Mutations in epidermis-specific HD-ZIP IV genes affect floral organ identity in Arabidopsis thaliana


For correspondence (e-mail


Development of the epidermis involves members of the class–IV homeodomain-leucine zipper (HD-ZIP IV) transcription factors. The Arabidopsis HD-ZIP IV family consists of 16 members, among which PROTODERMAL FACTOR 2 (PDF2) and ARABIDOPSIS THALIANA MERISTEM LAYER 1 (ATML1) play an indispensable role in the differentiation of shoot epidermal cells; however, the functions of other HD-ZIP IV genes that are also expressed specifically in the shoot epidermis remain to be fully elucidated. We constructed double mutant combinations of these HD-ZIP IV mutant alleles and found that the double mutants of pdf2-1 with homeodomain glabrous1-1 (hdg1-1), hdg2-3, hdg5-1 and hdg12-2 produced abnormal flowers with sepaloid petals and carpelloid stamens in association with the reduced expression of the petal and stamen identity gene APETALA 3 (AP3). Expression of another petal and stamen identity gene PISTILATA (PI) was less affected in these mutants. We confirmed that AP3 expression in pdf2-1 hdg2-3 was normally induced at the initial stages of flower development, but was attenuated both in the epidermis and internal cell layers of developing flowers. As the expression of PDF2 and these HD-ZIP IV genes during floral organ formation is exclusively limited to the epidermal cell layer, these double mutations may have non-cell-autonomous effects on AP3 expression in the internal cell layers. Our results suggest that cooperative functions of PDF2 and other members of the HD-ZIP IV family in the epidermis are crucial for normal development of floral organs in Arabidopsis.


Homeodomain-leucine zipper (HD-ZIP) proteins are transcription factors present only in the plant kingdom. They have a homeodomain (HD), which is a conserved 60-amino-acid motif for DNA binding, and a leucine zipper motif, which mediates their homo- and hetero-dimerization. The class-IV HD-ZIP (HD-ZIP IV) proteins are characterized by an internal loop in the middle of the leucine zipper motif, which is thus called a zipper-loop-zipper (ZLZ) domain (Schrick et al., 2004). They also contain a steroidogenic acute regulatory protein-related lipid transfer (START) domain and a START-adjacent (SAD) domain in the C terminus. The HD-ZIP IV family in Arabidopsis thaliana consists of 16 genes: GLABRA2 (GL2), ARABIDOPSIS THALIANA MERISTEM LAYER 1 (ATML1), ANTHOCYANINLESS 2 (ANL2), PROTODERMAL FACTOR 2 (PDF2) and HOMEODOMAIN GLABROUS 1 (HDG1)–HDG12, among which HDG6 is identical to the previously identified FWA (Nakamura et al., 2006). Expression analyses with promoter-GUS gene fusions have revealed that ATML1, PDF2, HDG2, HDG5, HDG11 and HDG12 are expressed predominantly in the epidermal layer of shoot meristems and organs (Nakamura et al., 2006), suggesting that many members of the HD-ZIP IV family regulate gene expression in the epidermis. However, only four single knock-out mutants are known to show morphological alterations. The gl2 mutant is defective in trichome and seed mucilage formation, and has ectopic root hairs in non-hair cell files (Rerie et al., 1994; Di Cristina et al., 1996). In anl2, anthocyanin accumulation is reduced in the shoot, and several extra cells are formed between cortical and epidermal layers of the root (Kubo et al., 1999; Kubo and Hayashi, 2011). The hdg11 mutant has trichomes with increased branching, and the phenotype is enhanced in hdg11 hdg12 double mutants (Nakamura et al., 2006), whereas hdg2 has trichomes with smooth cell walls (Marks et al., 2009). ATML1 and PDF2, which are closely similar in sequence to each other, are both expressed specifically in the outermost cell layer (L1) of shoot apical meristems (Lu et al., 1996; Sessions et al., 1999; Abe et al., 2003). While each T-DNA insertion mutant of ATML1 and PDF2 shows the wild-type phenotype, the atml1-1 pdf2-1 double mutant has severe defects in the differentiation of shoot epidermal cells, indicating that they play redundant but critical roles in the formation of the shoot epidermis (Abe et al., 2003). So far, however, there has been no further information on double mutants within the family that display abnormal phenotypes, except for pdf2-1 hdg3-1 and atml-1 hdg3-1, which show slight defects in cotyledon development (Nakamura et al., 2006), and the precise functions of HD-ZIP IV members and functional redundancy among them remain largely to be elucidated.

