Reactive oxygen species mediate conical cell shaping in Arabidopsis thaliana petals

Plants have evolved diverse cell types with distinct sizes, shapes, and functions. For example, most flowering plants contain specialized petal conical epidermal cells that are thought to attract pollinators and influence light capture and reflectance, but the molecular mechanisms controlling conical cell shaping remain unclear. Here, through a genetic screen in Arabidopsis thaliana, we demonstrated that loss-of-function mutations in ANGUSTIFOLIA (AN), which encodes for a homolog of mammalian CtBP/BARs, displayed conical cells phenotype with wider tip angles, correlating with increased accumulation of reactive oxygen species (ROS). We further showed that exogenously supplied ROS generated similar conical cell phenotypes as the an mutants. Moreover, reduced endogenous ROS levels resulted in deceased tip sharpening of conical cells. Furthermore, through enhancer screening, we demonstrated that mutations in katanin (KTN1) enhanced conical cell phenotypes of the an-t1 mutants. Genetic analyses showed that AN acted in parallel with KTN1 to control conical cell shaping. Both increased or decreased ROS levels and mutations in AN suppressed microtubule organization into well-ordered circumferential arrays. We demonstrated that the AN-ROS pathway jointly functioned with KTN1 to modulate microtubule ordering, correlating with the tip sharpening of conical cells. Collectively, our findings revealed a mechanistic insight into ROS homeostasis regulation of microtubule organization and conical cell shaping.

Introduction protein [30][31][32][33], caused enhanced conical cell phenotypes of an-t1. AN acted in parallel with KTN1 to control conical cell tip sharpening. Together, our results suggest that the AN-ROS pathway jointly functions with KTN1 to modulate microtubule ordering and conical cell shaping.

