Increased gene expression variability hinders the formation of regional mechanical conflicts leading to reduced organ shape robustness

Significance

There is more and more evidence that transcriptional noise exists in both unicellular and multicellular systems, but the contribution of transcriptional noise to organ robustness is poorly understood. Using the model plant Arabidopsis thaliana, we demonstrate that mutation in VIP3, a component of the polymerase-associated factor 1 complex (Paf1C), leads to increased gene expression variability and concomitantly, increased shape variability in sepals, a highly reproducible organ. Through quantitative imaging of the sepal, we find that local growth heterogeneity is increased and the stereotypical growth pattern is disrupted in the vip3 mutant. We propose that increased gene expression variability leads to increased local growth heterogeneity, which fuels local conflicts instead of regional conflicts, and eventually results in more variable organ shapes.

Abstract

To relate gene networks and organ shape, one needs to address two wicked problems: i) Gene expression is often variable locally, and shape is reproducible globally; ii) gene expression can have cascading effects on tissue mechanics, with possibly counterintuitive consequences for the final organ shape. Here, we address such wicked problems, taking advantage of simpler plant organ development where shape only emerges from cell division and elongation. We confirm that mutation in VERNALIZATION INDEPENDENCE 3 (VIP3), a subunit of the conserved polymerase–associated factor 1 complex (Paf1C), increases gene expression variability in Arabidopsis. Then, we focused on the Arabidopsis sepal, which exhibits a reproducible shape and stereotypical regional growth patterns. In vip3 sepals, we measured higher growth heterogeneity between adjacent cells. This even culminated in the presence of negatively growing cells in specific growth conditions. Interestingly, such increased local noise interfered with the stereotypical regional pattern of growth. We previously showed that regional differential growth at the wild-type sepal tip triggers a mechanical conflict, to which cells resist by reinforcing their walls, leading to growth arrest. In vip3, the disturbed regional growth pattern delayed organ growth arrest and increased final organ shape variability. Altogether, we propose that gene expression variability is managed by Paf1C to ensure organ robustness by building up mechanical conflicts at the regional scale, instead of the local scale.

Stochastic gene expression is increasingly reported in living organisms, from bacteria growing in the same Petri dish (1) to multicellular organisms, such as Drosophila (2) or Arabidopsis (3, 4). The instructive role of such noise is increasingly studied in cell biology, notably in relation to developmental robustness (5, 6). For instance, transcriptional noise can be exploited by cells for fate exploration and commitment (7–9). Notch-dependent cell fate in Drosophila is progressively acquired over time, largely through initially stochastic processes (10). Similar fate acquisition from initially fluctuating expression levels has been reported in Arabidopsis (3). Weak signals can also become unmasked, thanks to stochastic resonance, i.e., adding noise makes them pass the threshold of detection (11). At a more integrated level in both space and time, fluctuation in cellular growth rate and orientation allows its averaging over time, and this contributes to organ shape reproducibility in Arabidopsis (12).

Whereas noise can contribute to patterning and morphogenesis in a bottom–up way, it is also classically thought to be channeled by global cues, notably to ensure shape robustness. For instance, precise timing of organ initiation reduces organ shape variability in Arabidopsis (13). Morphogen gradients synchronize heterogeneous cell populations, and can even coordinate growth arrest through dilution as the organ becomes larger (14). Mechanical stress fields have been proposed to provide additional synchronizing cues (15–17). In particular, differential growth between domains has been proposed to trigger mechanical conflicts, ultimately leading to growth arrest in the Drosophila wing disc (18) and Arabidopsis sepal (19).

Beyond growth and development, noise management ultimately applies to transcriptional control. The causes of noise in gene expression are numerous, from chromatin modifications (20) to DNA topology (1) or ribosomal activity (21). More fundamentally, transcriptional noise is a product of the gene network topology (22), also consistent with the idea that “the cell is not a machine” (5). Interestingly, a genetic screen considering transcriptional noise as a quantitative trait identified the progression of the transcriptional elongation phase by RNA polymerase II as a leading cause of noise in gene expression in yeast. This notably involves the polymerase-associated factor 1 complex (Paf1C) (23). Whether this also applies to multicellular systems is unknown, and it is the focus of the present study.

In Eukaryotes, Paf1C is recruited to active Pol II elongation machinery. Through different mechanisms, Paf1C regulates multiple aspects of transcription including elongation and termination, as well as histone modification and chromatin structure (24, 25). The roles of Paf1C are systemic. For example, the complex regulates transcriptional elongation of nearly all genes in yeast (26, 27). Yeast and mammalian Paf1Cs consist of five and six subunits, respectively, and their six homologs in Arabidopsis have been identified (28–30) (SI Appendix, Fig. S1A). Mutants in the Arabidopsis Paf1C subunits were first uncovered, thanks to their early flowering phenotype (30). Careful analysis of the phenotypes of the mutants—vip3 (vernalization independence 3) being the best documented—revealed multiple developmental defects as well as increased phenotypic variability, even for phyllotaxis, one of the most stereotypical patterns in plant development (31–33) (SI Appendix, Fig. S1B).

Here, we provide evidence of increased gene expression variability in vip3 and explore the impact on organ shape variability. By using the model plant Arabidopsis, we take advantage of the absence of cell movement in plant organs, as well as the stereotypical growth and shape of sepals, the outermost floral organs (34). We find that increased gene expression variability not only correlates with increased growth heterogeneity, but it also disrupts regional patterns of growth, leading to increased organ shape variability.

Results

Mutation in VIP3 Increases the Variability in Linker Histone H1 Expression between Sister Guard Cells.

To see whether mutation in VIP3 could lead to increased gene expression variability and could serve as a mutant background to explore the roles of transcriptional noise in organ shape robustness, we first quantify gene expression variability in sister guard cells (described here), and in more complex tissue contexts (described next).

Stomata are epidermal valves formed by a pair of specialized guard cells, which originate from a common guard mother cell following a symmetric division (Fig. 1A). Because of their small sizes and proximity, they likely experience very similar internal and external cues. Given their common developmental origin, close position, and synchronous differentiation, guard cells offer an excellent system to study gene expression variability.

Fig. 1.