There are also accumulating reports of epidermis-related functions of HD-ZIP IV genes in other plants. In Zea mays (maize), OUTER CELL LAYER 4 (OCL4) has been suggested to regulate trichome patterning and anther development (Vernoud et al., 2009), whereas OCL1 may be involved in lipid metabolism and cuticle biosynthesis (Javelle et al., 2010). The cotton GbML1 regulates fiber development (Zhang et al., 2010). In Solanum lycopersicum (tomato), Woolly (Wo), the closest homolog of Arabidopsis PDF2, is essential for embryo development, and its dominant allele is known to exhibit the woolly trichome phenotype (Yang et al., 2011). However, these genes represent just a part of the HD-ZIP IV family, and we are still far from a comprehensive understanding of the HD-ZIP IV family.

The epidermis is organized in a continuous monolayer of cells that covers the plant body. Epidermal cells in shoot organs are exclusively derived from L1, the outermost layer of shoot apical meristems, which continues to undergo anticlinal cell division (Barton and Poethig, 1993). Shoot epidermis plays a critical role in organ separation and defense responses against drought or pathogens, as well as in the integrity of organs. In flowers, the Arabidopsis cytochrome P450 KLUH (KLU) expressed in the epidermis stimulates organ growth non-cell-autonomously, and may be one of the major players in the coordination of the final size of floral organs (Anastasiou et al., 2007; Eriksson et al., 2010). When floral organ identity genes in Arabidopsis such as AGAMOUS (AG), SEPALLATA 3 (SEP3), APETALA 3 (AP3) and PISTILLATA (PI) are expressed specifically in the L1, the flowers show similar modifications to those with constitutive expression of the respective genes, suggesting that the differentiation and maturation of floral organs can be partially directed from the L1 cells (Urbanus et al., 2010). Although AP3 expressed only in the epidermis is not sufficient for full restoration of floral organ identity in ap3 mutants, it acts non-autonomously to recover the overall shape of petals, confirming the important contribution of the epidermis to organ development (Jenik and Irish, 2001; Urbanus et al., 2010).

To examine further the function of the HD-ZIP IV genes in Arabidopsis, we generated their double mutants and found that the combinations of pdf2-1 with mutant alleles of HDG1, HDG2, HDG5 or HDG12 resulted in abnormal floral organ formation, suggesting the importance of the interplay in the epidermis between PDF2 and these HD-ZIP IV proteins during flower development.


A previous study has revealed that double mutants of HD-ZIP IV gene pairs with high sequence similarity such as HDG2–HDG3 and HDG4–HDG5 (Figure S1) do not necessarily show an abnormal phenotype (Nakamura et al., 2006). We therefore focused on the study of double mutant combinations of pdf2 or atml1 alleles with the alleles of other HD-ZIP IV genes.

pdf2-1 hdg2-3 produces flowers with sepaloid petals and carpelloid stamens

HDG2 is the second closest homolog to ATML1 and PDF2 (Figure S1). The mutant alleles, hdg2-2 and hdg2-3, have a T-DNA inserted in the START domain coding region and in the HD coding region of HDG2, respectively, and the latter may represent a null allele (Figure S2; Nakamura et al., 2006). Thus, we first generated double mutant combinations of hdg2-3 with pdf2-1. In the course of detailed phenotypic analysis, we found that one or two of the two short stamens were occasionally absent in pdf2-1 and hdg2-3 single mutant flowers (Figure 1a,b). pdf2-1 hdg2-3 flowers had smaller petals and stamens than the wild type, and fertility was severely reduced (Figures 1c and 2a–c). The mutant petals were greenish, and rather resembled sepals (Figure 2d). Tissue sections of the petal revealed that inner cells were often composed of more than two layers of large cells, like sepals (Figure 2e–g). We further observed by using scanning electron microscopy (SEM) that the epidermal cells of the adaxial surface of pdf2-1 hdg2-3 petals were flat and long, like sepals, rather than conical, like normal petals (Figure 3a–c). The abaxial surface of these petals was composed of cells with normal shape and clusters of elongated cells with scattering guard cells (Figure 3d–f). Stamens of pdf2-1 hdg2-3 flowers frequently had enlarged anthers and carpelloid structures such as stigmatic papillae and ovules, or filaments without anthers (Figure 3g–k). These abnormalities were observed at higher frequency in the four long stamens than in the two short stamens (Figure 1b). We also confirmed that the phenotype of pdf2-1 hdg2-2 flowers was almost identical with that of pdf2-1 hdg2-3 flowers (Figure 1).

Figure 1.

Phenotypes of wild-type and pdf2 hdg stamens.(a) Variation of the number of stamens per flower. (b) Percentage of phenotypes observed in long and short stamens of the wild type and a series of mutants. Fertile stamens had yellow mature anthers with pollen grains. Infertile stamens had greenish anthers without any obvious pollen grains. The stamens with carpel-like structures such as stigmatic papillae and ovule were counted as carpelloid. (c) Length of filaments of fertile and infertile stamens. Error bars indicate standard deviations.

