Author response: A digital 3D reference atlas reveals cellular growth patterns shaping the Arabidopsis ovule

Article Figures and data Abstract Introduction Results Discussion Materials and methods Appendix 1 Data availability References Decision letter Author response Article and author information Metrics Abstract A fundamental question in biology is how morphogenesis integrates the multitude of processes that act at different scales, ranging from the molecular control of gene expression to cellular coordination in a tissue. Using machine-learning-based digital image analysis, we generated a three-dimensional atlas of ovule development in Arabidopsis thaliana, enabling the quantitative spatio-temporal analysis of cellular and gene expression patterns with cell and tissue resolution. We discovered novel morphological manifestations of ovule polarity, a new mode of cell layer formation, and previously unrecognized subepidermal cell populations that initiate ovule curvature. The data suggest an irregular cellular build-up of WUSCHEL expression in the primordium and new functions for INNER NO OUTER in restricting nucellar cell proliferation and the organization of the interior chalaza. Our work demonstrates the analytical power of a three-dimensional digital representation when studying the morphogenesis of an organ of complex architecture that eventually consists of 1900 cells. Introduction How organs attain their species-specific size and shape in a reproducible manner is an important question in biology. Tissue morphogenesis constitutes a multi-scale process that occurs in three dimensions (3D). Thus, quantitative cell and developmental biology must address not only molecular processes but also cellular and tissue-level properties. It necessitates the quantitative 3D analysis of cell size, cell shape, and cellular topology, an approach that has received much less attention (Boutros et al., 2015; Jackson et al., 2019). Morphogenesis involves the coordination of cellular behavior between cells or complex populations of cells, leading to emergent properties of the tissue that are not directly encoded in the genome (Coen et al., 2017; Coen and Rebocho, 2016; Gibson et al., 2011; Jackson et al., 2019). For example, plant cells are linked through their cell walls. The physical coupling of plant cells can cause mechanical stresses that may control tissue shape by influencing growth patterns and gene expression (Bassel et al., 2014; Hamant et al., 2008; Hervieux et al., 2016; Kierzkowski et al., 2012; Landrein et al., 2015; Louveaux et al., 2016; Sampathkumar et al., 2014; Sapala et al., 2018; Sassi et al., 2014; Uyttewaal et al., 2012). Concepts involving the minimization of mechanical stresses caused by differential growth within a tissue have been employed to explain morphogenesis of different plant organs with curved shapes (Lee et al., 2019; Liang and Mahadevan, 2011; Rebocho et al., 2017). Developmental changes in the appearance of organs, such as leaves and sepals, are often assessed by focusing on the organ surface (Hervieux et al., 2016; Hong et al., 2016; Kierzkowski et al., 2019). This strategy, however, neglects internal cellular growth patterns. Cellular patterns in deeper tissue layers have classically been studied using 2D sectioning techniques with modern variations, for example in the study of the hypocotyl, relying on automated quantitative histology (Sankar et al., 2014). However, 2D analysis of cellular patterns can also result in misconceptions as was noticed for the early Arabidopsis embryo (Yoshida et al., 2014). For a more complete understanding of morphogenesis, a 3D cellular level description and quantification of the entire tissue under study is essential (Hong et al., 2018; Kierzkowski and Routier-Kierzkowska, 2019; Sapala et al., 2019). Digital 3D organs with cellular resolution are a desired goal, however, it remains a challenge to generate such accurate digital representations. Substantial efforts in animal developmental biology often still lack single-cell resolution (Anderson et al., 2019; Asadulina et al., 2012; Dreyer et al., 2010; Lein et al., 2007; Rein et al., 2002). Notable exceptions are Caenorhabditis elegans (Long et al., 2009) and the early embryo of the ascidian Phallusia mammillata (Guignard et al., 2020; Sladitschek et al., 2020). Model plants, such as Arabidopsis thaliana, are uniquely suited for this task. Plants feature a relatively small number of different cell types. Moreover, plant cells are immobile. As a consequence, one can often observe characteristic cell division patterns associated with the formation and organization of tissues and organs. A cellular level 3D digital organ atlas has been obtained for the early stages of the Arabidopsis embryo (Yoshida et al., 2014) which has very few cells. Mostly, complete