Gene expression variability in guard cells and floral buds. (A) Stomatal development in Arabidopsis. Sister guard cells are formed through a symmetrical cell division. Using a fluorescent protein targeted to the nucleus, the difference in gene expression (nuclear signal) between the two guard cells can be calculated. Adapted from ref. 35. (B) A large area of the cotyledon in WT and vip3-1 showing the expression of pH1.2::H1.2-eGFP in guard cells. (Scale bar, 50 µm.) (C) Close-up, representative confocal images of stomatal guard cells and expression of pH1.2::H1.2-eGFP in WT and vip3-1 (red signal: PI staining). (Scale bar, 20 µm.) (D) Difference (in percentage) in fluorescent signal between two sister guard cells. H1.1 (pH1.1::H1.1-eGFP), H1.2 (pH1.2::H1.2-eGFP), H1.3 (pH1.3::H1.3-eGFP), AtML1 (pAtML1::H2B-3xGFP::tRBCS). (E) Close-up, representative confocal images of stomatal guard cells and expression of pH1.2::H1.2-eGFP in WT and vip3-1 sepals (Left), and difference (in percentage) in H1.2 fluorescent signal between two sister guard cells (Right). n_WT(H1.2) = 73, n_vip3-1(H1.2) = 79, Kruskal–Wallis test. (Scale bars, 10 µm.) (F) A typical Arabidopsis inflorescence with the shoot apical meristem (SAM) in the center and several flower buds of different stages (membrane marker pUBQ10:Lti6b-TdTomato). Young flower buds are used to measure local variability in gene expression. Only cells in the center of the floral bud (white circle) are taken into account. Local variability in gene expression is measured as the coefficient of variation (CoV% = SD/mean*100%) in nuclear signal intensity of a group of cells consisting of a cell and all of its contacting neighbors. [Scale bars, 10 µm (Left) and 20 µm (Right).] (G) Representative image of pSTM::CFP-N7 expression in stage 3 floral buds (Left) and coefficient of variation (%) of signal intensity between neighboring cells (Right) for the WT and vip3-1 samples. n_WT = 224 cells, n_vip3-1 = 295 cells, P = 2.15e-55, Kruskal–Wallis test. (Scale bars, 20 µm.)

We chose to analyze the expression of the whole H1 linker histone gene family: H1.1 (pH1.1::H1.1-eGFP), H1.2 (pH1.2::H1.2-eGFP), and H1.3 (pH1.3::H1.3-eGFP). H1 genes are expressed in stomatal guard cells and are required for stomata functioning (36) (Fig. 1B). We reasoned that this family was small enough to get an overview of their expression variability between WT and mutant, while being functionally relevant for stomata. We also included one of the regulators of epidermal cell identity AtML1 (pAtML1::H2B-3xGFP::tRBCS), which is expressed in all epidermal cells. The use of GFP with a nuclear localization signal for AtML1 or translational fusion for histone markers allowed us to compare expression levels between two sister guard cells using the total signal in nuclei as a proxy (Materials and Methods). Here, we discuss the differences in the median as the parameter is less sensitive to outliers.

Since the two vip3 mutant alleles are well established and exhibit similar phenotypes [SI Appendix, Fig. S1 C and D, (31–33)], most of the work in this study was performed on vip3-1. Fig. 1C shows representative images of stomatal guard cells and H1.2 expression in WT and vip3-1. Whereas the signal of H1.1 and H1.2 between two sister guard cells differed by 15% and 28%, respectively, in the WT, the figures for vip3-1 were 34% and 63%, respectively, i.e., double those of the WT [n_WT(H1.1) = 30, n_vip3-1(H1.1)=32, n_WT(H1.2) = 80, n_vip3-1(H1.2) = 80, p_H1.1 = 0.019, p_H1.2 = 5.75e⁻⁰⁶, Kruskal–Wallis test] (Fig. 1D). In the normal growth condition, H1.3 was only expressed in a few guard cells in both backgrounds. We therefore placed 8-d-old seedlings in the dark for 24 h to induce H1.3 expression as noted in ref. 36. In these conditions, the variability of H1.3 expression in vip3-1 was again roughly twice as high as in WT [Δ_WT(H1.3) = 45%, Δ_vip3-1(H1.3) =109%, n_WT(H1.3) = 80, n_vip3-1(H1.3) = 84, P = 1.61e⁻⁰⁵, Kruskal–Wallis test] (Fig. 1D). Note that we did not observe any significant difference in AtML1 expression variability between sister guard cells in the WT and vip3-1 [Δ_WT(AtML1) = 13.4% vs. Δ_vip3-1(AtML1) = 13.1%, P = 0.291, Kruskal–Wallis test] (Fig. 1D). This is consistent with the finding that in yeast Paf1C binds to RNA polymerase II at different degrees, depending on genes (26). The rather stable AtML1 expression level is also consistent with the finding that VIP3 is not affecting all pathways with the same strength in Arabidopsis (32), while also providing a baseline for our analysis.

The analyses above were done on cotyledons. Since later on we work mostly with sepals, we tested whether the increased gene expression variability is also observed in the sepal. Using pH1.2::H1.2-eGFP and the same method as described above, we found that H1.2 expression was significantly more variable in the pair of sister guard cells in vip3-1 sepals compared to those in the WT [Δ_WT(H1.2) = 28%, Δ_vip3-1(H1.3) = 48%, n_WT(H1.2) = 73, n_vip3-1(H1.2) = 79, P = 0.0256, Kruskal–Wallis test] (Fig. 1E).

In short, using sister guard cells as stereotypical and synchronous cells in both cotyledons and sepals, we show that H1 expression is locally more variable in vip3-1, when compared to the WT.

Mutation in VIP3 Increases the Local Variability in Gene Expression between Adjacent Cells in Floral and Inflorescence Meristems.

Guard cells are an ideal system to analyze gene expression variability because they have a comparable history. Yet, results may not be applicable to other tissues where adjacent cells undergo stochastic cell divisions. To challenge our finding, we thus assessed gene expression variability in other developmental contexts, including floral and inflorescence meristems (Fig. 1F). Here, we analyzed expression of SHOOT MERISTEMLESS (STM) (pSTM::CFP-N7), AGAMOUS (pAG::AG-2xVenus), and AtML1 (pAtML1::H2B-3xGFP::tRBCS).

