Wood formation is a complex developmental process, exhibiting a continuous and iterative growth habit because of the existence of stem cells in the vascular cambium (Smetana et al., 2019). The process is controlled by hierarchical gene regulatory networks that regulate specific pathways. Stem-differentiating xylem (SDX), a wood-forming tissue that is part of the inner vascular cambium, differentiates into different types of cells. However, we currently lack distinct genetic markers defining sub-stages or cell types of the SDX differentiation processes. An understanding of the networks and key genes in specific cell types is important for improved feedstock sustainability.

To decipher gene markers and the networks in specific cell types of wood formation tissue, protoplasts were prepared for scRNA-seq, which are obtained from the surface of a freshly debarked stem segment of Populus trichocarpa (Figure 1a–e; Lin et al., 2014). Data were pre-filtered at the cell and gene level, resulting in a pool of 12 466 cells with a median of 1566 genes per cell. The uniform manifold approximation and projection (UMAP) algorithm yielded largely overlapping distributions of cells from each of the biological replicates, indicating a high degree of reproducibility (Figure 1f). These unsupervised analyses grouped SDX cells into 12 cell clusters (Figure 1g). The analysis of cell cycle genes revealed that the clustering was not dominated by cell cycle status (Figure 1h). Looking through the clustering profiles, we found that the specialized cells exhibited relatively distinct transcriptome profiles (Figure 1g). Notably, our gene expression and signature analyses provide fresh insights into cluster-specific properties.

We then used the following three strategies to annotate cell types in the SDX. First, we used marker genes identified from several well-established high-resolution transcriptome analyses of poplar. Second, we performed RNA in situ hybridization assays (RISH) for eight cluster-specific genes. Third, marker genes were selected to perform transgenic experiment to assign some cell clusters. The single-cell transcriptome atlas enabled us to identify most of the major cell types and several markers in the cluster cloud (Figure 1i,j). The expression pattern of these genes in RISH not only confirmed the above annotations but also enabled us to assign unknown cell clusters (Figure 1k). POPTR_007G044500 (cluster 1) and POPTR_009G101900 (cluster 4) displayed a higher RNA abundance in vessel cells and approximately no signals in other cells. Obvious RNA signals were detected for POPTR_005G150300 (cluster 2) in ray cells, indicating the cluster 2 belongs to ray cells. RNA signals for POPTR_002G104600 (cluster 7) were mainly observed in the cells that belong to phloem cells. Cluster 0 is composed of cambium cells and expressed marker genes such as PtCRF4 (POPTR-015G023200), PtLBD1 (POPTR-008G043900) and PtAIL5 (POPTR-018G091600); clusters 5, 8, 11 are composed of fibre cells and contained marker genes such as PtWPP2 (POPTR-014G021900) and PtNEV (POPTR-001G406300); clusters 1, 3 and 4 are composed of vessel cells and contained marker genes PtCSLC12 (POPTR-005G146900), PtNAC82 (POPTR-007G099400) and PtWAT1 (POPTR-002G029100). PtWAT1 (POPTR-002G029100) and PtCesA8 (POPTR-004G059600) were identified to be upregulated and expressed in vessel and fibre cells in our clustering analysis. To test the applicability of their promoter for cluster-specific expression, we generated transgenic P. tomentosa expressing β-glucuronidase (GUS) under the control of the PtWAT1 and PtCesA8 promoter (PtWAT1pro::GUS, PtCesA8pro::GUS) and examined GUS staining patterns in various tissues (Figure 1l,m). Generally, GUS staining of the PtWAT1pro::GUS and PtCesA8pro::GUS lines displayed vessel and fibre specificity, respectively (Figure 1n,o). This is consistent with their transcriptional signals of cell clusters (Figure 1p,q). We also identified several novel marker genes that were highly and specifically enriched in their respective clusters.

