Cell type-specific dynamics underlie cellular growth variability in plants

The development of organs in multicellular organisms is a complex process involving the coordinated behavior of many individual cells. One might expect that the reproducibility of organ forms would be generated through uniform behaviors at the cellular level. This is indeed the case in developing cotyledons where non-dividing cells exhibit rather homogeneous growth (Zhang et al., 2011). Such uniformity can also be observed in elongating organs, such as anther filaments or roots, where the anisotropic extension of the neighboring cells located at a given position is relatively homogeneous (Fridman et al., 2021; Silveira et al., 2022). However, most of developing organs display significant differences in growth at the cellular level (Elsner et al., 2012; Hong et al., 2016; Kierzkowski et al., 2019; Roeder et al., 2012; Sapala et al., 2018; Silveira et al., 2022; Tauriello et al., 2015).

The origin and role of cellular growth variability are still unclear. It can emerge through the random fluctuations of molecular regulators that can trigger differences between initially similar cells, including heterogeneity in cell wall mechanical properties, turgor pressure, or cell endoreduplication level (Hong et al., 2016; Long et al., 2020; Meyer et al., 2017; Roeder et al., 2010). For example, variability in the timing of cell division, which is particularly prevalent in organs with endoreduplicating cells, leads to local differences in growth between adjacent cells (Roeder et al., 2010; Tauriello et al., 2015). As plant cells are connected to each other by cell walls, mechanical feedback can further amplify or attenuate local growth differences (Coen et al., 2017; Hamant et al., 2008; Uyttewaal et al., 2012). Random fluctuations in cellular growth caused by stochastic changes in cellular mechanical properties have been proposed to be important for temporal growth averaging, which could ensure the emergence of reproducible sepal shapes (Hong et al., 2016).

Local growth differences between adjacent cells could also emerge from the combination of tightly controlled global and local patterning mechanisms. Cell differentiation in organs such as leaves and sepals proceeds in a basipetal direction, leading to the progressive establishment of a growth gradient with cells located at the tip expanding more slowly than those at the base of the organ (Hervieux et al., 2016; Kierzkowski et al., 2019; Kuchen et al., 2012). In parallel with the establishment of global growth gradients, local patterning mechanisms ensure homogeneous spacing of specialized cells, such as trichomes or stomata (Lee and Bergmann, 2019; Torii, 2021; Zuch et al., 2022). These cells follow specific developmental programs that often lead to a large increase in their size; for example, trichomes exhibit fast cellular growth at initiation (Hervieux et al., 2017; Yang and Ye, 2013). Together with the non-synchronous establishment of various cell types (Andriankaja et al., 2012; Donnelly et al., 1999), cell-specific growth could contribute significantly to local growth heterogeneity. However, it is still unknown how and if specialized cells contribute to overall cellular growth variability and impact plant organ development.

Here, we analyzed cellular growth variability at different scales by following growth dynamics in the developing organs of Arabidopsis thaliana. We found that cellular growth variability differs between leaf regions after their specification. Higher local growth variability was associated with the presence of stomata lineages in the leaf blade. Quantification of cellular dynamics in the leaves of a mutant lacking stomata, as well as other plant organs of wild-type plants, further supports the view that local growth variability is mainly caused by stomatal differentiation. We show that stomatal lineages followed specific growth trajectories independently of global organ behavior. Thus, the cell-autonomous behavior of specialized cells underlies the local growth variability in otherwise homogeneously growing tissue. These growth differences are locally buffered to ensure reproducible development.

We used confocal time-lapse imaging to follow the growth of the first true leaf of Arabidopsis thaliana at cellular resolution for 5 days, with time intervals of 12 h. The development of the leaf was comparable to prior studies, suggesting that our experimental setup did not have a substantial impact on normal leaf development (Fig. 1, Fig. S1) (Fox et al., 2018; Kuchen et al., 2012; Zhang et al., 2020). We extracted the 3D surface of all cells in the abaxial epidermis and computed cellular growth rates, cell divisions and growth anisotropy using MorphoGraphX software (see ‘Materials and Methods’ section for details) (Barbier de Reuille et al., 2015; Strauss et al., 2022).