atlases have been made for larger organs with simple layered structures like roots and hypocotyls (Bassel et al., 2014; Montenegro-Johnson et al., 2015; Pasternak et al., 2017; Schmidt et al., 2014; Yoshida et al., 2014). The applied approach, however, was incompatible with fluorescent stains. Full 3D processing of live imaged data sets has also been possible in some systems, such as the shoot apex but is limited by light penetration to outer layers (Montenegro-Johnson et al., 2019; Refahi et al., 2020; Willis et al., 2016). The ovule is the female reproductive organ of higher plants. It harbors the embryo sac with the egg cell that is protected by two integuments, lateral determinate tissues that develop into the seed coat following fertilization. The Arabidopsis ovule has been established as a model to study several important aspects of tissue morphogenesis including primordium formation, the establishment of the female germ line, and integument formation (Chaudhary et al., 2018; Gasser and Skinner, 2019; Nakajima, 2018; Schmidt et al., 2015). The ovule exhibits an elaborate tissue architecture exemplified by its multiple cell types and extreme curvature visible in the highly asymmetric growth of the two integuments. This property makes it ideal for addressing the complexity of morphogenetic processes. Qualitative descriptions of ovule development exist (Christensen et al., 1997; Robinson-Beers et al., 1992; Schneitz et al., 1995) but a quantitative cellular characterization is only available for the tissue that forms the germ line (Lora et al., 2017; Hernandez-Lagana et al., 2020). To make the next step in the study of ovule morphogenesis therefore requires 3D digital ovules with cellular resolution over all developmental stages. Here, we constructed a canonical 3D digital atlas of Arabidopsis ovule development with cellular resolution. The atlas covers all stages from early primordium outgrowth to the mature pre-fertilization ovule and provides quantitative information about various cellular parameters. It also provides a proof-of-concept analysis of spatial gene expression in 3D with cellular resolution. Our quantitative phenotypic analysis revealed a range of novel aspects of ovule morphogenesis and a new function for the regulatory gene INNER NO OUTER. Results Arabidopsis ovules become apparent as finger-like protrusions that emanate from the placental tissue of the carpel (Robinson-Beers et al., 1992; Schneitz et al., 1995; Figure 1A). The ovule is a composite of three clonally distinct radial layers (Jenik and Irish, 2000; Schneitz et al., 1995). Thus, its organization into L1 (epidermis), L2 (first subepidermal layer), and L3 (innermost layer) follows a general principle of plant organ architecture (Satina et al., 1940). Following primordium formation, three proximal-distal (PD) pattern elements can be recognized: the distal nucellus, central chalaza, and proximal funiculus, respectively (Figure 1A,B). The nucellus produces the megaspore mother cell (MMC), a large L2-derived cell that eventually undergoes meiosis. Only one of the meiotic products, the functional megaspore, survives and continues development. It develops into the eight-nuclear, seven-celled haploid embryo sac. The embryo sac, or female gametophyte, carries the egg cell. The chalaza is characterized by two integuments that initiate at its flanks. The two sheet-like integuments are determinate lateral organs of epidermal origin that undergo planar or laminar growth (Jenik and Irish, 2000; Schneitz et al., 1995; Truernit and Haseloff, 2008). The integuments grow around the nucellus in an asymmetric fashion eventually forming a hood-like structure and contributing to the curved shape (anatropy) of the mature ovule. Each of the two integuments initially forms a bilayered structure of regularly arranged cells. Eventually, the inner integument forms a third layer. The outer integument consists of two cell layers throughout its development. The two integuments leave open a small cleft, the micropyle, through which a pollen tube can reach the interior of the ovule (Figure 1A,C). The funiculus represents a stalk-like structure that carries the vasculature and connects the ovule to the placenta. Ovules eventually orient along the apical-basal or long axis of the gynoecium (pistil) with the micropyle facing toward the stigma (anterior half of ovule oriented gynapically), whereas the opposite side of the ovule faces the bottom of the gynoecium (posterior half, oriented gynbasally) (Simon et al., 2012; Figure 1B, see below). Figure 1 Download asset Open asset Stage-specific 3D digital ovules with cellular and tissue resolution. (A) 3D rendering of confocal z-stacks of SR2200-stained cell walls of ovule depicting ovule development from initiation at stage 1-I to maturity at stage 3-VI. (B) The