Figure 2.

Floral phenotype of pdf2-1 hdg2-3. (a–c) Mature flowers of wild type (a), hdg2-3 (b) and pdf2-1 hdg2-3 (c). Both sepals and petals were removed from a flower on the right-hand side (a–c). Only sepals were removed from the flower in the centre (c). (d) From left to right: a wild-type sepal, a wild-type petal and three petals typically observed in pdf2-1 hdg2-3. (e–g) Cross section of a wild-type petal (e), a wild-type sepal (f) and a pdf2-1 hdg2-3 petal (g). Scale bars: 1 mm (a–d); 100 μm (e–g).

Figure 3.

SEM images of petals and stamens. (a–c) Adaxial surface of a wild-type petal (a), a pdf2-1 hdg2-3 petal (b) and a wild-type sepal (c). (d–f) Abaxial surface of a wild-type petal (d), a pdf2-1 hdg2-3 petal (e) and a wild-type sepal (f). (g) Wild-type stamen. (h–l) pdf2-1 hdg2-3 stamens with stigmatic papillae (h), with an abnormally enlarged anther and stigmatic papillae (i), with ectopic ovule-like structures (j, k), and with no anther (l). Scale bars: 10 μm (a–f); 100 μm (g–l).

pdf2-1 hdg1-1, pdf2-1 hdg5-1 and pdf2-1 hdg12-2 also show the homeotic phenotype

We found that hdg1-1 also had flowers lacking one or two of the two short stamens, whereas hdg1-2 had normal flowers (Figure 1a). The T-DNA insert in hdg1-1 is located in the intron splitting the START domain coding region, and that in hdg1-2 is in the exon encoding the ZLZ domain of HDG1 (Figure S2) (Nakamura et al., 2006). pdf2-1 hdg1-1 showed partial homeotic conversions of petals into sepals and stamens into carpels, a phenotype similar to that of pdf2-1 hdg2-3 (Figures 1b and 4a,b), whereas pdf2-1 hdg1-2 showed no homeotic phenotype but instead had flowers lacking in one or two of the short stamens with higher frequency than pdf2-1 (Figures 1 and 4c,d). Expression of the GUS reporter gene fused to a HDG1 promoter has previously been detected only in the epidermis of stamen filaments (Nakamura et al., 2006); however, we confirmed that the longer promoter could also direct GUS expression in the epidermis of stems, flower buds and the nucellus (Figure S3).

Figure 4.

Floral phenotypes of hdg single and pdf2-1 hdg double mutants. (a, b) hdg1-1 (a) and pdf2-1 hdg1-1 (b). (c, d) hdg1-2 (c) and pdf2-1 hdg1-2 (d). (e, f) hdg5-1 (e) and pdf2-1 hdg5-1 (f). (g, h) hdg12-2 (g) and pdf2-1 hdg12-2 (h). Scale bars: 1 mm.

The hdg5-1 allele, which carries a T-DNA insertion in the START domain coding region of HDG5 (Figure S2), showed a wild-type phenotype, but pdf2-1 hdg5-1 also produced flowers with sepaloid petals and carpelloid stamens, resulting in reduced fertility (Figures 1 and 4e,f). On the contrary, double mutants of pdf2-1 with hdg5-2, which has a T-DNA insertion in the SAD domain coding sequence (Figure S2), were indistinguishable from pdf2-1 plants (Figure 1).

We also found that pdf2-1 hdg12-2 flowers exhibited only weak conversion of petals and stamens to sepals and carpels, respectively, with low frequency compared with the double mutants described above (Figures 1 and 4g,h). The shape of epidermal cells of the adaxial surface of pdf2-1 hdg12-2 petals was intermediate between those of petals and sepals, and guard cells were observed among them (Figure S4).

Furthermore, we generated double mutants between pdf2-1 and alleles of the other HD-ZIP IV genes reported in Nakamura et al. (2006), but observed no homeotic conversions of floral organs in these mutants (Figure S5, and data not shown). pdf2-1 hdg11-1 hdg12-2 triple mutants showed no difference in flower morphology from pdf2-1 hdg12-2 (Figure 1).

pdf2-2 and atml1 alleles are not involved in the homeotic conversion of floral organs

In addition to pdf2-1, we obtained a new allele of PDF2, which has a T-DNA insertion in the exon for the more C-terminal part of the START domain than pdf2-1, and is named pdf2-2 (Figure S2). Interestingly, pdf2-2 had flowers lacking one of the two short stamens with higher frequency than pdf2-1, but pdf2-2 hdg2-3 showed no additional abnormality (Figure 1). pdf2-2 was also crossed with hdg1-1, hdg5-1 and hdg12-2 alleles, but these double mutants showed no homeotic conversions of floral organs, and manifested only the pdf2-2 phenotype, namely the lack of a short stamen (Figure 1).