We first investigated cellular growth dynamics at the level of the entire organ in three independent time-lapse series. Our time-lapse series started 1 day after leaf initiation (DAI) when the primordia were small and rod-shaped. Although some temporal differences in growth dynamics could be observed between independent time-lapse series, they all displayed similar growth patterns and developed into leaves of comparable shapes at 6 DAI (Fig. 1, Fig. S1). At early stages (1-2.5 DAI), leaf primordia exhibited fast, homogeneous and anisotropic growth associated with high cell proliferation (Fig. 1A,C,E, Fig. S1), in accordance with previous reports (Burian et al., 2016; Kierzkowski et al., 2019). As the primordia developed (3-6 DAI), we registered a general decline in cell division, cell area expansion and growth anisotropy (Fig. 1B,D,F), and cell size increased (Fig. 1H). Fast cellular expansion and proliferation became progressively restricted to the basal and lateral parts of the leaf. Cell dynamics started to diverge between leaf regions from 3 DAI, with slower expansion and more directional growth (higher anisotropy) in the midrib/petiole, and faster and more isotropic expansion in the blade (Fig. 1A,C,E, Fig. S1A-C,E-G). In the leaf margin, growth and cell divisions terminated relatively early, suggesting its very early specification from 2.5 DAI (Fig. 1A,C,E, Fig. S1A-C,E-G). The distribution of cell sizes in the organ coincided with the cell division patterns. Cells at the margin, midrib/petiole and distal part of the blade tended to become bigger, whereas cells in the proximal, proliferating part of the leaf blade remained smaller (Fig. 1E, Fig. S1D,H). This led to a steep rise in cell size variability that increased from a ratio of around 3 (biggest cell/smallest cell) at 1 DAI to more than 20 at 6 DAI (Fig. 1H). Cellular growth rates progressively decreased in all leaf series, but we observed large differences in cellular growth across the organ. For all time points, the observed maximal growth in the fastest expanding cells was at least two and a half times higher than in the slowest growing cells (Fig. 1B). Growth anisotropy also differed greatly between cells ranging from nearly isotropic growth in some cells to nearly two times faster elongation in one direction compared with the other (Fig. 1D). Therefore, we conclude that cellular growth is highly heterogeneous at the organ level during the first 6 days of leaf development.

Different leaf regions displayed characteristic developmental patterns after their specification. The midrib/petiole region tended to grow slower and preferentially along the proximal-distal (tip-base) axis of the leaf, whereas the growth in the leaf blade was generally faster and more isotropic (Fig. 1A,C, Fig. S1A,B,E,F) (Fox et al., 2018; Kierzkowski et al., 2019; Kuchen et al., 2012). Therefore, we explored whether cellular growth heterogeneity differs between those distinct leaf regions. We used leaf geometry, growth anisotropy (from 3 to 6 DAI), cell sizes and cell geometries to delimit the midrib/petiole and leaf blade regions at the end of the time-lapse series (Fig. S2A-C; see ‘Materials and Methods’ section for details). Using lineage relations, we then traced back the origin of each region to 3 DAI, when the vast majority of mother cells of each clone gave rise to sectors located in either the midrib/petiole or leaf blade (Fig. 2A, Fig. S2D). Cells located at the leaf margin were excluded from the leaf blade as their growth followed distinct developmental characteristics with slower growth rates, early increase in cell sizes and early exit from cell proliferation (Fig. 1A,B,E,F, Fig. S1). Cellular growth rates tended to be higher in the leaf blade compared with the midrib/petiole region, especially at earlier developmental stages (Fig. 2B). Interestingly, the distribution of cellular growth rates was significantly broader in the leaf blade at all developmental stages (Fig. 2B). This indicates that region-specific characteristics can impact cellular growth heterogeneity.

Growth in the leaf follows a basipetal gradient with cells located close to the tip decreasing their growth earlier than cells located at the leaf base (Fig. 1A, Fig. S1A,E) (Donnelly et al., 1999; Fox et al., 2018; Kierzkowski et al., 2019). The divergence in growth distribution between regions could be caused by the different gradient dynamics in the leaf blade compared with the midrib/petiole regions. To evaluate this possibility, we first computed the averaged growth heat maps (average of growth rate of each cell and its adjacent neighbors, weighted by cell area), which reveal the organ-wide growth gradient by diminishing local growth differences between adjacent cells (Fig. 2C). We observed proximal-distal gradients of growth in both regions, but they seemed to follow different and dynamic trajectories (Fig. 2D-F). In the leaf blade, cellular growth increased in an almost linear fashion from the tip to the base (Fig. 2E). By contrast, the growth was more homogeneous along the midrib and often increased in the petiole (Fig. 2F). These differences could contribute to the divergence in growth variability between the leaf blade and midrib/petiole regions.