Because there are many cells in floral meristems and SAMs (shoot apical meristems) (as opposed to only a pair of guard cells), we calculated the coefficient of variation [CoV(%), defined as the SD divided by the mean in percentage] in gene expression within a group of adjacent cells (Materials and methods) (Fig. 1F, last diagram).

STM regulates meristem functions, and its expression defines meristematic tissues (37). In the epidermis (L1 layer) of floral buds of stages 2 to 3, STM expression is strong in the center but much weaker in emerging sepals. While nuclei in WT floral buds showed a homogenous signal, it was much less so in vip3-1 with nuclei having strong and weak signals next to each other (Fig. 1 G, Left). We quantified pSTM::CFP-N7 signal intensity and confirmed that the STM signal varied just slightly between cells in the WT, with the median CoV of local signal intensity at 7.3% (5 samples, 224 cells). However, signal variability was more than double in the mutant, with the average CoV at 18.2% (7 samples, 295 cells, P = 2.15e⁻⁵⁵, Kruskal–Wallis test). The difference across samples in the mutant was also more prominent than in the WT (Fig. 1 G, Right). Note that we found a similar trend when considering the L2 layer (median CoV of local signal intensity in the WT = 16.5%, in vip3-1 = 21.2%, P = 3.74e⁻¹¹ Kruskal–Wallis test) (SI Appendix, Fig. S1E). Yet, because the baseline in the WT was much higher in the L2 than in the L1 (median CoV of local signal intensity in the epidermis = 7.3%, in the layer 2 = 16.5%), arguably because of the presence of more neighbors, we focused the analysis on the epidermis, which is also generally thought to act as a growth-limiting layer in young tissues (38).

Similar trends were also observed for the key regulator of reproductive organ development AG in floral meristems (SI Appendix, Fig. S1F), and for AtML1 in floral and SAMs (SI Appendix, Fig. S1 G and H, respectively).

Increased Gene Expression Variability in vip3 Is Not Related to Average Gene Expression Levels.

Consistent with the law of large numbers, global gene expression variability should increase if global gene expression levels decreased. To check whether the observed increase in gene expression variability in the vip3-1 mutant is not simply caused by a general reduction in global gene expression, we reanalyzed our transcriptomic data previously obtained on WT and vip3-1 inflorescences, which contained SAMs and flower meristems (32). In vip3-1, we found that expression levels of 1,644 genes were significantly increased and 2,069 genes significantly decreased when compared to the WT (32). More specifically, among the genes selected for the gene expression variability analyses above (i.e., H1.1, H1.2, H1.3, STM, AG, and ATML1), STM and ATML1 showed an increase in average gene expression in vip3-1, H1.3 a decrease in vip3-1, while the average expression levels of H1.1 and H1.2 were not significantly different between WT and vip3-1. Since we detected an increased variability in gene expression for most of those genes (except ATML1 in guard cells), the observed increase in gene expression variability in the vip3-1 mutant is unlikely to be due to a general reduction in gene expression.

Evidence for Increased Molecular Noise in vip3.

If gene expression is significantly more variable in vip3, other molecular factors should also be noisier in that background. To check this, we used a chemical stain specific for the superoxide radical (O₂⁻), a major ROS molecule. We found that vip3-1 sepals exhibit a patchy ROS pattern (O₂⁻) in very young buds, in contrast to WT floral buds, which showed no staining (Fig. 2 A, Left). Such a patchy pattern was even more striking in vip3 carpels (Fig. 2 A, Right), which was observed in both vip3 alleles (SI Appendix, Fig. S2A).

Fig. 2.

Increased molecular noise (ROS) and organ variability in vip3-1. (A) NBT staining for superoxide in WT and vip3-1 inflorescences. Superoxide production in the mutant is patchy in young flower buds as well as in the carpels (arrows). (Scale bars, 0.1 mm.) (B) Top-view of WT and vip3-1 inflorescences showing the early opening of floral buds and outward-curing of the sepals in the mutant. (Scale bar, 1 mm.) (C) Close-up images of WT and vip3-1 floral buds of three different stages. (Scale bar, 1 mm.) (D) Confocal image of a young WT floral bud to indicate the position of the abaxial sepal (i.e., the sepal farther away from the SAM). Abaxial sepals are used throughout the present study. (Scale bar, 20 µm.) (E) Quantification of WT and vip3-1 abaxial sepal area (mm²). vip3-1 abaxial sepals are bigger and more variable in size. (F) Plots showing the contours of WT and vip3-1 abaxial sepals. The contours are normalized to the area. The red outline is the average shape. vip3-1 sepals are more variable in shape compared to the WT. (G) Quantification of Sepal shape variability S₂ (squared deviation of sepal outlines) confirming the higher variability in shape in vip3-1.

Because of its ubiquitous role in development, we also checked whether auxin signaling is more variable in the mutant as well, using the DR5rev::3XVENUS-N7 auxin signaling reporter. Overall signal intensity was low and highly variable in the WT, hindering the identification of a clear-cut difference with vip3 (SI Appendix, Fig. S2B). Yet, in stage-14 flowers, the DR5rev::3XVENUS-N7 signal was absent in the first few layers of WT carpels, but it exhibited a patchy pattern in the vip3-1 carpels (SI Appendix, Fig. S2C). These data echo what we observed with ROS staining in the carpels, and further support our conclusion on increased molecular variability in vip3.

In summary, working with various reporter lines and staining in different tissue contexts (guard cells in cotyledons and sepals, group of adjacent cells in floral and inflorescence meristems), we demonstrated that mutation in VIP3 leads to increased gene expression variability. Next, we investigated whether this may contribute to the organ shape robustness in the mutant.

VIP3 Loss of Function Increases Variability in Abaxial Sepal Shape and Size.

Through quantitative approaches, vip3 mutants have already been shown to exhibit variable developmental features, such as in floral termination (32) and phyllotaxis at the SAM (31). Yet, because such complex phenotypes involve coordination between several organs, it would be difficult to link molecular noise and phenotypic variability. Therefore, we focused on a simpler organ, the sepal. Each Arabidopsis flower produces four sepals, and each sepal exhibits a remarkably reproducible shape during its development. Thus, it is an ideal system to relate gene expression variability and global shape robustness (34).