Loss of AN function causes wide-angled tips of conical cells, correlating with increased accumulation of ROS
To identify regulatory components involved in controlling conical cell expansion in the petal blade epidermis, we screened more than 500 mutant homozygous lines from the Arabidopsis Biological Resource Center (ABRC) for mutants with abnormal conical cell phenotypes [21]. From this screen, we identified one mutant line, SALK_026489C, previously named an-t1 [34], with a T-DNA insertion causing a null allele of the gene AN (AT1G01510) (S1A and S1B Fig).
The an-t1 mutant showed reduced tip sharpening of conical cells, exhibiting a phenotype of conical cells with wider tip angles (Fig 1A-1E). As shown in Fig 1A, we quantified cone parameters (cone angles and cone heights) of conical cells from wild type with normal conical tips and mutant with swollen conical tips. Quantification data showed that conical cells of the an-t1 mutant had increased cone angles but no alternation in cone heights compared with those of the wild type (Fig 1A-1E). To further confirm the role of AN in conical cell shaping, one additional T-DNA insertion null mutant of the AN gene, an-t2 (WiscDslox329F05) (S1A and S1B Fig), was obtained from the ABRC. The an-t2 mutant exhibited a conical cell phenotype similar to an-t1 (Fig 1A-1E). Next, we performed the complementation experiment for an-t1 mutant line by introducing the coding sequence of the AN gene fused with the GFP tag under the control of AN's promoter into an-t1. We obtained more than 30 transgenic lines that could fully complement the previously reported an-t1 mutant phenotypes [23,24], in terms of narrow cotyledons and leaves, less-lobed pavement cells, and reduced trichome branches. One transgenic line, an-t1 COM #1, was selected for phenotypic analyses (Figs 1A-1E and S1C-S1F), and was shown to express the AN gene fused with GFP (S1G Fig). Expression of AN pro ::GUS showed strong signals in petals throughout petal developmental stages (S1H Fig). Furthermore, conical cells of the an-t1 mutant had similar geometric shape as the wild type at petal development stages 8-10, but displayed increased tip angles at petal development stage 11 and beyond compared with the wild type (S1I-S1K Fig). Taken together, these results demonstrated that AN plays a role in promoting the tip sharpening of conical cells at late developmental stages.
Given that a previous report has shown that loss of AN function causes accumulation of high ROS levels in leaves by an unknown mechanism [34], we investigated whether the widerangled tips of conical cells of an mutants attributed to this abnormal ROS accumulation. Thus, we compared ROS levels between wild type and an mutants. ROS contain many molecules, with superoxide radical (O 2 • -), the precursor for most other ROS, and hydrogen peroxide (H 2 O 2 ), being the two major components [35]. We investigated O 2 •and H 2 O 2 distribution using nitroblue tetrazolium (NBT) [36] and 3,3'-diaminobenzidine (DAB) staining [37], respectively. We found that both the an-t1 and an-t2 mutants accumulated higher levels of O 2 •and H 2 O 2 in their petals compared with those of the wild type (S2A and S2B Fig). Furthermore, we examined the O 2 •and H 2 O 2 distribution at the cellular resolution. Using the fluorescent dye dihydroethidium (DHE) and 2',7'-dichlorofluorescein diacetate (CM-H2DCFDA) [38,39] to monitor O 2 •and H 2 O 2 production by laser scanning confocal microscopy, respectively, we found that both the an-t1 and an-t2 mutants' mature conical cells had significantly higher levels of O 2 •and H 2 O 2 compared with the wild type (Fig 1F-1I). Moreover, the an-t1 COM #1 complementation line could fully rescue the ROS levels of an-t1 (Fig 1F-1I). Interestingly, qRT-PCR analysis revealed that the AN expression was down-regulated by H 2 O 2 For the quantification of cone height, ns indicating no significant difference (Mann-Whitney U test, P > 0.05) between the data sets from an-t1, an-t2 and an-t1 COM 1# compared with WT Col-0 (P = 0.2033, P = 0.1293, and P = 0.51330, respectively). For the quantification of cone angle, asterisks indicate a significant difference (Mann-Whitney U test, ��� P < 0.001) between the data sets from an-t1 and an-t2 compared with WT Col-0 (P = 0.0009, P = 0.0006, respectively), and ns indicates no significant difference (Mann-Whitney U test, P > 0.05) between the data sets from an-t1 COM 1# compared with WT Col-0 (P = 0.7822 , asterisk indicates a significant difference (Mann-Whitney U test, � P < 0.05) between the data sets from an-t1 and an-t2 compared with WT Col-0 (P = 0.024, P = 0.024, respectively), and ns indicating no significant difference (Mann-Whitney U test, P > 0.05) between the data sets from an-t1 COM 1# compared with WT Col-0 (P = 0.6423). For the quantification of average fluorescent intensity of CM-H2DCFDA-detected H 2  To investigate whether the AN-ROS pathway is a general mechanism important for cell shape in many different cell types, we next compared ROS levels in leaf pavement cells and trichomes between an mutants and the wild type. The an-t1 mutant accumulated higher levels of O 2 •and H 2 O 2 in their leaf pavement cells and trichomes compared with those of the wild type (S6 Fig). These findings suggested that AN plays a general role in regulating ROS levels in diverse cell types.

ROS homeostasis is required for normal conical cell expansion
The observations that mature petals of an mutants displayed reduced tip sharpening of conical cells and increased ROS levels raise the hypothesis that ROS act downstream of AN to modulate conical cell expansion in petal adaxial epidermis. To test this hypothesis, we next sought to determine whether exogenously supplied ROS was sufficient to generate conical cells with wider tip angles similar to those of an mutants. We then used 50

Endogenously increased ROS levels by Rho GTPase ROP2 overexpression suppresses tip sharpening of conical cells
To further confirm the role of ROS in modulating conical cell shape, we investigated whether endogenously increased ROS levels could also lead to the phenotype of wide-angled tips of conical cells. The plant Rho GTPase, which functions as a central molecular switch to control diverse cellular processes [40], has been shown to positively regulate endogenous ROS levels through directly binding to NADPH oxidase [41]. Thus, we investigated conical cell phenotypes in transgenic lines overexpressing ROP2. In agreement with the role of Rho GTPase in ROS production [41], analyses of petals in transgenic Arabidopsis plants overexpressing ROP2 (ROP2 OX) or a constitutively active mutant ROP2 gene (CA-ROP2) [42] under the control of the 35S promoter showed significantly increased ROS levels throughout the late phases of petal development (S9 Fig). In agreement with the observation that wide-angled tips of conical cells are correlated with accumulation of high levels of ROS ( Fig 3A-3D), mature petals from both ROP2 OX and CA-ROP2 plants displayed swollen conical cells, with higher levels of both O 2 •and H 2 O 2 ( Fig   3E-3H). However, this phenotype was more severe than the an-t1 and an-t2 mutants. This may reflect that ROP GTPase signaling acts through multiple downstream components [40].