The fully developed leaf epidermis in Arabidopsis is composed of different cell types, including pavement cells, trichomes, stomata and marginal cells (Andriankaja et al., 2012; Zuch et al., 2022). Stomata are much more prevalent in the leaf blade (Larkin et al., 1996; Torii, 2021). They could contribute to higher growth variability in this region as they follow specific developmental programs (Robinson et al., 2011; Yang and Ye, 2013). We thus explored the contribution of local cellular behaviors to the overall growth variability in the leaf. We computed local growth variability expressed as the difference between absolute and averaged growth for each cell (Fig. 3A) (see ‘Materials and Methods’ section for details). Local growth variability was significantly higher in the leaf blade compared with the midrib/petiole regions (Fig. 3B). This indicates that local differences in cellular behaviors contribute to higher growth variability in the leaf blade. Interestingly, cells displaying the highest local growth variability were often differentiated stomata or cells derived from asymmetric cell division in stomata lineages (Fig. 3A, inset), suggesting that stomata development could underlie the local cell variability observed in the leaf blade.

To evaluate this possibility, we investigated the contribution of different cell types to the local growth variability. To that end, we first identified five different cell types based on their shape, their location within the leaf, their ability to divide and their lineage relations. Dividing cells were classified either as meristematic, if their divisions were symmetric giving rise to all other cell types except stomata, or meristemoid, if their divisions were asymmetric and led to the development of stomata. Note that asymmetric divisions of meristemoids also gave rise to the pavement cells, which are the closest neighbors to the guard cells. Stomata were identified as two guard cells no longer dividing. In all statistical analyses, each pair of guard cells was treated as a single stoma. Elongated non-dividing cells located at the border of the leaf blade were classified as marginal cells. Finally, non-dividing cells located outside the leaf margin, other than stomata, were classified as pavement cells (Fig. 3C; see ‘Materials and Methods’ section for details). Growth of meristematic and pavement cells displayed low local variability oscillating around 4-5% (Fig. 3D). By contrast, in both meristemoid and stomata cells the local growth variability was significantly higher (8-18% depending on the time point) (Fig. 3A, inset, and 3C,D). This indicates that stomata lineages may follow cell-specific growth trajectories that are at least partially independent of the average leaf growth behavior.

If high local growth variability in the leaf blade is caused by stomata development, removing stomata may eliminate this regional difference. To assess this possibility, we analyzed cellular growth in the speechless (spch) knockout mutant, which lacks stomata lineages (MacAlister et al., 2007). To this end, we used previously published time-lapse imaging data of the first true leaves of the spch mutant (Fox et al., 2018). Removing stomata did not abolish the global gradients of growth in the leaf (Fig. 4A). As in the wild type (Fig. 2), we still observed relatively linear gradients of growth in the leaf blade with higher growth rates in its more proximal regions (Fig. 4B,C). Similar to wild type, the growth was rather homogeneous along the midrib and sometimes increased in the petiole region in the spch mutant (Fig. 4D). Thus, behavior specific to stomatal lineages does not seem to contribute to the establishment of differential growth gradients in the leaf. By contrast, abolishing stomata development led to an equalization of the local growth variability in the leaf blade and midrib/petiole regions (Fig. 4E,F). This strongly suggests that cell type-specific growth trajectories of stomatal lineages underlie high local growth variability between adjacent cells that can be observed in the leaf blade of wild-type plants.

Next, we evaluated whether stomata lineages contribute more broadly to the local growth variability in organs other than leaves. To do so, we followed growth at a cellular resolution in various floral organs that are believed to be modified versions of leaf-like structures (Pelaz et al., 2001). First, we tracked the source of local growth variability in sepals, which are known to display highly heterogeneous growth rates between adjacent cells and a strong basipetal gradient of growth (Hervieux et al., 2016; Hong et al., 2016; Tauriello et al., 2015). Cellular growth in sepals was relatively heterogeneous with maximal growth rates of the fastest expanding cells at least three times higher than the slowest growing cells at any given time point (Fig. S3A,D). Local growth variability increased slightly during the development of the sepal from 3 to 7 DAI (Fig. 5A,B), coinciding with a progressive establishment of stomata lineages (Fig. 5C). As in the leaf, meristemoids and differentiated stomata displayed the highest local growth variability (around 10-18%) (Fig. 5A,B,D). Interestingly, local growth variability was relatively low for both small meristematic cells and non-dividing pavement cells (around 3-6%) (Fig. 5D).