As PDF2 is functionally redundant with ATML1 (Abe et al., 2003), we next examined double mutants between atml1 alleles and hdg1-1, hdg2-3, hdg5-1 or hdg12-2. atml1-1 and atml1-3 carry a T-DNA insertion in the exon encoding the SAD domain and the homeodomain, respectively (Figure S2; Abe et al., 2003; Roeder et al., 2012), but no combination of these showed homeotic changes in floral organs (Figure S5).

We further confirmed that hdg1-1 hdg2-3, hdg1-1 hdg5-1, hdg1-1 hdg12-2, hdg2-3 hdg5-1, hdg2-3 hdg12-2 and hdg5-1 hdg12-2 also showed normal development of flowers, except for an occasional lack of short stamens.

AP3 expression is reduced in mutant flowers

According to the ABC model, the homeotic conversion of petals into sepals and stamens into carpels is caused by a loss of function of the class-B floral identity genes (Coen and Meyerowitz, 1991). Thus, we examined expression levels of the two class-B genes, AP3 and PI, in the above-described phenotypic mutants. AP3 expression was considerably reduced in pdf2-1 hdg2-3, pdf2-1 hdg1-1 and pdf2-1 hdg5-1 inflorescences, and was moderately reduced in pdf2-1 hdg12-2 (Figure 5), suggesting a correlation of the severity of the homeotic phenotype with the level of AP3 expression. On the other hand, reduction in PI expression was detected only in pdf2-1 hdg1-1 and pdf2-1 hdg2-3 inflorescences (Figure 5). Expression levels of the class-A floral identity gene AP2, the class-C gene AG and the class-E gene SEP3, which are also required for specifying the organ identity of petals and stamens (Honma and Goto, 2001; Theißen, 2001), were not reduced in these mutant inflorescences (Figure S6).

Figure 5.

Expression analysis of AP3 and PI in mutant inflorescences. Real-time RT-PCR was performed using inflorescence apices containing floral buds younger than stage 13. Error bars indicate standard deviations of three independent samples. Asterisks indicate a significant difference from the wild type (Student's t-test: *P < 0.05; **P < 0.1).

Real-time RT-PCR experiments using RNA prepared from flower buds at different stages revealed that the level of AP3 expression was much lower in pdf2-1 hdg1-1 and pdf2-1 hdg2-3 than in the wild type throughout different stages of flower development, whereas the level of PI expression gradually decreased in these mutants at later stages (Figure 6a). However, in situ hybridization experiments revealed that, as is the case with wild-type flowers (Smyth et al., 1990; Jack et al., 1992), the AP3 transcript was detected in the mutant flowers at stage 3 (Figure 6b–e), suggesting that the onset of AP3 expression is unaffected in these mutants. After stage 3, the AP3 transcript was detected only in the proximal part of petals and stamens (Figure 6d, arrows). PI expression is detected in whorls 2, 3 and 4 of wild-type flowers at stage 3, and then restricted to whorls 2 and 3, and is maintained together with AP3 expression throughout the development of petals and stamens (Goto and Meyerowitz, 1994). PI expression patterns in pdf2-1 hdg1-1 and pdf2-1 hdg2-3 flowers were similar to those in wild-type flowers (Figure 6f–i). As the initiation of AP3 expression involves meristem identity genes such as LEAFY (LFY), AP1 and UNUSUAL FLORAL ORGANS (UFO) (Weigel and Meyerowitz, 1993; Levin and Meyerowitz, 1995; Lee et al., 1997), we examined the expression patterns of these genes and confirmed that they were not altered in pdf2-1 hdg2-3 and pdf2-1 hdg1-1 (Figure S6).

Figure 6.

Temporal and spatial expression patterns of AP3 and PI. (a) Temporal expression of AP3 and PI during flower development. Flowers were sorted into four groups according to their size. More than two biological replicates, each containing flowers from more than 150 independent inflorescences, were analyzed. (b–e) In situ hybridization of AP3 in the wild type (b), in pdf2-1 hdg1-1 (c) and in pdf2-1 hdg2-3 (d). Arrows indicate signals at presumptive whorls 2 and 3 at stage 3 (d), and arrowheads indicate signals at the basal part of petals and stamens at stage 5 (left) and stage 6 (right) (c). Wild-type sections were also examined with the AP3 sense probe (e). (f–i) In situ hybridization of PI in the wild type (f), in pdf2-1 hdg1-1 (g) and in pdf2-1 hdg2-3 (h). Wild-type sections were also examined with the AP3 sense probe (i). (j–l) Confocal microscopy images of wild-type (j), pdf2-1 hdg1-1 (k) and pdf2-1 hdg2-3 (l, m) influorescnces expressing pAP3::GFP. An arrow indicates GFP fluorescence in a developing petal. Numbers indicate floral stages. Scale bars: 100 μm.