Identification of potential AN interactors
Given that both an mutants and CA-ROP2 plants exhibited a similar conical cell phenotype, correlating with accumulation of high ROS levels, we therefore explored the relationship between AN and ROP2. Thus, we tested whether AN functions to affect ROS production by either directly or indirectly activating ROP2. We then examined ROP2 activity in the inflorescences of wild type and an mutant lines. A comparison of the activation status of ROP2 in the wild type and the mutants an-t1 and an-t2 showed no significant differences (S10 Fig), whereas the spk1-4 mutant with a mutation in SPIKE1 as a control had reduced ROP2 activity as described previously (S10 Fig) [43]. This result suggested that AN may not act upstream of ROP2 in regulating ROS production.
To investigate the molecular mechanism by which AN regulates ROS levels, we identified potential interacting proteins of GFP-AN using a GFP-trap-immunoprecipitation based approach coupled with mass spectrometry (MS)-based protein identification. Proteins were isolated from whole-inflorescence tissues, excluding mature flowers and developing siliques, of a transgenic line expressing AN-GFP under the 35S promoter to pull-down GFP-AN protein complexes using GFP-binding agarose beads. Protein samples pulled-down from a transgenic line expressing GFP alone under the 35S promoter were used as a control. Co-immunoprecipitation (Co-IP)-MS analyses were conducted from three independent harvests of inflorescence tissues from both p35S:: GFP-AN line and p35S::GFP line to compose a biological triplicate. Gel analysis of the pull-down samples demonstrated a high specificity and efficiency for GFP-AN protein enrichment (S11 Fig). Pulled-down protein samples were analyzed by MS, and stringent data analyses identified proteins that are exclusive in all three GFP-AN replicates but not present in any of the control pull-downs. To identify proteins present in GFP-AN complexes that can be relevant to a role in ROS production, we chose targets already annotated with gene ontology (GO) biological processes related to oxidation-reduction process and oxidoreductase. Notably, among the identified AN-interacting proteins (S1 Table), we obtained 49 proteins directly responsible for ROS homeostasis (S1 Table). These ROS-related proteins include CATALASE 2 (CAT2) and CATALASE 3 (CAT3), which can catalyzes the breakdown of H 2 O 2 [44]. In addition, several NAD(P) superfamily proteins involved in the oxidation-reduction process were identified as putative AN interactors (S1 Table). To confirm the role of CAT2 in regulating conical cell shape, we identified one mutant line cat2-1 (SALK_076998) from the ABRC stock center, with a T-DNA insertion causing a null allele of the gene CAT2 (AT4G35090) (S12A and S12B Fig). The cat2-1 mutant exhibited a phenotype of conical cells with wider tip angles compared with the wild type (S12C-S12F Fig). We next sought to investigate whether AN and CAT2/CAT3 could physically interact. We used co-IP experiments to detect their interactions in vivo. We generated transgenic lines expressing CAT2-GFP and CAT3-GFP, respectively, and used GFP-Trap agarose beads to pull-down protein samples for the co-IP analyses. However, in our experimental conditions, the western-blotting results showed that AN may not physically interact with CAT2 and CAT3 in planta (S13 Fig). Although the direct interactions between AN and ROS-related proteins need to be further determined, the list of potential AN interactors may suggest a regulation of ROS production by AN.