To confirm our observations in sepals and leaves, we tracked growth in the abaxial epidermis of the anthers (top part of the male floral reproductive organs where pollen develops). Growth was relatively homogeneous between different anther regions, consistent with a less-pronounced basipetal gradient of growth in anthers compared with other organs (Fig. S3B,E) (Silveira et al., 2022). Although the cellular growth rates of the anther progressively decreased from 6 to 11 DAI, many cells belonging to stomatal lineages followed the opposite trend and temporarily accelerated their growth (Fig. S3B,E). As for leaves and sepals, stomatal lineages in anthers displayed very high local growth variability (between 10 and 20% from 7 to 11 DAI) whereas local growth variability in meristematic and pavement cells was consistently low (around 2-3%) (Fig. 5E-H).

In contrast to all other floral organs, stomata are generally absent in the petal epidermis (Pillitteri and Dong, 2013). We followed the growth of the petal abaxial epidermis between 7 and 8 DAI. At this stage, petals grew slightly faster in the upper/lateral parts and slower in the more proximal/median regions (Fig. S3C). Cellular growth rates between adjacent cells were relatively homogeneous with the local differences rarely exceeding 20% (Fig. 5I, Fig. S3C). Local growth variability calculated for all cells in the petal was low (around 3.5%) and comparable to what we observed for meristematic and pavement cells in other organs (Fig. 5J). Altogether, these data strongly support our idea that cell type-specific behavior of stomatal lineages is the main source of the local growth variability observed during the development of various plant organs.

How do stomata contribute to local growth variability? One possibility is that they follow a stochastic growth program, as opposed to other epidermal cells for which growth is largely determined by spatial and temporal cues. Another possibility is that stomata follow a predefined set of growth dynamics, which differs from that of neighboring cells, owing to their specialization. In the leaf epidermis, a basipetal front of primary cell cycle arrest of general cell proliferation is followed by a secondary arrest front of stomatal lineages (Donnelly et al., 1999; White, 2006). This means that cells located at the same distance from the leaf tip would have different timing of differentiation depending on the cell type, which could lead to local growth variability. To estimate these spatial-temporal differences, we computed heat maps of cell age at the last time point of our time-lapse series (Fig. 6A-C). This confirmed the early specification of the marginal cells as they were on average 2 days older compared with all other cell types (Fig. 6A, Fig. S4). The average ages of stomata and pavement cells at 6 DAI were comparable (Fig. S4), but a spatial delay of stomata differentiation was observed. The oldest pavement cells in the distal part of the leaf blade were 4 days old, whereas the oldest stomata at comparable locations were only 2 days old (Fig. 6B,C). This strongly suggests that temporal misalignments of cell type differentiation might contribute to local growth variability between cells. However, the comparable age of stomata and pavement cells in the more proximal portion of the leaf blade at 6 DAI (Fig. 6B,C, Fig. S4) suggests that other stomata-specific behaviors also increase the local growth variability of those cells.

Stomatal lineages follow a highly controlled developmental program (Robinson et al., 2011; Torii, 2021). We hypothesized that cells within stomata lineages may also display very different growth trajectories compared with their neighbors. To evaluate this idea, we normalized the behavior of all cell lineages by aligning each of them according to days relative to cell type differentiation (DRD), which represents the time when a specific cell type (i.e. stomata, pavement or marginal cells) appear for the first time during our time-lapse series. In general, the growth of the meristematic cells was relatively high, and they maintained a small size before differentiating into specific cell types (Fig. 6D-I). Both pavement and marginal cells slowly decreased their growth rates after differentiation from meristematic cells (Fig. 6D,E). Prolonged growth after the differentiation of marginal and pavement cells led to a significant expansion of their sizes (Fig. 6G,H). By contrast, the transition from meristematic to meristemoid cells was often associated with an increase in their growth rates and a simultaneous decrease in their sizes (Fig. 6F,I). Growth rates in stomata were initially very high and sharply fell over the next 2 days after their differentiation (Fig. 6F). Thus, stomatal lineages displayed very different growth dynamics from all other epidermal cells, sharply increasing and quickly ceasing their growth.

