The ASYMMETRIC LEAVES Complex Employs Multiple Modes of Regulation to Affect Adaxial-Abaxial Patterning and Leaf Complexity OPEN

Flattened leaf architecture is not a default state but depends on positional information to precisely coordinate patterns of cell division in the growing primordium. This information is provided, in part, by the boundary between the adaxial (top) and abaxial (bottom) domains of the leaf, which are speci ﬁ ed via an intricate gene regulatory network whose precise circuitry remains poorly de ﬁ ned. Here, we examined the contribution of the ASYMMETRIC LEAVES (AS) pathway to adaxial-abaxial patterning in Arabidopsis thaliana and demonstrate that AS1-AS2 affects this process via multiple, distinct regulatory mechanisms. AS1-AS2 uses Polycomb-dependent and -independent mechanisms to directly repress the abaxial determinants MIR166A , YABBY5 , and AUXIN RESPONSE FACTOR3 ( ARF3 ), as well as a nonrepressive mechanism in the regulation of the adaxial determinant TAS3A . These regulatory interactions, together with data from prior studies, lead to a model in which the sequential polarization of determinants, including AS1-AS2, explains the establishment and maintenance of adaxial-abaxial leaf polarity. Moreover, our analyses show that the shared repression of ARF3 by the AS and trans -acting small interfering RNA (ta-siRNA) pathways intersects with additional AS1-AS2 targets to affect multiple nodes in leaf development, impacting polarity as well as leaf complexity. These data illustrate the surprisingly multifaceted contribution of AS1-AS2 to leaf development showing that, in conjunction with the ta-siRNA pathway, AS1-AS2 keeps the Arabidopsis leaf both ﬂ at and simple.


INTRODUCTION
The formation of a stable, precisely defined boundary between two distinct cell fates is a fundamental feature of plant and animal development. Such cell fate boundaries coordinate the further patterning and growth of the tissue or organ. Accordingly, failure to maintain a stable cell fate boundary can result in dramatic developmental defects. Likewise, adaxial-abaxial leaf polarity, which directs the acquisition of distinct cell fates within the leaf's adaxial/ top and abaxial/bottom domains, must be carefully controlled. The boundary between these domains drives the flattened outgrowth of the leaf (Waites and Hudson, 1995), and slight perturbations in either adaxial or abaxial identity cause progressive leaf curling with severe consequences for physiological function (Lang et al., 2004;Zhang et al., 2009;Nakata et al., 2012). Adaxial-abaxial leaf polarity in particular poses an unusual and mechanistically challenging problem, namely, how to create a stable boundary within the plane of a long and wide, but shallow, structure. This requires precise coordination of division and differentiation patterns for hundreds of cells throughout primordium development.
The two sets of transcription factors are expressed in complementary domains on the top and bottom side of the leaf, respectively. The positional information needed to delineate these domains is provided in part by the small RNAs miR166 and tasiR-ARF. The latter is generated through a specialized trans-acting small interfering RNA (ta-siRNA) pathway. In Arabidopsis, this pathway employs a dedicated miR390-ARGONAUTE7 (AGO7) complex to target TAS3A transcripts, which are then made double-stranded through the action of SUPPRESSOR OF GENE SILENCING3 and RNA-DEPENDENT RNA POLYMERASE6. These double-stranded RNAs are subsequently processed by DICER-LIKE4 into 21-nucleotide phased small RNAs, a subset of which, the tasiR-ARFs, go on to regulate ARF3 and ARF4 (reviewed in Chapman and Carrington, 2007). Importantly, while tasiR-ARF biogenesis is restricted to the two adaxial-most cell layers of developing primordia, movement of these small RNAs creates an abaxially dissipating gradient that through a possible dosedependent readout limits expression of the ARF3 and ARF4 targets to the bottom side of leaf primordia Skopelitis et al., 2012). Likewise, miR166 forms a gradient across the developing leaf, but this gradient has its maximum on the abaxial side (Juarez et al., 2004;Nogueira et al., 2007;Yao et al., 2009). miR166 guides the cleavage of HD-ZIPIII transcripts, limiting expression of these adaxial determinants to the top side of primordia (McConnell et al., 2001;Emery et al., 2003;Juarez et al., 2004).
In addition to the opposing activities of these small RNAs, the transcription factors at the center of the adaxial-abaxial polarity network are characterized by mutually antagonistic behavior (Waites and Hudson, 1995;Kerstetter et al., 2001;Eshed et al., 2001;Emery et al., 2003). This leads to mutual exclusivity of adaxial versus abaxial cell fates and contributes to clean separation at the domain level (Husbands et al., 2009). While mutual antagonism between polarity determinants was inferred from early genetic studies, the mechanistic basis for this remains largely unknown. The recent finding that the HD-ZIPIII and KAN proteins act oppositely on a common set of targets (Merelo et al., 2013;Reinhart et al., 2013;Huang et al., 2014) suggests one possible mechanism by which mutual antagonism may occur. However, the clean separation of cell fates and maintenance of a stable boundary likely requires a more immediate and direct mechanism. This could be achieved by direct repressive interactions between the polarity determinants themselves, for example, the direct repression of AS2, and possibly PHB, by KAN1 (Wu et al., 2008;Huang et al., 2014).
Ironically, mutual antagonism makes identifying such direct interactions between the transcription factors at the core of the polarity network more difficult, as the direct contributions of a given adaxial or abaxial determinant are not easily untangled from indirect effects related to general imbalances in the polarity network. Further compounding this, most polarity determinants have roles in development outside of adaxial-abaxial patterning. For instance, ARF3 and ARF4 act as repressors of the auxin-signaling network that affects a wide range of developmental processes Hunter et al., 2006;Marin et al., 2010). Likewise, AS1-AS2 maintains the stable silencing of meristem fate-promoting KNOX genes in developing primordia Lodha et al., 2013). Specifically, AS1-AS2 was shown to bind discrete ciselements in the promoters of BREVIPEDICELLUS (BP) and KNAT2, leading to recruitment of Polycomb complexes that stably repress KNOX gene expression in determinant cells. Such a silencing mechanism would fit within the established paradigm of mutual antagonism between adaxial and abaxial determinants.
With this in mind, we set out to map the contributions of the Arabidopsis AS pathway to adaxial fate acquisition using complementary biochemical and genetic approaches. We show that AS1-AS2 binds the promoters of both adaxial and abaxial determinants and exploits distinct mechanisms to regulate target activity. AS1-AS2 employs both Polycomb-dependent and -independent mechanisms to directly repress activity of the abaxial determinants MIR166A, YAB5, and ARF3 and further uses a nonrepressive, possibly protective, mechanism in the regulation of TAS3A on the adaxial side. These regulatory interactions, together with data from prior studies, suggest a model in which the sequential polarization of polarity determinants, including AS1-AS2, explains the establishment and maintenance of adaxialabaxial leaf polarity. Moreover, our analyses show that the shared repression of ARF3 by the AS and ta-siRNA pathways intersects with additional AS1-AS2 targets to affect multiple nodes in leaf development. Loss of ARF3 regulation by mutation of both the AS and ta-siRNA pathways leads to enhanced abaxialized defects reflecting the loss of multiple adaxial-promoting factors. Double mutants also show a dramatic and unexpected reduction in TAS3A transcript levels, predicting the presence of a feedback loop in which targets of tasiR-ARF, in conjunction with AS1-AS2 binding at the promoter of TAS3A, polarize tasiR-ARF biogenesis. Finally, the AS and ta-siRNA pathways converge to suppress leaf complexity via the specific activity of ARF3, a function that is not shared by ARF4. These data illustrate the surprisingly multifaceted contribution of AS1-AS2 to leaf development showing that, in conjunction with the ta-siRNA pathway, AS1-AS2 keeps the Arabidopsis leaf both flat and simple.

