Epidermal identity is maintained by cell–cell communication via a universally active feedback loop in Arabidopsis thaliana



The transcription factors ARABIDOPSIS THALIANA MERISTEM L1 (ATML1) and PROTODERMAL FACTOR2 (PDF2) are indispensable for epidermal cell-fate specification in Arabidopsis embryos. However, the mechanisms of regulation of these genes, particularly their relationship with cell–cell signalling pathways, although the subject of considerable speculation, remain unclear. Here we demonstrate that the receptor kinase ARABIDOPSIS CRINKLY4 (ACR4) positively affects the expression of ATML1 and PDF2 in seedlings. In contrast, ATML1- and PDF2-containing complexes directly and negatively affect both their own expression and that of ACR4. By modelling the resulting feedback loop, we demonstrate a network structure that is capable of maintaining robust epidermal cell identity post-germination. We show that a second seed-specific signalling pathway involving the subtilase ABNORMAL LEAFSHAPE1 (ALE1) and the receptor kinases GASSHO1 (GSO1) and GASSHO2 (GSO2) acts in parallel to the epidermal loop to control embryonic surface formation via an ATML1/PDF2-independent pathway. Genetic interactions between components of this linear pathway and the epidermal loop suggest that an intact embryo surface is necessary for initiation and/or stabilization of the epidermal loop, specifically during early embryogenesis.


The specification of epidermal cell fate and the subsequent development of epidermal characteristics are critical elements for plant embryo development and post-germination survival (Javelle et al., 2011b). Embryonic epidermal cell-fate specification occurs early in the development of the zygotic embryo, and has been shown to be affected in a range of mutant backgrounds. In particular, the HD-ZIP class IV transcription factors ARABIDOPSIS THALIANA MERISTEM L1 (ATML1) and PROTODERMAL FACTOR2 (PDF2) have been shown to act redundantly to specify epidermal identity in the apical regions of the developing embryo (Abe et al., 2003). Double pdf2-1 atml1-1 mutant embryos exhibit a striking loss of epidermal cell fate and are non-viable. Consistent with a role in regulating epidermal functions, expression of ATML1 and PDF2 is restricted to the L1/epidermal cell layer from early embryogenesis onwards, and the presence of L1 boxes (cognate HD-ZIP IV binding sites) in the promoters of both genes led to the hypothesis that they may be regulated by their own gene products (Sessions et al., 1999; Abe et al., 2001, 2003; Nakamura et al., 2006; Yu et al., 2008; Javelle et al., 2010). These proteins belong to a large family, members of which show predominantly epidermal expression patterns in both Arabidopsis (Nakamura et al., 2006) and other species such as maize (Zea mays) (Javelle et al., 2011a,b), and for which extensive functional redundancy has been proposed.

Genetic and ablation studies performed in multiple systems over several decades have indicated an important role for cell–cell communication in the specification and maintenance of epidermal cell fate (Ingram, 2004; Javelle et al., 2011b). Consistent with these observations, several receptor kinases have been implicated in embryonic surface formation. Loss of function of either ARABIDOPSIS CRINKLY4 (ACR4) or ABNORMAL LEAFSHAPE2 (ALE2) leads to epidermal defects in seedlings and during later plant development (Gifford et al., 2003; Watanabe et al., 2004; Tanaka et al., 2007). The epistatic relationship between mutant alleles of ACR4 and ALE2, together with in vitro data showing intra-molecular phosphorylation between their gene products, indicates that these proteins act together in the same pathway (Tanaka et al., 2007). ACR4 expression is restricted to the L1 cell layer, and is lost in pdf2-1 atml1-1 double mutants. These observations, together with the presence of an L1 box in the ACR4 promoter, formed the basis for the proposal that ACR4 expression may be positively regulated by ATML1 and PDF2 (Abe et al., 2003; Tanaka et al., 2007). However, loss of epidermal cells in pdf2-1 atml1-1 mutant seedlings makes these results difficult to interpret.

A second pair of receptor kinases, GASSHO1 (GSO1) and GASSHO2 (GSO2), act redundantly to mediate production of an intact embryonic cuticle (Tsuwamoto et al., 2008). During seed development, the expression of GSO1 and GSO2 appears to be restricted to the developing embryo. Although both genes are also expressed post-germination, no defects in the cuticles of gso1-1 gso2-1 double mutants have been reported in post-embryonic organs (Tsuwamoto et al., 2008). Consistent with this seed-specific role, a recent study (Xing et al., 2013) has shown that GSO1 and GSO2 act in the same developmental pathway as ABNORMAL LEAF-SHAPE1 (ALE1), a subtilisin-like serine protease (Tanaka et al., 2001), to regulate embryo surface formation. ALE1 is expressed exclusively in the embryo-surrounding region of the Arabidopsis endosperm. ale1 mutants exhibit defects specifically in embryonic surface formation, showing mild seedling desiccation intolerance and seedling permeability to the hydrophilic dye toluidine blue.

Functional overlap between seed-specific and more global signalling pathways regulating epidermal development has been shown genetically. Although the phenotypes of both acr4 and ale1 mutant seedlings are subtle, and are only revealed reliably by treatment with hydrophilic dyes, ale1 acr4 double mutants are non-viable due to loss of epidermal identity and integrity in the apical part of the embryo, leading to embryonic arrest (Tanaka et al., 2007). Expression of ATML1 and other epidermal markers is lost in ale1 acr4 double mutants. A similarly strong synergistic interaction is observed between ale2 mutants and ale1 mutants (Tanaka et al., 2007), and between mutants in ZHOUPI (encoding a transcription factor that regulates ALE1 expression) and acr4 mutants (Yang et al., 2008). These interactions are consistent with the presence of two distinct but functionally overlapping pathways that are necessary for embryonic epidermal functions: one requiring endosperm-derived signals and involving ALE1, GSO1 and GSO2 (Xing et al., 2013), and the other that is active throughout plant development, and involves the ACR4 and ALE2 receptor kinases (Tanaka et al., 2007).