We noticed that sepals in vip3 mutants usually curve outward, and their shapes also appeared more variable than those in the WT (Fig. 2 B and C and SI Appendix, Fig. S1 C and D). In theory, sepal shape variability may be an indirect consequence of organ initiation defects in flowers (32). To avoid such effects and to focus on the intrinsic control of sepal morphogenesis, we only selected flowers with four sepals for all analyses. We also focused on abaxial sepals, which are the first to emerge from floral bud meristems in both genotypes (Fig. 2D). To quantify the variability in vip3-1 sepal shape, we employed the SepalContour tool (12). This tool allows us to extract several features of sepals, including sepal area, sepal contour, and other shape descriptors such as length and width.

Regarding sepal size, WT flowers produced abaxial sepals of remarkably comparable area (Fig. 2E; mean ± SD 1.39 ± 0.11 mm²; n_WT = 40 sepals). In contrast, the vip3-1 mutant produced abaxial sepals of highly variable areas (Fig. 2E; mean ± SD 1.57 ± 0.35 mm²; n_vip3-1 = 40 sepals). We found the CoV for the vip3-1 sepal area to be 3 times larger than that of the WT (CoV_WT = 7.8%, CoV_vip3-1 = 22.2%).

We next analyzed sepal shape variability in vip3 based on the sepal contour. To account for size variability between sepals, the contours of all sepals were extracted and normalized by their area to make the analysis independent of organ size and then plotted (Fig. 2F). Quantification of variability in sepal contour indicated that the median shape variability S₂ for vip3-1 abaxial sepals was about 4 times higher than that of WT (Fig. 2G; median ± SE 0.98 ± 0.14 10⁻³ for WT vs. 4.1 ± 1.37 10⁻³ for vip3-1, P = 7.05e⁻⁸, Kruskal–Wallis test). A similar increase in shape variability was also observed in vip3-2 abaxial sepals (SI Appendix, Fig. S2 D and E).

Altogether, these analyses indicate that VIP3 loss of function leads to increased variability in sepal shape and size. Next, we explored the nexus between increased gene expression variability and variable final organ shapes.

Growth in the vip3-1 Abaxial Sepal Is Locally More Heterogeneous.

In the simplest scenario, which we could call “error propagation”, variability in gene expression would increase cell growth variability, leading to final organ shape variability. To test that scenario, we next quantify cell growth in developing abaxial sepals.

Mature abaxial sepals in vip3-1 are more variable in shape and size, yet abaxial primordia in both backgrounds initiate with comparable size variability (SI Appendix, Fig. S2 F and G). This further confirms that the variable abaxial sepal sizes and shapes in vip3-1 relate to sepal growth, and not to defects in floral patterning or early sepal initiation (at least when selecting vip3-1 flowers with four sepals).

To check how growth contributes to variable sepals in the vip3-1 mutant, we tracked cell growth of WT and vip3-1 abaxial sepals over a period of 7 d. Abaxial sepals expressing the membrane marker pUBQ10:Lti6b-TdTomato were imaged every 24 h using a confocal microscope. At the first time point, they were at stage 4 according to (39). Here, we focus on local growth heterogeneity between neighboring cells (Fig. 3). The representative growth map for WT and vip3-1 sepals over the period is shown in the later part (Fig. 4). We describe the larger growth kinetics and pattern in more details there.

Fig. 3.

Growth analyses of vip3-1 organs reveal increased growth heterogeneity. (A and B) Local variation of growth rates (%) of individual WT and vip3-1 sepal samples over 7 d. Note that WT samples show fluctuations around a stable level, but vip3-1 does not. The error bars are SD. (C) Representative heat map showing local variation of growth rates in WT and vip3-1 sepals on Days 6 to 7 where the variation differs the most between two genotypes. (Scale bar, 100 µm.) (D) Local variation of growth rates in SAMs in WT and vip3-1. n_WT = 4, n_vip3-1 = 5 SAMs, Kruskal–Wallis test. (E) Difference (%) in cell size between two sister guard cells of the same stomata. n_WT = 171, n_vip3-1 = 169 pairs of guard cells, Kruskal–Wallis test. (F and G) Changes in cell surface area of small cells in WT and vip3-1 over time (arrows). (Scale bars, 20 µm.) (H) Growth kinetics (in cell surface area) of small cells in WT and vip3-1. Note the reduction in surface area in vip3-1 cells, which was not seen in the WT. Each line represents a cell. n_WT = 40 cells, n_vip3-1 = 37 cells. Reduction in cell surface area in the mutant was observed in three biological repeats. (I) 3D reconstruction of some selected cells in G on Day 2 and Day 3, top view. The same cell on Day 2 and Day 3 has the same color. Note the reduction in surface as viewed from the top of the cells indicated by the arrows (the same cells in J); the brown cell on Day 2 could not be detected anymore on Day 3. (J) The same cells in I, bottom view. Note that the brown cell has disappeared in this view as well. (K) Heat map showing changes in cell volume from Day 2 to Day 3, projected to Day 2. The scale is from 0 to 2.5 (times). Volume change < 1 means a volume contraction, which happens to the cells indicated by the arrows. The two cells indicated by the lower arrow are counted as one for this measurement. (Scale bar for I–K, 10 µm.) See also SI Appendix, Fig. S4 G–I. (L) Orthogonal views of the squeezed cells in vip3-1 along the red lines shown in panel G. Note the reduction or disappearance of the space between anticlinal walls.

Fig. 4.

Growth analysis of vip3-1 sepals over 7 d reveals a failure to arrest growth at the later stage. (A and B) Heat map of cell growth rate (in area) over 24-h intervals of WT (A) and vip3-1 (B) sepals, projected on the second time point (e.g., growth rate of a sepal from Day 1 to Day 2 is projected on the Day 2 sepal). The arrow indicates the main axis of the sepal, pointing to the tip. The same scale and magnification are used. (C and D) Cell growth rates over 24-h intervals of individual WT and vip3-1 sepals. While WT samples exhibit a declining growth rate over time, vip3-1 sepals maintain and even increase their growth at late stages. For averaged trends, see SI Appendix, Fig. S5A. The error bars are SD.