AS1-AS2 Directly Regulates Adaxial-Abaxial Polarity Determinants
To better understand the molecular basis for the mutual exclusivity of adaxial and abaxial cell fates, we set out to define the role AS1 and AS2 play in regulating adaxial-abaxial polarity. These transcription factors have been shown to bind cis-regulatory elements with consensus sequences CWGTTD (Motif I) and KMKTTGAHW (Motif II) in the promoters of BP and KNAT2 , prompting us to scan for similar sequence motifs in the regulatory regions of key polarity determinants. Putative AS1-AS2 targets were selected based on adherence to the following criteria: First, putative targets must include both Motifs I and II, but as the consensus sequences for these AS1-AS2 binding elements are based on data from just two targets, specifications for these motifs were relaxed to be as inclusive as possible of putative targets. As such, Motif I became CNGTTD, the consensus Myb binding site and, in line with most known plant transcription factor binding motifs (Franco-Zorrilla et al., 2014), Motif II was reduced to the core element KTTGAH. Second, Motifs I and II must be found on the same strand of DNA, with Motif I being located upstream of Motif II. Third, as the spacing between Motifs I and II at BP and KNAT2 is highly variable and the maximum functional distance between them is not known, a spacing limit of 175 nucleotides, three times the maximum distance between motifs I and II at BP or KNAT2, was used to maximize the number of potential targets. Fourth, even though AS1-AS2 occupy two pairs of binding sites in the BP and KNAT2 promoters , polarity genes containing a single binding site comprising Motifs I and II were still considered putative targets.
Using these criteria, we identified 34 potential AS1-AS2 binding sites in the regulatory regions of 18 core components of the adaxial-abaxial polarity network (Supplemental Table 1). To assess whether the AS1-AS2 complex occupies these cis-elements in planta, we utilized a transgenic line expressing an estradiolinducible translational AS2-YFP fusion (OlexA:AS2-YFP; Lodha et al., 2013) in chromatin immunoprecipitation (ChIP) assays. OlexA:AS2-YFP seedlings grown in the absence of estradiol appear completely wild-type (Supplemental Figure 1A) and show no YFP fluorescence, but upon induction with 5 µM estradiol, OlexA: AS2-YFP seedlings develop an adaxialized leaf phenotype and accumulate AS2-YFP RNA and protein (Supplemental Figures 1B  to 1D). ChIP assays on chromatin prepared from such seedlings showed significant enrichment of AS2-YFP at binding sites in the promoters of four of the predicted AS1-AS2 targets: ARF3, MIR166A, YAB5, and TAS3A ( Figure 1). ChIP analyses of mocktreated seedlings did not show this enrichment, substantiating that AS2 binds the promoters of key polarity determinants (Supplemental Figure 1F). Furthermore, consistent with data showing AS1 and AS2 operate as a complex (Lin et al., 2003;Guo et al., 2008), ChIP assays on seedlings expressing a functional AS1-HA fusion showed specific localization of AS1 to these ARF3, MIR166A, YAB5, and TAS3A promoter regions (Supplemental Figure 1E). Importantly, while a recent genome-wide analysis of AS1 activity identified ARF3 as a potential direct target (Iwasaki et al., 2013), the binding site proposed to mediate this regulation is distinct from the AS1-AS2 target site defined here. Moreover, the presence and position of the AS1-AS2 binding site identified in this study neatly explains the observation that deletion of the proposed AS1 target site from an ARF3:GUS transcriptional reporter does not phenocopy its pattern of misexpression in as1.
Considering that AS1 and AS2 promote adaxial identity, binding of these proteins to regulatory regions of the abaxial determinants ARF3, MIR166A, and YAB5 is conceptually consistent with the reported function of AS1-AS2 in a repressor complex and with mutual antagonism between adaxial and abaxial factors Lodha et al., 2013). However, a role for AS1-AS2 in the regulation of ARF3 will need to be more complex, as ARF3 is transcribed throughout developing leaf primordia and only becomes polarly localized in response to tasiR-ARF activity . Moreover, AS1-AS2 also binds the promoter of TAS3A. Expression of this polarity determinant on the adaxial side of leaves overlaps that of AS2 Iwakawa et al., 2007;Chitwood et al., 2009), further arguing that AS1-AS2 employs distinct mechanisms to regulate these targets.
Given this complexity, we sought to characterize the effect AS1-AS2 binding has on the regulation of these targets by examining expression in wild-type and as2-4 loss-of-function mutants ( Figure  2A). Transcript levels of YAB5 and MIR166A are significantly increased in as2, suggesting that AS1-AS2 acts to repress these abaxial determinants. To substantiate this finding, we generated a MIR166A:GUS reporter and compared its expression in the wild type and as2. Consistent with previous findings (Yao et al., 2009), the MIR166A:GUS reporter marks the vasculature as well as the abaxial epidermis of wild-type leaf primordia ( Figure 2B; Supplemental Figures 2A and 2B). While MIR166A expression in the vasculature is unchanged in as2, the MIR166A:GUS reporter is ectopically expressed in the rest of the blade, showing GUS activity in both the adaxial and abaxial epidermal layer of as2 primordia ( Figure 2C; Supplemental Figure 2B). Furthermore, deletion of the observed AS1-AS2 binding sites from the MIR166A:GUS reporter recapitulates this ectopic activity ( Figure 2D). It follows that AS1-AS2 acts directly at the MIR166A promoter to repress transcription in the adaxial epidermis of developing leaf primordia, thereby pointing to a role for these transcription factors in polarizing the spatial distribution of miR166, which in turn delineates the expression domain of the HD-ZIPIII adaxial determinants (Emery et al., 2003). Like MIR166A, YAB5 expression is restricted to the abaxial side of wild-type leaves (Iwakawa et al., 2007;Sarojam et al., 2010) and is upregulated in as2 mutants (Figure 2A), suggesting that AS1-AS2 binding to the YAB5 promoter may similarly contribute to the polarized activity of this abaxial determinant.
Consistent with the inference that AS1-AS2 is unlikely to act in a transcriptional repressor complex that defines the spatial expression domains of ARF3 and TAS3A, transcript levels for both these targets are unchanged in as2 leaf primordia ( Figure 2A). However, formation of a normal flattened leaf requires the maintenance of a stable adaxial-abaxial boundary throughout leaf development, presenting the possibility that AS1-AS2 contributes to the temporal regulation of ARF3 and TAS3A expression. To assess this possibility, we examined the effect of as2 on the expression of the ARF3:ARF3-GUS reporter and TAS3A gene trap in older seedling leaves Garcia et al., 2006). ARF3:ARF3-GUS activity in the wild type is strongest in young primordia, where accumulation is limited to the abaxial side ( Figure  2E; Chitwood et al., 2009). As the leaf matures, ARF3:ARF3-GUS activity subsides basipetally, with weak reporter activity persisting in the vasculature of differentiated leaves ( Figures 2E and 2G). By contrast, as2 retains ARF3:ARF3-GUS expression outside the vasculature, even in fully matured leaves ( Figures 2E and 2H). However, as in wild-type leaf primordia, ARF3:ARF3-GUS reporter activity in mature as2 leaves remains restricted to the abaxial side ( Figure 2H). Thus, AS1-AS2 also functions as a repressor complex ChIP analysis shows AS2-YFP occupies a subset of predicted AS1-AS2 binding sites in the promoters of both abaxial (dark blue) and adaxial (lavender) determinants. Significant enrichment is detected at ARF3, MIR166A, YAB5, and TAS3A. Quantitative PCR values (means 6 SE) are shown as fold enrichment over the ACT2 negative control (horizontal red line) and calculated from at least three independent biological replicates. Student's t test: *P < 0.05; ** P < 0.01. Diverse Regulatory Properties of AS1-AS2 3 of 15 at ARF3 but regulates its expression in a temporal rather than a spatial manner. Finally, in contrast to both ARF3 and MIR166A, the spatial and temporal pattern of TAS3A expression is not obviously changed in as2 ( Figures 2F, 2I, and 2J), indicating a more complex relationship between AS1-AS2 and this direct target.
MIR166A and YAB5 Accumulate H3K27me3 in an AS1-AS2-Dependent Manner AS1-AS2 mediates the stable repression of BP and KNAT2 in developing leaves via recruitment of Polycomb Repressive Com-plex2 (PRC2; Lodha et al., 2013). PRC2 deposits the H3K27me3repressive chromatin mark at target loci to ensure their correct spatiotemporal expression, thus suggesting a potential mechanism by which AS1-AS2 might regulate MIR166A, YAB5, and ARF3. To investigate this possibility, we assessed the levels of H3K27me3 present at these genes in developing leaves of 12-d-old seedlings. Maximum enrichment of H3K27me3 at BP and KNAT2 is seen between the proximal AS1-AS2 binding site and the transcription start site . With this in mind, we analyzed levels of H3K27me3 at multiple sites in the promoters of MIR166A, YAB5, and ARF3 ( Figure 3A). As expected, chromatin at the known PRC2 target SHOOTMERISTEMLESS (STM) is highly enriched for H3K27me3 in comparison to the levels of this histone mark at the actively transcribed ACTIN2 (ACT2) gene ( Figure 3B; Lodha et al., 2013). A significant enrichment for H3K27me3 was also detected at all sites tested along the promoters of MIR166A and YAB5 ( Figure  3B). Levels of this silencing mark are lower at MIR166A than STM, perhaps reflecting the fact that STM activity is silenced in all cells of the primordium, whereas repression of MIR166A is polar and may be limited to the adaxial epidermis. This might also explain the H3K27me3 profile seen at YAB5, which remains active throughout the abaxial domain (Sarojam et al., 2010). Interestingly, no enrichment for H3K27me3 was detected at the ARF3 promoter ( Figure 3B). Analysis of additional regions upstream of the AS1-AS2 binding site and within the ARF3 coding sequence itself also revealed no H3K27me3 deposition (Supplemental Figure 3). This finding is consistent with data from genome-wide H3K27me3 profiling studies (Zhang et al., 2007;Roudier et al., 2011) and supports the idea that distinct mechanisms are employed to repress expression of MIR166A and YAB5, versus ARF3.
To determine whether H3K27me3 at MIR166A and YAB5 is deposited in an AS1-AS2-dependent manner, we compared the H3K27me3 profiles at these loci in the wild type and as2. Levels of H3K27me3 at ACT2 and STM are not significantly changed in as2 ( Figure 3B); the latter being consistent with the fact that H3K27me3 deposition at STM occurs in an AS1-AS2-independent manner . H3K27me3 levels along the promoters of both MIR166A and YAB5, however, are significantly reduced in as2 (A) qRT-PCR analysis shows MIR166A and YAB5 transcript levels are elevated in young leaves of as2, while ARF3 and TAS3A levels remain unchanged. Expression values (means 6 SE) normalized to the wild type were calculated based on at least three independent biological replicates (Student's t test; *P < 0.05).
(B) to (D) MIR166A:GUS reporter activity is restricted to the abaxial epidermis of wild-type primordia (B) but is ectopically expressed in the adaxial epidermis (arrows) in as2 (C) and in leaves of wild-type plants carrying a MIR166AD:GUS transgene with deleted AS1-AS2 binding sites (D). (E) ARF3:ARF3-GUS reporter activity subsides to virtually undetectable levels in mature leaves of the wild type but remains high in mature leaves of as2. (F) By contrast, TAS3A gene trap activity is similar in wild-type and as2 seedlings.