different polarities of the ovule: the proximal-distal axis and the anterior-posterior (gynapical-gynbasal) axis are indicated. (C) 3D rendering of confocal z-stacks with multi-view of an ovule depicting the quality of the raw microscopic image. (D) Mid-section clip plane from the TO-PRO-3 channel displaying a two-nuclear embryo sac and mitotic nuclei. (E) Pipeline generating 3D digital ovules: raw data, PlantSeg cell contour prediction, 3D GASP segmentation, cell type annotation and quantitative analysis. (F) Mid-sagittal section of ovules from stages 1-I to 3-VI showing the cell type organization in wild-type ovules. Stages 1-I to 2-II includes radial L1, L2, L3 labeling. From stage 2-III, individual cell type labels are assigned according to the specific tissue. (G) 3D view of a mature ovule with cell type labels. The inner tissues are extracted from the 3D ovule after removing the overlying tissues and visualized separately. Different colors represent different tissue type labels. ii1/ii2, oi1/oi2 designate the integument layers as described in Beeckman et al., 2000. Number of 3D digital ovules scored: 10 (stages 2-III, 2-IV, 2-V, 3-I, 3-II, 3-IV, 3-VI), 11 (3-III, 3-V), 13 (stage 2-II), 23 (stage 1-I), 49 (stage 2-I), 66 (stage 1-II). Abbreviations: ab, abaxial; ad, adaxial; ch, chalaza; es, embryo sac; fu, funiculus; ii, inner integument; mp, micropyle; nu, nucellus; oi, outer integument. Scale bars: 20 μm. Figure 1—source data 1 Includes information of the available wild-type dataset of ovules at different developmental stages and their respective IDs. https://cdn.elifesciences.org/articles/63262/elife-63262-fig1-data1-v1.xlsx Download elife-63262-fig1-data1-v1.xlsx Generating stage-specific 3D digital ovules with cellular resolution Ovules are buried within the gynoecium and live imaging of Arabidopsis ovule development is not feasible, except for a short period of time and with a focus on a given cell (Tofanelli et al., 2019; Valuchova et al., 2020). Thus, we resorted to imaging cohorts of fixed specimens. We obtained z-stacks of optical sections of fixed and cleared ovules at different stages by laser scanning confocal microscopy (CLSM). The image stacks were further handled using MorphoGraphX (MGX) software (Barbier de Reuille et al., 2015; Strauss et al., 2019). The imaging method has recently been described in detail (Tofanelli et al., 2019). In short, we dissected and fixed ovules of different stages, cleared the ovules using ClearSee (Kurihara et al., 2015) and simultaneously stained the cell wall and nuclei using the fluorescent stains SR2200 (Musielak et al., 2015) and TO-PRO-3 iodide (TO-PRO-3) (Bink et al., 2001; Van Hooijdonk et al., 1994), respectively (Figure 1C,D). We processed the raw 3D datasets using PlantSeg, a deep learning pipeline for 3D segmentation of dense plant tissue at cellular resolution (Wolny et al., 2020; Figure 1E) (Materials and methods). The pipeline includes two major steps: cell wall stain-based cell boundary prediction performed by a convolutional neural network (CNN) and 3D cell segmentation based on the respective cell boundary predictions. Even after extensive optimizations, the procedure still resulted in some mistakes. We found that two distinct groups of cells represented the main sources of error. The first group included the MMC and its direct lateral neighbors at stages 2-III to 2-V. The second group encompassed the cells of the late embryo sac (stages 3-V/3-VI) (Tofanelli et al., 2019). We believe the reason for these errors lies in poor staining with SR2200 and could be due to their cell walls being particularly thin or of a biochemical composition recalcitrant to staining. Poor staining of the embryo sac cells could be explained by the observation that they exhibit only partially formed cell walls (Mansfield et al., 1991). An additional minor source of errors related to remaining general segmentation mistakes that were partly manually corrected. For further analysis, we selected ovules devoid of apparent segmentation errors for stages 1 to 2-II and 3-I to 3-IV. For stages 2-III to 2-V, we included ovules containing no more than five under-segmented (uncorrected) cells in the region occupied by the MMC and its lateral neighboring cells (≤10% of nucellar cells). Regarding mature ovules (stages 3-V/3-VI) we included ovules devoid of apparent segmentation errors in the sporophytic tissue. Cell-type labeling of 3D digital ovules with cellular resolution Following the generation of 3D cell meshes, we added specific labels to individual cells, thereby describing tissue types, such as radial cell layers (L1, L2, L3), nucellus, internal tissue of the chalaza, inner or outer integument, or the funiculus (see Supplementary file 1 for cell types). Staining with TO-PRO-3 also allowed the identification of cells