Local growth heterogeneity here is defined as the CoV in growth rates of cells within a group; the group is formed by a cell and all other neighbors contacting it (40). From the sepals in the time-series experiment just described, we used MorphoGraphX (MGX) to find all the groups and calculate their local growth heterogeneity, expressed as CoV (%) of local growth rates. Data for individual WT and vip3-1 samples are shown in Fig. 3 A and B, respectively.

We found that during the period, local growth heterogeneity in WT sepals, though experiencing a slow but steady downward trend, was remarkably stable: The average CoV of local growth rates for each sample fluctuate around 14.5% during the 7-d period (Fig. 3A). Individual samples showed slight fluctuation in local growth rates between days. This is particularly notable, knowing that the average cell growth rate in the WT steadily reduces from ~176 to ~49% within the same period (Figs. 3A and 4).

In vip3-1 sepals, we observed more fluctuations in local growth heterogeneity between time points (Fig. 3B). Like in the WT sepals, local growth heterogeneity in the mutant also experienced a downward trend in the first 4 d, then quickly increased to ~17.4% in the last 3 d when the mutant sepal shape progressively deviated from that of the WT (Fig. 3B). Such increased variability was also observed when considering the growth history of individual samples: Kinetics of local growth heterogeneity in the vip3-1 sepals were more erratic than those in the WT (Fig. 3 A and B).

We then mapped local growth heterogeneity onto the sepal meshes, with a focus on the last time point where the mutant shows significantly higher growth heterogeneity (CoV = 12.7% for WT vs. 17.4% for vip3-1, P = 2.17e⁻¹¹³, Kruskal–Wallis test). Fig. 3C illustrates how local growth heterogeneity is more intense and widespread in vip3-1, when compared to the WT.

Stomatal guard cells with their distinct mode of development are also a source of growth heterogeneity (41). To check the contribution of stomata in local growth heterogeneity observed above, we did the same analyses for sepals from Day 4 to Day 7 (when stomata appeared) without taking into account stomata. We found that vip3-1 sepals still have significantly higher local growth heterogeneity at the last time point (CoV = 11.0% for WT vs. 14.9% for vip3-1, P = 4.87e⁻¹⁰⁸, Kruskal–Wallis test) (SI Appendix, Fig. S3 A and B).

Although we worked mostly on sepals, we wanted to check whether the increase in local growth heterogeneity also happened in other developmental contexts. We performed the same analyses on SAMs growing over a 24-h period and found that cells in vip3-1 SAMs also displayed higher local variation of growth rates (Fig. 3D). Regarding stomatal guard cells, we found that the difference in size between two sister guard cells is significantly higher in the mutant compared with the WT counterparts (median difference = 7.3% for WT, = 12.5% for vip3-1, P = 3.2e⁻⁶ Kruskal–Wallis test, n_WT = 171, n_vip3-1 = 169 pairs) (Fig. 3E). Although this is a snapshot of mature stomata instead of a growth kinetics, it suggests that a growth difference between two sister guard cells in the mutant is higher compared to those in the WT.

To conclude, these analyses reveal an increased heterogeneity in growth rates between adjacent cells in vip3-1, consistent with the error propagation scenario.

High Growth Heterogeneity in vip3 Can Be Further Increased in the Absence of Microtubules.

The observed increase in growth heterogeneity in vip3 is rather modest. In fact, the most striking result is rather the very stable growth heterogeneity level in the WT, by comparison. We reasoned that growth heterogeneity in vip3 might be partly compensated. For example, it is well established that defective cell division can be compensated by promoting cell growth rate (42), or that defective cellulose synthesis can be compensated by the formation of thicker walls with more matrix (43). To check this, we altered the mechanical feedback operating in plant cells by using oryzalin, a microtubule depolymerizing drug. In these conditions, cells cannot guide the deposition of cellulose to resist tensile stress because cortical microtubules are absent, and they cannot mechanically reinforce their tissue through oriented cell divisions since cell division does not occur anymore (44, 45). Because compensatory mechanisms are impaired in such conditions, we reasoned that this could increase growth heterogeneity in vip3.

WT and vip3-1 sepals were treated with 20 µg/mL oryzalin for 3 h every 24 h, and samples were imaged right before each oryzalin treatment (see SI Appendix, Fig. S4A for the experiment plan). We analyzed the impact of depolymerizing microtubules on growth between adjacent cells on Day 2 to Day 3 where the impact of oryzalin is clear (SI Appendix, Fig. S4B). Oryzalin-treated sepals in both genotypes kept growing during the 3-d period, even though cell division was prohibited.

In this condition, vip3-1 sepals also display a higher local variation of growth rates (CoV_WT = 16.8% vs. CoV_vip3-1 = 20.8%, SI Appendix, Fig. S4 C–E). More interestingly, we observed some small cells (as viewed from the top, surface area < 40 µm²) in the WT that did not grow much (Fig. 3F, arrow) next to cells that grew a lot more, as previously shown in other contexts (45). Strikingly, this behavior was enhanced in vip3-1: We could detect smaller cells that were progressively contracted as their neighbors kept growing (Fig. 3G, arrows). More specifically, we detected 37 mutant cells showing reduction in surface area, while WT cells kept growing, as illustrated in Fig. 3H (n_WT = 40 for WT, n_vip3-1= 37). To check whether these 37 cells also experience reduction in volume, we performed three-dimensional (3D) reconstruction and volumetric growth measurement (arrows, Fig. 3 I–K). The analysis showed that 8 among 37 cells (i.e., ~22%) experienced a reduction in volume (SI Appendix, Fig. S4F). In three cases (beside these 37 cells), mutant cells even seemed to disappear. Orthogonal views of these disappearing cells (along the red lines in Fig. 3G) showed that the space between anticlinal walls of these cells reduced drastically (Fig. 3L and SI Appendix, Fig. S4 G and H, see Fig. 3 I–K for 3D segmentation, and SI Appendix, Fig. S4 I and J for another example). We never detected such a behavior in the WT. These results suggest that microtubule-assisted cell wall reinforcement can constrain local growth heterogeneity and that the potential for higher local growth variability observed in vip3 in normal conditions is underestimated. More importantly, these tests confirm that growth is indeed more heterogeneous in vip3.