of 15
The Plant Cell ( Figure 3B). Given that AS1-AS2 physically interacts with multiple PRC2 components, these data present a model wherein AS1-AS2 guides PRC2 to the promoters of MIR166A and YAB5 to catalyze the trimethylation of H3K27, much like the mechanism of repression of BP and KNAT2 . This silencing mechanism prevents the accumulation of MIR166A and YAB5 on the adaxial side of developing primordia and identifies additional direct repressive interactions contributing to the mutual antagonism between adaxial and abaxial cell fates. Finally, while AS1-AS2 is required for the correct temporal repression of ARF3, this involves a distinct and PRC2-independent mechanism.
Convergent Regulation of ARF3 by the AS and ta-siRNA Pathways Affects Leaf Architecture To gain insight into the significance of ARF3 regulation by AS1-AS2, we took into consideration that ARF3 is also spatially regulated by tasiR-ARF ). Loss of this convergent regulation may be required to unmask the contribution of AS1-AS2. Indeed, a strongly synergistic phenotype is observed when as1 or as2 alleles are introduced into plants mutant for ta-siRNA biogenesis Xu et al., 2006). However, the basis for this phenotype is not fully understood. Given the sensitivity of the leaf to ARF3 dosage Hunter et al., 2006), our data suggest expanded ARF3 activity not only spatially, but also temporally due to loss of AS1-AS2 activity, as a potential explanation. An alternative, non-mutually exclusive explanation derives from the mutual antagonism between adaxial and abaxial fates. Misregulation of abaxial determinants resulting from loss of AS1-AS2, in combination with ectopic expression of the abaxial determinants ARF3 and ARF4, may confer enhanced abaxialized characteristics on double mutant leaves, resulting in adaxialto-abaxial cell fate switches. Therefore, to elucidate the consequences of AS1-AS2mediated regulation of ARF3, we began to assay the nature of the synergistic interaction between the AS and ta-siRNA pathways by constructing a double mutant between the as2-4 and sgs3-1 null alleles. Compared with the wild type ( Figures 4A, 4E, and 4I), as2 seedlings are slightly dwarfed and develop rumpled, rounded leaves with a reduced petiole ( Figures 4B, 4F, and 4J), whereas sgs3 seedlings form elongated, more rectangular, and downwardcurled leaves ( Figures 4C, 4G, and 4K). as2 sgs3 double mutants, like as2, have a reduced stature but produce smaller leaves with ridged blades that contain small outgrowths at their base ( Figure  4D). These defects become progressively more severe, such that mature leaves of 27-d-old as2 sgs3 plants appear highly dissected, with multiple outgrowths originating at the base of their blades ( Figures 4H and 4L). These outgrowths are partially flattened and appear to arise at or near the margins of the main axis. Moreover, outgrowth formation in as2 sgs3 leaves appears a reiterative process as, over time, secondary and tertiary outgrowths develop, always from proximal, marginal regions of established outgrowths ( Figure 4L).
To understand the basis for the synergistic as2 sgs3 phenotype, we first examined how regulation of ARF3 dosage might contribute. Expression analysis showed that ARF3 transcript levels in sgs3 are elevated ;2-fold ( Figure 4M). ARF3 expression in as2 sgs3 is increased even further, consistent with convergent regulation of ARF3 by the AS and ta-siRNA pathways. However, tasiR-ARF also targets transcripts of ARF2 and ARF4 (Williams et al., 2005). To distinguish whether the synergistic phenotype of as2 sgs3 is due to misregulation of multiple tasiR-ARF targets, or ARF3 specifically, we further analyzed the expression levels of ARF2 and ARF4. ARF2 transcript levels in 12-d-old seedling leaves did not vary significantly for any genotype, while ARF4 transcript levels are unchanged in as2 and increased ;4-fold in sgs3. However, in contrast to ARF3, transcript levels for ARF4 are not significantly different in developing as2 sgs3 versus sgs3 leaves ( Figure 4M).
This finding correlates with the fact that only ARF3 is a direct target of AS1-AS2 and suggests that misregulation of ARF3 specifically contributes to the synergistic leaf phenotype of as2 sgs3. This possibility was verified genetically, as loss of ARF3 function in as2 sgs3 completely suppresses their dissection, giving rise to triple mutant plants that resemble as2 in appearance  Table 4. (B) Levels of H3K27me3 at YAB5, MIR166A, and STM are enriched over the ACT2 negative control (black asterisks), whereas levels of H3K27me3 at ARF3 are indistinguishable from ACT2. H3K27me3 levels at YAB5 and MIR166A are reduced in as2 leaves relative to the wild type (red asterisks), whereas levels of H3K27me3 at the ACT2 and STM controls remain unchanged. H3K27me3 values (means 6 SE; n $ 3) are normalized to H3 levels to correct for potential variations in nucleosome density and then further normalized to the STM positive control. Student's t test: *P < 0.05; **P < 0.01. Diverse Regulatory Properties of AS1-AS2 ( Figure 4N). By contrast, as2 sgs3 arf4 triple mutants are indistinguishable from as2 sgs3 double mutants (Supplemental Figure  4A). In addition, we crossed as2 to a line expressing a tasiR-ARFresistant form of ARF3 under its native promoter (ARF3:ARF3mut; Fahlgren et al., 2006), which uncouples ARF3 regulation from other aspects of the sgs3 phenotype. ARF3:ARF3mut seedlings develop leaves that curl downward to a greater extent than sgs3 ( Figure 4O). Also, the phenotype of as2 ARF3:ARF3mut mutants is initially more severe than that of as2 sgs3, in that the degree of leaf dissection in as2 ARF3:ARF3mut seedlings is increased compared with as2 sgs3 (Figures 4D and 4P;Supplemental Figures 4B and 4C). However, 27-d-old as2 ARF3:ARF3mut and as2 sgs3 plants are largely indistinguishable (Figures 4H and 4Q). As expected, ARF4 transcript levels are unaffected in both ARF3:ARF3mut and as2 ARF3:ARF3mut seedlings ( Figure 4R). ARF3 transcripts, on the other hand, accumulate to significantly higher levels in (E) to (H) Twenty-seven-day-old wild-type (E), as2 (F), sgs3 (G), and as2 sgs3 (H) plants show this phenotype becomes progressively more severe.
(I) to (L) Compared with representative leaves of 27-d-old wild type (I), as2 (J), and sgs3 (K), leaves of as2 sgs3 (L) have a highly dissected morphology, developing secondary outgrowths (arrowheads) at the base of primary outgrowths (arrows). (M) qRT-PCR shows transcript levels of ARF2 are unchanged in young leaves of 12-d-old as2, sgs3, and as2 sgs3 seedlings relative to the wild type, whereas ARF3 and ARF4 expression is increased specifically in sgs3 and as2 sgs3. Note only ARF3 transcript levels are increased further in as2 sgs3 over sgs3. Note that ARF3 transcript levels are increased in as2 ARF3:ARF3mut over ARF3:ARF3mut. ARF4 transcript levels, by contrast, do not differ between any genotype. Expression values (means 6 SE) normalized to the wild type were calculated based on at least three independent biological replicates (Student's t test: ns, not significant; *P < 0.05; **P < 0.01). Asterisks in (D), (H), (L), (P), and (Q) indicate primary leaf blade; arrows indicate primary outgrowth; arrowheads indicate secondary outgrowth. Bars = 1 cm.