Mutations in KTN1 result in enhanced conical cell phenotypes of an-t1
To identify genetic components that function together with AN in controlling conical cell expansion, we mutagenized an-t1 seeds with ethyl methanesulfonate (EMS) and conducted a genetic screen for mutants with enhanced conical cell expansion defects of an-t1. From this screen, we identified two enhancer mutants. Backcrosses to an-t1 and subsequent genetic analyses revealed that these two mutants were allelic and both harbored a recessive mutation. Sequencing of the KTN1 genomic DNA in these two mutants revealed a G-to-A mutation, resulting in amino acid alterations in the conserved ATPase domain of the KTN1 genes ( Fig 4A). Thus, these two mutants were named ktn1-7and ktn1-8, respectively. Notably, these mutants displayed short roots, compact rosette leaves, and dwarf plants (S14A-S14C Fig), reminiscent of the ktn1-4 mutant phenotypes [45]. The ktn1-8 allele had a weaker phenotype compared with those of the ktn1-4 allele (S14A-S14C Fig), suggesting that it is a novel weak allele. A transgenic line expressing a KTN1 pro ::GUS construct showed strong GUS signals in petals throughout petal development stages (Fig 4B), consistent with the role of KTN1 in petal conical cell shaping.
Conical cells of the ktn1-7 and ktn1-8 single mutants had wider tip angles than the wild type, but displayed similar phenotypes to an-t1 (Figs 4C-4F and S14D-S14G). Notably, both the an-t1 ktn1-7 and an-t1 ktn1-8 mutants had dramatically enhanced cell expansion defects, exhibiting extremely swollen petal adaxial epidermal cells, with a hemispherical shape instead of a conical shape, compared with the wild type and even the an-t1 mutant (Figs 4C-4F and S14D-S14G). Taken together, these findings suggest that AN and KTN1 have a genetic interaction, and may act in parallel and functionally related processes during epidermal cell shaping.
We next sought to investigate whether AN and KTN1 could physically interact. We used coimmunoprecipitation experiments to detect their interactions in vivo by transiently coexpressing 35S::GFP-AN with 35S::KTN1-Myc in Nicotiana benthamiana leaves. Transiently coexpressed 35S::GFP and 35S::KTN1-Myc were used as negative controls. In our experimental conditions, the result showed that KTN1-Myc was not detected in either the immunoprecipitated AN-GFP complex or the GFP complex (S14H Fig), indicating that AN may not physically interact with KTN1 in planta. Consistently, we did not detect the KTN1 protein from our Co-IP MS analyses in the AN-GFP line.