We next investigated how growth variability, caused by differential behavior of specialized cells such as stomata, could influence organ development. Stomata are very small compared with the surrounding pavement cells (Andriankaja et al., 2012), and they occupied only a limited fraction of the epidermis surface at the end of our time-lapse series (e.g. ∼4% of the leaf surface at 6 DAI) (Fig. 7A). The absence of stomata did not strongly affect the early growth and development of the first true leaf, which displayed comparable growth gradients, shapes and sizes in the wild type and the spch mutant (Figs 2C,E,F, 4A,C,D, 7B) (Fox et al., 2018). Therefore, we hypothesized that differential growth of specialized cells such as stomata is likely buffered through local interactions with their neighbors to ensure proper organ development.

To evaluate this possibility, we first investigated whether rapidly growing stomata can influence the growth and/or morphology of adjacent cells. In an isotropically extending leaf blade, asynchronous differentiation of stomata occurs next to pavement cells that undergo strong lobing during their growth (Sapala et al., 2018). To visualize the interactions between stomata and their neighbors clearly, we followed cellular growth during late development of carpels (a part of the female reproductive organ in plants). In this system, stomata differentiate within homogeneously expanding tissue whereby non-lobed epidermal cells grow slowly along the proximal-distal direction (Eldridge et al., 2016; Ripoll et al., 2019). As in the case of leaves, sepals and anthers, stomatal specification in carpels was associated with a strong increase in their cellular growth rates (Fig. 7C). Interestingly, we observed that cellular growth was slightly but significantly lower in cells directly neighboring differentiating stomata compared with other epidermal cells, therefore contributing to cell growth variability (Fig. 7D, Fig. S5). This suggests that fast-growing stomata could locally compress the slow-growing adjacent epidermal cells. To reveal such local effects, we computed subcellular deformation gradients using a morphing function in MorphoGraphX 2.0 (Strauss et al., 2022). First, we used cellular junctions as reference points to compute the deformation gradient and morph cell outlines between 0 h and 72 h. The morphing represents how cells would grow during this time interval, assuming that cell deformation was entirely determined by the displacement of junctions, i.e. there would be no local influence of neighbors. Real and morphed cellular outlines overlapped except at initiating stomata (Fig. 7E). Next, we used the complete outlines of the cells to map the deformation occurring between 0 h and 72 h, revealing a local compression in a small region immediately surrounding the stomata (Fig. 7F). The areal growth of small cells in direct contact with stomata was more affected than the growth of larger neighbors, as the area of local compression covered a more significant portion of small cells (Fig. 7C). Therefore, we conclude that local cell growth differences driven by stomata lineages are buffered by their immediate neighbors, which undergo compression at the subcellular scale.

To investigate further whether local growth variability can be buffered in the context of cells other than stomata, we followed the development of trichomes in the adaxial leaf epidermis. As previously observed in sepals (Hervieux et al., 2017), we measured fast growth rates during trichome initiation (Fig. 7G). This fast increase slowed down very quickly (around 1 day after trichome differentiation) and was followed by a relatively slow expansion of the trichome. Strikingly, surrounding fast-growing cells quickly adapted to accommodate this slow expansion of the trichome base. Over 2 days (from 3 to 5 DAI), neighboring cells developed into anisotropic sectors radiating away from the center of the trichome (Fig. 7H). The maximal growth within the sectors increased but the minimal growth was relatively unchanged (Fig. 7I,J), clearly demonstrating that slow-growing trichomes influence the growth of adjacent cells. Altogether, our data suggest that local growth variability driven by specialized cells is buffered by their neighbors to ensure reproducible organ development.

Growth variability during organ development may reflect random fluctuations in cellular behaviors or be a result of highly controlled patterning mechanisms (Hong et al., 2018; Lee and Bergmann, 2019; Long et al., 2020; Meyer et al., 2017; Robinson et al., 2011). Variability may also be derived from experimental errors in the quantification of cellular geometries (Bassel and Smith, 2016). Here, we have shown that the local variability in cellular growth observed during aerial organ development in Arabidopsis is strongly associated with the presence of specialized cells, such as stomata. We suggest that the default organ growth is relatively homogeneous with smooth gradients of cellular growth rates along the proximal-distal axis of the organ controlled by positional information (Andriankaja et al., 2012; Fox et al., 2018; Das Gupta and Nath, 2015; Horiguchi et al., 2005; Kuchen et al., 2012; Mansfield et al., 2018). We propose two independent mechanisms of how stomata could disturb those smooth gradients to increase local growth variability. First, stomatal lineages follow cell type-specific growth trajectories that are distinct from other epidermal cells. Second, the divergent developmental gradients of stomatal patterning and differentiation of pavement cells cause a spatial mismatch in the timing of the appearance of various cell types (Fig. 6B,C) (Andriankaja et al., 2012; Donnelly et al., 1999). This spatial asynchronization of cellular differentiation combined with the distinct cell-autonomous behavior of stomatal lineages produces strong differences in growth rates between neighboring cells. Thus, even if stochastic component is a contributing factor, cellular growth variability in plants seems to be largely deterministic. Local mismatches in cell expansion mainly result from a combination of cell-autonomous growth programs and differences in the timing of cell type differentiation, which emerge from organ patterning mechanisms.