We also introduced the GFP gene driven by the AP3 promoter into pdf2-1 hdg2-3 and pdf2-1 hdg1-1. Similar to the pattern observed by in situ hybridization, the GFP fluorescence in wild-type plants transformed with the pAP3::GFP construct was detected in whorls 2 and 3 at stage 3, and was maintained in petals and stamens until the later stages of flower development (Figure 6j). In pdf2-1 hdg1-1 and pdf2-1 hdg2-3, the GFP signal appeared in the presumptive whorls 2 and 3 of floral meristems at stage 3 and, unlike in the wild type, the signal became restricted to the basal region of petals and stamens at stages 6–9 (Figure 6k–m). We observed only sporadic signals of GFP in the middle to upper part of the petal at stage 10 (Figure 6l), suggesting that AP3 is not uniformly expressed in these mutants.

Transgenic expression of AP3 or PDF2 rescues the phenotype

We examined whether or not the transgenic expression of AP3 could restore the development of petals and stamens in pdf2-1 hdg2-3. When the AP3 full-length cDNA was fused to the CaMV 35S promoter (35S::AP3) and introduced into pdf2-1 hdg2-3 plants, the gross appearance of petals and stamens were indistinguishable from that of the wild type (Figure 7a,b), and fertility was fully restored. Epidermal cells of the petals in pdf2-1 hdg2-3 carrying the 35S::AP3 construct were conical in the adaxial epidermis and uniformly rectangular in the abaxial epidermis (Figure 7c,d), like those in the wild type.

Figure 7.

Floral phenotype of plants transformed with 35S::AP3 or pPDF2::PDF2. (a, b) Inflorescences of pdf2-1 carrying 35S::AP3 (a) and pdf2-1 hdg2-3 carrying 35S::AP3 (b). (c, d) SEM images of adaxial surface (c) and abaxial surface (d) of a petal in pdf2-1 hdg2-3 carrying 35S::AP3. (e–h) Inflorescences of pdf2-1 carrying pPDF2::PDF2 (e), pdf2-1 hdg2-3 carrying pPDF2::PDF2, and showing a wild-type phenotype (f), and pdf2-1 hdg2-3 carrying pPDF2::PDF2, producing short petals (g, h). The sepals and petals were removed from a flower on the right-hand side (h). Scale bars: 5 mm (a, b, e–g); 10 μm (c, d); and 1 mm (h).

We also performed complementation experiments with the wild-type PDF2. The PDF2 full-length cDNA was fused to the native PDF2 promoter (pPDF2::PDF2) and introduced into pdf2-1 hdg2-3. We obtained five transgenic lines, and two of them showed normal development of flowers as pdf2-1 transformed with the same construct, but the other three developed flowers with smaller petals than wild-type flowers (Figure 7e–g). Even in these flowers, stamens were fully fertile and showed no carpelloid features (Figure 7h).

pdf2-1 transcripts are present in the epidermis

T-DNA insertion mutants of PDF2 and other HD-ZIP IV genes could produce incomplete transcripts from the respective alleles. The presence of the pdf2-1 transcript was examined by RT-PCR with specific primers designed to amplify the region that is upstream of the T-DNA insertion and encodes the HD-ZLZ domain. The resulting PCR products were detected at similar levels in the wild type and in the pdf2-1 hdg double mutants (Figure 8a). In situ hybridization using a probe for the 5′ untranslated region (5′–UTR) of PDF2 showed that the pdf2-1 transcript was expressed in the L1 cells of inflorescence and floral meristems, a pattern identical to that of the wild-type PDF2 transcript (Figure 8b). RT-PCR experiments were also performed using specific primers for the HD-ZLZ domain coding sequences of HDG1, HDG2, HDG5 and HDG12, and revealed that the transcripts from hdg1-1 and hdg5-2 are present, but that those from hdg1-2, hdg2-3 and hdg12-2 are not, in their respective double mutants with pdf2-1 (Figure 8a). As the T-DNA insertion in hdg1-1 is located in an intron of HDG1, the full-length HDG1 transcript could be produced but no RT-PCR products were amplified from hdg1-1 by using the primers encompassing the insertion site.

Figure 8.