These results leave important questions unanswered. It has been suggested that ATML1 and PDF2 may represent a point of convergence for the ACR4/ALE2 and GSO1/GSO2/ALE1 signalling pathways, as ale1 acr4 and ale1 ale2 double mutants lose expression of ATML1 during embryogenesis, and because the phenotypes of ale1 acr4 and ale1 ale2 double mutants are very similar to those of pdf2 atml1 double mutants (Tanaka et al., 2007; Javelle et al., 2011b). However, this possibility has not yet been investigated experimentally. Here we test whether ATML1 and PDF2 act downstream of the ACR4/ALE2 signalling pathway, the GSO1/GSO2 signalling pathway, or both, during embryogenesis, and shed light on the functional relationships between the signalling pathways involved in maintenance of epidermal fate during Arabidopsis embryogenesis.


ATML1 and PDF2 show strong dosage sensitivity during embryogenesis, and a more severe double mutant phenotype than previously described

ATML1 and PDF2 have previously been shown to act redundantly to specify epidermal cell fate, but using mutant alleles that may not be fully null (Kamata et al., 2013). We therefore studied this interaction using the pdf2-2 allele, which contains a T-DNA insertion in the 6th exon of PDF2 (SALK_109425c) (Peterson et al., 2013), and a predicted null allele of ATML1, atml1-3 (SALK_033404) (Roeder et al., 2012; Kamata et al., 2013). Quantitative PCR analysis using primers upstream of the T-DNA insertions showed strongly reduced transcript levels in both backgrounds compared to wild-type (Figure S1). Both pdf2-2 single mutant seedlings and atml1-3 single mutant seedlings showed toluidine blue permeability comparable to that in acr4-2 and ale1-4 mutants, which have previously been shown to display mild epidermal defects (Watanabe et al., 2004; Xing et al., 2013) (Figure 1b), suggesting that redundancy between PDF2 and ATML1 may not be complete during embryogenesis.

Figure 1.

Characterization of ATML1 and PDF2 alleles.

(a) Genomic structure of ATML1 and PDF2 showing the positions of T-DNA insertions present in the alleles atml1-3 and pdf2-2.

(b) Comparison of permeability to toluidine blue of atml1-3, pdf2-2, ale1-4, acr4-2 and wild-type seedlings. The results are from biological quadruplicates, each containing 20 seedlings. Error bars indicate standard deviation. Asterisks indicate statistically significant differences (two-tailed t test, < 0.05) compared to wild-type stained control.

Consistent with previously published results (Abe et al., 2003), we were unable to generate pdf2-2 atml1-3 double mutant plants. pdf2-2 mutant plants additionally heterozygous for atml1-3, or atml1-3 mutant plants additionally heterozygous for pdf2-2, produced a high proportion of empty, shrivelled seeds. Moreover, many mis-shapen seeds were produced, particularly in pdf2-2 mutant plants additionally heterozygous for atml1-3 (Figure S2). When self-seed from these plants was germinated, we observed a high proportion of abnormal but viable seedlings (62%, n = 312), which, when tested, were always of the parental genotype (n = 48). Abnormal seedlings were characterized by the production of small, lumpy cotyledons that did not expand normally (Figure 2a–c). The majority of these seedlings survived if cultivated under very humid conditions for 7–10 days after transfer to soil. A smaller proportion of visibly abnormal seedlings were also produced in the progeny of atml1-3 mutant plants additionally heterozygous for pdf2-2, and, when tested, these were always found to be of the parental genotype (n = 24).

Figure 2.

Haploinsufficiency of ATML1 in a pdf2–2 mutant background.

(a–c) Self-progeny of an individual homozygous for pdf2–2 and heterozygous for atml1-3. (a) Seedling genotyped as homozygous for pdf2-2 carrying only the wild-type ATML1 allele. (b,c) Seedlings genotyped as homozygous for pdf2-2 and heterozygous for atml1-3, showing reduced cotyledon size and non-opposite cotyledon positions.

(d–i) Seeds from a self-pollinated individual homozygous for pdf2-2 and heterozygous for atml1-3. The images in (d–f) are all seeds from within one silique, and those in (g–i) are all from an older silique. (d,g) Normal-looking embryos at the heart and early-torpedo stages, respectively. Approximately 25% of embryos resemble those in (d) and (g). (e,h) Embryos with an intermediate phenotype, where the root pole is normal but the embryo shows epidermal defects (such as cell bulging), particularly in the apical zone. Approximately 50% of embryos resemble those in (e) and (h). (f,i) Extremely abnormal embryos, with a disorganized root pole and epidermis. Embryos with this phenotype do not develop further, and were observed in approximately 25% of seeds. Mature seeds from this plant are shown in Figure S2C.

(j,k) Typical epidermal defects, particularly abnormal division planes, observed at the globular stage (indicated by arrows).

Siliques of pdf2-2 mutant plants additionally heterozygous for atml1-3 were analysed at the early-heart to late-heart stage. Of 346 seeds, 95 contained embryos arrested at the globular stage with severe epidermal abnormalities. These embryos did not produce either cotyledon primordia or an organized root pole and hypocotyl (Figure 2f,i), and arrested much earlier than previously reported for pdf2-1 atml1-1 double mutants. Eighty of the 346 seeds produced normal-looking embryos with cotyledon primordia and a smooth epidermal cell layer (Figure 2d,g). The remaining 171 seeds produced embryos with an intermediate phenotype (Figure 2e,h) comprising an irregular epidermal cell layer in which oblique and periclinal cell divisions were observed (Figure 2j,k). The majority of these seeds produced two cotyledons and an embryo axis, but developed more slowly and had shorter cotyledons than the more normal-looking embryos. We predict that these embryos were of the parental genotype (χ2 value for this hypothesis of 1.347 with two degrees of freedom; two-tailed P value = 0.5100). Thus, in the absence of PDF2, one copy of ATML1 is insufficient to support normal epidermal development during embryogenesis.