of 15
The Plant Cell ARF3:ARF3mut seedlings than in sgs3 ( Figures 4M and 4R), and these levels are further increased in as2 ARF3:ARF3mut double mutants. These data highlight the importance of ARF3 regulation by AS1-AS2 to leaf development. While loss of ta-siRNA regulation conditions misexpression of both ARF3 and ARF4, only prolonged ectopic expression of the AS1-AS2 target ARF3 is required for the altered leaf morphology seen in AS and ta-siRNA pathway double mutants. Supporting this, ARF3 is correctly positioned to regulate the formation of ectopic outgrowths, as the ARF3:ARF3-GUS reporter marks their sites of initiation in as2 sgs3 (Supplemental Figures 4D and 4E). However, ectopic ARF3 expression alone is not sufficient to condition the as2 sgs3 phenotype, as leaves of ARF3:ARF3mut do not show this defect ( Figure 4O; Fahlgren et al., 2006). Prolonged ARF3 activity is likewise insufficient, as as2 mutants, or plants constitutively expressing ARF3mut, also fail to develop this phenotype ( Figure 4F; Hunter et al., 2006). This indicates that another aspect of the as2 phenotype converges with altered ARF3 expression to regulate leaf architecture.

Contributions of ARF3 to Leaf Development beyond Adaxial-Abaxial Patterning
Indeed, an alternative, or complementary, hypothesis to explain the synergistic phenotype of leaves compromised for both the AS and ta-siRNA pathways is a further increase in abaxial identity. To evaluate this hypothesis, we compared the expression level of central adaxial and abaxial polarity determinants in developing wild-type, as2, sgs3, and as2 sgs3 leaves ( Figures 5A and 5B). Beyond the direct targets MIR166A and YAB5, none of the genes tested show a significant change in expression in as2 versus wildtype seedlings. Likewise, only transcript levels for the direct ta-siRNA targets ARF3 and ARF4 are significantly changed in sgs3. However, in as2 sgs3, transcript levels for YAB5 and ARF3 are increased above the levels seen in as2 and sgs3. In addition, expression of the abaxial determinants FIL and KAN1 is also elevated, suggesting as2 sgs3 leaves are abaxialized to a greater extent than either single mutant. Consistent with this finding, and the mutually antagonistic relationship between adaxial and abaxial cell fates, a concomitant reduction in expression of the adaxial determinants AGO7, TAS3A, PHB, REV, and AS2 is detected in as2 sgs3. Of particular note, the transcript levels of TAS3A and AGO7 are dramatically reduced specifically in the as2 sgs3 double mutant ( Figure 5B), hinting at a more complex relationship between ARF3 and the ta-siRNA pathway that patterns its activity.
A greater degree of abaxialization of developing as2 sgs3 leaves was also evident from analysis of FIL and TAS3A reporters. In agreement with the described FIL mRNA expression pattern (Siegfried et al., 1999), the FIL:GUS reporter accumulates on the abaxial side of wild-type primordia ( Figure 5C). This expression domain is unchanged in as2 and sgs3 ( Figures 5D and 5E), but FIL expression extends into the adaxial domain of as2 sgs3 primordia, consistent with a switch from adaxial to abaxial identity in these cells ( Figure 5F). Moreover, whereas the TAS3A gene trap is adaxially expressed in wild-type, as2, and sgs3 leaf primordia, its activity is severely reduced in as2 sgs3 (Figures 5G to 5J). Extremely weak TAS3A gene trap activity specifically in as2 sgs3 mutants is consistent with the dramatic reduction of TAS3A transcript levels in this genotype ( Figure 5B) and demonstrates that repression of this adaxial determinant occurs at the transcriptional level. This finding is particularly intriguing given that TAS3A is a direct target of AS1-AS2 and suggests that the contribution of these DNA binding factors to TAS3A regulation is only apparent when ARF3 and/or ARF4 is ectopically expressed. (A) qRT-PCR on young leaves of 12-d-old seedlings reveals transcript levels of many key abaxial determinants are synergistically elevated in as2 sgs3 over as2 and sgs3 single mutants. This elevation is suppressed for some, but not all, in as2 sgs3 arf3. (B) Conversely, transcript levels of many adaxial determinants are synergistically reduced in 12-d-old as2 sgs3 leaves relative to as2 and sgs3 and are partially returned in as2 sgs3 arf3. Note the dramatic reductions in TAS3A and AGO7 transcript levels specifically in as2 sgs3, which remain low in as2 sgs3 arf3. Expression values (means 6 SE) normalized to the wild type were calculated based on at least three independent biological replicates (Student's t test: *P < 0.05; **P < 0.01).
(C) to (F) FIL:GUS reporter activity is restricted to the abaxial domain of wild-type (C), as2 (D), and sgs3 (E) primordia but is ectopically expressed in the adaxial domain (arrow) of young as2 sgs3 leaves (F). (G) to (J) TAS3A gene trap activity in the two adaxial-most cell layers of wild-type (G), as2 (H), and sgs3 (I) primordia is barely detectable (arrow) in as2 sgs3 (J). Note: Images in (G) and (H) are taken from Figures 2J and 2K, respectively. Bars = 50 µm in (C) to (J).