Detection of HD-ZIP IV transcripts in pdf2 hdg mutants. (a) Semiquantitative RT-PCR of HD-ZIP IV genes in pdf2 hdg mutants. PCR was performed using gene-specific primers designed to amplify the region encompassing the HD-ZIP domain of each gene. (b) In situ hybridization of PDF2 in the wild type and in pdf2-1. The 5′ untranslated region (5′-UTR) of PDF2 mRNA was used as a probe. No signal was detected in the wild type with the sense probe. Scale bar: 100 μm.


HDG2 is involved in determining the identity of petals and stamens

Previous studies have reported that pdf2-1 and hdg2-3 single mutants show no obvious phenotype (Abe et al., 2003; Nakamura et al., 2006). We found, however, that the short stamens were occasionally absent in these mutants (Figure 1a). Furthermore, pdf2-1 hdg2-3 was shown to cause partial homeotic conversions of petals and stamens into sepals and carpels, respectively. As hdg2-3 carries a T-DNA insertion in the HD coding exon of HDG2, and no transcripts including this exon were detectable (Figure 7a), it may represent a loss-of-function mutation, suggesting that HDG2 together with PDF2 is required for the identity of petals and stamens.

We note that, although pdf2-2 also sometimes had a reduced number of stamens, pdf2-2 hdg2-3 showed no homeotic phenotype. The site of T-DNA insertion in pdf2-1 is located in the midst of the START domain coding sequence of PDF2, whereas that in pdf2-2 is in its C-terminal part. Thus, it is conceivable that the pdf2-1 allele, which could produce PDF2 truncated in the START domain, has some negative effects on the determination of stamen and petal identity.

We also observed that neither atml1-1 nor atml1-3 in combination with hdg2-3 showed homeotic flower defects. It is possible that, unlike PDF2, ATML1 does not play a role in determining floral organ identity. Alternatively, because these atml1 alleles do not represent the mutant disrupting the START domain of ATML1, like pdf2-1, their defects might be recovered by PDF2.

HDG1, HDG5, and HDG12 also function in flower development

hdg1-1 has a T-DNA in an intron interrupting the START domain-coding sequence of HDG1. hdg1-1 also often had a reduced number of the short stamens, whereas hdg1-2, which carries a T-DNA insertion in the ZLZ coding region, and is likely to represent a loss-of-function allele, had normal flowers. These results suggest that hdg1-1 also has some negative effects on floral organ development. The severe phenotype, such as carpelloid stamens and the lack of anthers, in pdf2-1 hdg1-1 may reflect synergistic effects of these alleles. On the other hand, the reduction in the number of the short stamens in pdf2-1 was enhanced by hdg1-2 rather than by hdg1-1 (Figure 1a). This might be because of the only moderate downregulation of AP3 in pdf2-1 hdg1-2 (Figure 5a). Indeed, a very weak mutant allele of AP3, ap3-11, produces fertile stamens but in reduced numbers (Yi and Jack, 1998). However, the possibility should not be excluded that the effect of pdf2 and hdg mutations on the number of stamens is through the modulation of other factors than AP3. Taken together, we suggest that HDG1 and PDF2 play a cooperative role in the formation of stamens. The preferential expression of HDG1 in the epidermis of developing stamens also supports that HDG1 functions in stamens (Figure S3). However, given the fact that neither hdg1-1 hdg2-3 nor hdg1-2 hdg2-3 showed any additional phenotype, the role of HDG1 might be subsidiary to that of PDF2 in floral organ development.

We also observed homeotic conversions of petals and stamens in pdf2-1 hdg5-1. hdg5-1 carries a T-DNA insertion in the middle of the START domain coding exon of HDG5, suggesting that the property of hdg5-1 is also similar to that of pdf2-1; however, the effect of hdg5-1 may be moderate compared with that of pdf2-1 and hdg1-1, because hdg5-1 single mutants had normal flowers. Furthermore, hdg5-1 had no additional effect on the phenotype of hdg2-3 and hdg1-1, suggesting again that PDF2 plays a principal role among the HD-ZIP IV family in flower development. As hdg5-2, in which the START domain coding region is predicted to be intact, showed no homeotic phenotypes in the double mutant with pdf2-1, it is likely to represent a weaker allele. It is also possible that again the START domain has a negative regulatory role in the activation of HD-ZIP IV proteins.

The T-DNA insertion in hdg12-2 is located in the middle of the ZLZ domain coding exon of HDG12, suggesting that hdg12-2 is a loss-of-function allele. Thus, the relatively weak phenotype observed in pdf2-1 hdg12-2 flowers suggests that HDG12 is less involved in flower development than HDG2.