To order cells along a reconstructed “trajectory” of SDX cell differentiation, Monocle 2 and STREAM algorithms were employed to disentangle and visualize complex branching trajectories. The vascular cambium is composed of two cell types including fusiform initials and ray initials, and cambium cells divide periclinally inwards to xylem mother cells and outwards to phloem mother cells. The SDX tissue differentiation is essential to understand the wood formation process. It is interesting to note that when considering all clusters of SDX together (phloem cells and unknown clusters were not included), the result revealed that cluster 0 (cambium cells) topologically bifurcated into two trajectories (Figure 1r). A similar output was observed by employing the STREAM software (Figure 1s). The vascular cambium has typically two morphologically distinct types of initials: the axillary elongated fusiform initials that will lead to the formation of the axial system (including vessel and fibre cells) and the smaller isodiametrical ray initials giving rise to the radially orientated parenchymatous rays, supporting that, one trajectory demonstrated gradual transitions from cells of cambium, to fibre and vessel cells and the other from cambium, to fibre and vessel cells and ray cells.

To study the evolutionary effects of polyploidy on the transcriptional network underlying wood formation, we reanalysed the functional genomic and transcriptome data for a large number of duplicated gene pairs formed by ancient polyploidy events in poplar (Sundell et al., 2017). Of the 2932 WGD pairs in our marker gene set, we identified 1142 paralogs that were expressed in the same cell type/cluster, suggesting the regulatory divergence contributes to the process of cambial growth and wood formation. TFs were more central in the network than other genes and may serve essential functional roles in cambial growth and wood formation. Notably, compared with the WGD-derived TFs, other WGD paralogous pairs showed more narrow transcription profiles with expression restricted to the same cell type/cluster (Figure 1t; P = 1.450297e-10; Fisher’s exact test). This may reflect the existence of strong genetic control over the TFs, as they perform central roles in regulatory networks.

Altogether, the profiling of distinct cell-type-specific transcription in multicellular organisms is essential for elucidating unique developmental regulators and functional genes that give cells their distinctive forms and functions. The data presented here show that profiling of SDX using high-throughput scRNA-seq of thousands of cells offers an unparalleled view of the high heterogeneity and provide key regulators in the wood formation process.

木材的形成是一个复杂的发育过程，由于维管形成层中干细胞的存在，表现出连续和迭代的生长习性（Smetana et al ., 2019）。该过程由调节特定途径的分层基因调节网络控制。茎分化木质部 (SDX) 是一种木材形成组织，是内部血管形成层的一部分，可分化成不同类型的细胞。然而，我们目前缺乏定义 SDX 分化过程的子阶段或细胞类型的独特遗传标记。了解特定细胞类型中的网络和关键基因对于提高原料可持续性很重要。

为了破译木材形成组织的特定细胞类型中的基因标记和网络，为 scRNA-seq 制备了​​原生质体，这些原生质体是从毛果杨新鲜去皮茎段的表面获得的（图 1a-e；Lin等人，2014）。数据在细胞和基因水平进行了预过滤，产生了 12 466 个细胞池，每个细胞的中位数为 1566 个基因。统一流形近似和投影 (UMAP) 算法从每个生物复制中产生了大部分重叠的细胞分布，表明高度可重复性 (图 1f)。这些无监督分析将 SDX 细胞分为 12 个细胞簇（图 1g）。细胞周期基因的分析表明，聚类不受细胞周期状态的支配（图 1h）。通过聚类图谱，我们发现特化细胞表现出相对不同的转录组图谱（图 1g）。值得注意的是，我们的基因表达和特征分析为集群特定属性提供了新的见解。