Diverse Regulatory Properties of AS1-AS2 7 of 15
This prospect is in line with the earlier notion that AS1-AS2mediated regulation of TAS3A employs a mechanism distinct from that utilized to regulate MIR166A, YAB5, and ARF3. These data show that as2 sgs3 mutant leaves are molecularly abaxialized to a greater extent than leaves of either as2 or sgs3. To investigate whether this synergistic effect on adaxial-abaxial polarity is primarily responsible for the apparent dissection of as2 sgs3 leaves, we examined expression of adaxial and abaxial determinants also in as2 sgs3 arf3 triple mutants that show a complete suppression of this defect ( Figure 4N). Transcripts for select polarity determinants, including KAN1, REV, and AS2, no longer show a significant difference in expression in developing as2 sgs3 arf3 leaves compared with the wild type ( Figures 5A and  5B). However, transcript levels of the abaxial determinants FIL, YAB5, and ARF4, and the adaxial determinants AGO7, TAS3A, and PHB, remain significantly changed ( Figures 5A and 5B), indicating that as2 sgs3 arf3 triple mutants remain partially abaxialized at the molecular level, despite a complete suppression of the deep dissection of their leaves. This finding is consistent with the fact that ARF4, like ARF3, promotes abaxial identity (Pekker et al., 2005). It further shows that misexpression of ARF4 in an as2 background is sufficient to repress TAS3A and AGO7 expression, though not to the same extent as when ARF3 is simultaneously misexpressed. However, the fact that as2 sgs3 arf3 leaves remain partially abaxialized implies that a synergistic effect of mutations in both the AS and ta-siRNA pathways on adaxial-abaxial polarity is insufficient to condition the phenotype seen in as2 sgs3 and as2 ARF3:ARF3mut leaves. These observations point to a role for AS1-AS2 in the regulation of leaf dissection that is mediated via its direct target ARF3 but is distinct from its contribution to adaxialabaxial patterning. This regulation further identifies a function for ARF3 that is not shared by its close homolog ARF4.

Outgrowths on as2 sgs3 Leaves are Modified Serrations with Proximodistal Polarity
The notion that regulation of ARF3 by AS1-AS2, as well as tasiR-ARF, affects leaf morphology in a manner distinct from their role in adaxial-abaxial patterning is supported by the appearance of the as2 sgs3 outgrowths themselves. While mutants with defects in adaxial-abaxial polarity frequently develop ectopic outgrowths, these are new mediolateral axes that arise from ectopic adaxialabaxial boundaries on the upper or lower surface of the leaf blade and extend minimally (Waites and Hudson, 1995;Eshed et al., 2001;Pekker et al., 2005). Outgrowths on as2 sgs3 leaves, by contrast, appear to arise at or near the margins, are highly elaborated, and form in a reiterative fashion ( Figure 4L).
To further address the nature of these outgrowths, we analyzed their morphogenesis using scanning electron microscopy. Young leaves of 12-d-old as2 sgs3 seedlings show elaborations at the proximal margins of the blade that are not present in the wild type, as2, or sgs3 ( Figures 6A to 6D). Analysis of earlier primordial stages suggests these marginal outgrowths initiate as large dome-shaped structures at sites normally occupied by serrations, which by contrast, emerge as relatively flat and pointed structures ( Figures  6G and 6H). Moreover, unlike the simple protrusions seen in adaxialabaxial polarity mutants, these as2 sgs3 outgrowths eventually develop into structures with a well-defined proximodistal axis.
Epidermal cells of the petiole and blade of Arabidopsis leaves can be easily distinguished, as petiole cells have a rectangular morphology, oriented parallel to the proximodistal axis of the leaf, whereas blade cells are irregularly shaped with interdigitated lobes (Figure 6, insets). Despite their altered morphology, as2, sgs3, and as2 sgs3 leaves develop clearly defined petiole and blade regions ( Figures 6B  to 6D). Importantly, this morphology is also evident in as2 sgs3 outgrowths, as cells in the flattened distal area show blade characteristics while cells in the proximal regions resemble petiole cells ( Figure 6E). Even secondary outgrowths on as2 sgs3 leaves show