Impaired Boundaries between Growth Domains in vip3.

So far, we have shown that increased local variability in gene expression correlates with increased local growth heterogeneity in vip3-1. Yet, this does not explain how this could translate into a variable organ shape. In particular, many regional compensatory mechanisms have been previously shown to channel organ development (46). To reveal the nexus between local growth heterogeneity and final organ shape variability, we next analyzed global growth patterns in WT and vip3-1 sepals.

The representative growth map for WT and vip3-1 sepals over the 7-d period is displayed in Fig. 4 A and B. The stereotypical growth pattern of WT sepals has already been reported (19) and was observed here again (Fig. 4A). First, we quantified average growth rates of WT and vip3-1 sepals. In the first 4 d (Days 1 to 4), the overall growth rate gradually decreased (from 176% on Day 1 to 2 to 119% on Day 3 to 4, i.e., 32% reduction), and it decreased faster afterward (Fig. 4C). Whereas all analyzed WT sepals displayed rather consistent growth kinetics, growth kinetics in vip3-1 sepals were more variable (Fig. 4 C and D). Compared to the WT, growth rates of vip3-1 sepals decreased more quickly in the first 4 d (from 134% on Days 1 to 2 to 79% on Days 3 to 4, 41% reduction). More surprisingly, we observed that sepal growth rates then stopped decreasing and eventually increased between Day 4 and Day 7 in the mutant. The contrasting behaviors in growth between WT and vip3-1 sepals are illustrated in SI Appendix, Fig. S5A, which displays the average sepal growth rates of the two genotypes. This observation departs from the simplest error propagation scenario and suggests that another mechanism must contribute to final organ shape variability in vip3-1.

To dissect this response, we analyzed the regional growth pattern along the sepal proximodistal axis. WT sepals exhibited the typical growth dynamics as already described in ref. 19 (Fig. 4A). In the first 3 d, sepals displayed a sharp growth gradient from the tip to the base, with the tip growing much faster than the rest of the sepal. From Day 3 onward, the growth rate at the tip plummeted and the maximum of the growth rate shifted to cells in the middle and later on at the base of sepals.

vip3-1 sepals displayed a comparable growth pattern in the first days compared to WT ones, with the tip growing faster than the rest of the sepals (Fig. 4B). However, the difference between the slow-growing area and fast-growing area was less prominent in the first 5 d (Days 1 to 5) when compared to the WT (Fig. 4 A and B and SI Appendix, Fig. S5 B and C). Note that growth kinetics of the mutant sepals was more variable between individuals too and especially in the first days of the experiment period (Fig. 4 C and D). For example, two mutant samples exhibited a strong reduction in growth rates from Day 1 to Day 4, two samples only exhibited a slight reduction comparable to that of the WT, while one sample instead exhibited an increase in growth rate during Days 2 to 4.

Interestingly, in the last 2 d (Days 6 to 7), cells at the tips of vip3-1 sepals reactivated their growth, while WT tips grew very slowly (Fig. 4 A and B and SI Appendix, Fig. S5D). All samples consistently exhibited the same trend, i.e., an increase in growth rates compared to the previous day (Fig. 4D), an increase in local growth heterogeneity (Fig. 3B), and the reactivation of growth at the tip (Fig. 4B, see also SI Appendix, Fig. S5E for a growth map of another mutant sepal).

Thus, in addition to higher local growth variability between adjacent cells, vip3-1 is also unable to generate a sharp growth arrest front at the sepal tip, as observed in the WT. The sharp difference between slow-growing tip and fast-growing middle area in the WT sepal (Days 3 to 5 in Fig. 4A) has previously been proposed to act as a mechanical signal for growth arrest in the sepal: The mechanical conflict arising from differential growth prescribes a transverse tensile stress pattern, to which cells resist by reinforcing their cell walls, ultimately leading to growth arrest at the tip (19) (Fig. 5A). We wondered whether this mechanism is impaired in vip3.

Fig. 5.

Defective regional conflict between sepal body and tip leads to variable organ shapes in vip3-1. (A) Differential growth between the center domain and the tip domain of the sepal prescribes a maximal transverse tensile stress direction, with which cortical microtubules align. The resulting cellulose-dependent mechanical anisotropy in the wall resists stress and contributes to growth arrest (reproduced from ref. 19). (B) Organization of cortical microtubules (CMTs) in WT sepals. The transverse CMT pattern at the tip matches the predicted supracellular maximal tensile stress direction shown in panel (D). (Scale bar, 50 µm.) (C) Organization of cortical microtubules (CMT) in vip3-1 sepals. In contrast to the WT, vip3-1 sepals do not display a consistent transverse CMT band at the tip. (Scale bar, 50 µm.) (D) The working model linking transcriptional noise and organ shape robustness. Paf1C-dependent transcription controls gene expression variability. Reduced local gene expression variability allows the formation of regions with consistent growth rates, and thus sharp boundaries between them. The associated supracellular mechanical conflict acts as a signal for growth arrest, hence contributing to reproducible shapes. In the vip3-1 mutant where Paf1C function is impaired, transcriptional defects lead to increased transcriptional noise. Increased local gene expression variability fuels local growth conflicts between adjacent cells. Consequently, growth conflicts between regions are diffused and growth arrest is delayed, contributing to variable shapes.

No Supracellular Cortical Microtubule Alignment at the Sepal Tip in vip3-1.

Our data are consistent with a scenario in which increased gene expression variability leads to growth heterogeneity (error propagation), which itself prevents the build-up of regional conflicts and thus, growth arrest at the tip.

The impaired growth arrest could suggest that the presence of weaker cell walls is sufficient to explain the variable sepal shapes in vip3. To check this, we analyzed sepal shapes in mutants affected in cell wall composition [xxt1xxt2 (47)], cell wall sensing [feronia (48)] and microtubule dynamics [katanin, with a slower response to stress (40), and nek6, with a faster response to stress (49)]. In contrast to vip3-1, these mutants did not exhibit a significant increase in sepal shape variability (SI Appendix, Fig. S6 A and B). This shows that global defects in the cell wall (properties and/or sensing) are not sufficient to increase final sepal shape variability, consistent with our scenario where regional conflicts play that role.