ATML1 and PDF2 act in the same pathway as ACR4 to regulate epidermal development

acr4-2 mutants show slightly increased toluidine blue permeability of cotyledons and leaves compared with wild-type (Figure 1b). In addition, floral organ development is abnormal, and ovule development is impaired due to lack of integument outgrowth and occasional ovule fusions (Gifford et al., 2003). We crossed single pdf2-2 and atml1-3 mutants into the acr4-2 mutant background in order to test their functional relationships. Single pdf2-2, atml1-3 and ale1-4 mutants show comparable and subtle epidermal phenotypes (Figure 1b). However, in contrast to the embryo lethality previously observed for ale1 acr4 double mutants (Tanaka et al., 2007), double mutants between acr4-2 and pdf2-2 or atml1-3 were fully viable. However, acr4-2 pdf2-2 double mutants were sterile due to extensive fusion of ovules (Figure S3A,D). acr4-2 atml1-3 double mutants showed similar ovule phenotypes to acr4-2 pdf2-2 double mutants, but produced a few viable seeds per silique (usually 2–5). More than 50% of acr4-2 pdf2-2 (from segregating populations) and acr4-2 atml1-3 double mutant seedlings produced cotyledons in non-opposite positions (Figure 3e and Figure S3F), and increased levels (approximately 8%) of monocotyledony were observed in acr4-2 pdf2 seedlings (Figure S3E). The progeny of acr4-2 atml1-3 double mutants showed a slight increase in toluidine blue permeability, but values were not statistically significant compared to either single mutant (Figure 3a).

Figure 3.

ACR4 acts in the same genetic pathway as ATML1 and PDF2.

(a) Tests of toluidine blue permeability in acr4-2, atml1-3 and acr4-2 atml1-3 seedlings. The results are from biological quadruplicates, each containing 20 seedlings. Error bars indicate standard deviation. Double mutant seedlings are not significantly more permeable than single mutant seedlings (two-tailed t test, > 0.1).

(b–e) Seedling phenotypes of single and double mutants. (b) Col-0; (c) atml1-3; (d) acr4-2; (e) acr4-2 atml1-3 showing non-opposite cotyledon positioning. Genetic interactions between pdf2-2 and acr4-2 are shown in Figure S3.

(f) Quantitative PCR analysis of PDF2 transcript levels in wild-type and acr4-2 mutant backgrounds.

(g) Quantitative PCR analysis of ATML1 transcript levels in wild-type and acr4-2 mutant backgrounds.

Transcript levels are expressed relative to those of EIF4. Asterisks indicate significant decreases in transcript levels compared to wild-type genotypes (two-tailed t test, < 0.05).

In summary, combining acr4-2 with either atml1-3 or pdf2-2 mutations, although it exacerbates post-germination acr4 phenotypes such as ovule fusion, does not result in strong additive or synergistic effects at the level of seedling survival and embryonic cuticle permeability. Our genetic data are therefore consistent with a model in which PDF2 and ATML1 act together with ACR4 to regulate embryonic epidermal identity. To investigate this relationship at the molecular level, we quantified the expression levels of ATML1 and PDF2 in acr4-2 mutant seedlings. A slight but significant decrease in the expression levels of both genes was observed in acr4-2 mutants compared to wild-type seedlings (Figure 3f,g), supporting the hypothesis that ACR4-mediated signalling is necessary for maintaining expression levels of ATML1 and PDF2.

L1 boxes in the ACR4 promoter regulate epidermal expression in embryos

Several previous publications have suggested that ATML1 and PDF2 positively regulate the expression of ACR4 via an L1 box in the ACR4 promoter (Abe et al., 2003; Tanaka et al., 2007). We tested the importance of the L1 box in the ACR4 promoter by generating plants in which GFP expression was driven by a full-length wild-type promoter or by two versions of the ACR4 promoter in which the L1 box had been mutated (Figure S4). As previously reported, the wild-type promoter drove GFP expression in the embryonic epidermis and root pole from the globular stage of development onwards (Gifford et al., 2003). In contrast, although both mutated versions of the promoter were able to drive expression of GFP in the embryonic root pole, neither resulted in a strong epidermal GFP signal in the embryo, suggesting that the L1 box in the ACR4 promoter is required for normal epidermal expression of ACR4 during embryogenesis (Figure S4).

ATML1 and PDF2 negatively regulate their own expression and that of ACR4 in seedlings

In order to investigate the effects of reducing endogenous transcript levels of ATML1 and PDF2, combinations of atml1-3 and pdf2-2 were tested. Levels of ACR4 expression in either atml1-3 or pdf2-2 single mutants were not significantly different to those in wild-type plants, consistent with the redundancy of these two genes. However, in the progeny of a pdf2-2 mutant additionally heterozygous for atml1-3, a small but significant increase in ACR4 expression levels was consistently observed (Figure 4a,e), suggesting negative regulation of ACR4 expression by these transcription factors.

Figure 4.

ATML1 and PDF2 negatively affect their own expression, and that of ACR4 in seedling tissues.

(a) Quantitative PCR analysis of total PDF2 expression levels in seedlings from self-fertilized wild-type, atml1-3, pdf2-2, pdf2-2 heterozygous for atml1-3, and two independent lines expressing pPDF1:FLAG-PDF2.

(b,c) pPDF1:GFP-PDF2 expression in inflorescence meristem tissue and a heart stage embryo, respectively.

(d) pPDF1:GFP-ATML1 expression in a heart stage embryo.

(e) Quantitative PCR analysis of expression of ACR4 in the samples analysed in (a).

(f) Quantitative PCR analysis of expression of endogenous PDF2 and ATML1, respectively, in wild-type plants seedlings, and lines over-expressing FLAG-PDF2. Single and double asterisks indicate significant increases and decreases in expression compared with wild-type controls, respectively (two-tailed test, < 0.05).