The AN-ROS pathway cooperates with KTN1 to organize microtubule orientation during conical cell expansion
Recent reports have suggested that ROS influence cytoskeleton dynamics [46], and that H 2 O 2 directly activates the MAPK cascade to modulate the activities of MAP65 proteins, consequently affecting microtubule orientation [46][47][48][49], although the underlying molecular mechanisms remain to be further explored. We next tested whether ROS are essential for cortical For the quantification of cone height, ns indicates no significant difference (Mann-Whitney U test, P > 0.05) between the data sets compared with WT Col-0 (P = 0.2633, P = 0.11099, P = 0.31530, respectively). For the quantification of cone angle, asterisks indicate a significant difference (Mann-Whitney U test, �� P < 0.01) (for an-t1 vs WT, P = 0.00359; for an-t1 ktn1-7 vs WT, P = 0.00129; for ktn1-7 vs WT, P = 0.0017; for an-t1 ktn1-7 vs an-t1, P = 0.00223; for an-t1 ktn1-7 vs ktn1-7, P = 0.0022). Values are given as the mean ± SD of more than 300 cells from 20 petals.
https://doi.org/10.1371/journal.pgen.1007705.g004 microtubule organization during conical cell shaping. Firstly, we investigated whether exogenously supplied H 2 O 2 could lead to alterations in microtubule orientation using a microtubule reporter line expressing GFP-tagged α-tubulin 6 (GFP-TUA6) [50]. We used 100 mM H 2 O 2 to treat flower buds of stage 7 from inflorescences of the GFP-TUA6 reporter line, and the same treatment was repeated four additional times, with 24 hours between each application, to generate long-term H 2 O 2 effects. After treatments, we visualized microtubule arrays from the top view of the non-folded mature petals in confocal Z sections, allowing for the top view of conical cells. We quantified the microtubule alignment with "OrientationJ", a ImageJ plug-in, used for calculating the directional coherency coefficient of the fibers [51]. A coherency coefficient close to "1" represents a strongly coherent orientation of the microtubules. We found that adaxial epidermal cells without H 2 O 2 treatments had well-ordered circumferential microtubule arrays aligned perpendicular to the axis of conical outgrowth (Fig 5A and 5B), consistently with our previous report [21]. In contrast, adaxial epidermal cells of mature petals with H 2 O 2 treatments displayed randomly oriented microtubules with reduced coherency (Fig 5A and 5B). This result suggested that high levels of ROS accumulation inhibited microtubule ordering during conical cell shaping. Consistently with these findings, we observed randomly oriented microtubules in the mature conical cells of the CA-ROP2 line (S15A and S15B Fig), which was shown to accumulate higher ROS levels. We next investigated the effects of eliminating endogenous ROS on microtubule organization. Notably, we found that eliminating either O 2 •or H 2 O 2 led to mature conical cells with disordered microtubule arrays with reduced coherency (S15C and S15D Fig). Thus, these results suggested that ROS homeostasis mediated microtubule orientation into well-ordered circumferential arrays in petal conical cells, although the underlying mechanism remains to be further explored.
To test whether ROS play a general role in regulating cell morphogenesis and microtubule organization in different cell types, we next investigated the effects on microtubule arrangements in both cotyledon pavement cells and leaf trichomes after exogenous H 2 O 2 application. Notably, analysis of cotyledon pavement cell phenotypes after H 2 O 2 treatment showed that, H 2 O 2 application induced a phenotype of reduced lobe length in pavement cells, showing severe defects of interdigitated growth in a H 2 O 2 dose-dependent manner (S16A-S16D Fig).
It has been reported that microtubule arrays play crucial roles in leaf pavement cell shaping [52,53]. Consistently, H 2 O 2 application induced increased alignment of microtubules in pavement cells in a H 2 O 2 dose-dependent manner (S16E and S16F Fig). Exogenous H 2 O 2 treatment leaded to generate smaller leaves with shorter trichomes but no alternation of trichome branching (S17A-S17E Fig). Furthermore, H 2 O 2 treatment resulted in increased alignment of microtubules in trichomes (S17F and S17G Fig), which was similar to the result observed in leaf pavement cells. The observations that exogenous H 2 O 2 treatment resulted in disordered microtubule arrays in petal conical cells and increased microtubule ordering in both pavement cells and trichomes, respectively, suggested that ROS may play diverse roles in modulating microtubule organizations in different cell types, although the underlying molecular mechanisms need to be further investigated.
Based on the above-mentioned results showing that ROS played a role in mediating microtubule ordering, AN inhibited ROS production during conical cell development, and that AN and KTN1 acted in parallel pathways to modulate conical cell expansion, we hypothesized that AN acts through ROS to modulate microtubule orientation and that the AN-ROS and KTN1 pathways converge at a node to affect microtubule ordering during conical cell expansion. To test this hypothesis, we first compared microtubule organization patterns of the an-t1 mutant with the wild type. Microtubule arrays in wild-type cells became increasingly ordered over the course of conical cell development and displayed well-ordered circumferential arrays at stage 14 of petal development (Fig 5C-5D), consistently with our previous report [21]. In contrast, an-t1 mutant conical cells that were shown to accumulate high ROS levels had randomly oriented microtubule arrays with reduced coherency throughout early and late developmental stages (Fig 5C-5D), consistently with the observation that high levels of ROS accumulation caused wide-angled tips of conical cells with disordered microtubule arrays. Furthermore, in agreement with the previous report [21], the ktn1-4 mutant conical cells had randomly oriented microtubule arrays, similar to those observed in the an-t1 conical cells (Fig 5C-5D). Despite the conical cells of the an-t1 ktn1-4 double mutant displayed more severe defects than the single mutants, the an-t1 ktn1-4 double mutant conical cells displayed randomly oriented microtubule arrays, similar to those observed in the an-t1 or the ktn1-4 single mutant (S18 Fig). Given that it is well established that KTN1 directly affects microtubule ordering through its severing activity at both microtubule nucleation and crossover sites [54,55], KTN1 may function in microtubule orientation independently of ROS. As predicted, analysis of ROS accumulation in mature petals of the ktn1-4 mutant showed no significant differences as compared to the wild type (S19 Fig). Taken together, these results suggest that the AN-ROS pathway and KTN1 acted in parallel to modulate microtubule organization and conical cell shaping.