undergoing mitosis (Figure 1D). Available computational pipelines for near-automatic, geometry-based cell type identification (Montenegro-Johnson et al., 2015; Montenegro-Johnson et al., 2019; Schmidt et al., 2014) failed to provide reasonably good and consistent results. This was likely due to the ovule exhibiting a more complex tissue architecture. We therefore performed cell type labeling by a combination of semi-automated and manual cell type labeling. The entire procedure, from imaging to the final segmented and labeled digital ovule, takes about 45 min per z-stack. For younger ovules, the procedure takes even less time. In summary, the combined efforts resulted in a high-quality reference set of 158 hand-curated 3D digital ovules of the wild-type Col-0 accession (≥10 samples per stage). They feature cellular resolution, cover all stages, and include annotated cell types and cellular features (Figure 1F; Video 1). Video 1 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Mature 3D digital ovule with cell and tissue resolution. Video represents the summary of the workflow and multidimensional view of a 3D digital ovule of stage 3-IV generated from a z stack of cell wall images. Different colors represent 3D cells grouped according to respective tissue type labels. To the end, the 3D surfaces of inner tissues are extracted from the 3D ovule after removing the overlying tissues and visualized separately. Overall assessment of ovule development The different stages of Arabidopsis ovule development were defined previously (Schneitz et al., 1995). Throughout this work, we used a more precise definition of the subdivision of stage 1 into stage 1-I and 1-II. The subdivision was based on the first appearance of the signal of a reporter for WUSCHEL expression that became robustly apparent when ovule primordia consisted of 50 cells (see below). We first determined the average number of cells per ovule and stage (Figure 2A,B) and found an incremental increase in cell number for every consecutive stage of development until ovules at stage 3-VI exhibited an average of about 1900 cells (1897 ± 179.9 (mean ± SD)) (Table 1). We also assessed the mean volume per ovule for each stage by summing up the cell volumes of all cells in a given ovule (Figure 2A,C; Table 1). We measured a mean total volume for ovules at stage 3-VI of about 5 × 105 μm3 (4.9 × 105 ± 0.7 × 105). Figure 2 Download asset Open asset Ovule developmental stages and overall growth patterns. (A) 3D cell mesh view of wild-type ovules at different stages displaying heatmaps of cell volume ranging from 0 to 1200 µm3. (B, C) Plots depicting the total number of cells and total volume of individual ovules from early to late stages of development, respectively. Number of 3D digital ovules scored: 10 (stages 2-III, 2-IV, 2-V, 3-I, 3-II, 3-IV, 3-VI), 11 (3-III, 3-V), 13 (stage 2-II), 23 (stage 1-I), 49 (stage 2-I), 66 (stage 1-II). Mean ± SD is shown. Scale bar: 20 μm. Figure 2—source data 1 Includes the list of ovule IDs, stage, total number of cells and total volume of the available wild-type dataset. https://cdn.elifesciences.org/articles/63262/elife-63262-fig2-data1-v1.xlsx Download elife-63262-fig2-data1-v1.xlsx Table 1 Cell numbers and total volumes of ovules at different stages. Stage*N cellsVolume (x104 μm3)N mitotic cells% mitotic cells1-I39.6 ± 5.30.5 ± 0.091.0 ± 0.00.7 ± 1.21-II74.0 ± 17.11.0 ± 0.21.3 ± 0.50.7 ± 0.92-I176.9 ± 31.52.5 ± 0.43.1 ± 2.11.8 ± 1.22-II220.6 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 5.30.5 ± ± ± ± ± Number of 3D digital ovules scored: 10 (stages 3-II, 3-IV, 3-VI), 11 (stages 3-V), 13 (stage 2-II), (stage 1-I), (stage 1-II). represent mean ± tissue growth patterns and ovule development To a first into growth we performed a quantitative cellular analysis of different tissues following primordium formation (Table We growth of the nucellus the embryo except at the of stage with the funiculus growth at stage 3-II, the cell numbers and tissue volume of the chalaza and integuments stages 2 and a description see with Appendix 1 and Appendix Table 2 Cell numbers and total volumes of the major ovule cellsVolume (x104 μm3)N cellsVolume (x104 μm3)N cellsVolume (x104 μm3)N cellsVolume (x104 μm3)N cellsVolume (x104 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± of 3D digital ovules scored: 10 (stages 3-II, 3-IV, 3-VI), 11 (stages represent mean ± To more into the growth ovule development, we assessed its To this end, we established a we growth for the developmental period by of individual live to the plant at different stages (stages according to et al., Figure We generated a growth by a to a of over time (Figure file we dissected ovules from of various determined the stage of the ovules for each and used the growth for an of the of the different ovule stages (see Materials