To test the specific role of regional conflicts in final sepal shape variability, we analyzed the organization of cortical microtubules, the directional behavior of which reflects the ability of the cell to reinforce its cell wall through guided cellulose deposition in the direction of maximal tensile stress. In the WT, at the stage where differential growth between the tip and the middle of the sepal generates a mechanical conflict (Fig. 5A), we could observe a band of transverse cortical microtubules at the boundary between the two domains, as previously shown (Fig. 5B, (19). In the vip3-1 mutant, cortical microtubules appeared denser and as thicker bundles. Note that this was also observed in hypocotyls and cotyledons (SI Appendix, Fig. S6 C–F). More interestingly, the consistent multicellular transverse cortical microtubule band was largely disrupted in the sepal (Fig. 5C).

Altogether, this provides a scenario in which local variability in gene expression in vip3-1 leads to increased local heterogeneity in cell growth. This impedes the formation of sharp boundaries between distinct regional growth domains in sepals. Because the resulting mechanical conflicts are diffused at the regional scale, growth arrest at the tip is impaired. In the end, the organ shape becomes more variable.

Discussion

In this study, we find that VIP3 in the Paf1 complex reduces gene expression variability in a multicellular context, and we provide evidence for the requirement of noise management to allow mechanical conflicts to develop between regions, instead of between adjacent cells. We propose that channeling growth heterogeneity to a regional scale fuels organ shape reproducibility (Fig. 5D).

The Level of Gene Expression Variability in Plants Is Under Genetic Control.

There are several techniques to quantify gene expression variability in single cells, but a fluorescent protein targeted to the nucleus is the most practical approach for multicellular systems like plants at the moment. Note that studies on transcriptional noise commonly employ the dual reporter strategy to separate intrinsic and extrinsic noise (e.g., to see whether noise of the two reporters covary or not) (1, 4, 20). Here, we used a single reporter approach, which informs us about the noisiness of transcription regardless of sources.

In a multicellular system, each cell has a different history, and thus the observed gene expression variability can rather reflect that history. We circumvented this problem by using sister guard cells, for which the origin and history are by definition very close. Working with different reporter lines, we confirmed that there is a certain level of noise in gene expression in WT Arabidopsis and that the noise levels vary depending on the genes (3, 4). This was also validated when considering other tissues and, in particular, SAMs and floral buds.

More importantly, we demonstrated that mutation in VIP3, a component of the Paf1 complex indeed leads to increased gene expression variability in different contexts. This is in agreement with the pioneer work showing increased transcriptional noise due to mutation in Paf1C subunits in yeast (23). Among the gene targets, we were particularly surprised to see strong local variability in the expression of STM, a key regulator of meristem maintenance. This observation opens many new questions: Could redundant factors (KNAT genes) buffer such variability in part by displaying “expression averaging” at a regional scale or instead amplify it locally by adding even more stochasticity? Could certain master regulators of development, like STM, be oversensitive to transcriptional regulators, and thus be more likely to be noisy?

There is evidence showing that VIP3 can participate in two independent complexes, Paf1C in the nucleus and SKI (Superkiller) in the cytosol (50). These complexes control mRNA production and turnover, respectively. However, while the only available mutant of the SKI complex Atski2 does not produce any visible flower development defects (50), mutants of different Paf1C subunits show similar flower development defects (30), strongly suggesting that flower development defects in vip3-1 are the result of compromised Paf1C, not SKI, activity. Nevertheless, here, we used vip3-1 as a background in which transcription is noisier than the WT regardless of the precise mechanisms. With the identification of VIP3, and probably other subunits of the Paf1 complex, as regulators of gene expression variability, the corresponding mutants now offer a model system to study the effects of increased gene expression variability on many cell and developmental questions.

Noisy Gene Expression Is Associated with Increased Local Growth Heterogeneity.

Here, we studied one of these questions: the nexus between gene expression variability and organ shape reproducibility. In the simplest hypothesis, increased gene expression variability would lead to increased growth heterogeneity, eventually resulting in more variable organ shapes. To do so, we focused on sepals, an excellent system to study organ developmental robustness in Arabidopsis due to their highly reproducible shape and size (34). A first striking result was the observation that local variation of growth rates of cells in WT sepals is remarkably similar across time points, slightly fluctuating around 14%. This suggests that in the WT, there may exist a limit for local growth heterogeneity because growth of one cell is constrained by contacting cells. This could be a major limiting factor for many other developmental features such as size, aspect ratio, or flatness, and this remains to be studied.

In the mutant sepals, we observed a significant increase in growth heterogeneity between adjacent cells at the end of the imaging period when sepals between WT and vip3-1 become progressively more different. The increased growth heterogeneity in vip3-1 sepals observed in the normal condition is however rather modest, probably because it is compensated by cell wall reinforcement. When sepals are treated with oryzalin to remove cortical microtubules, and thus hinder the ability of the cell to reinforce its cell wall, we found that the growth conflict is further amplified, leading to the reduction in outer wall surface area and volume in some cells in the mutant, culminating in apparent disappearance in the most extreme, but rare, cases. Reduction in volume has not been reported before in plants, except for the very peculiar cell contraction step in the zygote, shortly after fertilization (51). Apparent cell loss in the Oryzalin experiment was even rarer, most likely because cell walls are stiff structures, which can mechanically resist the pressure from other cells. How could negative growth happen is unclear. This must involve differences in wall properties as well as water fluxes and may echo what happens in two-dimensional foams where differences in pressures can lead to cell shrinkage (45). Our observation echoes the “cell competition” phenomenon in animals, in which slow-growing cells are outcompeted and eventually eliminated by fast-growing cells. The death of slow-growing cells may be caused by both chemicals and mechanical compression (52). The observation also shows that in some rare cases, tissue topology in plants can be changed despite the immobility of the cells.

Increased Local Growth Heterogeneity between Adjacent Cells Hinders the Buildup of Larger Scale Conflicts, Resulting in More Variable Organs.