of 15
The Plant Cell distinct petiole and blade morphologies ( Figure 6F). Consistent with the enhanced abaxialized nature of this genotype, blades of individual outgrowths on as2 sgs3 leaves do show characteristics of adaxial-abaxial polarity mutants that are not obvious in either as2 or sgs3. These develop a rough upper surface with ridged blade-like outgrowths associated with reduced adaxial identity ( Figures 6D  and 6E), and occasionally individual outgrowths take on a trumpetshaped morphology ( Figure 6F). However, both expression and morphogenetic analyses indicate that the primary basis for the marginal proximodistal outgrowths on as2 sgs3 leaves is not a defect in adaxial-abaxial patterning.
The AS and ta-siRNA Pathways Converge to Suppress Leaf Complexity As defective adaxial-abaxial polarity is insufficient to explain the marginal outgrowths of as2 sgs3, what then might underlie their biogenesis, and how might AS1-AS2 regulation of ARF3 contribute? Considering ARF3 modulates the auxin response, and auxin maxima are known to drive outgrowth in a variety of developmental contexts, including the formation of leaves from the meristem, and leaflets or serrations from the leaf margin Reinhardt et al., 2003;Heisler et al., 2005;Koenig and Sinha, 2010;Bilsborough et al., 2011), auxin signaling emerges as a plausible contributor. We therefore examined the pattern of PIN-FORMED1 (PIN1) localization during primordium development using the PIN1:PIN1-GFP reporter . Within the sixth leaf of 10-and 12-d-old wild-type, as2, and sgs3 seedlings, PIN1-GFP is expressed at sites of incipient serrations (Supplemental Figures 5A and 6). Consistent with previous reports (Koenig and Sinha, 2010;Bilsborough et al., 2011), PIN1-GFP localization within the epidermis of serrations is polarized, predicting auxin is transported toward a convergence point at the serration tip. As normal leaves mature, PIN1-GFP expression dissipates and is undetectable in the same leaf of 16-d-old seedlings ( Figure 7A; Supplemental Figure 6). The PIN1-GFP dynamics in sgs3 are indistinguishable from the wild type ( Figure  7C; Supplemental Figure 6). In as2, however, PIN1-GFP activity persists and marks new serrations at the base of the sixth leaf even in 16-and 18-d-old plants ( Figure 7B; Supplemental Figure 6). Prolonged PIN1 activity is even more pronounced in as2 sgs3 leaves, where PIN1-GFP is detectable 27 d postgermination (Supplemental Figure 6). At 16 d, primary outgrowths on the sixth leaf of as2 sgs3 are significantly more elaborated than the serrations that occupy these sites in other genotypes and acquire discrete PIN1-GFP foci at their base ( Figure 7D). The pattern of PIN1-GFP polarization at these sites is reminiscent of that seen in young leaf primordia (Supplemental Figure 5B), consistent with these primary outgrowths possessing a functional proximodistal axis ( Figures 6E and 6F). Furthermore, these foci mark the sites of secondary outgrowths, which, upon elaboration, also generate new PIN1-GFP foci at their base (Supplemental Figure 5C). The spatial pattern and prolonged expression of PIN1-GFP thus correlates with the reiterative nature of outgrowths on leaves perturbed in both the AS and ta-siRNA pathways, predicting a role for polar auxin transport in their formation. To directly test this prediction, the severe as2 ARF3:ARF3mut double mutant was grown on media containing the auxin transport inhibitor 1-N-naphthylphthalamic acid (NPA). Wild-type, as2, and ARF3:ARF3mut seedlings show minor changes in leaf shape in response to NPA treatment (Figures 7E to 7G and 7I to 7K). Formation of elaborated outgrowths on as2 ARF3:ARF3mut leaves, by contrast, is completely suppressed ( Figures 7H and 7L), indicating that PIN1-derived auxin transport is required for their formation.
The fact that the outgrowths in as2 sgs3 are elaborated serrations with a well-defined proximal-distal axis suggests parallels between these outgrowths and leaflets of compound-leafed species. Formation of leaflets in most compound-leafed species is also associated with expression of indeterminacy-promoting KNOX genes in leaf primordia (Bharathan et al., 2002;Barkoulas et al., 2008). In this regard, it is interesting that ectopic ARF3 activity alone is insufficient to drive the formation of proximodistal outgrowths and that another aspect of the as2 phenotype is required. AS1-AS2 mediates the repression of both BP and KNAT2 in developing leaves , though only the levels of BP change dramatically in as2 (Ori et al., 2000;Semiarti et al., 2001;Lodha et al., 2013). Indeed, while unaffected in sgs3, BP is strongly misexpressed in as2 sgs3 and as2 sgs3 arf3 leaves (Supplemental Figure 5D).
Taken together, these data reveal that the direct regulation of ARF3 by AS1-AS2 contributes in a surprisingly multifaceted way to leaf development. The AS pathway converges with the ta-siRNA (A) to (D) Confocal imaging of 16-d-old PIN1:PIN1-GFP seedlings shows that, unlike the wild type (A) and sgs3 (C), the fifth leaves of as2 (B) and as2 sgs3 (D) seedlings retain foci of PIN1-GFP activity at their margin. In as2 sgs3, this PIN1-GFP signal correlates with formation of secondary outgrowths (arrowheads) from primary outgrowths (arrow). Diverse Regulatory Properties of AS1-AS2 9 of 15 pathway to limit ARF3 expression both spatially and temporally. Loss of this convergent regulation results in an enhanced abaxialized phenotype, characterized by a dramatic reduction in TAS3A transcription, as well as an ARF3-dependent increase in leaf complexity. These phenotypes show that the shared repression of ARF3 by the AS and ta-siRNA pathways intersects with additional AS1-AS2 targets to affect multiple nodes in leaf development, thereby keeping the Arabidopsis leaf both flat and simple.

AS1-AS2 Uses Distinct Mechanisms to Regulate Its Polarity Targets
The control of adaxial-abaxial leaf polarity must be extremely robust. In addition to differentiating distinct cell types on the leaf's top and bottom faces, the boundary between adaxial and abaxial fates provides positional information that drives the flattened outgrowth of the leaf blade (Waites and Hudson, 1995;Nakata et al., 2012). Flatness is not a default state but results from precise coordination of cell division and differentiation, such that even slight perturbations in adaxial-abaxial polarity lead to leaf curling and other morphological changes. Formation of the flat leaf thus poses an interesting mechanistic challenge, namely, how to create a stable boundary throughout the plane of a long and wide, yet shallow, structure. The clean separation of adaxial and abaxial fates, at both the cell and the domain level, relies on positive and negative feedback regulation between polarity determinants that reinforce initial cell fate decisions (Husbands et al., 2009). However, few of these regulatory interactions are understood at the mechanistic level. In particular, the molecular relationships between the conserved transcription factors at the core of the adaxial-abaxial polarity network remain poorly understood.
Here, we investigated the contribution of AS1-AS2 to adaxialabaxial patterning and show that MIR166A, YAB5, ARF3, and TAS3A are direct targets for this adaxial determinant. Interestingly, AS1-AS2 makes discrete contributions to the regulation of these targets and employs distinct molecular mechanisms to regulate their expression. At MIR166A, AS1-AS2 prevents transcription in the adaxial epidermal layer and is thus required for the correct spatial regulation of this microRNA. AS1-AS2 may also help define the spatial expression domain of YAB5. Like MIR166A, YAB5 expression is normally restricted to the abaxial side of developing leaves (Sarojam et al., 2010) and increases in as2 mutants. Repression of both MIR166A and YAB5 by AS1-AS2 involves a PRC2-dependent mechanism, perhaps reminiscent of the AS1-AS2-based repression of the KNOX targets BP and KNAT2 .
The spatial restriction of abaxial factors like MIR166A, and possibly YAB5, by the adaxial factor AS1-AS2 reveals new direct interactions within the polarity network that are in line with the mutually exclusive nature of these two cell fates. The effect of AS1-AS2 on ARF3 regulation, on the other hand, appears to be temporal in nature, ensuring abaxially localized ARF3 subsides as the leaf matures. This raises the question of how adaxial determinants might directly repress a target in the complementary abaxial domain. One explanation follows from the dynamic expression pattern of AS2, which overlaps with AS1 throughout initiating leaf primordia before being localized to the adaxial side (Iwakawa et al., 2007). This initially uniform accumulation, which nucleates epigenetic change at BP and KNAT2 , may similarly initiate epigenetic change at ARF3, although not via PRC2. Given the duration of ARF3 expression, this repression would also have to proceed more slowly, perhaps employing a mechanism similar to the H3K9me2-depositing complex GIANT KILLER, which gradually silences ARF3 in the inflorescence (Ng et al., 2009). Alternatively, AS1-AS2 is known to form complexes with several distinct protein partners, including other chromatin regulators, presenting several potential mechanisms via which to mediate the temporal regulation of ARF3 (Yang et al., 2008;Luo et al., 2012;Rast and Simon, 2012).
Interestingly, AS1-AS2 also binds the promoter of TAS3A, an adaxial determinant. The expression domain of TAS3A mirrors that of AS2, and its transcriptional output remains unchanged in as2 mutants. These data indicate that AS1-AS2 does not always function in a repressor complex and has varied effects on target transcription, a behavior shared by many transcription factors (reviewed in Biggin, 2011). Moreover, meta-analyses of transcription factor occupancy across the genome reveal that these proteins are often present at considerably more loci than they are known to regulate. In particular, the many weak, low-occupancy interactions that transcription factors make may not directly regulate transcription at nearby genes. Rather, such interactions are thought to affect transcription indirectly by promoting or inhibiting the recruitment of other transcription factors (Biggin, 2011).
Our findings suggest that AS1-AS2 may, in fact, contribute to the regulation of TAS3A via this latter mechanism. Transcript levels of TAS3A are reduced far beyond that of other adaxial determinants when ARF3 or ARF4 are adaxially expressed in an as2 background. Like other ARF proteins, ARF3 and ARF4 bind auxin response elements, but repress rather than activate target transcription (Tiwari et al., 2003). Interestingly, the TAS3A promoter contains a canonical auxin response and enhancer element pair immediately flanked by the AS1-AS2 binding sites, presenting the possibility that AS1-AS2 uses a "protective mechanism" to prevent downregulation of TAS3A on the adaxial side via stearic hindrance of ARF binding to its promoter. In this scenario, only in an as1 or as2 background would ARF3 and ARF4 be able to bind its cis-elements and downregulate TAS3A on the adaxial side. Additionally, repression of TAS3A may result from cooperativity between ARF3/ARF4 and factors that are misexpressed only in as2. One such candidate is KAN1, which physically interacts with ARF3 (Kelley et al., 2012), and whose expression extends into the adaxial side of as2 leaves (Wu et al., 2008). The fact that TAS3A expression persists in as2 seems to favor the requirement of a cofactor during the ARF3/ARF4-mediated repression of TAS3A. This hypothesis also nicely explains why ubiquitous ARF3 expression is insufficient to trigger adaxial-to-abaxial cell fate switches on its own Hunter et al., 2006).
Thus, AS1-AS2 employs at least three distinct mechanisms to regulate its polarity targets: (1) a Polycomb-mediated mechanism in the spatial regulation of targets such as MIR166A and YAB5; (2) a non-Polycomb-mediated, but still potentially epigenetic, mechanism in the temporal repression of ARF3; and (3) a nonrepressive, and possibly protective, mode of regulation at TAS3A.