Surprisingly, variations in the abundance of specialized cells have little impact on organ development. Removing stomata from the leaf epidermis in spch mutants reduces local growth variability but does not significantly affect early leaf development (Figs 4, 7B) (Fox et al., 2018). Likewise, the presence of giant cells in the epidermis seems irrelevant for proper sepal size and shape, and reproducible organ development is independent of trichome formation (Arteaga et al., 2021; Schwarz and Roeder, 2016; Wester et al., 2009). This suggests that local differences in growth rates of specialized cells observed at a given time are buffered during development to ensure reproducible organogenesis.

How could such growth buffering occur? Averaging local variation in cellular growth over time has been shown to be necessary to achieve reproducible shape in plants (Hong et al., 2016; Tsugawa et al., 2017). Here, we have shown that during their specification stomata grow quickly compared with the adjacent pavement cells, but also stop expanding faster than their neighbors (Fig. 6E,F). Over a longer period of time, this will generally lead to a more homogeneous growth. However, stomata appear at different times and within various developmental contexts. For example, when they differentiate in a tissue that is about to cease its expansion (such as the tip of the leaf or the anther), a stomata's fast growth may not be fully compensated for by the prolonged expansion of adjacent epidermal cells. In this case, differential growth between connected cells can generate mechanical conflicts that are able to alter rates and orientations of cellular growth (Echevin et al., 2019; Rebocho et al., 2017). The neighboring cells must physically accommodate this opposing behavior by either changing their geometry or specified growth rates to ensure reproducible development. For example, mechanical compression has been shown to influence growth at the boundary between the shoot apical meristem and the emerging organ primordium via an auxin-independent mechanotransduction pathway (Landrein et al., 2015). Stomata are symplastically isolated from their neighbors and display higher turgor pressure than do pavement cells (Franks et al., 1995, 2001; Kong et al., 2012). They also influence the cortical microtubule organization in surrounding cells (Sampathkumar et al., 2014). We have observed that fast-growing stomata can influence the geometry of adjacent epidermal cells (Fig. 7E) by exerting a local compression on their immediate surroundings (Fig. 7F). However, as stomata are small compared with their neighbors, local compression only slightly affects the growth rates of adjacent cells, in particular cells that are much larger than stomata. The observed local effect of stomata on their neighbors could favor the hypothesis of a passive response (cells being ‘pushed’) over the hypothesis of an active response, which we would expect to entail growth re-orientation or growth reduction in entire cells. The precise mechanistic basis of such compression will require further investigation. One possibility is that cells developing into stomata lineages increase their turgor pressure, which could be a driving force of their fast growth.

Now consider the opposite scenario, when a slow-growing, specialized cell is established within the context of a tissue that will grow for an extended period. Instead of compressing its neighbors, such a cell is expected to exert a physical pulling force on adjacent cells. We would not observe this effect for stomata as they are relatively small, and our time-lapse data did not cover the late expansion of organs. However, we have seen this behavior during trichome establishment. Trichomes are initiated during early development (from 1 DAI) of the first true leaf. As previously reported (Hervieux et al., 2017), trichome initiation was associated with only a brief increase in cellular growth rates lasting around 12-24 h (Fig. 7F). However, we found that the subsequent expansion of the trichome base was very slow (Fig. 7G). This slow planar growth occurs within the context of the fast-growing adaxial epidermis. The maximal growth of cells adjacent to the slowly expanding trichome significantly increased radiating away from the center of the trichome (Fig. 7H). The reorientation of the cortical microtubule in neighboring cells caused by the rapid expansion of the initiating trichomes (Hervieux et al., 2017) could facilitate this subsequent reorientation of growth. Alternatively, slow-growing trichomes could pull on the surrounding cells, which would explain why the growth of the neighbors in the radial direction is increased (Fig. 7I), but their growth in the other direction is similar to that of the rest of the epidermis (Fig. 7J). Thus, in addition to the previous hypothesis that local mechanical shielding is involved in filtering the fast growth of the initiating trichomes (Hervieux et al., 2017), our data suggest that an opposite mechanism could operate during subsequent trichome expansion, whereby the slow growth of the trichome base is compensated for by the increased directional growth of adjacent cells.