and methods). It was previously noticed that early stage ovules not develop within a (Schneitz et al., 1995; et al., 2020; et al., 2020). In line with these we ovules of different stages within a given file was for all stages and ovules within a were found to be two to consecutive stages. We determined the number of ovules per stage for a given and defined the stage with the number of ovules as Stages and were grouped as about numbers of stages. From the growth we the time between the and a given This time an of the of a given ovule stage (Figure Figure Download asset Open asset of ovule development. (A) depicting at different time 0 to Number of scored: (B) showing the increase in over time. Mean ± SD are represented as The is by the The model and are indicated. (C) between and of ovule stages. The at the represent the range of the given ovule stage The respective time were from the in The within the the of the respective ovule stages. Number of scored: Scale bar: Figure data 1 Includes the information on measured for each stage and the in of ovule stages. Download growth we that ovules from placental tissue in with a of about and that by the time of a of about Ovules from early stage 1 to the of stage within a period of or The three main stages in their 1 about stage 2 was and stage was or respectively. provide a for ovule development when to (Schneitz et al., 1995). Using the mean ovule volume per stage (Table we average growth for time For up to the of stage we an average growth of × of × for the including stages to and of × for the including stages 3-I to 3-IV. The growth to × for stage We growth (Figure In a first we growth and cell proliferation two consecutive stages by the respective of the stage (Figure growth a behavior ovule development and for several stage we determined growth however, we noticed that from stages to 2-II the cell proliferation was higher than the growth growth by cell the opposite was for stages to 3-IV that ovule growth these stages was due to cell In a second we the growth of the ovule by the between the mean total volume or mean total number of cells for two consecutive stages (Figure The in Figure and each This is due to the relatively minor in the of the individual ovule stages. Figure Download asset Open asset of Arabidopsis ovule development. Plots the volume or cell number (A) growth to the growth of the from stage 1-I to 1-II. (B) growth between two consecutive stages. and depicting tissue growth the different ovule stages. Stages are indicated. growth cell proliferation Number of 3D digital ovules scored: 10 (stages 2-III, 2-IV, 2-V, 3-I, 3-II, 3-IV, 3-VI), 11 (3-III, 3-V), 13 (stage 2-II), 23 (stage 1-I), 49 (stage 2-I), 66 (stage 1-II). 10 (stages 2-III, 2-IV, 2-V, 3-I, 3-II, 3-IV, 3-VI), 11 (stages 3-V), 13 (stage 2-II), (stage 1-I), (stage 1-II). Number of 3D digital ovules scored: 10 (stages 2-III, 2-IV, 2-V, 3-I, 3-II, 3-IV, 3-VI), 11 (3-III, Abbreviations: ch, chalaza; es, embryo sac; fu, funiculus; ii, inner integument; nu, nucellus; oi, outer integument. Figure data 1 Includes the growth and the growth and proliferation and the tissues growth per each ovule developmental Download we the growth of the major tissues (Figure We the between the mean total volume or mean total cell number of a tissue for two consecutive stages and it by the ovule The the tissue growth to overall ovule one that the tissue is at a higher than overall ovule growth and one the We that the L3 the growth and the L1 the stage 1-I (Figure stages to all three layers to the overall From stages 2-III to the nucellus, chalaza, integuments, and eventually the embryo sac changes development (Figure For example, up to stage 3-I the outer integument exhibited more growth than the inner integument, from stage 3-I on the growth pattern was In summary, the ovule growth to be in of overall growth and the respective of individual tissues development. of WUSCHEL expression in the nucellus 3D digital ovules a of spatial gene expression patterns and with cellular resolution. As we the expression of WUSCHEL in the ovule. cell development in the shoot et al., 2015; et al., is also ovule development and early pattern formation in the chalaza and integument initiation as gene expression in the chalaza and lack integuments et al., et al., Moreover, is also for MMC formation et al., but its expression must be from the MMC to through et al., 2017). A combination of in reporter gene and that is in the of the Thus, it is that its functions in a fashion et al., et al., 2011; et al., et al., 2017). However, it remains when expression first and at stage it may become its expression is to the nucellus, and expression is limited to the nucellar or is in L2 cells the as To address these we of a reporter line a reporter for et al., 2017). We generated a total of 3D