Proper growth arrest, among other factors, contributes to the robustness of the final organ shape and size. Mechanical conflicts have been demonstrated to be a cue for growth arrest in different model organisms (53). In Arabidopsis sepals, a sharp difference in growth rates between the tip and the body results in transverse tensile stress at the tip, as revealed by computational modeling and the observation of consistent transverse cortical microtubule pattern. Mechanical stress has been proposed to be a signal for organ growth arrest from the sepal tip (19). In this present study, the WT sepals show a stereotypical growth pattern as described in ref. 19 but vip3-1 sepals do not exhibit a sharp difference in growth rates between body and tip. Consistent with that, we observed that i) cells at the tip of vip3-1 sepals failed to form a coherent band of transverse cortical microtubules, and ii) cells at the tip do not arrest growth properly; they resume growth. These observations not only support the growth arrest model presented in ref. 19, but they also suggest that increased growth heterogeneity between adjacent cells is incompatible with the formation of a sharp boundary between larger growth domains (the tip and the body), hindering the formation of tissue-scale mechanical conflicts (regional conflicts) and resulting growth arrest. Consequently, the final organ shape and size become more variable.

Local growth heterogeneity has been linked to organ shape robustness in past studies. In Arabidopsis, the SAM maintains a certain level of growth heterogeneity between cells, and reduced growth heterogeneity in the katanin mutant has been associated with the loss of the typical dome shape and sharp organ boundary (40). Sepal cells also displayed significant growth heterogeneity, and reduced growth heterogeneity was found in the ftsh4 mutant with more variable organ shapes (12). Our present study shows that increased local growth heterogeneity is associated with more variable organs. These counterintuitive examples suggest that a certain level of local growth heterogeneity is necessary for organ robustness.

Our scenario could be tested in other contexts too, and in particular, in the Drosophila wing disc, where differential growth also triggers growth arrest through a mechanotransduction (Hippo) pathway (18). One would predict that increased gene expression variability, e.g., Paf1c-dependent as demonstrated in this study, could also diffuse mechanical conflicts at the local scale, generating more variable wing shapes. In the end, the introduction of transcriptional noise as an instructive cue is an invitation to revisit many cell and developmental questions with the lens of multiscale dynamics.

Materials And Methods

Extended Methods and Materials are available in SI Appendix.

Plant Materials and Growth Conditions.

Otherwise stated, all experiments were performed on the Co-0 ecotype. The vip3-1 (Salk_139885) and vip3-2 (Salk_083364) mutants were described in ref. 32, bot1-7 mutant in WS-4 ecotype (54), fer-4 in ref. 55, and xxt1 xxt2 double mutant in ref. 47. The plasma membrane marker line pUBQ10::Lti6b-tdTomato was described in ref. 56, the microtubule reporter line p35S::GFP-MBD in ref. 57, pH1.1::H1.1-EGFP, pH1.2::H1.2-EGFP in ref. 58, pH1.3::H1.3-EGFP in ref. 36, pSTM::CFP-N7 in ref. 59, pAG::AG-2xVenus in ref. 32, DR5rev::3xVENUS-N7 in ref. 60. The pAtML1::H2B-3xGFP::tRBCS line was generated by GreenGate assembly (61) of the following elements: Histone H2B fused to three GFPs and the RuBisCO small subunit terminator, under the AtML1 promoter (62).

For all analyses on sepals, plants were grown on soil at 20 °C in short-day conditions (8-h light/16-h dark) for 3 wk then transferred to long-day conditions (16-h light/8-h dark cycle).

Live Imaging of Abaxial Sepals and SAMs.

Preparation of samples was done as described in ref. 63, section Plants grown on soil. All confocal images were done using an SP8 laser-scanning confocal microscope (Leica) equipped with a long-distance 25× (NA 0.95) water-dipping objective and a resonant scanner module (without the resonant scanner for CMT imaging). For CFP: excitation laser 448 nm, collected range 460 to 490 nm. For GFP: 488 nm, 500 to 520 nm. For Propidium Iodide (PI): 514 nm, 610 to 650 nm. For tdTomato protein: 552 nm, 570 to 605 nm. For Venus protein: 488 nm, 520 to 545 nm.

Image Analysis.

For growth analyses, the MGX 3D image analysis software was used (64). There are several guides to use the software such as (63–65). To reveal the general growth patterns of the sepals as depicted in Fig. 4, the range of growth rates was set to 1-3 for all time points.

Local variation of growth rate is defined here as the CoV (CoV = SD/mean*100%) in growth rate within a group of cells consisting of a seed cell in the center and all its contacting neighbors, as described in ref. 40.

To assess gene expression variability in floral meristems and SAMs, images were taken using a confocal microscope (Leica SP8) with z-step of 0.2 µm from 2 channel: one for nuclear signal (GFP or Venus), and one for PI staining to detect cell walls. Combining the information from the two channel, CoV% in nuclear signal intensity within a group of nuclei/cells (which signify local variation of signal intensity) could be calculated.

Sepal Area Measurements.

Sepals of stage 14 flowers (39) were dissected and placed as flat as possible on a double-sided tape on top of a microscope slide. They were photographed on a black background using a binocular equipped with a camera. A Python program in Linux called SepalContour described in ref. 12 was used to extract sepal contours and morphological parameters such as sepal area, length, width, aspect ratio, and circularity.

Quantification of Sepal Shape Variability.

Sepal contours extracted using the SepalContour tool described above were then used as input for shape analyses. The Contour Analysis program was written in Mathematica and described in ref. 12. From the input contours, the program returns a graph of all original contours, a graph of contours normalized by their area and the average contour, and for each sepal contour, an S₂ value depicting the squared deviation of that contour from the average contour. These S₂ values were used to estimate sepal shape variability within and between genotypes (12).

Measuring Stomatal Guard Cell Sizes.

Stomatal guard cells taken by confocal microscopy using PI staining were segmented and measured using the MorphoLibJ plugin in ImageJ (66).

Detection of Reactive Oxygen Species.

In situ superoxide radicals (O₂⁻) were detected using and nitroblue tetrazolium (NBT) as described in refs. 12 and 67. After staining, the samples were photographed using a Leica binocular.

Oryzalin Treatment.

Oryzalin treatment on sepals was done as described in ref. 19.

Statistical Analyses and Data Visualization.

Statistical analyses were performed in R (68). Graphs were created in R using ggpubr (69) and rstatix (70) packages, or the online tool PlotsOfData (71), or in Microsoft Excel.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Data, Materials, and Software Availability

All data measurements used in this paper are available in supporting information.