Among the other HD-ZIP IV mutants investigated, pdf2-1 hdg3-1 showed an increased frequency of infertile and filamentous stamens, whereas the hdg3-1 single mutant was identical to the wild type (Figure S5). This implies that HDG3 is also involved in the proper development of stamens together with PDF2, and is consistent with a previous report suggesting that HDG3 functions in anther dehiscence (Li et al., 2007).

pdf2-1 has non-cell-autonomous effects on AP3 expression

The reduction of AP3 expression was not limited to epidermal cells of the floral meristem in pdf2-1 hdg2-3, whereas the pdf2-1 transcript was detected specifically in the epidermis of the inflorescence meristem (Figures 6d,l,m and 8b). This suggests that pdf2-1 could have non-cell-autonomous effects on the determination of floral organ identity, although the possibility that the pdf2-1 gene product directly binds to the AP3 promoter and represses its expression cannot be excluded. In maize, the fusion protein of the ZmOCL1 HD-ZLZ domains with the repressor domain of the Drosophila Engrailed protein reduces the expression of GA20 oxidase and probably the gibberellin (GA) content (Khaled et al., 2005). Synthesis of functional GAs may be required for the maturation of petals and stamens, along with the promotion of AP3 expression (Yu et al., 2004; Plackett et al., 2011). Although no GA-deficient mutants cause abnormalities in floral organ identity (Goto and Pharis, 1999), evidence of a link between GA signaling and the floral homeotic genes is mounting (Plackett et al., 2011). It might be possible that GA synthesis is reduced in the epidermis of pdf2-1 hdg2-3 flowers, thereby leading to a reduced expression of AP3 non-cell-autonomously. There are some reports suggesting the existence of epidermis-derived non-cell-autonomous signals that regulate whole organ growth, but the identity of these signals remains elusive (Savaldi-Goldstein et al., 2007; Savaldi-Goldstein and Chory, 2008). Anastasiou et al. (2007) proposed a possibility that KLU, a cytochrome P450 monooxygenese, can modify fatty-acid-related molecules and regulate organ growth non-cell-autonomously. Considering that epidermis-expressed HD-ZIP IV proteins regulate the expression of genes related to lipid metabolism, such as BDG, FDH and LTP (Abe et al., 2003; Wu et al., 2011), it is also possible that extracellular lipid composition is important for non-autonomous signaling, and is greatly affected in the epidermis of floral organs in the double mutants with the homeotic phenotype.

AP3 expression is maintained by the direct interaction of the AP3/PI heterodimer with the AP3 promoter region in an autoregulatory fashion (Hill et al., 1998; Tilly et al., 1998; Lamb et al., 2002). PI expression was less affected in the double mutants with the homeotic phenotype than AP3 expression, and only gradually decreased as flowers matured (Figure 5j). Thus, the inhibitory effect of pdf2-1 and/or hdg mutations may not be on the autoregulatory mechanism of AP3/PI expression, but may instead be specific to AP3.

In conclusion, our results suggest that cooperative functions of PDF2 with at least HDG1, HDG2, HDG5 and HDG12 in the epidermis are involved in determining the identity of petals and stamens; however, because of the limitations of mutant alleles available for analysis, the reasons for the phenotypes detected in the double mutant combinations remain not fully understood. There might also be additional HD-ZIP IV members that participate in floral organ formation, although they could have been missed because of the lack of appropriate mutant alleles. For future work, it will be necessary to generate further multiple mutant combinations and RNAi-induced knock-down plants to define the precise roles of each gene.

Experimental Procedures

Plant material and growth conditions

Arabidopsis thaliana ecotype Columbia-0 (Col-0) was used as the wild type. Most of the T-DNA insertion alleles have been described previously (Abe et al., 2003; Nakamura et al., 2006). atml1-3, pdf2-2 and hdg5-2 alleles were derived from the SALK insertion line, SALK_033408 (Roeder et al., 2012), the SAIL lines, SAIL_503_E09 and SAIL_238_A05, respectively, obtained from ABRC ( All seeds were surface-sterilized and cold-treated for 2–3 days at 4°C in the dark to synchronize germination. Plants were grown on rockwool cubes embedded in vermiculite under continuous light at 23°C. Genotypes of double mutants were determined by PCR of genomic DNA with the gene-specific primers described in previous studies (Abe et al., 2003; Nakamura et al., 2006; Table S1).

Anatomy of floral organs

For anatomy, early-formed (first to tenth) flowers on the primary inflorescence were examined under stereomicroscopy (≥ 15). For cross sections, flowers were fixed in FAA [45% (v/v) ethanol, 5% (v/v) formaldehyde and 5% (v/v) acetic acid] for 1 h, dehydrated in an ethanol series and embedded in Technovit 7100 (Kulzer, according to the manufacturer's instructions. Sections of 5 μm thick were cut with a microtome and stained with 0.05% (w/v) toluidine blue O in 1% (w/v) sodium tetraborate solution.