To increase expression levels of PDF2 or ATML1 in the epidermis, constructs encoding GFP–ATML1, FLAG–ATML1, GFP–PDF2 and FLAG–PDF2 fusions were generated under the control of the PDF1 promoter, which shows a strong epidermis-specific expression pattern and is thought to be a target of positive regulation by ATML1 and PDF2 (Abe et al., 2001, 2003). For GFP fusions, expression of nuclear-localized GFP was detected in both embryo and shoot epidermal tissues, in patterns that were strongly reminiscent of those described for endogenous ATML1 and PDF2 expression (Figure 4b–d). Western blot analysis showed the presence of full-length fusion proteins for all constructs generated (Figure 5b). Expression levels of ATML1 and PDF2 were analysed in transgenic lines by quantitative PCR. Most transgenic lines showed total expression levels (transgene + native gene) that were not significantly different to wild-type native gene expression levels, and showed wild-type phenotypes. However, higher expression levels were observed for several primary transformants expressing each construct. Lines in which FLAG–PDF2 was over-expressed were fertile, but lines over-expressing ATML1 fusions showed strongly reduced male fertility, and no GFP–PDF2 lines showing strong over-expression were obtained in subsequent generations. The phenotypes associated with high expression levels of all constructs were similar, and included the production of narrow leaves, which, in extreme cases, showed both epinastic curling and spiralling (Figure S5B). To test functionality, transgenes for each construct were introduced into the pdf2-2 atml1-3 double mutant backgrounds by crossing stable transgenic lines with plants homozygous for atml1-3 and heterozygous for pdf2-2. Double pdf2-2 atml1-3 mutant plants were obtained in the F2 progeny of crosses to pPDF1:FLAG-PDF2 and pPDF1:GFP-ATML1 transgenic lines which had been shown to have high transgene expression levels (Figure S5C,D). Complemented double mutant plants showed moderate fertility and produced F3 and F4 generations. The lack of rescue for pPDF1:FLAG-ATML1 and pPDF1:GFP-PDF2 may reflect the fact that the transgenic lines used in crosses did not express transgenes at high enough levels (Figure S5E). Although we cannot exclude the possibility that the fusion proteins produced in these lines are non-functional, the fact that strong over-expression phenotypes were observed in the T1 generation makes this possibility unlikely.

Figure 5.

PDF2 binds directly to the ACR4 promoter and can homo- and heterodimerize.

(a) Chromatin immunoprecipitation experiments showing enrichment of the proximal promoter region of ACR4, ATML1 and PDF2 relative to a fragment of the ACTIN7 promoter in chromatin isolated from atml1-3 pdf2-2 double mutants complemented with either GFP–ATML1 or FLAG–PDF2, and immunoprecipitated with either an anti-GFP antibody or an anti-FLAG antibody. Relative enrichment refers to enrichment in the immunoprecipitated chromatin sample relative to that in the relevant input (pre-immunoprecipitation) chromatin sample. Error bars indicate standard deviation of values obtained from three biological replicates.

(b) Western blot analysis of FLAG- and GFP-tagged versions of ATML1 and PDF2 in protein extracts of single and double transgenic plants. The levels of fusion protein expression tend to decrease in double transgenic backgrounds. The presence of a given protein in the line used is indicated above. The Western blots are for proteins immunoprecipitated from these extracts using anti-GFP antibody-conjugated beads. For each Western blot, the antibody used is indicated.

Subsequent analyses of the transcriptional effects of over-expression were therefore performed in two lines over-expressing FLAG–PDF2. In these lines, total PDF2 expression increased to up to three times wild-type expression levels (Figure 4a). Expression of ACR4 in backgrounds over-expressing FLAG–PDF2 was significantly decreased compared to that in wild-type plants, again consistent with negative feedback by PDF2 on the ACR4 promoter (Figure 4e). Because the presence of L1 boxes in the promoters of PDF2 and ATML1 has led several authors to propose that these genes may by self-regulated, we tested the effects of over-expressing PDF2 on expression of the endogenous ATML1 and PDF2 genes, using primer combinations extending into the 5′ UTR of each gene. Both endogenous genes were significantly down-regulated in PDF2 over-expressing plants (Figure 4f).

PDF2 binds directly to the promoters of ACR4, PDF2 and ATML1, homodimerizes, and forms heterodimers with ATML1 in planta

Because of the presence of an L1 box regulatory sequence in the promoter of ACR4, we suspected that regulation by ATML1 and PDF2 may be direct. Chromatin immunoprecipitation (ChIP) was performed in the pdf2-2 atml1-3 background complemented with pPDF1:FLAG-PDF2 or pPDF1:GFP-ATML1. In both cases, enrichment of the L1 box-containing extreme proximal regions of the ACR4, PDF2 and ATML1 promoters was observed compared to a control promoter, only when immunoprecipitation was performed with the antibody corresponding to the expressed tagged protein, consistent with direct binding of PDF2 and ATML1 to each of these sequences (Figure 5a).

The presence of a leucine zipper in their sequences means that ATML1 and/or PDF2 may potentially form homodimers or and heterodimers both between themselves and with other family members, adding a further layer of complexity to their regulatory role. To test whether ATML1 and PDF2 are capable of dimerization, we crossing all the tagged lines described above to generate transgenic plants expressing combinations of FLAG and GFP-tagged ATML1 and PDF2 proteins. Using co-immunoprecipitation, we were able to demonstrate homodimerization of PDF2 and heterodimerization of PDF2 with ATML1 in planta (Figure 5b). Homodimerization of ATML1 was not detected, possibly due to low levels of transgene expression in double transformants.

A negative feedback model for epidermal identity in seedlings provides robust cell fate maintenance

Based on the results above, we developed a simple mathematical model (shown in Figure 6a) in which ACR4 is the receptor for an unknown ligand (L). When bound by this ligand, ACR4 activates a signalling pathway, resulting in production of both ATML1 and PDF2. ATML1 and PDF2 form dimers, potentially of the three forms ATML1:ATML1, ATML1:PDF2 and PDF2:PDF2. In absence of further knowledge of the mechanistic details of the interactions between ATML1 and PDF2, parsimony leads to considering mass action laws for the formation of dimers. These dimers are transcriptional repressors of ACR4. The dimers formed by ATML1 and PFD2 also repress their own transcription. For the purposes of this model, it is assumed that ATML1 and PDF2 are involved in exactly the same reactions, and hence the equations describing their dynamics are exactly identical. For simplicity, their common dynamics are therefore described with a single variable, P. There is then only one other variable in the model, namely the concentration of ACR4 (A). In the absence of further knowledge of the mechanistic details of the various interactions between ACR4 and ATML1/PDF2, sigmoid laws, namely Hill functions, are used to describe the kinetics of both transcriptional regulation by ATML1/PDF2 and regulation via the ACR4 signal. In addition, it is assumed that, as for all proteins, both ACR4 and ATML1/PDF2 are subject to degradation. In the absence of knowledge regarding this process, it is represented by a linear decay. As the ligand that binds ACR4 is unknown, it is represented by a time-dependent quantity λ(t). Finally, both variables A and P are normalized to values between 0 and 1. The model then consists of the system of ordinary differential equations shown in Figure 6(b).