of 15
The Plant Cell

Further Roles for AS1-AS2 in Adaxial-Abaxial Patterning
How might these molecular interactions contribute to adaxialabaxial patterning? Adaxial-abaxial polarity is already evident within the incipient primordium (or P0), during which key polarity determinants, such as the HD-ZIPIII transcription factors, show a polar localization (Juarez et al., 2004;Heisler et al., 2005). The outcomes of surgical experiments, however, predict a brief developmental window during which organ polarity is resolved (Sussex, 1951;Reinhardt et al., 2005). Once in place, this polar axis is then stably maintained until the leaf differentiates. Given the severity of mutants expressing a miR166-resistant HD-ZIPIII allele (McConnell et al., 2001), spatial regulation of HD-ZIPIII expression, as patterned by polarized miR166 activity, forms a critical component in the establishment of adaxial-abaxial polarity. The detection of MIR166A:GUS reporter activity at the earliest stages of leaf development (Supplemental Figure 2A), and the strongly adaxialized phenotype of plants ubiquitously expressing AS2 (Lin et al., 2003), support a prominent role for this MIR166 member in particular early in adaxial-abaxial patterning. This indicates that MIR166A expression must somehow be polarized at leaf initiation, which raises an intriguing question: How could AS1-AS2 silence MIR166A specifically on the adaxial side while silencing KNOX targets uniformly across the primordium? This might imply the existence of additional polarized cofactors that promote silencing of MIR166A in adaxial cells or block its repression in abaxial cells. However, the adaxialized phenotype resulting from ectopic expression of AS2 (Lin et al., 2003) demonstrates AS1-AS2 can function on both sides of developing primordia. Rather, the polarized repression of MIR166A may reflect inherent properties of chromatin dynamics.
The switch from active to repressive chromatin is thought to involve a bistable epigenetic state that is resolved once nucleation of PRC2 elevates H3K27me3 levels beyond a certain threshold, triggering the spread of this repressive mark across the locus (Angel et al., 2011). AS1 and AS2 are only briefly coexpressed throughout a new primordium before AS2 expression is restricted to the adaxial side (Iwakawa et al., 2007). Thus, the polarized repression of MIR166A may reflect a requirement for persistent AS1-AS2 occupancy before H3K27me3 levels can surpass the threshold needed to trigger stable silencing. This hypothesis can explain why prolonged misexpression of AS2 across developing primordia conditions a strongly adaxialized phenotype (Lin et al., 2003). That repression of BP and KNAT2 is nonpolar, despite employing a similar regulatory mechanism, might suggest that the establishment of a stable state occurs more rapidly at these loci. In this regard, it is intriguing that both KNOX promoters contain two AS1-AS2 binding sites that act cooperatively to mediate this silencing , whereas AS1-AS2 occupies a single site at MIR166A. This paradigm also fits the regulatory dynamics of YAB5, which is similarly bound by AS1-AS2 at a single site and remains active in abaxial cells (Sarojam et al., 2010).
A key question then is how the activity of AS2 becomes polarized in developing primordia. KAN proteins directly repress AS2 expression (Wu et al., 2008), making them likely candidates to set up this polarized activity. Indeed, KAN dynamics appear complementary to AS2, with transcripts accumulating on the abaxial side of developing leaves (Kerstetter et al., 2001;Wu et al., 2008). Furthermore, KAN proteins are excluded from the P0, which is defined by high levels of the KAN1-repressed target PIN1, and instead accumulate in a ring below initiating organs (Merelo et al., 2013;Yadav et al., 2014).
Taken together, these observations suggest a model that incorporates expression dynamics of key adaxial and abaxial determinants to explain the establishment and resolution of polarity (Figure 8). The exclusion of KAN proteins from the P0 allows the nonpolar accumulation of AS1-AS2 and the initial nucleation of PRC2-mediated H3K27me3 at the promoter of MIR166A on both sides of the primordium. As the primordium develops, KAN expression extends into its abaxial side (Yadav et al., 2014), directly repressing AS2 expression and restricting further deposition of H3K27me3 at MIR166A solely to the adaxial side. This polarizes the production of mature miR166 and consequently limits HD-ZIPIII accumulation to the adaxial side. However, this model implies that MIR166A itself is not expressed in the P0, raising the question of where miR166, which patterns HD-ZIPIII activity in the incipient primordium, originates. One possibility is that miR166 originates in the KAN-expressing region below the incipient leaf, an idea first posited in maize (Zea mays; Juarez et al., 2004). Indeed, MIR166A is a direct target of KAN1 and is not downregulated by this repressor protein (Merelo et al., 2013). Mobility of mature miR166 from a primordiumindependent source could provide the positional information required to pattern HD-ZIPIII activity until the mechanisms within the primordium itself resolve to maintain adaxial-abaxial polarity. This can also explain why loss of AS1-AS2-directed silencing of MIR166A within the primordium does not condition a strong The incipient primordium (left panel) is polarized by a primordiumindependent source of miR166, which moves from the internode below into the P0 and limits HD-ZIPIII activity to the adaxial side. This initial polarity is resolved from an externally to an internally patterned process as the primordium develops and KAN proteins become active on the abaxial side. KAN proteins, possibly in complex with ARF3, restrict AS2 expression and thereby AS1-AS2 activity to the adaxial side, polarizing its downstream effects (middle panel). Once resolved, polarity is then stably maintained throughout primordium development through opposing miR166 and tasiR-ARF gradients acting on the domain level and mutually antagonistic interactions and positive feedback regulation of polarity determinants at the cellular level (right panel). More detailed descriptions of these interactions are presented in the text. T-bars denote direct repressive interactions, whereas the dotted arrow denotes a nonrepressive, possibly protective, interaction. The dumbbell denotes protein-protein interaction.
Diverse Regulatory Properties of AS1-AS2 11 of 15 HD-ZIPIII loss-of-function phenotype, as AS1-AS2 is not predicted to act outside the incipient leaf. The model thus predicts that sequential polarization of polarity factors is critical to adaxial-abaxial patterning. This process is established by polarity determinants originating from outside the incipient leaf, but as the primordium develops, polar expression patterns of adaxial and abaxial determinants within the primordium itself resolve. These are then stably maintained throughout leaf development by further positive and negative feedback between the polarity determinants (Figure 8).
A protective mechanism of action for AS1-AS2 at TAS3A (see above) is also in line with our current understanding of the spatiotemporal dynamics of the polarity network. ARF3 is present throughout the incipient leaf, as tasiR-ARF biogenesis is delayed until later in primordium development . The exclusion of KAN activity from the P0, however, predicts ARF3directed TAS3A repression is blocked at this developmental stage, as AS1-AS2 is active throughout. Subsequent extension of KAN activity into the primordium would then limit the effect of AS1-AS2 on TAS3A to the adaxial side and permit ARF3/ARF4mediated repression, possibly aided by a cofactor, on the abaxial side ( Figure 8). This creates a feedback loop wherein the targets of tasiR-ARF polarize its biogenesis, thereby reinforcing the small RNA gradient that patterns them.
Thus, the dynamic localization of AS1-AS2 contributes to the polarization of MIR166A as well as the TAS3A-derived tasiR-ARF gradient. Intercellular movement of these small RNAs provides positional information to separate adaxial and abaxial fates at the domain level, by polarizing HD-ZIPIII and ARF3/ARF4 accumulation, respectively. As such, AS1-AS2 stabilizes the separation of the adaxial and abaxial domains as polarity in the developing primordium transitions from an externally to an internally patterned process ( Figure 8). In addition, AS1-AS2 reinforces adaxial identity at the cellular level through direct, Polycomb-mediated repression of the abaxial determinant YAB5. However, as YABBY genes display considerable variability in expression patterns across species (Husbands et al., 2009), the regulatory relationships between them and orthologous AS1-AS2 complexes remain an open and intriguing question. Taken together, our data reveal how the versatile regulatory properties of AS1-AS2 impact the robust placement of the adaxial-abaxial boundary, contributing to the production of a flat leaf.