Taken together, this work provides strong evidence that local growth variability, achieved through cell type-specific growth patterns and accompanying local buffering, is a tightly regulated inherent growth module rather than an exclusively stochastic process. It enables functional complexity (i.e. many cell types) to be established across a tissue while maintaining smooth and robust organogenesis through the integration of individual and collective cell behaviors. Our findings may thus serve as a template for further studies of complex structure establishment in multicellular organisms.

A. thaliana Columbia-0 (Col-0) was used as the wild-type ecotype in our study. The pUBQ10::myr:YFP line used for time-lapse experiments was described previously (Willis et al., 2016). The speechless mutant in a Col-0 background carrying the fluorescent plasma membrane marker was described previously (Fox et al., 2018; MacAlister et al., 2007). Seeds were stratified for 2 days at 4°C in the dark to induce synchronous germination. Plants were grown on soil in a growth chamber at 22±1°C under long-day conditions (16 h of illumination, 95 µmol m⁻² s⁻¹) and 60-70% relative humidity.

Two-day-old plants were dissected with fine tweezers to remove one cotyledon and expose the initiating primordia of the first true leaves. Plants with undamaged leaf primordium, roots and hypocotyls were used for growth tracking. They were transferred onto Ø60 mm Petri dishes filled with ½ MS medium supplemented with 1.5% agar, 1% sucrose and 0.1% plant protective medium (Plant Cell Technology). Hypocotyl and the root were carefully placed and immobilized into the cut created in the medium. Leaves were imaged every 12 h for up to 1 week. If needed, plants were repositioned before the confocal scan such that the same side of the sample was always observed. Owing to the unfolding of the leaf, the last imaging was often performed on a glass slide after dissecting the leaf with an injection needle and flattening it under a coverslip. To avoid the samples being crushed, two coverslip spacers were placed at both extremities of a microscope slide, and the sample was positioned in between, under a third coverslip. Confocal imaging was performed with a Zeiss LSM800 upright confocal microscope equipped with a long working-distance water-immersion 40×objective (1 NA, Apochromat). Excitation was performed using a diode laser at 488 nm for YFP, and the signal was collected between 500 and 600 nm. Confocal stacks were acquired at 512×512 resolution with 1 µm distance in the z-dimension. For samples that were larger than the objective field of view, multiple overlapping stacks were acquired. Between consecutive imaging, samples were transferred to a growth chamber and cultured in vitro under long-day conditions (16 h illumination, 80 µmol m⁻² s⁻¹). Imaging of sepals and anthers was previously described (Hervieux et al., 2016; Silveira et al., 2022). For spch mutant, previously acquired time-lapse data were used (Fox et al., 2018). For imaging of petals and carpels, inflorescences were removed from 5-week-old soil-grown plants and flower buds at stages 11 and 12, respectively (Smyth et al., 1990), were dissected under a stereomicroscope. Using fine tweezers, sepals were removed to expose the petal or the carpel. Dissected flowers were imaged as previously described (Silveira et al., 2022).

Confocal images were processed with the 3D image analysis software MorphoGraphX (Barbier de Reuille et al., 2015; Strauss et al., 2022). For samples bigger than the objective field of view, images were stitched by manually superimposing individual stacks. For the adaxial leaf epidermis stack, parts of the trichomes protruding from the surface were digitally removed before surface detection. The surface of each organ was detected using the ‘Edge detect’ tool with a threshold of 6000-18,000, followed by the ‘Edge detect angle’ tool with a threshold of 3000-5000. The surface was extracted by creating an initial mesh of 5-9 µm cube size. This mesh was subdivided three times and smoothed to eliminate local irregularities. Additional smoothing was performed on the mesh at the positions of stomata, as the YFP signal in these cells was intense, leading to inaccurate detection of the surface. The signal of the cell membrane of the epidermal layer (∼2-4 µm) was then projected to visualize cell boundaries. The ‘Auto segmentation’ tool in MorphoGraphX was used to segment cells at early time points. At later stages, after stomata differentiation, manual segmentation was performed for entire samples. Parents were manually attributed to find corresponding cells at two consecutive time points. The ‘Check correspondence’ tool was then used to identify the parenting errors, which were manually corrected.