Discussion
ROS function as signaling molecules for organ growth and normal cellular processes such as cell growth and cell division and differentiation [26][27][28][29]. Our results provide definitive evidence for a role of ROS in modulating conical cell expansion of petal adaxial epidermal cells.
Previous reports showed that AN promotes pavement cell interdigitation in leaves, correlating with negatively regulating cortical microtubule ordering [23,24], although the detailed molecular mechanism is unclear. Our findings in this study showed that AN promotes tip sharpening of conical cells in petals, correlating with positively regulating microtubule ordering, and that AN negatively regulates ROS levels, which in turn affects microtubule organization. Notably, we demonstrated that AN also plays a negative role in regulation of ROS levels in leaf pavement cells and trichomes. Therefore, we proposed that the AN-ROS-microtubule pathway is a general mechanism important for cell shaping in many different cell types.
In contrast to the role of AN in negatively regulating ROS production, ROP2 plays a role in positively regulating ROS production during conical cell shaping. AN and ROP2 may act antagonistically to regulate ROS homeostasis, although the molecular mechanisms by which AN suppresses and ROP2 promotes ROS production, respectively, remain to be further explored. Therefore, we propose that an endogenous balance between ROS accumulation and removal must be achieved and tightly regulated to generate microtubule reorientation and normal conical cell expansion of petal adaxial epidermal cells. Notably, loss of KTN1 function does not result in alternation in ROS levels in conical cells, suggesting that KTN1-mediated microtubule re-orientation may act in parallel with ROS signals during conical cell tip sharpening.
Emerging evidence suggests that ROS and redox cues have effects on microtubule behaviors [46][47][48][49]. Consistently with these reports, we demonstrated that both H 2 O 2 treatment and endogenously increased ROS levels induced by either loss of AN function or ROP2 overexpression resulted in reduced microtubule ordering of the conical cell. Furthermore, the effects of diverse ROS types on cell wall properties have been well studied [46]. ROS play critical roles in both cell-wall stiffening and loosening by promoting the formation of crosslinks between cell wall polysaccharides and glycoproteins, or by cleaving cell wall polysaccharides, respectively. Therefore, based on these findings, we cannot rule out the possibility that ROS also play a role in conical cell shaping by directly influencing cell wall properties. How ROS-mediated signaling regulates microtubule orientation remains unclear and will require extensive research in the future.
Given that previous reports have shown that KTN1-mediated microtubule severing plays critical roles in promoting microtubule rearrangements in response to mechanical stress in both the A. thaliana shoot apical meristem and cotyledon pavement cells [56,57], it is possible that mechanical forces could generate a signal to induce microtubule orientation in a KTN1-dependent manner during conical anisotropic expansion of petal adaxial epidermal cells. Interestingly, cells can respond to mechanical signals generated by cell geometry, providing a pervasive feedback on growth [58]. Future studies should aim to uncover the role of mechanical forces during conical cell shaping.
On the basis of our findings, we proposed a model to explain the molecular mechanism underlying ROS-dependent microtubule orientation in the regulation of conical cell shaping (Fig 6). According to this model, the AN-ROS pathway cooperates with KTN1 to jointly reorient microtubules from random to well-ordered transverse arrays, which is critical for the tip sharpening of conical cells. We hypothesized that the re-orientation of microtubules from random to well-ordered arrays may orient the deposition of cellulose microfilaments and generate the cell wall reinforcement throughout conical cell development [12][13][14][15][16][17][18][19], consequently maintaining conical cell expansion and forming the final characteristic cell shape. Also, we hypothesized that mechanical forces could generate a signal to induce microtubule orientation. Conical cells could respond to mechanical cues generated by cell geometry, providing a feedback loop to define the final cell shape.