Figure 6.

A negative feedback model for epidermal identity in seedlings provides robust cell fate maintenance.

(a) Schematic representations of the interactions presented in the model. ‘L’ represents the unknown ligand of ACR4.

(b) The system of ordinary differential equations describing this system. The parameters appearing in this ordinary differential equation system are the threshold parameters θA and θP, accounting for ACR4 transcriptional regulation by ATML1/PDF2 and ATML1/PDF2 activation through ACR4-mediated signalling, respectively, the steepness parameter η for ACR4-mediated activation (with a value of 2 for ATML1/PDF2 regulation in this model, as a result of regulation by dimers), the decay rates δA and δP, which appear as global factors due to normalization of both variables to between 0 and 1, and the association constant, γ, for the ATML1/PDF2 dimers.

(c) Time evolution of variables for this model. The top curve represents an imposed ligand concentration, subject to a transient decrease between times 500 and 1000 (arbitrary units). Variables A and P are shown in the middle and lower graphs, respectively, for 20 randomly chosen initial conditions. Parameters are θA = 0:3, θP = 0:8, η  =  4, δA = δP = 0:1, γ  =  0.5.

Twenty numerical simulations of the model were performed for various values of the parameters. As the values of these parameters are currently unknown, the simulations were performed to investigate the qualitative behaviour of the system. Simulations consistently converged towards a unique stable steady state, where variables A and P took positive values, typically neither very close to 0 or 1. The effect of ACR4 signalling and the dynamics of a ligand were considered in the simulations, and a constant high input of ligand again led to asymptotic dynamics whereby both A and P took positive values. The asymptotic state of the system was very robust, with a unique positive steady state whose values were only moderately shifted by fluctuations of the ligand concentration. A graphical representation of such simulations is shown in Figure 6(c).

A model in which ATML1 and PDF2 activate their own expression and that of ACR4, similar to the system previously suggested in the literature, was also tested (Figure S6). Interestingly, this model behaved as a bistable switch for a number of simulations, provided ACR4-mediated induction of ATML1/PDF2 was steep enough (i.e. high enough values of n), although unrealistically high steepness was not required (> 2 was high enough for most choices of other parameters that were tried). In this case, the system showed hysteretic behaviour, as characteristic of bistable systems. Indeed, a transient decrease in the ligand concentration often induced an irreversible switch of the system towards the (0,0) steady state.

GSO1, GSO2 and ALE1 act in a separate developmental pathway to ACR4, ATML1 and PDF2

Double gso1-1 gso2-1 mutants, as previously described (Tsuwamoto et al., 2008), show a severe defect in embryonic surface development that leads to extreme permeability of cotyledon surfaces and embryo adhesion to the endosperm. gso1-1 gso2-1 double mutants are indistinguishable from ale1-4 gso1-1 gso2-1 triple mutants, suggesting that GSO1 and GSO2 act in the same pathway as ALE1 (Xing et al., 2013). In contrast, ale1 mutants show a strong synergistic interaction with mutants in the ACR4 gene, leading to embryo lethality (Tanaka et al., 2007). We tested the relationship between GSO1/GSO2, ALE1 and ACR4 function. Triple acr4-2 gso1-1 gso2-1 mutants were never obtained in our experiments. When the immature siliques of gso1-1 gso2-1 double mutants additionally heterozygous for acr4-2 were examined, they were found to contain aborted seeds (101 of 420 seeds examined), which contained small arrested embryos showing severe epidermal defects, similar to those reported in ale1 acr4 or ale1 ale2 double mutants (Figure S7A,B). This supports the hypothesis that GSO1 and GSO2 act in a separate pathway to ACR4, consistent with previously described genetic interactions.

To test whether the GSO1/GSO2 pathway involves ATML1 and PDF2, we crossed single pdf2-2 and atml1-3 mutants into the gso1-1 gso2-1 and ale1-4 mutant backgrounds. As similar interactions were observed for atml1-3 and pdf2-2, only those obtained for pdf2-2 are described here. Double ale1-4 pdf2-2 mutant plants were obtained, but, although their post-germination growth was indistinguishable from that of wild-type, the self-seed of this genotype showed a seed abortion rate of 28% (n = 367) (Figure S7C-E). In addition, the seedlings of ale1-4 pdf2-2 mutants were rarely normal. Of 230 seedlings examined, only 20% resembled ale1-4 seedlings, with 17% having three cotyledons (example in Figure 7g) and 23% having only one cotyledon (example in Figure 7h). In most seedlings, cotyledons were deformed, produced in non-opposite positions, and showed dramatic notching in the distal margin (example in Figure 7i). Viable seedlings were highly permeable to toluidine blue compared to single mutants or wild-type (Figure 7j). Thus PDF2 and ALE1 show a synergistic interaction during embryogenesis, and some ale1-4 pdf2-2 double mutants arrest during embryo development.

Figure 7.

ATML1 and PDF2 act in a separate pathway to GSO1, GSO2 and ALE1.

(a–d) Embryos from cleared siliques of pdf2-2 gso1-1 double mutants additionally heterozygous for gso2-1.

(a,c). Normal-looking embryos at the late-heart and walking-stick stages, respectively. (b,d) Presumed triple pdf2-2 gso1-1 gso2-1 mutant embryos from the same siliques as those shown in (a) and (c), respectively.

(e) pdf2-2 mutant seedling.

(f) ale1-4 mutant seedling.

(g–i) ale1-4 pdf2-2 double mutant seedlings.

(j) Tests of toluidine blue permeability in pdf2-2, ale1-4 and ale1-4 pdf2-2 double mutant seedlings. The results are from biological quadruplicates, each containing 20 seedlings. Error bars indicate standard deviation. Single asterisks represent significant increases compared to wild-type controls, and double asterisks represent significant differences compared to both wild-type and single mutants (two-tailed t test, < 0.05).