Cell lineages over multiple days were computed as described previously (Kierzkowski et al., 2019). Shape analysis was performed using the ‘Lobeyness’ plugin in MorphoGraphX (Sapala et al., 2018). The blade, petiole/midrib and margin regions were identified by manually labeling the cells at the latest time point based on their position, growth rates, proliferation rates and shapes. The margin region included giant cells located at the leaf edge that exhibited a highly anisotropic growth (Fig. S3A-C). The petiole/midrib region was located in the median part of the leaf and was composed of elongated cells that grow relatively slowly along the proximal-distal axis of the leaf (Fig. S3A-C). The leaf blade was located between the margin and the petiole/midrib and was composed of cells of variable sizes, including highly lobed pavement cells that grew isotropically (Fig. S3A-C). The origin of each region was then traced back in time using reverse lineage tracking. Cell types were attributed manually at each time point, based on their shape, location within the organ, and cell lineages. Cells that were not dividing were classified as ‘marginal cells’ if they were located at the leaf margin; ‘stomata’ if they were bean-shaped and surrounding the developing stomatal pore; or pavement cells if they were located outside the leaf margin. The guard cells for each stoma were merged together to avoid segmentation errors and were thus considered as a single cell for all statistical analysis. Dividing cells were classified either as meristematic, if their divisions were symmetric, or meristemoid, if their divisions were asymmetric and led to the development of stomata.

Cell area, area extension, growth anisotropy, and cell proliferation were calculated for each cell as described previously (Barbier de Reuille et al., 2015). Averaged growth was calculated using the ‘Heatmap Stats’ plugin from MorphoGraphX. The average was based on the growth rate of each cell and its adjacent neighbors, weighted by cell area. Distance from the tip was calculated by manually selecting the tip of the leaf or midrib and using the ‘Cell distance’ plugin in MorphoGraphX, choosing the ‘Euclidean’ parameter to get the shortest distance between cells centers following the organ geometry (Strauss et al., 2022). Heat maps between two time points were displayed on the second time point. All figures were assembled using Adobe Photoshop software.

Morphs (Fig. 7E) were computed in MorphoGraphX 2.0 between two carpel meshes acquired with 72 h interval (Strauss et al., 2022). The morphed mesh is a prediction of how the cells at 0 h would deform to achieve the shapes observed at 72 h, based on a continuous mapping of the displacement of landmarks. First, only the cell junctions were taken as landmarks, so that cell outlines were stretched based on a global deformation field of the entire cell (Fig. 7E). To compute local deformations all around the stomata (Fig. 7F), additional landmarks were manually added along stomata outlines before computing the morph. In order to avoid morphing artifacts caused by cell local curvature, the morph was calculated based on cell outlines and junctions only. The deformation gradient obtained the morph process based on vertices and was visualized using the ‘Product Stretches’ function in MorphoGraphX 2.0 (Strauss et al., 2022).

Analysis of growth by cell type was performed using a Python script. Cell lineages were acquired by correspondence between parents' labels over the whole experiment duration. Using labels at each time point, growth rate, region and cell type were extracted for all lineages. Data were aligned according to DRD with 0 DRD as the moment when a specific cell type appears in the lineage. Local growth variability was computed using Python script as the difference between the absolute and the averaged growth for each cell, normalized by the mean cellular growth in the organ at a specific time point.

Data were extracted and plotted using Python scripts. For boxplots, the lines indicate the median and dashed lines represent the mean, box limits represent the first and third quartiles and whiskers represent 90% of the values. The violin plot outlines illustrate the kernel probability density. The width of the shaded area represents the distribution of the data; violin plots contain 90% of the values. For bar plots, bars correspond to the relative proportion of each value per time point.

Statistical analysis for comparison between leaf blade and midrib/petiole regions was performed using the Kolmogorov–Smirnov test. For comparison between cell types, the Kruskal–Wallis test was used to assess whether at least one mean distribution was different. If the test was significant, all pairwise comparisons were carried out with the Dwass–Steel–Critchlow–Fligner post-hoc test.

We thank Viraj Alimchandani, Kristoffer Jonsson and David Morse for the critical reading of the manuscript and Andrea Gomez Felipe, Jerome Burkiewicz for help with data analysis. We thank Enrico Coen, Samantha Fox and Karen Lee from John Innes Center for providing confocal time-lapse data for spch mutant leaves.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200783.