Plant materials and growth conditions
All A. thaliana seeds used in this study were of the Columbia-0 (Col-0) ecotype. The ktn1-4 (SAIL_343_D12), an-t1 (SALK_026489C), and an-t2 (CS851381) were obtained from the Arabidopsis Biological Resource Centre. The seeds were sown in petri dishes on Murashige and A model depicting the molecular basis of ROS-dependent microtubule orientation and conical cell tip sharpening. We propose that AN and KTN1 jointly function to modulate microtubule orientation and conical cell shaping. AN functions to suppress ROS production, respectively. Both increased and decreased ROS suppress microtubule orientation into well-ordered circumferential arrays. We hypothesized that mechanical forces could generate a signal to induce microtubule orientation, and that conical cells could respond to mechanical cues generated by cell geometry, providing a feedback loop. Skoog medium agar supplemented with 1% (w/v) sucrose. Plants were grown in a growth room at 22˚C under 16-hr light/8-hr dark cycles.

Gene constructs and generation of transgenic plants
The sequences of primers used in this study are listed in S2 Table. The full length coding sequences (CDS) of the AN gene was amplified and cloned into PH35S-GFP-GW to construct p35S::GFP-AN. For Co-IP-LC-MS/MS analysis, the Ti plasmid expressing recombinant GFP-AN protein was introduced into rdr6-11 plants [59], which suppresses gene silencing. For the complementation experiment, the AN promoter was amplified from wild-type genomic DNA, and was inserted at Hind III and Xba I sites of vector p35S::GFP-AN to replace the 35S promoter, generating pAN::GFP-AN. For GUS activity assays, the promoter of AN was amplified and cloned into pCAMBIA1301. For co-IP assay, KTN1 CDS was cloned into pGWB517 to generate 35S::KTN1-Myc. CAT2 and CAT3 CDS was cloned into pGWB505 to generate 35S::CAT2-GFP and 35S::CAT3-GFP, respectively. The Ti plasmid expressing recombinant proteins were introduced into rdr6-11 to generate stable transgenic lines.

Confocal microscopy
For confocal imaging of conical cells from the side, petal blades were folded back, allowing for the side view of conical cells, and stained with a solution containing 10 μg/ml propidium iodide for more than 10 min. Petal samples were imaged with a Zeiss LSM 880 confocal laser scanning microscope (excitation at 514 nm, emission 550-700nm). For live-confocal imaging of cortical microtubules, non-folded petals stably expressing GFP-TUA6 were imaged with a Zeiss LSM 880 confocal laser scanning microscope (excitation at 488, emission 500-570nm). Serial optical sections were taken at 0.5-μm increments with a 40 × water or 63 × oil lens, and then were projected on a plane (i.e. maximum intensity) using Zeiss LSM 880 software. For CM-H2DCFDA staining, petal samples were incubated in 50 mM phosphatic buffer solution (PBS, pH 7.4) containing 10 μM CM-H2DCFDA (Invitrogen, C6827) for 30 min, and then the samples were washed for three times with PBS, and observed with the Zeiss LSM 880 microscope (excitation 488 nm, emission 500-570 nm) or the Zeiss observer A1 inverted microscope. For Dihydroethidium (DHE) staining, petal samples were incubated into 50 mM PBS (pH 7.4) buffer solution containing 40 μM DHE (Sigma, D7008) for 30min, and then visualized with the Zeiss LSM 880 microscope (excitation 514, emission 520-600 nm) or the Zeiss observer A1 inverted microscope.

Scanning electron microscopy
Petals from flower development stage 14 were dissected and directly observed with a TM-3030 table-top scanning electron microscope (Hitachi).