(k,l) Normal-looking and arrested embryos, respectively, from the silique of an ale1-4 pdf2-2 double mutant.

In order to characterize the point at which embryo defects were observed in ale1-4 pdf2-2 double mutants, developing seeds were cleared and studied by light microscopy. Clear defects including irregularities in epidermal cell division patterns and a delay in cotyledon initiation were observed in many embryos at the early-heart stage of development. A proportion of embryos (20–30%) arrested at the late-globular/early-heart stage (Figure 7l), and showed extreme disorganization of embryo cell divisions, including lack of defined cotyledon primordia. Most embryos continued to develop, although defects in epidermal organization and cotyledon number and shape were frequent, as were adhesion to the endosperm and abnormal embryo bending, as previously reported for ale1-4 single mutants (Tanaka et al., 2001; Xing et al., 2013). When samples of mature seeds were observed, only a very few (approximately 10%) presented a wild-type seed shape (Figure S7E).

It was not possible to obtain triple pdf2-2 gso1-1 gso2-1 mutants. When the seeds of pdf2-2 gso1-1 mutants additionally heterozygous for gso2-1 were observed, approximately 25% of seeds (93 of 394 seeds counted) appeared non-viable. This figure corresponds to the approximately 25% white seeds observed when immature siliques were examined (53 of 220 seeds examined). When the developing seeds of pdf2-2 gso1-1 mutants additionally heterozygous for gso2-1 were observed, approximately 25% of embryos were found to have arrested at time points between the late-globular and heart stage. These embryos showed severe disorganization of the epidermal cell layer, and generally lacked cotyledon primordial, although occasional single cotyledons were initiated prior to arrest (Figure 7b,d). The phenotypes were reminiscent of those observed in acr4-2 gso1-1 gso2-1 mutants. In conclusion, PDF2 shows similar synergistic interactions with ALE1, GSO1 and GSO2 to those shown by ACR4.

Our genetic results suggest that ATML1 and PDF2 act in a separate pathway to GSO1,GSO2 and ALE1. We therefore tested ATML1 and PDF2 expression in gso-1 gso2-1 double mutant and ale1-4 mutant seedlings. We found no change in the expression of ATML1, PDF2 or ACR4 in either mutant background compared to wild-type seedlings (Figure S8A–C). Because ALE1 function is restricted to seeds, we also tested the expression of ATML1 and PDF2 in ale1-4 mutant siliques during seed development, and again were unable to detect any significant difference in expression levels compared to those in the wild-type (Figure S8D,E).


ALE1, GSO1 and GSO2 are necessary for embryonic surface formation during seed development (Tanaka et al., 2001, 2007; Tsuwamoto et al., 2008; Xing et al., 2013). Although ale1 and gso1 gso2 double mutants are principally defective in embryo surface formation, expression of epidermal identity genes such as ATML1 is lost in ale1 acr4 double mutants (Tanaka et al., 2007). We determined whether this was because both the ACR4 signalling pathway and the GSO1/GSO2 signalling pathway are required to stimulate expression of ATML1 and PDF2 during embryogenesis. Although we show that plant development is extremely sensitive to the level of expression of ATML1 and PDF2, transcription of these genes does not depend on GSO1/GSO2 or ALE1 function. Indeed, similar synergistic genetic interactions were observed between pdf2 or atml1 mutants and either ale1 or gso1 gso2 double mutants to those previously observed between acr4 mutants and either ale1 or gso1 gso2 double mutants, suggesting that ALE1, GSO1 and GSO2 act in a parallel pathway to ATML1 and PDF2.

In contrast, our results suggest that ATML1 and PDF2 act in the same pathway as ACR4. We dissected this pathway by integrating mathematical modelling approaches with classical genetic analysis. Our genetic and molecular analysis confirmed that, in contrast to the published view, ATML1 and PDF2 negatively affect both their own transcription and the transcription of ACR4, at least in seedling tissue, generating a double-negative feedback loop. Our simulations show that this network structure is capable of maintaining robust and stable expression levels of all three genes, even in the face of strongly varying signal input. Negative feedback control of cell signalling and transcription factor networks has been shown to generate systems that are robust in the face of considerable noise, both in other biological systems and in pure modelling experiments (Kitano, 2004; Stelling et al., 2004). This result is consistent with the fact that epidermal identity is stably established in the L1 cells of plants at the seedling stage of development.

Paradoxically, our results in this study, and the results from other studies (Abe et al., 2001; Takada and Jurgens, 2007), show that the L1 boxes present in the promoters of ACR4, ATML1 and PDF1 are important for maintaining expression in the epidermis during embryogenesis. However, we complemented pdf2-2 atml1-3 double mutants with tagged versions of ATML1 and PDF2 expressed under the control of PDF1 promoter, showing that the PDF1 promoter drives expression independently of the activity of ATML1 and PDF2 during embryogenesis. This promoter may therefore be positively regulated by other members of the 16-strong HD-ZIP IV protein family, several of which are reported to be strongly expressed during early embryogenesis, and all of which are likely to bind to L1-box-like sequences (Nakamura et al., 2006). The expression of L1 box-containing genes, including ATML1, PDF2 and ACR4, is therefore also likely to be regulated by a complex and tissue-specific balance of the activity of these and other transcription factors.

Our work is the first to suggest an inhibitory role for members of the HD-ZIP IV family of transcription factors, which are generally considered to be positive regulators of transcription (Peterson et al., 2013). In this context, it is interesting to note that the three genes that we have studied each have single L1 box very close to their transcriptional start site (within 110 bp of the start site in each case). It is therefore possible that the regulatory mechanism by which ATML1 and PDF2 govern this central feedback loop differs from that towards other transcriptional targets. For example, they may interfere with the basic transcriptional machinery once high enough levels are present within the cell. It should also be noted that, although our results suggest strong functional similarities and redundancy between ATML1 and PDF2, it is possible that their different dimeric combinations have subtle functional specificities that this study has not uncovered.