The AS and ta-siRNA Pathways Intersect at Multiple Developmental Nodes
Our analyses further show that AS1-AS2 intersects with the ta-siRNA pathway to regulate ARF3, contributing to leaf development in a surprisingly multifaceted way. Loss of both pathways results in enhanced abaxialized defects, which are not simply explained by the synergistic misexpression of their shared target ARF3, but instead require misregulation of additional polarity targets of AS1-AS2. Similarly, the synergistic loss of TAS3A transcription occurs in response to ectopic ARF3/ARF4 expression, but only when AS1-AS2 activity is simultaneously compromised. In this paradigm, however, the contribution that AS1-AS2 makes is mechanistically distinct (see above).
Double mutants for the AS and ta-siRNA pathways also show a previously unappreciated convergence at leaf complexity. While links between adaxial-abaxial polarity and leaf complexity have been noted (Kim et al., 2003;Yifhar et al., 2012), our data indicate that altered adaxial-abaxial patterning is unlikely to drive this phenotype. Much like the examples described above, the increase in leaf complexity derives from the misregulation of their shared target ARF3 and the presence of additional factor(s) normally repressed by AS1-AS2. Candidates of particular interest include the AS1-AS2 target YAB5, which promotes outgrowth at the leaf margin (Sarojam et al., 2010); KAN1, which physically interacts with ARF3 (Kelley et al., 2012); and the KNOX targets, which already have a well-established role in compound leaf development (Bharathan et al., 2002;Barkoulas et al., 2008). Interestingly, reduced auxin signaling deepens serrations of as1 leaves in a BPdependent manner . When considered in the light of our data, it is thus possible that complexity emerges from the intersection of ARF3-dependent repression of auxin signaling and KNOX gene activity at the margins of developing leaves.
At what point during compound leaf development might intersection of AS1-AS2 target(s), like the KNOX genes, and ARF3 occur? Increases to leaf complexity derive from PIN1-generated auxin maxima found along the margin of developing primordia whose positioning is achieved, at least in part, through a combination of CUC2 and Aux/IAA activity (Barkoulas et al., 2008;Koenig and Sinha 2010;Bilsborough et al., 2011). As ARF3 requires polar auxin transport to mediate its effect on leaflet formation, and PIN1-GFP localization is unchanged in sgs3, ARF3 is unlikely to act at this point in the pathway. Once auxin maxima are positioned, isolation of outgrowth is reinforced through the activities of CUC2 and RCO, which repress rates of cell division in flanking regions (Koenig and Sinha 2010;Bilsborough et al., 2011;Sicard et al., 2014;Vlad et al., 2014), presenting another point where ARF3 may play its role. However, ARF3 is expressed throughout the developing outgrowth, rather than at its base, and does not affect leaf dissection when overexpressed in isolation, consistent with a different mode of action than either CUC2 or RCO. Instead, we propose that ARF3, in conjunction with additional AS pathway targets, promotes organogenic potential at marginal auxin maxima, such that its prolonged activity results in elaboration of serrations into leaflets. Such a role for ARF3 may be evolutionarily conserved, as hypomorphic mutations in ta-siRNA biogenesis components in tomato (Solanum lycopersicum) can condition mild increases in leaf complexity (Yifhar et al., 2012). Thus, our analyses show that the shared repression of ARF3 by the AS and ta-siRNA pathways intersects with multiple additional AS1-AS2 targets to affect discrete nodes in leaf development.
Taken together with its effects on MIR166A, YAB5, and TAS3A, this study illustrates the surprisingly multifaceted contribution of AS1-AS2 to leaf development, showing that, in conjunction with the ta-siRNA pathway, these transcription factors affect adaxialabaxial polarity as well as leaf complexity, thereby keeping the Arabidopsis leaf both flat and simple.

Plant Materials and Growth Conditions
Plants were grown at 22°C under long-day conditions in soil or on 1% agarose plates containing 13 Murashige and Skoog medium supplemented 12 of 15 The Plant Cell

Supplemental Data
Supplemental Figure 1. Promoter regions of polarity determinants occupied by AS2 are also bound by AS1.
Supplemental Figure 2. AS1-AS2 controls the expression domain of its direct target MIR166A.
Supplemental Figure 3. No enrichment of H3K27me3 is detected at ARF3.
Supplemental Figure 4. ARF3 expression is detectable in outgrowths on as2 sgs3 leaves and correlates with the degree of leaf dissection.
Supplemental Table 1. Sequences and locations of AS1-AS2 Motifs I and II.
Supplemental Table 2. Primer pairs used in promoter:GUS fusion constructs.
Supplemental Table 4. Primer pairs used in H3K27me3 ChIP analysis.
Supplemental Table 5. Primers used for quantitative RT-PCR.