Enhancer mutant screening of conical cell phenotype of an-t1
Approximately 10,000 seeds of an-t1 were mutagenized using ethyl methane sulfonate. M 2 seeds were harvested from self-fertilized M 1 plants individually, and M 2 lines were screened for enhanced an-t1 petal conical cells phenotypes. Among 2,000 independent EMS-mutagenized an-t1 lines, two candidate enhancers were identified and described in this study. Candidate mutants were backcrossed to an-t1 three times before further phenotypic analyses.

ROP2 activity assay
About 0.1 g of WT and mutant inflorescences were collected and frozen in liquid nitrogen, respectively. Total proteins were extracted using extraction buffer (25 mM HEPES, pH 7.4, 10 mM MgCl 2 , 10 mM KCl, 5 mM dithiothreitol, 5 mM Na 3 VO 4 , 5 mM NaF,1 mM phenylmethylsulfonyl fluoride, 1% Triton X-100, and protease inhibitor cocktail). Twenty micrograms of MBP-RIC1-conjugated agarose beads were added to the protein extracts and incubated at 4˚Cfor 2h on a rocker. The beads were washed four times in wash buffer (25 mM HEPES, pH 7.4, 1 mM EDTA, 5 mM MgCl 2 ,1 mM dithiothreitol, and 0.5% Triton X-100) at 4˚C. GTP-bound ROP proteins associated with the MBP-RIC1 beads were boiled and used for analysis by western blot with a ROP2 antibody that was generated against the peptides QFFIDHPGAVPITTNQG (Abicode, China).

LC-MS/MS
A. thaliana seedlings expressing GFP-AN (in rdr6-11 background) and, as control, GFP (35S promoter) were grown under continuous light in MS medium. Two grams 5-day-old seedlings were collected and ground in liquid N 2 , and total proteins were extracted with the buffer (25 mM Hepes-KOH, pH7.4, 10 mM MgCl 2 , 100 mM NaCl, 5mM NaF, 15% glycerol, 1% Triton X-100, proteinase inhibitor cocktail). After centrifuging at 14,000g for 15 min at 4˚C, the supernatant was mixed with GFP-Trap agarose beads (gta-100, Chromotek) and rotated at 4˚C for 4 hour. The immunoprecipitates were then separated in SDS-PAGE and digested with 0.025 mg/mL trypsin. The samples were put into a Thermo Scientific EASY trap column (100 μM × 2 cm, 5 μM, 100 Å, C18) for separation, and analyzed with Obitrap Fusion mass spectrometer (Thermo Finnigan, San Jose, CA). Each sample was analyzed for 60 min with a resolution of 120,000, the scanning range of 375-1500m/z, AGC target of 4e5 and injection time of 50 ms. Simultaneously, the MS2 scanning was performed with the following parameter: resolution (50,000), activation type (HCD), injection time (105ms), AGC target (1e5). The raw data operated with Proteome Discoverer 2.1 (Thermo Scientific) were searched against with protein database (TAIR10_pep_20101214.fasta), and processed with FDR (false discovery rate)�0.01 at both the peptide and protein level.
For the co-immunoprecipitation assays of AN and CAT2/CAT3, A. thaliana inflorescences expressing CAT2-GFP or CAT3-GFP (in rdr6-11 background) and, as control, GFP (35S promoter) were used for protein extraction with the buffer (25 mM Hepes-KOH, pH7.4, 10 mM MgCl 2 , 100 mM NaCl, 5mM NaF, 15% glycerol, 1% Triton X-100, proteinase inhibitor cocktail). After centrifuging at 14,000g for 15 min at 4˚C, the supernatants were mixed with GFP-Trap agarose beads (gta-100, Chromotek) and rotated at 4˚C for 4 hour. The bound proteins were eluted from the beads with 2×SDS-PAGE sample buffer by heating at 100˚C for 5 min, and analyzed by immunoblot. The primary antibody used was anti-AN (1:2,000). For the generation of Anti-AN antibody, the amino acid sequence from 280 to 490 a.a. of AN was amplified and cloned into expression plasmid pGEX-4T (GST tag), then transformed the vector into Escherichia coli for protein expression. The purified protein was injected into rabbits. Then, Antiserum was extracted and purified.