Although making quantitative assessments of gene expression levels during early zygotic embryogenesis is almost impossible, the observations above throw further light on developmental mechanisms during this process. If ATML1- and PDF2-containing complexes at the promoters of ACR4, ATML1 and PDF2 during early embryogenesis produce a net positive rather than a negative feedback, our modelling shows that the resulting system tends to bistability. Bistable switches, generated by positive feedback loops, have been implicated in robust developmental decision-making in a variety of organisms including Drosophila and mammals (Freeman, 2000; Stelling et al., 2004). During early Arabidopsis embryogenesis, such a mechanism may play an important role in developmental decisions necessary for correct cell-fate specification, being potentially involved, for example, in the permanent loss of ATML1 expression (and thus protodermal cell fate) in internalized daughter cells generated by the first periclinal divisions of the embryo proper at the dermatogen stage of embryogenesis (Lu et al., 1996; Takada and Jurgens, 2007).

The above hypothesis may also provide insight into how the function of the ALE1/GSO1/GSO2 pathway affects epidermal identity so profoundly in mutants with only mild epidermal phenotypes, whereas it appears to only affect embryonic surface formation in wild-type plants. The early arrest and severe developmental phenotypes shown by pdf2-2 gso1-1 gso2-1 triple mutants and acr4-2 gso1-1 gso2-1 triple mutants (this study), and to a lesser extent ale1 acr4 double mutants (Tanaka et al., 2007), suggest that the GSO1/GSO2-mediated pathway is required early in embryogenesis to maintain epidermal cell fate in backgrounds where epidermal cell fate specification is compromised. Surface defects in the ale1-4 mutant are less severe than those in the gso1-1 gso2-1 double mutant and, interestingly, only a proportion of ale1-4 pdf-2 double mutants arrest early in embryogenesis, whereas the rest, although showing embryo and seedling defects, complete embryogenesis and go on to produce phenotypically normal adult plants. This observation implies that the presence of an intact embryonic surface is particularly important for progression through the globular and early-heart stages of embryogenesis, and indicates the presence of a developmental threshold that is necessary for stabilization of epidermal cell fate at this stage in development. ACR4-mediated signalling probably requires the perception of an extracellular ligand within the epidermal cell layer. An intriguing possibility is that inappropriate apoplastic continuity between the embryo and endosperm may compromise accumulation of this as yet unidentified molecule. In genetic backgrounds with defects in embryonic surface formation, this may prevent stabilization of the ACR4–ATML1/PDF2 feedback loop during early embryogenesis, and thus precipitate catastrophic and irreversible loss of epidermal identity.

In summary, our results have clarified the relationships of two parallel signalling pathways implicated in embryonic surface formation in Arabidopsis thaliana. We provide strong evidence for a feedback loop capable of maintaining robust epidermal cell fate during post-embryonic development. Genetic analysis has highlighted a potential role for production of an intact embryonic surface in stabilizing this loop early in embryogenesis. The important but technically challenging next step in providing confirmation of how epidermal identity is regulated at various developmental stages will comprise parameterization of our model in various tissues, including early embryos.

Experimental procedures

Plant materials and growth conditions

Mutant lines used in this study have been described previously, with the exception of pdf2-2 (SALK_109425c), which was provided by the Nottingham Arabidopsis Stock Centre (http://arabidopsis.info/). All plant lines were plated on MS medium (with or without selection), stratified for 3 days at 4°C, and germinated in a growth cabinet under long (16 h) days at 21°C for 7–10 days before transferring to soil under identical growth-room conditions.

Transgene construction

The entire genomic region containing ATML1, starting at the ATG codon and ending at the 3′ end of the 3′ UTR, was amplified using primers gATML1B2r and gATML1B3 (Table 1) and recombined directly into pDONR P2R–P3 (Invitrogen , www.lifetechnologies.com/fr) to give pENTR-ATML1-L2R-L3. PDF2 was cloned similarly using primers gPDF2B2r and gPDF2B3 (Table 1) to give pENTR-PDF2-L2R-L3. A single FLAG tag was made by denaturing a mixture of the primers FLAGf and FLAGr (Table 1), re-annealing and recombining directly into pDONR221 (Invitrogen) to give pENTR-FLAG-L1-L2. The entry vector pENTR-pPDF1-L4-L1R, was produced by amplifying genomic DNA using primers pPDF1B4 and pPDF1B1R (Table 1) and recombining directly into pDONR P4–P1R (Invitrogen). The entry vector pENTR-GFP-L1-L2 containing the GFPS65C open reading frame (Heim et al., 1995) without a stop codon, has been described previously (Fobis-Loisy et al., 2007). Triple recombinations to produce binary vectors expressing N–terminally tagged proteins under the control of the PDF1 promoter were performed using destination vectors pART27 (Gleave, 1992) (kanamycin resistance) or pH7m34GW (Karimi et al., 2002) (hygromycin resistance).

Table 1. Primers used in this study
NameSequence (5′→3′)
Genotyping primers
Quantitative PCR primers
Cloning primers

For ACR4 promoter analysis, the full-length promoter region of ACR4 was amplified using primers pACR4F and pACR4R (Table 1), and cloned directionally into pENTR/D–TOPO (Invitrogen). Mutated versions of the promoter were generated by amplifying 5′ fragments using primers pACR4F and either L1mut1R or L1mut2R and 3′ fragments using primers pACR4R and either L1mut1F or L1mut2F (Table 1). Matching 5′ and 3′ fragments were annealed, subjected to a second round of PCR using primers pACR4F and pACR4R and cloned into pENTR/D–TOPO (Invitrogen). All promoter fragments were then recombined into pKGWFS7 (Karimi et al., 2002). For each promoter fragment, 10-20 single-insertion homozygous T3 lines were established, and representative GFP expression profiles were obtained by confocal microscopy. Agrobacterium-mediated plant transformation was performed by floral dipping (Clough and Bent, 1998)

Seed clearing

To visualize developing embryos, siliques were opened with needles, and the seeds were removed with forceps into a drop of clearing solution (8 g chloral hydrate, 2 ml water, 1 ml glycerol). Cover slips were applied, and the samples were incubated at 4°C overnight.