For SEM, flowers were fixed in FAA for 1 h at 23°C and dehydrated in an ethanol series. After exchange of ethanol to isoamyl acetate by a graded ethanol–isoamyl acetate series, samples were critical-point dried in liquid CO2, coated with platinum and viewed on a SEM JEOL JSM-6510LV (JEOL,

Confocal microscopy

The apical part of the primary inflorescence was embedded in 7% agar and sectioned with a vibratome. Confocal microscopy was performed on an inverted fluorescent microscope IX70 (Olympus, equipped with a confocal unit. Fluorescence of pAP3::GFP was exited with a wavelength of 488 nm and detected using a CCD camera through a 500–550-nm band-pass filter. Autofluorescence of chlorophyll was detected at 615–680 nm.

Plasmid construction

For making pAP3::GFP, the AP3 promoter of 830 bp in length was amplified by PCR using Ex Taq polymerase (TaKaRa, and cloned into pGEM-T Easy vector (Promega, Subsequently, the fragment was cut with HindIII and BglII, and cloned into the pBI101-based Ti plasmid vector that carries a GFP reporter gene (Matsuhara et al., 2000). For making 35S::AP3, full-length AP3 cDNA was amplified by RT-PCR with total RNA extracted from wild-type flowers, cloned into pGEM-T Easy vector, cut with BamHI and then cloned into the BamHI site of pBI121 (Clontech, The resulting 35S::AP3 fusion gene was cloned as a HindIII/EcoRI fragment into pBarMAP vector (Breuninger et al., 2008). For pPDF2::PDF2 construction, the PDF2 promoter sequence of 1591 bp was amplified from wild-type genomic DNA and the full-length PDF2 cDNA was amplified by RT-PCR. After the resulting fragments were respectively cloned into pGEM-T Easy, the XbaI/ClaI fragment of the PDF2 promoter and the ClaI/BglII fragment of the PDF2 cDNA were sequentially cloned into the modified pBarMap. The primer pairs used are listed in Table S1.

Plant transformation

The Ti plasmids were introduced into Agrobacterium tumefaciens C58C1 by electroporation. For pAP3::GFP plants, Col-0 plants were transformed by the floral-dip method (Clough and Bent, 1998). The pAP3::GFP transgene was further introduced into pdf2-1 hdg2-3 and pdf2-1 hdg1-1 mutants by crossing. Transformation with 35S::AP3 and pPDF2::PDF2 constructs was performed by floral spray methods (Chung et al., 2000), and the transformants were selected by spraying a 1 : 10 dilution of BASTA.

RNA extraction and RT-PCR

Total RNA was prepared with the RNeasy plant mini kit (Qiagen, First-strand cDNA was synthesized from 2 μg of total RNA with an oligo(dT) primer using the PrimeScript RT-PCR kit (TaKaRa). Real-time PCR was performed using the DNA Engine Opticon2 System (Bio-Rad, with KAPA SYBR Fast qPCR kit (KAPA Biosystems, Gene-specific primers are listed in Table S1. The ACTIN8 (ACT8) gene was used as an internal control to normalize the expression data for each gene. For the detection of the HD-ZIP IV transcripts in double mutant combinations, each cDNA fragment was amplified by 22 cycles of PCR for ACT8, 26 cycles for HDG1, HDG2, HDG5 and HDG12, and 28 cycles for PDF2, respectively. The PCR conditions were 94°C for 2 min, followed by cycles of 94°C for 30 sec, 53°C for 30 sec, and 72°C for 60 sec.

In situ hybridization

Inflorescences were fixed in FAA, dehydrated in an ethanol series, passed through a xylene–ethanol series, embedded in Paraplast (Sigma-Aldrich, and sectioned at 7 μm thickness. Gene-specific RNA probes for AP3 and PI were prepared as described previously (Jack et al., 1992; Goto and Meyerowitz, 1994). For the pdf2-1 transcript, a PDF2 cDNA fragment was amplified with the primers used for semi-quantitative RT-PCR (Table S1) and cloned into pGEM-T Easy. The fragment was then transferred to the EcoRI site of pBluescript SK + (Stratagene, now Agilent,, and linearized with BamHI and XhoI to prepare antisense and sense probes, respectively. Probe labeling, hybridization and immunological detection were performed as previously described (Abe et al., 1999).


We are grateful to Minako Ueda for kindly providing the pBarMap vector, Hiroyasu Motose for useful advice and Hiroo Fukuda for supporting our experiments using confocal microscopy. This work was supported by Grants-in-Aid for Scientific Research on Priority Areas (grant nos 21027028 and 23012032) from the Ministry of Education, Culture, Sports, Science and Technology, Japan. We have no conflicts of interest to declare.