysis by integrating Stereo-seq and scRNA-seq datasets. Error bars represent standard deviation (SD) from independent samples.
(B) Dotplot showing the expression level of genes encoding proteins involved in nutrient metabolism (top) and biosynthesis and signaling of phytohormones (bottom) in different cell types of germinating rice embryos (6–48 HAI from left to right). The circle size indicates the relative percentage of cells expressing the gene, and the color represents the relative expression level.
(C) The spatial expression pattern of the genes encoding amylase and lipid transfer protein in germinating rice embryos (6–48 HAI, from left to right).