Cleared seeds were visualization under differential interference contrast optics using a Zeiss AX10 microscope (http://microscopy.zeiss.com/microscopy/fr_fr/). Seedlings and dry seeds were photographed under a Leica MZ12 stereomicroscope (http://support.leica-microsystems.com/france/). Confocal images were obtained using a Zeiss LSM700 confocal laser scanning microscope. Cryo-scanning electron micrographs were obtained using a Hirox SH-3000 (http://www.hirox-europe.com/) benchtop scanning electron microscope equipped with a COOLSTAGE specimen cooling unit MK3 (Deben UK Ltd, http://deben.co.uk/).

Toluidine blue staining

Toluidine blue staining was performed as described by Xing et al. (2013).

Quantitative gene expression analysis

Seedlings were grown in Costar 3516 six-well cell-culture cluster plates (Corning , http://www.corning.com/lifesciences/emea/) containing 2.5 ml MS medium (0.5% sucrose). Genotypes were distributed randomly (four biological replicates per genotype) to minimize edge effects. Plates were stratified for 3 days, and incubated with gentle horizontal circular agitation for 10 days under growth room conditions. RNA extraction and quantitative PCR analysis were performed as described by Xing et al. (2013). Primers used for quantitative PCR analysis are shown in Table 1.


Plant DNA was extracted using a rapid CTAB isolation technique (Stewart and Via, 1993). ale1-4 genotyping was performed as previously described (Tanaka et al., 2001; Yang et al., 2008; Xing et al., 2013). gso1-1 and gso2-1 mutant alleles were genotyped as described by Tsuwamoto et al. (2008). Genotyping of atml1-3 was performed using duplex PCR with atml1-F, atml1-R and a SALK left border primer, except in the presence of ATML1-containing transgenes, when atml1-L was replaced by atml1-UTR. pdf2-2 was genotyped using duplex PCR with pdf2-F and pdf2-R and a SALK left border primer, except in the presence of PDF2-containing transgenes, when pdf2-R was replaced with pdf2-UTR.

Chromatin immunoprecipitation

Chromatin immunoprecipitation of GFP-tagged PDF2 was performed essentially as described previously (Gendrel et al., 2002), with modifications as described by Schubert et al. (2006). ChIP was performed on 1-3 g of inflorescence tip material, and quantitative PCR analysis of input and immunoprecipitation samples was performed as described above. Experiments were performed using three independent biological replicates. After chromatin purification and sonication, each chromatin sample was split in two and immunoprecipitated using either polyclonal antiGFP antibody A11122 (Invitrogen) or monoclonal M2 anti-FLAG antibody (Roche, www.roche-applied-science.com/). In either case, antibodies were pre-bound to a 50/50 mix of Protein A- and Protein G-linked Dynabeads (Life Technologies, http://www.lifetechnologies.com/fr/), so that the only difference between immunoprecipitations was the antibody used. ATML1, PDF2 and ACR4 promoter fragments adjacent to L1 boxes were detected using primers designed using the Roche Universal Probes Library (http://qpcr.probefinder.com/) online tool. Enrichment is calculated by comparison to a fragment of ACTIN7 amplified as described by Schubert et al. (2006). Relative enrichment is expressed as a ratio of the enrichment in input (pre-immunoprecipitation) chromatin samples to that in immunoprecipitated chromatin samples.

Western blotting and co-immunoprecipitation

Inflorescences were snap-frozen in liquid nitrogen, thoroughly ground in pre-cooled porcelain mortars, and proteins were extracted using an extraction buffer containing 50 mm Tris/HCl pH 7.5, 150 mm NaCl, 1% Nonidet P–40 and 1 × protease inhibitor cocktail P9599 (Sigma Aldrich, http://www.sigmaaldrich.com/france.html). Samples were incubated on ice for 30 min, diluted with extraction buffer to give an Nonidet P–40 concentration of 0.2%, centrifuged twice for 10 min at 14 000 g, and protein concentrations in supernatants were determined by Bio-Rad protein assay (http://www.bio-rad.com/). For immunoprecipitation, 500 μg of total proteins was incubated with 50 μl anti-GFP magnetic microbeads (Miltenyi Biotech, https://www.miltenyibiotec.com/) for 2 h at 4°C with gentle rotation. Beads were captured on microcolumns in a μMACS separator (Miltenyi Biotech). Columns were rinsed four times with extraction buffer containing 0.1% NP-40, and the immunocomplex was eluted using pre-heated (95°C) 4 × Laemmli buffer. Samples were resolved on 7.5% polyacrylamide/0.1% SDS gels. Proteins were transferred onto nylon membranes using an iBlot® dry blotting system (Invitrogen). Membranes were blocked for 1 h at room temperature using 1 x PBS/0.2% Tween-20 (Sigma-Aldrich) + 5% non-fat milk (Regilait, http://www.regilait.com/), and incubated overnight at 4°C with anti-FLAG® M2 (Sigma Aldrich) or anti-GFP (Roche) primary antibodies at dilutions of 1:2000 and 1:1000, respectively. Secondary horseradish peroxidase-conjugated anti-mouse immunoglobulins (Promega, http://france.promega.com/) were used at a dilution of 1:10 000. Blots were incubated with Super Signal West Femto reagents (Thermo Scientific, http://www.thermoscientific.com/) according to the manufacturer's instructions, and then exposed to Super RX film (Fujifilm, http://www.fujifilm.eu/fr/).


R.S.-B. was supported by a Marie Curie Initial Training Network (Signals and Regulatory Networks in Early Plant Embryogenesis, SIREN). This work also forms part of a Chaire D'Excellence by the Agence Nationale de la recherche awarded to G.I and supporting A.C and R.G. We would like to thank Pradeep Das for the vector pENTR-pPDF1-L4-L1R, Teva Vernoux and Enrico Coen (Department of Cell and Developmental Biology, John Innes Centre, Norwich, UK) for critical comments on the manuscript, Alexis Lacrois, Yannick Rasmus and Priscilla Angelot for help with plant culture, and Anne-Marie Thierry, Hervé Leyral and Isabelle Desbouchages for technical assistance.