然后，我们使用以下三种策略来注释 SDX 中的细胞类型。首先，我们使用了从几个成熟的杨树高分辨率转录组分析中鉴定的标记基因。其次，我们对八个簇特异性基因进行了 RNA 原位杂交测定 (RISH)。第三，选择标记基因进行转基因实验以分配一些细胞簇。单细胞转录组图谱使我们能够识别集群云中的大多数主要细胞类型和几个标记（图 1i，j）。这些基因在 RISH 中的表达模式不仅证实了上述注释，而且使我们能够分配未知的细胞簇（图 1k）。POPTR_007G044500（第 1 组）和 POPTR_009G101900（第 4 组）在血管细胞中显示出更高的 RNA 丰度，而在其他细胞中几乎没有信号。在射线细胞中检测到POPTR_005G150300（簇2）的明显RNA信号，表明簇2属于射线细胞。POPTR_002G104600（第 7 组）的 RNA 信号主要在属于韧皮部细胞的细胞中观察到。簇 0 由形成层细胞和表达的标记基因组成，例如PtCRF4 (POPTR-015G023200)、PtLBD1 (POPTR-008G043900) 和PtAIL5 (POPTR-018G091600)；簇 5、8、11 由纤维细胞组成，含有PtWPP2 (POPTR-014G021900) 和PtNEV (POPTR-001G406300) 等标记基因；簇 1、3 和 4 由血管细胞组成，包含标记基因PtCSLC12 (POPTR-005G146900)、PtNAC82 (POPTR-007G099400) 和PtWAT1 (POPTR-002G029100)。PtWAT1 (POPTR-002G029100) 和PtCesA8在我们的聚类分析中，（POPTR-004G059600）被鉴定为在血管和纤维细胞中上调和表达。为了测试它们的启动子对簇特异性表达的适用性，我们在PtWAT1和PtCesA8启动子（PtWAT1pro ::GUS，PtCesA8pro ::GUS）的控制下产生了表达β-葡萄糖醛酸酶（GUS ）的转基因毛白杨并检查了 GUS 染色各种组织中的模式（图 1l，m）。通常，PtWAT1pro ::GUS 和PtCesA8pro的 GUS 染色::GUS 线分别显示血管和纤维特异性（图 1n，o）。这与它们的细胞簇转录信号一致（图1p，q）。我们还鉴定了几种新的标记基因，它们在各自的簇中高度且特异性地富集。

为了沿着 SDX 细胞分化的重建“轨迹”排列细胞，使用 Monocle 2 和 STREAM 算法来解开和可视化复杂的分支轨迹。维管形成层由梭形初始和射线初始两种细胞类型组成，形成层细胞沿周向向内分裂为木质部母细胞，向外分裂为韧皮部母细胞。SDX 组织分化对于了解木材形成过程至关重要。有趣的是，当同时考虑所有 SDX 簇（韧皮部细胞和未知簇不包括在内）时，结果显示簇 0（形成层细胞）在拓扑上分为两个轨迹（图 1r）。使用 STREAM 软件观察到类似的输出（图 1s）。

为了研究多倍体对木材形成的转录网络的进化影响，我们重新分析了杨树古老多倍体事件形成的大量重复基因对的功能基因组和转录组数据（Sundell等，2017）。在我们的标记基因组中的 2932 个 WGD 对中，我们鉴定了 1142 个在相同细胞类型/簇中表达的旁系同源物，这表明调节分歧有助于形成层生长和木材形成的过程。TFs 在网络中比其他基因更重要，并且可能在形成层生长和木材形成中发挥重要的功能作用。值得注意的是，与 WGD 衍生的 TF 相比，其他 WGD 旁系同源对显示出更窄的转录谱，其表达仅限于相同的细胞类型/簇（图 1t；P  = 1.450297e-10；Fisher 精确检验）。这可能反映了对 TF 的强大遗传控制的存在，因为它们在监管网络中发挥着核心作用。

总而言之，多细胞生物中不同细胞类型特异性转录的分析对于阐明赋予细胞独特形式和功能的独特发育调节因子和功能基因至关重要。此处提供的数据表明，使用数千个细胞的高通量 scRNA-seq 对 SDX 进行分析提供了无与伦比的高异质性视图，并提供了木材形成过程中的关键调节剂。